NS32532-20_技术文档

May 1991

NS32532-20/NS32532-25/NS32532-30

High-Performance 32-Bit Microprocessor

General Description

The NS32532 is a high-performance 32-bit microprocessor

Features

Y

Software compatible with the Series 32000 family

32-bit architecture and implementation

4-GByte uniform addressing space

On-chip memory management unit with 64-entry

translation look-aside buffer

Y

in the Series 32000 family. It is software compatible with

É

the previous microprocessors in the family but with a greatly

enhanced internal implementation.

The high-performance specifications are the result of a four-

stage instruction pipeline, on-chip instruction and data

caches, on-chip memory management unit and a signifi-

cantly increased clock frequency. In addition, the system

interface provides optimal support for applications spanning

a wide range, from low-cost, real-time controllers to highly

sophisticated, general purpose multiprocessor systems.

Y

4-Stage instruction pipeline

512-Byte on-chip instruction cache

1024-Byte on-chip data cache

High-performance bus

Ð Separate 32-bit address and data lines

Ð Burst mode memory accessing

Ð Dynamic bus sizing

The NS32532 integrates more than 370,000 transistors fab-

ricated in a 1.25 mm double-metal CMOS technology. The

advanced technology and mainframe-like design of the de-

vice enable it to achieve more than 10 times the throughput

of the NS32032 in typical applications.

Y

Extensive multiprocessing support

Floating-point support via the NS32381 or NS32580

1.25 mm double-metal CMOS technology

175-pin PGA package

In addition to generally improved performance, the

NS32532 offers much faster interrupt service and task

switching for real-time applications.

Block Diagram

TL/EE/9354–1

FIGURE 1

Series 32000É and TRI-STATEÉ are registered trademarks of National Semiconductor Corporation.

C

1995 National Semiconductor Corporation

TL/EE/9354

RRD-B30M105/Printed in U. S. A.

Table of Contents

1.0 PRODUCT INTRODUCTION

2.0 ARCHITECTURAL DESCRIPTION

2.1 Register Set

3.0 FUNCTIONAL DESCRIPTION (Continued)

3.1.3 Instruction Pipeline

3.1.3.1 Branch Prediction

3.1.3.2 Memory Mapped I/O

2.1.1 General Purpose Registers

2.1.2 Address Registers

3.1.3.3 Serializing Operations

3.1.4 Slave Processor Instructions

3.1.4.1 Regular Slave Instruction Protocol

3.1.4.2 Pipelined Slave Instruction Protocol

3.1.4.3 Instruction Flow and Exceptions

3.1.4.4 Floating-Point Instructions

3.1.4.5 Custom Slave Instructions

2.1.3 Processor Status Register

2.1.4 Configuration Register

2.1.5 Memory Management Registers

2.1.6 Debug Registers

2.2 Memory Organization

2.2.1 Address Mapping

3.2 Exception Processing

3.2.1 Exception Acknowledge Sequence

3.2.2 Returning from an Exception Service Procedure

3.2.3 Maskable Interrupts

2.3 Modular Software Support

2.4 Memory Management

3.2.3.1 Non-Vectored Mode

2.4.1 Page Tables Structure

3.2.3.2 Vectored Mode: Non-Cascaded Case

3.2.3.3 Vectored Mode: Cascaded Case

3.2.4 Non-Maskable Interrupt

3.2.5 Traps

2.4.2 Virtual Address Spaces

2.4.3 Page Table Entry Formats

2.4.4 Physical Address Generation

2.4.5 Address Translation Algorithm

3.2.6 Bus Errors

2.5 Instruction Set

3.2.7 Priority Among Exceptions

2.5.1 General Instruction Format

2.5.2 Addressing Modes

3.2.8 Exception Acknowledge Sequences:

Detailed Flow

3.2.8.1 Maskable/Non-Maskable Interrupt

Sequence

2.5.3 Instruction Set Summary

3.0 FUNCTIONAL DESCRIPTION

3.2.8.2 Abort/Restartable Bus Error Sequence

3.1 Instruction Execution

3.2.8.3 SLAVE/ILL/SVC/DVZ/FLG/BPT/UND

Trap Sequence

3.1.1 Operating States

3.2.8.4 Trace Trap Sequence

3.1.2 Instruction Endings

3.1.2.1 Completed Instructions

3.1.2.2 Suspended Instructions

3.1.2.3 Terminated Instructions

3.1.2.4 Partially Completed Instructions

2

Table of Contents (Continued)

3.0 FUNCTIONAL DESCRIPTION (Continued)

4.0 DEVICE SPECIFICATIONS (Continued)

3.2.8.5 Integer-Overflow Trap Sequence

3.2.8.6 Debug Trap Sequence

4.4.1 Definitions

4.4.2 Timing Tables

3.2.8.7 Non-Restartable Bus Error Sequence

4.4.2.1 Output Signals: Internal Propagation

Delays

3.3 Debugging Support

4.4.2.2 Input Signal Requirements

3.3.1 Instruction Tracing

4.4.3 Timing Diagrams

3.3.2 Debug Trap Capability

3.4 On-Chip Caches

APPENDIX A: INSTRUCTION FORMATS

3.4.1 Instruction Cache (IC)

B: COMPATIBILITY ISSUES

B.1 Restrictions on Compatibility

B.2 Architecture Extensions

B.3 Integer-Overflow Trap

B.4 Self-Modifying Code

3.4.2 Data Cache (DC)

3.4.3 Cache Coherence Support

3.4.4 Translation Look-aside Buffer (TLB)

3.5 System Interface

3.5.1 Power and Grounding

3.5.2 Clocking

B.5 Memory-Mapped I/O

3.5.3 Resetting

C: INSTRUCTION SET EXTENSIONS

C.1 Processor Service Instructions

C.2 Memory Management Instructions

C.3 Instruction Definitions

3.5.4 Bus Cycles

3.5.4.1 Bus Status

3.5.4.2 Basic Read and Write Cycles

3.5.4.3 Burst Cycles

3.5.4.4 Cycle Extension

3.5.4.5 Interlocked Bus Cycles

3.5.4.6 Interrupt Control Cycles

3.5.4.7 Slave Processor Bus Cycles

D: INSTRUCTION EXECUTION TIMES

D.1 Internal Organization and Instruction

Execution

D.2 Basic Execution Times

3.5.5 Bus Exceptions

D.2.1 Loader Timing

3.5.6 Dynamic Bus Configuration

3.5.6.1 Instruction Fetch Sequences

3.5.6.2 Data Read Sequences

3.5.6.3 Data Write Sequences

3.5.7 Bus Access Control

D.2.2 Address Unit Timing

D.2.3 Execution Unit Timing

D.3 Instruction Dependencies

D.3.1 Data Dependencies

D.3.1.1 Register Interlocks

D.3.1.2 Memory Interlocks

3.5.8 Interfacing Memory-Mapped I/O Devices

3.5.9 Interrupt and Debug Trap Requests

3.5.10 Cache Invalidation Requests

3.5.11 Internal Status

D.3.2 Control Dependencies

D.4 Storage Delays

D.4.1 Instruction Cache Misses

D.4.2 Data Cache Misses

4.0 DEVICE SPECIFICATIONS

4.1 Pin Descriptions

D.4.3 TLB Misses

4.1.1 Supplies

D.4.4 Instruction and Operand Alignment

4.1.2 Input Signals

4.1.3 Output Signals

4.1.4 Input/Output Signals

D.5 Execution Time Calculations

D.5.1 Definitions

D.5.2 Notes on Table Use

4.2 Absolute Maximum Ratings

4.3 Electrical Characteristics

4.4 Switching Characteristics

D.5.3 T Evaluation

eff

D.5.4 Instruction Timing Example

D.5.5 Execution Timing Tables

D.5.5.1 Basic and Memory

Management Instructions

D.5.5.2 Floating-Point Instructions,

CPU Portion

3

List of Illustrations

CPU Block Diagram ÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀ 1

NS32532 Internal Registers ÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀ 2-1

Processor Status Register (PSR) ÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀ 2-2

Configuration Register (CFG) ÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀ 2-3

Page Table Base Registers (PTBn) ÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀ 2-4

Memory Management Control Register (MCR) ÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀ 2-5

Memory Management Status Register (MSR) ÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀ 2-6

Debug Condition Register (DCR) ÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀ 2-7

Debug Status Register (DSR) ÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀ 2-8

NS32532 Address Mapping ÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀ 2-9

NS32532 Run-Time Environment ÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀ 2-10

Two-Level Page Tables ÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀ 2-11

Page Table Entries (PTE’s) ÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀ 2-12

Virtual to Physical Address Translation ÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀ 2-13

General Instruction Format ÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀ 2-14

Index Byte Format ÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀ 2-15

Displacement Encodings ÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀ 2-16

Operating States ÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀ 3-1

NS32532 Internal Instruction Pipeline ÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀ 3-2

Memory References for Consecutive Instructions ÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀ 3-3

Memory References after Serialization ÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀ 3-4

Regular Slave Instruction Protocol: CPU Actions ÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀ 3-5

ID and Operation Word ÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀ 3-6

Slave Processor Status Word ÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀ 3-7

Instruction Flow in Pipelined Floating-Point Mode ÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀ 3-8

Interrupt Dispatch Table ÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀ 3-9

Exception Acknowledge Sequence: Direct-Exception Mode Disabled ÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀ 3-10

Exception Acknowledge Sequence: Direct-Exception Mode Enabled ÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀ 3-11

Return From Trap (RETTn) Instruction Flow: Direct-Exception Mode Disabled ÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀ 3-12

Return From Interrupt (RETI) Instruction Flow: Direct-Exception Mode Disabled ÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀ 3-13

Exception Processing Flowchart ÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀ 3-14

Service Sequence ÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀ 3-15

Instruction Cache Structure ÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀ 3-16

Data Cache Structure ÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀ 3-17

TLB Model ÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀ 3-18

Power and Ground Connections ÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀ 3-19

Bus Clock Synchronization ÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀ 3-20

Power-On Reset Requirements ÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀ 3-21

General Reset Timing ÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀ 3-22

Basic Read Cycle ÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀ 3-23

Write Cycle ÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀ 3-24

Burst Read cycles ÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀ 3-25

Cycle Extension of a Basic Read Cycle ÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀ 3-26

Slave Processor Write Cycle ÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀ 3-27

Slave Processor Read Cycle ÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀ 3-28

Bus Retry During a Basic Read Cycle ÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀ 3-29

Basic Interface for 32-Bit Memories ÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀ 3-30

Basic Interface for 16-Bit Memories ÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀ 3-31

Hold Acknowledge: (Bus Initially Idle) ÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀ 3-32

Typical I/O Device Interface ÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀ 3-33

4

List of Illustrations (Continued)

NS32532 Interface Signals ÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀ 4-1

175-Pin PGA Package ÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀ 4-2

Output Signals Specification Standard ÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀ4-3

Input Signals Specification StandardÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀ4-4

Basic Read Cycle Timing ÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀ 4-5

Write Cycle Timing ÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀ 4-6

Interlocked Read and Write Cycles ÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀ 4-7

Burst Read Cycles ÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀ 4-8

External Termination of Burst Cycles ÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀ 4-9

Bus Error or Retry During Burst Cycles ÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀ 4-10

Extended Retry Timing ÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀ 4-11

HOLD Timing (Bus Initially Idle) ÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀ 4-12

HOLD Acknowledge Timing (Bus Initially Not Idle) ÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀ 4-13

Slave Processor Read Timing ÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀ 4-14

Slave Processor Write Timing ÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀ 4-15

Slave Processor Done ÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀ 4-16

FSSR Signal Timing ÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀ 4-17

Cache Invalidation Request ÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀ 4-18

INT and NMI Signals Sampling ÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀ 4-19

Debug Trap Request ÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀ 4-20

PFS Signal Timing ÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀ 4-21

ISF Signal Timing ÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀ 4-22

Break Point Signal Timing ÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀ 4-23

Clock Waveforms ÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀ 4-24

Bus Clock Synchronization ÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀ 4-25

Power-On Reset ÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀ 4-26

Non-Power-On Reset ÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀ 4-27

LPRi/SPRi Instruction Formats ÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀ C-1

CINV Instruction Format ÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀ C-2

LMR/SMR Instruction Formats ÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀ C-3

List of Tables

Access Protection Levels ÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀ 2-1

NS32532 Addressing Modes ÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀ 2-2

NS32532 Instruction Set Summary ÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀ 2-3

Floating-Point Instruction Protocol ÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀ 3-1

Custom Slave Instruction Protocols ÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀ 3-2

Summary of Exception Processing ÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀ 3-3

Interrupt Sequences ÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀ 3-4

Cacheable/Non-Cacheable Instruction Fetches from a 32-Bit Bus ÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀ 3-5

Cacheable/Non-Cacheable Instruction Fetches from a 16-Bit Bus ÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀ 3-6

Cacheable/Non-Cacheable Instruction Fetches from an 8-Bit Bus ÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀ 3-7

Cacheable/Non-Cacheable Data Reads from a 32-Bit Bus ÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀ 3-8

Cacheable/Non-Cacheable Data Reads from a 16-Bit Bus ÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀ 3-9

Cacheable/Non-Cacheable Data Reads from an 8-Bit Bus ÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀ 3-10

Data Writes to a 32-Bit Bus ÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀ 3-11

Data Writes to a 16-Bit Bus ÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀ 3-12

Data Writes to an 8-Bit Bus ÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀ 3-13

LPRi/SPRi New ‘Short’ Field Encodings ÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀ C-1

LMR/SMR ‘Short’ Field Encodings ÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀ C-2

Additional Address Unit Processing Time for Complex Addressing Modes ÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀÀD-1

5

1.0 Product Introduction

The NS32532 is an extremely sophisticated microprocessor

in the Series 32000 family with a full 32-bit architecture and

implementation optimized for high-performance applica-

tions.

Large, Uniform Addressing. The NS32532 has 32-bit ad-

dress pointers that can address up to 4 gigabytes without

requiring any segmentation; this addressing scheme pro-

vides flexible memory management without added-on ex-

pense.

By employing a number of mainframe-like features, the de-

vice can deliver 15 MIPS peaks performance with no wait

states at a frequency of 30 MHz.

Modular Software Support. Any software package for the

Series 32000 family can be developed independent of all

other packages, without regard to individual addressing. In

addition, ROM code is totally relocatable and easy to ac-

cess, which allows a significant reduction in hardware and

software costs.

The NS32532 is fully software compatible will all the other

Series 32000 CPUs. The architectural features of the Series

32000 family and particularly the NS32532 CPU, are de-

scribed briefly below.

Software Processor Concept. The Series 32000 architec-

ture allows future expansions of the instruction set that can

be executed by special slave processors, acting as exten-

sions to the CPU. This concept of slave processors is

unique to the Series 32000 family. It allows software com-

patibility even for future components because the slave

hardware is transparent to the software. With future ad-

vances in semiconductor technology, the slaves can be

physically integrated on the CPU chip itself.

Powerful Addressing Modes. Nine addressing modes

available to all instructions are included to access data

structures efficiently.

Data Types. The architecture provides for numerous data

types, such as byte, word, doubleword, and BCD, which may

be arranged into a wide variety of data structures.

Symmetric Instruction Set. While avoiding special case

instructions that compilers can’t use, the Series 32000 ar-

chitecture incorporates powerful instructions for control op-

erations, such as array indexing and external procedure

calls, which save considerable space and time for compiled

code.

To summarize, the architectural features cited above pro-

vide three primary performance advantages and character-

istics:

High-level language support

#

Memory-to-Memory Operations. The Series 32000 CPUs

represent two-address machines. This means that each op-

erand can be referenced by any one of the addressing

modes provided.

Easy future growth path

Application flexibility

2.0 Architectural Description

2.1 REGISTER SET

This powerful memory-to-memory architecture permits

memory locations to be treated as registers for all usefull

operations. This is important for temporary operands as well

as for context switching.

The NS32532 CPU has 28 internal registers grouped ac-

cording to functions as follows: 8 general purpose, 7 ad-

dress, 1 processor status, 1 configuration, 7 memory man-

agement and 4 debug. All registers are 32 bits wide except

for the module and processor status, which are each 16 bits

wide. Figure 2-1 shows the NS32532 internal registers.

Memory Management. The NS32532 on-chip memory

management unit provides advanced operating system sup-

port functions, including dynamic address translation, virtual

memory management, and memory protection.

Address

32 Bits

General Purpose

w 32 Bits x

w

x

PC

R0

SP0

R1

SP1

R2

FP

R3

SB

R4

INTBASE

R5

MOD

R6

R7

Processor Status

PSR

Debug

DCR

DSR

CAR

BPC

Memory Management

PTB0

PTB1

IVAR0

IVAR1

TEAR

MCR

Configuration

MSR

CFG

FIGURE 2-1. NS32532 Internal Registers

6

2.0 Architectural Description (Continued)

2.1.1 General Purpose Registers

INTBASEÐInterrupt Base. The INTBASE register holds

the address of the dispatch table for interrupts and traps

(Section 3.2.1).

There are eight registers (R0–R7) used for satisfying the

high speed general storage requirements, such as holding

temporary variables and addresses. The general purpose

registers are free for any use by the programmer. They are

32 bits in length. If a general purpose register is specified for

an operand that is eight or 16 bits long, only the low part of

the register is used; the high part is not referenced or modi-

fied.

MODÐModule. The MOD register holds the address of the

module descriptor of the currently executing software mod-

ule. The MOD register is 16 bits long, therefore the module

table must be contained within the first 64 kbytes of memo-

ry.

2.1.3 Processor Status Register

2.1.2 Address Registers

The Processor Status Register (PSR) holds status informa-

tion for the microprocessor.

The seven address registers are used by the processor to

implement specific address functions. A description of them

follows.

The PSR is sixteen bits long, divided into two eight-bit

halves. The low order eight bits are accessible to all pro-

grams, but the high order eight bits are accessible only to

programs executing in Supervisor Mode.

PCÐProgram Counter. The PC register is a pointer to the

first byte of the instruction currently being executed. The PC

is used to reference memory in the program section.

C

The C bit indicates that a carry or borrow occurred after

an addition or subtraction instruction. It can be used

with the ADDC and SUBC instructions to perform multi-

ple-precision integer arithmetic calculations. It may

have a setting of 0 (no carry or borrow) or 1 (carry or

borrow).

SP0, SP1ÐStack Pointers. The SP0 register points to the

lowest address of the last item stored on the INTERRUPT

STACK. This stack is normally used only by the operating

system. It is used primarily for storing temporary data, and

holding return information for operating system subroutines

and interrupt and trap service routines. The SP1 register

points to the lowest address of the last item stored on the

USER STACK. This stack is used by normal user programs

to hold temporary data and subroutine return information.

T

L

The T bit causes program tracing. If this bit is set to 1, a

TRC trap is executed after every instruction (Section

3.3.1).

The L bit is altered by comparison instructions. In a

comparison instruction the L bit is set to ‘‘1’’ if the sec-

ond operand is less than the first operand, when both

operands are interpreted as unsigned integers. Other-

wise, it is set to ‘‘0’’. In Floating-Point comparisons, this

bit is always cleared.

When a reference is made to the selected Stack Pointer

(see PSR S-bit), the terms ‘SP Register’ or ‘SP’ are used.

SP refers to either SP0 or SP1, depending on the setting of

the S bit in the PSR register. If the S bit in the PSR is 0, SP

refers to SP0. If the S bit in the PSR is 1 then SP refers to

SP1.

V

F

The V-bit enables generation of a trap (OVF) when an

integer arithmetic operation overflows.

The NS32532 also allows the SP1 register to be directly

loaded and stored using privileged forms of the LPRi and

SPRi instructions, regardless of the setting of the PSR S-bit.

When SP1 is accessed in this manner, it is referred to as

‘USP Register’ or simply ‘USP’.

The F bit is a general condition flag, which is altered by

many instructions (e.g., integer arithmetic instructions

use it to indicate overflow).

Z

The Z bit is altered by comparison instructions. In a

comparison instruction the Z bit is set to ‘‘1’’ if the sec-

ond operand is equal to the first operand; otherwise it is

set to ‘‘0’’.

Stacks in the Series 32000 family grow downward in memo-

ry. A Push operation pre-decrements the Stack Pointer by

the operand length. A Pop operation post-increments the

Stack Pointer by the operand length.

N

The N bit is altered by comparison instructions. In a

comparison instruction the N bit is set to ‘‘1’’ if the sec-

ond operand is less than the first operand, when both

operands are interpreted as signed integers. Otherwise,

it is set to ‘‘0’’.

FPÐFrame Pointer. The FP register is used by a procedure

to access parameters and local variables on the stack. The

FP register is set up on procedure entry with the ENTER

instruction and restored on procedure termination with the

EXIT instruction.

U

If the U bit is ‘‘1’’ no privileged instructions may be exe-

cuted. If the U bit is ‘‘0’’ then all instructions may be

The frame pointer holds the address in memory occupied by

the old contents of the frame pointer.

e

Supervisor Mode; when U

executed. When U

0 the processor is said to be in

e

SBÐStatic Base. The SB register points to the global vari-

ables of a software module. This register is used to support

relocatable global variables for software modules. The SB

register holds the lowest address in memory occupied by

the global variables of a module.

1 the processor is said to

15

8 7

0

I

P

S

U

N

Z

F

V

L

T

C

FIGURE 2-2. Processor Status Register (PSR)

7

2.0 Architectural Description (Continued)

be in User Mode. A User Mode program is restricted

from executing certain instructions and accessing cer-

tain registers which could interfere with the operating

system. For example, a User Mode program is prevent-

ed from changing the setting of the flag used to indicate

its own privilege mode. A Supervisor Mode program is

assumed to be a trusted part of the operating system,

hence it has no such restrictions.

F

Floating-point instruction set. This bit indicates

whether a floating-point unit (FPU) is present to exe-

cute floating-point instructions. If this bit is 0 when the

CPU executes

a floating-point instruction, a Trap

(UND) occurs. If this bit is 1, then the CPU transfers

the instruction and any necessary operands to the

FPU using the slave-processor protocol described in

Section 3.1.4.1.

S

The S bit specifies whether the SP0 register or SP1

register is used as the Stack Pointer. The bit is automat-

ically cleared on interrupts and traps. It may have a

setting of 0 (use the SP0 register) or 1 (use the SP1

register).

M

Memory management instruction set. This bit en-

ables the execution of memory management instruc-

tions. If this bit is 0 when the CPU executes an LMR,

SMR, RDVAL, or WRVAL instruction, a Trap (UND)

occurs. If this bit is 1, the CPU executes LMR, SMR,

RDVAL, and WRVAL instructions using the on-chip

MMU.

P

I

The P bit prevents a TRC trap from occuring more than

once for an instruction (Section 3.3.1). It may have a

setting of 0 (no trace pending) or 1 (trace pending).

C

Custom instruction set. This bit indicates whether a

custom slave processor is present to execute custom

instructions. If this bit is 0 when the CPU executes a

custom instruction, a Trap (UND) occurs. If this bit is

1, the CPU transfers the instruction and any neces-

sary operands to the custom slave processor using

the slave-processor protocol described in Section

3.1.4.1.

e

0,

If I

1, then all interrupts will be accepted. If I

only the NMI interrupt is accepted. Trap enables are not

affected by this bit.

2.1.4 Configuration Register

The Configuration Register (CFG) is 32 bits wide, of which

ten bits are implemented. The implemented bits enable vari-

ous operating modes for the CPU, including vectoring of

interrupts, execution of slave instructions, and control of the

on-chip caches. In the NS32332 bits 4 through 7 of the CFG

register selected between the 16-bit and 32-bit slave proto-

cols and between 512-byte and 4-Kbyte page sizes. The

NS32532 supports only the 32-bit slave protocol and

4-Kbyte page size: consequently these bits are forced to 1.

DE

DC

Direct-Exception mode enable. This bit enables the

Direct-Exception mode for processing exceptions.

When this mode is selected, the CPU response time

to interrupts and other exceptions is significantly im-

proved. Refer to Section 3.2.1 for more information.

Data Cache enable. This bit enables the on-chip Data

Cache to be accessed for data reads and writes. Re-

fer to Section 3.4.2 for more information.

When the CFG register is loaded using the LPRi instruction,

bits 14 through 31 should be set to 0. Bits 4 through 7 are

ignored during loading, and are always returned as 1’s when

CFG is stored via the SPRi instruction. When the SETCFG

instruction is executed, the contents of the CFG register bits

0 through 3 are loaded from the instruction’s short field, bits

4 through 7 are ignored and bits 8 through 13 are forced to

0.

LDC Lock Data Cache. This bit controls whether the con-

tents of the on-chip Data Cache are locked to fixed

e

memory locations (LDC 1), or updated when a data

read is missing from the cache (LDC 0).

e

IC

Instruction Cache enable. This bit enables the on-

chip Instruction Cache to be accessed for instruction

fetches. Refer to Section 3.4.1 for more information.

The format of the CFG register is shown in Figure 2-3. The

various control bits are described below.

LIC Lock Instruction Cache. This bit controls whether the

contents of the on-chip Instruction Cache are locked

I

Interrupt vectoring. This bit controls whether maska-

e

to fixed memory locations (LIC 1), or updated when

an instruction fetch is missing from the cache

e

(LIC 0).

e

ble interrupts are handled in nonvectored (I 0) or

vectored (I 1) mode. Refer to Section 3.2.3 for more

e

information.

PF

Pipelined Floating-point execution. This bit indicates

whether the floating-point unit uses the pipelined

slave protocol. When PF is 1 the pipelined protocol is

selected. PF is ignored if the F bit is 0. Refer to Sec-

tion 3.1.4.2 for more information.

31

14 13

8 7

0

Reserved

PF

LIC

IC

LDC

DC

DE

1

C

M

F

I

FIGURE 2-3. Configuration Register (CFG) Bits

13 to 31 are Reserved; Bits 4 to 7 are Forced to 1.

8

2.0 Architectural Description (Continued)

2.1.5 Memory Management Registers

DS Dual Space. While this bit is 1, then PTB1 contains the

level-1 page table base address of all addresses spec-

ified in User-Mode, and PTB0 contains the level-1

page table base address of all addresses specified in

Supervisor Mode. While this bit is 0, then PTB0 con-

tains the level-1 page table base address of all ad-

dresses specified in both User and Supervisor Modes.

The NS32532 provides 7 registers to support memory man-

agement functions. They are accessed by means of the

LMR and SMR instructions. All of them can be read and

written except IVAR0 and IVAR1 that are write-only. A de-

scription of the memory management registers is given in

the following sections.

AO Access Level Override. When this bit is set to 1, User-

Mode accesses are given Supervisor Mode privilege.

PTB0, PTB1ÐPage Table Base Pointers. The PTBn regis-

ters hold the physical addresses of the level-1 page tables

used in address translation. The least significant 12 bits are

permanently zero, so that each register always points to a

4-Kbyte boundary in memory.

31

4 3

0

Reserved

AO DS TS TU

When either PTB0 or PTB1 is loaded by executing an LMR

instruction, the MMU automatically invalidates all entries in

the TLB that had been translated using the old value in the

selected PTBn register.

FIGURE 2-5. Memory Management

Control Register (MCR)

MSRÐMemory Management Status. The MSR register

provides status information related to the occurrence of a

translation exception. Only eight bits are implemented. Bits

8 to 31 are ignored when MSR is loaded and are returned

as zeroes when it is read as a 32-bit word. MSR is only

updated by the MMU when a protection violation or page

fault is detected while translating an address for a reference

required to execute an instruction. It is not updated if a page

fault is detected during either an operand or an instruction

prefetch, if the data being prefetched is not needed due to a

change in the instruction execution sequence. The format of

MSR is shown in Figure 2-6. Details on the function of each

bit are given below.

The format of the PTBn registers is shown in Figure 2-4.

31

12 11

0

Base Address

000000000000

FIGURE 2-4. Page Table Base Registers (PTBn)

IVAR0, IVAR1ÐInvalidate Virtual Address. The Invalidate

Virtual Address registers are write-only registers. When a

virtual address is written to IVAR0 or IVAR1 using the LMR

instruction, the translation for that virtual address is purged,

if present, from the TLB. This must be done whenever a

Page Table Entry has been changed in memory, since the

TLB might otherwise contain an incorrect translation value.

TEX Translation Exception. This two-bit field specifies the

cause of the current address translation exception.

(Trap(ABT)). Combinations appearing in this field

are summarized below.

Another technique for purging TLB entries is to load a PTBn

register. Turning off translation (clearing the MCR TU and/

or TS bits) does not purge any entries from the TLB.

00 No Translation Exception

01 First Level PTE Invalid

10 Second Level PTE Invalid

11 Protection Violation

TEARÐTranslation Exception Address Register. The

TEAR register is loaded by the on-chip MMU when a trans-

lation exception occurs. It contains the 32-bit virtual address

that caused the translation exception.

During address translation, if a protection violation

and an invalid PTE are detected at the same time,

the TEX field is set to indicate a protection violation.

TEAR is not updated if a page fault is detected while pre-

fetching an instruction that is not executed because the pre-

vious instruction caused a trap.

DDT Data Direction. This bit indicates the direction of the

transfer that the CPU was attempting when the

translation exception occurred.

MCRÐMemory Management Control. The MCR register

controls the operation of the MMU. Only four bits are imple-

mented. Bits 4 to 31 are reserved for future use and must be

loaded with zeroes.

l

e

DDT

0

1

Read Cycle

Write Cycle

When MCR is read as a 32-bit word, bits 4 to 31 are re-

turned as zeroes. The format of MCR is shown inFigure 2-5.

Details on the control bits are given below.

UST User/Supervisor. This bit indicates whether the

Translation Exception was caused by a User-Mode

or Supervisor Mode reference. If UST is 1, then the

exception was caused by a User-Mode reference;

otherwise it was caused by a Supervisor Mode refer-

ence.

TU Translate User. While this bit is 1, address translation

is enabled for User-Mode memory references. While

this bit is 0, address translations is disabled for User-

Mode memory references.

TS Translate Supervisor. While this bit is 1, address trans-

lation is enabled for Supervisor Mode memory refer-

ences. While this bit is 0, address translation is dis-

abled for Supervisor-Mode memory references.

9

2.0 Architectural Description (Continued)

31

8 7

4 3

UST

0

Reserved

STT

DDT

TEX

FIGURE 2-6. Memory Management Status Register (MSR)

STT CPU Status. This four bit field is set on an address PCE PC-match enable

translation exception according to the following en-

codings.

UD

SD

Enable debug conditions in User-Mode

Enable debug conditions in Supervisor Mode

1000 Sequential Instruction Fetch

1001 Non-Sequential Instruction Fetch

1010 Data Transfer

DEN Enable debug conditions

The following 2 bits control testing features that can be

used during initial system debugging. These features are

unique to the NS32532 implementation of the Series 32000

architecture; as such, they may not be supported in future

implementations. For normal operation these 2 bits should

be set to 0.

1011 Read Read-Modify-Write Operand

1100 Read for Effective Address

If a reference for an Interrupt-Acknowledge or End-

of-Interrupt bus cycle (either Master of Cascaded)

causes a Translation Exception, then the value of

the STT-field is undefined.

SI

Single-Instruction mode enable. This bit, when set

to 1, inhibits the overlapping of instruction’s execu-

tion.

2.1.6 Debug Registers

BCP Branch Condition Prediction disable. When this bit is

1, the branch prediction mechanism is disabled. See

Section 3.1.3.1.

The NS32532 contains 4 registers dedicated for debugging

functions.

These registers are accessed using privileged forms of the

LPRi and SPRi instructions.

DSRÐDebug Status Register. The DSR Register indicates

debug conditions that have been detected. When the CPU

detects an enabled debug condition, it sets the correspond-

ing bit (BC, BEX, BCA) in the DSR to 1. When an address-

compare condition is detected, then the RD-bit is loaded to

indicate whether a read or write reference was performed.

Software must clear all the bits in the DSR when appropri-

ate. The format of the DSR is shown in Figure 2-8; the vari-

ous fields are described below.

DCRÐDebug Condition Register. The DCR Register en-

ables detection of debug conditions. The format of the DCR

is shown in Figure 2-7; the various bits are described below.

A debug condition is enabled when the related bit is set to 1.

CBE0 Compare Byte Enable 0; when set, BYTE0 of an

aligned double-word is included in the address com-

parison

RD

Indicates whether the last address-compare condi-

e

0)

CBE1 Compare Byte Enable 1; when set, BYTE1 of an

aligned double-word is included in the address com-

parison

e

tion was for a read (RD

reference

1) or write (RD

BPC PC-match condition detected

BEX External condition detected

CBE2 Compare Byte Enable 2; when set, BYTE2 of an

aligned double-word is included in the address com-

parison

BCA Address-compare condition detected

CBE3 Compare Byte Enable 3; when set, BYTE3 of an

aligned double-word is included in the address com-

parison

Note 1: The content of the DSR register is not defined if a debug condition

was detected on a floating-point instruction in pipelined mode and a

trap was generated by a previous floating-point instruction.

Note 2: If an address compare is detected on a read and a write for the

e

VNP Compare virtual address (VNP

e

1) or physical ad-

same instruction then the RD-bit will remain clear.

dress (VNP

0)

CARÐCompare Address Register. The CAR Register

contains the address that is compared to operand reference

addresses to detect an address-compare condition. The ad-

dress must be double-word aligned; that is, the two least-

significant bits must be 0. The CAR is 32 bits wide.

CWR Address-compare enable for write references

CRD Address-compare enable for read references

CAE Address-compare enable

TR

Enable Trap (DBG) when a debug condition is de-

tected

15

8 7

0

Reserved

CAE CRD CWR VNP CBE3 CBE2 CBE1 CBE0

31

24 23

DEN

16

Res

SD

UD

PCE

TR

BCP

SI

FIGURE 2-7. Debug Condition Register (DCR)

28 27

BCA

0

RD

BPC

BEX

Reserved

FIGURE 2-8. Debug Status Register (DSR)

10

2.0 Architectural Description (Continued)

BPCÐBreakpoint Program Counter. The BPC Register

contains the address that is compared with the PC contents

to detect a PC-match condition. The BPC Register is 32 bits

wide.

stored at the lowest address and the most significant word

of the double-word is stored at the address two higher. In

memory, the address of a double-word is the address of its

least significant byte, and a double-word may start at any

address.

2.2 MEMORY ORGANIZATION

The NS32532 implements full 32-bit virtual addresses. This

allows the CPU to access up to 4 Gbytes of virtual memory.

The memory is a uniform linear address space. Memory lo-

31

24 23

16 15

8

7

0

a

A 1

A

3

A

2

A

cations are numbered sequentially starting at zero and end-

32

MSB

LSB

b

ing at 2

1. The number specifying a memory location is

called an address. The contents of each memory location is

a byte consisting of eight bits. Unless otherwise noted, dia-

grams in this document show data stored in memory with

the lowest address on the right and the highest address on

the left. Also, when data is shown vertically, the lowest ad-

dress is at the top of a diagram and the highest address at

the bottom of the diagram. When bits are numbered in a

diagram, the least significant bit is given the number zero,

and is shown at the right of the diagram. Bits are numbered

in increasing significance and toward the left.

Double-Word at Address A

Although memory is addressed as bytes, it is actually orga-

nized as double-words. Note that access time to a word or a

double-word depends upon its address, e.g. double-words

that are aligned to start at addresses that are multiples of

four will be accessed more quickly than those not so

aligned. This also applies to words that cross a double-word

boundary.

2.2.1 Address Mapping

Figure 2-9 shows the NS32532 address mapping.

7

0

The NS32532 supports the use of memory-mapped periph-

eral devices and coprocessors. Such memory-mapped de-

vices can be located at arbitrary locations in the address

space except for the upper 8 Mbytes of virtual memory (ad-

dresses between FF800000 (hex) and FFFFFFFF (hex), in-

clusive), which are reserved by National Semiconductor

Corporation. Nevertheless, it is recommended that high-per-

formance peripheral devices and coprocessors be located

in a specific 8 Mbyte region of virtual memory (addresses

between FF000000 (hex) and FF7FFFFF (hex), inclusive),

that is dedicated for memory-mapped I/O. This is because

the NS32532 detects references to the dedicated locations

and serializes reads and writes. See Section 3.1.3.3. When

making I/O references to addresses outside the dedicated

region, external hardware must indicate to the NS32532

that special handling is required.

A

Byte at Address A

Two contiguous bytes are called a word. Except where not-

ed, the least significant byte of a word is stored at the lower

address, and the most significant byte of the word is stored

at the next higher address. In memory, the address of a

word is the address of its least significant byte, and a word

may start at any address.

15

8

7

0

a

A

1

A

MSB

LSB

In this case a small performance degradation will also re-

sult. Refer to Section 3.1.3.2 for more information on memo-

ry-mapped I/O.

Word at Address A

Two contiguous words are called a double-word. Except

where noted, the least significant word of a double-word is

Address (Hex)

00000000

Memory and I/O

FF000000

FF800000

Memory-Mapped I/O

Reserved by NSC

Interrupt Control

FFFFFE00

FFFFFFFF

FIGURE 2-9. NS32532 Address Mapping

11

2.0 Architectural Description (Continued)

2.3 MODULAR SOFTWARE SUPPORT

The Module Table is located within the first 64 kbytes of

virtual memory. This table contains a Module Descriptor

(also called a Module Table Entry) for each module in the

address space of the program. A Module Descriptor has

four 32-bit entries corresponding to each component of a

module:

The NS32532 provides special support for software mod-

ules and modular programs.

Each module in a NS32532 software environment consists

of three components:

1. Program Code Segment.

The Static Base entry contains the address of the begin-

ning of the module’s static data segment.

#

This segment contains the module’s code and constant

data.

The Link Table Base points to the beginning of the mod-

ule’s Link Table.

#

2. Static Data Segment.

Used to store variables and data that may be accessed

by all procedures within the module.

The Program Base is the address of the beginning of the

code and constant data for the module.

#

3. Link Table.

A fourth entry is currently unused but reserved.

#

The MOD Register in the CPU contains the address of the

Module Descriptor for the currently executing module.

This component contains two types of entries: Absolute

Addresses and Procedure Descriptors.

An Absolute Address is used in the external addressing

mode, in conjunction with a displacement and the current

MOD Register contents to compute the effective address

of an external variable belonging to another module.

The Static Base Register (SB) contains a copy of the Static

Base entry in the Module Descriptor of the currently execut-

ing module, i.e., it points to the beginning of the current

module’s static data area.

The Procedure Descriptor is used in the call external pro-

cedure (CXP) instruction to compute the address of an

external procedure.

This register is implemented in the CPU for efficiency pur-

poses. By having a copy of the static base entry or chip, the

CPU can avoid reading it from memory each time a data

item in the static data segment is accessed.

Normally, the linker program specifies the locations of the

three components. The Static Data and Link Table typically

reside in RAM; the code component can be either in RAM or

in ROM. The three components can be mapped into non-

contiguous locations in memory, and each can be indepen-

dently relocated. Since the Link Table contains the absolute

addresses of external variables, the linker need not assign

absolute memory addresses for these in the module itself;

they may be assigned at load time.

In an NS32532 software environment modules need not be

linked together prior to loading. As modules are loaded, a

linking loader simply updates the Module Table and fills the

Link Table entries with the appropriate values. No modifica-

tion of a module’s code is required. Thus, modules may be

stored in read-only memory and may be added to a system

independently of each other, without regard to their individu-

al addressing. Figure 2-10 shows a typical NS32532 run-

time environment.

To handle the transfer of control from one module to anoth-

er, the NS32532 uses a module table in memory and two

registers in the CPU.

TL/EE/9354–2

Note: Dashed lines indicate information copied to registers during transfer of control between modules.

FIGURE 2-10. NS32532 Run-Time Environment

12

2.0 Architectural Description (Continued)

2.4 MEMORY MANAGEMENT

Level-2 Page Tables contain 1024 32-bit Page Table en-

tries, and so occupy 4 Kbytes (1 page). Each Level-2 Page

Table Entry points to a final 4-Kbyte physical page frame. In

other words, its PFN provides the Page Frame Number por-

tion (bits 12–31) of the translated address (Figure 2-13 ).

The OFFSET field of the translated address is taken directly

from the corresponding field of the virtual address.

The Memory Mangement Unit of the NS32532 provides

support for demand-paged virtual memory. The MMU trans-

lates 32-bit virtual addresses into 32-bit physical addresses.

The page size is 4096 bytes.

The mapping from virtual to physical addresses is defined

by means of sets of tables in physical memory. These tables

are found by the MMU using one of its two Page Table Base

registers: PTB0 or PTB1. Which register is used depends on

the currently selected address space. See Section 2.4.2.

2.4.2 Virtual Address Spaces

When the Dual Space option is selected for address transla-

tion in the MCR (Section 2.1.5) the on-chip MMU uses two

maps: one for translating addresses presented to it in Su-

pervisor Mode and another for User Mode addresses. Each

map is referenced by the MMU using one of the two Page

Table Base registers: PTB0 or PTB1. The MMU determines

the map to be used by applying the following rules.

Translation efficiency is improved by means of an on-chip

64-entry translation look-aside buffer (TLB). Refer to Sec-

tion 3.4.4 for details.

If the MMU detects a protection violation or page fault while

translating an address for a reference required to execute

an instruction, a translation exception (Trap (ABT)) will re-

sult.

e

1) While the CPU is in Supervisor Mode (U/S pin

0), the

CPU is said to be generating virtual addresses belonging

to Address Space 0, and the MMU uses the PTB0 regis-

ter as its reference for looking up translations from mem-

ory.

2.4.1 Page Tables Structure

The page tables are arranged in a two-level structure, as

shown in Figure 2-11. Each of the MMU’s PTBn registers

may point to a Level-1 page table. Each entry of the Level-1

page table may in turn point to a Level-2 page table. Each

Level-2 page table entry contains translation information for

one page of the virtual space.

e

2) While the CPU is in User Mode (U/S pin

1), and the

MCR DS bit is set to enable Dual Space translation, the

CPU is said to be generating virtual addresses belonging

to Address Space 1, and the MMU uses the PTB1 regis-

ter to look up translations.

The Level-1 page table must remain in physical memory

while the PTBn register contains its address and translation

is enabled. Level-2 Page Tables need not reside in physical

memory permanently, but may be swapped into physical

memory on demand as is done with the pages of the virtual

space.

3) If Dual Space translation is not selected in the MCR,

there is no Adress Space 1, and all virtual addresses gen-

erated in both Supervisor and User modes are consid-

ered by the MMU to be in Address Space 0. The privilege

level of the CPU is used then only for access level check-

ing.

The Level-1 Page Table contains 1024 32-bit Page Table

Entries (PTE’s) and therefore occupies 4 Kbytes. Each entry

of the Level-1 Page Table contains a field used to construct

the physical base address of a Level-2 Page Table. This

field is a 20-bit PFN field, providing bits 12–31 of the physi-

cal address. The remaining bits (0–11) are assumed zero,

placing a Level-2 Page Table always on a 4-Kbyte (page)

boundary.

Note: When the CPU executes a Dual-Space Move instruction (MOVUSi or

MOVSUi), it temporarily enters User Mode by switching the state of

the U/S pin. Accesses made by the CPU during this time are treated

by the MMU as User-Mode accesses for both mapping and access

level checking. It is possible, however, to force the MMU to assume

Supervisor Mode privilege on such accesses by setting the Access

Override (AO) bit in the MCR (Section 2.1.5).

TL/EE/9354–3

FIGURE 2-11. Two-Level Page Tables

13

2.0 Architectural Description (Continued)

2.4.3 Page Table Entry Formats

R

Referenced. This is a status bit, set by the MMU and

cleared by the operating system, that indicates

whether the page mapped by this PTE has been ref-

erenced within a period of time determined by the

operating system. It is intended to assist in imple-

menting memory allocation strategies. In a Level-1

PTE, the R bit indicates only that the Level-2 Page

Table has been referenced for a translation, without

necessarily implying that the translation was suc-

cessful. In a Level-2 PTE, it indicates that the page

mapped by the PTE has been sucessfully referenced.

Figure 2-12 shows the formats of Level-1 and Level-2 Page

Table Entries (PTE’s).

The bits are defined as follows:

V

Valid. The V bit is set and cleared only by software.

l

e

V

1

The PTE is valid and may be used for

translation by the MMU.

l

e

V

0

The PTE does not represent a valid trans-

lation. Any attempt to use this PTE to trans-

late and address will cause the MMU to

generate an Abort trap.

l

e

R

1

The page has been referenced since the

R bit was last cleared.

PL Protection Level. This two-bit field establishes the

types of accesses permitted for the page in both User

Mode and Supervisor Mode, as shown in Table 2-1.

l

e

R

0

The page has not been referenced since

the R bit was last cleared.

The PL field is modified only by software. In a Level-1

PTE, it limits the maximum access level allowed for all

pages mapped through that PTE.

M

Modified. This is a status bit, set by the MMU when-

ever a write cycle is successfully performed to the

page mapped by this PTE. It is initialized to zero by

the operating system when the page is brought into

physical memory.

TABLE 2-1. Access Protection Levels

Protection Level Bits (PL)

U/S

l

e

M

1

The page has been modified since it was

last brought into physical memory.

Mode

00

no

01

no

10

11

l

e

M

0

The page has not been modified since it

was last brought into physical memory.

User

1

read

only

full

access access

access

In Level-1 Page Table Entries, this bit po-

sition is undefined, and is unaltered.

Supervisor

0

read

only

full

USR User bits. These bits are ignored by the MMU and

their values are not changed.

access access access

NU

CI

Not Used. These bits are reserved by National for

future enhancements. Their values should be set to

zero.

They can be used by the user software.

PFN Page Frame Number. This 20-bit field provides bits

12–31 of the physical address. See Figure 2-13.

Cache Inhibit. This bit appears only in Level-2 PTE’s.

It is used to specify non-cacheable pages.

31

12 11

9 8

0

PFN

USR

NU

R

NU

PL

V

First Level PTE

31

12 11

8

9

0

M

R

CI

NU

Second Level PTE

FIGURE 2-12. Page Table Entries (PTE’s)

14

2.0 Architectural Description (Continued)

TL/EE/9354–4

FIGURE 2-13. Virtual to Physical Address Translation

2.4.4 Physical Address Generation

by 4) to the base address taken from the Level-1 Page Ta-

ble Entry. The PFN field of the selected entry provides the

entire Page Frame Number of the translated address.

When a virtual address is presented to the MMU and the

translation information is not in the TLB, the MMU performs

a page table lookup in order to generate the physical ad-

dress.

The offset field of the virtual address is then appended to

this frame number to generate the final physical address.

The Page Table structure is traversed by the MMU using

fields taken from the virtual address. This sequence is dia-

grammed in Figure 2-13.

2.4.5. Address Translation Algorithm

The MMU either translates the 32-bit virtual address to a 32-

bit physical address or generates an abort trap to report a

translation error. The algorithm used by the MMU to perform

the translation is compatible with that of the NS32382. Re-

fer to Appendix C for differences between the two MMUs.

Bits 12–31 of the virtual address hold the 20-bit Page Num-

ber, which in the course of the translation is replaced with

the 20-bit Page Frame Number of the physical address. The

virtual Page Number field is further divided into two fields,

INDEX 1 and INDEX 2.

In the description that follows, the symbol ‘U’ takes the val-

ue 1 for a User-Mode memory reference. A reference is a

User-Mode reference in the following cases:

Bits 0–11 constitute the OFFSET field, which identifies a

byte’s position within the accessed page. Since the byte

position within a page does not change with translation, this

value is not used, and is simply echoed by the MMU as bits

0–11 of the final physical address.

1. The reference is performed while executing in User-

Mode.

2. The reference is for the source operand of a MOVUS

instruction.

The 10-bit INDEX 1 field of the virtual address is used as an

index into the Level-1 Page Table, selecting one of its 1024

entries. The address of the entry is computed by adding

INDEX 1 (scaled by 4) to the contents of the current Page

Table Base register. The PFN field of that entry gives the

base address of the selected Level-2 Page Table.

3. The reference is for the destination operand of a MOVSU

instruction.

The following notations are used in the algorithm.

A B

x

A concatenated with B

#

ll

A.B

x

B is a field inside register A

object pointed to by address A

The INDEX 2 field of the virtual address (10 bits) is used as

the index into the Level-2 Page Table, by adding it (scaled

(A)

(A).B

x

B field of the object pointed to by address A

15

2.0 Architectural Description (Continued)

Each access is associated with one of two Address Spaces

(AS), defined as follows:

Ð TEAR

w

virtual address,

e

Ð clock MSR with MSR.TEX

Ð terminate translation;

e

11,

10,

e

AS

U AND MCR.DS

e

If (PTE.V

0) then

#

If AS

1, Page Table Base Register 1 (PTB1) is used to

e

select the first-level page table. If AS

select the first-level page table.

0, PTB0 is used to

/* PTE2 Invalid */

virtual address,

Ð TEAR

w

Ð clock MSR with MSR.TEX

Ð terminate translation;

The access-level is a 2-bit value used to specify the privi-

lege level of an access. It is determined as follows:

e

BIT1

BIT0

U AND (NOT(MCR.A0))

#

e

If ((read AND NOT interlocked) AND PTE.R

Read-Modify-Write a double-word interlocked (PTE Poin-

0) then

#

1 for write, or read with ‘RMW’ status

0 otherwise

e

ter).R

1;

START TRANSLATION:

e

If ((write OR interlocked read) AND (PTE.R

0 OR

#

e

0)

PTE.M

terlocked (PTE Pointer).R

6. Generate Physical address:

0) then Read-Modify-Write a double-word in-

If (U

0 AND MCR.TS

0 OR U

1 AND MCR.TU

e

1;

1, (PTE Pointer).M

then

/* address translation disabled */

(physical address virtual address; CIOUT pin

0 in all MMU generated accesses */

physical address

w

PTE.CI

PTE.PFN OFFSET

ll

#

e

w

0);

CIOUT pin

w

e

/* Note: CIOUT

7. Update Translation Buffer:

else BEGIN /* (see also Figure 2-13 ) */

1. Select PTB:

Select entry for replacement;

#

TLB. Virtual Page Number

w

INDEX1 INDEX2;

ll

#

e

1) then

If (MCR.DS

1 AND U

PTB0, AS

#

Ð PTB

TLB.AS

w

TLB. Physical Frame Number

AS;

e

PTB1,

w

PTE.PFN

e

Ð AS

1;

TLB.PL

TLB.CI

TLB.M

w

Effective PL

PTE.CI

e

0);

else (PTB

#

2. Fetch first level PTE:

(PTE Pointer) .M

e

PTE Pointer

PTB.BASE ADDRESS INDEX1 00;

ll

#

ll

Enable entry

PTE

w

Effective PL

(PTE Pointer); /* Fetch PTE1 */

PTE.PL

END

w

Note 1: The TEAR and MSR are only updated when a Trap (ABT) occurs. It

is possible that the MMU detects a page fault or protection violation

on a reference for an instruction that is not executed, for example

on a prefetch. In that event, Trap (ABT) does not occur, and the

TEAR and MSR are not updated.

3. Validate First Level PTE:

k

If (PTE.PL

access level) then

#

/* Protection Exception */

Ð TEAR

w

Ð clock MSR with MSR.TEX

Ð terminate translation;

virtual address,

Note 2: If the MMU is translating a virtual address to check protection while

executing a RDVAL or WRVAL instruction, then Trap (ABT) occurs

only if the level-1 PTE is invalid and the access is permitted by the

PL-field. These instructions will not generate an abort if the F bit

value can be determined from Level-1 PTE.

e

11,

e

If (PTE.V

0) then

/* PTE1 Invalid */

virtual address,

#

2.5 INSTRUCTION SET

Ð TEAR

w

2.5.1 General Instruction Format

e

Ð clock MSR with MSR.TEX

Ð terminate translation;

01,

Figure 2-14 shows the general format of a Series 32000

instruction. The Basic Instruction is one to three bytes long

and contains the Opcode and up to two 5-bit General Ad-

dressing Mode (‘‘Gen’’) fields. Following the Basic Instruc-

tion field is a set of optional extensions, which may appear

depending on the instruction and the addressing modes se-

lected.

e

If (PTE.R

0) then

Ð Write a Byte (PTE Pointer) .R

Effective PL PTE.PL

4. Fetch second level PTE:

#

e

1;

w

#

e

PTE Pointer

PTE.PFN INDEX2 00;

ll

#

Index Bytes appear when either or both Gen fields specify

Scaled Index. In this case, the Gen field specifies only the

Scale Factor (1, 2, 4 or 8), and the Index Byte specifies

which General Purpose Register to use as the index, and

which addressing mode calculation to perform before index-

ing. See Figure 2-15.

ll

PTE

w

(PTE Pointer); /* Fetch PTE2 */

k

If (PTE.PL

effective PL) then

PTE.PL;

5. Validate Second Level PTE:

Ð Effective PL

w

k

If (PTE.PL

access level) then

#

/* Protection Exception */

16

2.0 Architectural Description (Continued)

TL/EE/9354–5

FIGURE 2-14. General Instruction Format

TL/EE/9354–6

FIGURE 2-15. Index Byte Format

Following Index Bytes come any displacements (addressing

constants) or immediate values associated with the select-

ed addressing modes. Each Disp/Imm field may contain

one or two displacements, or one immediate value. The size

of a Displacement field is encoded with the top bits of that

field, as shown in Figure 2-16, with the remaining bits inter-

preted as a signed (two’s complement) value. The size of an

immediate value is determined from the Opcode field. Both

Displacement and Immediate fields are stored most signifi-

cant byte first. Note that this is different from the memory

representation of data (Section 2.2).

PC, SP, SB or FP. These registers point to data areas gen-

erally needed by high-level languages.

b

a

Byte Displacement: Range 64 to 63

b

a

Word Displacement: Range 8192 to 8191

Some instructions require additional, ‘implied’’ immediates

and/or displacements, apart from those associated with ad-

dressing modes. Any such extensions appear at the end of

the instruction, in the order that they appear within the list of

operands in the instruction definition (Section 2.5.3).

2.5.2 Addressing Modes

The CPU generally accesses an operand by calculating its

Effective Address based on information available when the

operand is to be accessed. The method to be used in per-

forming this calculation is specified by the programmer as

an ‘‘addressing mode.’’

Double Word Displacement:

29 24 29

b

Range (2

b

a

b

1)*

2

) to

(2

Addressing modes are designed to optimally support high-

level language accesses to variables. In nearly all cases, a

variable access requires only one addressing mode, within

the instruction that acts upon that variable. Extraneous data

movement is therefore minimized.

Addressing Modes fall into nine basic types:

Register: The operand is available in one of the eight Gen-

eral Purpose Registers. In certain Slave Processor instruc-

tions, an auxiliary set of eight registers may be referenced

instead.

TL/EE/9354–7

Register Relative: A General Purpose Register contains an

address to which is added a displacement value from the

instruction, yielding the Effective Address of the operand in

memory.

FIGURE 2-16. Displacement Encodings

*Note: The pattern ‘‘11100000’’ for the most significant byte of the displace-

ment is reserved by National for future enhancements. Therefore, it

should never be used by the user program. This causes the lower

29

24

2

29

b

2 .

b

limit of the displacement range to be (2

b

) instead of

Memory Space: Identical to Register Relative above, ex-

cept that the register used is one of the dedicated registers

17

2.0 Architectural Description (Continued)

Memory Relative: A pointer variable is found within the

memory space pointed to by the SP, SB or FP register. A

displacement is added to that pointer to generate the Effec-

tive Address of the operand.

Format tables (Appendix A). The Instruction column gives

the instruction as coded in assembly language, and the De-

scription column provides a short description of the function

provided by that instruction. Further details of the exact op-

erations performed by each instruction may be found in the

Instruction Set Reference Manual.

Immediate: The operand is encoded within the instruction.

This addressing mode is not allowed if the operand is to be

written.

Notations:

Absolute: The address of the operand is specified by a

displacement field in the instruction.

e

i

Integer length suffix: B

Byte

W

D

Word

External: A pointer value is read from a specified entry of

the current Link Table. To this pointer value is added a dis-

placement, yielding the Effective Address of the operand.

Double Word

e

f

Floating Point length suffix: F

L

Standard Floating

Long Floating

Top of Stack: The currently-selected Stack Pointer (SP0 or

SP1) specifies the location of the operand. The operand is

pushed or popped, depending on whether it is written or

read.

e

gen

specified.

General operand. Any addressing mode can be

e

short

(see Appendix A for encodings).

A 4-bit value encoded within the Basic Instruction

Scaled Index: Although encoded as an addressing mode,

Scaled Indexing is an option on any addressing mode ex-

cept Immediate or another Scaled Index. It has the effect of

calculating an Effective Address, then multiplying any Gen-

eral Purpose Register by 1, 2, 4 or 8 and adding it into the

total, yielding the final Effective Address of the operand.

e

imm

ed after any addressing extensions.

Implied immediate operand. An 8-bit value append-

e

disp

bits. All three lengths legal.

Displacement (addressing constant): 8, 16 or 32

e

reg

areg

Configuration.

Any General Purpose Register: R0–R7.

Table 2-2 is a brief summary of the addressing modes. For a

complete description of their actions, see the Instruction Set

Reference Manual.

e

Any Processor Register: Address, Debug, Status,

e

mreg

Any Memory Management Register.

2.5.3 Instruction Set Summary

e

tion Dependent).

creg

A Custom Slave Processor Register (Implementa-

Table 2-3 presents a brief description of the NS32532 in-

struction set. The Format column refers to the Instruction

e

the Basic Instruction (see Appendix A for encodings).

cond

Any condition code, encoded as a 4-bit field within

18

2.0 Architectural Description (Continued)

TABLE 2-2. NS32532 Addressing Modes

ENCODING

Register

00000

MODE

ASSEMBLER SYNTAX

EFFECTIVE ADDRESS

Register 0

Register 1

Register 2

Register 3

Register 4

Register 5

Register 6

Register 7

R0, F0, L0

R1, F1, L1

R2, F2, L2

R3, F3, L3

R4, F4, L4

R5, F5, L5

R6, F6, L6

R7, F7, L7

None: Operand is in the

specified register.

00001

00010

00011

00100

00101

00110

00111

Register Relative

01000

a

Register.

Register 0 relative

Register 1 relative

Register 2 relative

Register 3 relative

Register 4 relative

Register 5 relative

Register 6 relative

Register 7 relative

disp(R0)

disp(R1)

disp(R2)

disp(R3)

disp(R4)

disp(R5)

disp(R6)

disp(R7)

Disp

01001

01010

01011

01100

01101

01110

01111

Memory Relative

10000

a

Pointer; Pointer found at

Frame memory relative

Stack memory relative

Static memory relative

disp2(disp1(FP))

disp2(disp1(SP))

disp2(disp1(SB))

Disp2

a

Register. ‘‘SP’’ is either

10001

address Disp1

10010

SP0 or SP1, as selected in PSR.

Reserved

10011

(Reserved for Future Use)

Immediate

10100

value

None. Operand is input from

instruction queue.

Absolute

10101

@

Absolute

External

disp

Disp.

External

10110

a

Pointer; Pointer is found

EXT(disp1)

TOS

disp2

Disp2

at Link Table Entry number Disp1.

Top of Stack

10111

Top of stack

Top of current stack, using either

User or Interrupt Stack Pointer,

as selected in PSR. Automatic

Push/Pop included.

Memory Space

11000

a

Register; ‘‘SP’’ is either

Frame memory

Stack memory

Static memory

Program memory

disp(FP)

disp(SP)

disp(SB)

Disp

11001

SP0 or SP1, as selected in PSR.

11010

a

disp

11011

*

Scaled Index

11100

a

[

]

Index, bytes

mode Rn:B

EA (mode)

Rn.

c

a

[

11101

Index, words

mode Rn:W

2

Rn.

c

[

11110

Index, double words

Index, quad words

mode Rn:D

4

c

[

mode Rn:Q

11111

8

‘‘Mode’ and ‘n’ are contained

within the Index Byte.

EA (mode) denotes the effective

address generated using mode.

19

2.0 Architectural Description (Continued)

TABLE 2-3. NS32532 Instruction Set Summary

MOVES

Format

Operation

MOVi

Operands

gen,gen

Description

4

2

7

4

Move a value.

MOVQi

short,gen

gen,gen,disp

gen,gen

Extend and move a signed 4-bit constant.

Move Multiple: disp bytes (1 to 16).

Move with zero extension.

Move with sign extension.

Move Effective Address.

MOVMi

MOVZBW

MOVZiD

MOVXBW

MOVXiD

ADDR

gen,gen

INTEGER ARITHMETIC

Format

Operation

ADDI

ADDQi

ADDCi

SUBi

Operands

gen,gen

short,gen

gen,gen

Description

4

2

4

6

7

Add.

Add signed 4-bit constant.

Add with carry.

Subtract.

SUBCi

NEGi

ABSi

Subtract with carry (borrow).

Negate (2’s complement).

Take absolute value.

Multiply.

MULi

QUOi

REMi

DIVi

Divide, rounding toward zero.

Remainder from QUO.

Divide, rounding down.

Remainder from DIV (Modulus).

Multiply to Extended Integer.

Divide Extended Integer.

MODi

MEIi

DEIi

PACKED DECIMAL (BCD) ARITHMETIC

Format

Operation

ADDPi

Operands

gen,gen

Description

6

Add Packed.

SUBPi

Subtract Packed.

INTEGER COMPARISON

Format

Operation

CMPi

Operands

gen,gen

Description

4

2

7

Compare.

CMPQi

CMPMi

short,gen

gen,gen,disp

Compare to signed 4-bit constant.

Compare Multiple: disp bytes (1 to 16).

LOGICAL AND BOOLEAN

Format

Operation

ANDi

Operands

gen,gen

gen

Description

4

6

2

Logical AND.

ORi

Logical OR.

BICi

Clear selected bits.

XORi

Logical Exclusive OR.

Complement all bits.

COMi

NOTi

Boolean complement: LSB only.

Save condition code (cond) as a Boolean variable of size i.

Scondi

SHIFTS

Format

Operation

LSHi

Operands

gen,gen

Description

6

Logical Shift, left or right.

Arithmetic Shift, left or right.

Rotate, left or right.

ASHi

ROTi

20

2.0 Architectural Description (Continued)

TABLE 2-3. NS32532 Instruction Set Summary (Continued)

BITS

Format

Operation

TBITi

Operands

gen,gen

Description

4

6

8

Test bit.

SBITi

Test and set bit.

SBITIi

CBITi

CBITIi

IBITi

Test and set bit, interlocked.

Test and clear bit.

Test and clear bit, interlocked.

Test and invert bit.

Find first set bit.

FFSi

BIT FIELDS

Bit fields are values in memory that are not aligned to byte boundaries. Examples are PACKED arrays and records used in

Pascal. ‘‘Extract’’ instructions read and align a bit field. ‘‘Insert’’ instructions write a bit field from an aligned source.

Format

Operation

EXTi

Operands

Description

8

7

8

reg,gen,gen,disp

gen,gen,imm,imm

reg,gen,gen

Extract bit field (array oriented).

Insert bit field (array oriented).

Extract bit field (short form).

Insert bit field (short form).

Convert to Bit Field Pointer.

INSi

EXTSi

INSSi

CVTP

ARRAYS

Format

Operation

CHECKi

INDEXi

Operands

reg,gen,gen

Description

8

Index bounds check.

Recursive indexing step for multiple-dimensional arrays.

STRINGS

String instructions assign specific functions to

the General Purpose Registers:

R4 - Comparison Value

Options on all string instructions are:

B (Backward):

Decrement string pointers after each step

rather than incrementing.

End instruction if String 1 entry

matches R4.

R3 - Translation Table Pointer

R2 - String 2 Pointer

U (Until match):

W (While match):

R1 - String 1 Pointer

End instruction if String 1 entry

does not match R4.

R0 - Limit Count

All string instructions end when R0 decrements to zero.

Description

Format

Operation

MOVSi

MOVST

CMPSi

Operands

options

5

Move String 1 to String 2.

Move string, translating bytes.

Compare String 1 to String 2.

Compare translating, String 1 bytes.

Skip over String 1 entries.

5

CMPST

SKPSi

SKPST

Skip, translating bytes for Until/While.

21

2.0 Architectural Description (Continued)

TABLE 2-3. NS32532 Instruction Set Summary (Continued)

JUMPS AND LINKAGE

Format

Operation

JUMP

BR

Operands

gen

Description

3

0

3

2

3

1

3

1

Jump.

disp

Branch (PC Relative).

Bcond

CASEi

ACBi

JSR

disp

Conditional branch.

gen

Multiway branch.

short,gen,disp

gen

Add 4-bit constant and branch if non-zero.

Jump to subroutine.

BSR

disp

Branch to subroutine.

CXP

disp

Call external procedure.

CXPD

SVC

gen

Call external procedure using descriptor.

Supervisor Call.

FLAG

BPT

Flag Trap.

Breakpoint Trap.

[

]

ENTER

EXIT

reg list ,disp

Save registers and allocate stack frame (Enter Procedure).

Restore registers and reclaim stack frame (Exit Procedure).

Return from subroutine.

]

reg list

RET

disp

RXP

Return from external procedure call.

Return from trap. (Privileged)

Return from interrupt. (Privileged)

RETT

RETI

CPU REGISTER MANIPULATION

Format

Operation

SAVE

Operands

Description

[

]

1

2

reg list

Save General Purpose Registers.

RESTORE

LPRi

reg list

Restore General Purpose Registers.

Load Processor Register. (Privileged if PSR, INTBASE, USP, CFG

or Debug Registers).

areg,gen

2

SPRi

areg,gen

Store Processor Register. (Privileged if PSR, INTBASE, USP, CFG

or Debug Registers).

3

5

ADJSPi

gen

Adjust Stack Pointer.

BISPSRi

BICPSRi

SETCFG

Set selected bits in PSR. (Privileged if not Byte length)

Clear selected bits in PSR. (Privileged if not Byte length)

Set Configuration Register. (Privileged)

[

]

option list

FLOATING POINT

Format

11

9

Operation

Operands

gen,gen

gen

Description

MOVf

Move a Floating Point value.

MOVLF

MOVFL

MOVif

ROUNDfi

TRUNCfi

FLOORfi

ADDf

Move and shorten a Long value to Standard.

9

Move and lengthen a Standard value to Long.

9

Convert any integer to Standard or Long Floating.

9

Convert to integer by rounding.

9

Convert to integer by truncating, toward zero.

9

Convert to largest integer less than or equal to value.

11

12

9

Add.

SUBf

Subtract.

MULf

Multiply.

DIVf

Divide.

CMPf

Compare.

NEGf

Negate.

ABSf

Take absolute value.

Polynomial Step.

Dot Product.

Binary Scale.

Binary Log.

Square Root

Multiply and Accumulate

Load FSR.

POLYf

DOTf

SCALBf

LOGBf

SQRTf

MACf

LFSR

9

SFSR

gen

Store FSR.

22

2.0 Architectural Description (Continued)

TABLE 2-3. NS32532 Instruction Set Summary (Continued)

MEMORY MANAGEMENT

Format

Operation

LMR

Operands

mreg,gen

gen

Description

14

8

Load Memory Management Register. (Privileged)

Store Memory Management Register. (Privileged)

Validate address for reading. (Privileged)

Validate address for writing. (Privileged)

Move a value from Supervisor

SMR

RDVAL

WRVAL

MOVSUi

gen

gen,gen

Space to User Space. (Privileged)

Move a value from User Space

8

MOVUSi

gen,gen

to Supervisor Space. (Privileged)

MISCELLANEOUS

Format

Operation

Operands

Description

1

NOP

WAIT

DIA

No Operation.

Wait for interrupt.

Diagnose. Single-byte ‘‘Branch to Self’’ for hardware

breakpointing. Not for use in programming.

Cache Invalidate. (Privileged)

[ ]

options ,gen

14

CINV

CUSTOM SLAVE

Operation

Format

15.5

15.1

15.0

Operands

gen,gen

gen

Description

CCAL0c

CCAL1c

CCAL2c

CCAL3c

CMOV0c

CMOV1c

CMOV2c

CMOV3c

CCMP0c

CCMP1c

CCV0ci

CCV1ci

CCV2ci

CCV3ic

CCV4DQ

CCV5QD

LCSR

Custom Calculate.

Custom Move.

Custom Compare.

Custom Convert.

Load Custom Status Register.

Store Custom Status Register.

Load Custom Register. (Privileged)

Store Custom Register. (Privileged)

SCSR

gen

LCR

creg,gen

SCR

23

3.0 Functional Description

This chapter provides details on the functional characteris-

tics of the NS32532 microprocessor.

The chapter is divided into five main sections:

Instruction Execution, Exception Processing, Debugging,

On-Chip Caches and System Interface.

3.1 INSTRUCTION EXECUTION

To execute an instruction, the NS32532 performs the fol-

lowing operations:

Fetch the instruction

#

Read source operands, if any (1)

#

Calculate results

#

Write result operands, if any

#

Modify flags, if necessary

#

Update the program counter

#

Under most circumstances, the CPU can be conceived to

execute instructions by completing the operations above in

strict sequence for one instruction and then beginning the

sequence of operations for the next instruction. However,

due to the internal instruction pipelining, as well as the oc-

currence of exceptions, the sequence of operations per-

formed during the execution of an instruction may be al-

tered. Furthermore, exceptions also break the sequentiality

of the instructions executed by the CPU.

Details on the effects of the internal pipelining, as well as

the occurrence of exceptions on the instruction execution,

are provided in the following sections.

TL/EE/9354–8

FIGURE 3-1. Operating States

tion is detected, the CPU enters the Processing-An-Excep-

tion state.

Note: 1 In this and following sections, memory locations read by the CPU to

calculate effective addresses for Memory-Relative and External ad-

dressing modes are considered like source operands, even if the

effective address is being calculated for an operand with access

class of write.

The CPU enters the Halted state when a bus error or abort

is detected while the CPU is processing an exception, there-

by preventing the transfer of control to an appropriate ex-

ception service procedure. The CPU remains in the Halted

state until reset occurs. A special status identifying this state

is presented on the system interface.

3.1.1 Operating States

The CPU has five operating states regarding the execution

of instructions and the processing of exceptions: Reset, Ex-

ecuting Instructions, Processing An Exception, Waiting-For-

An-Interrupt, and Halted. The various states and transitions

between them are shown in Figure 3-1.

Note: When the Direct-Exception mode is enabled, the CPU does not save

the MOD Register contents nor does it read the module linkage infor-

mation for the exception service procedure. Refer to Section 3.2 for

details.

Whenever the RST signal is asserted, the CPU enters the

reset state. The CPU remains in the reset state until the

RST signal is driven inactive, at which time it enters the

Executing-Instructions state. In the Reset state the contents

of certain registers are initialized. Refer to Section 3.5.3 for

details.

3.1.2 Instruction Endings

The NS32532 checks for exceptions at various points while

executing instructions. Certain exceptions, like interrupts,

are in most cases recognized between instructions. Other

exceptions, like Divide-By-Zero Trap, are recognized during

execution of an instruction. When an exception is recog-

nized during execution of an instruction, the instruction ends

in one of four possible ways: completed, suspended, termi-

nated, or partially completed. Each type of exception caus-

es a particular ending, as specified in Section 3.2.

In the Executing-Instructions state, the CPU executes in-

structions. It will exit this state when an exception is recog-

nized or a WAIT instruction is encountered. At which time it

enters the Processing-An-Exception state or the Waiting-

For-An-Interrupt state respectively.

While in the Processing-An-Exception state, the CPU saves

the PC, PSR and MOD register contents on the stack and

reads the new PC and module linkage information to begin

execution of the exception service procedure (see note).

3.1.2.1 Completed Instructions

When an exception is recognized after an instruction is

completed, the CPU has performed all of the operations for

that instruction and for all other instructions executed since

the last exception occurred. Result operands have been

written, flags have been modified, and the PC saved on the

Interrupt Stack contains the address of the next instruction

to execute. The exception service procedure can, at its con-

clusion, execute the RETT instruction (or the RETI instruc-

tion for vectored interrupts), and the CPU will begin execut-

ing the instruction following the completed instruction.

Following the completion of all data references required to

process an exception, the CPU enters the Executing-In-

structions state.

In the Waiting-For-An-Interrupt state, the CPU is idle. A spe-

cial status identifying this state is presented on the system

interface (Section 3.5). When an interrupt or a debug condi-

24

3.0 Functional Description (Continued)

3.1.2.2 Suspended Instructions

are the contents of the Stack Pointers. The result operands

of other instructions executed since the last serializing oper-

ation may not have been written to memory. A terminated

instruction cannot be completed.

An instruction is suspended when one of several trap condi-

tions or a restartable bus error is detected during execution

of the instruction. A suspended instruction has not been

completed, but all other instructions executed since the last

exception occurred have been completed. Result operands

and flags due to be affected by the instruction may have

been modified, but only modifications that allow the instruc-

tion to be executed again and completed can occur. For

certain exceptions (Trap (ABT), Trap (UND), Trap (ILL), and

bus errors) the CPU clears the P-flag in the PSR before

saving the copy that is pushed on the Interrupt Stack. The

PC saved on the Interrupt Stack contains the address of the

suspended instruction.

3.1.2.4 Partially Completed Instructions

When a restartable bus error, interrupt, abort, or debug con-

dition is recognized during execution of a string instruction,

the instruction is said to be partially completed. A partially

completed instruction has not completed, but all other in-

structions executed since the last exception occurred have

been completed. Result operands and flags due to be af-

fected by the instruction may have been modified, but the

values stored in the string pointers and other general-pur-

pose registers used during the instruction’s execution allow

the instruction to be executed again and completed.

For example, the RESTORE instruction pops up to 8 gener-

al-purpose registers from the stack. If an invalid page table

entry is detected on one of the references to the stack, then

the instruction is suspended. The general-purpose registers

due to be loaded by the instruction may have been modified,

but the stack pointer still holds the same value that it did

when the instruction began.

The CPU clears the P-flag in the PSR before saving the

copy that is pushed on the Interrupt Stack. The PC saved on

the Interrupt Stack contains the address of the partially

completed instruction. The exception service procedure

can, at its conclusion, simply execute the RETT instruction

(or the RETI instruction for vectored interrupts), and the

CPU will resume executing the partially completed instruc-

tion.

To complete a suspended instruction, the exception service

procedure takes either of two actions:

1. The service procedure can simulate the suspended in-

struction’s execution. After calculating and writing the in-

struction’s results, the flags in the PSR copy saved on the

Interrupt Stack should be modified, and the PC saved on

the Interrupt Stack should be updated to point to the next

instruction to execute. The service procedure can then

execute the RETT instruction, and the CPU begins exe-

cuting the instruction following the suspended instruction.

This is the action taken when floating-point instructions

are simulated by software in systems without a hardware

floating-point unit.

3.1.3 Instruction Pipeline

The NS32532 executes instructions in a heavily pipelined

fashion. This allows a significant performance enhancement

since the operations of several instructions are performed

simultaneously rather than in a strictly sequential manner.

The CPU provides a four-stage internal instruction pipeline.

As shown in Figure 3-2, a write buffer, that can hold up to

two operands, is also provided to allow write operations to

be performed off-line.

2. The suspended instruction can be executed again after

the service procedure has eliminated the trap condition

that caused the instruction to be suspended. The service

procedure should execute the RETT instruction at its con-

clusion; then the CPU begins executing the suspended

instruction again. This is the action taken by a debugger

when it encounters a BPT instruction that was temporarily

placed in another instruction’s location in order to set a

breakpoint.

Note 1: Although the NS32532 allows a suspended instruction to be execut-

ed again and completed, the CPU may have read a source operand

for the instruction from a memory-mapped peripheral port before

the exception was recognized. In such a case, the characteristics of

the peripheral device may prevent correct reexecution of the in-

struction.

Note 2: It may be necessary for the exception service procedure to alter the

P-flag in the PSR copy saved on the Interrupt Stack: If the exception

service procedure simulates the suspended instruction and the P-

flag was cleared by the CPU before saving the PSR copy, then the

saved T-flag must be copied to the saved P-flag (like the floating-

point instruction simulation described above). Or if the exception

service procedure executes the suspended instruction again and

the P-flag was not cleared by the CPU before saving the PSR copy,

then the saved P-flag must be cleared (like the breakpoint trap de-

scribed above). Otherwise, no alteration to the saved P-flag is nec-

essary.

TL/EE/9354–9

FIGURE 3-2. NS32532 Internal Instruction Pipeline

3.1.2.3 Terminated Instructions

Due to the pipelining, operations like fetching one instruc-

tion, reading the source operands of a second instruction,

calculating the results of a third instruction and storing the

results of a fourth instruction, can all occur in parallel.

An instruction being executed is terminated when reset or a

nonrestartable bus error occurs. Any result operands and

flags due to be affected by the instruction are undefined, as

25

3.0 Functional Description (Continued)

The order of memory references performed by the CPU may

also differ from that related to a strictly sequential instruc-

tion execution. In fact, when an instruction is being execut-

ed, some of the source operands may be read from memory

before the instruction is completely fetched. For example,

the CPU may read the first source operand for an instruction

before it has fetched a displacement used in calculating the

address of the second source operand. The CPU, however,

always completes fetching an instruction and reading its

source operands before writing its results. When more than

one source operand must be read from memory to execute

an instruction, the operands may be read in any order. Simi-

larly, when more than one result operand is written to mem-

ory to execute an instruction, the operands may be written

in any order.

formed in the order implied by the program. Refer to Section

3.1.3.2 for details.

It is also to be noted that the CPU does not check for de-

pendencies between the fetching of an instruction and the

writing of previous instructions’ results. Therefore, special

care is required when executing self-modifying code.

3.1.3.1 Branch Prediction

One problem inherent to all pipelined machines is what is

called ‘‘Pipeline Breakage’’.

This occurs every time the sequentiality of the instructions is

broken, due to the execution of certain instructions or the

occurrence of exceptions.

The result of a pipeline breakage is a performance degrada-

tion, due to the fact that a certain portion of the pipeline

must be flushed and new data must be brought in.

An instruction is fetched only after all previous instructions

have been completely fetched. However, the CPU may be-

gin fetching an instruction before all of the source operands

have been read and results written for previous instructions.

The NS32532 provides a special mechanism, called branch

prediction, that helps minimize this performance penalty.

When a conditional branch instruction is decoded in the ear-

ly stages of the pipeline, a prediction on the execution of the

instruction is performed.

The source operands for an instruction are read only after

all previous instructions have been fetched and their source

operands read. A source operand for an instruction may be

read before all results of previous instructions have been

written, except when the source operand’s value depends

on a result not yet written. The CPU compares the physical

address and length of a source operand with those of any

results not yet written, and delays reading the source oper-

and until after writing all results on which the source oper-

and depends. Also, the CPU ensures that the interlocked

read and write references to execute an SBITIi or CBITIi

instruction occur after writing all results of previous instruc-

tions and before reading any source operands for subse-

quent instructions.

More precisely, the prediction mechanism predicts back-

ward branches as taken and forward branches as not taken,

except for the branch instructions BLE and BNE that are

always predicted as taken.

Thus, the resulting probability of correct prediction is fairly

high, especially for branch instructions placed at the end of

loops.

The sequence of operations performed by the loader and

execution units in the CPU is given below:

Loader detects branches and calculates destination ad-

dresses

#

The result operands for an instruction are written after all

results of previous instructions have been written.

Loader uses branch opcode and direction to select be-

tween sequential and non-sequential streams

#

The description above is summarized in Figure 3-3, which

shows the precedence of memory references for two con-

secutive instructions.

Loader saves address for alternate stream

#

Execution unit resolves branch decision

Due to the branch predicition, some special care is required

when writing self-modifying code. Refer to the appropriate

section in Appendix B for more information on this subject.

3.1.3.2 Memory-Mapped I/O

The characteristics of certain peripheral devices and the

overlapping of instruction execution in the pipeline of the

NS32532 require that special handling be applied to memo-

ry-mapped I/O references. I/O references differ from mem-

ory references in two significant ways, imposing the follow-

ing requirements:

TL/EE/9354–10

FIGURE 3-3. Memory References for

Consecutive Instructions

1. Reading from a peripheral port can alter the value read

on the next reference to the same port or another port in

the same device. (A characteristic called here ‘‘destruc-

tive-reading’’.) Serial communication controllers and

FIFO buffers commonly operate in this manner. As ex-

plained in ‘‘Instruction Pipeline’’ above, the NS32532 can

read the source operands for one instruction while the

previous instruction is executing. Because the previous

instruction may cause a trap, an interrupt may be recog-

nized, or the flow of control may be otherwise altered, it is

a requirement that destructive-reading of source oper-

ands before the execution of an instruction be avoided.

(An arrow from one reference to another indicates that

the first reference always precedes the second.)

Another consequence of overlapping the operations for sev-

eral instructions, is that the CPU may fetch an instruction

and read its source operands, even though the instruction is

not executed (e.g., due to the occurrence of an exception).

In such a case, the MMU may update the R-bit in Page

Table Entries used in referring to the fetched instruction and

its source operands.

Special care is needed in the handling of memory-mapped

I/O devices. The CPU provides special mechanisms to en-

sure that the references to these devices are always per-

26

3.0 Functional Description (Continued)

2. Writing to a peripheral port can alter the value read from

another port of the same device. (A characteristic called

here ‘‘side-effects of writing’’). For example, before read-

ing the counter’s value from the NS32202 Interrupt Con-

trol Unit it is first necessary to freeze the value by writing

to another control register.

serializing operation takes place. This is necessary since

the privilege level might have changed and the instructions

following the LPRW instruction must be fetched again with

the new privilege level and possibly with a different MMU

mapping. See Section 2.4.2.

The CPU serializes instruction execution after executing one

of the following instructions: BICPSRW, BISPSRW, BPT,

CINV, DIA, FLAG (trap taken), LMR, LPR (CFG, INTBASE,

PSR, UPSR, DCR, BPC, DSR, and CAR only), RETT, RETI,

and SVC. Figure 3-4 shows the memory references after

serialization.

However, as mentioned above, the NS32532 can read the

source operands for one instruction before writing the re-

sults of previous instructions unless the addresses indicate

a dependency between the read and write references. Con-

sequently, it is a requirement that read and write references

to peripheral that exhibit side-effects of writing must occur in

the order dictated by the instructions.

Note 1: LPRB UPSR can be executed in User Mode to serialize instruction

execution.

Note 2: After an instruction that writes a result to memory is executed, the

updating of the result’s memory location may be delayed until the

next serializing operation.

The NS32532 supports 2 methods for handling memory-

mapped I/O. The first method is more general; it satisfies

both requirements listed above and places no restriction on

the location of memory-mapped peripheral devices. The

second method satisfies only the requirement for side ef-

fects of writing, and it restricts the location of memory-

mapped I/O devices, but it is more efficient for devices that

do not have destructive-read ports.

Note 3: When reset or a nonrestartable bus error exception occurs, the CPU

discards any results that have not yet been written to memory.

The first method for handling memory-mapped I/O uses two

signals: IOINH and IODEC. When the NS32532 generates a

read bus cycle, it asserts the output signal IOINH if either of

the I/O requirements listed above is not satisfied. That is,

IOINH is asserted during a read bus cycle when (1) the read

reference is for an instruction that may not be executed or

(2) the read reference occurs while a write reference is

pending for a previous instruction. When the read reference

is to a peripheral device that implements ports with destruc-

tive-reading or side-effects of writing, the input signal

IODEC must be asserted; in addition, the device must not

be selected if IOINH is active. When the CPU detects that

the IODEC input signal is active while the IOINH output sig-

nal is also active, it discards the data read during the bus

cycle and serializes instruction execution. See the next sec-

tion for details on serializing operations. The CPU then gen-

erates the read bus cycle again, this time satisfying the re-

quirements for I/O and driving IOINH inactive.

TL/EE/9354–11

FIGURE 3-4. Memory References after Serialization

3.1.4 Slave Processor Instructions

The NS32532 recognizes two groups of instructions being

executable by external slave processors:

Floating Point Instructions

#

Custom Slave Instructions

#

Each Slave Instruction Set is enabled by a bit in the Configu-

ration Register (Section 2.1.4). Any Slave Instruction which

does not have its corresponding Configuration Register bit

set will trap as undefined, without any Slave Processor com-

munication attempted by the CPU. This allows software sim-

ulation of a non-existent Slave Processor.

Note that the Memory Management Instructions, like Float-

ing Point and Custom Slave Instructions, have to be en-

abled through an appropriate bit in the configuration register

in order to be executable.

The second method for handling memory-mapped I/O uses

a dedicated region of virtual memory. The NS32532 treats

all references to the memory range from address FF000000

to address FFFFFFFF inclusive in a special manner.

However, they are not considered here as Slave Instruc-

tions, since the NS32532 integrates the MMU on-chip and

the execution of them does not follow the protocol of the

Slave Instructions.

While a write to a location in this range is pending, reads

from locations in the same range are delayed. However,

reads from locations with addresses lower than FF000000

may occur. Similarly, reads from locations in the above

range may occur while writes to locations outside of the

range are pending.

3.1.4.1 Regular Slave Instruction Protocol

Slave Processor instructions have a three-byte Basic In-

struction field, consisting of an ID Byte followed by an Oper-

ation Word. The ID Byte has three functions:

It is to be noted that the CPU may assert IOINH even when

the reference is within the dedicated region. Refer to Sec-

tion 3.5.8 for more information on the handling of I/O devic-

es.

1) It identifies the instruction as being a Slave Processor

instruction.

2) It specifies which Slave Processor will execute it.

3.1.3.3 Serializing Operations

3) It determines the format of the following Operation Word

of the instruction.

After executing certain instructions or processing an excep-

tion, the CPU serializes instruction execution. Serializing in-

struction execution means that the CPU completes writing

all previous instructions’ results to memory, then begins

fetching and executing the next instruction.

Upon receiving a Slave Processor instruction, the CPU initi-

ates the sequence outlined in Figure 3-5 . While applying

Status code 11111 (Broadcast ID Section 3.5.4.1), the CPU

transfers the ID Byte on bits D24–D31, the operation

For example, when a new value is loaded into the PSR by

executing an LPRW instruction, the pipeline is flushed and a

27

3.0 Functional Description (Continued)

TL/EE/9354–12

FIGURE 3-5. Regular Slave Instruction Protocol: CPU Actions

28

3.0 Functional Description (Continued)

31

0

ID BYTE

OPCODE (LOW)

OPCODE (HIGH)

XXXXXXXX

FIGURE 3-6. ID and Operation Word

31

15

TS

7

0

ZERO

N

Z

0

L

0

Q

FIGURE 3-7. Slave Processor Status Word

word on bits D8–D23 in a swapped order of bytes and a

e

3.1.4.2 Pipelined Slave Instruction Protocol

non-used byte XXXXXXXX (X

(Figure 3-6 ).

don’t care) on bits D0–D7

In order to increase performance of floating-point instruc-

tions while maintaining full software compatibility with the

Series 32000 architecture, the NS32532 incorporates a

pipelined floating-point protocol. This protocol is designed

to operate in conjunction with the NS32580 FPC, or any

other floating-point slave which conforms to the protocol

and the Series 32000 architecture. The protocol is enabled

by the PF bit in the CFG register.

All slave processors observe the bus cycle and inspect the

identification code. The slave selected by the identification

code continues with the protocol; other slaves wait for the

next slave instruction to be broadcast.

After transferring the slave instruction, the CPU sends to the

slave any source operands that are located in memory or

the General-Purpose registers. The CPU then waits for the

slave to assert SDN or FSSR. While the CPU is waiting, it

can perform bus cycles to fetch instructions and read

source operands for instructions that follow the slave in-

struction being executed. If there are no bus cycles to per-

form, the CPU is idle with a special Status indicating that it is

waiting for a slave processor. After the slave asserts SDN or

FSSR, the CPU follows one of the two sequences described

below.

The basic methods of transferring data and control informa-

tion between the CPU and the FPC, are the same as in the

regular slave protocol.

However, in pipelined mode, the CPU may send a new float-

ing-point instruction to the FPC before the previous instruc-

tion has been completed.

Although the CPU can advance as many as four floating-

point instructions before receiving a completion pulse on

SDN for the first instruction, full exception recovery is as-

sured. This is accomplished through a FIFO mechanism

which maintains the addresses of all the floating-point in-

structions sent to the FPC for execution.

If the slave asserts SDN, then the CPU checks whether the

instruction stores any results to memory or the General-Pur-

pose registers. The CPU reads any such results from the

slave by means of 1 or 2 bus cycles and updates the desti-

nation.

Pipelined execution can occur only for instructions which do

not require a result to be read from the FPC.

If the slave asserts FSSR, then the NS32532 reads a 32-bit

status word from the slave. The CPU checks bit 0 in the

slave’s status word to determine whether to update the PSR

flags or to process an exception. Figure 3-7 shows the for-

mat of the slave’s status word.

In cases where a result is to be read back, the CPU will wait

for instruction completion before issuing the next instruc-

tion. Floating-point instructions can be divided into two

groups, depending on the amount of pipelining permitted.

Group A. Fully-Pipelined Instructions

If the Q bit in the status word is 0, the CPU updates the N, Z

and L flags in the PSR.

Instructions in this group can be sent to the FPC before

previous group A instructions are completed. No instruction

completion indication from the FPC is required in order to

continue to another group A or group B instruction.

If the Q bit in the status word is set to 1, the CPU processes

either a Trap (UND) if TS is 1 or a Trap (SLAVE) if TS is 0.

Note 1: Only the floating-point and custom compare instructions are allowed

to return a value of 0 for the Q bit when the FSSR signal is activat-

ed. All other instructions must always set the Q bit to 1 (to signal a

Trap), when activating FSSR.

Group A contains floating-point instructions satisfying all of

the following conditions.

1. The destination operand is in a floating-point register.

2. The source operand is not of type TOS or IMM.

3. The instruction format is either 11 or 12.

Group B. Half-Pipelined Instructions

Note 2: While executing an LMR or CINV instruction, the CPU displays the

operation code and source operand using slave processor write bus

cycles, as described in the protocol above. Nevertheless, the CPU

does not wait for SDN or FSSR to be asserted while executing

these instructions. This information can be used to monitor the con-

tents of the on-chip TLB, Instruction Cache, and Data Cache.

Group B instructions can begin execution before previous

group A instructions are completed. However, they cannot

complete before the FPC signals completion of all the previ-

ous floating-point instructions.

Note 3: The slave processor must be ready to accept new slave instruction

at any time, even while the slave is executing another instruction or

waiting for the CPU to read results. For example, the CPU may

terminate an instruction being executed by a slave because a non-

restartable bus error is detected while the MMU is updating a Page

Table Entry for an instruction being prefetched.

Group B contains floating-point instructions satisfying at

least one of the following conditions.

Note 4: If a slave instruction stores a result to memory, the CPU checks

whether Trap (ABT) would occur on the store operation before read-

ing the result from the slave. For quad-word destination operands,

the CPU checks that both double-words of the destination can be

stored without an abort before reading either double-word of the

result from the slave.

1. The destination operand is either in memory or in a CPU

register (this includes the CMPf instruction which modifies

the PSR register).

2. The source operand is of type TOS or IMM.

3. The instruction format is 9.

29

3.0 Functional Description (Continued)

TL/EE/9354–73

FIGURE 3-8. Instruction Flow in Pipelined Floating-Point Mode

30

3.0 Functional Description (Continued)

Note: Non-floating-point instructions cannot be pipelined. They can begin

execution only after all other instructions have been completed. The

CPU cannot proceed to other instructions before their execution is

completed.

The Returned Value Type and Destination column gives the

size of any returned value and where the CPU places it. The

PSR-Bits-Affected column indicates which PSR bits, if any,

are updated from the Slave Processor Status Word (Figure

3-7).

3.1.4.3 Instruction Flow and Exceptions

When operating in pipelined mode, the CPU will push the

address of group A instructions into a five-entry FIFO after

the ID, opcode and source operands have been sent to the

FPC. The address will be pushed into the FIFO only if no

exception is detected during the transfer of the source oper-

ands needed for the execution of the instruction.

Any operand indicated as being of type ‘‘f’’ will not cause a

transfer if the Register addressing mode is specified. This is

because the Floating Point Registers are physically on the

Floating Point Unit and are therefore available without CPU

assistance.

3.1.4.5 Custom Slave Instructions

Group A instructions are only stalled when the FIFO is full,

in which case the CPU will wait before sending the next

instruction. Group B instructions can begin execution while

some entries are still in the FIFO, but cannot complete be-

fore the FIFO is empty (i.e., before all previous instructions

are completed). Non-floating-point instructions cannot begin

execution until the FIFO is empty. When a normal comple-

tion indication is received, the instruction address at the bot-

tom of the FIFO is dropped. If a trap indication is received

and the FIFO is not empty, the instruction address at the

bottom of the FIFO is copied to the PC register and the

floating-point exception is serviced. The remaining entries in

the FIFO are discarded.

Provided in the NS32532 is the capability of communicating

with a user-defined, ‘‘Custom’’ Slave Processor. The in-

struction set provided for a Custom Slave Processor defines

the instruction formats, the operand classes and the com-

munication protocol. Left to the user are the interpretations

of the Op Code fields, the programming model of the Cus-

tom Slave and the actual types of data transferred. The pro-

tocol specifies only the size of an operand, not its data type.

Table 3-2 lists the relevant information for the Custom Slave

instruction set. The designation ‘‘c’’ is used to represent an

operand which can be a 32-bit (‘‘D’’) or 64-bit (‘‘Q’’) quantity

in any format; the size is determined by the suffix on the

mnemonic. Similarly, an ‘‘i’’ indicates an integer size (Byte,

Word, Double Word) selected by the corresponding mne-

monic suffix.

A floating-point exception may be received and serviced at

any time after the CPU has sent the ID and opcode for the

first instruction and until the FPC has signalled completion

for the last instruction.

Any operand indicated as being of type ‘‘c’’ will not cause a

transfer if the register addressing mode is specified. It is

assumed in this case that the slave processor is already

holding the operand internally.

Other exceptions may occur while the FIFO is not empty.

This may be the case when an interrupt is received or a

translation exception is detected in the access of an oper-

and needed for the execution of the next floating-point in-

struction. These exceptions will be processed as soon as

the FIFO becomes empty, and after any floating-point ex-

ception has been acknowledged.

For the instruction encodings, see Appendix A.

3.2 EXCEPTION PROCESSING

Exceptions are special events that alter the sequence of

instruction execution. The CPU recognizes three basic types

of exceptions: interrupts, traps and bus errors.

In the event of a non-restartable bus error, the acknowledge

will occur immediately. The CPU will flush the internal FIFO

and will reset the FPC by performing a dummy read of the

slave status word. This operation is performed for both the

regular and pipelined floating-point protocol and regardless

of whether any floating-point instruction is pending in the

FPC instruction queue.

An interrupt occurs in response to an event signalled by

activating the NMI or INT input signals. Interrupts are typi-

cally requested by peripheral devices that require the CPU’s

attention.

Traps occur as a result either of exceptional conditions

(e.g., attempted division by zero) or of specific instructions

whose purpose is to cause a trap to occur (e.g., supervisor

call instruction).

The CPU may cancel the last instruction sent to the FPC by

sending another ID and opcode, before the last source op-

erand for that instruction has been sent. Figure 3-8 shows

the instruction flow in pipelined floating-point mode.

A bus error exception occurs when the BER signal is acti-

vated during an instruction fetch or data transfer required by

the CPU to execute an instruction.

3.1.4.4 Floating Point Instructions

Table 3-1 gives the protocols followed for each Floating

Point instruction. The instructions are referenced by their

mnemonics. For the bit encodings of each instruction, see

Appendix A.

When an exception is recognized, the CPU saves the PC,

PSR and optionally the MOD register contents on the inter-

rupt stack and then it transfers control to an exception serv-

ice procedure.

The Operand class columns give the Access Class for each

general operand, defining how the addressing modes are

interpreted (see Instruction Set Reference Manual).

Details on the operations performed in the various cases by

the CPU to enter and exit the exception service procedure

are given in the following sections.

The Operand Issued columns show the sizes of the oper-

ands issued to the Floating Point Unit by the CPU. ‘‘D’’ indi-

cates a 32-bit Double Word. ‘‘i’’ indicates that the instruction

It is to be noted that the reset operation is not treated here

as an exception. Even though, like any exception, it alters

the instruction execution sequence.

e

Double Word). ‘‘f’’ indicates that the instruction

specifies an integer size for the operand (B

e

Byte, W

The reason being that the CPU handles reset in a signifi-

cantly different way than it does for exceptions.

Word, D

specifies a Floating Point size for the operand (F

e

32-bit

e

Refer to Section 3.5.3 for details on the reset operation.

Standard Floating, L

64-bit Long Floating).

31

3.0 Functional Description (Continued)

TABLE 3-1. Floating Point Instruction Protocols

Operand 1

Class

read.f

read.F

read.L

read.i

read.D

N/A

Operand 2

Class

rmw.f

Operand 1

Operand 2

Returned Value

Type and Dest.

f to Op.2

N/A

PSR Bits

Affected

none

N, Z, L

none

Mnemonic

Issued

ADDf

f

SUBf

rmw.f

f

MULf

rmw.f

f

DIVf

rmw.f

f

MOVf

write.f

read.f

write.i

write.L

write.F

write.f

N/A

f

N/A

f

ABSf

f

NEGf

f

CMPf

f

FLOORfi

TRUNCfi

ROUNDfi

MOVFL

MOVLF

MOVif

LFSR

f

N/A

f

i to Op.2

L to Op.2

F to Op.2

f to Op.2

N/A

f

F

L

i

D

SFSR

POLYf

DOTf

write.D

read.f

rmw.f

N/A

D to Op.2

f to F0

read.f

f

f to F0

SCALBf

LOGBf

SQRTf

MACf

f

f to Op.2

f to F1

write.f

read.f

N/A

f

TABLE 3-2. Custom Slave Instruction Protocols

Operand 1

Class

read.c

read.i

Operand 2

Class

rmw.c

write.c

read.c

write.i

write.c

write.Q

write.D

N/A

Operand 1

Operand 2

Issued

c

Returned Value

Type and Dest.

c to Op.2

N/A

PSR Bits

Affected

none

N,Z,L

none

Mnemonic

Issued

CCAL0c

CCAL1c

CCAL2c

CCAL3c

CMOV0c

CMOV1c

CMOV2c

CMOV3c

CCMP0c

CCMP1c

CCV0ci

CCV1ci

CCV2ci

CCV3ic

CCV4DQ

CCV5QD

LCSR

c

N/A

c

N/A

c

N/A

i to Op.2

c to Op.2

Q to Op.2

D to Op.2

N/A

c

i

read.D

read.Q

read.D

N/A

D

Q

D

N/A

D

N/A

SCSR

write.D

N/A

D to Op.2

N/A

LCR*

SCR*

read.D

write.D

N/A

D to Op.1

Note:

e

D

Double Word

e

i

Integer size (B,W,D) specified in mnemonic.

e

c

*

Custom size (D:32 bits or Q:64 bits) specified in mnemonic.

Privileged instruction: will trap if CPU is in User Mode.

e

N/A

Not Applicable to this instruction.

32

3.0 Functional Description (Continued)

3.2.1 Exception Acknowledge Sequence

reads the double-word entry from the Interrupt Dispatch ta-

c

4’. See Figures 3-9

and 3-10. The CPU uses this entry to call the exception

service procedure, interpreting the entry as an external pro-

cedure descriptor.

a

ble at address ‘INTBASE

vector

When an exception is recognized, the CPU goes through

three major steps:

1) Adjustment of Registers. Depending on the source of the

exception, the CPU may restore and/or adjust the con-

tents of the Program Counter (PC), the Processor Status

Register (PSR) and the currently-selected Stack Pointer

(SP). A copy of the PSR is made, and the PSR is then set

to reflect Supervisor Mode and selection of the Interrupt

Stack. Trap (TRC) and Trap (OVF) are always disabled.

Maskable interrupts are also disabled if the exception is

caused by an interrupt, Trap (DBG), Trap (ABT) or bus

error.

A new module number is loaded into the MOD register from

the least-significant word of the descriptor, and the static-

base pointer for the new module is read from memory and

loaded into the SB register. Then the program-base pointer

for the new module is read from memory and added to the

most-significant word of the module descriptor, which is in-

terpreted as an unsigned value. Finally, the result is loaded

into the PC register.

2) Vector Acquisition. A vector is either obtained from the

data bus or is supplied internally by default.

Direct-Exception Mode Enabled

The Direct-Exception mode is enabled when the DE bit in

the CFG register is set to 1. In this case the CPU first

pushes the saved PSR copy along with the contents of the

PC register on the Interrupt Stack. The word stored on the

Interrupt Stack between the saved PSR and PC register is

reserved for future use; its contents are undefined. The CPU

then reads the double-word entry from the Interrupt Dis-

3) Service Call. The CPU performs one of two sequences

common to all exceptions to complete the acknowledge

process and enter the appropriate service procedure.

The selection between the two sequences depends on

whether the Direct-Exception mode is disabled or en-

abled.

a

c

4’. The CPU

patch Table at address ‘INTBASE

vector

Direct-Exception Mode Disabled

uses this entry to call the exception service procedure, inter-

preting the entry as an absolute address that is simply load-

ed into the PC register. Figure 3-11 provides a pictorial of

the acknowledge sequence. It is to be noted that while the

The Direct-Exception mode is disabled while the DE bit in

the CFG register is 0 (Section 2.1.4). In this case the CPU

first pushes the saved PSR copy along with the contents of

the MOD and PC registers on the interrupt stack. Then it

TL/EE/9354–13

FIGURE 3-9. Interrupt Dispatch Table

33

3.0 Functional Description (Continued)

TL/EE/9354–14

TL/EE/9354–15

FIGURE 3-10. Exception Acknowledge Sequence.

Direct-Exception Mode Disabled.

34

3.0 Functional Description (Continued)

TL/EE/9354–16

TL/EE/9354–17

FIGURE 3-11. Exception Acknowledge Sequence.

Direct-Exception Mode Enabled.

direct-exception mode is enabled, the CPU can respond

more quickly to interrupts and other exceptions because

fewer memory references are required to process an excep-

tion. The MOD and SB registers, however, are not initialized

before the CPU transfers control to the service procedure.

Consequently, the service procedure is restricted from exe-

cuting any instructions, such as CXP, that use the contents

of the MOD or SB registers in effective address calcula-

tions.

mode procedures, RETT can also adjust the Stack Pointer

(SP) to discard a specified number of bytes from the original

stack as surplus parameter space.

RETI is used to return from a maskable interrupt service

procedure. A difference of RETT, RETI also informs any

external interrupt control units that interrupt service has

completed. Since interrupts are generally asynchronous ex-

ternal events, RETI does not discard parameters from the

stack.

Both of the above instructions always restore the Program

Counter (PC) and the Processor Status Register from the

interrupt stack. If the Direct-Exception mode is disabled,

they also restore the MOD and SB register contents. Fig-

ures 3-12 and 3-13 show the RETT and RETI instruction

flows when the Direct-Exception mode is disabled.

3.2.2 Returning from an Exception Service Procedure

To return control to an interrupted program, one of two in-

structions can be used: RETT (Return from Trap) and RETI

(Return from Interrupt).

RETT is used to return from any trap, non-maskable inter-

rupt or bus error service procedure. Since some traps are

often used deliberately as a call mechanism for supervisor

35

3.0 Functional Description (Continued)

TL/EE/9354–18

FIGURE 3-12. Return from Trap (RETT n) Instruction Flow.

Direct-Exception Mode Disabled.

3.2.3 Maskable Interrupts

3.2.3.2 Vectored Mode: Non-Cascaded Case

The INT pin is a level-sensitive input. A continuous low level

is allowed for generating multiple interrupt requests. The in-

put is maskable, and is therefore enabled to generate inter-

rupt requests only while the Processor Status Register I bit

is set. The I bit is automatically cleared during service of an

INT, NMI, Trap (DBG), Trap (ABT) or Bus Error request, and

is restored to its original setting upon return from the inter-

rupt service routine via the RETT or RETI instruction.

In the Vectored mode, the CPU uses an Interrupt Control

Unit (ICU) to prioritize many interrupt requests. Upon receipt

of an interrupt request on the INT pin, the CPU performs an

‘‘Interrupt Acknowledge, Master’’ bus cycle (Section

3.5.4.6) reading a vector value from the low-order byte of

the Data Bus. This vector is then used as an index into the

Dispatch Table in order to find the External Procedure De-

scriptor for the proper interrupt service procedure. The serv-

ice procedure eventually returns via the Return from Inter-

rupt (RETI) instruction, which performs an End of Interrupt

bus cycle, informing the ICU that it may re-prioritize any in-

terrupt requests still pending. The ICU provides the vector

number again, which the CPU uses to determine whether it

needs also to inform a Cascaded ICU (see below).

The INT pin may be configured via the SETCFG instruction

e

as either Non-Vectored (CFG Register bit I

e

0) or Vec-

tored (bit I

1).

3.2.3.1 Non-Vectored Mode

In the Non-Vectored mode, an interrupt request on the INT

pin will cause an Interrupt Acknowledge bus cycle, but the

CPU will ignore any value read from the bus and use instead

a default vector of zero. This mode is useful for small sys-

tems in which hardware interrupt prioritization is unneces-

sary.

In a system with only one ICU (16 levels of interrupt), the

vectors provided must be in the range of 0 through 127; that

is, they must be positive numbers in eight bits. By providing

36

3.0 Functional Description (Continued)

TL/EE/9354–19

FIGURE 3-13. Return from Interrupt (RETI) Instruction Flow.

Direct-Exception Mode Disabled.

37

3.0 Functional Description (Continued)

a negative vector number, an ICU flags the interrupt source

as being a Cascaded ICU (see below).

‘‘Interrupt Acknowledge, Master’’ bus cycle (Section

3.5.4.6) when processing of this interrupt actually begins.

The Interrupt Acknowledge cycle differs from that provided

for Maskable Interrupts in that the address presented is

Note: During a return from interrupt the CPU looks at bit 7 of the vector

number from the master ICU. If bit 7 is 0, bits 0 through 6 are ignored.

FFFFFF00 . The vector value used for the Non-Maskable

16

Interrupt is taken as 1, regardless of the value read from the

bus.

3.2.3.3 Vectored Mode: Cascaded Case

In order to allow more levels of interrupt, provision is made

in the CPU to transparently support cascading. Note that

the Interrupt output from a Cascaded ICU goes to an Inter-

rupt Request input of the Master ICU, which is the only ICU

which drives the CPU INT pin. Refer to the ICU data sheet

for details.

The service procedure returns from the Non-Maskable In-

terrupt using the Return from Trap (RETT) instruction. No

special bus cycles occur on return.

3.2.5 Traps

In a system which uses cascading, two tasks must be per-

formed upon initialization:

Traps are processing exceptions that are generated as di-

rect results of the execution of an instruction.

1) For each Cascaded ICU in the system, the Master ICU

must be informed of the line number on which it receives

the cascaded requests.

The return address saved on the stack by any trap except

Trap (TRC) and Trap (DBG) is the address of the first bye of

the instruction during which the trap occurred.

2) A Cascade Table must be established in memory. The

Cascade Table is located in a NEGATIVE direction from

the location indicated by the CPU Interrupt Base (INT-

BASE) Register. Its entries are 32-bit addresses, pointing

to the Vector Registers of each of up to 16 Cascaded

ICUs.

When a trap is recognized, maskable interrupts are not dis-

abled except for the case of Trap (ABT) and Trap (DBG).

There are 11 trap conditions recognized by the NS32532 as

described below.

Trap (ABT): An abort trap occurs when an invalid page ta-

ble entry or a protection level violation is detected for any of

the memory references required to execute an instruction.

Figure 3-9 illustrates the position of the Cascade Table. To

find the Cascade Table entry for a Cascaded ICU, take its

Master ICU line number (0 to 15) and subtract 16 from it,

Trap (SLAVE): An exceptional condition was detected by

the Floating Point Unit or another Slave Processor during

the execution of a Slave Instruction. This trap is requested

via the Status Word returned as part of the Slave Processor

Protocol (Section 3.1.4.1).

b

giving an index in the range 16 to 1. Multiply this value

by 4, and add the resulting negative number to the contents

of the INTBASE Register. The 32-bit entry at this address

must be set to the address of the Hardware Vector Register

of the Cascaded ICU. This is referred to as the ‘‘Cascade

Address.’’

Trap (ILL): Illegal operation. A privileged operation was at-

e

tempted while the CPU was in User Mode (PSR bit U

1).

Trap (SVC): The Supervisor Call (SVC) instruction was exe-

cuted.

Upon receipt of an interrupt request from a Cascaded ICU,

the Master ICU interrupts the CPU and provides the nega-

tive Cascade Table index instead of a (positive) vector num-

ber. The CPU, seeing the negative value, uses it as an index

into the Cascade Table and reads the Cascade Address

from the referenced entry. Applying this address, the CPU

performs an‘‘Interrupt Acknowledge, Cascaded’’ bus cycle,

reading the final vector value. This vector is interpreted by

the CPU as an unsigned byte, and can therefore be in the

range of 0 through 255.

Trap (DVZ): An attempt was made to divide an integer by

zero. (The FPU trap is used for Floating Point division by

zero.)

Trap (FLG): The FLAG instruction detected a ‘‘1’’ in the

PSR F bit.

Trap (BPT): The Breakpoint (BPT) instruction was execut-

ed.

Trap (TRC): The instruction just completed is being traced.

In returning from a Cascaded interrupt, the service proce-

dure executes the Return from Interrupt (RETI) instruction,

as it would for any Maskable Interrupt. The CPU performs

an ‘‘End of Interrupt, Master’’ bus cycle, whereupon the

Master ICU again provides the negative Cascade Table in-

dex. The CPU, seeing a negative value, uses it to find the

corresponding Cascade Address from the Cascade Table.

Applying this address, it performs an ‘‘End of Interrupt, Cas-

caded’’ bus cycle, informing the Cascaded ICU of the com-

pletion of the service routine. The byte read from the Cas-

caded ICU is discarded.

Refer to Section 3.3.1 for details.

Trap (UND): An Undefined-Instruction trap occurs when an

attempt to execute an instruction is made and one or more

of the following conditions is detected:

1. The instruction is undefined. Refer to Appendix A for a

description of the codes that the CPU recognizes to be

undefined.

2. The instruction is a floating point instruction and the F-bit

in the CFG register is 0.

3. The instruction is a custom slave instruction and the C-bit

in the CFG register is 0.

Note: If an interrupt must be masked off, the CPU can do so by setting the

corresponding bit in the interrupt mask register of the interrupt con-

troller.

4. The instruction is a memory-management instruction and

the M-bit in the CFG register is 0.

However, if an interrupt is set pending during the CPU instruction that

masks off that interrupt, the CPU may still perform an interrupt ac-

knowledge cycle following that instruction since it might have sampled

the INT line before the ICU deasserted it. This could cause the ICU to

provide an invalid vector. To avoid this problem the above operation

should be performed with the CPU interrupt disabled.

5. An LMR or SMR instruction is executed while the U-flag

in the PSR is 0 and the most significant bit of the instruc-

tion’s short field is 0.

6. The reserved general adressing mode encoding (10011)

is used.

3.2.4 Non-Maskable Interrupt

The Non-Maskable Interrupt is triggered whenever a falling

edge is detected on the NMI pin. The CPU performs an

7. Immediate addressing mode is used for an operand that

has access class different from read.

38

3.0 Functional Description (Continued)

8. Scaled Indexing is used and the basemode is also Scaled

Indexing.

The NS32532 does not respond to bus errors indicated for

instructions that are not executed. For example, no bus er-

ror exception occurs in response to asserting the BER sig-

nal during a bus cycle to prefetch an instruction that is not

executed because the previous instruction caused a trap.

9. The instruction is a floating-point or custom slave instruc-

tion that the FPU or custom slave detects to be unde-

fined. Refer to Section 3.1.4.1 for more information.

An exception to this rule occurs if the bus error is detected

during an MMU write cycle to update the R-bit in a page

table entry.

Trap (OVF): An Integer-Overflow trap occurs when the V-bit

in the PSR register is set to 1 and an Integer-Overflow con-

dition is detected during the execution of an instruction. An

Integer-Overflow condition is detected in the following cas-

es:

In this case the CPU recognizes the bus error and considers

it as non-restartable even though the bus cycle that caused

it belongs to a non-executed instruction.

1. The F-flag is 1 following execution of an ADDi, ADDQi,

ADDCi, SUBi, SUBCi, NEGi, ABSi, or CHECKi instruction.

If a bus error is detected during a data transfer required for

the processing of another exception or during the ICU read

cycle of a RETI instruction, then the CPU considers it as a

fatal bus error and enters the ‘HALTED’ state.

2. The product resulting from a MULi instruction cannot be

represented exactly in the destination operand’s location.

3. The quotient resulting from a DEIi, DIVi, or QUOi instruc-

tion cannot be represented exactly in the destination op-

erand’s location.

Note 1: If the address and control signals associated with the last bus cycle

that caused a bus error are latched by external hardware, then the

information they provide can be used by the service procedure for

restartable bus errors to analyze and resolve the exception recog-

nized by the CPU. This can be accomplished because upon detect-

ing a restartable bus error, the NS32532 stops making memory ref-

erences for subsequent instructions until it determines whether the

instruction that caused the bus error is executed and the exception

is processed.

4. The result of an ASHi instruction cannot be represented

exactly in the destination operand’s location.

5. The sum of the ‘INC’ value and the ‘INDEX’ operand for

an ACBi instruction cannot be represented exactly in the

index operand’s location.

Note 2: When a non-restartable bus error is recognized, the service proce-

dure must execute the CINV and LMR instructions to invalidate the

on-chip caches and TLB. This is necessary to maintain coherence

between them and external memory.

Trap (DBG): A debug trap occurs when one or more of the

conditions selected by the settings of the bits in the DCR

register is detected. This trap can also be requested by acti-

vating the input signal DBG. Refer to Section 3.3.2 for more

information.

Note 3: If the instruction causing a non-restartable bus error is followed by a

slave instruction, the service procedure should reset the slave by

reading the slave status register.

Note 1: Following execution of the WAIT instruction, then a Trap (DBG) can

be pending for a PC-match condition. In such an event, the Trap

(DBG) is processed immediately.

3.2.7 Priority Among Exceptions

The CPU checks for specific exceptions at various points

while executing an instruction. It is possible that several ex-

ceptions occur simultaneously. In that event, the CPU re-

sponds to the exception with highest priority.

Note 2: If an attempt is made to execute a memory-management instruction

while in User-Mode and the M-bit in the CFG register is 0, then Trap

(UND) occurs.

Note 3: If an attempt is made to execute a privileged custom instruction

while in User-Mode and the C-bit in the CFG register is 0, then Trap

(UND) occurs.

Figure 3-14 shows an exception processing flowchart. A

non-restartable bus error is assigned highest priority and is

serviced immediately regardless of the execution state of

the CPU.

Note 4: While operating in User-Mode, if an attempt is made to execute a

privileged instruction with an undefined use of a general addressing

mode (either the reserved encoding is used or else scaled-index or

immediate modes are incorrectly used), the Trap (UND) occurs.

Before executing an instruction, the CPU checks for pend-

ing Trap (DBG), interrupts, and Trap (TRC), in that order. If a

Trap (DBG) is pending, then the CPU processes that excep-

tion, otherwise the CPU checks for pending interrupts. At

this point, the CPU responds to any pending interrupt re-

quests; nonmaskable interrupts are recongized with higher

priority than maskable interrupts. If no interrupts are pend-

ing, then the CPU checks the P-flag in the PSR to determine

whether a Trap (TRC) is pending. If the P-flag is 1, a Trap

(TRC) is processed. If no Trap (DBG), interrupt or Trap

(TRC) is pending, the CPU begins executing the instruction.

Note 5: If an undefined instruction or illegal operation is detected, then no

data references are performed for the instruction.

Note 6: For certain instructions that are relatively long to execute, such as

DEID, the CPU checks for pending interrupts during execution of the

instruction. In order to reduce interrupt latency, the NS32532 can

suspend executing the instruction and process the interrupt. Refer

to Section B.5 in Appendix B for more information about recognizing

interrupts in this manner.

3.2.6 Bus Errors

A bus error exception occurs when the BER signal is assert-

ed in response to an instruction fetch or data transfer that is

required to execute an instruction.

While executing an instruction, the CPU may recognize up

to four exceptions:

Two types of bus errors are recognized: Restartable and

Non-Restartable. Restartable bus errors are recognized dur-

ing read bus cycles, except for MMU read cycles (from Page

Tables) needed to translate the address of a result being

stored into memory. All other bus errors are non-restartable.

1. trap (ABT)

2. restartable bus error

3. trap (DBG) or interrupt, if the instruction is interruptible

4. one of 7 mutually exclusive traps: SLAVE, ILL, SVC, DVZ,

FLG, BPT, UND

The CPU responds to restartable bus errors by suspending

the instruction that it was executing. When a non-restartable

bus error is detected, the CPU responds immediately and

the instruction being executed is terminated. See Section

3.1.2.3.

Trap (ABT) and restartable bus error have equal priority; the

CPU responds to the first one detected.

If no exception is detected while the instruction is executing,

then the instruction is completed and the PC is updated to

point to the next instruction. If a Trap (OVF) is detected,

then it is processed at this time.

The PC value saved on the stack is undefined.

39

3.0 Functional Description (Continued)

TL/EE/9354–20

FIGURE 3-14. Exception Processing Flowchart

40

3.0 Functional Description (Continued)

b

While executing the instruction, the CPU checks for enabled

debug conditions. If an enabled debug condition is met, a

Trap (DBG) is held pending until after the instruction is com-

pleted (see Note 3). If another exception is detected before

the instruction is completed, the pending Trap (DBG) is re-

moved and the DSR register is not updated.

7. If ‘‘Byte’’ is in the range 16 through 1, then the inter-

rupt source is Cascaded. (More negative values are re-

served for future use.) Perform the following:

a. Read the 32-bit Cascade Address from memory. The

a

address is calculated as INTBASE

4* Byte.

b. Read ‘‘Vector,’’ applying the Cascade Address just

read and Status Code 00101 (Interrupt Acknowledge,

Cascaded).

Note 1: Trap (DBG) can be detected simultaneously with Trap (OVF). In this

event, the Trap (OVF) is processed before the Trap (DBG).

Note 2: An address-compare debug condition can be detected while pro-

cessing a bus error, interrupt, or trap. In this event, the Trap (DBG)

is held pending until after the CPU has processed the first excep-

tion.

8. Perform Service (Vector, Return Address), Figure 3-15.

3.2.8.2 Abort/Restartable Bus Error Sequence

1. Suspend instruction and restore the currently selected

Stack Pointer to its original contents at the beginning of

the instruction.

Note 3: Between operations of a string instruction, the CPU responds to

pending operand address compare and external debug conditions

as well as interrupts. If

a PC-match debug condition is detected

while executing a string instruction, then Trap (DBG) is held pending

until the instruction has completed.

2. Clear the PSR P bit.

3. Copy the PSR into a temmporary register, then clear PSR

bits T, V, U, S and I.

3.2.8 Exception Acknowledge Sequences: Detailed Flow

For purposes of the following detailed discussion of excep-

tion acknowledge sequences, a single sequence called

‘‘service’’ is defined in Figure 3-15.

4. Set ‘‘Vector’’ to the value corresponding to the exception

type:

e

Abort:

Vector

2

Upon detecting any interrupt request, trap or bus error con-

dition, the CPU first performs a sequence dependent upon

the type of exception. This sequence will include saving a

copy of the Processor Status Register and establishing a

vector and a return address. The CPU then performs the

service sequence.

Restartable Bus Error: Vector

11

5. Set ‘‘Return Address’’ to the address of the first byte of

the suspended instruction.

6. Perform Service (Vector, Return Address), Figure 3-15.

3.2.8.3 SLAVE/ILL/SVC/DVZ/FLG/BPT/UND Trap

Sequence

3.2.8.1 Maskable/Non-Maskable Interrupt Sequence

This sequence is performed by the CPU when the NMI pin

receives a falling edge, or the INT pin becomes active with

the PSR I bit set. The interrupt sequence begins either at

the next instruction boundary or, in the case of an interrupt-

ible instruction (e.g., string instruction), at the next interrupt-

ible point during its execution.

1. Restore the currently selected Stack Pointer and the

Processor Status Register to their original values at the

start of the trapped instruction.

2. Set ‘‘Vector’’ to the value corresponding to the trap type.

e

SLAVE: Vector

3.

4.

5.

6.

7.

8.

10.

ILL:

Vector

1. If an interruptible instruction was interrupted and not yet

completed:

SVC:

DVZ:

FLG:

BPT:

UND:

a. Clear the Processor Status Register P bit.

b. Set ‘‘Return Address’’ to the address of the first byte of

the interrupted instruction.

Otherwise, set ‘‘Return Address’’ to the address of the

next instruction.

3. If Trap (ILL) or Trap (UND)

a. Clear the Processor Status Register P bit.

2. Copy the Processor Status Register (PSR) into a tempo-

rary register, then clear PSR bits T, V, U, S, P and I.

4. Copy the Processor Status Register (PSR) into a tempo-

rary register, then clear PSR bits T, V, U, S and P.

3. If the interrupt is Non-Maskable:

5. Set ‘‘Return Address’’ to the address of the first byte of

the trapped instruction.

a. Read

a

byte from address FFFFFF00 , applying

16

Status Code 00100 (Interrupt Acknowledge, Master).

Discard the byte read.

6. Perform Service (Vector, Return Address), Figure 3-15.

b. Set ‘‘Vector’’ to 1.

c. Go to Step 8.

3.2.8.4 Trace Trap Sequence

1. In the Processor Status Register (PSR), clear the P bit.

4. If the interrupt is Non-Vectored:

2. Copy the PSR into a temporary register, then clear PSR

bits T, V, U and S.

a. Read

a byte from address FFFFFE00 , applying

16

Status Code 00100 (Interrupt Acknowledge, Master).

Discard the byte read.

3. Set ‘‘Vector’’ to 9.

4. Set ‘‘Return Address’’ to the address of the next instruc-

tion.

b. Set ‘‘Vector’’ to 0.

c. Go to Step 8.

5. Perform Service (Vector, Return Address), Figure 3-15.

5. Here the interrupt is Vectored. Read ‘‘Byte’’ from address

3.2.8.5 Integer-Overflow Trap Sequence

FFFFFE00 , applying Status Code 00100 (Interrupt Ac-

16

knowledge, Master).

1. Copy the PSR into a temporary register, then clear PSR

bits T, V, U, S and P.

t

6. If ‘‘Byte’’ 0, then set ‘‘Vector’’ to ‘‘Byte’’ and go to Step

2. Set ‘‘Vector’’ to 13.

8.

3. Set ‘‘Return Address’’ to the address of the next instruc-

tion.

41

3.0 Functional Description (Continued)

4. Perform Service (Vector, Return Address), Figure 3-15.

3.3 DEBUGGING SUPPORT

The NS32532 provides serveral features to assist in pro-

gram debugging.

3.2.8.6 Debug Trap Sequence

A debug condition can be recognized either at the next in-

struction boundary or, in the case of an interruptible instruc-

tion, at the next interruptible point during its execution.

Besides the Breakpoint (BPT) instruction that can be used

to generate soft breaks, the CPU also provides instruction

tracing as well as debug trap (or hardware breakpoints) ca-

pabilities. Details on these features are provided in the fol-

lowing sub-sections.

1. If PC-match condition, then go to Step 3.

2. If a String instruction was interrupted and not yet com-

pleted:

3.3.1 Instruction Tracing

a. Clear the Processor Status Register P bit.

Instruction tracing is a very useful feature that can be used

during debugging to single-step through selected portions of

a program. Tracing is enabled by setting the T-bit in the PSR

Register. When enabled, the CPU generates a Trace Trap

(TRC) after the execution of each instruction.

b. Set ‘‘Return Address’’ to the address of the first byte of

the instruction.

c. Go to Step 4.

3. Set ‘‘Return Address’’ to the address of the next instruc-

tion.

At the beginning of each instruction, the T bit is copied into

the PSR P (Trace ‘‘Pending’’) bit. If the P bit is set at the end

of an instruction, then the Trace Trap is activated. If any

other trap or interrupt request is made during a traced in-

struction, its entire service procedure is allowed to complete

before the Trace Trap occurs. Each interrupt and trap se-

quence handles the P bit for proper tracing, guaranteeing

only one Trace Trap per instruction, and guaranteeing that

the Return Address pushed during a Trace Trap is always

the address of the next instruction to be traced.

4. Set ‘‘Vector’’ to 14.

5. Copy the Processor Status Register (PSR) into a tempo-

rary register, then clear PSR bits T, V, U, S, P and I.

6. Perform Service (Vector, Return Address), Figure 3-15.

Note: In case of PC-match or address-compare on write, the Trap (DBG)

may occur before the instruction is executed.

3.2.8.7 Non-Restartable Bus Error Sequence

1. Set ‘‘Vector’’ to 12.

Due to the fact that some instructions can clear the T and P

bits in the PSR, in some cases a Trace Trap may not occur

at the end of the instruction. This happens when one of the

privileged instructions BICPSRW or LPRW PSR is executed.

2. Set ‘‘Return Address’’ to ‘‘Undefined’’.

3. Copy the Processor Status Register (PSR) into a tempo-

rary register, then clear PSR bits T, V, U, S, P and I.

4. Perform a dummy read of the Slave Status Word to reset

the Slave Processor.

5. Perform Service (Vector, Return Address), Figure 3-15.

TABLE 3-3. Summary of Exception Processing

Instruction

Ending

Cleared Before

Saving PSR

Cleared After

Saving PSR

Exception

Restartable Bus Error

Nonrestartable Bus Error

Suspended

Terminated

P

Undefined

TVUSI

TVUS

Interrupt

Before Instruction

None/P*

TVUSPI

ABT

ILL, UND

SLAVE, SVC, DVZ, FLG, BPT

OVF

TRC

DBG

Suspended

Completed

Before Instruction

P

None

P

TVUSI

TVUS

TVUSP

TVUS

None/P*

TVUSPI

*Note: The P bit of the saved PSR is cleared in case the exception is acknowledged before the instruction is completed (e.g., interrupted string instruction). This is

to avoid a mid-instruction trace trap upon return from the Exception Service Routine.

Service (Vector, Return Address):

1) Push the PSR copy onto the Interrupt Stack as a 16-bit value.

2) If Direct-Exception mode is selected, then go to step 4.

3) Push MOD Register into the Interrupt Stack as a 16-bit value.

a

c

4’.

4) Read 32-bit Interrupt Dispatch Table (IDT) entry at address ‘INTBASE

5) If Direct-Exception mode is selected, then go to Step 10.

vector

6) Move the L.S. word of the IDT entry (Module Field) into the MOD register.

a

7) Read the Program Base pointer from memory address ‘MOD

Program Counter.

8’, and add to it the M.S. word of the IDT entry (Offset Field), placing the result in the

8) Read the new Static Base pointer from the memory address contained in MOD, placing it into the SB Register.

9) Go to Step 11.

10) Place IDT entry in the Program Counter.

11) Push the Return Address onto the Interrupt Stack as a 32-bit quantity.

12) Serialize: Non-sequentially fetch first instruction of Exception Service Routine.

Note: Some of the Memory Accesses indicated in the service sequence may be performed in an order different from the one shown.

FIGURE 3-15. Service Sequence

42

3.0 Functional Description (Continued)

In other cases, it is still possible to guarantee that a Trace

Trap occurs at the end of the instruction, provided that spe-

cial care is taken before returning from the Trace Trap Serv-

ice Procedure. In case a BICPSRB instruction has been ex-

ecuted, the service procedure should make sure that the T

bit in the PSR copy saved on the Interrupt Stack is set be-

fore executing the RETT instruction to return to the program

begin traced. If the RETT or RETI instructions have to be

traced, the Trace Trap Service Procedure should set the P

and T bits in the PSR copy on the Interrupt Stack that is

going to be restored in the execution of such instructions.

higher priority trap (i.e., ABORT) is detected, the BP signal

may or may not be asserted.

Note 1: The assertion of BP is not affected by the setting of the TR bit in the

DCR register.

Note 2: While executing the MOVUS and MOVSU instructions, the com-

pare-address condition is enabled for the User space memory refer-

ence under control of the UD-bit in the DCR.

Note 3: When the LPRi instruction is executed to load a new value into the

BPC, CAR or DCR, it is undefined whether the address-compare

and PC-match conditions, in effect while executing the instruction,

are detected under control of the old or new contents of the loaded

register. Therefore, any LPRi instruction that alters the control of the

address-compare or PC-match conditions should use register or im-

mediate addressing mode for the source operand.

Note: If instruction tracing is enabled while the WAIT instruction is executed,

the Trap (TRC) occurs after the next interrupt, when the interrupt

service procedure has returned.

Note 4: If an exception occurred during the previous instruction, trap (DBG)

may be taken prior to instruction execution.

3.3.2 Debug Trap Capability

3.4 ON-CHIP CACHES

The CPU recognizes three different conditions to generate a

Debug Trap:

The NS32532 provides three on-chip caches: the Instruc-

tion Cache (IC), the Data Cache (DC) and the Translation

Look-aside Buffer (TLB).

1) Address Compare

2) PC Match

The first two are used to hold the contents of frequently

used memory locations, while the TLB holds address-trans-

lation information.

3) External

These conditions can be enabled and monitored through

the CPU Debug Registers.

The IC and DC can be individually enabled by setting appro-

priate bits in the CFG Register (See Section 2.1.4); the TLB

is automatically enabled when address-translation is en-

abled.

An address-compare condition is detected when certain

memory locations are either read or written. The double-

word address used for the comparison is specified in the

CAR Register. The address-compare condition can be sep-

arately enabled for each of the bytes in the specified dou-

ble-word, under control of the CBE bits of the DCR Register.

The VNP bit in the DCR controls whether virtual or physical

addresses are compared. The CRD and CWR bits in the

DCR separately enable the address compare condition for

read and write references; the CAE bit in the DCR can be

used to disable the compare-address condition indepen-

dently from the other control bits. The CPU examines the

address compare condition for all data reads and writes,

reads of memory locations for effective address calcula-

tions, Interrupt-Acknowledge and End-of-Interrupt bus cy-

cles, and memory references for exception processing. An

address-compare condition is not detected for MMU refer-

ences to Page Table Entries.

The CPU also provides a locking feature that allows the

contents of the IC and DC to be locked to specific memory

locations. This is accomplished by setting the LIC and LDC

bits in the CFG register.

Cache locking can be successfully used in real-time applica-

tions to guarantee fast access to critical instruction and data

areas.

Details on the organization and function of each of the

caches are provided in the following sections.

Note: The size and organization of the on-chip caches may change in future

Series 32000 microprocessors. This however, will not affect software

compatibility.

3.4.1 Instruction Cache (IC)

The basic structure of the instruction cache (IC) is shown in

Figure 3-16.

The PC-match condition is detected when the address of

the instruction equals the value specified in the BPC regis-

ter. The PC-match condition is enabled by the PCE bit in the

DCR.

The IC stores 512 bytes of code in a direct-mapped organi-

zation with 32 sets. Direct-mapped means that each set

contains only one block, thus each memory location can be

loaded into the IC in only one place.

Detection of address-compare and PC-match conditions is

enabled for User and Supervisor Modes by the UD and SD

bits in the DCR. The DEN-bit can be used to disable detec-

tion of these two conditions independently from the other

control bits.

Each block contains a 23-bit tag, which holds the most-sig-

nificant bits of the physical address for the locations stored

in the block, along with 4 double-words and 4 validity bits

(one for each double-word).

An external condition is recognized whenever the DBG sig-

nal is activated.

A 4-double-word instruction buffer is also provided, which is

loaded either from a selected cache block or from external

memory. Instructions are read from this buffer by the loader

unit and transferred to an 8-byte instruction queue.

When the CPU detects an address-compare or PC-match

condition while executing an instruction or processing an

exception, then Trap (DBG) occurs if the TR bit in the DCR

is 1. When an external debug condition is detected, Trap

(DBG) occurs regardless of the TR bit. The cause of the

Trap (DBG) is indicated in the DSR Register.

The IC may or may not be enabled to cache an instruction

being fetched by the CPU. It is enabled when the IC bit in

the CFG Register is set to 1 and either the address transla-

tion is disabled or the CI bit in the Level-2 PTE used to

translate the virtual address of the instruction is set to 0.

When an address-compare or PC-match condition is detect-

ed while executing an instruction, the CPU asserts the BP

signal at the beginning of the next instruction, synchronous-

ly with PFS. If the instruction is not completed because a

If the IC is disabled, the CPU bypasses it during the instruc-

tion fetch and its contents are not affected. The instruction

is read directly from external memory into the instruction

buffer.

43

3.0 Functional Description (Continued)

TL/EE/9354–21

FIGURE 3-16. Instruction Cache Structure

3.4.2 Data Cache (DC)

When the IC is enabled, the instruction address bits 4 to 8

are used to select the IC set where the instruction may be

stored. The tag corresponding to the single block in the set

is compared with the 23 most-significant bits of the instruc-

tion’s physical address. The 4 double-words in this block are

loaded into the instruction buffer and the 4 validity bits are

also retrieved. Bits 2 and 3 of the instruction’s physical ad-

dress select one of these double-words and the associated

validity bit.

The Data Cache (DC) stores 1,024 bytes of data in a two-

way set associative organization as shown in Figure 3-17.

Each of the 32 sets has 2 cache blocks. Each block con-

tains a 23-bit tag, which holds the most-significant bits of

the physical address for the locations stored in the block,

along with 4 double-words and 4 validity bits (one for each

double-word).

The DC is enabled for a data read when all of the following

conditions are satisfied.

If the tag matches and the selected double-word is valid, a

cache ‘hit’ occurs and the double-word is directly trans-

ferred to the instruction queue for decoding; otherwise a

cache ‘miss’ will result.

The DC bit in the CFG Register is set to 1.

#

Either the address translation is disabled or the CI bit in

the Level-2 PTE used to translate the virtual address of

the data reference is set to 0.

#

In the latter case, if the cache is not locked, the CPU will

take the following actions.

The reference is not an interlocked read resulting from

executing a CBITI or SBITI instruction.

#

First, if the tag of the selected block does not match, the tag

is loaded with the 23 most-significant bits of the instruction

address and all the validity bits are cleared. Then, the in-

struction is read from external memory into the instruction

buffer.

If the DC is disabled, the CPU bypasses it during the data

read and its contents are not affected. The data is read

directly from external memory. The DC is also bypassed for

MMU reads from Page Table entries during address transla-

tion and for Interrupt-Acknowledge and End-of-Interrupt bus

cycles.

If the CIIN input signal is not active during the fetching of the

missing instruction, then the IC is updated and the instruc-

tion double-words fetched from memory are stored into it

with the validity bits set.

When the DC is enabled for a data read, the address bits 4

to 8 are used to select the DC set where the data may be

stored.

If the cache is locked, its contents are not affected, as the

CPU reads the missing instruction from external memory.

The tags corresponding to the two blocks in the set are

compared to the 23 most-significant bits of the physical ad-

dress. Bits 2 and 3 of the address select one double-word in

each block and the associated validity bit.

Whenever the CPU accesses external memory, whether or

not the IC is enabled, it always fetches instruction double-

words in a non-wrap-around fashion. Refer to Sections

3.5.4.3 and 3.5.6 for more information.

If one of the tag matches and the selected double-word in

the corresponding block is valid, a cache ‘hit’ occurs and

the data is used to execute the instruction; otherwise a

cache ‘miss’ will result. In the latter case, if the cache is not

locked, the CPU will take the following actions.

The contents of the instruction cache can be invalidated by

software through the CINV instruction or by hardware

through the appropriate cache invalidation input signals.

Clearing the IC bit in the CFG Register also invalidates the

instruction cache. Refer to Sections 3.5.10 and C.3 for de-

tails.

Note: If the IC is enabled for a certain instruction and a ‘miss’ occurs due to

a tag mismatch, the CPU will clear all the validity bits of the selected

tag before fetching the instruction from external memory. If the CIIN

input signal is activated during the fetching of that instruction, the

validity bits are not set and the IC is not updated.

44

3.0 Functional Description (Continued)

TL/EE/9354–22

FIGURE 3-17. Data Cache Structure

First, if the tag of either block in the set matches the data

address, that block is selected for updating. Otherwise, if

neither tag matches, then the least recently used block is

selected; its tag is loaded with the 23 most-significant bits of

the data address, and all the validity bits are cleared.

vidual pages using the CI-bit in the level-2 Page Table En-

tries. The CINV instruction can be executed to invalidate

entriely the Instruction Cache and/or Data Cache; the CINV

instruction can also be executed to invalidate

16-byte block in either or both caches.

a single

Then, the data is read from external memory; up to 4 dou-

ble-word bits are read into the cache in a wrap-around fash-

ion. Refer to Sections 3.5.4.3 and 3.5.6 for more informa-

tion.

In hardware, the use of the caches can be inhibited for indi-

vidual locations using the CIIN input signal. A cache invali-

dation request can cause the entire Instruction Cache and/

or Data Cache to be invalidated; a cache invalidation re-

quest can also cause invalidation of a single set in either or

both caches. Refer to Section 3.5.7 for more information.

If the CIIN and IODEC input signals are both inactive during

the bus cycles performed to read the missing data, then the

DC is updated, as each double-word is read from memory,

and the corresponding validity bit is set. If the cache is

locked, its contents are not affected, as the CPU reads the

missing data from external memory.

An external ‘‘Bus Watcher’’ circuit can also be used to help

maintain cache coherence. The Bus Watcher observes the

CPU’s bus cycles to maintain a copy of the on-chip cache

tags while also monitoring writes to main memory by DMA

controllers and other microprocessors in the system. When

the Bus Watcher detects that a location in one of the on-

chip caches has been modified in main memory, it issues an

invalidation request to the CPU. The CPU provides the nec-

essary information on the system interface to help maintain

an external copy of the on-chip tags.

The DC is enabled for a data write whenever the DC bit in

the CFG Register is set to 1, including interlocked writes

resulting from executing the CBITI and SBITI instructions,

and MMU writes to Page Table entries during address trans-

lation.

The DC does not use write allocation. This means that, dur-

ing a write, if a cache ‘hit’ occurs, the DC is updated, other-

wise it is unaffected. The data is always written through to

external memory.

The status codes differentiate between instruction fetches

and data reads.

The set, affected during the bus access (if CIOUT is low), as

well as the tag can be determined from the address bits A4

through A8 and A9 through A31 respectively.

The contents of the data cache can be invalidated by soft-

ware through the CINV instruction or by hardware through

the appropriate cache invalidation input signals. Clearing

the DC bit in the CFG Register also invalidates the data

cache. Refer to Sections 3.5.10 and C.3 for details.

During a data read the CPU also indicates, by means of the

CASEC signal, which block in the set is being updated.

Whenever a CINV instruction is executed, the operation

code and operand appear on the system interface using

slave processor bus cycles. Thus, invalidations of the on-

chip caches by software can be monitored externally.

Note: If the DC is enabled for a certain data reference and a ‘‘miss’’ occurs

due to tag mismatch, the CPU will clear all the validity bits for the least

recently used tag before reading the data from external memory. If

either CIIN or IODEC are activated during the data read bus cycles,

the validity bits are not set and the DC is not updated.

Note, however, that the software is responsible for commu-

nicating to the external circuitry the values of the cache en-

able and lock bits in the CFG Register, since the CPU does

not generate any special cycle (e.g., Slave Cycle) when the

CFG Register is loaded.

3.4.3 Cache Coherence Support

The NS32532 provides several mechanisms for maintaining

coherence between the on-chip caches and external mem-

ory. In software, the use of caches can be inhibited for indi-

45

3.0 Functional Description (Continued)

3.4.4 Translation Look-aside Buffer (TLB)

were not already set. For these reasons, there is no need to

replicate either the V bit or the R bit in the TLB entries.

The Translation Look-aside Buffer is an on-chip fully asso-

ciative memory. It provides direct virtual to physical mapping

for 64 pages, thus minimizing the time needed to perform

the address translation.

Whenever a Page Table Entry in memory is altered by soft-

ware, it is necessary to purge any matching entry from the

TLB, otherwise the corresponding addresses would be

translated according to obsolete information. TLB entries

may be selectively purged by writing a virtual address to one

of the IVARn registers using the LMR instruction. The TLB

entry (if any) that matches that virtual address is then

purged, and its space is made available for another transla-

tion. Purging is also performed whenever an address space

is remapped by altering the contents of the PTB0 or PTB1

register. When this is done, all the TLB entries correspond-

ing to the address space mapped by that register are

purged. Turning translation on or off (via the MCR TU and

TS bits) does not affect the contents of the TLB.

The efficiency of the on-chip MMU is greatly increased by

the TLB, which bypasses the much longer Page Table look-

up in over 99% of the accesses made by the CPU.

Entries in the TLB are allocated and replaced automatically;

the operating system is not involved. The TLB entries can-

not be read or written by software; however, they can be

purged from it under program control.

Figure 3-18 shows a model of the TLB. Information is

placed into the TLB whenever a Page Table lookup is per-

e

formed. If the retrieved mapping is valid (V

1 in both

levels of the Page Tables), and the access attempted is

permitted by the protection level, an entry of the TLB is

loaded from the information retrieved from memory.

It is possible to maintain an external copy of the valid con-

tents of the on-chip TLB by observing the CPU’s system

interface during the replacement and invalidation of TLB en-

tries. Whenever the CPU replaces a TLB entry, the page

tables are accessed in external memory using bus cycles

with a special Status. Because a FIFO replacement algo-

rithm is used, it is possible to determine which entry is being

replaced by using a 6-bit counter that is incremented when-

ever a Level-1 PTE is accessed. The contents of the new

entry can be found as follows:

The on-chip MMU places the Virtual Page Number (VPN)

and the Address Space qualifier (AS) into the tag portion of

the TLB entry.

The value portion of the entry is loaded from the Page Ta-

bles as follows:

The PFN field (20 bits) as well as the CI and M bits are

loaded from the Level-2 Page Table Entry (PTE2).

#

VPN appears on A2 through A11 during the PTE1 and

#

The PL field (2 bits) is loaded to reflect the most restric-

#

PTE2 accesses. The most-significant 10 bits appear dur-

ing the PTE1 access, and the least-significant 10 bits

appear during the PTE2 access.

tive of the protection levels imposed by the PL fields of

the Level-1 and Level-2 Page Table Entries (PTE1 and

PTE2).

AS can be determined from the U/S signal during the

PTE1 access.

#

Not shown in the figure is an additional bit associated with

each TLB entry which indicates whether the entry is valid.

PFN, M and CI can be determined from the PTE2 value

#

Address translation can be either enabled or disabled for a

memory reference. If translation is disabled, then the TLB is

bypassed and the physical address is identical to the virtual

address.

read on the Data Bus. PL can be determined from the

most restrictive of the PTE1 and PTE2 values read on

the Data Bus.

Whenever a LMR instruction is executed, the operation

code and operand appear on the system interface using

slave processor bus cycles. Thus, the information is avail-

able externally to determine the translation modes con-

trolled by the MCR and to identify that a TLB entry has been

invalidated.

When translation is enabled and a virtual address needs to

be translated, the high-order 20 bits (VPN) and the Address

Space qualifier are compared associatively to the corre-

sponding fields in all entries of the TLB.

For a read reference, if the tag portion of a valid TLB entry,

completely matches the input values, then the value portion

of the entry is used to complete the address translation and

protection checking.

When the PTB0 register is loaded by executing the ‘LMR

PTB0 src’ instruction, the internal FIFO pointer is also reset

to point to the first TLB entry.

For a write reference, if a valid entry with a matching tag is

present in the TLB, then the M bit is examined. If the M bit is

1, the value portion of the entry is used to complete the

address translation and protection checking. If the M bit is 0,

the entry is invalidated.

Note that the contents of the TLB maintained externally in-

clude copies of all valid entries in the on-chip TLB, but the

external copy may include some entries that are invalid in

the on-chip TLB. For example, when the TLB is searched

for a write reference and a matching entry is found with the

M bit clear, then the on-chip entry is invalidated and a miss

is processed. It is not possible to detect externally that the

old matching entry on-chip has been invalidated.

In either case, if a protection level violation is detected, a

translation exception (Trap (ABT)) is generated. When no

matching entry is found or a matching entry is invalidated

because the M bit is 0 in a write reference, a Page Table

lookup is performed. The virtual address is translated ac-

cording to the algorithm given in Section 2.4.5 and the

translation information is loaded into the TLB.

3.5 SYSTEM INTERFACE

This section provides general information on the NS32532

interface to the external world. Descriptions of the CPU re-

quirements as well as the various bus characteristics are

provided here. Details on other device characteristics in-

cluding timing are given in Chapter 4.

The recipient entry is selected by an on-chip circuit that im-

plements a First-In-First-Out (FIFO) algorithm.

Note that for a translation to be loaded into the TLB it is

necessary that the Level-1 and Level-2 Page Table Entries

3.5.1 Power and Grounding

e

be valid (V bit

1). Also, it is guaranteed that in the pro-

The NS32532 requires a single 5-volt power supply, applied

on 21 pins. The logic voltage pins (VCCL1 to VCCL6) supply

cess of loading a TLB entry (during a Page Table lookup)

the Level-1 and Level-2 R bits will be set in memory if they

46

3.0 Functional Description (Continued)

TL/EE/9354–23

*AS represents the virtual address space qualifier.

FIGURE 3-18. TLB Model

3.5.2 Clocking

the power to the on-chip logic. The buffer voltage pins

(VCCB1 to VCCB14) supply the power to the output drivers

of the chip. The bus clock power pin (VCCCLK) is the power

supply for the on-chip clock drivers. All the voltage pins

should be connected together by a power (VCC) plane on

the printed circuit board.

The NS32532 requires a single-phase input clock signal

(CLK) with frequency twice the CPU’s operating frequency.

This clock signal is internally divided by two to generate two

non-overlapping phases PHI1 and PHI2. One single-phase

clock signal BCLK in phase with PHI1 and its complement

BCLK, are also generated and output by the CPU for timing

reference.

The NS32532 grounding connections are made on 20 pins.

The logic ground pins (GNDL1 to GNDL6) are the ground

pins for the on-chip logic. The buffer ground pins (GNDB1 to

GNDB13) are the ground pins for the output drivers of the

chip. The bus clock ground pin (GNDCLK) is the ground

connection for the on-chip clock drivers. All the ground pins

should be connected together by a ground plane on the

printed circuit board.

Following power-on, the phase relationship between BCLK

and CLK is undefined. Nevertheless, in some systems it

may be necessary to synchronize the CPU bus timing to an

external reference. The SYNC input signal can be used to

initialize the phase relationship between CLK and BCLK.

SYNC can also be used to stretch BCLK (Low) while CLK is

toggling.

Both power and ground connections are shown in Figure

3-19.

SYNC is sampled on each rising edge of CLK. As shown in

Figure 3-20, whenever SYNC is sampled low, BCLK stops

toggling and stays low. On the first rising edge that SYNC is

sampled high, BCLK is driven high and then toggles on each

subsequent rising edge of CLK.

Every rising edge of BCLK defines a transition in the timing

state (‘‘T-State’’) of the CPU.

One T-State represents the execution of one microinstruc-

tion within the CPU and/or one step of an external bus

transfer.

Note: The CPU requirement on the maximum period of BCLK must be satis-

fied when SYNC is asserted at times other than reset.

3.5.3 Resetting

The RST input pin is used to reset the NS32532. The CPU

samples RST synchronously on the rising edge of BCLK.

Whenever a low level is detected, the CPU responds imme-

diately. Any instruction being executed is terminated; any

results that have not yet been written to memory are dis-

carded; and any pending bus errors, interrupts, and traps

are eliminated. The internal latches for the edge-sensitive

NMI and DBG signals are cleared.

TL/EE/9354–24

FIGURE 3-19. Power and Ground Connections

TL/EE/9354–25

FIGURE 3-20. Bus Clock Synchronization

47

3.0 Functional Description (Continued)

The CPU stores the PC contents in the R0 Register and the

PSR contents in the least-significant word of R1, leaving the

most-significant word undefined. The PC is then cleared to 0

and so are all the implemented bits in the PSR, MSR, MCR

and CFG registers. The DEN-bit in the DCR Register is also

cleared to 0. After reset, the remaining implemented bits in

DCR and the contents of all other registers are undefined.

The CPU begins executing the instruction at Address 0.

3.5.4.1 Bus Status

The CPU presents five bits of Bus Status information on

pins ST0–ST4. The various combinations on these pins in-

dicate why the CPU is performing a bus cycle, or, if it is idle

on the bus, then why is it idle.

The Bus Status pins are interpreted as a five-bit value, with

ST0 the least significant bit. Their values decode as follows:

00000 The bus is idle because the CPU does not yet need

to access the bus.

On application of power, RST must be held low for at least

is stable. This is to ensure that all on-chip

50 ms after V

CC

00001 The bus is idle because the CPU is waiting for an

interrupt following execution of the WAIT instruc-

tion.

voltages are completely stable before operation. Whenever

a Reset is applied, it must also remain active for not less

than 64 BCLK cycles. See Figures 3-21 and 3-22.

00010 The bus is idle because the CPU has halted after

detecting an abort or bus error while processing an

exception.

While in the Reset state, the CPU drives the signals ADS,

BE0–3, BMT, CONF and HLDA inactive. The data bus is

floated and the state of all other output signals is undefined.

00011 The bus is idle because the CPU is waiting for a

Slave Processor to complete executing an instruc-

tion.

Note 1: If HOLD is active at the time RST is deasserted, the CPU acknowl-

edges HOLD before performing any bus cycle.

Note 2: If SYNC is asserted while the CPU is being reset, then BCLK does

not toggle. Consequently, SYNC must be high for at least 128 CLK

cycles while RST is low.

00100 Interrupt Acknowledge, Master.

The CPU is reading an interrupt vector to acknowl-

edge an interrupt request.

00101 Interrupt Acknowledge, Cascaded.

The CPU is reading an interrupt vector to acknowl-

edge a maskable interrupt request from a Cascad-

ed Interrupt Control Unit.

00110 End of Interrupt, Master.

The CPU is performing a read cycle to indicate that

it is executing a Return from Interrupt (RETI) in-

struction at the completion of an interrupt’s service

procedure.

TL/EE/9354–26

00111 End of Interrupt, Cascaded.

FIGURE 3-21. Power-On Reset Requirements

The CPU is performing a read cycle from a Cascad-

ed Interrupt Control Unit to indicate that it is execut-

ing a Return from Interrupt (RETI) instruction at the

completion of an interrupt’s service procedure.

01000 Sequential Instruction Fetch.

The CPU is fetching the next double-word in se-

quence from the instruction stream.

TL/EE/9354–27

01001 Non-Sequential Instruction Fetch.

FIGURE 3-22. General Reset Timing

The CPU is fetching the first double-word of a new

sequence of instruction. This will occur as a result

of any JUMP or BRANCH, any exception, or after

the execution of certain instructions.

3.5.4 Bus Cycles

The NS32532 CPU will perform bus cycles for one of the

following reasons:

01010 Data Transfer.

1. To fetch instructions from memory.

The CPU is reading or writing an operand for an

instruction, or it is referring to memory while pro-

cessing an exception.

2. To write or read data to or from memory or peripheral

devices. Peripheral input and output are memory mapped

in the Series 32000 family.

01011 Read RMW Class Operand.

3. To read and update Page Table Entries in memory to

perform memory management functions.

The CPU is reading an operand with access class

of read-modify-write.

4. To acknowledge an interrupt and allow external circuitry

to provide a vector number, or to acknowledge comple-

tion of an interrupt service routine.

01100 Read for Effective Address Calculation.

The CPU is reading a pointer from memory in order

to calculate an effective address for Memory Rela-

tive or External addressing modes.

5. To transfer information to or from a Slave Processor.

In terms of bus timing, cases 1 through 4 above are identi-

cal. For timing specifications, see Section 4. The only exter-

nal difference between them is the 5-bit code placed on the

Bus Status pins (ST0–ST4). Slave Processor cycles differ in

that separate control signals are applied (Section 3.5.4.7).

01101 Access PTE1 by MMU.

The CPU is reading or writing a Level-1 Page Table

Entry while the on-chip MMU is translating virtual

address.

48

3.0 Functional Description (Continued)

01110 Access PTE2 by MMU.

The CPU is reading or writing a Level-2 Page Table

Entry while the on-chip MMU is translating a virtual

address.

11101 Transfer Slave Processor Operand.

The CPU is transferring an operand to or from a

Slave Processor.

11110 Read Slave Processor Status.

The CPU is reading a status word from a slave

processor after the slave processor has activated

the FSSR signal.

a

11111 Broadcast Slave Processor ID

OPCODE.

The CPU is initiating the execution of a Slave In-

struction by transferring the first 3 bytes of the in-

struction, which specify the Slave Processor identi-

fication and operation.

3.5.4.2 Basic Read and Write Cycles

The sequence of events occurring during a basic CPU ac-

cess to either memory or peripheral device is shown in Fig-

ure 3-23 for a read cycle, and Figure 3-24 for a write cycle.

The cases shown assume that the selected memory or pe-

ripheral device is capable of communicating with the CPU at

full speed. If not, then cycle extension may be requested

through the RDY line. See Section 3.5.4.4.

A full speed bus cycle is performed in two cycles of the

BCLK clock, labeled T1 and T2. For both read and write bus

cycles the CPU asserts ADS during the first half of T1 indi-

cating the beginning of the bus cycle. From the beginning of

T1 until the completion of the bus cycle the CPU drives the

Address Bus and other relevant control signals as indicated

in the timing diagrams. For cacheable data read cycles the

CPU also drives the CASEC signal to indicate the block in

the DC set where the data will be stored. If the bus cycle is

not cancelled (e.g., state T2 is entered in the next clock

cycle), the confirm signal (CONF) is asserted in the middle

of T1. Note that due to a bus cycle cancellation, the BMT

signal may be asserted at the beginning of T1, and then

deasserted before the time in which it is guaranteed valid

(see Section 4.4.2).

A confirmed bus cycle is completed at the end of T2, unless

a cycle extension is requested. Following state T2 is either

state T1 of the next bus cycle, or an idle T-state, if the CPU

has no bus cycle to perform.

In case of a read cycle the CPU samples the data bus at the

end of state T2.

TL/EE/9354–28

If a bus exception is detected, the data is ignored.

FIGURE 3-23. Basic Read Cycle

For write bus cycles, valid data is output from the middle of

T1 until the end of the cycle. When a write bus cycle is

immediately followed by another write cycle, the CPU keeps

driving the bus with the data related to the previous cycle

until the middle of state T1 of the second bus cycle.

The CPU always inserts an idle state before a write cycle

when the write immediately follows a read cycle.

Note: The CPU can initiate a bus cycle with a T1-state and then cancel the

cycle, such as when a TLB miss or a Cache hit occurs. In such a case,

the CONF signal remains High and the BMT signal is driven High; the

T1-state is followed by another T1-state or an idle T-state.

49

3.0 Functional Description (Continued)

3.5.4.3 Burst Cycles

The NS32532 is capable of performing burst cycles in order

to increase the bus transfer rate. Burst is only available in

instruction fetch cycles and data read cycle from 32-bit wide

memories. Burst is not supported in operand write cycles or

slave cycles.

The sequence of events for burst cycles is shown in Figure

3-25. The case shown assumes that the selected memory is

capable of communicating with the CPU at full speed. If not,

then cycle extension can be requested through the RDY

line. See Section 3.5.4.4.

A Burst cycle is composed of two parts. The first part is a

regular cycle (opening cycle), in which the CPU outputs the

new status and asserts all the other relevant control signals.

In addition, the Burst Out Signal (BOUT) is activated by the

CPU indicating that the CPU can perform Burst cycles. If the

selected memory allows Burst cycles, it will notify the CPU

by activating the burst in signal (BIN). BIN is sampled by the

CPU in the middle of T2 on the falling edge of BCLK. If the

memory does not allow burst (BIN high), the cycle will termi-

nate at the end of T2 and BOUT will go inactive immediate-

ly. If the memory allows burst (BIN low), and the CPU has

not deasserted BOUT, the second part of the Burst cycle

will be performed and BOUT will remain active until termina-

tion of the Burst.

The second part consists of up to 3 nibbles, labeled T2B. In

each of them a data item is read by the CPU. For each

nibble in the burst sequence the CPU forces the 2 least-sig-

nificant bits of the address to 0 and increments address bits

2 and 3 to select the next double-word; all the byte enable

signals (BE0–3) are activated.

As shown in Figures 3-25 and 4-8 (in Section 4), the CPU

samples RDY at the end of each nibble and extends the

access time for the burst transfer if RDY is inactive.

The CPU initiates burst read cycles in the following cases.

e

1. An instruction must be fetched (Status

01000 or

01001), and the instruction address does not fall within

the last double-word in an aligned 16-byte block (e.g.,

address bits 2 and 3 are not both equal to 1).

e

01100), and all of the following conditions are met.

2. A data item must be read (Status

01010, 01011 or

e

The data cache is enabled and not locked. (DC

e

1

#

and LDC

0 in the CFG register.)

The addressed page is cacheable as indicated in the

Level-2 Page Table Entry.

#

TL/EE/9354–29

The bus cycle is not an interlocked data access per-

formed while executing a CBITI or SBITI instruction.

FIGURE 3-24. Write Cycle

The Burst sequence will be terminated when one of the

following events occurs.

1. The last instruction double-word in an aligned 16-byte

block has been fetched.

2. The CPU detects that the instructions being prefetched

are no longer needed due to an alteration of the flow of

control. This happens, for example, when a Branch in-

struction is executed or an exception occurs.

3. 4 double-words of data have been read by the CPU. The

double-words are transferred within an aligned 16-byte

block in a wrap-around order. For example, if a source

operand is located at address 104 , then the burst read

16

cycle transfers the double-words at 104, 108, 10C, and

100, in that order.

50

3.0 Functional Description (Continued)

TL/EE/9354–30

FIGURE 3-25. Burst Read Cycles

51

3.0 Functional Description (Continued)

4. The BIN signal is deasserted.

Note 2: The CPU may assert ILO before a read cycle that is cancelled (for

example, due to a TLB miss). In such a case, the CPU deasserts

ILO before performing any additional bus cycles.

5. BRT is asserted to signal a bus retry.

6. IODEC is asserted or the BW0–1 signals indicate a bus

width other than 32-bits. The CPU samples these signals

during state T2 of the opening cycle. During T2B-states

BW0–1 are ignored and IODEC must be kept HIGH.

3.5.4.6 Interrupt Control Cycles

The CPU generates Interrupt-Acknowledge bus cycles in re-

sponse to non-maskable interrupt and enabled maskable

interrupt requests.

The CPU uses only the values of the above signals sampled

during the last state of the transfer when the cycle is ex-

tended. See Section 3.5.4.4.

The CPU also generates one or two End-of-Interrupt bus

cycles during execution of the Return-from-Interrupt (RETI)

instruction.

Note: A burst sequence is not stopped by the assertion of either BER or

CIIN. See Note 3 in Section 3.5.5.

The timing for the interrupt control cycles is the same as for

the basic memory read cycle shown in Figure 3-23 ; only the

status presented on pins ST0–4 is different. These cycles

are single-byte read cycles, and they always bypass the

data cache.

3.5.4.4 Cycle Extension

To allow sufficient access time for any speed of memory or

peripheral device, the NS32532 provides for extension of a

bus cycle. Any type of bus cycle except a slave processor

cycle can be extended.

Table 3-4 shows the interrupt control sequences associated

with each interrupt and with the return from its service pro-

cedure.

A bus cycle can be extended by causing state T2 for a

normal cycle or state T2B for a Burst cycle to be repeated.

3.5.4.7 Slave Processor Bus Cycles

At the end of each T2 or T2B state, on the rising edge of

BCLK, the RDY line is sampled by the CPU. If RDY is active,

then the transfer cycle will be completed. If RDY is inactive,

then the bus cycle is extended by repeating the T-state for

another clock cycle. These additional T-states inserted by

the CPU in this manner are called ‘WAIT’ states.

The NS32532 performs bus cycles to transfer information to

or from slave processors while executing floating-point or

custom-slave instructions.

The CPU uses slave write bus cycles to broadcast the iden-

tification and operation codes of a slave instruction as well

as to transfer operands from memory or general purpose

registers to a slave.

During a transfer the CPU samples the input control signals

BIN, BER, BRT, BW0–1, CIIN and IODEC.

Figure 3-27 shows the timing for a slave write bus cycle.

The CPU asserts SPC during T1; the status is valid during

T1 and T2. The operation code or operand is output on the

data bus from the middle of T1 until the end of T2.

When wait states are inserted, only the values of these sig-

nals sampled during the last wait state are significant.

Figures 3-26 and 4-8 (in Section 4) illustrate both a normal

read cycle and a Burst cycle with wait states added through

the RDY pin.

The CPU uses a slave read bus cycle to transfer a result

operand from a slave to either memory or a general purpose

register. A slave read cycle is also used to read a status

word when the FSSR signal is asserted. Figure 3-28 shows

the timing for a slave read bus cycle.

Note: If RST is asserted during a bus cycle, then the cycle is terminated

without regard of RDY.

3.5.4.5 Interlocked Bus Cycles

The NS32532 supports indivisible read-modify-write trans-

actions by asserting the ILO signal during consecutive read

and write operations. See Figure 4-7 in Section 4.

During T1 and T2 the CPU drives the status lines and as-

serts SPC. The data from the slave is sampled at the end of

T2.

Interlocked transactions are always preceded and followed

by one or more idle T-states.

The CPU will never perform another slave cycle immediately

following a slave read cycle.

The ILO signal is asserted in the middle of the idle T-state

preceding state T1 of the read operation, and is deasserted

in the middle of one of the idle T-states following completion

of the write operation, including any retried bus cycles.

Slave processor data transfers are always 32 bits wide. If

the operand is a single byte, then it is transferred on D0

through D7. If it is a word, then it is transferred on D0

through D15.

No other bus operations (e.g., instruction fetches) will occur

while an interlocked transaction is taking place.

When two operands are transferred, operand 1 is trans-

ferred before operand 2. For double-precision operands, the

least-significant double-word is transferred before the most-

significant double-word.

Interlocked transactions are required in multiprocessor sys-

tems to handle shared resources. The CPU uses them to

reference data while executing a CBITIi or SBITIi instruction,

during which a single byte of data is read and written. They

are also used when the on-chip MMU is updating a Level-2

Page Table Entry during a Page Table Lookup.

During a slave bus cycle the output signals BE0–3 are un-

defined while the input signals BW0–1 and RDY are ig-

nored.

BER and BRT must be kept high.

In this case a double-word is read and written. If the Level-2

Page Tables are located in a memory area whose width is

other than 32 bits, multiple interlocked reads followed by

multiple interlocked writes will result. The ILO signal is al-

ways released for one or more clock cycles in the middle of

two consecutive interlocked transactions.

Note 1: If a bus error is detected during an interlocked read cycle, the sub-

sequent interlocked write cycle will not be performed, and ILO is

deasserted before the next bus cycle begins.

52

3.0 Functional Description (Continued)

TL/EE/9354–31

3-26. Cycle Extension of a Basic Read Cycle

53

3.0 Functional Description (Continued)

TABLE 3-4. Interrupt Sequences

Data Bus

â

V

W

Byte 0

Cycle Status

Address

DDIN

BE3

BE2

BE1

BE0

Byte 3 Byte 2 Byte 1

A. Non-Maskable Interrupt Control Sequences

Interrupt Acknowledge

1

00100 FFFFFF00

0

1

0

X

16

Interrupt Return

None: Performed through Return from Trap (RETT) instruction.

B. Non-Vectored Interrupt Control Sequences

Interrupt Acknowledge

1

00100 FFFFFE00

0

1

0

X

16

Interrupt Return

1

00110 FFFFFE00

C. Vectored Interrupt Sequences: Non-Cascaded

Interrupt Acknowledge

1

00100 FFFFFE00

0

1

0

X

Vector:

16

Range: 0–127

Interrupt Return

1

00110 FFFFFE00

Vector: Same as

in Previous Int.

Ack. Cycle

16

D. Vectored Interrupt Sequences: Cascaded

Interrupt Acknowledge

1

00100 FFFFFE00

0

1

0

X

Cascade Index:

b

range 16 to

16

b

1

(The CPU here uses the Cascade Index to find the Cascade Address)

2

001101

Cascade

Address

0

See Note

Vector, range 16–255; on appropriate byte of

data bus.

Interrupt Return

00110 FFFFFE00

1

X

Cascade Index:

Same as in

16

previous Int.

Ack. Cycle

(The CPU here uses the Cascade Index to find the Cascade Address)

2

00111

Cascade

Address

0

See Note

X

e

Note: BE0–BE3 signals will be activated according to the cascaded ICU address

X

Don’t Care

54

3.0 Functional Description (Continued)

TL/EE/9354–32

TL/EE/9354–33

FIGURE 3-27. Slave Processor Write Cycle

FIGURE 3-28. Slave Processor Read Cycle

3.5.5 Bus Exceptions

When BER is sampled active, the CPU completes the bus

cycle normally. If a bus error occurs during a bus cycle for a

reference required to execute an instruction, then a bus er-

ror exception is recognized. However, if an error occurs dur-

ing an acknowledge cycle of another exception or during

the ICU read cycle of a RETI instruction, the CPU interprets

the event as a fatal bus error and enters the ‘halted’ state.

The NS32532 has the capability of handling errors occurring

during the execution of a bus cycle. These errors can be

either correctable or incorrectable, and the CPU can be no-

tified of their occurrence through the input signals BRT and/

or BER.

Bus Retry

In this state the CPU floats its address and data buses and

places a special status code on the ST0–4 lines. The CPU

can exit this condition only through a hardware reset. Refer

to Section 3.2.6 for more details on bus error.

If a bus error can be corrected, the CPU may be requested

to repeat the erroneous bus cycle. The request is done by

asserting the BRT signal. BRT is sampled at the end of

state T2 or T2B.

Note 1: If the erroneous bus cycle is extended by means of wait states, then

the CPU uses the values of BRT and/or BER sampled during the

last wait state.

When the CPU detects that BRT is active, it completes the

bus cycle normally, but ignores the data read in case of a

read cycle, and maintains a copy of the data to be written in

case of a write cycle. Then, after a delay of two clock cy-

cles, it will start executing the bus cycle again.

Note 2: If the CPU samples both BRT and BER active, BRT has higher

priority. The bus error indication is ignored, and the bus cycle is

repeated.

Note 3: If BER is asserted during a bus cycle of a multi-cycle data transfer,

the CPU completes the entire transfer normally, but the data will be

ignored. The CPU also ignores any subsequent assertion of BER

during the same data transfer.

If the transfer cycle is multiple (e.g., for non-aligned data),

only the problematic part will be repeated.

For instance, if a non-aligned double-word is being trans-

ferred and the second half of the transfer fails, only the

second part will be repeated.

Note 4: Neither BRT nor BER should be asserted during the T2 state of a

slave processor bus cycle.

3.5.6 Dynamic Bus Configuration

The same applies for a retry during a burst sequence. The

repeated cycle will begin where the read operation failed

(rather than the first address of the burst) and will finish the

original burst.

The NS32532 is tuned to operate with 32-bit wide memory

and peripheral devices. The bus also supports 8-bit and

16-bit data widths, but at reduced efficiency. The CPU can

switch from one bus width to another dynamically; the only

restriction is that the bus width cannot change for locations

within an aligned 16-byte block.

Figures 3-29 and 4-10 (in Section 4) show the BRT timing

for a basic access cycle and for burst cycles respectively.

The CPU always waits for BRT to be HIGH before repeating

the bus cycle. While BRT is LOW, the CPU places all the

The CPU determines the bus width in effect for a bus cycle

by using the values of the BW0 and BW1 signals sampled

during the last T2 state. Values of BW0 and BW1 sampled

before the last T2 state or during T2B states are ignored.

Whenever a bus width other than 32-bit is detected by the

CPU, two idle states are inserted before the next bus cycle

is initiated. These idle states are only inserted once during

an operand access, even if more than two bus cycles are

needed to complete the access.

output signals shown inFigure 4-11 in a TRI-STATE condi-

tion.

É

Bus Error

If a bus error is incorrectable the CPU may be requested to

interrupt the current process and branch to an appropriate

procedure to handle the error. The request is performed by

activating the BER signal. BER is sampled by the CPU at

the end of state T2 or T2B on the rising edge of BCLK.

55

3.0 Functional Description (Continued)

TL/EE/9354–34

FIGURE 3-29. Bus Retry During a Basic Read Cycle

56

3.0 Functional Description (Continued)

The various combinations for BW0 and BW1 are shown be-

low.

The following subsections provide detailed descriptions of

the access sequences performed in the various cases.

Note: Although the NS32532 ignores the BIN signal for 8-bit and 16-bit bus

widths, it is recommended that BIN be asserted only if the system

supports burst transfers. This is to ensure compatibility with future

versions of the CPU that might support burst transfers for 8-bit and

16-bit buses.

BW1

BW0

0

1

0

1

0

1

Reserved

8-Bit Bus

16-Bit Bus

32-Bit Bus

The bus width must always be 32 bits during slave cycles.

An important feature of the NS32532 is that it does not

impose any restrictions on the data alignment, regardless of

the bus width.

Bus accesses are performed in double-word units. Access-

es of data operands that cross double-word boundaries are

decomposed into two or more aligned double-word access-

es.

The CPU provides four byte enable signals (BE0–3) which

facilitate individual byte accessing on either a 32-bit or a

16-bit bus.

Figures 3-30 and 3-31 show the basic interfaces for 32-bit

and 16-bit memories. An 8-bit memory interface (not shown)

is even simpler since it does not use any of the BE0–3

signals and its single bank is always enabled whenever the

memory is selected. Each byte location in this case is se-

lected by address bits A0–31.

The NS32532 does not keep track of the bus width used in

previous instruction fetches or data accesses. At the begin-

ning of every memory transaction, the CPU always assumes

that the bus is 32-bit wide and the BE0–3 signals are acti-

vated accordingly.

The BOUT signal is also asserted during instruction fetches

or data reads if the conditions for bursting are satisfied. If

the bus is other than 32-bit wide, the BIN signal is ignored

and BOUT is deasserted at the beginning of the T state

following T2, since burst cycles are not allowed for 8-bit or

16-bit buses.

TL/EE/9354–36

FIGURE 3-31. Basic Interface for 16-Bit Memories

3.5.6.1 Instruction Fetch Sequences

The CPU performs two types of instruction fetch cycles: se-

quential and non-sequential. These can be distinguished

from each other by the differing status combinations on pins

ST0–4. For non-sequential instruction fetches the CPU

presents on the address bus the exact byte address of the

first instruction in the instruction stream that is about to be-

gin; for sequential instruction fetches, the address of the

next aligned instruction double-word is presented on the ad-

dress bus. The CPU always activates all byte enable signals

(BE0–3) for both sequential and non-sequential fetches.

BOUT is also asserted during T2 if the addressed double-

word is not the last in an aligned 16-byte block. Tables 3-5

to 3-7 show the fetch sequence for the various bus widths.

32-Bit Bus Width

The CPU reads the entire double-word present on the data

bus into its internal instruction buffer.

If BOUT and BIN are both active, the CPU reads up to 3

consecutive double-words using burst cycles. Burst cycles

are used for instruction fetches regardless of whether the

accesses are cacheable.

TL/EE/9354–35

FIGURE 3-30. Basic Interface for 32-Bit Memories

Note: The CACH signal must be asserted during cacheable read accesses.

57

3.0 Functional Description (Continued)

@

6

Example: JUMP

5

Example JUMP

The CPU performs a fetch cycle at address 5 with BE0–3

all active.

A fetch cycle is performed at address 6 with BE0–3 all

active.

#

Two burst cycles are then performed and addresses 8 and

12 are output while BE0–3 are kept active.

The word at address 4 is then fetched if the access is

cacheable.

#

16-Bit Bus Width

8-Bit Bus Width

The word on the least-significant half of the data bus is read

by the CPU. This is either the even or the odd word within

the required instruction double-word, as determined by ad-

dress bit 1.

The instruction byte on the bus lines D0–7 is fetched. The

CPU performs three consecutive cycles to read the remain-

ing bytes within the required double-word, while keeping

BE0–3 all active. The 4 bytes are then assembled into a

double-word and transferred into the instruction buffer. For

a non-sequential fetch, if the access is not cacheable, the

CPU will only read the upper bytes within the instruction

double-word starting with the byte at the instruction ad-

dress.

The CPU then complements address bit 1, clears address

bit 0 and initiates a bus cycle to read the other word, while

keeping all the BE0–3 signals active.

These two words are then assembled into a double-word

and transferred into the instruction buffer.

@

Example: JUMP

7

In case of a non-sequential fetch, if the access is not cache-

able and the instruction address selects the odd word within

the instruction double-word, the even word is not fetched.

The CPU performs a fetch cycle at address 7 with BE0–3

all active.

#

Bytes at addresses 4, 5 and 6 are then fetched consecu-

tively if the access is cacheable.

#

TABLE 3-5. Cacheable/Non-Cacheable Instruction Fetches from a 32-Bit Bus

1. In a burst access four bytes are fetched with the L.S. bits of the address set to 00.

2. A ‘C’ on the data bus refers to cacheable fetches and indicates that the byte is placed in the instruction cache. An ‘I’ refers

to non-cacheable fetches and indicates that the byte is ignored.

Number

of Bytes

Address

LSB

Address

Bus

Bytes to be Fetched

BE0–3

Data Bus

1

2

3

4

11

10

01

00

B0

B1

B2

B3

Ð

B0

B1

B2

Ð

B0

A

L L L L

B0

B1

B2

B3

C/I

B0

B1

C/I

B0

B1

B2

B0

B1

TABLE 3-6. Cacheable/Non-Cacheable Instruction Fetches from a 16-Bit Bus

1. A bus access marked with ‘*’ in the ‘Address Bus’ column is performed only if the fetch is cacheable.

Number

of Bytes

Address

LSB

Address

Bus

Bytes to be Fetched

BE0–3

Data Bus

1

2

3

4

11

10

01

00

B0

B1

B2

B3

Ð

B0

B1

B2

Ð

B0

A

L L L L

Ð

B0

C

C/I

C

b

*A

A

3

2

1

2

Ð

A

L L L L

Ð

B1

C

B0

C

b

B0

B1

A

L L L L

Ð

B0

B2

C/I

B1

a

A

L L L L

Ð

B1

B3

B0

B2

a

A

58

3.0 Functional Description (Continued)

TABLE 3-7. Cacheable/Non-Cacheable Instruction Fetches from an 8-Bit Bus

Number

of Bytes

Address

LSB

Address

Bus

Bytes to be Fetched

BE0–3

Data Bus

1

2

3

4

11

10

01

00

B0

B1

B2

B3

Ð

B0

B1

B2

Ð

B0

A

L L L L

Ð

B0

C

b

* A

3

2

1

Ð

C

A

L L L L

Ð

B0

B1

C

a

b

A

* A

1

2

1

C

B0

B1

A

L L L L

Ð

B0

B1

B2

C

a

b

A

1

2

1

* A

A

L L L L

Ð

B0

B1

B2

B3

a

A

1

2

3

3.5.6.2 Data Read Sequences

16-Bit Bus Width

The CPU starts a data read access by placing the exact

address of the operand on the address bus. The byte en-

able lines are activated to select only the bytes required by

the instruction being executed. This prevents spurious ac-

cesses to peripheral devices that might be sensitive to read

accesses, such as those which exhibit the characteristic of

destructive reading. If the on-chip data cache is internally

enabled for the read access, the BOUT signal is asserted at

the beginning of state T2. BOUT will be deasserted if the

data cache is externally inhibited (through CIIN or IODEC),

or the bus width is other than 32 bits. During cacheable

accesses the CPU always reads all the bytes in the double-

word, whether or not they are needed to execute the in-

struction, and stores them into the data cache. The external

memory, in this case, must place the data on the bus re-

gardless of the state of the byte enable signals.

The word on the least-significant half of the data bus is read

by the CPU. The CPU can then perform another access

cycle with address bit 1 complemented and address bit 0

cleared to read the other word within the addressed double-

word.

If the access is cacheable, the entire double-word is read

and stored into the cache.

If the access is not cacheable, the CPU ignores the bytes in

the double-word not selected by BE0–3. In this case, the

second access cycle is not performed, unless selected

bytes are contained in the second word.

@

Example: MOVB 5, R0

The CPU reads a word at address 5 while keeping BE1

active.

#

If the access is not cacheable, the CPU ignores byte 0.

#

If the data cache is either internally or externally inhibited

during the access, the CPU ignores the bytes not selected

by the BE0–3 signals. Data read sequences for the various

bus widths are shown in tables 3-8 to 3-10.

If the access is cacheable, the CPU performs another ac-

cess cycle, with BE0–3 all active, to read the word at

address 6.

#

8-Bit Bus Width

32-Bit Bus Width

The data byte on the bus lines D0–7 is read by the CPU.

The CPU can then perform up to 3 access cycles to read

the remaining bytes in the double-word.

The entire double-word present on the bus is read by the

CPU. If the access is cacheable and the memory allows

burst accesses, the CPU reads up to 3 additional double-

words within the aligned 16-byte block containing the first

byte of the operand. These burst accesses are performed in

a wrap-around fashion within the 16-byte block.

If the access is cacheable, the entire double-word is read

and stored into the cache.

If the access is not cacheable, the CPU will only perform

those access cycles needed to read the selected bytes.

@

Example: MOVW 5, R0

@

Example: MOVW 5, R0

The CPU reads a double-word at address 5 while keeping

BE1 and BE2 active.

#

The CPU reads the byte at address 5 while keeping BE1

and BE2 active.

#

If the access is not-cacheable, BOUT is deasserted and

the data bytes 0 and 3 are ignored.

#

If the access is not cacheable, the CPU activates BE2 and

reads the byte at address 6.

#

If the access is cacheable, the CPU performs burst cycles

with BE0–3 all active, to read the double-words at ad-

dresses 8, 12, and 0.

#

If the access is cacheable, the CPU performs three bus

cycles with BE0–3 all active, to read the bytes at address-

es 6, 7 and 4.

#

59

3.0 Functional Description (Continued)

TABLE 3-8. Cacheable/Non-Cacheable Data Reads from a 32-Bit Bus

1. In a burst access four bytes are read with the L.S. bits of the address set to 00.

2. A ‘C’ on the data bus refers to cacheable reads and indicates that the byte is placed in the data cache. An ‘I’ refers to non-

cacheable reads and indicates that the byte is ignored.

Number

of Bytes

Address

LSB

Address

Bus

Bytes to be Read

BE0–3

Data Bus

C/I

1

2

3

4

00

01

10

11

00

01

10

00

01

00

Ð

B0

Ð

B0

Ð

B0

Ð

B0

Ð

B0

A

H H H L

H H L H

H L H H

L H H H

H H L L

H L L H

L L H H

H L L L

L L L H

L L L L

C/I

B0

C/I

B0

C/I

B1

B0

C/I

B1

B0

B1

B0

C/I

B0

C/I

B0

C/I

B1

B0

B2

B1

B2

Ð

B0

Ð

BO

Ð

B1

B0

Ð

C/I

B1

Ð

B1

B0

B2

B1

B2

C/I

B0

B1

Ð

B1

B0

B1

C/I

B2

B3

C/I

B0

B3

TABLE 3-9. Cacheable/Non-Cacheable Data Reads from a 16-Bit Bus

1. A bus access marked with ‘*’ in the ‘Address Bus’ column is performed only if the read is cacheable.

Number

of Bytes

Address

LSB

Address

Bus

BE0–3

Non Cach.

Data to be Read

Data Bus

Cach.

1

2

3

4

00

01

10

11

00

01

10

00

01

00

Ð

B0

Ð

B0

Ð

B0

Ð

B0

Ð

B0

A

H H H L

L L L L

H H H L

H H L H

H L H H

L H H H

H H L L

Ð

C/I

C

B0

C

a

* A

A

2

1

2

3

2

1

2

1

2

Ð

A

H H L H

L L L L

Ð

B0

C

C/I

C

a

Ð

B0

Ð

A

H L H H

L L L L

Ð

C/I

C

B0

C

b

B0

Ð

A

L H H H

L L L L

Ð

B0

C

C/I

C

b

Ð

B1

B0

Ð

A

H H L L

L L L L

Ð

B1

C

B0

C

a

Ð

B1

B0

B2

B1

B2

A

H L L H

L L L L

H L L H

H L H H

Ð

B0

C/I

B1

a

C/I

B1

Ð

A

L L H H

L L L L

L L H H

Ð

B1

C

B0

C

b

* A

B1

B0

B1

A

H L L L

L L L L

H L L L

Ð

B1

B0

B2

a

A

H L H H

C/I

B2

B3

A

L L L H

L L L L

L L L H

L L H H

Ð

B0

B2

C/I

B1

a

A

L L L L

Ð

B1

B3

B0

B2

a

A

L L H H

60

3.0 Functional Description (Continued)

TABLE 3-10. Cacheable/Non-Cacheable Data Reads from an 8-Bit Bus D8–12

Number

of Bytes

Address

LSB

Address

Bus

BE0–3

Non Cach.

Data to be Read

Data Bus

Cach.

1

2

3

4

00

01

10

11

00

01

10

00

01

00

Ð

B0

Ð

B0

Ð

B0

Ð

B0

Ð

B0

A

H H H L

L L L L

H H H L

H H L H

H L H H

L H H H

Ð

B0

C

a

*A

1

2

3

C

A

H H L H

L L L L

Ð

B0

C

a

b

*A

1

2

1

C

Ð

B0

Ð

A

H L H H

L L L L

Ð

B0

C

a

b

*A

1

2

1

C

B0

Ð

A

L H H H

L L L L

Ð

B0

C

b

*A

3

2

1

C

Ð

B1

B0

Ð

A

H H L L

L L L L

H H L L

H H L H

Ð

B0

B1

C

a

A

*A

1

2

3

C

Ð

B1

B0

B2

B1

B2

A

H L L H

L L L L

H L L H

H L H H

Ð

B0

B1

C

a

b

A

*A

1

2

1

C

B1

Ð

A

L L H H

L L L L

L L H H

L H H H

Ð

B0

B1

C

a

b

A

*A

1

2

1

C

B1

B0

B1

A

H L L L

L L L L

H L L L

H L L H

H L H H

Ð

B0

B1

B2

C

a

A

1

2

3

*A

B2

B3

A

L L L H

L L L L

L L L H

L L H H

L H H H

Ð

B0

B1

B2

C

a

b

A

1

2

1

*A

A

L L L L

L L L H

L L H H

L H H H

Ð

B0

B1

B2

B3

a

A

1

2

3

3.5.6.3 Data Write Sequences

32-Bit Bus Width

In a write access the CPU outputs the operand address and

asserts only the byte enable lines needed to select the spe-

cific bytes to be written.

The CPU performs only one access cycle to write the se-

lected bytes within the addressed double-word.

@

Example: MOVB R0,

6

In addition, the CPU duplicates the data to be written on the

appropriate bytes of the data bus in order to handle 8-bit

and 16-bit buses.

The CPU duplicates byte 2 of the data bus into byte 0 and

performs a write cycle at address 6 with BE2 active.

#

16-Bit Bus Width

The various access sequences as well as the duplication of

data are summarized in tables 3-11 to 3-13.

Up to two access cycles are needed to complete the write

operation.

61

3.0 Functional Description (Continued)

@

Example: MOVW R0,

5

signals. By asserting HOLD, an external device requests ac-

cess to the bus. On receipt of HLDA from the CPU, the

device may perform bus cycles, as the CPU at this point has

placed all the output signals shown in Figure 3-32 into the

TRI-STATE condition.

The CPU duplicates byte 1 of the data bus into byte 0 and

performs a write cycle at address 5 with BE1 and BE2

active.

#

A write at address 6 is then performed with BE2 active

and the original byte 2 of the data bus placed on byte 0.

#

To return control of the bus to the CPU, the external device

sets HOLD inactive, and the CPU acknowledges return of

the bus by setting HLDA inactive.

8-Bit Bus Width

Up to 4 access cycles are needed in this case to complete

the write operation.

The CPU samples HOLD in the middle of each T-state on

the falling edge of BCLK. If HOLD is asserted when the bus

is idle between access sequences, then the bus is granted

immediately (see Figure 3-31). If HOLD is asserted during

an access sequence, then the bus is granted immediately

after the access sequence, including any retried bus cycles,

has completed (see Figure 4-13). Note that an access se-

quence can be composed of several bus cycles if the bus

width is 8 or 16 bits.

@

Example: MOVB R0,

7

The CPU duplicates byte 3 of the data bus into bytes 0

and 1, and then performs a write cycle at address 7 with

BE3 active.

#

3.5.7 Bus Access Control

The NS32532 has the capability of relinquishing its control

of the bus upon request from a DMA device or another CPU.

This capability is implemented with the HOLD and HLDA

TABLE 3-11. Data Writes to a 32-Bit Bus

1. Bytes on the data bus marked with ‘ ’ are undefined.

#

Number

of Bytes

Address

LSB

Address

Data to be Written

BE0–3

Data Bus

Bus

1

2

3

4

00

01

10

11

00

01

10

00

01

00

Ð

B0

Ð

B0

Ð

B0

Ð

B0

Ð

B0

A

H H H L

H H L H

H L H H

L H H H

H H L L

H L L H

L L H H

H L L L

L L L H

L L L L

B0

#

A

B0

#

Ð

B0

Ð

A

B0

#

B0

Ð

A

B0

B1

B0

B1

B0

B1

#

Ð

B1

B0

Ð

A

#

Ð

B1

B0

B2

B1

B2

A

B1

B0

B2

B1

B2

#

B1

Ð

A

B1

B0

B1

A

#

B2

B3

A

B2

B3

A

TABLE 3-12. Data Writes to a 16-Bit Bus

Address

Number

of Bytes

Address

LSB

Data to be Written

BE0–3

Data Bus

Bus

1

2

00

01

10

11

00

01

Ð

B0

Ð

B0

Ð

B1

Ð

B0

Ð

B0

Ð

B0

Ð

A

H H H L

H H L H

H L H H

L H H H

H H L L

B0

#

A

B0

A

B0

#

Ð

A

B0

B1

#

B1

B0

A

#

A

H L L H

H L H H

B1

B0

B1

#

a

A

1

#

2

3

10

00

B1

Ð

B0

B2

Ð

A

L L H H

B1

B0

B1

B0

B1

B0

A

H L L L

B2

B1

B0

B2

#

a

A

2

1

2

H L H H

#

3

4

01

00

B2

B3

B1

B2

B0

B1

Ð

A

L L L H

L L H H

B2

B1

B0

B2

B0

B1

a

#

B0

A

L L L L

B3

B2

B1

B3

B0

B2

a

L L H H

#

62

3.0 Functional Description (Continued)

TABLE 3-13. Data Writes to an 8-Bit Bus

Number

of Bytes

Address

LSB

Address

Bus

Data to be Written

BE0–3

Data Bus

1

2

00

01

10

11

00

Ð

B0

Ð

B0

Ð

B0

Ð

B0

Ð

B0

A

H H H L

H H L H

H L H H

L H H H

B0

#

B0

#

Ð

B0

#

B1

H H L L

H H L H

B1

B0

B1

#

a

A

1

#

2

3

01

10

00

Ð

B1

Ð

B1

B0

B2

B0

Ð

A

H L L H

H L H H

B1

B0

B1

#

a

#

A

L L H H

L H H H

B1

B0

B1

B0

B1

a

#

B1

B0

A

H L L L

H L L H

H L H H

B2

#

B1

#

B0

B1

B2

#

a

A

1

2

3

4

01

00

B2

B3

B1

B2

B0

B1

Ð

A

L L L H

L L H H

L H H H

B2

#

B1

#

B0

#

B0

B1

B2

a

A

1

2

B0

A

L L L L

L L L H

L L H H

L H H H

B3

#

B2

#

B1

#

B0

B1

B2

B3

a

A

1

2

3

#

63

3.0 Functional Description (Continued)

TL/EE/9354–37

FIGURE 3-32. Hold Acknowledge. (Bus Initially Idle.)

Note: The status indicates ‘IDLE’ while the bus is granted. If the cause of the IDLE changes (e.g., CPU starts waiting for an interrupt), the status also changes.

The CPU will never grant the bus between interlocked read

and write bus cycles.

When IODEC is active during a bus cycle for which IOINH is

asserted, the CPU discards the data and applies the special

handling required for I/O devices. Figure 3-33 shows a pos-

sible implementation of an I/O device interface where the

address mapping of the I/O devices is fixed.

Note: If an external device requires a very short latency to get control of the

bus, the bus retry signal (BRT) can be used instead of hold. See

Section 3.5.5.

3.5.8 Interfacing Memory-Mapped I/O Devices

In an open system configuration, IODEC could be generated

by the decoding logic of each I/O device subsystem.

In Section 3.1.3.2 it was mentioned that some special pre-

cautions are needed when interfacing I/O devices to the

NS32532 due to its internal pipelined implementation. Two

special signals are provided for this purpose: IOINH and

IODEC. The CPU asserts IOINH during a read bus cycle to

indicate that the bus cycle should be ignored if an I/O de-

vice is selected. The system responds by asserting IODEC

to indicate to the CPU that an I/O device has been select-

ed. IODEC is sampled by the CPU in the middle of state T2.

If the cycle is extended, then the CPU uses the IODEC val-

ue sampled during the last wait state. If a bus error or a bus

retry occurs, the sampled IODEC value is ignored. IODEC

must be kept high during burst transfer cycles.

When the on-chip MMU is enabled, the CIOUT signal could

also be used for this purpose, since I/O devices are located

in noncacheable areas. In this case however, a small per-

formance degradation could result, due to the fact that the

special I/O handling is also applied on references to non-

cacheable program and/or data areas.

Note 1: When IODEC is active in response to a read bus cycle, the CPU

treats the reference as noncacheable.

Note 2: IOINH is kept inactive during write cycles.

64

3.0 Functional Description (Continued)

INVIC, INVDC, INVSET and CIA0–CIA6 are all sampled

synchronously by the CPU on the rising edge of BCLK. The

CPU can respond to cache invalidation requests at a rate of

one per BCLK cycle.

As shown in Figures 3-16 and 3-17, the validity bits of the

on-chip caches are dual-ported. One port is used for ac-

cessing and updating the caches, while the other port is

used independently for invalidation requests. Consequently,

invalidation of the on-chip caches occurs with no interfer-

ence to on-going cache accesses or bus cycles.

TL/EE/9354–38

FIGURE 3-33. Typical I/O Device Interface

3.5.9 Interrupt and Debug Trap Requests

A cache invalidation request can occur during a read bus

cycle for a location affected by the invalidation. In such a

case, the data will be invalid in the cache if the invalidation

request occurs after the T2- or T2B-state of the bus cycle.

Three signals are provided by the CPU to externally request

interrupts and/or a debug trap. INT and NMI are for maska-

ble and non-maskable interrupts respectively. DBG is used

for requesting an external debug trap.

Note: In the case of the Data Cache, the cache location will also be invali-

dated if the invalidation occurs during T2 or T2B of the read cycle.

Refer to Figure 4-18 in Section 4 for timing details.

The CPU samples INT and NMI on every other rising edge

of BCLK, starting with the second rising edge of BCLK after

RST goes high.

3.5.11 Internal Status

The NS32532 provides information on the system interface

concerning its internal activity.

NMI is edge-sensitive; a high-to-low transition on it is detect-

ed by the CPU and stored in an internal latch, so that there

is no need to keep it asserted until it is acknowledged.

The U/S signal indicates the Address Space for a memory

reference (See Section 2.4.2).

INT is level-sensitive and, as such, once asserted, it must

be kept asserted until it is acknowledged.

Note that U/S does not necessarily reflect the value of the

U bit in the PSR register. For example, U/S is high during

the memory access used to store the destination operand of

a MOVSU instruction.

The DBG signal, like NMI, is edge-sensitive; it differs from

NMI in that the CPU samples it on each rising edge of

BCLK. DBG can be asserted asynchronously to the CPU

clock, but it should be at least 1.5 clock cycles wide in order

to be recognized.

The PFS signal is asserted for one BCLK cycle when the

CPU begins executing a new instruction. The ISF signal is

driven High along with PFS if the new instruction does not

follow the previous instruction in sequence. More specifical-

ly, ISF is High along with PFS after processing an exception

or after executing one of the following instructions: ACB

(branch taken), Bcond (branch taken), BR, BSR, CASE,

CXP, CXPD, DIA, JSR, JUMP, RET, RETT, RETI, and RXP.

If DBG meets the specified setup and hold times, it will be

recognized on the rising edge of BCLK deterministically.

Refer to Figures 4-19 and 4-20 for more details on the tim-

ing of the above signals.

Note: If the NMI signal is pulsed to request a non-maskable interrupt, it may

be necessary to keep it asserted for a minimum of two clock cycles to

guarantee its detection, unless extra logic ensures that the pulse oc-

curs around the BCLK sampling edge.

The BP signal is asserted for one BCLK cycle when an ad-

dress-compare or PC-match condition is detected. If the BP

signal is asserted one BCLK cycle after PFS, it indicates

that an address-compare debug condition has been detect-

ed. If BP is asserted at any other time, it indicates that a PC-

Match debug condition has been detected.

3.5.10 Cache Invalidation Requests

The contents of the on-chip Instruction and Data Caches

can be invalidated by external requests from the system. It

is possible to invalidate a single set or all sets in the Instruc-

tion Cache, Data Cache or both. The input signals INVIC

and INVDC request invalidation of the Instruction Cache

and Data Cache respectively. The input signal INVSET indi-

cates whether the invalidation applies to a single set (16

bytes for the Instruction Cache and 32 bytes for the Data

Cache) or to the entire cache. When only a single set is

invalidated, the set number is specified on CIA0–CIA6.

While executing an LMR or CINV instruction, the CPU dis-

plays the operation code and source operand using slave

processor write bus cycles. This information can be used to

monitor the contents of the on-chip TLB, Instruction Cache

and Data Cache.

During idle bus cycles, the signals ST0–ST4 indicate wheth-

er the CPU is waiting for an interrupt, waiting for a Slave

Processor to complete executing an instruction or halted.

65

4.0 Device Specifications

TL/EE/9354–39

FIGURE 4-1. NS32532 Interface Signals

4.1 NS32532 PIN DESCRIPTIONS

Descriptions of the NS32532 pins are given in the following

sections.

4.1.2 Input Signals

CLK

Clock.

Included are also references to portions of the functional

description, Section 3.

Input Clock used to derive all CPU Timing.

SYNC

Synchronize.

Figure 4-1 shows the NS32532 interface signals grouped

according to related functions.

When SYNC is active, BCLK will stop tog-

gling. This signal can be used to synchronize

two or more CPUs (Section 3.5.2).

Note: An asterisk next to the signal name indicates a TRI-STATE condition

for that signal when HOLD is acknowledged or during an extended

retry.

HOLD

Hold Request.

When active, causes the CPU to release the

bus for DMA or multiprocessing purposes

(Section 3.5.7).

4.1.1 Supplies

VCCL1–6

Logic Power.

a

5V positive supplies for on-chip logic.

Note:

VCCB1–14 Buffers Power.

If the HOLD signal is generated asynchronously, its set

up and hold times may be violated. In this case it is rec-

ommended to synchronize it with the falling edge of

BCLK to minimize the possibility of metastable states.

a

buffers.

5V positive supplies for on-chip output

VCCCLK

Bus Clock Power.

The CPU provides only one synchronization stage to min-

imize the HLDA latency. This is to avoid speed degrada-

tions in cases of heavy HOLD activity (i.e. DMA controller

cycles interleaved with CPU cycles).

a

ers.

5V positive supply for on-chip clock driv-

GNDL1–6

Logic Ground.

RST

INT

Reset.

Ground references for on-chip logic.

When RST is active, the CPU is initialized to

a known state (Section 3.5.3).

GNDB1–13 Buffers Ground.

Ground references for on-chip output buffers.

Bus Clock Ground.

Ground reference for on-chip clock drivers.

Interrupt.

GNDCLK

A low level on this signal requests a maska-

ble interrupt (Section 3.5.9).

NMI

Nonmaskable Interrupt.

A High-to-Low transition of this signal re-

a nonmaskable interrupt (Section

quests

3.5.9).

66

4.0 Device Specifications (Continued)

DBG

Debug Trap Request.

10Ð16 Bits

11Ð32 Bits

Bus Retry.

A High-to-Low transition of this signal re-

quests a debug trap (Section 3.5.9).

BRT

BER

CIA0–6

Cache Invalidation Address Bus.

When active, the CPU will reexecute the last

bus cycle (Section 3.5.5).

Bits 0 through 4 specify the set address to

invalidate in the on-chip caches. CIA0 is the

least significant. Bits 5 and 6 are reserved

(Section 3.5.10).

Bus Error.

When active, indicates that an error occurred

during a bus cycle. It is treated by the CPU as

the highest priority exception after reset.

INVSET

INVDC

Invalidate Set.

When Low, only a set in the on-chip cache(s)

is invalidated; when High, the entire cache(s)

is (are) invalidated.

4.1.3 Output Signals

BCLK

HLDA

Bus Clock.

Output clock for bus timing (Section 3.5.2).

Bus Clock Inverse.

Invalidate Data Cache.

When Low, the Data Cache contents are in-

validated. INVSET determines whether a sin-

gle set or the entire Data Cache is invalidat-

ed.

Inverted output clock.

Hold Acknowledge.

Activated by the CPU in response to the

HOLD input to indicate that the CPU has re-

leased the bus.

INVIC

CIIN

Invalidate Instruction Cache.

When Low, the Instruction Cache contents

are invalidated. INVSET determines whether

a single set or the entire Instruction Cache is

invalidated.

PFS

ISF

Program Flow Status.

A pulse on this signal indicates the beginning

of execution for each instruction (Section

3.5.11).

Cache Inhibit In.

When active, indicates that the location refer-

enced in the current bus cycle is not cache-

able. CIIN must not change within an aligned

16-byte block.

Internal Sequential Fetch.

Indicates along with PFS that the instruction

beginning execution is sequential (ISF Low)

or non-sequential (ISF High).

IODEC

FSSR

I/O Decode.

U/S

BP

User/Supervisor.

Indicates to the CPU that a peripheral device

is addressed by the current bus cycle. The

value of IODEC must not change within an

aligned 16-byte block (Section 3.5.8).

User or supervisor mode status.

Break Point.

This signal is activated when the CPU de-

tects a PC or operand-address match debug

condition (Section 3.3.2).

Force Slave Status Read.

When asserted, indicates that the slave

status word should be read by the CPU (Sec-

tion 3.1.4.1). An external 10 kX resistor

should be connected between FSSR and

CASEC

CIOUT

*Cache Section.

For cacheable data read bus cycles indicates

the Section of the on-chip Data Cache where

the data will be placed; undefined for other

bus cycles. This signal can be used for exter-

nal monitoring of the data cache contents.

V

.

CC

SDN

Slave Done.

Used by a slave processor to signal the com-

a slave instruction (Section

Cache Inhibit Out.

pletion of

3.1.4.1). An external 10 kX resistor should be

This signal reflects the state of the CI bit in

the second level page table entry (PTE). It is

used to specify non-cacheable pages. It is

held low while address translation is disabled

and for MMU references to page table en-

tries.

connected between SDN and V

.

CC

BIN

Burst In.

When active, indicates to the CPU that the

memory supports burst cycles (Section

3.5.4.3).

IOINH

I/O Inhibit.

RDY

Ready.

Indicates that the current bus cycle should

a peripheral device is ad-

While this signal is not active, the CPU ex-

tends the current bus cycle to support a slow

memory or peripheral device.

be ignored if

dressed.

SPC

Slave Processor Control.

Data strobe for slave processor transfers.

*Burst Out.

BW0–1

Bus Width.

These lines define the bus width (8, 16 or 32

bits) for each data transfer; BW0 is the least

significant bit. The bus width must not

change within an aligned 16-byte blockÐen-

codings are:

BOUT

When active, indicates that the CPU is re-

questing to perform burst cycles.

ILO

Interlocked Operation.

00ÐReserved

01Ð8 Bits

When active, indicates that interlocked cy-

cles are being performed (Section 3.5.4.5).

67

4.0 Device Specifications (Continued)

DDIN

*Data Direction.

00101ÐInterrupt Acknowledge, Cascaded.

00110ÐEnd of Interrupt, Master.

00111ÐEnd of Interrupt, Cascaded.

01000ÐSequential Instruction Fetch.

01001ÐNon-Sequential Instruction Fetch.

01010ÐData Transfer.

Indicates the direction of a data transfer. It is

low for reads and high for writes.

CONF

*Confirm Bus Cycle.

When active, indicates that a bus cycle initia-

ted by ADS is valid; that is, the bus cycle has

not been cancelled (Section 3.5.4.2).

01011ÐRead Read-Modify-Write Operand.

01100ÐRead for Effective Address.

01101ÐAccess PTE1 by MMU.

01110ÐAccess PTE2 by MMU.

01111

BMT

ADS

*Begin Memory Transaction.

When Stable Low indicates that the current

bus cycle is valid; that is, the bus cycle has

not been cancelled (Section 3.5.4.2).

*Address Strobe.

#

When active, indicates that a bus cycle has

begun and a valid address is on the address

bus.

Reserved.

#

11100

BE0–3

ST0–4

*Byte Enables.

*

Used to selectively enable data transfers on

bytes 0–3 of the data bus.

11101ÐTransfer Slave Operand.

11110ÐRead Slave Status Word.

11111ÐBroadcast Slave ID.

*Address Bus.

Status.

Bus cycle status code; ST0 is the least signif-

icant. Encodings are:

A0–31

Used by the CPU to output a 32-bit address

at the beginning of a bus cycle. A0 is the

least significant.

00000ÐIdle: CPU Inactive on Bus.

00001ÐIdle: WAIT Instruction.

00010ÐIdle: Halted.

4.1.4 Input/Output Signals

00011ÐIdle: The bus is idle while the slave

processor is executing an instruction.

D0–31

*Data Bus.

Used by the CPU to input or output data dur-

ing a read or write cycle respectively.

00100ÐInterrupt Acknowledge, Master.

4.2 ABSOLUTE MAXIMUM RATINGS

All Input or Output Voltages with

Respect to GND

b

a

0.5V to 7V

If Military/Aerospace specified devices are required,

please contact the National Semiconductor Sales

Office/Distributors for availability and specifications.

Power Dissipation

4 W

Note: Absolute maximum ratings indicate limits beyond

which permanent damage may occur. Continuous operation

at these limits is not intended; operation should be limited to

those conditions specified under Electrical Characteristics.

a

0 C to 95 C

Case Temperature Under Bias

Storage Temperature

§

65 C to 150 C

§

b

a

§

e

a

0 to 95 C, V

e

g

5V 5%, GND

4.3 ELECTRICAL CHARACTERISTICS T

0V

Typ

§

CASE

CC

Symbol

Parameter

High Level Input Voltage

Low Level Input Voltage

High Level Output Voltage

Conditions

Min

Max

Units

a

V

2.0

V

CC

0.5

V

IH

b

0.5

0.8

IL

e b

I

400 mA

2.4

OH

OL

OH

Low Level Output Voltage

A0–11, D0–31, DDIN

CONF, BMT

e

I

4 mA

0.4

V

OL

s

6 mA

16 mA

2 mA

BCLK, BCLK

All Other Outputs

s

b

I

Input Load Current

0

V

IN

V

20

mA

L

CC

s

Leakage Current (Output and

0.4

V

IN

V

CC

L

I/O pins in TRI-STATE/Input Mode)

C

I

CLK Input Capacitance

15

pF

IN

@

650 30 MHz

@

800 30 MHz

e

A

Active Supply Current

I

0, T

25 C,

§

CC

OUT

@

550 25 MHz

@

675 25 MHz

V

5V

mA

CC

@

450 20 MHz

@

575 20 MHz

68

4.0 Device Specifications (Continued)

Connection Diagram

TL/EE/9354–40

Bottom View

Order Number NS32532-20, NS32532-25 or NS32532-30

FIGURE 4-2. 175-Pin PGA Package

NS32532 Pinout Descriptions

Desc

Reserved

BP

A1 D26

B16 GNDB13 D14

GNDL6

VCCL5

D13

J14

J15

J16

K1

GNDL5

CONF

RDY

N9

A0

R6

R7

A2 Reserved C1 VCCB14

A3 Reserved C2 D23

D15

D16

E1

N10 VCCB9

N11 CIOUT

N12 SPC

R8

A4 VCCL2

C3 IOINH

VCCB6

A23

HOLD

VCCB11

GNDB10

D4

R9

ISF

A5 Reserved C4 ILO

E2

K2

N13 BE3

R10

R11

R12

R13

R14

R15

R16

S1

RST

A6 PFS

A7 SDN

C5 GNDB3

C6 D24

E3

GNDL4

GNDB11

D11

K3

N14 VCCB10

N15 ADS

NMI

E14

E15

E16

F1

K14

K15

K16

L1

GNDB1

Reserved

VCCB2

INVIC

A8 Reserved C7 D22

D6

N16 BW1

A9 BCLK

A10 VCCCLK

A11 SYNC

C8 D20

C9 A30

D12

A16

P1

P2

BER

CIIN

D2

A22

VCCB7

GNDB6

A10

C10 CASEC

C11 Reserved

C12 D21

C13 D19

C14 D18

C15 A29

F2

A21

L2

P3

Reserved (1) A12 CIA0

F3

VCCL3

D8

L3

P4

A13

A8

CIA1

A13 CIA6

A14 VCCL6

A15 D29

B1 D27

B2 D25

B3 U/S

F14

F15

F16

G1

L14

L15

L16

M1

M2

M3

A6

P5

S2

CIA4

D9

A2

P6

A5

S3

VCCB1

Reserved

VCCB4

Reserved

VCCB3

FSSR

D10

ST3

P7

A3

S4

A20

GNDB8

VCCL4

BE1

P8

A1

S5

C16 A31

G2

GNDB5

A17

P9

ST2

ST1

ST0

BOUT

DDIN

BE2

BE0

BMT

BRT

IODEC

D1

S6

D1 VCCB5

G3

P10

P11

P12

P13

P14

P15

P16

R1

S7

B4 Reserved D2 GNDB12 G14

D5

M14 GNDB9

M15 BW0

M16 BIN

S8

B5 Reserved D3 D17

G15

G16

H1

D7

S9

B6 GNDL3

B7 GNDB2

B8 DBG

D4 D16

D5 A27

D6 A28

VCCB12

A19

S10

S11

S12

S13

S14

S15

S16

INT

N1

N2

N3

N4

N5

N6

N7

N8

Reserved

D0

VCCL1

GNDL2

INVSET

INVDC

CIA3

H2

A18

B9 Reserved D7 GNDB4

H3

A14

D3

B10 BCLK

B11 GNDCLK

B12 CLK

D8 VCCB13

D9 D15

H14

H15

H16

J1

A11

A15

A12

A9

VCCB8

GNDB7

ST4

R2

D10 D14

D11 A26

D12 A25

D13 A24

R3

CIA5

B13 CIA2

B14 D31

A7

R4

D30

J2

HLDA

A4

R5

D28

B15 GNDL1

J3

Note 1: This pin should be grounded.

All other reserved pins should be left open.

69

4.0 Device Specifications (Continued)

4.4 SWITCHING CHARACTERISTICS

4.4.1 Definitions

ABBREVIATIONS:

L.E.Ðleading edge R.E.Ðrising edge

T.E.Ðtraining edge F.E.Ðfalling edge

All the timing specifications given in this section refer to

0.8V or 2.0V on all the signals as illustrated in Figures 4-3

and 4-4, unless specifically stated otherwise.

TL/EE/9354–42

FIGURE 4-4. Input Signals Specification Standard

TL/EE/9354–41

FIGURE 4-3. Output Signals Specification Standard

70

4.0 Device Specifications (Continued)

4.4.2 Timing Tables

4.4.2.1 Output Signals: Internal Propagation Delays, NS32532-20, NS32532-25, NS32532-30

Maximum times assume capacitive loading of 100 pF on the clock signals and 50 pF on all the other signals. A minimum

capacitance load of 50 pF on BCLK and BCLK is also assumed.

#

NS32532-20

NS32532-25

NS32532-30

Name

Figure

4–24

Description

Bus Clock Period

BCLK High Time

BCLK Low Time

BCLK Rise Time

BCLK Fall Time

BCLK High Time

BCLK Low Time

BCLK Rise Time

BCLK Fall Time

Reference/Conditions

Units

Min

Max

Min

Max

Min

Max

t

R.E., BCLK to Next

R.E., BCLK

BC

p

50

100

40

100

33.3

100

ns

At 2.0V on BCLK

(Both Edges)

0.5 t

BCp

0.5 t

BCp

0.5 t

BCp

BC

h

b

5

4

3.65

At 0.8V on BCLK

(Both Edges)

0.5 t

b

0.5 t

b

0.5 t

BCp

BC

l

BCp

5

BCp

4

b

3.65

(1)

0.8V to 2.0V on

R.E., BCLK

BC

r

5

4

3

(1)

2.0V to 0.8V on

F.E., BCLK

BC

f

At 2.0V on BCLK

(Both Edges)

0.5 t

b

0.5 t

b

0.5 t

BCp

NBC

h

BCp

5

BCp

4

b

3.65

At 0.8V on BCLK

(Both Edges)

0.5 t

b

0.5 t

b

0.5 t

BCp

NBC

l

BCp

5

BCp

4

b

3.65

(1)

0.8V to 2.0V on

R.E., BCLK

NBC

r

5

4

3

(1)

2.0V to 0.8V on

F.E., BCLK

NBC

f

5

4

3

CLK to BCLK

R.E. Delay

2.0V on R.E., CLK to

2.0V on R.E., BCLK

CBC

dr

20

17

15

CLK to BCLK

F.E. Delay

2.0V on R.E., CLK to

0.8V on F.E., BCLK

CBC

df

CLK to BCLK

R.E. Delay

2.0V on R.E., CLK to

0.8V on R.E., BCLK

CNBC

dr

CLK to BCLK

F.E. Delay

2.0V on R.E., CLK to

0.8V on F.E., BCLK

CNBC

df

Bus Clocks Skew 2.0V on R.E., BCLK to

0.8V on F.E., BCLK

BCNBC

rf

b

a

b

a

b

a

2

1

Bus Clocks Skew 0.8V on F.E., BCLK to

2.0V on R.E., BCLK

BCNBC

fr

b

4–5, 4–6 Address Bits 0–31 After R.E., BCLK T1

Valid

A

v

11

9

8

4–5, 4–6 Address Bits 0–31 After R.E., BCLK T1 or Ti

Hold

A

h

0

4–11, 4–12 Address Bits 0–31 After F.E., BCLK Ti

Floating

A

f

21

17

13

4–11, 4–12 Address Bits 0–31 After F.E., BCLK Ti

Not Floating

An

f

Note 1: Guaranteed by characterization. Due to tester conditions this parameter is not 100% tested.

71

4.0 Device Specifications (Continued)

4.4.2.1 Output Signals: Internal Propagation Delays, NS32532-20, NS32532-25, NS32532-30 (Continued)

NS32532-20

NS32532-25

NS32532-30

Name

Figure

4–8

Description

Reference/Conditions

Units

Min

Max

Min

Max

Min

Max

t

Address Bits A2, A3 After R.E., BCLK T2B

Valid (Burst Cycle)

AB

v

11

9

8

ns

4–8

Address Bits A2, A3 After R.E., BCLK T2B

Hold (Burst Cycle)

AB

h

0

ns

4–6, 4–15 Data Out Valid

After R.E., BCLK T1

0.5 t

BCp

a

13 ns

0.5 t

BCp

a

12 ns

0.5 t

BCp

a

11 ns

DO

v

0.5 t

BCp

0.5 t

BCp

0.5 t

BCp

t

4–6, 4–15 Data Out Hold

After R.E., BCLK T1 or Ti

Before SPC T.E.

0

8

ns

DO

h

4–15

4–7

Data Out Setup

(Slave Write)

DO

spc

12

10

t

Data Bus Floating

After R.E., BCLK

T1 or Ti

DO

f

21

17

13

ns

t

4–7

Data Bus

After F.E., BCLK T1

DO

nf

0

Not Floating

t

4–5, 4–7 BMT Signal Valid

4–5, 4–7 BMT Signal Hold

After R.E., BCLK T1

After R.E., BCLK T2

30

21

25

17

21

13

ns

BMT

v

BMT

h

4–11, 4–12 BMT Signal Floating After F.E., BCLK Ti

BMT

f

4–11, 4–12 BMT Signal

Not Floating

After F.E., BCLK Ti

BMT

hf

ns

t

4–5, 4–8 CONF Signal Active After R.E., BCLK T1

0.5 t

a

0.5 t

a

0.5 t

a

CONF

a

BC

11

BC

9

BC

8

p

0.5 t

BC

0.5 t

BC

0.5 t

BC

p

t

4–5, 4–8 CONF Signal Inactive After R.E., BCLK T1 or Ti

4–11, 4–12 CONF Signal Floating After F.E., BCLK Ti

11

9

8

ns

CONF

ia

21

17

13

CONF

f

4–11, 4–12 CONF Signal

Not Floating

After F.E., BCLK Ti

CONF

nf

0

ns

t

4–5, 4–8 ADS Signal Active

After R.E., BCLK T1

11

9

8

ns

ADS

a

4–5, 4–8 ADS Signal Inactive After F.E., BCLK T1

4–6 ADS Pulse Width At 0.8V (Both Edges)

4–11, 4–12 ADS Signal Floating After F.E., BCLK Ti

ADS

ia

15

0

12

0

10

0

ADS

w

21

11

21

17

9

13

8

ADS

f

4–11, 4–12 ADS Signal

Not Floating

After F.E., BCLK Ti

ADS

nf

ns

t

4–6, 4–8 BE Signals Valid

n

After R.E., BCLK T1

BE

v

4–6, 4–8 BE Signals Hold

n

After R.E., BCLK T1,

Ti or T2B

BE

h

0

t

4–11, 4-12 BE Signals Floating After F.E., BCLK Ti

n

17

13

BE

f

4–11, 4–12 BE Signals

n

Not Floating

After F.E., BCLK Ti

BE

nf

0

t

4–5, 4–6 DDIN Signal Valid

4–5, 4–6 DDIN Signal Hold

After R.E., BCLK T1

11

21

9

8

ns

DDIN

v

After R.E., BCLK T1 or Ti

h

f

4–11, 4–12 DDIN Signal Floating After F.E., BCLK Ti

17

13

4–11, 4–12 DDIN Signal

Not Floating

After F.E., BCLK Ti

nf

ns

t

4–14, 4–15 SPC Signal Active

After R.E., BCLK T1

19

15

12

SPC

a

72

4.0 Device Specifications (Continued)

4.4.2.1 Output Signals: Internal Propagation Delays, NS32532-20, NS32532-25, NS32532-30 (Continued)

NS32532-20 NS32532-25 NS32532-30

Name

Figure

Description

Reference/Conditions

Units

Min

Max

Min

Max

Min

Max

t

4–14, 4–15 SPC Signal Inactive

After R.E., BCLK Ti, T1 or T2

Before SPC L.E.

19

15

12

ns

SPC

ia

(1)

4–14

DDIN Valid to

SPC Active

DDSPC

0

t

4–12, 4–13 HLDA Signal Active

4–12 HLDA Signal Inactive

After F.E., BCLK Ti

After R.E., BCLK T1

After R.E., BCLK T1 or Ti

After R.E., BCLK T2

15

11

9

10

8

ns

HLDA

a

HLD

Aia

4–5, 4–14 Status (ST0–4) Valid

4–5, 4–14 Status (ST0–4) Hold

4–8, 4–9 BOUT Signal Active

4–8, 4–9 BOUT Signal Inactive

ST

v

0

ST

h

15

21

12

17

11

13

BOUT

a

After R.E., BCLK

Last T2B, T1 or Ti

BOUT

ia

ns

t

4–11, 4–12 BOUT Signal Floating

After F.E., BCLK Ti

BOUT

f

4–11, 4–12 BOUT Signal

Not Floating

BOUT

nf

0

t

4–7

Interlock Signal Active After F.E., BCLK Ti

Interlock Signal Inactive After F.E., BCLK Ti

11

15

11

9

8

ns

ILO

a

9

8

ILO

ia

4–21

4–22

4–23

4–5

PFS Signal Active

PFS Signal Inactive

ISF Signal Active

ISF Signal Inactive

BP Signal Active

BP Signal Inactive

U/S Signal Valid

U/S Signal Hold

After F.E., BCLK

11

9

10

8

PFS

a

After F.E., Next BCLK

After F.E., BCLK

PFS

ia

ISF

a

After F.E., Next BCLK

After F.E., BCLK

ISF

ia

BP

a

After F.E., Next BCLK

After R.E., BCLK T1

After R.E., BCLK T1 or Ti

After F.E., BCLK T1

After R.E., BCLK T1 or Ti

BP

ia

US

v

4–5

0

US

h

4–5

CASEC Signal Valid

CASEC Signal Hold

15

21

11

17

10

13

CAS

v

4–5

CAS

h

4–11, 4–12 CASEC Signal Floating After F.E., BCLK Ti

CAS

f

4–11, 4–12 CASEC Signal

Not Floating

After F.E., BCLK Ti

CAS

nf

0

ns

t

4–5

CIOUT Signal Valid

After R.E., BCLK T1

15

11

10

ns

CIO

IOI

v

CIOUT Signal Hold

IOINH Signal Valid

IOINH Signal Hold

After R.E., BCLK T1 or Ti

After R.E., BCLK T1

0

h

v

After R.E., BCLK T1 or Ti

IOI

h

73

4.0 Device Specifications (Continued)

4.4.2.2 Input Signal Requirements: NS32532-20, NS32532-25, NS32532-30

NS32532-20 NS32532-25 NS32532-30

Name

Figure

4–24

Description

Reference/Conditions

Units

Min

Max

Min

Max

Min

Max

t

Input Clock Period R.E., CLK to Next

R.E., CLK

C

p

h

l

25

50

20

50

16.6

50

ns

CLK High Time

At 2.0V on CLK

(Both Edges)

0.5 t

b

0.5 t

b

0.5 t

b

C

p

5

4

CLK Low Time

At 0.8V on CLK

(Both Edges)

0.5 t

b

0.5 t

b

0.5 t

C

b

4

5

t

(1)

4–24

CLK Rise Time

CLK Fall Time

0.8V to 2.0V on R.E., CLK

2.0V to 0.8V on F.E., CLK

Before R.E., BCLK T1 or Ti

After R.E., BCLK T1 or Ti

5

4

3

ns

C

r

f

4–5, 4–14 Data In Setup

4–5, 4–14 Data In Hold

12

1

10

1

8

1

DI

s

h

4–5

RDY Setup Time

Before R.E., BCLK T2(W),

T1 or Ti

RDY

s

19

1

15

1

12

1

ns

t

4–5

RDY Hold Time

Ater R.E., BCLK T2(W),

T1 or Ti

RDY

h

t

4–5

BW0–1 Setup Time Before F.E., BCLK T2 or T2(W)

BW0–1 Hold Time After F.E., BCLK T2 or T2(W)

19

1

15

1

12

1

ns

BW

s

BW

h

4–12, 4–13 HOLD Setup Time Before F.E., BCLK

19

1

15

1

12

1

HOLD

s

4–12

4–8

HOLD Hold Time

BIN Setup Time

BIN Hold Time

After F.E., BCLK

HOLD

h

Before F.E., BCLK T2 or T2(W)

After F.E., BCLK T2 or T2(W)

Before R.E., BCLK T1 or Ti

After R.E., BCLK T1 or Ti

Before R.E., BCLK T1 or Ti

After R.E., BCLK T1 or Ti

18

1

14

1

11

1

BIN

s

4–8

BIN

h

4–6, 4–8 BER Setup Time

4–6, 4–8 BER Hold Time

4–6, 4–8 BRT Setup Time

4–6, 4–8 BRT Hold Time

19

1

15

1

12

1

BER

s

BER

h

19

1

15

1

12

1

BRT

s

BRT

h

4–5

IODEC Setup Time Before F.E., BCLK T2 or T2(W)

18

1

14

1

11

1

IOD

s

IODEC Hold Time

After F.E., BCLK T2 or T2(W)

After VCC Reaches 4.5V

IOD

h

(1)

4–26

Power Stable to

R.E. of RST

PWR

50

40

30

ms

t

4–27

RST Setup Time

RST Pulse Width

Before R.E., BCLK

14

64

12

64

11

64

ns

RST

s

At 0.8V (Both Edges)

t

BC

RST

w

p

Note 1: Guaranteed by characterization. Due to tester conditions this parameter is not 100% tested.

74

4.0 Device Specifications (Continued)

4.4.2.2 Input Signal Requirements: NS32532-20, NS32532-25, NS32532-30 (Continued)

NS32532-20

NS32532-25

NS32532-30

Name

Figure

Description

Reference/Conditions

Units

Min

19

1

Max

Min

15

1

Max

Min

12

1

Max

t

4–5

CIIN Setup Time

CIIN Hold Time

INT Setup Time

INT Hold Time

Before F.E., BCLK T2

After F.E., BCLK T2

Before R.E., BCLK

After R.E., BCLK

Before R.E., BCLK

After R.E., BCLK

Before R.E., BCLK

After R.E., BCLK

Before R.E., BCLK

After R.E., BCLK

Before R.E., CLK

After R.E., CLK

ns

CII

s

t

4–5

CII

h

4–19

4–16

4–17

4–25

4–18

4–20

12

1

10

1

9

INT

s

1

h

NMI Setup Time

NMI Hold Time

18

1

15

1

14

1

NMI

s

h

SDN Setup Time

SDN Hold Time

FSSR Setup Time

FSSR Hold Time

SYNC Setup Time

SYNC Hold Time

CIA0–6 Setup Time

CIA0–6 Hold Time

12

1

10

1

9

SD

s

1

SD

h

12

1

10

1

9

FSSR

s

1

FSSR

h

10

1

8

7

SYNC

s

1

h

Before R.E., BCLK

After R.E., BCLK

12

1

10

1

9

CIA

s

1

CIA

h

INVSET Setup Time Before R.E., BCLK

12

1

11

1

9

INVS

s

INVSET Hold Time

INVIC Setup Time

INVIC Hold Time

INVDC Setup Time

INVDC Hold Time

DBG Setup Time

DBG Hold Time

After R.E., BCLK

Before R.E., BCLK

After R.E., BCLK

Before R.E., BCLK

After R.E., BCLK

Before R.E., BCLK

After R.E., BCLK

1

INVS

INVI

h

12

1

10

1

9

s

1

INVI

h

12

1

10

1

9

INVD

DBG

s

1

h

12

1

10

1

9

s

1

DBG

h

75

4.0 Device Specifications (Continued)

4.4.3 Timing Diagrams

TL/EE/9354–43

FIGURE 4-5. Basic Read Cycle Timing

76

4.0 Device Specifications (Continued)

TL/EE/9354–44

Note: An Idle State is always inserted before a Write Cycle when the Write immediately follows a confirmed Read Cycle.

FIGURE 4-6. Write Cycle Timing

77

4.0 Device Specifications (Continued)

TL/EE/9354–45

FIGURE 4-7. Interlocked Read and Write Cycles

78

4.0 Device Specifications (Continued)

TL/EE/9354–46

FIGURE 4-8. Burst Read Cycles

79

4.0 Device Specifications (Continued)

TL/EE/9354–47

FIGURE 4-9. External Termination of Burst Cycles

TL/EE/9354–48

FIGURE 4-10. Bus Error or Retry During Burst Cycles

Note: Two idle state are always inserted by the CPU following the assertion of BRT.

80

4.0 Device Specifications (Continued)

TL/EE/9354–49

FIGURE 4-11. Extended Retry Timing

81

4.0 Device Specifications (Continued)

TL/EE/9354–50

FIGURE 4-12. Hold Timing (Bus Initially Idle)

82

4.0 Device Specifications (Continued)

TL/EE/9354–52

FIGURE 4-14. Slave Processor Read Timing

TL/EE/9354–51

FIGURE 4-13. HOLD Acknowledge Timing

(Bus Initially Not Idle)

TL/EE/9354–54

FIGURE 4-16. Slave Processor Done

TL/EE/9354–53

FIGURE 4-15. Slave Processor Write Timing

TL/EE/9354–55

FIGURE 4-17. FSSR Signal Timing

83

4.0 Device Specifications (Continued)

TL/EE/9354–56

FIGURE 4-18. Cache Invalidation Request

Note 1: CIA0–6 and INVSET are only relevant when INVIC and/or INVDC are asserted.

TL/EE/9354–57

FIGURE 4-19. INT and NMI Signals Sampling

Note 1: INT and NMI are sampled on every other rising edge of BCLK, starting with the second rising edge of BCLK after RST goes high.

Note 2: INT is level sensitive, and once asserted, it should not be deasserted until it is acknowledged.

TL/EE/9354–58

TL/EE/9354–59

FIGURE 4-20. Debug Trap Request

FIGURE 4-21. PFS Signal Timing

TL/EE/9354–60

TL/EE/9354–61

FIGURE 4-22. ISF Signal Timing

FIGURE 4-23. Break Point Signal Timing

84

4.0 Device Specifications (Continued)

TL/EE/9354–62

FIGURE 4-24. Clock Waveforms

TL/EE/9354–63

FIGURE 4-25. Bus Clock Synchronization

TL/EE/9354–64

FIGURE 4-26. Power-On Reset

TL/EE/9354–65

FIGURE 4-27. Non-Power-On Reset

85

Appendix A: Instruction Formats

NOTATIONS:

Options: in String Instructions

e

i

Integer Type Field

e

U/W

B

T

B

00 (Byte)

01 (Word)

11 (Double Word)

e

T

B

Translated

Backward

00: None

01: While Match

11: Until Match

W

D

e

U/W

e

f

c

Floating Point Type Field

e

F

L

1 (Std. Floating: 32 bits)

0 (Long Floating: 64 bits)

Configuration bits, in SETCFG Instruction:

Custom Type Field

e

D

Q

1 (Double Word)

0 (Quad Word)

C

M

F

I

mreg: MMU Register number, in LMR, SMR.

e

op

Operation Code

Valid encodings shown with each format.

0000

#

e

gen, gen 1, gen 2 General Addressing Mode Field

See Section 2.2 for encodings.

Trap (UND)

#

e

reg

General Purpose Register Number

Condition Code Field

e

0111

1000

1001

1010

1011

1100

1101

1110

1111

*

cond

Reserved

MCR

e

1

0000

0001

0010

0011

0100

0101

0110

0111

1000

1001

1010

1011

1100

1101

1110

1111

EQual: Z

e

Not Equal: Z

Carry Set: C

Carry Clear: C

0

MSR

1

TEAR

PTB0

e

0

e

Lower or Same: L

HIgher: L

1

PTB1

e

0

IVAR0

IVAR1

e

Greater Than: N

Less or Equal: N

1

0

e

Flag Set: F

1

7

0

e

0 and Z

Flag Clear: F

0

cond

1

0

1

0

e

Higher or Same: L

e

0

LOwer: L

e

0 and Z

e

1 or Z

1

Format 0

e

Greater or Equal: N

e

1 or Z

Less Than: N

0

Bcond

(BR)

e

1

(Unconditionally True)

(Unconditionally False)

0

op

0

e

short

Short Immediate value. May contain:

quick: Signed 4-bit value, in MOVQ, ADDQ,

CMPQ, ACB.

Format 1

BSR

RET

-0000

-0001

-0010

-0011

-0100

-0101

-0110

-0111

ENTER

EXIT

NOP

WAIT

DIA

-1000

-1001

-1010

-1011

-1100

-1101

-1110

-1111

cond: Condition Code (above), in Scond.

areg: CPU Dedicated Register, in LPR, SPR.

CXP

e

0000

0001

0010

0011

0100

UPSR

DCR

BPC

RXP

RETT

RETI

SAVE

FLAG

SVC

DSR

CAR

RESTORE

BPT

e

0101–0111

(Reserved)

e

1000

1001

1010

1011

1100

1101

1110

1111

FP

SP

SB

15

8

7

0

gen

short

op

1

i

USP

CFG

PSR

Format 2

ADDQ

CMPQ

SPR

-000

ACB

-100

-101

-110

INTBASE

MOD

-001

-010

-011

MOVQ

LPR

Scond

86

Appendix A: Instruction Formats (Continued)

15

8 7

0

gen

op

1 1 1 1 1

i

Format 3

TL/EE/9354–66

CXPD

-0000

ADJSP

JSR

-1010

-1100

-1110

Format 8

BICPSR

JUMP

-0010

-0100

-0110

EXT

-0 00

INDEX

FFS

-1 00

-1 01

CASE

CVTP

INS

-0 01

BISPSR

-0 10

Trap (UND) on XXX1, 1000

CHECK

MOVSU

MOVUS

-0 11

e

-110, reg

001

011

15

8 7

0

gen 1

gen 2

op

i

23

16 15

gen 2

8 7

0

Format 4

gen 1

op

Format 9

-000

f

i

0 0 1 1 1 1 1 0

ADD

CMP

BIC

-0000

-0001

-0010

-0100

-0101

-0110

SUB

-1000

-1001

-1010

-1100

-1101

-1110

ADDR

AND

MOVif

LFSR

ROUND

TRUNC

SFSR

-100

-101

-110

-111

ADDC

MOV

OR

SUBC

TBIT

XOR

-001

-010

-011

MOVLF

MOVFL

FLOOR

23

16 15

8 7

0

0 0 0 0 0 short

0

op

i

0 0 0 0 1 1 1 0

TL/EE/9354–67

Format 5

Format 10

MOVS

CMPS

-0000

-0001

SETCFG

SKPS

-0010

-0011

Trap (UND) Always

23

16 15

gen 2

8 7

0

Trap (UND) on 1XXX, 01XX

gen 1

op

0

f

1 0 1 1 1 1 1 0

23

16 15

8 7

0

gen 1

gen 2

op

i

0 1 0 0 1 1 1 0

Format 11

-0000 DIVf

ADDf

-1000

-1001

-1010

-1011

-1100

-1101

-1110

-1111

Format 6

-0000

MOVf

CMPf

Note 3

SUBf

-0001

-0010

-0011

-0100

-0101

-0110

-0111

Note 1

Note 3

Note 1

MULf

ROT

NEG

NOT

-1000

-1001

ASH

-0001

-0010

-0011

-0100

-0101

-0110

-0111

CBIT

Trap (UND)

SUBP

ABS

-1010

-1011

-1100

-1101

-1110

-1111

CBITI

Trap (UND)

LSH

NEGf

Note 2

Note 1

ABSf

Note 2

Note 1

COM

SBIT

IBIT

SBITI

ADDP

23

16 15

gen 2

8 7

0

gen 1

op

i

1 1 0 0 1 1 1 0

Format 7

MOVM

-0000

MUL

MEI

-1000

-1001

CMPM

-0001

-0010

-0011

-0100

-0101

-0110

-0111

INSS

Trap (UND)

DEI

-1010

-1011

-1100

-1101

-1110

-1111

EXTS

MOVXBW

MOVZBW

MOVZiD

MOVXiD

QUO

REM

MOD

DIV

87

Appendix A: Instruction Formats (Continued)

Format 15.1

23

16 15

8

7

0

CCV3

LCSR

CCV5

CCV4

-000

CCV2

CCV1

SCSR

CCV0

-100

-101

-110

-111

gen 1

gen 2

op

0

f

1

1 1 1 1 1 1 0

-001

-010

-011

Format 12

Note 2

SQRTf

POLYf

DOTf

-0000

Note 2

Note 1

MACf

-1000

-1001

-1010

-1011

-1100

-1101

-1110

-1111

23

16 15

gen 2

8

-0001

-0010

-0011

-0100

-0101

-0110

-0111

101

gen 1

op

x

c

Note 1

Note 2

Note 1

Note 2

Note 1

SCALBf

LOGBf

Note 2

Note 1

Format 15.5

CCAL3

CCAL0

CMOV0

CCMP0

CCMP1

CCAL1

CMOV2

Note 2

Note 1

-0000

-0001

-0010

-0011

-0100

-0101

-0110

-0111

-1000

-1001

-1010

-1011

-1100

-1101

-1110

-1111

CMOV3

Note 3

Note 1

CCAL2

CMOV1

Note 2

Note 1

TL/EE/9354–68

Format 13

Trap (UND) Always

23

16 15

short

8

7

0

23

16 15

gen 2

8

gen 1

0

op

i

0

0 0 1 1 1 1 0

111

gen 1

op

x

c

Format 14

Format 15.7

RDVAL

WRVAL

-0000

-0001

LMR

SMR

-0010

-0011

-1001

Note 2

Note 1

Note 3

Note 2

Note 1

Note 2

Note 1

-0000

Note 2

-1000

-1001

-1010

-1011

-1100

-1101

-1110

-1111

CINV

Trap (UND) on 01XX, 1000, 101X, 11XX

-0001

-0010

-0011

-0100

-0101

-0110

-0111

Note 1

Note 3

Note 1

Note 2

Note 1

Note 2

Note 1

23

16 15

8 7

0

n n n 1 0 1 1 0

ID Byte

Operation Word

e

If nnn

010, 011, 100, 110 then Trap (UND) Always.

Format 15

(Custom Slave)

nnn

Operation Word Format

TL/EE/9354–69

23

16 15

8

Format 16

000

gen 1

short

x

op

i

Trap (UND) Always

Format 15.0

LCR

SCR

-0010

-0011

TL/EE/9354–70

Format 17

Trap (UND) on all others

Trap (UND) Always

23

16 15

gen 2

8

001

gen 1

op

c

i

TL/EE/9354–71

88

erence Manual, then the results produced by the

NS32532 may differ from those of the NS32032.

Appendix A: Instruction

Formats ^(Continued)

3. Either the program does not depend on the use of a

Memory Management Unit (MMU), or it is written for op-

eration with the NS32382 MMU and does not use the

bus-error or debugging features of the NS32382.

Format 18

Trap (UND) Always

4. The program does not depend on the detection of bus

errors according to the implementation of the NS32332.

For example, the NS32532 distinguishes between re-

startable and nonrestartable bus errors by transferring

control to the appropriate bus-error exception service

procedure through one of two distinct entries in the In-

terrupt Dispatch Table. In contrast, the NS32332 uses a

single entry in the Interrupt Dispatch Table for all bus

errors.

TL/EE/9354–72

Format 19

Trap (UND) Always

Implied Immediate Encodings:

7

0

5. The program does not modify itself. Refer to Section B.4

for more information.

r7

r6

r5

r4

r3

r2

r1

r0

6. The program does not depend on the execution of cer-

tain complex instructions to be non-interruptible. Refer

to Section B.5 on. ‘‘Memory-Mapped I/O’’ for more in-

formation.

Register Mark, Appended to SAVE, ENTER

7

0

r0

r1

r2

r3

r4

r5

r6

r7

7. The program does not use the custom slave instructions

CATSTO and CATST1, as they are not supported by the

NS32532 and will result in a Trap (UND) when their exe-

cution is attempted.

Register Mark, Appended to RESTORE, EXIT

7

0

B.2 ARCHITECTURE EXTENSIONS

offset

length - 1

The NS32532 implements the following extensions of the

Series 32000 architecture using previously reserved control

bits, instruction encodings, and memory locations. Exten-

sions implemented earlier in the NS32332, such as 32-bit

addressing, are not listed.

Offset/Length Modifier Appended to INSS, EXTS

Note 1: Opcode not defined; CPU treats like MOV or CMOV . First operand

c

f

has access class of read; second operand has access class of write; f or c

field selects 32- or 64-bit data.

Note 2: Opcode not defined; CPU treats like ADD or CCAL . First operand

c

f

1. The DC, LDC, IC, and LIC bits in the CFG register have

been defined to control the on-chip Instruction and Data

Caches. The DE-bit in the CFG register has been de-

fined to enable Direct-Exception Mode.

has access class of read;, second operand has access class of read-modify-

write; f or c field selects 32- or 64-bit data.

Note 3: Opcode not defined; CPU treats like CMP or CCMP . First operand

c

f

has access class of read;, second operand has access class of read; f or c

field selects 32- or 64-bit data.

2. The V-flag in the PSR register has been defined to en-

able the Integer-Overflow Trap.

Appendix B. Compatibility Issues

3. The DCR, BPC, DSR, and CAR registers have been de-

fined to control debugging features. Access to these

registers has been added to the definition of the LPR

and SPR instructions.

The NS32532 is compatible with the Series 32000 architec-

ture implemented by the NS32032, NS32332, and previous

microprocessors in the family. Compatibility means that

within certain limited constraints, programs that execute on

one of the earlier Series 32000 microprocessors will pro-

duce identical results when executed on the NS32532.

Compatibility applies to privileged operating systems pro-

grams, as well as to non-privileged applications programs.

This appendix explains both the restrictions on compatibility

with previous Series 32000 microprocessors and the exten-

sions to the architecture that are implemented by the

NS32532.

4. Access to the CFG and SP1 registers has been added

to the definition of the LPR and SPR instructions.

5. The CINV instruction has been defined to invalidate

control of the on-chip Instruction and Data Caches.

6. Direct-Exception Mode has been added to support fast-

er interrupt service time and systems without module

tables.

7. A new entry has been added to the Interrupt Dispatch

Table for supporting vectors to distinguish between re-

startable and nonrestartable bus errors. Two additional

entries support Trap (OVF) and Trap (DBG).

B.1 RESTRICTIONS ON COMPATIBILITY

If the following restrictions are observed, then a program

that executes on an earlier Series 32000 microprocessor

will produce identical results when executed on the

NS32532 in an appropriately configured system:

8. Restrictions have been eliminated for recovery from

Trap (ABT) for operands with access class of write that

cross page boundaries. Restrictions still exist however,

for the operands of the MOVMi instruction.

1. The program is not time-dependent. For example, the

program should not use instruction loops to control real-

time delays.

B.3 INTEGER OVERFLOW TRAP

A new trap condition is recognized for integer arithmetic

overflow. Trap (OVF) is enabled by the V-flag in the PSR.

This new trap is important because detection of integer

overflow conditions is required for certain programming lan-

guages, such as ADA, and the PSR flags do not indicate the

occurrence of overflow for ASHi, DIVi and MULi instructions.

2. The program does not use any encodings of instruc-

tions, operands, addresses, or control fields identified to

be reserved or undefined. For example, if the count op-

erand’s value for an LSHi instruction is not within the

range specified by the Series 32000 Instruction Set Ref-

89

Appendix C. Instruction Set Extensions (Continued)

C.1 PROCESSOR SERVICE INSTRUCTIONS

15

8 7

0

The CFG register, User Stack Pointer (SP1), and Debug

Registers can be loaded and stored using privileged forms

of the LPRi and SPRi instructions.

gen

src

short 1 1 0 1 1

i

procreg

8 7

LPRi

When the SETCFG instruction is executed, the CFG register

bits 0 through 3 are loaded from the instruction’s short field,

bits 4 through 7 are forced to 1, and bits 8 through 12 are

forced to 0.

gen

short 0 1 0 1 1

procreg SPRi

FIGURE C-1. LPRi/SPRi Instruction Formats

dest

The contents of the on-chip Instruction Cache and Data

Cache can be invalidated by executing the privileged in-

struction CINV. While executing the CINV instruction, the

CPU generates 2 slave bus cycles on the system interface

to display the first 3 bytes of the instruction and the source

operand. External circuitry can thereby detect the execution

of the CINV instruction for use in monitoring the contents of

the on-chip caches.

TABLE C-1. LPRi/SPRi New ‘Short’ Field Encodings

Register

procreg

DCR

BPC

short field

0001

Debug Condition Register

Breakpoint Program Counter

Debug Status Register

Compare Address Register

User Stack Pointer

0010

DSR

0011

C.2 MEMORY MANAGEMENT INSTRUCTIONS

CAR

0100

The NS32532 on-chip MMU does not implement the BAR,

BDR, BEAR, and BMR registers of the NS32382. These

registers are used in the NS32382 to support bus error and

debugging features. When an attempt is made to access

one of these 4 registers by executing an LMR or SMR in-

struction, a Trap (UND) occurs. More generally, a Trap

(UND) occurs whenever an attempt is made to execute an

LMR or SMR instruction and the most-significant bit of the

short-field is 0.

USP

1011

Configuration Register

CFG

1100

Cache Invalidate

[

]

options , src

Syntax: CINV

gen

read. D

While executing an LMR instruction, the CPU generates 2

slave bus cycles on the system interface to display the first

3 bytes of the instruction and the source operand. External

circuitry can thereby detect the execution of an LMR in-

struction for use in monitoring the contents of the on-chip

Translation Lookaside Buffer.

The CINV instruction invalidates the contents of locations in

the on-chip Instruction Cache and Data Cache. The instruc-

tion can be used to invalidate either the entire contents of

the on-chip caches or only a 16-byte block. In the latter

case, the 28 most-significant bits of the source operand

specify the physical address of the aligned 16-byte block;

the 4 least-significant bits of the source operand are ig-

nored. If the specified block is not located in the on-chip

caches, then the instruction has no effect. If the entire

cache contents is to be invalidated, then the source oper-

and is read, but its value is ignored.

Like the NS32382 MMU, the F-flag in the PSR is set and no

Trap (ABT) occurs when a RDVAL or WRVAL instruction is

executed and the Protection Level in the Level-1 Page Ta-

ble Entry indicates that the access is not allowed. In the

NS32082 MMU, an abort occurs when the Level-1 PTE is

invalid, regardless of the Protection Level.

Options are specified by listing the letters A (invalidate All), I

(Instruction Cache), and D (Data Cache). If neither the I nor

D option is specified, the instruction has no effect.

C.3 INSTRUCTION DEFINITIONS

This section provides a description of the operations and

encodings of the new NS32532 privileged instructions.

In the instruction encoding, the options are represented in

the A, I, and D fields as follows:

Load and Store Processor Registers

A: 0Ðinvalidate only a 16-byte block

1Ðinvalidate the entire cache

Syntax: LPRi

procreg,

short

src

gen

I: 0Ðdo not affect the Instruction Cache

1Ðinvalidate the Instruction Cache

read.i

SPRi

procreg

short

dest

gen

write.i

D: 0Ðdo not affect the Data Cache

1Ðinvalidate the Data Cache

Flags Affected: None

The LPRi and SPRi instructions can be used to load and

store the User Stack Pointer (USP or SP1), the Configura-

tion Register (CFG) and the Debug Registers in addition to

the Processor Registers supported by the previous Series

32000 CPUs. Access to these registers is privileged.

Traps:

Illegal Operation Trap (ILL) occurs if an at-

tempt is made to execute this instruction

while the U-flag is 1.

Examples:

Figure C-1 and Table C-1 show the instruction formats and

the new ‘short’ field encodings for LPRi and SPRi.

[

]

1. CINV A, D, I , R3 1E A7 1B

[ ]

2. CINV I , R3

1E 27 19

Flags Affected: No flags affected by loading or storing the

USP, CFG, or Debug Registers.

Example 1 invalidates the entire Instruction Cache and Data

Cache.

Traps:

Illegal Instruction Trap (ILL) occurs if an

attempt is made to load or store the USP,

CFG or Debug Registers while the U-flag

is 1.

Example 2 invalidates the 16-byte block whose physical ad-

dress in the Instruction Cache is contained in R3.

90

Appendix C. Instruction Set Extensions (Continued)

C.1 PROCESSOR SERVICE INSTRUCTIONS

15

8 7

0

The CFG register, User Stack Pointer (SP1), and Debug

Registers can be loaded and stored using privileged forms

of the LPRi and SPRi instructions.

gen

src

short 1 1 0 1 1

i

procreg

8 7

LPRi

When the SETCFG instruction is executed, the CFG register

bits 0 through 3 are loaded from the instruction’s short field,

bits 4 through 7 are forced to 1, and bits 8 through 12 are

forced to 0.

gen

short 0 1 0 1 1

procreg SPRi

FIGURE C-1. LPRi/SPRi Instruction Formats

dest

The contents of the on-chip Instruction Cache and Data

Cache can be invalidated by executing the privileged in-

struction CINV. While executing the CINV instruction, the

CPU generates 2 slave bus cycles on the system interface

to display the first 3 bytes of the instruction and the source

operand. External circuitry can thereby detect the execution

of the CINV instruction for use in monitoring the contents of

the on-chip caches.

TABLE C-1. LPRi/SPRi New ‘Short’ Field Encodings

Register

procreg

DCR

BPC

short field

0001

Debug Condition Register

Breakpoint Program Counter

Debug Status Register

Compare Address Register

User Stack Pointer

0010

DSR

0011

C.2 MEMORY MANAGEMENT INSTRUCTIONS

CAR

0100

The NS32532 on-chip MMU does not implement the BAR,

BDR, BEAR, and BMR registers of the NS32382. These

registers are used in the NS32382 to support bus error and

debugging features. When an attempt is made to access

one of these 4 registers by executing an LMR or SMR in-

struction, a Trap (UND) occurs. More generally, a Trap

(UND) occurs whenever an attempt is made to execute an

LMR or SMR instruction and the most-significant bit of the

short-field is 0.

USP

1011

Configuration Register

CFG

1100

Cache Invalidate

Syntax: CINV

options, src

gen

read. D

While executing an LMR instruction, the CPU generates 2

slave bus cycles on the system interface to display the first

3 bytes of the instruction and the source operand. External

circuitry can thereby detect the execution of an LMR in-

struction for use in monitoring the contents of the on-chip

Translation Lookaside Buffer.

The CINV instruction invalidates the contents of locations in

the on-chip Instruction Cache and Data Cache. The instruc-

tion can be used to invalidate either the entire contents of

the on-chip caches or only a 16-byte block. In the latter

case, the 28 most-significant bits of the source operand

specify the physical address of the aligned 16-byte block;

the 4 least-significant bits of the source operand are ig-

nored. If the specified block is not located in the on-chip

caches, then the instruction has no effect. If the entire

cache contents is to be invalidated, then the source oper-

and is read, but its value is ignored.

Like the NS32382 MMU, the F-flag in the PSR is set and no

Trap (ABT) occurs when a RDVAL or WRVAL instruction is

executed and the Protection Level in the Level-1 Page Ta-

ble Entry indicates that the access is not allowed. In the

NS32082 MMU, an abort occurs when the Level-1 PTE is

invalid, regardless of the Protection Level.

Options are specified by listing the letters A (invalidate All), I

(Instruction Cache), and D (Data Cache). If neither the I nor

D option is specified, the instruction has no effect.

C.3 INSTRUCTION DEFINITIONS

This section provides a description of the operations and

encodings of the new NS32532 privileged instructions.

In the instruction encoding, the options are represented in

the A, I, and D fields as follows:

Load and Store Processor Registers

A: 0Ðinvalidate only a 16-byte block

1Ðinvalidate the entire cache

Syntax: LPRi

procreg,

short

src

gen

I: 0Ðdo not affect the Instruction Cache

1Ðinvalidate the Instruction Cache

read.i

SPRi

procreg

short

dest

gen

write.i

D: 0Ðdo not affect the Data Cache

1Ðinvalidate the Data Cache

Flags Affected: None

The LPRi and SPRi instructions can be used to load and

store the User Stack Pointer (USP or SP1), the Configura-

tion Register (CFG) and the Debug Registers in addition to

the Processor Registers supported by the previous Series

32000 CPUs. Access to these registers is privileged.

Traps:

Illegal Operation Trap (ILL) occurs if an at-

tempt is made to execute this instruction

while the U-flag is 1.

Examples:

Figure C-1 and Table C-1 show the instruction formats and

the new ‘short’ field encodings for LPRi and SPRi.

1. CINV A, D, I, R3 1E A7 1B

2. CINV I, R3 1E 27 19

Flags Affected: No flags affected by loading or storing the

USP, CFG, or Debug Registers.

Example 1 invalidates the entire Instruction Cache and Data

Cache.

Traps:

Illegal Instruction Trap (ILL) occurs if an

attempt is made to load or store the USP,

CFG or Debug Registers while the U-flag

is 1.

Example 2 invalidates the 16-byte block whose physical ad-

dress in the Instruction Cache is contained in R3.

91

Appendix C. Instruction Set Extensions (Continued)

23

15

8 7

0

23

16 15

short 0 0 0 1 1 1 1 0 0 0 1 1 1 1 0

mmureg SMR

8 7

0

gen

src

0 A

I

D 0 1 0 0 1 1 1 0 0 0 1 1 1 1 0

CINV

FIGURE C-2. CINV Instruction Format

gen

options

dest

FIGURE C-3. LMR/SMR Instruction Formats

16 15 8 7

short 0 0 0 1 0 1 1 0 0 0 1 1 1 1 0

mmureg LMR

Load and Store Memory Management Register

23

0

Syntax: LMR

SMR

mmreg,

short

src

gen

read.D

gen

src

mmureg,

short

dest

gen

write.D

Appendix D. Instruction

Execution Times

The NS32532 achieves its performance by using an ad-

vanced implementation incorporating a 4-stage Instruction

Pipeline, a Memory Management Unit, an Instruction Cache

and a Data Cache into a single integrated circuit.

The LMR and SMR instruction load and store the on-chip

MMU registers as 32-bit quantities to and from any general

operand. For reasons of system security, these instructions

are privileged. In order to be executable, they must also be

enabled by setting the M bit in the CFG register.

As a consequence of this advanced implementation, the

performance evaluation for the NS32532 is more complex

than for the previous microprocessors in the Series 32000

family. In fact, it is no longer possible to determine the exe-

cution time for an instruction using only a set of tables for

operations and addressing modes. Rather, it is necessary to

consider dependencies between the various instructions ex-

ecuting in the pipeline, as well as the occurrence of misses

for the on-chip caches.

The instruction formats as well as the ‘short’ field encodings

are shown in Figure C-3 and Table C-2 respectively.

It is to be noted that the IVAR0 and IVAR1 registers are

write-only, and as such, they can only be loaded by the LMR

instruction.

Flags Affected: none

Traps:

Undefined Instruction Trap (UND) occurs if

an attempt is made to execute this instruc-

tion while either of the following conditions

is true:

The following sections explain the method to evaluate the

performance of the NS32532 by calculating various timing

parameters for an instruction sequence. Due to the high

degree of parallelism in the NS32532, the evaluation tech-

niques presented here include some simplifications and ap-

proximations.

1. The M-bit in the CFG register is 0.

2. The U-Flag in the PSR is 0 and the

most-significant bit of the short field is 0.

Illegal Instruction Trap (ILL) occurs if an at-

tempt is made to execute this instruction

while the M-bit in the CFG register and the

U-flag in the PSR are both 1.

D.1 INTERNAL ORGANIZATION

AND INSTRUCTION EXECUTION

The NS32532 is organized internally as 8 functional units as

shown in Figure 1. The functional units operate in parallel to

execute instructions in the 4-stage pipeline. The structure of

this pipeline is shown in Figure 3-2. The Instruction Fetch

and Instruction Decode pipeline stages are implemented in

the loader along with the 8-byte instruction queue and the

buffer for a decoded instruction. The Address Calculation

pipeline stage is implemented in the address unit. The Exe-

cute pipeline stage is implemented in the Execution Unit

along with the write data buffer that holds up to two results

directed to memory.

TABLE C-2. LMR/SMR ‘Short’ Field Encodings

Register

mmureg

short field

Memory Management

Control Reg

MCR

1001

Memory Management

Status Reg

MSR

1010

1011

1100

1101

1110

1111

Translation Exception

Address Reg

TEAR

PTB0

PTB1

IVAR0

IVAR1

The Address Unit and Execution Unit can process instruc-

tions at a peak rate of 2 clock cycles per instruction, en-

abling a sustained pipeline throughput at 30 MHz of

15 MIPS (million instructions per second) for sequences of

register-to-register, immediate-to-register, register-to-mem-

ory and memory-to-register instructions. Nevertheless, the

execution of instructions in the pipeline is reduced from the

peak throughput of 2 cycles by the following causes of de-

lay:

Page Table Base

Register 0

Page Table Base

Register 1

Invalidate Virtual

Address 0

Invalidate Virtual

Address 1

1. Complex operations, like division, require more than 2 cy-

cles in the Execution Unit, and complex addressing

modes, like memory relative, require more than 2 cycles

in the Address Unit.

92

Appendix D. Instruction Execution Times (Continued)

2. Dependencies between instructions can limit the flow

through the pipeline. A data dependency can arise when

the result of one instruction is the source of a following

instruction. Control dependencies arise when branching

instructions are executed. Section D.3 describes the

types of instruction dependencies that impact perform-

ance and explains how to calculate the pipeline delays.

2 cycles in total to process a double-precision floating-

point immediate value.

#

D.2.2 Address Unit Timing

The processing time of the Address Unit depends on the

instruction’s operation and the number and type of its gen-

eral addressing modes. The basic time for most instructions

is 2 cycles. A relatively small number of instructions require

an additional address unit time, as shown in the timing ta-

bles in Section D.5.5. Non-pipelined floating-point instruc-

tions as well as Custom-Slave instructions require an addi-

tional 3 cycles plus 2 cycles for each quad-word operand in

memory.

3. Cache and TLB misses can cause the flow of instructions

through the pipeline to be delayed, as can non-aligned

references. Section D.4 explains the performance impact

for these forms of storage delays.

The effective time T needed to execute an instruction is

eff

given by the following formula:

For instructions with 2 general addressing modes, 2 addi-

tional cycles are required when both addressing modes re-

fer to memory. Certain general addressing modes require an

additional processing time, as shown in Table D-1. For ex-

ample, the instruction MOVD 4(8(FP)), TOS requires 7 cy-

cles in the Address Unit; 2 cycles for the basic time, an

additional 2 cycles because both modes refer to memory,

and an additional 3 cycles for Memory Relative addressing

mode.

e

a

T T

T

eff

T

e

d

s

T

e

is the execution time in the pipeline in the absence of

data dependencies between instructions and storage de-

lays, T is the delay due to data dependencies, and T is the

d s

effect of storage delays.

D.2 BASIC EXECUTION TIMES

Instruction flow in sequence through the pipeline stages im-

plemented by the Loader, Address Unit, and Execution Unit.

In almost all cases, the Loader is at least as fast at decod-

ing an instruction as the Address Unit is at processing the

instruction. Consequently, the effects of the Loader can be

ignored when analyzing the smooth flow of instructions in

the pipeline, and it is only necessary to consider the times

for the Address Unit and Execution Unit. The time required

by the Loader to fetch and decode instructions is significant

only when there are control dependencies between instruc-

tions or Instruction Cache misses, both of which are ex-

plained later.

TABLE D-1. Additional Address Unit Processing

Time for Complex Addressing Modes

Additional

Mode

Cycles

Memory Relative

External

3

8

2

Scaled Indexing

D.2.3 Execution Unit Timing

The time for the pipeline to advance from one instruction to

the next is typically determined by the maximum time of the

Address Unit and Execution Unit to complete processing of

the instruction on which they are operating. For example, if

the Execution Unit is completing instruction n in 2 cycles

The Execution Unit processing times for the various

NS32532 instructions are provided in Section D.5.5. Certain

operations cause a break in the instruction flow through the

pipeline.

Some of these operation simply stop the Address Unit,

while others flush the instruction queue as well. The infor-

mation on how to evaluate the penalty resulting from in-

struction flow breaks is provided in the following sections.

a

and the Address Unit is completing instruction n 1 in 4

cycles, then the pipeline will advance in 4 cycles. For certain

instructions, such as RESTORE, the Address Unit waits until

the Execution Unit has completed the instruction before

proceeding to the next instruction. When such an instruction

is in the Execution Unit, the time for the pipeline to advance

is equal to the sum of the time for the Execution Unit to

complete instruction n and the time for the Address Unit to

D.3 INSTRUCTION DEPENDENCIES

Interactions between instructions in the pipeline can cause

delays. Two types of interactions can arise, as described

below.

a

complete instruction n 1. The processing times for the

Loader, Address Unit, and Execution Unit are explained be-

low.

D.3.1 Data Dependencies

In certain circumstances the flow of instructions in the pipe-

line will be delayed when the result of an instruction is used

as the source of a succeeding instruction. Such interlocks

are automatically detected by the microprocessor and han-

dled with complete transparency to software.

D.2.1 Loader Timing

The Loader can process an instruction field on each clock

cycle, where a field is one of the following:

An opcode of 1 to 3 bytes including addressing mode

specifiers.

#

D.3.1.1 Register Interlocks

When an instruction uses a base register that is the destina-

tion of either of the previous 2 instructions, a delay occurs.

The delay is 3 cycles when, as in the following example, the

base register is modified by the immediately preceding in-

struction. Modifications of the Stack Pointer resulting from

the use of TOS addressing mode do not cause any delay.

Also, there is no delay for a data dependency when the

instruction that modifies the register is one for which the

Address Unit stops.

Up to 2 index bytes, if scaled index addressing mode is

used.

#

A displacement.

#

An immediate value of 8, 16 or 32 bits.

#

The Loader requires additional time in the following cases:

1 additional cycle when 2 consecutive double-word fields

begin at an odd address.

#

93

Appendix D. Instruction Execution Times (Continued)

n: ADDD R1,R0

; modify R0

The delay is 2 cycles when the memory location is modified

2 instructions before its use as a source operand or effec-

tive address, as shown in this example.

n01: MOVD 4(R0),R2 ; R0 is base register,

delay 3 cycles

n: ADDQD 1,4(SP) ; modify 4(SP)

The delay is 1 cycle when the register is modified 2 instruc-

tions before its use as a base register, as shown in this

example.

n01: MOVD R0,R1

; no reference to 4(SP)

n02: CMPD 10, 4(SP); read 4(SP),

n: ADDD R1,R0

; modify R0

2 cycles delay

n01: MOVD 4(SP),R3 ; R0 not used

n02: MOVD 4(R0),R2 ; R0 is base register,

delay 1 cycle

Certain sequences of read and write references can cause

a delay of 1 cycle although there is no data dependency

between the references. This arises because the Data

Cache is occupied for 2 cycles on write references. In the

absence of data dependencies, read references are given

priority over write references. Therefore, this delay only oc-

curs when an instruction with destination in memory is fol-

lowed 2 instructions later by an instruction that refers to

memory (read or write) and 3 instructions later by an instruc-

tion that reads from memory. Here is an example:

When an instruction uses an index register that is the desti-

nation of the previous instruction, a delay of 1 cycle occurs,

as shown in the example below. If the register is modified 2

or more instructions prior to its use as an index register,

then no delay occurs.

n: ADDD R1,R0

n01: MOVD 4(SP)[R0:B],R2

; R0 is index register,

; modify R0

n: MOVD R0,4(SP) ; memory write

n01: MOVD R6,R7

; any instruction

delay 1 cycle

n02: MOVD 8(SP),R0 ; memory read or write

n03: MOVD 12(SP),R1; memory read

delayed 1 cycle

Bypass circuitry in the Execution Unit generally avoids delay

when a register modified by one instruction is used as the

source operand of the following instruction, as in the follow-

ing example.

D.3.2 Control Dependencies

The flow of instructions through the pipeline is delayed

when the address from which to fetch an instruction de-

pends on a previous instruction, such as when a conditional

branch is excuted. The Loader includes special circuitry to

handle branch instructions (ACB, BR, Bcond, and BSR) that

serves to reduce such delays. When a branch instruction is

decoded, the Loader calculates the destination address and

selects between the sequential and non-sequential instruc-

tion streams. The non-sequential stream is selected for un-

conditional branches. For conditional branches the selec-

tion is based on the branch’s direction (forward or back-

ward) as well as the tested condition. The branch is predict-

ed taken in any of the following cases.

n: ADDD R1,R0

; modify R0

n01: MOVD R0,R2

; R0 is source register,

no delay

For the uncommon case where the operand in the source

register is larger than the destination of the previous instruc-

tion, a delay of 2 cycles occurs. Here is an example.

n: ADDB R1,R0

; modify byte in R0

; R0 dw source operand,

2 cycle delay

n01: MOVD R0,R2

Note: The Address Unit does not make any differentiation between CPU

and FPU registers. Therefore, register interlocks can occur between

integer and floating-point instructions.

The branch is backward.

#

D.3.1.2 Memory Interlocks

The tested condition is either NE or LE.

When an instruction reads a source operand (or address for

effective address calculation) from memory that depends on

the destination of either of the previous 2 instructions, a

delay occurs. The CPU detects a dependency between a

read and a write reference in the following cases, which

include some false dependencies in addition to all actual

dependencies:

Measurements have shown that the correct stream is se-

lected for 64% of conditional branches and 71% of total

branches.

If the Loader selects the non-sequential stream, then the

destination address is transferred to the Instruction Cache.

For conditional branches, the Loader saves the address of

the alternate stream (the one not selected). When a condi-

tional branch instruction reaches the Execution Unit, the

condition is resolved, and the Execution Unit signals the

Loader whether or not the branch was taken. If the branch

had been incorrectly predicted, the Instruction Cache be-

gins fetching instructions from the correct stream.

Either reference crosses a double-word boundary

#

Address bits 0 through 11 are equal

#

Address bits 2 through 11 are equal and either reference

is for a word

#

Address bits 2 through 11 are equal and either reference

is for a double-word

#

The delay for handling a branch instruction depends on

whether the branch is taken and whether it is predicted cor-

rectly. Unconditional branches have the same delay as cor-

rectly predicted, taken conditional branches.

The delay for a memeory interlock is 4 cycles when, as in

the following example, the memory location is modified by

the immediately preceding instruction.

Another form of delay occurs when 2 consecutive condition-

al branch instructions are executed. This delay of 2 cycles

arises from contention for the register that holds the alter-

nate stream address in the Loader.

n: ADDQD 1,4(SP) ; modify 4(SP)

n01: CMPD 10,4(SP) ; read, 4(SP),

4 cycle delay

Control dependencies also arise when JUMP, RET, and oth-

er non-branch instructions alter the sequential execution of

instructions.

94

Appendix D. Instruction Execution Times (Continued)

D.4 STORAGE DELAYS

should be separately evaluated through a careful examina-

tion of the instruction sequence.

The flow of instructions in the pipeline can be delayed by

off-chip memory references that result from misses in the

on-chip storage buffers and by misalignment of instructions

and operands. These considerations are explained in the

following sections. The delays reported assume no wait

states on the external bus and no interference between in-

struction and data references.

The following assumptions are made:

Ð The entire instruction, with displacements and immedi-

ate operands, is present in the instruction queue when

needed.

Ð All memory operands are available to the Execution Unit

and Address Unit when needed.

D.4.1 Instruction Cache Misses

Ð Memory writes are performed at full speed through the

write buffer.

An Instruction Cache miss causes a 5 cycle gap in the fetch-

ing of instructions. When the miss occurs for a non-sequen-

tial instruction fetch, the pipeline is idle for the entire gap, so

the delay is 5 cycles. When the miss occurs for a sequential

fetch, the pipeline is not idle for the entire gap because

instructions that have been prefetched ahead and buffered

can be executed. The delay for misses on non-sequential

instruction fetches can be estimated to be approximately

half the gap, or 2.5 cycles.

Ð Where possible, the values of operands are taken into

consideration when they affect instruction timing, and a

range of times is given. When this is not done, the worst

case is assumed.

D.5.1 Definitions

T

Time required by the Execution Unit to execute an

instruction.

eu

T

Total processing time in the Address Unit.

au

ad

D.4.2 Data Cache Misses

Extra time needed by the Address Unit, in addition

to the basic time, to process more complex cases.

A Data Cache miss causes a delay of 2 cycles. When a

burst read cycle is used to fill the cache block, then 3 addi-

tional cycles are required to update the Data Cache. In case

a burst cycle is used and either of the 2 instructions follow-

ing the instruction that caused the miss also reads from

memory, then an additional delay occurs: 3 cycle delay

when the instruction that reads from memory immediately

follows the miss, and 2 cycle delay when the memory read

occurs 2 instructions after the miss.

T

can be evaluated as follows:

ad

e

a

T T

y1

T

x

ad

y2

e

2 if the instruction has two general operands

and both of them are in memory.

x

0 otherwise.

T

and T are related to operands 1 and 2 re-

y2

spectively. Their values are given below.

y1

D.4.3 TLB Misses

e

T

3 if Memory Relative

8 if External

y(1, 2)

There is a delay for the MMU to translate a virtual address

whenever there is a TLB miss for an instruction fetch, data

read or data write and whenever the M-bit in the Page Table

Entry (PTE) must be set for a data write that hits in the TLB.

The delay for the MMU to handle a TLB miss is 15 cycles

when no update to the PTEs is necessary. When only the

Level-1 PTE must be updated, the delay is 17 cycles; when

only the Level-2 PTE must be updated, the delay is 22 cy-

cles. When both PTEs must be updated, the delay is 24

cycles.

2 if Scaled Indexing

0 if any other addressing mode

The following parameters are only used for floating-point

execution time calculations.

T

anp

Additional Address Unit time needed to process

floating-point instructions in non-pipelined mode.

(Section D.2.2).

T

may be totally hidden for pipelined instruc-

anp

tions. For non-pipelined instructions it can be cal-

culated as follows:

D.4.4 Instruction and Operand Alignment

When a data reference (either read or write) crosses a dou-

ble-word boundary, there is a delay of 2 cycles.

e

a

2 * (Number of 64-bit operands in

memory)

T

anp

3

When the opcode for a non-sequential instruction crosses a

double-word boundary, there is a delay of 1 cycle. No delay

occurs in the same situation for a sequential instruction.

There is also a delay of 2 cycles when an instruction fetch is

located on a different page from the previous fetch and

there is a hit in the Instruction Cache. This delay, which is

due to the time required to translate the new page’s ad-

dress, also occurs following any serializing operation.

T

tcs

Time required to transfer ID and Opcode, if no op-

erand needs to be transferred to the slave. Other-

wise, it is the time needed to transfer the last 32

bits of operand data to the slave. In the latter case

the transfer of ID and Opcode as well as any oper-

and data except the last 32 bits is included in the

Execution Unit timing.

T

tsc

Time required by the CPU to complete the floating-

point instruction upon receiving the DONE signal

from the slave. This includes the time to process

the DONE signal itself in addition to the time need-

ed to read the result (if any) from the slave.

D.5 EXECUTION TIME CALCULATIONS

This section provides the necessary information to calculate

the T portion of the effective time required by the CPU to

e

execute an instruction.

The effects of data dependencies and storage delays are

not taken into account in the evaluation of T , rather, they

e

95

Appendix D. Instruction Execution Times (Continued)

l

This parameter is related to the floating-point oper-

and size as follows:

6. The keyword defined for the Bcond instruction have the

following meaning:

e

Standard floating (32 bits): l

Long floating (64 bits):

0

1

BTPC Branch Taken, Predicted Correctly

l

BTPI

Branch Taken, Predicted Incorrectly

BNTPC Branch Not Taken, Predicted Correctly

BNTPI Branch Not Taken, Predicted Incorrectly

D.5.2 Notes on Table Use

1. In the T column the notation n1

eu

mum, n2 maximum.

x

n2 means n1 mini-

D.5.3 T Evaluation

eff

2. In the notes column, notations held within angle brackets

kl

The T portion of the effective execution time for a certain

e

instruction in an instruction sequence is obtained by per-

forming the following steps:

indicate alternatives in the operand addressing

modes which affect the execution time. A table entry

which is affected by the operand addressing may have

multiple values, corresponding to the alternatives. This

addressing notations are:

1. Label the current and previous instruction in the se-

b

quence with n and n 1 respectively.

2. Obtain from the tables the values of T and T for in-

eu au

k l

I

Immediate

b

struction n and T for instruction n 1.

eu

k

l

R

CPU register

3. For floating-point instructions, obtain the values of T

tcs

l

M

Memory

and T

.

tsc

4. Use the following formula to determine the execution time

T .

l

F

FPU register, either 32 or 64 bits

Memory, except Top of Stack

Top of Stack

l

m

e

l

e

a

b

(n)

flt

b

T

e

T

(n)

dpf

Break (n 1))

func (T (n), T (n 1), T (n 1),

au eu flt

a

T (n)

eu

T

l

x

Any addressing mode

T

dpf

is the delay incurred before an instruction can begin

l

ab

a and b represent the addressing modes of oper-

ands 1 and 2 respectively. Both of them can be

execution. It must be considered only when the floating-

point pipelined mode is enabled.

k

l

any addressing mode. (e.g.,

memory to CPU register).

MR

means

For a non-floating-point instruction, it represents the time

needed to complete all the instructions in the FIFO. For a

floating-point instruction, it is only relevant if the FIFO is

full, and represents the time to complete the first instruc-

tion in the FIFO.

3. The notation ‘Break K’ provides pipeline status informa-

tion after executing the instruction to which ‘Break K’ ap-

plies. The value of K is interpreted as follows:

e

K

0

The Address Unit was stopped by the instruction

but the pipeline was not flushed. The Address

Unit can start processing the next instruction im-

mediately.

func provides the amount of processing time in the Ad-

dress Unit that cannot be hidden. Its definition is given

below.

s

e

b

func

0

if T (n)

au

(T (n 1)

eu

l

k

K

0

The pipeline was flushed by the instruction. The

Address Unit must wait for K cycles before it can

start processing the next instruction.

a

b

(n 1))

T

flt

AND NOT Break (n 1)

b

The Address Unit was stopped at the beginning

of the instruction but it was restarted K cycles

l

b

a

b

(n 1)

b

T

(n)

T

if T (n)

au

(T (n 1)

eu

au

eu

l

a

b

(n 1))

T

flt

AND NOT Break (n 1)

before the end of it. The Address Unit can start

processing the next instruction cycles before

b

K

l

a

T

0

K

if (T (n)

au

K)

0

the end of the instruction to which ‘Break K’ ap-

plies.

au

b

AND Break (n 1)

4. Some instructions must wait for pending writes to com-

plete before being able to execute. The number of cycles

that these instructions must wait for, is between 6 and 7

for the first operand in the write buffer and 2 for the sec-

ond operand, if any.

s

K)

a

if (T (n)

au

0

b

AND Break (n 1)

b

K is the value associated with Break (n 1).

5. The CBITIi and SBITIi instructions will execute a RMW

access after waiting for pending writes. The extra time

required for the RMW access is only 3 cycles since the

read portion is overlapped with the time in the Execution

Unit.

96

Appendix D. Instruction Execution Times (Continued)

T

only applies to floating-point instructions and is al-

ways 0 for other instructions. It is evaluated as follows:

function. In this example there are no data dependencies or

storage buffer misses; only the basic instruction execution

times in the pipeline, control dependencies, and instruction

alignment are considered.

flt

if pipelined mode is disabled, then

e

a

T

tsc

T

t

tcs

T

fpu

flt

else

The following is the source of the procedure in C.

e

Tflt

0

if group A instruction.

if group B instruction.

unsigned fib(x)

a

max (T , T

)

T

tsc

prv tcs

int

x;

T

is the execution time in the Floating-Point

fpu

À

Unit. T is the time needed by the CPU and FPU

prv

l

if (x

else

2)

to complete all the floating-point instructions in the

FIFO.

return (fib(x11) 0 fib(x12));

5. Calculate the total execution time T by using the follow-

eff

ing formula:

return(1);

e

a

T T

T

e

Ó

eff

d

s

Where T and T are dependent on the instruction se-

s

d

The assembly code for the procedure with comments indi-

cating the execution time is shown below. The procedure

requires 26 cycles to execute when the actual parameter is

less than or equal to 2 (branch taken) and 99 cycles when

the actual parameter is equal to 3 (recursive calls).

quence, and can be obtained using the information pro-

vided in Section D.4.

D.5.4 Instruction Timing Example

This section presents a simple instruction timing example

for a procedure that recursively evaluates the Fibonacci

fib:

movd

cmpqd

bge

r3,tos

r4,tos

r1,r3

$(2),r3

.L1

; 2 cycles

; 2 cycles, Break 4 If Branch Taken

; 2 cycles

movd

addqd

bsr

r3,r1

$(12),rl ; 2 cycles

fib

; 3 cycles

movd

addqd

bsr

r0,r4

r3,r1

; 4 cycles 0 4 Cycles due to RET

; 2 cycles

$(11),r1 ; 2 cycles

fib

; 3 cycles

a

addd

movd

ret

r4,r0

tos,r4

tos,r3

$(0)

; 4 cycles 0 1 cycle alignment

; 2 cycles

4 cycles due to RET

; 2 cycles

; 4 cycles, break 4

.align 4

movqd

movd

ret

a

L1:

$(1),r0

tos,r4

tos,r3

$(0)

; 4 cycles

; 2 cycles

4 cycles due to BGE

; 4 cycles, Break 4

97

Appendix D. Instruction Execution Times (Continued)

D.5.5 Execution Timing Tables

The following tables provide the execution timing information for all the NS32532 instructions. The table for the floating-point

instructions provides only the CPU portion of the total execution time. The FPU execution times can be found in the NS32381

and NS32580 datasheets.

D.5.5.1 Basic and Memory Management Instructions

Mnemonic

ABSi

T

Notes

Mnemonic

T

au

Notes

eu

au

eu

a

b

5

2

T

CHECKi

10

2

T

Break 3.

ad

If SRC is out

of bounds and

the V bit in the

PSR is set,

then add trap

time.

ACBi

If incorrect prediction

then Break 1

ad

a

ADDi

2

9

2

4

T

ad

ADDCi

ADDPi

ADDQi

ADDR

ADJSPi

a

CINV

10

2

T

Wait for

pending

writes.

ad

Break 5

a

e

5

3

2

T

i

B, W

D

Break 0

ad

CMPi

ad

a

e

number

of elements.

Break 0

CMPMi

6

8 * n

n

a

ANDi

ASHi

2

9

2

T

ad

T

ad

B

COND

2

x

2

3

2

BTPC

a

CMPQi

CMPSi

2

T

ad

2

BTPI

Break 2

a

e

n number

of elements.

Break 0

7

13 * n

2

T

ad

2

BNTPC

2

BNTPI

(see Note 5 in

Section D.5.2)

a

e

number

CMPST

6

20 * n

2

T

ad

n

of elements.

Break 0

a

BICi

2

6

2

T

ad

a

BICPSRi

Wait for pending writes.

Break 5

ad

a

COMi

CVTP

CXP

2

5

2

4

T

ad

a

BISPSRi

BPT

6

2

T

ad

Wait for pending writes.

Break 5

17

21

13

Break 5

30

21

2

Modular

Direct

a

T

CXPD

DEIi

11

5

ad

2

a

e

i 0/4/12 for

28

4 * i

T

ad

Break 5

B/W/D.

Break 0

BR

2

x

7

3

2

DIA

3

2

Break 5

a

BSR

CASEi

CBITi

3

2

T

ad

a

e

for B/W/D

DIVi

(30

x

40)

4 * i

2

T

i

0/4/12

ad

a

Break 5

ad

k

l

10

14

2

R

a

e

number

ENTER

15

2 * n

3

n

a

T

M

Break 0

ad

of registers

saved.

k

l

CBITIi

18

2

M

ad

Wait for pending writes.

Execute interlocked

Break 0

a

e

number

EXIT

EXTi

8

2 * n

2

8

n

RMW access. Break 5

of registers

restored

k

l

12

R

a

13

8

6

T

M

ad

b

Break

3

k

l

EXSi

11

14

6

R

a

T

M

Break

98

Appendix D. Instruction Execution Times (Continued)

D.5.5.1 Basic and Memory Management Instructions (Continued)

Mnemonic

T

Notes

number

Mnemonic

T

au

Notes

eu

au

eu

a

e

of bytes

a

FFSi

11

3 * i

2

T

i

MOVSVi

9

2

T

Wait for

pending writes.

Break 5

ad

FLAG

4

32

21

2

No trap

a

Trap, Modular

Trap, Direct

If trap then:

MOVUSi

11

T

Wait for

pending writes.

Break 5

À

pending writes;

wait for

a

MOVXii

MOVZii

MULi

2

T

ad

T

ad

T

ad

Ó

Break 5

a

e

0/4/12

for B/W/D.

General case.

If MULD and

13

2 * i

i

k

l

IBITi

10

14

2

R

M

a

2

5

T

Break 0

ad

a

INDEXi

INSi

43

T

a

24

2

T

ad

k

l

s

SRC 255

15

18

8

R

0

l

a

8

6

T

M

ad

NEGi

NOP

NOTi

ORi

2

ad

k

l

INSSi

14

19

6

R

2

a

T

M

a

3

2

T

ad

T

ad

T

ad

Break 0

Break 5

2

a

JSR

3

9

4

2

T

ad

e

for B/W/D

QUOi

(30

(32

x

40) 2

i

0/4/12

JUMP

LMR

a

4 * i

11

Wait for

pending writes.

Break 5

a

RDVAL

10

2

T

Wait for

pending writes.

Break 5

ad

a

e

FP,

SP, USP, SP, MOD.

LPRi

6

5

2

T

CPU Reg

e

for B/W/D

ad

REMi

x

42) 2

T

i

0/4/12

a

4 * i

Break 0

CPU Reg

INTBASE, DSR,

BPC, UPSR.

Wait for pending

writes.

a

e

number

RESTORE

7

2 * n

2

n

e

CFG,

ad

of registers

restored.

Break 0

RET

4

3

Break 4

Break 5

CPU Reg

PSR CAR. Wait for

pending writes.

Break 5

RETI

19

13

29

22

5

Noncascaded, Modular

Noncascaded, Direct

Cascaded, Modular

Cascaded, Direct

a

e

DCR,

7

3

2

T

ad

Wait for

pending writes.

Break 5

a

LSHi

MEIi

2

5

T

ad

a

e

0/4/12

for B/W/D.

13

2 * i

i

ad

RETT

14

8

5

Modular

Direct

Break 0

a

e

for B/W/D

MODi

(34

a

x

49)

4 * i

2

T

i

0/4/12

ad

Wait for

pending writes.

Break 5

a

MOVi

2

T

ad

a

e

number

MOVMi

5

4 * n

n

a

ad

ROTi

7

8

3

2

T

ad

of elements.

Break 0

RXP

5

Break 5

a

SCONDi

SAVE

ad

a

MOVQi

MOVSi

2

T

ad

a

e

number

8

2 * n

2

n

e

of elements.

n

number

of registers.

Break 0

a

12

14

4 * n

8 * n

2

T

ad

T

ad

No options.

B, W and/or U

Options in effect.

k

l

SBITi

10

14

2

a

R

M

2

T

ad

Break 0

a

e

n number

of elements.

MOVST

16

9 * n

2

T

ad

Break 0

99

Appendix D. Instruction Execution Times (Continued)

D.5.5.1 Basic and Memory Management Instructions (Continued)

Mnemonic

T

Notes

Mnemonic

T

au

Notes

eu

10

18

au

eu

k

l

a

e

SBITIi

2

a

R

M

SPRi

5

3

2

T

CPU Reg

PSR, CAR

e

all others

ad

2

T

ad

a

SUBi

2

6

2

T

ad

T

ad

T

ad

Wait for pending

writes. Execute

interlocked RMW

access.

SUBCi

SUBPi

SVC

Break 5

32

21

2

Modular

Direct

SETCFG

SKPSi

6

2

Break 5

Wait for

pending writes.

Break 5

a

e

n number of

elements.

Break 0

8

6 * n

2

T

ad

k

l

TBITi

WAIT

8

11

2

R

M

a

e

n number of

elements.

Break 0

SKPST

SMR

6

20 * n

a

2

T

Break 0

ad

3

10

2

Wait for pending

writes. Wait

for interrupt

7

Wait for

pending writes.

Break 5

a

WRVAL

XORi

2

T

Wait for

pending writes.

Break 5

ad

a

T

100

Appendix D. Instruction Execution Times (Continued)

D.5.5.2 Floating-Point Instructions, CPU Portion

Mnemonic

T

eu

T

tcs

T

tsc

Group

Notes

au

k

l

FF

a

MOVf, NEGf,

ABSf, SQRTf,

LOGBf

2

4

6

11

13

2

T

2

1

3

A

B

anp

l

a

3 * l

4 * l

7 * l

T

MF

ad

l

IF

TF

l

a

b

a

T

ad

T

ad

2 * l

FM Break (1 l)

k

l

IM Break (1

a

b

a

MM

,

l)

k

l

a

ADDf, SUBf,

MULf, DIVf,

SCALBf

2

T

anp

T

anp

T

anp

T

anp

T

anp

T

anp

2

1

3

A

B

FF

MF

l

a

4

6

17

19

3 * l

7 * l

10 * l

l

IF

TF

l

a

b

a

T

ad

T

ad

2 * l

FM Break (1 l)

k

l

IM Break (1

b

MM

,

l)

k

l

a

b

1

ROUNDfi, TRUNCfi,

FLOORfi

11

13

2

T

anp

T

anp

T

anp

T

anp

2

3

2 * l

B

FR Break

l

a

b

a

IR Break

4 * l

7 * l

T

ad

T

ad

T

ad

FM Break (1

l)

b

l

k

l

MR

MM

,

1

IM Break (1

l

a

l

k

l

a

CMPf

18

20

23

25

2

T

anp

T

anp

T

anp

T

anp

2

B

FF

l

a

3 * l

6 * l

T

ad

T

ad

T

ad

MF

FM

l

k

l

k

l

k

MM

,

IM

,

MI

,

II

Break 3

k

l

FF

a

POLYf, DOTf,

MACf

2

4

6

2

T

anp

T

anp

T

anp

T

anp

2

1

A

B

A

l

a

3 * l

T

ad

MF

IF

l

k

l

TF

3 * l

,

l

FM Break (1

a

b

a

l)

11

4 * l

T

ad

k

Break (1

l

k

l

k

l

IM ,

l

a

13

7 * l

2

T

anp

2

1

B

MM

,

MI

,

ad

b

a

l)

k

l

RF

a

MOVif

LFSR

6

13

6

13

2

T

anp

T

anp

T

anp

T

anp

2

1

B

l

a

b

T

ad

T

ad

T

ad

RM Break

k

1

l

k

IM Break (1

l

a

3 * l

7 * l

MF

MM

,

IF

,

TF

l

k

l

b

l)

k

k l

l

a

6

2

T

anp

T

anp

T

anp

T

anp

2

1

B

R

l

a

6

3 * l

T

ad

M

I

k

l

T

a

b

Break

SFSR

11

2

T

2

3

B

1

anp

ad

k

l

a

MOVFL

4

6

2

T

anp

T

anp

2

1

B

FF

MF

k

l

k

l

TF

,

IF

,

ad

k

l

FM Break 0

a

15

17

2

T

anp

T

anp

T

ad

T

ad

2

B

l

k l

IM Break 0

MM

,

k

l

a

MOVLF

4

9

2

T

anp

T

anp

2

1

B

FF

MF

k

l

k

l

TF

a

T

ad

,

IF

,

k

l

FM Break 0

a

15

20

2

T

anp

T

anp

T

ad

T

ad

2

B

l

k l

IM Break 0

MM

,

101

Ý

Lit. 114272

Physical Dimensions inches (millimeters)

Hermetic Pin Grid Array (U)

Order Number NS32532-20, NS32532-25 or NS32532-30

NS Package Number U175A

LIFE SUPPORT POLICY

NATIONAL’S PRODUCTS ARE NOT AUTHORIZED FOR USE AS CRITICAL COMPONENTS IN LIFE SUPPORT

DEVICES OR SYSTEMS WITHOUT THE EXPRESS WRITTEN APPROVAL OF THE PRESIDENT OF NATIONAL

SEMICONDUCTOR CORPORATION. As used herein:

1. Life support devices or systems are devices or

systems which, (a) are intended for surgical implant

into the body, or (b) support or sustain life, and whose

failure to perform, when properly used in accordance

with instructions for use provided in the labeling, can

be reasonably expected to result in a significant injury

to the user.

2. A critical component is any component of a life

support device or system whose failure to perform can

be reasonably expected to cause the failure of the life

support device or system, or to affect its safety or

effectiveness.

National Semiconductor

Corporation

National Semiconductor

Europe

National Semiconductor

Hong Kong Ltd.

National Semiconductor

Japan Ltd.

a

1111 West Bardin Road

Arlington, TX 76017

Tel: 1(800) 272-9959

Fax: 1(800) 737-7018

Fax:

(

49) 0-180-530 85 86

@

13th Floor, Straight Block,

Ocean Centre, 5 Canton Rd.

Tsimshatsui, Kowloon

Hong Kong

Tel: (852) 2737-1600

Fax: (852) 2736-9960

Tel: 81-043-299-2309

Fax: 81-043-299-2408

Email: cnjwge tevm2.nsc.com

a

Deutsch Tel:

English Tel:

Fran3ais Tel:

Italiano Tel:

(

49) 0-180-530 85 85

49) 0-180-532 78 32

49) 0-180-532 93 58

49) 0-180-534 16 80

National does not assume any responsibility for use of any circuitry described, no circuit patent licenses are implied and National reserves the right at any time without notice to change said circuitry and specifications.


型号：	NS32532-20
厂家：	ETC
描述：
文件：	总102页 (文件大小：1116K)
中文：	中文翻译
下载：	下载PDF数据表文档文件

NS32532-20 [ETC]

相关型号：

NS32532-25

NS32532-30

NS32580-20

NS32580-25

NS32817D-70

NS32817D-80

NS32817D-80

NS32817N-70

NS32817V-70X

NS32817V-80

NS32817V-80X

NS32818D-70