Intel MultiProcessor User Manual

MultiProcessor

Specification

Version 1.4

May 1997

Revision History

Revision

Revision History

Date

Pre-release Version 1.0. Formerly called “PC+MP Specification”

10/27/93

4/11/94

-001

Version 1.1. Resolves conflicts with MCA-based systems. The following

changes have been made:

1. Two MP feature information bytes were moved from the BIOS System

Configuration Table to the RESERVED area of the MP Floating Pointer

Structure.

2. If the MP Floating Pointer Structure is present, it indicates that the system

is MP-compliant, in accordance with this specification. (Previously, this was

indicated by bit 0 of MP Feature Information Byte 1.)

3. One more hardware default system configuration was added for MCA+PCI

with the integrated APIC.

-002

-003

-004

Minor technical corrections.

6/1/94

9/1/94

7/1/95

Added Appendix D: Multiple I/O APIC Multiple PCI Bus Systems.

Version 1.4. Added extended configuration table to improve support for

multiple PCI bus configurations and improve future expandability. Made

clarifications to Appendix B.

-005

-006

Added Appendix E: Errata. Includes information for hierarchical PCI bus.

08/15/96

05/12/97

Appendix E Update. Included Byte 2, Bit 6 information to MP Floating Point

Structure section (4.1).

iii

Table of Contents

Chapter 1 Introduction

1.1

1.2

1.3

1.4

1.5

1.6

1.7

Goals ........................................................................................................ 1-1

Features of the Specification .................................................................... 1-2

Scope........................................................................................................ 1-2

Target Audience ....................................................................................... 1-3

Organization of This Document ................................................................ 1-3

Conventions Used in This Document ....................................................... 1-4

For More Information................................................................................ 1-4

Chapter 2 System Overview

2.1

Hardware Overview .................................................................................. 2-2

2.1.1 System Processors...................................................................... 2-2

2.1.2 Advanced Programmable Interrupt Controller ............................. 2-3

2.1.3 System Memory........................................................................... 2-4

2.1.4 I/O Expansion Bus ....................................................................... 2-4

BIOS Overview ......................................................................................... 2-5

Operating System Overview..................................................................... 2-5

2.2

2.3

Chapter 3 Hardware Specification

3.1

3.2

3.3

3.4

3.5

3.6

System Memory Configuration ................................................................. 3-1

System Memory Cacheability and Shareability......................................... 3-2

External Cache Subsystem ...................................................................... 3-4

Locking ..................................................................................................... 3-4

Posted Memory Write ............................................................................... 3-5

Multiprocessor Interrupt Control ............................................................... 3-5

3.6.1 APIC Architecture ........................................................................ 3-5

3.6.2 Interrupt Modes............................................................................ 3-6

3.6.2.1 PIC Mode...................................................................... 3-7

3.6.2.2 Virtual Wire Mode ......................................................... 3-9

3.6.2.3 Symmetric I/O Mode................................................... 3-11

3.6.3 Assignment of System Interrupts to the APIC Local Unit........... 3-12

3.6.4 Floating Point Exception Interrupt.............................................. 3-12

3.6.5 APIC Memory Mapping.............................................................. 3-12

Contents

3.6.6 APIC Identification ..................................................................... 3-13

3.6.7 APIC Interval Timers.................................................................. 3-13

3.6.8 Multiple I/O APIC Configurations ............................................... 3-13

RESET Support ...................................................................................... 3-14

3.7.1 System-wide RESET ................................................................. 3-14

3.7.2 System-wide INIT....................................................................... 3-15

3.7.3 Processor-specific INIT.............................................................. 3-15

System Initial State................................................................................. 3-16

Support for Fault-resilient Booting .......................................................... 3-16

3.7

3.8

3.9

Chapter 4 MP Configuration Table

4.1

4.2

4.3

MP Floating Pointer Structure................................................................... 4-3

MP Configuration Table Header ............................................................... 4-5

Base MP Configuration Table Entries....................................................... 4-6

4.3.1 Processor Entries......................................................................... 4-7

4.3.2 Bus Entries................................................................................. 4-10

4.3.3 I/O APIC Entries......................................................................... 4-12

4.3.4 I/O Interrupt Assignment Entries................................................ 4-12

4.3.5 Local Interrupt Assignment Entries............................................ 4-15

Extended MP Configuration Table Entries.............................................. 4-17

4.4.1 System Address Space Mapping Entries................................... 4-18

4.4.2 Bus Hierarchy Descriptor Entries............................................... 4-21

4.4.3 Compatibility Bus Address Space Modifier Entries.................... 4-22

4.4

Chapter 5 Default Configurations

5.1

5.2

5.3

Discrete APIC Configurations................................................................... 5-2

Integrated APIC Configurations................................................................ 5-4

Assignment Of I/O Interrupts To The APIC I/O Unit ................................. 5-6

5.3.1 EISA and IRQ13 .......................................................................... 5-7

5.3.2 Level-triggered Interrupt Support................................................. 5-7

Assignment Of System Interrupts To The APIC Local Unit ...................... 5-7

5.4

Contents

Appendix A System BIOS Programming Guidelines

A.1

A.2

A.3

A.4

BIOS Post Initialization ............................................................................A-1

Controlling the Application Processors....................................................A-2

Programming the APIC for Virtual Wire Mode.........................................A-2

Constructing the MP Configuration Table................................................A-4

Appendix B Operating System Programming Guidelines

B.1

B.2

B.3

B.4

Operating System Boot-up .......................................................................B-1

Operating System Booting and Self-configuration....................................B-2

Interrupt Mode Initialization and Handling ................................................B-2

Application Processor Startup ..................................................................B-3

B.4.1 USING INIT IPI ............................................................................B-4

B.4.2 USING STARTUP IPI...................................................................B-5

AP Shutdown Handling.............................................................................B-5

Other IPI Applications...............................................................................B-6

B.6.1 Handling Cache Flush..................................................................B-6

B.6.2 Handling TLB Invalidation............................................................B-6

B.6.3 Handling PTE Invalidation............................................................B-6

Spurious APIC Interrupts..........................................................................B-6

Supporting Unequal Processors...............................................................B-7

B.5

B.6

B.7

B.8

Appendix C System Compliance Checklist

Appendix D Multiple I/O APIC Multiple PCI Bus Systems

D.1

Interrupt Routing with Multiple APICs.......................................................D-1

D.1.1 Variable Interrupt Routing............................................................D-1

D.1.2 Fixed Interrupt Routing ................................................................D-2

Bus Entries in Systems with More Than One PCI Bus .............................D-3

I/O Interrupt Assignment Entries for PCI Devices.....................................D-3

D.2

D.3

Appendix E Errata

Glossary

vii

Contents

Figures

1-1. Conceptual Overview...................................................................... 1-1

1-2. Memory Layout Conventions .......................................................... 1-4

2-1. Multiprocessor System Architecture................................................ 2-2

2-2. APIC Configuration ......................................................................... 2-3

3-1. System Memory Address Map........................................................ 3-2

3-2. PIC Mode....................................................................................... 3-8

3-3. Virtual Wire Mode via Local APIC................................................... 3-9

3-4. Virtual Wire Mode via I/O APIC..................................................... 3-10

3-5. Symmetric I/O Mode ..................................................................... 3-11

3-6. Multiple I/O APIC Configurations .................................................. 3-14

4-1. MP Configuration Data Structures .................................................. 4-1

4-2. MP Floating Pointer Structure......................................................... 4-3

4-3. MP Configuration Table Header...................................................... 4-5

4-4. Processor Entry............................................................................... 4-7

4-5. Bus Entry....................................................................................... 4-10

4-6. I/O APIC Entry............................................................................... 4-12

4-7. I/O Interrupt Entry ......................................................................... 4-13

4-8. Local Interrupt Entry...................................................................... 4-15

4-9. System Address Space Entry ....................................................... 4-18

4-10. Example System with Multiple Bus Types and Bridge Types ....... 4-19

4-11. Bus Hierarchy Descriptor Entry..................................................... 4-21

4-12. Compatibility Bus Address Space Modifier Entry.......................... 4-23

5-1. Default Configuration for Discrete APIC.......................................... 5-3

5-2. Default Configuration for Integrated APIC....................................... 5-5

Tables

1-1. Document Organization .................................................................. 1-3

3-1. Memory Cacheability Map............................................................... 3-3

3-2. APIC Versions................................................................................. 3-6

4-1. MP Floating Pointer Structure Fields .............................................. 4-3

4-2. MP Configuration Table Header Fields........................................... 4-6

4-3. Base MP Configuration Table Entry Types..................................... 4-7

4-4. Processor Entry Fields.................................................................... 4-8

4-5. Intel486™ and Pentium^®Processor Signatures ............................. 4-9

viii

Contents

4-6. Feature Flags from CPUID Instruction............................................ 4-9

4-7. Bus Entry Fields............................................................................ 4-10

4-8. Bus Type String Values................................................................. 4-11

4-9. I/O APIC Entry Fields.................................................................... 4-12

4-10. I/O Interrupt Entry Fields............................................................... 4-14

4-11. Interrupt Type Values.................................................................... 4-15

4-12. Local Interrupt Entry Fields........................................................... 4-16

4-13. Extended MP Configuration Table Entry Types............................ 4-17

4-14. System Address Space Mapping Entry Fields.............................. 4-19

4-15. Bus Hierarchy Descriptor Entry Fields .......................................... 4-22

4-16. Compatibility Bus Address Space Modifier Entry Fields ............... 4-24

4-17. Predefined Range Lists................................................................. 4-24

5-1. Default Configurations..................................................................... 5-2

5-2. Default Configuration Interrupt Assignments .................................. 5-6

5-3. Assignment of System Interrupts to APIC Local Unit...................... 5-7

D-1. I/O Interrupt Entry Source Bus IRQ Field for PCI Devices..............D-3

Examples

A-1. Programming Local APIC for Virtual Wire Mode.............................A-3

B-1. Universal Start-up Algorithm ...........................................................B-3

Introduction

The MultiProcessor Specification, hereafter known as the “MP specification,” defines an

enhancement to the standard to which PC manufacturers design DOS-compatible systems.

MP-capable operating systems will be able to run without special customization on multiprocessor

systems that comply with this specification. End users who purchase a compliant multiprocessor

system will be able to run their choice of operating systems.

The MP specification covers PC/AT-compatible MP platform designs based on Intel processor

architectures and Advanced Programmable Interrupt Controller (APIC) architectures. The term

“PC/AT-compatible” here refers to the software-visible components of the PC/AT, not to hardware

features, such as the bus implementation, that are not visible to software. An implementation of

this specification may incorporate one or more industry standard buses, such as ISA, EISA, MCA,

PCI, or other OEM-specific buses.

1.1 Goals

The intent of this specification is to establish an MP Platform interface standard that extends the

performance of the existing PC/AT platform beyond the traditional single processor limit, while

maintaining 100% PC/AT binary compatibility.

The ultimate goal is to enable scalable, high-end workstations and enterprise server systems that

provide computer users with superior price/performance and have the ability to execute all existing

AT binaries, as well as MP-ready software packages on shrink-wrapped MP operating systems.

Figure 1-1 shows that at the heart of the specification are the data structures that define the

configuration of the MP system. The BIOS constructs the MP configuration data structures,

presenting the hardware in a known format to the standard device drivers or to the hardware

abstraction layer of the operating system. The specification details default hardware

configurations, and, for added flexibility, outlines extensions to the standard BIOS.

OPERATING SYSTEM

DRIVER/

HARDWARE

ABSTRACTION LAYER

MP BIOS

CONFIGURATION

DATA

STRUCTURES

HARDWARE

Figure 1-1. Conceptual Overview

Version 1.4

1-1

MultiProcessor Specification

1.2 Features of the Specification

The MP specification includes the following features:

•

A multiprocessor extension to the PC/AT platform that runs all existing uniprocessor shrink-

wrapped binaries, as well as MP binaries.

•

Support for symmetric multiprocessing with one or more processors that are Intel architecture

instruction set compatible, such as the CPUs in the Intel486™ and the Pentium processor

family.

•

Support for symmetric I/O interrupt handling with the APIC, a multiprocessor interrupt

controller.

•

Flexibility to use a BIOS with minimal MP-specific support.

An optional MP configuration table to communicate configuration information to an MP

operating system.

•

Incorporation of ISA and other industry standard buses, such as EISA, MCA, VL and PCI

buses in MP-compliant systems.

Requirements that make secondary cache and memory bus implementation transparent to

software.

1.3 Scope

There are many vendors building innovative MP systems based on Intel architectures today. Intel

supports and encourages vendors to develop advanced approaches to the technological

requirements of today's computing environments. In no way does the MP specification prevent

system manufacturers from adding their own unique value to MP systems. This specification does

not define the only way that multiprocessor systems can be implemented. Vendors may, for

example, create noncompliant, high-performance, scalable multiprocessor systems that do not have

the PC/AT compatibility required by this specification. Hardware vendors will always have the

option of offering one or more customized operating systems that support the unique, value-added

capabilities that they have designed into their systems. End users will be able to benefit from the

additional capabilities that these vendors offer.

The MP specification benefits PC vendors who wish to offer MP-enabled systems without having

to invest in the customization of one or more operating systems. This specification focuses on

providing a standard mechanism to add MP support to personal computers built around the PC/AT

hardware standard. With that goal, the specification covers only the minimum set of capabilities

required to extend the PC/AT platform to be MP-capable. The hardware required to implement the

MP specification is kept to a minimum, as follows:

•

One or more processors that are Intel architecture instruction set compatible, such as the CPUs

in the Intel486 or Pentium processor family.

•

One or more APICs, such as the Intel 82489DX Advanced Programmable Interrupt Controller

or the integrated APIC, such as that on the Intel Pentium 735\90 and 815\100 processors,

together with a discrete I/O APIC unit.

•

The necessary supporting electronics to duplicate the initialization and signaling protocol

described in this specification.

PC/AT-compliant hardware.

1-2

Version 1.4

Introduction

In addition to the hardware requirements, this document also specifies MP features that are visible

to the BIOS and operating system. However, it is important to understand that as hardware

technology progresses, the functions performed by the BIOS may change in accordance with the

hardware technology. ONLY THE INTERFACE TO THE OPERATING SYSTEM LEVEL

IS EXPECTED TO REMAIN CONSTANT.

This specification does not address issues relating to the processor's System Management Mode

(SMM).

1.4 Target Audience

This document is intended for the following users:

•

OEMs who will be creating PC/AT-compatible, MP-ready hardware based on this

specification.

BIOS developers, either those who create general-purpose BIOS products or those who modify

these products to suit specific MP hardware.

Operating-system developers who will be adapting MP operating systems to run on the class of

MP-ready platform specified here.

1.5 Organization of This Document

Table 1-1 shows the organization of this document.

Table 1-1. Document Organization

Chapter

Description

Overview of the MultiProcessor Specification

Specification of the MP hardware

Specification of MP configuration information available to OS

Specification of default hardware configurations

Guidelines for MP BIOS programming

Appendix A

Appendix B

Appendix C

Appendix D

Guidelines for MP operating system programming

Checklist for hardware compliance to the specification

Clarifications for multiple I/O APIC, multiple PCI bus systems

Glossary

Definitions of specialized terms used in this document

Version 1.4

1-3

MultiProcessor Specification

1.6 Conventions Used in This Document

Signal names that are followed by the character # represent active low signals. For example,

FERR# is active when at its low-voltage state.

Throughout this document, the Intel 82489DX APIC is referred to as the “discrete APIC.” The

term “integrated APIC” is used to refer to an APIC integrated with other system components, such

as the Pentium 735\90 and 815\100 processors. This specification uses the term APIC to refer to

both discrete and integrated versions.

The processors of the Intel486 and Pentium processor family are “little endian” machines. This

means that the low-order byte of a multibyte data item in memory is at the lowest address, while

the high-order byte is at the highest address. Illustrations of data structures in memory show the

lowest addresses at the bottom and the highest addresses at the top of the illustration, as shown in

Figure 1-2. Bit positions are numbered from right to left.

INCREASING

ADDRESSES

24 23

16 15

RESERVED

0CH

RESERVED

08H

04H

00H

RESERVED

TWO-BYTE FIELD

ONE-BYTE

FIELD

THREE-BYTE FIELD

24 23

16 15

HIGH-ORDER BITS

LOW-ORDER BITS

Figure 1-2. Memory Layout Conventions

In some memory layout descriptions, certain fields are marked RESERVED. Software should

initialize these fields as binary zeros, but should otherwise treat them as having a future, though

unknown effect. SOFTWARE SHOULD AVOID ANY DEPENDENCE ON THE VALUES

IN THE RESERVED FIELDS.

1.7 For More Information

For more information, refer to any of the following documents:

•

82489DX Advanced Programmable Interrupt Controller (data book), Intel order number

290446

•

Intel486 Microprocessor Programmer's Reference Manual, Intel order number 240486

Intel Processor Identification with the CPUID Instruction AP-485, Intel order number 241618

Pentium Processor User’s Manual, Intel order number 241428

1-4

Version 1.4

System Overview

In the realm of multiprocessor architectures, there are several conceptual models for tying together

computing elements, and there are a variety of interconnection schemes and details of

implementation. Figure 2-1 shows the general structure of a design based on the MP specification.

The MP specification’s model of multiprocessor systems incorporates a tightly-coupled, shared-

memory architecture with a distributed interprocessor and I/O interrupt capability. It is fully

symmetric; that is, all processors are functionally identical and of equal status, and each processor

can communicate with every other processor. There is no hierarchy, no master-slave relationship,

no geometry that limits communication only to “neighboring” processors. The model is symmetric

in two important respects:

•

Memory symmetry. Memory is symmetric when all processors share the same memory space

and access that space by the same addresses. Memory symmetry offers a very important

feature—the ability for all processors to execute a single copy of the operating system. Any

existing system and application software will execute the same, regardless of the number of

processors installed in a system.

•

I/O symmetry. I/O is symmetric when all processors share access to the same I/O subsystem

(including I/O ports and interrupt controllers) and any processor can receive interrupts from

any source. Some multiprocessor systems that have symmetric access to memory are actually

asymmetric with regard to I/O interrupts, because they dedicate one processor to interrupt

functions. I/O symmetry helps eliminate the potential of an I/O bottleneck, thereby increasing

system scalability.

With both memory and I/O symmetry, a system that complies with the MP specification can

achieve hardware scalability as well as support software standardization. Based on this kind of

scalable architecture, systems developers can produce systems that span a wide range of price and

performance, and that all execute the same binaries.

Version 1.4

2-1

MultiProcessor Specification

CPU

HIGH-BANDWIDTH MEMORY BUS

ICC BUS

APIC

SHARED

MEMORY

MODULE

GRAPHICS

FRAME

BUFFER

I/O

INTERFACE

APIC

ICC

ADVANCED PROGRAMMABLE

INTERRUPT CONTROLLER

COMMUNICATIONS

I/O EXPANSION BUS

Figure 2-1. Multiprocessor System Architecture

2.1 Hardware Overview

The MP specification defines a system architecture based on the following hardware components:

•

One or more processors that are Intel architecture instruction set compatible, such as the CPUs

in the Intel486 and the Pentium processor family.

•

One or more APICs, such as the Intel 82489DX Advanced Programmable Interrupt Controller

or the integrated APIC on the Pentium 735\90 and 815\100 processors.

•

Software-transparent cache and shared memory subsystem.

Software-visible components of the PC/AT platform.

2.1.1 System Processors

To maintain compatibility with existing PC/AT software products, this specification is based on

the Intel486 and the Pentium processor family. To achieve a minimum level of MP system

performance, two or more processors that are Intel architecture instruction set compatible are

required.

Figure 2-2 gives a different point of view of a compliant system, showing the configuration of the

APICs with respect to the CPUs. While all processors in a compliant system are functionally

identical, this specification classifies them into two types: the bootstrap processor (BSP) and the

application processors (AP). Which processor is the BSP is determined by the hardware or by the

hardware in conjunction with the BIOS. This differentiation is for convenience and is in effect

2-2

Version 1.4

System Overview

only during the initialization and shutdown processes. The BSP is responsible for initializing the

system and for booting the operating system; APs are activated only after the operating system is

up and running. CPU1 is designated as the BSP. CPU2, CPU3, and so on, are designated as the

APs.

BSP

AP1

AP2

CPU 1

CPU 2

CPU 3

LOCAL

APIC

LOCAL

APIC

LOCAL

APIC

ICC BUS

INTERRUPT

REQUESTS

INTERRUPT

REQUESTS

I/O

APIC

I/O

APIC

Figure 2-2. APIC Configuration

2.1.2 Advanced Programmable Interrupt Controller

The Advanced Programmable Interrupt Controller (APIC) is based on a distributed architecture in

which interrupt control functions are distributed between two basic functional units, the local unit

and the I/O unit. The local and I/O units communicate through a bus called the Interrupt Controller

Communications (ICC) bus, as shown in Figure 2-2.

In a multiprocessor system, multiple local and I/O APIC units operate together as a single entity,

communicating with one another over the ICC bus. The APIC units are collectively responsible

for delivering interrupts from interrupt sources to interrupt destinations throughout the

multiprocessor system.

The APICs help achieve the goal of scalability by doing the following:

•

Off-loading interrupt-related traffic from the memory bus, making the memory bus more

available for processor use.

•

Helping processors share the interrupt processing load with other processors.

The APICs help achieve the goal of AT-compatibility by cooperating with 8259A-equivalent PICs

in the system.

Version 1.4

2-3

MultiProcessor Specification

The local APIC units also provide interprocessor interrupts (IPIs), which allow any processor to

interrupt any other processor or set of processors. There are several types of IPIs. Among them,

the INIT IPI and the STARTUP IPI are specifically designed for system startup and shutdown.

Each local APIC has a Local Unit ID Register and each I/O APIC has an I/O Unit ID Register.

The ID serves as a physical name for each APIC unit. It is used by software to specify destination

information for I/O interrupts and interprocessor interrupts, and is also used internally for accessing

the ICC bus.

Due to the distributed architecture, the APIC local and I/O units can be implemented in either a

single chip, such as Intel’s 82489DX interrupt controller, or they can be integrated with other parts

of the system’s components. For example, the local APIC may be integrated with the CPU chip,

such as Intel’s Pentium processors (735\90, 815\100), and the I/O APIC may be integrated with the

I/O chipset, such as Intel’s 82430 PCI-EISA bridge chipset.

2.1.3 System Memory

A system that complies with the MP specification uses the standard AT memory architecture. All

memory is allocated for system memory with the exception of addresses 0A_0000h through

0F_FFFFh and 0FFFE_0000h through 0FFFF_FFFFh, which are reserved for I/O devices and the

BIOS.

Compared to a uniprocessor system, a symmetric multiprocessor system imposes a high demand

for memory bus bandwidth. The demand is proportional to the number of processors on the

memory bus. To reduce memory bus bandwidth limitations, an implementation of this

specification should use a secondary cache that has high-performance features, such as a write-back

update policy and a snooping cache-consistency protocol. A secondary cache can push the

scalability limit upward by reducing bus traffic and increasing bus bandwidth.

While both secondary caches and memory bus controllers are critical components for high

performance in a symmetric multiprocessor system, this specification does not detail their

implementation, requiring only that they be totally software transparent.

2.1.4 I/O Expansion Bus

The MP specification provides a multiprocessor extension to the industry-standard PC/AT

platform. The term “industry-standard PC/AT platform” here refers to the software-visible

components of the PC/AT. The standard does not designate a specific bus architecture. All

industry-standard buses, such as ISA, EISA, MCA, VL, and PCI, can be incorporated in the

design. Systems developers can implement one or more bus types in their designs, depending on

the systems’ I/O performance or capacity requirements.

2-4

Version 1.4

System Overview

2.2 BIOS Overview

A BIOS functions as an insulator between the hardware on one hand, and the operating system and

applications software on the other. A standard uniprocessor BIOS performs the following

functions:

•

Tests system components.

Builds configuration tables to be used by the operating system.

Initializes the processor and the rest of the system to a known state.

Provides run-time device-oriented services.

For a multiprocessor system, the BIOS may perform the following additional functions:

•

Pass configuration information to the operating system that identifies all processors and other

multiprocessing components of the system.

•

Initialize all processors and the rest of the multiprocessing components to a known state.

This specification allows a wide range of capability in the BIOS. At the minimal end of the

capability scale, the system developer can simply insert an MP floating pointer structure in the

standard BIOS. The cost of this level of simplicity in the BIOS, however, is that the system

developer has less flexibility in the design of the hardware. At the maximal end of the BIOS

capability scale might be a BIOS that dynamically configures the system to provide resilience in

the face of component malfunctions.

BIOS developers should read Chapters 3, 4, 5, and Appendix A to understand the tradeoffs

between hardware and BIOS capabilities.

2.3 Operating System Overview

Enabling the creation of shrink-wrapped MP operating systems is one of the principal goals of this

specification. This goal is achieved by allowing a flexible balance between the capabilities of the

hardware and those of the BIOS. The potentially vast variety of hardware configurations is

reduced by the BIOS to a few simple scenarios that can be readily handled by the low-level boot-

up phase of the operating system.

Operating-system developers should read Chapters 3, 4, and 5, and Appendixes A and B to fully

understand the interface between the BIOS and the operating system.

Version 1.4

2-5

Hardware Specification

This section outlines the minimal set of common hardware features necessary for the operating

system to operate on multiple hardware platforms. The MP hardware specification defines how the

components mentioned in Chapter 2 are implemented. Compliance to the specification involves

the following aspects of hardware implementation:

•

System memory configuration

System memory cacheability and shareability

External cache implementation requirements

Control of memory contention (locking)

Ordering of posted memory writes

Multiprocessor interrupt control

Reset support

Interval timer usage

Support for fault-resilient booting

While the bulk of the MP hardware specification pertains to multiprocessor interrupt control, other

areas also require some attention. The following sections take up each of these topics in turn.

3.1 System Memory Configuration

The MP memory specifications are based on the standard PC/AT memory map, which currently

has a physical memory space of four gigabytes, as shown in see Figure 3-1. Physical memory

should begin at 0 and be contiguous. Memory-mapped I/O devices should be mapped at the top of

physical memory. The APIC default base memory addresses defined by this specification are

0FEC0_0000h and 0FEE0_0000h.

Version 1.4

3-1

MultiProcessor Specification

4GB

FFFF_FFFFH

BIOS PROM

LOCAL APIC

I/O APIC

FFFE_0000H

FEF0_0000H

FEE0_0000H

FED0_0000H

FEC0_0000H

MEMORY-MAPPED

I/O SPACE

EXTENDED

MEMORY REGION

1MB

0010_0000H

000F_0000H

000E_0000H

000D_0000H

000C_0000H

SHADOWED BIOS

SHADOWED

EXPANSION BIOS

EXPANSION ROM

ROM EXTENSIONS

VIDEO BUFFER

640K 000A_0000H

SYSTEM-BASED

MEMORY

0000_0000H

PART OF THIS SPECIFICATION

UNSHADED ADDRESS REGIONS ARE FOR REFERENCE ONLY AND SHOULD NOT BE CONSTRUED AS THE SOLE

DEFINITION OF A PC/AT-COMPATIBLE ADDRESS SPACE.

Figure 3-1. System Memory Address Map

3.2 System Memory Cacheability and Shareability

The cacheability and shareability of the physical memory space are defined in Table 3-1. The

address space reserved for the local APIC is used by each processor to access its own local APIC.

The address space reserved for the I/O APIC must be shareable by all processors to permit dynamic

reconfiguration.

3-2

Version 1.4

Hardware Specification

Table 3-1. Memory Cacheability Map

Addresses

Shared by All

Processors? Cacheable? Comment

(in hex)

Size

Description

0000_0000h –

0009_FFFFh

640KB Main memory

Yes

000A_0000h –

000B_FFFFh

128KB Display buffer for

video adapters

Yes

000C_0000h –

000D_FFFFh

128KB ROM BIOS for add-on Yes

cards

Yes

000E_0000h –

000F_FFFFh

128KB System ROM BIOS

Yes

0010_0000h –

0FEBF_FFFFh

Main memory

Yes

Maximum address

depends on total memory

installed in the system.

Not specified.

Memory-mapped I/O

devices

Yes²

Yes

Not

specified

Top unused memory

0FEC0_0000h –

0FECF_FFFFh¹

APIC I/O unit

Refer to the register

description in the APIC

data book.

0FED0_0000h –

0FEDF_FFFFh

Reserved for

memory-mapped I/O

devices

Yes²

Not

specified

0FEE0_0000h-

0FEEF_FFFFh¹

APIC Local Unit

Refer to the register

description in the APIC

data book.

0FEF0_0000h –

0FFFD_FFFFh

Reserved for

memory-mapped I/O

devices

Yes²

Yes

Not

specified

0FFFE_0000h – 128KB Initialization ROM

0FFFF_FFFFh

Not

specified

NOTES:

1. These addresses are part of this specification. The other address regions in this table are shown for reference only, and

should not be construed as the sole definition of a PC/AT-compatible address space format or cache.

2. Any memory-mapped device should be shareable unless the nature of the device is that there is one device per

processor.

Version 1.4

3-3

MultiProcessor Specification

3.3 External Cache Subsystem

Intel-compatible processors support multiprocessing both on the processor bus and on a memory

bus, both with and without secondary cache units. Due to the high bandwidth demands of

multiprocessor systems, external caches are often employed to improve performance. The

existence and implementation details of external caches are not a part of this specification.

However, when external caches are used, they must conform to certain requirements with regard to

the following design issues:

•

Maintaining cache coherency—When one processor accesses data cached in another

processor’s cache, it must not receive incorrect data. If it modifies data, all other processors

that access that data also must not receive stale data. External caches must maintain coherency

among themselves, and with the main memory, internal caches, and other bus master DMA

devices.

•

Cache flushing—The processor can generate special flush and write-back bus cycles that must

be used by external caches in a manner that maintains cache coherency. The actual responses

are implementation-specific and may vary from design to design. A program can initiate

hardware cache flushing by executing a WBINVD instruction. This instruction is only

guaranteed to flush the caches of the local processor. See Appendix B for system-wide

flushing mechanisms. Given that cache coherency is maintained by hardware, there is no need

for software to issue cache flush instructions under normal circumstances.

•

Reliable communication—All processors must be able to communicate with each other in a

way that eliminates interference when more than one processor accesses the same area in

memory simultaneously. The processor uses the LOCK# signal for this purpose. External

caches must ensure that all locked operations are visible to other processors.

Write ordering—In some circumstances, it is important that memory writes be observed

externally in precisely the same order as programmed. External write buffers must maintain

the write ordering of the processor.

3.4 Locking

To protect the integrity of certain critical memory operations, Intel-compatible processors provide

an output signal called LOCK#. For any given memory access, LOCK# is asserted once, but may

remain asserted for as many memory bus cycles as required to complete the memory operation. It

is the responsibility of the system hardware designers to use this signal to control memory accesses

among processors.

A compliant system in multiprocessor mode must guarantee atomicity of locked-aligned memory

operations; however, the implementation is not specified in this specification. A compliant system

must lock at least the area of memory defined by the destination operand. A specific

implementation may lock a broader area—it may even lock the entire bus. Therefore, software

must consider this behavior.

To guarantee AT compatibility, locking of misaligned memory operations over other

AT-compatible buses in the compliant system must be strictly implemented in accordance with the

bus specifications. A compliant system may not be required to support the misaligned memory

3-4

Version 1.4

Hardware Specification

operations over its internal shared memory bus, if it is AT compatible. Operating system and

software developers must ensure that data is aligned if locked access is required, because lock

operations on misaligned data are not guaranteed to work on all platforms.

3.5 Posted Memory Write

When controlling I/O devices, it is important that memory and I/O operations be carried out in the

order programmed. Intel-compatible processors do not buffer I/O writes; thus, strict ordering

among I/O operations is enforced by the processors.

To optimize memory performance, processors and chipsets often implement write buffers and

writeback caches. Intel-compatible processors guarantee processor ordering on all internal cache

and write buffer accesses. However, chipsets must also guarantee processor ordering on all

external memory accesses.

For systems based on the integrated APIC, posting of memory writes may result in spurious

interrupts for memory-mapped I/O devices using level-triggered interrupts. I/O device drivers

must serialize instructions to ensure that the device interrupt clear command reaches the device

before the EOI command reaches the APIC and handles the spurious interrupt in case one occurs.

3.6 Multiprocessor Interrupt Control

In an MP-compliant system, interrupts are controlled through the APIC. The following sections

describe the APIC architecture and the three interrupt modes allowed in an MP-compliant system.

3.6.1 APIC Architecture

The Intel Advanced Programmable Interrupt Controller (APIC) is based on a distributed

architecture. Interrupt control functions are distributed between two basic functional units: the

local unit and the I/O unit. The local and I/O units communicate through a bus called the ICC bus.

The I/O unit senses an interrupt input, addresses it to a local unit, and sends it over the ICC bus.

The local unit that is addressed accepts the message sent by the I/O unit.

In an MP-compliant system, one local APIC per CPU is required. Depending on the total number

of interrupt lines in an MP system, one or more I/O APICs may be used. The bus interrupt line

assignments can be implementation-specific and can be defined by the MP configuration table

described in Chapter 4.

The Intel 82489DX APIC is a “discrete APIC” implementation. The programming interface of the

82489DX APIC units serves as the base of the MP specification. Each APIC has a version register

that contains the version number of a specific APIC implementation. The version register of the

82489DX family has a version number of “0x,” where x is a four-bit hexadecimal number. Version

number “1x” refers to Pentium processors with integrated APICs, such as the Pentium 735\90 and

815\100 processors, and x is a four-bit hexadecimal number.

The integrated APIC maintains the same programming interface as the 82489DX APIC. Table 3-2

describes the features specific to the integrated APIC.

Version 1.4

3-5

MultiProcessor Specification

Table 3-2. APIC Versions

Local APIC Version

APIC Type

82489DX APIC

Integrated APIC, i.e.,

Pentium processors (735\90,

815\100)

STARTUP IPI. See Appendix B.4.2 for details.

Programmable interrupt input polarity

NOTE:

x is a 4-bit hexadecimal number.

To encourage future extendibility and innovation, the Intel APIC architecture definition is limited

to the programming interface of the APIC units. The ICC bus protocol and electrical specifications

are considered implementation-specific. That is, while different versions of APIC implementations

may execute the same binary software, different versions of APIC components may be

implemented with different bus protocols or electrical specifications. Care must be taken when

using different versions of the APIC in a system.

The APIC architecture is designed to be scaleable. The 82489DX APIC has an 8-bit ID register

that can address from one to 255 APIC devices. Furthermore, the Logical Destination register for

the 82489DX APIC supports 32 bits, which can address up to 32 devices. For small system

implementations, the APIC ID register can be reduced to the least significant 4 bits and the Logical

Destination register can be reduced to the most significant 8 bits.

To ensure software compatibility with all versions of APIC implementations, software developers

must follow the following programming guidelines:

1. Assign an 8-bit APIC ID starting from zero.

2. Assign logical destinations starting from the most significant byte of the 32-bit register.

3. Program the APIC spurious vector to hexadecimal “xF,” where x is a 4-bit hexadecimal

number.

The following features are only available in the integrated APIC:

1. The I/O APIC interrupt input signal polarity can be programmable.

2. A new interprocessor interrupt, STARTUP IPI is defined.

In general, the operating system must use the STARTUP IPI to wake up application processors in

systems with integrated APICs, but must use INIT IPI in systems with the 82489DX APIC. Refer

to Appendix B, Section B.4, for application processor startup.

3.6.2 Interrupt Modes

The MP specification defines three different interrupt modes as follows:

1. PIC Mode—effectively bypasses all APIC components and forces the system to operate in

single-processor mode.

2. Virtual Wire Mode—uses an APIC as a virtual wire, but otherwise operates the same as PIC

Mode.

3. Symmetric I/O Mode—enables the system to operate with more than one processor.

3-6

Version 1.4

Hardware Specification

The first two interrupt modes, PIC Mode and Virtual Wire Mode, provide PC/AT-compatibility.

At least one of these modes must be implemented in systems that comply with the MP

specification. In these modes, full DOS compatibility with the uniprocessor PC/AT is provided by

using the APICs in conjunction with standard 8259A-equivalent programmable interrupt

controllers (PICs).

The third mode, Symmetric I/O Mode, must be implemented in addition to either PIC Mode or

Virtual Wire Mode. An MP operating system is booted under either one of the two PC/AT-

compatible modes. Later the operating system switches to Symmetric I/O Mode as it enters

multiprocessor mode.

The interrupt modes are implemented by a combination of hardware and software. The hardware

and programming specifications for each of these modes are further defined in the following

subsections. BIOS programmers should refer to Appendix A, which includes information about

programming the APIC for Virtual Wire Mode. Operating system programmers should refer to

Appendix B, which provides more information about initializing the APIC for Symmetric I/O

Mode.

3.6.2.1 PIC Mode

PIC Mode is software compatible with the PC/AT because it actually employs the same hardware

interrupt configuration. As Figure 3-2 illustrates, the hardware for PIC Mode bypasses the APIC

components by using an interrupt mode configuration register (IMCR). This register controls

whether the interrupt signals that reach the BSP come from the master PIC or from the local APIC.

Before entering Symmetric I/O Mode, either the BIOS or the operating system must switch out of

PIC Mode by changing the IMCR.

Version 1.4

3-7

MultiProcessor Specification

BSP

AP1

AP2

CPU 1

CPU 2

CPU 3

IMCR

NMI INTR

LOCAL

APIC

LOCAL

APIC

LOCAL

APIC

REG.

MARK

LINTIN0

LINTIN1

LINTIN0

LINTIN1

LINTIN0

LINTIN1

LINTIN0

RESET

ICC BUS

NMI

8259A-

EQUIVALENT

PICS

INTR

I/O

APIC

INTERRUPT INPUTS

SHADED AREAS INDICATE UNUSED CIRCUITS. DOTTED LINE SHOWS INTERRUPT PATH.

Figure 3-2. PIC Mode

The IMCR is supported by two read/writable or write-only I/O ports, 22h and 23h, which receive

address and data respectively. To access the IMCR, write a value of 70h to I/O port 22h, which

selects the IMCR. Then write the data to I/O port 23h. The power-on default value is zero, which

connects the NMI and 8259 INTR lines directly to the BSP. Writing a value of 01h forces the

NMI and 8259 INTR signals to pass through the APIC.

The IMCR must be cleared after a system-wide INIT or RESET to enable the PIC Mode as default.

(Refer to Section 3.7 for information on the INIT and RESET signals.)

The IMCR is optional if PIC Mode is not implemented. The IMCRP bit of the MP feature

information bytes (refer to Chapter 4) enables the operating system to detect whether the IMCR is

implemented.

3-8

Version 1.4

Hardware Specification

3.6.2.2 Virtual Wire Mode

Virtual Wire Mode provides a uniprocessor hardware environment capable of booting and running

all DOS shrink-wrapped software.

In Virtual Wire Mode, as shown in Figure 3-3, the 8259A-equivalent PIC fields all interrupts, and

the local APIC of the BSP becomes a virtual wire, which delivers interrupts from the PIC to the

BSP via the local APIC’s local interrupt 0 (LINTIN0). The LINTIN0 pin of the local APIC is

programmed as ExtINT, specifying to the APIC that the PIC is to serve as an external interrupt

controller. Whenever the local APIC finds that a particular interrupt is of type ExtINT, it asserts

the ExtINTA transaction along with the PINT interrupt to the processor. In this case, the I/O APIC

is not used.

BSP

AP1

AP2

CPU 1

CPU 2

CPU 3

NMI INTR

LOCAL

APIC

LOCAL

APIC

LOCAL

APIC

REG.

MARK

LINTIN0

LINTIN1

LINTIN0

LINTIN1

LINTIN0

LINTIN1

LINTIN0

RESET

ICC BUS

NMI

8259A-

EQUIVALENT

PICS

INTR

I/O

APIC

INTERRUPT INPUTS

SHADED AREAS INDICATE UNUSED CIRCUITS. DOTTED LINE SHOWS INTERRUPT PATH.

Figure 3-3. Virtual Wire Mode via Local APIC

Version 1.4

3-9

MultiProcessor Specification

Figure 3-3 shows how Virtual Wire Mode can be implemented through the BSP’s local APIC. It is

also permissible to program the I/O APIC for Virtual Wire Mode, as shown in Figure 3-4. In this

case the interrupt signal passes through both the I/O APIC and the BSP’s local APIC.

BSP

AP1

AP2

CPU 1

CPU 2

CPU 3

NMI INTR

LOCAL

APIC

LOCAL

APIC

LOCAL

APIC

REG.

MARK

LINTIN0

LINTIN1

LINTIN0

LINTIN1

LINTIN0

LINTIN1

LINTIN0

RESET

ICC BUS

NMI

8259A-

EQUIVALENT

INTR

PICS

I/O

APIC

INTERRUPT INPUTS

SHADED AREAS INDICATE UNUSED CIRCUITS. DOTTED LINE SHOWS INTERRUPT PATH.

Figure 3-4. Virtual Wire Mode via I/O APIC

3-10

Version 1.4

Hardware Specification

3.6.2.3 Symmetric I/O Mode

Some MP operating systems operate in Symmetric I/O Mode. This mode requires at least one I/O

APIC to operate. In this mode, I/O interrupts are generated by the I/O APIC. All 8259 interrupt

lines are either masked or work together with the I/O APIC in a mixed mode. See Figure 3-5 for

an overview of Symmetric I/O Mode.

BSP

AP1

AP2

CPU 1

CPU 2

CPU 3

NMI INTR

LOCAL

APIC

LOCAL

APIC

LOCAL

APIC

REG.

MARK

LINTIN0

LINTIN1

LINTIN0

LINTIN1

LINTIN0

LINTIN1

LINTIN0

RESET

ICC BUS

NMI

8259A-

EQUIVALENT

PICS

INTR

I/O

APIC

INTERRUPT INPUTS

SHADED AREAS INDICATE UNUSED CIRCUITS. DOTTED LINE SHOWS INTERRUPT PATH.

Figure 3-5. Symmetric I/O Mode

The APIC I/O unit has general-purpose interrupt inputs that can be individually programmed to

different operating modes. The I/O APIC interrupt line assignments are system implementation

specific. Refer to Chapter 4 for custom implementations and to Chapter 5 for default

configurations.

The hardware must support a mode of operation in which the system can switch easily to

Symmetric I/O mode from PIC or Virtual Wire mode. When the operating system is ready to

switch to MP operation, it writes a 01H to the IMCR register, if that register is implemented, and

enables I/O APIC Redirection Table entries. The hardware must not require any other action on the

part of software to make the transition to Symmetric I/O mode.

Version 1.4

3-11

MultiProcessor Specification

3.6.3 Assignment of System Interrupts to the APIC Local Unit

The APIC local unit has two general-purpose interrupt inputs, which are reserved for system

interrupts. These interrupt inputs can be individually programmed to different operating modes.

Like the I/O APIC interrupt lines, the local APIC interrupt line assignments of a non-PC/AT-

compatible system are system implementation specific. Refer to Chapter 4 for custom

implementations and to Chapter 5 for default configurations.

3.6.4 Floating Point Exception Interrupt

For PC/AT compatibility, the bootstrap processor must support DOS-compatible FPU execution

and exception handling while running in either of the PC/AT-compatible modes. This means that

floating point error signals from the BSP must be routed to the interrupt request 13 signal, IRQ13,

when the system is in PIC or virtual wire mode. While floating point error signals from an

application processor need not be routed to IRQ13, platform designers may choose to connect the

two. For example, connecting the floating point error signal from application processors to IRQ13

can be useful in the case of a platform that supports dynamic choice of BSP during boot.

In symmetric mode, a compliant system supports only on-chip floating point units, with error

signaling via interrupt vector 16. Operating systems must use interrupt vector 16 to manage

floating point exceptions when the system is in symmetric mode.

It is recommended that hardware platforms be designed to block delivery of floating point

exception signals from the processors once the system has switched into symmetric mode to avoid

delivery of superfluous interrupts. If done, such blocking must be implemented in a manner that is

transparent to the operating system. However, the operating system must still be prepared to

handle interrupts generated through assertion of floating point error signals, because on some

platforms these signals may still be routed to IRQ13 even after the switch to symmetric mode.

3.6.5 APIC Memory Mapping

In a compliant system, all APICs must be implemented as memory-mapped I/O devices. APIC

base addresses are at the top of the memory address space. All APIC local units are mapped to the

same addresses, which are not shared. Each processor accesses its local APIC via these memory

addresses. The default base address for the local APICs is 0FEE0_0000h.

Unlike the local APICs, the I/O APICs are mapped to give shared access from all processors,

providing full symmetric I/O access. The default base address for the first I/O APIC is

0FEC0_0000h. Subsequent I/O APIC addresses are assigned in 4K increments. For example, the

second I/O APIC is at 0FEC0_1000h.

Non-default APIC base addresses can be used if the MP configuration table is provided. (Refer to

Chapter 4.) However, the local APIC base address must be aligned on a 4K boundary, and the I/O

APIC base address must be aligned on a 1K boundary.

3-12

Version 1.4

Hardware Specification

3.6.6 APIC Identification

Systems developers must assign APIC local unit IDs and ensure that all are unique. There are two

acceptable ways to assign local APIC IDs, as follows:

•

By hardware. The ID of each APIC local unit is sampled from the appropriate pins at RESET.

By the BIOS. Software can override the APIC IDs assigned by hardware by writing to the

Local Unit ID Register. This is possible only with help from the hardware; for example, using

board ID registers that the software can read to determine which board has the BSP.

Local APIC IDs must be unique, and need not be consecutive.

The MP operating system can use the local APIC ID to determine on which processor it is

executing.

The ID of each I/O APIC unit is set to zero during RESET. It is the responsibility of the operating

system to verify the uniqueness of the I/O APIC ID and to assign a unique ID if a conflict is found.

The assignment of APIC IDs for I/O units must always begin from the lowest number that is

possible after the assignment of local APIC IDs. The operating system must not attempt to change

the ID of an APIC I/O unit if the preset ID number is acceptable.

3.6.7 APIC Interval Timers

The 82489DX APIC local unit contains a 32-bit wide programmable timer with the following two

independent clock input sources:

1. The CLK pin provides the clock signal that drives the 82489DX APIC’s internal operation.

2. The TMBASE pin allows an independent clock signal to be connected to the 82489DX APIC

for use by the timer functions.

The interval timers of the integrated APIC have only one clock input source, CLK. To maintain

consistency, developers of compliant systems based on the 82489DX must choose CLK as the

source of the 82489DX APIC timer clock. TMBASE must be left disabled. An MP operating

system may use the IRQ8 real-time clock as a reference to determine the actual APIC timer clock

speed.

Special consideration must be made for systems with stop clock (STPCLK#) capability. Timer

interrupts are ignored while STPCLK# is asserted. The system time-of-day clock may need to be

reset when STPCLK# is deasserted.

3.6.8 Multiple I/O APIC Configurations

Systems may provide more than one I/O APIC for increased interrupt scalability in Symmetric I/O

Mode. To preserve PC/AT compatibility in PIC or Virtual Wire mode, interrupts connected to

additional I/O APICs must also be connected to the 8259A programmable interrupt controller.

Figure 3-6 represents one possible interrupt connection scheme for a system with two I/O APICs.

The non-ISA interrupts are connected to both the 8259A IRQ inputs and the inputs of the

associated I/O APIC. To prevent non-ISA interrupts from appearing at inputs on both I/O APICs,

the hardware must provide a means to disable the interrupt routing network when the system

switches to symmetric I/O mode with the second I/O APIC enabled.

More information about the implementation of multiple I/O APIC configurations is presented in

Appendix D.

Version 1.4

3-13

MultiProcessor Specification

REG.

MARK

INTR/LINT0

ICC BUS

I/O APIC

NON-ISA INTERRUPT

INTERRUPT

ROUTING

NETWORK

I/O APIC

ISA INTERRUPT

8259A-

EQUIVALENT

PICS

Figure 3-6. Multiple I/O APIC Configurations

3.7 RESET Support

To bring all circuitry in a computer system to an initial state, computer systems require a system-

wide reset capability. To support multiple processors, a compliant system requires a processor-

specific reset or initialization capability in addition to the typical system-wide reset and

initialization capabilities.

•

The term “RESET” refers to the system-wide hard reset. It refers to the RESET signal on both

Pentium and Intel486 processors or the RESET signal of the 82489DX APIC. (See

Section 3.7.1.)

•

The term “INIT,” refers to either a system-wide soft reset/initialization or a processor-specific

initialization. For example, the term “INIT” may refer to the INIT signal on the Pentium

processor or to the RESET signal on the Intel486 processor. (See Sections 3.7.2. and 3.7.3.)

Because the INIT signal is available on the Pentium processor but not on the Intel486 processor,

the remainder of this document uses the above-mentioned special definitions for the terms “INIT”

and “RESET”:

3.7.1 System-wide RESET

The system-wide RESET, as defined by this specification, refers to a hard or cold reset, which

sets all circuitry, including processor, coprocessor, add-in cards, and control logic, to an initial

state. A hard reset is the reset signal that is sent to all components of the system during a power-on

3-14

Version 1.4

Hardware Specification

or by the front panel reset button (if the system is so equipped). This type of reset operates without

regard to cycle boundaries, and, for example, is connected to the RESET pin of Pentium

processors.

3.7.2 System-wide INIT

The system-wide INIT, as defined by this specification, refers to a soft or warm reset that

initializes only portions of the processor. This type of reset initializes the microprocessor in such a

way that the reset does not corrupt any pending cycles, but waits instead for a cycle boundary, and

does not invalidate the contents of caches and floating point registers. This type of reset request is

connected to the INIT signal of newer processors, such as the Pentium processors. On Intel486

processors, the RESET pin is used for this function, as well as for hard resets, but the RESET pin

does not provide the advantages of the INIT pin. There are many possible ways to assert a soft

reset, including:

•

A write either to a port of the 8042 Keyboard Controller or to some other port provided for the

same purpose by a chipset.

•

A shutdown special bus cycle. Usually a chipset senses a shutdown cycle and asserts a soft

reset to the processor.

In a compliant system, the standard PC/AT-platform resets mentioned above, both hard and soft,

must be directed to all processors in the system, except in the case of fault-tolerant MP systems, in

which a soft reset may be handled on a per-processor basis.

3.7.3 Processor-specific INIT

A processor-specific INIT is one of the basic multiprocessor support functions of a compliant

multiprocessor system, along with processor startup and shutdown. With it, the BSP can

selectively initialize an AP for subsequent startup or recover an AP from a fatal system error. This

type of INIT function is exclusively used by the MP operating system or BIOS self-test routine.

The system must be designed so that the processor-specific INIT can be initiated by software

programming; it is not necessary that it be initiated by hardware.

A compliant system supports the processor-specific INIT via a special interprocessor interrupt (IPI)

mechanism called INIT IPI. For the 82489DX APIC, INIT IPI is an IPI that has the delivery mode

RESET, which delivers the signal to all processors listed in the destination by asserting/deasserting

the addressed APIC local unit’s PRST output pin. When the PRST signal is connected to the INIT

pin of the Pentium processor or to the RESET pin of the Intel486 processor, the INIT IPI forces the

processor to begin executing at the reset vector.

For systems based on the Intel486 processor, the 82489DX APIC’s PRST line must be the only

line connected to the processor’s RESET input, so that the INIT IPI resets the targeted processor

only. The system reset signal is connected to the local 82489DX APIC’s RESET input. Assertion

of the system reset signal then causes all of the local 82489DX APICs to assert their PRST outputs,

thereby resetting all the processors.

For integrated APIC versions of the Pentium processor, INIT IPI asserts and deasserts the internal

INIT signal of the Pentium processor.

Version 1.4

3-15

MultiProcessor Specification

3.8 System Initial State

The system initial state is the state before the BIOS gives control to the operating system. It is

identical to the system initial state of a typical PC/AT system, with the additional MP components

in the following state:

1. All local APICs are disabled, except for the local APIC of the BSP if the system starts in

Virtual Wire Mode.

2. All pending I/O APIC interrupts are cleared and disabled.

3. If IMCR is present, it should be set to 0 or 1 depending on the interrupt mode chosen for

startup.

4. All APs are in Real Mode.

5. All APs are in HALT state or off the system bus.

The BIOS must disable interrupts to all processors and set the APICs to the system initial state

before giving control to the operating system. The operating system is responsible for initializing

all of the APICs.

3.9 Support for Fault-resilient Booting

OEMs may choose not to implement a fault-resilient booting capability in their systems. However,

if such a capability is provided, systems developers must observe the following guidelines:

•

BSP determination may be performed by special hardware or by the BIOS, but it must be

totally transparent to the operating system.

•

NMI and INTR must be connected to the BSP.

FERR# and IGNNE# signals from the designated BSP must be used to support IRQ13.

The A20M# signal from the designated BSP must be used to support the masking of physical

address bit 20 on the BSP (to support DOS compatibility).

3-16

Version 1.4

MP Configuration Table

The operating system must have access to some information about the multiprocessor

configuration. The MP specification provides two methods for passing this information to the

operating system: a minimal method for configurations that conform to one of a set of common

hardware defaults, and a maximal method that provides the utmost flexibility in hardware design.

Figure 4-1 shows the general layout of the MP configuration data structures.

24 23

16 15

VARIABLE NUMBER OF

VARIABLE-LENGTH

ENTRIES

ENTRY LENGTH DEPENDS ON ENTRY TYPE

ENTRY TYPE

CONFIGURATION

TABLE

02CH

FIXED-LENGTH

HEADER

00H

24 23

16 15

FLOATING

POINTER

STRUCTURE

PHYSICAL ADDRESS POINTER

Figure 4-1. MP Configuration Data Structures

Version 1.4

4-1

MultiProcessor Specification

The following two data structures are used:

1. The MP Floating Pointer Structure. This structure contains a physical address pointer to the

MP configuration table and other MP feature information bytes. When present, this structure

indicates that the system conforms to the MP specification. This structure must be stored in at

least one of the following memory locations, because the operating system searches for the MP

floating pointer structure in the order described below:

a. In the first kilobyte of Extended BIOS Data Area (EBDA), or

b. Within the last kilobyte of system base memory (e.g., 639K-640K for systems with 640

KB of base memory or 511K-512K for systems with 512 KB of base memory) if the

EBDA segment is undefined, or

c. In the BIOS ROM address space between 0F0000h and 0FFFFFh.

2. The MP Configuration Table. This table is optional. The table is composed of a base section

and an extended section. The base section contains entries that are completely backwards

compatible with previous versions of this specification. The extended section contains

additional entry types. The MP configuration table contains explicit configuration information

about APICs, processors, buses, and interrupts. The table consists of a header, followed by a

number of entries of various types. The format and length of each entry depends on its type.

When present, this configuration table must be stored either in a non-reported system RAM or

within the BIOS read-only memory space.

An MP configuration table is not required if the system design corresponds to one of the default

configurations listed in Chapter 5. Note that these defaults are only for designs that are always

equipped with two processors. Systems that support a variable number of installed processors

must supply a complete MP configuration table that indicates the correct number of installed

processors. For example, systems that ship with only one processor but provide for a later upgrade

with a second processor may not use a default MP configuration.

The following is a list of the suggested memory spaces for the MP configuration table:

a. In the first kilobyte of Extended BIOS Data Area (EBDA), or

b. Within the last kilobyte of system base memory if the EBDA segment is undefined, or

c. At the top of system physical memory, or

d. In the BIOS read-only memory space between 0E0000h and 0FFFFFh.

The BIOS reports the base memory size in a two-byte location (40:13h) of the BIOS data area.

The base memory size is reported in kilobytes minus 1K, which is used by the EBDA segment or

for other purposes.

The exact starting address of the EBDA segment for EISA or MCA systems can be found in a two-

byte location (40:0Eh) of the BIOS data area. If system memory is used, the BIOS must not report

this memory as part of the available memory space.

These two MP configuration data structures can be located in the ROM space only if the system is

not dynamically reconfigurable. The MP configuration information is intended to be read-only to

the operating system.

Strings in the configuration tables are coded in ASCII. The first character of the string is stored at

the lowest address of the string field. If the string is shorter than its field, the extra character

locations are filled with space characters. Strings are not null terminated.

4-2

Version 1.4

MP Configuration Table

4.1 MP Floating Pointer Structure

An MP-compliant system must implement the MP floating pointer structure, which is a variable

length data structure in multiples of 16 bytes. Currently, only one 16-byte data structure is defined.

It must span a minimum of 16 contiguous bytes, beginning on a 16-byte boundary, and it must be

located within the physical address as specified in the previous section. To determine whether the

system conforms to the MP specification, the operating system must search for the MP floating

pointer structure in the order specified in the previous section. Figure 4-2 shows the format of this

structure, and Table 4-1 explains each of the fields. See also Appendix E, for more information.

24 23

16 15

MP FEATURE

BYTES 2-5

0CH

MP FEATURE

BYTE 1

CHECKSUM

SPEC_REV

LENGTH

08H

PHYSICAL ADDRESS POINTER

SIGNATURE

04H

00H

_ (5Fh)

P (50h)

M (4Dh)

_ (5Fh)

24 23

16 15

Figure 4-2. MP Floating Pointer Structure

Table 4-1. MP Floating Pointer Structure Fields

Offset Length

Field

(in bytes:bits) (in bits)

Description

SIGNATURE

The ASCII string represented by “_MP_” which

serves as a search key for locating the pointer

structure.

PHYSICAL ADDRESS

POINTER

The address of the beginning of the MP

configuration table. All zeros if the MP

configuration table does not exist.

LENGTH

The length of the floating pointer structure table

in paragraph (16-byte) units. The structure is

16 bytes or 1 paragraph long; so this field

contains 01h.

SPEC_REV

CHECKSUM

The version number of the MP specification

supported. A value of 01h indicates Version 1.1.

A value of 04h indicates Version 1.4.

A checksum of the complete pointer structure.

All bytes specified by the length field, including

CHECKSUM and reserved bytes, must add up to

zero.

Version 1.4

4-3

MultiProcessor Specification

Table 4-1. MP Floating Pointer Structure Fields (continued)

Offset

Length

Field

(in bytes:bits) (in bits)

Description

Bits 0-7:

MP System Configuration Type.

MP FEATURE

INFORMATION BYTE 1

When these bits are all zeros, the MP

configuration table is present. When nonzero,

the value indicates which default configuration

(as defined in Chapter 5) is implemented by the

system.

Bits 0-6:

MP FEATURE

INFORMATION BYTE 2

12:0

12:7

Reserved for future MP definitions.

Bit 7:

IMCRP. When the IMCR presence bit is

set, the IMCR is present and PIC Mode is

implemented; otherwise, Virtual Wire Mode is

implemented.

MP FEATURE

INFORMATION BYTES 3-5

Reserved for future MP definitions. Must be

zero.

The MP feature information byte 1 specifies the MP system default configuration type. If nonzero,

the system configuration conforms to one of the default configurations. The default configurations,

specified in Chapter 5, may only be used to describe systems that always have two processors

installed.

Bit 7 of MP feature information byte 2, the IMCR present bit, is used by the operating system to

determine whether PIC Mode or Virtual Wire Mode is implemented by the system.

The physical address pointer field contains the address of the beginning of the MP configuration

table. If it is nonzero, the MP configuration table can be accessed at the physical address provided

in the pointer structure. This field must be all zeros if the MP configuration table does not exist.

4-4

Version 1.4

MP Configuration Table

4.2 MP Configuration Table Header

Figure 4-3 shows the format of the header of the MP configuration table, and Table 4-2 explains

each of the fields.

24 23

16 15

EXTENDED

TABLE

RESERVED

EXTENDED TABLE LENGTH

28H

24H

20H

1CH

CHECKSUM

MEMORY-MAPPED ADDRESS OF LOCAL APIC

OEM TABLE SIZE

ENTRY COUNT

OEM TABLE POINTER

space

PRODUCT ID STRING

space

10H

space

OEM ID STRING

space

08H

04H

00H

CHECKSUM

SPEC_REV

BASE TABLE LENGTH

SIGNATURE

M (4Dh) C (43h)

P (50h)

24 23

16 15

Figure 4-3. MP Configuration Table Header

Version 1.4

4-5

MultiProcessor Specification

Table 4-2. MP Configuration Table Header Fields

Offset

Length

Field

(in bytes) (in bits) Description

SIGNATURE

The ASCII string representation of “PCMP,” which

confirms the presence of the table.

BASE TABLE LENGTH

SPEC_REV

The length of the base configuration table in bytes,

including the header, starting from offset 0. This field

aids in calculation of the checksum.

The version number of the MP specification. A value

of 01h indicates Version 1.1. A value of 04h indicates

Version 1.4.

CHECKSUM

OEM ID

A checksum of the entire base configuration table. All

bytes, including CHECKSUM and reserved bytes,

must add up to zero.

A string that identifies the manufacturer of the system

hardware.

PRODUCT ID

A string that identifies the product family.

OEM TABLE POINTER

A physical-address pointer to an OEM-defined

configuration table. This table is optional. If not

present, this field is zero.

OEM TABLE SIZE

ENTRY COUNT

The size of the base OEM table in bytes. If the table

is not present, this field is zero.

The number of entries in the variable portion of the

base table. This field helps the software identify the

end of the table when stepping through the entries.

ADDRESS OF LOCAL APIC 36

The base address by which each processor accesses

its local APIC.

EXTENDED TABLE

LENGTH

Length in bytes of the extended entries that are

located in memory at the end of the base

configuration table. A zero value in this field indicates

that no extended entries are present.

EXTENDED TABLE

CHECKSUM

Checksum for the set of extended table entries,

including only extended entries starting from the end

of the base configuration table. When no extended

table is present, this field is zero.

4.3 Base MP Configuration Table Entries

A variable number of variable length entries follow the header of the MP configuration table. The

first byte of each entry identifies the entry type. Each entry type has a known, fixed length. The

total length of the MP configuration table depends upon the configuration of the system. Software

must step through each entry in the base table until it reaches ENTRY COUNT. The entries are

sorted on ENTRY TYPE in ascending order. Table 4-3 gives the meaning of each value of

ENTRY TYPE.

4-6

Version 1.4

MP Configuration Table

Table 4-3. Base MP Configuration Table Entry Types

Length

Entry Description

Entry Type Code* (in bytes)

Comments

Processor

One entry per processor.

One entry per bus.

Bus

I/O APIC

One entry per I/O APIC.

One entry per bus interrupt source.

I/O Interrupt Assignment

Local Interrupt Assignment

One entry per system interrupt

source.

* All other type codes are reserved.

4.3.1 Processor Entries

Figure 4-4 shows the format of each processor entry, and Table 4-4 defines the fields.

28 27

24 23

20 19

16 15

12 11

RESERVED

10H

0CH

08H

04H

00H

RESERVED

FEATURE FLAGS

CPU SIGNATURE

CPU FLAGS

LOCAL APIC

VERSION #

ENTRY TYPE

LOCAL APIC ID

B E

RESERVED

P N

24 23

28 27

20 19

16 15

12 11

Figure 4-4. Processor Entry

In systems that use the MP configuration table, the only restriction placed on the assignment of

APIC IDs is that they be unique. They do not need to be consecutive. For example, it is possible

for only APIC IDs 0, 2, and 4 to be present.

Version 1.4

4-7

MultiProcessor Specification

Table 4-4. Processor Entry Fields

Offset

Length

Field

(in bytes:bits) (in bits) Description

ENTRY TYPE

LOCAL APIC ID

A value of 0 identifies a processor entry.

The local APIC ID number for the particular

processor.

LOCAL APIC VERSION #

CPU FLAGS: EN

Bits 0–7 of the local APIC’s version register.

3:0

If zero, this processor is unusable, and the

operating system should not attempt to access this

processor.

CPU FLAGS: BP

3:1

Set if specified processor is the bootstrap

processor.

CPU SIGNATURE:

STEPPING

4:0

4:4

5:0

Refer to Table 4-5 for values.

MODEL

FAMILY

FEATURE FLAGS

The feature definition flags for the processor as

returned by the CPUID instruction. Refer to

Table 4-6 for values. If the processor does not

have a CPUID instruction, the BIOS must assign

values to FEATURE FLAGS according to the

features that it detects.

The configuration table is filled in by the BIOS after it executes a CPU identification procedure on

each of the processors in the system. Whenever possible, the complete 32-bit CPU signature

should be filled with the values returned by the CPUID instruction. The CPU signature includes

but is not limited to, the stepping, model, and family fields. If the processor does not have a

CPUID instruction, the BIOS must fill these and future reserved fields with information returned

by the processor in the EDX register after a processor reset. See the Pentium Processor User’s

Manual and Intel Processor Identification with the CPUID Instruction (AP-485) for details on the

CPUID instruction.

4-8

Version 1.4

MP Configuration Table

Table 4-5. Intel486 and Pentium Processor Signatures

Stepping^a

Family

0000

0100

0101

Model

0000

Description

0000

xxxx

Not a valid CPU signature.

Intel486 DX Processor

Intel486 SX Processor

Intel487 Processor

0000 and 0001

0010

0011

IntelDX2™ Processor

0100

Intel486 SL Processor

IntelSX2™ Processor

0101

1000

IntelDX4™ Processor

0001

Pentium Processors (510\60, 567\66)

Pentium Processors (735\90, 815\100)

0010

Values not shown are reserved for future processors.

Refer to the documentation of each new processor for its family and model values.

1111

Not a valid CPU signature. Indicates a processor that is not

an Intel architecture-compatible processor (a graphics

controller, for example)

Intel releases information about stepping numbers as needed.

Table 4-6. Feature Flags from CPUID Instruction

Bit

Name

Description

Comments

FPU

On-chip Floating The processor contains an FPU that supports the Intel387

Point Unit

processor floating point instruction set.

Reserved.

1–6

MCE

Machine Check

Exception

Exception 18 is defined for machine checks, including CR4.MCE for

controlling the feature. This feature does not define the model-

specific implementations of machine-check error logging reporting

and processor shutdowns. Machine-check exception handlers may

have to depend on processor version to do model-specific

processing of the exception or test for the presence of the standard

machine-check feature.

CX8

CMPXCHG8B

Instruction

The 8 byte (64 bits) compare-and-exchange instruction is supported

(implicitly locked and atomic). Introduced by the Pentium

processor.

APIC

On-chip APIC

Indicates that an integrated APIC is present and hardware enabled.

(Software disabling does not affect this bit.)

10–31

Reserved.

Version 1.4

4-9

MultiProcessor Specification

4.3.2 Bus Entries

Bus entries identify the kinds of buses in the system. Because there may be more than one bus in a

system, each bus is assigned a unique bus ID number by the BIOS. The bus ID number is used by

the operating system to associate interrupt lines with specific buses. Figure 4-5 shows the format

of a bus entry, and Table 4-7 explains the fields of each entry.

24 23

16 15

BUS TYPE STRING

04H

00H

ENTRY TYPE

BUS ID

24 23

16 15

Figure 4-5. Bus Entry

Table 4-7. Bus Entry Fields

Offset

Length

Field

(in bytes) (in bits)

Description

Entry type 1 identifies a bus entry.

ENTRY TYPE

BUS ID

An integer that identifies the bus entry. The BIOS assigns

identifiers sequentially, starting at zero.

BUS TYPE

STRING

A string that identifies the type of bus. Refer to Table 4-8 for

values. These are 6-character ASCII (blank-filled) strings

used by the MP specification.

4-10

Version 1.4

MP Configuration Table

Table 4-8. Bus Type String Values

Bus Type String

CBUS

CBUSII

EISA

Description

Corollary CBus

Corollary CBUS II

Extended ISA

FUTURE

INTERN

ISA

IEEE FutureBus

Internal bus

Industry Standard Architecture

Multibus I

MBI

MBII

Multibus II

MCA

Micro Channel Architecture

MPI

MPSA

NUBUS

PCI

MPSA

Apple Macintosh NuBus

Peripheral Component Interconnect

PC Memory Card International Assoc.

DEC TurboChannel

VESA Local Bus

VMEbus

PCMCIA

VME

XPRESS

Express System Bus

Each bus in a system must have a unique BUS ID if any one of the following criteria are true:

1. The bus does not share its memory address space with another bus.

2. The bus does not share its I/O address space with another bus.

3. The bus does not share interrupt lines with another bus.

4. Any aspect of the bus as an independent entity is software visible (such as PCI configuration

space).

Special consideration must be given when assigning a BUS ID for local buses such as VL, which

are designed to work in conjunction with another bus. If the bus looks like a part of another bus

because it uses a subset of that bus's interrupts and address space, rendering it totally invisible to

software, it does not need its own bus entry in the table. The two buses are then considered a single

logical bus.

Version 1.4

4-11

MultiProcessor Specification

4.3.3 I/O APIC Entries

The configuration table contains one or more entries for I/O APICs. Figure 4-6 shows the format

of each I/O APIC entry, and Table 4-9 explains each field.

24 23

16 15

04H

00H

MEMORY-MAPPED ADDRESS OF I/O APIC

I/O APIC

I/O APIC FLAGS

ENTRY TYPE

I/O APIC ID

VERSION #

RESERVED

24 23

16 15

Figure 4-6. I/O APIC Entry

Table 4-9. I/O APIC Entry Fields

Offset

Length

Field

(in bytes:bits) (in bits)

Description

ENTRY TYPE

A value of 2 identifies an I/O APIC entry.

The ID of this I/O APIC.

I/O APIC ID

I/O APIC VERSION #

I/O APIC FLAGS: EN

Bits 0–7 of the I/O APIC’s version register.

3:0

If zero, this I/O APIC is unusable, and the

operating system should not attempt to access

this I/O APIC.

At least one I/O APIC must be enabled.

I/O APIC ADDRESS

Base address for this I/O APIC.

4.3.4 I/O Interrupt Assignment Entries

These entries indicate which interrupt source is connected to each I/O APIC interrupt input. There

is one entry for each I/O APIC interrupt input that is connected. Figure 4-7 shows the format of

each entry, and Table 4-10 explains each field. Appendix D provides the semantics for encoding

PCI interrupts.

The MP specification enables significantly more interrupt sources than the standard AT

architecture by using I/O APICs. When using I/O APICs, it is preferable that the buses do not

share interrupts with the other buses. Bus implementations that share interrupts, such as the PCI

and VL local buses, support their bus interrupts by overloading them into another bus space. These

buses can be supported in one of the following two ways:

1. Interrupt Assignment Entries for each of the bus interrupts are listed in the MP configuration

table. Each interrupt destination matches the destination of another interrupt source interrupt

that this interrupt shares. For example, if PCI-Device1/INTA# has the same vector as

ISA-IRQ2, then both Interrupt Assignment Entries for these vectors would refer to the same

destination I/O APIC and INTIN#.

4-12

Version 1.4

MP Configuration Table

2. No Interrupt Assignment Entries are declared for any of the bus source interrupts, and the

operating system uses some other bus-specific knowledge of bus interrupt schemes in order to

support the bus. This operating system bus-specific knowledge is beyond the scope of this

specification.

24 23

16 15

DESTINATION

I/O APIC INTIN#

DESTINATION

I/O APIC ID

SOURCE BUS

IRQ

SOURCE

BUS ID

04H

00H

I/O INTERRUPT FLAG

INTERRUPT

TYPE

ENTRY TYPE

RESERVED

24 23

16 15

Figure 4-7. I/O Interrupt Entry

Version 1.4

4-13

MultiProcessor Specification

Table 4-10. I/O Interrupt Entry Fields

Offset

Length

Field

(in bytes:bits) (in bits)

Description

ENTRY TYPE

Entry type 3 identifies an I/O interrupt

entry.

INTERRUPT TYPE

See Table 4-11 for values.

2:0

Polarity of APIC I/O input signals:

00 =

Conforms to

specifications of bus (for

example, EISA is active-

low for level-triggered

interrupts)

01 =

10 =

11 =

Active high

Reserved

Active low

Must be 00 if the 82489DX is used.

2:2

Trigger mode of APIC I/O input signals:

00 =

Conforms to

specifications of bus (for

example, ISA is edge-

triggered)

01 =

10 =

11 =

Edge-triggered

Reserved

Level-triggered

SOURCE BUS ID

SOURCE BUS IRQ

Identifies the bus from which the interrupt

signal comes.

Identifies the interrupt signal from the

source bus. Values are mapped onto

source bus signals, starting from zero. A

value of 0, for example, would indicate

IRQ0 of an ISA bus. See Section D.3 for

PCI bus semantics.

DESTINATION I/O APIC ID

Identifies the I/O APIC to which the signal

is connected. If the ID is 0FFh, the signal

is connected to all I/O APICs.

DESTINATION I/O APIC INTIN#

Identifies the INTINn pin to which the

signal is connected.

4-14

Version 1.4

MP Configuration Table

Table 4-11. Interrupt Type Values

Interrupt Type*

Description

Comments

INT

Signal is a vectored interrupt; vector is supplied by APIC

redirection table.

NMI

Signal is a nonmaskable interrupt.

SMI

Signal is a system management interrupt.

ExtINT

Signal is a vectored interrupt; vector is supplied by external PIC.

For example, if an 8259 is used as the external PIC, the source is

the 8259 INTR output line, and the vector is supplied by the 8259.

* All other values are reserved.

4.3.5 Local Interrupt Assignment Entries

These configuration table entries tell what interrupt source is connected to each local interrupt

input of each local APIC. Figure 4-8 shows the format of each entry, and Table 4-12 explains each

field.

24 23

16 15

DESTINATION

LOCAL APIC

LINTIN#

DESTINATION

LOCAL APIC ID

SOURCE BUS

IRQ

SOURCE

BUS ID

04H

00H

LOCAL INTERRUPT FLAG

INTERRUPT

TYPE

ENTRY TYPE

RESERVED

24 23

16 15

Figure 4-8. Local Interrupt Entry

Version 1.4

4-15

MultiProcessor Specification

Table 4-12. Local Interrupt Entry Fields

Offset (in

Length

Field

bytes:bits)

(in bits) Description

ENTRY TYPE

Entry type 4 identifies a local interrupt

entry.

INTERRUPT TYPE

See Table 4-11 for values

Polarity of APIC local input signals:

2:0

00 =

Conforms to

specifications of bus

(for example, EISA is

active-low for level

triggered interrupts)

Active high

01 =

10 =

11 =

Reserved

Active low

Must be 00 if the 82489DX is used.

2:2

Trigger mode of APIC local input signals:

00 =

Conforms to

specifications of bus

(for example, ISA is

edge triggered)

Edge-triggered

Reserved

01 =

10 =

11 =

Level-triggered

SOURCE BUS ID

SOURCE BUS IRQ

Identifies the bus from which the interrupt

signal came.

Identifies the interrupt signal from the

source bus. Values are mapped onto

source bus signals, starting from zero. A

value of 0, for example, would indicate

IRQ0 of an ISA bus. See Section D.3 for

PCI bus semantics.

DESTINATION LOCAL APIC ID

Identifies the local APIC to which the

signal is connected. If the ID is 0FFh, the

signal is connected to all local APICs.

DESTINATION LOCAL APIC

LINTIN#

Identifies the LINTINn pin to which the

signal is connected, where n = 0 or 1.

4-16

Version 1.4

MP Configuration Table

4.4 Extended MP Configuration Table Entries

A variable number of variable-length entries are located in memory, immediately following entries

in the base section of the MP configuration table described in Section 4.3. These entries compose

the extended section of the MP configuration table. Each entry in the extended section of the table

has three elements:

•

ENTRY TYPE

ENTRY LENGTH

DATA

ENTRY TYPE and ENTRY LENGTH make up a header that is present in all extended

configuration table entries. The ENTRY TYPE and ENTRY LENGTH fields are eight bit

unsigned values. ENTRY LENGTH specifies the total length of the given configuration entry. The

length of the DATA field is two bytes less than the value specified by ENTRY LENGTH.

This scheme ensures that operating systems can parse all the entries in the extended MP

configuration table area, even if an entry type is unrecognized. Operating systems that find an

unknown entry type when parsing this section of the table should ignore the content and move on

to the next entry, until the offset from the end of the base table reaches the length that is specified

in the EXTENDED TABLE LENGTH field of the configuration table header. The ability to skip

entries with unrecognized type codes beyond those listed in Table 4-13 is essential since it is

anticipated that more types of entries will be added to this list over time.

The total length of the Extended MP configuration table depends upon the configuration of the

system. The entries are sorted on ENTRY TYPE in ascending order. Table 4-13 gives the meaning

of each value of ENTRY TYPE.

Table 4-13. Extended MP Configuration Table Entry Types

Entry Type Length

Entry Description

Code*

(in bytes) Comments

SYSTEM ADDRESS SPACE

MAPPING

128

Entry to declare system visible memory

or I/O space on a bus.

BUS HIERARCHY DESCRIPTOR

129

130

Entry to describe I/O bus interconnection.

COMPATIBILITY BUS ADDRESS

SPACE MODIFIER

Entry to describe predefined address

ranges that modify the memory or I/O

space visible on a bus in order to support

ISA compatibility.

*All other type codes are reserved.

Version 1.4

4-17

MultiProcessor Specification

4.4.1 System Address Space Mapping Entries

System Address Space Mapping entries define the system addresses that are visible on a particular

bus. Each bus defined in the Base Table can have any number of System Address Space Mapping

entries included in the Extended Table. Thus, individual buses can be configured to support

different address ranges, thereby decreasing the amount of bus traffic on a given bus and increasing

the overall system performance. Each consecutive address range that is usable by the operating

system to access devices on a given bus has a System Address Space Mapping entry. Figure 4-9

shows the format of each entry, and Table 4-14 explains each field. See also Appendix E, for more

information.

28 27

24 23

20 19

16 15

12 11

ADDRESS LENGTH

10H

0CH

08H

04H

00H

ADDRESS BASE

ENTRY TYPE

128

ADDRESS TYPE

BUS ID

ENTRY LENGTH

28 27

24 23

20 19

16 15

12 11

Figure 4-9. System Address Space Entry

4-18

Version 1.4

MP Configuration Table

Table 4-14. System Address Space Mapping Entry Fields

Offset

(in

Length

(in bits)

Field

Description

bytes:bits)

ENTRY TYPE

ENTRY LENGTH

BUS ID

Entry type 128 identifies a System Address

Space Mapping Entry.

A value of 20 indicates that an entry of this type

is twenty bytes long.

The BUS ID for the bus where the system

address space is mapped. This number

corresponds to the BUS ID as defined in the

base table bus entry for this bus.

ADDRESS TYPE

System address type used to access bus

addresses must be:

0 = I/O address

1 = Memory address

2 = Prefetch address

All other numbers are reserved.

Starting address

ADDRESS BASE

LENGTH

Number of addresses which are visible to the

bus

If any main memory address is mapped to a software visible bus, such as PCI, it must be explicitly

declared using a System Address Space Mapping entry.

In the case of a bus that is directly connected to the main system bus, system address space records

and compatibility base address modifiers must be provided as needed to fully describe the complete

set of addresses that are mapped to that bus. For example, in Figure 4-10, complete explicit

descriptions must be provided for PCI BUS 0 and PCI BUS 1 even if one of the buses is

programmed for subtractive decode.

Processor 0

Processor 1

Memory

Controller

System Bus

PCI

Host Bridge

PCI

Host Bridge

PCI Bus 0

PCI Bus 1

PCI-to-PCI

Bridge

EISA Bridge

EISA Bus

PCI Bus 2

Version 1.4

4-19

MultiProcessor Specification

Figure 4-10. Example System with Multiple Bus Types and Bridge Types

Since all device settings must fall within supported System Address Space mapping for a given bus

in order to be usable by the operating system, buses that do not support dynamically configurable

devices (i.e., ISA, EISA) should support all possible addresses to that bus.

In general, the MP configuration table must provide entries to describe system address space

mappings for all I/O buses present in the system. There are two exceptions to this rule:

1. For buses that are connected via PCI-to-PCI-bridge-specification-compliant bridges: in this

case, the system address space mappings for such buses may be omitted from the configuration

table. In such cases, the operating system is expected to discover the address space mapping by

querying the PCI-to-PCI-bridge-specification-compliant bridge directly.

2. For buses that are connected via a parent I/O bus and for which the subtractive decode bit is set

(refer to Section 4.4.2 for details).

Typically, this would mean that a minimal description of resources allocated to PCI buses in a

system need only include System Address Space Mapping entries for PCI bus zero and any

additional peer PCI buses (if present) where these buses are connected by PCI bridges that are

specific to the chipset or host bus.

Note

System Address Mappings are unidirectional in nature. They only describe system addresses that

propagate to the target bus from any given processor. For DMA support on the target bus, all memory

addresses that contain real memory should be accessible directly by either the bus or by a bus-specific

DMA controller. For buses with fewer than 32-bit address lines, all real memory at addresses that the

bus can generate must be accessible for DMA.

4-20

Version 1.4

MP Configuration Table

4.4.2 Bus Hierarchy Descriptor Entry

If present, Bus Hierarchy Descriptor entries define how I/O buses are connected relative to each

other in a system with more than one I/O bus. Bus Hierarchy Descriptors are used to supplement

System Address Mapping entries to describe how addresses propagate to particular buses in

systems where address decoding cannot be completely described by System Address Space

Mapping entries alone. Entries of this type are required for each bus that is connected to the

system hierarchically below another I/O bus. For example, given the system described in

Figure 4-10, bus hierarchy entries are required for the EISA bus and the PCI BUS 2 since both

have parent buses that are themselves I/O buses.

The Bus Hierarchy entry provides information about where in a hierarchical connection scheme a

given bus is connected and the type of address decoding performed for that bus. Figure 4-11 shows

the format of each entry, and Table 4-15 explains each field. See also Appendix E, for more

information.

28 27

24 23

20 19

16 15

12 11

RESERVED

PARENT BUS

04H

00H

BUS INFO

ENTRY TYPE

129

BUS ID

ENTRY LENGTH

RESERVED

28 27

24 23

20 19

16 15

12 11

Figure 4-11. Bus Hierarchy Descriptor Entry

Version 1.4

4-21

MultiProcessor Specification

Table 4-15 Bus Hierarchy Descriptor Entry Fields

Offset

Length

Field

(in bytes:bits) (in bits)

Description

ENTRY TYPE

Entry type 129 identifies a Bus Hierarchy

Descriptor Entry.

ENTRY LENGTH

BUS ID

A value of 8 indicates that this entry type is eight

bytes long.

The BUS ID identity of this bus. This number

corresponds to the BUS ID as defined in the

base table bus entry for this bus.

BUS INFORMATION:SD 3:0

Subtractive Decode Bus. If set, all addresses

visible on the parent bus but not claimed by

another device on the parent bus (including

bridges to other buses) are useable on this bus.

PARENT BUS

Parent Bus. This number corresponds to the

BUS ID as defined in the base table bus entry for

the parent bus of this bus

For buses where the BUS INFORMATION:SD bit is set, System Address Mappings may not be

needed. Since the bus is defined as being subtractive decode, the range of addresses that appear on

the bus can be derived from address decoding information for parent and peer buses.

4.4.3 Compatibility Bus Address Space Modifier Entry

The Compatibility Bus Address Space Modifier defines a set of predefined address ranges that

should either be added or removed from the supported address map ranges for a given bus. This

entry type is used in combination with System Address Space Mapping entries to complete the

description of memory and I/O ranges that are visible on a bus that incorporates support for ISA

device compatibility.

4-22

Version 1.4

MP Configuration Table

For example, a host bus bridge for a PCI bus that provides ISA compatibility may decode a

predefined range of addresses used for ISA device support in addition to the addresses used for PCI

devices on that bus. A Compatibility Bus Address Space Modifier can be used in this case to add

these predefined address ranges to the list specified by System Address Space Mapping entries for

that PCI bus. As a corollary, in a system where two peer PCI buses are included, one of which

provides ISA compatibility, a Compatibility Bus Address Space Modifier can be used to subtract

these predefined ranges from the address space assigned to the PCI bus that does not support ISA

devices to avoid any potential conflict.

The same effect can be achieved by using System Address Space Mapping entries to completely

describe the address ranges supported on a bus, including those ranges that might otherwise be

described by a Compatibility Bus Address Space Modifier entry. However, given the number of

discrete address ranges that are used for ISA device compatibility, using an approach based solely

on System Address Space Mapping entries may result in a significantly larger number of

configuration table entries and a corresponding increase in table size.

Figure 4-12 shows the format of each entry, and Table 4-16 explains each field.

28 27

24 23

20 19

16 15

12 11

04H

00H

PREDEFINED RANGE LIST

ADDRESS MOD

RESERVED

ENTRY TYPE

130

BUS ID

ENTRY LENGTH

28 27

24 23

20 19

16 15

12 11

Figure 4-12. Compatibility Bus Address Space Modifier Entry

Version 1.4

4-23

MultiProcessor Specification

Table 4-16. Compatibility Bus Address Space Modifier Entry Fields

Offset

(in

Length

(in bits)

Field

Description

bytes:bits)

ENTRY TYPE

ENTRY LENGTH

BUS ID

Entry type 130 identifies a Compatibility Bus

Address Space Modifier Entry.

A value of 8 indicates that an entry of this type is

eight bytes long.

Bus for address space mappings are to be

modified. This number corresponds to the BUS

ID as defined in the base table entry for this bus.

ADDRESS MODIFIER:PR

3:0

If this bit is set to one, the address ranges

specified by PREDEFINED RANGE LIST are to

be subtracted from the address space

associated with the bus. If this bit is set to zero,

the specified address ranges are to be added to

the address space associated with the bus.

PREDEFINED RANGE LIST 4

A number that indicates the list of predefined

address space ranges that this record will

modify for the bus.

PREDEFINED RANGE LIST may take one of the values from Table 4-17. The value of

PREDEFINED RANGE LIST indicates the set of address ranges that are to be either added to or

subtracted from the address range associated with the BUS ID.

Table 4-17. Predefined Range Lists

List

Value

Address Ranges

X100-X3FF

ISA Compatible I/O Range

X500-X7FF

X900-XBFF

XD00-XFFF

VGA Compatible I/O Range

X3B0 - X3BB

X3C0 - X3DF

X7B0 - X7BB

X7C0 - X7DF

XBB0 - XBBB

XBC0 - XBDF

XFB0 - XFBB

XFC0 - XFDF

All addresses in Table 4-17 are in hexadecimal notation. In each case the X represents any

hexadecimal digit, 0-F. As a result, the ISA Compatible I/O Range describes 64 distinct ranges.

4-24

Version 1.4

Default Configurations

The MP specification defines several default MP system configurations. The purpose of these

defaults is to simplify BIOS design. If a system conforms to one of the default configurations, the

BIOS will not need to provide the MP configuration table. The operating system will have the

default MP configuration table predefined internally.

Default system configuration types are defined by MP feature information byte 1, which is part of

the MP floating pointer structure. The physical address pointer field of the MP floating pointer

structure must be zero if one of the default configurations is selected. (Refer to Chapter 4 for a

detailed description of the MP floating pointer structure.)

To use a default configuration, a system must meet the following basic criteria:

1. The system supports two processors.

2. Both processors must execute the common Intel architecture instruction set.

3. The local APICs are located at base memory address 0FEE0_0000h.

4. The local APIC IDs are assigned consecutively by hardware starting from zero.

5. An I/O APIC is present at base memory address 0FEC0_0000h.

6. Either PIC Mode or Virtual Wire Mode is implemented as the power-on default interrupt

mode.

The default system configurations include configurations that use the discrete APIC, such as the

Intel 82489DX or its equivalent, and configurations that use the integrated APIC, such as the

Pentium processors (735\90, 815\100).

Each default configuration has a unique code. Table 5-1 specifies the configuration associated with

each code.

Version 1.4

5-1

MultiProcessor Specification

Table 5-1. Default Configurations

Default

Number

Bus

APIC

Type

Config Code of CPUs Type

Variant

Schematic

ISA

82489DX

As in Figure 5-1, but without

EISA logic.

EISA

82489DX

Neither timer

IRQ0 nor DMA

chaining

As in Figure 5-1, but without

IRQ0 and IRQ13 connection to

the I/O APIC.

EISA

MCA

82489DX

As in Figure 5-1.

As in Figure 5-1, but without

EISA bus logic, with inverters

before I/O APIC inputs 1-15.

ISA + PCI

Integrated

As in Figure 5-2, but without

EISA logic.

EISA + PCI Integrated

MCA + PCI Integrated

As in Figure 5-2.

As in Figure 5-2, but without

EISA bus logic, with inverters

before I/O APIC inputs 1-15.

8-255

Reserved for MP future use.

The default system configurations are designed to support dual-processor systems with fixed

configurations. Systems with dynamically configurable components, for example, a uniprocessor

system with an upgrade socket for the second processor, must always generate the MP

configuration table. Failure to do so may cause the operating system to install the wrong modules

due to erroneous configuration information.

5.1 Discrete APIC Configurations

Figure 5-1 shows the default configuration for systems that use the discrete 82489 APIC. The

Intel486 processor is shown as an example; however, this configuration can also employ Pentium

processors. In Pentium processor systems, PRST is connected to INIT instead of to RESET.

5-2

Version 1.4

Default Configurations

BSP

AP2

INTEL486

CPU 1

INTEL486

CPU 3

IMCR

NMI INTR RESET

ExtINTA

PNMI PINT PRST

LOCAL

82489DX APIC

LOCAL

82489DX APIC

REG.

MARK

INTA

TRAP

LINTIN0

LINTIN1

LINTIN0

LINTIN1

NMI

INTR

RESET

ICC BUS

I/O BUS

NMI

IRQ1

8254 TIMER

IRQ8#

I/O

82489DX

APIC

IRQ13

EISA DMA CHAINING

GLUE

FROM BSP

FERR#

FERR

SAMPLING

IGNNE#

MASTER

8259A PIC

INTR

ABFULL

(PS/2 MOUSE)

ABFULL

SAMPLING

INTA

EDGE/LEVEL TRIGGER

POLARITY CONTROL

SLAVE

8259A PIC

IRQ3-7,

9-12,14,15

IRQx

LITMx

LITM3-7,

9-12,14,15

SHADED AREAS:

A: OPTIONAL IF VIRTUAL WIRE MODE IS IMPLEMENTED

B,C: MAY NOT BE EXTERNALIZED WITH SOME EISA CHIPSETS

C,D: EISA BUS SPECIFIC

Figure 5-1. Default Configuration for Discrete APIC

Version 1.4

5-3

MultiProcessor Specification

The INTA TRAP and GLUE in the figure are the additional hardware interface logic needed for the

82489DX APIC. INTA TRAP conditions all interrupt acknowledge cycles with ExtINTA to steer

the vector either from the 8259A PIC or the APIC. INTA TRAP is also responsible for preventing

the interrupt acknowledge cycle from reaching the 8259A PIC, in case ExtINTA is negated when

PINT is activated. During an interrupt acknowledge cycle with ExtINTA active, the APIC does

not return RDY#. Therefore, the ready generation logic should also take into consideration the

status of ExtINTA to steer the ready signal either from the APIC or from external bus logic,

depending upon the source of the interrupt vector.

The GLUE logic converts the INTR level-triggered interrupt output signal to a signal that is

acceptable to edge-triggered input pins, such as INTIN0 or LINTIN0.

If the AP used in these configurations does not automatically HALT after RESET or INIT, the AP

must be prevented from executing the BIOS by external hardware or by the BIOS itself.

5.2 Integrated APIC Configurations

Figure 5-2 shows the default configuration for systems that use processors with the integrated

APIC, such as Pentium processors (735\90 or 815\100). The local APIC is configured as part of

the processor unit. The APICEN input is used to enable or disable the internal local APIC. If the

internal local APIC is used, the BIOS must initialize the APIC to Virtual Wire Mode during

system initialization.

Both Pentium processors used in the dual-processor (DP) system design are identical. The AP

functions exactly the same as the BSP processor, except that the AP will go to HALT after the

assertion of the RESET or INIT signals. It will remain halted until the MP operating system sends

a STARTUP IPI to bring it on line.

5-4

Version 1.4

Default Configurations

BSP

PENTIUM (735\90, 815\100)

CPU2

PENTIIUM (735\90, 815\100)

CPU1

LOCAL

APIC

APICEN

LOCAL

APIC

APICEN

REG.

MARK

INTR/LINT0

NMI/LINT1

INIT

NMI

INIT

SMI#

ICC BUS

IRQ1

8254 TIMER

IRQ8#

I/O

APIC

INT8

IRQ13

EISA DMA CHAINING

FROM BSP

FERR#

FERR

SAMPLING

IGNNE#

MASTER

8259A PIC

INTR

ABFULL

(PS/2 MOUSE) SAMPLING

ABFULL

PIRQ0-3

EDGE/LEVEL TRIGGER

POLARITY CONTROL

SLAVE

8259A PIC

IRQ3-7,

IRQx

9-12,14,15

LITMx

PIRQ

MAPPING

IMCR

LITM3-7,

3-7,9-11,14,15

9-12,14,15

SHADED AREAS:

A,B: MAY NOT BE EXTERNALIZED WITH SOME EISA CHIPSETS

B,C: EISA BUS SPECIFIC

D: PCI BUS SPECIFIC

Figure 5-2. Default Configuration for Integrated APIC

Two local interrupt input pins, LINT0 and LINT1, are shared with the INTR and NMI pins,

respectively. The LINT0, LINT1, SMI# and INIT signals are switched by APICEN, and they

Version 1.4

5-5

MultiProcessor Specification

should be cross-connected between the BSP and AP processors. Although the INIT pin is cross-

connected between BSP and AP, a targeted INIT IPI initializes only the targeted processor, because

the INIT IPI does not cause the INIT pin to change state.

The interconnection of I/O APIC interrupt lines is the same as for the 82489DX APIC

configuration. However, for PCI system implementations based on the Intel PCI chipset, the PCI

PIRQx lines are mapped to the ISA IRQx via a mapping register. This type of implementation

makes PCI interrupt lines appear as ISA interrupt lines, which are transparent to the operating

system. All PCI systems defined in the default configurations are of this type. No I/O interrupt

assignment entries are declared for PCI interrupts, as described in Section 4.3.4.

5.3 Assignment of I/O Interrupts to the APIC I/O Unit

The typical APIC I/O unit has 16 general-purpose interrupt inputs. Table 5-2 shows how the

interrupt request line (IRQ) assignments are connected to the I/O APIC in each of the default

configurations.

Table 5-2. Default Configuration Interrupt Assignments

First I/O

APIC

INTINx

Config

Comments

INTIN0

INTIN1

8259A

INTR

8259A

INTR

8259A

INTR

8259A

INTR

8259A

INTR

8259A

INTR

N/C

INTR output from

master 8259A or

equivalent

IRQ1

Keyboard controller

buffer full

INTIN2

INTIN3

INTIN4

INTIN5

INTIN6

INTIN7

INTIN8

INTIN9

INTIN10

INTIN11

INTIN12

INTIN13

IRQ0

IRQ3

IRQ4

IRQ5

IRQ6

IRQ7

IRQ8

IRQ9

IRQ10

IRQ11

IRQ12

IRQ13

N/C

IRQ0

IRQ3

IRQ4

IRQ5

IRQ6

IRQ7

IRQ8

IRQ9

IRQ10

IRQ11

IRQ12

IRQ13

IRQ0

IRQ3

IRQ4

IRQ5

IRQ6

IRQ7

IRQ8

IRQ9

IRQ10

IRQ11

IRQ12

IRQ13

IRQ0

IRQ3

IRQ4

IRQ5

IRQ6

IRQ7

IRQ8

IRQ9

IRQ10

IRQ11

IRQ12

IRQ13

IRQ0

IRQ3

IRQ4

IRQ5

IRQ6

IRQ7

IRQ8

IRQ9

IRQ10

IRQ11

IRQ12

IRQ13

IRQ0

IRQ3

IRQ4

IRQ5

IRQ6

IRQ7

IRQ8

IRQ9

IRQ10

IRQ11

IRQ12

IRQ13

8254 Timer

IRQ3

IRQ4

IRQ5

IRQ6

IRQ7

IRQ8

IRQ9

IRQ10

IRQ11

IRQ12

N/C

Real time clock

Floating point

exception and

DMA chaining

INTIN14

INTIN15

IRQ14

IRQ15

IRQ14

IRQ15

IRQ14

IRQ15

IRQ14

IRQ15

IRQ14

IRQ15

IRQ14

IRQ15

IRQ14

IRQ15

NOTE:

N/C designates not connected.

5-6

Version 1.4

Default Configurations

Certain EISA chipsets do not bring out the IRQ0, 8254 timer interrupt, and IRQ13 EISA DMA

chaining interrupt signals. If these signals are not directly available, INTIN2 and INTIN13 should

be disabled. Refer to Section 5.3.1 for more details.

5.3.1 EISA and IRQ13

IRQ13 is a shared interrupt as defined in the EISA bus specification. Because a compliant system

supports only the on-chip floating point unit, IRQ13 carries only the EISA chaining interrupt.

If IRQ13 is not connected to the I/O APIC, the EISA chaining interrupt may be handled as a

mixed-mode operation. Mixed mode means that the APIC and 8259A-equivalent PIC are

connected in a cascading manner via INTIN0, and INTIN0 is programmed for ExtINT and edge-

triggered mode. If all other interrupts are masked off in the PIC, INTIN0 only receives the DMA

chaining interrupt.

An MP operating system should disable the I/O APIC INTIN13 and configure the I/O APIC to

mixed mode if the EISA DMA chaining signal is not available at the I/O APIC.

5.3.2 Level-triggered Interrupt Support

Several AT-compatible buses, such as EISA and MCA, support active-low, level-triggered

interrupts. If these types of buses are to be incorporated in a compliant system, external inverters

must be implemented to ensure that signals presented to the 82489DX APIC are active-high and

level-triggered. See Section 4.3.4 on I/O Interrupt Assignment Flags.

For EISA implementations, the external interrupt polarity control inverters must be controlled by

the EISA edge/level-triggered polarity control registers (4D0h-4D1h). MCA does not have this

line must be implemented.

5.4 Assignment of System Interrupts to the APIC Local Unit

The APIC local unit has two general-purpose interrupt inputs that are reserved for system

interrupts. Table 5-3 shows how the interrupt request line (IRQ) assignments are connected to the

local APIC in each of the default configurations.

Table 5-3 Assignment of System Interrupts to APIC Local Unit

All Local

APICs

LINTINx

Config

Comments

LINTIN0

LINTIN1

8259A

INTR

8259A

INTR

8259A

INTR

8259A

INTR

8259A

INTR

8259A

INTR

8259A

INTR

INTR output from

master 8259A or

equivalent

NMI

Nonmaskable

interrupt

Version 1.4

5-7

MultiProcessor Specification

The 8259A INTR output signal is connected to the LINTIN0 of all local APICs, which makes

INTR dynamically routable via software. NMI is connected to the LINTIN1 of all local APICs,

which makes NMI dynamically routable via software.

In PIC-Mode configurations, the NMI signal is delivered to the local interrupt input 1 (LINTIN1)

of all local APICs and the input of a 2-to-1 MUX. When the system is operated in PIC Mode, the

NMI is sent to the BSP directly via the MUX. The BIOS and the operating system must leave the

LINTIN1 of all local APICs disabled to ensure that the BSP is the only processor that receives the

NMI.

In PIC-Mode configurations, the 8259 INTR signal also follows the same convention, connecting

to the local interrupt line 0 (LINTIN0) of all local APICs and the input of the second 2-to-1 MUX.

When the system is operated in PIC Mode, the 8259 INTR is sent to the BSP directly via the

MUX. The BIOS and the operating system must leave the LINTIN0 of all local APICs disabled to

ensure that the BSP is the only processor that receives the 8259 INTR signal.

When the system is operated in Symmetric I/O Mode, the operating system may enable the

LINTIN0 and LINTIN1 of any or all local APICs as necessary.

5-8

Version 1.4

System BIOS

Programming Guidelines

Depending on the MP components in a multiprocessor system, the system BIOS may have the

following additional responsibilities:

1. Put the APs to sleep, so that they do not all try to execute the same BIOS code as the BSP.

This is necessary, because BIOS code is not typically multithreaded for multiprocessing.

2. Initialize the APICs and other MP components (if any).

3. Build the MP configuration table to communicate information to the operating system about

the APICs and APs.

Note that the above activities can be implemented by the hardware. The BIOS is not required to

perform these activities if the hardware makes them unnecessary. For example, the system BIOS

for one of the default configurations defined in Chapter 5 needs to ensure only that the MP feature

information bytes identify the configuration. In all other respects, the BIOS can be the same as a

standard PC/AT BIOS.

Support for the shutdown status byte (0Fh) of the PC/AT CMOS RAM is required. The startup of

APs by the operating system depends on the jump to warm reset vector (40:67h) capability, as

defined by the shutdown status byte. Appendix B explains this in more detail.

A.1 BIOS Post Initialization

Once system power is applied or the reset button is pressed (if the system is so equipped), a

hardware circuit generates a system RESET sequence to put all the system hardware into an initial

state. All active processors start to execute instructions and enter the POST (power-on self test)

procedure of the BIOS, which is responsible for initializing all components in a system to a known

state and for constructing various system tables in the BIOS data area (400h-4FFh) for the

operating system to use.

For compliant systems that match one of the default configurations listed in Chapter 5, the work

performed by the BIOS POST is minimal. The MP feature information bytes must identify the

default configuration type and determine whether PIC Mode or Virtual Wire Mode is implemented.

During the system INIT or soft reset cycle, both local and I/O APICs must be reinitialized by the

INIT signal and by the BIOS. This is required because the operating system will always assume

that all components in the system are initialized to a known state. For the APIC, this means that

all APIC registers are cleared and the local APIC ID register is initialized by the BIOS or the

hardware.

Upon warm reset, the BIOS must initialize all APICs to the power on state if the warm reset signal

does not physically reset the APICs.

Version 1.4

A-1

MultiProcessor Specification

A.2 Controlling the Application Processors

Provision must be made to prevent all processors from executing the BIOS after a power-on

RESET. System developers may choose to do this by the hardware alone or by cooperation

between hardware and the BIOS. In the latter case, the BIOS may be used for selecting the BSP

and placing all APs to sleep after POST. The BIOS may use the APIC ID as a means by which to

identify each processor and select the proper code sequence to execute. Only the selected BSP

continues to load the operating system after the POST routine.

A.3 Programming the APIC for Virtual Wire Mode

The APICs do not require BIOS programming if the default interrupt mode at start-up is PIC

Mode. Special programming is needed only if the startup interrupt mode is Virtual Wire Mode.

Because Virtual Wire Mode must run all existing uniprocessor software, the system BIOS must

initialize and enable the BSP’s APIC first. The local unit must be programmed to function as a

“virtual wire,” which delivers the CPU interrupt from the 8259A-equivalent PIC to the BSP via its

local APIC.

The External Interrupt (ExtINT) delivery mode must be used so that the APICs and 8259A-

equivalent PICs can function together in the same system. For interrupts that are programmed for

ExtINT delivery mode, there is no need to issue an EOI to the APIC; only the 8259A PIC requires

an EOI as usual. Also, because the 8259A delivers the vector to the processor for ExtINT delivery

mode, the interrupt vector in the APIC’s redirection table is ignored.

To program the APIC to Virtual Wire Mode, the system BIOS must program the APIC to enable

the LINT0 of the BSP’s local APIC for edge-triggered ExtINT delivery mode, and LINT1 for

level-triggered NMI delivery mode. There is no need to program the I/O APIC if it is not used in

Virtual Wire Mode.

Example A-1 is an example of programming the LINTIN0 and LINTIN1 to support Virtual Wire

Mode.

A-2

Version 1.4

System BIOS Programming Guidelines

;-----------------------------------------------------------------------;

; InitLocalAPIC( )

;

;-----------------------------------------------------------------------;

;

Initialize the local APIC to virtual wire mode.

;-----------------------------------------------------------------------;

SVR

LVT1

LVT2

equ

0FEE000F0H

0FEE00350H

0FEE00360H

APIC_ENABLED

equ

000000100H

public InitLocalAPIC

InitLocalAPIC proc near

push

esi

; save regs used for APIC init

; ensure NMI disabled

mov

out

al,080h

070h,al

al,021h

al,0ffh

021h,al

al,0a1h

; read primary imr

; save settings

; mask all off

push

mov

out

push

mov

out

; read secondary imr

; save settings

; mask all off

al,0ffh

0a1h,al

extrn pmode_on : near

call pmode_on

; switch into real big mode

;

; The APIC spurious interrupt must point to a vector whose lower

; nibble is 0F, that is 0xF, where x is 0 - F. Here we use Int 00FH,

; which handles spurious interrupts and supplies the necessary IRET.

; This vector is assumed to have already been initialized in memory.

;

; Enable the APIC via SVR and set the spurious interrupt to use Int 00F

;

mov

and

esi,SVR

eax,[esi]

eax,0FFFFFF0FH

; read SVR

; clear spurious vector (use vector

00FH)

mov

eax,APIC_ENABLED

[esi],eax

; bit 8 = 1

; write SVR

;

; Program LVT1 as ExtInt, which delivers the signal to the INTR signal of all

; processors' cores listed in the destination as an interrupt that originated

; in an externally-connected interrupt controller.

;

Example A-1. Programming Local APIC for Virtual Wire Mode

Version 1.4

A-3

MultiProcessor Specification

mov

and

esi,LVT1

eax,[esi]

eax,0FFFE00FFH

eax,000005700H

[esi],eax

; read LVT1

; not masked, edge, active high

; ExtInt

mov

; write LVT1

;

; Program LVT2 as NMI, which delivers the signal on the NMI signal of all

; processors' cores listed in the destination.

;

mov

and

esi,LVT2

eax,[esi]

eax,0FFFE00FFH

eax,000005400H

[esi],eax

; read LVT2

; not masked, edge, active high

; NMI

; write LVT2

mov

extrn pmode_off : near

call

pmode_off

; switch back to real mode

pop

out

pop

out

0a1h,al

; restore imr settings

; restore secondary imr

021h,al

; restore primary imr

; this routine leaves NMI disabled

; restore regs used in APIC init

; (unless also saved for CPUID)

pop

ret

esi

InitLocalAPIC endp

Example A-1. Programming Local APIC for Virtual Wire Mode (continued)

A.4 Constructing the MP Configuration Table

For a compliant system, one of the main functions of the system BIOS is to construct the MP

floating pointer structure and the MP configuration table. Because the MP configuration table is

optional, the BIOS must set the MP feature information bytes in the MP floating pointer structure

to indicate whether an MP configuration table is present.

If the MP configuration table is required, the BIOS constructs it in conjunction with the BSP and

APs. The BIOS is responsible for synchronizing the activities of the APs during the construction

of the table. The BIOS may need some synchronization during processor initialization so that each

processor may be brought up in the proper order. The mechanism for synchronization is not

specified; however, the procedure described in the following paragraphs of this section uses AP

status flags as an example of a synchronization mechanism. This procedure also initializes the APs

serially. System developers may employ other mechanisms and may initialize all processors in

parallel to minimize the system start-up time.

The BIOS maintains an initialized AP status flag for each AP. Each AP will begin executing the

same BIOS code as the BSP, but will eventually be put in a HALT state or held in a loop until the

BSP enables its AP status flag.

A-4

Version 1.4

System BIOS Programming Guidelines

The BSP is responsible for positioning the MP configuration table. The table can be located within

any unreported, hidden system memory space or within the BIOS ROM region. The BIOS can

select any unused space in those regions. For example, some PC/AT systems implement the

Extended BIOS Data Segment, a 1-Kbyte block usually positioned at the top of the PC’s 640K

base memory. The configuration table may be placed in this portion of the memory map if enough

free space is available.

The BIOS initializes the floating table pointer with the configuration table’s memory address.

Then the BSP constructs the MP configuration table based on the total number of processors,

buses, I/O APICs, and interrupt sources in the system. It sets all AP processor entry bits to zero.

Next, the BSP enables each AP in turn by setting its AP status flag. Each AP follows these steps:

1. It executes a CPU identification procedure, reads its local APIC ID from the Local Unit ID

2. Just prior to exiting the BIOS, the AP clears its status flag to signal the BSP to enable the next

AP initialization process.

3. The AP, which is in Real Mode with interrupts disabled, executes a HLT instruction, and

enters the HALT state.

After the APs complete the initialization process, the BSP continues filling in the rest of the MP

configuration table entries. For I/O interrupt assignment entries, each line should have its trigger

mode and polarity configured according to the device installed.

Once the MP table configuration process is complete, the BSP must calculate the checksum for the

MP configuration table. The MP configuration table should be fully populated at this point. A

disabled processor entry indicates that an AP failed in the initialization self-test. System

developers may choose either to suspend system start up or to continue with the partially functional

system.

The BSP is the only processor that is responsible for loading the operating system. It is up to the

MP operating system to decide when to bring the APs on-line.

Version 1.4

A-5

Operating System

Programming Guidelines

The goal of the MP specification is to transfer enough information about the hardware environment

to the operating system that a single, shrink-wrapped, operating-system binary can boot-up and

fully utilize a wide variety of multiprocessor systems. The following sections explain how the

operating system can take advantage of this specification to handle these operations:

1. Operating-system boot-up.

2. Self configuration.

3. Interrupt mode initialization.

4. Application processor startup.

5. Application processor shutdown.

6. Dynamic interrupt masking.

7. Support for unequal processors.

B.1 Operating System Boot-up

While all processors in an MP-compliant system are functionally identical, one of the processors

will be designated as the boot processor (BSP) at system initialization by the system hardware or

by the system hardware in conjunction with the BIOS. The rest of the processors are designated as

the application processors (APs). The BSP is responsible for booting the operating system. Once

the MP operating system is up and running, the BSP functions as an AP.

Usually a processor is designated as the BSP because it is capable of controlling all system

hardware, including AP startup and shutdown. The operating system must determine and

remember the APIC ID of the designated BSP, so it can keep the BSP operating as the last running

processor during system shutdown. The BSP is not necessarily the first processor, especially in

fault-tolerant MP systems in which any available processor can be designated as the BSP.

At the time that the first instruction of the operating system is executed, the APs are in the

following state:

•

The APs have been restrained (either by the BIOS or by the hardware) from executing

operating system code.

•

The APs are in a halted condition with interrupts disabled. This means that the AP’s local

APICs are passively monitoring the APIC bus and will react only to INIT or STARTUP

interprocessor interrupts (IPIs).

Version 1.4

B-1

MultiProcessor Specification

The operating system’s first task is to determine whether the system conforms to the MP

specification. This is done by searching for the MP floating pointer structure. If a valid floating

pointer structure is detected, it indicates that the system is MP-compliant, and the operating system

should continue to look for the MP configuration table. If the system is not MP-compliant, the

operating system may attempt other means of MP system detection, if it is capable of doing so, or

treat the system as a uniprocessor system.

B.2 Operating System Booting and Self-configuration

An MP configuration table is required by the MP specification, with the exception of the default

system configurations defined in Chapter 5. The table should be treated as read-only by the

operating system. If the MP configuration table exists, the BSP should access the processor entries

in the table to configure the operating system.

The BSP should later configure the operating system based on the bus, I/O APIC, IRQs, and

system interrupt assignment entries of the configuration table. Note that certain types of buses are

mutually exclusive, such as EISA with MCA, or ISA with MCA. The operating system may report

such errors in the configuration table, if they occur.

If the MP configuration table does not exist, the BSP configures the operating system for the

default system configuration indicated by the default configuration bits of the MP feature

information bytes. In this case, only two processors and one I/O APIC exist in the system; both

processors have the same type and features.

The operating system contains a set of predefined internal configuration tables that represent the

default configurations described in Chapter 5. For MP-compliant systems that use one of the

default configurations, the operating system derives the required configuration information from

the corresponding predefined table.

To wake up the AP, the BSP should use the universal algorithm defined in Section B.4.

B.3 Interrupt Mode Initialization and Handling

At the time the operating system boots, the interrupt structure is configured for DOS compatibility.

The system may be running either in PIC Mode or in Virtual Wire Mode. If it is in PIC Mode, the

NMI and INTR will bypass the BSP’s local APIC when the Interrupt Mode Configuration Register

(IMCR) has a value of zero. The operating system should not try to read the IMCR because it may

not exist.

The operating system should switch over to Symmetric I/O Mode to start multiprocessor operation.

If the IMCRP bit of the MP feature information bytes is set, the operating system must set the

IMCR to APIC mode. The operating system should not write to the IMCR unless the IMCRP bit

is set.

B-2

Version 1.4

Operating System Programming Guidelines

Then the operating system should enable its own local APIC, thereby allowing IPI communications

with other APIC-based processors. At this time, the APs’ local APICs have interrupts disabled.

Interrupts must remain disabled at the APs’ local APICs while the BSP is enabling the I/O APIC

and bringing the system to the normal operating state. Otherwise, an I/O interrupt may be

delivered to the uninitialized AP, resulting in the loss of the interrupt.

It is the responsibility of the operating system to assign unique IDs to I/O APIC units.

B.4 Application Processor Startup

An AP may be started either by the BSP or by another active AP. The operating system causes

application processors to start executing their initial tasks in the operating system code by using the

following universal algorithm. The algorithm detailed below consists of a sequence of

interprocessor interrupts and short programmatic delays to allow the APs to respond to the wakeup

commands. The algorithm shown here in pseudo-code assumes that the BSP is starting an AP for

documentation convenience. The BSP must initialize BIOS shutdown code to 0AH and the warm

reset vector (DWORD based at 40:67) to point to the AP startup code prior to executing the

following sequence:

BSP sends AP an INIT IPI

BSP DELAYs (10mSec)

If (APIC_VERSION is not an 82489DX) {

BSP sends AP a STARTUP IPI

BSP DELAYs (200µSEC)

BSP sends AP a STARTUP IPI

BSP DELAYs (200µSEC)

}

BSP verifies synchronization with executing AP

Example B-1. Universal Start-up Algorithm

If the MP configuration table exists, it provides the IDs of the application processor local APICs.

These IDs should be used as the destination addresses in targeted IPIs.

If the MP configuration table does not exist on an MP-compliant system, the system must be of

default configuration type. The MP specification requires local APIC IDs to be numbered

sequentially, starting at zero for all default configurations. As a result, the BSP can determine the

AP’s local APIC ID in default, two-processor configurations by reading its own local APIC ID.

Since there are only two possible local APIC IDs in this case, zero and one, when the APIC ID of

the BSP is one, the APIC ID of the AP is zero, and vice versa. This is important, because a BSP

cannot start up an AP unless it already knows the local APIC ID.

Both INIT IPI and STARTUP IPI are open-ended commands. The operating system is responsible

for determining whether its local APIC unit successfully dispatches one of these commands. The

operating system must do so for an INIT and STARTUP IPI because the APIC either does not

automatically retry or guarantee delivery for one of these special messages.

A local APIC unit indicates successful dispatch of an IPI by resetting the Delivery Status bit in the

Interrupt Command Register (ICR). The operating system polls the delivery status bit after

sending an INIT or STARTUP IPI until the command has been dispatched.

Version 1.4

B-3

MultiProcessor Specification

A period of 20 microseconds should be sufficient for IPI dispatch to complete under normal

operating conditions. If the IPI is not successfully dispatched, the operating system can abort the

command. Alternatively, the operating system can retry the IPI by writing the lower 32-bit double

word of the ICR. This “time-out” mechanism can be implemented through an external interrupt, if

interrupts are enabled on the processor, or through execution of an instruction or time-stamp

counter spin loop.

Optionally, the operating system can make use of the information provided by the integrated APIC

error register. Integrated local APIC units on Intel Architecture processors provide an APIC error

include SEND_ACCEPT and RECEIVE_ACCEPT errors that are generated when no processor

responds to a message on the APIC bus. An example of this type of failure includes sending a

STARTUP IPI to a destination APIC ID for a non-existent processor. Other errors are usually

indicative of hardware problems and include SEND and RECEIVE CHECKSUM errors.

The operating system is responsible for determining whether the IPI was received by the targeted

AP and executed successfully. For example, the operating system can define a status flag for each

processor. After being awakened, a processor sets its status flag, indicating to the operating system

that it is present and running. If the status flag does not change value after a certain time, the

operating system should treat the processor as not present or not functional.

The following two sections describe specifics relating to the use of INIT and STARTUP IPIs that

can be used to guide the implementation of the universal algorithm presented here.

B.4.1 USING INIT IPI

INIT IPIs can be used with systems based on the 82489DX APIC, or on systems that are based on

multiple Pentium (735/90, 815/100) processors. INIT IPI is an Interprocessor Interrupt with trigger

mode set to level and delivery mode set to “101” (bits 8 to 10 of the ICR). INIT IPIs should

always be programmed as level triggered; the operating system must perform two writes to the ICR

to assert and then deassert this delivery mode.

An INIT IPI is an IPI that has its delivery mode set to RESET. Upon receiving an INIT IPI, a local

APIC causes an INIT at its processor. The processor resets its state, except that caches, floating

point unit, and write buffers are not cleared. Then the processor starts executing from a fixed

location, which is the reset vector location. To cause the processor to jump to a different location,

the INIT IPI must be used as part of a warm-reset.

The warm reset is a feature of every standard PC/AT BIOS. It allows the INIT signal to be

asserted without actually causing the processor to run through its entire BIOS initialization

procedure (POST). This feature is used, for example, to return an 80286 processor to Real Mode.

The key components of warm reset are the following:

•

Shutdown code. One of the first actions of the BIOS POST procedure is to read the shutdown

code from location 0Fh of the CMOS RAM. This code can have any of several values that

indicate the reason that an INIT was performed. A value of 0Ah indicates a warm reset.

•

Warm-reset vector. When POST finds a shutdown code of 0Ah, it executes an indirect jump

via the warm-reset vector, which is a doubleword pointer in system RAM location 40:67h.

B-4

Version 1.4

Operating System Programming Guidelines

By putting an appropriate pointer in the warm-reset vector, setting the shutdown code to 0Ah, then

causing an INIT, the BIOS (or the operating system) can cause the current processor to jump

immediately to any location. Because all processors in an MP system share the same system

memory, and because the INIT IPI gives one processor the power to cause an INIT at another, the

operating system can cause any processor to jump immediately to any location.

B.4.2 USING STARTUP IPI

STARTUP IPIs are used with systems based on Intel processors with local APIC versions of 1.x or

higher. These local APICs recognize the STARTUP IPI, which is an APIC Interprocessor Interrupt

with trigger mode set to edge and delivery mode set to “110” (bits 8 through 10 of the ICR).

The STARTUP IPI causes the target processor to start executing in Real Mode from address

000VV000h, where VV is an 8-bit vector that is part of the IPI message. Startup vectors are

limited to a 4-kilobyte page boundary in the first megabyte of the address space. Vectors A0-BF

are reserved; do not use vectors in this range. STARTUP IPIs are not maskable, do not cause any

change of state in the target processor (except for the change to the instruction pointer), and can be

issued only one time after RESET or after an INIT IPI reception or pin assertion. A STARTUP IPI

neither requires the targeted APIC to be enabled nor the interrupt table to be programmed. If the

target processor is in the halted state immediately after RESET or INIT, a STARTUP IPI causes it

to leave that state and start executing. The effect is to set CS:IP to VV00:0000h.

For an operating system to use a STARTUP IPI to wake up an AP, the address of the AP

initialization routine (or of a branch to that routine) must be in the form of 000VV000h. Sending a

STARTUP IPI with VV as its vector causes the AP to jump immediately to and begin executing

the operating system’s AP initialization routine.

The operating system should not issue a STARTUP IPI to an 82489DX since the STARTUP IPI

will be ignored instead of forcing the targeted processor to execute from the given address.

B.5 AP Shutdown Handling

An AP may be shut down by itself, by the BSP, or by another active AP. Shutting down an AP

with an active task or a bound device driver is not permitted. Only the BSP may shut itself down,

and it must be the last processor to shut down.

There are several possible alternatives for shutdown handling. Two of the possibilities are

presented here by way of example.

In the simplest case, the operating system can define an IPI for shutting down and taking APs off-

line. The exact usage and behavior of an IPI that is defined for this purpose is operating system

specific and is not defined in this specification.

Should the BSP need to take an AP off-line or to place the AP back into a HALT state, the BSP

can also use the INIT IPI with Warm Restart. The operating system places the address of a HLT

instruction in the warm-reset vector (40:67), sets the CMOS shut-down code to 0Ah, then sends an

INIT IPI to the AP. The INIT IPI causes the AP to enter the BIOS POST routine, where it

immediately jumps to the warm-reset vector and executes the operating system’s HLT instruction.

Only one processor can execute the shutdown routine at any given time, due to the use of the

shutdown code. The operating system must not rely on any code being executed after the delivery

Version 1.4

B-5

MultiProcessor Specification

of an INIT IPI used to shut down an AP. As a result, the operating system must ensure that any

required state information is captured and that caches are flushed as necessary before sending the

INIT IPI.

In order to do a complete system shutdown, followed by a warm restart if necessary, the operating

system should return the system to a state similar to that at power-on. This includes disabling the

Local APIC interrupts (LINT0/LINT1/Local APIC Timer/Error interrupt) on all processors,

disabling the Local APIC on all APs and disabling all interrupts at all the I/O APICs in the system.

The operating system can use an IPI or an NMI to signal to all APs for per-processor shutdown

handling. The operating system may then set the CMOS shutdown code to 0Ah and perform a

keyboard controller reset.

B.6 Other IPI Applications

The operating system may use IPIs for other run-time duties, such as handling the various

processor caches.

B.6.1 Handling Cache Flush

The MP specification requires that hardware maintain cache coherency. Cache flushing by the

operating system should not be required under normal circumstances. The only need for cache

flushing by the operating system is prior to powering down a processor.

Should a system-wide cache flush be necessary, the operating system should use the broadcast IPI

mechanism to request that each of the processors write back and invalidate its own cache

subsystem and then synchronize upon the completion of that activity.

B.6.2 Handling TLB Invalidation

The operating system should use the IPI mechanism to request that each of the processors

invalidate its TLBs. The operating system may use a broadcast IPI for this purpose. The BSP and

APs should synchronize the completion of their actions either via memory-based semaphores or via

targeted return IPIs. The actual IPI vector is operating system dependent.

B.6.3 Handling PTE Invalidation

The operating system should use the IPI mechanism to request that each processor invalidate a

specific page-table entry (PTE) if it is cached in that processor’s TLBs. The operating system may

use a broadcast IPI for this purpose. The BSP and APs should synchronize the completion of their

actions either via memory-based semaphores or via targeted return IPIs. The actual IPI vector is

operating system dependent.

B.7 Spurious APIC Interrupts

For the 8259, there is a time window in which a spurious interrupt may be misinterpreted as a

genuine interrupt. For example, if an interrupt goes inactive just after the first INTA cycle but

before the second INTA cycle, the 8259 will also signal this spurious interrupt as a genuine

B-6

Version 1.4

Operating System Programming Guidelines

interrupt. The distributed APIC architecture, by its nature, is more vulnerable to spurious interrupt,

because the device interrupt may be latched and recognized without the INTA cycle.

To ensure that spurious interrupts are handled properly, it is strongly recommended that the device

drivers must read the status register before servicing the device. In cases where spurious interrupts

do occur, the device drivers may simply ignore them.

Note that this spurious interrupt is not related in any way to the Local APIC Spurious Interrupt

generated by the Local APIC Error Register (0xFEE00370)

B.8 Supporting Unequal Processors

Some MP operating systems that exist today do not support processors of different types, speeds,

or capabilities. However, as processor lifetimes increase and new generations of processors arrive,

the potential for dissimilarity among processors increases. The MP specification addresses this

potential by providing an MP configuration table to help the operating system configure itself.

Operating system writers should factor in processor variations, such as processor type, family,

model, and features, to arrive at a configuration that maximizes overall system performance. At a

minimum, the MP operating system should remain operational and should support the common

features of unequal processors.

Version 1.4

B-7

System Compliance Checklist

Any "NO" answer indicates non-compliance.

Condition

YES

1. PC/AT Compatibility

Does system contain all necessary MP-compatible circuitry?

Will system boot and run DOS and Microsoft Windows?

2. Memory Subsystem

Are system memory address map, cacheability, and shareability consistent with

definitions in Table 3-1?

Are memory-mapped I/O devices located at the top of the memory address space?

If the system has external cache:

Is cache coherence maintained by hardware?

Is cache flushing supported by hardware?

Is cache flushing limited to the local caches?

Are all locked operations visible to all processors?

Are locked operations guaranteed on aligned memory operations?

Are memory writes observed externally in same order as programmed?

3. Multiprocessor Interrupt Control

Does each processor have its own local APIC?

Is there a processor designated for booting?

Does system support either PIC Mode or Virtual Wire Mode at power-on?

Is Symmetric I/O Mode implemented?

Are all APIC IDs unique?

Are all local APIC IDs assigned by hardware or BIOS?

Do local APIC IDs begin with zero?

4. Reset Support

Does system implement software-initiated processor-specific INIT (INIT IPI)?

Do system soft and hard resets reset all processors?

5. APIC Interval Timer

On 82489DX, is CLK the only clock source of the APIC timer?

6. Fault Resilient Booting

If fault resilient booting feature is implemented:

Is fault resilient booting feature software transparent?

Are system signals: NMI, INTR, FERR#, IGNNE#, and A20M connected to BSP?

7. MP Floating Pointer Structure and MP Configuration Table

Is the MP floating pointer structure implemented?

If the MP feature byte 1 (default configuration type) is nonzero:

Does the system correctly implement one of the default configurations?

If the MP feature byte 1 (default configuration type) is zero:

Has the system created a correct MP configuration table?

If the IMCR presence bit is set, is PIC Mode implemented?

If the IMCR presence bit is not set, is Virtual-Wire Mode implemented?

Version 1.4

C-1

Multiple I/O APIC

Multiple PCI Bus Systems

The information in this specification describes the majority of multiprocessor systems. This

appendix provides clarifications for implementors who are considering designs with more than one

I/O APIC. In particular, a number of proposed systems will incorporate multiple I/O APICs in

order to support multiple PCI buses. This appendix provides guidance for implementors who wish

to be sure that their designs comply with this specification.

D.1 Interrupt Routing with Multiple APICs

Two basic approaches to routing interrupts can be used when the system has more than one I/O

APIC:

•

The fixed routing scheme uses the same routing in both PIC or Virtual Wire Mode and

symmetric I/O mode.

•

The variable approach changes the routing when switching to symmetric I/O mode.

This section applies when a PCI interrupt is connected both to an I/O APIC input of its own and to

the I/O APIC input of the EISA/ISA IRQ to which the interrupt is routed when in PIC or Virtual

Wire Mode. This double routing is typically used to preserve PC AT compatibility at system start-

up, allowing a system to boot from a disk connected to a PCI controller on a second PCI bus, for

example. To prevent double delivery of this PCI interrupt once the system switches to symmetric

I/O mode for an MP operating system, the duplicate routing must either be turned off or concealed

from the operating system. If a PCI interrupt is only connected via an EISA/ISA IRQ, the

EISA/ISA entry in the MP configuration table is sufficient to describe the routing.

The variable routing method described below is preferred since it is more flexible and offers best

use of available system resources. Fixed routing is described here for compatibility with existing

systems that do not implement a variable routing strategy.

D.1.1 Variable Interrupt Routing

In systems with variable interrupt routing, all PCI interrupts map to EISA/ISA IRQs when in PIC

or Virtual Wire Mode. When switched to symmetric I/O mode, the system disables this routing

and delivers the PCI interrupt through I/O APIC inputs different from those used by the EISA/ISA

IRQs.

If IMCR is implemented, the hardware design can use this bit to enable/disable the routing of the

PCI interrupts to EISA/ISA IRQs. On systems with IMCR, this operation might be the only one

that is required of the operating system when switching to symmetric I/O mode, other than the

actual programming of the I/O and Local APICs.

Version 1.4

D-1

MultiProcessor Specification

If IMCR is implemented but the system includes one or more I/O APICs that are not controlled

through IMCR, the hardware must accomplish routing changes for such I/O APICs by some other

means when the system switches into symmetric I/O mode. These routing changes must be done

without requiring any additional intervention from software.

For systems without the IMCR register, the routing of the PCI interrupts to the EISA/ISA IRQ

must be automatically disabled as the I/O APICs are programmed. Therefore, when the operating

system programs the I/O APICs in accordance with the MP configuration table, the hardware must

detect this operation and disable the routing mechanism without additional intervention by the

operating system. This operation can be done globally for an entire system as soon as any APIC

interrupts are enabled or it can be done on an interrupt-by-interrupt basis.

D.1.2 Fixed Interrupt Routing

Several implementations of fixed interrupt routing are possible, depending on hard wiring, via

jumpers for example, or software means, such as chipset-specific registers. Since these

implementations have no mechanism to disable the PCI interrupt to EISA/ISA IRQ routing, the

MP configuration table must be set up carefully to avoid problems with duplicate interrupt

mappings in symmetric I/O mode.

To avoid such problems on systems with fixed routing, PIC or Virtual Wire Mode interrupt

routings must not be used by software when the system is in symmetric mode, since these routings

cannot be disabled or altered. This situation implies two restrictions that must be placed on the

way PCI interrupts are routed to EISA/ISA IRQs and on that way the MP configuration table is

built:

•

If a PCI interrupt is routed to an EISA/ISA IRQ that is used by an EISA/ISA device, that PCI

interrupt must be delivered through the same I/O APIC input as that EISA/ISA IRQ. The

connection from the PCI interrupt to the additional I/O APIC input must not be entered in the

MP configuration table.

•

If a PCI interrupt is routed to an EISA/ISA IRQ which is NOT used by an EISA/ISA device,

that PCI interrupt must be delivered through its individual I/O APIC connection, and the

connection of that EISA/ISA IRQ to its I/O APIC input should not be entered in the MP

configuration table.

As an example, take a system with two I/O APICs, a PCI bus, and an EISA bus. Each EISA IRQ

is connected to an input of one I/O APIC and each PCI interrupt is connected to the other I/O

APIC. A fixed interrupt routing design could connect all PCI interrupts to a single EISA interrupt.

This design does not give optimum performance, because PCI interrupts must be shareable, but it

does allow all interrupts to be properly handled.

If an EISA device is connected to the same interrupt, the MP configuration table would not contain

any entries for the second I/O APIC. The second I/O APIC is not used.

If no EISA device uses that interrupt, however, the MP configuration table could contain an entry

for each PCI interrupt, describing its connection to the second I/O APIC. The EISA IRQ used by

the PCI interrupts in PIC mode would not have an entry in the table. All other EISA IRQs would

have an entry in the table describing connections to the first I/O APIC.

D-2

Version 1.4

Multiple I/O APIC Multiple PCI Bus Systems

Fixed interrupt routing also implies a restriction on software that is implicit but important in the

context of systems with more than one I/O APIC. The operating system must program I/O APICs

to handle only the interrupts for which the MP configuration table contains corresponding I/O

interrupt assignment entries. If the configuration table contains no entry for a given I/O APIC

input, that interrupt must be left in the masked state.

D.2 Bus Entries in Systems with More Than One PCI Bus

To accommodate systems with more than one PCI bus within the confines of version 1.1 of this

specification, construction of the bus entries on the MP configuration table must be handled in a

very particular sequence:

1. Begin with bus entries for the PCI buses. Start at bus zero, using the actual PCI bus number as

the bus ID for the bus entry.

2. Add entries for other buses. These entries can use bus ID numbers left vacant by the PCI bus

entries.

This sequence implies that bus ID numbers do not have to increase sequentially by increments of

one; the requirement is that they must appear in ascending order by bus ID number. This specific

interpretation of the information presented in Table 4-7 ensures consistency between the

information in the MP configuration table and the model for systems with multiple PCI buses that

is presented in the formal PCI specification, which allows for more flexibility in bus numbering.

This numbering scheme requires bus entries in the MP configuration table to be sorted

appropriately. For example, bus entries should appear in the order PCI (0), EISA (1), and PCI(4)

in a system with three buses, two PCI buses numbered 0 and 4, and a single EISA bus numbered

as 1.

D.3 I/O Interrupt Assignment Entries for PCI Devices

Section 4.3.4 defines the format of interrupt assignment entries. The example presented there does

not, however, completely explain the semantics of the source bus IRQ field for PCI devices.

For PCI devices, the semantics for encoding PCI interrupts should mirror the PCI specification as

follows:

Table D-1. I/O Interrupt Entry Source Bus IRQ Field for PCI Devices

Offset

(in bytes:bits)

Length

(in bits)

Field

Description

SOURCE BUS IRQ

5:0

Identifies the PCI interrupt signal, where

0x0 corresponds to INT_A#, 0x1 to

INT_B#, 0x2 to INT_C# and 0x3 to

INT_D#.

SOURCE BUS IRQ

RESERVED

5:2

5:7

Gives the PCI Device Number where the

interrupt originates.

Reserved for future use.

Version 1.4

D-3

Errata

The following sections provided here are intended to replace the corresponding sections in the

main body of the specification. The sections provided here are shown in their entirety to provide

context for the changes. Changes contained below are marked with double underlined text which

shows changes relative to the version of the sections that are provided in the main body of the

specification.

4.1 MP Floating Pointer Structure

An MP-compliant system must implement the MP floating pointer structure, which is a variable

length data structure in multiples of 16 bytes. Currently, only one 16-byte data structure is defined.

It must span a minimum of 16 contiguous bytes, beginning on a 16-byte boundary, and it must be

located within the physical address as specified in the previous section. To determine whether the

system conforms to the MP specification, the operating system must search for the MP floating

pointer structure in the order specified in the previous section. Figure 4-2 shows the format of this

structure, and Table 4-1 explains each of the fields.

24 23

16 15

MP FEATURE

BYTES 2-5

0CH

MP FEATURE

BYTE 1

CHECKSUM

SPEC_REV

LENGTH

08H

PHYSICAL ADDRESS POINTER

SIGNATURE

04H

00H

_ (5Fh)

P (50h)

M (4Dh)

_ (5Fh)

24 23

16 15

Figure 4-2. MP Floating Pointer Structure

Version 1.4

E-1

MultiProcessor Specification

Table 4-1. MP Floating Pointer Structure Fields

Offset

Length

Field

(in bytes:bits) (in bits)

Description

SIGNATURE

The ASCII string represented by “_MP_” which

serves as a search key for locating the pointer

structure.

PHYSICAL ADDRESS

POINTER

The address of the beginning of the MP

configuration table. All zeros if the MP

configuration table does not exist.

LENGTH

The length of the floating pointer structure table

in paragraph (16-byte) units. The structure is 16

bytes or 1 paragraph long; so this field contains

01h.

SPEC_REV

CHECKSUM

The version number of the MP specification

supported. A value of 01h indicates Version 1.1.

A value of 04h indicates Version 1.4.

A checksum of the complete pointer structure.

All bytes specified by the length field, including

CHECKSUM and reserved bytes, must add up to

zero.

Bits 0-7:

MP System Configuration Type.

MP FEATURE

INFORMATION BYTE 1

When these bits are all zeros, the MP

configuration table is present. When nonzero,

the value indicates which default configuration

(as defined in Chapter 5) is implemented by the

system.

Bits 0-5:

MP FEATURE

INFORMATION BYTE 2

12:0

12:6

Reserved for future MP definitions.

Bit 6:

Multiple Clock Sources. When set, this bit

indicates that the processors derive clock signals

from different sources. Otherwise when not set

this bit indicates that all processors share a

single clock source.

12:7

Bit 7:

IMCRP. When the IMCR presence bit is

set, the IMCR is present and PIC Mode is

implemented; otherwise, Virtual Wire Mode is

implemented.

MP FEATURE

INFORMATION BYTES 3-5

Reserved for future MP definitions. Must be

zero.

The MP feature information byte 1 specifies the MP system default configuration type. If nonzero,

the system configuration conforms to one of the default configurations. The default configurations,

specified in Chapter 5, may only be used to describe systems that always have two processors

installed.

Bit 6 of MP feature information byte 2, the Multiple Clock Source bit, is used by the operating

system to determine how processors derive clock sources. If the system design does not provide

for a single clock source shared by all processors then this bit is set. Otherwise this bit is zero to

indicate that all processors derive their clock from a common source. Bit 7 of MP feature

E-2

Version 1.4

MP Configuration Table

information byte 2, the IMCR present bit, is used by the operating system to determine whether

PIC Mode or Virtual Wire Mode is implemented by the system.

The physical address pointer field contains the address of the beginning of the MP configuration

table. If it is nonzero, the MP configuration table can be accessed at the physical address provided

in the pointer structure. This field must be all zeros if the MP configuration table does not exist.

4.4.1 System Address Space Mapping Entries

System Address Space Mapping entries define the system addresses that are visible on a particular

bus. Each bus defined in the Base Table can have any number of System Address Space Mapping

entries included in the Extended Table. Thus, individual buses can be configured to support

different address ranges, thereby decreasing the amount of bus traffic on a given bus and increasing

the overall system performance. Each consecutive address range that is usable by the operating

system to access devices on a given bus has a System Address Space Mapping entry. Figure 4-9

shows the format of each entry, and Table 4-14 explains each field.

28 27

24 23

20 19

16 15

12 11

ADDRESS LENGTH

10H

0CH

08H

04H

00H

ADDRESS BASE

ENTRY TYPE

128

ADDRESS TYPE

BUS ID

ENTRY LENGTH

28 27

24 23

20 19

16 15

12 11

Figure 4-9. System Address Space Entry

Version 1.4

E-3

MultiProcessor Specification

Table 4-14. System Address Space Mapping Entry Fields

Offset

(in bytes:bits)

Length

(in bits)

Field

Description

ENTRY TYPE

Entry type 128 identifies a System Address Space

Mapping Entry.

ENTRY LENGTH

BUS ID

A value of 20 indicates that an entry of this type is

twenty bytes long.

The BUS ID for the bus where the system address

space is mapped. This number corresponds to the

BUS ID as defined in the base table bus entry for

this bus.

ADDRESS TYPE

System address type used to access bus

addresses must be:

0 = I/O address

1 = Memory address

2 = Prefetch address

All other numbers are reserved.

Starting address

ADDRESS BASE

LENGTH

Number of addresses which are visible to the bus

If any main memory address is mapped to a software visible bus, such as PCI, it must be explicitly

declared using a System Address Space Mapping entry.

In the case of a bus that is directly connected to the main system bus, system address space records

and compatibility base address modifiers must be provided as needed to fully describe the complete

set of addresses that are mapped to that bus. For example, in Figure 4-10, complete explicit

descriptions must be provided for PCI BUS 0 and PCI BUS 1 even if one of the buses is

programmed for subtractive decode.

Processor 0

Processor 1

Memory

Controller

System Bus

PCI

Host Bridge

PCI

Host Bridge

PCI Bus 0

PCI Bus 1

PCI-to-PCI

Bridge

EISA Bridge

EISA Bus

PCI Bus 2

Figure 4-10. Example System with Multiple Bus Types and Bridge Types

E-4

Version 1.4

MP Configuration Table

Since all device settings must fall within supported System Address Space mapping for a given bus

in order to be usable by the operating system, buses that do not support dynamically configurable

devices (i.e., ISA, EISA) should support all possible addresses to that bus.

In general, the MP configuration table must provide entries to describe system address space

mappings for all I/O buses present in the system. There are two exceptions to this rule:

1. For buses that are connected via PCI-to-PCI-bridge-specification-compliant bridges: in this

case, the system address space mappings for such buses may be omitted from the configuration

table if bus hierarchy entries for these buses are also omitted (refer to Section 4.4.2 for

additional information). In such cases, the operating system is expected to discover the address

space mapping by querying the PCI-to-PCI-bridge-specification-compliant bridge directly.

2. For buses that are connected via a parent I/O bus and for which the subtractive decode bit is set

(refer to Section 4.4.2 for details).

Typically, this would mean that a minimal description of resources allocated to PCI buses in a

system need only include System Address Space Mapping entries for PCI bus zero and any

additional peer PCI buses (if present) where these buses are connected by PCI bridges that are

specific to the chipset or host bus.

Note

System Address Mappings are unidirectional in nature. They only describe system addresses that

propagate to the target bus from any given processor. For DMA support on the target bus, all memory

addresses that contain real memory should be accessible directly by either the bus or by a bus-specific

DMA controller. For buses with fewer than 32-bit address lines, all real memory at addresses that the

bus can generate must be accessible for DMA.

4.4.2 Bus Hierarchy Descriptor Entry

If present, Bus Hierarchy Descriptor entries define how I/O buses are connected relative to each

other in a system with more than one I/O bus. Bus Hierarchy Descriptors are used to supplement

System Address Mapping entries to describe how addresses propagate to particular buses in

systems where address decoding cannot be completely described by System Address Space

Mapping entries alone.

In general, entries of this type are required for each bus that is connected to the system

hierarchically below another I/O bus. The one exception to this is for PCI buses that are connected

behind a PCI to PCI bridge specification compliant bridge. In this case the bus hierarchy entry

may be omitted as the operating system is expected to discover such buses without the need to refer

to the configuration table. Since bus hierarchy descriptor and system address space records are

intended to work in concert, the configuration table may chose to omit both types of records for a

bus behind a compliant bridge or it may provide both. The table must not however provide one

without the other for buses behind compliant bridges. For example, given the system described in

Figure 4-10, bus hierarchy entries are required for the EISA bus and may be provided for PCI

BUS 2 since both have parent buses that are themselves I/O buses. If the configuration table does

include a bus hierarchy entry for PCI BUS 2, then a complete set of corresponding system address

space records must also be provided.

The Bus Hierarchy entry provides information about where in a hierarchical connection scheme a

given bus is connected and the type of address decoding performed for that bus. Figure 4-11 shows

the format of each entry, and Table 4-15 explains each field.

Version 1.4

E-5

MultiProcessor Specification

28 27

24 23

20 19

16 15

12 11

RESERVED

PARENT BUS

04H

00H

BUS INFO

ENTRY TYPE

129

BUS ID

ENTRY LENGTH

RESERVED

28 27

24 23

20 19

16 15

12 11

Figure 4-11. Bus Hierarchy Descriptor Entry

Table 4-15 Bus Hierarchy Descriptor Entry Fields

Offset Length

Field

(in bytes:bits) (in bits) Description

ENTRY TYPE

Entry type 129 identifies a Bus Hierarchy Descriptor

Entry.

ENTRY LENGTH

BUS ID

A value of 8 indicates that this entry type is eight

bytes long.

The BUS ID identity of this bus. This number

corresponds to the BUS ID as defined in the base

table bus entry for this bus.

BUS INFORMATION:SD 3:0

Subtractive Decode Bus. If set, all addresses visible

on the parent bus but not claimed by another device

on the parent bus (including bridges to other

buses) are useable on this bus.

PARENT BUS

Parent Bus. This number corresponds to the BUS

ID as defined in the base table bus entry for the

parent bus of this bus

For buses where the BUS INFORMATION:SD bit is set, System Address Mappings may not be

needed. Since the bus is defined as being subtractive decode, the range of addresses that appear on

the bus can be derived from address decoding information for parent and peer buses.

E-6

Version 1.4

Glossary

82489DX: The 82489DX Advanced Programmable Interrupt Controller (APIC).

8259A: The 8259A Programmable Interrupt Controller (PIC) or its equivalent.

AP: Application processor, one of the processors not responsible for system initialization.

APIC: Advanced Programmable Interrupt Controller, either the 82489DX APIC or the integrated

APIC on Pentium processors.

BIOS: Basic Input/Output Subsystem.

BSP: Bootstrap processor, the processor responsible for system initialization.

Cache coherency: A property of a cache/memory system that guarantees that a request for an item

from memory will retrieve the most up-to-date value of that item, regardless of what cache or

memory location currently holds that value.

CMOS RAM: The battery backed-up configuration memory of the PC/AT motherboard.

DP: A dual processor system is one with two processors.

ExtINT: A delivery mode of the Local Vector Table of a local APIC that causes delivery of a

signal to the INT pin of the processor as an interrupt that originated in an externally connected

8259A-equivalent PIC. The ExtINTA output signal is also asserted. The INTA cycle that

corresponds to the ExtINT delivery should be routed to the external PIC that is expected to supply

the vector.

Flush: Write back all modified lines of a cache.

INIT: Unless otherwise specified, the processor-specific reset or system-wide soft reset functions.

This definition is functional and sometimes bears no relationship to the actual signal name. For

example, the term "INIT" may refer to the INIT signal on the Pentium processor or to the RESET

signal on the Intel486 processor.

INIT IPI: A type of APIC interprocessor interrupt whose delivery mode is set to RESET. Upon

delivery to the specified destinations, the destination APICs assert their PRST output signals.

When the PRST lines are connected to the INIT or RESET inputs of their respective processors, an

INIT IPI causes reinitialization of the destination processors.

Invalidate: Change the state of a cache line to the Invalid state.

IPI: Interprocessor interrupt.

MESI: A cache coherency protocol named after the states that cache lines may have: Modified,

Exclusive, Shared, Invalid.

MP: A multiprocessor system is one with two or more processors.

PIC: Programmable Interrupt Controller.

Version 1.4

Glossary-1

MultiProcessor Specification

PIC Mode: One of three interrupt modes defined by the MP specification. In this mode the APICs

are effectively disabled, while interrupts are generated by 8259A-equivalent PICs and delivered

directly to the BSP. This is a uniprocessor compatibility mode.

POST: Power-On Self Test, the first BIOS procedure executed after a RESET or INIT.

RESET: The system-wide hard reset. This definition is functional. It may refer to the RESET

signal on both Pentium and Intel486 processors or the RESET signal of the 82489DX APIC.

Shutdown code: The value of CMOS RAM location 0Fh, which indicates that reason that a

RESET was performed.

STARTUP IPI: A type of APIC interprocessor interrupt that is similar to an NMI with an

embedded vector. It does not cause any change of state, but merely causes the targeted processor

to start executing in Real Mode from address 000VV000h, VV being an 8-bit vector which is part

of the IPI message. Startup vectors are limited to a 4K page boundary in the first 1 MB of the

address space. STARTUP IPIs are not maskable, and can be issued at any time. The benefit of

STARTUP-IPI compared to NMI is that it does not require the targeted APIC to be enabled, and it

does not require the interrupt table to be programmed. Thus, the operating system's initialization

procedure can use it to wake up an AP that has been sleeping since RESET or INIT. The

STARTUP IPI is not supported by the 82489DX APIC.

Symmetric I/O Mode: One of three interrupt modes defined by the MP specification. In this

mode, the APICs are fully functional, and interrupts are generated and delivered to the processors

by the APICs. Any interrupt can be delivered to any processor. This is the only multiprocessor

interrupt mode.

Symmetry: The relationship of equality among components of a multiprocessor system in which

no processor is special with respect to its access to memory, interrupts, or I/O. For interrupts,

symmetry means that any interrupt from any source can be routed to any processor and handled

there. For I/O, it means that all I/O control registers, be they in memory space, I/O space, or some

other special address space, are accessible to all processors. Per-processor control hardware, such

as interrupt controllers or processor identification registers, must be at the same physical address

for all processors.

Virtual Wire Mode: One of three interrupt modes defined by the MP specification. In this mode

interrupts are generated by the 8259A-equivalent PICs, but delivered to the BSP by an APIC that is

programmed to act as a "virtual wire"; that is, the APIC is logically indistinguishable from a

hardwired connection. This is a uniprocessor compatibility mode.

Warm reset: A technique that allows the RESET or INIT signal to be asserted without actually

causing the BIOS to run through its entire initialization procedure. If a value of 0Ah is placed in

the shutdown code, the first instructions of the BIOS POST procedure read the warm-reset vector

from system RAM location 40:67h, and jump to that address.

Write-back: A cache update policy wherein a modified cache line is not written back to main

memory until the last possible instant—when another processor needs to access the data or when

its location in the cache is needed for other data.

Glossary-2

Version 1.4

Order Number: 242016-006

Printed in U.S.A.

Intel GA 965P DS3 User Manual
JVC AV 21E14 User Manual
JVC AV 21MS30 User Manual
KitchenAid KGCS166GWH3 User Manual
LG Electronics CT 29M65VE User Manual
NEC E700 User Manual
Oricom T1000 User Manual
Panasonic TX 29AS1X User Manual
Philips 25PT3323 User Manual
Philips 170B4BG User Manual