Digital PDFs

DS-0038B-TE

March 2002

356 pages

Original

4.1MB

Document:	21264/EV68A Microprocessor Hardware Reference Manual
Order Number:	DS-0038B-TE
Revision:	0
Pages:	356
Original Filename:

OCR Text

21264/EV68A Microprocessor
Hardware Reference Manual
Part Number: DS–0038B–TE

This manual is directly derived from the internal 21264/EV68A Specifications, Revision 1.1. You can access this hardware reference manual in PDF format from the
following site:

ftp://ftp.compaq.com/pub/products/alphaCPUdocs

Revision/Update Information:

Revision 1.1, March 2002

Compaq Computer Corporation
Shrewsbury, Massachusetts

March 2002
The information in this publication is subject to change without notice.
COMPAQ COMPUTER CORPORATION SHALL NOT BE LIABLE FOR TECHNICAL OR EDITORIAL
ERRORS OR OMISSIONS CONTAINED HEREIN, NOR FOR INCIDENTAL OR CONSEQUENTIAL DAMAGES RESULTING FROM THE FURNISHING, PERFORMANCE, OR USE OF THIS MATERIAL. THIS
INFORMATION IS PROVIDED “AS IS” AND COMPAQ COMPUTER CORPORATION DISCLAIMS ANY
WARRANTIES, EXPRESS, IMPLIED OR STATUTORY AND EXPRESSLY DISCLAIMS THE IMPLIED WARRANTIES OF MERCHANTABILITY, FITNESS FOR PARTICULAR PURPOSE, GOOD TITLE AND AGAINST
INFRINGEMENT.
This publication contains information protected by copyright. No part of this publication may be photocopied or
reproduced in any form without prior written consent from Compaq Computer Corporation.
© Compaq Computer Corporation 2002.
All rights reserved. Printed in the U.S.A.
COMPAQ, the Compaq logo, the Digital logo, and VAX Registered in United States Patent and Trademark Office.
Pentium is a registered trademark of Intel Corporation.
Other product names mentioned herein may be trademarks and/or registered trademarks of their respective companies.

21264/EV68A Hardware Reference Manual

Table of Contents
Preface
1

Introduction
1.1
1.1.1
1.1.2
1.1.3
1.2

The Architecture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Addressing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Integer Data Types. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Floating-Point Data Types . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
21264/EV68A Microprocessor Features . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

1–1
1–2
1–2
1–2
1–3

Internal Architecture
2.1
21264/EV68A Microarchitecture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.1.1
Instruction Fetch, Issue, and Retire Unit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.1.1.1
Virtual Program Counter Logic . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.1.1.2
Branch Predictor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.1.1.3
Instruction-Stream Translation Buffer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.1.1.4
Instruction Fetch Logic . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.1.1.5
Register Rename Maps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.1.1.6
Integer Issue Queue . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.1.1.7
Floating-Point Issue Queue . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.1.1.8
Exception and Interrupt Logic . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.1.1.9
Retire Logic . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.1.2
Integer Execution Unit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.1.3
Floating-Point Execution Unit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.1.4
External Cache and System Interface Unit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.1.4.1
Victim Address File and Victim Data File . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.1.4.2
I/O Write Buffer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.1.4.3
Probe Queue. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.1.4.4
Duplicate Dcache Tag Array . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.1.5
Onchip Caches. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.1.5.1
Instruction Cache . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.1.5.2
Data Cache . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.1.6
Memory Reference Unit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.1.6.1
Load Queue . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.1.6.2
Store Queue . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.1.6.3
Miss Address File . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.1.6.4
Dstream Translation Buffer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.1.7
SROM Interface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.2
Pipeline Organization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.2.1
Pipeline Aborts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.3
Instruction Issue Rules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

21264/EV68A Hardware Reference Manual

2–1
2–2
2–2
2–3
2–5
2–6
2–6
2–6
2–7
2–8
2–8
2–8
2–10
2–11
2–11
2–11
2–11
2–11
2–11
2–11
2–12
2–12
2–13
2–13
2–13
2–13
2–13
2–13
2–16
2–16

iii

2.3.1
Instruction Group Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.3.2
Ebox Slotting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.3.3
Instruction Latencies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.4
Instruction Retire Rules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.4.1
Floating-Point Divide/Square Root Early Retire . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.5
Retire of Operate Instructions into R31/F31 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.6
Load Instructions to R31 and F31 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.6.1
Normal Prefetch: LDBU, LDF, LDG, LDL, LDT, LDWU, HW_LDL Instructions . . . . . . .
2.6.2
Prefetch with Modify Intent: LDS Instruction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.6.3
Prefetch, Evict Next: LDQ and HW_LDQ Instructions . . . . . . . . . . . . . . . . . . . . . . . . . .
2.7
Special Cases of Alpha Instruction Execution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.7.1
Load Hit Speculation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.7.2
Floating-Point Store Instructions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.7.3
CMOV Instruction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.8
Memory and I/O Address Space Instructions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.8.1
Memory Address Space Load Instructions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.8.2
I/O Address Space Load Instructions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.8.3
Memory Address Space Store Instructions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.8.4
I/O Address Space Store Instructions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.9
MAF Memory Address Space Merging Rules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.10
Instruction Ordering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.11
Replay Traps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.11.1
Mbox Order Traps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.11.1.1
Load-Load Order Trap . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.11.1.2
Store-Load Order Trap . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.11.2
Other Mbox Replay Traps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.12
I/O Write Buffer and the WMB Instruction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.12.1
Memory Barrier (MB/WMB/TB Fill Flow) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.12.1.1
MB Instruction Processing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.12.1.2
WMB Instruction Processing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.12.1.3
TB Fill Flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.13
Performance Measurement Support—Performance Counters . . . . . . . . . . . . . . . . . . . . . . .
2.14
Floating-Point Control Register . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.15
AMASK and IMPLVER Instruction Values . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.15.1
AMASK. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.15.2
IMPLVER . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.16
Design Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Hardware Interface
3.1
3.2
3.3
3.4
3.5

21264/EV68A Microprocessor Logic Symbol . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
21264/EV68A Signal Names and Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Pin Assignments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Mechanical Specifications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
21264/EV68A Packaging . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

3–1
3–3
3–8
3–17
3–18

Cache and External Interfaces
4.1
Introduction to the External Interfaces. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.1.1
System Interface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.1.1.1
Commands and Addresses. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.1.2
Second-Level Cache (Bcache) Interface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.2
Physical Address Considerations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.3
Bcache Structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.3.1
Bcache Interface Signals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.3.2
System Duplicate Tag Stores. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

2–17
2–18
2–20
2–21
2–22
2–22
2–23
2–23
2–23
2–24
2–24
2–24
2–26
2–26
2–27
2–27
2–27
2–28
2–29
2–30
2–30
2–31
2–31
2–31
2–31
2–32
2–32
2–32
2–33
2–33
2–34
2–35
2–35
2–37
2–38
2–38
2–38

4–1
4–3
4–4
4–4
4–4
4–7
4–7
4–7

21264/EV68A Hardware Reference Manual

4.4
Victim Data Buffer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.5
Cache Coherency . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.5.1
Cache Coherency Basics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.5.2
Cache Block States . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.5.3
Cache Block State Transitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.5.4
Using SysDc Commands . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.5.5
Dcache States and Duplicate Tags . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.6
Lock Mechanism . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.6.1
In-Order Processing of LDx_L/STx_C Instructions . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.6.2
Internal Eviction of LDx_L Blocks. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.6.3
Liveness and Fairness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.6.4
Managing Speculative Store Issues with Multiprocessor Systems . . . . . . . . . . . . . . . .
4.7
System Port . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.7.1
System Port Pins . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.7.2
Programming the System Interface Clocks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.7.3
21264/EV68A-to-System Commands . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.7.3.1
Bank Interleave on Cache Block Boundary Mode . . . . . . . . . . . . . . . . . . . . . . . . .
4.7.3.2
Page Hit Mode . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.7.4
21264/EV68A-to-System Commands Descriptions . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.7.5
ProbeResponse Commands (Command[4:0] = 00001) . . . . . . . . . . . . . . . . . . . . . . . . .
4.7.6
SysAck and 21264/EV68A-to-System Commands Flow Control . . . . . . . . . . . . . . . . . .
4.7.7
System-to-21264/EV68A Commands . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.7.7.1
Probe Commands (Four Cycles) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.7.7.2
Data Transfer Commands (Two Cycles). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.7.8
Data Movement In and Out of the 21264/EV68A . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.7.8.1
21264/EV68A Clock Basics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.7.8.2
Fast Data Mode . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.7.8.3
Fast Data Disable Mode . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.7.8.4
SysDataInValid_L and SysDataOutValid_L . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.7.8.5
SysFillValid_L . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.7.8.6
Data Wrapping . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.7.9
Nonexistent Memory Processing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.7.10
Ordering of System Port Transactions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.7.10.1
21264/EV68A Commands and System Probes . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.7.10.2
System Probes and SysDc Commands . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.8
Bcache Port . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.8.1
Bcache Port Pins . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.8.2
Bcache Clocking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.8.2.1
Setting the Period of the Cache Clock . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.8.3
Bcache Transactions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.8.3.1
Bcache Data Read and Tag Read Transactions . . . . . . . . . . . . . . . . . . . . . . . . . .
4.8.3.2
Bcache Data Write Transactions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.8.3.3
Bubbles on the Bcache Data Bus . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.8.4
Pin Descriptions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.8.4.1
BcAdd_H[23:4] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.8.4.2
Bcache Control Pins . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.8.4.3
BcDataInClk_H and BcTagInClk_H . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.8.5
Bcache Banking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.8.6
Disabling the Bcache for Debugging . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.9
Interrupts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

4–8
4–8
4–8
4–9
4–10
4–11
4–13
4–14
4–15
4–15
4–15
4–16
4–16
4–17
4–18
4–19
4–19
4–20
4–21
4–24
4–25
4–26
4–26
4–28
4–30
4–30
4–31
4–33
4–34
4–35
4–36
4–38
4–40
4–40
4–42
4–42
4–43
4–44
4–45
4–47
4–47
4–48
4–49
4–50
4–51
4–51
4–53
4–53
4–53
4–54

Internal Processor Registers
5.1
5.1.1
5.1.2
5.1.3

Ebox IPRs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Cycle Counter Register – CC . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Cycle Counter Control Register – CC_CTL . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Virtual Address Register – VA . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

21264/EV68A Hardware Reference Manual

5–3
5–3
5–3
5–4

5.1.4
5.1.5
5.2
5.2.1
5.2.2
5.2.3
5.2.4
5.2.5
5.2.6
5.2.7
5.2.8
5.2.9
5.2.10
5.2.11
5.2.12
5.2.13
5.2.14
5.2.15
5.2.16
5.2.17
5.2.18
5.2.19
5.2.20
5.2.21
5.2.22
5.3
5.3.1
5.3.2
5.3.3
5.3.4
5.3.5
5.3.6
5.3.7
5.3.8
5.3.9
5.3.10
5.3.11
5.4
5.4.1
5.4.2
5.4.3
5.4.4
5.4.5

5–4
5–5
5–6
5–6
5–6
5–7
5–7
5–7
5–8
5–8
5–9
5–9
5–10
5–11
5–12
5–13
5–15
5–15
5–18
5–21
5–21
5–21
5–21
5–21
5–23
5–25
5–25
5–26
5–26
5–27
5–27
5–27
5–28
5–28
5–29
5–30
5–31
5–32
5–33
5–33
5–33
5–38
5–41

Privileged Architecture Library Code
6.1
6.2
6.3
6.4
6.4.1
6.4.2
6.4.3
6.4.4
6.5
6.5.1
6.5.2

Virtual Address Control Register – VA_CTL . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Virtual Address Format Register – VA_FORM. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Ibox IPRs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
ITB Tag Array Write Register – ITB_TAG . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
ITB PTE Array Write Register – ITB_PTE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
ITB Invalidate All Process (ASM=0) Register – ITB_IAP . . . . . . . . . . . . . . . . . . . . . . . .
ITB Invalidate All Register – ITB_IA . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
ITB Invalidate Single Register – ITB_IS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
ProfileMe PC Register – PMPC . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Exception Address Register – EXC_ADDR . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Instruction Virtual Address Format Register — IVA_FORM. . . . . . . . . . . . . . . . . . . . . .
Interrupt Enable and Current Processor Mode Register – IER_CM. . . . . . . . . . . . . . . .
Software Interrupt Request Register – SIRR . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Interrupt Summary Register – ISUM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Hardware Interrupt Clear Register – HW_INT_CLR . . . . . . . . . . . . . . . . . . . . . . . . . . .
Exception Summary Register – EXC_SUM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
PAL Base Register – PAL_BASE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Ibox Control Register – I_CTL . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Ibox Status Register – I_STAT. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Icache Flush Register – IC_FLUSH . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Icache Flush ASM Register – IC_FLUSH_ASM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Clear Virtual-to-Physical Map Register – CLR_MAP . . . . . . . . . . . . . . . . . . . . . . . . . . .
Sleep Mode Register – SLEEP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Process Context Register – PCTX . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Performance Counter Control Register – PCTR_CTL . . . . . . . . . . . . . . . . . . . . . . . . . .
Mbox IPRs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
DTB Tag Array Write Registers 0 and 1 – DTB_TAG0, DTB_TAG1 . . . . . . . . . . . . . . .
DTB PTE Array Write Registers 0 and 1 – DTB_PTE0, DTB_PTE1 . . . . . . . . . . . . . . .
DTB Alternate Processor Mode Register – DTB_ALTMODE. . . . . . . . . . . . . . . . . . . . .
Dstream TB Invalidate All Process (ASM=0) Register – DTB_IAP . . . . . . . . . . . . . . . .
Dstream TB Invalidate All Register – DTB_IA . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Dstream TB Invalidate Single Registers 0 and 1 – DTB_IS0,1 . . . . . . . . . . . . . . . . . . .
Dstream TB Address Space Number Registers 0 and 1 – DTB_ASN0,1 . . . . . . . . . . .
Memory Management Status Register – MM_STAT . . . . . . . . . . . . . . . . . . . . . . . . . . .
Mbox Control Register – M_CTL . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Dcache Control Register – DC_CTL . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Dcache Status Register – DC_STAT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Cbox CSRs and IPRs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Cbox Data Register – C_DATA . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Cbox Shift Register – C_SHFT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Cbox WRITE_ONCE Chain Description . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Cbox WRITE_MANY Chain Description . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Cbox Read Register (IPR) Description . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

PALcode Description . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
PALmode Environment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Required PALcode Function Codes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Opcodes Reserved for PALcode . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
HW_LD Instruction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
HW_ST Instruction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
HW_RET Instruction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
HW_MFPR and HW_MTPR Instructions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Internal Processor Register Access Mechanisms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
IPR Scoreboard Bits. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Hardware Structure of Explicitly Written IPRs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

6–1
6–2
6–3
6–3
6–3
6–4
6–5
6–6
6–7
6–8
6–8

21264/EV68A Hardware Reference Manual

6.5.3
Hardware Structure of Implicitly Written IPRs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
6.5.4
IPR Access Ordering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
6.5.5
Correct Ordering of Explicit Writers Followed by Implicit Readers. . . . . . . . . . . . . . . . .
6.5.6
Correct Ordering of Explicit Readers Followed by Implicit Writers. . . . . . . . . . . . . . . . .
6.6
PALshadow Registers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
6.7
PALcode Emulation of the FPCR . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
6.7.1
Status Flags . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
6.7.2
MF_FPCR . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
6.7.3
MT_FPCR . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
6.8
PALcode Entry Points . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
6.8.1
CALL_PAL Entry Points . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
6.8.2
PALcode Exception Entry Points . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
6.9
Translation Buffer (TB) Fill Flows . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
6.9.1
DTB Fill . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
6.9.2
ITB Fill . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
6.10
Performance Counter Support . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
6.10.1
General Precautions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
6.10.2
Aggregate Mode Programming Guidelines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
6.10.2.1
Aggregate Mode Precautions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
6.10.2.2
Operation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
6.10.2.3
Aggregate Counting Mode Description . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
6.10.2.3.1
Cycle counting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
6.10.2.3.2
Retired instructions cycles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
6.10.2.3.3
Bcache miss or long latency probes cycles . . . . . . . . . . . . . . . . . . . . . . . . . . .
6.10.2.3.4
Mbox replay traps cycles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
6.10.2.4
Counter Modes for Aggregate Mode. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
6.10.3
ProfileMe Mode Programming Guidelines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
6.10.3.1
ProfileMe Mode Precautions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
6.10.3.2
Operation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
6.10.3.3
ProfileMe Counting Mode Description . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
6.10.3.3.1
Cycle counting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
6.10.3.3.2
Inum retire delay cycles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
6.10.3.3.3
Retired instructions cycles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
6.10.3.3.4
Bcache miss or long latency probes cycles . . . . . . . . . . . . . . . . . . . . . . . . . . .
6.10.3.3.5
Mbox replay traps cycles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
6.10.3.4
Counter Modes for ProfileMe Mode . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

6–9
6–9
6–10
6–11
6–11
6–11
6–12
6–12
6–12
6–12
6–12
6–13
6–14
6–14
6–16
6–17
6–18
6–18
6–18
6–19
6–20
6–20
6–20
6–20
6–20
6–20
6–20
6–20
6–21
6–23
6–23
6–23
6–23
6–23
6–23
6–24

Initialization and Configuration
7.1
7.1.1
7.1.2
7.1.3
7.1.4
7.1.5
7.2
7.3
7.4
7.5
7.6
7.7
7.8
7.9
7.10
7.11
7.11.1
7.11.2

Power-Up Reset Flow and the Reset_L and DCOK_H Pins . . . . . . . . . . . . . . . . . . . . . . . . .
Power Sequencing and Reset State for Signal Pins . . . . . . . . . . . . . . . . . . . . . . . . . . .
Clock Forwarding and System Clock Ratio Configuration . . . . . . . . . . . . . . . . . . . . . . .
PLL Ramp Up. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
BiST and SROM Load and the TestStat_H Pin . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Clock Forward Reset and System Interface Initialization . . . . . . . . . . . . . . . . . . . . . . . .
Fault Reset Flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Energy Star Certification and Sleep Mode Flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Warm Reset Flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Array Initialization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Initialization Mode Processing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
External Interface Initialization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Internal Processor Register Power-Up Reset State . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
IEEE 1149.1 Test Port Reset . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Reset State Machine . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Phase-Lock Loop (PLL) Functional Description . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Differential Reference Clocks. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
PLL Output Clocks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

21264/EV68A Hardware Reference Manual

7–1
7–3
7–4
7–6
7–6
7–7
7–8
7–9
7–11
7–12
7–12
7–14
7–14
7–16
7–16
7–19
7–19
7–19

vii

7.11.2.1
7.11.2.2
7.11.2.3
7.11.2.4

Electrical Characteristics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
DC Characteristics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Power Supply Sequencing and Avoiding Potential Failure Mechanisms . . . . . . . . . . . . . . .
AC Characteristics. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

9–1
9–2
9–5
9–6

Operating Temperature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Heat Sink Specifications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Thermal Design Considerations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

10–1
10–3
10–6

Testability and Diagnostics
11.1
11.2
11.2.1
11.2.2
11.3
11.4
11.5
11.5.1

viii

8–2
8–2
8–2
8–3
8–3
8–4
8–4
8–4
8–4
8–5
8–5
8–5
8–5
8–6
8–7
8–7
8–7
8–7
8–7
8–8
8–9
8–9
8–10

Thermal Management
10.1
10.2
10.3

Data Error Correction Code. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Icache Data or Tag Parity Error. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Dcache Tag Parity Error . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Dcache Data Single-Bit Correctable ECC Error . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Load Instruction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Store Instruction (Quadword or Smaller) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Dcache Victim Extracts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Dcache Store Second Error . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Dcache Duplicate Tag Parity Error . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Bcache Tag Parity Error . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Controlling Bcache Block Parity Calculation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Bcache Data Single-Bit Correctable ECC Error . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Icache Fill from Bcache . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Dcache Fill from Bcache . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Bcache Victim Read . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Bcache Victim Read During a Dcache/Bcache Miss . . . . . . . . . . . . . . . . . . . . . . .
Bcache Victim Read During an ECB Instruction . . . . . . . . . . . . . . . . . . . . . . . . . . .
Memory/System Port Single-Bit Data Correctable ECC Error. . . . . . . . . . . . . . . . . . . . . . . .
Icache Fill from Memory. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Dcache Fill from Memory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Bcache Data Single-Bit Correctable ECC Error on a Probe . . . . . . . . . . . . . . . . . . . . . . . . .
Double-Bit Fill Errors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Error Case Summary. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Electrical Data
9.1
9.2
9.3
9.4

7–19
7–19
7–19
7–20

Error Detection and Error Handling
8.1
8.2
8.3
8.4
8.4.1
8.4.2
8.4.3
8.5
8.6
8.7
8.8
8.9
8.9.1
8.9.2
8.9.3
8.9.3.1
8.9.3.2
8.10
8.10.1
8.10.2
8.11
8.12
8.13

GCLK . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Differential 21264/EV68A Clocks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Nominal Operating Frequency . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Power-Up/Reset Clocking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Test Pins . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
SROM/Serial Diagnostic Terminal Port . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
SROM Load Operation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Serial Terminal Port . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
IEEE 1149.1 Port. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
TestStat_H Pin . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Power-Up Self-Test and Initialization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Built-in Self-Test . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

11–1
11–2
11–2
11–2
11–3
11–4
11–5
11–5

21264/EV68A Hardware Reference Manual

11.5.2
SROM Initialization. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
11.5.2.1
Serial Instruction Cache Load Operation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
11.6
Notes on IEEE 1149.1 Operation and Compliance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Alpha Instruction Set
A.1
A.2
A.2.1
A.2.2
A.3
A.4
A.5
A.6
A.7
A.8

11–5
11–6
11–7

Alpha Instruction Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Reserved Opcodes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Opcodes Reserved for Compaq. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Opcodes Reserved for PALcode . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
IEEE Floating-Point Instructions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
VAX Floating-Point Instructions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Independent Floating-Point Instructions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Opcode Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Required PALcode Function Codes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
IEEE Floating-Point Conformance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

A–1
A–8
A–8
A–9
A–9
A–11
A–11
A–12
A–13
A–14

21264/EV68A Boundary-Scan Register
B.1
B.1.1

Boundary-Scan Register . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
BSDL Description of the Alpha 21264/EV68A Boundary-Scan Register . . . . . . . . . . . .

Serial Icache Load Predecode Values

PALcode Restrictions and Guidelines
D.1
D.2
D.3
D.4
D.5
D.6
D.7
D.8
D.9
D.10
D.11
D.12
D.13
D.14
D.15
D.16
D.17
D.18
D.19
D.20
D.21
D.22
D.23
D.24
D.25

B–1
B–1

Restriction 1 : Reset Sequence Required by Retire Logic and Mapper . . . . . . . . . . . . . . .
D–1
Restriction 2 : No Multiple Writers to IPRs in Same Scoreboard Group . . . . . . . . . . . . . . .
D–8
Restriction 4 : No Writers and Readers to IPRs in Same Scoreboard Group . . . . . . . . . .
D–8
Guideline 6 : Avoid Consecutive Read-Modify-Write-Read-Modify-Write . . . . . . . . . . . .
D–9
Restriction 7 : Replay Trap, Interrupt Code Sequence, and STF/ITOF . . . . . . . . . . . . . . .
D–9
Restriction 9 : PALmode Istream Address Ranges . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
D–10
Restriction 10: Duplicate IPR Mode Bits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
D–10
Restriction 11: Ibox IPR Update Synchronization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
D–11
Restriction 12: MFPR of Implicitly-Written IPRs EXC_ADDR, IVA_FORM, and EXC_SUM
D–11
Restriction 13 : DTB Fill Flow Collision . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
D–11
Restriction 14 : HW_RET . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
D–11
Guideline 16 : JSR-BAD VA . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
D–12
Restriction 17: MTPR to DTB_TAG0/DTB_PTE0/DTB_TAG1/DTB_PTE1 . . . . . . . . . . . . .
D–12
Restriction 18: No FP Operates, FP Conditional Branches, FTOI, or STF in Same Fetch Block as
HW_MTPR . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
D–12
Restriction 19: HW_RET/STALL After Updating the FPCR by way of MT_FPCR in PALmode D–12
Guideline 20 : I_CTL[SBE] Stream Buffer Enable . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
D–12
Restriction 21: HW_RET/STALL After HW_MTPR ASN0/ASN1. . . . . . . . . . . . . . . . . . . . . .
D–12
Restriction 22: HW_RET/STALL After HW_MTPR IS0/IS1. . . . . . . . . . . . . . . . . . . . . . . . . .
D–13
Restriction 23: HW_ST/P/CONDITIONAL Does Not Clear the Lock Flag. . . . . . . . . . . . . . .
D–13
Restriction 24: HW_RET/STALL After HW_MTPR IC_FLUSH, IC_FLUSH_ASM, CLEAR_MAP
.......................................................................
D–14
Restriction 25: HW_MTPR ITB_IA After Reset . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
D–14
Guideline 26: Conditional Branches in PALcode . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
D–14
Restriction 27: Reset of ‘Force-Fail Lock Flag’ State in PALcode . . . . . . . . . . . . . . . . . . . . .
D–15
Restriction 28: Enforce Ordering Between IPRs Implicitly Written by Loads and Subsequent Loads
.......................................................................
D–15
Guideline 29 : JSR, JMP, RET, and JSR_COR in PALcode . . . . . . . . . . . . . . . . . . . . . . . . .
D–15

21264/EV68A Hardware Reference Manual

D.26
D.27
D.28
D.29
D.30
D.31
D.32
D.33
D.34
D.35
D.36
D.37
D.38
D.39
D.40
D.41
D.42
D.43
D.44

Restriction 30 : HW_MTPR and HW_MFPR to the Cbox CSR . . . . . . . . . . . . . . . . . . . . . . .
D–15
Restriction 31 : I_CTL[VA_48] Update . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
D–17
Restriction 32 : PCTR_CTL Update . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
D–17
Restriction 33 : HW_LD Physical/Lock Use. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
D–18
Restriction 34 : Writing Multiple ITB Entries in the Same PALcode Flow . . . . . . . . . . . . . . .
D–18
Guideline 35 : HW_INT_CLR Update . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
D–18
Restriction 36 : Updating I_CTL[SDE]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
D–18
Restriction 37 : Updating VA_CTL[VA_48] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
D–18
Restriction 38 : Updating PCTR_CTL . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
D–18
Guideline 39: Writing Multiple DTB Entries in the Same PAL Flow. . . . . . . . . . . . . . . . . . . .
D–19
Restriction 40: Scrubbing a Single-Bit Error . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
D–19
Restriction 41: MTPR ITB_TAG, MTPR ITB_PTE Must be in the Same Fetch Block . . . . .
D–21
Restriction 42: Updating VA_CTL, CC_CTL, or CC IPRs . . . . . . . . . . . . . . . . . . . . . . . . . . .
D–21
Restriction 43: No Trappable Instructions Along with HW_MTPR. . . . . . . . . . . . . . . . . . . . .
D–21
Restriction 44: Not Applicable to the 21264/EV68A . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
D–21
Restriction 45: No HW_JMP or JMP Intructions in PALcode . . . . . . . . . . . . . . . . . . . . . . . .
D–21
Restriction 46: Avoiding Livelocks in Speculative Load CRD Handlers . . . . . . . . . . . . . . . .
D–22
Restriction 47: Cache Eviction for Single-Bit Cache Errors . . . . . . . . . . . . . . . . . . . . . . . . .
D–22
Restriction 48: MB Bracketing of Dcache Writes to Force Bad Data ECC and Force Bad Tag Parity
.......................................................................
D–24

21264/EV68A-to-Bcache Pin Interface
E.1
E.2
E.3

Forwarding Clock Pin Groupings. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Late-Write Non-Bursting SSRAMs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Dual-Data Rate SSRAMs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

E–1
E–2
E–3

Glossary
Index

21264/EV68A Hardware Reference Manual

Figures
2–1
2–2
2–3
2–4
2–5
2–6
2–7
2–8
2–9
2–10
2–11
2–12
2–13
3–1
3–2
3–3
3–4
4–1
4–2
4–3
4–4
4–5
4–6
5–1
5–2
5–3
5–4
5–5
5–6
5–7
5–8
5–9
5–10
5–11
5–12
5–13
5–14
5–15
5–16
5–17
5–18
5–19
5–20
5–21
5–22
5–23
5–24
5–25
5–26
5–27
5–28
5–29
5–30
5–31
5–32
5–33

21264/EV68A Block Diagram . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Branch Predictor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Local Predictor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Global Predictor. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Choice Predictor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Integer Execution Unit—Clusters 0 and 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Floating-Point Execution Units . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Pipeline Organization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Pipeline Timing for Integer Load Instructions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Pipeline Timing for Floating-Point Load Instructions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Floating-Point Control Register . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Typical Uniprocessor Configuration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Typical Multiprocessor Configuration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
21264/EV68A Microprocessor Logic Symbol . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Package Dimensions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
21264/EV68A Top View (Pin Down) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
21264/EV68A Bottom View (Pin Up). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
21264/EV68A System and Bcache Interfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
21264/EV68A Bcache Interface Signals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Cache Subset Hierarchy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
System Interface Signals. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Fast Transfer Timing Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
SysFillValid_L Timing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Cycle Counter Register . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Cycle Counter Control Register. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Virtual Address Register . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Virtual Address Control Register . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Virtual Address Format Register (VA_48 = 0, VA_FORM_32 = 0) . . . . . . . . . . . . . . . . . . . .
Virtual Address Format Register (VA_48 = 1, VA_FORM_32 = 0) . . . . . . . . . . . . . . . . . . . .
Virtual Address Format Register (VA_48 = 0, VA_FORM_32 = 1) . . . . . . . . . . . . . . . . . . . .
ITB Tag Array Write Register . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
ITB PTE Array Write Register . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
ITB Invalidate Single Register . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
ProfileMe PC Register. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Exception Address Register . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Instruction Virtual Address Format Register (VA_48 = 0, VA_FORM_32 = 0) . . . . . . . . . . .
Instruction Virtual Address Format Register (VA_48 = 1, VA_FORM_32 = 0) . . . . . . . . . . .
Instruction Virtual Address Format Register (VA_48 = 0, VA_FORM_32 = 1) . . . . . . . . . . .
Interrupt Enable and Current Processor Mode Register . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Software Interrupt Request Register . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Interrupt Summary Register . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Hardware Interrupt Clear Register . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Exception Summary Register . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
PAL Base Register . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Ibox Control Register. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Ibox Status Register . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Process Context Register . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Performance Counter Control Register . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
DTB Tag Array Write Registers 0 and 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
DTB PTE Array Write Registers 0 and 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
DTB Alternate Processor Mode Register . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Dstream Translation Buffer Invalidate Single Registers . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Dstream Translation Buffer Address Space Number Registers 0 and 1 . . . . . . . . . . . . . . . .
Memory Management Status Register . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Mbox Control Register. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Dcache Control Register . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

21264/EV68A Hardware Reference Manual

2–3
2–4
2–4
2–5
2–5
2–9
2–10
2–14
2–24
2–25
2–36
2–39
2–39
3–2
3–17
3–18
3–19
4–3
4–7
4–9
4–17
4–32
4–36
5–3
5–3
5–4
5–4
5–5
5–6
5–6
5–6
5–7
5–7
5–8
5–8
5–9
5–9
5–9
5–10
5–11
5–11
5–12
5–14
5–15
5–16
5–19
5–22
5–23
5–25
5–26
5–26
5–27
5–28
5–28
5–29
5–31

5–34
5–35
5–36
5–37
6–1
6–2
6–3
6–4
6–5
6–6
7–1
7–2
7–3
7–4
7–5
10–1
10–2
10–3
11–1
11–2
11–3

xii

Dcache Status Register. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Cbox Data Register . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Cbox Shift Register . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
WRITE_MANY Chain Write Transaction Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
HW_LD Instruction Format . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
HW_ST Instruction Format . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
HW_RET Instruction Format . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
HW_MFPR and HW_MTPR Instructions Format . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Single-Miss DTB Instructions Flow Example. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
ITB Miss Instructions Flow Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Power-Up Timing Sequence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Fault Reset Sequence of Operation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Sleep Mode Sequence of Operation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Example for Initializing Bcache . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
21264/EV68A Reset State Machine State Diagram . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Type 1 Heat Sink . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Type 2 Heat Sink . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Type 3 Heat Sink . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
TestStat_H Pin Timing During Power-Up Built-In Self-Test (BiST) . . . . . . . . . . . . . . . . . . .
TestStat_H Pin Timing During Built-In Self-Initialization (BiSI) . . . . . . . . . . . . . . . . . . . . . . .
SROM Content Map . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

5–32
5–33
5–33
5–39
6–4
6–4
6–6
6–6
6–14
6–16
7–3
7–9
7–11
7–13
7–17
10–3
10–4
10–5
11–5
11–5
11–6

21264/EV68A Hardware Reference Manual

Tables
1–1
2–1
2–2
2–3
2–4
2–5
2–6
2–7
2–8
2–9
2–10
2–11
2–12
2–13
2–14
2–15
2–16
3–1
3–2
3–3
3–4
3–5
3–6
4–1
4–2
4–3
4–4
4–5
4–6
4–7
4–8
4–9
4–10
4–11
4–12
4–13
4–14
4–15
4–16
4–17
4–18
4–19
4–20
4–21
4–22
4–23
4–24
4–25
4–26
4–27
4–28
4–29
4–30
4–31
4–32
4–33

Integer Data Types . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Pipeline Abort Delay (GCLK Cycles). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Instruction Name, Pipeline, and Types . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Instruction Group Definitions and Pipeline Unit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Instruction Class Latency in Cycles. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Minimum Retire Latencies for Instruction Classes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Instructions Retired Without Execution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Rules for I/O Address Space Load Instruction Data Merging . . . . . . . . . . . . . . . . . . . . . . . .
Rules for I/O Address Space Store Instruction Data Merging . . . . . . . . . . . . . . . . . . . . . . . .
MAF Merging Rules. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Memory Reference Ordering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
I/O Reference Ordering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
TB Fill Flow Example Sequence 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
TB Fill Flow Example Sequence 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Floating-Point Control Register Fields. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
21264/EV68A AMASK Values. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
AMASK Bit Assignments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Signal Pin Types Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
21264/EV68A Signal Descriptions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
21264/EV68A Signal Descriptions by Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Pin List Sorted by Signal Name. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Pin List Sorted by PGA Location . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Ground and Power (VSS and VDD) Pin List . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Translation of Internal References to External Interface Reference . . . . . . . . . . . . . . . . . . .
21264/EV68A-Supported Cache Block States . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Cache Block State Transitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
System Responses to 21264/EV68A Commands . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
System Responses to 21264/EV68A Commands and Reactions . . . . . . . . . . . . . . . . . . . . .
System Port Pins . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Programming Values for System Interface Clocks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Program Values for Data-Sample/Drive CSRs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Forwarded Clocks and Frame Clock Ratio . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Bank Interleave on Cache Block Boundary Mode of Operation . . . . . . . . . . . . . . . . . . . . . .
Page Hit Mode of Operation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
21264/EV68A-to-System Command Fields Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Maximum Physical Address for Short Bus Format . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
21264/EV68A-to-System Commands Descriptions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Programming INVAL_TO_DIRTY_ENABLE[1:0]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Programming SET_DIRTY_ENABLE[2:0] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
21264/EV68A ProbeResponse Command . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
ProbeResponse Fields Descriptions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
System-to-21264/EV68A Probe Commands . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
System-to-21264/EV68A Probe Commands Fields Descriptions . . . . . . . . . . . . . . . . . . . . .
Data Movement Selection by Probe[4:3] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Next Cache Block State Selection by Probe[2:0] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Data Transfer Command Format . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
SysDc[4:0] Field Description . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
SYSCLK Cycles Between SysAddOut and SysData. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Cbox CSR SYSDC_DELAY[4:0] Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Four Timing Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Data Wrapping Rules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
System Wrap and Deliver Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Wrap Interleave Order. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Wrap Order for Double-Pumped Data Transfers. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
21264/EV68A Commands with NXM Addresses and System Response . . . . . . . . . . . . . . .
21264/EV68A Response to System Probe and In-Flight Command Interaction . . . . . . . . . .

21264/EV68A Hardware Reference Manual

1–2
2–16
2–17
2–18
2–20
2–21
2–23
2–28
2–29
2–30
2–30
2–31
2–34
2–34
2–36
2–38
2–38
3–3
3–3
3–6
3–8
3–12
3–16
4–5
4–9
4–10
4–10
4–11
4–17
4–18
4–18
4–19
4–19
4–20
4–20
4–21
4–21
4–23
4–24
4–24
4–25
4–26
4–27
4–27
4–27
4–28
4–29
4–32
4–33
4–34
4–36
4–37
4–37
4–38
4–39
4–41

xiii

4–34
4–35
4–36
4–37
4–38
4–39
4–40
4–41
4–42
4–43
4–44
4–45
4–46
4–47
5–1
5–2
5–3
5–4
5–5
5–6
5–7
5–8
5–9
5–10
5–11
5–12
5–13
5–14
5–15
5–16
5–17
5–18
5–19
5–20
5–21
5–22
5–23
5–24
5–25
5–26
6–1
6–2
6–3
6–4
6–5
6–6
6–7
6–8
6–9
6–10
6–11
6–12
6–13
6–14
7–1
7–2
7–3
7–4
7–5
xiv

Rules for System Control of Cache Status Update Order . . . . . . . . . . . . . . . . . . . . . . . . . . .
Range of Maximum Bcache Clock Ratios . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Bcache Port Pins . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
BC_CPU_CLK_DELAY[1:0] Values . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
BC_CLK_DELAY[1:0] Values . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Program Values to Set the Cache Clock Period (Single-Data) . . . . . . . . . . . . . . . . . . . . . . .
Program Values to Set the Cache Clock Period (Dual-Data Rate) . . . . . . . . . . . . . . . . . . . .
Data-Sample/Drive Cbox CSRs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Programming the Bcache to Support Each Size of the Bcache . . . . . . . . . . . . . . . . . . . . . .
Programming the Bcache Control Pins . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Control Pin Assertion for RAM_TYPE A . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Control Pin Assertion for RAM_TYPE B . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Control Pin Assertion for RAM_TYPE C . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Control Pin Assertion for RAM_TYPE D . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Internal Processor Registers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Cycle Counter Control Register Fields Description . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Virtual Address Control Register Fields Description . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
ProfileMe PC Fields Description . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
IER_CM Register Fields Description . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Software Interrupt Request Register Fields Description . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Interrupt Summary Register Fields Description. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Hardware Interrupt Clear Register Fields Description . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Exception Summary Register Fields Description . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
PAL Base Register Fields Description . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Ibox Control Register Fields Description . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Ibox Status Register Fields Description . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
IPR Index Bits and Register Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Process Context Register Fields Description . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Performance Counter Control Register Fields Description . . . . . . . . . . . . . . . . . . . . . . . . . .
Performance Counter Control Register Input Select Fields. . . . . . . . . . . . . . . . . . . . . . . . . .
DTB Alternate Processor Mode Register Fields Description. . . . . . . . . . . . . . . . . . . . . . . . .
Memory Management Status Register Fields Description . . . . . . . . . . . . . . . . . . . . . . . . . .
Mbox Control Register Fields Description . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Dcache Control Register Fields Description . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Dcache Status Register Fields Description . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Cbox Data Register Fields Description . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Cbox Shift Register Fields Description . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Cbox WRITE_ONCE Chain Order . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Cbox WRITE_MANY Chain Order . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Cbox Read IPR Fields Description . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Required PALcode Function Codes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Opcodes Reserved for PALcode . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
HW_LD Instruction Fields Descriptions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
HW_ST Instruction Fields Descriptions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
HW_RET Instruction Fields Descriptions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
HW_MFPR and HW_MTPR Instructions Fields Descriptions . . . . . . . . . . . . . . . . . . . . . . . .
Paired Instruction Fetch Order . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
PALcode Exception Entry Locations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
IPRs Used for Performance Counter Support . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Aggregate Mode Returned IPR Contents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Aggregate Mode Performance Counter IPR Input Select Fields . . . . . . . . . . . . . . . . . . . . . .
CMOV Decomposed . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
ProfileMe Mode Returned IPR Contents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
ProfileMe Mode PCTR_CTL Input Select Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
21264/EV68A Reset State Machine Major Operations . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Signal Pin Reset State . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Pin Signal Names and Initialization State . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Power-Up Flow Signals and Their Constraints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Effect on IPRs After Fault Reset . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

4–42
4–43
4–43
4–45
4–45
4–46
4–46
4–47
4–51
4–51
4–51
4–52
4–52
4–52
5–1
5–4
5–5
5–8
5–10
5–11
5–12
5–13
5–14
5–15
5–16
5–19
5–21
5–22
5–23
5–25
5–26
5–28
5–30
5–31
5–32
5–33
5–33
5–34
5–39
5–41
6–3
6–3
6–4
6–5
6–6
6–7
6–9
6–13
6–18
6–19
6–20
6–21
6–22
6–24
7–1
7–3
7–5
7–7
7–8

21264/EV68A Hardware Reference Manual

7–6
7–7
7–8
7–9
7–10
7–11
7–12
8–1
8–2
8–3
9–1
9–2
9–3
9–4
9–5
9–6
9–7
9–8
9–9
9–10
9–11
9–12
9–13
10–1
10–2
10–3
10–4
10–5
10–6
11–1
11–2
11–3
11–4
A–1
A–2
A–3
A–4
A–5
A–6
A–7
A–8
A–9
A–10
A–11
E–1
E–2
E–3
E–4
E–5

Effect on IPRs After Transition Through Sleep Mode . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Signals and Constraints for the Sleep Mode Sequence . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Effect on IPRs After Warm Reset . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
WRITE_MANY Chain CSR Values for Bcache Initialization . . . . . . . . . . . . . . . . . . . . . . . . .
Internal Processor Registers at Power-Up Reset State . . . . . . . . . . . . . . . . . . . . . . . . . . . .
21264/EV68A Reset State Machine State Descriptions . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Differential Reference Clock Frequencies in Full-Speed Lock . . . . . . . . . . . . . . . . . . . . . . .
21264/EV68A Error Detection Mechanisms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
64-Bit Data and Check Bit ECC Code . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Error Case Summary. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Maximum Electrical Ratings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Signal Types . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
VDD (I_DC_POWER) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Input DC Reference Pin (I_DC_REF) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Input Differential Amplifier Receiver (I_DA). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Input Differential Amplifier Clock Receiver (I_DA_CLK) . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Pin Type: Open-Drain Output Driver (O_OD) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Bidirectional, Differential Amplifier Receiver, Open-Drain Output Driver (B_DA_OD) . . . . .
Pin Type: Open-Drain Driver for Test Pins (O_OD_TP) . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Bidirectional, Differential Amplifier Receiver, Push-Pull Output Driver (B_DA_PP) . . . . . . .
Push-Pull Output Driver (O_PP) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Push-Pull Output Clock Driver (O_PP_CLK). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
AC Specifications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Operating Temperature at Heat Sink Center (Tc) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
qca at Various Airflows for 21264/EV68A . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Maximum Ta for 21264/EV68A @ 750 MHz and @ 1.7 V with Various Airflows . . . . . . . . .
Maximum Ta for 21264/EV68A @ 833 MHz and @ 1.7 V with Various Airflows . . . . . . . . .
Maximum Ta for 21264/EV68A @ 875 MHz and @ 1.7 V with Various Airflows . . . . . . . . .
Maximum Ta for 21264/EV68A @ 940 MHz and @ 1.7 V with Various Airflows . . . . . . . . .
Dedicated Test Port Pins. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
IEEE 1149.1 Instructions and Opcodes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
TAP Controller State Machine . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Icache Bit Fields in an SROM Line . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Instruction Format and Opcode Notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Architecture Instructions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Opcodes Reserved for Compaq . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Opcodes Reserved for PALcode . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
IEEE Floating-Point Instruction Function Codes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
VAX Floating-Point Instruction Function Codes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Independent Floating-Point Instruction Function Codes . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Opcode Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Key to Opcode Summary Used in Table A–8 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Required PALcode Function Codes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Exceptional Input and Output Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Bcache Forwarding Clock Pin Groupings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Late-Write Non-Bursting SSRAMs Data Pin Usage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Late-Write Non-Bursting SSRAMs Tag Pin Usage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Dual-Data Rate SSRAM Data Pin Usage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Dual-Data Rate SSRAM Tag Pin Usage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

21264/EV68A Hardware Reference Manual

7–10
7–11
7–11
7–12
7–14
7–17
7–20
8–1
8–2
8–10
9–1
9–2
9–3
9–3
9–3
9–3
9–4
9–4
9–4
9–4
9–5
9–5
9–7
10–1
10–2
10–2
10–2
10–2
10–2
11–1
11–3
11–4
11–7
A–1
A–2
A–8
A–9
A–9
A–11
A–12
A–12
A–13
A–13
A–15
E–1
E–2
E–2
E–3
E–4

Preface
Audience
This manual is for system designers and programmers who use the Alpha 21264/
EV68A microprocessor (referred to as the 21264/EV68A).

Content
This manual contains the following chapters and appendixes:
Chapter 1, Introduction, introduces the 21264/EV68A and provides an overview of the
Alpha architecture.
Chapter 2, Internal Architecture, describes the major hardware functions and the internal chip architecture. It describes performance measurement facilities, coding rules, and
design examples.
Chapter 3, Hardware Interface, lists and describes the internal hardware interface signals, and provides mechanical data and packaging information, including signal pin
lists.
Chapter 4, Cache and External Interfaces, describes the external bus functions and
transactions, lists bus commands, and describes the clock functions.
Chapter 5, Internal Processor Registers, lists and describes the internal processor register set.
Chapter 6, Privileged Architecture Library Code, describes the privileged architecture
library code (PALcode).
Chapter 7, Initialization and Configuration, describes the initialization and configuration sequence.
Chapter 8, Error Detection and Error Handling, describes error detection and error handling.
Chapter 9, Electrical Data, provides electrical data and describes signal integrity issues.
Chapter 10, Thermal Management, provides information about thermal management.
Chapter 11, Testability and Diagnostics, describes chip and system testability features.
Appendix A, Alpha Instruction Set, summarizes the Alpha instruction set.
Appendix B, 21264/EV68A Boundary-Scan Register, presents the BSDL description
of the 21264/EV68A boundary-scan register.

21264/EV68A Hardware Reference Manual

xvii

Appendix C, Serial Icache Load Predecode Values, provides a pointer to the Alpha
Motherboards Software Developer’s Kit (SDK), which contains this information.
Appendix D, PALcode Restrictions and Guidelines, lists restrictions and guidelines
that must be adhered to when generating PALcode.
Appendix E, 21264/EV68A-to-Bcache Pin Interface, provides the pin interface
between the 21264/EV68A and Bcache SSRAMs.
The Glossary lists and defines terms associated with the 21264/EV68A.
An Index is provided at the end of the document.
Documentation Included by Reference

The companion volume to this manual, the Alpha Architecture Reference Manual,
Fourth Edition, can be accessed from the following website: ftp.compaq.com/
pub/products/alphaCPUdocs.

xviii

21264/EV68A Hardware Reference Manual

Terminology and Conventions
This section defines the abbreviations, terminology, and other conventions used
throughout this document.
Abbreviations
•

Binary Multiples
The abbreviations K, M, and G (kilo, mega, and giga) represent binary multiples
and have the following values.
K
M
G

=
=
=

210 (1024)
220 (1,048,576)
230 (1,073,741,824)

For example:
2KB
4MB
8GB
2K pixels
4M pixels
•

=
=
=
=
=

2 kilobytes
=
4 megabytes =
8 gigabytes =
2 kilopixels =
4 megapixels =

2 × 210 bytes
4 × 220 bytes
8 × 230 bytes
2 × 210 pixels
4 × 220 pixels

Register Access
The abbreviations used to indicate the type of access to register fields and bits have
the following definitions:

Abbreviation

Meaning

IGN

Ignore
Bits and fields specified are ignored on writes.

MBZ

Must Be Zero
Software must never place a nonzero value in bits and fields specified as
MBZ. A nonzero read produces an Illegal Operand exception. Also, MBZ
fields are reserved for future use.

RAZ

Read As Zero
Bits and fields return a zero when read.

Read Clears
Bits and fields are cleared when read. Unless otherwise specified, such bits
cannot be written.

RES

Reserved
Bits and fields are reserved by Compaq and should not be used; however,
zeros can be written to reserved fields that cannot be masked.

Read Only
The value may be read by software. It is written by hardware. Software write
operations are ignored.

RO,n

Read Only, and takes the value n at power-on reset.
The value may be read by software. It is written by hardware. Software write
operations are ignored.

21264/EV68A Hardware Reference Manual

xix

Abbreviation

Meaning

Read/Write
Bits and fields can be read and written.

RW,n

Read/Write, and takes the value n at power-on reset.
Bits and fields can be read and written.

W1C

Write One to Clear
If read operations are allowed to the register, then the value may be read by
software. If it is a write-only register, then a read operation by software
returns an UNPREDICTABLE result. Software write operations of a 1 cause
the bit to be cleared by hardware. Software write operations of a 0 do not
modify the state of the bit.

W1S

Write One to Set
If read operations are allowed to the register, then the value may be read by
software. If it is a write-only register, then a read operation by software
returns an UNPREDICTABLE result. Software write operations of a 1 cause
the bit to be set by hardware. Software write operations of a 0 do not modify
the state of the bit.

Write Only
Bits and fields can be written but not read.

WO,n

Write Only, and takes the value n at power-on reset.
Bits and fields can be written but not read.

•

Sign extension
SEXT(x) means x is sign-extended to the required size.

Addresses

Unless otherwise noted, all addresses and offsets are hexadecimal.
Aligned and Unaligned

The terms aligned and naturally aligned are interchangeable and refer to data objects
that are powers of two in size. An aligned datum of size 2n is stored in memory at a
byte address that is a multiple of 2n; that is, one that has n low-order zeros. For example, an aligned 64-byte stack frame has a memory address that is a multiple of 64.
A datum of size 2n is unaligned if it is stored in a byte address that is not a multiple of
2n.
Bit Notation

Multiple-bit fields can include contiguous and noncontiguous bits contained in square
brackets ([]). Multiple contiguous bits are indicated by a pair of numbers separated by a
colon [:]. For example, [9:7,5,2:0] specifies bits 9,8,7,5,2,1, and 0. Similarly, single bits
are frequently indicated with square brackets. For example, [27] specifies bit 27. See
also Field Notation.
Caution

Cautions indicate potential damage to equipment or loss of data.

21264/EV68A Hardware Reference Manual

Data Units

The following data unit terminology is used throughout this manual.
Term

Words

Bytes

Bits

Other

Byte

—

Word

—

Longword

Dword

Quadword

2 longword

Do Not Care (X)

A capital X represents any valid value.
External

Unless otherwise stated, external means not contained in the chip.
Field Notation

The names of single-bit and multiple-bit fields can be used rather than the actual bit
numbers (see Bit Notation). When the field name is used, it is contained in square
brackets ([]). For example, RegisterName[LowByte] specifies RegisterName[7:0].
Note

Notes emphasize particularly important information.
Numbering

All numbers are decimal or hexadecimal unless otherwise indicated. The prefix 0x indicates a hexadecimal number. For example, 19 is decimal, but 0x19 and 0x19A are hexadecimal (also see Addresses). Otherwise, the base is indicated by a subscript; for
example, 1002 is a binary number.
Ranges and Extents

Ranges are specified by a pair of numbers separated by two periods (..) and are inclusive. For example, a range of integers 0..4 includes the integers 0, 1, 2, 3, and 4.
Extents are specified by a pair of numbers in square brackets ([]) separated by a colon
(:) and are inclusive. Bit fields are often specified as extents. For example, bits [7:3]
specifies bits 7, 6, 5, 4, and 3.
Register Figures

The gray areas in register figures indicate reserved or unused bits and fields.
Bit ranges that are coupled with the field name specify the bits of the named field that
are included in the register. The bit range may, but need not necessarily, correspond to
the bit Extent in the register. See the explanation above Table 5–1 for more information.
Signal Names

The following examples describe signal-name conventions used in this document.

21264/EV68A Hardware Reference Manual

xxi

AlphaSignal[n:n]

Boldface, mixed-case type denotes signal names that are
assigned internal and external to the 21264/EV68A (that
is, the signal traverses a chip interface pin).

AlphaSignal_x[n:n]

When a signal has high and low assertion states, a lowercase italic x represents the assertion states. For example,
SignalName_x[3:0] represents SignalName_H[3:0] and
SignalName_L[3:0].

UNDEFINED

Operations specified as UNDEFINED may vary from moment to moment, implementation to implementation, and instruction to instruction within implementations. The
operation may vary in effect from nothing to stopping system operation.
UNDEFINED operations may halt the processor or cause it to lose information. However, UNDEFINED operations must not cause the processor to hang, that is, reach an
unhalted state from which there is no transition to a normal state in which the machine
executes instructions.
UNPREDICTABLE

UNPREDICTABLE results or occurrences do not disrupt the basic operation of the processor; it continues to execute instructions in its normal manner. Further:
•

Results or occurrences specified as UNPREDICTABLE may vary from moment to
moment, implementation to implementation, and instruction to instruction within
implementations. Software can never depend on results specified as UNPREDICTABLE.

•

An UNPREDICTABLE result may acquire an arbitrary value subject to a few constraints. Such a result may be an arbitrary function of the input operands or of any
state information that is accessible to the process in its current access mode.
UNPREDICTABLE results may be unchanged from their previous values.
Operations that produce UNPREDICTABLE results may also produce exceptions.

•

An occurrence specified as UNPREDICTABLE may happen or not based on an
arbitrary choice function. The choice function is subject to the same constraints as
are UNPREDICTABLE results and, in particular, must not constitute a security
hole.
Specifically, UNPREDICTABLE results must not depend upon, or be a function of,
the contents of memory locations or registers that are inaccessible to the current
process in the current access mode.
Also, operations that may produce UNPREDICTABLE results must not:
–

Write or modify the contents of memory locations or registers to which the current process in the current access mode does not have access, or

–

Halt or hang the system or any of its components.

For example, a security hole would exist if some UNPREDICTABLE result
depended on the value of a register in another process, on the contents of processor
temporary registers left behind by some previously running process, or on a
sequence of actions of different processes.

xxii

21264/EV68A Hardware Reference Manual

Do not care. A capital X represents any valid value.

21264/EV68A Hardware Reference Manual

xxiii

1
Introduction
This chapter provides a brief introduction to the Alpha architecture, Compaq’s RISC
(reduced instruction set computing) architecture designed for high performance. The
chapter then summarizes the specific features of the Alpha 21264/EV68A microprocessor (hereafter called the 21264/EV68A) that implements the Alpha architecture. Appendix A provides a list of Alpha instructions.
The companion volume to this document, the Alpha Architecture Reference Manual,
Fourth Edition, contains the complete architecture information.

1.1 The Architecture
The Alpha architecture is a 64-bit load and store RISC architecture designed with particular emphasis on speed, multiple instruction issue, multiple processors, and software
migration from many operating systems.
All registers are 64 bits long and all operations are performed between 64-bit registers.
All instructions are 32 bits long. Memory operations are either load or store operations.
All data manipulation is done between registers.
The Alpha architecture supports the following data types:

•

8-, 16-, 32-, and 64-bit integers

•

IEEE 32-bit and 64-bit floating-point formats

•

VAX architecture 32-bit and 64-bit floating-point formats

In the Alpha architecture, instructions interact with each other only by one instruction
writing to a register or memory location and another instruction reading from that register or memory location. This use of resources makes it easy to build implementations
that issue multiple instructions every CPU cycle.
The 21264/EV68A uses a set of subroutines, called privileged architecture library code
(PALcode), that is specific to a particular Alpha operating system implementation and
hardware platform. These subroutines provide operating system primitives for context
switching, interrupts, exceptions, and memory management. These subroutines can be
invoked by hardware or CALL_PAL instructions. CALL_PAL instructions use the
function field of the instruction to vector to a specified subroutine. PALcode is written
in standard machine code with some implementation-specific extensions to provide
direct access to low-level hardware functions. PALcode supports optimizations for multiple operating systems, flexible memory-management implementations, and multiinstruction atomic sequences.

21264/EV68A Hardware Reference Manual

Introduction

1–1

The Architecture

The Alpha architecture performs byte shifting and masking with normal 64-bit, register-to-register instructions. The 21264/EV68A performs single-byte and single-word
load and store instructions.

1.1.1 Addressing
The basic addressable unit in the Alpha architecture is the 8-bit byte. The 21264/
EV68A supports a 48-bit or 43-bit virtual address (selectable under IPR control).
Virtual addresses as seen by the program are translated into physical memory addresses
by the memory-management mechanism. The 21264/EV68A supports a 44-bit physical
address.

1.1.2 Integer Data Types
Alpha architecture supports the four integer data types listed in Table 1–1.
Table 1–1 Integer Data Types
Data Type

Description

Byte

A byte is 8 contiguous bits that start at an addressable byte boundary.
A byte is an 8-bit value.

Word

A word is 2 contiguous bytes that start at an arbitrary byte boundary.
A word is a 16-bit value.

Longword

A longword is 4 contiguous bytes that start at an arbitrary byte boundary. A
longword is a 32-bit value.

Quadword

A quadword is 8 contiguous bytes that start at an arbitrary byte boundary.

Note:

Alpha implementations may impose a significant performance penalty
when accessing operands that are not naturally aligned. Refer to the Alpha
Architecture Handbook, Version 4 for details.

1.1.3 Floating-Point Data Types
The 21264/EV68A supports the following floating-point data types:

•

Longword integer format in floating-point unit

•

Quadword integer format in floating-point unit

•

IEEE floating-point formats

•

1–2

Introduction

–

S_floating

–

T_floating

VAX floating-point formats
–

F_floating

–

G_floating

–

D_floating (limited support)

21264/EV68A Hardware Reference Manual

21264/EV68A Microprocessor Features

1.2 21264/EV68A Microprocessor Features
The 21264/EV68A microprocessor is a superscalar pipelined processor. It is packaged
in a 587-pin PGA carrier and has removable application-specific heat sinks. A number
of configuration options allow its use in a range of system designs ranging from
extremely simple uniprocessor systems with minimum component count to high-performance multiprocessor systems with very high cache and memory bandwidth.
The 21264/EV68A can issue four Alpha instructions in a single cycle, thereby minimizing the average cycles per instruction (CPI). A number of low-latency and/or highthroughput features in the instruction issue unit and the onchip components of the memory subsystem further reduce the average CPI.
The 21264/EV68A and associated PALcode implements IEEE single-precision and
double-precision, VAX F_floating and G_floating data types, and supports longword
(32-bit) and quadword (64-bit) integers. Byte (8-bit) and word (16-bit) support is provided by byte-manipulation instructions. Limited hardware support is provided for the
VAX D_floating data type.
Other 21264/EV68A features include:

•

The ability to issue up to four instructions during each CPU clock cycle.

•

A peak instruction execution rate of four times the CPU clock frequency.

•

An onchip, demand-paged memory-management unit with translation buffer, which,
when used with PALcode, can implement a variety of page table structures and translation algorithms. The unit consists of a 128-entry, fully-associative data translation
buffer (DTB) and a 128-entry, fully-associative instruction translation buffer (ITB),
with each entry able to map a single 8KB page or a group of 8, 64, or 512 8KB
pages. The allocation scheme for the ITB and DTB is round-robin. The size of each
translation buffer entry’s group is specified by hint bits stored in the entry. The
DTB and ITB implement 8-bit address space numbers (ASN), MAX_ASN=255.

•

Two onchip, high-throughput pipelined floating-point units, capable of executing
both VAX and IEEE floating-point data types.

•

An onchip, 64KB virtually-addressed instruction cache with 8-bit ASNs
(MAX_ASN=255).

•

An onchip, virtually-indexed, physically-tagged dual-read-ported, 64KB data
cache.

•

Supports a 48-bit or 43-bit virtual address (program selectable).

•

Supports a 44-bit physical address.

•

An onchip I/O write buffer with four 64-byte entries for I/O write transactions.

•

An onchip, 8-entry victim data buffer.

•

An onchip, 32-entry load queue.

•

An onchip, 32-entry store queue.

•

An onchip, 8-entry miss address file for cache fill requests and I/O read
transactions.

•

An onchip, 8-entry probe queue, holding pending system port probe commands.

21264/EV68A Hardware Reference Manual

Introduction

1–3

21264/EV68A Microprocessor Features

•

An onchip, duplicate tag array used to maintain level 2 cache coherency.

•

A 64-bit data bus with onchip parity and error correction code (ECC) support.

•

Support for an external second-level (Bcache) cache. The size and some timing
parameters of the Bcache are programmable.

•

An internal clock generator providing a high-speed clock used by the 21264/
EV68A, and two clocks for use by the CPU module.

•

Onchip performance counters to measure and analyze CPU and system performance.

•

Chip and module level test support, including an instruction cache test interface to
support chip and module level testing.

•

A 2.0-V external interface.

Refer to Chapter 9 for 21264/EV68A dc and ac electrical characteristics. Refer to the
Alpha Architecture Handbook, Version 4, Appendix E, for waivers and any other
implementation-dependent information.

1–4

Introduction

21264/EV68A Hardware Reference Manual

2
Internal Architecture
This chapter provides both an overview of the 21264/EV68A microarchitecture and a system designer’s view of the 21264/EV68A implementation of the Alpha architecture. The
combination of the 21264/EV68A microarchitecture and privileged architecture library
code (PALcode) defines the chip’s implementation of the Alpha architecture. If a certain
piece of hardware seems to be “architecturally incomplete,” the missing functionality is
implemented in PALcode. Chapter 6 provides more information on PALcode.
This chapter describes the major functional hardware units and is not intended to be a
detailed hardware description of the chip. It is organized as follows:

•

21264/EV68A microarchitecture

•

Pipeline organization

•

Instruction issue and retire rules

•

Load instructions to R31/F31 (software-directed instruction prefetch)

•

Special cases of Alpha instruction execution

•

Memory and I/O address space

•

Miss address file (MAF) and load-merging rules

•

Instruction ordering

•

Replay traps

•

I/O write buffer and the WMB instruction

•

Performance measurement support

•

Floating-point control register

•

AMASK and IMPLVER instruction values

•

Design examples

2.1 21264/EV68A Microarchitecture
The 21264/EV68A microprocessor is a high-performance third-generation implementation of the Compaq Alpha architecture. The 21264/EV68A consists of the following
sections, as shown in Figure 2–1:

•

Instruction fetch, issue, and retire unit (Ibox)

•

Integer execution unit (Ebox)

21264/EV68A Hardware Reference Manual

Internal Architecture

2–1

21264/EV68A Microarchitecture

•

Floating-point execution unit (Fbox)

•

Onchip caches (Icache and Dcache)

•

Memory reference unit (Mbox)

•

External cache and system interface unit (Cbox)

•

Pipeline operation sequence

2.1.1 Instruction Fetch, Issue, and Retire Unit
The instruction fetch, issue, and retire unit (Ibox) consists of the following subsections:

•

Virtual program counter logic

•

Branch predictor

•

Instruction-stream translation buffer (ITB)

•

Instruction fetch logic

•

Integer and floating-point issue queues

•

Exception and interrupt logic

•

Retire logic

2.1.1.1 Virtual Program Counter Logic

The virtual program counter (VPC) logic maintains the virtual addresses for instructions that are in flight. There can be up to 80 instructions, in 20 successive fetch slots, in
flight between the register rename mappers and the end of the pipeline. The VPC logic
contains a 20-entry table to store these fetched VPC addresses.

2–2

Internal Architecture

21264/EV68A Hardware Reference Manual

21264/EV68A Microarchitecture

Figure 2–1 21264/EV68A Block Diagram
Instruction Cache

Ibox
Fetch Unit

Virtual Address

VPC
Queue

Four
Instructions

ITB

Next Address

Physical
Address

Predecode

Retire
Unit

Branch
Predictor

Integer Issue Queue
(20 Entries)

Decode and
Rename Registers

FP Issue Queue
(15 Entries)

128

Cbox

Cache
Data

Probe
Queue
Ebox

128

Fbox
Address
ALU 0
(L0)

INT
UNIT
0
(U0)

Integer Registers 0
(80 Registers)

INT
UNIT
1
(U1)

FP
ADD
DIV
SQRT

Address
ALU 1
(L1)

Integer Registers 1
(80 Registers)

FP
MUL

FP Registers
(72 Registers)

Duplicate
Tag Store

Cache
Index
20
System
Bus

IOWB

Victim
Buffer

System
Address
15

Arbiter

Data

Mbox
DTB
(Dual-ported, 128-entry)

Load
Queue

Physical
Address

Store
Queue

Miss Address
File

128

Data

Dual-Ported Data Cache
FM-05642-AI4

2.1.1.2 Branch Predictor

The branch predictor is composed of three units: the local, global, and choice predictors. Figure 2–2 shows how the branch predictor generates the predicted branch
address.

21264/EV68A Hardware Reference Manual

Internal Architecture

2–3

21264/EV68A Microarchitecture

Figure 2–2 Branch Predictor
Local
Predictor

Global
Predictor

Choice
Predictor

Predicted
Branch
Address
FM-05810.AI4

Local Predictor

The local predictor uses a 2-level table that holds the history of individual branches.
The 2-level table design approaches the prediction accuracy of a larger single-level
table while requiring fewer total bits of storage. Figure 2–3 shows how the local predictor generates a prediction. Bits [11:2] of the VPC of the current branch are used as
the index to a 1K entry table in which each entry is a 10-bit value. This 10-bit value is
used as the index to a 1K entry table of 3-bit saturating counters. The value of the saturating counter determines the predication, taken/not-taken, of the current branch.
Figure 2–3 Local Predictor
VPC[11:2]

Local
History
Table
1K x 10

10
10
Index

Local
Predictor
1K x 3

+/-

3
3
1
Local Branch Prediction
FM-05811.AI4

Global Predictor

The global predictor is indexed by a global history of all recent branches. The global
predictor correlates the local history of the current branch with all recent branches. Figure 2–4 shows how the global predictor generates a prediction. The global path history
is comprised of the taken/not-taken state of the 12 most-recent branches. These 12
states are used to form an index into a 4K entry table of 2-bit saturating counters. The
value of the saturating counter determines the predication, taken/not-taken, of the current branch.

2–4

Internal Architecture

21264/EV68A Hardware Reference Manual

21264/EV68A Microarchitecture

Figure 2–4 Global Predictor
Global
Path
History
12
Index

Global
Predictor
4K x 2

+/-

2
2
1
Global Branch Prediction
FM-05812.AI4

Choice Predictor

The choice predictor monitors the history of the local and global predictors and chooses
the best of the two predictors for a particular branch. Figure 2–5 shows how the choice
predictor generates its choice of the result of the local or global prediction. The 12-bit
global path history (see Figure 2–4) is used to index a 4K entry table of 2-bit saturating
counters. The value of the saturating counter determines the choice between the outputs
of the local and global predictors.
Figure 2–5 Choice Predictor
Global
Path
History
12

2
Choice
Predictor
4K x 2

Choice Prediction
FM-05813.AI4

2.1.1.3 Instruction-Stream Translation Buffer

The Ibox includes a 128-entry, fully-associative instruction-stream translation buffer
(ITB) that is used to store recently used instruction-stream (Istream) address translations and page protection information. Each of the entries in the ITB can map 1, 8, 64,
or 512 contiguous 8KB pages. The allocation scheme is round-robin.
The ITB supports an 8-bit ASN and contains an ASM bit. The Icache is virtually
addressed and contains the access-check information, so the ITB is accessed only for
Istream references that miss in the Icache.
Istream transactions to I/O address space are UNDEFINED.

21264/EV68A Hardware Reference Manual

Internal Architecture

2–5

21264/EV68A Microarchitecture

2.1.1.4 Instruction Fetch Logic

The instruction prefetcher (predecode) reads an octaword, containing up to four naturally aligned instructions per cycle, from the Icache. Branch prediction and line prediction bits accompany the four instructions. The branch prediction scheme operates most
efficiently when only one branch instruction is contained among the four fetched
instructions. The line prediction scheme attempts to predict the Icache line that the
branch predictor will generate, and is described in Section 2.2.
An entry from the subroutine return prediction stack, together with set prediction bits
for use by the Icache stream controller, are fetched along with the octaword. The Icache
stream controller generates fetch requests for additional Icache lines and stores the
Istream data in the Icache. There is no separate buffer to hold Istream requests.
2.1.1.5 Register Rename Maps

The instruction prefetcher forwards instructions to the integer and floating-point register rename maps. The rename maps perform the two functions listed here:

•

Eliminate register write-after-read (WAR) and write-after-write (WAW) data
dependencies while preserving true read-after-write (RAW) data dependencies, in
order to allow instructions to be dynamically rescheduled.

•

Provide a means of speculatively executing instructions before the control flow
previous to those instructions is resolved. Both exceptions and branch
mispredictions represent deviations from the control flow predicted by the
instruction prefetcher.

The map logic translates each instruction’s operand register specifiers from the virtual
register numbers in the instruction to the physical register numbers that hold the corresponding architecturally-correct values. The map logic also renames each instruction’s
destination register specifier from the virtual number in the instruction to a physical
register number chosen from a list of free physical registers, and updates the register
maps.
The map logic can process four instructions per cycle. It does not return the physical
register, which holds the old value of an instruction’s virtual destination register, to the
free list until the instruction has been retired, indicating that the control flow up to that
instruction has been resolved.
If a branch mispredict or exception occurs, the map logic backs up the contents of the
integer and floating-point register rename maps to the state associated with the instruction that triggered the condition, and the prefetcher restarts at the appropriate VPC. At
most, 20 valid fetch slots containing up to 80 instructions can be in flight between the
register maps and the end of the machine’s pipeline, where the control flow is finally
resolved. The map logic is capable of backing up the contents of the maps to the state
associated with any of these 80 instructions in a single cycle.
The register rename logic places instructions into an integer or floating-point issue
queue, from which they are later issued to functional units for execution.
2.1.1.6 Integer Issue Queue

The 20-entry integer issue queue (IQ), associated with the integer execution units
(Ebox), issues the following types of instructions at a maximum rate of four per cycle:

2–6

Internal Architecture

21264/EV68A Hardware Reference Manual

21264/EV68A Microarchitecture

•

Integer operate

•

Integer conditional branch

•

Unconditional branch – both displacement and memory format

•

Integer and floating-point load and store

•

PAL-reserved instructions: HW_MTPR, HW_MFPR, HW_LD, HW_ST,
HW_RET

•

Integer-to-floating-point (ITOFx) and floating-point-to-integer (FTOIx)

Each queue entry asserts four request signals—one for each of the Ebox subclusters. A
queue entry asserts a request when it contains an instruction that can be executed by the
subcluster, if the instruction’s operand register values are available within the subcluster.
There are two arbiters—one for the upper subclusters and one for the lower subclusters.
(Subclusters are described in Section 2.1.2.) Each arbiter picks two of the possible 20
requesters for service each cycle. A given instruction only requests upper subclusters or
lower subclusters, but because many instructions can only be executed in one type or
another this is not too limiting.
For example, load and store instructions can only go to lower subclusters and shift
instructions can only go to upper subclusters. Other instructions, such as addition and
logic operations, can execute in either upper or lower subclusters and are statically
assigned before being placed in the IQ.
The IQ arbiters choose between simultaneous requesters of a subcluster based on the
age of the request—older requests are given priority over newer requests. If a given
instruction requests both lower subclusters, and no older instruction requests a lower
subcluster, then the arbiter assigns subcluster L0 to the instruction. If a given instruction
requests both upper subclusters, and no older instruction requests an upper subcluster,
then the arbiter assigns subcluster U1 to the instruction. This asymmetry between the
upper and lower subcluster arbiters is a circuit implementation optimization with negligible overall performance effect.
2.1.1.7 Floating-Point Issue Queue

The 15-entry floating-point issue queue (FQ) associated with the Fbox issues the following instruction types:

•

Floating-point operates

•

Floating-point conditional branches

•

Floating-point stores

•

Floating-point register to integer register transfers (FTOIx)

Each queue entry has three request lines—one for the add pipeline, one for the multiply
pipeline, and one for the two store pipelines. There are three arbiters—one for each of
the add, multiply, and store pipelines. The add and multiply arbiters pick one requester
per cycle, while the store pipeline arbiter picks two requesters per cycle, one for each
store pipeline.

21264/EV68A Hardware Reference Manual

Internal Architecture

2–7

21264/EV68A Microarchitecture

The FQ arbiters pick between simultaneous requesters of a pipeline based on the age of
the request—older requests are given priority over newer requests. Floating-point store
instructions and FTOIx instructions in even-numbered queue entries arbitrate for one
store port. Floating-point store instructions and FTOIx instructions in odd-numbered
queue entries arbitrate for the second store port.
Floating-point store instructions and FTOIx instructions are queued in both the integer
and floating-point queues. They wait in the floating-point queue until their operand register values are available. They subsequently request service from the store arbiter.
Upon being issued from the floating-point queue, they signal the corresponding entry in
the integer queue to request service. Upon being issued from the integer queue, the
operation is completed.
2.1.1.8 Exception and Interrupt Logic

There are two types of exceptions: faults and synchronous traps. Arithmetic exceptions
are precise and are reported as synchronous traps.
The four sources of interrupts are listed as follows:

•

Level-sensitive hardware interrupts sourced by the IRQ_H[5:0] pins

•

Edge-sensitive hardware interrupts generated by the serial line receive pin,
performance counter overflows, and hardware corrected read errors

•

Software interrupts sourced by the software interrupt request (SIRR) register

•

Asynchronous system traps (ASTs)

Interrupt sources can be individually masked. In addition, AST interrupts are qualified
by the current processor mode.
2.1.1.9 Retire Logic

The Ibox fetches instructions in program order, executes them out of order, and then
retires them in order. The Ibox retire logic maintains the architectural state of the
machine by retiring an instruction only if all previous instructions have executed without generating exceptions or branch mispredictions. Retiring an instruction commits the
machine to any changes the instruction may have made to the software-visible state.
The three software-visible states are listed as follows:

•

Integer and floating-point registers

•

Memory

•

Internal processor registers (including control/status registers and translation
buffers)

The retire logic can sustain a maximum retire rate of eight instructions per cycle, and
can retire up to as many as 11 instructions in a single cycle.

2.1.2 Integer Execution Unit
The integer execution unit (Ebox) is a 4-path integer execution unit that is implemented
as two functional-unit “clusters” labeled 0 and 1. Each cluster contains a copy of an 80entry, physical-register file and two “subclusters”, named upper (U) and lower (L). Figure 2–6 shows the integer execution unit. In the figure, iop_wr is the cross-cluster bus
for moving integer result values between clusters.
2–8

Internal Architecture

21264/EV68A Hardware Reference Manual

21264/EV68A Microarchitecture

Figure 2–6 Integer Execution Unit—Clusters 0 and 1
iop_wr
iop_wr

iop_wr
iop_wr
Load/Store Data
Load/Store Data
eff_VA

eff_VA
FM-05643.AI4

Most instructions have 1-cycle latency for consumers that execute within the same cluster. Also, there is another 1-cycle delay associated with producing a value in one cluster
and consuming the value in the other cluster. The instruction issue queue minimizes the
performance effect of this cross-cluster delay. The Ebox contains the following
resources:

•

Four 64-bit adders that are used to calculate results for integer add instructions
(located in U0, U1, L0, and L1)

•

The adders in the lower subclusters that are used to generate the effective virtual
address for load and store instructions (located in L0 and L1)

•

Four logic units

•

Two barrel shifters and associated byte logic (located in U0 and U1)

•

Two sets of conditional branch logic (located in U0 and U1)

•

Two copies of an 80-entry register file

•

One pipelined multiplier (located in U1) with 7-cycle latency for all integer multiply
operations

•

One fully-pipelined unit (located in U0), with 3-cycle latency, that executes the following instructions:
–

CTLZ, CTPOP, CTTZ

–

PERR, MINxxx, MAXxxx, UNPKxx, PKxx

21264/EV68A Hardware Reference Manual

Internal Architecture

2–9

21264/EV68A Microarchitecture

The Ebox has 80 register-file entries that contain storage for the values of the 31 Alpha
integer registers (the value of R31 is not stored), the values of 8 PALshadow registers,
and 41 results written by instructions that have not yet been retired.
Ignoring cross-cluster delay, the two copies of the Ebox register file contain identical
values. Each copy of the Ebox register file contains four read ports and six write ports.
The four read ports are used to source operands to each of the two subclusters within a
cluster. The six write ports are used as follows:

•

Two write ports are used to write results generated within the cluster.

•

Two write ports are used to write results generated by the other cluster.

•

Two write ports are used to write results from load instructions. These two ports
are also used for FTOIx instructions.

2.1.3 Floating-Point Execution Unit
The floating-point execution unit (Fbox) has two paths. The Fbox executes both VAX
and IEEE floating-point instructions. It supports IEEE S_floating-point and T_floatingpoint data types and all rounding modes. It also supports VAX F_floating-point and
G_floating-point data types, and provides limited support for D_floating-point format.
The basic structure of the floating-point execution unit is shown in Figure 2–7.
Figure 2–7 Floating-Point Execution Units

Floating-Point
Execution Units

FP Mul

Reg
FP Add
FP Div
SQRT
LK98-0004A

The Fbox contains the following resources:

•

72-entry physical register file

•

Fully-pipelined multiplier with 4-cycle latency

•

Fully-pipelined adder with 4-cycle latency

•

Nonpipelined divide unit associated with the adder pipeline

•

Nonpipelined square root unit associated with the adder pipeline

The 72 Fbox register file entries contain storage for the values of the 31 Alpha floatingpoint registers (F31 is not stored) and 41 values written by instructions that have not
been retired.

2–10

Internal Architecture

21264/EV68A Hardware Reference Manual

21264/EV68A Microarchitecture

The Fbox register file contains six reads ports and four write ports. Four read ports are
used to source operands to the add and multiply pipelines, and two read ports are used
to source data for store instructions. Two write ports are used to write results generated
by the add and multiply pipelines, and two write ports are used to write results from
floating-point load instructions.

2.1.4 External Cache and System Interface Unit
The interface for the system and external cache (Cbox) controls the Bcache and system
ports. It contains the following structures:

•

Victim address file (VAF)

•

Victim data file (VDF)

•

I/O write buffer (IOWB)

•

Probe queue (PQ)

•

Duplicate Dcache tag (DTAG)

2.1.4.1 Victim Address File and Victim Data File

The victim address file (VAF) and victim data file (VDF) together form an 8-entry victim buffer used for holding:

•

Dcache blocks to be written to the Bcache

•

Istream cache blocks from memory to be written to the Bcache

•

Bcache blocks to be written to memory

•

Cache blocks sent to the system in response to probe commands

2.1.4.2 I/O Write Buffer

The I/O write buffer (IOWB) consists of four 64-byte entries and associated address
and control logic used for buffering I/O write data between the store queue and the system port.
2.1.4.3 Probe Queue

The probe queue (PQ) is an 8-entry queue that holds pending system port cache probe
commands and addresses.
2.1.4.4 Duplicate Dcache Tag Array

The duplicate Dcache tag (DTAG) array holds a duplicate copy of the Dcache tags and
is used by the Cbox when processing Dcache fills, Icache fills, and system port probes.

2.1.5 Onchip Caches
The 21264/EV68A contains two onchip primary-level caches.
2.1.5.1 Instruction Cache

The instruction cache (Icache) is a 64KB virtual-addressed, 2-way set-predict cache.
Set prediction is used to approximate the performance of a 2-set cache without slowing
the cache access time. Each Icache block contains:

•

16 Alpha instructions (64 bytes)

21264/EV68A Hardware Reference Manual

Internal Architecture

2–11

21264/EV68A Microarchitecture

•

Virtual tag bits [47:15]

•

8-bit address space number (ASN) field

•

1-bit address space match (ASM) bit

•

1-bit PALcode bit to indicate physical addressing

•

Valid bit

•

Data and tag parity bits

•

Four access-check bits for the following modes: kernel, executive, supervisor, and
user (KESU)

•

Additional predecoded information to assist with instruction processing and fetch
control

2.1.5.2 Data Cache

The data cache (Dcache) is a 64KB, 2-way set-associative, virtually indexed, physically
tagged, write-back, read/write allocate cache with 64-byte blocks. During each cycle
the Dcache can perform one of the following transactions:

•

Two quadword (or shorter) read transactions to arbitrary addresses

•

Two quadword write transactions to the same aligned octaword

•

Two non-overlapping less-than-quadword writes to the same aligned quadword

•

One sequential read and write transaction from and to the same aligned octaword

Each Dcache block contains:

•

64 data bytes and associated quadword ECC bits

•

Physical tag bits

•

Valid, dirty, shared, and modified bits

•

Tag parity bit calculated across the tag, dirty, shared, and modified bits

•

One bit to control round-robin set allocation (one bit per two cache blocks)

The Dcache contains two sets, each with 512 rows containing 64-byte blocks per row
(that is, 32K bytes of data per set). The 21264/EV68A requires two additional bits of
virtual address beyond the bits that specify an 8KB page, in order to specify a Dcache
row index. A given virtual address might be found in four unique locations in the
Dcache, depending on the virtual-to-physical translation for those two bits. The 21264/
EV68A prevents this aliasing by keeping only one of the four possible translated
addresses in the cache at any time.

2.1.6 Memory Reference Unit
The memory reference unit (Mbox) controls the Dcache and ensures architecturally
correct behavior for load and store instructions. The Mbox contains the following structures:

2–12

•

Load queue (LQ)

•

Store queue (SQ)

Internal Architecture

21264/EV68A Hardware Reference Manual

Pipeline Organization

•

Miss address file (MAF)

•

Dstream translation buffer (DTB)

2.1.6.1 Load Queue

The load queue (LQ) is a reorder buffer for load instructions. It contains 32 entries and
maintains the state associated with load instructions that have been issued to the Mbox,
but for which results have not been delivered to the processor and the instructions
retired. The Mbox assigns load instructions to LQ slots based on the order in which
they were fetched from the Icache, then places them into the LQ after they are issued by
the IQ. The LQ helps ensure correct Alpha memory reference behavior.
2.1.6.2 Store Queue

The store queue (SQ) is a reorder buffer and graduation unit for store instructions. It
contains 32 entries and maintains the state associated with store instructions that have
been issued to the Mbox, but for which data has not been written to the Dcache and the
instruction retired. The Mbox assigns store instructions to SQ slots based on the order
in which they were fetched from the Icache and places them into the SQ after they are
issued by the IQ. The SQ holds data associated with store instructions issued from the
IQ until they are retired, at which point the store can be allowed to update the Dcache.
The SQ also helps ensure correct Alpha memory reference behavior.
2.1.6.3 Miss Address File

The 8-entry miss address file (MAF) holds physical addresses associated with pending
Icache and Dcache fill requests and pending I/O space read transactions.
2.1.6.4 Dstream Translation Buffer

The Mbox includes a 128-entry, fully associative Dstream translation buffer (DTB) used
to store Dstream address translations and page protection information. Each of the entries
in the DTB can map 1, 8, 64, or 512 contiguous 8KB pages. The allocation scheme is
round-robin. The DTB supports an 8-bit ASN and contains an ASM bit.

2.1.7 SROM Interface
The serial read-only memory (SROM) interface provides the initialization data load
path from a system SROM to the Icache. Refer to Chapter 7 for more information.

2.2 Pipeline Organization
The 7-stage pipeline provides an optimized environment for executing Alpha instructions. The pipeline stages (0 to 6) are shown in Figure 2–8 and described in the following paragraphs.

21264/EV68A Hardware Reference Manual

Internal Architecture

2–13

Pipeline Organization

Figure 2–8 Pipeline Organization

Branch
Predictor

ALU
Shifter
Integer
Register
Rename
Map

Integer
Issue
Queue
(20)

Integer
Register
File

ALU Shifter
Multiplier

ALU

Address
64KB
Data
Cache

Four
Instructions

Instruction
Cache
(64KB)
(2-Set)

FloatingPoint
Register
Rename
Map

System
Bus
(64 Bits)

Address
ALU

FloatingPoint
Issue
Queue
(15)

FloatingPoint
Register
File

Floating-Point
Add, Divide,
and Square Root
Floating-Point
Multiply

Bus
Interface
Unit

Cache
Bus
(128 Bits)

Physical
Address
(44 Bits)
FM-05575.AI4

Stage 0 — Instruction Fetch

The branch predictor uses a branch history algorithm to predict a branch instruction target address.
Up to four aligned instructions are fetched from the Icache, in program order. The
branch prediction tables are also accessed in this cycle. The branch predictor uses tables
and a branch history algorithm to predict a branch instruction target address for one
branch or memory format JSR instruction per cycle. Therefore, the prefetcher is limited
to fetching through one branch per cycle. If there is more than one branch within the
fetch line, and the branch predictor predicts that the first branch will not be taken, it will
predict through subsequent branches at the rate of one per cycle, until it predicts a taken
branch or predicts through the last branch in the fetch line.
The Icache array also contains a line prediction field, the contents of which are applied
to the Icache in the next cycle. The purpose of the line predictor is to remove the pipeline bubble which would otherwise be created when the branch predictor predicts a
branch to be taken. In effect, the line predictor attempts to predict the Icache line which
the branch predictor will generate. On fills, the line predictor value at each fetch line is
initialized with the index of the next sequential fetch line, and later retrained by the
branch predictor if necessary.
Stage 1 — Instruction Slot

The Ibox maps four instructions per cycle from the 64KB 2-way set-predict Icache.
Instructions are mapped in order, executed dynamically, but are retired in order.

2–14

Internal Architecture

21264/EV68A Hardware Reference Manual

Pipeline Organization

In the slot stage, the branch predictor compares the next Icache index that it generates to
the index that was generated by the line predictor. If there is a mismatch, the branch
predictor wins—the instructions fetched during that cycle are aborted, and the index
predicted by the branch predictor is applied to the Icache during the next cycle. Line
mispredictions result in one pipeline bubble.
The line predictor takes precedence over the branch predictor during memory format
calls or jumps. If the line predictor was trained with a true (as opposed to predicted)
memory format call or jump target, then its contents take precedence over the target
hint field associated with these instructions. This allows dynamic calls or jumps to be
correctly predicted.
The instruction fetcher produces the full VPC address during the fetch stage of the pipeline. The Icache produces the tags for both Icache sets 0 and 1 each time it is accessed.
That enables the fetcher to separate set mispredictions from true Icache misses. If the
access was caused by a set misprediction, the instruction fetcher aborts the last two
fetched slots and refetches the slot in the next cycle. It also retrains the appropriate set
prediction bits.
The instruction data is transferred from the Icache to the integer and floating-point register map hardware during this stage. When the integer instruction is fetched from the
Icache and slotted into the IQ, the slot logic determines whether the instruction is for
the upper or lower subclusters. The slot logic makes the decision based on the
resources needed by the (up to four) integer instructions in the fetch block. Although all
four instructions need not be issued simultaneously, distributing their resource usage
improves instruction loading across the units. For example, if a fetch block contains
two instructions that can be placed in either cluster followed by two instructions that
must execute in the lower cluster, the slot logic would designate that combination as
EELL and slot them as UULL. Slot combinations are described in Section 2.3.2 and
Table 2–3.
Stage 2 — Map

Instructions are sent from the Icache to the integer and floating-point register maps during the slot stage and register renaming is performed during the map stage. Also, each
instruction is assigned a unique 8-bit number, called an inum, which is used to identify
the instruction and its program order with respect to other instructions during the time
that it is in flight. Instructions are considered to be in flight between the time they are
mapped and the time they are retired.
Mapped instructions and their associated inums are placed in the integer and floatingpoint queues by the end of the map stage.
Stage 3 — Issue

The 20-entry integer issue queue (IQ) issues instructions at the rate of four per cycle.
The 15-entry floating-point issue queue (FQ) issues floating-point operate instructions,
conditional branch instructions, and store instructions, at the rate of two per cycle. Normally, instructions are deleted from the IQ or FQ two cycles after they are issued. For
example, if an instruction is issued in cycle n, it remains in the FQ or IQ in cycle n+1
but does not request service, and is deleted in cycle n+2.

21264/EV68A Hardware Reference Manual

Internal Architecture

2–15

Instruction Issue Rules

Stage 4 — Register Read

Instructions issued from the issue queues read their operands from the integer and floating-point register files and receive bypass data.
Stage 5 — Execute

The Ebox and Fbox pipelines begin execution.
Stage 6 — Dcache Access

Memory reference instructions access the Dcache and data translation buffers. Normally load instructions access the tag and data arrays while store instructions only
access the tag arrays. Store data is written to the store queue where it is held until the
store instruction is retired. Most integer operate instructions write their register results
in this cycle.

2.2.1 Pipeline Aborts
The abort penalty as given is measured from the cycle after the fetch stage of the
instruction which triggers the abort to the fetch stage of the new target, ignoring any
Ibox pipeline stalls or queuing delay that the triggering instruction might experience.
Table 2–1 lists the timing associated with each common source of pipeline abort.
Table 2–1 Pipeline Abort Delay (GCLK Cycles)
Abort Condition

Penalty
(Cycles)

Branch misprediction

Integer or floating-point conditional branch
misprediction.

JSR misprediction

Memory format JSR or HW_RET.

Mbox order trap

Load-load order or store-load order.

Other Mbox replay traps

—

DTB miss

—

ITB miss

—

Integer arithmetic trap

—

Floating-point arithmetic
trap

13+latency

Add latency of instruction. See Section 2.3.3 for
instruction latencies.

Comments

2.3 Instruction Issue Rules
This section defines instruction classes, the functional unit pipelines to which they are
issued, and their associated latencies.

2–16

Internal Architecture

21264/EV68A Hardware Reference Manual

Instruction Issue Rules

2.3.1 Instruction Group Definitions
Table 2–2 lists the instruction class, the pipeline assignments, and the instructions
included in the class.
Table 2–2 Instruction Name, Pipeline, and Types
Class
Name

Pipeline

Instruction Type

ild

L0, L1

All integer load instructions

fld

L0, L1

All floating-point load instructions

ist

L0, L1

All integer store instructions

fst

FST0, FST1, L0, L1 All floating-point store instructions

lda

L0, L1, U0, U1

LDA, LDAH

mem_misc

WH64, ECB, WMB

rpcc

RPCC

RS, RC

mxpr

L0, L1
(depends on IPR)

HW_MTPR, HW_MFPR

icbr

U0, U1

Integer conditional branch instructions

jsr

BR, BSR, JMP, CALL, RET, COR, HW_RET,
CALL_PAL

iadd

L0, U0, L1, U1

Instructions with opcode 1016, except CMPBGE

ilog

L0, U0, L1, U1

AND, BIC, BIS, ORNOT, XOR, EQV, CMPBGE

ishf

U0, U1

Instructions with opcode 1216

cmov

L0, U0, L1, U1

Integer CMOV — either cluster

imul

Integer multiply instructions

imisc

CTLZ, CTPOP, CTTZ , PERR, MINxxx, MAXxxx,
PKxx, UNPKxx

fcbr

Floating-point conditional branch instructions

fadd

All floating-point operate instructions except multiply,
divide, square root, and conditional move instructions

fmul

Floating-point multiply instruction

fcmov1

Floating-point CMOV—first half

fcmov2

Floating-point CMOV— second half

fdiv

Floating-point divide instruction

fsqrt

Floating-point square root instruction

nop

None

TRAP, EXCB, UNOP - LDQ_U R31, 0(Rx)

21264/EV68A Hardware Reference Manual

Internal Architecture

2–17

Instruction Issue Rules

Table 2–2 Instruction Name, Pipeline, and Types (Continued)
Class
Name

Pipeline

ftoi

FST0, FST1, L0, L1 FTOIS, FTOIT

itof

L0, L1

ITOFS, ITOFF, ITOFT

mx_fpcr

Instructions that move data from the floating-point
control register

Instruction Type

2.3.2 Ebox Slotting
Instructions that are issued from the IQ, and could execute in either upper or lower
Ebox subclusters, are slotted to one pair or the other during the pipeline mapping stage
based on the instruction mixture in the fetch line. The codes that are used in Table 2–3
are as follows:

•

U—The instruction only executes in an upper subcluster.

•

L—The instruction only executes in a lower subcluster.

•

E—The instruction could execute in either an upper or lower subcluster.

Table 2–3 defines the slotting rules. The table field Instruction Class 3, 2, 1 and 0 identifies each instruction’s location in the fetch line by the value of bits [3:2] in its PC.
Table 2–3 Instruction Group Definitions and Pipeline Unit

2–18

Instruction Class
3210

Slotting
3210

Instruction Class
3210

Slotting
3210

EEEE

ULUL

LLLL

EEEL

ULUL

LLLU

EEEU

ULLU

LLUE

LLUU

EELE

ULLU

LLUL

EELL

UULL

LLUU

EELU

ULLU

LUEE

LULU

EEUE

ULUL

LUEL

LUUL

EEUL

ULUL

LUEU

LULU

EEUU

LLUU

LULE

LULU

ELEE

ULUL

LULL

ELEL

ULUL

LULU

ELEU

ULLU

LUUE

LUUL

ELLE

ULLU

LUUL

ELLL

ULLL

LUUU

ELLU

ULLU

UEEE

ULUL

ELUE

ULUL

UEEL

ULUL

ELUL

ULUL

UEEU

ULLU

Internal Architecture

21264/EV68A Hardware Reference Manual

Instruction Issue Rules

Table 2–3 Instruction Group Definitions and Pipeline Unit (Continued)
Instruction Class
3210

Slotting
3210

Instruction Class
3210

Slotting
3210

ELUU

LLUU

UELE

ULLU

EUEE

LULU

UELL

UULL

EUEL

LUUL

UELU

ULLU

EUEU

LULU

UEUE

ULUL

EULE

LULU

UEUL

ULUL

EULL

UULL

UEUU

ULUU

EULU

LULU

ULEE

ULUL

EUUE

LUUL

ULEL

ULUL

EUUL

LUUL

ULEU

ULLU

EUUU

LUUU

ULLE

ULLU

LEEE

LULU

ULLL

U L LL

LEEL

LUUL

ULLU

LEEU

LULU

ULUE

ULUL

LELE

LULU

ULUL

LELL

LULL

ULUU

LELU

LULU

UUEE

UULL

LEUE

LUUL

UUEL

UULL

LEUL

LUUL

UUEU

UULU

LEUU

LLUU

UULE

UULL

LLEE

LLUU

UULL

LLEL

LLUL

UULU

LLEU

LLUU

UUUE

UUUL

LLLE

LLLU

UUUL

—

UUUU

21264/EV68A Hardware Reference Manual

Internal Architecture

2–19

Instruction Issue Rules

2.3.3 Instruction Latencies
After an instruction is placed in the IQ or FQ, its issue point is determined by the availability of its register operands, functional unit(s), and relationship to other instructions
in the queue. There are register producer-consumer dependencies and dynamic functional unit availability dependencies that affect instruction issue. The mapper removes
register producer-producer dependencies.
The latency to produce a register result is generally fixed. The one exception is for load
instructions that miss the Dcache. Table 2–4 lists the latency, in cycles, for each
instruction class.
Table 2–4 Instruction Class Latency in Cycles
Class

Latency

Comments

ild

3
13+

Dcache hit.
Dcache miss, latency with 6-cycle Bcache. Add additional Bcache loop latency if
Bcache latency is greater than 6 cycles.

fld

4
14+

Dcache hit.
Dcache miss, latency with 6-cycle Bcache. Add additional Bcache loop latency if
Bcache latency is greater than 6 cycles.

lda

Possible 1-cycle Ebox cross-cluster delay.

mem_misc —

Does not produce register value.

ist

—

Does not produce register value.

fst

—

Does not produce register value.

rpcc

Possible 1-cycle cross-cluster delay.

—

mxpr

1 or 3

HW_MFPR:

icbr

—

Conditional branch. Does not produce register value.

ubr

Unconditional branch. Does not produce register value.

jsr

—

iadd

Possible 1-cycle Ebox cross-cluster delay.

ilog

Possible 1-cycle Ebox cross-cluster delay.

ishf

Possible 1-cycle Ebox cross-cluster delay.

cmov1

Only consumer is cmov2. Possible 1-cycle Ebox cross-cluster delay.

cmov2

Possible 1-cycle Ebox cross-cluster delay.

imul

Possible 1-cycle Ebox cross-cluster delay.

imisc

Possible 1-cycle Ebox cross-cluster delay.

fcbr

—

Does not produce register value.

2–20

Ebox IPRs = 1.
Ibox and Mbox IPRs = 3.
HW_MTPR does not produce a register value.

Internal Architecture

21264/EV68A Hardware Reference Manual

Instruction Retire Rules

Table 2–4 Instruction Class Latency in Cycles (Continued)
Class

Latency

Comments

fadd

4
6

Consumer other than fst or ftoi.
Consumer fst or ftoi.
Measured from when an fadd is issued from the FQ to when an fst or ftoi is issued
from the IQ.

fmul

4
6

Consumer other than fst or ftoi.
Consumer fst or ftoi.
Measured from when an fmul is issued from the FQ to when an fst or ftoi is issued
from the IQ.

fcmov1

Only consumer is fcmov2.

fcmov2

4
6

Consumer other than fst.
Consumer fst or ftoi.
Measured from when an fcmov2 is issued from the FQ to when an fst or ftoi is
issued from the IQ.

fdiv

12
9
15
12

Single precision - latency to consumer of result value.
Single precision - latency to using divider again.
Double precision - latency to consumer of result value.
Double precision - latency to using divider again.

fsqrt

18
15
33
30

Single precision - latency to consumer of result value.
Single precision - latency to using unit again.
Double precision - latency to consumer of result value.
Double precision - latency to using unit again.

ftoi

—

itof

—

nop

—

Does not produce register value.

2.4 Instruction Retire Rules
An instruction is retired when it has been executed to completion, and all previous
instructions have been retired. The execution pipeline stage in which an instruction
becomes eligible to be retired depends upon the instruction’s class.
Table 2–5 gives the minimum retire latencies (assuming that all previous instructions
have been retired) for various classes of instructions.
Table 2–5 Minimum Retire Latencies for Instruction Classes
Instruction Class

Retire Stage

Comments

Integer conditional branch

—

Integer multiply

7/13

Latency is 13 cycles for the MUL/V instruction.

Integer operate

—

Memory

—

Floating-point add

—

Floating-point multiply

—

21264/EV68A Hardware Reference Manual

Internal Architecture

2–21

Retire of Operate Instructions into R31/F31

Table 2–5 Minimum Retire Latencies for Instruction Classes (Continued)
Instruction Class

Retire Stage

Comments

Floating-point DIV/SQRT

11 + latency

Add latency of unit reuse for the instruction indicated in Table
2–4. For example, latency for a single-precision fdiv would be
11 plus 9 from Table 2–4. Latency is 11 if hardware detects that
no exception is possible (see Section 2.4.1).

Floating-point conditional
branch

Branch instruction mispredict is reported in stage 7.

BSR/JSR

JSR instruction mispredict is reported in stage 8.

2.4.1 Floating-Point Divide/Square Root Early Retire
The floating-point divider and square root unit can detect that, for many combinations
of source operand values, no exception can be generated. Instructions with these operands can be retired before the result is generated. When detected, they are retired with
the same latency as the FP add class. Early retirement is not possible for the following
instruction/operand/architecture state conditions:

•

Instruction is not a DIV or SQRT.

•

SQRT source operand is negative.

•

Divide operand exponent_a is 0.

•

Either operand is NaN or INF.

•

Divide operand exponent_b is 0.

•

Trapping mode is /I (inexact).

•

INE status bit is 0.

Early retirement is also not possible for divide instructions if the resulting exponent has
any of the following characteristics (EXP is the result exponent):

•

DIVT, DIVG: (EXP >= 3FF16) OR (EXP <= 216)

•

DIVS, DIVF: (EXP >= 7F16) OR (EXP <= 38216)

2.5 Retire of Operate Instructions into R31/F31
Many instructions that have R31 or F31 as their destination are retired immediately
upon decode (stage 3). These instructions do not produce a result and are removed from
the pipeline as well. They do not occupy a slot in the issue queues and do not occupy a
functional unit. Table 2–6 lists these instructions and some of their characteristics. The
instruction type in Table 2–6 is from Table C-6 in Appendix C of the Alpha Architecture
Handbook, Version 4.

2–22

Internal Architecture

21264/EV68A Hardware Reference Manual

Load Instructions to R31 and F31

Table 2–6 Instructions Retired Without Execution
Instruction Type

Notes

INTA, INTL, INTM, INTS

All with R31 as destination.

FLTI, FLTL, FLTV

All with F31 as destination. MT_FPCR is not included
because it has no destination—it is never removed from the
pipeline.

LDQ_U

All with R31 as destination.

MISC

TRAPB and EXCB are always removed. Others are never
removed.

FLTS

All (SQRT, ITOF) with F31 as destination.

2.6 Load Instructions to R31 and F31
This section describes how the 21264/EV68A processes software-directed prefetch
transactions and load instructions with a destination of R31 and F31.
Prefetches allocate a MAF entry. How the MAF entry is allocated is what distinguishes
the type of prefetch. A normal prefetch is equivalent to a normal load MAF (that is, a
MAF entry that puts the block into the Dcache in a readable state). A prefetch with
modify intent is equivalent to a normal store MAF (that is, a MAF entry that puts the
block into the Dcache in a writeable state). A prefetch, evict next, is equivalent to a normal load MAF, with the additional behavior described in Section 2.6.3.
A prefetch is not performed if the prefetch hits in the Dcache (as if it were a normal
load).
Load operations to R31 and F31 may generate exceptions. These exceptions must be
dismissed by PALcode.
The following sections describe the operational prefetch behavior of these instructions.

2.6.1 Normal Prefetch: LDBU, LDF, LDG, LDL, LDT, LDWU, HW_LDL Instructions
The 21264/EV68A processes these instructions as normal cache line prefetches. If the
load instruction hits the Dcache, the instruction is dismissed, otherwise the addressed
cache block is allocated into the Dcache.
The HW_LDL instruction construct equates to the HW_LD instruction with the LEN
field clear. See Table 6–3.

2.6.2 Prefetch with Modify Intent: LDS Instruction
The 21264/EV68A processes an LDS instruction, with F31 as the destination, as a
prefetch with modify intent transaction (ReadBlkMod command). If the transaction hits
a dirty Dcache block, the instruction is dismissed. Otherwise, the addressed cache block
is allocated into the Dcache for write access, with its dirty and modified bits set.

21264/EV68A Hardware Reference Manual

Internal Architecture

2–23

Special Cases of Alpha Instruction Execution

2.6.3 Prefetch, Evict Next: LDQ and HW_LDQ Instructions
The 21264/EV68A processes this instruction like a normal prefetch transaction (ReadBlkSpec command), with one exception—if the load misses the Dcache, the addressed
cache block is allocated into the Dcache, but the Dcache set allocation pointer is left
pointing to this block. The next miss to the same Dcache line will evict the block. For
example, this instruction might be used when software is reading an array that is known
to fit in the offchip Bcache, but will not fit into the onchip Dcache. In this case, the
instruction ensures that the hardware provides the desired prefetch function without displacing useful cache blocks stored in the other set within the Dcache.
The HW_LDQ instruction construct equates to the HW_LD instruction with the LEN
field set. See Table 6–3.

2.7 Special Cases of Alpha Instruction Execution
This section describes the mechanisms that the 21264/EV68A uses to process irregular
instructions in the Alpha instruction set, and cases in which the 21264/EV68A processes instructions in a non-intuitive way.

2.7.1 Load Hit Speculation
The latency of integer load instructions that hit in the Dcache is three cycles. Figure 2–
9 shows the pipeline timing for these integer load instructions. In Figure 2–9:
Symbol

Meaning

Q
R
E
D
B

Issue queue
Register file read
Execute
Dcache access
Data bus active

Figure 2–9 Pipeline Timing for Integer Load Instructions
Hit
Cycle Number

ILD

Instruction 1
Instruction 2

FM-05814.AI4

There are two cycles in which the IQ may speculatively issue instructions that use load
data before Dcache hit information is known. Any instructions that are issued by the IQ
within this 2-cycle speculative window are kept in the IQ with their requests inhibited
until the load instruction’s hit condition is known, even if they are not dependent on the
load operation. If the load instruction hits, then these instructions are removed from the
queue. If the load instruction misses, then the execution of these instructions is aborted
and the instructions are allowed to request service again.

2–24

Internal Architecture

21264/EV68A Hardware Reference Manual

Special Cases of Alpha Instruction Execution

For example, in Figure 2–9, instruction 1 and instruction 2 are issued within the speculative window of the load instruction. If the load instruction hits, then both instructions
will be deleted from the queue by the start of cycle 7—one cycle later than normal for
instruction 1 and at the normal time for instruction 2. If the load instruction misses, both
instructions are aborted from the execution pipelines and may request service again in
cycle 6.
IQ-issued instructions are aborted if issued within the speculative window of an integer
load instruction that missed in the Dcache, even if they are not dependent on the load
data. However, if software misses are likely, the 21264/EV68A can still benefit from
scheduling the instruction stream for Dcache miss latency. The 21264/EV68A includes
a saturating counter that is incremented when load instructions hit and is decremented
when load instructions miss. When the upper bit of the counter equals zero, the integer
load latency is increased to five cycles and the speculative window is removed. The
counter is 4 bits wide and is incremented by 1 on a hit and is decremented by two on a
miss.
Since load instructions to R31 do not produce a result, they do not create a speculative
window when they execute and, therefore, never waste IQ-issue cycles if they miss.
Floating-point load instructions that hit in the Dcache have a latency of four cycles. Figure 2–10 shows the pipeline timing for floating-point load instructions. In Figure 2–10:
Symbol

Meaning

Q
R
E
D
B

Issue queue
Register file read
Execute
Dcache access
Data bus active

Figure 2–10 Pipeline Timing for Floating-Point Load Instructions
Hit
Cycle Number

FLD

Instruction 1
Instruction 2

R
Q

FM-05815.AI4

The speculative window for floating-point load instructions is one cycle wide.
FQ-issued instructions that are issued within the speculative window of a floating-point
load instruction that has missed, are only aborted if they depend on the load being successful.
For example, in Figure 2–10 instruction 1 is issued in the speculative window of the
load instruction.
If instruction 1 is not a user of the data returned by the load instruction, then it is
removed from the queue at its normal time (at the start of cycle 7).

21264/EV68A Hardware Reference Manual

Internal Architecture

2–25

Special Cases of Alpha Instruction Execution

If instruction 1 is dependent on the load instruction data and the load instruction hits,
instruction 1 is removed from the queue one cycle later (at the start of cycle 8). If the
load instruction misses, then instruction 1 is aborted from the Fbox pipeline and may
request service again in cycle 7.

2.7.2 Floating-Point Store Instructions
Floating-point store instructions are duplicated and loaded into both the IQ and the FQ
from the mapper. Each IQ entry contains a control bit, fpWait, that when set prevents
that entry from asserting its requests. This bit is initially set for each floating-point store
instruction that enters the IQ, unless it was the target of a replay trap. The instruction’s
FQ clone is issued when its Ra register is about to become clean, resulting in its IQ
clone’s fpWait bit being cleared and allowing the IQ clone to issue and be executed by
the Mbox. This mechanism ensures that floating-point store instructions are always
issued to the Mbox, along with the associated data, without requiring the floating-point
register dirty bits to be available within the IQ.

2.7.3 CMOV Instruction
For the 21264/EV68A, the Alpha CMOV instruction has three operands, and so presents a special case. The required operation is to move either the value in register Rb or
the value from the old physical destination register into the new destination register,
based upon the value in Ra. Since neither the mapper nor the Ebox and Fbox data paths
are otherwise required to handle three operand instructions, the CMOV instruction is
decomposed by the Ibox pipeline into two 2-operand instructions:
The Alpha architecture instruction

CMOV Ra, Rb

Becomes the 21264/EV68A instructions

⇒ Rc

CMOV1 Ra, oldRc

CMOV2 newRc1, Rb

⇒ newRc1

⇒ newRc2

The first instruction, CMOV1, tests the value of Ra and records the result of this test in
a 65th bit of its destination register, newRc1. It also copies the value of the old physical
destination register, oldRc, to newRc1.
The second instruction, CMOV2, then copies either the value in newRc1 or the value in
Rb into a second physical destination register, newRc2, based on the CMOV predicate
bit stored in newRc1.
In summary, the original CMOV instruction is decomposed into two dependent instructions that each use a physical register from the free list.
To further simplify this operation, the two component instructions of a CMOV instruction are driven through the mappers in successive cycles. Hence, if a fetch line contains
n CMOV instructions, it takes n+1 cycles to run that fetch line through the mappers.
For example, the following fetch line:
ADD CMOVx SUB CMOVy
Results in the following three map cycles:
ADD CMOVx1
CMOVx2 SUB CMOVy1
CMOVy2
2–26

Internal Architecture

21264/EV68A Hardware Reference Manual

Memory and I/O Address Space Instructions

The Ebox executes integer CMOV instructions as two distinct 1-cycle latency operations. The Fbox add pipeline executes floating-point CMOV instructions as two distinct
4-cycle latency operations.

2.8 Memory and I/O Address Space Instructions
This section provides an overview of the way the 21264/EV68A processes memory and
I/O address space instructions.
The 21264/EV68A supports, and internally recognizes, a 44-bit physical address space
that is divided equally between memory address space and I/O address space. Memory
address space resides in the lower half of the physical address space (PA[43]=0)
and I/O address space resides in the upper half of the physical address space
(PA[43]=1).
The IQ can issue any combination of load and store instructions to the Mbox at the rate
of two per cycle. The two lower Ebox subclusters, L0 and L1, generate the
48-bit effective virtual address for these instructions.
An instruction is defined to be newer than another instruction if it follows that instruction in program order and is older if it precedes that instruction in program order.

2.8.1 Memory Address Space Load Instructions
The Mbox begins execution of a load instruction by translating its virtual address to a
physical address using the DTB and by accessing the Dcache. The Dcache is virtually
indexed, allowing these two operations to be done in parallel. The Mbox puts information about the load instruction, including its physical address, destination register, and
data format, into the LQ.
If the requested physical location is found in the Dcache (a hit), the data is formatted
and written into the appropriate integer or floating-point register. If the location is not in
the Dcache (a miss), the physical address is placed in the miss address file (MAF) for
processing by the Cbox. The MAF performs a merging function in which a new miss
address is compared to miss addresses already held in the MAF. If the new miss address
points to the same Dcache block as a miss address in the MAF, then the new miss
address is discarded.
When Dcache fill data is returned to the Dcache by the Cbox, the Mbox satisfies the
requesting load instructions in the LQ.

2.8.2 I/O Address Space Load Instructions
Because I/O space load instructions may have side effects, they cannot be performed
speculatively. When the Mbox receives an I/O space load instruction, the Mbox places
the load instruction in the LQ, where it is held until it retires. The Mbox replays retired
I/O space load instructions from the LQ to the MAF in program order, at a rate of one
per GCLK cycle.

21264/EV68A Hardware Reference Manual

Internal Architecture

2–27

Memory and I/O Address Space Instructions

The Mbox allocates a new MAF entry to an I/O load instruction and increases I/O bandwidth by attempting to merge I/O load instructions in a merge register. Table 2–7 shows
the rules for merging data. The columns represent the load instructions replayed to the
MAF while the rows represent the size of the load in the merge register.
Table 2–7 Rules for I/O Address Space Load Instruction Data Merging
Merge Register/
Replayed Instruction

Load Byte/Word Load Longword

Load Quadword

Byte/Word

No merge

Longword

No merge

Merge up to 32 bytes

No merge

Quadword

No merge

Merge up to 64 bytes

In summary, Table 2–7 shows some of the following rules:

•

Byte/word load instructions and different size load instructions are not allowed to
merge.

•

A stream of ascending non-overlapping, but not necessarily consecutive, longword
load instructions are allowed to merge into naturally aligned 32-byte blocks.

•

A stream of ascending non-overlapping, but not necessarily consecutive, quadword
load instructions are allowed to merge into naturally aligned 64-byte blocks.

•

Merging of quadwords can be limited to naturally-aligned 32-byte blocks based on
the Cbox WRITE_ONCE chain 32_BYTE_IO field.

•

Issued MB, WMB, and I/O load instructions close the I/O register merge window.
To minimize latency, the merge window is also closed when a timer detects no I/O
store instruction activity for 1024 cycles.

After the Mbox I/O register has closed its merge window, the Cbox sends I/O read
requests offchip in the order that they were received from the Mbox.

2.8.3 Memory Address Space Store Instructions
The Mbox begins execution of a store instruction by translating its virtual address to a
physical address using the DTB and by probing the Dcache. The Mbox puts information about the store instruction, including its physical address, its data and the results of
the Dcache probe, into the store queue (SQ).
If the Mbox does not find the addressed location in the Dcache, it places the address
into the MAF for processing by the Cbox. If the Mbox finds the addressed location in a
Dcache block that is not dirty, then it places a ChangeToDirty request into the MAF.
A store instruction can write its data into the Dcache when it is retired, and when the
Dcache block containing its address is dirty and not shared. SQ entries that meet these
two conditions can be placed into the writable state. These SQ entries are placed into
the writable state in program order at a maximum rate of two entries per cycle. The
Mbox transfers writable store queue entry data from the SQ to the Dcache in program
order at a maximum rate of two entries per cycle. Dcache lines associated with writable
store queue entries are locked by the Mbox. System port probe commands cannot evict
these blocks until their associated writable SQ entries have been transferred into the
Dcache. This restriction assists in STx_C instruction and Dcache ECC processing.

2–28

Internal Architecture

21264/EV68A Hardware Reference Manual

Memory and I/O Address Space Instructions

SQ entry data that has not been transferred to the Dcache may source data to newer load
instructions. The Mbox compares the virtual Dcache index bits of incoming load
instructions to queued SQ entries, and sources the data from the SQ, bypassing the
Dcache, when necessary.

2.8.4 I/O Address Space Store Instructions
The Mbox begins processing I/O space store instructions, like memory space store
instructions, by translating the virtual address and placing the state associated with the
store instruction into the SQ.
The Mbox replays retired I/O space store entries from the SQ to the IOWB in program
order at a rate of one per GCLK cycle. The Mbox never allows queued I/O space store
instructions to source data to subsequent load instructions.
The Cbox maximizes I/O bandwidth when it allocates a new IOWB entry to an I/O
store instruction by attempting to merge I/O store instructions in a merge register. Table
2–8 shows the rules for I/O space store instruction data merging. The columns represent
the load instructions replayed to the IOWB while the rows represent the size of the store
in the merge register.
Table 2–8 Rules for I/O Address Space Store Instruction Data Merging
Merge Register/
Replayed Instruction

Store
Byte/Word

Store Longword

Store Quadword

Byte/Word

No merge

Longword

No merge

Merge up to 32 bytes

No merge

Quadword

No merge

Merge up to 64 bytes

Table 2–8 shows some of the following rules:

•

Byte/word store instructions and different size store instructions are not allowed to
merge.

•

A stream of ascending non-overlapping, but not necessarily consecutive, longword
store instructions are allowed to merge into naturally aligned 32-byte blocks.

•

A stream of ascending non-overlapping, but not necessarily consecutive, quadword
store instructions are allowed to merge into naturally aligned 64-byte blocks.

•

Merging of quadwords can be limited to naturally-aligned 32-byte blocks based on
the Cbox WRITE_ONCE chain 32_BYTE_IO field.

•

After the IOWB merge register has closed its merge window, the Cbox sends I/O space
store requests offchip in the order that they were received from the Mbox.

21264/EV68A Hardware Reference Manual

Internal Architecture

2–29

MAF Memory Address Space Merging Rules

2.9 MAF Memory Address Space Merging Rules
Because all memory transactions are to 64-byte blocks, efficiency is improved by merging several small data transactions into a single larger data transaction. Table 2–9 lists
the rules the 21264/EV68A uses when merging memory transactions into 64-byte naturally aligned data block transactions. Rows represent the merged instruction in the
MAF and columns represent the new issued transaction.
Table 2–9 MAF Merging Rules
MAF/New

LDx

STx

STx_C

WH64

ECB

Istream

LDx

Merge

—

STx

Merge

—

STx_C

—

Merge

—

WH64

—

Merge

—

ECB

—

Merge

—

Istream

—

Merge

In summary, Table 2–9 shows that only like instruction types, with the exception of
load instructions merging with store instructions, are merged.

2.10 Instruction Ordering
In the absence of explicit instruction ordering, such as with MB or WMB instructions,
the 21264/EV68A maintains a default instruction ordering relationship between pairs of
load and store instructions.
The 21264/EV68A maintains the default memory data instruction ordering as shown in
Table 2–10 (assume address X and address Y are different).
Table 2–10 Memory Reference Ordering

2–30

First Instruction in Pair

Second Instruction in Pair

Reference Order

Load memory to address X

Maintained (litmus test 1)

Load memory to address X

Load memory to address Y

Not maintained

Store memory to address X

Maintained

Store memory to address X

Store memory to address Y

Maintained

Load memory to address X

Store memory to address X

Maintained

Load memory to address X

Store memory to address Y

Not maintained

Store memory to address X

Load memory to address X

Maintained

Store memory to address X

Load memory to address Y

Not maintained

Internal Architecture

21264/EV68A Hardware Reference Manual

Replay Traps

The 21264/EV68A maintains the default I/O instruction ordering as shown in Table 2–
11 (assume address X and address Y are different).
Table 2–11 I/O Reference Ordering
First Instruction in Pair

Second Instruction in Pair

Reference Order

Load I/O to address X

Maintained

Load I/O to address X

Load I/O to address Y

Maintained

Store I/O to address X

Maintained

Store I/O to address X

Store I/O to address Y

Maintained

Load I/O to address X

Store I/O to address X

Maintained

Load I/O to address X

Store I/O to address Y

Not maintained

Store I/O to address X

Load I/O to address X

Maintained

Store I/O to address X

Load I/O to address Y

Not maintained

2.11 Replay Traps
There are some situations in which a load or store instruction cannot be executed due to
a condition that occurs after that instruction issues from the IQ or FQ. The instruction is
aborted (along with all newer instructions) and restarted from the fetch stage of the
pipeline. This mechanism is called a replay trap.

2.11.1 Mbox Order Traps
Load and store instructions may be issued from the IQ in a different order than they
were fetched from the Icache, while the architecture dictates that Dstream memory
transactions to the same physical bytes must be completed in order. Usually, the Mbox
manages the memory reference stream by itself to achieve architecturally correct
behavior, but the two cases in which the Mbox uses replay traps to manage the memory
stream are load-load and store-load order traps.
2.11.1.1 Load-Load Order Trap

The Mbox ensures that load instructions that read the same physical byte(s) ultimately
issue in correct order by using the load-load order trap. The Mbox compares the
address of each load instruction, as it is issued, to the address of all load instructions in
the load queue. If the Mbox finds a newer load instruction in the load queue, it invokes
a load-load order trap on the newer instruction. This is a replay trap that aborts the target of the trap and all newer instructions from the machine and refetches instructions
starting at the target of the trap.
2.11.1.2 Store-Load Order Trap

The Mbox ensures that a load instruction ultimately issues after an older store instruction that writes some portion of its memory operand by using the store-load order trap.
The Mbox compares the address of each store instruction, as it is issued, to the address
of all load instructions in the load queue. If the Mbox finds a newer load instruction in
the load queue, it invokes a store-load order trap on the load instruction. This is a replay
trap. It functions like the load-load order trap.

21264/EV68A Hardware Reference Manual

Internal Architecture

2–31

I/O Write Buffer and the WMB Instruction

The Ibox contains extra hardware to reduce the frequency of the store-load trap. There
is a 1-bit by 1024-entry VPC-indexed table in the Ibox called the stWait table. When an
Icache instruction is fetched, the associated stWait table entry is fetched along with the
Icache instruction. The stWait table produces 1 bit for each instruction accessed from
the Icache. When a load instruction gets a store-load order replay trap, its associated bit
in the stWait table is set during the cycle that the load is refetched. Hence, the trapping
load instruction’s stWait bit will be set the next time it is fetched.
The IQ will not issue load instructions whose stWait bit is set while there are older unissued store instructions in the queue. A load instruction whose stWait bit is set can be
issued the cycle immediately after the last older store instruction is issued from the
queue. All the bits in the stWait table are unconditionally cleared every 16384 cycles, or
every 65536 cycles if I_CTL[ST_WAIT_64K] is set.

2.11.2 Other Mbox Replay Traps
The Mbox also uses replay traps to control the flow of the load queue and store queue,
and to ensure that there are never multiple outstanding misses to different physical
addresses that map to the same Dcache or Bcache line. Unlike the order traps, however,
these replay traps are invoked on the incoming instruction that triggered the condition.

2.12 I/O Write Buffer and the WMB Instruction
The I/O write buffer (IOWB) consists of four 64-byte entries with the associated
address and control logic used to buffer I/O write data between the store queue (SQ)
and the system port.

2.12.1 Memory Barrier (MB/WMB/TB Fill Flow)
The Cbox CSR SYSBUS_MB_ENABLE bit determines if MB instructions produce
external system port transactions. When the SYSBUS_MB_ENABLE bit equals 0, the
Cbox CSR MB_CNT[3:0] field contains the number of pending uncommitted transactions. The counter will increment for each of the following commands:

•

RdBlk, RdBlkMod, RdBlkI

•

RdBlkSpec (valid), RdBlkModSpec (valid), RdBlkSpecI (valid)

•

RdBlkVic, RdBlkModVic, RdBlkVicI

•

CleanToDirty, SharedToDirty, STChangeToDirty, InvalToDirty

•

FetchBlk, FetchBlkSpec (valid), Evict

•

RdByte, RdLw, RdQw, WrByte, WrLW, WrQW

The counter is decremented with the C (commit) bit in the Probe and SysDc commands
(see Section 4.7.7). Systems can assert the C bit in the SysDc fill response to the commands that originally incremented the counter, or attached to the last probe seen by that
command when it reached the system serialization point. If the number of uncommitted
transactions reaches 15 (saturating the counter), the Cbox will stall MAF and IOWB
processing until at least one of the pending transactions has been committed. Probe processing is not interrupted by the state of this counter.

2–32

Internal Architecture

21264/EV68A Hardware Reference Manual

I/O Write Buffer and the WMB Instruction

2.12.1.1 MB Instruction Processing

When an MB instruction is fetched in the predicted instruction execution path, it stalls
in the map stage of the pipeline. This also stalls all instructions after the MB, and control of instruction flow is based upon the value in Cbox CSR SYSBUS_MB_ENABLE
as follows:

•

If Cbox CSR SYSBUS_MB_ENABLE is clear, the Cbox waits until the IQ is
empty and then performs the following actions:
a. Sends all pending MAF and IOWB entries to the system port.
b. Monitors Cbox CSR MB_CNT[3:0], a 4-bit counter of outstanding committed
events. When the counter decrements from one to zero, the Cbox marks the
youngest probe queue entry.
c. Waits until the MAF contains no more Dstream references and the SQ, LQ, and
IOWB are empty.
When all of the above have occurred and a probe response has been sent to the system for the marked probe queue entry, instruction execution continues with the
instruction after the MB.

•

If Cbox CSR SYSBUS_MB_ENABLE is set, the Cbox waits until the IQ is empty
and then performs the following actions:
a. Sends all pending MAF and IOWB entries to the system port
b. Sends the MB command to the system port
c. Waits until the MB command is acknowledged, then marks the youngest entry
in the probe queue
d. Waits until the MAF contains no more Dstream references and the SQ, LQ, and
IOWB are empty
When all of the above have occurred and a probe response has been sent to the system for the marked probe queue entry, instruction execution continues with the
instruction after the MB.
Because the MB instruction is executed speculatively, MB processing can begin
and the original MB can be killed. In the internal acknowledge case, the MB may
have already been sent to the system interface, and the system is still expected to
respond to the MB.

2.12.1.2 WMB Instruction Processing

Write memory barrier (WMB) instructions are issued into the Mbox store-queue, where
they wait until they are retired and all prior store instructions become writable. The
Mbox then stalls the writable pointer and informs the Cbox. The Cbox closes the IOWB
merge register and responds in one of the following two ways:

•

If Cbox CSR SYSBUS_MB_ENABLE is clear, the Cbox performs the following
actions:
a. Stalls further MAF and IOWB processing.
b. Monitors Cbox CSR MB_CNT[3:0], a 4-bit counter of outstanding committed
events. When the counter decrements from one to zero, the Cbox marks the
youngest probe queue entry.

21264/EV68A Hardware Reference Manual

Internal Architecture

2–33

I/O Write Buffer and the WMB Instruction

c. When a probe response has been sent to the system for the marked probe queue
entry, the Cbox considers the WMB to be satisfied.

•

If Cbox CSR SYSBUS_MB_ENABLE is set, the Cbox performs the following
actions:
a. Stalls further MAF and IOWB processing.
b. Sends the MB command to the system port.
c. Waits until the MB command is acknowledged by the system with a SysDc
MBDone command, then sends acknowledge and marks the youngest entry in
the probe queue.
d. When a probe response has been sent to the system for the marked probe queue
entry, the Cbox considers the WMB to be satisfied.

2.12.1.3 TB Fill Flow

Load instructions (HW_LDs) to a virtual page table entry (VPTE) are processed by the
21264/EV68A to avoid litmus test problems associated with the ordering of memory
transactions from another processor against loading of a page table entry and the subsequent virtual-mode load from this processor.
Consider the sequence shown in Table 2–12. The data could be in the Bcache. Pj should
fetch datai if it is using PTEi.
Table 2–12 TB Fill Flow Example Sequence 1
Pi

Write Datai

Load/Store datai

Write PTEi

Load-PTE
<write TB>
Load/Store (restart)

Also consider the related sequence shown in Table 2–13. In this case, the data could be
cached in the Bcache; Pj should fetch datai if it is using PTEi.
Table 2–13 TB Fill Flow Example Sequence 2
Pi

Write Datai

Istream read datai

Write PTEi

Load-PTE
<write TB>
Istream read (restart) - will miss the Icache

The 21264/EV68A processes Dstream loads to the PTE by injecting, in hardware, some
memory barrier processing between the PTE transaction and any subsequent load or
store instruction. This is accomplished by the following mechanism:
1. The integer queue issues a HW_LD instruction with VPTE.

2–34

Internal Architecture

21264/EV68A Hardware Reference Manual

Performance Measurement Support—Performance Counters

2. The integer queue issues a HW_MTPR instruction with a DTB_PTE0, that is datadependent on the HW_LD instruction with a VPTE, and is required in order to fill
the DTBs. The HW_MTPR instruction, when queued, sets IPR scoreboard bits [4]
and [0].
3. When a HW_MTPR instruction with a DTB_PTE0 is issued, the Ibox signals the
Cbox indicating that a HW_LD instruction with a VPTE has been processed. This
causes the Cbox to begin processing the MB instruction. The Ibox prevents any
subsequent memory operations being issued by not clearing the IPR scoreboard bit
[0]. IPR scoreboard bit [0] is one of the scoreboard bits associated with the
HW_MTPR instruction with DTB_PTE0.
4. When the Cbox completes processing the MB instruction (using one of the above
sequences, depending upon the state of SYSBUS_MB_ENABLE), the Cbox signals the Ibox to clear IPR scoreboard bit [0].
The 21264/EV68A uses a similar mechanism to process Istream TB misses and fills to
the PTE for the Istream.
1. The integer queue issues a HW_LD instruction with VPTE.
2. The IQ issues a HW_MTPR instruction with an ITB_PTE that is data-dependent
upon the HW_LD instruction with VPTE. This is required in order to fill the ITB.
The HW_MTPR instruction, when queued, sets IPR scoreboard bits [4] and [0].
3. The Cbox issues a HW_MTPR instruction for the ITB_PTE and signals the Ibox
that a HW_LD/VPTE instruction has been processed, causing the Cbox to start processing the MB instruction. The Mbox stalls Ibox fetching from when the HW_LD/
VPTE instruction finishes until the probe queue is drained.
4. When the 21264/EV68A is finished (SYS_MB selects one of the above sequences),
the Cbox directs the Ibox to clear IPR scoreboard bit [0]. Also, the Mbox directs the
Ibox to start prefetching.
Inserting MB instruction processing within the TB fill flow is only required for multiprocessor systems. Uniprocessor systems can disable MB instruction processing by
deasserting Ibox CSR I_CTL[TB_MB_EN].

2.13 Performance Measurement Support—Performance Counters
The 21264/EV68A provides hardware support for two methods of obtaining program
performance feedback information. The two methods do not require program modification. The first method offers similar capabilities to earlier microprocessor performance
counters. The second method supports the new ProfileMe way of statistically sampling
individual instructions during program execution to develop a model of program execution. Both methods use the same hardware registers.
See Section 6.10 for information about counter control.

2.14 Floating-Point Control Register
The floating-point control register (FPCR) is shown in Figure 2–11.

21264/EV68A Hardware Reference Manual

Internal Architecture

2–35

Floating-Point Control Register

Figure 2–11 Floating-Point Control Register
63 62 61 60 59 58 57 56 55 54 53 52 51 50 49 48 47

SUM
INED
UNFD
UNDZ
DYN
IOV
INE
UNF
OVF
DZE
INV
OVFD
DZED
INVD
DNZ

LK99-0050A

The floating-point control register fields are described in Table 2–14.
Table 2–14 Floating-Point Control Register Fields
Name

Extent

Type

Description

SUM

[63]

Summary bit. Records bit-wise OR of FPCR exception bits.The summary bit is
not directly modified by writes to bit 63 of the FPCR, but is indirectly modified
by changes to FPCR bits 57–52.

INED

[62]

Inexact Disable. If this bit is set and a floating-point instruction that enables
trapping on inexact results generates an inexact value, the result is placed in the
destination register and the trap is suppressed.

UNFD

[61]

Underflow Disable. The 21264/EV68A hardware cannot generate IEEE compliant denormal results. UNFD is used in conjunction with UNDZ as follows:

UNDZ

2–36

[60]

Internal Architecture

UNFD

UNDZ

Result

Underflow trap.

Trap to supply a possible denormal result.

Underflow trap suppressed. Destination is written
with a true zero (+0.0).

Underflow to zero. When UNDZ is set together with UNFD, underflow traps
are disabled and the 21264/EV68A places a true zero in the destination register.
See UNFD, above.

21264/EV68A Hardware Reference Manual

AMASK and IMPLVER Instruction Values

Table 2–14 Floating-Point Control Register Fields (Continued)
Name

Extent

Type

Description

DYN

[59:58]

Dynamic rounding mode. Indicates the rounding mode to be used by an IEEE
floating-point instruction when the instruction specifies dynamic rounding
mode:
Bits

Meaning

Chopped

Minus infinity

Normal

Plus infinity

IOV

[57]

Integer overflow. A CVTGQ, CVTTQ, or CVTQL overflowed the destination
precision.

INE

[56]

Inexact result. A floating-point arithmetic or conversion operation gave a result
that differed from the mathematically exact result.

UNF

[55]

Underflow. A floating-point arithmetic or conversion operation gave a result
that underflowed the destination exponent.

OVF

[54]

Overflow. A floating-point arithmetic or conversion operation gave a result that
overflowed the destination exponent.

DZE

[53]

Divide by zero. An attempt was made to perform a floating-point divide with a
divisor of zero.

INV

[52]

Invalid operation. An attempt was made to perform a floating-point arithmetic
operation and one or more of its operand values were illegal.

OVFD

[51]

Overflow disable. If this bit is set and a floating-point arithmetic operation generates an overflow condition, then the appropriate IEEE nontrapping result is
placed in the destination register and the trap is suppressed.

DZED

[50]

Division by zero disable. If this bit is set and a floating-point divide by zero is
detected, the appropriate IEEE nontrapping result is placed in the destination
register and the trap is suppressed.

INVD

[49]

Invalid operation disable. If this bit is set and a floating-point operate generates
an invalid operation condition and 21264/EV68A is capable of producing the
correct IEEE nontrapping result, that result is placed in the destination register
and the trap is suppressed.

DNZ

[48]

Denormal operands to zero. If this bit is set, treat all Denormal operands as a
signed zero value with the same sign as the Denormal operand.

Reserved

[47:0]1

—

Alpha architecture FPCR bit 47 (DNOD) is not implemented by the 21264/EV68A.

2.15 AMASK and IMPLVER Instruction Values
The AMASK and IMPLVER instructions return the supported architecture extensions
and processor type , respectively.

21264/EV68A Hardware Reference Manual

Internal Architecture

2–37

Design Examples

2.15.1 AMASK
The 21264/EV68A returns the AMASK instruction values provided in Table 2–15. The
I_CTL register reports the 21264/EV68A pass level (see I_CTL[CHIP_ID], Section
5.2.15).
Table 2–15 21264/EV68A AMASK Values
21264/EV68A Pass Level

AMASK Feature Mask Value

See I_CTL[CHIP_ID], Table 5–11

130716

The AMASK bit definitions provided in Table 2–15 are defined in Table 2–16.
Table 2–16 AMASK Bit Assignments
Bit

Meaning

Support for the byte/word extension (BWX)
The instructions that comprise the BWX extension are LDBU, LDWU, SEXTB,
SEXTW, STB, and STW.

Support for the square-root and floating-point convert extension (FIX)
The instructions that comprise the FIX extension are FTOIS, FTOIT, ITOFF, ITOFS,
ITOFT, SQRTF, SQRTG, SQRTS, and SQRTT.

Support for the count extension (CIX)
The instructions that comprise the CIX extension are CTLZ, CTPOP, and CTTZ.

Support for the multimedia extension (MVI)
The instructions that comprise the MVI extension are MAXSB8, MAXSW4,
MAXUB8, MAXUW4, MINSB8, MINSW4, MINUB8, MINUW4, PERR, PKLB,
PKWB, UNPKBL, and UNPKBW.

Support for precise arithmetic trap reporting in hardware. The trap PC is the same as
the instruction PC after the trapping instruction is executed.

Support for using a prefetch with modify intent to improve the performance of the
first attempt to acquire a lock. When clear, indicates possible prefetch error with
locks, described in waiver 10 to the Alpha Architecture and in the prefetch section of
the appropriate processor (21264/EV6 and 21264/EV67) documents.

2.15.2 IMPLVER
For the 21264/EV68A, the IMPLVER instruction returns the value 2.

2.16 Design Examples
The 21264/EV68A can be designed into many different uniprocessor and multiprocessor system configurations. Figures 2–12 and 2–13 illustrate two possible configurations. These configurations employ additional system/memory controller chipsets.
Figure 2–12 shows a typical uniprocessor system with a second-level cache. This system configuration could be used in standalone or networked workstations.

2–38

Internal Architecture

21264/EV68A Hardware Reference Manual

Design Examples

Figure 2–12 Typical Uniprocessor Configuration
L2 Cache
Tag
Store

Data
Store

21272 Core
Logic Chipset

21264

Duplicate
Tag Store
(Optional)

Tag
Address
Out

Control
Chips

Address
Address
In

Data Slice
Chips

Data
Host PCI
Bridge Chip

Data

DRAM
Arrays

Address
Data

64-bit PCI Bus
FM-05573-EV67

Figure 2–13 shows a typical multiprocessor system, each processor with a second-level
cache. Each interface controller must employ a duplicate tag store to maintain cache
coherency. This system configuration could be used in a networked database server
application.
Figure 2–13 Typical Multiprocessor Configuration
21264

21272 Core
Logic Chipset

L2
Cache

DRAM
Arrays

Address
Control
Chip

21264

Data Slice
Chips

Data

DRAM
Arrays

L2
Cache
Address
Host PCI
Bridge Chip

Host PCI
Bridge Chip

Data

64-bit PCI Bus
64-bit PCI Bus
FM-05574-EV67

21264/EV68A Hardware Reference Manual

Internal Architecture

2–39

3
Hardware Interface
This chapter contains the 21264/EV68A microprocessor logic symbol and provides
information about signal names, their function, and their location. This chapter also
describes the mechanical specifications of the 21264/EV68A. It is organized as follows:

•

The 21264/EV68A logic symbol

•

The 21264/EV68A signal names and functions

•

Lists of the signal pins, sorted by name and PGA location

•

The specifications for the 21264/EV68A mechanical package

•

The top and bottom views of the 21264/EV68A pinouts

3.1 21264/EV68A Microprocessor Logic Symbol
Figure 3–1 show the logic symbol for the 21264/EV68A chip.

21264/EV68A Hardware Reference Manual

Hardware Interface

3–1

21264/EV68A Microprocessor Logic Symbol

Figure 3–1 21264/EV68A Microprocessor Logic Symbol
21264
Bcache Interface

System Interface
SysAddIn_L[14:0]

BcAdd_H[23:4]

SysAddInClk_L

BcData_H[127:0]

SysAddOut_L[14:0]

BcCheck_H[15:0]

SysAddOutClk_L

BcDataInClk_H[7:0]

SysVref

BcDataOutClk_[3:0]

SysData_L[63:0]

BcDataOE_L

SysCheck_L[7:0]

BcDataWr_L

SysDataInClk_H[7:0]

BcTag_H[42:20]

SysDataOutClk_L[7:0]

BcTagInClk_H

SysDataInValid_L

BcTagOutClk_x

SysDataOutValid_L

BcVref

SysFillValid_L

BcTagDirty_H
BcTagParity_H
BcTagShared_H
BcTagValid_H
BcTagOE_L
BcTagWr_L
BcLoad_L

ClkIn_x

Clocks

FrameClk_x
EV6Clk_x
2.5 V

PLL_VDD
Miscellaneous
IRQ_H[5:0]
ClkFwdRst_H
SromData_H
Tms_H
Trst_L
Tick_H
Tdi_H
PllBypass_H

SromClk_H

MiscVref

SromOE_L

Reset_L

TestStat_H

DCOK_H

Tdo_H
FM-05646b

3–2

Hardware Interface

21264/EV68A Hardware Reference Manual

21264/EV68A Signal Names and Functions

3.2 21264/EV68A Signal Names and Functions
Table 3–1 defines the 21264/EV68A signal types referred to in this section.
Table 3–1 Signal Pin Types Definitions
Signal Type

Definition

Inputs
I_DC_REF

Input DC reference pin

I_DA

Input differential amplifier receiver

I_DA_CLK

Input clock pin

Outputs
O_OD

Open drain output driver

O_OD_TP

Open drain driver for test pins

O_PP

Push/pull output driver

O_PP_CLK

Push/pull output clock driver

Bidirectional
B_DA_OD

Bidirectional differential amplifier receiver with open drain output

B_DA_PP

Bidirectional differential amplifier receiver with push/pull output

Other
Spare

Reserved to COMPAQ1

NoConnect

No connection — Do not connect to these pins for any revision of the
21264/EV68A. These pins must float.

All Spare connections are Reserved to COMPAQ to maintain compatibility between
passes of the chip. Designers should not use these pins.

Table 3–2 lists all signal pins in alphabetic order and provides a full functional description of the pins. Table 3–4 lists the signal pins and their corresponding pin grid array
(PGA) locations in alphabetic order for the signal type. Table 3–5 lists the pin grid array
locations in alphabetical order.
Table 3–2 21264/EV68A Signal Descriptions
Signal

Type

Count

Description

BcAdd_H[23:4]

O_PP

These signals provide the index to the Bcache.

BcCheck_H[15:0]

B_DA_PP

ECC check bits for BcData_H[127:0].

BcData_H[127:0]

B_DA_PP

128

Bcache data signals.

BcDataInClk_H[7:0]

I_DA

Bcache data input clocks. These clocks are used with high
speed SDRAMs, such as DDRs, that provide a clock-out with
data-output pins to optimize Bcache read bandwidths. The
21264/EV68A internally synchronizes the data to its logic with
clock forward receive circuits similar to the system interface.

BcDataOE_L

O_PP

Bcache data output enable. The 21264/EV68A asserts this signal during Bcache read operations.

21264/EV68A Hardware Reference Manual

Hardware Interface

3–3

21264/EV68A Signal Names and Functions

Table 3–2 21264/EV68A Signal Descriptions (Continued)
Signal

Count

Description

BcDataOutClk_H[3:0] O_PP
BcDataOutClk_L[3:0]

Bcache data output clocks. These free-running clocks are differential copies of the Bcache clock and are derived from the
21264/EV68A GCLK. Their period is a multiple of the GCLK
and is fixed for all operations. They can be configured so that
their rising edge lags BcAdd_H[23:4] by 0 to 2 GCLK cycles.
The 21264/EV68A synchronizes tag output information with
these clocks.

BcDataWr_L

O_PP

Bcache data write enable. The 21264/EV68A asserts this signal
when writing data to the Bcache data arrays.

BcLoad_L

O_PP

Bcache burst enable.

BcTag_H[42:20]

B_DA_PP

Bcache tag bits.

BcTagDirty_H

B_DA_PP

Tag dirty state bit. During cache write operations, the 21264/
EV68A will assert this signal if the Bcache data has been modified.

BcTagInClk_H

I_DA

Bcache tag input clock. The 21264/EV68A uses this input
clock to latch the tag information on Bcache read operations.
This clock is used with high-speed SDRAMs, such as DDRs,
that provide a clock-out with data-output pins to optimize
Bcache read bandwidths. The 21264/EV68A internally synchronizes the data to its logic with clock forward receive circuits similar to the system interface.

BcTagOE_L

O_PP

Bcache tag output enable. This signal is asserted by the 21264/
EV68A for Bcache read operations.

BcTagOutClk_H
BcTagOutClk_L

O_PP

Bcache tag output clock. These clocks “echo” the clock-forwarded BcDataOutClk_x[3:0] clocks.

BcTagParity_H

B_DA_PP

Tag parity state bit.

BcTagShared_H

B_DA_PP

Tag shared state bit. The 21264/EV68A will write a 1 on this
signal line if another agent has a copy of the cache line.

BcTagValid_H

B_DA_PP

Tag valid state bit. If set, this line indicates that the cache line
is valid.

BcTagWr_L

O_PP

Tag RAM write enable. The 21264/EV68A asserts this signal
when writing a tag to the Bcache tag arrays.

BcVref

I_DC_REF

Bcache tag reference voltage.

ClkFwdRst_H

I_DA

Systems assert this synchronous signal to wake up a powereddown 21264/EV68A. The ClkFwdRst_H signal is clocked
into a 21264/EV68A register by the captured FrameClk_x
signals. Systems must ensure that the timing of this signal
meets 21264/EV68A requirements (see Section 4.7.2).

ClkIn_H
ClkIn_L

I_DA_CLK 2

Differential input signals provided by the system.

DCOK_H

I_DA

dc voltage OK. Must be deasserted until dc voltage reaches
proper operating level. After that, DCOK_H is asserted.

EV6Clk_H
EV6Clk_L

O_PP_CLK 2

3–4

Type

Hardware Interface

Provides an external test point to measure phase alignment of
the PLL.

21264/EV68A Hardware Reference Manual

21264/EV68A Signal Names and Functions

Table 3–2 21264/EV68A Signal Descriptions (Continued)
Signal

Type

Count

FrameClk_H
FrameClk_L

I_DA_CLK 2

A skew-controlled differential 50% duty cycle copy of the system clock. It is used by the 21264/EV68A as a reference, or
framing, clock.

IRQ_H[5:0]

I_DA

These six interrupt signal lines may be asserted by the system.
The response of the 21264/EV68A is determined by the system
software.

MiscVref

I_DC_REF

Voltage reference for the miscellaneous pins
(see Table 3–3).

PllBypass_H

I_DA

When asserted, this signal will cause the two input clocks
(ClkIn_x) to be applied to the 21264/EV68A internal circuits,
instead of the 21264/EV68A global clock (GCLK).

PLL_VDD

2.5 V

2.5-V dedicated power supply for the 21264/EV68A PLL.

Reset_L

I_DA

System reset. This signal protects the 21264/EV68A from
damage during initial power-up. It must be asserted until
DCOK_H is asserted. After that, it is deasserted and the
21264/EV68A begins its reset sequence.

SromClk_H

O_OD_TP

Serial ROM clock. Supplies the clock that causes the SROM to
advance to the next bit. The cycle time for this clock is 256
times the cycle time of the GCLK (internal 21264/EV68A
clock).

SromData_H

I_DA

Serial ROM data. Input data line from the SROM.

SromOE_L

O_OD_TP

Serial ROM enable. Supplies the output enable to the SROM.

SysAddIn_L[14:0]

I_DA

Time-multiplexed command/address/ID/Ack from system to
the 21264/EV68A.

SysAddInClk_L

I_DA

Single-ended forwarded clock from system for
SysAddIn_L[14:0] and SysFillValid_L.

SysAddOut_L[14:0]

O_OD

Time-multiplexed command/address/ID/mask from the 21264/
EV68A to the system bus.

SysAddOutClk_L

O_OD

Single-ended forwarded clock output for
SysAddOut_L[14:0].

SysCheck_L[7:0]

B_DA_OD

Quadword ECC check bits for SysData_L[63:0].

SysData_L[63:0]

B_DA_OD

Data bus for memory and I/O data.

SysDataInClk_H[7:0]

I_DA

Single-ended system-generated clocks for clock forwarded
input system data.

SysDataInValid_L

I_DA

When asserted, marks a valid data cycle for data transfers to
the 21264/EV68A.

SysDataOutClk_L[7:0] O_OD

Single-ended 21264/EV68A-generated clocks for clock forwarded output system data.

SysDataOutValid_L

I_DA

When asserted, marks a valid data cycle for data transfers from
the 21264/EV68A.

SysFillValid_L

I_DA

When asserted, this bit indicates validation for the cache fill
delivered in the previous system SysDc command.

21264/EV68A Hardware Reference Manual

Description

Hardware Interface

3–5

21264/EV68A Signal Names and Functions

Table 3–2 21264/EV68A Signal Descriptions (Continued)
Signal

Type

Count

Description

SysVref

I_DC_REF

System interface reference voltage.

Tck_H

I_DA

IEEE 1149.1 test clock.

Tdi_H

I_DA

IEEE 1149.1 test data-in signal.

Tdo_H

O_OD_TP

IEEE 1149.1 test data-out signal.

TestStat_H

O_OD_TP

Test status pin. System reset drives the test status pin low.
The TestStat_H pin is forced high at the start of the Icache
BiST. If the Icache BiST passes, the pin is deasserted at the end
of the BiST operation; otherwise, it remains high.
The 21264/EV68A generates a timeout reset signal if an
instruction is not retired within one billion cycles.
The 21264/EV68A signals the timeout reset event by outputting a 256 GCLK cycle wide pulse on TestStat_H.

Tms_H

I_DA

IEEE 1149.1 test mode select signal.

Trst_L

I_DA

IEEE 1149.1 test access port (TAP) reset signal.

Table 3–3 lists signals by function and provides an abbreviated description.
Table 3–3 21264/EV68A Signal Descriptions by Function
Signal

Type

Count

Description

BcAdd_H[23:4]

O_PP

Bcache index.

BcCheck_H[15:0]

B_DA_PP

ECC check bits for BcData_H[127:0].

BcData_H[127:0]

B_DA_PP

128

Bcache data.

BcDataInClk_H[7:0]

I_DA

Bcache data input clocks.

BcDataOE_L

O_PP

Bcache data output enable.

BcDataOutClk_H[3:0] O_PP
BcDataOutClk_L[3:0]

Bcache data output clocks.

BcDataWr_L

O_PP

Bcache data write enable.

BcLoad_L

O_PP

Bcache burst enable.

BcTag_H[42:20]

B_DA_PP

Bcache tag bits.

BcTagDirty_H

B_DA_PP

Tag dirty state bit.

BcTagInClk_H

I_DA

Bcache tag input clock.

BcTagOE_L

O_PP

Bcache tag output enable.

BcTagOutClk_H
BcTagOutClk_L

O_PP

Bcache tag output clocks.

BcTagParity_H

B_DA_PP

Tag parity state bit.

BcTagShared_H

B_DA_PP

Tag shared state bit.

BcTagValid_H

B_DA_PP

Tag valid state bit.

BcTagWr_L

O_PP

Tag RAM write enable.

BcVref Domain

3–6

Hardware Interface

21264/EV68A Hardware Reference Manual

21264/EV68A Signal Names and Functions

Table 3–3 21264/EV68A Signal Descriptions by Function (Continued)
Signal

Type

Count

Description

BcVref

I_DC_REF

Tag data input reference voltage.

SysAddIn_L[14:0]

I_DA

Time-multiplexed SysAddIn, system-to-21264/EV68A.

SysAddInClk_L

I_DA

Single-ended forwarded clock from system for
SysAddIn_L[14:0] and SysFillValid_L.

SysAddOut_L[14:0]

O_OD

Time-multiplexed SysAddOut, 21264/EV68A-to-system.

SysAddOutClk_L

O_OD

Single-ended forwarded-clock.

SysCheck_L[7:0]

B_DA_OD

Quadword ECC check bits for SysData_L[63:0].

SysData_L[63:0]

B_DA_OD

Data bus for memory and I/O data.

SysDataInClk_H[7:0]

I_DA

Single-ended system-generated clocks for clock forwarded
input system data.

SysDataInValid_L

I_DA

When asserted, marks a valid data cycle for data transfers to
the 21264/EV68A.

SysDataOutClk_L[7:0] O_OD

Single-ended 21264/EV68A-generated clocks for clock forwarded output system data.

SysDataOutValid_L

I_DA

When asserted, marks a valid data cycle for data transfers
from the 21264/EV68A.

SysFillValid_L

I_DA

Validation for fill given in previous SysDC command.

SysVref

I_DC_REF

System interface reference voltage.

ClkIn_H
ClkIn_L

I_DA_CLK

Differential input signals provided by the system.

EV6Clk_H
EV6Clk_L

O_PP_CLK

Provides an external test point to measure phase alignment of
the PLL.

FrameClk_H
FrameClk_L

I_DA_CLK

A skew-controlled differential 50% duty cycle copy of the
system clock. It is used by the 21264/EV68A as a reference,
or framing, clock.

PLL_VDD

2.5 V

2.5-V dedicated power supply for the 21264/EV68A PLL.

ClkFwdRst_H

I_DA

Systems assert this synchronous signal to wake up a powereddown 21264/EV68A. The ClkFwdRst_H signal is clocked
into a 21264/EV68A register by the captured FrameClk_x
signals.

DCOK_H

I_DA

dc voltage OK. Must be deasserted until dc voltage reaches
proper operating level. After that, DCOK_H is asserted.

IRQ_H[5:0]

I_DA

These six interrupt signal lines may be asserted by the system.

MiscVref

I_DC_REF

Reference voltage for miscellaneous pins.

PllBypass_H

I_DA

When asserted, this signal will cause the input clocks
(ClkIn_x) to be applied to the 21264/EV68A internal circuits,
instead of the 21264/EV68A’s global clock (GCLK).

SysVref Domain

Clocks and PLL

MiscVref Domain

21264/EV68A Hardware Reference Manual

Hardware Interface

3–7

Pin Assignments

Table 3–3 21264/EV68A Signal Descriptions by Function (Continued)
Signal

Type

Count

Description

Reset_L

I_DA

SromClk_H

O_OD_TP

Serial ROM clock.

SromData_H

I_DA

Serial ROM data.

SromOE_L

O_OD_TP

Serial ROM enable.

Tck_H

I_DA

IEEE 1149.1 test clock.

Tdi_H

I_DA

IEEE 1149.1 test data-in signal.

Tdo_H

O_OD_TP

IEEE 1149.1 test data-out signal.

TestStat_H

O_OD_TP

Test status pin.

Tms_H

I_DA

IEEE 1149.1 test mode select signal.

Trst_L

I_DA

IEEE 1149.1 test access port (TAP) reset signal.

3.3

Pin Assignments
The 21264/EV68A package has 587 pins aligned in a pin grid array (PGA) design.
There are 380 functional signal pins, 1 dedicated 2.5-V pin for the PLL, 112 ground
VSS pins, and 94 VDD pins. Table 3–4 lists the signal pins and their corresponding pin
grid array (PGA) locations in alphabetical order for the signal type. Table 3–5 lists the
pin grid array locations in alphabetical order.

Table 3–4 Pin List Sorted by Signal Name
Signal Name

PGA Location Signal Name

PGA Location

BcAdd_H_10

B30

BcAdd_H_11

D30

BcAdd_H_12

C31

BcAdd_H_13

H28

BcAdd_H_14

G29

BcAdd_H_15

A33

BcAdd_H_16

E31

BcAdd_H_17

D32

BcAdd_H_18

B34

BcAdd_H_19

A35

BcAdd_H_20

B36

BcAdd_H_21

H30

BcAdd_H_22

C35

BcAdd_H_23

E33

BcAdd_H_4

B28

BcAdd_H_5

E27

BcAdd_H_6

A29

BcAdd_H_7

G27

BcAdd_H_8

C29

BcAdd_H_9

F28

BcCheck_H_0

BcCheck_H_1

AB4

BcCheck_H_10

AW1

BcCheck_H_11

BD10

BcCheck_H_12

E45

BcCheck_H_13

AC45

BcCheck_H_14

AT44

BcCheck_H_15

BB36

BcCheck_H_2

AT2

BcCheck_H_3

BC11

BcCheck_H_4

M38

BcCheck_H_5

AB42

BcCheck_H_6

AU43

BcCheck_H_7

BC37

BcCheck_H_8

BcCheck_H_9

AA3

BcData_H_0

B10

BcData_H_1

D10

BcData_H_10

BcData_H_100

D42

BcData_H_101

D44

BcData_H_102

H40

BcData_H_103

H42

BcData_H_104

G45

BcData_H_105

L43

3–8

Hardware Interface

21264/EV68A Hardware Reference Manual

Pin Assignments

Table 3–4 Pin List Sorted by Signal Name (Continued)
Signal Name

PGA Location Signal Name

PGA Location

BcData_H_106

L45

BcData_H_107

N45

BcData_H_108

T44

BcData_H_109

U45

BcData_H_11

BcData_H_110

W45

BcData_H_111

AA43

BcData_H_112

AC43

BcData_H_113

AD44

BcData_H_114

AE41

BcData_H_115

AG45

BcData_H_116

AK44

BcData_H_117

AL43

BcData_H_118

AM42

BcData_H_119

AR45

BcData_H_12

BcData_H_120

AP40

BcData_H_121

BA45

BcData_H_122

AV42

BcData_H_123

BB44

BcData_H_124

BB42

BcData_H_125

BC41

BcData_H_126

BA37

BcData_H_127

BD40

BcData_H_13

BcData_H_14

BcData_H_15

BcData_H_16

AC1

BcData_H_17

AD2

BcData_H_18

AE3

BcData_H_19

AG1

BcData_H_2

BcData_H_20

AK2

BcData_H_21

AL3

BcData_H_22

AR1

BcData_H_23

AP2

BcData_H_24

AY2

BcData_H_25

BB2

BcData_H_26

AW5

BcData_H_27

BB4

BcData_H_28

BB8

BcData_H_29

BE5

BcData_H_3

BcData_H_30

BB10

BcData_H_31

BE7

BcData_H_32

G33

BcData_H_33

C37

BcData_H_34

B40

BcData_H_35

C41

BcData_H_36

C43

BcData_H_37

E43

BcData_H_38

G41

BcData_H_39

F44

BcData_H_4

BcData_H_40

K44

BcData_H_41

N41

BcData_H_42

M44

BcData_H_43

P42

BcData_H_44

U43

BcData_H_45

V44

BcData_H_46

Y42

BcData_H_47

AB44

BcData_H_48

AD42

BcData_H_49

AE43

BcData_H_5

BcData_H_50

AF42

BcData_H_51

AJ45

BcData_H_52

AK42

BcData_H_53

AN45

BcData_H_54

AP44

BcData_H_55

AN41

BcData_H_56

AW45

BcData_H_57

AU41

BcData_H_58

AY44

BcData_H_59

BA43

BcData_H_6

BcData_H_60

BC43

BcData_H_61

BD42

BcData_H_62

BB38

BcData_H_63

BE41

BcData_H_64

C11

BcData_H_65

BcData_H_66

BcData_H_67

BcData_H_68

BcData_H_69

BcData_H_7

BcData_H_70

BcData_H_71

BcData_H_72

BcData_H_73

BcData_H_74

BcData_H_75

BcData_H_76

BcData_H_77

BcData_H_78

BcData_H_79

BcData_H_8

BcData_H_80

AB2

BcData_H_81

AC3

BcData_H_82

AD4

BcData_H_83

AF4

BcData_H_84

AJ3

BcData_H_85

AK4

BcData_H_86

AN1

BcData_H_87

AM4

BcData_H_88

AU5

BcData_H_89

BA1

21264/EV68A Hardware Reference Manual

Hardware Interface

3–9

Pin Assignments

Table 3–4 Pin List Sorted by Signal Name (Continued)
Signal Name

PGA Location Signal Name

PGA Location

BcData_H_9

BcData_H_90

BA3

BcData_H_91

BC3

BcData_H_92

BD6

BcData_H_93

BA9

BcData_H_94

BC9

BcData_H_95

AY12

BcData_H_96

A39

BcData_H_97

D36

BcData_H_98

A41

BcData_H_99

B42

BcDataInClk_H_0

BcDataInClk_H_1

BcDataInClk_H_2

AH2

BcDataInClk_H_3

BC5

BcDataInClk_H_4

F38

BcDataInClk_H_5

U39

BcDataInClk_H_6

AH44

BcDataInClk_H_7

AY40

BcDataOE_L

A27

BcDataOutClk_H_0

BcDataOutClk_H_1

AU3

BcDataOutClk_H_2

J43

BcDataOutClk_H_3

AR43

BcDataOutClk_L_0

BcDataOutClk_L_1

AV4

BcDataOutClk_L_2

K42

BcDataOutClk_L_3

AT42

BcDataWr_L

D26

BcLoad_L

F26

BcTag_H_20

E13

BcTag_H_21

H16

BcTag_H_22

A11

BcTag_H_23

B12

BcTag_H_24

D14

BcTag_H_25

E15

BcTag_H_26

A13

BcTag_H_27

G17

BcTag_H_28

C15

BcTag_H_29

H18

BcTag_H_30

D16

BcTag_H_31

B16

BcTag_H_32

C17

BcTag_H_33

A17

BcTag_H_34

E19

BcTag_H_35

B18

BcTag_H_36

A19

BcTag_H_37

F20

BcTag_H_38

D20

BcTag_H_39

E21

BcTag_H_40

C21

BcTag_H_41

D22

BcTag_H_42

H22

BcTagDirty_H

C23

BcTagInClk_H

G19

BcTagOE_L

H24

BcTagOutClk_H

C25

BcTagOutClk_L

D24

BcTagParity_H

B22

BcTagShared_H

G23

BcTagValid_H

B24

BcTagWr_L

E25

BcVref

F18

ClkFwdRst_H

BE11

ClkIn_H

AM8

ClkIn_L

AN7

DCOK_H

AY18

EV6Clk_H

AM6

EV6Clk_L

AL7

FrameClk_H

AV16

FrameClk_L

AW15

IRQ_H_0

BA15

IRQ_H_1

BE13

IRQ_H_2

AW17

IRQ_H_3

AV18

IRQ_H_4

BC15

IRQ_H_5

BB16

MiscVref

AV22

NoConnect

BB14

NoConnect

BD2

PLL_VDD

AV8

PllBypass_H

BD12

Reset_L

BD16

Spare

AJ1

Spare

V38

Spare

AT4

Spare

BE9

Spare

BD4

Spare

AJ43

Spare

AR3

Spare

E39

Spare

BA39

Spare

BC21

SromClk_H

AW19

SromData_H

BC17

SromOE_L

BE17

SysAddIn_L_0

BD30

SysAddIn_L_1

BC29

SysAddIn_L_10

BB24

SysAddIn_L_11

AV24

SysAddIn_L_12

BD24

SysAddIn_L_13

BE23

SysAddIn_L_14

AW23

SysAddIn_L_2

AY28

SysAddIn_L_3

BE29

SysAddIn_L_4

AW27

3–10

Hardware Interface

21264/EV68A Hardware Reference Manual

Pin Assignments

Table 3–4 Pin List Sorted by Signal Name (Continued)
Signal Name

PGA Location Signal Name

PGA Location

SysAddIn_L_5

BA27

SysAddIn_L_6

BD28

SysAddIn_L_7

BE27

SysAddIn_L_8

AY26

SysAddIn_L_9

BC25

SysAddInClk_L

BB26

SysAddOut_L_0

AW33

SysAddOut_L_1

BE39

SysAddOut_L_10

BE33

SysAddOut_L_11

AW29

SysAddOut_L_12

BC31

SysAddOut_L_13

AV28

SysAddOut_L_14

BB30

SysAddOut_L_2

BD36

SysAddOut_L_3

BC35

SysAddOut_L_4

BA33

SysAddOut_L_5

AY32

SysAddOut_L_6

BE35

SysAddOut_L_7

AV30

SysAddOut_L_8

BB32

SysAddOut_L_9

BA31

SysAddOutClk_L

BD34

SysCheck_L_0

SysCheck_L_1

AA5

SysCheck_L_2

AK8

SysCheck_L_3

BA13

SysCheck_L_4

L39

SysCheck_L_5

AA41

SysCheck_L_6

AM40

SysCheck_L_7

AY34

SysData_L_0

F14

SysData_L_1

G13

SysData_L_10

SysData_L_11

SysData_L_12

SysData_L_13

SysData_L_14

SysData_L_15

SysData_L_16

AB8

SysData_L_17

AC7

SysData_L_18

AD8

SysData_L_19

AE5

SysData_L_2

F12

SysData_L_20

AH6

SysData_L_21

AH8

SysData_L_22

AJ7

SysData_L_23

AL5

SysData_L_24

AP8

SysData_L_25

AR7

SysData_L_26

AT8

SysData_L_27

AV6

SysData_L_28

AV10

SysData_L_29

AW11

SysData_L_3

H12

SysData_L_30

AV12

SysData_L_31

AW13

SysData_L_32

F32

SysData_L_33

F34

SysData_L_34

H34

SysData_L_35

G35

SysData_L_36

F40

SysData_L_37

G39

SysData_L_38

K38

SysData_L_39

J41

SysData_L_4

H10

SysData_L_40

M40

SysData_L_41

N39

SysData_L_42

P40

SysData_L_43

T38

SysData_L_44

V40

SysData_L_45

W41

SysData_L_46

W39

SysData_L_47

Y40

SysData_L_48

AB38

SysData_L_49

AC39

SysData_L_5

SysData_L_50

AD38

SysData_L_51

AF40

SysData_L_52

AH38

SysData_L_53

AJ39

SysData_L_54

AL41

SysData_L_55

AK38

SysData_L_56

AN39

SysData_L_57

AP38

SysData_L_58

AR39

SysData_L_59

AT38

SysData_L_6

SysData_L_60

AY38

SysData_L_61

AV36

SysData_L_62

AW35

SysData_L_63

AV34

SysData_L_7

SysData_L_8

SysData_L_9

SysDataInClk_H_0

SysDataInClk_H_1

SysDataInClk_H_2

AF6

SysDataInClk_H_3

AY6

SysDataInClk_H_4

E37

SysDataInClk_H_5

R43

SysDataInClk_H_6

AG41

SysDataInClk_H_7

AV40

SysDataInValid_L

BD22

SysDataOutClk_L_0 G11

SysDataOutClk_L_1 U7

SysDataOutClk_L_2 AG7

SysDataOutClk_L_3 AY8

SysDataOutClk_L_4 H36

21264/EV68A Hardware Reference Manual

Hardware Interface

3–11

Pin Assignments

Table 3–4 Pin List Sorted by Signal Name (Continued)
Signal Name

PGA Location Signal Name

PGA Location

SysDataOutClk_L_5 R41

SysDataOutClk_L_6 AH40

SysDataOutClk_L_7 AW39

SysDataOutValid_L

BB22

SysFillValid_L

BC23

SysVref

BA25

Tck_H

BE19

Tdi_H

BA21

Tdo_H

BB20

TestStat_H

BA19

Tms_H

BD18

Trst_L

AY20

Table 3–5 Pin List Sorted by PGA Location
PGA Location Signal Name

PGA Location Signal Name

A11

BcTag_H_22

A13

BcTag_H_26

A17

BcTag_H_33

A19

BcTag_H_36

A27

BcDataOE_L

A29

BcAdd_H_6

A33

BcAdd_H_15

A35

BcAdd_H_19

A39

BcData_H_96

A41

BcData_H_98

BcData_H_2

BcData_H_65

AA3

BcCheck_H_9

AA41

SysCheck_L_5

AA43

BcData_H_111

AA5

SysCheck_L_1

AB2

BcData_H_80

AB38

SysData_L_48

AB4

BcCheck_H_1

AB42

BcCheck_H_5

AB44

BcData_H_47

AB8

SysData_L_16

AC1

BcData_H_16

AC3

BcData_H_81

AC39

SysData_L_49

AC43

BcData_H_112

AC45

BcCheck_H_13

AC7

SysData_L_17

AD2

BcData_H_17

AD38

SysData_L_50

AD4

BcData_H_82

AD42

BcData_H_48

AD44

BcData_H_113

AD8

SysData_L_18

AE3

BcData_H_18

AE41

BcData_H_114

AE43

BcData_H_49

AE5

SysData_L_19

AF4

BcData_H_83

AF40

SysData_L_51

AF42

BcData_H_50

AF6

SysDataInClk_H_2

AG1

BcData_H_19

AG41

SysDataInClk_H_6

AG45

BcData_H_115

AG7

SysDataOutClk_L_2 AH2

BcDataInClk_H_2

AH38

SysData_L_52

AH40

SysDataOutClk_L_6 AH44

BcDataInClk_H_6

AH6

SysData_L_20

AH8

SysData_L_21

AJ1

Spare

AJ3

BcData_H_84

AJ39

SysData_L_53

AJ43

Spare

AJ45

BcData_H_51

AJ7

SysData_L_22

AK2

BcData_H_20

AK38

SysData_L_55

AK4

BcData_H_85

AK42

BcData_H_52

AK44

BcData_H_116

AK8

SysCheck_L_2

AL3

BcData_H_21

AL41

SysData_L_54

AL43

BcData_H_117

AL5

SysData_L_23

AL7

EV6Clk_L

AM4

BcData_H_87

AM40

SysCheck_L_6

AM42

BcData_H_118

AM6

EV6Clk_H

AM8

ClkIn_H

AN1

BcData_H_86

AN39

SysData_L_56

AN41

BcData_H_55

AN45

BcData_H_53

AN7

ClkIn_L

AP2

BcData_H_23

AP38

SysData_L_57

AP40

BcData_H_120

AP44

BcData_H_54

AP8

SysData_L_24

3–12

Hardware Interface

21264/EV68A Hardware Reference Manual

Pin Assignments

Table 3–5 Pin List Sorted by PGA Location (Continued)
PGA Location Signal Name

PGA Location Signal Name

AR1

BcData_H_22

AR3

Spare

AR39

SysData_L_58

AR43

BcDataOutClk_H_3

AR45

BcData_H_119

AR7

SysData_L_25

AT2

BcCheck_H_2

AT38

SysData_L_59

AT4

Spare

AT42

BcDataOutClk_L_3

AT44

BcCheck_H_14

AT8

SysData_L_26

AU3

BcDataOutClk_H_1

AU41

BcData_H_57

AU43

BcCheck_H_6

AU5

BcData_H_88

AV10

SysData_L_28

AV12

SysData_L_30

AV16

FrameClk_H

AV18

IRQ_H_3

AV22

MiscVref

AV24

SysAddIn_L_11

AV28

SysAddOut_L_13

AV30

SysAddOut_L_7

AV34

SysData_L_63

AV36

SysData_L_61

AV4

BcDataOutClk_L_1

AV40

SysDataInClk_H_7

AV42

BcData_H_122

AV6

SysData_L_27

AV8

PLL_VDD

AW1

BcCheck_H_10

AW11

SysData_L_29

AW13

SysData_L_31

AW15

FrameClk_L

AW17

IRQ_H_2

AW19

SromClk_H

AW23

SysAddIn_L_14

AW27

SysAddIn_L_4

AW29

SysAddOut_L_11

AW33

SysAddOut_L_0

AW35

SysData_L_62

AW39

SysDataOutClk_L_7 AW45

BcData_H_56

AW5

BcData_H_26

AY12

BcData_H_95

AY18

DCOK_H

AY2

BcData_H_24

AY20

Trst_L

AY26

SysAddIn_L_8

AY28

SysAddIn_L_2

AY32

SysAddOut_L_5

AY34

SysCheck_L_7

AY38

SysData_L_60

AY40

BcDataInClk_H_7

AY44

BcData_H_58

AY6

SysDataInClk_H_3

AY8

SysDataOutClk_L_3 B10

BcData_H_0

B12

BcTag_H_23

B16

BcTag_H_31

B18

BcTag_H_35

B22

BcTagParity_H

B24

BcTagValid_H

B28

BcAdd_H_4

B30

BcAdd_H_10

B34

BcAdd_H_18

B36

BcAdd_H_20

BcData_H_68

B40

BcData_H_34

B42

BcData_H_99

BcData_H_67

BA1

BcData_H_89

BA13

SysCheck_L_3

BA15

IRQ_H_0

BA19

TestStat_H

BA21

Tdi_H

BA25

SysVref

BA27

SysAddIn_L_5

BA3

BcData_H_90

BA31

SysAddOut_L_9

BA33

SysAddOut_L_4

BA37

BcData_H_126

BA39

Spare

BA43

BcData_H_59

BA45

BcData_H_121

BA9

BcData_H_93

BB10

BcData_H_30

BB14

NoConnect

BB16

IRQ_H_5

BB2

BcData_H_25

BB20

Tdo_H

BB22

SysDataOutValid_L

BB24

SysAddIn_L_10

BB26

SysAddInClk_L

BB30

SysAddOut_L_14

BB32

SysAddOut_L_8

BB36

BcCheck_H_15

BB38

BcData_H_62

BB4

BcData_H_27

BB42

BcData_H_124

BB44

BcData_H_123

BB8

BcData_H_28

BC11

BcCheck_H_3

BC15

IRQ_H_4

BC17

SromData_H

BC21

Spare

BC23

SysFillValid_L

21264/EV68A Hardware Reference Manual

Hardware Interface

3–13

Pin Assignments

Table 3–5 Pin List Sorted by PGA Location (Continued)
PGA Location Signal Name

PGA Location Signal Name

BC25

SysAddIn_L_9

BC29

SysAddIn_L_1

BC3

BcData_H_91

BC31

SysAddOut_L_12

BC35

SysAddOut_L_3

BC37

BcCheck_H_7

BC41

BcData_H_125

BC43

BcData_H_60

BC5

BcDataInClk_H_3

BC9

BcData_H_94

BD10

BcCheck_H_11

BD12

PllBypass_H

BD16

Reset_L

BD18

Tms_H

BD2

NoConnect

BD22

SysDataInValid_L

BD24

SysAddIn_L_12

BD28

SysAddIn_L_6

BD30

SysAddIn_L_0

BD34

SysAddOutClk_L

BD36

SysAddOut_L_2

BD4

Spare

BD40

BcData_H_127

BD42

BcData_H_61

BD6

BcData_H_92

BE11

ClkFwdRst_H

BE13

IRQ_H_1

BE17

SromOE_L

BE19

Tck_H

BE23

SysAddIn_L_13

BE27

SysAddIn_L_7

BE29

SysAddIn_L_3

BE33

SysAddOut_L_10

BE35

SysAddOut_L_6

BE39

SysAddOut_L_1

BE41

BcData_H_63

BE5

BcData_H_29

BE7

BcData_H_31

BE9

Spare

C11

BcData_H_64

C15

BcTag_H_28

C17

BcTag_H_32

C21

BcTag_H_40

C23

BcTagDirty_H

C25

BcTagOutClk_H

C29

BcAdd_H_8

BcData_H_4

C31

BcAdd_H_12

C35

BcAdd_H_22

C37

BcData_H_33

C41

BcData_H_35

C43

BcData_H_36

BcData_H_3

BcData_H_66

D10

BcData_H_1

D14

BcTag_H_24

D16

BcTag_H_30

BcData_H_71

D20

BcTag_H_38

D22

BcTag_H_41

D24

BcTagOutClk_L

D26

BcDataWr_L

D30

BcAdd_H_11

D32

BcAdd_H_17

D36

BcData_H_97

BcData_H_69

D42

BcData_H_100

D44

BcData_H_101

SysDataInClk_H_0

BcData_H_7

E13

BcTag_H_20

E15

BcTag_H_25

E19

BcTag_H_34

E21

BcTag_H_39

E25

BcTagWr_L

E27

BcAdd_H_5

BcData_H_5

E31

BcAdd_H_16

E33

BcAdd_H_23

E37

SysDataInClk_H_4

E39

Spare

E43

BcData_H_37

E45

BcCheck_H_12

BcDataInClk_H_0

F12

SysData_L_2

F14

SysData_L_0

F18

BcVref

BcCheck_H_0

F20

BcTag_H_37

F26

BcLoad_L

F28

BcAdd_H_9

F32

SysData_L_32

F34

SysData_L_33

F38

BcDataInClk_H_4

F40

SysData_L_36

F44

BcData_H_39

SysData_L_6

Spare

BcData_H_73

G11

SysDataOutClk_L_0 G13

SysData_L_1

G17

BcTag_H_27

G19

BcTagInClk_H

G23

BcTagShared_H

G27

BcAdd_H_7

G29

BcAdd_H_14

G33

BcData_H_32

G35

SysData_L_35

3–14

Hardware Interface

21264/EV68A Hardware Reference Manual

Pin Assignments

Table 3–5 Pin List Sorted by PGA Location (Continued)
PGA Location Signal Name

PGA Location Signal Name

G39

SysData_L_37

G41

BcData_H_38

G45

BcData_H_104

BcData_H_70

SysData_L_5

H10

SysData_L_4

H12

SysData_L_3

H16

BcTag_H_21

H18

BcTag_H_29

H22

BcTag_H_42

H24

BcTagOE_L

H28

BcAdd_H_13

H30

BcAdd_H_21

H34

SysData_L_34

H36

SysDataOutClk_L_4

BcData_H_72

H40

BcData_H_102

H42

BcData_H_103

BcData_H_6

BcData_H_8

J41

SysData_L_39

J43

BcDataOutClk_H_2

BcDataOutClk_H_0

BcData_H_9

K38

SysData_L_38

BcDataOutClk_L_0

K42

BcDataOutClk_L_2

K44

BcData_H_40

SysData_L_7

BcData_H_75

BcData_H_10

L39

SysCheck_L_4

L43

BcData_H_105

L45

BcData_H_106

SysCheck_L_0

BcData_H_11

M38

BcCheck_H_4

M40

SysData_L_40

M44

BcData_H_42

SysData_L_8

BcCheck_H_8

BcData_H_76

N39

SysData_L_41

N41

BcData_H_41

N45

BcData_H_107

BcData_H_74

SysData_L_9

SysDataInClk_H_1

P40

SysData_L_42

P42

BcData_H_43

SysData_L_10

BcDataInClk_H_1

R41

SysDataOutClk_L_5 R43

SysDataInClk_H_5

BcData_H_12

T38

SysData_L_43

Spare

T44

BcData_H_108

SysData_L_11

BcData_H_13

BcData_H_77

U39

BcDataInClk_H_5

U43

BcData_H_44

U45

BcData_H_109

SysDataOutClk_L_1 V2

BcData_H_14

V38

Spare

V40

SysData_L_44

V44

BcData_H_45

SysData_L_13

SysData_L_12

BcData_H_79

W39

SysData_L_46

W41

SysData_L_45

W45

BcData_H_110

BcData_H_78

SysData_L_14

BcData_H_15

Y40

SysData_L_47

Y42

BcData_H_46

SysData_L_15

21264/EV68A Hardware Reference Manual

Hardware Interface

3–15

Pin Assignments

Table 3–6 lists the 21264/EV68A ground and power (VSS and VDD, respectively) pin
list.
Table 3–6 Ground and Power (VSS and VDD) Pin List
Signal
VSS

VDD

3–16

PGA Location
A15

A21

A25

A31

A37

A43

AA1

AA39

AA45

AA7

AC41

AC5

AE1

AE39

AE45

AE7

AG3

AG39

AG43

AG5

AJ41

AJ5

AL1

AL39

AL45

AN3

AN43

AN5

AR41

AR5

AU1

AU39

AU45

AU7

AW21

AW25

AW3

AW31

AW37

AW41

AW43

AW7

AW9

AY14

BA11

BA17

BA23

BA29

BA35

BA41

BA5

BA7

BC1

BC13

BC19

BC27

BC33

BC39

BC45

BC7

BE15

BE21

BE25

BE3

BE31

BE37

BE43

C13

C19

C27

C33

C39

C45

DS8

E11

E17

E23

E29

E35

E41

G15

G21

G25

G31

G37

G43

J39

J45

L41

N43

R39

R45

T42

U41

W43

—

A23

AB40

AB6

AD40

AD6

AF2

AF38

AF44

AF8

AH4

AH42

AK40

AK6

AM2

AM38

AM44

AP4

AP42

AP6

AT40

AT6

AV14

AV2

AV20

AV26

AV32

AV38

AV44

AY10

AY16

AY22

AY24

AY30

AY36

AY4

AY42

B14

B20

B26

B32

B38

B44

BB12

BB18

BB28

BB34

BB40

BB6

BD14

BD20

BD26

BD32

BD38

BD44

BD8

D12

D18

D28

D34

D40

F10

F16

F22

F24

F30

F36

F42

H14

H20

H26

H32

H38

H44

K40

M42

P38

P44

T40

V42

Y38

Y44

—

Hardware Interface

21264/EV68A Hardware Reference Manual

Mechanical Specifications

3.4

Mechanical Specifications
This section shows the 21264/EV68A mechanical package dimensions without a heat
sink. For heat sink information and dimensions, refer to Chapter 10.
Figure 3–2 shows the package physical dimensions without a heat sink.

Figure 3–2 Package Dimensions
1.27 mm (.050 in) Typ
4.32 mm (.170 in) Typ
2.54 mm (.100 in) Typ
BE
BC
BA
AW
AU
AR
AN
AL
AJ
AG
AE
AC
AA
W
U
R
N
L
J
G
E
C
A

1.377 mm (.055 in) Typ

Standoff (4x)
587x 1.40 mm (.055 in) Typ

BD
BB
AY
AV

1.27 mm (.050 in) Typ

AT
AP
AM
AK
AH

Lid

1/4-20 Stud (2x)

AD
AB
Y
V
T

27.94 mm
(1.100 in)

P
M
K

7.62 mm (.300 in) Typ

H
F
D
B

.13 mm
(.005 in) R

02 04 06 08 10 12 14 16 18 20 22 24 26 28 30 32 34 36 38 40 42 44
01 03 05 07 09 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45

27.94 mm
(1.100 in)

1.905 mm (.075 in) Typ

59.94 mm (2.360 in) Typ
29.62 mm
(1.180 in) Typ

29.62 mm
(1.180 in) Typ

25.40 mm
(1.000 in) Typ
53.85 mm
(2.120 in) Typ

21264/EV68A Hardware Reference Manual

FM-05662.AI4

Hardware Interface

3–17

21264/EV68A Packaging

3.5

21264/EV68A Packaging
Figure 3–3 shows the 21264/EV68A pinout from the top view with pins facing down.

Figure 3–3 21264/EV68A Top View (Pin Down)

BE
BC
BA
AW
AU
AR
AN
AL
AJ
AG
AE
AC
AA
W
U
R
N
L
J
G
E
C
A

BD
BB
AY
AV
AT
AP
AM
AK
AH
AF
AD
AB
Y
V
T
P
M
K
H
F
D
B

21264/EV68A
Top View
(PinDown)

44 42 40 38 36 34 32 30 28 26 24 22 20 18 16 14 12 10 08 06 04 02
45 43 41 39 37 35 33 31 29 27 25 23 21 19 17 15 13 11 09 07 05 03 01
FM-05644

3–18

Hardware Interface

21264/EV68A Hardware Reference Manual

21264/EV68A Packaging

Figure 3–4 shows the 21264/EV68A pinout from the bottom view with pins facing up.
Figure 3–4 21264/EV68A Bottom View (Pin Up)

BE
BC
BA
AW
AU
AR
AN
AL
AJ
AG
AE
AC
AA
W
U
R
N
L
J
G
E
C
A

BD
BB
AY
AV
AT
AP
AM
AK
AH
AF
AD
AB
Y
V
T
P
M
K
H
F
D
B

21264/EV68A
Bottom View
(PinUp)

02 04 06 08 10 12 14 16 18 20 22 24 26 28 30 32 34 36 38 40 42 44
01 03 05 07 09 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45
FM-05645

21264/EV68A Hardware Reference Manual

Hardware Interface

3–19

4
Cache and External Interfaces
This chapter describes the 21264/EV68A cache and external interface, which includes
the second-level cache (Bcache) interface and the system interface. It also describes
locks, interrupt signals, and ECC/parity generation. It is organized as follows:

•

Introduction to the external interfaces

•

Physical address considerations

•

Bcache structure

•

Victim data buffer

•

Cache coherency

•

Lock mechanism

•

System port

•

Bcache port

•

Interrupts

Chapter 3 lists and defines all 21264/EV68A hardware interface signal pins. Chapter 9
describes the 21264/EV68A hardware interface electrical requirements.

4.1 Introduction to the External Interfaces
A 21264/EV68A-based system can be divided into three major sections:

•

21264/EV68A microprocessor

•

Second-level Bcache

•

System interface logic
–
–
–

Optional duplicate tag store
Optional lock register
Optional victim buffers

The 21264/EV68A external interface is flexible and mandates few design rules, allowing a wide range of prospective systems. The external interface is composed of the
Bcache interface and the system interface.

•

Input clocks must have the same frequency as their corresponding output clock. For
example, the frequency of SysAddInClk_L must be the same as
SysAddOutClk_L.

21264/EV68A Hardware Reference Manual

Cache and External Interfaces

4–1

Introduction to the External Interfaces

•

The Bcache interface includes a 128-bit bidirectional data bus, a 20-bit unidirectional address bus, and several control signals.
–

The BcDataOutClk_x[3:0] clocks are free-running and are derived from the
internal GCLK. The period of BcDataOutClk_x[3:0] is a programmable multiple of GCLK.

–

The Bcache turns the BcDataOutClk_x[3:0] clocks around and returns them
to the 21264/EV68A as BcDataInClk_H[7:0]. Likewise, BcTagOutClk_x
returns as BcTagInClk_H.

–

The Bcache interface supports a 64-byte block size.

The system interface includes a 64-bit bidirectional data bus, two 15-bit
unidirectional address buses, and several control signals.
–

The SysAddOutClk_L clock is free-running and is derived from the internal
GCLK. The period of SysAddOutClk_L is a programmable multiple of
GCLK.

–

The SysAddInClk_L clock is a turned-around copy of SysAddOutClk_L.

Figure 4–1 shows a simplified view of the external interface. The function and purpose
of each signal is described in Chapter 3.

4–2

Cache and External Interfaces

21264/EV68A Hardware Reference Manual

Introduction to the External Interfaces

Figure 4–1 21264/EV68A System and Bcache Interfaces
SysAddIn_L[14:0]
SysAddInClk_L
SysAddOut_L[14:0]
SysAddOutClk_L
SysVref
SysData_L[63:0]
SysCheck_L[7:0]
SysDataInClk_H[7:0]
SysDataOutClk_L[7:0]
SysDataInValid_L
SysDataOutValid_L
SysFillValid_L
BcAdd_H[23:4]
21264

[23:4]
Data

[23:6]

[23:6]
Tag

Status

System

BcLoad_L
BcData_H[127:0]
BcCheck_H[15:0]
BcDataInClk_H[7:0]
BcDataOutClk_ x [3:0]
BcDataOE_L
BcDataWr_L
BcTag_H[42:20]
BcTagInClk_H
BcTagOutClk_ x
BcVref
BcTagWr_L
BcTagOE_L
BcTagValid_H
BcTagDir ty_H
BcTagShared_H
BcTagParity_H
IRQ_H[5:0]
FM-05818B-EV67

4.1.1 System Interface
This section introduces the system (external) bus interface. The system interface is
made up of two unidirectional 15-bit address buses, 64 bidirectional data lines, eight
bidirectional check bits, two single-ended unidirectional clocks, and a few control pins.
The 15-bit address buses provide time-shared address/command/ID in two or four
GCLK cycles. The Cbox controls the system interface.

21264/EV68A Hardware Reference Manual

Cache and External Interfaces

4–3

Physical Address Considerations

4.1.1.1 Commands and Addresses

The system sends probe and data movement commands to the 21264/EV68A. The
21264/EV68A can hold up to eight probe commands from the system. The system controls the number of outstanding probe commands and must ensure that the 21264/
EV68A 8-entry probe queue does not overflow.
The Cbox contains an 8-entry miss buffer (MAF) and an 8-entry victim buffer (VAF).
A miss occurs when the 21264/EV68A probes the Bcache but does not find the
addressed block. The 21264/EV68A can queue eight cache misses to the system in its
MAF.

4.1.2 Second-Level Cache (Bcache) Interface
The 21264/EV68A Cbox provides control signals and an interface for a second-level
cache, the Bcache. The 21264/EV68A supports a Bcache from 1MB to 16MB, with 64byte blocks. A 128-bit data bus is used for transfers between the 21264/EV68A and the
Bcache. The Bcache must be comprised of synchronous static RAMs (SSRAMs) and
must contain either one, two, or three internal registers. All Bcache control and address
pins are clocked synchronously on Bcache cycle boundaries. The Bcache clock rate
varies as a multiple of the CPU clock cycle in half-cycle increments from 1.5 to 4.0,
and in full-cycle increments of 5, 6, 7, and 8 times the CPU clock cycle. The 1.5 multiple is only available in dual-data mode.

4.2 Physical Address Considerations
The 21264/EV68A supports a 44-bit physical address space that is divided equally
between memory space and I/O space. Memory space resides in the lower half of the
physical address space (PA[43] = 0) and I/O space resides in the upper half of the physical address space (PA[43] = 1). The 21264/EV68A recognizes these spaces internally.
The 21264/EV68A-generated external references to memory space are always of a
fixed 64-byte size, though the internal access granularity is byte, word, longword, or
quadword. All 21264/EV68A-generated external references to memory or I/O space
are physical addresses that are either successfully translated from a virtual address or
produced by PALcode. Speculative execution may cause a reference to nonexistent
memory. Systems must check the range of all addresses and report nonexistent
addresses to the 21264/EV68A.
Table 4–1 describes the translation of internal references to external interface references. The first column lists the instructions used by the programmer, including load
(LDx) and store (STx) instructions of several sizes. The column headings are described
here:

4–4

•

DcHit (block was found in the Dcache)

•

DcW (block was found in a writable state in the Dcache)

•

BcHit (block was found in the Bcache)

•

BcW (block was found in a writable state in the Bcache)

•

Status and Action (status at end of instruction and action performed by the 21264/
EV68A)

Cache and External Interfaces

21264/EV68A Hardware Reference Manual

Physical Address Considerations

Prefetches (LDL, LDF, LDG, LDT, LDBU, LDWU) to R31 use the LDx flow, and
prefetch with modify intent (LDS) uses the STx flow. If the prefetch target is addressed
to I/O space, the upper address bit is cleared, converting the address to memory space
(PA[42:6] ). Notes follow the table.
Table 4–1 Translation of Internal References to External Interface Reference
Instruction

DcHit

DcW

BcHit

BcW

Status and Action

LDx Memory

Dcache hit, done.

LDx Memory

Bcache hit, done.

LDx Memory

Miss, generate RdBlk command.

LDx I/O

RdBytes, RdLWs, or RdQWs based on size.

Istream Memory

Dcache hit, Istream serviced from Dcache.

Istream Memory

Bcache hit, Istream serviced from Bcache.

Istream Memory

Miss, generate RdBlkI command.

STx Memory

Store Dcache hit and writable, done.

STx Memory

Store hit and not writable, set dirty flow (note 1).

STx Memory

Store Bcache hit and writable, done.

STx Memory

Store hit and not writable, set-dirty flow (note 1).

STx Memory

Miss, generate RdBlkMod command.

STx I/O

WrBytes, WrLWs, or WrQWs based on size.

STx_C Memory

Fail STx_C.

STx_C Memory

STx_C hit and not writable, set dirty flow (note 1).

STx_C I/O

Always succeed and WrQws or WrLws are generated,
based on the size.

WH64 Memory

Hit, done.

WH64 Memory

WH64 hit not writable, set dirty flow (note 1).

WH64 Memory

WH64 hit dirty, done.

WH64 Memory

WH64 hit not writable, set dirty flow (note 1).

WH64 Memory

Miss, generate InvalToDirty command (note 2).

WH64 I/O

NOP the instruction. WH64 is UNDEFINED for I/O
space.

ECB Memory

Generate evict command (note 3).

ECB I/O

NOP the instruction. ECB instruction is UNDEFINED
for I/O space.

MB/WMB
TBFill Flows

Generate MB command (note 4). Also see Section 3.2.5.

21264/EV68A Hardware Reference Manual

Cache and External Interfaces

4–5

Physical Address Considerations

Table 4–1 notes:
1. Set Dirty Flow: Based on the Cbox CSR SET_DIRTY_ENABLE[2:0], SetDirty
requests can be either internally acknowledged (called a SetModify) or sent to the
system environment for processing. When externally acknowledged, the shared status information for the cache block is also broadcast. The commands sent externally are SharedToDirty or CleanToDirty. Based on the Cbox CSR
ENABLE_STC_COMMAND[0], the external system can be informed of a STx_C
generating a SetDirty using the STCChangeToDirty command. See Table 4–16 for
more information.
2. InvalToDirty: Based on the Cbox CSR INVAL_TO_DIRTY_ENABLE[1:0], InvalToDirty requests can be either internally acknowledged or sent to the system environment as InvalToDirty commands. This Cbox CSR provides the ability to convert
WH64 instructions to RdModx operations. See Table 4–15 for more information.
3. Evict: There are two aspects to the commands that are generated by an ECB
instruction: first, those commands that are generated to notify the system of an evict
being performed; second, those commands that are generated by any victim that is
created by servicing the ECB.
–

If Cbox CSR ENABLE_EVICT[0] is clear, no command is issued by the
21264/EV68A on the external interface to notify the system of an evict being
performed. If Cbox CSR ENABLE_EVICT[0] is set, the 21264/EV68A issues
an Evict command on the system interface only if a Bcache index match to the
ECB address is found in the 21264/EV68A cache system.
Note that whenever ENABLE_EVICT[0] is true (in the write-many chain),
BC_CLEAN_VICTIM must also be true (in the write-once chain). Otherwise,
the 21264/EV68A could respond miss to a probe, rather than hit, before an
Evict command has been sent off chip, but after the Evict command has
removed a (clean) block from the internal caches and the Bcache. That behavior might cause systems that maintain an external duplicate copy of the Bcache
tags to become confused, because the system could receive the probe response
indicating the miss before it receives the Evict command.

–

The 21264/EV68A can issue the commands CleanVictimBlk and WrVictimBlk
for a victim that is created by an ECB. CleanVictimBlk is issued only if Cbox
CSR BC_CLEAN_VICTIM is set and there is a Bcache index match valid but
not dirty in the 21264/EV68A cache system. WrVictimBlk is issued for any
Bcache match of the ECB address that is dirty in the 21264/EV68A cache system.

4. MB: Based on the Cbox CSR SYSBUS_MB_ENABLE, the MB command can be
sent to the pins.
Each of these CSRs is programmed appropriately, based on the cache coherence protocol used by the system environment. For example, uniprocessor systems would prefer
to internally acknowledge most of these transactions. In contrast, multiprocessor systems may require notification and control of any change in cache state. The 21264/
EV68A and the external system must cooperate to maintain cache coherence. Section
4.5 explains the 21264/EV68A part of the cache coherency protocol.

4–6

Cache and External Interfaces

21264/EV68A Hardware Reference Manual

Bcache Structure

4.3 Bcache Structure
The 21264/EV68A Cbox provides control signals and an interface for a second-level
cache (Bcache).
The 21264/EV68A supports a Bcache from 1MB to 16MB, with 64-byte blocks. A 128bit bidirectional data bus is used for transfers between the 21264/EV68A and the
Bcache. The Bcache is fully synchronous and the synchronous static RAMs (SSRAMs)
must contain either one, two, or three internal registers. All Bcache control and address
pins are clocked synchronously on Bcache cycle boundaries. The Bcache clock rate
varies as a multiple of the CPU clock cycle in half-cycle increments from 1.5 to 4.0,
and in full-cycle increments of 5, 6, 7, and 8 times the CPU clock cycle. The 1.5 multiple is only available in dual-data mode.

4.3.1 Bcache Interface Signals
Figure 4–2 shows the 21264/EV68A system interface signals.
Figure 4–2 21264/EV68A Bcache Interface Signals
BcData_H[127:0]
21264

BcCheck_H[15:0]
BcDataInClk_H[7:0]
BcDataOutClk_x[3:0]
BcDataOE_L
BcDataWr_L
BcAdd_H[23:4]
BcTag_H[42:20]
BcTagInClk_H
BcTagOutClk_x
BcVref
BcTagDirty_H
BcTagParity_H
BcTagShared_H
BcTagValid_H
BcTagOE_L
BcTagWr_L
BcLoad_L
FM-05650

4.3.2 System Duplicate Tag Stores
The 21264/EV68A provides Bcache state support for systems with and without duplicate tag stores, and will take different actions on this basis. The system sets the Cbox
CSR DUP_TAG_ENA[0], indicating that it has a duplicate tag store for the Bcache.
Systems using the DUP_TAG_ENA[0] bit must also use the Cbox CSR
BC_CLEAN_VICTIM[0] bit to avoid deadlock situations.
Systems using a Bcache duplicate tag store can accelerate system performance by:

21264/EV68A Hardware Reference Manual

Cache and External Interfaces

4–7

Victim Data Buffer

•

Issuing probes and SysDc fill commands to the 21264/EV68A out-of-order with
respect to their order at the system serialization point

•

Filtering out all probe misses from the 21264/EV68A cache system

If a probe misses in the 21264/EV68A cache system (Bcache miss and VAF miss), the
21264/EV68A stalls probe processing with the expectation that a SysDc fill will allocate this block. Because of this, in duplicate tag mode, the 21264/EV68A can never
generate a probe miss response.
When Cbox CSR DUP_TAG_ENA[0] equals 0, the 21264/EV68A delivers a miss
response for probes that do not hit in its cache system.

4.4 Victim Data Buffer
The 21264/EV68A has eight victim data buffers (VDBs). They have the following
properties:

•

The VDBs are used for both victims (fills that are replacing dirty cache blocks) and
for system probes that require data movement. The CleanVictimBlk command
(optional) assigns and uses a VDB.

•

Each VDB has two valid bits that indicate the buffer is valid for a victim or valid
for a probe or valid for both a victim and a probe. Probe commands that match the
address of a victim address file (VAF) entry with an asserted probe-valid bit (P)
will stall the 21264/EV68A probe queue. No ProbeResponses will be returned until
the P bit is clear.

•

The release victim buffer (RVB) bit, when asserted, causes the victim valid bit, on
the victim data buffer (VDB) specified in the ID field, to be cleared. The RVB bit
will also clear the IOWB when systems move data on I/O write transactions. In this
case, ID[3] equals one.

•

The release probe buffer (RPB) bit, when asserted (with a WriteData or ReleaseBuffer SysDc command), clears the P bit in the victim buffer entry specified in the
ID field.

•

Read data commands and victim write commands use IDs 0-7, while IDs 8-11 are
used to address the four I/O write buffers.

4.5 Cache Coherency
This section describes the basics and protocols of the 21264/EV68A cache coherency
scheme.

4.5.1 Cache Coherency Basics
The 21264/EV68A systems maintain the cache hierarchy shown in Figure 4–3.

4–8

Cache and External Interfaces

21264/EV68A Hardware Reference Manual

Cache Coherency

Figure 4–3 Cache Subset Hierarchy
System
Main Memory
Bcache

Dcache
Icache

FM-05824.AI4

The following tasks must be performed to maintain cache coherency:

•

Istream data from memory spaces may be cached in the Icache and Bcache. Icache
coherence is not maintained by hardware—it must be maintained by software using
the CALL_PAL IMB instruction.

•

The 21264/EV68A maintains the Dcache as a subset of the Bcache. The Dcache is
set-associative but is kept a subset of the larger externally implemented directmapped Bcache.

•

System logic must help the 21264/EV68A to keep the Bcache coherent with main
memory and other caches in the system.

•

The 21264/EV68A requires the system to allow only one change to a block at a
time. This means that if the 21264/EV68A gains the bus to read or write a block, no
other node on the bus should be allowed to access that block until the data has been
moved.

•

The 21264/EV68A provides hardware mechanisms to support several cache coherency protocols. The protocols can be separated into two classes: write invalidate
cache coherency protocol and flush cache coherency protocol.

4.5.2 Cache Block States
Table 4–2 lists the cache block states supported by the 21264/EV68A.
Table 4–2 21264/EV68A-Supported Cache Block States
State Name

Description

Invalid

The 21264/EV68A does not have a copy of the block.

Clean

This 21264/EV68A holds a read-only copy of the block, and no other agent in the system
holds a copy. Upon eviction, the block is not written to memory.

21264/EV68A Hardware Reference Manual

Cache and External Interfaces

4–9

Cache Coherency

Table 4–2 21264/EV68A-Supported Cache Block States (Continued)
State Name

Description

Clean/Shared

This 21264/EV68A holds a read-only copy of the block, and at least one other agent in the
system may hold a copy of the block. Upon eviction, the block is not written to memory.

Dirty

This 21264/EV68A holds a read-write copy of the block, and must write it to memory after it
is evicted from the cache. No other agent in the system holds a copy of the block.

Dirty/Shared

This 21264/EV68A holds a read-only copy of the dirty block, which may be shared with
another agent. The block must be written back to memory when it is evicted.

4.5.3 Cache Block State Transitions
Cache block state transitions are reflected by 21264/EV68A-generated commands to
the system. Cache block state transitions can also be caused by system-generated commands to the 21264/EV68A (probes). Probes control the next state for the cache block.
The next state can be based on the previous state of the cache block. Table 4–3 lists the
next state for the cache block.
Table 4–3 Cache Block State Transitions
Next State

Action Based on Probe Hit

No change

Do not update cache state. Useful for DMA transactions that sample data but
do not want to update tag state.

Clean

Independent of previous state, update next state to Clean.

Clean/Shared

Independent of previous state, update next state to Clean/Shared. This transaction is useful for systems that update memory on probe hits.

T1:
Clean
Dirty

⇒ Clean/Shared
⇒ Dirty/Shared

Based on the dirty bit, make the block clean or dirty shared. This transaction
is useful for systems that do not update memory on probe hits.

⇒
⇒

If the block is Clean or Dirty/Shared, change to Clean/Shared. If the block is
Dirty, change to Invalid. This transaction is useful for systems that use the
Dirty/Shared state as an exclusive state.

T3:
Clean Clean/Shared
Invalid
Dirty
Dirty/Shared Clean/Shared

⇒

The cache state transitions caused by 21264/EV68A-generated commands are under the
full control of the system environment using the SysDc (system data control) commands. Table 4–4 lists these commands.
Table 4–4 System Responses to 21264/EV68A Commands
Response Type

21264/EV68A Action

SysDc ReadData

Fill block with the associated data and update tag with clean cache status.

SysDc ReadDataDirty

Fill block with the associated data and update tag with dirty cache status.

SysDc ReadDataShared

Fill block with the associated data and update tag with shared cache status.

SysDc ReadDataShared/Dirty

Fill block with the associated data and update tag with dirty/shared status.

SysDc ReadDataError

Fill block with all-ones reference pattern and update tag with invalid status.

SysDc ChangeToDirtySuccess

Unconditionally update block with dirty cache status.

SysDc ChangeToDirtyFail

Do not update cache status and fail any associated STx_C instructions.

4–10

Cache and External Interfaces

21264/EV68A Hardware Reference Manual

Cache Coherency

4.5.4 Using SysDc Commands
Note the following:

•

The conventional response for RdBlk commands is SysDc ReadData or ReadDataShared.

•

The conventional response for a RdBlkMod command is SysDc ReadDataDirty.

•

The conventional response for ChangeToDirty commands is
ChangeToDirtySuccess or ChangeToDirtyFail.

However, the system environment is not limited to these responses. Table 4–5 shows all
21264/EV68A commands, system responses, and the 21264/EV68A reaction. The
21264/EV68A commands are described in the following list:

•

Rdx commands are generated by load or Istream references.

•

RdBlkModx commands are generated by store references.

•

The ChxToDirty command group includes CleanToDirty, SharedToDirty, and STCChangeToDirty commands, which are generated by store references that hit in the
21264/EV68A cache system.

•

InvalToDirty commands are generated by WH64 instructions that miss in the
21264/EV68A cache system.

•

FetchBlk and FetchBlkSpec are noncached references to memory space that have
missed in the 21264/EV68A cache system.

•

Rdiox commands are noncached references to I/O address space.

•

Evict and STCChangeToDirty commands are generated by ECB and STx_C
instructions, respectively.

Table 4–5 shows the system responses to 21264/EV68A commands and 21264/EV68A
reactions.
Table 4–5 System Responses to 21264/EV68A Commands and Reactions
21264/EV68A
CMD

SysDc

21264/EV68A Action

Rdx

ReadData
ReadDataShared

This is a normal fill. The cache block is filled and marked clean or
shared based on SysDc.

Rdx

ReadDataShared/Dirty

The cache block is filled and marked dirty/shared. Succeeding store
commands cannot update the block without external reference.

Rdx

ReadDataDirty

The cache block is filled and marked dirty.

Rdx

ReadDataError

The cache block access was to NXM address space. The 21264/
EV68A delivers an all-ones pattern to any load command and evicts
the block from the cache (with associated victim processing). The
cache block is marked invalid.

Rdx

ChangeToDirtySuccess Both SysDc responses are illegal for read commands.
ChangeToDirtyFail

21264/EV68A Hardware Reference Manual

Cache and External Interfaces

4–11

Cache Coherency

Table 4–5 System Responses to 21264/EV68A Commands and Reactions (Continued)
21264/EV68A
CMD

SysDc

21264/EV68A Action

RdBlkModx

ReadData
ReadDataShared
ReadDataShared/Dirty

The cache block is filled and marked with a nonwritable status. If the
store instruction that generated the RdBlkModx command is still
active (not killed), the 21264/EV68A will retry the instruction, generating the appropriate ChangeToDirty command. Succeeding store
commands cannot update the block without external reference.

RdBlkModx

ReadDataDirty

The 21264/EV68A performs a normal fill response, and the cache
block becomes writable.

RdBlkModx

ChangeToDirtySuccess Both SysDc responses are illegal for read/modify commands.
ChangeToDirtyFail

RdBlkModx

ReadDataError

The cache block command was to NXM address space. The 21264/
EV68A delivers an all-ones pattern to any dependent load command,
forces a fail action on any pending store commands to this block, and
any store to this block is not retried. The Cbox evicts the cache block
from the cache system (with associated victim processing). The cache
block is marked invalid.

ChxToDirty

ReadData
ReadDataShared
ReadDataShared/Dirty

The original data in the Dcache is replaced with the filled data. The
block is not writable, so the 21264/EV68A will retry the store instruction and generate another ChxToDirty class command. To avoid a
potential livelock situation, the STC_ENABLE CSR bit must be set.
Any STx_C instruction to this block is forced to fail. In addition, a
Shared/Dirty response causes the 21264/EV68A to generate a victim
for this block upon eviction.

ChxToDirty

ReadDataDirty

The data in the Dcache is replaced with the filled data. The block is
writable, so the store instruction that generated the original command
can update this block. Any STx_C instruction to this block is forced
to fail. In addition, the 21264/EV68A generates a victim for this
block upon eviction.

ChxToDirty

ReadDataError

Impossible situation. The block must be cached to generate a ChxToDirty command. Caching the block is not possible because all NXM
fills are filled noncached.

ChToDirty

ChangeToDirtySuccess Normal response. ChangeToDirtySuccess makes the block writable.
The 21264/EV68A retries the store instruction and updates the
Dcache. Any STx_C instruction associated with this block is allowed
to succeed.

ChxToDirty

ChangeToDirtyFail

The MAF entry is retired. Any STx_C instruction associated with the
block is forced to fail. If a STx instruction generated this block, the
21264/EV68A retries and generates either a RdBlkModx (because the
reference that failed the ChangeToDirty also invalidated the cache by
way of an invalidating probe) or another ChxToDirty command.

InvalToDirty

ReadData
ReadDataShared
ReadDataShared/Dirty

The block is not writable, so the 21264/EV68A will retry the WH64
instruction and generate a ChxToDirty command.

InvalToDirty

ReadDataError

The 21264/EV68A doesn’t send InvalToDirty commands offchip
speculatively. This NXM condition is a hard error. Systems should
perform a machine check.

InvalToDirty

ReadDataDirty
The block is writable. Done.
ChangeToDirtySuccess

4–12

Cache and External Interfaces

21264/EV68A Hardware Reference Manual

Cache Coherency

Table 4–5 System Responses to 21264/EV68A Commands and Reactions (Continued)
21264/EV68A
CMD

SysDc

21264/EV68A Action

InvalToDirty

ChangeToDirtyFail

Illegal. InvalToDirty instructions must provide a cache block.

Fetchx
Rdiox

ReadData
ReadDataShared
ReadDataShared/Dirty
ReadDataDirty

The 21264/EV68A delivers the data block, independent of its
status, to waiting load instructions and does not cache the block in the
21264/EV68A cache system.

Fetchx

ReadDataError

The cache block address was to an NXM address space. The 21264/
EV68A delivers the all-ones patterns to any dependent load instructions and does not cache the block in the 21264/EV68A cache system.

Rdiox

ReadDataError

The cache block access was to NXM address space. The 21264/
EV68A delivers an all-ones pattern to any load command and does
not cache the block in the 21264/EV68A cache system.

Evict

ChangeToDirtyFail

Retiring the MAF entry is the only legal response.

STCChangeTo
Dirty

ReadDataX
ChangeToDirtyFail

All fill and ChangeToDirtyFail responses will fail the STx_C requirements.

STCChangeTo
Dirty

ChangeToDirtySuccess The STx_C instruction succeeds.

MBDone

Acknowledgment for MB.

The 21264/EV68A sends a WrVictimBlk command to the system when it evicts a Dirty
or Dirty/Shared cache block. The 21264/EV68A may be configured to send a CleanVictimBlk to the system (by way of the Cbox CSR BC_CLEAN_VICTIM[0]) when evicting a clean or shared block. Both commands allocate buffers in the VAF (victim address
file). This buffer is a coherent part of the 21264/EV68A cache system. Write data control and deallocation of the VAF can be directly controlled by using the SysDc WriteData and ReleaseBuffer commands.

4.5.5 Dcache States and Duplicate Tags
Each Dcache block contains an extra state bit (modified bit), beyond those required to
support the cache protocol. If set, this bit indicates that the associated block should be
written to the Bcache when it is evicted from the Dcache. The modified bit is set in two
cases:
1. When a block is filled into the Dcache from memory its modified bit is set, ensuring that it also gets written back into the Bcache at some future time.
2. When the processor writes to a dirty Dcache block the modified bit is set, indicating
it should be written to the Bcache when evicted.
The contents of the modified bit are functionally invisible to the external cache environment, but knowledge of the bits function is useful to programmers optimizing the
scheduling of the Bcache data bus.
The Cbox contains a duplicate copy of the Dcache tag array. In contrast to the Dcache
tag array (DTAG), which is virtually indexed, the Cbox copy of the Dcache tag array
(CTAG) is physically-indexed. The Cbox uses the CTAG array entries in the following
situations.

21264/EV68A Hardware Reference Manual

Cache and External Interfaces

4–13

Lock Mechanism

1. When the Mbox requests a Dcache fill, the Cbox uses the CTAG array entry to find
if the Dcache already contains the requested physical address in another virtuallyindexed Dcache line. If it does, the Cbox invalidates that cache line after first writing the data back to the Bcache if it was in the modified state. The Cbox also checks
to see if the Dcache contains an address different from the requested address, but
maps to the same Bcache line. If it does, the Dcache line is evicted in order to keep
the Dcache a subset of the Bcache.
2. When the Ibox requests an Icache fill, the Cbox uses the CTAG array entries to find
if the Dcache contains the requested physical address in the modified state. If it
does, the Cbox forces the line to be written back to the Bcache before servicing the
Icache fill request. The Cbox also checks to see if the Dcache contains an address
different from the requested address but which maps to the same Bcache line. In
this case the Istream request will miss the Bcache, and the Cbox will
service the request by launching a noncached Fetch command to the system port
and will not put the Istream block into the Bcache. This mechanism allows the
21264/EV68A to use a cache resident lock flag for LDx_L/STx_C instructions.
3. The Cbox uses the CTAG array entries to find whether probe addresses are held in
the Dcache without interrupting load/store instruction processing in the processor
core.

4.6 Lock Mechanism
The 21264/EV68A does not contain a dedicated lock register, nor are system components required to do so.
When a load-lock (LDx_L) instruction executes, data is accessed from the Dcache or
Bcache. If there is a cache miss, data is accessed from memory with a RdBlk command.
Its associated cache line is filled into the Dcache in the clean state, if it is not already
there.
When the store-conditional (STx_C) instruction executes, it is allowed to succeed if its
associated cache line is still present in the Dcache and can be made writable; otherwise,
it fails.
This algorithm is successful because another agent in the system writing to the cache
line between the load-lock and the store-conditional cache line would make the cache
line invalid. This mechanism’s coherence is based on the following four items:
1. LDx_L instructions are processed in-order in relation to the associated STx_C.
2. Once a block is locked by way of an LDx_L instruction, no internal agent can evict
the block from the Dcache as a side-effect of its processing.
3. Any external agent that intends to update the contents of the stored block must use
an invalidating probe command to inform the 21264/EV68A.
4. The system is the only agent with sufficient information to manage the tasks of fairness and liveness. However, to enable these tasks, the 21264/EV68A only generates
external commands for nonspeculative STx_C instructions, and once given a success indication from the system, must faithfully update the Dcache with the STx_C
value.
The system is entirely responsible for item number three. The 21264/EV68A plays an
active role in items one, two, and four.

4–14

Cache and External Interfaces

21264/EV68A Hardware Reference Manual

Lock Mechanism

4.6.1 In-Order Processing of LDx_L/STx_C Instructions
The 21264/EV68A uses the stWait logic in the IQ to ensure that LDx_L/STx_C pairs
are issued in order. The stWait logic treats an Ldx_L instruction like Stx instructions.
STx_C instructions are always loaded into the IQ with their associate stWait bit set.
Thus, a STx_C instruction is not issued until the older LDx_L is out of the IQ.

4.6.2 Internal Eviction of LDx_L Blocks
The 21264/EV68A prevents the eviction of cache blocks in the Dcache due to either of
the following references:

•

Istream references with a Bcache index that matches the Dcache block and a
Bcache tag that mismatches the Dcache block.
To avoid evictions of LDx_L blocks, Istream references that match the index of a
block in the Dcache are converted to noncached references.

•

Ldx or Stx references with a Dcache index that matches the block.
In the Alpha architecture, Dstream references between a LDx_L/STx_C pair force
the value of the STx_C success flag to be UNPREDICTABLE. The 21264/EV68A
forces all STx_C instructions that interrupt an LDx_L/STx_C pair to fail in program order.
There should be no Dstream references between LDx_L/STx_C pairs; however, the
out-of-order nature of the 21264/EV68A can introduce Dstream references between
LDx_L/STx_C pairs. To prevent load or store instructions older than the LDx_L
from evicting the LDx_L cache block, the Mbox invokes a replay trap on the
incoming load or store instruction, which also aborts the LDx_L. These instructions
are issued in program order in the next iteration of the trap retry down the pipeline.
To prevent newer load or store instructions from evicting the locked cache line, the
Ibox ensures that a STx_C is issued before any newer load or store instruction by
placing the STx_C into the IQ and stalling all subsequent instructions in the map
stage of the pipe until the IQ is empty.
Branch instructions between the LDx_L/STx_C pair may be mispredicted, introducing load and store instructions that evict the locked cache block. To prevent that
from happening, there is a bit in the instruction fetcher that is set for a LDx_L reference and cleared on any other memory reference. When this bit is set, the branch
predictor predicts all branches to fall through.

4.6.3 Liveness and Fairness
To prevent a livelock condition, the 21264/EV68A processes the STx_C as follows:
1. If a STx_C misses the Dcache, then no system port transaction is started and the
STx_C fails.
2. If a STx_C hits a block that is not dirty, then a ChangeToDirty (Shared or Clean) is
launched after the STx_C retires and all older store queue entries are in the writable
state. This ensures that once the ChangeToDirty command is launched on behalf of
the STx_C, the STx_C will be executed to completion if the ChangeToDirty command succeeds.

21264/EV68A Hardware Reference Manual

Cache and External Interfaces

4–15

System Port

If the ChangeToDirty command succeeds, the STx_C enters the writable state, and the
Mbox locks the Dcache line. The Mbox does not release the Dcache line until the
STx_C data is transferred to the Dcache. This ensures that no other agent, by way of a
probe, can take the block before the STx_C can update the locked block.

4.6.4 Managing Speculative Store Issues with Multiprocessor Systems
The 21264/EV68A provides two mechanisms to manage an inherent potential side
effect of speculative execution with multiprocessor systems — a livelock condition
caused by a speculative store that misses in one processor affecting the execution of a
LDx_L/STx_C pair in another processor. The potential livelock condition in multiprocessor systems can be effectively controlled by placing processors in a conservative
mode, where speculative store MAFs are blocked. The 21264/EV68A manages conservative mode with the Mbox IPR, M_CTL[SMC], described in Table 5–19.

•

M_CTL[SMC] can be set to place the 21264/EV68A in full-time conservative
mode.

•

M_CTL[SMC] can be set to place the 21264/EV68A in periodic conservative
mode, timed by two counters: an 8-bit primary counter that tracks branch mispredicts and conditional branch retires, and a backup counter that places the 21264/
EV68A in conservative mode for a period of 16K cycles every 2 million cycles.
The 8-bit counter is enabled by placing M_CTL[SMC] in periodic conservative
mode. The backup counter takes effect whenever the 8-bit counter is enabled. Further, the backup counter can be reset to 0 by clearing a previously set
M_CTL[SMC], allowing synchronization between processors.

4.7 System Port
The system port is the 21264/EV68A’s connection to either a memory or I/O controller
or to a shared multiprocessor system controller. System port interface signals are shown
in Figure 4–4.
The system port supports transactions between the 21264/EV68A and the system. Systems must receive and drive signals that are asserted low. Transaction commands are
communicated on signal lines SysAddOut_L[14:0] (21264/EV68A-to-system) and
SysAddIn_L[14:0] (system-to-21264/EV68A). Transaction data is transferred on a
bidirectional data bus over pins SysData_L[63:0] with ECC on pins SysCheck_L[7:0].

4–16

Cache and External Interfaces

21264/EV68A Hardware Reference Manual

System Port

Figure 4–4 System Interface Signals
SysAddIn_L[14:0]
21264

SysAddInClk_L
SysAddOut_L[14:0]
SysAddOutClk_L
SysVref
SysData_L[63:0]
SysCheck_L[7:0]
SysDataInClk_H[7:0]
SysDataOutClk_L[7:0]
SysDataInValid_L
SysDataOutValid_L
SysFillValid_L
IRQ_H[5:0]
FM-05652-EV67

4.7.1 System Port Pins
Table 3–1 defines the 21264/EV68A signal types referred to in this section. Table 4–6
lists the system port pin groups along with their type, number, and functional description.
Table 4–6 System Port Pins
Pin Name

Type

Count Description

IRQ_H[5:0]

I_DA

These six interrupt signal lines may be asserted by the system.

SysAddIn_L[14:0]

I_DA

Time-multiplexed SysAddIn, system-to-21264/EV68A.

SysAddInClk_L

I_DA

Single-ended forwarded clock from system for
SysAddIn_L[14:0] and SysFillValid_L.

SysAddOut_L[14:0]

O_OD

Time-multiplexed SysAddOut, 21264/EV68A-to-system.

SysAddOutClk_L

O_OD

Single-ended forwarded clock.

SysVref

I_DC_REF

System interface reference voltage.

SysCheck_L[7:0]

B_DA_OD

Quadword ECC check bits for SysData_L[63:0].

SysData_L[63:0]

B_DA_OD

Data bus for memory and I/O data.

SysDataInClk_H[7:0]

I_DA

Single-ended system-generated clocks for clock forwarded
input system data.

SysDataInValid_L

I_DA

When asserted, marks a valid data cycle for data transfers to
the 21264/EV68A.

SysDataOutClk_L[7:0]

O_OD

Single-ended 21264/EV68A-generated clocks for clock forwarded output system data.

SysDataOutValid_L

I_DA

When asserted, marks a valid data cycle for data transfers
from the 21264/EV68A.

SysFillValid_L

I_DA

Validation for fill given in previous SysDc command.

21264/EV68A Hardware Reference Manual

Cache and External Interfaces

4–17

System Port

4.7.2 Programming the System Interface Clocks
The system forwarded clocks are free running and derived from the 21264/EV68A
GCLK. The period of the system forwarded clocks is controlled by three Cbox CSRs,
based on the bit-rate ratio (similar to the Bcache bit-rate ratio) except that all transfers
are dual-data.

•

SYS_CLK_LD_VECTOR[15:0]

•

SYS_BPHASE_LD_VECTOR[3:0]

•

SYS_FDBK_EN[7:0]

Table 4–7 lists the programming values used to program the system interface clocks.
Table 4–7 Programming Values for System Interface Clocks
System Transfer

SYS_CLK_LD_VECTOR1 SYS_BPHASE_LD_VECTOR1 SYS_FDBK_EN 1

1.5X-DD

9249

2.0X-DD

3333

2.5X-DD

8C63

3.0X-DD

71C7

3.5X-DD

C387

4.0X-DD

0F0F

5.0X-DD

7C1F

6.0X-DD

F03F

7.0X-DD

C07F

8.0X-DD

00FF

These are hexadecimal values.

In addition to programming of the clock CSRs, the data-sample/drive Cbox CSRs at the
pads have to be set appropriately. Table 4–8 shows the programmed values for these
system CSRs. In Table 4–8, each system forwarded clock is the inversion of the lowassertion signal at the corresponding pin.
Table 4–8 Program Values for Data-Sample/Drive CSRs
CBOX CSR

Description

SYS_DDM_FALL_EN[0]

Enables the update of 21264/EV68A system outputs based on the falling edge
of the system forwarded clock. (Always asserted)

SYS_DDM_RISE_EN[0]

Enables the update of 21264/EV68A system outputs based on the rising edge
of the system forwarded clock. (Always asserted)

SYS_DDM_RD_FALL_EN[0]

Enables the sampling of incoming data on the falling edge of the incoming
forwarded clock. (Always asserted)

4–18

Cache and External Interfaces

21264/EV68A Hardware Reference Manual

System Port

Table 4–8 Program Values for Data-Sample/Drive CSRs (Continued)
CBOX CSR

Description

SYS_DDM_RD_RISE_EN[0]

Enables the sampling of incoming data on the rising edge of the incoming
forwarded clock. (Always asserted)

SYS_DDMF_ENABLE

Enables the falling edge of the system forwarded clock. (Always asserted)

SYS_DDMR_ENABLE

Enables the rising edge of the system forwarded clock. (Always asserted)

Table 4–9 lists the program values for CSR SYS_FRAME_LD_VECTOR[4:0] that set
the ratio between the forwarded clocks and the frame clock.
Table 4–9 Forwarded Clocks and Frame Clock Ratio
Clock Ratio

Transfer Mode

Value1

1:1

All

2:1

3.0X, 3.5X, 8.0X

2:1

1.5X, 2.0X, 2.5X 4.0X, 5.0X, 6.0X 7.0X

4:1

1.5X, 4.0X, 5.0X, 6.0X, 7.0X

4:1

3.0X, 3.5X

4:1

2.0X, 2.5X

These are hexadecimal values.

4.7.3 21264/EV68A-to-System Commands
This section describes the 21264/EV68A-to-system commands format and operation.
The command, address, ID, and mask bits are transmitted in four consecutive cycles on
SysAddOut_L[14:0]. The 21264/EV68A sends the command information in one of the
two following modes as selected by the Cbox CSR bit.

•

Bank interleave on cache block boundary mode—SYSBUS_FORMAT[0] = 0

•

Page hit mode—SYSBUS_FORMAT[0] = 1

The physical address (PA) bits arrangements for the two modes is shown in Tables 4–10
and 4–11. The purpose of the two modes is to give the system the PA bits that allow it to
select the memory bank and drive the RAS address as soon as possible.
4.7.3.1 Bank Interleave on Cache Block Boundary Mode

Table 4–10 shows the command format for the bank interleave on cache block boundary mode of operation (21264/EV68A-to-system).
Table 4–10 Bank Interleave on Cache Block Boundary Mode of Operation
SysAddOut_L[14:2]

Cycle 1

Command[4:0]

PA[34:28]

21264/EV68A Hardware Reference Manual

SysAddOut_L[1] SysAddOut_L[0]

PA[36]

PA[38]

Cache and External Interfaces

4–19

System Port

Table 4–10 Bank Interleave on Cache Block Boundary Mode of Operation (Continued)
SysAddOut_L[14:2]

Cycle 2
Cycle 3

Cycle 4

SysAddOut_L[1] SysAddOut_L[0]

PA[27:22], PA[12:6]

PA[35]

PA[37]

PA[40]

PA[42]

PA[39]

PA[41]

Mask[7:0]

ID[2:0]

PA[21:13], PA[5:3]

4.7.3.2 Page Hit Mode

Table 4–11 shows the command format for page hit mode (21264/EV68A-to-system).
Table 4–11 Page Hit Mode of Operation
SysAddOut_L[14:2]

Cycle 1

Command[4:0]

Cycle 2

PA[31:25]

PA[24:12]

Cycle 3

Cycle 4

Mask[7:0]

ID[2:0]

PA[34:32], PA[11:3]

SysAddOut_L[1] SysAddOut_L[0]

PA[32]

PA[33]

PA[11]

PA[34]

PA[35]

PA[37]

PA[36]

PA[38]

Table 4–12 describes the field definitions for Tables 4–10 and 4–11.
Table 4–12 21264/EV68A-to-System Command Fields Definitions
SysAddOut Field

Definition

When set, reports a miss to the system for the oldest probe.
When clear, has no meaning.

Command[4:0]

The 5-bit command field is defined in Table 4–14.

SysAddOut[1:0]

This field is needed for systems with greater than 32GB of memory, up to a maximum of 8
Terabyte (8TB). Cost-focused systems can tie these bits high and use a 13-bit command/
address field.

When set, reports that the oldest probe has missed in cache. Also, this bit is set for systemto-21264/EV68A probe commands that hit but have no data movement (see the CH bit,
below).
When clear, has no meaning.
M1 and M2 are not asserted simultaneously. Reporting probe results as soon as possible is
critical to high-speed operation, so when a result is known the 21264/EV68A uses the earliest opportunity to send an M signal to the system. M bit assertion can occur either in a
valid command or a NZNOP.

ID[2:0]

The ID number for the MAF, VDB, or WIOB associated with the command.

If set, validates this command.
In speculative read mode (optional), RV = 1 validates the command and RV = 0 indicates
a NOP.
For all nonspeculative commands RV = 1.

Mask[7:0]

The byte, LW, or QW mask field for the corresponding I/O commands.

The cache hit bit is asserted, along with M2, when probes with no data movement hit in
the Dcache or Bcache. This response can be generated by a probe that explicitly indicates
no data movement or a ReadIfDirty command that hits on a valid but clean or shared
block.

4–20

Cache and External Interfaces

21264/EV68A Hardware Reference Manual

System Port

System designers can minimize pin count for systems with a small memory by configuring both the bank interleave on cache block boundary mode and the page hit mode
formats into a short bus format. The pin SysAddOut_L[1] and/or SysAddOut_L[0]
are not used (selected by Cbox CSR SYS_BUS_SIZE[1:0]). Table 4–13 lists the values
for SYSBUS_FORMAT and SYS_BUS_SIZE[1:0] and shows the maximum physical
memory size.
Table 4–13 Maximum Physical Address for Short Bus Format
SYSBUS_
FORMAT

SYSBUS_
SIZE[1:0]

Maximum PA

Comment

Bank interleave + full address

Bank interleave + SysAddOut_L[0] unused

Illegal

Illegal combination

Bank interleave + both SysAddOut_L[1:0] are used for I/O

Page hit mode + full address

Page hit mode + SysAddOut_L[0] unused

Illegal

Illegal combination

Page hit mode + both SysAddOut_L[1:0] are unused

Because addresses above the maximum PA are not visible to the external system, any
memory transaction generated to addresses above the maximum PA are detected and
converted to transactions to NXM (nonexistent memory) and processed internally by
the 21264/EV68A.

4.7.4 21264/EV68A-to-System Commands Descriptions
Table 4–14 describes the 21264/EV68A-to-system commands.
Table 4–14 21264/EV68A-to-System Commands Descriptions
Command

Command
[4:0]

NOP

00000

The 21264/EV68A drives this command on idle cycles during reset. After
the clock forward reset period, the first NZNOP is generated and this
command is no longer generated.

ProbeResponse

00001

Returns probe status and ID number of the VDB entry holding the
requested cache block.

NZNOP

00010

This nonzero NOP helps to parse the command packet.

VDBFlushRequest

00011

VDB flush request. The 21264/EV68A sends this command to the system
when an internally generated transaction Bcache index matches a Bcache
victim or probe in the VDB. The system should flush VDB entries
associated with all probe and WrVictimBlk transactions that occurred
before this command.

MB1

00111

Indicates an MB was issued, optional when Cbox CSR
SYSBUS_MB_ENA[0] is set.

ReadBlk

10000

Memory read.

Function

21264/EV68A Hardware Reference Manual

Cache and External Interfaces

4–21

System Port

Table 4–14 21264/EV68A-to-System Commands Descriptions (Continued)
Command

Command
[4:0]

Function

ReadBlkMod

10001

Memory read with modify intent.

ReadBlkI

10010

Memory read for Istream.

FetchBlk

10011

Noncached memory read.

10100

Speculative memory read (optional).

10101

Speculative memory read with modify intent (optional).

ReadBlkSpecI2

10110

Memory read for Istream (optional).

10111

Speculative memory noncached ReadBlk (optional).

11000

Memory read with a victim (optional).

11001

Memory read with modify intent, with a victim (optional).

ReadBlkVicI3

11010

Memory read for Istream with a victim (optional).

WrVictimBlk

00100

Write-back of dirty block.

CleanVictimBlk

ReadBlkSpec2
ReadBlkModSpec

FetchBlkSpec

ReadBlkVic3
ReadBlkModVic

00101

Address of a clean victim (optional).

Evict

00110

Invalidate evicted block at the given Bcache index (optional).

ReadBytes

01000

I/O read, byte mask.

ReadLWs

01001

I/O read, longword mask.

ReadQWs

01010

I/O read, quadword mask.

WrBytes

01100

I/O write, byte mask.

WrLWs

01101

I/O write, longword mask.

WrQWs

01110

I/O write, quadword mask.

11100

Sets a block dirty that was previously clean (optional for duplicate tags).

11101

Sets a block dirty that was previously shared (optional for multiprocessor
systems).

STCChangeToDirty6

11110

Sets a block dirty that was previously clean or shared for a STx_C
instruction (optional for multiprocessor systems).

InvalToDirtyVic3,5

11011

Invalid to dirty with a victim (optional).

InvalToDirty5

11111

WH64 Acts like a ReadBlkMod without the fill cycles (optional).

CleanToDirty6
SharedToDirty

4–22

Cache and External Interfaces

21264/EV68A Hardware Reference Manual

System Port

Table 4–14 footnotes:
1. Systems can optionally enable MB instructions to the external system by asserting
Cbox CSR SYSBUS_MB_ENABLE. This mode is described in Section 2.12.1.
2. To minimize load-to-use memory latency, systems can optionally enable speculative transactions to memory space by asserting the Cbox CSR
SPEC_READ_ENABLE[0]. If the Cbox system command queue is empty, a
bypass between the Bcache interface and the system interface is enabled (in combination with this mode). When the next new transaction is delivered by the Mbox,
the Cbox starts MAF memory references to the system interface before the results
of Bcache hit is known. The RV bit is deasserted on a Bcache hit, or in
BC_RDVICTIM[0] mode (see footnote 3, below), and for Bcache miss transactions
that generate a victim (clean or dirty). Otherwise, the RV bit is asserted.
3. Systems can optionally enable RdBlkVic, RdBlkModVic, and InvalToDirtyVic
commands using Cbox CSR BC_RDVICTIM[0]. In this mode of operation
RdBlkxVic command cycles are always followed immediately by the WrVictimBlk
commands. Also, when CleanVictimBlk commands are enabled, they
immediately follow RdBlkVic, RdBlkModVic, and InvalToDirtyVic commands.
4. Systems can optionally enable Evict commands by asserting the Cbox CSR
ENABLE_EVICT. In this mode, all ECB instructions will generate an Evict command, and in combination with BC_RDVICTIM[0] mode, the WriteVictim or
CleanVictim (when Cbox CSR BC_CLEAN_VICTIM[0] is asserted) is associated
with the Evict command is atomically sent after the Evict command.
5. Optionally, systems can enable InvalToDirty commands by programming Cbox
CSR INVAL_TO_DIRTY_ENABLE[1:0]. Table 4–15 shows how to program
INVAL_TO_DIRTY_ENABLE[1:0].
6. Optionally, systems can enable CleanToDirty or SharedToDirty commands by
using Cbox CSR SET_DIRTY_ENABLE[2:0]. These three bits control the Cbox
action upon a block that was hit in the Dcache with a status of dirty/shared, clean/
shared, or clean respectively.
Table 4–15 Programming INVAL_TO_DIRTY_ENABLE[1:0]
INVAL_TO_DIRTY_ENABLE[1:0]

Cbox Action

WH64 instructions are converted to RdModx commands at the interface.
Beyond this point, no other agent sees the WH64 instruction. This mode is
useful for microprocessors that do not want to support InvalToDirty transactions.

WH64 instructions are enabled, but they are acknowledged within the
21264/EV68A.

WH64 instructions are enabled, and generate InvalToDirty transactions at
the 21264/EV68A pins.

21264/EV68A Hardware Reference Manual

Cache and External Interfaces

4–23

System Port

Table 4–16 Programming SET_DIRTY_ENABLE[2:0]
SET_DIRTY_ENABLE
[2,0] (DS,CS,C)

Cbox Action

000

Everything acknowledged internally (uniprocessor).

001

Only clean blocks generate external acknowledge (CleanToDirty commands only).

010

Only clean/shared blocks generate external acknowledge (SharedToDirty command
only).

011

Clean and clean/shared blocks generate external acknowledge.

100

Only dirty/shared blocks generate external acknowledge (SharedToDirty commands
only).

101

Only dirty/shared and clean blocks generate external acknowledge.

110

Only dirty/shared and clean/shared blocks generate external acknowledge.

111

All transactions generate external acknowledge.

Systems that require an explicit indication of ChangeToDirty status changes initiated by STx_C instructions can assert Cbox CSR STC_ENABLE[0]. When this
register field = 000, CleanToDirty and SharedToDirty commands are used. The distinction between a ChangeToDirty command generated by a STx_C instruction and
one generated by a STx instruction is important to systems that want to service
ChangeToDirty commands with dirty data from a source processor. In this case, the
distinction between a locked exclusive instruction and a normal instruction is critical to avoid livelock for a LDx_L/STx_C sequence.

4.7.5 ProbeResponse Commands (Command[4:0] = 00001)
The 21264/EV68A responds to system probes that did not miss with a 4-cycle transfer
on SysAddOut_L[14:0]. As shown in Table 4–14, the Command[4:0] field for a ProbeResponse command equals 00001. Table 4–17 shows the format of the 21264/EV68A
ProbeResponse command.
Table 4–17 21264/EV68A ProbeResponse Command
SysAddOut_L[14:2]

Cycle 1

Cycle 2
Cycle 3
Cycle 4

4–24

00001

Status[1:0]

0
0

X
X

Cache and External Interfaces

SysAddOut_L[1] SysAddOut_L[0]

VDB X
[2:0]

MAF X
[2:0]

21264/EV68A Hardware Reference Manual

System Port

Table 4–18 describes the ProbeResponse command fields.
Table 4–18 ProbeResponse Fields Descriptions
ProbeResponse Field

Description

Command[4:0]

The value 00001 identifies the command as a ProbeResponse.

Indicates that data movement should occur (copy of probe valid bit). See Section 4.4.

Write victim sent bit.

VDB[2:0]

ID number of the VDB entry containing the requested cache block. This field is valid
when either the DM bit or the VS bit equals 1.

MAF address sent.

MAF[2:0]

This field indicates the SharedToDirty, CleanToDirty, or
STCChangetoDirty MAF entry that matched the full probe address.

Status[1:0]

Result of probe:
Status[1:0] Probe state
00
HitClean
01
HitShared
10
HitDirty
11
HitSharedDirty

The system uses the SysDc signal lines to retrieve data for probes that requested a cache
block from the 21264/EV68A. See Section 4.7.7.2 for more information about 2-cycle
data transfer commands. Probes that respond with M1, M2, or CH=1 will not be
reported to the system in a probe response command.

4.7.6 SysAck and 21264/EV68A-to-System Commands Flow Control
Controlling the flow of 21264/EV68A-to-system commands is a joint task of the
21264/EV68A and the system. The flow is controlled using the A bit, which is asserted
by the system, and the Cbox CSR SYSBUS_ACK_LIMIT[4:0] counter. The counter
has the following properties:

•

The 21264/EV68A increments its command-outstanding counter when it sends a
command to the system. The 21264/EV68A decrements the counter by one each
time the A bit (SysAddIn_L[14]) is asserted in a system-to-21264/EV68A command. The A bit is transmitted during cycle four of a probe mode command or during cycle two of a SysDc command.

•

The 21264/EV68A stops sending new commands when the counter hits the maximum count specified by Cbox CSR SYSBUS_ACK_LIMIT[4:0]. When this
counter is programmed to zero, the CMD_ACK count is ignored (unlimited commands are allowed in-flight).

•

Because RdBlkxVic and WrVictimBlk commands are atomic when the CSR
BC_RDVICTIM[0] is set, the 21264/EV68A does not send a RdBlkxVic command
if the SYSBUS_ACK_LIMIT[4:0] is equal to one less than the maximum outstanding count. The limit cannot be programmed with a value of one when RdBlkxVic
commands are enabled unless the Cbox CSR RDVIC_ACK_INHIBIT command is
also asserted (see Table 5–24).

21264/EV68A Hardware Reference Manual

Cache and External Interfaces

4–25

System Port

•

There is no mechanism for the system to reject a 21264/EV68A-to-system command. ProbeResponse, VDBFlushReq, NOP, NZNOP, and RdBlkxSpec (with a
clear RV bit) commands do not require a response from the system. Systems must
provide adequate resources for responses to all probes sent to the 21264/EV68A.

•

Systems that program the Cbox CSR BC_RDVICTIM[0] to immediately follow
victim write transactions with read transactions and allocate combined resources
for the pair, may find it useful to increment the SYSBUS_ACK_LIMIT[4:0]
counter only once for the pair. These systems may assert Cbox CSR
RDVIC_ACK_INHIBIT, which does not increment the
SYSBUS_ACK_LIMIT[4:0] count for RdBlkVic, RdBlkModVic, and RdBlkVicI
commands.

•

Systems that maintain victim data buffers may find it useful to limit the number of
outstanding WrVictimBlk commands. This can be accomplished by using the Cbox
CSR SYSBUS_VIC_LIMIT[2:0]. When the number of outstanding WrVictim
commands or CleanVictim commands reaches this programmed limit, the Cbox
stops generating victim commands on the system port. Because victim and read
commands are atomic when BC_RDVICTIM[0] = 1, the RdBlkxVic commands are
stalled when the victim limit is reached. Programming the
SYSBUS_VIC_LIMIT[2:0] to zero disables this limit.

4.7.7 System-to-21264/EV68A Commands
The system can send either probes (4-cycle) or data movement (2-cycle) commands to
the 21264/EV68A. Signal pin SysAddIn_L[14] in the first command cycle indicates
the type of command being sent (1 = probe, 0 = data transfer). Sections 4.7.7.1 and
4.7.7.2 describe the formats of the two types of commands.
4.7.7.1 Probe Commands (Four Cycles)

Probes are always 4-cycle commands that contain a field to indicate a valid SysDc command. The format of the 4-cycle command is shown below.
The SysAddIn_L[1:0] signal lines are optional and are used for memory
designs greater than 32GB. The position of the address bits matches the
selected format of the SysAddOut bus. The example below shows the bank
interleave format.

Note:

Table 4–19 shows the format of the system-to-21264/EV68A probe commands.
Table 4–19 System-to-21264/EV68A Probe Commands
SysAddIn_L[14:2]

Cycle 1 1
Cycle 2
Cycle 3 0
Cycle 4 C

4–26

SysAddIn_L[1] SysAddIn_L[0]

Probe[4:0]

PA[34:28]

PA[27:22], PA[12:6]
SysDc[4:0]

RVB

RPB

PA[21:13], PA[5:3]

Cache and External Interfaces

ID[3:0]

PA[36]

PA[38]

PA[35]

PA[37]

PA[40]

PA[42]

PA[39]

PA[41]

21264/EV68A Hardware Reference Manual

System Port

Table 4–20 describes the system-to-21264/EV68A probe commands fields descriptions.
Table 4–20 System-to-21264/EV68A Probe Commands Fields Descriptions
SysAddIn_L[14:0]
Field
Description

Probe[4:0]

Probe type and next tag state (see Tables 4–21 and 4–22).

SysDc[4:0]

Controls data movement in and out of the 21264/EV68A. See Table 4–24 for a list of data
movement types.

RVB

Clears the victim or I/O write buffer (IOWB) valid bit specified in ID[3:0].

RPB

Clears probe valid bit specified in ID[2:0].

Command acknowledge. When set, the 21264/EV68A decrements its command outstanding counter (SYSBUS_ACK_LIMIT[4:0]).

ID[3:0]

Identifies the victim data buffer (VDB) number or the I/O write buffer (IOWB) number.
Bit [3] is only asserted for the IOWB.

Commit bit. This bit decrements the uncommitted event counter (MB_CNTR) used for
MB acknowledge.

The probe command field Probe[4:0] has two sections, Probe[4:3] and Probe[2:0].
Table 4–21 lists the data movement selected by Probe[4:3].
Table 4–21 Data Movement Selection by Probe[4:3]
Probe[4:3]

Data Movement Function

NOP

Read if hit, supply data to system if block is valid.

Read if dirty, supply data to system if block is valid/dirty.

Read anyway, supply data to the system at index of probe.

Table 4–22 lists the next cache block state selected by Probe[2:0].
Table 4–22 Next Cache Block State Selection by Probe[2:0]
Probe[2:0]

Next Tag State

000

NOP

001

Clean

010

Clean/Shared

011

Clean/Shared
Transition31: Clean
Invalid
Dirty
Clean/Shared
Dirty/Shared

100

⇒
⇒

⇒

Dirty/Shared

21264/EV68A Hardware Reference Manual

Cache and External Interfaces

4–27

System Port

Table 4–22 Next Cache Block State Selection by Probe[2:0] (Continued)
Probe[2:0]

Next Tag State

101

Invalid

110

Transition12: Clean
Dirty

111

Reserved

⇒ Clean/Shared
⇒ Dirty/Shared

Transition3 is useful in nonduplicate tag systems that want to give writable status to the reader and do
not know if the block is clean or dirty.
2
Transition1 is useful in nonduplicate tag systems that do not update memory on ReadBlk hits to a
dirty block in another processor.

The 21264/EV68A holds pending probe commands in a 8-entry deep probe queue. The
system must count the number of probes that have been sent and ensure that the probes
do not overrun the 21264/EV68A queue. The 21264/EV68A removes probes from the
internal probe queue when the probe response is sent.
The 21264/EV68A expects to hit in cache on a probe response, so it always fetches a
cache block from the Bcache on system probes. This can become a performance problem for systems that do not monitor the Bcache tags, so the 21264/EV68A provides
Cbox CSR PRB_TAG_ONLY[0], which only accesses Bcache tags for system probes.
For a Bcache hit, the 21264/EV68A retries the probe reference to get the associated
data. In this mode, the 21264/EV68A has a cache-hit counter that maintains some history of past cache hits in order to fetch the data with the tag in the cases where streamed
transactions are being performed to the host processor.
4.7.7.2 Data Transfer Commands (Two Cycles)

Data transfer commands use a 2-cycle format on SysAddIn_L[14:0]. The SysDc[4:0]
field indicates success or failure for ChangeToDirty and MB commands, and error conditions as shown in Table 4–24.
The pattern of data is controlled by the SysDataInValid_L and SysDataOutValid_L
signals. These signals are valid each cycle of data transfer, indicating any gaps in the
data cycle pattern. The SysDataInValid_L and SysDataOutValid_L signals are
described in Section 4.7.8.4. Table 4–23 shows the format of the data transfer command.
Table 4–23 Data Transfer Command Format
SysAddIn_L[14:2]

Cycle 1

Cycle 2

4–28

SysDc[4:0]

RVB

Cache and External Interfaces

RPB
X

SysAddIn_L[1] SysAddIn_L[0]

ID[3:0]

21264/EV68A Hardware Reference Manual

System Port

Table 4–24 describes the SysDc[4:0] field.
Table 4–24 SysDc[4:0] Field Description
SysDc[4:0] Command

SysDc[4:0]

Description

NOP

00000

NOP, SysData is ignored by the 21264/EV68A.

ReadDataError

00001

Data is returned for read commands. The system drives the SysData
bus, I/O, or memory NXM.

ChangeToDirtySuccess

00100

No data. SysData is ignored by the 21264/EV68A. This command is
also used for the InvalToDirty response.

ChangeToDirtyFail

00101

No data. SysData is ignored by the 21264/EV68A. This command is
also used for the Evict response.

MBDone

00110

Memory barrier operation completed.

ReleaseBuffer

00111

Command to alert the 21264/EV68A that the RVB, RPB, and ID
field are valid.

ReadData
(System Wrap)

100xx

Data returned for read commands. The system drives SysData. The
system uses SysDc[1:0] to control the wrap order. See Section
4.7.8.6 for a description of the data wrapping scheme.

ReadDataDirty
(System Wrap)

101xx

Data is returned for Rdx and RdModx commands. The ending tag
status is dirty. The system uses SysDc[1:0] to define the wrap order.

ReadDataShared
(System Wrap)

110xx

Data is returned for read commands. The system drives the data. The
tag is marked shared. The system uses SysDc[1:0] to control the
wrap order.

ReadDataShared/Dirty
(System Wrap)

111xx

Data is returned for the RdBlk command. The ending tag status is
Shared/Dirty. The system uses SysDc[1:0] to control the wrap order.

WriteData

010xx

Data is sent for 21264/EV68A write commands or system probes.
The 21264/EV68A drives during the SysData cycles. The lower two
bits of the command specify the octaword address around which the
21264/EV68A wraps the data.

The A bit in the first cycle indicates that the command is acknowledged. When A = 1, the
21264/EV68A decrements its command outstanding counter, but the A bit is not necessarily related to the current SysDc command.
Probe commands can combine a SysDc command along with MBDone. In that event,
the probe is considered ahead of the SysDc command. If the SysDc command allows
the 21264/EV68A to retire an instruction before an MB, or allows the 21264/EV68A
itself to retire an MB (SysDc is MBDone), that MB will not complete until the probe is
executed.
The system can select the ending cache status for a cache fill operation by specifying
the status in one of the following SysDc commands:
ReadData (Clean)

ReadDataShared (Clean/Shared)

ReadDataDirty (Dirty)

ReadDataShared/Dirty (Shared/Dirty)

The system returns ReadDataShared or ReadData for ReadBlk commands, and ReadDataDirty for a ReadMod command. However, other combinations are possible, but
should be used only after a careful study of the situation.

21264/EV68A Hardware Reference Manual

Cache and External Interfaces

4–29

System Port

The ChangeToDirtySuccess and ChangeToDirtyFail commands cannot be issued in the
shadow of SysDc cache fill commands (ReadDataError, ReadData, ReadDataDirty,
ReadDataShared, and ReadDataShared/Dirty). Each cache fill command allocates eight
cycles on the SysData bus. Systems are required to ensure that any future SysDc commands do not cause conflicts with those eight SysData bus cycles. In addition, the system must not issue ChangeToDirtySuccess or ChangeToDirtyFail commands in the six
SysAddrIn cycles after any of the ReadDatax commands because doing so will overload internal MAF resources in the 21264/EV68A.
Because of an internal 21264/EV68A constraint, a minimum memory latency of
4 × BCACHE_CLK_PERIOD is imposed. This latency is measured from A3 of the outgoing command (the last cycle) to the delivery of the SysDc command to the processor.

4.7.8 Data Movement In and Out of the 21264/EV68A
There are two modes of operation for data movement in and out of the 21264/EV68A:
fast mode and fast mode disable. The data movement mode is selected using Cbox CSR
FAST_MODE_DISABLE[0]. Fast data mode allows movement of data from the
21264/EV68A to bypass protocol and achieve the lowest possible latency for probe’s
data, write victim data, and I/O write data. Rules and conditions for the two modes are
listed and described in Sections 4.7.8.2 and 4.7.8.3. Before discussing data movement
operation, 21264/EV68A clock basics are described in Section 4.7.8.1.
4.7.8.1 21264/EV68A Clock Basics

The 21264/EV68A uses a clock forwarding technique to achieve very high bandwidth
on its pin interfaces. The clock forwarding technique has three main principles:
1. Local point-to-point transfers can be made safely, and at very high bandwidth, if the
sender can provide the receiver with a forward clock (FWD_CLK) to latch the
transmitted data at the receiver.
–

The SysAddOutClk_L and SysDataOutClk_L[7:0] pins provide the forwarding clocks for transfers out of the 21264/EV68A.

–

The SysAddInClk_L and SysDataInClk_H[7:0] pins provide the forwarding
clocks for transfers into the 21264/EV68A.

2. If only one state element was used to capture the transmitted data, and the skew
between the two clock systems was greater then the bit-rate of the transfer, the data
valid time of the transmitted data would not be sufficient to safely transfer the
latched data into the receivers clock domain. In order to avoid this problem, the
receiver provides a queue that is manipulated in the transmitter’s time domain.
Using this queue, the data valid window of the transmitted data is extended (to an
arbitrary size based on the queue size), and the transfer to the receiver’s clock
domain can be safely made by delaying the unloading of this queue element beyond
the skew between the two clock domains. The internal clock that unloads this queue
is labelled INT_FWD_CLK. INT_FWD_CLK is timed at both the rising and falling edges of the external clock, thus appearing to run at twice the external clock’s
frequency.
3. The first two points provide the steady state basis for clock forwarded transfers;
however, both the sender and receiver must be correctly initialized to enable coherent and predictable transfers. This clock initialization is performed during system
initialization using the ClkFwdRst_H and FrameClk_H signals.

4–30

Cache and External Interfaces

21264/EV68A Hardware Reference Manual

System Port

If both the sender and the receiver are sampling at the same rate, these three principles
are sufficient to safely make point-to-point transfers using clock forwarding. However,
it is often desirable for systems to align clock-forwarded transactions on a slower
SYSCLK that is the basis of all non-processor system transactions.
The 21264/EV68A supports three ratios for SYSCLK to INT_FWD_CLK:
one-to-one (1-1), two-to-one (2-1), and four-to-one (4-1). Using one of these ratios, the
21264/EV68A starts transactions on SYSCLK boundaries. This ratio is programmed
into the 21264/EV68A using the Cbox CSR SYS_FRAME_LD_VECTOR[4:0]. This
ratio is independent of the frequency of FrameClk_H.
For data movement, the 21264/EV68A reacts to SysDc commands when they are
resolved into the 21264/EV68A’s clock domain. This occurs when the 21264/EV68A’s
INT_FWD_CLK unloads the SysDc command from the clock forwarding queue. This
moment is determined by the amount of delay programmed into the clock forwarding
silo (by way of Cbox CSR SYS_RCV_MUX_CNT_PRESET[1:0]). Thus, all the timing relationships are relative to this unload point in time, which will be referred to as
the point the command is perceived by 21264/EV68A.
4.7.8.2 Fast Data Mode

The 21264/EV68A is the default driver of the bidirectional SysData bus1. As the
21264/EV68A is processing WrVictim, ProbeResponse (only the hit case), and IOWB
commands to the system, accompanying data is made available at the clock-forwarded
bus.
Because there is a bandwidth difference between address (4 cycles) and data (8 cycles)
transfers, the 21264/EV68A tries to fully use fast data mode by delaying the next
SysAddOut write command until a fast data mode slot is available on the SysDataOut
bus.
SysDc commands (cache fill or explicit write commands) that collide with the fast data
on the SysData bus have higher priority, and so may interrupt the successful completion
of the fast transfer. Systems are responsible for detecting and replaying all interrupted
fast transfers. There are no gaps in a fast transfer and no data wrapping (the first cycle
contains QW0, addressed by PA[5:3] = 000).
The system must release victim buffers, and probe buffers and IOWB entries by sending a SysDc command with the appropriate RVB/RPB bit for both successful fast data
transfers and for transfers that have been replayed. Fast data transfers have two parts:
1. SysAddOut command with the probe response, WrVictim, or Wr(I/O)
2. Data
The command precedes data by at least one SYSCLK period. Table 4–25 shows the
number of SYSCLK cycles between SysAddOut and SysData for all system clock
ratios (clock forwarded bit times) and system framing clock multiples.

1 The SysData bus contains SysData_L[63:0] and SysCheck_L[7:0].

21264/EV68A Hardware Reference Manual

Cache and External Interfaces

4–31

System Port

Table 4–25 SYSCLK Cycles Between SysAddOut and SysData
GCLK/INT_FWD_CLK (Data Rate Ratio)

System framing clock ratio

1.5X

2.0X

2.5X

3.0X

3.5X

4.0X

5.0X

6.0X

7.0X

8.0X

Figure 4–5 show a simple example of a fast transfer. The data rate ratio is 1.5X with a
4:1 SYSCLK to INT_FWD_CLK ratio.
Figure 4–5 Fast Transfer Timing Example
SysAddOut_L[14:0]

Probe Response

SysData_L[63:0]

SYSCLK

SysAddOutClk_L

INT_FWD_CLK

GCLK
FM05822B.AI4

In fast data mode, movement of data into the 21264/EV68A requires turning around the
SysData bus that is being actively driven by the 21264/EV68A. Given a SysDc fill
command (ReadDataError, ReadData, ReadDataShared, ReadDataShared/Dirty, ReadDataDirty), the 21264/EV68A responds as follows:
1. Three GCLK cycles after perceiving the SysDc fill command, the 21264/EV68A
turns off its drivers, interrupting any ongoing fast data write transactions.
2. The 21264/EV68A drivers stay off until the last piece of fill data is received, or a
new SysDc write command overrides the current SysDc fill command. It is the
responsibility of the external system to schedule SysDc fill or write commands so
that there is no conflict on the SysData bus.
3. The 21264/EV68A samples fill data in the GCLK clock domain, 10 +
SYSDC_DELAY GCLK cycles after perceiving the SysDc fill command. The
Cbox CSR SYSDC_DELAY[4:0] provides GCLK granularity for precisely placing
fills into the processor pipeline discussed in Section 2.2.

4–32

Cache and External Interfaces

21264/EV68A Hardware Reference Manual

System Port

Table 4–26 shows four example configurations and shows their use of the
SYSDC_DELAY[4:0].
Table 4–26 Cbox CSR SYSDC_DELAY[4:0] Examples
System

Bit Rate

System Framing Clock Ratio1

SYSDC_DELAY

System 1

1.5X

4:1

5 (3 SYSCLK cycles)

System 2

2.0X

2:1

2 (3 SYSCLK cycles)

System 3

2.5X

2:1

0 (2 SYSCLK cycles)

System 4

2:1

6 (2 SYSCLK cycles)

The system framing clock ratio is the number of INT_FWD_CLK cycles per
SYSCLK cycles.

System 1 has six GCLKs to every SYSCLK and only sends 4-cycle commands to the
21264/EV68A. Thus, a period of three SYSCLKs between the SysDc command and
data leaves a period of 15 GCLKs between SysDc and data (SysDc is in the middle of
the 4-cycle command). A SYSDC_DELAY[4:0] of five would align sampling and
receipt of SysData.
System 2 has four GCLKs in every SYSCLK, so leading data by three SYSCLK cycles,
and programming the SYSDC_DELAY[4:0] to two, aligns sampling and receiving.
Timing for systems 3 and 4 is derived in a similar manner.
Note:

The maximum valid value for SYSDC_DELAY must be less than the minimum number of GCLK cycles between two consecutive SYSDC commands to the 21264/EV68A.

If a fast data transfer is interrupted and fails to complete, the system must use the conventional protocol to send a SysDc WriteData command to the 21264/EV68A, removing the desired data buffer. Section 4.7.8.3 describes the timing events for transferring
data from the 21264/EV68A to the system.
4.7.8.3 Fast Data Disable Mode

The system controls all data movement to and from the 21264/EV68A. Movement of
data into and out of the 21264/EV68A is preceded by a SysDc command. The 21264/
EV68A drivers are only enabled for the duration of an 8-cycle transfer of data from the
21264/EV68A to the system. Systems must ensure that there is no overlap of enabled
drivers and that there is adequate settle time on the SysData bus.
Given a SysDc fill command, the 21264/EV68A samples data 10 + SYSDC_DELAY
GCLK cycles after the command is perceived within the 21264/EV68A clock domain.
Because there is no linkage with the output driver, fills into the 21264/EV68A are not
affected by the SYS_RCV_MUX_PRESET[1:0] value.
In both modes, given a SysDc write command, the 21264/EV68A looks for the next
SYSCLK edge 8.5 cycles after perceiving the SysDc write command in its clock
domain. Because the SysDc write command must be perceived before its use, SysDc
write commands are dependent upon the amount of delay introduced by Cbox CSR
SYS_RCV_MUX_CNT_PRESET[1:0].

21264/EV68A Hardware Reference Manual

Cache and External Interfaces

4–33

System Port

Table 4–27 lists information for the four timing examples. In Table 4–27, note the following:

•

SysDc write commands are not affected by the SYSDC_DELAY parameter.

•

The SYS_RCV_MUX_PRESET adds delay at the rate of one INT_FWD_CLK at a
time. For example, adding the delay of one bit time to system 1 adds 1.5 GCLK
cycles to the delay and drives the SysDc write command-to-data relationship from
one to two SYSCLKs.

•

For write transfers, the 21264/EV68A drivers are enabled on the preceding GCLK
BPHASE, before the start of a write transfer, and disabled on the succeeding GCLK
BPHASE at the end of the write transfer. The write data is enveloped by the 21264/
EV68A drivers to guarantee that every data transfer has the same data valid window.

Table 4–27 Four Timing Examples
System

Bit Rate

System Framing Clock Ratio1

Write Data

System 1

1.5X

4:1

2 SYSCLKs

System 2

2.0X

2:1

3 SYSCLKs

System 3

2.5X

2:1

2 SYSCLKs

System 4

2:1

2 SYSCLKs

The system framing clock ratio is the number of INT_FWD_CLK cycles per
SYSCLK cycles.

The four examples described here assume no skew for the 2.0X and 4.0X cases and one
bit time of skew for the 1.5X and 2.5X cases.
For system 1, the distance between SysDc and the first SYSCLK is nine GCLK cycles
but the additional delay of one bit time (1.5 GCLKs) puts the actual delay after perceiving the SysDc command to 7.5 GCLKS, which misses the 8.5 cycle constraint. Therefore, the 21264/EV68A drives data two SYSCLKs after receiving the SysDc write
command.
For system two, the distance between SysDc and the second SYSCLK is eight GCLK
cycles, which also misses the 8.5 cycle constraint, so the 21264/EV68A drives data
three SYSCLK cycles after receiving the SysDc write command (12 cycles).
The other two cases are derived in a similar manner.
4.7.8.4 SysDataInValid_L and SysDataOutValid_L

The SysDataValid signals (SysDataInValid_L and SysDataOutValid_L) are driven by
the system and control the rate of data delivery to and from the 21264/EV68A.
SysDataInValid_L
The SysDataInValid_L signal controls the flow of data into the 21264/EV68A, and

may be used to introduce an arbitrary number of cycles between octaword transfers into
the 21264/EV68A. The rules for using SysDataInValid_L follow:

4–34

Cache and External Interfaces

21264/EV68A Hardware Reference Manual

System Port

1. The SysDataInValid_L signal must be asserted for both cycles of a SysDc fill
command, and two quadwords of data must be delivered to the 21264/EV68A in
succeeding bit-clock cycles with the appropriate timing in reference to the SysDc
fill command (SYSDC_DELAY + 10 CPU cycles).
2. Any number of bubble cycles can be introduced within the fill by deasserting
SysDataInValid_L between octaword transfers.
3. The transfer of fill data can continue by asserting SysDataInValid_L for at least
two bit-clock cycles, and delivering data SYSDC_DELAY + 10 CPU cycles after
the assertion of SysDataInValid_L.
4. The 21264/EV68A must see SysDataInValid_L asserted for eight data cycles in
order to complete a fill. When the eighth cycle of an asserted SysDataInValid_L is
perceived by the 21264/EV68A, the transfer is complete.
5. Systems that do not use SysDataInValid_L may tie the pin to the asserted state.
If SYSDC_DELAY is greater than the bit-time of a transfer, the SysDataInValid_L
signal must be internally pipelined. To enable the correct sampling of
SysDataInValid_L, the 21264/EV68A provides a delay, with Cbox CSR
DATA_VALID_DELAY[1:0], that is equal to SYSDC_DELAY[4:0]/bit-time. For
example, consider system 1 in Table 4–26, which has a SYSDC_DELAY of five
GCLKs. Running at a bit-time of 1.5X, the DATA_VALID_DELAY[1,0] is programmed with a value of three.
SysDataOutValid_L

Systems that use a ratio of 1:1 for SYSCLK:INT_FWD_CLK may control the flow of
data out of the 21264/EV68A by using SysDataOutValid_L as follows:
1. The SysDataOutValid_L pin must be asserted for at least the first cycle of the
SysDc write command that initiates a write transfer.
2. Any number of bubble cycles may be introduced between quadword transfers by
deasserting SysDataOutValid_L.
3. The 21264/EV68A must see the SysDataOutValid_L signal asserted for eight data
cycles to complete a write transaction, and when the eighth cycle of an asserted
SysDataOutValid_L is perceived by the 21264/EV68A, the transfer is complete.
4.7.8.5 SysFillValid_L

The SysFillValid_L pin, when asserted, validates the current memory and I/O data
transfer into the 21264/EV68A. The system designer may tie this pin to the asserted
state (validating all fills), or use it to enable or cancel fills as they progress. The 21264/
EV68A samples SysFillValid_L at D1 time (when the 21264/EV68A samples the second data cycle).
If SysFillValid_L is asserted at D1 time, the fill will continue uninterrupted. If it is not
asserted, the 21264/EV68A cancels the fill, but expects all eight QWs of data to arrive
at its system bus before continuing to the next fill. Also, the 21264/EV68A maintains
the state of the MAF, expecting another valid fill to the same MAF entry. Figure 4–6
illustrates SysFillValid_L timing.

21264/EV68A Hardware Reference Manual

Cache and External Interfaces

4–35

System Port

Figure 4–6 SysFillValid_L Timing
SysAddIn_L[14:0]

SysDc

Transport Delay on Address
T3

Command Receiver

SysFillValid_L

SysData_L[63:0]

FM-05823B.FH8

4.7.8.6 Data Wrapping

All data movement between the 21264/EV68A and the system is composed of 64 bytes
in eight cycles on the data bus. All 64 bytes of memory data are valid. This applies to
memory read transactions, memory write transactions, and system probe read transactions. The wrap order is interleaved. The internal data bus, which delivers data to the
functional units and the Dcache, is 16 bytes wide, and so, no transfers happen until two
data cycles occur on the interface.
Table 4–28 lists the rules for data wrapping. I/O read and write addresses on the
SysAddOut bus point to the desired byte, word, LW, or QW, with a combination of
SysAddOut_L[5:3] and the mask field [7:0].
Table 4–28 Data Wrapping Rules
Command

Significant Address Mask
Bits
Type

Rules

ReadQW and
WrQW

SysAddOut_L[5:3]

SysAddOut_L[5:3] contains the exact PA bits of the first
LDQ or STQ to the block. The mask bits point to the valid
QWs merged in ascending order.

ReadLW and
WrLW

SysAddOut_L[5:3]

SysAddOut_L[5:3] contain the exact PA bits of the first
LDL or STL to the block. The mask bits point to the valid
LWs merged in ascending order within one hexword.

LDByte/Word
and
STByte/Word

SysAddOut_L[5:3]

Byte

SysAddOut_L[5:3] contain the exact QW PA bits of the
LDByte/Word or STByte/Word instruction. The mask bits
point to the valid byte in the QW.

The order in which data is provided to the 21264/EV68A (for a memory or I/O fill) or
moved from the 21264/EV68A (write victims or probe reads) can be determined by the
system. The system chooses to reflect back the same low-order address bits and the corresponding octaword found in the SysAddOut field or the system chooses any other
starting point within the block.
SysDc commands for the ReadData, ReadDataShared, and WriteData groups require
that systems define the position of the first QW by inserting the appropriate value of
SysAddOut_L[5:3] into bits [1:0] of the command field. The recommended starting

4–36

Cache and External Interfaces

21264/EV68A Hardware Reference Manual

System Port

point is the QW pointed to by the 21264/EV68A; however, some systems may find it
more beneficial to begin the transfer elsewhere. The system must always indicate the
starting point to the 21264/EV68A. The wrap order for subsequent QWs is interleaved.
Table 4–29 defines the method for systems to specify wrap and deliver data.
Table 4–29 System Wrap and Deliver Data
Source/
Destination

SysDc[4:2]

SysDc[1:0]

Size

Rules

Memory

100 (ReadData)

SysAddOut_L[5:4]

Block (64 Bytes)

See Note 1

Memory

101(ReadDataDirty)

SysAddOut_L[5:4]

Block (64 Bytes)

See Note 1

Memory

110 (ReadDataShared)

SysAddOut_L[5:4]

Block (64 Bytes)

See Note 1

Memory

111(Read DataShared/Dirty)

SysAddOut_L[5:4]

Block (64 Bytes)

See Note 1

Memory

010 (WriteData)

SysAddOut_L[5:4]

Block (64 Bytes)

See Note 1

I/O

100 (ReadData)

SysAddOut_L[5:4]

QW (8-64 Bytes)

See Note 1

I/O

100 (ReadData)

SysAddOut_L[4:3]

LW(4-32 Bytes)

See Note 2

I/O

100 (ReadData)

SysAddOut_L[4:3]

Byte/Word

See Note 2

I/O

010 (WriteData)

SysAddOut_L[5:4]

QW (8-64 Bytes)

See Note 1

I/O

010 (WriteData)

SysAddOut_L[5:4]

LW(4-32 Bytes)

See Note 1

I/O

010 (WriteData)

SysAddOut_L[5:4]

Byte/Word

See Note 1

Note 1:

Transfers to and from the 21264/EV68A have eight data cycles for a total
of eight quadwords. The starting point is defined by the system. The preferred starting point is the one pointed to by SysAddOut_L[5:4]. Systems
can insert the SysAddOut_L[5:4] into the SysDc[1:0] field of the command. See Table 4–30 for the wrap order.

Note 2:

LW and byte/word read transfers differ from all other transfers. The system
unloads only four QWs of data into eight data cycles by sending each QW
twice (referred to as double-pumped data transfer). The first QW returned
is determined by SysAddOut_L[4:3]. The system again may elect to
choose its own starting point for the transfer and insert that value into
SysDc[1:0]. See Table 4–31 for the wrap order.

Table 4–30 defines the interleaved scheme for the wrap order.
Table 4–30 Wrap Interleave Order
PA Bits [5:3] of Transferred QW

First quadword

000

010

100

110

Second quadword

001

011

101

111

Third quadword

010

000

110

100

Fourth quadword

011

001

111

101

Fifth quadword

100

110

000

010

21264/EV68A Hardware Reference Manual

Cache and External Interfaces

4–37

System Port

Table 4–30 Wrap Interleave Order (Continued)
PA Bits [5:3] of Transferred QW

Sixth quadword

101

111

001

011

Seventh quadword

110

100

010

000

Eighth quadword

111

101

011

001

Table 4–31 defines the wrap order for double-pumped data transfers.
Table 4–31 Wrap Order for Double-Pumped Data Transfers
PA [5:3] of Transferred QW

First quadword

x00

x01

x10

x11

Second quadword

x00

x01

x10

x11

Third quadword

x01

x00

x11

x10

Fourth quadword

x01

x00

x11

x10

Fifth quadword

x10

x11

x00

x01

Sixth quadword

x10

x11

x00

x01

Seventh quadword

x11

x10

x01

x00

Eighth quadword

x11

x10

x01

x00

4.7.9 Nonexistent Memory Processing
Like its predecessors, the 21264/EV68A can generate references to nonexistent (NXM)
memory or I/O space. However, unlike the earlier Alpha microprocessor implementations, the 21264/EV68A can generate speculative references to memory space. To
accommodate the speculative nature of the 21264/EV68A, the system must not generate or lock error registers because of speculative references. The 21264/EV68A translates all memory references through the translation lookaside buffer (TLB) and, in some
cases, the 21264/EV68A may generate speculative references (instruction execution
down mispredicted paths) to NXM space. In these cases, the system sends a SysDc
ReadDataError and the 21264/EV68A does the following:

4–38

•

Delivers an all-ones pattern to all load instructions to the NXM address

•

Force-fails all store instructions to the NXM address (much like a STx_C
failure)

•

Invalidates the cache block at the same index by way of an atomic Evict
command

Cache and External Interfaces

21264/EV68A Hardware Reference Manual

System Port

Table 4–32 shows each 21264/EV68A command, with NXM addresses, and the appropriate system response.
Table 4–32 21264/EV68A Commands with NXM Addresses and System Response
21264/EV68A
Command NXM
Address

System/21264/EV68A Response

ProbeResponse

Probe responses for addresses to NXM space are of UNPREDICTABLE status. Although
the final status of a ReadDataError is Invalid, the 21264/EV68A fills the block Valid/
Clean and uses an atomic Evict command to invalidate the block. Systems that send
probes to NXM space to the 21264/EV68A must disregard the probe result.

RdBlk
RdBlkSpec
RdBlkVic

Load references to NXM space can be speculative. In this case, systems should respond
with a SysDc ReadDataError fill that the 21264/EV68A uses to service the original load/
Istream command. If the original load command was speculative, the 21264/EV68A will
remove the load instruction that generated the NXM command, and start processing
instructions down the correctly predicted path. If the command was not speculative, there
must be an error in the operating system mapping of a virtual address to an illegal physical address, and the 21264/EV68A provides an all ones pattern as a signature for this bug.
The NXM block is not cached in the Dcache or Bcache.

RdBlkI
RdBlkSpecI
RdBlkVicI

Istream references to NXM space can be speculative. In this case, systems should respond
with a SysDc ReadDataError fill, which the 21264/EV68A will use to service and execute the original Istream reference. If the original Istream reference was speculative, the
21264/EV68A will remove the instructions started after the mispredicted instruction that
generated the NXM reference, and start instruction processing down the correctly predicted path. If the reference was not speculative, there must be an error in the operating
system mapping of a virtual address to an illegal physical address, and the 21264/EV68A
provides an all ones pattern as a signature for this bug. The NXM block is not cached in
the Bcache, but can be cached in the Icache.

RdBlkMod
RdBlkModSpec
RdBlkModVic

Store instructions to NXM space initiate RdBlkMod commands. Again, speculative store
instructions are removed. Nonspeculative store instructions are forced to fail, much like
STx_C instructions that fail. The NXM block is not cached in the Dcache or Bcache.

WrVictimBlk

Dirty Victims to NXM space are illegal. Systems should perform a machine check, with
the 21264/EV68A indicating a severe error.

CleanVictimBlk

The 21264/EV68A can generate CleanVictimBlk commands to NXM space if the Cbox
CSR BC_CLEAN_VICTIM[0] bit is asserted and a SysDc ReadDataError has been generated. Systems that use clean victims must faithfully deallocate the CleanVictim VAF
entry.

Evict

If the Cbox CSR ENABLE_EVICT is asserted, the 21264/EV68A will generate Evict
commands to NXM space. Systems may use this command to invalidate their duplicate
tags. Systems must respond with SysDc ChangeToDirtyFail to retire the NXM MAF
entry.

RdBytes
RdLWs
RdQWs

Load instructions to I/O space are not speculative, so an I/O reference to NXM space is
an error. Systems must respond with ReadDataError and should generate a machine
check to indicate an operating system error.

WrBytes
WrLWs
WrQWs

Store instructions to I/O space are not speculative, so an I/O reference to NXM space is
an error. Systems must respond by deallocating the appropriate IOWB entries, and should
generate a machine check to indicate an operating system error.

FetchBlk
FetchBlkSpec

Loads to noncached memory in NXM space may be speculative. Systems must respond
with a SysDc ReadDataError to retire the MAF entry.

21264/EV68A Hardware Reference Manual

Cache and External Interfaces

4–39

System Port

Table 4–32 21264/EV68A Commands with NXM Addresses and System Response (Continued)
21264/EV68A
Command NXM
Address

System/21264/EV68A Response

CleanToDirty
ChangeToDirty commands to NXM space are impossible in the 21264/EV68A because
SharedToDirty
all NXM references to memory space are atomically filled with an Invalid cache status.
STCChangeToDirty
InvalToDirty
InvalToDirtyVic

InvalToDirty commands are not speculative, so InvalToDirty commands to NXM space
indicate an operating system error. Systems should respond with a SysDc ReadDataError,
and should generate a machine check to indicate error.

4.7.10 Ordering of System Port Transactions
This section describes ordering of system port transactions. The two classes of transactions are listed here:

•

21264/EV68A commands and system probes

•

System probes and SysDc transfers

4.7.10.1 21264/EV68A Commands and System Probes

This section describes the interaction of 21264/EV68A-generated commands and system-generated probes that reference the same cache block. Some definitions are presented here:

•

ProbeResponses generated by the 21264/EV68A respond to all system-generated
probe commands. System-generated data transfer commands respond to all 21264/
EV68A-generated data transfer commands.

•

The victim address file (VAF) and victim data buffer (VDB) entries each have independent valid bits for both a victim and a probe.

•

Probe results indicate a hit on a VAF/VDB and when a WrVictim command has
been sent to the system. Systems can decide whether to move the buffer once or
twice.

•

ProbeResponses are issued in the order that the system-generated probes were
received; however, there is no requirement for the system to retain order when issuing release buffer commands.

•

Probe processing can stall inside the 21264/EV68A when the probe entry index
matches PA[19:6] of a previous probe entry in the VAF.

•

The 21264/EV68A reserves one VAF entry for probe processing, so that VAF-full
conditions cannot stall the processing of probes at the head of the queue.

Table 4–33 lists all interactions between pending internal 21264/EV68A commands
and the Probe[2:0] command field, Next Cache Block State, described in Table 4–22.
Table 4–33 shows the 21264/EV68A response to system probe and in-flight command
interaction. In the table, note the following:

•

4–40

ReadBlkVic and ReadBlkModVic commands do not appear in Table 4–33. If there
is interaction between the probe and the victim, it is the same as a WrVictimBlk
command.

Cache and External Interfaces

21264/EV68A Hardware Reference Manual

System Port

•

Probes that invalidate locked blocks do not generate a ReadBlkMod command. The
21264/EV68A fails the STx_C instruction as defined in the Alpha Architecture
Handbook, Version 4.

•

All read commands (RdBlk, RdBlkMod, Fetch, InvalToDirty) do not interact
because the 21264/EV68A does not yet own the block.

Table 4–33 21264/EV68A Response to System Probe and In-Flight Command Interaction
Pending Internal
21264/EV68A
Command

21264/EV68A Response to System Probe and In-Flight Command Interaction

ReadBlk
ReadBlkMod
FetchBlk
InvalToDirty
WrVictimBlk

This case assumes that a WrVictimBlk command has been sent to the system and another
agent has performed a load/store instruction to the same address. The 21264/EV68A provides VAF hit information with the probe response so that the system can manage the race
condition between the WrVictimBlk command from this processor and a possible WrVictimBlk command from the probing processor. This race condition can be managed by
either forcing the completion of the WrVictimBlk command to memory before allowing
the progress by the probing processor, or by killing the WrVictimBlk command in this
processor.

CleanToDirty
SharedToDirty

This case assumes that a SetDirty command has been sent to the system environment
because of a store instruction that hit in the 21264/EV68A caches and that another processor has performed a load/store instruction to the same address. The 21264/EV68A provides MAF hit information so that the system can correctly respond to the Set/Dirty
command. If the next state of the probe was Invalid (the other processor performed a store
instruction), and the probe reached the system serialization point before the Set/Dirty
command, the system must either fail the Set/Dirty command or provide the updated data
from the other processor.

STCChangeToDirty This case is similar to case 2, except that the initiating instruction for the Set/Dirty command is a STx_C. An address match with an invalidating probe must fail the Set/Dirty
command. Delivering the updated data from the other processor is not an option because
of the requirements of the LDx_L/STx_C instruction pair.

21264/EV68A Hardware Reference Manual

Cache and External Interfaces

4–41

Bcache Port

4.7.10.2 System Probes and SysDc Commands

Ordering of cache transactions at the system serialization point must be reflected in the
21264/EV68A cache system. Table 4–34 shows the rules that a system must follow to
control the order of cache status update within the 21264/EV68A cache structures
(including the VAF) at the 21264/EV68A pins.
Table 4–34 Rules for System Control of Cache Status Update Order
First

Second

Rule

Probe

To control the sequence of cache status updates between probes, systems
can present the probes in order to the 21264/EV68A, and the 21264/
EV68A will update the appropriate cache state (including the VAF) in
order.

Probe

SysDc MAF To ensure that a probe updates the internal cache status before a SysDc

MAF transaction (including fills and ChangeToDirtySuccess commands),
systems must wait for the probe response before presenting the SysDc
MAF command to the 21264/EV68A. To ensure that a probe updates a
VAF entry before a SysDc VAF (release buffer), systems must wait for the
probe response.
Probe

SysDc VAF

Same as Probe/SysDc MAF, above.

SysDc MAF

Probe

To ensure that a SysDc MAF command updates the 21264/EV68A cache
system before a probe to the same address, systems must deliver the D1
(the second QW of data delivered to the 21264/EV68A) before or in the
same cycle as the A3 of the probe (the last cycle of the 4-cycle probe command). This rule also applies to ChangeToDirtySuccess commands that
have a virtual D0 and D1 transaction.

SysDc MAF

SysDc MAF SysDc MAF transactions can be ordered into the 21264/EV68A by order-

ing them appropriately at the 21264/EV68A interface.
SysDc MAF

SysDc VAF

SysDc MAF transactions and SysDc VAF transactions cannot interact
within the 21264/EV68A because the 21264/EV68A does not generate
MAF transactions to the same address as existing VAF transactions.

SysDc VAF

Probe

To ensure that a SysDc VAF invalidates a VAF entry before a probe to the
same address, the SysDc VAF command must precede the first cycle of the
4-cycle probe command.

SysDc VAF

SysDc MAF SysDc MAF transactions and SysDc VAF transactions cannot interact

within the 21264/EV68A because the 21264/EV68A does not generate
MAF transactions to the same address as existing VAF transactions.
SysDc VAF

SysDc VAF

SysDc VAF transactions can be ordered into the 21264/EV68A by ordering them appropriately at the 21264/EV68A interface.

4.8 Bcache Port
The 21264/EV68A supports a second-level cache (Bcache) with 64-byte blocks. The
Bcache size can be 1MB, 2MB, 4MB, 8MB, or 16MB. The Bcache port has a 144-bit
data bus that is used for data transfers between the 21264/EV68A and the Bcache. All
Bcache control and address signal lines are clocked synchronously on Bcache clock cycle
boundaries.

4–42

Cache and External Interfaces

21264/EV68A Hardware Reference Manual

Bcache Port

The Bcache supports the following multiples of the GCLK period: 1.5X (dual-data
mode only), 2X, 2.5X, 3X, 3.5X, 4X, 5X, 6X, 7X, and 8X. However, the 21264/EV68A
imposes a maximum Bcache clock period based on the SYSCLK ratio. Table 4–35 lists
the range of maximum Bcache clock periods. Section 4.7.8.2 describes fast mode.
Table 4–35 Range of Maximum Bcache Clock Ratios
SYSCLK Ratio

Bcache Clock Ratio with Fast Mode
Enabled

Bcache Clock Ratio with Fast Mode
Disabled

1.5X

4.0X

7.0X

2.0X

4.0X

7.0X

2.5X

5.0X

8.0X

3.0X

6.0X

8.0X

3.5X

7.0X

8.0X

4.0X

7.0X

8.0X

5.0X

8.0X

6.0X

8.0X

7.0X

8.0X

The 21264/EV68A provides a range of programmable Cbox CSRs to manipulate the
Bcache port pins so that a variety of industry-standard SSRAMs can communicate efficiently with the 21264/EV68A. The following SSRAMs can be used:.

•

Nonburst mode Reg/Reg late-write SSRAMs

•

Burst mode Reg/Reg late-write dual-data SSRAMs

4.8.1 Bcache Port Pins
Table 3–1 defines the 21264/EV68A signal types referred to in this section. Table 4–36
lists the Bcache port pin groups along with their type, number, reference clock, and
functional description.
Table 4–36 Bcache Port Pins
Pin Name

Type

Count Reference Clock

Description

BcAdd_H[23:4]

O_PP

Bcache index

BcCheck_H[15:0]

B_DA_PP

BcData_H[127:0]

B_DA_PP

128

BcDataInClk_H[7:0]

I_DA

Bcache data input clocks

BcDataOE_L

O_PP

Int_Index_BcClk

Bcache data output enable/chip
select

BcDataOutClk_H[3:0] O_PP
BcDataOutClk_L[3:0]

Bcache data clocks— high and low
version

Int_Index_BcClk

⇒ output ECC check bits for BcData
⇒ input
Int_Data_BcClk ⇒ output Bcache data
BcDataInClk_H ⇒ input
Int_Data_BcClk
BcDataInClk_H

21264/EV68A Hardware Reference Manual

Cache and External Interfaces

4–43

Bcache Port

Table 4–36 Bcache Port Pins (Continued)
Pin Name

Type

Count Reference Clock

Description

BcDataWr_L

O_PP

Int_Index_BcClk

Bcache data write enable

BcLoad_L

O_PP

Int_Index_BcClk

Bcache burst enable

BcTag_H[42:20]

B_DA_PP

BcTagDirty_H

B_DA_PP

BcTagInClk_H

I_DA

Tag input data reference clock

BcTagOE_L

O_PP

Int_Index_BcClk

Bcache tag output enable/chip
select

BcTagOutClk_H
BcTagOutClk_L

O_PP

Bcache tag clock— high and low
versions

BcTagParity_H

B_DA_PP

Int_Data_BcClk
BcTagInClk_H

BcTagShared_H

B_DA_PP

BcTagValid_H

B_DA_PP

BcVref

I_DC_REF 1

Input reference voltage for tag data

BcTagWr_L

O_PP

Int_Index_BcClk

Bcache data write enable

⇒ output Bcache tag data
⇒ input
Int_Data_BcClk ⇒ output Bcache tag dirty bit
BcTagInClk_H ⇒ input
Int_Data_BcClk
BcTagInClk_H

⇒ output Bcache tag parity bit
⇒ input
Int_Data_BcClk ⇒ output Bcache tag shared bit
BcTagInClk_H ⇒ input
Int_Data_BcClk ⇒ output Bcache tag valid bit
BcTagInClk_H ⇒ input

4.8.2 Bcache Clocking
For clocking, the Bcache port pins can be divided into three groups.
1. The Bcache index pins (address and control) are referenced to Int_Add_BcClk, an
internal version of the Bcache forwarded clock. The index pins are valid for the
whole period of the Int_Add_BcClk. The index pins are:
BcAdd_H[23:4]
BcDataOE_L
BcDataWr_L
BcLoad_L
BcTagOE_L
BcTagWr_L
2. The data pins, when driven as outputs, are referenced to Int_Data_BcClk, another
internal version of the Bcache forwarded clock. The data pins, when used as inputs,
can be referenced to the incoming Bcache clocks, BcDataInClk_H[7:0] and
BcTagInClk_H. Int_Data_BcClk can be delayed relative to Int_Add_BcClk from
0 to 3 GCLK cycles by using Cbox CSR BC_CPU_CLK_DELAY[1:0]. The data
pins are:
BcCheck_H[15:0]
BcData_H[127:0]
BcTag_H[42:20]
BcTagDirty_H
BcTagParity_H

4–44

Cache and External Interfaces

21264/EV68A Hardware Reference Manual

Bcache Port

BcTagShared_H
BcTagValid_H
3. The Bcache clock pins (BcDataOutClk_x[3:0] and BcTagOutClk_x) clock the
index and data pins at the SSRAMs. These clocks can be delayed from
Int_Data_BcClk from 0 to 2 GCLK phases (half cycles) using Cbox CSR
BC_CPU_CLK_DELAY[1:0].
Table 4–37 provides the BC_CPU_CLK_DELAY[1:0] values, which is the delay
from BC_ADDRESS to BC_WRITE_DATA (and BC_CLOCK_OUT) in GCLK
cycles.
Table 4–37 BC_CPU_CLK_DELAY[1:0] Values
BC_CPU_CLK_DELAY[1:0] Value

GCLK Cycles of Delay

In the 21264/EV68A topology, the index pins are loaded by all the SSRAMs, while the
clock and data pins see a limit load. This arrangement requires a relatively large amount
of delay between the index pins and the Bcache clock pins to meet the setup constraints
at the SSRAMs. The 21264/EV68A Cbox CSRs can provide a programmable amount
of delay between the index and clock pins by using Cbox CSRs
BC_CPU_CLK_DELAY[1:0] and BC_CLK_DELAY[1:0].
Table 4–38 provides the BC_CLK_DELAY[1:0] values, which is the delay from
BC_WRITE_DATA to BC_CLOCK_OUT, in GCLK phases.
Table 4–38 BC_CLK_DELAY[1:0] Values
BC_CLK_DELAY[1:0] Value

GCLK Phases

Invalid (turns off BC_CLOCK_OUT)

With BC_CPU_CLK_DELAY[1:0] and BC_CLK_DELAY[1:0], a 500-MHz 21264/
EV68A can provide up to 8 ns (3 × 2 + 2) of delay between the index and the outgoing
forwarded clocks. The relative loading difference between the data and the clock is
minimal, so Cbox CSR BC_CLK_DELAY[1:0] alone is sufficient to provide the delay
needed for the setup constraint at the Bcache data register.
4.8.2.1 Setting the Period of the Cache Clock

The free running Bcache clocks are derived from the 21264/EV68A GCLK. The period
of the Bcache clocks is programmed using the following three Cbox CSRs:
1. BC_CLK_LD_VECTOR[15:0]
2. BC_BPHASE_LD_VECTOR[3:0]

21264/EV68A Hardware Reference Manual

Cache and External Interfaces

4–45

Bcache Port

3. BC_FDBK_EN[7:0]
To program these three CSRs, the programmer must know the bit-rate of the Bcache
data, and whether only the rising edge or both edges of the clock are used to latch data.
For example, a 200-MHz late-write SSRAM has a data period of 5 ns. For a 2-ns
GCLK, the READCLK_RATIO must be set to 2.5X. This part is called a 2.5X SD (single-data part).
Table 4–39 shows how the three CSRs are programmed for single-data devices.
Table 4–39 Program Values to Set the Cache Clock Period (Single-Data)
Bcache Transfer

BC_CLK_LD_VECTOR1

BC_BPHASE_LD_VECTOR1

BC_FDBK_EN1

2.0X-SD

5555

2.5X-SD

94A5

3.0X-SD

9249

3.5X-SD

4C99

4.0X-SD

3333

5.0X-SD

8C63

6.0X-SD

71C7

7.0X-SD

C387

8.0X-SD

0F0F

These are hexadecimal values.

With the exception of the 2.5X-SD and 3.5X-SD cases, the clock waveform generated
by the 21264/EV68A for the forwarded clocks has a 50-50 duty cycle. In the 2.5X-SD
case, the 21264/EV68A produces an asymmetric clock that is high for two GCLK
phases and low for three phases. Likewise, for the 3.5X-SD case, the 21264/EV68A
produces an asymmetric clock that is high for three GCLK phases and low for four
GCLK phases. Also, for both of these cases, the 21264/EV68A will only start transactions on the rising edge of the GCLK and the Bcache clock. The 1.5X-SD case is not
supported.
A dual-data rate (DDR) SSRAM’s data rate is derived in a similar manner, except that
because both edges of the clock are used, the SSRAM clock generated is 2X the period
of the data. This part is called a 2.5X DDR SSRAM.
Table 4–40 shows how the three CSRs are programmed for dual-data devices.
Table 4–40 Program Values to Set the Cache Clock Period (Dual-Data Rate)
Bcache
Transfer

BC_CLK_LD_VECTOR1

BC_BPHASE_LD_VECTOR1 BC_FDBK_EN1

1.5X-DD

9249

2.0X-DD

3333

2.5X-DD

8C63

3.0X-DD

71C7

4–46

Cache and External Interfaces

21264/EV68A Hardware Reference Manual

Bcache Port

Table 4–40 Program Values to Set the Cache Clock Period (Dual-Data Rate) (Continued)
Bcache
Transfer

BC_CLK_LD_VECTOR1

BC_BPHASE_LD_VECTOR1 BC_FDBK_EN1

3.5X-DD

C387

4.0X-DD

0F0F

5.0X-DD

7C1F

6.0X-DD

F03F

7.0X-DD

C07F

8.0X-DD

00FF

These are hexadecimal values.

In addition to programming the clock CSRs, the data-sample/drive Cbox CSRs, at the
pads, must be set appropriately. Table 4–41 lists these CSRs and provides their programmed value.
Table 4–41 Data-Sample/Drive Cbox CSRs
CBOX CSR

Description

BC_DDM_FALL_EN[0]

Enables the update of the 21264/EV68A’s Bcache outputs referenced to the
falling edge of the Bcache forwarded clock. Dual-data RAMs assert this
CSR.

BC_TAG_DDM_FALL_EN[0]

Enables the update of the 21264/EV68A’s Bcache tag outputs referenced to
the falling edge of the Bcache forwarded clock. Alway deasserted.

BC_DDM_RISE_EN[0]

Enables the update of the 21264/EV68A’s Bcache outputs referenced to the
rising edge of the Bcache forwarded clock. Always asserted.

BC_TAG_DDM_RISE_EN[0]

Enables the update of the 21264/EV68A’s Bcache tag outputs referenced to
the rising edge of the Bcache forwarded clock. Always asserted.

BC_DDMF_ENABLE[0]

Enables the rising edge of the Bcache forwarded clock. Always asserted.

BC_DDMR_ENABLE[0]

Enables the falling edge of the Bcache forwarded clock. Always asserted.

BC_FRM_CLK[0]

Forces the 21264/EV68A to only start Bcache transactions on the rising edge
of Bcache clocks that also coincide with the rising edge of GCLK. Must be
asserted for all dual-data parts and single-data parts at 2.5X and 3.5X.

BC_CLKFWD_ENABLE[0]

Enables clock forward enable. Always asserted.

4.8.3 Bcache Transactions
The Cbox uses the programmed clock values to start data read, tag read, data write, and
tag write transactions on the rising edge of a Bcache clock. The Cbox can also be configured to introduce a programmable number of bubbles when changing between write
and read commands. The following three sections describe these Bcache transactions.
4.8.3.1 Bcache Data Read and Tag Read Transactions

The 21264/EV68A always reads four pieces of data (64 bytes) from the Bcache during
a data read transaction, and always interrogates the tag array on the first cycle. Once
started, data read transactions are never cancelled. Assuming that the appropriate values

21264/EV68A Hardware Reference Manual

Cache and External Interfaces

4–47

Bcache Port

have been programmed for the Bcache clock period, and with satisfactory delay parameters for the SSRAM setup/hold Bcache address latch requirements, a Bcache read
command proceeds through the 21264/EV68A Cbox as follows:
1. When the 21264/EV68A clocks out the first address value on the Bcache index pins
with the appropriate Int_Add_BcClk value, the Cbox loads the values of Cbox CSR
BC_LAT_DATA_PATTERN[31:0] and Cbox CSR
BC_LAT_TAG_PATTERN[23:0] into two shift registers, which shift during every
GCLK cycle.
2. The address and control pins are latched into the SSRAMs. During the next cycle,
the SSRAMs provide data and tag information to the 21264/EV68A.
3. Using the returning forwarded clocks (BcDataInClk_H[7:0], BcTagInClk_H), the
data/tag information is loaded into the 21264/EV68A clock forwarding queue for
the Bcache.
4. Based on the value of BC_RCV_MUX_PRESET_CNT[1,0] (the unload pointer),
the result of a Bcache write command is loaded into a 21264/EV68A GCLK
(BPHASE) register.
5. The Cbox CSR BC_LAT_DATA_PATTERN[31:0] and
BC_LAT_TAG_PATTERN[23:0] contain the GCLK frequency at which the output
of the clock forward FIFO can be consumed by the processor. This provides GCLK
granularity for the Bcache interface, so that the 21264/EV68A can minimize
latency to the Bcache. When the values based on these Cbox CSRs are shifted
down to the bottom of the shift register, the processor samples the Bcache data and
delivers it to the consumers of load data in the 21264/EV68A functional units.
For example, when a 2.5X-SD SSRAM has a latency of eight GCLK cycles from
BcAdd_H[23:4] to the output of Bcache FIFO, Cbox CSR
BC_LAT_DATA_PATTERN[31:0] is programmed to 94816 and Cbox CSR
BC_LAT_TAG_PATTERN[23:0] is programmed to 816. The data pattern contains the
placement for four pieces of data and the aggregate rate of the data is 2.5X. In addition,
bit one of the BC_LAT_DATA_PATTERN is placed at a GCLK latency of six GCLK
cycles, which is the minimum latency supported by the 21264/EV68A. The
BC_LAT_TAG_PATTERN contains the placement of the tag data to the 21264/EV68A.
A shift of one to the left increases the latency of the Bcache transfer to nine GCLK
cycles, and a shift to the right reduces the latency of the Bcache transfer to seven GCLK
cycles.
The Cbox performs isolated tag read transactions in response to system probe commands. In addition, when using burst-mode SSRAMs, the Cbox can combine a separate
tag read transaction with the tail end of a data read transaction, thus optimizing Bcache
bandwidth. A Bcache tag read transaction proceeds exactly like a Bcache data read
transaction, except that only the BC_LAT_TAG PATTERN is used to update the tag
shift register.
4.8.3.2 Bcache Data Write Transactions

During a data write transaction, the 21264/EV68A always writes four pieces of data (64
bytes of data and 8 bytes of ECC) to the Bcache, and always writes the tag array during
the first cycle. Once started, data write operations are never cancelled. Given the appro-

4–48

Cache and External Interfaces

21264/EV68A Hardware Reference Manual

Bcache Port

priate programming of the Bcache clock period and delay parameters to satisfy SSRAM
setup/hold requirements of the Bcache address latch, a Bcache write transaction proceeds through the Cbox as follows:
1. The Cbox transmits the index and write control signals during an Int_Adr_BcClk
edge.
2. The data is placed on Bcache data, tag, and tag status pins on the appropriate
Int_Data_BcClk edge from 0 to 7 Bcache bit-times later, based on the Cbox CSR
BC_LATE_WRITE_NUM[2:0]. The BC_LATE_WRITE_NUM[2:0] supports the
late-write SSRAM, which optimize Bcache data bus bandwidth by minimizing
bubbles between read and write transactions. For example, single-data late-write
SSRAMs would need this CSR programmed to a value of one, and dual-data latewrite SSRAMs would need this CSR programmed to a value of two.
3. The difference between the data delivery (Int_Data_BcClk) and forwarded clocks
out provides the setup for the data at the Bcache data flip-flop.
4. For Bcache writes, the 21264/EV68A drivers are enabled on the GCLK BPHASE
preceding the start of a write transfer, and disabled on the succeeding GCLK
BPHASE at the end of a write transfer. Thus, the write data is enveloped by the
21264/EV68A drivers to guarantee that every data transfer has the same data-valid
window.
4.8.3.3 Bubbles on the Bcache Data Bus

When changing between read and write transactions on the bidirectional bus, it is often
necessary to introduce NOP cycles (bubbles) to allow the bus to settle and to drain the
Bcache read pipeline. The Cbox provides two CSRS, BC_RD_WR_BUBBLES[5:0]
and BC_WR_RD_BUBBLES[3:0], to help control the bubbles between read and write
transactions.
The optimum parameters for these CSRs are determined by formulas that include the
following terms:
Term

Description

bcfrm

Bcache frame clock.

•
•

In dual-data mode, bcfrm is twice the ratio.
In single-data mode, the value for bcfrm is determined by whether
the ratio is even or odd:
– When the ratio is even, bcfrm is equal to the ratio.
– When the ratio is odd, bcfrm is twice the ratio.

For example, in single-data mode:

GCLK

Ratio

Bcfrm

2
2.5

2
5

The processor clock.

21264/EV68A Hardware Reference Manual

Cache and External Interfaces

4–49

Bcache Port

Term

Description

Ratio

The number of GCLK cycles per peak Bcache bandwidth transfer. For example, a
ratio of 2.5 means the peak Bcache bandwidth is 16 bytes for every 2.5 GCLK
cycles.

rd_wr

The minimum spacing required between the read and write indices at the data/tag
pins, expressed as GCLK cycles.

wr_rd

The minimum spacing required between the write and read indices at the data/tag
pins, expressed as GCLK cycles.

The Relationship Between Write-to-Read — BC_WR_RD_BUBBLES and wr_rd

The following formulas calculate the relationship between the Cbox CSR
BC_WR_RD_BUBBLES and wr_rd:
wr_rd = (BC_WR_RD_BUBBLES – 1) * bcfrm

or
BC_WR_RD_BUBBLES = ((wr_rd + bcfrm – 1) / bcfrm) + 1

There is never a need to use a value of 0 or 1 for BC_WR_RD_BUBBLES.
If wr_rd = 4*ratio, then value 3 would be the minimum
BC_WR_RD_BUBBLES value when bcfrm = 2*ratio, and value 5 would be the
minimum BC_WR_RD_BUBBLES value when bcfrm = ratio.
There is a special case for ratio = 2.0 in single-data mode. In this case, the formula is:
wr_rd = (BC_WR_RD_BUBBLES – 2) * bcfrm

The Relationship Between Read-to-Write — BC_RD_WR_BUBBLES and rd_wr

Use the following formula to calculate the value for the Cbox CSR
BC_RD_WR_BUBBLES that produces the minimum rd_wr restriction:
BC_RD_WR_BUBBLES = rd_wr – 6

Note that a value for BC_RD_WR_BUBBLES of zero really means 64 GCLK cycles.
In that case, amend the formula. For example, it is impossible to have rd_wr = 6 in
the 1.5x dual-data rate mode case.

4.8.4 Pin Descriptions
This section describes the characteristics of the Bcache interface pins.

4–50

Cache and External Interfaces

21264/EV68A Hardware Reference Manual

Bcache Port

4.8.4.1 BcAdd_H[23:4]

The BcAdd_H[23:4] pins are high drive outputs that provides the index for the Bcache.
The 21264/EV68A supports Bcache sizes of 1MB, 2MB, 4MB, 8MB, and 16MB.
Table 4–42 lists the values to be programmed into Cbox CSRs BC_ENABLE[0] and
BC_SIZE[3:0] to support each size of the Bcache.
Table 4–42 Programming the Bcache to Support Each Size of the Bcache
BC_ENABLE[0]

BC_SIZE[3:0]

Bcache Size

0000

1MB

0001

2MB

0011

4MB

0111

8MB

1111

16MB

When the Cbox CSR BC_BANK_ENABLE[0] is not set, the unused BcAdd_H[23:4]
pins are tied to zero. For example, when configured as a 4MB cache, the 21264/EV68A
never changes BcAdd_H[23:22] from logic zero, and when BC_BANK_ENABLE[0]
is asserted, the 21264/EV68A drives the complement of the MSB index on the next
higher BcAdd_H pin.
4.8.4.2 Bcache Control Pins

The Bcache control pins (BcLoad_L, BcDataWr_L, BcDataOE_L, BcTagWr_L,
BcTagOE_L) are controlled using Cbox CSRs BC_BURST_MODE_ENABLE[0] and
BC_PENTIUM_MODE[0].
Table 4–43 shows the four combinations of Bcache control pin behavior obtained using
the two CSRs.
Table 4–43 Programming the Bcache Control Pins
BC_PENTIUM_MODE

BC_BURST_MODE_ENABLE

RAM_TYPE

RAM_TYPE A

RAM_TYPE B

Unsupported

Table 4–44 lists the combination of control pin assertion for RAM_TYPE A.
Table 4–44 Control Pin Assertion for RAM_TYPE A
TYPE_A

NOP

RA0

RA1

RA2

RA3

NOP

WA0 WA1

WA2

WA3

NOP

BcLoad_L

BcDataOE_L

BcDataWr_L

BcTagOE_L

BcTagWr_L

21264/EV68A Hardware Reference Manual

Cache and External Interfaces

4–51

Bcache Port

Table 4–45 lists the combination of control pin assertion for RAM_TYPE B.
Table 4–45 Control Pin Assertion for RAM_TYPE B
TYPE_B

NOP

RA0

RA1

RA2

RA3

NOP

WA0

WA1

WA2

WA3

NOP

BcLoad_L

BcDataOE_L

BcDataWr_L

BcTagOE_L

BcTagWr_L

Table 4–46 lists the combination of control pin assertion for RAM_TYPE C.
Table 4–46 Control Pin Assertion for RAM_TYPE C
TYPE_C

NOP RA0

RA1

RA2

RA3

NOP

WA0 WA1 WA2 WA3 NOP

BcLoad_L

BcDataOE_L

BcDataWr_L

BcTagOE_L

BcTagWr_L

Table 4–47 lists the combination of control pin assertion for RAM_TYPE D.
Table 4–47 Control Pin Assertion for RAM_TYPE D
TYPE_D

NOP

RA0

RA1

RA2

RA3 NOP

NOP

WA0

WA1

WA2

WA3

NOP

BcLoad_L

BcDataOE_L

BcDataWr_L

BcTagOE_L

BcTagWr_L

Notes:
1. The NOP condition for RAM_TYPE B is consistent with bursting nonPentium
style SSRAMs.
2. In both RAM_TYPE A and RAM_TYPE B, the pins BcDataOE_L and BcTagOE_L
function changes from output-enable control to chip-select control.
3. In both RAM_TYPE C and RAM_TYPE D SSRAMs, the pins BcDataOE_L and
BcTagOE_L function as an asynchronous output enable that envelopes the Bcache
read data by providing an extra cycle of output enable.
Using these Cbox CSRs, late-write nonbursting and dual-data rate SSRAMs can be
connected to the 21264/EV68A as described in Appendix E.

4–52

Cache and External Interfaces

21264/EV68A Hardware Reference Manual

Bcache Port

4.8.4.3 BcDataInClk_H and BcTagInClk_H

The BcDataInClk_H[7:0] and BcTagInClk_H pins are used to capture tag data and
data from the Bcache data and tag RAMs respectively. Dual-data rate SSRAMs provide
a clock output with the data output pins to minimize skew between the data and clock,
thus allowing maximum bandwidth. The 21264/EV68A internally synchronizes the
data to its GCLK with clock forward receive circuitry similar to that in the system interface. For nonDDR SSRAMs, systems can connect the Bcache data and tag output clock
pins to the Bcache data and tag input clock pins.

4.8.5 Bcache Banking
Bcache banking is possible by decoding the index MSB (as determined by Cbox CSR
BC_SIZE[3:0]) and asserting Cbox CSR BC_BANK_ENABLE[0]. To facilitate banking, the 21264/EV68A provides the complement of the MSB bit in the next higher
unused index bit. For example, when configured as an 8MB cache with banking
enabled, the 21264/EV68A drives the inversion of PA[22] on BcAdd_H[23] for use as
a chip enable in a banked configuration. Because there is no higher index bit available
for 16MB caches, this scheme only works for cache sizes of 1MB, 2MB, 4MB, and
8MB.
Setting BC_RD_RD_BUBBLE to 1 introduces one Bcache clock cycle of delay
between consecutive read transactions, regardless of whether or not they are read transactions to the same bank.
Setting BC_WR_WR_BUBBLE to 1 introduces one Bcache clock cycle of delay
between consecutive write transactions, regardless of whether or not they are write
transactions to the same bank.
Setting BC_SJ_BANK_ENABLE to 1 introduces one Bcache clock cycle of delay
between consecutive read transactions to a different bank (based on the MSB of the
index), even if BC_RD_RD_BUBBLE is set to 0. No additional delay is inserted
between consecutive read transactions to the same bank or between consecutive write
transactions.

4.8.6 Disabling the Bcache for Debugging
The Bcache is a required component for a 21264/EV68A-based system. However, for
debug purposes, the 21264/EV68A can be operated with the Bcache disabled. The
Bcache can be disabled by clearing all of the BC_ENABLE bits in the Cbox
WRITE_MANY CSR. When disabling the Bcache, the following additional steps must
be taken:
1. The various Bcache control bits in the Cbox WRITE_ONCE chain must be programmed to a valid combination (normally the same settings that would be used if
the Bcache were enabled).
2. The Bcache must still be initialized (using BC_INIT mode) during the reset PAL
flow, after which the Bcache should be left disabled.
3. Error Detection and Correction should be disabled by clearing DC_DAT_ERR_EN
(bit 7 of the DC_CTL IPR), or the following bits in the Cbox WRITE_ONCE chain
must be programmed to the indicated values:
BC_CLK_DELAY[1:0]
BC_CPU_CLK_DELAY[1:0]

21264/EV68A Hardware Reference Manual

= 0x1
= 0x1

Cache and External Interfaces

4–53

Interrupts

BC_CPU_LATE_WRITE_NUM[1:0] = 0x1
BC_LATE_WRITE_NUM[2:0]
= 0x0
BC_LATE_WRITE_UPPER
= 0
DUP_TAG_ENABLE
= 0

4.9 Interrupts
The system may request interrupts by way of the IRQ_H[5:0] pins. These six interrupt
sources are identical. They may be asynchronous, are level sensitive, and can be individually masked by way of the EIE field of the CM_IER IPR. The system designer
determines how these signals are used and selects their relative priority.

4–54

Cache and External Interfaces

21264/EV68A Hardware Reference Manual

5
Internal Processor Registers
This chapter describes 21264/EV68A internal processor registers (IPRs). They are separated into the following circuit logic groups: Ebox, Ibox, Mbox, and Cbox.
The gray areas in register figures indicate reserved fields. Bit ranges that are coupled
with the field name specify those bits in that named field that are included in the IPR.
For example, in Figure 5–2, the field named COUNTER[31:4] contains bits 31 through
4 of the COUNTER field from Section 5.1.1. The bit range of COUNTER[31:4] in the
IPR is also listed in the column Extent in Table 5–2. In many cases, such as this one, the
bit ranges correspond. However, the bit range of the named field need not always correspond to the Extent in the IPR. For example, in Figure 5–14, the field VA[47:13] resides
in IPR IVA_FORM[37:3] under the stated conditions.
The register contents after initialization are listed in Section 7.8.
Table 5–1 lists the 21264/EV68A internal processor registers.
Table 5–1 Internal Processor Registers
MT/MF
Issued
from Ebox
Access Pipe

Latency
for
MFPR
(Cycles)

Mnemonic

Index
(Binary)

ScoreBoard
Bit

Cycle counter

1100 0000

Cycle counter control

CC_CTL

1100 0001

—

Virtual address

1100 0010

4, 5, 6, 7

Virtual address control

VA_CTL

1100 0100

—

Virtual address format

VA_FORM

1100 0011

4, 5, 6, 7

ITB tag array write

ITB_TAG

0000 0000

—

ITB PTE array write

ITB_PTE

0000 0001

4, 0

—

ITB invalidate all process (ASM=0)

ITB_IAP

0000 0010

—

ITB invalidate all

ITB_IA

0000 0011

—

ITB invalidate single

ITB_IS

0000 0100

4, 6

—

ProfileMePC

PMPC

0000 0101

—

Exception address

EXC_ADDR

0000 0110

—

Ebox IPRs

Ibox IPRs

21264/EV68A Hardware Reference Manual

Internal Processor Registers

5–1

Table 5–1 Internal Processor Registers (Continued)
MT/MF
Issued
from Ebox
Access Pipe

Latency
for
MFPR
(Cycles)

Mnemonic

Index
(Binary)

ScoreBoard
Bit

Instruction VA format

IVA_FORM

0000 0111

Current mode

0000 1001

Interrupt enable

IER

0000 1010

Interrupt enable and current mode

IER_CM

0000 10xx

Software interrupt request

SIRR

0000 1100

Interrupt summary

ISUM

0000 1101

—

Hardware interrupt clear

HW_INT_CLR

0000 1110

—

Exception summary

EXC_SUM

0000 1111

—

PAL base address

PAL_BASE

0001 0000

Ibox control

I_CTL

0001 0001

Ibox status

I_STAT

0001 0110

Icache flush

IC_FLUSH

0001 0011

—

Icache flush ASM

IC_FLUSH_ASM

0001 0010

—

Clear virtual-to-physical map

CLR_MAP

0001 0101

4, 5, 6, 7

—

Sleep mode

SLEEP

0001 0111

4, 5, 6, 7

—

Process context register

PCTX

01xn nnnn

Process context register

PCTX

01xx xxxx

Performance counter control

PCTR_CTL

0001 0100

DTB tag array write 0

DTB_TAG0

0010 0000

2, 6

—

DTB tag array write 1

DTB_TAG1

1010 0000

1, 5

—

DTB PTE array write 0

DTB_PTE0

0010 0001

0, 4

—

DTB PTE array write 1

DTB_PTE1

1010 0001

3, 7

—

DTB alternate processor mode

DTB_ALTMODE

0010 0110

—

DTB invalidate all process (ASM = 0) DTB_IAP

1010 0010

—

DTB invalidate all

DTB_IA

1010 0011

—

DTB invalidate single (array 0)

DTB_IS0

0010 0100

—

DTB invalidate single (array 1)

DTB_IS1

1010 0100

—

DTB address space number 0

DTB_ASN0

0010 0101

—

DTB address space number 1

DTB_ASN1

1010 0101

—

Memory management status

MM_STAT

0010 0111

—

Mbox control

M_CTL

0010 1000

—

Dcache control

DC_CTL

0010 1001

—

Dcache status

DC_STAT

0010 1010

Mbox IPRs

5–2

Internal Processor Registers

21264/EV68A Hardware Reference Manual

Ebox IPRs

Table 5–1 Internal Processor Registers (Continued)
MT/MF
Issued
from Ebox
Access Pipe

Latency
for
MFPR
(Cycles)

Mnemonic

Index
(Binary)

ScoreBoard
Bit

Cbox data

C_DATA

0010 1011

Cbox shift control

C_SHFT

0010 1100

Cbox IPRs

1When n equals 1, that process context field is selected (FPE, PPCE, ASTRR, ASTER, ASN).

5.1 Ebox IPRs
This section describes the internal processor registers that control Ebox functions.

5.1.1 Cycle Counter Register – CC
The cycle counter register (CC) is a read-write register. The lower half of CC is a
counter that, when enabled by way of CC_CTL[32], increments once each CPU cycle.
The upper half of the register is 32 bits of register storage that may be used as a counter
offset as described in the Alpha Architecture Handbook, Version 4 under Processor Cycle
Counter (PCC) Register.
A HW_MTPR instruction to the CC writes the upper half of the register and leaves the
lower half unchanged. The RPCC instruction returns the full 64-bit value of the register.
Figure 5–1 shows the cycle counter register.
Figure 5–1 Cycle Counter Register
63

32 31

OFFSET
COUNTER

LK99-0008A

5.1.2 Cycle Counter Control Register – CC_CTL
The cycle counter control register (CC_CTL) is a write-only register through which the
lower half of the CC register may be written and its associated counter enabled and disabled. Figure 5–2 shows the cycle counter control register.
Figure 5–2 Cycle Counter Control Register
63

33 32 31

4 3

CC_ENA
COUNTER[31:4]

21264/EV68A Hardware Reference Manual

LK99-0009A

Internal Processor Registers

5–3

Ebox IPRs

Table 5–2 describes the CC_CTL register fields.
Table 5–2 Cycle Counter Control Register Fields Description
Name

Extent

Type

Description

Reserved

[63:33]

—

CC_ENA

[32]

Counter Enable.
When set, this bit allows the cycle counter to increment.

COUNTER[31:4]

[31:4]

CC[31:4] may be written by way of this field. Write transactions
to CC_CTL result in CC[3:0] being cleared.

Reserved

[3:0]

—

5.1.3 Virtual Address Register – VA
The virtual address register (VA) is a read-only register. When a DTB miss or fault
occurs, the associated effective virtual address is written into the VA register. VA is not
written when a LD_VPTE gets a DTB miss or Dstream fault. Figure 5–3 shows the virtual address register.
Figure 5–3 Virtual Address Register
63

VA[63:0]

LK99-0010A

5.1.4 Virtual Address Control Register – VA_CTL
The virtual address control register (VA_CTL) is a write-only register that controls the
way in which the faulting virtual address stored in the VA register is formatted when it
is read by way of the VA_FORM register. It also contains control bits that affect the
behavior of the memory pipe virtual address sign extension checkers and the behavior
of the Ebox extract, insert, and mask instructions. Figure 5–4 shows the virtual address
control register.
Figure 5–4 Virtual Address Control Register
63

30 29

3 2 1 0

VPTB[63:30]
VA_FORM_32
VA_48
B_ENDIAN
LK99-0014A

5–4

Internal Processor Registers

21264/EV68A Hardware Reference Manual

Ebox IPRs

Table 5–3 describes the virtual address control register fields.
Table 5–3 Virtual Address Control Register Fields Description
Name

Extent

Type

Description

VPTB[63:30]

[63:30]

Virtual Page Table Base.
See the VA_FORM register section for details.

Reserved

[29:3]

—

VA_FORM_32

[2]

This bit is used to control address formatting when reading the
VA_FORM register. See the section on the VA_FORM register for
details.

VA_48

[1]

WO,0

This bit controls the format applied to effective virtual addresses
by the VA_FORM register and the memory pipe virtual address
sign extension checkers. When VA_48 is clear, the 43-bit virtual
address format is used, and when VA_48 is set, the 48-bit virtual
address format is used.
When VA_48 is set, the sign extension checkers generate an
access control violation (ACV) if VA[63:0] ≠ SEXT (VA[47:0]).
When VA_48 is clear, the sign extension checkers generate an
ACV if VA[63:0] ≠ SEXT(VA[42:0]).

B_ENDIAN

[0]

Big Endian Mode.
When set, the shift amount (Rbv[2:0]) is inverted for EXTxx,
INSxx, and MSKxx instructions. The lower bits of the physical
address for Dstream accesses are inverted based upon the length
of the reference as follows:
Byte:
Invert bits [2:0]
Word:
Invert bits [2:1]
Longword:
Inverts bit [2]

5.1.5 Virtual Address Format Register – VA_FORM
The virtual address format register (VA_FORM) is a read-only register. It contains the
virtual page table entry address derived from the faulting virtual address stored in the
VA register. It also contains the virtual page table base and associated control bits stored
in the VA_CTL register.
Figure 5–5 shows VA_FORM when VA_CTL(VA_48) equals 0 and
VA_CTL(VA_FORM_32) equals 0.
Figure 5–5 Virtual Address Format Register (VA_48 = 0, VA_FORM_32 = 0)
63

33 32

3 2

VPTB[63:33]
VA[42:13]

LK99-0011A

Figure 5–6 shows VA_FORM when VA_CTL(VA_48) equals 1 and
VA_CTL(VA_FORM_32) equals 0.

21264/EV68A Hardware Reference Manual

Internal Processor Registers

5–5

Ibox IPRs

Figure 5–6 Virtual Address Format Register (VA_48 = 1, VA_FORM_32 = 0)
63

43 42

38 37

3 2

VPTB[63:43]
SEXT(VA[47])
VA[47:13]

LK99-0012A

Figure 5–7 shows VA_FORM when VA_CTL(VA_48) equals 0 and
VA_CTL(VA_FORM_32) equals 1.
Figure 5–7 Virtual Address Format Register (VA_48 = 0, VA_FORM_32 = 1)
63

30 29

22 21

3 2

VPTB[63:30]
VA[31:13]

LK99-0013A

5.2 Ibox IPRs
This section describes the internal processor registers that control Ibox functions.

5.2.1 ITB Tag Array Write Register – ITB_TAG
The ITB tag array write register (ITB_TAG) is a write-only register. The ITB tag array
is written by way of this register. A write transaction to ITB_TAG writes a register outside the ITB array. When a write to the ITB_PTE register is retired, the contents of both
the ITB_TAG and ITB_PTE registers are written into the ITB entry. The specific ITB
entry that is written is determined by a round-robin algorithm; the algorithm writes to
entry number 0 as the first entry after the 21264/EV68A is reset. Figure 5–8 shows the
ITB tag array write register.
Figure 5–8 ITB Tag Array Write Register
63

48 47

13 12

VA[47:13]

LK99-0015A

5.2.2 ITB PTE Array Write Register – ITB_PTE
The ITB PTE array write register (ITB_PTE) is a write-only register through which the
ITB PTE array is written. A round-robin allocation algorithm is used. A write to the
ITB_PTE array, when retired, results in both the ITB_TAG and ITB_PTE arrays being
written. The specific entry that is written is chosen by the round-robin algorithm
described above. Figure 5–9 shows the ITB PTE array write register.

5–6

Internal Processor Registers

21264/EV68A Hardware Reference Manual

Ibox IPRs

Figure 5–9 ITB PTE Array Write Register
63

44 43

13 12 11 10 9 8 7 6 5 4 3

PFN[43:13]
URE
SRE
ERE
KRE
GH[1:0]
ASM

LK99-0016A

5.2.3 ITB Invalidate All Process (ASM=0) Register – ITB_IAP
The ITB invalidate all process register (ITB_IAP) is a pseudo register that, when written to, invalidates all ITB entries whose ASM bit is clear. An explicit write to
IC_FLUSH_ASM is required to flush the Icache of blocks with ASM equal to zero.

5.2.4 ITB Invalidate All Register – ITB_IA
The ITB invalidate all register (ITB_IA) is a pseudo register that, when written to,
invalidates all ITB entries. An explicit write to IC_FLUSH is required to flush the
Icache.

5.2.5 ITB Invalidate Single Register – ITB_IS
The ITB invalidate single register (ITB_IS) is a write-only register. Writing a virtual
page number to this register invalidates any ITB entry that meets one of the following
criteria:

•

The ITB entry’s virtual page number matches ITB_IS[47:13] (or fewer bits if granularity hint bits are set in the ITB entry) and its ASN field matches the address
space number supplied in PCTX[46:39].

•

The ITB entry’s virtual page number matches ITB_IS[47:13] and its ASM bit is set.

Figure 5–10 shows the ITB invalidate single register.
Figure 5–10 ITB Invalidate Single Register
63

48 47

13 12

INVAL_ITB[47:13]

LK99-0017A

Note:

Because the Icache is virtually indexed and tagged, it is normally not necessary to flush the Icache when paging. Therefore, a write to ITB_IS will
not flush the Icache.

21264/EV68A Hardware Reference Manual

Internal Processor Registers

5–7

Ibox IPRs

5.2.6 ProfileMe PC Register – PMPC
The ProfileMe PC register (PMPC) is a read-only register that contains the PC of the
last profiled instruction. Additional information is available in the I_STAT and
PCTR_CTL register descriptions.
Usage of PMPC in performance monitoring is described in Section 6.10.
Figure 5–11 shows the ProfileMe PC register.
Figure 5–11 ProfileMe PC Register
63

2 1 0

PC[63:2]
PAL
LK99-0018A

Table 5–4 describes the ProfileMe PC register fields.
Table 5–4 ProfileMe PC Fields Description
Name

Extent

Type

Description

PC[63:2]

[63:2]

Address of the profiled instruction

Reserved

[1]

Read as zero

PAL

[0]

Indicates that the PC field contains a physical-mode PALmode
address

5.2.7 Exception Address Register – EXC_ADDR
The exception address register (EXC_ADDR) is a read-only register that is updated by
hardware when it encounters an exception or interrupt.
EXC_ADDR[0] is set if the associated exception occurred in PALmode. The exception
actions are listed here:

•

If the exception was a fault or a synchronous trap, EXC_ADDR contains the PC of
the instruction that triggered the fault or trap.

•

If the exception was an interrupt, EXC_ADDR contains the PC of the next instruction that would have executed if the interrupt had not occurred.

Figure 5–12 shows the exception address register.
Figure 5–12 Exception Address Register
63

2 1 0

PC[63:2]
PAL
LK99-0018A

5–8

Internal Processor Registers

21264/EV68A Hardware Reference Manual

Ibox IPRs

5.2.8 Instruction Virtual Address Format Register — IVA_FORM
The instruction virtual address format register (IVA_FORM) is a read-only register. It
contains the virtual PTE address derived from the faulting virtual address stored in the
EXC_ADDR register, and from the virtual page table base, VA_48 and VA_FORM_32
bits, stored in the I_CTL register.
Figure 5–13 shows IVA_FORM when I_CTL(VA_48) equals 0 and
I_CTL(VA_FORM_32) equals 0.
Figure 5–13 Instruction Virtual Address Format Register (VA_48 = 0, VA_FORM_32 = 0)
63

33 32

3 2

VPTB[63:33]
VA[42:13]

LK99-0019A

Figure 5–14 shows IVA_FORM when I_CTL(VA_48) equals 1 and
I_CTL(VA_FORM_32) equals 0.
Figure 5–14 Instruction Virtual Address Format Register (VA_48 = 1, VA_FORM_32 = 0)
63

43 42

38 37

3 2

VPTB[63:43]
SEXT(VA[47])
VA[47:13]

LK99-0020A

Figure 5–15 shows IVA_FORM when I_CTL(VA_48) equals 0 and
I_CTL(VA_FORM_32) equals 1.
Figure 5–15 Instruction Virtual Address Format Register (VA_48 = 0, VA_FORM_32 = 1)
63

30 29

22 21

3 2

VPTB[63:30]
VA[31:13]

LK99-0021A

5.2.9 Interrupt Enable and Current Processor Mode Register – IER_CM
The interrupt enable and current processor mode register (IER_CM) contains the interrupt enable and current processor mode bit fields. These bit fields can be written either
individually or together with a single HW_MTPR instruction. When bits [7:2] of the
IPR index field of a HW_MTPR instruction contain the value 0000102, this register is
selected. Bits [1:0] of the IPR index indicate which bit fields are to be written: bit[1]
corresponds to the IER field and bit[0] corresponds to the processor mode field. A
HW_MFPR instruction to this register returns the values in both fields. Figure 5–16
shows the interrupt enable and current processor mode register.

21264/EV68A Hardware Reference Manual

Internal Processor Registers

5–9

Ibox IPRs

Figure 5–16 Interrupt Enable and Current Processor Mode Register
63

39 38

33 32 31 30 29 28

14 13 12

5 4 3 2

EIEN[5:0]
SLEN
CREN
PCEN[1:0]
SIEN[15:1]
ASTEN
CM[1:0]

LK99-0022A

Table 5–5 describes the interrupt enable and current processor mode register fields.
Table 5–5 IER_CM Register Fields Description
Name

Extent

Type

Description

Reserved

[63:39]

—

EIEN[5:0]

[38:33]

External Interrupt Enable

SLEN

[32]

Serial Line Interrupt Enable

CREN

[31]

Corrected Read Error Interrupt Enable

PCEN[1:0]

[30:29]

Performance Counter Interrupt Enables

SIEN[15:1]

[28:14]

Software Interrupt Enables

ASTEN

[13]

AST Interrupt Enable
When set, enables those AST interrupt requests that are also
enabled by the value in ASTER.

Reserved

[12:5]

—

CM[1:0]

[4:3]

Current Mode
00
01
10
11

Reserved

[2:0]

—

Kernel
Executive
Supervisor
User

—

5.2.10 Software Interrupt Request Register – SIRR
The software interrupt request register (SIRR) is a read-write register containing bits to
request software interrupts. To generate a particular software interrupt, its corresponding bits in SIRR and IER[SIER] must both be set. Figure 5–17 shows the software
interrupt request register.

5–10

Internal Processor Registers

21264/EV68A Hardware Reference Manual

Ibox IPRs

Figure 5–17 Software Interrupt Request Register
63

29 28

14 13

SIR[15:1]

LK99-0023A

Table 5–6 describes the software interrupt request register fields.
Table 5–6 Software Interrupt Request Register Fields Description
Name

Extent

Type

Description

Reserved

[63:29]

—

SIR[15:1]

[28:14]

Software Interrupt Requests

Reserved

[13:0]

—

5.2.11 Interrupt Summary Register – ISUM
The interrupt summary register (ISUM) is a read-only register that records all pending
hardware, software, and AST interrupt requests that have their corresponding enable bit
set.
If a new interrupt (hardware, serial line, crd, or performance counters) occurs simultaneously with an ISUM read, the ISUM read returns zeros. That condition is normally
assumed to be a passive release condition. The interrupt is signaled again when the
PALcode returns to native mode. The effects of this condition can be minimized by
reading ISUM twice and ORing the results.
Usage of ISUM in performance monitoring is described in Section 6.10. Figure 5–18
shows the interrupt summary register.
Figure 5–18 Interrupt Summary Register
63

39 38

33 32 31 30 29 28

14 13

11 10 9 8

5 4 3 2

EI[5:0]
SL
CR
PC[1:0]
SI[15:1]
ASTU
ASTS
ASTE
ASTK

21264/EV68A Hardware Reference Manual

LK99-0024A

Internal Processor Registers

5–11

Ibox IPRs

Table 5–7 describes the interrupt summary register fields.
Table 5–7 Interrupt Summary Register Fields Description
Name

Extent

Type

Description

Reserved

[63:39]

—

EI[5:0]

[38:33]

External Interrupts

[32]

Serial Line Interrupt

[31]

Corrected Read Error Interrupts

PC[1:0]

[30:29]

Performance Counter Interrupts
PC0 when PC[0] is set.
PC1 when PC[1] is set.

SI[15:1]

[28:14]

Software Interrupts

Reserved

[13:11]

—

ASTU, ASTS

[10],[9]

AST Interrupts
For each processor mode, the bit is set if an associated AST
interrupt is pending. This includes the mode’s ASTER and
ASTRR bits and whether the processor mode value held in the
IER_CM register is greater than or equal to the value for the
mode.

Reserved

[8:5]

—

ASTE, ASTK

[4],[3]

Reserved

[2:0]

—

5.2.12 Hardware Interrupt Clear Register – HW_INT_CLR
The hardware interrupt clear register (HW_INT_CLR) is a write-only register used to
clear edge-sensitive interrupt requests. See Section D.31 for more information about the
PALcode restriction concerning this register. Figure 5–19 shows the hardware interrupt
clear register.
Figure 5–19 Hardware Interrupt Clear Register
63

33 32 31 30 29 28 27 26 25

SL
CR
PC[1:0]
MCHK_D
FBTP
LK99-0025A

5–12

Internal Processor Registers

21264/EV68A Hardware Reference Manual

Ibox IPRs

Table 5–8 describes the hardware interrupt clear register fields.
Table 5–8 Hardware Interrupt Clear Register Fields Description
Name

Extent

Type

Description

Reserved

[63:33]

—

[32]

W1C

Clears serial line interrupt request

[31]

W1C

Clears corrected read error interrupt request

PC[1:0]

[30:29]

W1C

Clears performance counter interrupt requests

MCHK_D

[28]

W1C

Clears Dstream machine check interrupt request

Reserved

[27]

—

FBTP

[26]

W1S

Forces the next Bcache hit that fills the Icache to generate bad
Icache fill parity

Reserved

[25:0]

—

5.2.13 Exception Summary Register – EXC_SUM
The exception summary register (EXC_SUM) is a read-only register that contains
information about instructions that have triggered traps. The register is updated at trap
delivery time. Its contents are valid only if it is read (by way of a HW_MFPR) in the
first fetch block of the exception handler. There are three types of traps for which this
register captures related information:

•

Arithmetic traps: The instruction generated an exceptional condition that should be
reported to the operating system, and/or the FPCR status bit associated with this
condition is clear and should be set by PALcode. Additionally, the REG field contains the register number of the destination specifier for the instruction that triggered the trap.

•

Istream ACV: The BAD_IVA bit of this register indicates whether the offending
Istream virtual address is latched into the EXC_ADDR register or the VA register.

•

Dstream exceptions: The REG field contains the register number of either the
source specifier (for stores) or the destination specifier (for loads) of the instruction
that triggered the trap.

Figure 5–20 shows the exception summary register.

21264/EV68A Hardware Reference Manual

Internal Processor Registers

5–13

Ibox IPRs

Figure 5–20 Exception Summary Register
63

48 47 46 45 44 43 42 41 40

14 13 12

8 7 6 5 4 3 2 1 0

SEXT(SET_IOV)
SET_IOV
SET_INE
SET_UNF
SET_OVF
SET_DZE
SET_INV
PC_OVFL
BAD_IVA
REG[4:0]
INT
IOV
INE
UNF
FOV
DZE
INV
SWC
LK99-0026A

Table 5–9 describes the exception summary register fields.
Table 5–9 Exception Summary Register Fields Description
Name

Extent

Type

Description

SEXT(SET_IOV)

[63:48]

RO, 0

Sign-extended value of bit 47, SET_IOV.

SET_IOV

[47]

PALcode should set FPCR[IOV].

SET_INE

[46]

PALcode should set FPCR[INE].

SET_UNF

[45]

PALcode should set FPCR[UNF].

SET_OVF

[44]

PALcode should set FPCR[OVF].

SET_DZE

[43]

PALcode should set FPCR[DZE].

SET_INV

[42]

PALcode should set FPCR[INV].

PC_OVFL

[41]

Indicates that EXC_ADDR was improperly sign extended for 48bit mode over/underflow IACV.

Reserved

[40:14]

RO, 0

Reserved for COMPAQ.

BAD_IVA

[13]

Bad Istream VA.
This bit should be used by the IACV PALcode routine to determine whether the offending I-stream virtual address is latched in
the EXC_ADDR register or the VA register. If BAD_IVA is clear,
EXC_ADDR contains the address; if BAD_IVA is set, VA contains the address.

5–14

Internal Processor Registers

21264/EV68A Hardware Reference Manual

Ibox IPRs

Table 5–9 Exception Summary Register Fields Description (Continued)
Name

Extent

Type

Description

REG[4:0]

[12:8]

Destination register of load or operate instruction that triggered
the trap OR source register of store that triggered the trap. These
bits may contain the Rc field of an operate instruction or the Ra
field of a load or store instruction. The value is UNPREDICTABLE
if the trap was triggered by an ITB miss, interrupt, OPCDEC, or
other non load/st/operate.

INT

[7]

Set to indicate Ebox integer overflow trap, clear to indicate Fbox
trap condition.

IOV

[6]

Indicates Fbox convert-to-integer overflow or Ebox integer overflow trap.

INE

[5]

Indicates floating-point inexact error trap.

UNF

[4]

Indicates floating-point underflow trap.

FOV

[3]

Indicates floating-point overflow trap.

DZE

[2]

Indicates divide by zero trap.

INV

[1]

Indicates invalid operation trap.

SWC

[0]

Indicates software completion possible. This bit is set if the
instruction that triggered the trap contained the /S modifier.

5.2.14 PAL Base Register – PAL_BASE
The PAL base register (PAL_BASE) is a read-write register that contains the base physical address for PALcode. Its contents are cleared by chip reset but are not cleared after
waking up from sleep mode or from fault reset. Figure 5–21 shows the PAL base register.
Figure 5–21 PAL Base Register
63

44 43

15 14

PAL_BASE[43:15]

LK99-0027A

Table 5–10 describes the PAL base register fields.
Table 5–10 PAL Base Register Fields Description
Name

Extent

Type

Description

Reserved

[63:44]

RO, 0

Reserved for COMPAQ.

PAL_BASE[43:15]

[43:15]

Base physical address for PALcode.

Reserved

[14:0]

RO, 0

Reserved for COMPAQ.

5.2.15 Ibox Control Register – I_CTL
The Ibox control register (I_CTL) is a read-write register that controls various Ibox
functions. Its contents are cleared by chip reset. Figure 5–22 shows the Ibox control
register.
21264/EV68A Hardware Reference Manual

Internal Processor Registers

5–15

Ibox IPRs

Figure 5–22 Ibox Control Register
63

48 47

30 29

24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 9 8 7 6 5

3 2 1 0

SEXT(VPTB[47])
VPTB[47:30]
CHIP_ID[5:0]
BIST_FAIL
TB_MB_EN
MCHK_EN
ST_WAIT_64K
PCT1_EN
PCT0_EN
SINGLE_ISSUE_H
VA_FORM_32
VA_48
SL_RCV
SL_XMIT
HWE
BP_MODE[1:0]
SBE[1:0]
SDE[1:0]
SPE[2:0]
IC_EN[1:0]
SPCE
LK99-0029A

Table 5–11 describes the Ibox control register fields.
Table 5–11 Ibox Control Register Fields Description
Name

Extent

Type

Description

SEXT(VPTB[47])

[63:48]

RW,0

Sign extended VPTB[47].

VPTB[47:30]

[47:30]

RW,0

Virtual Page Table Base. See Section 5.1.5 for details.

CHIP_ID[5:0]

[29:24]

This is a read-only field that supplies the revision ID number
for the 21264/EV68A part.
21264/EV68A pass 2.1 ID is 010010 2.

BIST_FAIL

[23]

RO,0

Indicates the status of BiST (clear = pass, set = fail),
described in Section 11.5.1.

TB_MB_EN

[22]

RW,0

When set, the hardware ensures that the virtual-mode loads
in DTB and ITB fill flows that access the page table and the
subsequent virtual mode load or store that is being retried are
‘ordered’ relative to another processor’s stores. This must be
set for multiprocessor systems in which no MB instruction is
present in the TB fill flow, unless there are other mechanisms present that ensure coherency.

MCHK_EN

[21]

RW,0

Machine check enable — set to enable machine checks.

5–16

Internal Processor Registers

21264/EV68A Hardware Reference Manual

Ibox IPRs

Table 5–11 Ibox Control Register Fields Description (Continued)
Name

Extent

Type

Description

ST_WAIT_64K

[20]

RW,0

The stWait table is used to reduce load/store order traps.
When set, the stWait table is cleared after 64K cycles. When
clear, the stWait table is cleared after 16K cycles. See Section 2.11.

PCT1_EN

[19]

RW,0

Enable performance counter #1. If this bit is one, the performance counter will count if either the system (SPCE) or process (PPCE) performance counter enable is asserted.

PCT0_EN

[18]

RW,0

Enable performance counter #0. If this bit is one, the performance counter will count if EITHER the system (SPCE) or
process (PPCE) performance counter enable is set.

SINGLE_ISSUE_H

[17]

RW,0

When set, this bit forces instructions to issue only from the
bottom-most entries of the IQ and FQ.

VA_FORM_32

[16]

RW,0

This bit controls address formatting on a read of the
IVA_FORM register.

VA_48

[15]

RW,0

This bit controls the format applied to effective virtual
addresses by the IVA_FORM register and the Ibox virtual
address sign extension checkers. When VA_48 is clear, 43bit virtual address format is used, and when VA_48 is set,
48-bit virtual address format is used. The effect of this bit on
the IVA_FORM register is identical to the effect of
VA_CTL[VA_48] on the VA_FORM register. See Section
5.1.5.
When VA_48 is set, the sign extension checkers generate an
ACV if va[63:0] ≠ SEXT(va[47:0]). When VA_48 is clear,
the sign extension checkers generate an ACV if va[63:0] ≠
SEXT(va[42:0]).
This bit also affects DTB_DOUBLE traps. If set, the DTB
double miss traps vector to the DTB_DOUBLE_4 entry
point.
DTB_DOUBLE PALcode flow selection is not affected by
VA_CTL[VA_48].

SL_RCV

[14]

See Section 11.2.

SL_XMIT

[13]

When set, drives a value on SromClk_H. See Section 11.2.

HWE

[12]

RW,0

If set, allow PALRES intructions to be executed in kernel
mode. Note that modification of the ITB while in kernel
mode/native mode may cause UNPREDICTABLE behavior.

BP_MODE[1:0]

[11:10]

RW,0

Branch Prediction Mode Selection.
BP_MODE[1], if set, forces all branches to be predicted to
fall through. If clear, the dynamic branch predictor is chosen.
BP_MODE[0]. If set, the dynamic branch predictor chooses
local history prediction. If clear, the dynamic branch predictor chooses local or global prediction based on the state of
the chooser.

SBE[1:0]

[9:8]

RW,0

Stream Buffer Enable.
The value in this bit field specifies the number of Istream
buffer prefetches (besides the demand-fill) that are launched
after an Icache miss. If the value is zero, only demand
requests are launched.

21264/EV68A Hardware Reference Manual

Internal Processor Registers

5–17

Ibox IPRs

Table 5–11 Ibox Control Register Fields Description (Continued)
Name

Extent

Type

Description

SDE[1:0]

[7:6]

RW,0

PALshadow Register Enable.
Enables access to the PALshadow registers. If SDE[1] is set,
R4-R7 and R20-R23 are used as PALshadow registers.
SDE[0] does not affect 21264/EV68A operation.

SPE[2:0]

[5:3]

RW,0

Super Page Mode Enable.
Identical to the SPE bits in the Mbox M_CTL SPE[2:0]. See
Section 5.3.9.

IC_EN[1:0]

[2:1]

RW,3

Icache Set Enable.
At least one set must be enabled. The entire cache may be
enabled by setting both bits. Zero, one, or two Icache sets
can be enabled.
This bit does not clear the Icache, but only disables fills to
the affected set.

SPCE

[0]

RW,0

System Performance Counting Enable.
Enables performance counting for the entire system if individual counters (PCTR0 or PCTR1) are enabled by setting
PCT0_EN or PCT1_EN, respectively.
Performance counting for individual processes can be
enabled by setting PCTX[PPCE]. See Section 5.2.21 for
more information.
See Section 6.10 for information about performance counting.

5.2.16 Ibox Status Register – I_STAT
The Ibox status register (I_STAT) is a read/write-1-to-clear register that contains Ibox
status information.
Usage of I_STAT in performance monitoring is described in Section 6.10.
Figure 5–23 shows the Ibox status register.

5–18

Internal Processor Registers

21264/EV68A Hardware Reference Manual

Ibox IPRs

Figure 5–23 Ibox Status Register
63

41 40 39 38 37

34 33 32

30 29 28 27

Reserved
MIS
TRP
LS0
TRAP TYPE[3:0]
ICM
OVR[2:0]
PAR
LAM
Reserved

LK99-0031A

Table 5–12 describes the Ibox status register fields.
Table 5–12 Ibox Status Register Fields Description
Name

Extent

Type

Description

Reserved

[63:41]

Reserved for COMPAQ.

MIS

[40]

ProfileMe Mispredict Trap.
If the I_STAT[TRP] bit is set, this bit indicates that the profiled instruction caused a mispredict trap. JSR/JMP/RET/COR or HW_JSR/
HW_JMP/HW_RET/HW_COR mispredicts do not set this bit but can be
recognized by the presence of one of these instructions at the PMPC location with the I_STAT[TRP] bit set. This identification is exact in all cases
except error condition traps. Hardware corrected Icache parity or Dcache
ECC errors, and machine check traps can occur on any instruction in the
pipeline.

TRP

[39]

ProfileMe Trap.
This bit indicates that the profiled instruction caused a trap. The trap type
field, PMPC register, and instruction at the PMPC location are needed to
distinguish all trap types.

LS0

[38]

ProfileMe Load-Store Order Trap.
If the profiled instruction caused a replay trap, this bit indicates that the
precise trap cause was an Mbox load-store order replay trap.
If clear, this bit indicates that the replay trap was any one of the following:
Mbox load-load order
Mbox load queue full
Mbox store queue full
Mbox wrong size trap (such as, STL → LDQ)
Mbox Bcache alias (2 physical addresses map to same Bcache line)
Mbox Dcache alias (2 physical addresses map to same Dcache line)
Icache parity error
Dcache ECC error

21264/EV68A Hardware Reference Manual

Internal Processor Registers

5–19

Ibox IPRs

Table 5–12 Ibox Status Register Fields Description (Continued)
Name

Extent

Type

Description

TRAP
TYPE[3:0]

[37:34]

ProfileMe Trap Types.
If the profiled instruction caused a trap (indicated by I_STAT[TRP]), this
field indicates the trap type as listed here:
Value
Trap Type
0
Replay
1
Invalid (unused)
2
DTB Double miss (3 level page tables)
3
DTB Double miss (4 level page tables)
4
Floating point disabled
5
Unaligned Load/Store
6
DTB Single miss
7
Dstream Fault
8
OPCDEC
9
Invalid (use PMPC, described below)
10
Machine Check
11
Invalid (use PMPC, described below)
12
Arithmetic
13
Invalid (use PMPC, described below)
14
MT_FPCR
15
Reset
Traps due to ITB miss, Istream access violation, or interrupts are not
reported in the trap type field because they do not cause pipeline aborts.
Instead, these traps cause pipeline redirection and can be distinguished by
examining the PMPC value for the presence of the corresponding PALcode entry offset addresses indicated below. In these cases, the ProfileMe
interrupt will normally be delivered when exiting the trap PALcode flow
and the EXC_ADDR register will contain the original PC that encountered the redirect trap.
PMPC[14:0]
Trap
0581
ITB miss
0481
Istream Access Violation
0681
Interrupt

ICM

[33]

ProfileMe Icache Miss.
This bit indicates that the profiled instruction was contained in an aligned
4-instruction Icache fetch block that requested a new Icache fill stream.

OVR[2:0]

[32:30]

ProfileMe Counter 0 Overcount.
This bit indicates a value (0-7) that must be subtracted from the counter 0
result to obtain an accurate count of the number of instructions retired in
the interval beginning three cycles after the profiled instruction reaches
pipeline stage 2 and ending four cycles after the profiled instruction is
retired.

PAR

[29]

W1C

Icache Parity Error.
This bit indicates that the Icache encountered a parity error on instruction
fetch. When a parity error is detected, the Icache is flushed, a replay trap
back to the address of the error instruction is generated, and a correctable
read interrupt is requested. See also I_STAT[LAM].

LAM

[28]

When set, indicates that an error in the line predictor set I_STAT[PAR].
I_STAT[LAM] is set only when I_STAT[PAR] is set; is subsequently
cleared when I_STAT[PAR] is cleared (when a 1 is written to
I_STAT[PAR] ).

Reserved

[28:0]

Reserved for COMPAQ.

5–20

Internal Processor Registers

21264/EV68A Hardware Reference Manual

Ibox IPRs

5.2.17 Icache Flush Register – IC_FLUSH
The Icache flush register (IC_FLUSH) is a pseudo register. Writing to this register
invalidates all Icache blocks. The cache is flushed when the next HW_RET/STALL
instruction is retired. See Section D.20 for more information.

5.2.18 Icache Flush ASM Register – IC_FLUSH_ASM
The Icache flush ASM register (IC_FLUSH_ASM) is a pseudo register. Writing to this
register invalidates all Icache blocks with their ASM bit clear.

5.2.19 Clear Virtual-to-Physical Map Register – CLR_MAP
The clear virtual-to-physical map register (CLR_MAP) is a pseudo register that, when
written, results in the clearing of the current map of virtual to physical registers. This
register must only be written after there are no register-borne dependencies present and
there are no unretired instructions. See an example in the PALcode restrictions.

5.2.20 Sleep Mode Register – SLEEP
The sleep mode register (SLEEP) is a pseudo register that, when written, results in the
PLL speed being reduced and the chip entering a low-power mode. This register must
only be written after a sequence of code has been run which saves all necessary state to
DRAM, flushes the caches, and unmasks certain interrupts so the chip can be woken up.
See Section 7.3 for details.

5.2.21 Process Context Register – PCTX
The process context register (PCTX) contains information associated with the context
of a process. Any combination of the bit fields within this register may be written with
a single HW_MTPR instruction. When bits [7:6] of the IPR index field of a
HW_MTPR instruction contain the value 012, this register is selected. Bits [4:0] of the
IPR index indicate which bit fields are to be written. Usage of PCTX in performance
monitoring is described in Section 6.10.
Table 5–13 lists the correspondence between IPR index bits and register fields.
Table 5–13 IPR Index Bits and Register Fields
IPR Index Bit

ASN

ASTER

ASTRR

PPCE

FPE

A HW_MFPR from this register returns the values in all of its component bit fields.
Figure 5–24 shows the process context register.

21264/EV68A Hardware Reference Manual

Internal Processor Registers

5–21

Ibox IPRs

Figure 5–24 Process Context Register
63

47 46

39 38

13 12

9 8

5 4 3 2 1 0

ASN[7:0]
ASTRR[3:0]
ASTER[3:0]
FPE
PPCE
LK99-0032A

Table 5–14 describes the process context register fields.
Table 5–14 Process Context Register Fields Description
Name

Extent

Type

Description

Reserved

[63:47]

—

ASN[7:0]

[46:39]

Address space number.

Reserved

[38:13]

—

ASTRR[3:0]

[12:9]

AST request register—used to request AST interrupts in
each of the four processor modes.
To generate a particular AST interrupt, its corresponding
bits in ASTRR and ASTER must be set, along with the
ASTE bit in IER.
Further, the value of the current mode bits in the PS register
must be equal to or higher than the value of the mode associated with the AST request.
The bit order with this field is:
User Mode
12
Supervisor Mode
11
Executive Mode
10
Kernel Mode
9

ASTER[3:0]

[8:5]

AST enable register—used to individually enable each of
the four AST interrupt requests.
The bit order with this field is:
User Mode
8
Supervisor Mode
7
Executive Mode
6
Kernel Mode
5

Reserved

[4:3]

—

5–22

Internal Processor Registers

21264/EV68A Hardware Reference Manual

Ibox IPRs

Table 5–14 Process Context Register Fields Description (Continued)
Name

Extent

Type

Description

FPE

[2]

RW,1

Floating-point enable—if clear, floating-point instructions
generate FEN exceptions. This bit is set by hardware on
reset.

PPCE

[1]

Process performance counting enable.
Enables performance counting for an individual process
with counters PCTR0 or PCTR1, which are enabled by setting PCT0_EN or PCT1_EN, respectively.
Performance counting for the entire system can be enabled
by setting I_CTL[SPCE]. See Section 5.2.15 for more information.
See Section 6.10 for information about performance counting.

Reserved

[0]

—

5.2.22 Performance Counter Control Register – PCTR_CTL
The performance counter control register (PCTR_CTL) is a read-write register that
controls the function of the performance counters for either aggregate counting or ProfileMe sampling counting.
Usage of PCTR_CTL in performance monitoring is described in Section 6.10.
Figure 5–25 shows the performance counter control register.
Figure 5–25 Performance Counter Control Register
63

48 47

28 27 26 25

6 5 4 3 2 1 0

SEXT(PCTR0_CTL[47])
PCTR0[19:0]
PM_STALLED
PM_KILLED_BM
PCTR1[19:0]
SL0
SL1[1:0]
VAL
TAK

LK99-0034A

Table 5–15 describes the performance counter control register fields.
Table 5–15 Performance Counter Control Register Fields Description
Name

Extent Type Description

SEXT(PCTR0_CTL[47]) [63:48] RO

When read, this field is sign extended from PCTR_CTL[47]. Writes
to this field are ignored.

21264/EV68A Hardware Reference Manual

Internal Processor Registers

5–23

Ibox IPRs

Table 5–15 Performance Counter Control Register Fields Description (Continued)
Name

Extent Type Description

PCTR0[19:0]

[47:28] RW

Performance counter 0.
PCTR0 is enabled by I_CTL[PCT0_EN] and either I_CTL[SPCE] or
PCTX[PPCE].
In Aggregate mode:
When enabled, PCTR0 is incremented at each cycle by the selected
input. (See Section 6.10.2 for more information.)
On overflow, if enabled by IER_CM[PCEN0],
ISUM[PC0] is set and an interrupt is triggered.
In ProfileMe mode:
On overflow, a count window is opened and PCTR0 is incremented
as described in Section 6.10.3. When the count window overflows, if
enabled by IER_CM[PCEN0], ISUM[PC0] is set and an interrupt is
triggered.
See Table 5–16 for counter modes.

PM_STALLED

[27]

The profiled instruction stalled for at least one cycle between the
fetch and map stages of the pipeline.

PM_KILLED_BM

[26]

The profiled instruction was killed during or before the cycle in
which it was mapped.

PCTR1[19:0]

[25:6]

Performance counter 1.
PCTR1 is enabled by I_CTL[PCT1_EN] and either I_CTL[SPCE] or
PCTX[PPCE].
In Aggregate mode:
When enabled, PCTR1 is incremented at each cycle by the selected
input. (See Section 6.10.2 for more information.)
On overflow, if enabled by IER_CM[PCEN1], ISUM[PC1] is set and
an interrupt is triggered.
In ProfileMe mode, how PCTR1 is incremented is described in Section 6.10.3.
In either case, PCTR1 is incremented no more than 1 per cycle.
See Table 5–16 for counter modes.

Reserved

[5]

Reads to this field return zero. Writes to this field are ignored.

SL0

[4]

Selector 0.
0 = Aggregate counting mode
1 = ProfileMe mode
See Table 5–16 for more information.

SL1[1:0]

[3:2]

Selector 1.
Selects counter PCTR0 and PCTR1 modes. See Table 5–16 for more
information.

5–24

Internal Processor Registers

21264/EV68A Hardware Reference Manual

Mbox IPRs

Table 5–15 Performance Counter Control Register Fields Description (Continued)
Name

Extent Type Description

VAL

[1]

Profiled instruction valid.
When set, indicates a nontrapping profiled instruction retired valid.
When clear, indicates that a nontrapping profiled instruction was
killed after the cycle in which it was mapped. Valid retire/abort status
for a trapping profiled instruction is determined by the trap type (see
I_STAT[TRAP_TYPE]).

TAK

[0]

ProfileMe conditional branch taken.
Indicates program branch direction, if the profiled instruction is a
conditional branch.

Table 5–16 Performance Counter Control Register Input Select Fields
SL0[4]

SL1[3:2]

Mode

PCTR0

PCTR1

Aggregate

Retired instructions

Cycle counting

Aggregate

Cycle counting

Not defined

Aggregate

Retired instructions

Bcache miss or long latency probes

Aggregate

Cycle counting

Mbox replay traps

ProfileMe

Retired instructions

Cycle counting

ProfileMe

Cycle counting

Inum retire delay

ProfileMe

Retired instructions

Bcache miss or long latency probes

ProfileMe

Cycle counting

Mbox replay traps

5.3 Mbox IPRs
This section describes the internal processor registers that control Mbox functions.

5.3.1 DTB Tag Array Write Registers 0 and 1 – DTB_TAG0, DTB_TAG1
The DTB tag array write registers 0 and 1 (DTB_TAG0 and DTB_TAG1) are writeonly registers through which the two memory pipe DTB tag arrays are written. Write
transactions to DTB_TAG0 and DTB_TAG1 write data to registers outside the DTB
arrays. When write transactions to the corresponding DTB_PTE registers are retired,
the contents of both the DTB_TAG and DTB_PTE registers are written into their
respective DTB arrays, at locations determined by the round-robin allocation algorithm.
Figure 5–26 shows the DTB tag array write registers 0 and 1.
Figure 5–26 DTB Tag Array Write Registers 0 and 1
63

48 47

VA[47:13]

21264/EV68A Hardware Reference Manual

13 12

LK99-0035A

Internal Processor Registers

5–25

Mbox IPRs

5.3.2 DTB PTE Array Write Registers 0 and 1 – DTB_PTE0, DTB_PTE1
The DTB PTE array write registers 0 and 1 (DTB_PTE0 and DTB_PTE1) are registers
through which the DTB PTE arrays are written. The entries to be written are chosen by
a round-robin allocation scheme. Write transactions to the DTB_PTE registers, when
retired, result in both the DTB_TAG and DTB_PTE arrays being written. Figure 5–27
shows the DTB PTE array write registers 0 and 1.
Figure 5–27 DTB PTE Array Write Registers 0 and 1
63 62

32 31

16 15 14 13 12 11 10 9 8 7 6 5 4 3 2 1 0

PA[43:13]
UWE
SWE
EWE
KWE
URE
SRE
ERE
KRE
GH[1:0]
ASM
FOW
FOR
LK99-0036A

5.3.3 DTB Alternate Processor Mode Register – DTB_ALTMODE
The DTB alternate processor mode register (DTB_ALTMODE) is a write-only register
whose contents specify the alternate processor mode used by some HW_LD and
HW_ST instructions. Figure 5–28 shows the DTB alternate processor mode register.
Figure 5–28 DTB Alternate Processor Mode Register
63

2 1 0

ALT_MODE[1:0]
LK99-0037A

Table 5–17 describes the DTB_ALTMODE register fields.
Table 5–17 DTB Alternate Processor Mode Register Fields Description
Name

Extent

Type

Description

Reserved

[63:2]

—

5–26

Internal Processor Registers

21264/EV68A Hardware Reference Manual

Mbox IPRs

Table 5–17 DTB Alternate Processor Mode Register Fields Description (Continued)
Name

Extent

Type

Description

ALT_MODE[1:0]

[1:0]

Alt_Mode:
ALT_MODE[1:0]
00
01
10
11

Mode
Kernel
Executive
Supervisor
User

5.3.4 Dstream TB Invalidate All Process (ASM=0) Register – DTB_IAP
The Dstream translation buffer invalidate all process (ASM=0) register (DTB_IAP) is a
write-only pseudo register. Write transactions to this register invalidate all DTB entries
in which the address space match (ASM) bit is clear.

5.3.5 Dstream TB Invalidate All Register – DTB_IA
The Dstream translation buffer invalidate all register (DTB_IA) is a write-only pseudo
register. Write transactions to this register invalidate all DTB entries and reset the DTB
not-last-used pointer to its initial state.

5.3.6 Dstream TB Invalidate Single Registers 0 and 1 – DTB_IS0,1
The Dstream translation buffer invalidate single registers (DTB_IS0 and DTB_IS1) are
write-only pseudo registers through which software may invalidate a single entry in the
DTB arrays. Writing a virtual page number to one of these registers invalidates any
DTB entry in the corresponding memory pipeline which meets one of the following criteria:

•

The DTB entry’s virtual page number matches DTB_IS[47:13] and its ASN field
matches DTB_ASN[63:56].

•

The DTB entry’s virtual page number matches DTB_IS[47:13] and its ASM bit is
set.

Figure 5–29 shows the Dstream translation buffer invalidate single registers.
Figure 5–29 Dstream Translation Buffer Invalidate Single Registers
63

48 47

VA[47:13]

21264/EV68A Hardware Reference Manual

13 12

LK99-0015A

Internal Processor Registers

5–27

Mbox IPRs

5.3.7 Dstream TB Address Space Number Registers 0 and 1 – DTB_ASN0,1
The Dstream translation buffer address space number registers (DTB_ASN0 and
DTB_ASN1) are write-only registers that should be written with the address space
number (ASN) of the current process. Figure 5–30 shows the Dstream translation buffer
address space number registers 0 and 1.
Figure 5–30 Dstream Translation Buffer Address Space Number Registers 0 and 1
63

56 55

ASN[7:0]
LK99-0038A

5.3.8 Memory Management Status Register – MM_STAT
The memory management status register (MM_STAT) is a read-only register.
When a Dstream TB miss or fault occurs, information about the error is latched in
MM_STAT. MM_STAT is not updated when a LD_VPTE gets a DTB miss instruction.
Figure 5–31 shows the memory management status register.
Figure 5–31 Memory Management Status Register
63

11 10 9

4 3 2 1 0

DC_TAG_PERR
OPCODE[5:0]
FOW
FOR
ACV
WR

LK99-0039A

Table 5–18 describes the memory management status register fields.
Table 5–18 Memory Management Status Register Fields Description
Name

Extent

Type

Description

Reserved

[63:11]

—

DC_TAG_PERR

[10]

This bit is set when a Dcache tag parity error occurred during the
initial tag probe of a load or store instruction. The error created a
synchronous fault to the D_FAULT PALcode entry point and is
correctable. The virtual address associated with the error is available in the VA register.

OPCODE[5:0]

[9:4]

Opcode of the instruction that caused the error.
HW_LD is displayed as 3 and HW_ST is displayed as 7.

FOW

[3]

This bit is set when a fault-on-write error occurs during a write
transaction and PTE[FOW] was set.

5–28

Internal Processor Registers

21264/EV68A Hardware Reference Manual

Mbox IPRs

Table 5–18 Memory Management Status Register Fields Description (Continued)
Name

Extent

Type

Description

FOR

[2]

This bit is set when a fault-on-read error occurs during a read
transaction and PTE[FOR] was set.

ACV

[1]

This bit is set when an access violation occurs during a transaction. Access violations include a bad virtual address.

[0]

This bit is set when an error occurs during a write transaction.

Note:

The Ra field of the instruction that triggered the error can be obtained from
the Ibox EXC_SUM register.

5.3.9 Mbox Control Register – M_CTL
The Mbox control register (M_CTL) is a write-only register. Its contents are cleared by
chip reset. Figure 5–32 shows the Mbox control register.
Figure 5–32 Mbox Control Register
63

6 5 4 3

1 0

SMC[1:0]
SPE[2:0]
LK99-0040A

21264/EV68A Hardware Reference Manual

Internal Processor Registers

5–29

Mbox IPRs

Table 5–19 describes the Mbox control register fields.
Table 5–19 Mbox Control Register Fields Description
Name

Extent

Type

Description

Reserved

[63:6]

—

SMC[1:0]

[5:4]

WO,0

Speculative miss control (see Section 4.6.4).
Bits

Meaning When Set

00
01

Allow full-time speculation.
Force full-time conservative mode. Make retries wait until retire,
force all new stores that do not hit dirty to retry, and cause prefetches
with modify intent (see Section 2.6.2) to behave like normal
prefetches.
Place 21264/EV68A in periodic conservative mode by using an 8-bit
counter to add by 4 each time a branch mispredict happens and subtract by one each time a conditional branch retires. Enter conservative mode if the MSB of the counter is set.
Place 21264/EV68A in periodic conservative mode by using an 8-bit
countner to add by 8 each time a branch mispredict happens and subtract by one each time a conditional branch retires. Enter conservative mode if the MSB of the counter is set.

SPE[2:0]

[3:1]

WO,0

Superpage mode enables.
SPE[2], when set, enables superpage mapping when VA[47:46] = 2. In this
mode, VA[43:13] are mapped directly to PA[43:13] and VA[45:44] are
ignored.
SPE[1], when set, enables superpage mapping when VA[47:41] = 7E 16. In
this mode, VA[40:13] are mapped directly to PA[40:13] and PA[43:41] are
copies of PA[40] (sign extension).
SPE[0], when set, enables superpage mapping when VA[47:30] = 3FFFE 16.
In this mode, VA[29:13] are mapped directly to PA[29:13] and PA[43:30] are
cleared.

Reserved

[0]

—

Note:

Superpage accesses are only allowed in kernel mode. Non-kernel mode references to superpages result in access violations.

5.3.10 Dcache Control Register – DC_CTL
The Dcache control register (DC_CTL) is a write-only register that controls Dcache
activity. The contents of DC_CTL are initialized by chip reset as indicated. Figure 5–33
shows the Dcache control register.

5–30

Internal Processor Registers

21264/EV68A Hardware Reference Manual

Mbox IPRs

Figure 5–33 Dcache Control Register
63

8 7 6 5 4 3 2 1 0

DCDAT_ERR_EN
DCTAG_PAR_EN
F_BAD_DECC
F_BAD_TPAR
F_HIT
SET_EN[1:0]

LK99-0041A

Table 5–20 describes the Dcache control register fields.
Table 5–20 Dcache Control Register Fields Description
Name

Extent

Type

Description

Reserved

[63:8]

—

DCDAT_ERR_EN

[7]

WO,0

Dcache data ECC and parity error enable.

DCTAG_PAR_EN

[6]

WO,0

Dcache tag parity enable.

F_BAD_DECC

[5]

WO,0

Force Bad Data ECC. When set, ECC data is not written into
the cache along with the block that is loaded by a fill or store.
Writing data that is different from that already in the block will
cause bad ECC to be present. Since the old ECC value will
remain, the ECC will be bad.

F_BAD_TPAR

[4]

WO,0

Force Bad Tag Parity. When set, this bit causes bad tag parity to
be put into the Dcache tag array during Dcache fill operations.

Reserved

[3]

—

F_HIT

[2]

WO,0

Force Hit. When set, this bit causes all memory space load and
store instructions to hit in the Dcache, independent of the
Dcache tag address compare. F_HIT does not force the status of
the block to register as DIRTY (the tag status bits are still consulted), so stores may still generate offchip activity.
In this mode, only one of the two sets may be enabled, and tag
parity checking must be disabled (set DCTAG_PER_EN to
zero).

SET_EN[1:0]

[1:0]

WO,3

Dcache Set Enable. At least one set must be enabled.

5.3.11 Dcache Status Register – DC_STAT
The Dcache status register (DC_STAT) is a read-write register. If a Dcache tag parity
error or data ECC error occurs, information about the error is latched in this register.
Figure 5–34 shows the Dcache status register.

21264/EV68A Hardware Reference Manual

Internal Processor Registers

5–31

Cbox CSRs and IPRs

Figure 5–34 Dcache Status Register
63

5 4 3 2 1 0

SEO
ECC_ERR_LD
ECC_ERR_ST
TPERR_P1
TPERR_P0
LK99-0042A

Table 5–21 describes the Dcache status register fields.
Table 5–21 Dcache Status Register Fields Description
Name

Extent

Type

Description

Reserved

[63:5]

—

SEO

[4]

W1C

Second error occurred. When set, this bit indicates that a second
Dcache store ECC error occurred within 6 cycles of the previous
Dcache store ECC error.

ECC_ERR_LD

[3]

W1C

ECC error on load. When set, this bit indicates that a single-bit ECC
error occurred while processing a load from the Dcache or any fill.

ECC_ERR_ST

[2]

W1C

ECC error on store. When set, this bit indicates that an ECC error
occurred while processing a store.

TPERR_P1

[1]

W1C

Tag parity error — pipe 1. When set, this bit indicates that a Dcache
tag probe from pipe 1 resulted in a tag parity error. The error is uncorrectable and results in a machine check.

TPERR_P0

[0]

W1C

Tag parity error — pipe 0. When set, this bit indicates that a Dcache
tag probe from pipe 0 resulted in a tag parity error. The error is uncorrectable and results in a machine check.

5.4 Cbox CSRs and IPRs
This section describes the Cbox CSRs and IPRs.
The Cbox configuration registers are split into three shift register chains:

5–32

•

The hardware allocates 367 bits for the WRITE_ONCE chain, of which the 21264/
EV68A uses 304 bits. During hardware reset (after BiST), 367 bits are always
shifted into the WRITE_ONCE chain from the SROM, MSB first, so that the
unused bits are shifted out the end of the WRITE_ONCE chain.

•

A 36-bit WRITE_MANY chain that is loaded using MTPR instructions to the Cbox
data register. Six bits of information are shifted into the WRITE_MANY chain during each write transaction to the Cbox data register.

•

A 60-bit Cbox ERROR_REG chain that is read by using MFFR instructions from
the Cbox data register in combination with MTPR instructions to the Cbox shift
register. Each write transaction to the Cbox shift register destructively shifts six bits
of information out of the Cbox error register.

Internal Processor Registers

21264/EV68A Hardware Reference Manual

Cbox CSRs and IPRs

5.4.1 Cbox Data Register – C_DATA
Figure 5–35 shows the Cbox data register.
Figure 5–35 Cbox Data Register
63

6 5

C_DATA[5:0]
LK99-0043A

Table 5–22 describes the Cbox data register fields.
Table 5–22 Cbox Data Register Fields Description
Name

Extent

Type

Description

Reserved

[63:6]

—

C_DATA[5:0]

[5:0]

Cbox data register. A HW_MTPR instruction to this register causes six
bits of data to be placed into a serial shift register. When the
HW_MTPR instruction is retired, the data is shifted into the Cbox. After
the Cbox shift register has been accessed, performing a HW_MFPR
instruction to this register will return six bits of data.

5.4.2 Cbox Shift Register – C_SHFT
Figure 5–36 shows the Cbox shift register.
Figure 5–36 Cbox Shift Register
63

1 0

C_SHIFT
LK99-0044A

Table 5–23 describes the Cbox shift register fields.
Table 5–23 Cbox Shift Register Fields Description
Name

Extent

Type

Description

Reserved

[63:1]

—

C_SHIFT

[0]

Writing a 1 to this register bit causes six bits of Cbox IPR data to shift into
the Cbox data register. Software can then use a HW_MFPR read operation
to the Cbox data register to read the six bits of data.

5.4.3 Cbox WRITE_ONCE Chain Description
The WRITE_ONCE chain order is contained in Table 5–24. In the table:

•

Many CSRs are duplicated for ease of hardware implementation. These CSRs are
indicated in italics. They must be written with values that are identical to the values
written to the original CSRs.

21264/EV68A Hardware Reference Manual

Internal Processor Registers

5–33

Cbox CSRs and IPRs

•

Only a brief description of each CSR is given. The functional description of these
CSRs is contained in Chapter 4.

•

The order of multibit vectors is [MSB:LSB], so the LSB is first bit in the Cbox
chain.

Table 5–24 describes the Cbox WRITE_ONCE chain order from LSB to MSB.
Table 5–24 Cbox WRITE_ONCE Chain Order
Cbox WRITE_ONCE Chain

Description

32_BYTE_IO[0]

Enable 32_BYTE I/O mode.

BC_CLK_RATIO[1]

Asserted when Bcache is at 1.5x ratio.

SKEWED_FILL_MODE[0]

Must be asserted for Bcache 1.5x ratio; for maximum performance,
can also be asserted for 3.0x and 3.5x ratios.

DCVIC_THRESHOLD[7:0]

Threshold of the number of Dcache victims that will accumulate
before streamed write transactions to the Bcache are initiated. The
Cbox can accumulate up to six victims for streamed Dcache processing. This register is programmed with the decoded value of the
threshold count.

BC_CLEAN_VICTIM[0]

Enable clean victims to the system interface.

SYS_BUS_SIZE[1:0]

Size of SysAddOut and SysAddOut buses.

SYS_BUS_FORMAT[0]

Indicates system bus format.

SYS_CLK_RATIO[4:1]

Speed of system bus.
Code
0001
0010
0100
1000

Multiplier
1.5X
2.0X
2.5X
3.0X

DUP_TAG_ENABLE[0]

Enable duplicate tag mode in the 21264/EV68A.

PRB_TAG_ONLY[0]

Enable probe-tag only mode in the 21264/EV68A.

FAST_MODE_DISABLE[0]

When asserted, disables fast data movement mode.

BC_RDVICTIM[0]

Enables RdVictim mode on the pins.

BC_CLEAN_VICTIM[0]

Duplicate CSR.

RDVIC_ACK_INHIBIT

Enable inhibition of incrementing acknowledge counter for RdVic
commands.

SYSBUS_MB_ENABLE

Enable MB commands offchip.

SYSBUS_ACK_LIMIT[0:4]

Sysbus acknowledge limit CSR.

SYSBUS_VIC_LIMIT[0:2]

Limit for victims.

BC_CLEAN_VICTIM[0]

Duplicate CSR.

BC_WR_WR_BUBBLE[0]

Write to write GCLK bubble.

BC_RD_WR_BUBBLES[0:5]

Read to write GCLK bubbles for the Bcache interface.

BC_RD_RD_BUBBLE[0]

Read to read GCLK bubble for banked Bcaches.

BC_SJ_BANK_ENABLE

Enable bank mode for Bcache.

5–34

Internal Processor Registers

21264/EV68A Hardware Reference Manual

Cbox CSRs and IPRs

Table 5–24 Cbox WRITE_ONCE Chain Order (Continued)
Cbox WRITE_ONCE Chain

Description

BC_WR_RD_BUBBLES[0:3]

Write to read GCLK bubbles.

DUP_TAG_ENABLE

Duplicate CSR.

SKEWED_FILL_MODE

Duplicate CSR.

BC_RDVICTIM

Duplicate CSR.

SKEWED_FILL_MODE

Duplicate CSR.

BC_RDVICTIM

Duplicate CSR.

BC_CLEAN_VICTIM

Duplicate CSR.

DUP_TAG_MODE

Duplicate CSR.

SKEWED_FILL_MODE

Duplicate CSR.

ENABLE_PROBE_CHECK

Enable error checking during probe processing.

SPEC_READ_ENABLE[0]

Enable speculative references to the system port.

SKEWED_FILL_MODE

Duplicate CSR.

SKEWED_FILL_MODE

Duplicate CSR.

MBOX_BC_PRB_STALL

Must be asserted when BC_RATIO = 4.0X, 5.0X, 6.0X, 7.0X, or
8.0X.

BC_LAT_DATA_PATTERN[0:31]

Bcache data latency pattern.

BC_LAT_TAG_PATTERN[0:23]

Bcache tag latency pattern.

BC_RDVICTIM

Duplicate CSR.

ENABLE_STC_COMMAND[0]

Enable STx_C instructions to the pins.

BC_LATE_WRITE_NUM[0:2]

Number of Bcache clocks to delay the data for Bcache write commands.

BC_CPU_LATE_WRITE_NUM[0:1]

Number of GCLK cycles to delay the Bcache clock/data from
index.

BC_BURST_MODE_ENABLE[0]

Burst mode enable signal.

BC_PENTIUM_MODE[0]

Enable Pentium mode RAM behavior.

BC_CLK_RATIO[1]

Duplicate CSR.

BC_FRM_CLK[0]

Force all Bcache transactions to start on rising edges of the A phase
of a GCLK.

BC_CLK_DELAY[0:1]

Delay of Bcache clock for 0,0,1,2 GCLK phases.

BC_DDMR_ENABLE[0]

Enables the rising edge of the Bcache forwarded clock (always
enabled).

BC_DDMF_ENABLE[0]

Enable the falling edge of the Bcache forwarded clock. (always
enabled).

BC_LATE_WRITE_UPPER[0]

Asserted when (BC_LATE_WRITE_NUM > 3) or
((BC_LATE_WRITE_NUM = 3) and
(BC_CPU_LATE_WRITE_NUM > 1)).

21264/EV68A Hardware Reference Manual

Internal Processor Registers

5–35

Cbox CSRs and IPRs

Table 5–24 Cbox WRITE_ONCE Chain Order (Continued)
Cbox WRITE_ONCE Chain

Description

BC_TAG_DDM_FALL_EN[0]

Enables the update of the 21264/EV68A Bcache tag outputs based
on the falling edge of the forwarded clock.

BC_TAG_DDM_RISE_EN[0]

Enables the update of the 21264/EV68A Bcache tag outputs based
on the rising edge of the forwarded clock.

BC_CLKFWD_ENABLE[0]

Enable clock forwarding on the Bcache interface.

BC_RCV_MUX_CNT_PRESET[0:1]

Initial value for the Bcache clock forwarding unload pointer FIFO.

BC_LATE_WRITE_UPPER[0]

Duplicate CSR.

SYS_DDM_FALL_EN[0]

Enables the update of the 21264/EV68A system outputs based on
the falling edge of the system forwarded clock.

SYS_DDM_RISE_EN[0]

Enables the update of the 21264/EV68A system outputs based on
the rising edge of the system forwarded clock.

SYS_CLKFWD_ENABLE[0]

Enables clock forwarding on the system interface.

SYS_RCV_MUX_CNT_PRESET[0:1]

Initial value for the system clock forwarding unload pointer FIFO.

SYS_CLK_DELAY[0:1]

Delay of 0 to 2 phases between the forwarded clock out and
address/data.

SYS_DDMR_ENABLE[0]

Enables the rising edge of the system forwarded clock (always
enabled).

SYS_DDMF_ENABLE[0]

Enables the falling edge of the system forwarded clock (always
enabled).

BC_DDM_FALL_EN[0]

Enables update of data/address on the rising edge of the system forwarded clock.

BC_DDM_RISE_EN[0]

Enables the update of data/address on the falling edge of the system
forwarded clock.

BC_CLKFWD_ENABLE

Duplicate CSR.

BC_RCV_MUX_CNT_PRESET[0:1]

Duplicate CSR.

BC_CLK_DELAY[0:1]

Duplicate CSR.

BC_DDMR_ENABLE

Duplicate CSR.

BC_DDMF_ENABLE

Duplicate CSR.

SYS_DDM_FALL_EN

Duplicate CSR.

SYS_DDM_RISE_EN

Duplicate CSR.

SYS_CLKFWD_ENABLE

Duplicate CSR.

SYS_RCV_MUX_CNT_PRESET[0:1]

Duplicate CSR.

SYS_CLK_DELAY[0:1]

Duplicate CSR.

SYS_DDMR_ENABLE

Duplicate CSR.

SYS_DDMF_ENABLE

Duplicate CSR.

BC_DDM_FALL_EN

Duplicate CSR.

BC_DDM_RISE_EN

Duplicate CSR.

5–36

Internal Processor Registers

21264/EV68A Hardware Reference Manual

Cbox CSRs and IPRs

Table 5–24 Cbox WRITE_ONCE Chain Order (Continued)
Cbox WRITE_ONCE Chain

Description

BC_CLKFWD_ENABLE

Duplicate CSR.

BC_RCV_MUX_CNT_PRESET[0:1]

Duplicate CSR.

SYS_DDM_FALL_EN

Duplicate CSR.

SYS_DDM_RISE_EN

Duplicate CSR.

SYS_CLKFWD_ENABLE

Duplicate CSR.

SYS_RCV_MUX_CNT_PRESET[0:1]

Duplicate CSR.

SYS_CLK_DELAY[0:1]

Duplicate CSR.

SYS_DDMR_ENABLE

Duplicate CSR.

SYS_DDMF_ENABLE

Duplicate CSR.

BC_DDM_FALL_EN

Duplicate CSR.

BC_DDM_RISE_EN

Duplicate CSR.

BC_CLKFWD_ENABLE

Duplicate CSR.

BC_RCV_MUX_CNT_PRESET[0:1]

Duplicate CSR.

BC_CLK_DELAY[0:1]

Duplicate CSR.

BC_DDMR_ENABLE

Duplicate CSR.

BC_DDMF_ENABLE

Duplicate CSR.

SYS_DDM_FALL_EN

Duplicate CSR.

SYS_DDM_RISE_EN

Duplicate CSR.

SYS_CLKFWD_ENABLE

Duplicate CSR.

SYS_RCV_MUX_CNT_PRESET[0:1]

Duplicate CSR.

SYS_CLK_DELAY[1:0]

Duplicate CSR.

SYS_DDMR_ENABLE

Duplicate CSR.

SYS_DDMF_ENABLE

Duplicate CSR.

BC_DDM_FALL_EN

Duplicate CSR.

BC_DDM_RISE_EN

Duplicate CSR.

BC_CLKFWD_ENABLE

Duplicate CSR.

BC_RCV_MUX_CNT_PRESET[1:0]

Duplicate CSR.

SYS_CLK_DELAY[0:1]

Duplicate CSR.

SYS_DDMR_ENABLE

Duplicate CSR.

SYS_DDMF_ENABLE

Duplicate CSR.

SYS_DDM_FALL_EN

Duplicate CSR.

SYS_DDM_RISE_EN

Duplicate CSR.

SYS_CLKFWD_ENABLE

Duplicate CSR.

SYS_RCV_MUX_CNT_PRESET[0:1]

Duplicate CSR.

21264/EV68A Hardware Reference Manual

Internal Processor Registers

5–37

Cbox CSRs and IPRs

Table 5–24 Cbox WRITE_ONCE Chain Order (Continued)
Cbox WRITE_ONCE Chain

Description

CFR_GCLK_DELAY[0:3]

Number of GCLK cycles to delay internal ClkFwdRst.

CFR_EV6CLK_DELAY[0:2]

Number of EV6Clk_x cycles to delay internal ClkFwdRst.

CFR_FRMCLK_DELAY[0:1]

Number of FrameClk_x cycles to delay internal ClkFwdRst.

BC_LATE_WRITE_NUM[0:2]

Duplicate CSR.

BC_CPU_LATE_WRITE_NUM[1:0]

Duplicate CSR.

JITTER_CMD[0]

Add one GCLK cycle to the SYSDC write path.

FAST_MODE_DISABLE[0]

Duplicate CSR.

SYSDC_DELAY[3:0]

Number of GCLK cycles to delay SysDc fill commands before
action by the Cbox.

DATA_VALID_DLY[1:0]

Number of Bcache clock cycles to delay signal SysDataInValid
before sample by the Cbox.

BC_DDM_FALL_EN

Duplicate CSR.

BC_DDM_RISE_EN

Duplicate CSR.

BC_CPU_CLK_DELAY[0:1]

Delay of Bcache clock for 0, 1, 2, 3 GCLK cycles.

BC_FDBK_EN[0:7]

CSR to program the Bcache forwarded clock shift register feedback
points.

BC_CLK_LD_VECTOR[0:15]

CSR to program the Bcache forwarded clock shift register load values.

BC_BPHASE_LD_VECTOR[0:3]

CSR to program the Bcache forwarded clock b-phase enables.

SYS_DDM_FALL_EN

Duplicate CSR.

SYS_DDM_RISE_EN

Duplicate CSR.

SYS_CPU_CLK_DELAY[0:1]

Delay of 0..3 GCLK cycles between the forwarded clock out and
address/data.

SYS_FDBK_EN[0:7]

CSR to program the system forwarded clock shift register feedback
points.

SYS_CLK_LD_VECTOR[0:15]

CSR to program the system forwarded clock shift register load values.

SYS_BPHASE_LD_VECTOR[0:3]

CSR to program the system forwarded clock b-phase enables.

SYS_FRAME_LD_VECTOR[0:4]

CSR to program the ratio between frame clock and system forwarded clock.

SYSDC_DELAY[4]

Fifth SYSDC_DELAY bit.

5.4.4 Cbox WRITE_MANY Chain Description
The WRITE_MANY chain order is contained in Table 5–25. Note the following:

5–38

•

Many CSRs are duplicated for ease of hardware implementation. These CSR names
are indicated in italics and have two leading asterisks.

•

Only a brief description of each CSR is given. The functional description of these
CSRs is contained in Chapter 3.

Internal Processor Registers

21264/EV68A Hardware Reference Manual

Cbox CSRs and IPRs

•

The order of multibit vectors is [MSB:LSB], so the LSB is first bit in the Cbox
chain.

Table 5–25 describes the Cbox WRITE_MANY chain order from LSB to MSB.
Table 5–25 Cbox WRITE_MANY Chain Order
Cbox WRITE_MANY Chain

Description

For Information:

BC_VALID_MODE

Control Bcache block parity calculation

Section 8.8

INIT_MODE[0]

Enable initialize mode

Section 7.6

BC_SIZE[3:0]

Bcache size

Table 4–42

BC_ENABLE[0]

Enable the Bcache

Table 4–42

BC_ENABLE

Duplicate CSR

Table 4–42

BC_SIZE[0:3]

Duplicate CSR

Table 4–42

BC_ENABLE

Duplicate CSR

Table 4–42

BC_ENABLE1

Duplicate CSR

Table 4–42

BC_ENABLE

Duplicate CSR

Table 4–42

INVAL_TO_DIRTY_ENABLE[1]

WH64 acknowledges

Table 4–15

ENABLE_EVICT

Enable issue evict

Table 4–1

BC_ENABLE

Duplicate CSR

Table 4–42

INVAL_TO_DIRTY_ENABLE[0]

WH64 acknowledges

Table 4–15

BC_ENABLE

Duplicate CSR

Table 4–42

BC_ENABLE

Duplicate CSR

Table 4–42

BC_ENABLE

Duplicate CSR

Table 4–42

SET_DIRTY_ENABLE[0]

SetDirty acknowledge programming

Table 4–16

INVAL_TO_DIRTY_ENABLE[0]

Duplicate CSR

Table 4–15

SET_DIRTY_ENABLE[2:1]

SetDirty acknowledge programming

Table 4–16

BC_BANK_ENABLE[0]

Enable bank mode for Bcache

Section 4.8.5

BC_SIZE[0:3]

Duplicate CSR

Table 4–42

INIT_MODE

Duplicate CSR

Section 7.6

BC_WRT_STS[0:3]

Write status for Bcache in initialize-mode
(Valid, Dirty, Shared, Parity)

Section 7.6

MBZ during initialization mode; see Section 7.6 for information.

Figure 5–37 shows an example of PALcode used to write to the WRITE_MANY chain.
Figure 5–37 WRITE_MANY Chain Write Transaction Example
;
; Initialize the Bcache configuration in the Cbox
;
;
BC_VALID_MODE = 1
;
BC_ENABLE = 1

21264/EV68A Hardware Reference Manual

Internal Processor Registers

5–39

Cbox CSRs and IPRs

;
INIT_MODE = 0
;
BC_SIZE = 0xF
;
INVALID_TO_DIRTY_ENABLE = 3
;
ENABLE_EVICT = 1
;
SET_DIRTY_ENABLE = 6
;
BC_BANK_ENABLE = 1
;
BC_WRT_STS = 0
;
; The value for the write_many chain is based on Table 5–25.
;
; The value is sampled from MSB, 6 bits at a time, as it is written
; to EV6__DATA. Therefore, before the value can be shifted in, it must be
; inverted on a by 6 basis. The code then writes out 6 bits at a time,
; shifting right by 6 after each write.
;
; So the following transformation is done on the write_many value:
;
;
[35:30]|[29:24]|[23:18]|[17:12]|[11:06]|[05:00] =>
;
[05:00]|[11:06]|[17:12]|[23:18]|[29:24]|[35:30]
;
;
WRITE_MANY chain = 0x07FBFFFFD
;
value to be shifted in = 0xF7FFEFFC1
;
; Before the chain can be written, I_CTL[SBE] must be disabled,
; and the code must be forced into the Icache.
;
ALIGN_CACHE_BLOCK <^x47FF041F>; align with nops
mb
lda
r0, ^x0086(r31)
hw_mtpr
r0, EV6__I_CTL
br
r0, .

; wait for MEM-OP’s to complete
; load I_CTL.....
; .....SDE=2, IC_EN=3, SBE=0
; create dest address

addq
r0, #17, r0
hw_mtpr r31, EV6__IC_FLUSH
bne
r31, .
hw_jmp_stall (r0)

; finish computing dest address
; flush the Icache
; separate retires
; force flush

ALIGN_CACHE_BLOCK <^x47FF041F>

; align with nops

bc_config:
mb
lda
ldah
zap

r1, ^xFFC1(r31)
r0, ^x7FFE(r31)
r1, #^x0c, r1

; pull this block in Icache
; data[15:00] = 0xFFC1
; data[31:16] = 0x7FFE
; clear out bits [31:16]

bis
r1, r0, r1
addq
r31, #6, r0
bc_config_shift_in:
hw_mtpr r1, EV6__DATA
subq
r0, #1, r0

; or in bits [31:16]
; shift in 6 x 6 bits

beq
r0, bc_config_done
srl
r1, #6, r1
br
r31, bc_config_shift_in
bc_config_done:
hw_mtpr r31, <EV6__MM_STAT ! 64>

; done if R0 is zero
; align next 6 bits
; continue shifting

beq
br
bis
bis

r31, bc_config_end
r31, .-4
r31, r31, r31
r31, r31, r31

; shift in 6 bits
; decrement R0

; wait until last shift
; predicts fall thru
; predict infinite loop
; nop
; nop

bc_config_end:

5–40

Internal Processor Registers

21264/EV68A Hardware Reference Manual

Cbox CSRs and IPRs

5.4.5 Cbox Read Register (IPR) Description
The Cbox read register is read 6 bits at a time. Table 5–26 shows the ordering from LSB
to MSB.
Table 5–26 Cbox Read IPR Fields Description
Name

Description

C_SYNDROME_1[7:0]

Syndrome for upper QW in OW of victim that was scrubbed.

C_SYNDROME_0[7:0]

Syndrome for lower QW in OW of victim that was scrubbed.

21264/EV68A Hardware Reference Manual

Internal Processor Registers

5–41

Table 5–26 Cbox Read IPR Fields Description (Continued)
Name

Description

C_STAT[4:0]

As follows:
Bits

Error Status

0 0 0 0 0 Either no error, or error on a speculative load, or a Bcache victim read
due to a Dcache/Bcache miss
0 0 0 0 1 BC_PERR (Bcache tag parity error)
0 0 0 1 0 DC_PERR (duplicate tag parity error)
0 0 0 1 1 DSTREAM_MEM_ERR
0 0 1 0 0 DSTREAM_BC_ERR
0 0 1 0 1 DSTREAM_DC_ERR
0 0 1 1 X PROBE_BC_ERR
0 1 0 0 0 Reserved
0 1 0 0 1 Reserved
0 1 0 1 0 Reserved
0 1 0 1 1 ISTREAM_MEM_ERR
0 1 1 0 0 ISTREAM_BC_ERR
0 1 1 0 1 Reserved
0 1 1 1 X Reserved
1 0 0 1 1 DSTREAM_MEM_DBL1
1 0 1 0 0 DSTREAM_BC_DBL1
1 1 0 1 1 ISTREAM_MEM_DBL1
1 1 1 0 0 ISTREAM_BC_DBL1
1

C_STS[3:0]

C_ADDR[6:42]

When the Cbox WRITE_ONCE chain bit
SKEWED_FILL_MODE[0] is clear, the error status is as specified;
when set, the error status applies only to first octaword of the fill, and
error status for the rest of the fill is generic DOUBLE_BIT_ERROR
(1XXXX).

If C_STAT equals xxx_MEM_ERR or xxx_BC_ERR, then C_STS contains the
status of the block as follows; otherwise, the value of C_STS is X:
Bit Value

Status of Block

7:4
3
2
1
0

Reserved
Parity
Valid
Dirty
Shared

Address of last reported ECC or parity error. If C_STAT value is
DSTREAM_DC_ERR, only bits 6:19 are valid.

6
Privileged Architecture Library Code
This chapter describes the 21264/EV68A privileged architecture library code (PALcode). The chapter is organized as follows:

•

PALcode description

•

PALmode environment

•

Required PALcode function codes

•

Opcodes reserved for PALcode

•

Internal processor register access mechanisms

•

PALshadow registers

•

PALcode emulation of FPCR

•

PALcode entry points

•

Translation buffer fill flows

•

Performance counter support

6.1 PALcode Description
PALcode is macrocode that provides an architecturally-defined, operating-system-specific programming interface that is common across all Alpha microprocessors. The
actual implementation of PALcode differs for each operating system. PALcode runs
with privileges enabled, instruction stream (Istream) mapping disabled, and interrupts
disabled. PALcode has privilege to use five special opcodes that allow functions such as
physical data stream (Dstream) references and internal processor register (IPR) manipulation.
PALcode can be invoked by the following events:

•

Reset

•

System hardware exceptions (MCHK, ARITH)

•

Memory-management exceptions

•

Interrupts

•

CALL_PAL instructions

PALcode has characteristics that make it appear to be a combination of microcode,
ROM BIOS, and system service routines, though the analogy to any of these other
items is not exact. PALcode exists for several major reasons:

21264/EV68A Hardware Reference Manual

Privileged Architecture Library Code

6–1

PALmode Environment

•

There are some necessary support functions that are too complex to implement
directly in a processor chip’s hardware, but that cannot be handled by a normal
operating system software routine. Routines to fill the translation buffer (TB),
acknowledge interrupts, and dispatch exceptions are some examples. In some architectures, these functions are handled by microcode, but the Alpha architecture is
careful not to mandate the use of microcode so as to allow reasonable chip implementations.

•

There are functions that must run atomically, yet involve long sequences of instructions that may need complete access to all of the underlying computer hardware.
An example of this is the sequence that returns from an exception or interrupt.

•

There are some instructions that are necessary for backward compatibility or ease
of programming; however, these are not used often enough to dedicate them to
hardware, or are so complex that they would jeopardize the overall performance of
the computer. For example, an instruction that does a VAX style interlocked memory access might be familiar to someone used to programming on a CISC machine,
but is not included in the Alpha architecture. Another example is the emulation of
an instruction that has no direct hardware support in a particular chip implementation.

In each of these cases, PALcode routines are used to provide the function. The routines
are nothing more than programs invoked at specified times, and read in as Istream code
in the same way that all other Alpha code is read. Once invoked, however, PALcode
runs in a special mode called PALmode.

6.2 PALmode Environment
PALcode runs in a special environment called PALmode, defined as follows:

•

Istream memory mapping is disabled. Because the PALcode is used to implement
translation buffer fill routines, Istream mapping clearly cannot be enabled. Dstream
mapping is still enabled.

•

The program has privileged access to all of the computer hardware. Most of the
functions handled by PALcode are privileged and need control of the lowest
levels of the system.

•

Interrupts are disabled. If a long sequence of instructions need to be executed
atomically, interrupts cannot be allowed.

An important aspect of PALcode is that it uses normal Alpha instructions for most of its
operations; that is, the same instruction set that nonprivileged Alpha programmers use.
There are a few extra instructions that are only available in PALmode, and will cause a
dispatch to the OPCDEC PALcode entry point if attempted while not in PALmode. The
Alpha architecture allows some flexibility in what these special PALmode instructions do.
In the 21264/EV68A, the special PALmode-only instructions perform the following functions:

6–2

•

Read or write internal processor registers (HW_MFPR, HW_MTPR)

•

Perform memory load or store operations without invoking the normal memorymanagement routines (HW_LD, HW_ST)

•

Return from an exception or interrupt (HW_RET)

Privileged Architecture Library Code

21264/EV68A Hardware Reference Manual

Required PALcode Function Codes

When executing in PALmode, there are certain restrictions for using the privileged
instructions because PALmode gives the programmer complete access to many of the
internal details of the 21264/EV68A. Refer to Section 6.4 for information on these special PALmode instructions.
Caution:

It is possible to cause unintended side effects by writing what appears to be
perfectly acceptable PALcode. As such, PALcode is not something that
many users will want to change. Before writing PALcode, at least become
familiar with the information in Appendix D.

6.3 Required PALcode Function Codes
Table 6–1 lists opcodes required for all Alpha implementations. The notation used is
oo.ffff, where oo is the hexadecimal 6-bit opcode and ffff is the hexadecimal 26-bit
function code.
Table 6–1 Required PALcode Function Codes
Mnemonic

Type

Function Code

DRAINA

Privileged

00.0002

HALT

Privileged

00.0000

IMB

Unprivileged

00.0086

6.4 Opcodes Reserved for PALcode
Table 6–2 lists the opcodes reserved by the Alpha architecture for implementation-specific use. These opcodes are privileged and are only available in PALmode.
Table 6–2 Opcodes Reserved for PALcode
Mnemonic

Opcode

Architecture
Mnemonic

Function

HW_LD

PAL1B

Dstream load instruction

HW_ST

PAL1F

Dstream store instruction

HW_RET

PAL1E

Return from PALcode routine

HW_MFPR

PAL19

Copies the value of an IPR into an integer GPR

HW_MTPR

PAL1D

Writes the value of an integer GPR into an IPR

These instructions generally produce an OPCDEC exception if executed while the processor is not in PALmode. If I_CTL[HWE] is set, these instructions can also be executed in kernel mode. Software that uses these instructions must adhere to the PALcode
restrictions listed in this section.

6.4.1 HW_LD Instruction
PALcode uses the HW_LD instruction to access memory outside the realm of normal
Alpha memory management and to perform special Dstream load transactions. Data
alignment traps are disabled for the HW_LD instruction.
Figure 6–1 shows the HW_LD instruction format.
21264/EV68A Hardware Reference Manual

Privileged Architecture Library Code

6–3

Opcodes Reserved for PALcode

Figure 6–1 HW_LD Instruction Format
31

26 25

21 20

OPCODE

16 15

13 12 11

DISP

TYPE
LEN

FM-05654.AI4

Table 6–3 describes the HW_LD instruction fields.
Table 6–3 HW_LD Instruction Fields Descriptions
Extent

Mnemonic Value

Description

[31:26]

OPCODE

1B16

The opcode value.

[25:21]

—

Destination register number.

[20:16]

—

Base register for memory address.

[15:13]

TYPE

0002

Physical — The effective address for the HW_LD instruction is physical.

0012

Physical/Lock — The effective address for the HW_LD instruction is
physical. It is the load lock version of the HW_LD instruction.

0102

Virtual/VPTE — Flags a virtual PTE fetch (LD_VPTE). Used by trap logic
to distinguish a single TB miss from a double TB miss. Kernel mode access
checks are performed.

1002

Virtual — The effective address for the HW_LD instruction is virtual.

1012

Virtual/WrChk — The effective address for the HW_LD instruction is
virtual. Access checks for fault-on-read (FOR), fault-on-write (FOW), read
and write protection.

1102

Virtual/Alt — The effective address for the HW_LD instruction is virtual.
Access checks use DTB_ALT_MODE IPR.

1112

Virtual/WrChk/Alt — The effective address for the HW_LD instruction is
virtual. Access checks for FOR, FOW, read and write protection. Access
checks use DTB_ ALT_MODE IPR.

Access length is longword.

Access length is quadword.

—

Holds a 12-bit signed byte displacement.

[12]

LEN

[11:0]

DISP

6.4.2 HW_ST Instruction
PALcode uses the HW_ST instruction to access memory outside the realm of normal
Alpha memory management and to do special forms of Dstream store instructions. Data
alignment traps are inhibited for HW_ST instructions. Figure 6–2 shows the HW_ST
instruction format.
Figure 6–2 HW_ST Instruction Format
31

26 25

OPCODE

21 20

16 15

13 12 11

DISP

TYPE
LEN

6–4

FM-05654.AI4

Privileged Architecture Library Code

21264/EV68A Hardware Reference Manual

Opcodes Reserved for PALcode

Table 6–4 describes the HW_ST instruction fields.
Table 6–4 HW_ST Instruction Fields Descriptions
Extent

Mnemonic Value

Description

[31:26]

OPCODE

1F16

The opcode value.

[25:21]

—

Write data register number.

[20:16]

—

Base register for memory address.

[15:13]

TYPE

0002

Physical — The effective address for the HW_ST instruction is physical.

0012

Physical/Cond — The effective address for the HW_ST instruction is
physical. Store conditional version of the HW_ST instruction. The lock
flag is returned in RA. Refer to PALcode restrictions for correct use of this
function.

0102

Virtual — The effective address for the HW_ST instruction is virtual.

1102

Virtual/Alt — The effective address for the HW_ST instruction is virtual.
Access checks use DTB_ ALT_MODE IPR.

All others Unused.
[12]
[11:0]

LEN
DISP

Access length is longword.

Access length is quadword.

—

Holds a 12-bit signed byte displacement.

6.4.3 HW_RET Instruction
The HW_RET instruction is used to return instruction flow to a specified PC. The RB
field of the HW_RET instruction specifies an integer GPR, which holds the new value
of the PC. Bit [0] of this register provides the new value of PALmode after the
HW_RET instruction is executed. Bits [15:14] of the instruction determine the stack
action.
Normally the HW_RET instruction succeeds a CALL_PAL instruction, or a trap handler that pushed its PC onto the prediction stack. In this mode, the HINT should be set
to ‘10’ to pop the PC and generate a predicted target address for the HW_RET instruction.
In some conditions, the HW_RET instruction is used in the middle of a PALcode flow
to cause a group of instructions to retire. In these cases, if the HW_RET instruction
does not have a corresponding instruction that pushed a PC onto the stack, the HINT
field should be set to ‘00’ to keep the stack from being modified.
In the rare circumstance that the HW_RET instruction might be used like a JSR or
JSR_COROUTINE, the stack can be managed by setting the HINT bits accordingly.
See Section D.25 for more information about the HW_RET instruction.
Figure 6–3 shows the HW_RET instruction format.

21264/EV68A Hardware Reference Manual

Privileged Architecture Library Code

6–5

Opcodes Reserved for PALcode

Figure 6–3 HW_RET Instruction Format
31

26 25

21 20

OPCODE

16 15 14 13 12

DISP

HINT
STALL

FM-05656.AI4

Table 6–5 describes the HW_RET instruction fields.
Table 6–5 HW_RET Instruction Fields Descriptions
Extent

Mnemonic Value

Description

[31:26]

OPCODE

1E16

The opcode value.

[25:21]

—

[20:16]

—

Target PC of the HW_RET instruction. Bit [0] of the register’s contents
determines the new value of PALmode.

[15:14]

HINT

HW_JMP — The PC is not pushed onto the prediction stack. The predicted
target is PC + (4*DISP[12:0]).

HW_JSR — The PC is pushed onto the prediction stack. The predicted
target is PC + (4*DISP[12:0]).

HW_RET — The prediction is popped off the stack and used as the target.

HW_COROUTINE — The prediction is popped off the stack and used as
the target. The PC is pushed onto the stack.

—

If set, the fetcher is stalled until the HW_RET instruction is retired or
aborted. The 21264/EV68A will:

[13]

STALL

• Force a mispredict
• Kill instructions that were fetched beyond the HW_RET instruction
• Refetch the target of the HW_RET instruction
• Stall until the HW_RET instruction is retired or aborted
If instructions beyond the HW_RET have been issued out of order, they
will be killed and refetched.
[12:0]

DISP

—

Holds a 13-bit signed longword displacement.

6.4.4 HW_MFPR and HW_MTPR Instructions
The HW_MFPR and HW_MTPR instructions are used to access internal processor registers. The HW_MFPR instruction reads the value from the specified IPR into the integer register specified by the RA field of the instruction. The HW_MTPR instruction
writes the value from the integer GPR, specified by the RB field of the instruction, into
the specified IPR. Figure 6–4 shows the HW_MFPR and HW_MTPR instructions format.
Figure 6–4 HW_MFPR and HW_MTPR Instructions Format
31

26 25

OPCODE

21 20

16 15

8 7

INDEX

SCBD_MASK
FM-05657.AI4

6–6

Privileged Architecture Library Code

21264/EV68A Hardware Reference Manual

Internal Processor Register Access Mechanisms

Table 6–6 describes the HW_MFPR and HW_MTPR instructions fields.
Table 6–6 HW_MFPR and HW_MTPR Instructions Fields Descriptions
Extent

Mnemonic

Value

Description

[31:26]

OPCODE

1916

The opcode value for the HW_MFPR instruction.

1D16

The opcode value for the HW_MTPR instruction.

[25:21]

—

Destination register for the HW_MFPR instruction. It should be R31
for the HW_MTPR instruction.

[20:16]

—

Source register for the HW_MTPR instruction. It should be R31 for the
HW_MFPR instruction.

[15:8]

INDEX

—

IPR index.

[7:0]

SCBD_MASK —

Specifies which IPR scoreboard bits in the IQ are to be applied to this
instruction. If a mask bit is set, it indicates that the corresponding IPR
scoreboard bit should be applied to this instruction.

6.5 Internal Processor Register Access Mechanisms
This section describes the hardware and software access mechanisms that are used for
the 21264s IPRs.
Because the Ibox reorders and executes instructions speculatively, extra hardware is
required to provide software with the correct view of the architecturally-defined state.
The Alpha architecture defines two classes of state: general-purpose registers and
memory. Register renaming is used to provide architecturally-correct register file
behavior. The Ibox and Mbox each have dedicated hardware that provides correct memory behavior to the programmer. Because the internal processor registers are implementation-specific, and their state is not defined by the Alpha architecture, access
mechanisms for these registers may be defined that impose restrictions and limitations
on the software that uses them.
For every IPR, each instruction type can be classified by how it affects and is affected
by the value held by that IPR.

•

Explicit readers are HW_MFPR instructions that explicitly read the value of the
IPR.

•

Implicit readers are instructions whose behavior is affected by the value of the IPR.
For example, each load instruction is an implicit reader of the DTB.

•

Explicit writers are HW_MTPR instructions that explicitly write a value into the
IPR.

•

Implicit writers are instructions that may write a value into the IPR as a side effect
of execution. For example, a load instruction that generates an access violation is
an implicit writer of the VA, MM_STAT, and EXC_ADDR IPRs. In the 21264/
EV68A, only instructions that generate an exception will act as implicit IPR writers.

Only certain IPRs, such as those with write-one-to-clear bits, are both implicitly and
explicitly written. The read-write semantics of these IPRs is controlled by software.

21264/EV68A Hardware Reference Manual

Privileged Architecture Library Code

6–7

Internal Processor Register Access Mechanisms

6.5.1 IPR Scoreboard Bits
In previous Alpha implementations, IPR registers were not scoreboarded in hardware.
Software was required to schedule HW_MTPR and HW_MFPR instructions for each
machine’s pipeline organization in order to ensure correct behavior. This software
scheduling task is more difficult in the 21264/EV68A because the Ibox performs
dynamic scheduling. Hence, eight extra scoreboard bits are used within the IQ to help
maintain correct IPR access order. The HW_MTPR and HW_MFPR instruction formats contain an 8-bit field that is used as an IPR scoreboard bit mask to specify which
of the eight IPR scoreboard bits are to be applied to the instruction.
If any of the unmasked scoreboard bits are set when an instruction is about to enter the
IQ, then the instruction, and those behind it, are stalled outside the IQ until all the
unmasked scoreboard bits are clear and the queue does not contain any implicit or
explicit readers that were dependent on those bits when they entered the queue. When
all the unmasked scoreboard bits are clear, and the queue does not contain any of those
readers, the instruction enters the IQ and the unmasked scoreboard bits are set.
HW_MFPR instructions are stalled in the IQ until all their unmasked IPR scoreboard
bits are clear.
When scoreboard bits [3:0] and [7:4] are set, their effect on other instructions is different, and they are cleared in a different manner.
If any of scoreboard bits [3:0] are set when a load or store instruction enters the IQ, that
load or store instruction will not be issued from the IQ until those scoreboard bits are
clear.
Scoreboard bits [3:0] are cleared when the HW_MTPR instructions that set them are
issued (or are aborted). Bits [7:4] are cleared when the HW_MTPR instructions that set
them are retired (or are aborted).
Bits [3:0] are used for the DTB_TAG and DTB_PTE register pairs within the DTB fill
flows. These bits can be used to order writes to the DTB for load and store instructions.
See Sections 5.3.1 and 6.9.1.
Bit [0] is used in both DTB and ITB fill flows to trigger, in hardware, a lightweight
memory barrier (TB-MB) to be inserted between a LD_VPTE and the corresponding
virtual-mode load instruction that missed in the TB.

6.5.2 Hardware Structure of Explicitly Written IPRs
IPRs that are written by software are physically implemented as two registers. When
the HW_MTPR instruction that writes the IPR executes, it writes its value to the first
register. When the HW_MTPR instruction is retired, the contents of the first register are
written into the second register. Instructions that either implicitly or explicitly read the
value of the IPR access the second register. Read-after-write and write-after-write
dependencies are managed using the IPR scoreboard bits. To avoid write-after-read
conflicts, the second register is not written until the writer is retired. The writer will not
be retired until the previous reader is retired, and the reader is retired after it has read its
value from the second register.
Some groups of IPRs are built using a single shared first register. To prevent writeafter-write conflicts, IPRs that share a first register also share scoreboard bits.

6–8

Privileged Architecture Library Code

21264/EV68A Hardware Reference Manual

Internal Processor Register Access Mechanisms

6.5.3 Hardware Structure of Implicitly Written IPRs
Implicitly written IPRs are physically built using only a single level of register, however the IPR has two hardware states associated with it:
1. Default State—The contents of the register may be written when an instruction generates an exception. If an exception occurs, write a new value into the IPR and go to
state 2.
2. Locked State—The contents of the register may only be overwritten by an excepting instruction that is older than the instruction associated with the contents of the
IPR. If such an exception occurs, overwrite the value of the IPR. When the triggering instruction, or instruction that is older than the triggering instruction, is killed
by the Ibox, go to state 1.

6.5.4 IPR Access Ordering
IPR access mechanisms must allow values to be passed through each IPR from a producer to its intended consumers.
Table 6–7 lists all of the paired instruction orderings between instructions of the four
IPR access types. It specifies whether access order must be maintained, and if so, the
mechanisms used to ensure correct ordering.
Table 6–7 Paired Instruction Fetch Order
Second
Instruction

First Instruction

Implicit Reader Implicit Writer

Explicit Reader

Explicit Writer

Implicit
Reader

Read transactions can be
reordered.

No IPRs in this class. Read transactions can
be reordered.

Implicit
Writer

No IPRs in this
class.

The hardware structure of implicitly
written IPRs handles
this case.

IPR-specific PALcode No IPRs in this class.
restrictions are
required for this case.
An interlock mechanism must be placed
between the explicit
reader and the implicit
writer (a read transaction).

Explicit
Reader

Read transactions can be
reordered.

If the reader is in the
PALcode routine
invoked by the
exception associated
with the writer, then
ordering is guaranteed.

Read transactions can
be reordered.

21264/EV68A Hardware Reference Manual

A variety of mechanisms are
used to ensure order:
scoreboard bits to stall issue of
reader; HW_RET_STALL to
stall reader; double write plus
buffer blocks to force retire and
allow for propagation delay.

Scoreboard bits stall issue of
reader until writer is retired.

Privileged Architecture Library Code

6–9

Internal Processor Register Access Mechanisms

Table 6–7 Paired Instruction Fetch Order (Continued)
Second
Instruction

Explicit
Writer

First Instruction

Reader reads
second register.
Writer cannot
write second
register until it
is retired.

Write-one-to-clear
bits, or performance
counter special case.
For example, performance counter increments are typically
not scoreboarded
against read transactions.

Reader reads second
Scoreboard bits stall second
register. Writer cannot writer in map stage until first
write second register
writer is retired.
until it is retired.

For convenience of implementation, there is no IPR scoreboard bit checking within the
same fetch block (octaword-aligned octaword).

•

Within one fetch block, there can be only one explicit writer (HW_MTPR) to an
IPR in a particular scoreboard group.

•

Within one fetch block, an explicit writer (HW_MTPR) to an IPR in a particular
scoreboard group cannot be followed by an explicit reader (HW_MFPR) to an IPR
in that same scoreboard group.

•

Within one fetch block, an explicit writer (HW_MTPR) to an IPR in a particular
scoreboard group cannot be followed by an implicit reader to an IPR in that scoreboard group. This case covers writes to DTB_PTE or DTB_TAG followed by a
LD, ST, or any memory operation, including HW_RETs without the ‘stall’ bit set.

6.5.5 Correct Ordering of Explicit Writers Followed by Implicit Readers
Across fetch blocks, the correct ordering of the explicit write of the DTB_PTE or
DTB_TAG followed by an implicit reader (memory operation) is guaranteed using the
IPR scoreboard bits.
However, there are cases where correct ordering of explicit writers followed by implicit
readers cannot be guaranteed using the IPR scoreboard mechanism. If the instruction
that implicitly reads the IPR does so before the issue stage of the pipeline, the scoreboard mechanism is not sufficient.
For example, modification of the ITB affects instructions before the issue state of the
pipeline. In this case, PALcode must contain a HW_RET instruction, with its stall bit
set, before any instruction that implicitly reads the IPR(s) in question. This prevents
instructions that are newer than the HW_RET instruction from being successfully
fetched, issued, and retired until after the HW_RET instruction is retired (or aborted).
There are also cases when the HW_RET with the STALL bit mechanism is not sufficient. There may be additional propagation delay past the retirement of the HW_RET
instruction. In these cases, instead of using a HW_RET, a suggested method of ensuring the ordering is coding a group of 5 fetch blocks, where the first contains the
HW_MTPR to the IPR, the second contains a HW_MTPR to the same IPR or one in the
same scoreboard group, and where the following 3 fetch blocks each contain at least
one non-NOP instruction. See Appendix D for a listing of cases where this method is
recommended.

6–10

Privileged Architecture Library Code

21264/EV68A Hardware Reference Manual

PALshadow Registers

6.5.6 Correct Ordering of Explicit Readers Followed by Implicit Writers
Certain IPRs that are updated as a result of faulting memory operations require PALcode assistance to maintain ordering against newer instructions. Consider the following
code sequence:
HW_MFPR IPR_MM_STAT
LDQ rx,(ry)

It is typically the case that these instructions would issue in-order:

•

The MFPR is data-ready and both instructions use a lower subcluster. However, the
HW_MFPRs (and HW_MTPRs) respond to certain resource-busy indications and
do not issue when the MBOX informs the IBOX that a certain set of resources
(store bubbles) are busy.

•

The LDs respond to a different set of resource-busy indications (load-bubbles) and
could issue around the HW_MFPR in the presence of the former. PALcode assistance is required to enforce the issue order.

One totally reliable method is to insert an MB (memory barrier) instruction before the
first load that occurs after the HW_MFPR MM_STAT. Another method would be to
force a register dependency between the HW_MFPR and the LD.

6.6 PALshadow Registers
The 21264/EV68A contains eight extra virtual integer registers, called shadow registers, which are available to PALcode for use as scratch space and storage for commonly used values. These registers are made available under the control of the SDE[1]
field of the I_CTL IPR. These shadow registers overlay R4 through R7 and R20
through R23, when the CPU is in PALmode and SDE[1] is set.
PALcode generally runs with shadow mode enabled. Any PALcode that supports
CALL_PAL instructions must run in that mode because the hardware writes a
PALshadow register with the return address of CALL_PAL instructions.
PALcode may occasionally be required to toggle shadow mode to obtain access to the
overlayed registers. See the PALcode restriction, Updating I_CTL[SDE], in Section
D.32.

6.7 PALcode Emulation of the FPCR
The FPCR register contains status and control bits. They are accessed by way of the
MT_FPCR and MF_FPCR instructions. The register is physically implemented like an
explicitly written IPR. It may be written with a value from the floating-point register
file by way of the MT_FPCR instruction. Architecturally-compliant FPCR behavior
requires PALcode assistance. The FPCR register must operate as listed here:
1. Correct operation of the status bits, which must be set when a floating-point
instruction encounters an exceptional condition, independent of whether a trap for
the condition is enabled.
2. Correct values must be returned when the FPCR is read by way of a MF_FPCR
instruction.

21264/EV68A Hardware Reference Manual

Privileged Architecture Library Code

6–11

PALcode Entry Points

3. Correct actions must occur when the FPCR is written by way of a MT_FPCR
instruction.

6.7.1 Status Flags
The FPCR status bits in the 21264/EV68A are set with PALcode assistance. Floatingpoint exceptions, for which the associated FPCR status bit is clear or for which the
associated trap is enabled, result in a hardware trap to the ARITH PALcode routine. The
EXC_SUM register contains information to allow this routine to update the FPCR
appropriately, and to decide whether to report the exception to the operating system.

6.7.2 MF_FPCR
The MF_FPCR is issued from the floating-point queue and executed by the Fbox. No
PALcode assistance is required.

6.7.3 MT_FPCR
The MT_FPCR instruction is issued from the floating-point queue. This instruction is
implemented as an explicit IPR write operation. The value is written into the first latch,
and when the instruction is retired, the value is written into the second latch. There is no
IPR scoreboarding mechanism in the floating-point queue, so PALcode assistance is
required to ensure that subsequent readers of the FPCR get the updated value.
After writing the first latch, the MT_FPCR instruction invokes a synchronous trap to
the MT_FPCR PALcode entry point. The PALcode can return using a HW_RET
instruction with its STALL bit set. This sequence ensures that the MT_FPCR instruction will be correctly ordered for subsequent readers of the FPCR.

6.8 PALcode Entry Points
PALcode is invoked at specific entry points, of which there are two classes:
CALL_PAL and exceptions.

6.8.1 CALL_PAL Entry Points
CALL_PAL entry points are used whenever the Ibox encounters a CALL_PAL instruction in the Istream. To speed the processing of CALL_PAL instructions, CALL_PAL
instructions do not invoke pipeline aborts but are processed as normal jumps to the offset from the contents of the PAL_BASE register, which is specified by the CALL_PAL
instruction’s function field.
The Ibox fetches a CALL_PAL instruction, bubbles one cycle, and then fetches the
instructions at the CALL_PAL entry point. For convenience of implementation, returns
from CALL_PAL are aided by a linkage register (much like JSRs). PALshadow register R23 is used as the linkage register. The Ibox loads the PC of the instruction after the
CALL_PAL instruction, into the linkage register. Bit [0] of the linkage register is set if
the CALL_PAL instruction was executed while the processor was in PALmode.
The Ibox pushes the value of the return PC onto the return prediction stack.
CALL_PAL instructions start at the following offsets:

6–12

•

Privileged CALL_PAL instructions start at offset 200016.

•

Nonprivileged CALL_PAL instructions start at offset 300016.

Privileged Architecture Library Code

21264/EV68A Hardware Reference Manual

PALcode Entry Points

Each CALL_PAL instruction includes a function field that is used to calculate the PC of
its associated PALcode entry point. The PALcode OPCDEC exception flow will be
invoked if the CALL_PAL function field satisfies any of the following requirements:

•

Is in the range of 4016 to 7F16 inclusive

•

Is greater than BF16

•

Is between 0016 and 3F16 inclusive, and IER_CM[CM] is not equal to the kernel
mode value 0

If none of the conditions above are met, the PALcode entry point PC is as follows:

•

PC[63:15] = PAL_BASE[63:15]

•

PC[14] = 0

•

PC[13] = 1

•

PC[12] = CALL_PAL function field [7]

•

PC[11:6] = CALL_PAL function field [5:0]

•

PC[5:1] = 0

•

PC[0] = 1 (PALmode)

6.8.2 PALcode Exception Entry Points
When hardware encounters an exception, Ibox execution jumps to a PALcode entry
point at a PC determined by the type of exception. The return PC of the instruction that
triggered the exception is placed in the EXC_ADDR register and onto the return prediction stack.
Table 6–8 shows the PALcode exception entry locations and their offset from the
PAL_BASE IPR.
Table 6–8 PALcode Exception Entry Locations
Entry Name

Type

Offset16

Description

DTBM_DOUBLE_3

Fault

100

Dstream TB miss on virtual page table entry fetch. Use threelevel flow.

DTBM_DOUBLE_4

Fault

180

Dstream TB miss on virtual page table entry fetch. Use fourlevel flow.

FEN

Fault

200

Floating point disabled.

UNALIGN

Fault

280

Unaligned Dstream reference.

DTBM_SINGLE

Fault

300

Dstream TB miss.

DFAULT

Fault

380

Dstream fault or virtual address sign check error.

OPCDEC

Fault

400

Illegal opcode or function field:
• Opcode 1, 2, 3, 4, 5, 6 or 7
• Opcode 1916, 1B16, 1D 16, 1E16 or 1F16 , not PALmode or
not I_CTL[HWE]
• Extended precision IEEE format
• Unimplemented function field of opcodes 1416 or 1C16

IACV

Fault

480

Istream access violation or virtual address sign check error.

21264/EV68A Hardware Reference Manual

Privileged Architecture Library Code

6–13

Translation Buffer (TB) Fill Flows

Table 6–8 PALcode Exception Entry Locations (Continued)
Entry Name

Type

Offset16

Description

MCHK

Interrupt

500

Machine check.

ITB_MISS

Fault

580

Istream TB miss.

ARITH

Synch. Trap 600

Arithmetic exception or update to FPCR.

INTERRUPT

Interrupt

Interrupts: hardware, software, and AST.

MT_FPCR

Synch. Trap 700

Invoked when a MT_FPCR instruction is issued.

RESET/WAKEUP

Interrupt

Chip reset or wake-up from sleep mode.

680

780

6.9 Translation Buffer (TB) Fill Flows
This section shows the expected PALcode flows for DTB miss and ITB miss. Familiarity with 21264/EV68A IPRs is assumed.

6.9.1 DTB Fill
Figure 6–5 shows single-miss DTB instructions flow.
Figure 6–5 Single-Miss DTB Instructions Flow Example
hw_mfprp23, EV6__EXC_ADDR

; (0L) get exception address

hw_mfprp4, EV6__VA_FORM

; (4-7,1L) get vpte address

hw_mfprp5, EV6__MM_STAT

; (0L) get miss info

hw_mfpr p7, EV6__EXC_SUM

; (0L) get exc_sum for ra

hw_mfpr p6, EV6__VA

; (4-7,1L) get original va

bic p7, #1, p7

; clear double miss flag

xor p4, p6, p4

; interlock p4 and p6

xor p4, p6, p4

; restore p4

trap__dtbm_single_vpte:
hw_ldq/v p4, (p4)

; (1L) get vpte

blt p_misc, trap__d1to1

; (xU) <63>=1 => 1-to-1

blbcp4, trap__invalid_dpte

; (xU) invalid => branch

and p4, #^x80, p7

; isolate mb bit

xor p7, #^x80, p7

; flip mb bit

ALIGN_FETCH_BLOCK <^x47FF041F>
PVC_VIOLATE <2>

; ignore scoreboard violation

hw_mtprp6, EV6__DTB_TAG0

; (2&6,0L) write tag0

hw_mtpr p6, EV6__DTB_TAG1

; (1&5,1L) write tag1

hw_mtprp4, <EV6__DTB_PTE0 ! ^x44>

; (0,4,2,6) (0L) write pte0

hw_mtprp4, <EV6__DTB_PTE1 ! ^x22>

; (3,7,1,5) (1L) write pte1

6–14

Privileged Architecture Library Code

21264/EV68A Hardware Reference Manual

Translation Buffer (TB) Fill Flows

ASSUME <tb_mb_en + pte_eco> ne 2
.if ne pte_eco
bne p7, trap__dtbm_single_mb

; branch for mb

hw_ret (p23)

; return

trap__dtbm_single_mb:
mb
hw_ret(p23)

; return

.iff
hw_ret(p23)

; return
; (assumes tb_mb_en on multi-processors)

.endc

The following list presents information about the single-miss DTB code example:

•

In Figure 6–5, where (x,y) or (y) appear in the comments, x specifies the scoreboard
bits and y specifies the Ebox subcluster.

•

r4 −r7 and r20 − r23 are PALshadow registers.

•

PALshadow r22 contains a flag that indicates whether the native code is running
“1−to−1”, that is, running in a mode where the physical address should be mapped
1−to−1 to the virtual address, rather than being taken from a page table.

•

IPR scoreboard bits [3:0] are used to order the restarted load or store instructions
for the DTB write transactions.

•

MM_STAT and VA will not be overwritten if the LD_VPTE instruction misses the
DTB. There is no issue order constraint.

•

The code is written to prevent a later execution of the DTB fill instruction from
being issued before a previous execution and corrupting the previous write to the
TB registers. The correct sequence of executions is accomplished by placing code
dependencies on scoreboard bits [7:4] in the path of the successive writers. This
prevents the successive writers from being issued before the previous writers are
retired.

•

When I_CTL[TB_MB_EN] = 1, the issue of MTPR DTB_PTE0 triggers, in hardware, a lightweight memory barrier (TB-MB). The lightweight memory barrier
enforces read-ordering of store instructions from another processor (I) to this processor’s (J) page table and this processor’s virtual memory area such that if this
processor sees the write to the PTE from (I) it will see the new data.

Processor I

Processor J

Wr Data

LD/ST

Wr PTE

LD-PTE, write TB
LD/ST

21264/EV68A Hardware Reference Manual

Privileged Architecture Library Code

6–15

Translation Buffer (TB) Fill Flows

•

The conditional branch is placed in the code so that all of the MTPR instructions
are issued and retired or none of them are issued and retired. This allows the TB fill
hardware to update the TB whenever it sees the retiring of PTE1 and to ignore
writes to TAG0/TAG1/PTE0/PTE1 in the interim between the issuing of those
writes and a retire of PTE1.

•

As an alternative to using I_CTL[TB_MB_EN] = 1 to enforce read ordering,
I_CTL[TB_MB_EN] can be set to 0 and the PALcode may use a bit in the PTE to
indicate whether to do an explicit MB.
The flow example in Figure 6–5 shows the code using pte_eco and the code not
using pte_eco. It assumes the following:

•

–

In a multi-processor configuration, if pte_eco is not enabled, it is necessary to
enable tb_mb_en.

–

In a uni-processor configuration, if pte_eco is not enabled, it is not necessary to
enable tb_mb_en.

–

At no time should pte_eco and tb_mb_en both be enabled.

The value in DTB_PTEx[GH] determines whether the scoreboard mechanism alone
is sufficient to guarantee all subsequent load/store instructions (implicit readers of
the DTB) are ordered relative to the creation of a new DTB entry; whether all subsequent loads and stores to the loaded address will hit in the DTB.
–

If DTB_PTEx[GH] is zero, the scoreboard mechanism alone is sufficient.

–

If DTB_PTEx[GH] is not zero, the scoreboard mechanism alone is not sufficient (although this is not a problem). In this case, the new DTB entry is not
visible to subsequent load/store instructions until after the MTPR DTB_PTE1
retires.
Issuing a HW_RET_STALL instead of a HW_RET would guarantee ordering,
but is not necessary. Code executes correctly without the stall although execution might result in two passes through the DTB miss flow, rather than one,
because the re-execution of the memory operation after the first DTB miss
might miss again.
This behavior is functionally correct because DTB loads that tag-match an
existing DTB entry are ignored by the 21264/EV68A and the second DTB miss
execution will load exactly the same entry as the first.

6.9.2 ITB Fill
Figure 6–6 shows the ITB miss instructions flow.
Figure 6–6 ITB Miss Instructions Flow Example
hw_mfpr

r4,

EV6__IVA_FORM

; (0L) get vpte address

hw_mfpr

r23,

EV6__EXC_ADDR

; (0L) get exception address

lda

r6,

^x0FFF(r31)

; (xU) create mask for prot

bis

r31, r31, r31

; (xU) fill out fetch block

trap__itb_miss_vpte:
hw_ldq/v

r4,

(r4)

and

r4, r6, r5

blt p_misc, trap__i1to1

6–16

Privileged Architecture Library Code

; (xL) get vpte
; (xL) get prot bits
; (xU) 1-to-1 => branch

21264/EV68A Hardware Reference Manual

Performance Counter Support

srl

r4,

#OSF_PTE__PFN__S, r6

; (xU) shift PFN to <0>

sll

r6,

#EV6__ITB_PTE__PFN__S, r6

; (xU) shift PFN into place

and

r4,

#<1@OSF_PTE__FOE__S>, r7

; (xL) get FOE bit

blbc

r4,

trap__invalid_ipte

; (xU) invalid => branch

bne

r7,

trap__foe

; (xU) FOE => branch

srl

r4, #7, r7

; check for mb bit

bis

r5, r6, r6

; (xL) PTE in ITB format

hw_mtpr

r23,

EV6__ITB_TAG

; (6,0L) write tag

hw_mtpr

r6,

EV6__ITB_PTE

; (0&4,0L) write PTE

ASSUME <tb_mb_en + pte_eco> ne 2
.if ne pte_eco
blbc

r7, trap__itb_miss_mb

; branch for mb

hw_ret_stall (r23); (0L)
trap__itb_miss_mb:
mb
.endc
hw_ret_stall (r23)

; (0L)

The following list presents information about the ITB miss flow code example:

•

In Figure 6–6, where (x,y) or (y) appear in the comments, x specifies the scoreboard
bits and y specifies the Ebox subcluster.

•

The ITB is only accessed on Icache misses.

•

r4 −r7 and r20 − r23 are PALshadow registers.

•

The HW_RET instruction should have its STALL bit set to ensure that the restarted
Istream does not read the ITB until the ITB is written.
As an alternative to using I_CTL[TB_MB_EN] = 1 to enforce read ordering,
I_CTL[TB_MB_EN] can be set to 0 and the PALcode may use a bit in the PTE to
indicate whether to do an explicit MB. The flow example in Figure 6–6 assumes
this alternative.

6.10 Performance Counter Support
The 21264/EV68A provides hardware support for two methods of obtaining program
performance feedback information. The two methods do not require program modification. Instead, performance monitoring utilities make calls to the PALcode to set up the
counters and contain interrupt handlers that call PALcode to retrieve the collected data.
The first method, Aggregate mode, offers capabilities that are similar to earlier microprocessor performance counters. This mode counts events when enabled, until it overflows, causing an interrupt that can retrieve the collected data. The second method,

21264/EV68A Hardware Reference Manual

Privileged Architecture Library Code

6–17

Performance Counter Support

ProfileMe mode, supports a new way of statistically sampling individual instructions
during program execution. This mode counts events triggered by a targeted inflight
instruction.
Counter support uses the hardware registers listed in Table 6–9.
Table 6–9 IPRs Used for Performance Counter Support
Register Name

Mnemonic

Relevant Fields

Described in Section

ProfileMe PC

PMPC

All fields

5.2.6

Interrupt enable and current processor mode

IER_CM

PCEN[1:0]

5.2.9

Interrupt summary

ISUM

PC[1:0]

5.2.11

Ibox control

I_CTL

SPCE, PCT0_EN, PCT1_EN

5.2.15

Ibox status

I_STAT

OVR, ICM, TRAP-TYPE,
LSO, TRP, MIS

5.2.16

Ibox process context

PCTX

PPCE

5.2.21

Performance counter support

PCTR_CTL All fields

5.2.22

6.10.1 General Precautions
Initialize both counters, (PCTR_CTL[PCTR0 and PCTR1]), to zero in reset PALcode
to avoid spurious interrupts when exiting initial PALcode. Counters must be written
twice during initialization to ensure that the overflow latch has been cleared (see the
PALcode restrictions in Sections D.28 and D.34).
The counters should never be left within one cycle of overflow when disabled because
that can cause some interrupts to be blocked in anticipation of an overflow interrupt
(see PALcode restriction 32).
If a counter is at the overflow threshold and a value is written to that counter, the
counter signals an overflow interrupt upon leaving PALmode, even if that counter is
disabled. To avoid that interrupt, the PALcode should clear the interrupt by writing to
HW_INT_CLR.
Interrupts are disabled in PALmode.
As a quirk of the implementation, while counting is disabled, a read of PCTR_CTL can
yield value+some increment, where value is the actual value in PCTR_CTL, and increment for PCTR0 is in the range 0..4 (retired instructions in that cycle), and increment
for PCTR1 is dependent on SL1.

6.10.2 Aggregate Mode Programming Guidelines
Use the following information to program counters in Aggregate mode.
6.10.2.1 Aggregate Mode Precautions

Counters continue to count after overflow.
Only the counters return useful data. See Table 6–11 for counting modes.
Counters can be read by a PALcode instruction at any time to get the aggregate count.

6–18

Privileged Architecture Library Code

21264/EV68A Hardware Reference Manual

Performance Counter Support

The legal range for PCTR0 when writing the IPR is 0:(2**20-16).
The legal range for PCTR1 when writing the IPR is 0:(2**20-4).
6.10.2.2 Operation

1. Setup
The following IPRs need to be set up by PALcode instructions.
IPR Name

Relevant Fields Meaning

IER_CM

PCEN[1:0]

Enable Interrupts.

PCTX

PPCE

Enable Process Performance Counting or use I_CTL[SPCE].

PCTR_CTL

SL0

Selects Aggregate or ProfileMe mode; set to 0 for Aggregate mode.

SL1

Selects PCTR0 and PCTR1 counting modes. See Table 6–11 for more information.

PCTR0[19:0]

Set counter 0 starting value [0:(2**20-16)]. See Section 6.10.1 for setup
precautions.

PCTR1[19:0]

Set counter 1 starting value [0:(2**20-4)]. See Section 6.10.1 for setup precautions.

SPCE

Enable System Performance Counting or use PCTX[PPCE].

PCT0_EN

Enable performance counter 0.

PCT1_EN

Enable performance counter 1.

I_CTL

2. Count
If PCTR0 and PCTR1 are enabled, will increment according to modes selected by
SL0 and SL1.
3. Overflow
If PCEN[1:0] is enabled, PC[1:0] is set when PCTR0 or PCTR1 overflows.
4. Hardware interrrupt
When PC[1:0] is set, the PALcode interrupt routine is entered. Interrupt is acknowledged and PALcode generates an interrupt to the operating system performance
monitoring utility.
5. Operating system interrupt handler
The handler should read the IPR PCTR_CTL, as shown in Table 6–10, to note
which counter overflowed in the handler's data structures. The handler may read the
counter to see how many events have happened since the overflow.
The handler may also choose to write the counters to control the frequency of interrupts.
Table 6–10 Aggregate Mode Returned IPR Contents
IPR

Field

Contents

PCTR_CTL

PCTR0[19:0]

Counter #0 value

PCTR1[19:0]

Counter #1 value

21264/EV68A Hardware Reference Manual

Privileged Architecture Library Code

6–19

Performance Counter Support

6.10.2.3 Aggregate Counting Mode Description
6.10.2.3.1 Cycle counting

Counts cycles.
PCTR0 is incremented by the number of cycles counted, that is, 1.
6.10.2.3.2 Retired instructions cycles

PCTR0 is incremented by up to 8 retired instructions per cycle when enabled via
I_CTL[PCT0_EN] and either I_CTL[SPCE] or PCTX[PPCE]. On overflow, an interrupt is triggered as ISUM[PC0] if enabled via IER_CM[PCEN0].
The 21264/EV68A can retire up to 11 instructions per cycle, which exceeds PCTR0's
maximum increment of 8 per cycle. However, no retires go uncounted because the
21264/EV68A cannot sustain 11 retires per cycle, and the 21264/EV68A corrects
PCTR0 in subsequent cycles.
A squashed instruction does not count as a retire.
6.10.2.3.3 Bcache miss or long latency probes cycles

This input counts the number of times the Bcache result was a miss.

Essentially, a long latency probe is a data request from other processes that cause
Bcache misses in a system.
This count is phase shifted three cycles early and thus includes events that occurred
three cycles before the start and before the end of the ProfileMe window.
6.10.2.3.4 Mbox replay traps cycles

This input counts Mbox replay traps.
6.10.2.4 Counter Modes for Aggregate Mode

Table 6–11 shows the counter modes that are used with Aggregate mode.
Table 6–11 Aggregate Mode Performance Counter IPR Input Select Fields
SL0[4]

SL1[3:2]

PCTR0

PCTR1

Retired instructions

Cycle counting

Not defined

Retired instructions

Bcache miss or long latency probes

Cycle counting

Mbox replay traps

6.10.3 ProfileMe Mode Programming Guidelines
Use the following information to program counters in ProfileMe mode.
6.10.3.1 ProfileMe Mode Precautions

Squashed NOPs count as valid fetched instructions.
Counter 1 must be explicitly cleared in the trap handler before each data collection.

6–20

Privileged Architecture Library Code

21264/EV68A Hardware Reference Manual

Performance Counter Support

The CMOV instruction is decomposed into two valid fetched instructions that, in the
absence of stalls, are fetched in consecutive cycles. See Table 6–12 for more information.
Table 6–12 CMOV Decomposed
Instruction

New Instructions

CMOV Ra, Rb--> Rc

CMOV1 Ra, oldRc −−> newRc1
CMOV2 newRc1, Rb −−> newRc2

6.10.3.2 Operation

1. Setup
The following IPRs need to be set up by using PALcode instructions.
IPR Name

Relevant Fields

Meaning

IER_CM

PCEN[1:0]

Enable Interrupts.

PCTX

PPCE

Enable Process Performance Counting or use I_CTL[SPCE].

PCTR_CTL

SL0

Selects Aggregate or ProfileMe mode; set to 1 for ProfileMe mode.

SL1

Selects PCTR0 and PCTR1 counting modes. See Table 6–14 for more information.

PCTR0[19:0]

Set counter 0 value (2**20-N). This selects approximately the Nth valid
fetched instruction as the profiled instruction. Because writes to PCTR0 are
incremented by 0..4, the profiled instruction is one of the (N-4)th to Nth valid
fetched instructions. See Section 6.10.1 for more setup precautions.

PCTR1[19:0]

Set counter 1 value = 0. See Section 6.10.1 for more setup precautions.

SPCE

Enable System Performance Counting or use PCTX[PPCE].

PCT0_EN

Enable performance counter 0.

PCT1_EN

Enable performance counter 1.

I_CTL

2. Open window
PCTR0 accumulates up to 4 valid fetched instructions per cycle when enabled via
I_CTL[PCT0_EN] and either I_CTL[SPCE] or PCTX[PPCE].
The valid fetched instruction that causes PCTR0 to overflow opens the window and
becomes the profiled instruction and covers a period of time near to when the
instruction was in flight. The first cycle of the window is the 5th cycle after the
instruction was fetched. A residual count of up to 7 valid fetched instructions is
accumulated in PCTR0 in the two cycles between overflow and the start of the ProfileMe window. This residual count is returned in I_STAT[overcount(2,0)].
3. Count
If PCTR0 and PCTR1 are enabled, they increment according to modes selected by
SL0 & SL1.
4. End window
The last cycle of the window depends on whether the instruction traps, retires,
aborts, and/or is squashed by the fetcher.
21264/EV68A Hardware Reference Manual

Privileged Architecture Library Code

6–21

Performance Counter Support

For instructions that cause a trap, the last cycle in the window is the 2nd cycle after
the trap. Mispredicted branches are included in this category.
For nontrapping instructions that retire, the last cycle in the window is the 2nd
cycle after the instruction retires.
For instructions that abort, the last cycle in the window is the 2nd cycle after the
trap that caused the abort.
For instructions that are squashed (such as TRAPB), the last cycle in the window is
approximately the 2nd cycle after the squashed instruction would have aborted or
retired.
Every non-squashed valid fetched instruction either aborts or retires, but not both.
In either case, the instruction may also trap.
PCTR0 is disabled from counting until PCTR_CTL is next written.
5. Interrupt PALcode
When ISUM field PC[1:0] is set, execution of PCTR0's or PCTR1's interrupt PALcode is performed.
6. Operating system interrupt handler
The handler should first read the IPRs in Table 6–13 and then write PCTR_CTL to set
up the next interrupt.
Table 6–13 ProfileMe Mode Returned IPR Contents
IPR Name

Meaning

PMPC[63:0] All

Profiled PC.

I_STAT

ICM

Instruction was in a new Icache fill stream.

TRP

Instruction caused a trap and was not in the shadow of
a younger trapping instruction.

MIS

Conditional branch mispredict.

TRAP TYPE

Exception type code.

LSO

Load-store order replay trap.

OVR

Counter 0 overcount.

VAL

Instruction retired valid.

TAK

Branch direction if instruction is a conditional branch.

PM_STALLED

Instruction stalled for at least one cycle between fetch
and map stages of pipeline.

PM_KILLED_BM

Instruction killed during or before cycle in which it
was mapped.

PCTR0[19:0]

Counter 0 value.

PCTR1[19:0]

Counter 1 value.

PCTR_CTL

6–22

Relevant Fields

Privileged Architecture Library Code

21264/EV68A Hardware Reference Manual

Performance Counter Support

6.10.3.3 ProfileMe Counting Mode Description
6.10.3.3.1 Cycle counting

In ProfileMe mode, either counter counts cycles during the window of the profiled
instruction.
6.10.3.3.2 Inum retire delay cycles

This input is used to measure a lower bound on the inum retire delay of the profiled
instruction. The maximum final value of PCTR1 is the length of the ProfileMe window
minus 2.
Counts cycles that a profiled instruction delayed the retire pointer advance during the
ProfileMe window. The 21264/EV68A tracks instructions in the pipeline by allocating
them "inums" near the front of the pipeline. All inums are retired in the order in which
they were allocated at the end of the pipeline.
Inums are allocated in batches of four, so there may be more inums allocated than there
are program instructions in flight. Every inum is retired in order, including those for
aborted instructions.
The "retire pointer" points to the next inum to be retired. An inum retires in the cycle
that the retire pointer advances past the inum.
Let X and Y be consecutive inums in the allocation order. The "inum retire delay" of Y
is [(cycle in which Y retired) – (cycle in which X retired)]. A large inum retire delay
indicates a possible performance bottleneck (for example, an instruction stalled on a
data cache miss).
6.10.3.3.3 Retired instructions cycles

When counting retired instructions in ProfileMe mode, the final count in PCTR0 may
include instructions that retired before the ProfileMe window and may exclude instructions that retired near the end of the ProfileMe window. These discrepancies are caused
by a variable delay between the time that an instruction retires and the time that PCTR0
is incremented for that retire. This discrepancy is in the range of plus or minus 4 retired
instructions.
6.10.3.3.4 Bcache miss or long latency probes cycles

This input counts the number of times the Bcache result was a miss.

This input counts Mbox replay traps.
PCTR1 is enabled to count Mbox replay traps that occur during a window that is the
ProfileMe window phase-shifted one cycle later. The first replay trap counted would be
the 7th cycle after the instruction is fetched.

21264/EV68A Hardware Reference Manual

Privileged Architecture Library Code

6–23

Performance Counter Support

6.10.3.4 Counter Modes for ProfileMe Mode

Table 6–14 shows the counter modes that are used with ProfileMe mode.
Table 6–14 ProfileMe Mode PCTR_CTL Input Select Fields

6–24

SL0[4]

SL1[3:2] PCTR0

PCTR1

Retired instructions

Cycle counting

Inum retire delay

Retired instructions

Bcache miss or long latency probes

Cycle counting

Mbox replay traps

Privileged Architecture Library Code

21264/EV68A Hardware Reference Manual

7
Initialization and Configuration
This chapter provides information on 21264/EV68A-specific microprocessor system
initialization and configuration. It is organized as follows:

•

Power-up reset flow

•

Fault reset flow

•

Energy star certification and sleep mode flow

•

Warm reset flow

•

Array initialization

•

Initialization mode processing

•

External interface initialization

•

Internal processor register (IPR) reset state

•

IEEE 1149.1 test port reset

•

Reset state machine state transitions

•

Phase-locked loop (PLL) functional description

Initialization is controlled by the reset state machine, which is responsible for four
major operations. Table 7–1 describes the four major operations.
Table 7–1 21264/EV68A Reset State Machine Major Operations
Operation

Function

Ramp up

Sequence the PLL input and output dividers (Xdiv and Zdiv) to gradually raise the internal
GCLK frequency and generate time intervals for the PLL to re-establish lock.

BiST/SROM

Receive a synchronous transfer on the ClkFwdRst_H pin in order to start built-in self-test and
SROM load at a predictable GCLK cycle.

Clock forward
interface

Receive a synchronous transfer on the ClkFwdRst_H pin in order to initialize the clock forwarding interface.

Ramp down

Sequence the PLL input and output dividers (Xdiv and Zdiv) to gradually lower the internal
GCLK frequency during sleep mode.

7.1 Power-Up Reset Flow and the Reset_L and DCOK_H Pins
The 21264/EV68A reset sequence is triggered using the two input signals Reset_L and
DCOK_H in a sequence that is described in Section 7.1.1. After Reset_L is deasserted,
the following sequence of operations takes place:

21264/EV68A Hardware Reference Manual

Initialization and Configuration

7–1

Power-Up Reset Flow and the Reset_L and DCOK_H Pins

1. The clock forwarding and system clock ratio configuration information is loaded
onto the 21264/EV68A. See Section 7.1.2.
2. The internal PLL is ramped up to operating frequency.
3. The internal arrays built-in self-test (BiST) is run, followed by Icache initialization
using an external serial ROM (SROM) interface.
The 21264/EV68A systems, unlike the Alpha 21064 and 21164 microprocessor
systems, are required to have an SROM. The SROM provides the only means to
configure the system port, and the SROM pins can be used as a software-controlled
UART.
The Icache must contain PALcode that starts at location 0x780. This code is used to
configure the 21264/EV68A IPRs as necessary before causing any offchip read or
write commands. This allows the 21264/EV68A to be configured to match the
external system implementation.
4. After configuring the 21264/EV68A, control can be transferred to code anywhere
in memory, including the noncacheable regions. The Icache can be flushed by a
write operation to the ITB invalidate-all register after control is transferred. This
transfer of control should be to addresses not loaded in the Icache by the SROM
interface or the Icache may provide unexpected instructions.
5. Typically, any state required by the PALcode is initialized and then the console is
started (switching out of PALmode and into native mode). The console code initializes and configures the system and boots an operating system from an I/O device
such as a disk or the network.
Figure 7–1 shows the sequence of events at power-up, or cold reset. In Figure 7–1, note
the following symbols for constraints and information:
Constraints:
A
B

Setup (A0) and hold (A1) for IRQ’s to be latched by DCOK (2 ns for each).
Enough time for Reset_L to propagate through 5 stages of RESET synchronizer (clocked by the internal framing clock, which is driven by EV6Clk_x). Worst case for the 21264/EV68A would be 5x8x15
= 600 GCLK cycles.
Min = 1 FrameClk cycle.

Information:
a
b
c
d
e
f
g

7–2

8 GCLK cycles from DCOK assertion to first “real” EV6Clk_x cycle.
Approximately 525 GCLK cycles for external framing clock to be sampled and captured.
1 FrameClk_x cycle.
3 FrameClk_x cycles.
Approximately 264 GCLK cycles to prevent first command from appearing too early.
Approximately 700,000 GCLK cycles for BiST + approximately 100,000 GCLK cycles fixed time +
approximately 50,000 GCLK cycles per line of Icache for SROM load.
16 GCLK cycles.

Initialization and Configuration

21264/EV68A Hardware Reference Manual

Power-Up Reset Flow and the Reset_L and DCOK_H Pins

Figure 7–1 Power-Up Timing Sequence
A0 A1

IRQ_H

valid

DCOK_H
a

Reset_L
B

state

WAIT_SETTLE

WAIT_NORMAL

f
RAMP1

RAMP2

WAIT_ClkFwdRst0

WAIT_BiST

WAIT_ClkFwdRst1

RUN

c
e

SromOE_L
ClkFwdRst_H

no min

internal ClkFwdRst
TestStat_H
g

external Clks
End of BiST

BiST Fails

BiST Passes

FM-06486B.FH8

7.1.1 Power Sequencing and Reset State for Signal Pins
Power sequencing and avoiding potential failure mechanisms is described in Section
9.3.
The reset state for the signal pins is listed in Table 7–2.
Table 7–2 Signal Pin Reset State
Signal

Reset State

Signal

Reset State

Bcache
BcAdd_H[23:4]

Tristated

BcCheck_H[15:0]

Tristated

BcTagInClk_H

NA (input)

BcData_H[127:0]

Tristated

BcTagOE_L

Tristated

BcDataInClk_H[7:0]

NA (input)

BcTagOutClk_x

Tristated

BcDataOE_L

Tristated

BcTagParity_H

Tristated

BcDataOutClk_x[3:0]

Tristated

BcTagShared_H

Tristated

BcDataWr_L

Tristated

BcTagValid_H

Tristated

BcLoad_L

Tristated

BcTagWr_L

Tristated

BcTag_H[42:20]

Tristated

BcVref

NA
(I_DC_REF)

BcTagDirty_H

Tristated

System Interface
IRQ_H[5:0]

NA (input)

SysDataInClk_H[7:0]

NA (input)

SysAddIn_L[14:0]

NA (input)

SysDataInValid_L

NA (input)

SysAddInClk_L

NA (input)

SysDataOutClk_L[7:0] Tristated

21264/EV68A Hardware Reference Manual

Initialization and Configuration

7–3

Power-Up Reset Flow and the Reset_L and DCOK_H Pins

Table 7–2 Signal Pin Reset State (Continued)
Signal

Reset State

Signal

Reset State

SysAddOut_L[14:0]

Initially, during power-up reset, state SysDataOutValid_L
is not defined. If not during powerup, preserves previous state. Then,
after the clock forward reset period
(as the external clocks start), signal
driven to NZNOP until the reset
state machine enters RUN, when it
is driven to NOP.

NA (input)

SysAddOutClk_L

Tristated

SysFillValid_L

NA (input)

SysCheck_L[7:0]

Tristated

SysVref

NA
(I_DC_REF)

SysData_L[63:0]

Tristated

Clocks
ClkFwdRst_H

NA (input)

FrameClk_x

NA (input)

ClkIn_H
ClkIn_L

NA (input)

PLL_VDD

NA
(I_DC_REF)

EV6Clk_H
EV6Clk_L

NA (input)

Miscellaneous
DCOK_H

Must be deasserted until dc voltage
reaches proper operating level.

Tck_H

NA (input)

PllBypass_H

NA (input)

Tdi_H

NA (input)

Reset_L

NA (input)

Tdo_H

Unspecified

SromClk_H

Tristated

TestStat_H

Tristated

SromData_H

NA (input)

Tms_H

NA (input)

SromOE_L

Tristated

Trst_L

NA (input)

In addition, as power is being ramped, Reset_L must be asserted — this allows the
21264/EV68A to reset internal state. Once the target voltage levels are attained, systems should assert DCOK_H. This indicates to the 21264/EV68A that internal logic
functions can be evaluated correctly and that the power-up sequence should be continued. Prior to DCOK_H being asserted, the logic internal to the 21264/EV68A is being
reset and the internal clock network is running (either clocked by the PLL VCO, which
is at a nominal speed, or by ClkIn_H, if the PLL is bypassed).
The reset state machine is in state WAIT_SETTLE.

7.1.2 Clock Forwarding and System Clock Ratio Configuration
When DCOK_H is asserted, the 21264/EV68A samples several pins and latches in
some initialization state, including the value of the PLL Ydiv divisor, which specifies
the ratio of the system clock to the internal clock (see Section 7.11.2.3), and enables the
charge pump on the phase-locked loop.

7–4

Initialization and Configuration

21264/EV68A Hardware Reference Manual

Power-Up Reset Flow and the Reset_L and DCOK_H Pins

Table 7–3 summarizes the pins and the suggested/required initialization state. Most of
this information is supplied by placing (switch-selectable or hardwired) weak pull-ups
or pull-downs on the IRQ_H pins. The IRQ_H pins are sampled on the rising edge of
DCOK_H, during which time the 21264/EV68A is in reset and is not generating any
system activity. During normal operation, the IRQ_H pins supply interrupt requests to
the 21264/EV68A.
It is possible to disable the 21264/EV68A PLL and source GCLK directly from
ClkIn_x. This mode is selected via PllBypass_H. The 21264/EV68A still produces a
divided-down clock on EV6Clk_x; this output clock, which tracks GCLK, can be used
in a feedback loop to generate a locked input clock via an external PLL. The input
clock can be locked against a slower speed system reference clock.
Table 7–3 Pin Signal Names and Initialization State
Signal Name

Sample Time

Function

Value

PllBypass_H

Continuous input Select ClkIn_x onto GCLK instead of internal
PLL.

0 Bypass1
1 Use PLL

ClkFwdRst_H Sampling method
according to
IRQ_H[4]

—

Reset_L

Continuous input

—

IRQ_H[5]

Rising edge of
DCOK_H

Select 1:1 FrameClk mode.
Internal FrameClk can be generated two ways:

0 Sample with
FrameClk_H
1 Use a copy of
EV6Clk_H

1
2

By sampling FrameClk_H. Used if
FrameClk_H is slower than ClkIn_H.
As a direct copy of EV6Clk_H. Used if
FrameClk_H is the same frequency as
ClkIn_H or is DC.

IRQ_H[4]

Rising edge of
DCOK_H

Select method of sampling ClkFwdRst_H to
produce internal ClkFwdRst — either with
external or internal copy of FrameClk_x.

0 Sample with External FrameClk_x
1 Sample with Internal Frameclk

IRQ_H[3:0]

Rising edge of
DCOK_H

Select Y div divisor value. This is the dividedown factor between GCLK and EV6Clk_x.

IRQ_H[3:0]

When the PLL is in use and the 21264/EV68A is
ramped-up to full speed, the VCO adjusts in
order to phase-align (and rate-match) EV6Clk_x
to ClkIn_x. When the PLL is not in use, and
ClkIn_x is bypassed onto GCLK, EV6Clk_x is
slower than ClkIn_x by the divisor Ydiv.

21264/EV68A Hardware Reference Manual

0011
0100
0101
0110
0111
0000
1000
1001
1010
1011
1100
1101
1110
1111

Divisor
3
4
5
6
7
8
9
10
11
12
13
14
15
16

Initialization and Configuration

7–5

Power-Up Reset Flow and the Reset_L and DCOK_H Pins

Table 7–3 Pin Signal Names and Initialization State (Continued)
Signal Name

Sample Time

Function

Value

DCOK_H

Continuous input

When deasserted, initializes the internal 21264/
EV68A reset state machine and keeps the PLL
internal oscillator running at a nominal speed.
Assertion, which implies power to the 21264/
EV68A is good, causes configuration information to be sampled.

—

The maximum permissible instantaneous change in ClkIn_x frequency is 333 MHz (to prevent current spikes).

7.1.3 PLL Ramp Up
After the configuration is loaded through the IRQ_H pins, the next phase in the power
up flow is the internal PLL ramp up sequence. Ramping up of the PLL is required to
guarantee that the dynamic change in frequency will not cause the supply on the 21264/
EV68A to fall due to the supply loop inductance. Clock control circuitry steps GCLK
from power-up/reset clocking to 1/16th operating frequency, to ½ operating frequency,
and finally normal operating frequency.
After the assertion of DCOK_H, the 21264/EV68A waits for the deassertion of
Reset_L from the system while the PLL attempts to achieve a lock. The PLL internal
ramp dividers are set to divide down the input clock by 16 and the PLL attempts to
achieve lock against an effective input frequency of ClkIn_x/16. Once lock is
achieved, the actual internal frequency (GCLK) is ClkIn_x*(Ydiv divisor value)/16.
There should be a minimum delay of 100 ms between the assertion of DCOK_H and
the deassertion of Reset_L to allow for this locking The reset state machine is in the
WAIT_NOMINAL state.
After the deassertion of Reset_L, the reset state machine goes into the RAMP1 state.
The 21264/EV68A ramps the internal frequency, by changing the effective input frequency of the PLL to ClkIn_x/2 for a sufficient lock interval (at most, 20 µs at 400
MHz). The state machine then goes into the RAMP2 state, changing the effective input
frequency to ClkIn/1 for an additional lock interval (about 20 µs). The lock periods are
generated by the internal duration counter, which is driven by GCLK. The counter
counts 4108 GCLK cycles during the ClkIn_x/2 lock interval. Note that GCLK is produced by the output of the PLL, which is locking to an input clock which is 1/2 of the
operating frequency — therefore, the 4108 cycle interval constitutes a 6-20 µs interval
when the operating frequency is 400–1250 MHz. Then, the counter counts 8205 GCLK
cycles during the ClkIn_x/1 lock interval.

7.1.4 BiST and SROM Load and the TestStat_H Pin
The 21264/EV68A uses the deassertion of ClkFwdRst_H (which must be deasserted
for a minimum of one FrameClk_H cycle and then reasserted) to begin built-in selftest (BiST). The reset state machine goes into the WAIT_BiST state. Details on BiST
are given in Chapter 11. The power-up BiST lasts approximately 700,000 cycles. The
result of the self-test is made available on the TestStat_H pin. The pin is forced low by
the system reset. It is then forced high during BiST.

7–6

Initialization and Configuration

21264/EV68A Hardware Reference Manual

Power-Up Reset Flow and the Reset_L and DCOK_H Pins

As BiST completes, the TestStat_H pin is held low for 16 GCLK cycles. Then, if BiST
succeeds, the pin remains low. Otherwise, it is asserted. After successfully completing
BiST, the 21264/EV68A then performs the SROM load sequence (described in Chapter
11). After the SROM load sequence is finished, the 21264/EV68A deasserts
SromOE_L.

7.1.5 Clock Forward Reset and System Interface Initialization
After the deassertion of SromOE_L, the reset state machine enters the
WAIT_ClkFwdRst1 state, where the 21264/EV68A waits for the system to deassert
ClkFwdReset_H. The 21264/EV68A samples the deasserting edge of
ClkFwdReset_H to take synchronous actions. It uses this synchronous event to reset
the clock forwarding interface, start the outgoing clocks, and deassert internal reset.
The chip then waits 264 cycles before issuing commands. The reset state machine is
then in RUN and the 21264/EV68A begins fetching code at address 0x780.
Table 7–4 lists signals relevant to the power-up flow, provides a short description of
each, and any relevant constraints.
Table 7–4 Power-Up Flow Signals and Their Constraints
Signal Name

Description

Constraint

ClkIn_x

Differential clocks that are
inputs to PLL or are
bypassed onto GCLK
directly

Clocks must be running before DCOK_H is
asserted.

PLL_VDD

VDD supply to PLL

PLL_VDD must lead VDD.

VDD

VDD supply to the 21264/
EV68A chip logic (except
PLL)

—

DCOK_H

Logic signal to the 21264/
EV68A that the VDD supply is good

—

Reset_L

RESET pin asserted by
SYSTEM to the 21264/
EV68A

Reset_L must be asserted prior to DCOK_H and
must remain asserted for at least 100 ms after
DCOK_H is asserted. This allows for PLL settling
time. Deassertion of Reset_L causes the 21264/
EV68A to ramp divisors to their final value and
begin BiST.

ClkFwdRst_H
Deassertion #1

Signal asserted by SYSTEM to synchronously
commence built-in self-test
and SROM load

ClkFwdRst_H must be deasserted after PLL has
achieved its lock in its final divisor value (about 20
µs). The deassertion causes built-in self-test to
begin on an internal clock cycle that corresponds to
one framing clock cycle after ClkFwdRst_H is
deasserted. ClkFwdRst_H can be asserted after
one frame clock cycle. See Figure 7–1.

ClkFwdRst_H
Deassertion #2

Signal asserted by SYSClkFwdRst_H must be deasserted when the Cbox
TEM to initialize and reset has loaded configuration information. This occurs
clock forwarding interfaces as the first part of the serial ROM load, after BiST
is run. Once ClkFwdRst_H is deasserted, the
interface is initialized and can receive probe
requests from the 21264/EV68A.

21264/EV68A Hardware Reference Manual

Initialization and Configuration

7–7

Fault Reset Flow

7.2 Fault Reset Flow
The fault reset sequence of operation is triggered by the assertion of the ClkFwdRst_H
signal line. Figure 7–2 shows the fault reset sequence of operation. The reset state
machine is initially in RUN state. ClkFwdRst_H is asserted by the system, which
causes the state machine to transition to the WAIT_FAULT_RESET state.
The 21264/EV68A internally resets a minimum amount of internal state. Note the
effects of that reset on the IPRs in Table 7–5.
Table 7–5 Effect on IPRs After Fault Reset
IPR

After Reset

PAL_BASE

Maintained (not reset)

I_CTL

Bit value = 3 (both Icaches are enabled)

PCTX[FPE]

Set

WRITE_MANY Cleared (That is, the WRITE_MANY chain is initialized and the Bcache is turned off.)
EXC_ADDR

Set to an address that is close to the PC

The 21264/EV68A then waits for ClkFwdRst_H to deassert twice:

•

One deassert to transition directly to the WAIT_ClkFwdRst1 state without performing any BiST

•

One deassert to initialize the clock forwarding interface

The 21264/EV68A then begins fetching code at PAL_BASE + 0x780.
Figure 7–2 shows the fault reset sequence of operation. In Figure 7–2, note the following symbols for constraints and information:
Constraints:
A

Min = 1 FrameClk_x cycle

Information:
a
b
c
e
f
g

7–8

Approximately 264 GCLK cycles
Approximately 525 GCLK cycles for external framing clock to be sampled and captured
1 FrameClk_x cycle plus 2 GCLK cycles
Next FrameClk_x rising edge
3 FrameClk_x cycles
Approximately 264 GCLK cycles to prevent first command from appearing too early

Initialization and Configuration

21264/EV68A Hardware Reference Manual

Energy Star Certification and Sleep Mode Flow

Figure 7–2 Fault Reset Sequence of Operation
internal clks aligned

state

RUN

WAIT_ClkFwdRst0

WAIT_FAULT_RESET

WAIT_ClkFwdRst1

RUN

SromOE_L
ClkFwdRst_H

no min

internal ClkFwdRst
external Clks
FM-06488B.AI4

7.3 Energy Star Certification and Sleep Mode Flow
The 21264/EV68A is Energy Star compliant. Energy Star is a program administered by
the Environmental Protection Agency to reduce energy consumption. For compliance,
a computer must automatically enter a low power sleep mode using 30 watts or less
after a specified period of inactivity. When the system is awakened, the user shall be
returned automatically to the same situation that existed prior to entering sleep mode.
During normal operation, the 21264/EV68A encounters inactive periods and enters a
mode that saves the entire active processor state to memory.
The PALcode is responsible for saving all necessary state to DRAM and flushing the
caches.
The sleep mode sequence of operations is triggered by the PALcode twice performing a
HW_MTPR to the Ibox SLEEP IPR. The first write prevents the assertion of
ClkFwdRst_H from fault-resetting the chip.
The PALcode then informs the system, in an implementation-dependent way, that it
may assert ClkFwdRst_H.
On the second HW_MTPR to the SLEEP IPR, the PLL begins to ramp down and the
21264/EV68A can then respond to the ClkFwdRst_H that was asserted by the system,
causing the outgoing clocks from the 21264/EV68A to stop.
The PLL ramp-down sequence takes exactly the same amount of time as the ramp up
sequence described in Section 7.1.3. The same internal duration counter is used and the
reset state machine transitions through the DOWN1, DOWN2, and DOWN3 states
which have similar PLL divisor ratios and clock speeds to the RAMP2, RAMP1, and
WAIT_NOMINAL states.

21264/EV68A Hardware Reference Manual

Initialization and Configuration

7–9

Energy Star Certification and Sleep Mode Flow

After the PLL has finished ramping down, the reset state machine enters the
WAIT_INTERRUPT state. Note the effects of the entry into that state on the IPRs
listed in Table 7–6.
Table 7–6 Effect on IPRs After Transition Through Sleep Mode
IPR

Effects After Transition Through Sleep Mode

PAL_BASE

Maintained (not reset)

I_CTL

Bit value = 3 (both Icaches are enabled)

PCTX[FPE]

Set

WRITE_MANY Cleared (That is, the WRITE_MANY chain is initialized and the Bcache is
turned off.)

Note that Interrupt enables are maintained during sleep mode, enabling the 21264/
EV68A to wake up. The 21264/EV68A waits for either an unmasked clock interrupt or
an unmasked device interrupt from the system.
When an enabled interrupt occurs, the PLL ramps back to full frequency. Subsequent to
that, the 21264/EV68A performs a built-in self-initialization (BiSI), a shortened built-in
self-test, which initializes the internal arrayed structures. The SROM is not reloaded.
Instead, the 21264/EV68A begins fetching code from the system at address
PAL_BASE + 0x780.
Figure 7–3 shows the sleep mode sequence of operations. In Figure 7–3, note the following constraint and informational symbols:
Constraints:
A

Min = 1 FrameClk_x cycle

Informational symbols:
a
b
c
d
e
f
g

7–10

Approximately 525 GCLK cycles for external framing clock to be sampled and captured
Next FrameClk_x rising edge
1 FrameClk_x cycle
3 FrameClk_x cycles
Approximately 264 GCLK cycles to prevent first command from appearing too early
Approximately 8192 GCLK cycles for BiSI
16 GCLK cycles

Initialization and Configuration

21264/EV68A Hardware Reference Manual

Warm Reset Flow

Figure 7–3 Sleep Mode Sequence of Operation
internal clks

state

RUN

DOWN1

DOWN2

DOWN3

WAIT_INTR

RAMP1

RAMP2

WAIT_ClkFwdRst0

SLEEP IPR

WAIT_BiSI

WAIT_ClkFwdRst1

RUN

Wake-up interrupt
SromOE_L
ClkFwdRst_H

no min

TestStat_H
internal ClkFwdRst
external Clks
FM-06487A.AI4

Table 7–7 describes each signal and constraint for the sleep mode sequence.
Table 7–7 Signals and Constraints for the Sleep Mode Sequence
Signal Name

Description

Constraint

ClkFwdRst_H

Signal asserted by the system to
ClkFwdRst_H must be asserted by the system
initialize and reset clock forwarding when entering sleep mode. The system deasserts
interfaces
ClkFwdRst_H no sooner than one FrameClk_H
cycle after sourcing an interrupt to the 21264/
EV68A.

Forwarded clocks

Bit clocks forwarded to/from the
21264/EV68A

Clocks stop running under ClkFwdRst_H.

System interrupt

Asynchronous interrupt which
causes the 21264/EV68A to exit
sleep mode

—

7.4 Warm Reset Flow
The warm reset sequence of operation is triggered by the assertion of the Reset_L signal line. The reset state machine is initially in RUN state. The 21264/EV68A then, by
default, ramps down the PLL (similar to the sleep flow sequence) and the reset state
machine ends up in the WAIT_RESET state.
Note the effects of entry into that state on the IPRs listed in Table 7–8.
Table 7–8 Effect on IPRs After Warm Reset
IPR

Effects After Warm Reset

PAL_BASE

Cleared

I_CTL

Cleared

PCTX[FPE]

Set

WRITE_MANY Cleared (That is, the WRITE_MANY chain is initialized and the Bcache is
turned off.)

21264/EV68A Hardware Reference Manual

Initialization and Configuration

7–11

Array Initialization

The 21264/EV68A waits until Reset_L is deasserted before transitioning from the
WAIT_RESET state. The 21264/EV68A ramps up the PLL until the state machine
enters the WAIT_ClkFwdRst0 state. Note that the system must assert ClkFwdRst_H
before the state machine enters the WAIT_ClkFwdRst0 state. Then, similarly to the
other flows, SromOE_L is asserted and the system waits for the deassertion of
ClkFwdRst_H.
On the deassertion of ClkFwdRst_H, the 21264/EV68A performs BiST and the SROM
loading procedure.
After BiST and SROM loading have completed, SromOE_L deasserts and the 21264/
EV68A waits for ClkFwdRst_H to deassert before starting the external clocks and, like
the other flows, waits for 264 cycles before starting instructions.

7.5 Array Initialization
The following arrays are initialized by BiST:

•
•
•

Icache and Icache tag
Dcache, Dcache tag, and Duplicate Dcache tag
Branch history table

The external second-level cache (Bcache) is disabled by Reset_L.
The Bcache must be initialized by PALcode before it is enabled.

7.6 Initialization Mode Processing
The initialization mode allows the 21264/EV68A to generate and manipulate cache
blocks before the system interface has been initialized. Within the 21264/EV68A, the
Cbox configuration registers are divided into the WRITE_ONCE and the
WRITE_MANY shift register chains (see Sections 5.4.3 and 5.4.4). The
WRITE_ONCE chain is loaded from the SROM during reset processing, and contains
information such as the clock forwarding setup values. The WRITE_MANY chain can
be written many times using MTPR instructions.
The WRITE_MANY chain contains the following CSRs that are important to initialization mode, which must be set to the values in Table 7–9 to initialize the Bcache.
Table 7–9 WRITE_MANY Chain CSR Values for Bcache Initialization

7–12

WRITE_MANY Chain CSRs

Required Value at Initialization Mode

BC_ENABLE

1
The duplicate bits for BC_ENABLE in [14:12] must
be 0 during initialization mode.

BC_SIZE[3:0]

The exact size or maximum size of the Bcache.

INVAL_TO_DIRTY_ENABLE[1:0]

SET_DIRTY_ENABLE[2:0]

INIT_MODE

Initialization and Configuration

21264/EV68A Hardware Reference Manual

Initialization Mode Processing

Table 7–9 WRITE_MANY Chain CSR Values for Bcache Initialization (Continued)
WRITE_MANY Chain CSRs

Required Value at Initialization Mode

EVICT_ENABLE

BC_WRT_STS[3:0]

BC_BANK_ENABLE

Except for INIT_MODE, all the CSR registers have been described in earlier sections.
When asserted, INIT_MODE has the following behavior:

•

Cache block updates to the Dcache set the block to the Clean state.

•

Updates to the Bcache use the BC_WRT_STS[3:0] bits.

•

WrVictimBlk command generation to the system interface are squashed.

Using the INVAL_TO_DIRTY_ENABLE and INIT_MODE registers, initialization
code loaded from the SROM can generate and delete blocks inside the 21264/EV68A
without system interaction. This behavior is very useful for initialization and startup
processing, when the system interfaces are not fully functional. Figure 7–4 shows a
code example for initializing Bcache.
Figure 7–4 Example for Initializing Bcache
Reset chip and load Icache with this code
set init_mode
;now all WrVictims are ignored
;bc_enable_a
1
;zeroblk_enable_a
1
;set_dirty_enable_a
0
;init_mode_a
1
;enable_evict_a
0
;bc_wrt_sts_a
0
;bc_bank_enable_a
0
;bc_size_a
15
;now all writes to Bcache actually invalidate
;the Bcache. (if space was needed for scratch
;pad, the status bits could just as
;well be Valid)
for 2 X bc_size
{ WH64 address }

;This loop generates legal ECC data, and
;invalidate tags which are written to the
;Bcache for all but the final 64KB of address.

turn_off_bcache:

;bc_enable_a
;init_mode_a
;bc_size_a
;zeroblk_enable_a
;enable_evict_a
;set_dirty_enable_a
;bc_bank_enable_a
;bc_wrt_sts_a

21264/EV68A Hardware Reference Manual

0
0
0
1
0
0
0
0

Initialization and Configuration

7–13

External Interface Initialization

SweepMemory:

;Write good parity/ecc to memory by
; writing a all memory locations. This is
;done by WH64 of memory addresses

turn_on_bcache:

;bc_enable_a
0
;bc_size_a
Actual Bcache size
;zeroblk_enable_a
3
;set_dirty_enable_a
6
;init_mode_a
0
;enable_evict_a
0
;bc_wrt_sts_a
0
;bc_bank_enable_a
0
;This loop generates legal ECC data, and
;invalidate tags which are written to the
;Bcache for all but the final 64KB of address.

for 2 X bc_size
{ WH64 address }
for 2 X dcache size
{ ECB address }
(done)

;and cleans up the Dcache also.

In addition to initialization, the dynamic programming ability of the WRITE_MANY
chain provides the basic tools to build various other software flows such as dynamically
changing the Bcache enable/size parameters for performance testing.

7.7 External Interface Initialization
After reset, the system interface is in the default configuration dictated by the reset state
of the IPR bits that select the configuration options.
The response to system interface commands and internally generated memory accesses
is determined by this default configuration. System environments that are not compatible with the default configuration must use the SROM Icache load feature to initially
load and execute a PALcode program to configure the external system interface unit
IPRs as needed.

7.8 Internal Processor Register Power-Up Reset State
Many IPR bits are not initialized by reset. They are located in error-reporting registers
and other IPR states. They must be initialized by initialization PALcode. Tables 7–5,
7–6, and 7–8, list the effects on IPRs by fault reset, transition through sleep mode, and
warm reset, respectively. Table 7–10 lists the state of all internal processor registers
(IPRs) immediately following power-up reset. The table also specifies which registers
need to be initialized by power-up PALcode.
Table 7–10 Internal Processor Registers at Power-Up Reset State
Mnemonic

Reset State Comments

Ibox IPRs
ITB_TAG

ITB tag array write

—

ITB_PTE

ITB PTE array write

—

7–14

Initialization and Configuration

21264/EV68A Hardware Reference Manual

Internal Processor Register Power-Up Reset State

Table 7–10 Internal Processor Registers at Power-Up Reset State (Continued)
Mnemonic

Reset State Comments

ITB_IAP

ITB invalidate-all (ASM=0)

—

ITB_IA

ITB invalidate all

Must be written to in PALcode.

ITB_IS

ITB invalidate single

—

PMPC

ProfileMePC

—

EXC_ADDR

Exception address

—

IVA_FORM

Instruction VA format

—

IER_CM

Interrupt enable current mode

Must be written to in PALcode.

SIRR

Software interrupt request

—

ISUM

Interrupt summary

—

HW_INT_CLR

Hardware interrupt clear

Must be cleared in PALcode.

EXC_SUM

Exception summary

—

PAL_BASE

PAL base address

Cleared

—

I_CTL

Ibox control

I_STAT

Ibox status

Must be cleared in PALcode.

IC_FLUSH

Icache flush

—

CLR_MAP

Clear virtual-to-physical map

—

SLEEP

Sleep mode

—

PCTX

Ibox process context

PCTR_CTL

Performance counter control

Must be cleared in PALcode.

Cycle counter

Must be cleared in PALcode.

CC_CTL

Cycle counter control

Must be cleared in PALcode.

Virtual address

—

VA_FORM

Virtual address format

—

VA_CTL

Virtual address control

Must be cleared in PALcode.

DTB_TAG0

DTB tag array write 0

Cleared

—

DTB_TAG1

DTB tag array write 1

Cleared

—

DTB_PTE0

DTB PTE array write 0

Cleared

—

DTB_PTE1

DTB PTE array write 1

Cleared

—

DTB_ALTMODE

DTB alternate processor mode

PALcode must initialize.

DTB_IAP

DTB invalidate all process
ASM = 0

—

DTB_IA

DTB invalidate all process

Must be written to in PALcode.

IC_EN = 3 All other bits are cleared on reset.

PCTX[FPE] is set. All other bits are X.

Ebox IPRs

Mbox IPRs

21264/EV68A Hardware Reference Manual

Initialization and Configuration

7–15

IEEE 1149.1 Test Port Reset

Table 7–10 Internal Processor Registers at Power-Up Reset State (Continued)
Mnemonic

Reset State Comments

DTB_IS0

DTB invalidate single (array 0)

—

DTB_IS1

DTB invalidate single (array 1)

—

DTB_ASN0

DTB address space number 0

Cleared

—

DTB_ASN1

DTB address space number 1

Cleared

—

MM_STAT

Memory management status

—

M_CTL

Mbox control

Cleared

—

DC_CTL

Dcache control

DC_STAT

Dcache status

Must be cleared in PALcode.

C_DATA

Cbox data

Must be read in PALcode.

C_SHFT

Cbox shift control

—

DC_CTL[7:2] are cleared at reset.
DC_CTL[1:0] are set at power up.

Cbox IPRs

7.9 IEEE 1149.1 Test Port Reset
Signal Trst_L must be asserted when powering up the 21264/EV68A. Trst_L must not
be deasserted prior to assertion of DCOK_H. Trst_L can remain asserted during normal operation of the 21264/EV68A.

7.10 Reset State Machine
The state diagram in Figure 7–5 summarizes how the 21264/EV68A transitions into
running code. Each state is described in Table 7–11. Table 7–11 describes outputs and
approximate state transition equations. Note that there are implicit transitions from
each state to an appropriate down-ramp state when Reset_L is asserted.

7–16

Initialization and Configuration

21264/EV68A Hardware Reference Manual

Reset State Machine

Figure 7–5 21264/EV68A Reset State Machine State Diagram
PLL Ramp Up
Reset_L
deasserted

RAMP1
[2,4]

Counter
finished

WAIT_
NOMINAL
[16,32]

DCOK_H
asserted

RAMP2
[1,2]

Counter
finished

WAIT_
SETTLE
[16,32]

WAIT_ClkFwd
Rst0

Counter
finished

ClkFwdRst_H

Enabled
Interrupt

COLD

Reset_L
asserted

WAIT_
INTERRUPT

Out of
Sleep
Mode

deasserted

Reset_L
deasserted

Reset_L
asserted

Out of
FAULT_
RESET*

WAIT_
BiST

FAULT_
RESET

WAIT_
BiSI

BiST
finished

WAIT_
RESET

BiSI
finished

ClkFwdRst_H

asserted
Numbers in "[,]" are
Xdiv and Zdiv divisors,
respectively

Counter
finished &
Sleep Mode

Counter
finished &
not Sleep Mode

WAIT_ClkFwd
Rst1

*No BiST/BiSI
on recovery from Fault
Reset

PLL Ramp Down

DOWN3
[16,32]

Counter
finished

DOWN2
[2,4]

ClkFwdRst_H

deasserted

RUN
Counter
finished

DOWN1
[1,2]

Sleep Mode
or Reset_L
asserted
LKG-10982A-98WF

Table 7–11 21264/EV68A Reset State Machine State Descriptions
State Name

Description

COLD

Chip cold. Transitioned to WAIT_SETTLE with assertion of Reset_L, PLL_VDD, and
VDD.

WAIT_SETTLE

PLL_VDD asserted; PLL at minimum frequency.

WAIT_NOMINAL

Triggered by assertion of DCOK_H. PLL achieves a lock at Xdiv and Zdiv divisors equal
16 and 32, respectively.

RAMP1

Triggered by Reset_L deassertion; Xdiv and Zdiv divisors are changed to 2 and 4, respectively, increasing the internal GCLK frequency. An internal duration counter is initialized to count 4108 GCLK cycles.

21264/EV68A Hardware Reference Manual

Initialization and Configuration

7–17

Reset State Machine

Table 7–11 21264/EV68A Reset State Machine State Descriptions (Continued)
State Name

Description

RAMP2

Triggered by the duration counter reaching 4108 cycles, the X div and Zdiv divisors are
changed to 1 and 2, respectively, and the frequency is increased. The duration counter is
reloaded to count 8205-cycles.

WAIT_ClkFwdRst0

Triggered by the duration counter reaching 8205 cycles (or by the deassertion of
Reset_L while in the WAIT_RESET state). 21264/EV68A asserts SromOE_L and
waits for SYSTEM to deassert ClkFwdReset_H. The deassertion must be synchronous
to a falling edge of FrameClk_H. 21264/EV68A uses this deassertion to begin BiST
and SROM load at a predictable time. 21264/EV68A samples and generates an internal,
aligned copy of FrameClk_H, and, in turn, uses this clock to sample ClkFwdReset_H.

WAIT_BiST

BiST and SROM load is started. The SROM first loads the Write-once chain and then
reads the number of bits of Icache data to load.

WAIT_BiSI

This state is entered when 'waking up' from sleep mode. 21264/EV68A receives an
external interrupt, ramps the PLL, synchronously samples a transition on
ClkFwdReset_H, and runs built-in self-initialization to clear the internal caches. Builtin self-test is not performed and the SROM is not loaded.

WAIT_ClkFwdRst1

Entered when the appropriate amount of BiST and SROM loading has been completed.
21264/EV68A deasserts SromOE_L and waits for SYSTEM to deassert
ClkFwdReset_H. The deassertion must be synchronous to a rising edge of
FrameClk_H. 21264/EV68A uses this synchronous event to reset the clock forwarding
interface and deassert internal reset. 21264/EV68A subsequently begins running code
(either preloaded in the SROM or located in memory) and begins system transactions.

RUN

Chip is running software, interface is reset, and system transactions can be processed.
From power-up, the Icache sets are enabled and contain bootstrap code loaded from the
SROM; 21264/EV68A executes code from Icache. From wake-up, the Icache sets are
disabled and 21264/EV68A fetches and executes code from DRAM.

WAIT_RESET

Triggered by duration counter reaching 264 cycles, or when Reset_L is asserted when in
WAIT_INTERRUPT state. 21264/EV68A waits in this state until Reset_L is deasserted,
at which point, the PLL starts to ramp up again.

FAULT_RESET

ClkFwdReset is asserted while the 21264/EV68A is running. The 21264/EV68A internally resets a minimum amount of internal state, waits for clock forward reset deassertion, and begins fetching code at PAL_BASE + 0x780.

DOWN1

21264/EV68A was in a state in which GCLK was at its highest speed and Reset_L was
asserted. Internal chip functions are reset and the internal duration counter is set to 8205
cycles. The purpose of this sequence is to down-ramp the clocks in anticipation of power
being removed. If power is not removed (that is, reset is being toggled), 21264/EV68A
ramps the clocks back to the original speed.
This state is also entered when software writes the I_CTL internal processor register to
sleep mode.

7–18

Initialization and Configuration

21264/EV68A Hardware Reference Manual

Phase-Lock Loop (PLL) Functional Description

Table 7–11 21264/EV68A Reset State Machine State Descriptions (Continued)
State Name

Description

DOWN2

Triggered by duration counter reaching 8205 cycles, the PLL ramps GCLK frequency
down by the first divider ratio (Xdiv and Zdiv equal 2 and 4, respectively). This has the
effect of halving the GCLK frequency. The duration counter is set to 4108 cycles.

DOWN3

Triggered by duration counter reaching 4108 cycles, the PLL ramps frequency down by
the second divider ratio (Xdiv and Zdiv equal 16 and 32, respectively). This has the
effect of reducing the frequency by a factor of 16 (of the original frequency). The internal counter is set to 264 cycles.

WAIT_INTERRUPT

Triggered by duration counter reaching 264 cycles, the 21264/EV68A waits for either an
unmasked clock interrupt or unmasked device interrupt from system. The interrupts are
wired to the interrupt request and enable internal registers. When an enabled interrupt
occurs, the PLL ramps back to full frequency. Subsequent to that, the built-in self-init
(BiSI) initializes arrayed structures. The SROM is not reloaded; instead, the 21264/
EV68A begins fetching code from the SYSTEM.

7.11 Phase-Lock Loop (PLL) Functional Description
The PLL multiplies the clock frequency of a differential input reference clock and
aligns the phase of its output to that differential input clock. Thus, the 21264/EV68A
can communicate synchronously on clock boundaries with clock periods that are
defined by the system.

7.11.1

Differential Reference Clocks
A skew-controlled, ac-coupled differential clock is provided to the PLL by way of
ClkIn_x . ClkIn_x are input signals to a differential amplifier. The frequency of
ClkIn_x can range from 80 MHz to 200 MHz. ClkIn_x can be sourced by a variety of
components that include PECL fanout parts or system PLLs. ClkIn_x are also the primary clock source for the 21264/EV68A when in PLL bypass mode.

7.11.2

PLL Output Clocks
The following sections summarize the PLL output clocks.

7.11.2.1

GCLK

The PLL provides an output clock, GCLK, with a frequency that can range from 400
MHz to 1.25 GHz under full-speed conditions. GCLK is the nominal onchip clock that
is distributed to the entire 21264/EV68A chip.
7.11.2.2

Differential 21264/EV68A Clocks

The EV6Clk_x output pads provide an external test point to measure the PLL phase
alignment. They do not provide a clock source. EV6Clk_x are square-wave signals
that drive rail-to-rail continually from 0 to VDD.
7.11.2.3

Nominal Operating Frequency

Under normal operating conditions, the frequency of the PLL output clock, GCLK, is a
simple function of the Y div divider value.

21264/EV68A Hardware Reference Manual

Initialization and Configuration

7–19

Phase-Lock Loop (PLL) Functional Description

Table 7–12 shows the allowable ClkIn_x frequencies for a given operating frequency
of the 21264/EV68A and the Ydiv divider. For example, to set the 21264/EV68A GCLK
frequency to 500 MHz with a ClkIn_x frequency of 166.7 MHz, the system must select
a Ydiv divider of 3 by placing the value 00112 on pins IRQ_H[3:0].
Table 7–12 Differential Reference Clock Frequencies in Full-Speed Lock
GCLK

Reference Clock Frequency (MHz) for Ydiv Dividers1

Period Frequency
(ns)
(MHz)

2.5

400

133.3 100

2.4

416.7

2.3
2.2

—

138.9 104.2 83.3

—

434.8

144.9 108.7 87.0

—

454.5

151.2 113.6 90.9

—

2.1

476.2

158.7 119.0 95.2

—

2.0

500

166.7 125.0 100

83.3

—

1.9

526.3

175.4 131.6 105.3 87.7

—

1.8

555.6

185.2 138.9 111.1 92.6

—

1.7

588.2

196.1 147.1 117.6 98.0

84.0

—

1.6

625

—

156.3 125.0 104.2 89.3

—

1.5

666.7

—

166.7 133.3 111.1 95.2

83.3

—

1.4

714.3

—

178.6 142.9 119.1 102.0 89.3

—

1.3

769.2

—

192.3 153.8 128.2 109.9 96.2

85.5

—

1.2

833.3

—

166.7 138.9 119.0 104.2 92.6

83.3

—

1.1

909.1

—

181.8 151.5 129.9 113.6 101

90.9

—

1.0

1000

—

200

166.7 142.9 125

90.9

83.3

—

0.9

1111.1

—

185.2 158.7 138.9 123.5 111.1 101.0 92.6

85.5

—

0.8

1250

—

113.6 104.2 96.2

89.3

83.3 —

111,1 100

178.6 156.3 138.9 125

Divider 16 is out of range for the 21264/EV68A and reserved for future use. Valid reference clock
(ClkIn_x) frequencies for the 21264/EV68A are specified in the range from 80 to 200. Divider values
that are out of that range are displayed as a dash “—”.
2
Dividers of 1 and 2 are to be used only in a PLL test mode.
7.11.2.4

Power-Up/Reset Clocking

During the power-up/reset sequence, when not in PLL bypass mode, there may be a
period of time when ClkIn_x is not yet running, but there is a voltage on PLL_VDD.
The signal DCOK_H is deasserted until power is good throughout the system. The
10% to 90% rise time of DCOK_H should be less than 2 ns. The deasserted state of
DCOK_H and the presence of PLL_VDD causes the PLL to generate a global clock
that is distributed throughout the 21264/EV68A with a frequency range of 1 MHz to
500 MHz. The presence of the global clock during this period avoids permanent damage to the 21264/EV68A.

7–20

Initialization and Configuration

21264/EV68A Hardware Reference Manual

8
Error Detection and Error Handling
This chapter gives an overview of the 21264/EV68A error detection and error handling
mechanisms, and is organized as follows:

•

Data error correction code

•

Icache data or tag parity error

•

Dcache tag parity error

•

Dcache data correctable ECC error

•

Dcache store second error

•

Dcache duplicate tag parity error

•

Bcache tag parity error

•

Bcache data correctable ECC error

•

Memory/system port data correctable ECC error

•

Bcache data correctable ECC error on a probe

•

Double-bit fill errors

•

Error case summary

Table 8–1 summarizes the 21264/EV68A error detection.
Table 8–1 21264/EV68A Error Detection Mechanisms
Component

Error Detection Mechanism

Bcache tag

Parity protected.

Bcache data array

Quadword-ECC protected.

Dcache tag array

Parity protected.

Dcache duplicate tag array

Parity protected.

Dcache data array

Quadword-ECC protected, however this mode of operation is
only supported in systems that have ECC enabled on both the
system and Bcache ports.

Icache tag array

Parity protected.

Icache data array

Parity protected.

System port data bus

Quadword-ECC protected.

21264/EV68A Hardware Reference Manual

Error Detection and Error Handling

8–1

Data Error Correction Code

8.1 Data Error Correction Code
The 21264/EV68A supports a quadword error correction code (ECC) for the system data
bus. ECC is generated by the 21264/EV68A for all memory write transactions
(WrVictimBlk) emitted from the 21264/EV68A and for all probe data. ECC is also
checked on every memory read transaction for single-bit correction and double-bit error
detection. Bcache data is checked for fills to the Dcache and Icache, and for Bcache-tosystem transfers that are initiated by a probe (if enabled by the CSR
ENABLE_PROBE_CHECK).
The 21264/EV68A ECC implementation corrects single-bit errors in hardware.
I/O write transaction data will not have a valid ECC (the ECC bits must be ignored by
the system). Also, ECC checking is not performed on I/O read data.
Error detection and correction can be enabled/disabled by way of Mbox IPR
DC_CTL[DCDAT_ERR_EN].
Table 8–2 shows the ECC code.
Table 8–2 64-Bit Data and Check Bit ECC Code
11 1111 1111 2222 2222 2233 3333 3333 4444 4444 4455 5555 5555 6666 CCCC CCCC
0123 4567 8901 2345 6789 0123 4567 8901 2345 6789 0123 4567 8901 2345 6789 0123 0123 4567
CB0

0111 0100 1101 0010 0111 0100 1101 0010 1000 1011 0010 1101 1000 1011 0010 1101 1000 0000

CB1

1110 1010 1010 1000 1110 1010 1010 1000 1110 1010 1010 1000 1110 1010 1010 1000 0100 0000

CB2

1001 1001 0110 0101 1001 1001 0110 0101 1001 1001 0110 0101 1001 1001 0110 0101 0010 0000

CB3

1100 0111 0001 1100 1100 0111 0001 1100 1100 0111 0001 1100 1100 0111 0001 1100 0001 0000

CB4

0011 1111 0000 0011 0011 1111 0000 0011 0011 1111 0000 0011 0011 1111 0000 0011 0000 1000

CB5

0000 0000 1111 1111 0000 0000 1111 1111 0000 0000 1111 1111 0000 0000 1111 1111 0000 0100

CB6

1111 1111 0000 0000 0000 0000 1111 1111 1111 1111 0000 0000 0000 0000 1111 1111 0000 0010

CB7

1111 1111 0000 0000 0000 0000 1111 1111 0000 0000 1111 1111 1111 1111 0000 0000 0000 0001

8.2 Icache Data or Tag Parity Error
The following actions are performed when an Icache data or tag parity error occurs.
1. When the hardware detects an error during an Icache read transaction, it traps and
replays the instructions that were fetched during the error, then flushes the entire
Icache so the re-fetched instructions do not come directly from the Icache.
2. I_STAT[PAR] is set.
3. A corrected read data (CRD) interrupt is posted, when enabled. (Pass 3 only)

8.3 Dcache Tag Parity Error
The primary copies of the Dcache tags are used only when servicing 21264/EV68A-generated load and store instructions. There are correctable and uncorrectable forms of this
error. If an issued load or store instruction detects a Dcache tag parity error, the following
actions are performed:
1. MM_STAT[DC_TAG_PERR] is set.
2. A Dstream fault (DFAULT) is taken.
8–2

Error Detection and Error Handling

21264/EV68A Hardware Reference Manual

Dcache Data Single-Bit Correctable ECC Error

3. The virtual address associated with the error is available in the VA register.
4. The PALcode flushes the error block by temporarily disabling
DC_CTL[DCTAG_PAR_EN] and evicting the block using two HW_LD instructions. The onchip duplicate tag provides the correct victim address and cache
coherence state.
If a retried load instruction detects the Dcache tag parity error, the memory reference
may have already been retired, so the EXC_ADDR is not available. In this case, the
error is uncorrectable and the Mbox performs the following actions:

•

Either DC_STAT[TPERR_P0] or DC_STAT[TPERR_P1] is set, indicating the
source of the error.

•

When enabled, a machine check (MCHK) is posted. The MCHK is taken when not
in PALmode.

8.4 Dcache Data Single-Bit Correctable ECC Error
The following operations may cause Dcache data ECC errors:

•

Load instructions

•

Stores of less than quadword length

•

Dcache victim read transactions

The hardware flow used for Dcache data ECC errors depends on the event that
caused the error.

8.4.1 Load Instruction
Loads that read data from the Dcache may do so either in the same cycle as the Dcache
tag probe (typical case) or in some subsequent cycle (load-queue retry). The hardware
functional flows for these two error cases differ slightly.
When a load instruction reads the Dcache data array in the same cycle as the tag array,
if an ECC error occurs on the LSD ECC error detectors, then the Ibox stops retiring
instructions and does not resume retiring until after hardware recovers from the error.
If an ECC error occurs on the LSD ECC error detectors, when a load instruction reads
the Dcache tag array before it reads the Dcache data array, then the load instruction may
have already been retired. In either case:

•

The incorrect data is written into the load instruction’s destination register;
however, the load queue retains the state associated with the load instruction.

•

A consumer of the load instruction’s data may be issued before the error is
recognized; however, the Ibox will invoke a replay trap at an instruction that is
older than (or equal to) any instruction that consumes the load instruction’s data,
and then stalls the replayed Istream in the map stage of the pipeline until the error is
corrected.

•

Given a READ_ERR read-type from the Mbox for the error load instruction, the
Cbox scrubs the block in the Dcache by evicting the block into the victim buffer
(thereby scrubbing it) and writing it back into the Dcache as follows:
–

C_STAT[DSTREAM_DC_ERR] is set.

21264/EV68A Hardware Reference Manual

Error Detection and Error Handling

8–3

Dcache Store Second Error

–

C_ADDR contains bits [19:6] of the Dcache address of the block that contains
the error (bits [42:20] of the physical address are not updated).

–

DC_STAT[ECC_ERR_LD] is set.

–

The load queue retries the load and rewrites the register.

–

A corrected read data (CRD) error interrupt is posted, when enabled.

Note:

Errors in speculative load instructions cause a CRD error interrupt
to be posted but the data is not scrubbed by hardware. The PALcode
cannot perform a scrub because C_STAT is zero and C_ADDR does not
contain the address of the error.

8.4.2 Store Instruction (Quadword or Smaller)
A store instruction that is a quadword or smaller could invoke a Dcache ECC error,
since the original quadword must be read to calculate the new check bits.

•

The Mbox scrubs the original quadword and replays the write transaction.

•

DC_STAT[ECC_ERR_ST] is set.

•

A corrected read data (CRD) error interrupt is posted, when enabled.

8.4.3 Dcache Victim Extracts
•

Dcache victims with an ECC error are scrubbed as they are written into the
victim data buffer.

•

No status is logged.

•

No exception is posted.

8.5 Dcache Store Second Error
A second store instruction error is logged when it occurs close behind the first.
Neither error is corrected.

•

DC_STAT[ECC_ERR_ST] is set.

•

DC_STAT[SEO] is set.

•

When enabled, a machine check (MCHK) is posted. The MCHK is taken when not
in PALmode.

8.6 Dcache Duplicate Tag Parity Error
The Dcache duplicate tag has the correct version of the Dcache coherence state for the
21264/EV68A, allowing it to be used for correct tag/status data when the Dcache tags
generate a parity error. These tags are parity protected also; however, the Dcache duplicate tag cell is designed to be much more tolerant of soft errors. The parity generators
for the duplicate tags are enabled whenever the Cbox performs a physically-indexed
read transaction of eight locations in the tag array. If an error is generated, the following
actions are taken:

•

8–4

Dcache duplicate tag parity errors are not recoverable.

Error Detection and Error Handling

21264/EV68A Hardware Reference Manual

Bcache Tag Parity Error

•

C_STAT[DC_PERR] is set.

•

C_ADDR contains bits [42:6] of the Dcache duplicate tag address of the block that
contains the error.

•

When enabled, a machine check (MCHK) is posted. The MCHK is taken when not
in PALmode.

8.7 Bcache Tag Parity Error
The Bcache tag parity is checked on all Bcache tag references, including references
invoked by system probes. If an error is detected, the following actions are taken:

•
•
•

Bcache tag parity errors are not recoverable.

•

When enabled, a machine check (MCHK) is posted. The MCHK is taken when not
in PALmode.

C_STAT[BC_PERR] is set.
C_ADDR contains bits [42:6] of the Bcache address of the block that contains the
error.

8.8 Controlling Bcache Block Parity Calculation
Parity is calculated for either valid Bcache blocks or all Bcache blocks. The calculation
is controlled by the value in Cbox CSR BC_VALID_MODE in the WRITE_MANY
chain, as follows:

•

If the MSB of BcTag_H is less than the value of Maximum PA in Table 4–13, then
BC_VALID_MODE=1 and parity is calculated for only valid Bcache blocks.

•

If the MSB of BcTag_H is greater than or equal to the value of Maximum PA in
Table 4–13, then BC_VALID_MODE=0 and parity is calculated for all Bcache
blocks.

For example, if BcTag_H[38:20] and Maximum PA is 36, then 38 is greater than or
equal to 36 and BC_VALID_MODE=0 and parity is calculated for all Bcache blocks.

8.9 Bcache Data Single-Bit Correctable ECC Error
The following actions may trigger Bcache data ECC errors:

•
•
•

Icache fill, data possibly used by Icache
Dcache fill, data possibly used by load instruction
Bcache victim during an ECB instruction or during a Dcache/Bcache miss

The recovery mechanism depends on the action that triggered the error.

8.9.1 Icache Fill from Bcache
For an Icache fill, the LSD ECC checkers detect the error, and bad Icache data parity is
generated for the octaword that contains the quadword in error. If an error is detected,
the following actions are taken:

•
•

The hardware flushes the Icache.
C_STAT[ISTREAM_BC_ERR] is set.

21264/EV68A Hardware Reference Manual

Error Detection and Error Handling

8–5

Bcache Data Single-Bit Correctable ECC Error

•

C_ADDR contains bits [42:6] of the Bcache fill address of the block that contains
the error.

•

C_SYNDROME_0[7:0] and C_SYNDROME_1[7:0] contain the syndrome of
quadword 0 and 1, respectively, of the octaword subblock that contains the error.

•

A machine check (MCHK) is posted and taken immediately. The PALcode machine
check handler performs a scrubbing operation as described in Section D.36 to
ensure that the origination point of the error is corrected.

Note:

A corrected read data (CRD) error interrupt is also posted in case this error
is in a speculative path and the MCHK is removed. The CRD PALcode
reads the status, to detect this condition, and scrubs the block. In the normal
MCHK flow, the PALcode clears the pending CRD error.

8.9.2 Dcache Fill from Bcache
If the quadword in error is not used to satisfy a load instruction, a hardware recovery
flow is not invoked. The quadword in error, and its associated check bits, are written
into the Dcache. However, status is logged as shown in the bulleted list below, and a
corrected read data (CRD) error interrupt is posted, when enabled. PALcode may elect
to correct the error by scrubbing the block. If the error is not corrected by PALcode
when it occurs, the error will be detected and corrected by a later load/victim operation.
If the quadword in error is used to satisfy a load instruction, then the flow is very similar to that used for a Dcache ECC error. The LSD ECC checker detects the error and the
21264/EV68A performs the following actions:

8–6

•

The load instruction’s destination register is written with incorrect data; however,
the load queue will retain the state associated with the load instruction.

•

A consumer of the load instruction’s data may be issued before the error is
recognized. The Ibox will invoke a replay trap at an instruction that is older than (or
equal to) any instruction that consumes the load instruction’s data. The 21264/
EV68A then stalls the replayed Istream in the map stage of the pipeline, until the
error is corrected.

•

With a READ_ERR read type from the Mbox for the load instruction in error, the
Cbox scrubs the block in the Dcache by evicting the block into the victim buffer
and writing it back into the Dcache.

•

C_STAT[DSTREAM_BC_ERR] is set.

•

C_ADDR contains bits [42:6] of the Bcache fill address of the block that contains
the error.

•

C_SYNDROME_0[7:0] and C_SYNDROME_1[7:0] contain the syndrome of
quadword 0 and 1, respectively, of the octaword subblock that contains the error.

•

The load queue retries the load instruction and rewrites the register.

•

DC_STAT[ECC_ERR_LD] is set.

•

A corrected read data (CRD) error interrupt is posted, when enabled.

Error Detection and Error Handling

21264/EV68A Hardware Reference Manual

Memory/System Port Single-Bit Data Correctable ECC Error

Note:

Errors in speculative load instructions cause a CRD error to be posted but
the data is not scrubbed by hardware. The PALcode cannot perform a scrub
operation because C_STAT is zero and C_ADDR does not contain the
address of the block in error.

8.9.3 Bcache Victim Read
A victim from the Bcache is written directly to the system port, without correction. The
ECC parity checker on the LSD detects the error and posts a corrected read data (CRD)
error interrupt. The Cbox error register is not updated.
8.9.3.1 Bcache Victim Read During a Dcache/Bcache Miss

While the Bcache is servicing a Dcache miss and that Bcache access is also a miss, and
an error occurs during that Bcache data access, the Cbox does not latch the error information. However, the Mbox correction state machine is activated and it invokes a CRD
error despite the fact that no correction is performed.
The Bcache access error is written out to memory and is subsequently detected and corrected by the next consumer of the data.

•

No correction is made.

•

No status is logged (C_STAT = 0).

•

A CRD error interrupt is posted, when enabled.

8.9.3.2 Bcache Victim Read During an ECB Instruction

A victim from the Bcache that occurs while an ECB instruction is being executed is
written directly to the system port without correction. No Cbox registers are set and no
exception is taken.

8.10 Memory/System Port Single-Bit Data Correctable ECC Error
The following actions may cause memory/system port data ECC errors:

•

Icache fill–data possibly used by Icache

•

Dcache fill–data possibly used by a load instruction

The recovery mechanism depends on the event that caused the error.

8.10.1 Icache Fill from Memory
For an Icache fill the LSD ECC generators detect the error, and bad Icache data
parity is generated for the octaword that contains the quadword in error.

•

The hardware flushes the Icache.

•

C_STAT[ISTREAM_MEM_ERR] is set.

•

C_ADDR contains bits [42:6] of the system memory fill address of the block that
contains the error.

•

C_SYNDROME_0[7:0] and C_SYNDROME_1[7:0] contain the syndrome of
quadword 0 and 1, respectively, of the octaword subblock that contains the error.

21264/EV68A Hardware Reference Manual

Error Detection and Error Handling

8–7

Memory/System Port Single-Bit Data Correctable ECC Error

•

Note:

Also, a corrected read data (CRD) error is posted, when enabled, in case
this error is in a speculative path and the MCHK is removed. The CRD
error PALcode reads the status to detect this condition and scrubs the block.
In the normal MCHK flow, the PALcode clears the pending CRD error.

8.10.2 Dcache Fill from Memory
If the quadword in error is not used to satisfy a load instruction, no hardware
recovery flow is invoked. The quadword in error, and its associated check bits, are written into the Dcache. However, status is logged as shown in the bulleted list below and a
corrected read data (CRD) error interrupt is posted, when enabled. PALcode may
choose to correct the error by scrubbing the block. If the error is not corrected by PALcode at the time, the error will be detected and corrected by a load/victim operation.
If the quadword in error is used to satisfy a load instruction, then the flow is very similar to that used for a Dcache ECC error:

•

The load instruction’s destination register is written with incorrect data; however,
the load queue will retain the state associated with the load instruction.

•

A consumer of the load instruction’s data may be issued before the error is
recognized; however, the Ibox will invoke a replay trap at an instruction that is
older than (or equal to) any instruction that consumes the load instruction’s data.
The Ibox stalls the replayed Istream in the map stage of the pipeline until the error
is corrected.

•

With a READ_ERR read type from the Mbox for the load instruction in error, the
Cbox scrubs the block in the Dcache by evicting the block into the victim buffer
and writing it back into the Dcache.

•

C_STAT[DSTREAM_MEM_ERR] is set.

•

C_ADDR contains bits [42:6] of the system memory fill address of the block that
contains the error.

•

C_SYNDROME_0[7:0] and C_SYNDROME_1[7:0] contain the syndrome of
quadword 0 and 1, respectively, of the octaword subblock that contains the error.

•

The load queue retries the load instruction and rewrites the register.

•

DC_STAT[ECC_ERR_LD] is set.

•

A corrected read data (CRD) error interrupt is posted, when enabled.

Note:

8–8

Errors in speculative load instructions cause a CRD error to be posted but
the data is not scrubbed by hardware. The PALcode cannot scrub the data
because C_STAT is zero, and C_ADDR does not have the address of the
block with the error.

Error Detection and Error Handling

21264/EV68A Hardware Reference Manual

Bcache Data Single-Bit Correctable ECC Error on a Probe

8.11 Bcache Data Single-Bit Correctable ECC Error on a Probe
The probed processor extracts the block from its Bcache, signaling a corrected read
data (CRD) error and latching error information. The single-bit ECC detected error data
is not corrected by the probed processor, but is forwarded to the requesting processor.
The requesting processor then detects a related system fill error as a result of this system probe transaction.

•

No hardware correction is performed.

•

C_STAT[PROBE_BC_ERR] is set.

•

C_ADDR contains bit [42:6] of the Bcache address of the block that contains the
error.

•

C_SYNDROME_0[7:0] and C_SYNDROME_1[7:0] contain the syndrome of
quadword 0 and 1, respectively, of the octaword subblock that contains the error.

•

A CRD error interrupt is posted, when enabled.

•

The PALcode on the probed processor may choose to scrub the error, though it will
probably be scrubbed by the requesting processor.

8.12 Double-Bit Fill Errors
Double-bit errors for fills are detected, but not corrected, in the 21264/EV68A. The following events may cause a double-bit fill error:

•

Icache fill from Bcache

•

Dcache fill from Bcache

•

Icache fill from memory

•

Dcache fill from memory

If an error is detected, the following actions are taken:

•

C_STAT is set to one of the following:
ISTREAM_BC_DBL (Icache fill from Bcache)
DSTEAM_BC_DBL (Dcache fill from Bcache)
ISTREAM_MEM_DBL (Icache fill from memory)
DSTREAM_MEM_DBL (Dcache fill from memory)

•

C_ADDR contains bits [42:6] of the system memory fill address of the block that
contains the error.

•

When enabled, a machine check (MCHK) is posted. The MCHK is taken when not
in PALmode.

•

A double-bit fill error from memory, marked by the data's corresponding ECC,
when written to cache, also writes the corresponding ECC to cache. Any consumer
of that error (such as another CPU) also consumes the corresponding ECC value.

Note:

C_ADDR may be inaccurate in heavy traffic conditions. C_STAT is accurate.

21264/EV68A Hardware Reference Manual

Error Detection and Error Handling

8–9

Error Case Summary

8.13 Error Case Summary
Table 8–3 summarizes the various error cases and their ramifications.
Table 8–3 Error Case Summary
Error

Exception Status

Hardware
Action

PALcode Action

Icache data or tag
parity error

CRD

ISTAT[PAR]

Icache flushed

Log as CRD

Dcache tag parity
error (on issue)

DFAULT

MM_STAT[DC_TAG_PERR]
VA[address]

—

Evict with two
HW_LDs and log as
CRD

Dcache tag parity
error (on retry)

MCHK1

DC_STAT[TPERR_P0] or
DC_STAT[TPERR_P1]

—

Log as MCHK

Dcache single-bit
ECC error on load

CRD

DC_STAT[ECC_ERR_LD]
Corrected and
C_STAT[DSTREAM_DC_ERR] scrubbed
C_ADDR[bits [19:6] of the error
address. [42:20] not updated.]

Log as CRD

Dcache single-bit
ECC error on
speculative load

CRD

DC_STAT[ECC_ERR_LD]
C_STAT contains zero

None

Log as CRD

Dcache single-bit
ECC error on small
store

CRD

DC_STAT[ECC_ERR_ST]

Corrected and
scrubbed

Log as CRD

Dcache single-bit
None
ECC error on victim
read

None

Corrected and
scrubbed

None

Dcache second error MCHK1
on store

DC_STAT[SEO]

No correction
on either store

Log as MCHK

Dcache duplicate tag MCHK1
parity error

C_STAT[DC_PERR]
C_ADDR[error address]

Uncorrectable

Log as MCHK

Bcache tag parity
error

MCHK1

C_STAT[BC_PERR]
C_ADDR[error address]

Uncorrectable

Log as MCHK

Bcache single-bit
error on Icache fill

MCHK
and CRD2

C_STAT[ISTREAM_BC_ERR]
C_ADDR[error address]
C_SYNDROME_0
C_SYNDROME_1

Icache flushed

Scrub error as described
in Section D.36.
Log as CRD

Bcache single-bit
error on Dcache fill

CRD

DC_STAT[ECC_ERR_LD]
C_STAT[DSTREAM_BC_ERR]
C_ADDR[error address]
C_SYNDROME_0
C_SYNDROME_1

Corrected and
scrubbed in
Dcache3

Scrub error as described
in Section D.36.
Log as CRD

Bcache victim read
on Dcache/Bcache
miss

CRD

DC_STAT[ECC_ERR_LD]
C_STAT contains 0

None

Log as CRD

Bcache victim read
on ECB

None

8–10

Error Detection and Error Handling

21264/EV68A Hardware Reference Manual

Error Case Summary

Table 8–3 Error Case Summary (Continued)
Hardware
Action

Error

Exception Status

PALcode Action

Memory single-bit
error on Icache fill

MCHK
and CRD2

C_STAT[ISTREAM_MEM_ERR]
C_ADDR[error address]
C_SYNDROME_0
C_SYNDROME_1

Icache flushed

Scrub error as described
in Section D.36.
Log as CRD

Memory single-bit
error on Dcache fill

CRD

DC_STAT[ECC_ERR_LD]
C_STAT[DSTREAM_MEM_ERR]
C_ADDR[error address]
C_SYNDROME_0
C_SYNDROME_1

Corrected and
scrubbed in
Dcache3

Scrub error as described
in Section D.36.
Log as CRD

Bcache single-bit
error on a probe hit

CRD

C_STAT[PROBE_BC_ERR]
C_ADDR[error address] 4
C_SYNDROME_0
C_SYNDROME_1

None

May scrub error as
described in Section
D.36 .
Log as CRD

Bcache double-bit
error on Icache fill

MCHK1

C_STAT[ISTREAM_BC_DBL]
C_ADDR[error address] 4

None

Log as MCHK

Bcache double-bit
error on Dcache fill

MCHK1

C_STAT[DSTREAM_BC_DBL]
C_ADDR[error address] 4

None

Log as MCHK

Memory double-bit
error on Icache fill

MCHK1

C_STAT[ISTREAM_MEM_DBL]
C_ADDR[error address] 4

None

Log as MCHK

Memory double-bit
error on Dcache fill

MCHK1

C_STAT[DSTREAM_MEM_DBL]
C_ADDR[error address] 4

None

Log as MCHK

Machine check taken in native mode. It is deferred while in PALmode.
CRD error posted in case the machine check is down a speculative path.
3
For a single-bit error on a non-target quadword, the error is not corrected in hardware,
but is corrected by PALcode during the scrub operation.
4
The contents of C_ADDR may not be accurate when there is heavy cache fill traffic.
2

21264/EV68A Hardware Reference Manual

Error Detection and Error Handling

8–11

9
Electrical Data
This chapter describes the electrical characteristics of the 21264/EV68A and its interface pins. The chapter contains both ac and dc electrical characteristics and power supply considerations, and is organized as follows:

•

Electrical characteristics

•

DC characteristics

•

Power supply sequencing

•

AC characteristics

9.1 Electrical Characteristics
Table 9–1 lists the maximum electrical ratings for the 21264/EV68A.
Table 9–1 Maximum Electrical Ratings
Characteristics

Ratings

Storage temperature

–55° C to +125° C (–67° F to 257° F)

Junction temperature

0° C to 100° C (32° F to 212° F)

Open-drain driver interface

2.0 V

Push-pull driver interface

1.7 V

Minimum dc voltage on signal pins

VSS – 400 mV

Maximum power @ indicated VDD
for the following frequencies:

Notes:

Frequency

Peak Power

750 MHz
833 MHz
875 MHz
940 MHz

60 W @ 1.7 V
67 W @ 1.7 V
70 W @ 1.7 V
75 W @ 1.7 V

Stresses above those listed under the given maximum electrical ratings may
cause permanent device failure. Functionality at or above these
limits is not implied. Exposure to these limits for extended periods of time
may affect device reliability.
Power data is preliminary and based on measurements from a limited set of
material.

21264/EV68A Hardware Reference Manual

Electrical Data

9–1

DC Characteristics

9.2 DC Characteristics
This section contains the dc characteristics for the 21264/EV68A. The 21264/EV68A
pins can be divided into 10 distinct electrical signal types. The mapping between these
signal types and the package pins is shown in Chapter 3. Table 9–2 shows the signal
types.
Table 9–2 Signal Types
Signal Type

Description

I_DC_POWER

Supply voltage pins (VDD/PLL_VDD)

I_DC_REF

Input dc reference pin

I_DA

Input differential amplifier receiver

I_DA_CLK

Input differential amplifier clock receiver

O_OD

Open-drain output driver

O_OD_TP

Open-drain driver for test pins

O_PP

Push-pull output driver

O_PP_CLK

Push-pull output clock driver

B_DA_OD

Bidirectional differential amplifier receiver — open-drain

B_DA_PP

Bidirectional differential amplifier receiver — push-pull

Tables 9–3 through 9–12 show the dc switching characteristics of each signal type.
Also, the following notes apply to Tables 9–3 through 9–12.
1. The differential voltage, Vdiff, is the absolute difference between the differential
input pins.
2. Delta V BIAS is defined as the open-circuit differential voltage on the appropriate
differential pairs. Test condition for these inputs are to let the input network self
bias and measure the open circuit voltage. The test load must be ≥ 1M ohm. In normal operation, these inputs are coupled with a 680-pF capacitor.
3. Functional operation of the 21264/EV68A with less than all VDD and VSS pins
connected is not implied.
4. The test load is a 50-ohm resistor to VDD/2. The resistor can be connected to the
21264/EV68A pin by a 50-ohm transmission line of any length.
5. DC test conditions set the minimum swing required. These dc limits set the trip
point precision.
6. Input pin capacitance values include 2.0 pF added for package capacitance.

9–2

Electrical Data

21264/EV68A Hardware Reference Manual

DC Characteristics

Note:

Current out of a 21264/EV68A pin is represented by a – symbol while a +
symbol indicates current flowing into a 21264/EV68A pin.

Table 9–3 VDD (I_DC_POWER)
Parameter Symbol

Description

Test Conditions

Minimum Maximum

VDD

Processor core supply voltage

—

1.6 V

1.7 V

Power (sleep)

Processor power required (sleep)

@ VDD = 1.7 V
Note 3

—

12 W1

PLL_VDD

PLL supply voltage

—

2.45 V

2.55 V

PLL_IDD

PLL supply current (running)

Freq = 940 MHz

—

25 mA

Power measured at 37.5 MHz while running the “Ebox aliveness test.”

Table 9–4 Input DC Reference Pin (I_DC_REF)
Parameter
Symbol

Description

Test Conditions

Minimum

Maximum

VREF

DC input reference voltage

—

600 mV

VDD – 100 mV

| II |

Input current

VSS ≤ V ≤ VDD

—

150 µA

Table 9–5 Input Differential Amplifier Receiver (I_DA)
Parameter
Symbol

Description

Test Conditions Minimum

Maximum

V IL

Low-level input voltage

Note 5

—

VREF – 200 mV

V IH

High-level input voltage

—

VREF + 200 mV —

| II |

Input current

VSS ≤ V ≤ VDD —

150 µA

CIN

Input-pin capacitance

Freq =10 MHz

5.7 pF
Note 6

—

Table 9–6 Input Differential Amplifier Clock Receiver (I_DA_CLK)
Parameter
Symbol

Description

Test Conditions

Minimum

Maximum

V diff

Differential input voltage

—

200 mv Note 1

—

| ∆ VBIAS |

Open-circuit differential

I ≤ ± 1 µA
Note 2

—

50 mV

| II |

Input current

VSS ≤ V ≤ VDD

—

150 µA

CIN

Input-pin capacitance

Freq =10 MHz

—

5.0 pF
Note 6

21264/EV68A Hardware Reference Manual

Electrical Data

9–3

DC Characteristics

Table 9–7 Pin Type: Open-Drain Output Driver (O_OD)
Parameter
Symbol

Description

Test
Conditions

Minimum

Maximum

V OL

Low-level output voltage

IOL = 50 mA

—

400 mV

|IOZ |

High impedance output current

0 < V < VDD

—

150 µA

COD

Open-drain pin capacitance

Freq = 10 MHz

—

5.7 pF
Note 6

Table 9–8 Bidirectional, Differential Amplifier Receiver, Open-Drain Output Driver (B_DA_OD)
Parameter
Symbol

Description

Test Conditions

Minimum

Maximum

V IL

Low-level input voltage

Note 5

—

VREF –200 mv

V IH

High-level input voltage

—

VREF + 200 mV —

V OL

Low-level output voltage

IOL = 50 mA

—

400 mV

| II |

Input current

VSS ≤ V ≤ VDD

—

150 µA1

CIN

Input-pin capacitance

Freq =10 MHz

—

5.7 pF
Note 6

Measurement taken with output driver disabled.

Table 9–9 Pin Type: Open-Drain Driver for Test Pins (O_OD_TP)
Parameter
Symbol

Description

Test
Conditions

Minimum

Maximum

V OL

Low-level output voltage

IOL = 15 mA

—

400 mV

| IOZ |

High-impedance output current

0 < V < VDD

—

150 µA

COD_TP

Pin capacitance

Freq = 10 MHz

—

5.2 pF
Note 6

Table 9–10 Bidirectional, Differential Amplifier Receiver, Push-Pull Output Driver (B_DA_PP)
Parameter
Symbol

Description

Test Conditions

Minimum

Maximum

V IL

Low-level input voltage

—

VREF – 200 mV

V IH

High-level input voltage

—

VREF + 200 mV —

V OL

Low-level output voltage

IOL = 6 mA

—

400 mV

V OH

High-level output voltage

IOH = –6 mA

VDD – 400 mV

—

| II |

Input current

VSS ≤ V ≤ VDD

—

150 µA1

CIN

Input-pin capacitance

Freq =10 MHz

—

6.0 pF
Note 6

9–4

Measurement taken with output driver disabled.

Electrical Data

21264/EV68A Hardware Reference Manual

Power Supply Sequencing and Avoiding Potential Failure Mechanisms

Table 9–11 Push-Pull Output Driver (O_PP)
Parameter
Symbol

Description

Test
Conditions

Minimum

Maximum

V OL

Low-level output voltage

IOL = 40 mA

—

500 mV

V OH

High-level output voltage

IOL = –40 mA

VDD – 500 mV

—

| IOZ |

High-impedance output current

0 < V < VDD

—

150 µA

COD

Open-drain pin capacitance

Freq = 10 MHz

—

6.0 pF
Note 6

Table 9–12 Push-Pull Output Clock Driver (O_PP_CLK)
Parameter
Symbol

Description

Test
Conditions

Minimum

Maximum

V OL

Low-level output voltage

Note 4

—

VDD/2 – 325 mV

V OH

High-level output voltage

Note 4

VDD/2 + 325 mV —

| IOZ |

High-impedance output
current

0 < V < VDD

—

40 mA1

Measured value includes current from onchip termination structures.

9.3 Power Supply Sequencing and Avoiding Potential Failure Mechanisms
Before the power-on sequencing can occur, systems should ensure that DCOK_H is
deasserted and Reset_L is asserted. Then, systems ramp power to the 21264/EV68A
PLL_VDD @ 2.5 V and the 21264/EV68A power planes (VDD @ 1.65 V, not to
exceed 1.70 V under any circumstances), with PLL_VDD leading VDD. Systems
should supply differential clocks to the 21264/EV68A on ClkIn_H and ClkIn_L. The
clocks should be running as power is supplied.
When enabling the power supply inputs in a system, three failure mechanisms must be
avoided:
1. Bidirectional signal buses must not conflict during power-up. A conflict on these
buses can generate high current conditions, which can compromise the reliability of
the associated chips.
2. Similarly, input receivers should not see intermediate voltage levels that can also
generate high current conditions, which can compromise the reliability of the
receiving chip.
3. Finally, no CMOS chip should see an input voltage that is higher than its internal
VDD. In such a condition, a reasonable level of charge can be injected into the bulk
of the die. This condition can expose the chip to a positive-feedback latchup
condition.
The 21264/EV68A addresses those three failure mechanisms by disabling all of its
outputs and bidirectional pins (with three exceptions) until the assertion of DCOK_H.
The three exceptions are Tdo_H, EV6Clk_L, and EV6Clk_H. Tdo_H is used only in

21264/EV68A Hardware Reference Manual

Electrical Data

9–5

AC Characteristics

the tester environment and does not need to be disabled. EV6Clk_L and EV6Clk_H
are outputs that are both generated and consumed by the 21264/EV68A; thus, VDD
tracks for both the producer and consumer.
On the push-pull interfaces:

•

Disabling all output drivers leaves the output signal at the DC bias point of the termination network.

•

Disabling the bidirectional drivers leaves the other consumers of the bus as the bus
master.

On the open-drain interfaces:

•

Disabling all output drivers leaves the output signal at the voltage of the open-drain
pull-up.

•

Disabling all bidirectional drivers leaves the other consumers of the bus as the bus
master.

To avoid failure mechanism number two, systems must sequence and control external
signal flow in such a way as to avoid zero differential into the 21264/EV68A input
receivers (I_DA, I_DA_CLK, B_DA_OD, B_DA_PP, and B_DA_PP). Finally, to
avoid failure mechanism number three, systems must sequence input and bidirectional
pins (I_DA, I_DA_CLK, B_DA_OD, B_DA_PP, and I_DC_REF) such that the 21264/
EV68A does not see a voltage above its VDD.
In addition, as power is being ramped, Reset_L must be asserted — this allows the
21264/EV68A to reset internal state. Once the target voltage levels are attained,
systems should assert DCOK_H. This indicates to the 21264/EV68A that internal
logic functions can be evaluated correctly and that the power-up sequence should be
continued. Prior to DCOK_H being asserted, the logic internal to the 21264/EV68A is
being reset and the internal clock network is running (either clocked by the VCO, which
is at a nominal speed, or by ClkIn_H, if the PLL is bypassed).
The reset state machine is in state WAIT_SETTLE.

9.4 AC Characteristics
Abbreviations:

The following abbreviations apply to Table 9–13:

•

TSU = Setup time

•

Duty cycle = Minimum clock duty cycle

•

TDH = Hold time

•

Slew rate = referenced to signal edge

AC Test Conditions:

The following conditions apply to the measurements that are listed in Table 9–13:

•

9–6

TSU, TDH, and TSkew values include the following onchip and package variations: min/max metal capacitance, GCLK skew, transistor gate length variation
(OCV), across-chip temperature and voltage variation, and package lead-length
variation.

Electrical Data

21264/EV68A Hardware Reference Manual

AC Characteristics

•

The AC specification values for TSU, TDH, and TSkew assume a ± 0.4 V input
signal amplitude, relative to Vref, one volt per nanosecond slew rates, no mismatch
in pin loading, and zero time-offset between forwarded clocks and data. Any deviation from those conditions can result in degraded chip performance.
–

VDD is in the range between 1.55 V and 1.65 V.

–

SysVref is 1.2 Volts.

–

BcVref is 0.75 Volts.

All output skew data is based on simulation into a 50-ohm transmission line that is
terminated with 50 ohms to VDD/2 for Bcache timing, and with 50 ohms to VDD
for all other timing.
Timings are measured at the pins as follows:
–
–
–
–

For open-drain outputs, timing is measured to (Vol + Vterm)/2. Where V term is
the offchip termination voltage for system signals.
For non-open-drain outputs, timing is measured to (Vol + Voh)/2.
For all inputs other than type I_DA_CLK, timing is measured to the point
where the input signal crosses VREF.
For type I_DA_CLK inputs, timing is measured when the voltage on the complementary inputs is equal.

Table 9–13 AC Specifications
Signal Name

Type

Reference Signal

TSU1

TDH2

TSkew

Duty Cycle

TSlew

SysAddIn_L[14:0

I_DA

SysAddInClk_L

400 ps

1.0 V/ns

SysFillValid_L

I_DA

SysAddInClk_L

400 ps

1.0 V/ns

SysDataInValid_L

I_DA

SysAddInClk_L

400 ps

1.0 V/ns

SysDataOutValid_L

I_DA

SysAddInClk_L

400 ps

1.0 V/ns

SysAddInClk_L

I_DA

45–55%

1.0 V/ns

SysAddOut_L[14:0]

O_OD

SysAddOutClk_L

± 300 ps

SysAddOutClk_L

O_OD

EV6Clk_x

± 350 ps

45-55%

SysData_L[63:0]

B_DA_OD

SysDataInClk_H[7:0]

400 ps

1.0 V/ns

SysCheck_L[7:0]

SysDataOutClk_L[7:0]

± 300 ps

SysDataInClk_H[7:0]

400 ps

1.0 V/ns

45-55%

1.0 V/ns

45-55%

—

45-55%

—

O_PP

38-63%8

—

O_PP

40-60%9

—

45-55%

1.0 V/ns

38-63%8

40-60%9

B_DA_OD

SysDataOutClk_L[7:0]

± 300 ps3

SysDataInClk_H[7:0]

I_DA

SysDataOutClk_L[7:0]

O_OD

EV6Clk_x

± 350 ps

BcAdd_H[23:4]
BcDataOE_L
BcLoad_L
BcDataWr_L
BcData_H[127:0]

O_PP
O_PP

B_DA_PP

BcTagOutClk_x
BcDataOutClk_x[3:0]

BcDataOutClk_x[3:0]

BcDataInClk_H[7:0]

21264/EV68A Hardware Reference Manual

5,6

± 300 ps

400 ps

± 300 ps

Electrical Data

9–7

AC Characteristics

Table 9–13 AC Specifications (Continued)
Signal Name

Type

Reference Signal

TSU1

TDH2

TSkew

Duty Cycle

BcDataInClk_H[7:0]

I_DA

45-55%

BcDataOutClk_H[3:0]

O_PP

EV6Clk_x

± 350 ps

BcDataOutClk_L[3:0]

O_PP

EV6Clk_x

± 350 ps

BcTag_H[42:20]

B_DA_PP

BcTagInClk_H

400 ps

BcTagDirty_H

B_DA_PP

BcTagInClk_H

400 ps

NA
1.0 V/ns

B_DA_PP

BcTagInClk_H

400 ps

BcTagShared_H

B_DA_PP

BcTagInClk_H

400 ps

BcTagInClk_H

400 ps

45-55%

± 350 ps

10 ns

10 ns11

100 mV/ns

I_DA

100 mV/ns

I_DA

100 mV/ns

I_DA

100 mV/ns
1.0 V/ns

BcTagOutClk_x

O_PP

EV6Clk_x

I_DA

PllBypass_H
ClkIn_x

DCOK_H

I_DA_CLK

FrameClk_x
EV6Clk_x

EV6Clk_x

I_DA

1.0 V/ns

45-55%

BcTagInClk_H

DCOK_H

± 300 ps

O_PP

BcTagWr_L

Reset_L

O_PP

400 ps

40-60%

BcTagOutClk_x
BcTagOE_L

IRQ_H[5:0]

1.0 V/ns
9

BcTagOutClk_x
B_DA_PP

1.0 V/ns

38-63%

BcTagParity_H

BcTagValid_H

1.0 V/ns

BcTagOutClk_x

TSlew

I_DA_CLK

ClkIn_x

O_PP_CLK ClkIn_x

40–60%16

400 ps

1.0 V/ns

±1.0 ns

YDiv±5%

1.0 V/ns

Cycle Compression Specification: See Note 19

ClkFwdRst_H

I_DA

FrameClk_x

400 ps

SromData_H

I_DA

SromClk_H

2.0 ns

SromOE_L

O_OD

EV6Clk_x

± 2.0 ns

SromClk_H20

O_OD

EV6Clk_x

± 7.0 ns

Tms_H

I_DA

Tck_H

2.0 ns

100 mV/ns

I_DA

Tck_H

100 mV/ns

Tdi_H

I_DA

Tck_H

2.0 ns

100 mV/ns

Tdo_H

O_OD

Tck_H

± 7.0 ns

Tck_H

I_DA

IEEE 1149.1 Port Freq. = 5.0
MHz Max.

45-55%

100 mV/ns

TestStat_H

O_OD

EV6Clk_x

± 4.0 ns

Trst_L

9–8

100 mV/ns

The TSU specified for all clock-forwarded signal groups is with respect to the associated clock.

Electrical Data

21264/EV68A Hardware Reference Manual

AC Characteristics

2
3
4
5
6
7

The TDH specified for all clock-forwarded signal groups is with respect to the associated clock.
The TSkew value applies only when the SYS_CLK_DELAY[0:1] entry in the Cbox WRITE_ONCE
chain (Table 5–24) is set to zero phases of delay between forwarded clock out and address/data.
The TSkew specified for SysData_L signals is only with respect to the associated clock.
These signals should be referenced to BcTagOutClk_x when measuring TSkew, provided that
BcTagOutClkl_x and BcDataOutClk_x have no programmed offset.
The TSkew value applies only when the BC_CLK_DELAY[0:1] entry in the Cbox WRITE_ONCE
chain (Table 5–24) is set to zero phases of delay for Bcache clock.
The TSkew specified for BcAdd_H signals is only with respect to the associated clock.

The duty cycle for 2.5X single data mode 2 GCLK phases high and 3 GCLK phases low.
The duty cycle for 3.5X single data mode 3 GCLK phases high and 4 GCLK phases low.
10
The TSkew specified for BcData_H signals is only with respect to the associated clock pair.
11
IRQ_H[5:0] must have their TSU and TDH times referenced to DCOK_H during power-up to ensure
the correct Y divider and resulting EV6Clk_x duty cycle. When the 21264/EV68A is executing
instructions IRQ_H[5:0] act as normal asynchronous pins to handle interrupts.
12
Reset_L is an asynchronous pin. It may be asserted asynchronously.
13
DCOK_H is an asynchronous pin. Note the minimum slew rate on the assertion edge.
14 PllBypass_H may not switch when ClkIn_x is running. This pin must either be deasserted during
power-up or the 21264/EV68A core power pin (VDD pins) indicating the 21264/EV68A’s internal
PLL will be used. Note that it is illegal to use PllBypass_H asserted during power-up unless a
ClkIn_x is present.
15 See Section 7.11.2 for a discussion of ClkIn_x as it relates to operating the 21264/EV68A’s internal
PLL versus running the 21264/EV68A in PLL bypass mode. ClkIn_x has specific input jitter requirements to ensure optimum performance of the internal 21264/EV68A PLL.
16 In PLL bypass mode, duty cycle deviation from 50%–50% directly degrades device operating frequency.
9

The TSU and TDH of FrameClk_x are referenced to the deasserting edge of ClkIn_x.
This signal is a feedback to the internal PLL and may be monitored for overall 21264/EV68A jitter. It
can also be used as a feedback signal to an external PLL when in PLL bypass mode. Proper termination of EV6Clk_x is imperative.
19
The cycle or phase cannot be more than 5% shorter than the nominal. Do not confuse this measurement with duty cycle.
20
The period for SromClk_H is 256 GCLK cycles.
21
When Trst_L is deasserted, Tms_H must not change state. Trst_L is asserted asynchronously but
may be deasserted synchronously.
18

21264/EV68A Hardware Reference Manual

Electrical Data

9–9

10
Thermal Management
This chapter describes the 21264/EV68A thermal management and thermal design
considerations, and is organized as follows:

•

Operating temperature

•

Heat sink specifications

•

Thermal design considerations

10.1 Operating Temperature
The 21264/EV68A is specified to operate when the temperature at the center of the heat
sink (Tc) is as shown in Table 10–1. Temperature Tc should be measured at the center of
the heat sink, between the two package studs. The GRAFOIL pad is the interface material between the package and the heat sink.
Table 10–1 Operating Temperature at Heat Sink Center (Tc)
Tc

Frequency

80.8° C

750 MHz

78.6° C

833 MHz

77.6° C

875 MHz

76.0° C

940 MHz

Note:

Compaq recommends using the heat sink because it reduces overheating
that causes thermal failure.

21264/EV68A Hardware Reference Manual

Thermal Management

10–1

Operating Temperature

Table 10–2 lists the values for the center of heat-sink-to-ambient (θca) for the 21264/
EV68A 587-pin PGA. Tables 10–3 through 10–6 show the allowable Ta (without
exceeding Tc) at various airflows.
Table 10–2 θca at Various Airflows for 21264/EV68A
Airflow (linear ft/min)

200

400

600

800

θca with heat sink type 1 (°C/W)

0.68

0.42

0.36

0.33

θca with heat sink type 2 (°C/W)

0.62

0.35

0.28

0.25

θca with heat sink type 31 (°C/W)
1

— 0.32 —

Heat sink type 3 has a 80 mm × 80 mm × 15 mm fan attached.

Table 10–3 Maximum Ta for 21264/EV68A @ 750 MHz and @ 1.7 V with Various Airflows
Airflow (linear ft/min)

200

400

600

800

Maximum Ta with heat sink type 1 (°C)

40.0

55.6

59.2

61.0

Maximum Ta with heat sink type 2 (°C)

43.6

59.8

64.0

65.8

Maximum Ta
1

with heat sink type 31 (°C)

— 61.6 —

Heat sink type 3 has a 80 mm × 80 mm × 15 mm fan attached.

Table 10–4 Maximum Ta for 21264/EV68A @ 833 MHz and @ 1.7 V with Various Airflows
Airflow (linear ft/min)

200

400

600

800

Maximum Ta with heat sink type 1 (°C)

33.0

50.4

54.4

56.5

Maximum Ta with heat sink type 2 (°C)

37.0

55.1

59.8

61.8

Maximum Ta with heat sink type 3 (°C)
1

— 57.1 —

Heat sink type 3 has a 80 mm × 80 mm × 15 mm fan attached.

Table 10–5 Maximum Ta for 21264/EV68A @ 875 MHz and @ 1.7 V with Various Airflows
Airflow (linear ft/min)

200

400

600

800

Maximum Ta with heat sink type 1 (°C)

30.0

48.2

52.4

54.5

Maximum Ta with heat sink type 2 (°C)

34.2

53.1

58.0

60.1

Maximum Ta with heat sink type 31 (°C)
1

— 55.2 —

Heat sink type 3 has a 80 mm × 80 mm × 15 mm fan attached.

Table 10–6 Maximum Ta for 21264/EV68A @ 940 MHz and @ 1.7 V with Various Airflows
Airflow (linear ft/min)

200

400

600

800

Maximum Ta with heat sink type 1 (°C)

25.0

44.5

49.0

51.3

Maximum Ta with heat sink type 2 (°C)

29.5

49.8

55.0

57.3

Maximum Ta
1

10–2

with heat sink type 31 (°C)

— 52.0 —

Heat sink type 3 has a 80 mm × 80 mm × 15 mm fan attached.

Thermal Management

21264/EV68A Hardware Reference Manual

Heat Sink Specifications

10.2 Heat Sink Specifications
Three heat sink types are specified. The mounting holes for all three are in line with the
cooling fins.
Figure 10–1 shows the heat sink type 1, along with its approximate dimensions.
Figure 10–1 Type 1 Heat Sink
4.10 in

1.55 in
2.05 in

1.00 in

4.10 in

0.10 in
0.05 in

1.50 in

0.30 in
LK99-0210a

21264/EV68A Hardware Reference Manual

Thermal Management

10–3

Heat Sink Specifications

Figure 10–2 shows the heat sink type 2, along with its approximate dimensions.
Figure 10–2 Type 2 Heat Sink
3.75 in
1.875 in

1.375 in
1.875 in

1.00 in

3.75 in

1.015 in

0.055 in
0.075 in
0.04 in

2.35 in

0.25 in
LK99-0211a

10–4

Thermal Management

21264/EV68A Hardware Reference Manual

Heat Sink Specifications

Figure 10–3 shows heat sink type 3, along with its approximate dimensions.
The cooling fins of heat sink type 3 are cross-cut. Also, an 80 mm × 80 mm × 15 mm
fan is attached to heat sink type 3.
Figure 10–3 Type 3 Heat Sink
3.20 in
1.600 in

3.20 in

1.00 in

1.600 in
0.200 in
1.100 in

0.575 in
0.20 in
0.060 in
0.080 in
0.040 in

0.250 in

0.100 in
0.05 in
0.55 in

0.090 in
2.00 in

0.885 in
0.30 in
3.200 in

2.605 in
LK99-0212a

21264/EV68A Hardware Reference Manual

Thermal Management

10–5

Thermal Design Considerations

10.3 Thermal Design Considerations
Follow these guidelines for printed circuit board (PCB) component placement:

•

Orient the 21264/EV68A on the PCB with the heat sink fins aligned with the airflow direction.

•

Avoid preheating ambient air. Place the 21264/EV68A on the PCB so that inlet air
is not preheated by any other PCB components.

•

Do not place other high power devices in the vicinity of the 21264/EV68A.

Do not restrict the airflow across the 21264/EV68A heat sink. Placement of other
devices must allow for maximum system airflow in order to maximize the performance
of the heat sink.

10–6

Thermal Management

21264/EV68A Hardware Reference Manual

11
Testability and Diagnostics
This chapter describes the 21264/EV68A user-oriented testability and diagnostic features. These features include automatic power-up self-test, Icache initialization from
external serial ROMs, and the serial diagnostic terminal port.
The boundary-scan register, which is another testability and diagnostic feature, is listed
in Appendix B. The boundary-scan register is compatible with IEEE Standard 1149.1.
This chapter is organized as follows:

•

Test pins

•

SROM/serial diagnostic terminal port

•

IEEE 1149.1 port

•

TestStat_H pin

•

Power-up self-test and initialization

•

Notes on IEEE 1149.1 operation and compliance

The 21264/EV68A has several manufacturing test features that are used only by the
factory, and they are beyond the scope of this chapter.

11.1 Test Pins
The 21264/EV68A test access ports include the IEEE 1149.1 test access port, a dualpurpose SROM/Serial diagnostic terminal port, and a test status output pin. Table 11–1
lists the test access port pins.
Table 11–1 Dedicated Test Port Pins
Pin Name

Type

Function

Tms_H

Input

IEEE 1149.1 test mode select

Tdi_H

Input

IEEE 1149.1 test data in

Trst_L

Input

IEEE 1149.1 test logic reset

Tck_H

Input

IEEE 1149.1 test clock

Tdo_H

Output

IEEE 1149.1 test data output

SromData_H

Input

SROM data/Diagnostic terminal data input

21264/EV68A Hardware Reference Manual

Testability and Diagnostics

11–1

SROM/Serial Diagnostic Terminal Port

Table 11–1 Dedicated Test Port Pins (Continued)
Pin Name

Type

Function

SromClk_H

Output

SROM clock/Diagnostic terminal data output

SromOE_L

Output

SROM enable/Diagnostic terminal enable

TestStat_H

Output

BiST status/timeout output

11.2 SROM/Serial Diagnostic Terminal Port
This port supports two functions. During power-up, it supports automatic initialization
of the Cbox configuration registers and the Icache from the system serial ROMs. After
power-up, it supports a serial diagnostic terminal.

11.2.1 SROM Load Operation
The following actions are performed while the SROM is loaded:

•

The SromOE_L pin supplies the output enable as well as the reset to the serial
ROM. (Refer to the serial ROM specifications for details.) The 21264/EV68A
asserts this signal low for the duration of the Icache load from the serial ROM.
When the load has been completed, the signal remains deasserted.

•

The SromClk_H pin supplies the clock to the SROM that causes it to advance to
the next bit. Simultaneously, it causes the existing data on the SromData_H pin to
be shifted into an internal shift register. The cycle time of this clock is 256 times the
CPU clock rate. (If the FASTROM flag is set, the rate is 16 times the CPU clock
rate.) The hold time on SromData_H is 2* CPU cycle time with respect to
SromClk_H.

•

The SromData_H pin reads data from the SROM.

Every data and tag bit in Icache is loaded by that sequence.

11.2.2 Serial Terminal Port
After the SROM data is loaded into the Icache, the three SROM interface signals can be
used as a software UART and the pins become parallel I/O pins that can drive a system
debug or diagnostic terminal by using an interface such as RS422.
The serial line interface is automatically enabled if the SromOE_L pin is wired to the
following pins:

•

An active high enable RS422 (or 26LS32) driver, driving to SromData_H

•

An active high enable RS422 (or 26LS31) receiver, driven from SromClk_H

After reset, SromClk_H is driven from the Ibox I_CTL[SL_XMIT]. This register is
cleared during reset, so it starts driving as a 0, but it can be written by software. The
data becomes available at the pin after the HW_MTPR instruction that wrote
I_CTL[SL_XMIT] is retired.

11–2

Testability and Diagnostics

21264/EV68A Hardware Reference Manual

IEEE 1149.1 Port

On the receive side, while in native mode, any transition on the Ibox I_CTL
[SL_RCV], driven from the SromData_H pin, results in a trap to the PALcode interrupt handler. When in PALmode, all interrupts are blocked. The interrupt routine then
begins sampling I_CTL [SL_RCV] under a software timing loop to input as much data
as needed, using the chosen serial line protocol.

11.3 IEEE 1149.1 Port
The IEEE 1149.1 Test Access Port consists of the Tdi_H, Tdo_H, Tms_H, Tck_H,
and Trst_L pins. These pins access the IEEE 1149.1 mandated public test features as
well as several private chip manufacturing test features.
The port meets all requirements of the standard except that there are no pull-ups on the
Tdi_H, Tms_H, and Trst_L pins, as required by the present standard.
The scope of 1149.1 compliant features on the 21264/EV68A is limited to the board
level assembly verification test. The systems that do not intend to drive this port must
terminate the port pins as follows: pull-ups on Tdi_H and Tms_H, pull-downs on
Tck_H and Trst_L.
The port logic consists of the usual standard compliant components, namely, the TAP
Controller State Machine, the Instruction Register, and the Bypass Register.
The Bypass Register provides a short shift path through the chip’s IEEE 1149.1 logic. It
is generally useful at the board level testing. It consists of a 1-bit shift register.
The Instruction Register holds test instructions. On the 21264/EV68A, this register is 5
bits wide. Table 11–2 describes the supported instructions. The instruction set supports
several public and private instructions. The public instructions operate and produce
behavior compliant with the standard. The private instructions are used for chip manufacturing test and must not be used outside of chip manufacturing.
Table 11–2 IEEE 1149.1 Instructions and Opcodes
Opcode

Instruction

Operation/Function

00xxx
01xxx
10xxx

Private

These instructions are for factory test use only. The user must
not load them as they may have a harmful effect on the
21264/EV68A.

11000

SAMPLE

IEEE 1149.1 SAMPLE instruction.

11001

HIGHZ

IEEE 1149.1 HIGHZ instruction.

11010

CLAMP

IEEE 1149.1 CLAMP instruction.

11011

EXTEST

IEEE 1149.1 EXTEST instruction.

11100
11101
11110

Private

These instructions are for factory test use only. The user must
not load them as they may have a harmful effect on the
21264/EV68A.

11111

BYPASS

IEEE 1149.1 BYPASS instruction.

Figure 11–3 shows the TAP controller state machine state diagram. The signal Tms_H
controls the state transitions that occur with the rising clock edge. TAP state machine
states are decoded and used for initiating various actions for testing.

21264/EV68A Hardware Reference Manual

Testability and Diagnostics

11–3

TestStat_H Pin

Table 11–3 TAP Controller State Machine

Test Logic
Reset
0
Run-Test/Idle

Select-DR-Scan

0
1

Capture-DR

Capture-IR
0

Shift-IR

Shift-DR

0
1

1
1

Exit1-DR

Values
shown
are for
TMS.

Select-IR-Scan

Exit1-IR

0
0

Pause-DR

Pause-IR
0

1
0

Exit2-DR

Exit2-IR

Update-DR
1

Scan Sequence

Update-IR
1

Scan Sequence
MK145508.AI4

11.4 TestStat_H Pin
The TestStat_H pin serves two purposes. During power-up, it indicates BiST pass/fail
status. After power-up, it indicates the 21264/EV68A timeout event.
The system reset forces TestStat_H to low. Tbox forces it high during the internal BiST
and array initialization operations. During result extraction (DoResult state), the Tbox
drives it low for 16 cycles. After that, the pin remains low if the BiST has passes, otherwise, it is asserted high and remains high until chip is reset again. Figure 11–1 pictorially shows the behavior of the pin during the power-up operations.
Note:

11–4

A system designer may sample the TestStat_H pin on the first rising edge
of the SromClk_H pin to determine BiST results. After the power-up during the normal chip operation, whenever the 21264/EV68A does not retire
an instruction for 2K CPU cycles, the pin is asserted high for 3 CPU cycles.

Testability and Diagnostics

21264/EV68A Hardware Reference Manual

Power-Up Self-Test and Initialization

Figure 11–1 TestStat_H Pin Timing During Power-Up Built-In Self-Test (BiST)
ClkFwdReset_L
Tbox_Reset_A_L
TBox Reset Engine

Idle

DoBist

DoResult

TestStatus_H

DoSROM

Idle

BiSTResult

BiSTResult OR T

LKG-10950A-98WF

Figure 11–2 TestStat_H Pin Timing During Built-In Self-Initialization (BiSI)
Tbox_Rst_A_L1
TBox Reset Engine1

Idle

DoMfgSelfinit

TestStatus_H1

Idle

TimeOut

ClkFwdRst_L1
LKG-10951A-98WF

11.5 Power-Up Self-Test and Initialization
Upon powering up, the 21264/EV68A automatically performs the self-test of all major
embedded RAM arrays and then loads the Cbox configuration registers and the instruction cache from the system SROM. The chip’s internal logic is held in reset during
these operations. See Chapter 9 for sequencing of power-up operations.

11.5.1 Built-in Self-Test
The power-up self-test is performed on the instruction cache and tag arrays, the data
cache and tag arrays, the triplicate tag arrays, and the various RAM arrays located in the
branch history table logic. The power-up self-test lasts for approximately 700,000 CPU
cycles. The result of self-test is made available as Pass/Fail status on the TestStat_H
pin (see Section 11.4).
The result of self-test is also available in an IPR bit. Software can read this status
through IPR I_CTL(23) (0 = pass, 1 = fail). See Section 5.2.15.
The power-up BiST leaves all bits in all arrays initialized to zeroes. The instruction
cache and the tag are reinitialized as part of the SROM initialization step. This is
detailed in Section 11.5.2.

11.5.2 SROM Initialization
Power-up initialization on the 21264/EV68A is different from previous generation
Alpha systems in two aspects. First, in the 21264/EV68A systems, the presence of
serial ROMs is mandatory as initialization of several Cbox configuration registers
depends on them. Second, it is possible to skip or partially fill Icache from serial
ROMs. Figure 11–3 shows the map of the data in serial ROMs.

21264/EV68A Hardware Reference Manual

Testability and Diagnostics

11–5

Power-Up Self-Test and Initialization

In the SROM represented in Figure 11–3, the length for fields Cbox Config
Data(0,n) plus MBZ(m,0) must equal 367 bits. (If Cbox Config Data(0,n) is
(0,366), MBZ would be zero.)
For the 21264/EV68A, Cbox Config Data is 304 bits; the value for n is 303.
Therefore, the value MBZ field for Pass 3 is:
MBZ(m,0) = 367 minus 304 = 63 = (62,0)
Tables 11–4 and 5–24 describe the details of the Icache and Cbox bit fields, respectively. Note that fetch_count(1,0) must be 3, which guarantees that the SROM never
partially loads an Icache block.
Figure 11–3 SROM Content Map
fetch [0](0,192) fetch[j-1](0,192) fetch[j](0,192) fetch_count(11,0) Cbox Config Data(0, n)
(first block)
(last block)

MBZ(m,0)

11.5.2.1 Serial Instruction Cache Load Operation

All Icache bits, including each block’s tag, address space number (ASN), address space
match (ASM), and valid and branch history bits are loaded serially from offchip serial
ROMs. Once the serial load has been invoked by the chip reset sequence, the cache is
loaded from the lower to the higher addresses.
The serial Icache fill invoked by the chip reset sequence operates internally at a fre--------------- .
quency of GCLK
256

Table 11–4 lists the Icache bit fields in an SROM line. Fetch bits are listed in the order
of shift direction (to down and to right). In Table 11–4:
Bit Type

Meaning

c
i
iq
tr
dv
ea
par-MBZ

Disp_add carry
Instruction
Iqueue predecodes
Trouble bits
Destination valid
Ea_src
Must be zero

The load occurs at the rate of 1 bit per 256 CPU cycles. The chip outputs a 50% duty
cycle clock on the SromClk_H pin.
The serial ROMs can contain enough Alpha code to complete the configuration of the
external interface (for example, set the timing on the external cache RAMs, and diagnose the path between the CPU chip and the real ROM).

11–6

Testability and Diagnostics

21264/EV68A Hardware Reference Manual

Notes on IEEE 1149.1 Operation and Compliance

The instruction cache lines are loaded in the reverse order. If the fetch_count(9,0) is
zero, then, no instruction cache lines are loaded. Since the valid bits are already cleared
by the BiST operation, the first instruction fetch is missed in the instruction cache and
the chip seeks instructions from the offchip memory.
Table 11–4 Icache Bit Fields in an SROM Line
Fetch Bit

Icache Data

Fetch Bit

Icache Data

Fetch Bit Icache Data

par-MBZ

172

lp_train

c[3]

c[0]

173:175

lp_src(2:0)

2:27

i[3](25,20,24,19,23,18,22,17 88:113
,21,16:0)

i[0](25,20,24,19,3,18,22,17, 176:181
21,16:0)

lp_idx(14:9)

c[2]

114

c[1]

182:186

lp_idx(8:4)

29:42

i[2](25,20,24,19,
23,18,22,17,21,16:12)

115:128

i[1] (25,20,24,19,
23,18,22,17,21,16:12)

187

lp_idx(15)

parity

129

parity

188:192

lp_ssp[4:0]

44:55

i[2](11:0)

130:141

i[1](11:0)

—

dv[3]

142

dv[0]

—

57:59

iq[3](2:0)

143:145

iq[0](2:0)

—

60:65

i[3](26:31)

146:151

i[0](26:31)

—

66,68

ea[3](2:0)

152:154

ea[0](2:0)

—

dv[2]

155

dv[1]

—

70,72

iq[2](2:0)

156:158

iq[1](2:0)

—

73:78

i[2](26:31)

159:164

i[1](26:31)

—

79:81

ea[2](2:0)

165:167

ea[1](2:0)

—

82:85

tr(7:4)

168:171

tr(0:3)

—

Refer to the Alpha Motherboards Software Developer’s Kit (SDK) for example C code
that calculates the predecode values of a serial Icache load.

11.6 Notes on IEEE 1149.1 Operation and Compliance
1. IEEE 1149.1 port pins on the 21264/EV68A are not pulled up or pulled down on
the chip. The necessary pull-up or pull-down function must be implemented on the
board.
2. Tms_H should not change when Trst_L is being deasserted.
References

IEEE Std. 1149.1-1993 A Test Access Port and Boundary Scan Architecture.
See Appendix B for a listing of the Boundary-Scan Register.

21264/EV68A Hardware Reference Manual

Testability and Diagnostics

11–7

A
Alpha Instruction Set
This appendix provides a summary of the Alpha instruction set and describes the
21264/EV68A IEEE floating-point conformance. It is organized as follows:

•

Alpha instruction summary

•

Reserved opcodes

•

IEEE floating-point instructions

•

VAX floating-point instructions

•

Independent floating-point instructions

•

Opcode summary

•

Required PALcode function codes

•

IEEE floating-point conformance

A.1 Alpha Instruction Summary
This section contains a summary of all Alpha architecture instructions. All values are in
hexadecimal radix. Table A–1 describes the contents of the Format and Opcode columns that are in Table A–2.
Table A–1 Instruction Format and Opcode Notation
Instruction Format

Format
Symbol

Opcode
Notation

Meaning

Branch

Bra

oo is the 6-bit opcode field.

Floating-point

F-P

oo.fff

oo is the 6-bit opcode field.
fff is the 11-bit function code field.

Memory

Mem

oo is the 6-bit opcode field.

oo.ffff

oo is the 6-bit opcode field.
ffff is the 16-bit function code in the displacement
field.

Memory/function code Mfc

21264/EV68A Hardware Reference Manual

Alpha Instruction Set

A–1

Alpha Instruction Summary

Table A–1 Instruction Format and Opcode Notation (Continued)
Instruction Format

Format
Symbol

Opcode
Notation

Memory/ branch

Mbr

oo.h

oo is the 6-bit opcode field.
h is the high-order 2 bits of the displacement field.

Operate

Opr

oo.ff

oo is the 6-bit opcode field.
ff is the 7-bit function code field.

PALcode

Pcd

oo is the 6-bit opcode field; the particular PALcode instruction is specified in the 26-bit function
code field.

Meaning

Qualifiers for operate instructions are shown in Table A–2. Qualifiers for IEEE and
VAX floating-point instructions are shown in Tables A–5 and A–6, respectively.
Table A–2 Architecture Instructions
Mnemonic

Format

Opcode

Description

ADDF

F-P

15.080

Add F_floating

ADDG

F-P

15.0A0

Add G_floating

ADDL

Opr

10.00

Add longword

ADDL/V

Opr

10.40

Add longword with integer overflow enable

ADDQ

Opr

10.20

Add quadword

ADDQ/V

Opr

10.60

Add quadword with integer overflow enable

ADDS

F-P

16.080

Add S_floating

ADDT

F-P

16.0A0

Add T_floating

AMASK

Opr

11.61

Architecture mask

AND

Opr

11.00

Logical product

BEQ

Bra

Branch if = zero

BGE

Bra

Branch if ≥ zero

BGT

Bra

Branch if > zero

BIC

Opr

11.08

Bit clear

BIS

Opr

11.20

Logical sum

BLBC

Bra

Branch if low bit clear

BLBS

Bra

Branch if low bit set

BLE

Bra

Branch if ≤ zero

BLT

Bra

Branch if < zero

BNE

Bra

Branch if ≠ zero

Bra

Unconditional branch

A–2

Alpha Instruction Set

21264/EV68A Hardware Reference Manual

Alpha Instruction Summary

Table A–2 Architecture Instructions (Continued)
Mnemonic

Format

Opcode

Description

BSR

Mbr

Branch to subroutine

CALL_PAL

Pcd

Trap to PALcode

CMOVEQ

Opr

11.24

CMOVE if = zero

CMOVGE

Opr

11.46

CMOVE if ≥ zero

CMOVGT

Opr

11.66

CMOVE if > zero

CMOVLBC

Opr

11.16

CMOVE if low bit clear

CMOVLBS

Opr

11.14

CMOVE if low bit set

CMOVLE

Opr

11.64

CMOVE if ≤ zero

CMOVLT

Opr

11.44

CMOVE if < zero

CMOVNE

Opr

11.26

CMOVE if ≠ zero

CMPBGE

Opr

10.0F

Compare byte

CMPEQ

Opr

10.2D

Compare signed quadword equal

CMPGEQ

F-P

15.0A5

Compare G_floating equal

CMPGLE

F-P

15.0A7

Compare G_floating less than or equal

CMPGLT

F-P

15.0A6

Compare G_floating less than

CMPLE

Opr

10.6D

Compare signed quadword less than or equal

CMPLT

Opr

10.4D

Compare signed quadword less than

CMPTEQ

F-P

16.0A5

Compare T_floating equal

CMPTLE

F-P

16.0A7

Compare T_floating less than or equal

CMPTLT

F-P

16.0A6

Compare T_floating less than

CMPTUN

F-P

16.0A4

Compare T_floating unordered

CMPULE

Opr

10.3D

Compare unsigned quadword less than or equal

CMPULT

Opr

10.1D

Compare unsigned quadword less than

CPYS

F-P

17.020

Copy sign

CPYSE

F-P

17.022

Copy sign and exponent

CPYSN

F-P

17.021

Copy sign negate

CTLZ

Opr

1C.32

Count leading zero

CTPOP

Opr

1C.30

Count population

CTTZ

Opr

1C.33

Count trailing zero

CVTDG

F-P

15.09E

Convert D_floating to G_floating

CVTGD

F-P

15.0AD Convert G_floating to D_floating

CVTGF

F-P

15.0AC

Convert G_floating to F_floating

21264/EV68A Hardware Reference Manual

Alpha Instruction Set

A–3

Alpha Instruction Summary

Table A–2 Architecture Instructions (Continued)
Mnemonic

Format

Opcode

Description

CVTGQ

F-P

15.0AF

Convert G_floating to quadword

CVTLQ

F-P

17.010

Convert longword to quadword

CVTQF

F-P

15.0BC

Convert quadword to F_floating

CVTQG

F-P

15.0BE

Convert quadword to G_floating

CVTQL

F-P

17.030

Convert quadword to longword

CVTQS

F-P

16.0BC

Convert quadword to S_floating

CVTQT

F-P

16.0BE

Convert quadword to T_floating

CVTST

F-P

16.2AC

Convert S_floating to T_floating

CVTTQ

F-P

16.0AF

Convert T_floating to quadword

CVTTS

F-P

16.0AC

Convert T_floating to S_floating

DIVF

F-P

15.083

Divide F_floating

DIVG

F-P

15.0A3

Divide G_floating

DIVS

F-P

16.083

Divide S_floating

DIVT

F-P

16.0A3

Divide T_floating

ECB

Mfc

18.E800 Evict cache block

EQV

Opr

11.48

EXCB

Mfc

18.0400 Exception barrier

EXTBL

Opr

12.06

Extract byte low

EXTLH

Opr

12.6A

Extract longword high

EXTLL

Opr

12.26

Extract longword low

EXTQH

Opr

12.7A

Extract quadword high

EXTQL

Opr

12.36

Extract quadword low

EXTWH

Opr

12.5A

Extract word high

EXTWL

Opr

12.16

Extract word low

FBEQ

Bra

Floating branch if = zero

FBGE

Bra

Floating branch if ≥ zero

FBGT

Bra

Floating branch if > zero

FBLE

Bra

Floating branch if ≤ zero

FBLT

Bra

Floating branch if < zero

FBNE

Bra

Floating branch if ≠ zero

FCMOVEQ

F-P

17.02A

FCMOVE if = zero

FCMOVGE

F-P

17.02D

FCMOVE if ≥ zero

A–4

Alpha Instruction Set

Logical equivalence

21264/EV68A Hardware Reference Manual

Alpha Instruction Summary

Table A–2 Architecture Instructions (Continued)
Mnemonic

Format

Opcode

Description

FCMOVGT

F-P

17.02F

FCMOVE if > zero

FCMOVLE

F-P

17.02E

FCMOVE if ≤ zero

FCMOVLT

F-P

17.02C

FCMOVE if < zero

FCMOVNE

F-P

17.02B

FCMOVE if ≠ zero

FETCH

Mfc

18.8000 Prefetch data

FETCH_M

Mfc

18.A000 Prefetch data, modify intent

FTOIS

F-P

1C.78

Floating to integer move, S_floating

FTOIT

F-P

1C.70

Floating to integer move, T_floating

IMPLVER

Opr

11.6C

Implementation version

INSBL

Opr

12.0B

Insert byte low

INSLH

Opr

12.67

Insert longword high

INSLL

Opr

12.2B

Insert longword low

INSQH

Opr

12.77

Insert quadword high

INSQL

Opr

12.3B

Insert quadword low

INSWH

Opr

12.57

Insert word high

INSWL

Opr

12.1B

Insert word low

ITOFF

F-P

14.014

Integer to floating move, F_floating

ITOFS

F-P

14.004

Integer to floating move, S_floating

ITOFT

F-P

14.024

Integer to floating move, T_floating

JMP

Mbr

1A.0

Jump

JSR

Mbr

1A.1

Jump to subroutine

JSR_COROUTINE Mbr

1A.3

Jump to subroutine return

LDA

Mem

Load address

LDAH

Mem

Load address high

LDBU

Mem

Load zero-extended byte

LDF

Mem

Load F_floating

LDG

Mem

Load G_floating

LDL

Mem

Load sign-extended longword

LDL_L

Mem

Load sign-extended longword locked

LDQ

Mem

Load quadword

LDQ_L

Mem

Load quadword locked

LDQ_U

Mem

Load unaligned quadword

21264/EV68A Hardware Reference Manual

Alpha Instruction Set

A–5

Alpha Instruction Summary

Table A–2 Architecture Instructions (Continued)
Mnemonic

Format

Opcode

Description

LDS

Mem

Load S_floating

LDT

Mem

Load T_floating

LDWU

Mem

Load zero-extended word

MAXSB8

Opr

1C.3E

Vector signed byte maximum

MAXSW4

Opr

1C.3F

Vector signed word maximum

MAXUB8

Opr

1C.3C

Vector unsigned byte maximum

MAXUW4

Opr

1C.3D

Vector unsigned word maximum

Mfc

18.4000 Memory barrier

MF_FPCR

F-P

17.025

Move from FPCR

MINSB8

Opr

1C.38

Vector signed byte minimum

MINSW4

Opr

1C.39

Vector signed word minimum

MINUB8

Opr

1C.3A

Vector unsigned byte minimum

MINUW4

Opr

1C.3B

Vector unsigned word minimum

MSKBL

Opr

12.02

Mask byte low

MSKLH

Opr

12.62

Mask longword high

MSKLL

Opr

12.22

Mask longword low

MSKQH

Opr

12.72

Mask quadword high

MSKQL

Opr

12.32

Mask quadword low

MSKWH

Opr

12.52

Mask word high

MSKWL

Opr

12.12

Mask word low

MT_FPCR

F-P

17.024

Move to FPCR

MULF

F-P

15.082

Multiply F_floating

MULG

F-P

15.0A2

Multiply G_floating

MULL

Opr

13.00

Multiply longword

MULL/V

Opr

13.40

Multiply longword with integer overflow enable

MULQ

Opr

13.20

Multiply quadword

MULQ/V

Opr

13.60

Multiply quadword with integer overflow enable

MULS

F-P

16.082

Multiply S_floating

MULT

F-P

16.0A2

Multiply T_floating

ORNOT

Opr

11.28

Logical sum with complement

PERR

Opr

1C.31

Pixel error

PKLB

Opr

1C.37

Pack longwords to bytes

A–6

Alpha Instruction Set

21264/EV68A Hardware Reference Manual

Alpha Instruction Summary

Table A–2 Architecture Instructions (Continued)
Mnemonic

Format

Opcode

Description

PKWB

Opr

1C.36

Pack words to bytes

Mfc

18.E000 Read and clear

RET

Mbr

1A.2

RPCC

Mfc

18.C000 Read process cycle counter

Mfc

18.F000 Read and set

S4ADDL

Opr

10.02

Scaled add longword by 4

S4ADDQ

Opr

10.22

Scaled add quadword by 4

S4SUBL

Opr

10.0B

Scaled subtract longword by 4

S4SUBQ

Opr

10.2B

Scaled subtract quadword by 4

S8ADDL

Opr

10.12

Scaled add longword by 8

S8ADDQ

Opr

10.32

Scaled add quadword by 8

S8SUBL

Opr

10.1B

Scaled subtract longword by 8

S8SUBQ

Opr

10.3B

Scaled subtract quadword by 8

SEXTB

Opr

1C.00

Sign extend byte

SEXTW

Opr

1C.01

Sign extend word

SLL

Opr

12.39

Shift left logical

SQRTF

F-P

14.08A

Square root F_floating

SQRTG

F-P

14.0AA Square root G_floating

SQRTS

F-P

14.08B

Square root S_floating

SQRTT

F-P

14.0AB

Square root T_floating

SRA

Opr

12.3C

Shift right arithmetic

SRL

Opr

12.34

Shift right logical

STB

Mem

Store byte

STF

Mem

Store F_floating

STG

Mem

Store G_floating

STL

Mem

Store longword

STL_C

Mem

Store longword conditional

STQ

Mem

Store quadword

STQ_C

Mem

Store quadword conditional

STQ_U

Mem

Store unaligned quadword

STS

Mem

Store S_floating

STT

Mem

Store T_floating

Return from subroutine

21264/EV68A Hardware Reference Manual

Alpha Instruction Set

A–7

Reserved Opcodes

Table A–2 Architecture Instructions (Continued)
Mnemonic

Format

Opcode

Description

STW

Mem

Store word

SUBF

F-P

15.081

Subtract F_floating

SUBG

F-P

15.0A1

Subtract G_floating

SUBL

Opr

10.09

Subtract longword

SUBL/V

Opr

10.49

Subtract longword with integer overflow enable

SUBQ

Opr

10.29

Subtract quadword

SUBQ/V

Opr

10.69

Subtract quadword with integer overflow enable

SUBS

F-P

16.081

Subtract S_floating

SUBT

F-P

16.0A1

Subtract T_floating

TRAPB

Mfc

18.0000 Trap barrier

UMULH

Opr

13.30

Unsigned multiply quadword high

UNPKBL

Opr

1C.35

Unpack bytes to longwords

UNPKBW

Opr

1C.34

Unpack bytes to words

WH64

Mfc

18.F800 Write hint — 64 bytes

WMB

Mfc

18.4400 Write memory barrier

XOR

Opr

11.40

Logical difference

ZAP

Opr

12.30

Zero bytes

ZAPNOT

Opr

12.31

Zero bytes not

A.2 Reserved Opcodes
This section describes the opcodes that are reserved in the Alpha architecture. They can
be reserved for Compaq or for PALcode.

A.2.1 Opcodes Reserved for Compaq
Table A–3 lists opcodes reserved for Compaq.
Table A–3 Opcodes Reserved for Compaq

A–8

Mnemonic

Opcode

Mnemonic

Opcode

OPC01

OPC05

OPC02

OPC06

OPC03

OPC07

OPC04

—

Alpha Instruction Set

21264/EV68A Hardware Reference Manual

IEEE Floating-Point Instructions

A.2.2 Opcodes Reserved for PALcode
Table A–4 lists the 21264/EV68A-specific instructions. See Chapter 2 for more
information.
Table A–4 Opcodes Reserved for PALcode
21264/EV68A
Mnemonic

Opcode

Architecture
Mnemonic

Function

HW_LD

PAL1B

Performs Dstream load instructions.

HW_ST

PAL1F

Performs Dstream store instructions.

HW_REI

PAL1E

Returns instruction flow to the program counter (PC) pointed
to by EXC_ADDR internal processor register (IPR).

HW_MFPR

PAL19

Accesses the Ibox, Mbox, and Dcache IPRs.

HW_MTPR

PAL1D

Accesses the Ibox, Mbox, and Dcache IPRs.

A.3 IEEE Floating-Point Instructions
Table A–5 lists the hexadecimal value of the 11-bit function code field for the IEEE
floating-point instructions, with and without qualifiers. The opcode for these
instructions is 1616.
Table A–5 IEEE Floating-Point Instruction Function Codes
Mnemonic

None

/UC

/UM

/UD

ADDS

080

000

040

0C0

180

100

140

1C0

ADDT

0A0

020

060

0E0

1A0

120

160

1E0

CMPTEQ

0A5

—

CMPTLT

0A6

—

CMPTLE

0A7

—

CMPTUN

0A4

—

CVTQS

0BC

03C

07C

0FC

—

CVTQT

0BE

03E

07E

0FE

—

See
below

—

CVTST

See
below

—

CVTTQ
CVTTS

0AC

02C

06C

0EC

1AC

12C

16C

1EC

DIVS

083

003

043

0C3

183

103

143

1C3

DIVT

0A3

023

063

0E3

1A3

123

163

1E3

MULS

082

002

042

0C2

182

102

142

1C2

MULT

0A2

022

062

0E2

1A2

122

162

1E2

21264/EV68A Hardware Reference Manual

Alpha Instruction Set

A–9

IEEE Floating-Point Instructions

Table A–5 IEEE Floating-Point Instruction Function Codes (Continued)
SQRTS

08B

00B

04B

0CB

18B

10B

14B

1CB

SQRTT

0AB

02B

06B

0EB

1AB

12B

16B

1EB

SUBS

081

001

041

0C1

181

101

141

1C1

SUBT

0A1

021

061

0E1

1A1

121

161

1E1

Mnemonic

/SU

/SUC

/SUM

/SUD

/SUI

/SUIC

/SUIM

/SUID

ADDS

580

500

540

5C0

780

700

740

7C0

ADDT

5A0

520

560

5E0

7A0

720

760

7E0

CMPTEQ

5A5

CMPTLT

5A6

CMPTLE

5A7

CMPTUN

5A4

CVTQS

7BC

73C

77C

7FC

CVTQT

7BE

73E

77E

7FE

CVTTS

5AC

52C

56C

5EC

7AC

72C

76C

7EC

DIVS

583

503

543

5C3

783

703

743

7C3

DIVT

5A3

523

563

5E3

7A3

723

763

7E3

MULS

582

502

542

5C2

782

702

742

7C2

MULT

5A2

522

562

5E2

7A2

722

762

7E2

SQRTS

58B

50B

54B

5CB

78B

70B

74B

7CB

SQRTT

5AB

52B

56B

5EB

7AB

72B

76B

7EB

SUBS

581

501

541

5C1

781

701

741

7C1

SUBT

5A1

521

561

5E1

7A1

721

761

7E1

Mnemonic

None

CVTST

2AC

6AC

Mnemonic

None

/VC

/SV

/SVC

/SVI

/SVIC

CVTTQ

0AF

02F

1AF

12F

5AF

52F

7AF

72F

Mnemonic

/VD

/SVD

/SVID

/VM

/SVM

/SVIM

CVTTQ

0EF

1EF

5EF

7EF

06F

16F

56F

76F

A–10 Alpha Instruction Set

21264/EV68A Hardware Reference Manual

VAX Floating-Point Instructions

Programming Note:

In order to use CMPTxx with software completion trap handling, it is necessary to
specify the /SU IEEE trap mode, even though an underflow trap is not possible. In order
to use CVTQS or CVTQT with software completion trap handling, it is necessary to
specify the /SUI IEEE trap mode, even though an underflow trap is not possible.

A.4 VAX Floating-Point Instructions
Table A–6 lists the hexadecimal value of the 11-bit function code field for the VAX
floating-point instructions. The opcode for these instructions is 1516.
Table A–6 VAX Floating-Point Instruction Function Codes
Mnemonic

None

/UC

/SC

/SU

/SUC

ADDF

080

000

180

100

480

400

580

500

ADDG

0A0

020

1A0

120

4A0

420

5A0

520

CMPGEQ

0A5

4A5

CMPGLE

0A7

4A7

CMPGLT

0A6

4A6

CVTDG

09E

01E

19E

11E

49E

41E

59E

51E

CVTGD

0AD

02D

1AD

12D

4AD

42D

5AD

52D

CVTGF

0AC

02C

1AC

12C

4AC

42C

5AC

52C

CVTGQ

See below

CVTQF

0BC

03C

CVTQG

0BE

03E

DIVF

083

003

183

103

483

403

583

503

DIVG

0A3

023

1A3

123

4A3

423

5A3

523

MULF

082

002

182

102

482

402

582

502

MULG

0A2

022

1A2

122

4A2

422

5A2

522

SQRTF

08A

00A

18A

10A

48A

40A

58A

50A

SQRTG

0AA

02A

1AA

12A

4AA

42A

5AA

52A

SUBF

081

001

181

101

481

401

581

501

SUBG

0A1

021

1A1

121

4A1

421

5A1

521

Mnemonic

None

/VC

/SC

/SV

/SVC

CVTGQ

0AF

02F

1AF

12F

4AF

42F

5AF

52F

A.5 Independent Floating-Point Instructions
Table A–7 lists the hexadecimal value of the 11-bit function code field for the floatingpoint instructions that are not directly tied to IEEE or VAX floating point. The opcode
for the following instructions is 17 16.

21264/EV68A Hardware Reference Manual

Alpha Instruction Set A–11

Opcode Summary

Table A–7 Independent Floating-Point Instruction Function Codes
Mnemonic

None

/SV

CPYS

—

CPYSE

020
022

—

CPYSN

021

—

CVTLQ

010

CVTQL

030

—
130

—
530

FCMOVEQ

02A

—

FCMOVGE

02D

—

FCMOVGT

02F

—

FCMOVLE

02E

—

FCMOVLT

02C

—

MF_FPCR

025

—

MT_FPCR

024

—

A.6 Opcode Summary
Table A–8 lists all Alpha opcodes from 00 (CALL_PAL) through 3F (BGT). In the
table, the column headings that appear over the instructions have a granularity of 816.
The rows beneath the Offset column supply the individual hexadecimal number to
resolve that granularity.
If an instruction column has a 0 in the right (low) hexadecimal digit, replace that 0 with
the number to the left of the backslash (\) in the Offset column on the instruction’s row.
If an instruction column has an 8 in the right (low) hexadecimal digit, replace that 8
with the number to the right of the backslash in the Offset column.
For example, the third row (2/A) under the 1016 column contains the symbol INTS*,
representing the all-integer shift instructions. The opcode for those instructions would
then be 1216 because the 0 in 10 is replaced by the 2 in the Offset column. Likewise, the
third row under the 1816 column contains the symbol JSR*, representing all jump
instructions. The opcode for those instructions is 1A because the 8 in the heading is
replaced by the number to the right of the backslash in the Offset column. The
instruction format is listed under the instruction symbol.
Table A–8 Opcode Summary
Offset

0/8

PAL*
(pal)

LDA
(mem)

1/9

Res

2/A

LDBU

INTA* MISC*
(op)
(mem)

LDF
(mem)

LDL
(mem)

BR
(br)

BLBC
(br)

LDAH
(mem)

INTL*
(op)

\PAL\

LDG
(mem)

LDQ
(mem)

FBEQ
(br)

BEQ
(br)

Res

INTS*
(op)

JSR*
(mem)

LDS
(mem)

LDL_L
(mem)

FBLT
(br)

BLT
(br)

A–12 Alpha Instruction Set

21264/EV68A Hardware Reference Manual

Required PALcode Function Codes

Table A–8 Opcode Summary (Continued)
Offset

3/B

Res

LDQ_U INTM* \PAL\
(mem) (op)

LDT
(mem)

LDQ_L
(mem)

FBLE
(br)

BLE
(br)

4/C

LDWU Res

ITFP*

FPTI*

STF
(mem)

STL
(mem)

BSR
(br)

BLBS
(br)

5/D

Res

STW

FLTV* \PAL\
(op)

STG
(mem)

STQ
(mem)

FBNE
(br)

BNE
(br)

6/E

Res

STB

FLTI*
(op)

\PAL\

STS
(mem)

STL_C
(mem)

FBGE
(br)

BGE
(br)

7/F

Res

STQ_U
(mem)

FLTL* \PAL\
(op)

STT
(mem)

STQ_C
(mem)

FBGT
(br)

BGT
(br)

Table A–9 explains the symbols used in Table A–8.
Table A–9 Key to Opcode Summary Used in Table A–8
Symbol

Meaning

FLTI*
FLTL*
FLTV*
FPTI*
INTA*
INTL*
INTM*
INTS*
ITFP*
JSR*
MISC*
PAL*
\PAL\
Res

IEEE floating-point instruction opcodes
Floating-point operate instruction opcodes
VAX floating-point instruction opcodes
Floating-point to integer register move opcodes
Integer arithmetic instruction opcodes
Integer logical instruction opcodes
Integer multiply instruction opcodes
Integer shift instruction opcodes
Integer to floating-point register move opcodes
Jump instruction opcodes
Miscellaneous instruction opcodes
PALcode instruction (CALL_PAL) opcodes
Reserved for PALcode
Reserved for Compaq

A.7 Required PALcode Function Codes
Table A–10 lists opcodes required for all Alpha implementations. The notation used is
oo.ffff, where oo is the hexadecimal 6-bit opcode and ffff is the hexadecimal 26-bit
function code.
Table A–10 Required PALcode Function Codes
Mnemonic

Type

Function Code

DRAINA

Privileged

00.0002

HALT

Privileged

00.0000

IMB

Unprivileged

00.0086

21264/EV68A Hardware Reference Manual

Alpha Instruction Set A–13

IEEE Floating-Point Conformance

A.8 IEEE Floating-Point Conformance
The 21264/EV68A supports the IEEE floating-point operations defined in the Alpha
System Reference Manual, Revision 8 and therefore also from the Alpha Architecture
Reference Manual, Fourth Edition. Support for a complete implementation of the IEEE
Standard for Binary Floating-Point Arithmetic (ANSI/IEEE Standard 754 1985) is provided by a combination of hardware and software. The 21264/EV68A provides several
hardware features to facilitate complete support of the IEEE standard.
The 21264/EV68A provides the following hardware features to facilitate complete support of the IEEE standard:

•

The 21264/EV68A implements precise exception handling in hardware, as denoted
by the AMASK instruction returning bit 9 set. TRAPB instructions are treated as
NOPs and are not issued.

•

The 21264/EV68A accepts both Signaling and Quiet NaNs as input operands and
propagates them as specified by the Alpha architecture. In addition, the 21264/
EV68A delivers a canonical Quiet NaN when an operation is required to produce a
NaN value and none of its inputs are NaNs. Encodings for Signaling NaN and
Quiet NaN are defined by the Alpha Architecture Reference Manual, Fourth Edition.

•

The 21264/EV68A accepts infinity operands and implements infinity arithmetic as
defined by the IEEE standard and the Alpha Architecture Reference Manual,
Fourth Edition.

•

The 21264/EV68A implements SQRT for single (SQRTS) and double (SQRTT)
precision in hardware.
In addition, the 21264/EV68A also implements the VAX SQRTF and
SQRTG instructions.

Note:

•

The 21264/EV68A implements the FPCR[DNZ] bit. When FPCR[DNZ] is set,
denormal input operand traps can be avoided for arithmetic operations that include
the /S qualifier. When FPCR[DNZ] is clear, denormal input operands for arithmetic
operations produce an unmaskable denormal trap. CPYSE/CPYSN, FCMOVxx,
and MF_FPCR/MT_FPCR are not arithmetic operations, and pass denormal values
without initiating arithmetic traps.

•

The 21264/EV68A implements the following disable bits in the floating-point control register (FPCR):
–

Underflow disable (UNFD)

–

Overflow disable (OVFD)

–

Inexact result disable (INED)

–

Division by zero disable (DZED)

–

Invalid operation disable (INVD)

A–14 Alpha Instruction Set

21264/EV68A Hardware Reference Manual

IEEE Floating-Point Conformance

If one of these bits is set, and an instruction with the /S qualifier set generates the
associated exception, the 21264/EV68A produces the IEEE nontrapping result and
suppresses the trap. These nontrapping responses include correctly signed
infinity, largest finite number, and Quiet NaNs as specified by the IEEE
standard.
The 21264/EV68A does not produce a denormal result for the underflow exception.
Instead, a true zero (+0) is written to the destination register. In the 21264/EV68A,
the FPCR underflow to zero (UNDZ) bit must be set if the underflow disable
(UNFD) bit is set. If desired, trapping on underflow can be enabled by the instruction and the FPCR, and software may compute the denormal value as defined in the
IEEE standard.
The 21264/EV68A records floating-point exception information in two places:

•

The FPCR status bits record the occurrence of all exceptions that are detected,
whether or not the corresponding trap is enabled. The status bits are cleared only
through an explicit clear command (MT_FPCR); hence, the exception information
they record is a summary of all exceptions that have occurred since the last time
they were cleared.

•

If an exception is detected and the corresponding trap is enabled by the instruction,
and is not disabled by the FPCR control bits, the 21264/EV68A will record the
condition in the EXC_SUM register and initiate an arithmetic trap.

The following items apply to Table A–11:

•

The 21264/EV68A traps on a denormal input operand for all arithmetic operations
unless FPCR[DNZ] = 1.

•

Input operand traps take precedence over arithmetic result traps.

•

The following abbreviations are used:
Inf: Infinity
QNaN: Quiet NaN
SNaN: Signalling NaN
CQNaN: Canonical Quiet NaN
For IEEE instructions with /S, Table A–11 lists all exceptional input and output
conditions recognized by the 21264/EV68A, along with the result and exception
generated for each condition.

Table A–11 Exceptional Input and Output Conditions
21264/EV68A Hardware
Supplied Result

Exception

Inf operand

±Inf

(none)

QNaN operand

QNaN

(none)

SNaN operand

QNaN

Invalid Op

Effective subtract of two Inf operands

CQNaN

Invalid Op

Alpha Instructions

ADDx SUBx INPUT

ADDx SUBx OUTPUT

21264/EV68A Hardware Reference Manual

Alpha Instruction Set A–15

IEEE Floating-Point Conformance

Table A–11 Exceptional Input and Output Conditions (Continued)
Alpha Instructions

21264/EV68A Hardware
Supplied Result

Exception

Exponent overflow

±Inf or ±MAX

Overflow

Exponent underflow

Underflow

Inexact result

Result

Inexact

A–16 Alpha Instruction Set

21264/EV68A Hardware Reference Manual

IEEE Floating-Point Conformance

Table A–11 Exceptional Input and Output Conditions (Continued)
21264/EV68A Hardware
Supplied Result

Exception

Inf operand

±Inf

(none)

QNaN operand

QNaN

(none)

SNaN operand

QNaN

Invalid Op

0 * Inf

CQNaN

Invalid Op

QNaN operand

QNaN

(none)

SNaN operand

QNaN

Invalid Op

0/0 or Inf/Inf

CQNaN

Invalid Op

A/0 (A not 0)

±Inf

Div Zero

A/Inf

±0

(none)

Inf/A

±Inf

(none)

+Inf operand

+Inf

(none)

QNaN operand

QNaN

(none)

SNaN operand

QNaN

Invalid Op

-A (A not 0)

CQNaN

Invalid Op

-0

(none)

root

Inexact

Inf operand

True or False

(none)

QNaN operand

False for EQ, True for UN

(none)

SNaN operand

False for EQ,True for UN

Invalid Op

Inf operand

True or False

(none)

QNaN operand

False

Invalid Op

SNaN operand

False

Invalid Op

Inf operand

Invalid Op

QNaN operand

Invalid Op

Alpha Instructions

MULx INPUT

MULx OUTPUT (same as ADDx)
DIVx INPUT

DIVx OUTPUT (same as ADDx)
SQRTx INPUT

SQRTx OUTPUT
Inexact result
CMPTEQ CMPTUN INPUT

CMPTLT CMPTLE INPUT

CVTfi INPUT

21264/EV68A Hardware Reference Manual

Alpha Instruction Set A–17

Table A–11 Exceptional Input and Output Conditions (Continued)
Alpha Instructions

21264/EV68A Hardware
Supplied Result

Exception

SNaN operand

Invalid Op

Inexact result

Result

Inexact

Integer overflow

Truncated result

Invalid Op

Result

Inexact

Inf operand

±Inf

(none)

QNaN operand

QNaN

(none)

SNaN operand

QNaN

Invalid Op

CVTfi OUTPUT

CVTif OUTPUT
Inexact result
CVTff INPUT

CVTff OUTPUT (same as ADDx)
FBEQ FBNE FBLT FBLE FBGT FBGE
LDS LDT
STS STT
CPYS CPYSN
FCMOVx

See Section 2.14 for information about the floating-point control register (FPCR).

B
21264/EV68A Boundary-Scan Register
This appendix contains the BSDL description of the 21264/EV68A boundary-scan register.

B.1 Boundary-Scan Register
The Boundary-Scan Register (BSR) on the 21264/EV68A is 367 bits long. It is
accessed by the three public (SAMPLE, EXTEST, CLAMP) instructions. The register
operation for the public instructions is compliant with the IEEE 1149.1 standard.
The boundary-scan register covers all input, output, and bidirectional pins with the
exception of the compliance enable pins and pins that are power-supply-type or analog
in nature. The BSDL for the boundary-scan register is given in Section B.1.1.

B.1.1 BSDL Description of the Alpha 21264/EV68A Boundary-Scan Register
-------------------------------------------------------------------------------- alpha21264b.bsdl
--The BSDL Description for EV6’s IEEE 1149.1 Circuits
-------------------------------------------------------------------------------Revision History
--Rev Date
Description
-- 1.0 Feb 99
First external release
------------------------------------------------------------------------------entity Alpha_21264b is-- (ref B.8)
generic (PHYSICAL_PIN_MAP :string := "PGA_EV6");-- (ref B.8.2)
port (-- (ref B.8.3)
TestStat_H
:out
bit
;
SromOE_L
:out
bit
;
SromClk_H
:out
bit
;
SromData_H
:in
bit
;
Reset_L
:in
bit
;
IRQ_H
:in
bit_vector (0 to 5) ;
DcOk_H
:linkage bit
; -- Compliance enable input
NoConnect_0
:linkage bit
; -- n/c
NoConnect_1
:linkage bit
; -- n/c
PllBypass_H
:linkage bit
;
FrameClk_H
:linkage bit
;
FrameClk_L
:linkage bit
;
ClkFwdRst_H
:in
bit
;
BcCheck_H
:inout
bit_vector (0 to 15);
BcData_H
:inout
bit_vector (0 to 127);
SysData_L
:inout
bit_vector (0 to 63) ;
SysCheck_L
:inout
bit_vector (0 to 7) ;
BcDataInClk_H
:in
bit_vector (0 to 7) ;
SysDataOutClk_L
:out
bit_vector (0 to 7) ;
Spare_7
:linkage bit_vector (0 to 7) ;

21264/EV68A Hardware Reference Manual

21264/EV68A Boundary-Scan Register

B–1

Boundary-Scan Register

SysDataInClk_H
:in
bit_vector (0 to 7) ;
BcDataOutClk_L
:out
bit_vector (0 to 3) ; -- JWB corrected
BcDataOutClk_H
:out
bit_vector (0 to 3) ; -- JWB corrected
ClkIn_H
:linkage bit
; -- Oscillator
ClkIn_L
:linkage bit
; -- Oscillator
PLL_VDD
:linkage bit
;
EV6Clk_H
:linkage bit
;
EV6Clk_L
:linkage bit
;
Spare_4
:linkage bit
;
Spare_5
:linkage bit
;
BcTag_H
:inout
bit_vector (20 to 42);
BcVref
:linkage bit
;
BcTagInClk_H
:in
bit
; -- Name in model:
BcTagClkIn_H
BcTagParity_H
:inout
bit
;
BcTagShared_H
:inout
bit
;
BcTagDirty_H
:inout
bit
;
BcTagValid_H
:inout
bit
;
BcTagOutClk_L
:out
bit
;
BcTagOutClk_H
:out
bit
;
BcTagOE_L
:out
bit
;
BcTagWr_L
:out
bit
;
BcDataWr_L
:out
bit
;
BcLoad_L
:out
bit
;
BcDataOE_L
:out
bit
;
BcAdd_H
:out
bit_vector (4 to 23) ;
SysAddOut_L
:out
bit_vector (0 to 14) ;
SysAddIn_L
:in
bit_vector (0 to 14) ;
SysAddInClk_L
:in
bit
;
SysAddOutClk_L
:out
bit
; --JWB added
SysVref
:linkage bit
; --JWB added
SysFillValid_L
:in
bit
;
SysDataInValid_L
:in
bit
;
SysDataOutValid_L
:in
bit
;
Spare_0
:linkage bit
; -- n/c
MiscVref
:linkage bit
; -Spare_2
:linkage bit
; -- n/c
Tdi_H
:in
bit
;
Tdo_H
:out
bit
;
Trst_L
:in
bit
;
Tck_H
:in
bit
;
Tms_H
:in
bit
;
VSS:linkage bit_vector (0 to 103);
VDD
:linkage bit_vector (0 to 93) );
use STD_1149_1_1994.all ;-- (ref B.8.4)
attribute COMPONENT_CONFORMANCE of Alpha_21264: entity is "STD_1149_1_1993";
attribute PIN_MAP of Alpha_21264 : entity is PHYSICAL_PIN_MAP ;
constant PGA_EV6 : PIN_MAP_STRING := "
" &
"SysAddIn_L
: (BD30, BC29, AY28, BE29, AW27, BA27, BD28, BE27, "&
"
AY26, BC25, BB24, AV24, BD24, BE23, AW23),
"&
"SysAddInClk_L
: BB26,
"&
"SysVref
: BA25,
"&
"SysFillValid_L
: BC23,
"&
"SysAddOut_L
: (AW33, BE39, BD36, BC35, BA33, AY32, BE35, AV30, "&
"
BB32, BA31, BE33, AW29, BC31, AV28, BB30),
"&
"SysAddOutClk_L
: BD34,
"&
"SysData_L
: (F14 , G13 , F12 , H12 , H10 , G7 , F6 , K8 , "&
"
M6 , N7 , P6 , T8 , V8 , V6 , W7 , Y6 , "&
"
AB8 , AC7 , AD8 , AE5 , AH6 , AH8 , AJ7 , AL5 , "&
"
AP8 , AR7 , AT8 , AV6 , AV10, AW11, AV12, AW13, "&
"
F32 , F34 , H34 , G35 , F40 , G39 , K38 , J41 , "&
"
M40 , N39 , P40 , T38 , V40 , W41 , W39 , Y40 , "&

B–2

21264/EV68A Boundary-Scan Register

21264/EV68A Hardware Reference Manual

Boundary-Scan Register

"
"
"SysCheck_L
"SysDataInClk_H

AB38, AC39, AD38, AF40, AH38, AJ39, AL41, AK38, "&
AN39, AP38, AR39, AT38, AY38, AV36, AW35, AV34),"&
: (L7 , AA5 , AK8 , BA13, L39 , AA41, AM40, AY34),"&
: (D8 , P4 , AF6 , AY6 , E37 , R43 , AG41, AV40),"&

:
"SysDataOutClk_L
"SysDataInValid_L
"SysDataOutValid_L
"BcAdd_H
"
"
"BcDataOE_L
"BcLoad_L
"BcDataWr_L
"BcData_H
"
"
"
"
"
"
"
"
"
"
"
"
"
"
"
"BcCheck_H
"
"BcDataInClk_H
"Spare_7
"BcDataOutClk_L
"BcDataOutClk_H
"BcTag_H
"
"
"BcTagValid_H
"BcTagDirty_H
"BcTagShared_H
"BcTagParity_H
"BcTagOE_L
"BcTagWr_L
"BcTagInClk_H
"BcVref
"BcTagOutClk_L
"BcTagOutClk_H
"IRQ_H
"Reset_L
"SromData_H
"SromCLK_H
"SromOE_L
"Tms_H
"Tck_H
"Trst_L
"Tdi_H
"Tdo_H
"TestStat_H
"ClkIn_H
"ClkIn_L
"FrameClk_H
"FrameClk_L
"PllBypass_H

: (G11 , U7 , AG7 , AY8 , H36 , R41 , AH40, AW39),"&
: BD22,
"&
: BB22,
"&
: (B28 , E27 , A29 , G27 , C29 , F28 , B30 , D30 ,"&
C31 , H28 , G29 , A33 , E31 , D32 , B34 , A35 ,"&
B36 , H30 , C35 , E33 ),
"&
: A27 ,
"&
: F26 ,
"&
: D26 ,
"&
: (B10 , D10 , A5 , C5 , C3 , E3 , H6 , E1
,"&
J3 , K2 , L3 , M2 , T2 , U1 , V2 , Y4
,"&
AC1 , AD2 , AE3 , AG1 , AK2 , AL3 , AR1 , AP2 ,"&
AY2 , BB2 , AW5 , BB4 , BB8 , BE5 , BB10, BE7 ,"&
G33 , C37 , B40 , C41 , C43 , E43 , G41 , F44 ,"&
K44 , N41 , M44 , P42 , U43 , V44 , Y42 , AB44 ,"&
AD42, AE43, AF42, AJ45, AK42, AN45, AP44, AN41 ,"&
AW45, AU41, AY44, BA43, BC43, BD42, BB38, BE41 ,"&
C11 , A7 , C9 , B6 , B4 , D4 , G5 , D2
,"&
H4 , G1 , N5 , L1 , N1 , U3 , W5 , W1
,"&
AB2 , AC3 , AD4 , AF4 , AJ3 , AK4 , AN1 , AM4 ,"&
AU5 , BA1 , BA3 , BC3 , BD6 , BA9 , BC9 , AY12 ,"&
A39 , D36 , A41 , B42 , D42 , D44 , H40 , H42 ,"&
G45 , L43 , L45 , N45 , T44 , U45 , W45 , AA43 ,"&
AC43, AD44, AE41, AG45, AK44, AL43, AM42, AR45 ,"&
AP40, BA45, AV42, BB44, BB42, BC41, BA37, BD40),"&
: (F2 , AB4 , AT2 , BC11, M38 , AB42, AU43, BC37 ,"&
M8 , AA3 , AW1 , BD10, E45 , AC45, AT44, BB36),"&
: (E7 , R3 , AH2 , BC5 , F38 , U39 , AH44, AY40),"&
: (F8 , T4 , AJ1 , BD4 , E39 , V38 , AJ43, BA39),"&
: (K4 , AV4 , K42 , AT42),
"&
: (J5 , AU3 , J43 , AR43),
"&
: (E13 , H16 , A11 , B12 , D14 , E15 , A13 , G17 ,"&
C15 , H18 , D16 , B16 , C17 , A17 , E19 , B18 ,"&
A19 , F20 , D20 , E21 , C21 , D22 , H22 ),
"&
: B24 ,
"&
: C23 ,
"&
: G23 ,
"&
: B22 ,
"&
: H24 ,
"&
: E25 ,
"&
: G19 ,
"&
: F18 ,
"&
: D24 ,
"&
: C25 ,
"&
: (BA15, BE13, AW17, AV18, BC15, BB16),
"&
: BD16,
"&
: BC17,
"&
: AW19,
"&
: BE17,
"&
: BD18,
"&
: BE19,
"&
: AY20,
"&
: BA21,
"&
: BB20,
"&
: BA19,
"&
: AM8 ,
"&
: AN7 ,
"&
: AV16,
"&
: AW15,
"&
: BD12,
"&

21264/EV68A Hardware Reference Manual

21264/EV68A Boundary-Scan Register

B–3

Boundary-Scan Register

"NoConnect_0
"NoConnect_1
"ClkFwdRst_H
"EV6Clk_H
"EV6Clk_L
"Spare_4
"Spare_5
"PLL_VDD
"Spare_0
"MiscVref
"Spare_2
"DCOK_H
"VSS:
"
"
"
"
"
"
"
"
"
"
"
"
“
"VDD
"
"
"
"
"
"
"
"
"
"
"

:
:
:
:
:
:
:
:
:
:
:
:

BB14,
"&
BD2 ,
"&
BE11,
"&
AM6 ,
"&
AL7 ,
"&
AT4 ,
"&
AR3 ,
"&
AV8 ,
"&
BC21,
"&
AV22,
"&
BE9 ,
"&
AY18,
"&
(C1 , W3 , AR5 , G9 , E17 , G25 , C33 , AA39, "&
BA41, R45 , J1 , AG3 , BA5 , AW9 , BA17, AW25, "&
BC33, AE39, A43 , AA45, R1 , AN3 , C7 , C19 , "&
BE25, E35 , AL39, G43 , AE45, AA1 , AW3 , J7 , "&
E11 , BC19, C27 , BA35, AU39, N43 , AL45, AE1 , "&
BE3 , R7 , BA11, A21 , BC27, A37 , BC39, W43 , "&
AU45, AL1 , E5 , AA7 , C13 , G21 , E29 , G37 , "&
E41 , AG43, BC45, AU1 , L5 , AE7 , BC13, AW21, "&
BA29, AW37, L41 , AN43, BC1 , U5 , AU7 , A15 , "&
BE21, A31 , BE37, U41 , AW43, A3 , AC5 , AW7 , "&
G15 , E23 , G31 , C39 , AC41, BE43, G3 , AJ5 , "&
BC7 , AY14, BA23, AW31, J39 , AJ41, C45 , N3 , "&
AN5 , A9 , BE15, A25 , BE31, R39 , AR41, J45 , "&
E9 , R5 , AG5 , BA7 , D38 , T42 , AG39, AW41) ,"&
: (B2 , V4 , AP6 , D12 , B20 , H26 , BD32, AM38, "&
BB40, Y44 , H2 , AH4 , AT6 , BB12, H20 , AV26, "&
D34 , AV38, F42 , AF44, P2 , AP4 , BB6 , B14 , "&
AV20, BD26, BB34, BD38, M42 , AM44, Y2 , AY4 , "&
B8 , H14 , BD20, D28 , F36 , D40 , V42 , AV44, "&
AF2 , D6 , P8 , AV14, F22 , BB28, AY36, K40 , "&
AH42, BD44, AM2 , K6 , Y8 , BD14, AY22, F30 , "&
B38 , T40 , AP42, AV2 , T6 , AF8 , F16 , A23 , "&
AY30, H38 , AB40, AY42, AB6 , BD8 , AY16, F24 , "&
B32 , P38 , AD40, B44 , F4 , AD6 , F10 , D18 , "&
AY24, H32 , Y38 , AK40, H44 , M4 , AK6 , AY10, "&
BB18, B26 , AV32, AF38, AT40, P44 )
";

constant numeric_EV6 : PIN_MAP_STRING := "
" &
"SysAddIn_L
: (559 , 536 , 468 , 580 , 445 , 490 , 558 , 579 , "&
"
467 , 534 , 511 , 421 , 556 , 577 , 443 ),
"&
"SysAddInClk_L
: 512 ,
"&
"SysVref
: 489 ,
"&
"SysFillValid_L
: 533 ,
"&
"SysAddOut_L
: (448 , 585 , 562 , 539 , 493 , 470 , 583 , 424 , "&
"
515 , 492 , 582 , 446 , 537 , 423 , 514 ),
"&
"SysAddOutClk_L
: 561 ,
"&
"SysData_L
: (118 , 140 , 117 , 161 , 160 , 137 , 114 , 189 , "&
"
204 , 213 , 220 , 237 , 253 , 252 , 261 , 268 , "&
"
285 , 293 , 301 , 308 , 332 , 333 , 341 , 356 , "&
"
381 , 389 , 397 , 412 , 414 , 437 , 415 , 438 , "&
"
127 , 128 , 172 , 151 , 131 , 153 , 190 , 183 , "&
"
207 , 214 , 223 , 238 , 255 , 263 , 262 , 271 , "&
"
286 , 294 , 302 , 319 , 334 , 342 , 359 , 350 , "&
"
374 , 382 , 390 , 398 , 473 , 427 , 449 , 426 ),"&
"SysCheck_L
: (197 , 276 , 349 , 483 , 198 , 279 , 367 , 471 ),"&
"SysDataInClk_H
: (70 , 219 , 316 , 457 , 107 , 232 , 327 , 429 ),"&
"SysDataOutClk_L
: (139 , 245 , 325 , 458 , 173 , 231 , 335 , 451 ),"&
"SysDataInValid_L
: 555 ,
"&
"SysDataOutValid_L
: 510 ,
"&
"BcAdd_H
: (35 , 102 , 14 , 147 , 58 , 125 , 36 , 81 , "&
"
59 , 169 , 148 , 16 , 104 , 82 , 38 , 17 , "&
"
39 , 170 , 61 , 105 ),
"&
"BcDataOE_L
: 13 ,
"&

B–4

21264/EV68A Boundary-Scan Register

21264/EV68A Hardware Reference Manual

Boundary-Scan Register

"BcLoad_L
"BcDataWr_L
"BcData_H
"
"
"
"
"
"
"
"
"
"
"
"
"
"
"
"BcCheck_H
"
"BcDataInClk_H
"Spare_7
"BcDataOutClk_L
"BcDataOutClk_H
"BcTag_H
"
"
"BcTagValid_H
"BcTagDirty_H
"BcTagShared_H
"BcTagParity_H
"BcTagOE_L
"BcTagWr_L
"BcTagInClk_H
"BcVref
"BcTagOutClk_L
"BcTagOutClk_H
"IRQ_H
"Reset_L
"SromData_H
"SromCLK_H
"SromOE_L
"Tms_H
"Tck_H
"Trst_L
"Tdi_H
"Tdo_H
"TestStat_H
"ClkIn_H
"ClkIn_L
"FrameClk_H
"FrameClk_L
"PllBypass_H
"NoConnect_0
"NoConnect_1
"ClkFwdRst_H
"EV6Clk_H
"EV6Clk_L
"Spare_4
"Spare_5
"PLL_VDD
"Spare_0
"MiscVref
"Spare_2
"DCOK_H

: 124 ,
"&
: 79 ,
"&
: (26 , 71 , 2
, 46 , 45 , 90 , 159 , 89 , "&
179 , 186 , 195 , 202 , 234 , 242 , 250 , 267 , "&
290 , 298 , 307 , 322 , 346 , 355 , 386 , 378 , "&
455 , 500 , 434 , 501 , 503 , 568 , 504 , 569 , "&
150 , 62 , 41 , 64 , 65 , 110 , 154 , 133 , "&
193 , 215 , 209 , 224 , 248 , 257 , 272 , 289 , "&
304 , 312 , 320 , 345 , 352 , 377 , 385 , 375 , "&
454 , 407 , 476 , 498 , 543 , 565 , 518 , 586 , "&
49 , 3
, 48 , 24 , 23 , 68 , 136 , 67 , "&
158 , 134 , 212 , 194 , 210 , 243 , 260 , 258 , "&
282 , 291 , 299 , 315 , 339 , 347 , 370 , 363 , "&
404 , 477 , 478 , 523 , 547 , 481 , 526 , 460 , "&
19 , 84 , 20 , 42 , 87 , 88 , 175 , 176 , "&
156 , 200 , 201 , 217 , 241 , 249 , 265 , 280 , "&
296 , 305 , 311 , 329 , 353 , 360 , 368 , 393 , "&
383 , 499 , 430 , 521 , 520 , 542 , 495 , 564 ),"&
: (112 , 283 , 394 , 527 , 206 , 288 , 408 , 540 , "&
205 , 275 , 432 , 549 , 111 , 297 , 401 , 517 ),"&
: (92 , 227 , 330 , 524 , 130 , 246 , 337 , 474 ),"&
: (115 , 235 , 338 , 546 , 108 , 254 , 344 , 496 ),"&
: (187 , 411 , 192 , 400 ),
"&
: (180 , 403 , 184 , 392 ),
"&
: (95 , 163 , 5
, 27 , 73 , 96 , 6
, 142 , "&
51 , 164 , 74 , 29 , 52 , 8
, 98 , 30 , "&
9
, 121 , 76 , 99 , 54 , 77 , 166 ),
"&
: 33 ,
"&
: 55 ,
"&
: 145 ,
"&
: 32 ,
"&
: 167 ,
"&
: 101 ,
"&
: 143 ,
"&
: 120 ,
"&
: 78 ,
"&
: 56 ,
"&
: (484 , 572 , 440 , 418 , 529 , 507 ),
"&
: 552 ,
"&
: 530 ,
"&
: 441 ,
"&
: 574 ,
"&
: 553 ,
"&
: 575 ,
"&
: 464 ,
"&
: 487 ,
"&
: 509 ,
"&
: 486 ,
"&
: 365 ,
"&
: 373 ,
"&
: 417 ,
"&
: 439 ,
"&
: 550 ,
"&
: 506 ,
"&
: 545 ,
"&
: 571 ,
"&
: 364 ,
"&
: 357 ,
"&
: 395 ,
"&
: 387 ,
"&
: 413 ,
"&
: 532 ,
"&
: 420 ,
"&
: 570 ,
"&
: 463 ,
"&

21264/EV68A Hardware Reference Manual

21264/EV68A Boundary-Scan Register

B–5

Boundary-Scan Register

"VSS
"
"
"
"
"
"
"
"
"
"
"
"
“
"VDD
"
"
"
"
"
"
"
"
"
"
"

: (44 , 259 , 388 , 138 , 97 , 146 , 60 , 278 , "&
497 , 233 , 178 , 323 , 479 , 436 , 485 , 444 , "&
538 , 310 , 21 , 281 , 226 , 371 , 47 , 53 , "&
578 , 106 , 358 , 155 , 313 , 274 , 433 , 181 , "&
94 , 531 , 57 , 494 , 406 , 216 , 361 , 306 , "&
567 , 229 , 482 , 10 , 535 , 18 , 541 , 264 , "&
409 , 354 , 91 , 277 , 50 , 144 , 103 , 152 , "&
109 , 328 , 544 , 402 , 196 , 309 , 528 , 442 , "&
491 , 450 , 199 , 376 , 522 , 244 , 405 , 7
, "&
576 , 15 , 584 , 247 , 453 , 1
, 292 , 435 , "&
141 , 100 , 149 , 63 , 295 , 587 , 135 , 340 , "&
525 , 461 , 488 , 447 , 182 , 343 , 66 , 211 , "&
372 , 4
, 573 , 12 , 581 , 230 , 391 , 185 ,"&
93 , 228 , 324 , 480 , 85 , 240 , 326 , 452 ),"&
: (22 , 251 , 380 , 72 , 31 , 168 , 560 , 366 , "&
519 , 273 , 157 , 331 , 396 , 505 , 165 , 422 , "&
83 , 428 , 132 , 321 , 218 , 379 , 502 , 28 , "&
419 , 557 , 516 , 563 , 208 , 369 , 266 , 456 , "&
25 , 162 , 554 , 80 , 129 , 86 , 256 , 431 , "&
314 , 69 , 221 , 416 , 122 , 513 , 472 , 191 , "&
336 , 566 , 362 , 188 , 269 , 551 , 465 , 126 , "&
40 , 239 , 384 , 410 , 236 , 317 , 119 , 11 , "&
469 , 174 , 287 , 475 , 284 , 548 , 462 , 123 , "&
37 , 222 , 303 , 43 , 113 , 300 , 116 , 75 , "&
466 , 171 , 270 , 351 , 177 , 203 , 348 , 459 , "&
508 , 34 , 425 , 318 , 399 , 225 )
";

attribute PORT_GROUPING of Alpha_21264b : entity is-- (Ref B.8.8. See Note 4.
"Differential_Voltage ( (CLKIN_H), (CLKIN_L) )" ;
attribute TAP_SCAN_CLOCK of Tck_H : signal is (5.0e6, LOW);
attribute TAP_SCAN_IN
of Tdi_H : signal is TRUE;
attribute TAP_SCAN_OUT
of Tdo_H : signal is TRUE;
attribute TAP_SCAN_MODE of Tms_H : signal is TRUE;
attribute TAP_SCAN_RESET of Trst_L : signal is TRUE;
attribute COMPLIANCE_PATTERNS of Alpha_21264b : entity is
"(DcOk_H), (1)" ;

-- (Ref B.8.10). See Note 4.

attribute INSTRUCTION_LENGTH of Alpha_21264b : entity is 5 ;
attribute INSTRUCTION_OPCODE of Alpha_21264b : entity is
"EXTEST
(11011),"&-- No longer mandated to be (00000)!
"SAMPLE
(11000),"&-- JWB changed "PRELOAD" to "SAMPLE"
"CLAMP
(11010),"&
"HIGHZ
(11001),"&
"DIE_ID
(11110),"&
"BYPASS
(11111)";
attribute INSTRUCTION_CAPTURE of Alpha_21264b : entity is "00001" ;
attribute INSTRUCTION_PRIVATE of Alpha_21264b : entity is "Private"; -- See Note 4.
attribute REGISTER_ACCESS of Alpha_21264b : entity is-- (ref B.8.13)
"BOUNDARY (EXTEST, SAMPLE)," &-- Redundant. Added for completeness
"BYPASS
(BYPASS, HIGHZ, CLAMP)," &-- ditto
"DIE_ID[32] (DIE_ID)";
attribute

BOUNDARY_LENGTH

of Alpha_21264b : entity is 367 ;

attribute BOUNDARY_REGISTER of Alpha_21264b : entity is
------------------------------------------------------------------------------ scan cell
safe
cntrl disable disable
-- cell type port
function | cell value state
----|-----|-----|-------------------|--------|---|----|------|-------------" 366 ( BC_2, TestStat_H,
OUTPUT2, x
), "& -" 365 ( BC_2, SromOE_L,
OUTPUT2, X
), "& -" 364 ( BC_2, SromClk_H,
OUTPUT2, x
), "& --

B–6

21264/EV68A Boundary-Scan Register

21264/EV68A Hardware Reference Manual

Boundary-Scan Register

" 363 ( BC_2, SromData_H,
INPUT,
" 362 ( BC_3, reset_L,
INPUT,
" 361 ( BC_3, IRQ_H(5),
INPUT,
" 360 ( BC_3, IRQ_H(4),
INPUT,
" 359 ( BC_3, IRQ_H(3),
INPUT,
" 358 ( BC_3, IRQ_H(2),
INPUT,
" 357 ( BC_3, IRQ_H(1),
INPUT,
" 356 ( BC_3, IRQ_H(0),
INPUT,
" 355 ( BC_3, ClkFwdRst_H,
INPUT,
" 354 ( BC_2, BcCheck_H(3),
BIDIR,
" 353 ( BC_2, BcCheck_H(11),
BIDIR,
" 352 ( BC_2, SysCheck_L(3),
BIDIR,
" 351 ( BC_2, BcData_H(31),
BIDIR,
" 350 ( BC_2, BcData_H(95),
BIDIR,
" 349 ( BC_2, SysData_L(31),
BIDIR,
" 348 ( BC_2, BcData_H(30),
BIDIR,
" 347 ( BC_2, BcData_H(94),
BIDIR,
" 346 ( BC_2, SysData_L(30),
BIDIR,
" 345 ( BC_2, BcData_H(29),
BIDIR,
" 344 ( BC_2, BcData_H(93),
BIDIR,
" 343 ( BC_2, SysData_L(29),
BIDIR,
" 342 ( BC_2, BcData_H(28),
BIDIR,
" 341 ( BC_2, BcData_H(92),
BIDIR,
" 340 ( BC_2, SysData_L(28),
BIDIR,
" 339 ( BC_3, *,
CONTROL,
" 338 ( BC_3, BcDataInClk_H(3),
INPUT,
" 337 ( BC_2, SysDataOutClk_L(3), OUTPUT2,
" 336 ( BC_3, *,
CONTROL,
" 335 ( BC_3, SysDataInClk_H(3), INPUT,
" 334 ( BC_2, BcData_H(27),
BIDIR,
" 333 ( BC_2, BcData_H(91),
BIDIR,
" 332 ( BC_2, SysData_L(27),
BIDIR,
" 331 ( BC_2, BcData_H(26),
BIDIR,
" 330 ( BC_2, BcData_H(90),
BIDIR,
" 329 ( BC_2, SysData_L(26),
BIDIR,
" 328 ( BC_2, BcData_H(25),
BIDIR,
" 327 ( BC_2, BcData_H(89),
BIDIR,
" 326 ( BC_2, SysData_L(25),
BIDIR,
" 325 ( BC_2, BcData_H(24),
BIDIR,
" 324 ( BC_2, BcData_H(88),
BIDIR,
" 323 ( BC_2, SysData_L(24),
BIDIR,
" 322 ( BC_2, BcDataOutClk_L(1), OUTPUT2,
" 321 ( BC_2, BcDataOutClk_H(1), OUTPUT2,
" 320 ( BC_2, BcCheck_H(2),
BIDIR,
" 319 ( BC_2, BcCheck_H(10),
BIDIR,
" 318 ( BC_2, SysCheck_L(2),
BIDIR,
" 317 ( BC_2, BcData_H(23),
BIDIR,
" 316 ( BC_2, BcData_H(87),
BIDIR,
" 315 ( BC_2, SysData_L(23),
BIDIR,
" 314 ( BC_2, BcData_H(22),
BIDIR,
" 313 ( BC_2, BcData_H(86),
BIDIR,
" 312 ( BC_2, SysData_L(22),
BIDIR,
" 311 ( BC_2, BcData_H(21),
BIDIR,
" 310 ( BC_2, BcData_H(85),
BIDIR,
" 309 ( BC_2, SysData_L(21),
BIDIR,
" 308 ( BC_2, BcData_H(20),
BIDIR,
" 307 ( BC_2, BcData_H(84),
BIDIR,
" 306 ( BC_2, SysData_L(20),
BIDIR,
" 305 ( BC_3, *,
CONTROL,
" 304 ( BC_3, BcDataInClk_H(2),
INPUT,
" 303 ( BC_2, SysDataOutClk_L(2), OUTPUT2,
" 302 ( BC_3, *,
CONTROL,
" 301 ( BC_3, SysDataInClk_H(2), INPUT,
" 300 ( BC_2, BcData_H(19),
BIDIR,
" 299 ( BC_2, BcData_H(83),
BIDIR,

21264/EV68A Hardware Reference Manual

x
x
x
x
x
x
x
x
x
x, 339,
x, 339,
x, 336,
x, 339,
x, 339,
x, 336,
x, 339,
x, 339,
x, 336,
x, 339,
x, 339,
x, 336,
x, 339,
x, 339,
x, 336,
0
x
x
0
x
x, 339,
x, 339,
x, 336,
x, 339,
x, 339,
x, 336,
x, 339,
x, 339,
x, 336,
x, 339,
x, 339,
x, 336,
x
x
x, 305,
x, 305,
x, 302,
x, 305,
x, 305,
x, 302,
x, 305,
x, 305,
x, 302,
x, 305,
x, 305,
x, 302,
x, 305,
x, 305,
x, 302,
0
x
x
0
x
x, 305,
x, 305,

0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,

0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,

0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,

0,
0,

), "& -), "& -), "& -), "& -), "& -), "& -), "& -), "& -), "& -Z
), "& -Z
), "& -WEAK1 ), "& -Z
), "& -Z
), "& -WEAK1 ), "& -Z
), "& -Z
), "& -WEAK1 ), "& -Z
), "& -Z
), "& -WEAK1 ), "& -Z
), "& -Z
), "& -WEAK1 ), "& -), "& -- bccell0
), "& -), "& -), "& -- sccell0
), "& -Z
), "& -Z
), "& -WEAK1 ), "& -Z
), "& -Z
), "& -WEAK1 ), "& -Z
), "& -Z
), "& -WEAK1 ), "& -Z
), "& -Z
), "& -WEAK1 ), "& -), "& -), "& -Z
), "& -Z
), "& -WEAK1 ), "& -Z
), "& -Z
), "& -WEAK1 ), "& -Z
), "& -Z
), "& -WEAK1 ), "& -Z
), "& -Z
), "& -WEAK1 ), "& -Z
), "& -Z
), "& -WEAK1 ), "& -), "& -- bccell1
), "& -), "& -), "& -- sccell1
), "& -Z
), "& -Z
), "& --

21264/EV68A Boundary-Scan Register

B–7

Boundary-Scan Register

" 298 ( BC_2, SysData_L(19),
BIDIR,
" 297 ( BC_2, BcData_H(18),
BIDIR,
" 296 ( BC_2, BcData_H(82),
BIDIR,
" 295 ( BC_2, SysData_L(18),
BIDIR,
" 294 ( BC_2, BcData_H(17),
BIDIR,
" 293 ( BC_2, BcData_H(81),
BIDIR,
" 292 ( BC_2, SysData_L(17),
BIDIR,
" 291 ( BC_2, BcData_H(16),
BIDIR,
" 290 ( BC_2, BcData_H(80),
BIDIR,
" 289 ( BC_2, SysData_L(16),
BIDIR,
" 288 ( BC_2, BcCheck_H(1),
BIDIR,
" 287 ( BC_2, BcCheck_H(9),
BIDIR,
" 286 ( BC_2, SysCheck_L(1),
BIDIR,
" 285 ( BC_2, BcData_H(15),
BIDIR,
" 284 ( BC_2, BcData_H(79),
BIDIR,
" 283 ( BC_2, SysData_L(15),
BIDIR,
" 282 ( BC_2, BcData_H(14),
BIDIR,
" 281 ( BC_2, BcData_H(78),
BIDIR,
" 280 ( BC_2, SysData_L(14),
BIDIR,
" 279 ( BC_2, BcData_H(13),
BIDIR,
" 278 ( BC_2, BcData_H(77),
BIDIR,
" 277 ( BC_2, SysData_L(13),
BIDIR,
" 276 ( BC_2, BcData_H(12),
BIDIR,
" 275 ( BC_2, BcData_H(76),
BIDIR,
" 274 ( BC_2, SysData_L(12),
BIDIR,
" 273 ( BC_3, *,
CONTROL,
" 272 ( BC_3, BcDataInClk_H(1),
INPUT,
" 271 ( BC_2, SysDataOutClk_L(1), OUTPUT2,
" 270 ( BC_3, *,
CONTROL,
" 269 ( BC_3, SysDataInClk_H(1), INPUT,
" 268 ( BC_2, BcData_H(11),
BIDIR,
" 267 ( BC_2, BcData_H(75),
BIDIR,
" 266 ( BC_2, SysData_L(11),
BIDIR,
" 265 ( BC_2, BcData_H(10),
BIDIR,
" 264 ( BC_2, BcData_H(74),
BIDIR,
" 263 ( BC_2, SysData_L(10),
BIDIR,
" 262 ( BC_2, BcData_H(9) ,
BIDIR,
" 261 ( BC_2, BcData_H(73),
BIDIR,
" 260 ( BC_2, SysData_L(9),
BIDIR,
" 259 ( BC_2, BcData_H(8) ,
BIDIR,
" 258 ( BC_2, BcData_H(72),
BIDIR,
" 257 ( BC_2, SysData_L(8),
BIDIR,
" 256 ( BC_2, BcDataOutClk_L(0), OUTPUT2,
" 255 ( BC_2, BcDataOutClk_H(0), OUTPUT2,
" 254 ( BC_2, BcCheck_H(0),
BIDIR,
" 253 ( BC_2, BcCheck_H(8),
BIDIR,
" 252 ( BC_2, SysCheck_L(0),
BIDIR,
" 251 ( BC_2, BcData_H(7) ,
BIDIR,
" 250 ( BC_2, BcData_H(71),
BIDIR,
" 249 ( BC_2, SysData_L(7),
BIDIR,
" 248 ( BC_2, BcData_H(6) ,
BIDIR,
" 247 ( BC_2, BcData_H(70),
BIDIR,
" 246 ( BC_2, SysData_L(6),
BIDIR,
" 245 ( BC_2, BcData_H(5) ,
BIDIR,
" 244 ( BC_2, BcData_H(69),
BIDIR,
" 243 ( BC_2, SysData_L(5),
BIDIR,
" 242 ( BC_2, BcData_H(4) ,
BIDIR,
" 241 ( BC_2, BcData_H(68),
BIDIR,
" 240 ( BC_2, SysData_L(4),
BIDIR,
" 239 ( BC_3, *,
CONTROL,
" 238 ( BC_3, BcDataInClk_H(0),
INPUT,
" 237 ( BC_2, SysDataOutClk_L(0), OUTPUT2,
" 236 ( BC_3, *,
CONTROL,
" 235 ( BC_3, SysDataInClk_H(0), INPUT,
" 234 ( BC_2, BcData_H(3) ,
BIDIR,

B–8

21264/EV68A Boundary-Scan Register

x, 302,
x, 305,
x, 305,
x, 302,
x, 305,
x, 305,
x, 302,
x, 305,
x, 305,
x, 302,
x, 273,
x, 273,
x, 270,
x, 273,
x, 273,
x, 270,
x, 273,
x, 273,
x, 270,
x, 273,
x, 273,
x, 270,
x, 273,
x, 273,
x, 270,
0
x
x
0
x
x, 273,
x, 273,
x, 270,
x, 273,
x, 273,
x, 270,
x, 273,
x, 273,
x, 270,
x, 273,
x, 273,
x, 270,
x
x
x, 239,
x, 239,
x, 236,
x, 239,
x, 239,
x, 236,
x, 239,
x, 239,
x, 236,
x, 239,
x, 239,
x, 236,
x, 239,
x, 239,
x, 236,
0
x
x
0
x
x, 239,

0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,

0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,

0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,

WEAK1 ), "& -Z
), "& -Z
), "& -WEAK1 ), "& -Z
), "& -Z
), "& -WEAK1 ), "& -Z
), "& -Z
), "& -WEAK1 ), "& -Z
), "& -Z
), "& -WEAK1 ), "& -Z
), "& -Z
), "& -WEAK1 ), "& -Z
), "& -Z
), "& -WEAK1 ), "& -Z
), "& -Z
), "& -WEAK1 ), "& -Z
), "& -Z
), "& -WEAK1 ), "& -), "& -- bccell2
), "& -), "& -), "& -- sccell2
), "& -Z
), "& -Z
), "& -WEAK1 ), "& -Z
), "& -Z
), "& -WEAK1 ), "& -Z
), "& -Z
), "& -WEAK1 ), "& -Z
), "& -Z
), "& -WEAK1 ), "& -), "& -), "& -Z
), "& -Z
), "& -WEAK1 ), "& -Z
), "& -Z
), "& -WEAK1 ), "& -Z
), "& -Z
), "& -WEAK1 ), "& -Z
), "& -Z
), "& -WEAK1 ), "& -Z
), "& -Z
), "& -WEAK1 ), "& -), "& -- bccell3
), "& -), "& -), "& -- sccell3
), "& -Z
), "& --

21264/EV68A Hardware Reference Manual

Boundary-Scan Register

" 233 ( BC_2, BcData_H(67),
" 232 ( BC_2, SysData_L(3),
" 231 ( BC_2, BcData_H(2) ,
" 230 ( BC_2, BcData_H(66),
" 229 ( BC_2, SysData_L(2),
" 228 ( BC_2, BcData_H(1) ,
" 227 ( BC_2, BcData_H(65),
" 226 ( BC_2, SysData_L(1),
" 225 ( BC_2, BcData_H(0) ,
" 224 ( BC_2, BcData_H(64),
" 223 ( BC_2, SysData_L(0),
" 222 ( BC_2, BcTag_H(20),
" 221 ( BC_2, BcTag_H(21),
" 220 ( BC_2, BcTag_H(22),
" 219 ( BC_2, BcTag_H(23),
" 218 ( BC_2, BcTag_H(24),
" 217 ( BC_2, BcTag_H(25),
" 216 ( BC_2, BcTag_H(26),
" 215 ( BC_2, BcTag_H(27),
" 214 ( BC_2, BcTag_H(28),
" 213 ( BC_2, BcTag_H(29),
" 212 ( BC_2, BcTag_H(30),
" 211 ( BC_2, BcTag_H(31),
" 210 ( BC_2, BcTag_H(32),
" 209 ( BC_2, BcTag_H(33),
" 208 ( BC_3, *,
" 207 ( BC_3, BcTagInClk_H,
" 206 ( BC_2, BcTag_H(34),
" 205 ( BC_2, BcTag_H(35),
" 204 ( BC_2, BcTag_H(36),
" 203 ( BC_2, BcTag_H(37),
" 202 ( BC_2, BcTag_H(38),
" 201 ( BC_2, BcTag_H(39),
" 200 ( BC_2, BcTag_H(40),
" 199 ( BC_2, BcTag_H(41),
" 198 ( BC_2, BcTag_H(42),
" 197 ( BC_2, BcTagParity_H,
" 196 ( BC_2, BcTagShared_H,
" 195 ( BC_2, BcTagDirty_H,
" 194 ( BC_2, BcTagValid_H,
" 193 ( BC_2, BcTagOutClk_L,
" 192 ( BC_2, BcTagOutClk_H,
" 191 ( BC_2, BcTagOE_L,
" 190 ( BC_2, BcTagWr_L,
" 189 ( BC_2, BcDataWr_L,
" 188 ( BC_2, BcLoad_L,
" 187 ( BC_2, BcDataOE_L,
" 186 ( BC_2, BcAdd_H(4),
" 185 ( BC_2, BcAdd_H(5),
" 184 ( BC_2, BcAdd_H(6),
" 183 ( BC_2, BcAdd_H(7),
" 182 ( BC_2, BcAdd_H(8),
" 181 ( BC_2, BcAdd_H(9),
" 180 ( BC_2, BcAdd_H(10),
" 179 ( BC_2, BcAdd_H(11),
" 178 ( BC_2, BcAdd_H(12),
" 177 ( BC_2, BcAdd_H(13),
" 176 ( BC_2, BcAdd_H(14),
" 175 ( BC_2, BcAdd_H(15),
" 174 ( BC_2, BcAdd_H(16),
" 173 ( BC_2, BcAdd_H(17),
" 172 ( BC_2, BcAdd_H(18),
" 171 ( BC_2, BcAdd_H(19),
" 170 ( BC_2, BcAdd_H(20),
" 169 ( BC_2, BcAdd_H(21),

21264/EV68A Hardware Reference Manual

BIDIR,
BIDIR,
BIDIR,
BIDIR,
BIDIR,
BIDIR,
BIDIR,
BIDIR,
BIDIR,
BIDIR,
BIDIR,
BIDIR,
BIDIR,
BIDIR,
BIDIR,
BIDIR,
BIDIR,
BIDIR,
BIDIR,
BIDIR,
BIDIR,
BIDIR,
BIDIR,
BIDIR,
BIDIR,
CONTROL,
INPUT,
BIDIR,
BIDIR,
BIDIR,
BIDIR,
BIDIR,
BIDIR,
BIDIR,
BIDIR,
BIDIR,
BIDIR,
BIDIR,
BIDIR,
BIDIR,
OUTPUT2,
OUTPUT2,
OUTPUT2,
OUTPUT2,
OUTPUT2,
OUTPUT2,
OUTPUT2,
OUTPUT2,
OUTPUT2,
OUTPUT2,
OUTPUT2,
OUTPUT2,
OUTPUT2,
OUTPUT2,
OUTPUT2,
OUTPUT2,
OUTPUT2,
OUTPUT2,
OUTPUT2,
OUTPUT2,
OUTPUT2,
OUTPUT2,
OUTPUT2,
OUTPUT2,
OUTPUT2,

x, 239,
x, 236,
x, 239,
x, 239,
x, 236,
x, 239,
x, 239,
x, 236,
x, 239,
x, 239,
x, 236,
x, 208,
x, 208,
x, 208,
x, 208,
x, 208,
x, 208,
x, 208,
x, 208,
x, 208,
x, 208,
x, 208,
x, 208,
x, 208,
x, 208,
0
x
x, 208,
x, 208,
x, 208,
x, 208,
x, 208,
x, 208,
x, 208,
x, 208,
x, 208,
x, 208,
x, 208,
x, 208,
x, 208,
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x

0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,

0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,

Z
), "& -WEAK1 ), "& -Z
), "& -Z
), "& -WEAK1 ), "& -Z
), "& -Z
), "& -WEAK1 ), "& -Z
), "& -Z
), "& -WEAK1 ), "& -Z
), "& -Z
), "& -Z
), "& -Z
), "& -Z
), "& -Z
), "& -Z
), "& -Z
), "& -Z
), "& -Z
), "& -Z
), "& -Z
), "& -Z
), "& -Z
), "& -), "& -- tccell0
), "& -Z
), "& -Z
), "& -Z
), "& -Z
), "& -Z
), "& -Z
), "& -Z
), "& -Z
), "& -Z
), "& -Z
), "& -Z
), "& -Z
), "& -Z
), "& -), "& -), "& -), "& -), "& -), "& -), "& -), "& -), "& -), "& -), "& -), "& -), "& -), "& -), "& -), "& -), "& -), "& -), "& -), "& -), "& -), "& -), "& -), "& -), "& -), "& --

21264/EV68A Boundary-Scan Register

B–9

Boundary-Scan Register

" 168 ( BC_2, BcAdd_H(22),
OUTPUT2,
" 167 ( BC_2, BcAdd_H(23),
OUTPUT2,
" 166 ( BC_2, SysData_L(32),
BIDIR,
" 165 ( BC_2, BcData_H(96),
BIDIR,
" 164 ( BC_2, BcData_H(32),
BIDIR,
" 163 ( BC_2, SysData_L(33),
BIDIR,
" 162 ( BC_2, BcData_H(97),
BIDIR,
" 161 ( BC_2, BcData_H(33),
BIDIR,
" 160 ( BC_2, SysData_L(34),
BIDIR,
" 159 ( BC_2, BcData_H(98),
BIDIR,
" 158 ( BC_2, BcData_H(34),
BIDIR,
" 157 ( BC_2, SysData_L(35),
BIDIR,
" 156 ( BC_2, BcData_H(99),
BIDIR,
" 155 ( BC_2, BcData_H(35),
BIDIR,
" 154 ( BC_3, SysDataInClk_H(4), INPUT,
" 153 ( BC_3, *,
CONTROL,
" 152 ( BC_2, SysDataOutClk_L(4), OUTPUT2,
" 151 ( BC_3, BcDataInClk_H(4),
INPUT,
" 150 ( BC_3, *,
CONTROL,
" 149 ( BC_2, SysData_L(36),
BIDIR,
" 148 ( BC_2, BcData_H(100),
BIDIR,
" 147 ( BC_2, BcData_H(36),
BIDIR,
" 146 ( BC_2, SysData_L(37),
BIDIR,
" 145 ( BC_2, BcData_H(101),
BIDIR,
" 144 ( BC_2, BcData_H(37),
BIDIR,
" 143 ( BC_2, SysData_L(38),
BIDIR,
" 142 ( BC_2, BcData_H(102),
BIDIR,
" 141 ( BC_2, BcData_H(38),
BIDIR,
" 140 ( BC_2, SysData_L(39),
BIDIR,
" 139 ( BC_2, BcData_H(103),
BIDIR,
" 138 ( BC_2, BcData_H(39),
BIDIR,
" 137 ( BC_2, SysCheck_L(4),
BIDIR,
" 136 ( BC_2, BcCheck_H(12),
BIDIR,
" 135 ( BC_2, BcCheck_H(4),
BIDIR,
" 134 ( BC_2, BcDataOutClk_H(2), OUTPUT2,
" 133 ( BC_2, BcDataOutClk_L(2), OUTPUT2,
" 132 ( BC_2, SysData_L(40),
BIDIR,
" 131 ( BC_2, BcData_H(104),
BIDIR,
" 130 ( BC_2, BcData_H(40),
BIDIR,
" 129 ( BC_2, SysData_L(41),
BIDIR,
" 128 ( BC_2, BcData_H(105),
BIDIR,
" 127 ( BC_2, BcData_H(41),
BIDIR,
" 126 ( BC_2, SysData_L(42),
BIDIR,
" 125 ( BC_2, BcData_H(106),
BIDIR,
" 124 ( BC_2, BcData_H(42),
BIDIR,
" 123 ( BC_2, SysData_L(43),
BIDIR,
" 122 ( BC_2, BcData_H(107),
BIDIR,
" 121 ( BC_2, BcData_H(43),
BIDIR,
" 120 ( BC_3, SysDataInClk_H(5), INPUT,
" 119 ( BC_3, *,
CONTROL,
" 118 ( BC_2, SysDataOutClk_L(5), OUTPUT2,
" 117 ( BC_3, BcDataInClk_H(5),
INPUT,
" 116 ( BC_3, *,
CONTROL,
" 115 ( BC_2, SysData_L(44),
BIDIR,
" 114 ( BC_2, BcData_H(108),
BIDIR,
" 113 ( BC_2, BcData_H(44),
BIDIR,
" 112 ( BC_2, SysData_L(45),
BIDIR,
" 111 ( BC_2, BcData_H(109),
BIDIR,
" 110 ( BC_2, BcData_H(45),
BIDIR,
" 109 ( BC_2, SysData_L(46),
BIDIR,
" 108 ( BC_2, BcData_H(110),
BIDIR,
" 107 ( BC_2, BcData_H(46),
BIDIR,
" 106 ( BC_2, SysData_L(47),
BIDIR,
" 105 ( BC_2, BcData_H(111),
BIDIR,
" 104 ( BC_2, BcData_H(47),
BIDIR,

B–10 21264/EV68A Boundary-Scan Register

x
x
x, 150,
x, 153,
x, 153,
x, 150,
x, 153,
x, 153,
x, 150,
x, 153,
x, 153,
x, 150,
x, 153,
x, 153,
x
0
x
x
0
x, 150,
x, 153,
x, 153,
x, 150,
x, 153,
x, 153,
x, 150,
x, 153,
x, 153,
x, 150,
x, 153,
x, 153,
x, 150,
x, 153,
x, 153,
x
x
x, 119,
x, 116,
x, 116,
x, 119,
x, 116,
x, 116,
x, 119,
x, 116,
x, 116,
x, 119,
x, 116,
x, 116,
x
0
x
x
0
x, 119,
x, 116,
x, 116,
x, 119,
x, 116,
x, 116,
x, 119,
x, 116,
x, 116,
x, 119,
x, 116,
x, 116,

0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,

0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,

0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,

), "& -), "& -WEAK1 ), "& -Z
), "& -Z
), "& -WEAK1 ), "& -Z
), "& -Z
), "& -WEAK1 ), "& -Z
), "& -Z
), "& -WEAK1 ), "& -Z
), "& -Z
), "& -), "& -), "& -- sccell4
), "& -), "& -), "& -- bccell4
WEAK1 ), "& -Z
), "& -Z
), "& -WEAK1 ), "& -Z
), "& -Z
), "& -WEAK1 ), "& -Z
), "& -Z
), "& -WEAK1 ), "& -Z
), "& -Z
), "& -WEAK1 ), "& -Z
), "& -Z
), "& -), "& -), "& -WEAK1 ), "& -Z
), "& -Z
), "& -WEAK1 ), "& -Z
), "& -Z
), "& -WEAK1 ), "& -Z
), "& -Z
), "& -WEAK1 ), "& -Z
), "& -Z
), "& -), "& -), "& -- sccell5
), "& -), "& -), "& -- bccell5
WEAK1 ), "& -Z
), "& -Z
), "& -WEAK1 ), "& -Z
), "& -Z
), "& -WEAK1 ), "& -Z
), "& -Z
), "& -WEAK1 ), "& -Z
), "& -Z
), "& --

21264/EV68A Hardware Reference Manual

Boundary-Scan Register

" 103 ( BC_2, SysCheck_L(5),
BIDIR,
" 102 ( BC_2, BcCheck_H(13),
BIDIR,
" 101 ( BC_2, BcCheck_H(5),
BIDIR,
" 100 ( BC_2, SysData_L(48),
BIDIR,
" 99 ( BC_2, BcData_H(112),
BIDIR,
" 98 ( BC_2, BcData_H(48),
BIDIR,
" 97 ( BC_2, SysData_L(49),
BIDIR,
" 96 ( BC_2, BcData_H(113),
BIDIR,
" 95 ( BC_2, BcData_H(49),
BIDIR,
" 94 ( BC_2, SysData_L(50),
BIDIR,
" 93 ( BC_2, BcData_H(114),
BIDIR,
" 92 ( BC_2, BcData_H(50),
BIDIR,
" 91 ( BC_2, SysData_L(51),
BIDIR,
" 90 ( BC_2, BcData_H(115),
BIDIR,
" 89 ( BC_2, BcData_H(51),
BIDIR,
" 88 ( BC_3, SysDataInClk_H(6), INPUT,
" 87 ( BC_3, *,
CONTROL,
" 86 ( BC_2, SysDataOutClk_L(6), OUTPUT2,
" 85 ( BC_3, BcDataInClk_H(6),
INPUT,
" 84 ( BC_3, *,
CONTROL,
" 83 ( BC_2, SysData_L(52),
BIDIR,
" 82 ( BC_2, BcData_H(116),
BIDIR,
" 81 ( BC_2, BcData_H(52),
BIDIR,
" 80 ( BC_2, SysData_L(53),
BIDIR,
" 79 ( BC_2, BcData_H(117),
BIDIR,
" 78 ( BC_2, BcData_H(53),
BIDIR,
" 77 ( BC_2, SysData_L(54),
BIDIR,
" 76 ( BC_2, BcData_H(118),
BIDIR,
" 75 ( BC_2, BcData_H(54),
BIDIR,
" 74 ( BC_2, SysData_L(55),
BIDIR,
" 73 ( BC_2, BcData_H(119),
BIDIR,
" 72 ( BC_2, BcData_H(55),
BIDIR,
" 71 ( BC_2, SysCheck_L(6),
BIDIR,
" 70 ( BC_2, BcCheck_H(14),
BIDIR,
" 69 ( BC_2, BcCheck_H(6),
BIDIR,
" 68 ( BC_2, BcDataOutClk_H(3), OUTPUT2,
" 67 ( BC_2, BcDataOutClk_L(3), OUTPUT2,
" 66 ( BC_2, SysData_L(56),
BIDIR,
" 65 ( BC_2, BcData_H(120),
BIDIR,
" 64 ( BC_2, BcData_H(56),
BIDIR,
" 63 ( BC_2, SysData_L(57),
BIDIR,
" 62 ( BC_2, BcData_H(121),
BIDIR,
" 61 ( BC_2, BcData_H(57),
BIDIR,
" 60 ( BC_2, SysData_L(58),
BIDIR,
" 59 ( BC_2, BcData_H(122),
BIDIR,
" 58 ( BC_2, BcData_H(58),
BIDIR,
" 57 ( BC_2, SysData_L(59),
BIDIR,
" 56 ( BC_2, BcData_H(123),
BIDIR,
" 55 ( BC_2, BcData_H(59),
BIDIR,
" 54 ( BC_3, SysDataInClk_H(7), INPUT,
" 53 ( BC_3, *,
CONTROL,
" 52 ( BC_2, SysDataOutClk_L(7), OUTPUT2,
" 51 ( BC_3, BcDataInClk_H(7),
INPUT,
" 50 ( BC_3, *,
CONTROL,
" 49 ( BC_2, SysData_L(60),
BIDIR,
" 48 ( BC_2, BcData_H(124),
BIDIR,
" 47 ( BC_2, BcData_H(60),
BIDIR,
" 46 ( BC_2, SysData_L(61),
BIDIR,
" 45 ( BC_2, BcData_H(125),
BIDIR,
" 44 ( BC_2, BcData_H(61),
BIDIR,
" 43 ( BC_2, SysData_L(62),
BIDIR,
" 42 ( BC_2, BcData_H(126),
BIDIR,
" 41 ( BC_2, BcData_H(62),
BIDIR,
" 40 ( BC_2, SysData_L(63),
BIDIR,
" 39 ( BC_2, BcData_H(127),
BIDIR,

21264/EV68A Hardware Reference Manual

x, 119,
x, 116,
x, 116,
x, 87,
x, 84,
x, 84,
x, 87,
x, 84,
x, 84,
x, 87,
x, 84,
x, 84,
x, 87,
x, 84,
x, 84,
x
0
x
x
0
x, 87,
x, 84,
x, 84,
x, 87,
x, 84,
x, 84,
x, 87,
x, 84,
x, 84,
x, 87,
x, 84,
x, 84,
x, 87,
x, 84,
x, 84,
x
x
x, 53,
x, 50,
x, 50,
x, 53,
x, 50,
x, 50,
x, 53,
x, 50,
x, 50,
x, 53,
x, 50,
x, 50,
x
0
x
x
0
x, 53,
x, 50,
x, 50,
x, 53,
x, 50,
x, 50,
x, 53,
x, 50,
x, 50,
x, 53,
x, 50,

0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,

0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,

0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,

WEAK1 ), "& -Z
), "& -Z
), "& -WEAK1 ), "& -Z
), "& -Z
), "& -WEAK1 ), "& -Z
), "& -Z
), "& -WEAK1 ), "& -Z
), "& -Z
), "& -WEAK1 ), "& -Z
), "& -Z
), "& -), "& -), "& -- sccell6
), "& -), "& -), "& -- bccell6
WEAK1 ), "& -Z
), "& -Z
), "& -WEAK1 ), "& -Z
), "& -Z
), "& -WEAK1 ), "& -Z
), "& -Z
), "& -WEAK1 ), "& -Z
), "& -Z
), "& -WEAK1 ), "& -Z
), "& -Z
), "& -), "& -), "& -WEAK1 ), "& -Z
), "& -Z
), "& -WEAK1 ), "& -Z
), "& -Z
), "& -WEAK1 ), "& -Z
), "& -Z
), "& -WEAK1 ), "& -Z
), "& -Z
), "& -), "& -), "& -- sccell7
), "& -), "& -), "& -- bccell7
WEAK1 ), "& -Z
), "& -Z
), "& -WEAK1 ), "& -Z
), "& -Z
), "& -WEAK1 ), "& -Z
), "& -Z
), "& -WEAK1 ), "& -Z
), "& --

21264/EV68A Boundary-Scan Register B–11

Boundary-Scan Register

" 38
" 37
" 36
" 35
" 34
" 33
" 32
" 31
" 30
" 29
" 28
" 27
" 26
" 25
" 24
" 23
" 22
" 21
" 20
" 19
" 18
" 17
" 16
" 15
" 14
" 13
" 12
" 11
" 10
" 9
" 8
" 7
" 6
" 5
" 4
" 3
" 2
" 1
" 0

( BC_2, BcData_H(63),
( BC_2, SysCheck_L(7),
( BC_2, BcCheck_H(15),
( BC_2, BcCheck_H(7),
( BC_2, SysAddOut_L(0),
( BC_2, SysAddOut_L(1),
( BC_2, SysAddOut_L(2),
( BC_2, SysAddOut_L(3),
( BC_2, SysAddOut_L(4),
( BC_2, SysAddOut_L(5),
( BC_2, SysAddOut_L(6),
( BC_2, SysAddOut_L(7),
( BC_2, SysAddOutClk_L,
( BC_2, SysAddOut_L(8),
( BC_2, SysAddOut_L(9),
( BC_2, SysAddOut_L(10),
( BC_2, SysAddOut_L(11),
( BC_2, SysAddOut_L(12),
( BC_2, SysAddOut_L(13),
( BC_2, SysAddOut_L(14),
( BC_3, SysAddIn_L(0),
( BC_3, SysAddIn_L(1),
( BC_3, SysAddIn_L(2),
( BC_3, SysAddIn_L(3),
( BC_3, SysAddIn_L(4),
( BC_3, SysAddIn_L(5),
( BC_3, SysAddIn_L(6),
( BC_3, SysAddIn_L(7),
( BC_3, SysAddIn_L(8),
( BC_3, SysAddInClk_L,
( BC_3, SysAddIn_L(9),
( BC_3, SysAddIn_L(10),
( BC_3, SysAddIn_L(11),
( BC_3, SysAddIn_L(12),
( BC_3, SysAddIn_L(13),
( BC_3, SysAddIn_L(14),
( BC_3, SysFillValid_L,
( BC_3, SysDataInValid_L,
( BC_3, SysDataOutValid_L,

BIDIR,
BIDIR,
BIDIR,
BIDIR,
OUTPUT2,
OUTPUT2,
OUTPUT2,
OUTPUT2,
OUTPUT2,
OUTPUT2,
OUTPUT2,
OUTPUT2,
OUTPUT2,
OUTPUT2,
OUTPUT2,
OUTPUT2,
OUTPUT2,
OUTPUT2,
OUTPUT2,
OUTPUT2,
INPUT,
INPUT,
INPUT,
INPUT,
INPUT,
INPUT,
INPUT,
INPUT,
INPUT,
INPUT,
INPUT,
INPUT,
INPUT,
INPUT,
INPUT,
INPUT,
INPUT,
INPUT,
INPUT,

x, 50,
x, 53,
x, 50,
x, 50,
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x

0,
0,
0,
0,

Z
), "& -WEAK1 ), "& -Z
), "& -Z
), "& -), "& -), "& -), "& -), "& -), "& -), "& -), "& -), "& -), "& -), "& -), "& -), "& -), "& -), "& -), "& -), "& -), "& -), "& -), "& -), "& -), "& -), "& -), "& -), "& -), "& -), "& -), "& -), "& -), "& -), "& -), "& -), "& -), "& -), "& -) ";

attribute DESIGN_WARNING of Alpha_21264b: entity is
"1. IEEE 1149.1 circuits on Alpha 21264b are designed primarily to support
"&
"
testing in off-line module manufacturing environment. The SAMPLE/PRELOAD"&
"
instruction support is designed primarily for supporting interconnection"&
"
verification test and not for at-speed samples of pin data.
"&
"2. TDO is Open-Drain signal.
"&
"3. Add comment on port pin electrical characteristics:
"&
"4. Comment out if compiler does not support this statement.
";
end Alpha_21264b;

B–12 21264/EV68A Boundary-Scan Register

21264/EV68A Hardware Reference Manual

C
Serial Icache Load Predecode Values
See the Alpha Motherboards Software Developer’s Kit (SDK) for information.

21264/EV68A Hardware Reference Manual

Serial Icache Load Predecode Values

C–1

D
PALcode Restrictions and Guidelines
D.1 Restriction 1 : Reset Sequence Required by Retire Logic and
Mapper
For convenience of implementation, the Ibox retire logic done status bits are not initialized during reset. Instead, as shown in the example below, the first batch of valid
instructions sweeps through inum-space and initializes these bits. The 80 status bits
(one for each inflight instruction) must be marked not done by the first 80 instructions
mapped after reset, and later marked done when those instructions are retired. Therefore, the first 20 fetch blocks must contain four valid instructions each, and must not
contain any retire logic NOP instructions.
reset:
** (1) Initialize 80 retirator "done" status bits and
** the integer and floating mapper destinations.
** (2) Do A MTPR ITB_IA, which turns on the mapper source
** enables.
** (3) Create a map stall to complete the ITB_IA.
**
** State after execution of this code:
**
retirator initialized
**
destinations mapped
**
source mapping enabled
**
itb flushed
**
** The PALcode need not assume the following since the SROM is not
** required to do these:
**
dtb
flushed
**
dtb_asn0
0
**
dtb_asn1
0
**
dtb_alt_mode
0
*/
/*
** Initialize retirator and destination map, doing 80 retires.
*/
addq
r31,r31,r0
/* initialize Int. Reg. 0*/
addq
r31,r31,r1
/* initialize Int. Reg. 1*/
addt
f31,f31,f0
/* initialize F.P. Reg. 0*/
mult
f31,f31,f1
/* initialize F.P. Reg. 1*/
addq
addq

r31,r31,r2
r31,r31,r3

/* initialize Int. Reg. 2*/
/* initialize Int. Reg. 3*/

21264/EV68A Hardware Reference Manual

PALcode Restrictions and Guidelines

D–1

Restriction 1 : Reset Sequence Required by Retire Logic and Mapper

D–2

addt
mult

f31,f31,f2
f31,f31,f3

/* initialize F.P. Reg. 2*/
/* initialize F.P. Reg. 3*/

addq
addq
addt
mult

r31,r31,r4
r31,r31,r5
f31,f31,f4
f31,f31,f5

/* initialize Int. Reg. 4*/
/* initialize Int. Reg. 5*/
/* initialize F.P. Reg. 4*/
/* initialize F.P. Reg. 5*/

addq
addq
addt
mult

r31,r31,r6
f31,r31,r7
f31,f31,f6
f31,f31,f7

/* initialize Int. Reg. 6*/
/* initialize Int. Reg. 7*/
/* initialize F.P. Reg. 6*/
/* initialize F.P. Reg. 7*/

addq
addq
addt
mult

r31,r31,r8
r31,r31,r9
f31,f31,f8
f31,f31,f9

/* initialize Int. Reg. 8*/
/* initialize Int. Reg. 9*/
/* initialize F.P. Reg. 8*/
/* initialize F.P. Reg. 9*/

addq
addq
addt
mult

r31,r31,r10
r31,r31,r11
f31,f31,f10
f31,f31,f11

/* initialize Int. Reg. 10*/
/* initialize Int. Reg. 11*/
/* initialize F.P. Reg. 10*/
/* initialize F.P. Reg. 11*/

addq
addq
addt
mult

r31,r31,r12
r31,r31,r13
f31,f31,f12
f31,f31,f13

/* initialize Int. Reg. 12*/
/* initialize Int. Reg. 13*/
/* initialize F.P. Reg. 12*/
/* initialize F.P. Reg. 13*/

addq
addq
addt
mult

r31,r31,r14
r31,r31,r15
f31,f31,f14
f31,f31,f15

/* initialize Int. Reg. 14*/
/* initialize Int. Reg. 15*/
/* initialize F.P. Reg. 14*/
/* initialize F.P. Reg. 15*/

addq
addq
addt
mult

r31,r31,r16
r31,r31,r17
f31,f31,f16
f31,f31,f17

/* initialize Int. Reg. 16*/
/* initialize Int. Reg. 17*/
/* initialize F.P. Reg. 16*/
/* initialize F.P. Reg. 17*/

addq
addq
addt
mult

r31,r31,r18
r31,r31,r19
f31,f31,f18
f31,f31,f19

/* initialize Int. Reg. 18*/
/* initialize Int. Reg. 19*/
/* initialize F.P. Reg. 18*/
/* initialize F.P. Reg. 19*/

addq
addq
addt
mult

r31,r31,r20
r31,r31,r21
f31,f31,f20
f31,f31,f21

/* initialize Int. Reg. 20*/
/* initialize Int. Reg. 21*/
/* initialize F.P. Reg. 20*/
/* initialize F.P. Reg. 21*/

addq
addq
addt
mult

r31,r31,r22
r31,r31,r23
f31,f31,f22
f31,f31,f23

/* initialize Int. Reg. 22*/
/* initialize Int. Reg. 23*/
/* initialize F.P. Reg. 22*/
/* initialize F.P. Reg. 23*/

addq
addq
addt
mult

r31,r31,r24
r31,r31,r25
f31,f31,f24
f31,f31,f25

/* initialize Int. Reg. 24*/
/* initialize Int. Reg. 25*/
/* initialize F.P. Reg. 24*/
/* initialize F.P. Reg. 25*/

addq

r31,r31,r26

/* initialize Int. Reg. 26*/

PALcode Restrictions and Guidelines

21264/EV68A Hardware Reference Manual

Restriction 1 : Reset Sequence Required by Retire Logic and Mapper

addq
addt
mult

r31,r31,r27
f31,f31,f26
f31,f31,f27

/* initialize Int. Reg. 27*/
/* initialize F.P. Reg. 26*/
/* initialize F.P. Reg. 27*/

addq
addq
addt
mult

r31,r31,r28
r31,r31,r29
f31,f31,f28
f31,f31,f29

/* initialize Int. Reg. 28*/
/* initialize Int. Reg. 29*/
/* initialize F.P. Reg. 28*/
/* initialize F.P. Reg. 29*/

addq
addt
addq
addq

r31,r31,r30
f31,f31,f30
r31,r31,r0
r31,r31,r0

/* initialize Int. Reg. 30*/
/* initialize F.P. Reg. 30*/
/* initialize retirator 63*/
/* initialize retirator 64*/

addq
addq
addq
addq

r31,r31,r0
r31,r31,r0
r31,r31,r0
r31,r31,r0

/* initialize retirator 65*/
/* initialize retirator 66*/
/* initialize retirator 67*/
/* initialize retirator 68*/

addq
addq
addq
addq

r31,r31,r0
r31,r31,r0
r31,r31,r0
r31,r31,r0

/* initialize retirator 69*/
/* initialize retirator 70*/
/* initialize retirator 71*/
/* initialize retirator 72*/

addq
addq
addq
addq

r31,r31,r0
r31,r31,r0
r31,r31,r0
r31,r31,r0

/* initialize retirator 73*/
/* initialize retirator 74*/
/* initialize retirator 75*/
/* initialize retirator 76*/

addq
addq
addq
addq

r31,r31,r0
r31,r31,r0
r31,r31,r0
r31,r31,r0

/* initialize retirator 77*/
/* initialize retirator 78*/
/* initialize retirator 79*/
/* initialize retirator 80*/

/* stop deleting*/
mtpr

r31,EV6__ITB_IA

mtpr
mtpr
mtpr

r31,EV6__DTB_IA
r31,EV6__VA_CTL
r31,EV6__M_CTL

/* flush the ITB (SCRBRD=4) *** this also
turns on mapper source enables ****/
/* flush the DTB (SCRBRD=7)*/
/* clear VA_CTL (SCRBRD=5)*/
/* clear M_CTL (SCRBRD=6)*/

/*
** Create a stall outside the IQ until the mtpr EV6__ITB_IA retires.
** We can use DTB_ASNx even though we don’t seem to follow the restriction on
** scoreboard bits (4-7).It’s okay because there are no real dstream
** operations happening.
*/
mtpr r31,EV6__DTB_ASN0
/* clear DTB_ASN0 (SCRBRD=4) creates a mapstall under the above mtpr to SCRBRD=4*/
mtpr r31,EV6__DTB_ASN1
/* clear DTB_ASN1 (SCRBRD=7)*/
mtpr r31,EV6__CC_CTL
/* clear CC_CTL (SCRBRD=5)*/
mtpr r31,EV6__DTB_ALT_MODE/* clear DTB_ALT_MODE (SCRBRD=6)*/

/*
** MAP_SHADOW_REGISTERS
**
** The shadow registers are mapped. This code may be done by the SROM

21264/EV68A Hardware Reference Manual

PALcode Restrictions and Guidelines

D–3

Restriction 1 : Reset Sequence Required by Retire Logic and Mapper

** or the PALcode, but it must be done in the manner and order below.
**
** It assumes that the retirator has been initialized, that the
** non-shadow registers are mapped, and that mapper source enables are on.
**
** Source enables are on. For fault-reset and wake from sleep, we need to
** ensure we are in the icache so we don’t fetch junk that touches the
** shadow sources before we write the destinations. For normal reset,
** we are already in the icache. However, so this macro is useful for
** all cases, force the code into the icache before doing the mapping.
**
** Assume for fault-reset, and wake from sleep case, the exc_addr is
** stored in r1.
*/
addq
r31,r31,r0
/* nop*/
addq
r31,r31,r0
/* nop*/
addq
r31,r31,r0
/* nop*/
br
r31, tch0
/* fetch in next block*/
.align 3
nxt0: lda
r0,0x0086(r31)
mtpr
r0,EV6__I_CTL
br
r31, nxt1
tch0: br
r31, tch1
nxt1: mtpr r31,EV6__IER_CM

D–4

/* load I_CTL.....*/
/* .....SDE=2, IC_EN=3 (SCRBRD=4)*/
/* continue executing in next block*/
/* fetch in next block*/
/* clear IER_CM (SCRBRD=4) creates a map-stall
under the above mtpr to SCRBRD=4*/
/* nop*/
/* continue executing in next block*/
/* fetch in next block*/

addq
br
tch1: br

r31,r31,r0
r31, nxt2
r31, tch2

nxt2: addq

r31,r31,r0

addq
br
tch2: br

r31,r31,r0
r31, nxt3
r31, tch3

/* 1st buffer fetch block for above mapstall*/
/* nop*/
/* continue executing in next block*/
/* fetch in next block*/

nxt3: addq
addq
br
tch3: br

r31,r31,r0
r31,r31,r0
r31, nxt4
r31, tch4

/* 2nd buffer fetch block for above map-stall*/
/* nop*/
/* continue executing in next block*/
/* fetch in next block*/

nxt4: addq

r31,r31,r0

addq
br
tch4: br

r31,r31,r0
r31, nxt5
r31, tch5

/* need 3rd buffer fetch block to get correct
SDE bit for next fetch block*/
/* nop*/
/* continue executing in next block*/
/* fetch in next block*/

nxt5: addq
addq
br
tch5: br

r31,r31,r4
r31,r31,r5
r31, nxt6
r31, tch6

/* initialize Shadow Reg. 0*/
/* initialize Shadow Reg. 1*/
/* continue executing in next block*/
/* fetch in next block*/

nxt6: addq
addq
br
tch6: br

r31,r31,r6
r31,r31,r7
r31, nxt7
r31, tch7

/* initialize Shadow Reg. 2*/
/* initialize Shadow Reg. 3*/
/* continue executing in next block*/
/* fetch in next block*/

nxt7: addq

r31,r31,r20

/* initialize Shadow Reg. 4*/

PALcode Restrictions and Guidelines

21264/EV68A Hardware Reference Manual

Restriction 1 : Reset Sequence Required by Retire Logic and Mapper

addq
br
tch7: br

r31,r31,r21
r31, nxt8
r31, tch8

/* initialize Shadow Reg. 5*/
/* continue executing in next block*/
/* fetch in next block*/

nxt8: addq
addq
br
tch8: br
nxt9:

r31,r31,r22
r31,r31,r23
r31, nxt9
r31, nxt0

/* initialize Shadow Reg. 6*/
/* initialize Shadow Reg. 7*/
/* continue executing in next block*/
/* go back to 1st block and start executing*/

/*
** INIT_WRITE_MANY
**
** Write the cbox write many chain, initializing the bcache configuration.
**
** This code is on a cache block boundary,
**
** *** the bcache is initialized OFF for the burnin test ***
*/
/*
** Because we aligned on and fit into a icache block, and because sbe=0,
** and because we do an mb at the beginning (which blocks further progress
** until the entire block has been fetched in), we don’t have to
** fool with pulling this code in before executing it.
*/
#undef bc_enable_a
#undef init_mode_a
#undef bc_size_a
#undef zeroblk_enable_a
#undef enable_evict_a
#undef set_dirty_enable_a
#undef bc_bank_enable_a
#undef bc_wrt_sts_a
#define bc_enable_a
#define init_mode_a
#define bc_size_a
#define zeroblk_enable_a
#define enable_evict_a
#define set_dirty_enable_a
#define bc_bank_enable_a
#define bc_wrt_sts_a

0
0
0
1
0
0
0
0

loadwm:
lda
r1, WRITE_MANY_CHAIN_H(r31)
sll
r1, 32, r1
/* data<35:32> */
LDLI(r1, WRITE_MANY_CHAIN_L, r1)
/* data<31:00> */
addq
r31,6,r0
/* shift in 6x 6-bits*/
mb
/* wait for all istream/dstream to complete*/
br
r31, bccshf
.align 6
bccshf:mtpr r1,EV6__DATA
subq
r0,1,r0
beq
r0,bccend
srl
r1,6,r1

21264/EV68A Hardware Reference Manual

/* shift in 6 bits*/
/* decrement R0*/
/* done if R0 is zero*/
/* align next 6 bits*/

PALcode Restrictions and Guidelines

D–5

Restriction 1 : Reset Sequence Required by Retire Logic and Mapper

br
bccend:mtpr
addq
addq

r31,bccshf
/* continue shifting*/
r31,EV6__EXC_ADDR + 16/* dummy IPR write - sets SCBD bit 4
r31,r31,r0
/* nop*/
r31,r31,r1
/* nop*/

mtpr

r31,EV6__EXC_ADDR + 16

beq
br
addq

r31, bccnxt
r31, .-4
r31,r31,r1

/* also a dummy IPR write /* stalls until above write
/* retires*/
/* predicts fall through in PALmode*/
/* fools ibox predictor into infinite loop*/
/* nop*/

bccnxt:addq r31,4,r0
/* load PCTX.....*/
mtpr
r0,EV6__PROCESS_CONTEXT
/* ..... FPE=1 (SCRBRD=4)*/
lda
r0,DC_CTL_INIT_K(r31)
/* load DC_CTL.....*/
mtpr
r0,EV6__DC_CTL
/* .....ECC_EN=0, FHIT=0, SET_EN=3
/* (SCRBRD=6)*/
addq
addq
lda
zap

r31,r31,r0
r31,r31,r1
r0,0xff61(r31)
r0,0xfc,r0

mtpr
mtpr
mtpr
mtpr

r31,EV6__DTB_TAG0 /* write DTB_TAG0 (SCRBRD=2,6)*/
r31,EV6__DTB_TAG1 /* write DTB_TAG1 (SCRBRD=1,5)*/
r0,EV6__DTB_PTE0 /* write DTB_PTE0 (SCRBRD=0,4)*/
r0,EV6__DTB_PTE1 /* write DTB_PTE1 (SCRBRD=3,7)*/

mtpr
lda
sll
itoft

r31,EV6__SIRR
r0,0x08FF(r31)
r0,52,r0
r0, f0

mt_fpcr f0
lda
r0,0x2086(r31)
ldah
r0,0x0050(r0)
mtpr

r0,EV6__I_CTL

mtpr
lda

r31,EV6__CC
r0,0x001F(r31)

/* nop*/
/* nop*/
/* R0 = ^xff61 (superpage) */
/* PTE protection for DTB write in next
block*/

/* clear SIRR (SCRBRD=4)*/
/* load FPCR.....*/
/* .....initial FPCR value*/
/* nop
itoftr0,f0; value = 0x8FF0000000000000*/
/* nop
mt_fpcrf0,f0,f0; do the load*/
/* load I_CTL.....*/
/* .....TB_MB_EN=1, CALL_PAL_R23=1, SL_XMIT=1,
/* SBE=0, SDE=2, IC_EN=3*/
/* value = 0x0000000000502086 (SCRBRD=4)*/

sll
mtpr

/* clear CC (SCRBRD=5)*/
/* write-one-to-clear bits in HW_INT_CLR,
/* I_STAT and DC_STAT*/
r0,28,r0
/* value = 0x00000001F0000000*/
r0,EV6__HW_INT_CLR/* clear bits in HW_INT_CLR (SCRBRD=4)*/

mtpr

r0,EV6__I_STAT

lda
mtpr
addq

r0,0x001F(r31)
r0,EV6__DC_STAT
r31,r31,r0

/* clear bits in I_STAT
/*(SCRBRD=4) creates a map-stall
/* under the above mtpr to SCRBRD=4*/
/* value = 0x000000000000001F*/
/* clear bits in DC_STAT (SCRBRD=6)*/
/* nop*/

mtpr

r31,EV6__PCTR_CTL /* 1st buffer fetch block for above map-stall
/* and 1st clear PCTR_CTL (SCRBRD=4)*/
bis
r31,1,r0
/* set up value for demon write*/
bis
r31,1,r0
/* set up value for demon write*/
mulq/v r31,r31,r0
/* nop*/
D–6

PALcode Restrictions and Guidelines

21264/EV68A Hardware Reference Manual

Restriction 1 : Reset Sequence Required by Retire Logic and Mapper

mtpr
bis
bis
mulq

r31,EV6__PCTR_CTL /* 2nd buffer fetch block for above map-stall
/* and 2nd clear PCTR_CTL (SCRBRD=4)*/
r31,1,r0
/* set up value for demon write*/
r31,1,r0
/* set up value for demon write*/
r31,r31,r0
/* nop*/

lda

r0,0x780(r31)

/* this is new initialization stuff to
prevent*/

/* ld/st below from going off-chip */

r31,1,r0

/* set up value for demon write*/

r1,0x780(r31)
r0,0x788(r31)

/* flush Pipe 0 LD logic*/
/* flush Pipe 1 LD logic*/
/* wait for LD’s to complete*/
/* wait for LD’s to complete*/

mb
whint
mb
bis
ldq_p
ldq_p
mb
mb

stq_p r1,0x780(r31)
/* flush Pipe 0 ST logic*/
stq_p r0,0x788(r31)
/* flush Pipe 1 ST logic*/
bis
r31, 32, r0
/* load loop count of 32*/
jsr_init_loop:
bsr
r31,jsr_init_loop_nxt
/* JSR to PC+4*/
jsr_init_loop_nxt:
stq_p r1,0x780(r31)
/* flush Pipe 0 ST logic*/
subq
r0,1,r0
/* decrement loop count*/
beq
r0,jsr_init_done /* done?*/
br
r31,jsr_init_loop /* continue loop*/
jsr_init_done:
lda
r0,0x03FF(r31)
sll
r0,52,r0
itoft r0,f0
addq
r31,r31,r1
mult
addt
divt
sqrtt

f0,f0,f0
f0,f0,f0
f0,f0,f0
f0,f0

/* create FP one..... */
/* .....value = 0x3FF0000000000000 */
/* put it into F0 reg */
/* nop (also clears R1) */
/* flush mul-pipe */
/* flush add-pipe */
/* flush div-pipe */
/* flush div-pipe */

cvtqt f0,f0
perr r31,r31,r0
maxuw4 r31,r31,r0
pkwb r31,r0

/* flush add-pipe (integer logic) */
/* flush MVI logic */
/* flush MVI logic */
/* flush MVI logic */

rc
addq
addq
addq

/* clear interrupt flag*/
/* nop (also clears R1)*/
/* nop (also clears R1)*/
/* nop (also clears R1)*/

r0
r31,r31,r1
r31,r31,r1
r31,r31,r1

/*
* This palbase init exists for the rare cases
* when this code is loaded into upper memory.
* That is the case when this code is loaded
* and executed in memory on a system that has
* already been initialized. This technique
* can sometimes be used to debug snippets of
* this code.
*/

21264/EV68A Hardware Reference Manual

PALcode Restrictions and Guidelines

D–7

Restriction 2 : No Multiple Writers to IPRs in Same Scoreboard Group

br
r31,palbase_init
palbase_init:
br
r0, br60
/* r0 <- current location */
br60: lda
r1, (EntryPoint-br60)(r0) /* r1 <- location of codebase */
mtpr
r1, EV6__PAL_BASE /* set up pal_base register */

bis
mtpr

r31, 2, r0
r0, EV6__VA_CTL

bis
mtpr

r31, 8, r0
r0, EV6__M_CTL

br
r0, jmp0
jmp0: addq
r0, (jmp1-jmp0+1), r0
hw_rets/jmp(r0)
jmp1:
lda
sll
mtpr

r1, 1(r31)
r1, 32, r1
r1, EV6__CC_CTL

/* r1 <- cc_ctl enable bit */
/* Enable/clear the cycle counter. */

/*
** Now initialize the dcache to allow the
** minidebugger so save gpr’s
*/

D.2 Restriction 2 : No Multiple Writers to IPRs in Same Scoreboard
Group
For convenience of implementation, only one explicit writer (HW_MTPR) to IPRs that
are in the same group can appear in the same fetch block (octaword-aligned octaword).
Multiple explicit writers to IPRs that are not in the same scoreboard group can appear.
If this restriction is violated, the IPR readers might not see the in-order state. Also, the
IPR might ultimately end up with a bad value.

D.3 Restriction 4 : No Writers and Readers to IPRs in Same Scoreboard Group
This restriction is made for the convenience of microprocessor implementation.
An explicit reader of an IPR in a particular scoreboard group cannot follow an explicit
writer (HW_MTPR) to an IPR in that same scoreboard group within one fetch block
(octaword-aligned octaword). Also within one fetch block, an implicit reader of an IPR
in a particular scoreboard group cannot follow an explicit writer (HW_MTPR) to an
IPR in that scoreboard group. This restriction covers writes to DTB_PTE or DTB_TAG
followed by LD, ST, or any memory operation, including all types of JMP instructions
and HW_RET instructions that do not have the STALL bit set.

D–8

PALcode Restrictions and Guidelines

21264/EV68A Hardware Reference Manual

Guideline 6 : Avoid Consecutive Read-Modify-Write-Read-Modify-Write

D.4 Guideline

6 : Avoid Consecutive Read-Modify-Write-ReadModify-Write

Avoid consecutive read-modify-write-read-modify-write sequences to IPRs in the same
scoreboard group.
The latency between the first write and the second read is determined by the retire
latency of the IPR. For convenience of implementation, the latency between the time
when the read is issued and when the final write is issued depends on the run-time contents of the issue queue. It is somewhere between four and nine cycles, even if there is
no data dependency between the read and write.

D.5 Restriction 7 : Replay Trap, Interrupt Code Sequence, and STF/
ITOF
On an Mbox replay trap, the 21264/EV68A Ibox guarantees that the refetched load or
store instruction that caused the trap is issued before any newer load or store instructions. For load and integer store instructions, this is a consequence of the natural operation of the issue queue. The refetched instruction enters the age-prioritized queue ahead
of newer load and store instructions and does not have any dependencies on dirty registers.
Because there is no overhead time for checking these register dependencies (that is, it is
known upon enqueueing that there are no dirty registers), the queue will issue the
refetched instruction in priority order. For floating-point store instructions, there is normally some overhead associated with checking the floating-point source register dirty
status, so the store instruction would normally wait before being issued. This would
have the undesired consequence of allowing newer load and store instructions to be
issued out of order. A deadlock can occur if issuing the instructions out-of-order causes
the floating-point store instruction to continually replay the trap. To avoid the deadlock
on a floating-point store instruction replay trap, the source register dirty status is not
checked (the source register is assumed to be clean because the store instruction was
issued previously).
The hardware mechanism that keeps track of replayed floating-point store instructions,
and cancels the dirty register check, requires some software restrictions to guarantee
that it is applied appropriately to the replayed instruction and not to other floating-point
store instructions. The hardware mechanism marks the position in the fetch block (low
two bits of the PC) where the replay trap occurred. This action cancels the dirty floating-point source register check of the next valid instruction enqueued to the integer
queue (integer, all load and store, and ITOF instructions) that has the same position in
the fetch block (normally the replayed STF). If the PC is somehow diverted to a PALcode flow, this hardware might inadvertently cancel the register check of some other
STF (or ITOF) instruction. Fortunately, there are a minimal number of reasons why the
PC might be diverted during a replay trap. They are interrupts and ITB fills.
The following PALcode example shows that an STF or ITOF instruction, in a given
position in a fetch block, must be preceded by a valid instruction that is issued out of
the integer queue in the same position in an earlier fetch block. Acceptable instruction
classes include load, integer store, and integer operate instructions that do not have R31
as a destination or branch.

21264/EV68A Hardware Reference Manual

PALcode Restrictions and Guidelines

D–9

Restriction 9 : PALmode Istream Address Ranges

Bad_interrupt_flow_entry:
ADDQ R31,R31,R0
STF Fa,(Rb) ; This STF might not undergo a dirty source register
; check and might give wrong results
ADDQ R31,R31,R0
ADDQ R31,R31,R0
................................
Good_interrupt_flow_entry:
ADDQ R31,R31,R0; Enables FP dirty source register
; check for (PC[1:0] == 00)
ADDQ R31,R31,R0; Enables FP dirty source register
; check for (PC[1:0] == 01)
ADDQ R31,R31,R0; Enables FP dirty source register
; check for (PC[1:0] == 10)
ADDQ R31,R31,R0; Enables FP dirty source register
; check for (PC[1:0] == 11)
ADDQ R31,R31,R0
STF Fa,(Rb); This STF will successfully undergo
; a dirty source register check
ADDQ R31,R31,R0
ADDQ R31,R31,R0

D.6 Restriction 9 : PALmode Istream Address Ranges
PALmode[physical] Istream addresses must ensure proper sign extension for the
selected value of I_CTL[VA_48]. When I_CTL[VA_48] is clear, indicating 43-bit virtual address format, PALmode[physical] Istream addresses must sign-extend address
bits above bit 42 although the physical address range is 44 bits. An illegal address can
only be generated by a PALmode JSR-type instruction or a HW_RET instruction
returning to a PALmode address.

D.7 Restriction 10: Duplicate IPR Mode Bits
The virtual address size is selectable by programming IPR bits I_CTL[VA_48]
and VA_CTL[VA_48]. These bit values should usually be equal when operating in
native (virtual) mode. The I_CTL[VA_48] bit determines the DTB double3/double4
PALcode entry, the JSR mispredict comparison width, the VPC address generation
width, the Istream ACV limits, and the IVA_FORM format selection. The
VA_CTL[VA_48] bit determines the VA_FORM format selection and the Dstream
ACV limits. IPR mode bits I_CTL[VA_FORM_32] and VA_CTL[VA_FORM_32]
should be consistent when executing in native mode.

D–10 PALcode Restrictions and Guidelines

21264/EV68A Hardware Reference Manual

Restriction 11: Ibox IPR Update Synchronization

D.8 Restriction 11: Ibox IPR Update Synchronization
When updating any Ibox IPR, a return to native (virtual) mode should use the HW_RET
instruction with the associated STALL bit set to ensure that the updated IPR value
affects all instructions following the return path. The new IPR value takes effect only
after the associated HW_MTPR instruction is retired.
For update to some IPR fields with propagation delay, such as I_CTL[SDE] and
PCTX[FPE], synchronization as described in Section D.32 is the preferred method of
synchronization.

D.9 Restriction 12: MFPR of Implicitly-Written IPRs EXC_ADDR,
IVA_FORM, and EXC_SUM
Implicitly written IPRs are non-renamed hardware registers that must be available for
subsequent traps. After any trap to PALcode, hardware protects the values from a second implicit write by locking these registers and delaying subsequent traps for a safe
(limited time). Their values can be read reliably by a HW_MFPR within the first four
instructions of a PALcode flow and prior to any taken branch in that PALcode flow,
whichever is earlier. These instructions should not include PALmode trapping instructions. After the delimiting instruction defined above retires, these registers are unlocked
and may change due to new exception conditions.
If a second exception occurs before the registers are unlocked, it will be either delayed
or forced to replay trap (a non-PALmode trap) until the register has been unlocked.
After being unlocked, a subsequent new path exception condition will be allowed to
reload the register and trap to PALcode. The 21264/EV68A may complete execution of
the first PALcode flow, encountering the second exception condition before the delimiting instruction is retired, hence the need for the locking mechanism to ensure visibility
of the initial register value.
The VA_FORM, VA, and MM_STAT registers are not included in this list of protected
IPRS. See Section D.24 for a description of how to protect these IPRs from subsequent
implicit writers.

D.10 Restriction 13 : DTB Fill Flow Collision
Two DTB fill flows might collide such that the HW_MTPR’s in the second fill could be
issued before all of the HW_MTPR’s in the first PALcode flow are retired. This can be
prevented by putting appropriate software scoreboard barriers in the PALcode flow.

D.11 Restriction 14 : HW_RET
There can be no HW_RET in the first fetch block of a PALcode routine, other
than CALL_PAL routines. With a HW_RET in the first fetch block of a PALcode routine, the HW_RET will be mispredicted and the JSR/RETURN stack could lose its synchronization.

21264/EV68A Hardware Reference Manual

PALcode Restrictions and Guidelines D–11

Guideline 16 : JSR-BAD VA

D.12 Guideline 16 : JSR-BAD VA
A JSR memory format instruction that generates a bad VA (IACV) trap requires PALcode assistance to determine the correct exception address. If the
EXC_SUM[BAD_IVA] is set, bits [63,1] of the exception address are valid in the VA
IPR and not the EXC_ADDR as usual. The PALmode bit, however, is always located in
EXC_ADDR[0] and must be combined, if necessary, by PALcode to determine the full
exception address.

D.13 Restriction 17: MTPR to DTB_TAG0/DTB_PTE0/DTB_TAG1/
DTB_PTE1
These four write operations must be executed atomically, that is, either all four must be
retired or none of them may be retired.

D.14 Restriction 18: No FP Operates, FP Conditional Branches,
FTOI, or STF in Same Fetch Block as HW_MTPR
No FP operate instructions (including Mx_FPCR), FP conditional branches, FTOI register move instructions, or FP store instructions are allowed in the same fetch block as
any HW_MTPR instructions. This includes ADDx/MULx/DIVx/SQRTx/FPConditionalBranch/STx/FTOIx, where x is any applicable FP data type, but does not include
LDx/ITOFx.

D.15 Restriction 19: HW_RET/STALL After Updating the FPCR by
way of MT_FPCR in PALmode
FPCR updating occurs in hardware based on the retirement of a nontrapping version of
MT_FPCR (in PALcode). Use a HW_RET/STALL after the nontrapping MT_FPCR to
achieve minimum latency (four cycles) between the retiring of the MT_FPCR and the
first FLOP that uses the updated FPCR.

D.16 Guideline 20 : I_CTL[SBE] Stream Buffer Enable
The I_CTL[SBE] bits should not be enabled when running with the Icache disabled to
avoid potentially long fill delays. When the Icache is disabled, the only method of supplying instructions is by way of a stream hit. If the fill is returned in non-sequential
wrap order, the stream will continue fetching through the entire page while waiting for
a hit. Normally the data will be found in the cache.

D.17 Restriction 21: HW_RET/STALL After HW_MTPR ASN0/ASN1
There must be a scoreboard bit-to-register dependency chain to prevent HW_MTPR
ASN0 or HW_MTPR ASN1 from being issued while any of scoreboard bits [7:4] are
set. The following example contains a code sequence that creates the dependency chain.
:Assume Ra holds value to write to ASN0/ASN1
HW_MFPR R0, VA, SCBD<7,6,5,4>
XOR R0, R0, R0
BIS R0, R9, R9

D–12 PALcode Restrictions and Guidelines

21264/EV68A Hardware Reference Manual

Restriction 22: HW_RET/STALL After HW_MTPR IS0/IS1

BIS R31, R31, R31
HW_MTPR R9, ASN0, SCBD<4>
HW_MTPR R9, ASN1, SCBD<7>

This sequence guarantees, through the register dependency on R0, that neither
HW_MTPR are issued before scoreboard bits [7:4] are cleared. In addition, there must
be a HW_RET/STALL after a HW_MTPR ASN0/HW_MTPR ASN1 pair. Finally,
these two writes must be executed atomically, that is, either both must be retired or neither may be retired.

D.18 Restriction 22: HW_RET/STALL After HW_MTPR IS0/IS1
There must be a scoreboard bit-to-register dependency chain to prevent either
HW_MTPR IS0 or HW_MTPR IS1 from issuing instructions while any of scoreboard
bits [7:4] are set. The following example contains a code sequence that creates the
dependency chain.
HW_MFPR R0, VA, SCBD<7,6,5,4>,R0
XOR R0, R0, R0
BIS R0, R9, R9
BIS R31 ,R31, R31
HW_MTPR R9, IS0, SCBD<6>
HW_MTPR R9, IS1, SCBD<7>

This sequence guarantees, through the register dependency on R0, that neither
HW_MTPR are issued before scoreboard bits [7:4] are cleared. There must be a
HW_RET/STALL after an HW_MTPR IS0/HW_MTPR IS1 pair. Also, these two
writes must be executed atomically, that is, either both must be retired or neither may be
retired.

D.19 Restriction 23: HW_ST/P/CONDITIONAL Does Not Clear the
Lock Flag
A HW_ST/P/CONDITIONAL will not clear the lock flag such that a successive storeconditional (either STx_C or HW_ST/C) might succeed even in the absence of a loadlocked instruction. In the 21264/EV68A, a store-conditional is forced to fail if there is
an intervening memory operation between the store-conditional and its address-matching LDxL. The following example shows the memory operations.
LDL/Q/F/G/S/T
STL/Q/F/G/S/T
LDQ_U (not to R31)
STQ_U

Absent from this list are HW_LD (any type), HW_ST (any type), ECB, and WH64.
Their absence implies that they will not force a subsequent store-conditional instruction
to fail. PALcode must insert a memory operation from the above list after a HW_ST/
CONDITIONAL in order to force a future store-conditional to fail if it was not preceded by a load-locked operation:
HW_LDxL

21264/EV68A Hardware Reference Manual

PALcode Restrictions and Guidelines D–13

Restriction 24: HW_RET/STALL After HW_MTPR IC_FLUSH, IC_FLUSH_ASM,

xxx
HW_ST/C -> R0
Bxx R0, try_again
STQ ; Force next ST/C to fail if no preceding LDxL
HW_RET

D.20 Restriction 24: HW_RET/STALL After HW_MTPR IC_FLUSH,
IC_FLUSH_ASM, CLEAR_MAP
There must be a HW_RET/STALL after a HW_MTPR IC_FLUSH, IC_FLUSH_ASM, or
CLEAR_MAP. The Icache flush associated with these instructions will not occur until
the HW_RET/STALL occurs and all outstanding Istream fetches have been completed.
Also, there must be a guarantee that the HW_MTPR IC_FLUSH or HW_MTPR
IC_FLUSH_ASM will not be retired simultaneously with the HW_RET/STALL. This
can be ensured by inserting a conditional branch between the two (BNE R31, 0 cannot
be mispredicted in PALmode), or by ensuring at least 10 instructions between the
MTPR instruction and the HW_RET/STALL containing at least one instruction in each
quad aligned group with a valid destination. Finally, the HW_RET/STALL that is used
for CLEAR_MAP cannot trigger a cache flush. That is, if both a CLEAR_MAP and
IC_FLUSH are desired, there must be two HW_RET/STALLs, one following each
HW_MTPR.

D.21 Restriction 25: HW_MTPR ITB_IA After Reset
An HW_MTPR ITB_IA is required in the reset PALcode to initialize the ITB. It is also
required that PALcode not be exited, even via a mispredicted path until this
HW_MTPR ITB_IA has been retired. PALmode can change temporarily after fetching
a HW_RET, regardless of the STALL qualifier, down a mispredicted path leading to use
of the ITB before it is actually initialized.
Unexpected instruction fetch and execution can occur following misprediction of any
memory format control instruction (JMP, JSR, RET, JSR_CO, or HW_JMP, HW_JSR,
HW_RET, HW_JSR_CO regardless of the STALL qualifier), or after any mispredicted
conditional branch instruction. If the unexpected instruction flow contains a HW_RET
instruction, PALmode may be exited prematurely.
One way to ensure that PALmode is not exited is to place the HW_MTPR ITB_IA at
least 80 instructions before any possible HW_RET instruction can be encountered via
any fetch path. Since memory format control instructions can mispredict to any cache
location, they should also be avoided within these 80 instructions.

D.22 Guideline 26: Conditional Branches in PALcode
To avoid pollution of the branch predictors and improve overall branch prediction accuracy, conditional branch instructions in PALcode will be predicted to not be taken. The
only exception to this rule are conditional branches within the first cache fetch (up to
four instructions) of all PALcode flows except CALL_PAL flows. Conditional branches
should be avoided in this window.

D–14 PALcode Restrictions and Guidelines

21264/EV68A Hardware Reference Manual

Restriction 27: Reset of ‘Force-Fail Lock Flag’ State in PALcode

D.23 Restriction 27: Reset of ‘Force-Fail Lock Flag’ State in PALcode
A virtual mode load or store is required in PALcode before the execution of any loadlocked or store-conditional instructions. The virtual-mode load or store may not be a
HW_LD, HW_ST, LDx_L, ECB, or WH64.

D.24 Restriction 28: Enforce Ordering Between IPRs Implicitly Written by Loads and Subsequent Loads
Certain IPRs, which are updated as a result of faulting memory operations, require software assistance to maintain ordering against newer instructions. Consider the following
code sequence:
HW_MFPR IPR_MM_STAT
LDQ rx, (ry)

These instructions would typically be issued in-order. The HW_MFPR is data-ready
and both instructions use a lower subcluster. However, the HW_MFPRs (and
HW_MTPRs) respond to certain resource-busy indications and are not issued when the
Mbox informs the Ibox that a certain set of resources (store-bubbles) are busy. The LDs
respond to a different set of resource-busy indications (load-bubbles) and could be
issued around the HW_MFPR in the presence of the former. Software assistance is
required to enforce the issue order. One sure way to enforce the issue order is to insert
an MB instruction before the first load that occurs after the HW_MFPR MM_STAT.
The VA, VA_FORM, and DC_CTL registers require a similar constraint. All LOAD
instructions except HW_LD might modify any or all of these registers. HW_LD does
not modify MM_STAT.

D.25 Guideline 29 : JSR, JMP, RET, and JSR_COR in PALcode
Unprivileged JSR, JMP, RET, and JSR_COR instructions will always mispredict when
used in PALcode. In addition, HW_RET to a PALmode target will always mispredict
since the JSR stack only predicts native-mode return addresses. HW_RET to a nativemode target uses the JSR stack for prediction and should usually be used when exiting
PALmode in order to maintain JSR stack alignment since all PALmode traps also push
the value of the EXC_ADDR on the JSR stack.
Privileged versions of the JSR type instructions (HW_JSR,HW_JMP,HW_JSR_COR)
can be used both within PALmode or to exit PALmode and generate a predicted target
based on their hint bits and the current processor PALmode state.

D.26 Restriction 30 : HW_MTPR and HW_MFPR to the Cbox CSR
External bus activity must be isolated from writes and reads to the Cbox CSR. This
requires that all Dstream and Istream fills must be avoided until after the HW_MTPR/
HW_MFPR updates are completed. An MB instruction can block Dstream activity, but
blocking all Istream fills, including prefetches, requires more extensive code. The following code example blocks all Istream fill requests and stalls instruction fetch until
after the desired MTPR/MFPR action is completed. This code disables Istream
prefetching by way of a HW_MTPR to I_CTL[SBE], IC_FLUSH, and
HW_RET_STALL sequence.

21264/EV68A Hardware Reference Manual

PALcode Restrictions and Guidelines D–15

Restriction 30 : HW_MTPR and HW_MFPR to the Cbox CSR

ALIGN_FETCH_BLOCK
sys__cbox:
mb
hw_mfpr p6, EV6__I_CTL
lda
p4, ^xFCFF(r31)
and
p6, p4, p4

; quiet the dstream
; (4,0L) get i_ctl
; mask for clearing SBE bits
; clear SBE bits

sbe_off_offset = <sys__cbox_sbe_off_done - sys__cbox_sbe_off>
hw_mtpr p4, EV6__I_CTL
br
p6, sys__cbox_sbe_off
sys__cbox_sbe_off:
addq
p6, #<sbe_off_offset+1>, p6
bsr
r31, .
ALIGN_FETCH_BLOCK <^x47FF041F>; align
hw_mtpr r31, EV6__IC_FLUSH
bne
r31, .
PVC_JSR sbe_off
hw_ret_stall (p6)
PVC_JSR sbe_off, dest=1
sys__cbox_sbe_off_done:
br
r31, sys__cbox_touch1

; (4,0L) write new i_ctl

; past stall in palmode
; stack push
; (4,0L) eliminate prefetches
; pvc #24
; synch and flush
; use ret, pop stack
; br stops predictor
; now pull in the next block

ALIGN_CACHE_BLOCK
sys__cbox_over1:
addq
r31, #11, p6
addq
r31, r31, p7
br
r31, sys__cbox_over2
sys__cbox_touch1:
br
r31, sys__cbox_touch2

; block 1
; initialize shift count (11x)
; initialize shift data
; go to block 2
;
; touch block 2

sys__cbox_over2:
hw_mtpr r31, EV6__SHIFT_CONTROL
subq
p6, #1, p6
br
r31, sys__cbox_over3
sys__cbox_touch2:
br
r31, sys__cbox_touch3

; block 2
; (6,0L) shift in 6 bits
; decrement shift count
; go to block 3
;
; touch block 3

sys__cbox_over3:
hw_mtpr r31, <EV6__MM_STAT ! 64 >
bis
p5, #1, p5
br
r31, sys__cbox_over4
sys__cbox_touch3:
br
r31, sys__cbox_touch4

; block 3
; (6,0L) wait for shift
; return in pal mode
; go to block 4
;
; touch block 4

sys__cbox_over4:
hw_mfpr p4, EV6__DATA
bis
r31, r31, r31
br
r31, sys__cbox_over5
sys__cbox_touch4:
br
r31, sys__cbox_touch5

; block 4
; (6,0L) read cbox data
; nop
; go to block 5
;
; touch block 5

sys__cbox_over5:
and
addq
br
sys__cbox_touch5:
br

; block 5
; clean to <5:0>
; accumulate shift data
; go to block 6
;
; touch block 6

p4, #^x3F, p4
p4, p7, p7
r31, sys__cbox_over6
r31, sys__cbox_touch6

D–16 PALcode Restrictions and Guidelines

21264/EV68A Hardware Reference Manual

Restriction 31 : I_CTL[VA_48] Update

sys__cbox_over6:
beq
bis
br
sys__cbox_touch6:
br
sys__cbox_over7:
bis
sll
br
sys__cbox_touch7:
br

p6, sys__cbox_over8
r31, r31, r31
r31, sys__cbox_over7
r31, sys__cbox_touch7

p7, r31, p20
p7, #6, p7
r31, sys__cbox_over2
r31, sys__cbox_touch8

sys__cbox_over8:
beq
r31, sys__cbox_cbox_done
PVC_VIOLATE <1006>
br
r31, .-4
bis
r31, r31, r31
sys__cbox_touch8:
br
r31, sys__cbox_over1
sys__cbox_cbox_done:
hw_mfpr p6, EV6__I_CTL
lda
p4, <3@EV6__I_CTL__SBE__S>(r31)
or
p6, p4, p4
bis
r31, r31, r31

; block 6
; branch if done
; nop
; go to block 7
;
; touch block 7
; block 7
; save before shifting
; shift data 6 bits left
; do next shift
;
; touch block 8
; block 8
; predict not taken
; predict back to infinite loop
;
;
; now start executing the shifts
; now restore i_ctl
; (4,0L) get i_ctl
; sbe bits
; set SBE bits

hw_mtpr p4, EV6__I_CTL

; (4,0L) restore i_ctl

PVC_JSR cbox, bsr=1, dest=1
hw_ret_stall (p5)

; return to caller with stall

D.27 Restriction 31 : I_CTL[VA_48] Update
The VA_48 virtual address format cannot be changed while executing a JSR, JMP,
GOTO, JSR_COROUTINE, or HW_RET instruction. A simple method of ensuring
that the address does not change is to write I_CTL twice, in two separate fetch blocks,
with the same data. The second write will stall the pipeline and ensure that the mode
cannot change, even down a mispredicted path, while a following JSR type instruction
might be using the address comparison logic.

D.28 Restriction 32 : PCTR_CTL Update
The performance counter must not be left in a state near overflow. If counting is disabled, the counters may produce multiple overflow signals if the counter output is not
updated due to the counter being disabled. A repeated overflow signal with counters
disabled can block other incoming interrupt requests while the overflow state persists.
To avoid this situation, reads or writes to the counters should not leave a value near
overflow. In normal operation, with counters enabled, a counter overflow will produce
an overflow pulse, clear the counter, and produce a performance counter interrupt.
Interrupts can only be blocked for one cycle.

21264/EV68A Hardware Reference Manual

PALcode Restrictions and Guidelines D–17

Restriction 33 : HW_LD Physical/Lock Use

D.29 Restriction 33 : HW_LD Physical/Lock Use
The HW_LD physical/lock instruction must be one of the first three instructions in a
quad-instruction aligned fetch block. A pipeline error can occur if the HW_LD physical/lock is fetched as the fourth instruction of the fetch block.

D.30 Restriction 34 : Writing Multiple ITB Entries in the Same PALcode Flow
Before a PALcode flow writes multiple ITB entries, additional scoreboard bits should
be set to avoid possible corruption of the TAG IPR prior to final update in the ITB. The
addition of scoreboard bits 0 and 4 to the standard scoreboard bit 6 for ITB_TAG will
prevent subsequent HW_MTPR ITB_TAG writes from changing the staging register
TAG value prior to retirement of the HW_MTPR ITB_PTE that triggers the final ITB
update.

D.31 Guideline 35 : HW_INT_CLR Update
When writing the HW_INT_CLR IPR to clear interrupt requests, it may be necessary to
write the same value twice in distinct fetch blocks to ensure that the interrupt request is
cleared before exiting PALcode. A second write will cause a scoreboard stall until the
first write retires, creating a convenient synchronization with the PALmode exit.

D.32 Restriction 36 : Updating I_CTL[SDE]
A software interlock is required between updates of the I_CTL[SDE] and a subsequent
instruction fetch that may use any destination registers. A suggested method of ensuring
this interlock is to use two MTPR I_CTL instructions in separate fetch blocks, followed
by three more fetch blocks of non-NOP instructions.

D.33 Restriction 37 : Updating VA_CTL[VA_48]
A software interlock is required between updates of the VA_CTL[VA_48] and following LD or ST instructions. This is necessary since the VA_CTL update will not occur
until the HW_MTPR VA_CTL instruction retires. A sufficient method of ensuring this
interlock is to write the VA_CTL with the same data in two successive fetch blocks,
causing a mapper stall. The dependant LD or ST instructions can be placed in any location of the second fetch block.

D.34 Restriction 38 : Updating PCTR_CTL
When updating the PCTR_CTL, it may be necessary to write the update value twice. If
the counter being updated is currently disabled by way of the respective I_CTL or
PCTX bits, the value must be written twice to ensure that the counter overflow is properly cleared. The overflow bit is conditionally latched using the same write enable as
the counter update, so an additional write of the counter value will ensure that the overflow logic accurately reflects the addition of the new counter value plus the input conditions. The new update value must not be within one cycle of overflow (within 16 for
SL0, within 4 for SL1) as required by Section D.28.

D–18 PALcode Restrictions and Guidelines

21264/EV68A Hardware Reference Manual

Guideline 39: Writing Multiple DTB Entries in the Same PAL Flow

D.35 Guideline 39: Writing Multiple DTB Entries in the Same PAL
Flow
If a PALcode flow intends to write multiple DTB entries (as would occur in a double
miss), it must take care to keep subsequent HW_MTPR DTB_TAGx writes from corrupting the staging register TAG values prior to retirement of the HW_MTPR
DTB_PTEx, which triggers the final DTB update.
For example, in the double miss DTB flow, the following code could be used to hold up
the return to the single miss flow (the numbers in parentheses are the scoreboard bits):
hw_mtpr r4, EV6__DTB_TAG0
hw_mtpr r4, EV6__DTB_TAG1
hw_mtpr r5, EV6__DTB_PTE0
hw_mtpr r5, EV6__DTB_PTE1

; (2&6) write tag0
; (1&5) write tag 1
; (0&4) write pte0
; (3&7) write pte1

bis r31, r31, r31
bis r31, r31, r31
bis r31, r31, r31
hw_mtpr r31, <EV6__MM_STAT ! ^x80>

; force new fetch block

hw_ret

; return to single miss

(r6)

; (7) wait for pte write

D.36 Restriction 40: Scrubbing a Single-Bit Error
On Bcache and Memory single bit errors on Icache fills, the hardware flushes the
Icache, but the PALcode must scrub the block in the Bcache and memory. On Bcache
and Memory single bit errors on Dcache fills, the hardware scrubs the Dcache as long
as the error was on a target quadword, but the PALcode must scrub the Dcache for nontarget quadwords, and must in general scrub the block in the Bcache and memory.
The scrub consists of reading each quadword in the block, with at least one exclusive
access load/store to ensure the corrected data will be scrubbed in Bcache and memory.
The scrub itself causes a CRD to be flagged, which is cleared by the PALcode before
exiting to native mode.
; Sample code for scrubbing a single bit error.
;
; Since we only have the block address, and the hardware only corrects
;
target quadwords, we read each quadword.
; In order to ensure eviction to bcache and memory, a store
;
is needed to mark the block dirty. An exclusive access is
;
used to ensure we scrub in main memory. Virtual access is
;
used because of restrictions in use of hw_ld/hw_st lock
;
instructions.
; After the scrub, read the cbox chain again.
; The scrub will cause a crd, but will get cleared with a write
; to hw_int_clr.
;
; Current state:
;
r5
base of crd logout frame
;
hw_ldq/p r4, MCHK_CRD__C_ADDR(r5)
; get address back
bis
r31, r31, r31
bis
r31, r31, r31
bis
r31, r31, r31

21264/EV68A Hardware Reference Manual

PALcode Restrictions and Guidelines D–19

Restriction 40: Scrubbing a Single-Bit Error

hw_mtpr r31, EV6__DTB_IA
lda
r20, ^x3301(r31)
bis
r31, r31, r31
bis
r31, r31, r31

; (7,1L) flush dtb
; set WE, RE

hw_mtpr r31, <EV6__MM_STAT ! ^x80>
srl
r4, #13, r6
sll
r6, #EV6__DTB_PTE0__PFN__S, r6
bis
r6, r20, r6

; wait for retire
; shift byte offset
; shift into position
; produce pte

hw_mtpr r4, EV6__DTB_TAG0
hw_mtpr r4, EV6__DTB_TAG1
hw_mtpr r6, EV6__DTB_PTE0
hw_mtpr r6, EV6__DTB_PTE1

; (2&6,0L) write tag0
; (1&5,1L) write tag1
; (0&4,0L) write pte0
; (3&7,1L) write pte1

mb
bis
bis
bis

; quiet before we start
r31, r31, r31
r31, r31, r31
r31, r31, r31

ldq
ldq
ldq
ldq
ldq
ldq
ldq
mb

r6, ^x00(r4)
r6, ^x08(r4)
r6, ^x10(r4)
r6, ^x18(r4)
r6, ^x20(r4)
r6, ^x28(r4)
r6, ^x30(r4)

; re-read the bad block QW #0
; re-read the bad block QW #1
; re-read the bad block QW #2
; re-read the bad block QW #3
; re-read the bad block QW #4
; re-read the bad block QW #5
; re-read the bad block QW #6
; no other mem-ops till done

ldq_l
stq_c
mb
and

r6, ^x38(r4)
r6, ^x38(r4)

; re-read the bad block QW #7
; now store it to force scrub

r6, r31, r6

; consumer of above

beq
r6, sys__crd_scrub_done
br
r31, .-4
sys__crd_scrub_done:
bsr
r7, sys__cbox
bis
r31, r31, r31
hw_mtpr r31, EV6__DTB_IA
bis
r31, r31, r31
bis
r31, r31, r31
bis
r31, r31, r31

; these 2 lines......
; .....stop pre-fetching
; clean the cbox error chain

; (7,1L) flush dtb

hw_mtpr r31, <EV6__MM_STAT ! ^x80>
; wait for retire
bis
r31, #1, r7
; get a 1
sll
r7, #EV6__HW_INT_CLR__CR__S, r7 ; shift into position
hw_mtpr r7, EV6__HW_INT_CLR
; (4,0L) clear crd
lda
r7, EV6__DC_STAT_W1C_CRD(r31)
hw_mtpr r7, EV6__DC_STAT
bis
r31, r31 ,r31
bis
r31, r31 ,r31

; W1C bits
; (6,0L)

hw_mtpr r31, <EV6__MM_STAT ! ^x50>

; stall till they retire

D–20 PALcode Restrictions and Guidelines

21264/EV68A Hardware Reference Manual

Restriction 41: MTPR ITB_TAG, MTPR ITB_PTE Must be in the Same Fetch Block

D.37 Restriction 41: MTPR ITB_TAG, MTPR ITB_PTE Must be in the
Same Fetch Block
Write the ITB_TAG and ITB_PTE registers in the same fetch block. This avoids a
mispredict path write of invalid data to the ITB_TAG register.

D.38 Restriction 42: Updating VA_CTL, CC_CTL, or CC IPRs
When writing to the VA_CTL, CC_CTL, or CC IPRs, write the same value twice in distinct fetch blocks. This ensures that the instruction is retired before any mispredict from
a younger branch, DTB miss trap, or hw_ret_stall.

D.39 Restriction 43: No Trappable Instructions Along with
HW_MTPR
There cannot be any mispredictable/trappable instructions together with an HW_MTPR
in the current fetch block.

D.40 Restriction 44: Not Applicable to the 21264/EV68A
D.41 Restriction 45: No HW_JMP or JMP Intructions in PALcode
Do not include HW_JMP or JMP instructions in PALcode; use HW_RET instead.
HW_JMP always predicts in PALmode, and may mispredict to random cache blocks.
This may cause speculative code to begin executing in PALmode and may have unexpected side effects such as I/O stream references.
HW_RET always predicts in native mode, and when it mispredicts, it avoids speculative execution in PALmode.

21264/EV68A Hardware Reference Manual

PALcode Restrictions and Guidelines D–21

Restriction 46: Avoiding Livelocks in Speculative Load CRD Handlers

D.42 Restriction 46: Avoiding Livelocks in Speculative Load CRD
Handlers
Speculative load CRD handlers that release from the interrupt without scrubbing a
cache block could suffer from the following livelock condition:
1. An initial error on a speculative load forces a CRD interrupt.
2. The CRD releases without scrubbing the block. A speculative load in the shadow of
the hw_ret (or hw_ret_stall) touches a Dcache location that has the single-bit error,
forcing a CRD.
3. The CRD handler is entered again immediately.
4. Go to (2).
This problem can be avoided if all jumps in the CRD handler path for speculative loads
use the following sequence:
mb

; make sure hw_ret goes

ALIGN_FETCH_BLOCK <^x47FF041F>
mulq
p6, #1, p6
mulq
p6, #1, p6
hw_mtpr p6, <EV6__MM_STAT ! ^x44>
PVC_VIOLATE<43>
hw_ret_stall (p23)

; Hold up loads
; Hold up loads
; Hold up loads
; Ignore restriction 43
; Return

This sequence prevents speculative loads from issuing in the shadow of the
hw_ret_stall. Note that it is a violation of restriction 4 to have in the same fetch block a
MTPR that specifies scoreboard bit 2 (an explicit writer in the memory operation
group) and a HW_RET (an implicit reader in the memory operation group). Under normal circumstances, the intention would be for a HW_RET to wait until the MTPR
issues, and that can only be enforced by putting the two instructions in different fetch
blocks. In this case, the intention is for the HW_RET to issue before the MTPR. The
hardware does not enforce the scoreboarding when the two instructions are in the same
fetch block, and thus the HW_RET can issue and mispredict before any speculative
loads (which are held up by the MTPR) can issue.

D.43 Restriction 47: Cache Eviction for Single-Bit Cache Errors
A live lock can occur if issuing instructions out-of-order causes a floating-point store
instruction (with sberr) to replay trap.
A hardware mechanism exists that keeps track of replayed floating-point store instructions, and cancels the dirty register check. See Section D.5 for more details.
If the floating-point store instruction has an sberr and the CRD_HANDLER is entered/
exited before the instruction is replayed, the mechanism will lose track of the instruction. When the instruction is replayed, the dirty register check is not canceled, and a
replay trap occurs, causing the floating-point store instruction to continually replay the
trap until the sberr is evicted from cache. The sberr will not evict, because the floatingpoint store instruction is killed by the replay trap. Killed instructions are not scrubed by
the Error Recovery Machine, and CBOX_ERR[C_ADDR] may not contain the address
of the sberr. Because CBOX_ERR[C_ADDR] is not guaranteed, the CRD_HANDLER
might not evict the sberr.

D–22 PALcode Restrictions and Guidelines

21264/EV68A Hardware Reference Manual

Restriction 47: Cache Eviction for Single-Bit Cache Errors

If "CBOX_ERR[C_ADDR]" has not changed when the CRD_HANDLER is reentered, or "CBOX_ERR[C_STAT] == 0x0", all cache locations should to be evicted to
avoid the live lock described above.
; Sample code for evicting cache.
; This method loads a 64K block, then exits the CRD_HANDLER
; to check if the sberr has been evicted. If not it loads the next 64K block.
; In the sample code below,
;
sx is a shadow register
;
ldi is a macro that loads a 64-bit constant into the specified register
full_scrub:
hw_ldq/p

s5, 104(r31)

ldi
ldi
ldi
ldi

s1, ^x200
s2, ^x1C0
s3, ^x10000
s4, ^x2000000

addq
ble

s3,s5,s5
s5, s4, <.+4>

bis
r31, r31, s5
hw_stq/p
s5, 104(r31)
subq
s5, s2, s5
mb
.align
blbc
br
.align

4, NOP_OPCODE
r31, <.+4>
r31, <.-4>
4, NOP_OPCODE

; Loop dec value
; Start offset
; Block size (64K -> size of dcache)
; 2X bcache size

; Skip next instruction if ADDR
; .le. 2X bcache
; Set ADDR = 0x0
; Store ADDR for next pass thru
; Make sure no speculative loads
; happen in the CRD handler
;|
;|
;V
; Make sure no speculative loads
; happen in the CRD handler

next_reread:
; ******** four cache blocks
; Evict dcache by prefetching to all dcache indexes.
; use 'hw_ldl r31 xxxx' Normal Prefetch
; Do not use 'hw_ldq/p r31 xxx' Prefetch,
; Evict Next because this will always access the same set in dcache.
hw_ldl/p
r31,^x1C0(s5)
; Re-read the bad block QW #0
hw_ldl/p
r31,^x180(s5)
; Re-read the bad block QW #0
hw_ldl/p
r31,^x140(s5)
; Re-read the bad block QW #0
hw_ldl/p
r31,^x100(s5)
; Re-read the bad block QW #0
hw_ldl/p
hw_ldl/p
hw_ldl/p
hw_ldl/p
subq
subq
ble
br
bsr

r31,^xC0(s5)
r31,^x80(s5)
r31,^x40(s5)
r31,^x00(s5)

s5, s1, s5
s3, s1, s3
s3, <.+4>
r31, next_reread
s7, sys__cbox

21264/EV68A Hardware Reference Manual

; Re-read the bad block QW #0
; Re-read the bad block QW #0
; Re-read the bad block QW #0
; Re-read the bad block QW #0
; Decrement addr
; Decrement counter

; Read and clean cbox error ipr

PALcode Restrictions and Guidelines D–23

Restriction 48: MB Bracketing of Dcache Writes to Force Bad Data ECC and Force

D.44 Restriction 48: MB Bracketing of Dcache Writes to Force Bad
Data ECC and Force Bad Tag Parity
Writes to DC_CTL[F_BAD_DECC] and DC_CTL[DCDAT_ERR_EN] must be bracketed by MB instructions to quiesce the memory system. The Istream must also be quiesced before and during the sequence, as described in Section D.26.

D–24 PALcode Restrictions and Guidelines

21264/EV68A Hardware Reference Manual

E
21264/EV68A-to-Bcache Pin Interface
This appendix provides the pin interface between the21264/EV68A and Bcache
SSRAMs.

E.1 Forwarding Clock Pin Groupings
Table E–1 lists the correspondance between the clock signals for the 21264/EV68A and
Bcache (late-write non-bursting and dual-data rate) SSRAMs.
Table E–1

Bcache Forwarding Clock Pin Groupings

Pad and Pin

Input Clock

Output Clocks

BcData_H[71:64,7:0]

BcDataInClk_H[0]

BcDataOutClk_x[0]

BcCheck_H[8,0]

BcDataInClk_H[0]

BcDataOutClk_x[0]

BcData_H[79:72,15:8]

BcDataInClk_H[1]

BcDataOutClk_x[0]

BcCheck_H[9,1]

BcDataInClk_H[1]

BcDataOutClk_x[0]

BcData_H[87:80,23:16]

BcDataInClk_H[2]

BcDataOutClk_x[1]

BcCheck_H[10,2]

BcDataInClk_H[2]

BcDataOutClk_x[1]

BcData_H[95:88,31:24]

BcDataInClk_H[3]

BcDataOutClk_x[1]

BcCheck_H[11,3]

BcDataInClk_H[3]

BcDataOutClk_x[1]

BcData_H[103:96,39:32]

BcDataInClk_H[4]

BcDataOutClk_x[2]

BcCheck_H[12,4]

BcDataInClk_H[4]

BcDataOutClk_x[2]

BcData_H[111:104,47:40]

BcDataInClk_H[5]

BcDataOutClk_x[2]

BcCheck_H[13,5]

BcDataInClk_H[5]

BcDataOutClk_x[2]

BcData_H[119:112,55:48]

BcDataInClk_H[6]

BcDataOutClk_x[3]

BcCheck_H[14.6]

BcDataInClk_H[6]

BcDataOutClk_x[3]

BcData_H[127:120,63:56]

BcDataInClk_H[7]

BcDataOutClk_x[3]

BcCheck_H[15,7]

BcDataInClk_H[7]

BcDataOutClk_x[3]

BcTag_H[42:20]

BcTagInClk_H

BcTagOutClk_x

BcTagParity_H

BcTagInClk_H

BcTagOutClk_x

21264/EV68A Hardware Reference Manual

21264/EV68A-to-Bcache Pin Interface

E–1

Late-Write Non-Bursting SSRAMs

Table E–1

Bcache Forwarding Clock Pin Groupings (Continued)

Pad and Pin

Input Clock

Output Clocks

BcTagShared_H

BcTagInClk_H

BcTagOutClk_x

BcTagDirty_H

BcTagInClk_H

BcTagOutClk_x

BcTagValid_H

BcTagInClk_H

BcTagOutClk_x

E.2 Late-Write Non-Bursting SSRAMs
Table E–2 provides the data pin connections between late-write non-bursting SSRAMs
and the 21264/EV68A or the system board. Table E–3 provides the same information
for the tag pins.

Data Pin Usage
Table E–2 Late-Write Non-Bursting SSRAMs Data Pin Usage
21264/EV68A Signal Name or Board Connection

Late-Write SSRAM Data Pin Name

BcAdd_H[21:4]

SA_H[17:0]

BcDataOutClk_H[3:0]

CK_H

Set from board to 1/2 the 21264/EV68A core voltage CK_L
BcData_H[127:0]/BcCheck_H[15:0]

DQx

BcDataWr_L

SW_L

Unconnected

Tck_H

Unconnected

Tdo_H

Unconnected

Tms_H

Unconnected

Tdi_H

From board, pull down to VSS

G_L

From board, pull down to VSS

SBx_L

From board, pull down to VSS or BcDataOE_L

SS_L (Vendor dependent)

Tag Pin Usage
Unused Bcache tag pins should be pulled to ground through a 200-ohm resistor.
Table E–3 Late-Write Non-Bursting SSRAMs Tag Pin Usage

E–2

21264/EV68A Signal Name or Board Connection

Late-Write SSRAM Tag Pin Name

BcAdd_H[22:6]

SA_H[16:0]

BcTag_H[42:20]

DQx

BcTagOE_L or from board, pull down to VSS

SS_L (Vendor dependent)

BcTagWr_L

SW_L

From board, pull down toVSS

SBx_L

BcTagOutClk_H

CK_H

21264/EV68A-to-Bcache Pin Interface

21264/EV68A Hardware Reference Manual

Dual-Data Rate SSRAMs

Table E–3 Late-Write Non-Bursting SSRAMs Tag Pin Usage (Continued)
21264/EV68A Signal Name or Board Connection

Late-Write SSRAM Tag Pin Name

Set from board to 1/2 the 21264/EV68A core voltage CK_L
Set from board to 1/2 the 21264/EV68A core voltage VREF1_H
VREF2_H
Set from board (implementation dependent)

ZQ_H

BcTagValid_H

DQx

BcTagDirty_H

DQx

BcTagShared_H

DQx

Unconnected

TMS_H

Unconnected

TDI_H

Unconnected

TCK_H

Unconnected

TDC_H

E.3 Dual-Data Rate SSRAMs
Table E–4 provides the data pin connections between dual-data rate SSRAMs and the
21264/EV68A or the system board. Table E–5 provides the same information for the
tag pins.

Data Pin Usage
Table E–4 Dual-Data Rate SSRAM Data Pin Usage
21264/EV68A Signal Name or Board
Connection

Dual-Data Rate SSRAM Data Pin Name

BcAdd_H[21:4]

SA_H[17:0]

BcData_H[33:20]/
BcCheck_H[15:0]

DQx

BcLoad_L

LD_L (B1)

BcDataWr_L

R/W_L(B2)

From board, pulled up to VDD

LBO_L

From board, pulled down to VSS

Q_L

BcDataInClk_H

CQ_H

BcDataOutClk_H

CK_H

BcDataOutClk_L

CK_L

Set from board to 1/2 the 21264/EV68A core
voltage

VREF1_H
VREF2_H

Set from board (implementation-dependent)

ZQ_H

Unconnected or terminated

CQ_L

From board, pulled up toVDD

TCK_H

Unconnected

TDO_H

21264/EV68A Hardware Reference Manual

21264/EV68A-to-Bcache Pin Interface

E–3

Dual-Data Rate SSRAMs

Table E–4 Dual-Data Rate SSRAM Data Pin Usage (Continued)
21264/EV68A Signal Name or Board
Connection

Dual-Data Rate SSRAM Data Pin Name

From board, pulled up to VDD

TMS_H

From board, pulled up to VDD

TDI_H

Unconnected or pulled down to VSS

TRST_L

BcDataOE_L

OE_L (G_L)

From board, pulled down to VSS

SD/DD_L (B3)

Tag Pin Usage
Unused Bcache tag pins should be pulled to ground through a 200-ohm resister.
Table E–5 Dual-Data Rate SSRAM Tag Pin Usage

E–4

21264/EV68A Signal Name or Board
Connection

Dual-Data Rate SSRAM Tag Pin Name

BcAdd_H[23:6]

SA_H[17:0]

BcTag_H[33:20]

DQx

BcTagOE_L

LD_L (B1)

BcTagWr_L

R/W_L (B2)

From board, pulled up to VDD

LBO_L

From board, pulled down to VSS

Q_L
SA[19:18]

BcTagInClk_H

CQ_H

BcTagOutClk_H

CK_H

BcTagOutClk_L

CK_L

Set from board to 1/2 core voltage

VREF1_H
VREF2_H

Set from board (implementation-dependent)

ZQ_H

BcTagValid_H

DQx

BcTagDirty_H

DQx

BcTagShared_H

DQx

BcTagParity_H

DQx

Unconnected or terminated

CQ_L

From board, pulled up to VDD

TCK_H

Unconnected

TDO_H

From board, pulled up to VDD

TMS_H

From board, pulled up to VDD

TDI_H

21264/EV68A-to-Bcache Pin Interface

21264/EV68A Hardware Reference Manual

Dual-Data Rate SSRAMs

Table E–5 Dual-Data Rate SSRAM Tag Pin Usage (Continued)
21264/EV68A Signal Name or Board
Connection

Dual-Data Rate SSRAM Tag Pin Name

Unconnected

TRST_L

From board, pulled down to VSS

OE_L (G_L)

From board, pulled up to VDD

SD/DD_L (B3)

21264/EV68A Hardware Reference Manual

21264/EV68A-to-Bcache Pin Interface

E–5

Glossary
This glossary provides definitions for specific terms and acronyms associated with the
Alpha 21264/EV68A microprocessor and chips in general.

abort
The unit stops the operation it is performing, without saving status, to perform some
other operation.

address space number (ASN)
An optionally implemented register used to reduce the need for invalidation of cached
address translations for process-specific addresses when a context switch occurs. ASNs
are processor specific; the hardware makes no attempt to maintain coherency across
multiple processors.

address translation
The process of mapping addresses from one address space to another.

ALIGNED
A datum of size 2**N is stored in memory at a byte address that is a multiple of 2**N
(that is, one that has N low-order zeros).

ALU
Arithmetic logic unit.

ANSI
American National Standards Institute. An organization that develops and publishes
standards for the computer industry.

ASIC
Application-specific integrated circuit.

ASM
Address space match.

ASN
See address space number.

assert
To cause a signal to change to its logical true state.

AST
See asynchronous system trap.

21264/EV68A Hardware Reference Manual

asynchronous system trap (AST)
A software-simulated interrupt to a user-defined routine. ASTs enable a user process to
be notified asynchronously, with respect to that process, of the occurrence of a specific
event. If a user process has defined an AST routine for an event, the system interrupts
the process and executes the AST routine when that event occurs. When the AST routine exits, the system resumes execution of the process at the point where it was interrupted.

bandwidth
Bandwidth is often used to express the rate of data transfer in a bus or an I/O channel.

barrier transaction
A transaction on the external interface as a result of an MB (memory barrier) instruction.

Bcache
See second-level cache.

bidirectional
Flowing in two directions. The buses are bidirectional; they carry both input and output
signals.

BiSI
Built-in self-initialization.

BiST
Built-in self-test.

bit
Binary digit. The smallest unit of data in a binary notation system, designated as 0 or 1.

bit time
The total time that a signal conveys a single valid piece of information (specified in ns).
All data and commands are associated with a clock and the receiver’s latch on both the
rise and fall of the clock. Bit times are a multiple of the 21264/EV68A clocks. Systems
must produce a bit time identical to 21264/EV68A’s bit time. The bit time is one-half
the period of the forwarding clock.

BIU
Bus interface unit. See Cbox.

Block exchange
Memory feature that improves bus bandwidth by paralleling a cache victim write-back
with a cache miss fill.

board-level cache
See second-level cache.

Glossary–2

21264/EV68A Hardware Reference Manual

boot
Short for bootstrap. Loading an operating system into memory is called booting.

BSR
Boundary-scan register.

buffer
An internal memory area used for temporary storage of data records during input or
output operations.

bugcheck
A software condition, usually the response to software’s detection of an “internal inconsistency,” which results in the execution of the system bugcheck code.

bus
A group of signals that consists of many transmission lines or wires. It interconnects
computer system components to provide communications paths for addresses, data, and
control information.

byte
Eight contiguous bits starting on an addressable byte boundary. The bits are numbered
right to left, 0 through 7.

byte granularity
Memory systems are said to have byte granularity if adjacent bytes can be written concurrently and independently by different processes or processors.

cache
See cache memory.

cache block
The smallest unit of storage that can be allocated or manipulated in a cache. Also
known as a cache line.

cache coherence
Maintaining cache coherence requires that when a processor accesses data cached in
another processor, it must not receive incorrect data and when cached data is modified,
all other processors that access that data receive modified data. Schemes for maintaining consistency can be implemented in hardware or software. Also called cache consistency.

cache fill
An operation that loads an entire cache block by using multiple read cycles from main
memory.

cache flush
An operation that marks all cache blocks as invalid.

21264/EV68A Hardware Reference Manual

cache hit
The status returned when a logic unit probes a cache memory and finds a valid cache
entry at the probed address.

cache interference
The result of an operation that adversely affects the mechanisms and procedures used to
keep frequently used items in a cache. Such interference may cause frequently used
items to be removed from a cache or incur significant overhead operations to ensure
correct results. Either action hampers performance.

cache line
See cache block.

cache line buffer
A buffer used to store a block of cache memory.

cache memory
A small, high-speed memory placed between slower main memory and the processor. A
cache increases effective memory transfer rates and processor speed. It contains copies
of data recently used by the processor and fetches several bytes of data from memory in
anticipation that the processor will access the next sequential series of bytes. The
21264/EV68A microprocessor contains two onchip internal caches. See also writethrough cache and write-back cache.

cache miss
The status returned when cache memory is probed with no valid cache entry at the
probed address.

CALL_PAL instructions
Special instructions used to invoke PALcode.

Cbox
External cache and system interface unit. Controls the Bcache and the system ports.

central processing unit (CPU)
The unit of the computer that is responsible for interpreting and executing instructions.

CISC
Complex instruction set computing. An instruction set that consists of a large number
of complex instructions. Contrast with RISC.

clean
In the cache of a system bus node, refers to a cache line that is valid but has not been
written.

clock
A signal used to synchronize the circuits in a computer.

Glossary–4

21264/EV68A Hardware Reference Manual

clock offset (or clkoffset)
The delay intentionally added to the forwarded clock to meet the setup and hold
requirements at the Receive Flop.

CMOS
Complementary metal-oxide semiconductor. A silicon device formed by a process that
combines PMOS and NMOS semiconductor material.

conditional branch instructions
Instructions that test a register for positive/negative or for zero/nonzero. They can also
test integer registers for even/odd.

control and status register (CSR)
A device or controller register that resides in the processor’s I/O space. The CSR initiates device activity and records its status.

CPI
Cycles per instruction.

CPU
See central processing unit.

CSR
See control and status register.

cycle
One clock interval.

data bus
A group of wires that carry data.

Dcache
Data cache. A cache reserved for storage of data. The Dcache does not contain instructions.

DDR
Dual-data rate. A dual-data rate SSRAM can provide data on both the rising and falling
edges of the clock signal.

denormal
An IEEE floating-point bit pattern that represents a number whose magnitude lies
between zero and the smallest finite number.

DIP
Dual inline package.

21264/EV68A Hardware Reference Manual

direct-mapping cache
A cache organization in which only one address comparison is needed to locate any
data in the cache, because any block of main memory data can be placed in only one
possible position in the cache.

direct memory access (DMA)
Access to memory by an I/O device that does not require processor intervention.

dirty
One status item for a cache block. The cache block is valid and has been written so that
it may differ from the copy in system main memory.

dirty victim
Used in reference to a cache block in the cache of a system bus node. The cache block
is valid but is about to be replaced due to a cache block resource conflict. The data must
therefore be written to memory.

DMA
See direct memory access.

DRAM
Dynamic random-access memory. Read/write memory that must be refreshed (read
from or written to) periodically to maintain the storage of information.

DTB
Data translation buffer. Also defined as Dstream translation buffer.

DTL
Diode-transistor logic.

dual issue
Two instructions are issued, in parallel, during the same microprocessor cycle. The
instructions use different resources and so do not conflict.

ECC
Error correction code. Code and algorithms used by logic to facilitate error detection
and correction. See also ECC error.

ECC error
An error detected by ECC logic, to indicate that data (or the protected “entity”) has
been corrupted. The error may be correctable (soft error) or uncorrectable (hard error).

ECL
Emitter-coupled logic.

EEPROM
Electrically erasable programmable read-only memory. A memory device that can be
byte-erased, written to, and read from. Contrast with FEPROM.

Glossary–6

21264/EV68A Hardware Reference Manual

external cache
See second-level cache.

FEPROM
Flash-erasable programmable read-only memory. FEPROMs can be bank- or bulkerased. Contrast with EEPROM.

FET
Field-effect transistor.

FEU
The unit within the 21264/EV68A microprocessor that performs floating-point calculations.

firmware
Machine instructions stored in nonvolatile memory.

floating point
A number system in which the position of the radix point is indicated by the exponent
part and another part represents the significant digits or fractional part.

flush
See cache flush.

forwarded clock
A single-ended differential signal that is aligned with its associated fields. The forwarded clock is sourced and aligned by the sender with a period that is two times the bit
time. Forwarded clocks must be 50% duty cycle clocks whose rising and falling edges
are aligned with the changing edge of the data.

FPGA
Field-programmable gate array.

FPLA
Field-programmable logic array.

FQ
Floating-point issue queue.

framing clock
The framing clock defines the start of a transmission either from the system to the
21264/EV68A or from the 21264/EV68A to the system. The framing clock is a powerof-2 multiple of the 21264/EV68A GCLK frequency, and is usually the system clock.
The framing clock and the input oscillator can have the same frequency. The
add_frame_select IPR sets that ratio of bit times to framing clock. The frame clock
could have a period that is four times the bit time with a add_frame_select of 2X.
Transfers begin on the rising and falling edge of the frame clock. This is useful for systems that have system clocks with a period too small to perform the synchronous reset
21264/EV68A Hardware Reference Manual

of the clock forward logic. Additionally, the framing clock can have a period that is
less than, equal to, or greater than the time it takes to send a full four cycle command/
address.

GCLK
Global clock within the 21264/EV68A.

granularity
A characteristic of storage systems that defines the amount of data that can be read and/
or written with a single instruction, or read and/or written independently.

hardware interrupt request (HIR)
An interrupt generated by a peripheral device.

high-impedance state
An electrical state of high resistance to current flow, which makes the device appear not
physically connected to the circuit.

hit
See cache hit.

Icache
Instruction cache. A cache reserved for storage of instructions. One of the three areas of
primary cache (located on the 21264/EV68A) used to store instructions. The Icache
contains 8KB of memory space. It is a direct-mapped cache. Icache blocks, or lines,
contain 32 bytes of instruction stream data with associated tag as well as a 6-bit ASM
field and an 8-bit branch history field per block. Icache does not contain hardware for
maintaining cache coherency with memory and is unaffected by the invalidate bus.

IDU
A logic unit within the 21264/EV68A microprocessor that fetches, decodes, and issues
instructions. It also controls the microprocessor pipeline.

IEEE Standard 754
A set of formats and operations that apply to floating-point numbers. The formats cover
32-, 64-, and 80-bit operand sizes.

IEEE Standard 1149.1
A standard for the Test Access Port and Boundary Scan Architecture used in boardlevel manufacturing test procedures.

Inf
Infinity.

INT nn
The term INTnn, where nn is one of 2, 4, 8, 16, 32, or 64, refers to a data field size of nn
contiguous NATURALLY ALIGNED bytes. For example, INT4 refers to a NATURALLY ALIGNED longword.

Glossary–8

21264/EV68A Hardware Reference Manual

interface reset
A synchronously received reset signal that is used to preset and start the clock forwarding circuitry. During this reset, all forwarded clocks are stopped and the presettable
count values are applied to the counters; then, some number of cycles later, the clocks
are enabled and are free running.

Internal processor register (IPR)
Special registers that are used to configure options or report status.

IOWB
I/O write buffer.

IPGA
Interstitial pin grid array.

IQ
Integer issue queue.

ITB
Instruction translation buffer.

JFET
Junction field-effect transistor.

latency
The amount of time it takes the system to respond to an event.

LCC
Leadless chip carrier.

LFSR
Linear feedback shift register.

load/store architecture
A characteristic of a machine architecture where data items are first loaded into a processor register, operated on, and then stored back to memory. No operations on memory
other than load and store are provided by the instruction set.

longword (LW)
Four contiguous bytes starting on an arbitrary byte boundary. The bits are numbered
from right to left, 0 through 31.

LQ
Load queue.

LSB
Least significant bit.

21264/EV68A Hardware Reference Manual

machine check
An operating system action triggered by certain system hardware-detected errors that
can be fatal to system operation. Once triggered, machine check handler software analyzes the error.

MAF
Miss address file.

main memory
The large memory, external to the microprocessor, used for holding most instruction
code and data. Usually built from cost-effective DRAM memory chips. May be used in
connection with the microprocessor’s internal caches and an external cache.

masked write
A write cycle that only updates a subset of a nominal data block.

MBO
See must be one.

Mbox
This section of the processor unit performs address translation, interfaces to the
Dcache, and performs several other functions.

MBZ
See must be zero.

MESI protocol
A cache consistency protocol with full support for multiprocessing. The MESI protocol
consists of four states that define whether a block is modified (M), exclusive (E), shared
(S), or invalid (I).

MIPS
Millions of instructions per second.

miss
See cache miss.

module
A board on which logic devices (such as transistors, resistors, and memory chips) are
mounted and connected to perform a specific system function.

module-level cache
See second-level cache.

MOS
Metal-oxide semiconductor.

MOSFET
Metal-oxide semiconductor field-effect transistor.
Glossary–10

21264/EV68A Hardware Reference Manual

MSI
Medium-scale integration.

multiprocessing
A processing method that replicates the sequential computer and interconnects the collection so that each processor can execute the same or a different program at the same
time.

must be one (MBO)
A field that must be supplied as one.

must be zero (MBZ)
A field that is reserved and must be supplied as zero. If examined, it must be assumed to
be UNDEFINED.

NaN
Not-a-Number. An IEEE floating-point bit pattern that represents something other than
a number. This comes in two forms: signaling NaNs (for Alpha, those with an initial
fraction bit of 0) and quiet NaNs (for Alpha, those with an initial fraction bit of 1).

NATURALLY ALIGNED
See ALIGNED.

NATURALLY ALIGNED data
Data stored in memory such that the address of the data is evenly divisible by the size of
the data in bytes. For example, an ALIGNED longword is stored such that the address
of the longword is evenly divisible by 4.

NMOS
N-type metal-oxide semiconductor.

NVRAM
Nonvolatile random-access memory.

OBL
Observability linear feedback shift register.

octaword
Sixteen contiguous bytes starting on an arbitrary byte boundary. The bits are numbered
from right to left, 0 through 127.

OpenVMS Alpha operating system
The version of the open VMS operating system for Alpha platforms.

operand
The data or register upon which an operation is performed.

21264/EV68A Hardware Reference Manual

output mux counter
Counter used to select the output mux that drives address and data. It is reset with the
Interface Reset and incremented by a copy of the locally generated forwarded clock.

PAL
Privileged architecture library. See also PALcode. See also Programmable array logic
(hardware). A device that can be programmed by a process that blows individual fuses
to create a circuit.

PALcode
Alpha privileged architecture library code, written to support Alpha microprocessors.
PALcode implements architecturally defined behavior.

PALmode
A special environment for running PALcode routines.

parameter
A variable that is given a specific value that is passed to a program before execution.

parity
A method for checking the accuracy of data by calculating the sum of the number of
ones in a piece of binary data. Even parity requires the correct sum to be an even number, odd parity requires the correct sum to be an odd number.

PGA
Pin grid array.

pipeline
A CPU design technique whereby multiple instructions are simultaneously overlapped
in execution.

PLA
Programmable logic array.

PLCC
Plastic leadless chip carrier or plastic-leaded chip carrier.

PLD
Programmable logic device.

PLL
Phase-locked loop.

PMOS
P-type metal-oxide semiconductor.

PQ
Probe queue.

Glossary–12

21264/EV68A Hardware Reference Manual

PQFP
Plastic quad flat pack.

primary cache
The cache that is the fastest and closest to the processor. The first-level caches, located
on the CPU chip, composed of the Dcache and Icache.

program counter
That portion of the CPU that contains the virtual address of the next instruction to be
executed. Most current CPUs implement the program counter (PC) as a register. This
register may be visible to the programmer through the instruction set.

PROM
Programmable read-only memory.

pull-down resistor
A resistor placed between a signal line and a negative voltage.

pull-up resistor
A resistor placed between a signal line to a positive voltage.

QNaN
Quiet Nan. See NaN.

quad issue
Four instructions are issued, in parallel, during the same microprocessor cycle. The
instructions use different resources and so do not conflict.

quadword
Eight contiguous bytes starting on an arbitrary byte boundary. The bits are numbered
from right to left, 0 through 63.

RAM
Random-access memory.

RAS
Row address select.

RAW
Read-after-write.

READ_BLOCK
A transaction where the 21264/EV68A requests that an external logic unit fetch read
data.

read data wrapping
System feature that reduces apparent memory latency by allowing read data cycles to
differ the usual low-to-high sequence. Requires cooperation between the 21264/EV68A
and external hardware.
21264/EV68A Hardware Reference Manual

read stream buffers
Arrangement whereby each memory module independently prefetches DRAM data
prior to an actual read request for that data. Reduces average memory latency while
improving total memory bandwidth.

receive counter
Counter used to enable the receive flops. It is clocked by the incoming forwarded clock
and reset by the Interface Reset.

receive mux counter
The receive mux counter is preset to a selectable starting point and incremented by the
locally generated forward clock.

reliability
The probability a device or system will not fail to perform its intended functions during
a specified time interval when operated under stated conditions.

reset
An action that causes a logic unit to interrupt the task it is performing and go to its initialized state.

RISC
Reduced instruction set computing. A computer with an instruction set that is paired
down and reduced in complexity so that most can be performed in a single processor
cycle. High-level compilers synthesize the more complex, least frequently used instructions by breaking them down into simpler instructions. This approach allows the RISC
architecture to implement a small, hardware-assisted instruction set, thus eliminating
the need for microcode.

ROM
Read-only memory.

RTL
Register-transfer logic.

SAM
Serial access memory.

SBO
Should be one.

SBZ
Should be zero.

scheduling
The process of ordering instruction execution to obtain optimum performance.
Glossary–14

21264/EV68A Hardware Reference Manual

SDRAM
Synchronous dynamic random-access memory.

second-level cache
A cache memory provided outside of the microprocessor chip, usually located on the
same module. Also called board-level, external, or module-level cache.

set-associative
A form of cache organization in which the location of a data block in main memory
constrains, but does not completely determine, its location in the cache. Set-associative
organization is a compromise between direct-mapped organization, in which data from
a given address in main memory has only one possible cache location, and fully associative organization, in which data from anywhere in main memory can be put anywhere in the cache. An “n-way set-associative” cache allows data from a given address
in main memory to be cached in any of n locations.

SIMM
Single inline memory module.

SIP
Single inline package.

SIPP
Single inline pin package.

SMD
Surface mount device.

SNaN
Signaling NaN. See Nan.

SRAM
See SSRAM.

SROM
Serial read-only memory.

SSI
Small-scale integration.

SSRAM
Synchronous static random-access memory.

stack
An area of memory set aside for temporary data storage or for procedure and interrupt
service linkages. A stack uses the last-in/first-out concept. As items are added to
(pushed on) the stack, the stack pointer decrements. As items are retrieved from
(popped off) the stack, the stack pointer increments.

21264/EV68A Hardware Reference Manual

STRAM
Self-timed random-access memory.

superpipelined
Describes a pipelined machine that has a larger number of pipe stages and more complex scheduling and control. See also pipeline.

superscalar
Describes a machine architecture that allows multiple independent instructions to be
issued in parallel during a given clock cycle.

system clock
The primary skew controlled clock used throughout the interface components to clock
transfer between ASICs, main memory, and I/O bridges.

tag
The part of a cache block that holds the address information used to determine if a
memory operation is a hit or a miss on that cache block.

target clock
Skew controlled clock that receives the output of the RECEIVE MUX.

TB
Translation buffer.

tristate
Refers to a bused line that has three states: high, low, and high-impedance.

TTL
Transistor-transistor logic.

UART
Universal asynchronous receiver-transmitter.

UNALIGNED
A datum of size 2**N stored at a byte address that is not a multiple of 2**N.

unconditional branch instructions
Instructions that change the flow of program control without regard to any condition.
Contrast with conditional branch instructions.

UNDEFINED
An operation that may halt the processor or cause it to lose information. Only privileged
software (that is, software running in kernel mode) can trigger an UNDEFINED operation. (This meaning only applies when the word is written in all upper case.)

Glossary–16

21264/EV68A Hardware Reference Manual

UNPREDICTABLE
Results or occurrences that do not disrupt the basic operation of the processor; the processor continues to execute instructions in its normal manner. Privileged or unprivileged software can trigger UNPREDICTABLE results or occurrences. (This meaning
only applies when the word is written in all upper case.)

UVPROM
Ultraviolet (erasable) programmable read-only memory.

VAF
See victim address file.

valid
Allocated. Valid cache blocks have been loaded with data and may return cache hits
when accessed.

VDF
See victim data file.

VHSIC
Very-high-speed integrated circuit.

victim
Used in reference to a cache block in the cache of a system bus node. The cache block
is valid but is about to be replaced due to a cache block resource conflict.

victim address file
The victim address file and the victim data file, together, form an 8-entry buffer used to
hold information for transactions to the Bcache and main memory.

victim data file
The victim address file and the victim data file, together, form an 8-entry buffer used to
hold information for transactions to the Bcache and main memory.

virtual cache
A cache that is addressed with virtual addresses. The tag of the cache is a virtual
address. This process allows direct addressing of the cache without having to go
through the translation buffer making cache hit times faster.

VLSI
Very-large-scale integration.

VPC
Virtual program counter.

VRAM
Video random-access memory.

21264/EV68A Hardware Reference Manual

WAR
Write-after-read.

word
Two contiguous bytes (16 bits) starting on an arbitrary byte boundary. The bits are numbered from right to left, 0 through 15.

write-back
A cache management technique in which write operation data is written into cache but
is not written into main memory in the same operation. This may result in temporary
differences between cache data and main memory data. Some logic unit must maintain
coherency between cache and main memory.

write-back cache
Copies are kept of any data in the region; read and write operations may use the copies,
and write operations use additional state to determine whether there are other copies to
invalidate or update.

WRITE_BLOCK
A transaction where the 21264/EV68A requests that an external logic unit process write
data.

write data wrapping
System feature that reduces apparent memory latency by allowing write data cycles to
differ the usual low-to-high sequence. Requires cooperation between the 21264/EV68A
and external hardware.

write-through cache
A cache management technique in which a write operation to cache also causes the
same data to be written in main memory during the same operation. Copies are kept of
any data in a region; read operations may use the copies, but write operations update the
actual data location and either update or invalidate all copies.

Glossary–18

21264/EV68A Hardware Reference Manual

Index
Numerics
21264/EV68A, features of, 1–3
32_BYTE_IO Cbox CSR
defined, 5–34

A
Abbreviations, xix
binary multiples, xix
register access, xix
AC characteristics, 9–6
Address conventions, xx
Aggregate mode, 6–18
Aligned convention, xx
Alpha instruction summary, A–1
AMASK instruction values , 2–37
ARITH synchronous trap, 6–14

B
B_DA_OD pin type, 3–3, 9–2
values for, 9–4
B_DA_PP pin type , 3–3, 9–2
values for, 9–4
BC_BANK_ENABLE Cbox CSR, 4–51, 5–39,
7–13
BC_BPHASE_LD_VECTOR Cbox CSR, 4–45
defined, 5–38
BC_BURST_MODE_ENABLE Cbox CSR, 4–51
defined, 5–35
BC_CLEAN_VICTIM Cbox CSR, 4–23
defined, 5–34
BC_CLK_DELAY Cbox CSR, 4–45
defined, 5–35
BC_CLK_LD_VECTOR Cbox CSR, 4–45
defined, 5–38
BC_CLKFWD_ENABLE Cbox CSR, 4–47
defined, 5–36

21264/EV68A Hardware Reference Manual

BC_CLOCK_OUT Cbox CSR, 4–45
BC_CPU_CLK_DELAY Cbox CSR, 4–44, 4–45
defined, 5–38
BC_CPU_LATE_WRITE_NUM Cbox CSR
defined, 5–35
BC_DDM_FALL_EN Cbox CSR, 4–47
defined, 5–36
BC_DDM_RISE_EN Cbox CSR, 4–47
defined, 5–36
BC_DDMF_ENABLE Cbox CSR, 4–47
defined, 5–35
BC_DDMR_ENABLE Cbox CSR, 4–47
defined, 5–35
BC_ENABLE Cbox CSR, 4–51, 5–39, 7–12
BC_FDBK_EN Cbox CSR, 4–46
defined, 5–38
BC_FRM_CLK Cbox CSR, 4–47
defined, 5–35
BC_LAT_DATA_PATTERN Cbox CSR, 4–48
defined, 5–35
BC_LAT_TAG_PATTERN Cbox CSR, 4–48
defined, 5–35
BC_LATE_WRITE_NUM Cbox CSR, 4–49
defined, 5–35
BC_LATE_WRITE_UPPER Cbox CSR
defined, 5–35
BC_PENTIUM_MODE Cbox CSR, 4–51
defined, 5–35
BC_PERR error status in C_STAT, 5–42
BC_RCV_MUX_CNT_PRESET Cbox CSR
defined, 5–36
BC_RCV_MUX_PRESET_CNT Cbox CSR, 4–48
BC_RD_RD_BUBBLE Cbox CSR
defined, 5–34
BC_RD_WR_BUBBLES Cbox CSR, 4–49
defined, 5–34
BC_RDVICTIM Cbox CSR, 4–23, 4–26
defined, 5–34
BC_SIZE Cbox CSR, 4–51, 5–39, 7–12
Index–1

BC_SJ_BANK_ENABLE Cbox CSR
defined, 5–34
BC_TAG_DDM_FALL_EN Cbox CSR, 4–47
defined, 5–36
BC_TAG_DDM_RISE_EN Cbox CSR, 4–47
defined, 5–36
BC_WR_RD_BUBBLES Cbox CSR, 4–49
defined, 5–35
BC_WR_WR_BUBBLE Cbox CSR, 4–53
defined, 5–34
BC_WRT_STS Cbox CSR, 5–39, 7–13
Bcache
banking, 4–53
bubbles on the data bus, 4–49
clocking, 4–44
control pins, 4–51
data read transactions, 4–47
data single-bit correctable ECC error, 8–5
data single-bit correctable ECC error on a probe,
8–9
data write transactions , 4–48
error case summary for, 8–10
filling Dcache error, 8–6
filling Icache error, 8–5
forwarding clock pin groupings, E–1
maximum clock ratio, 4–42
port, 4–42
port pins, 4–43
programming the size of, 4–51
setting clock period, 4–45
structure of, 4–7
tag parity errors, 8–5
tag read transactions, 4–47
victim read during an ECB instruction error,
8–7
victim read during Dcache/Bcache miss error,
8–7
victim read error, 8–7
BcAdd_H signal pins, 3–3, 4–43
characteristics, 4–51
BcCheck_H signal pins, 3–3, 4–43
BcData_H signal pins, 3–3, 4–43
BcDataInClk_H signal pins, 3–3, 4–43
using, 4–53
BcDataOE_L signal pin, 3–3, 4–43
BcDataOutClk_x signal pins, 3–4, 4–43
BcDataWr_L signal pin, 3–4, 4–44

BcTagParity_H signal pin, 3–4, 4–44
BcTagShared_H signal pin, 3–4, 4–44
BcTagValid_H signal pin, 3–4, 4–44
BcTagWr_L signal pin, 3–4, 4–44
BcVref signal pin, 3–4, 4–44
Bidirectional differential amplifier receiver open-drain. See B_DA_OD
Bidirectional differential amplifier receiver push-pull. See B_DA_PP
Binary multiple abbreviations, xix
BiST. See Built-in self-test
Bit notation conventions, xx
Bounder-scan register, B–1
Branch history table, initialized by BiST, 7–12
Branch mispredication, pipeline abort delay from,
2–16
Branch predictor, 2–3
BSDL description of the boundary-scan register,
B–1
Built-in self-test, 11–5
load, 7–6

C
C_ADDR Cbox read register field, 5–42
C_DATA Cbox data register, 5–33
at power-on reset state, 7–16
C_SHFT Cbox shift register, 5–33
at power-on reset state, 7–16
C_STAT Cbox read register field, 5–42
C_STS Cbox read register field, 5–42
C_SYNDROME_0 Cbox read register field, 5–41
C_SYNDROME_1 Cbox read register field, 5–41
Cache block states, 4–9
response to 21264/EV68A commands, 4–10
transitions, 4–10
Cache coherency, 4–8
CALL_PAL entry points, 6–12
Caution convention, xx

BcLoad_L signal pin, 3–4, 4–44
BcTag_H signal pins, 3–4, 4–44
BcTagDirty_H signal pin, 3–4, 4–44
BcTagInClk_H signal pin, 3–4, 4–44
using, 4–53
BcTagOE_L signal pin, 3–4, 4–44
BcTagOutClk_x signal pins, 3–4, 4–44

Index–2

21264/EV68A Hardware Reference Manual

Cbox
data register C_DATA, 5–33
described, 2–11, 4–3
duplicate Dcache tag array, 2–11
duplicate Dcache tag array with, 4–13
HW_MTPR and HW_MFPR to CSR, D–15
I/O write buffer, 2–11
internal processor registers, 5–3
probe queue, 2–11
read register, 5–41
shift register C_SHFT, 5–33
victim address file, 2–11
WRITE_MANY chain, 5–38
WRITE_MANY chain example, 5–39
WRITE_ONCE chain, 5–33
CC cycle counter register, 5–3
at power-on reset state, 7–15
CC_CTL cycle counter control register, 5–3
at power-on reset state, 7–15
CFR_EV6CLK_DELAY Cbox CSR, defined, 5–38

Conventions, xix
abbreviations, xix
address, xx
aligned, xx
bit notation, xx
caution, xx
data units, xxi
do not care, xxi
external, xxi
field notation, xxi
note, xxi
numbering, xxi
ranges and extents, xxi
register figures, xxi
signal names, xxi
unaligned, xx
X, xxi
CTAG, 4–13

CFR_FRMCLK_DELAY Cbox CSR, defined, 5–38

Data cache. See Dcache

CFR_GCLK_DELAY Cbox CSR, defined, 5–38

Data merging
load instructions in I/O address space, 2–28
store instructions in I/O address space, 2–29
Data transfer commands, system, 4–28

ChangeToDirtyFail, SysDc command, 4–10, 4–11,
4–12
ChangeToDirtySuccess, SysDc command, 4–10,
4–11, 4–12
Choice predictor, 2–5
ChxToDirty, 21264/EV68A command, 4–12
CLAMP public instruction, B–1
Clean cache block state, 4–9
Clean/Shared cache block state, 4–10
CleanToDirty, 21264/EV68A command, 4–22,
4–40
system probes, with, 4–41
CleanVictimBlk, 21264/EV68A command, 4–22,
4–39
ClkFwdRst_H signal pin, 3–4, 4–30
with system initialization, 7–7
ClkIn_x signal pins, 3–4
Clock forwarding, 7–4
CLR_MAP clear virtual-to-physical map register,
5–21
at power-on reset state, 7–15
CMOV instruction, special cases of, 2–26

Data types
floating point support, 1–2
integer supported, 1–2
supported, 1–1
Data units convention, xxi
Data wrap, 4–36
double-pumped, 4–38
interleaved, 4–37
DATA_VALID_DLY Cbox CSR, defined, 5–38
dc
characteristics of, 9–2
input pin capacitance defined, 9–2
test load defined, 9–2
DC_CTL Dcache control register, 5–30
at power-on reset state, 7–16
error correction and, 8–2
DC_PERR error status in C_STAT, 5–42
DC_STAT Dcache status register, 5–31
at power-on reset state, 7–16

COLD reset machine state, 7–17
Commands
21264/EV68A to system, 4–19
system to 21264/EV68A, 4–26
when to NXM , 4–38

21264/EV68A Hardware Reference Manual

Index–3

Dcache
described, 2–12
duplicate tag parity errors, 8–4
duplicate tags with, 4–13
error case summary for, 8–10
fill from Bcache error, 8–6
fill from memory errors , 8–8
initialized by BiST, 7–12
pipelined, 2–16
single-bit correctable ECC error, 8–3
store second error, 8–4
tag parity errors, 8–2
victim extracts, 8–4
Dcache data single-bit correctable ECC errors, 8–3
Dcache tag, initialized by BiST, 7–12
DCOK_H signal pin, 3–4
power-on reset flow, 7–1
DCVIC_THRESHOLD Cbox CSR, defined, 5–34
DFAULT fault, 6–13
Differential 21264/EV68A clocks, 7–19
Differential reference clocks, 7–19
Dirty cache block state, 4–10
Dirty/Shared cache block state, 4–10
Do not care convention, xxi
Double-bit fill errors, 8–9
DOWN1 reset machine state, 7–18
DOWN2 reset machine state, 7–19
DOWN3 reset machine state, 7–19
Dstream translation buffer, 2–13
See also DTB
DSTREAM_BC_DBL error status in C_STAT,
5–42

DTB_ASN0 address space number register 0
at power-on reset state, 7–16
DTB_ASN0 address space number registers 0, 5–28
DTB_ASN1 address space number register 1, 5–28
at power-on reset state, 7–16
DTB_IA invalidate-all process register, 5–27
at power-on reset state, 7–15
DTB_IAP invalidate-all (ASM=0) process register,
5–27
at power-on reset state, 7–15
DTB_IS0 invalidate single (array 0) register, 5–27
at power-on reset state, 7–16
DTB_IS1 invalidate single (array 1) register, 5–27
at power-on reset state, 7–16
DTB_PTE0 array write 0 register
at power-on reset state, 7–15
MTPR to, D–12
DTB_PTE0 array write register 0, 5–26
DTB_PTE1 array write 1 register, 5–26
at power-on reset state, 7–15
MTPR to, D–12
DTB_TAG0 array write 0 register, 5–25
at power-on reset state, 7–15
MTPR to, D–12
DTB_TAG1 array write 1 register, 5–25
at power-on reset state, 7–15
MTPR to, D–12
DTBM_DOUBLE_3 fault, 6–13
DTBM_DOUBLE_4 fault, 6–13
DTBM_SINGLE fault, 6–13
Dual-data rate SSRAM pin assignments, E–3
DUP_TAG_ENABLE Cbox CSR, defined, 5–34

DSTREAM_BC_ERR error status in C_STAT,
5–42

Duplicate Dcache tag array, 2–11

DSTREAM_DC_ERR error status in C_STAT,
5–42

Duplicate tag array, Cbox copy. See CTAG

DSTREAM_MEM_DBL error status in C_STAT,
5–42
DSTREAM_MEM_ERR error status in C_STAT,
5–42
DTAG. See Duplicate Dcache tag array
DTB entries, writing multiple in same PAL flow ,
D–19
DTB fill, 6–14
DTB, pipeline abort delay with, 2–16
DTB_ALTMODE alternate processor mode register,
5–26
at power-on reset state, 7–15

Index–4

Duplicate Dcache, initialized by BiST, 7–12
Duplicate tag stores, Bcache, 4–7

E
Ebox
cycle counter control register CC_CTL, 5–3
cycle counter register CC, 5–3
described, 2–8
executed in pipeline, 2–16
internal processor registers, 5–1
slotting, 2–18
subclusters, 2–18
virtual address control register VA_CTL, 5–4
virtual address format register VA_FORM, 5–5
virtual address register, 5–4
ECB instruction, external interface reference, 4–5

21264/EV68A Hardware Reference Manual

ECC
64-bit data and check bit code, 8–2
Dcache data single-bit correctable errors, 8–3
for system data bus, 8–2
memory/system port single-bit correctable
errors, 8–7
store instructions, 8–4
ENABLE_EVICT Cbox CSR, 4–23, 5–39
ENABLE_PROBE_CHECK Cbox CSR, 8–2
defined, 5–35
ENABLE_STC_COMMAND Cbox CSR, defined,
5–35

FetchBlkSpec, 21264/EV68A command, 4–22,
4–39
Field notation convention, xxi
Floating-point arithmetic trap, pipeline abort delay
with, 2–16
Floating-point control register, 2–35
PALcode emulation of, 6–11
Floating-point execution unit. See Fbox

Error case summary, 8–10

Floating-point instructions
IEEE, A–9
independent, A–11
VAX, A–11
Floating-point issue queue, 2–7

Error correction code. See ECC

Forwarding clock pin groupings, E–1

Error detection mechanisms, 8–1

FPCR. See Floating-point control register

EV6Clk_x signal pins, 3–4

FQ. See Floating-point issue queue

Evict, 21264/EV68A command, 4–13, 4–22, 4–39

FrameClk_x signal pins, 3–5, 4–30

Energy star certification, 7–9

EVICT_ENABLE Cbox CSR, 7–13

EXC_ADDR exception address register, 5–8
after fault reset, 7–8
at power-on reset state, 7–15
EXC_SUM exception summary register, 5–13
at power-on reset state, 7–15
Exception and interrupt logic, 2–8

Global predictor, 2–4

Exception condition summary, A–15

Heat sink center temperature, 10–1

External cache and system interface unit. See Cbox

Heat sink specifications, 10–3

External convention, xxi

HW_INT_CLR hardware interrupt clear register,
5–12
at power-on reset state, 7–15
updating, D–18
HW_LD PALcode instruction, 6–3, A–9, D–18

External interface initialization, 7–14
EXTEST public instruction, B–1

GCLK, 7–19

HW_MFPR PALcode instruction, 6–6, A–9

F31

HW_MTPR PALcode instruction, 6–6, A–9

load instructions with, 2–23
retire instructions with, 2–22
Fast data disable mode, 4–33

HW_REI PALcode instruction, A–9

Fast data mode, 4–30, 4–31

HW_ST PALcode instruction, 6–4, A–9

FAST_MODE_DISABLE Cbox CSR, 4–30
defined, 5–34
Fault reset flow, 7–8
Fault reset sequence of operations , 7–9
FAULT_RESET reset machine state, 7–18
Fbox
described, 2–10
executed in pipeline, 2–16
FEN fault, 6–13

HW_RET PALcode instruction, 6–5

I
I/O address space
instruction data merging, 2–29
load instruction data merging, 2–28
load instructions with, 2–27
store instructions with, 2–29
I/O write buffer, 2–11
defined, 2–32

FetchBlk, 21264/EV68A command, 4–22, 4–39
system probes, with, 4–41

21264/EV68A Hardware Reference Manual

Index–5

I_CTL Ibox control register, 5–15
after fault reset, 7–8
after warm reset, 7–11
at power-on reset state, 7–15
PALshadow registers, 6–11
through sleep mode, 7–10
VA_48 field update, D–17
I_DA pin type, 3–3, 9–2
values for, 9–3
I_DA_CLK pin type, 3–3, 9–2
values for, 9–3
I_DC_POWER pin type, 9–2
I_DC_REF pin type, 3–3, 9–2
values for, 9–3
I_STAT Ibox status register, 5–18
at power-on reset state, 7–15
IACV fault, 6–13
Ibox
branch predictor, 2–3
clear virtual-to-physical map register
CLR_MAP, 5–21
exception address register EXC_ADDR, 5–8
exception and interrupt logic, 2–8
exception summary register EXC_SUM, 5–13
floating-point issue queue, 2–7
hardware interrupt clear register HW_INT_CLR,
5–12
Ibox control register I_CTL, 5–15
Ibox process context register PCTX, 5–21
Ibox status register I_STAT, 5–18
Icache flush ASM register IC_FLUSH_ASM,
5–21
Icache flush register IC_FLUSH, 5–21
instruction fetch logic, 2–6
instruction virtual address format register
IVA_FORM, 5–9
instruction-stream translation buffer, 2–5
integer issue queue, 2–6
internal processor registers, 5–1
interrupt enable and current processor mode
register IER_CM, 5–9
interrupt summary register ISUM , 5–11
ITB invalidate single register ITB_IS, 5–7
ITB invalidate-all ASM (ASM=0) register
ITB_IAP, 5–7
ITB invalidate-all register ITB_IA, 5–7
ITB PTE array write register ITB_PTE, 5–6
ITB tag array write register ITB_TAG , 5–6
PAL base register PAL_BASE, 5–15
performance counter control register
PCTR_CTL, 5–23
ProfileMe register PMPC, 5–8
register rename maps, 2–6
retire logic, 2–8
retire logic and mapper, required sequence for,
D–1
sleep mode register SLEEP, 5–21
software interrupt request register SIRR, 5–10
subsections in, 2–2
virtual program counter logic, 2–2

Index–6

IC_FLUSH Icache flush register
at power-on reset state, 7–15
IC_FLUSH_ASM Icache flush ASM register, 5–21
Icache
data errors, 8–2
error case summary for, 8–10
fill from Bcache error, 8–5
fill from memory error, 8–7
flush register IC_FLUSH, 5–21
initialized by BiST, 7–12
tag, initialized by BiST, 7–12
IEEE 1149.1
notes for compliance to, 11–7
test port reset, 7–16
test port, operation of, 11–3
IEEE floating-point conformance, A–14
IEEE floating-point instruction opcodes, A–9
IER_CM interrupt enable and current processor mode
register, 5–9
at power-on reset state, 7–15
IMPLVER instruction values, 2–38
Independent floating-point function codes, A–11
INIT_MODE Cbox CSR, 5–39, 7–12
Initialization mode processing, 7–12
Input dc reference pin. See I_DC_REF pin type
Input differential amplifier clock receiver. See
I_DA_CLK pin type
Input differential amplifier receiver. See I_DA pin
type
Instruction fetch logic, 2–6
Instruction fetch, issue, and retire unit. See Ibox
Instruction fetch, pipelined, 2–14
Instruction issue rules, 2–16
Instruction latencies, pipelined, 2–20
Instruction ordering, 2–30
Instruction retire latencies, minimum, 2–21
Instruction retire rules
F31, 2–22
floating-point divide, 2–22
floating-point square root, 2–22
pipelined, 2–21
R31, 2–22
Instruction slot, pipelined, 2–14
Instruction-stream translation buffer, 2–5
Int_Add_BcClk internal forwarded clock, 4–44,
4–48
Int_Data_BcClk internal forwarded clock, 4–44,
4–49
INT_FWD_CLK clock queue, 4–30
Integer arithmetic trap, pipeline abort delay with,

21264/EV68A Hardware Reference Manual

2–16
Integer execution unit. See Ebox
Integer issue queue, 2–6
pipelined, 2–15
Internal processor registers, 5–1
accessing, 6–7
explicitly written, 6–8
implicitly written, 6–9
ordering access, 6–9
paired fetch order, 6–9
scoreboard bits for, 6–8
INTERRUPT interrupt, 6–14
INVAL_TO_DIRTY Cbox CSR, 4–23
programming, 4–23
INVAL_TO_DIRTY_ENABLE Cbox CSR, 5–39,
7–12

ITB_TAG array write register, 5–6
at power-on reset state, 7–14
IVA_FORM instruction virtual address format
register, 5–9
at power-on reset state, 7–15

J
JITTER_CMD Cbox CSR, defined, 5–38
JMP misprediction, in PALcode, D–15
JSR misprediction
in PALcode, D–15
pipeline abort delay with, 2–16
JSR_COR misprediction, in PALcode, D–15
Junction temperature, 9–1

InvalToDirty, 21264/EV68A command, 4–12,
4–22, 4–40
system probes, with, 4–41
InvalToDirtyVic, 21264/EV68A command, 4–22,
4–40

LDBU instruction, normal prefetch with, 2–23

IOWB. See I/O write buffer

LDF instruction, normal prefetch with, 2–23

L
Late-write non-bursting SSRAM pin assignments,
E–2

IPRs. See Internal processor registers

LDG instruction, normal prefetch with, 2–23

IQ. See Integer issue queue

LDQ instruction, prefetch with evict next, 2–24

IRQ_H signal pins, 3–5

LDS instruction, prefetch with modify intent, 2–23

Istream, 2–5

LDT instruction, normal prefetch with, 2–23

Istream memory references
translation to external references , 4–5
ISTREAM_BC_DBL error status in C_STAT, 5–42
ISTREAM_BC_ERR error status in C_STAT, 5–42
ISTREAM_MEM_DBL error status in C_STAT,
5–42

LDWU instruction, normal prefetch with, 2–23
LDx_L instructions
in-order processing for, 4–15
locking mechanism for, 4–14
Load hit speculation, 2–24

ISUM interrupt summary register, 5–11
at power-on reset state, 7–15
ITB, 2–5

Load instructions
ECC with, 8–3
I/O reference ordering, 2–31
Mbox order traps, 2–31
memory reference ordering, 2–30
translation to external interface, 4–5
Load queue, described, 2–13

ITB fill, 6–16

Load-load order trap, 2–31

ITB miss, pipeline abort delay with, 2–16

Local predictor, 2–4

ITB_IA invalidate-all register, 5–7
at power-on reset state, 7–15
ITB_IAP invalidate-all (ASM=0) register, 5–7
at power-on reset state, 7–15
ITB_IS invalidate single register, 5–7
at power-on reset state, 7–15
ITB_MISS fault, 6–14

Lock mechanism, 4–14

ISTREAM_MEM_ERR error status in C_STAT,
5–42

ITB_PTE array write register, 5–6
at power-on reset state, 7–14

Logic symbol, the 21264/EV68A, 3–2
LQ. See Load queue

M
M_CTL Mbox control register, 5–29
at power-on reset state, 7–16
MAF. See Miss address file
MB instruction processing, 2–33

21264/EV68A Hardware Reference Manual

Index–7

MB, 21264/EV68A command, 4–13, 4–21
MB_CNT Cbox CSR, operation, 2–32
MBDone, SysDc command, 4–13
Mbox
Dcache control register DC_CTL, 5–30
Dcache status register DC_STAT, 5–31
described, 2–12
Dstream translation buffer, 2–13
DTB address space number registers 0 and 1
DTB_ASNx, 5–28
DTB alternate processor mode register
DTB_ALTMODE, 5–26
DTB invalidate-all (ASM=0) process register
DTB_IAP, 5–27
DTB invalidate-all process register DTB_IA ,
5–27
DTB invalidate-single registers 0 and 1
DTB_ISx, 5–27
DTB PTE array write registers 0 and 1
DTB_PTEx, 5–26
DTB tag array write registers 0 and 1
DTB_TAGx, 5–25
internal processor registers, 5–2
load queue, 2–13
Mbox control register M_CTL, 5–29
memory management status register
MM_STAT, 5–28
miss address file, 2–13
order traps, 2–31
pipeline abort delay with order trap, 2–16
pipeline abort delays, 2–16
store queue, 2–13
MBOX_BC_PRB_STALL Cbox CSR, defined,
5–35
MCHK interrupt, 6–14
Mechanical specifications, 3–17
Memory
error case summary for, 8–11
filling Dcache errors, 8–8
filling Icache errors, 8–7
Memory address space
load instructions with, 2–27
merging rules, 2–30
store instructions with, 2–28
Memory barrier instructions
translation to external interface, 4–5
Memory barriers, 2–32

MM_STAT memory management status register,
5–28
at power-on reset state, 7–16
MT_FPCR instruction, 6–12
MT_FPCR synchronous trap, 6–14

N
NoConnect pin type, 3–3
Nonexistent memory
processing, 4–38
NOP, 21264/EV68A command, 4–21
Note convention, xxi
Numbering convention, xxi
NXM. See Nonexistent memory
NZNOP, 21264/EV68A command, 4–21

O
O_OD pin type, 3–3, 9–2
values for, 9–4
O_OD_TP pin type, 3–3, 9–2
values for, 9–4
O_PP pin type, 3–3, 9–2
values for, 9–5
O_PP_CLK pin type, 3–3, 9–2
values for, 9–5
OPCDEC fault, 6–13
Opcodes
IEEE floating-point, A–9
independent floating-point, A–11
reserved for Compaq, A–8
reserved for PALcode, A–9
summary of, A–12
VAX floating-point, A–11
Open-drain driver for test pins. See O_OD_TP
Open-drain output driver. See O_OD pin type
Operating temperature, 10–1

P
Packaging, 3–18

Memory reference unit. See Mbox

Paired instruction fetch order, 6–9

MF_FPCR instruction, 6–12

PAL_BASE register, 5–15
after fault reset, 7–8
after warm reset, 7–11
at power-on reset state, 7–15
through sleep mode, 7–10

Microarchitecture
summarized, 2–1
MiscVref signal pin, 3–5
Miss address file, 2–13
I/O address space loads, 2–27
memory address space loads, 2–27
memory address space stores , 2–28

Index–8

21264/EV68A Hardware Reference Manual

PALcode
conditional branches in, D–14
described, 6–1
entries points for, 6–12
exception entry points, 6–13
guidelines for, D–1
HW_LD instruction, 6–3
HW_MFPR instruction, 6–6
HW_MTPR instruction, 6–6
HW_RET instruction, 6–5
HW_ST instruction, 6–4
required function codes, 6–3
reserved opcodes for, 6–3
restrictions for, D–1
PALmode environment, 6–2
PALshadow registers, 6–11
PCTR_CTL performance counter control counter
register
updating, D–17
PCTR_CTL performance counter control register,
5–23
at power-on reset state, 7–15
updating, D–18
PCTX Ibox process context register, 5–21
after fault reset, 7–8
after warm reset, 7–11
at power-on reset state, 7–15
through sleep mode, 7–10
Phase-lock loop. See PLL
Physical address considerations, 4–4
Pipeline
abort delay, 2–16
Dcache access, 2–16
Ebox execution, 2–16
Ebox slotting, 2–18
Fbox execution, 2–16
instruction fetch, 2–14
instruction group definitions, 2–17
instruction issue rules, 2–16
instruction latencies, 2–20
instruction retire rules , 2–21
instruction slot, 2–14
issue queue, 2–15
organization, 2–13
register maps, 2–15
register reads, 2–16
PLL
description, 7–19
output clocks, 7–19
ramp up, 7–6
PLL_IDD, values for, 9–3
PLL_VDD signal pin, 3–5
PLL_VDD, values for, 9–3
PllBypass_H signal pin, 3–5
PMPC ProfileMe register, 5–8

21264/EV68A Hardware Reference Manual

Ports
IEEE 1149.1, 11–3
serial terminal, 11–2
SROM load, 11–2
Power
maximum, 9–1
sleep defined, 9–3
Power supply sequencing, 9–5
Power-on
flow signals and constraints, 7–7
reset flow, 7–1
self-test and initialization, 11–5
timing sequence, 7–3
PRB_TAG_ONLY Cbox CSR, 4–28
defined, 5–34
Privileged architecture library code
See PALcode
Probe commands, system, 4–26, 4–40
Probe queue, 2–11
PROBE_BC_ERR error status in C_STAT, 5–42
ProbeResponse, 21264/EV68A command, 4–21,
4–24, 4–39
ProfileMe mode, 6–20
Push-pull output clock driver. See O_PP_CLK
Push-pull output driver. See O_PP

R
R31
load instructions with, 2–23
retire instructions with, 2–22
speculative loads to, 2–25
RAMP1 reset machine state, 7–17
RAMP2 reset machine state, 7–18
Ranges and extents convention, xxi
RdBlk, 21264/EV68A command, 4–39
RdBlkI, 21264/EV68A command, 4–39
RdBlkMod, 21264/EV68A command, 4–39
RdBlkModSpec, 21264/EV68A command, 4–39
RdBlkModVic, 21264/EV68A command, 4–39
RdBlkSpec, 21264/EV68A command, 4–39
RdBlkSpecI, 21264/EV68A command, 4–39
RdBlkVic, 21264/EV68A command, 4–39
RdBlkVicI, 21264/EV68A command, 4–39
RdBytes, 21264/EV68A command, 4–39
RdLWs, 21264/EV68A command, 4–39
RdQWs, 21264/EV68A command, 4–39
RDVIC_ACK_INHIBIT Cbox CSR, 4–25, 4–26
defined, 5–34

Index–9

ReadBlk, 21264/EV68A command, 4–21
system probes, with, 4–41
ReadBlkI, 21264/EV68A command, 4–22

Security holes
with UNPREDICTABLE results, xxii
Serial terminal port, 11–2

ReadBlkMod, 21264/EV68A command, 4–22
system probes, with, 4–41
ReadBlkModSpec, 21264/EV68A command, 4–22

ReadBlkSpecI, 21264/EV68A command, 4–22

SET_DIRTY_ENABLE Cbox CSR, 4–23, 5–39,
7–12
programming, 4–24
SharedToDirty, 21264/EV68A command, 4–22 ,
4–40
system probes, with, 4–41
Signal name convention, xxi

ReadBlkVic, 21264/EV68A command, 4–22

Signal pin types, defined, 3–3

ReadBlkModVic, 21264/EV68A command, 4–22
ReadBlkSpec, 21264/EV68A command, 4–22

ReadBlkVicI, 21264/EV68A command, 4–22
ReadData, SysDc command, 4–10, 4–11, 4–12

Signal pins
test, 11–1
Single-bit error scribbing, D–19

ReadDataDirty, SysDc command, 4–10, 4–11, 4–12

Single-bit errors in hardware, correcting, 8–2

ReadDataError, SysDc command, 4–10, 4–11,
4–12, 4–13

SIRR software interrupt request register, 5–10
at power-on reset state, 7–15
Sleep mode
flow, 7–9
timing sequence, 7–11
SLEEP mode register, 5–21
at power-on reset state, 7–15
Spare pin type, 3–3

ReadBytes, 21264/EV68A command, 4–22

ReadDataShared, SysDc command, 4–10, 4–11,
4–12
ReadDataShared/Dirty, SysDc command, 4–10,
4–11, 4–12
ReadLWs, 21264/EV68A command, 4–22
ReadQWs, 21264/EV68A command, 4–22
Register access abbreviations, xix
Register figure conventions, xxi

SPEC_READ_ENABLE Cbox CSR, 4–23
defined, 5–35
SQ. See Store queue

SROM content map, 11–6

SROM initialization, 11–5

Replay traps, 2–31

SROM interface, in microarchitecture, 2–13

RESET interrupt, 6–14

SROM line, Icache bit fields in a, 11–6

Reset state machine
major operations of, 7–1
Reset_L signal pin, 3–5
power-on reset flow, 7–1
RET misprediction, in PALcode, D–15

SROM load, 7–6

Retire logic, 2–8, D–1

SROM load operation, 11–2
SromClk_H signal pin, 3–5, 11–2
SromData_H signal pin, 3–5, 11–2
SromOE_L signal pin, 3–5, 11–2

RW,n convention, xx

SSRAMs
dual-data rate pin assignments, E–3
late-write non-bursting pin assignments, E–2
STC_ENABLE Cbox CSR, 4–24

STCChangeToDirty, 21264/EV68A command,
4–13, 4–22, 4–40

RO,n convention, xix
RUN reset machine state, 7–18

SAMPLE public instruction, B–1

Storage temperature, 9–1

Scrubbing single-bit errors, D–19
I_CTL Ibox control register
updating I_CTL, D–18
Second-level cache. See Bcache

Index–10

21264/EV68A Hardware Reference Manual

Store instructions
Dcache ECC errors with, 8–4
I/O address space, 2–29
I/O reference ordering, 2–31
Mbox order traps, 2–31
memory address space, 2–28
memory reference ordering, 2–30
translation to external interface, 4–5
Store queue, 2–13
Store-load order trap, 2–31
STx_C instructions
in-order processing for, 4–15
locking mechanism for, 4–14
SUM bit. See Summary bit
Summary bit, in FPCR, 2–36

SysAddInClk_L signal pin, 3–5
SysAddOut_L signal pins, 3–5
SysAddOutClk_L signal pin, 3–5
SYSBUS_ACK_LIMIT Cbox CSR, 4–25
defined, 5–34
SYSBUS_FORMAT Cbox CSR, 4–21
SYSBUS_MB_ENABLE Cbox CSR, 4–23
defined, 5–34
operation, 2–32
SYSBUS_VIC_LIMIT Cbox CSR, 4–26
defined, 5–34
SysCheck_L signal pin, 3–5
SYSCLK, 4–31
SysData_L signal pin, 3–5

Supply voltage signal pins. See I_DC_POWER pin
type

SysDataInClk_H signal pin, 3–5

Synchronous static random-access memory. See
SSRAMs

SysDataInValid_L signal pin, 3–5
rules for, 4–34
SysDataOutClk_L signal pin, 3–5

SYS_BPHASE_LD_VECTOR Cbox CSR, 4–18
defined, 5–38
SYS_BUS_FORMAT Cbox CSR, defined, 5–34
SYS_BUS_SIZE Cbox CSR, 4–21
defined, 5–34
SYS_CLK_DELAY Cbox CSR, defined, 5–36
SYS_CLK_LD_VECTOR Cbox CSR, 4–18
defined, 5–38
SYS_CLK_RATIO Cbox CSR, defined, 5–34
SYS_CLKFWD_ENABLE Cbox CSR, defined,
5–36
SYS_CPU_CLK_DELAY Cbox CSR
defined, 5–38
SYS_DDM_FALL_EN Cbox CSR, 4–18
defined, 5–36
SYS_DDM_RD_FALL_EN Cbox CSR, 4–18

SysDataOutValid_L signal pin, 3–5
rules for, 4–35
SysDc commands, 4–11
system probes, with, 4–42
SysDc field, system to 21264/EV68A commands,
4–29
SYSDC_DELAY Cbox CSR, 4–32
defined, 5–38
SysFillValid_L signal pin, 3–5
rules for, 4–35
System clock ratio configuration, 7–4
System initialization, 7–7
System interface clocks, programming, 4–18
System port, 4–16
SysVref signal pin, 3–6

SYS_DDM_RD_RISE_EN Cbox CSR, 4–19
SYS_DDM_RISE_EN Cbox CSR, 4–18
defined, 5–36
SYS_DDMF_ENABLE Cbox CSR, 4–19
defined, 5–36
SYS_DDMR_ENABLE Cbox CSR, 4–19
defined, 5–36
SYS_FDBK_EN Cbox CSR, 4–18
defined, 5–38
SYS_FRAME_LD_VECTOR Cbox CSR, 4–19,
4–31
defined, 5–38
SYS_RCV_MUX_CNT_PRESET Cbox CSR, 4–31
defined, 5–36
SYS_RCV_MUX_PRESET Cbox CSR, 4–33

T
Tag parity errors, 8–2
TB fill flow, 2–34, 6–14
Tck_H signal pin, 3–6
Tdi_H signal pin, 3–6
Tdo_H signal pin, 3–6
Temperatures
maximium average per frequency, 10–2
operating, 10–1
Terminology, xix
TestStat_H signal pin, 3–6
purpose for, 11–4
with BiST and SROM load, 7–6

SysAddIn_L signal pins, 3–5

21264/EV68A Hardware Reference Manual

Index–11

Thermal design characteristics, 10–6

WAIT_ClkFwdRst0 reset machine state, 7–18

Tms_H signal pin, 3–6

WAIT_ClkFwdRst1 reset machine state, 7–18

Traps
load-load order , 2–31
Mbox order, 2–31
replay, 2–31
store-load order, 2–31
Trst_L signal pin, 3–6

WAIT_INTERRUPT reset machine state, 7–19

U
UNALIGN fault, 6–13

WAIT_NOMINAL reset machine state, 7–17
WAIT_RESET reset machine state, 7–18
WAIT_SETTLE reset machine state, 7–17
WAKEUP interrupt, 6–14
WAR, eliminating, 2–6
Warm reset flow, 7–11

UNDEFINED, xxii

WAW
eliminating, 2–6
WMB instruction processing, 2–33

UNPREDICTABLE, xxii

WO,n convention, xx

Unaligned convention, xx

V
VA virtual address register, 5–4
at power-on reset state, 7–15
VA_CTL virtual address control register, 5–4
at power-on reset state, 7–15
updating VA_48 field, D–18
VA_FORM virtual address format register, 5–5
at power-on reset state, 7–15
VAF. See Victim address file
VAX floating-point instruction opcodes, A–11
VBIAS defined, 9–2
VDB. See Victim data buffer
VDBFlushRequest, 21264/EV68A command, 4–21
VDD signal pin list, 3–16
VDD, values for, 9–3
VDF. See Victim data file
Vdiff defined, 9–2
Victim address file
described, 2–11
Victim address file, described, 2–11
Victim data buffer (VDB), 4–8
Virtual address support, 1–2

Wrap order
double-pumped, 4–38
interleaved, 4–37
WrBytes, 21264/EV68A command, 4–22, 4–39
Write hint instructions, translation to external
interface, 4–5
WRITE_MANY chain, 5–38
example, 5–39
values for Bcache initialization, 7–12
WRITE_MANY register
after fault reset, 7–8
after warm reset, 7–11
through sleep mode, 7–10
WRITE_ONCE chain description, 5–33
Write-after-read. See WAR
Write-after-write. See WAW
WrLWs, 21264/EV68A command, 4–22, 4–39
WrQWs, 21264/EV68A command, 4–22, 4–39
WrVictimBlk, 21264/EV68A command, 4–22,
4–39
system probes, with, 4–41

X
X convention, xxi

Virtual program counter logic, 2–2
VPC. See Virtual program counter logic
VREF, values for, 9–3
VSS signal pin list, 3–16

W
WAIT_BiSI reset machine state, 7–18
WAIT_BiST reset machine state, 7–18

Index–12

21264/EV68A Hardware Reference Manual