While reasonable efforts have been made to assure the accuracy of this document,
Motorola, Inc. a ssumes n o lia bility r esulti ng from any omissio ns in this docu ment, or from
the use of the information ob tained therein. Motorola reserv es the right to r evise this
document and to ma ke c hanges from time to ti me in t he content hereof witho ut o bligation
of Motorola to notify any person of such revision or changes.
No part of this material may be reproduce d or copied in any tangi ble medium, or stored in
a retrieval system, or transmitted in any form, or by any means, radio, electronic,
mechanical, photocopying, recording or facsimile, or otherwise, without the prior written
permission of Motorola, Inc.
It is possible that this pu blicati on ma y cont ain r eferenc e to, or i nformat ion a bout Mot oro la
products (machines and programs), programming, or services that are not announced in
your country. Such refere nces or informat ion must not be construed to mean that Moto rola
intends to announce such Motorola products, programming, or services in your country.
Restricted Rights Legend
If the documentation contained herein is supplied, directly or indirectly, to the U.S.
Government, the following notice shall apply unless otherwise agreed to in writing by
Motorola, Inc.
Use, duplication, or disclosure by the Government is subject to restrictions as set forth in
subparagraph (c)(1)(ii) of the Rights in Technical Data and Computer Software clause at
DFARS 252.227-7013.
Motorola, Inc.
Computer Group
2900 South Diablo Way
Tempe, Arizona 85282
Page 3
Preface
The MVME2600/2700 Series Single Board Computer Programmer’s Reference Guide
provides brief board level information, complete memory maps, and detailed ASIC chip
information includ ing regi ster b it desc riptions f or the MVME2600 and MVME2700 se ries
Single Board Computers. The information contained in this manual applies to the single
board computers built from some of the plug-together components in the following list:
OEM systems, supply addi ti ona l capability to an exi st ing compa ti bl e s yst em, or work in a
lab environment for experimental purposes.
A basic knowledge of computers and digital logic is assumed.
To use this manual, you should be familiar with the publications listed in Appendix A,
Related Documentation.
The following conventions are used in this document:
bold
is used for user input that you type just as it appears. Bold is also used for
commands, options and arguments to commands, and names of programs,
directories, and files.
italic
is used for names of variables to which you assign values. Italic is also used for
comments in screen displays and examples.
Page 4
courier
is used for system output (e.g., screen displays, reports), examples, and system
prompts.
<RETURN> or <CR>
represents the carriage return or enter key.
CTRL
represents t h e C ont ro l key. Execute c on trol c ha rac te rs b y pr es si ng th e CTRL key
and the letter simultaneously, e.g., CTRL-d.
The computer programs stored in the Read Only Memory of this device contain material
copyrighted by Motorola Inc., first published 1990, and may be used only under a license
such as the License f or Computer Progra ms (Article 14) contained in Moto rola’s Terms and
Conditions of Sale, Rev. 1/79.
All Motorola PWBs (printed wiring boards) are manufactured by UL-recognized
manufacturers, with a flammability rating of 94V-0.
This equipment generates, uses, and can radiate electro- magnetic
!
WARNING
energy. It may cause or be susceptible to electro-magnetic
interference (EMI) if not installed and used in a cabinet with
adequate EMI protection.
Motorola® and the Motorola symbol are registered trademarks of Motorola, Inc.
TM
PowerStack
, VMEmoduleTM, and VMEsystemTM are trademarks of Motorola, Inc.
TM
PowerPC
, PowerPC 603TM, and PowerPC 604TMare trademarks of IBM Corp, and are
used by Motorola, Inc. under license from IBM Corp.
TM
AIX
Timekeeper
is a trademark of IBM Corp.
TM
and ZeropowerTM are trademarks of Thompson Components.
All other products ment io ned i n this document are trademarks or re gi st ered trademarks of
their respective holders.
The following general safety precautions must be observed during all phases of operation, service, and repair of this
equipment. Failure to comply with these prec autions or with spe cific warnings els ewhere in this manua l violates saf ety
standards of design, manufacture, and intended use of the equipment. Motorola, Inc. assumes no liability for the
customer’s failure to comply with these requirements.
The safety precaut ions listed be low represent warnings of ce rtain danger s of which Mot orola is awar e. You, as the
user of the product, should follow these warnings and all other safety precautions necessary for the safe operation of
the equipment in your operating environment.
Ground the Instrument.
To minimize shock hazard, the equipment chassis and enclosure must be connected to an electrical ground. The
equipment is supp lied with a three- conductor ac pow er cable. The power cable must be plu gged into an appro ved
three-contact electrical outlet. The power jack and mating plug of the power cable meet International Electrotechnical
Commission (IEC) safety standards.
Do Not Operate in an Explosive Atmosphere.
Do not operate the equipment in the presence of flammable gases or fumes. Operation of any electrical equipment in
such an environment constitutes a definite safety hazard .
Keep Away From Live Circuits.
Operating personnel must not remove equipment covers. Only Factory Authorized Service Personnel or other
qualified maint enance personnel may remove equipment covers for internal subassembly or com ponent replacement
or any internal adjustment. Do not replace components with power cable connected. Under certain conditions,
dangerous voltages may exist even with the power cable removed. To avoid injuries, always disconnect power and
discharge circuit s before touching them.
Do Not Service or Adjust Alone.
Do not attempt internal service or adjustment unless another person capable of rendering first aid and resuscitation is
present.
Use Caution When Exposing or Handling the CRT.
Breakage of the Cathode-Ray Tube (CRT) causes a high-velocity scattering of glass fragments (implosion). To
prevent CRT implosion, avoid rough handling or jarring of the equipment. Handling of the CRT should be done only
by qualified mainte nance personnel using appr oved safety mask and gloves.
Do Not Substitute Parts or Modify Equipment.
Because of the dan ger of i ntrodu cing add ition al haza rds, do not ins tall sub stitut e parts or perf orm an y unauth orized
modification of the equipment. Con tact your local Mot or ola representative for service and re pa ir to ensure that safety
features are maintained.
Dangerous Procedure Warnings.
W arn ings , such as th e exa mple be low, precede potentially dangerous p roce dure s thro ughou t this ma nual . Instr uctio ns
contained in the warnings m ust be follow ed. You should also employ all ot her safety precautions w hich you dee m
necessary for the operation of the equi pm ent in your operating en vi ronment.
Dangerous voltages, capable of causing death, are present in this
!
WARNING
equipment. Use extreme caution when handling, testing, and
adjusting.
Table A-3. Related Specifications ........................................................................A-10
xviii
Page 18
1Board Description and Memory
Revision Note
This manual has been updated for the following reasons:
❏ To reflect that, since the differences between the MVME2600 and
the MVME2700 series SBCs are primarily related to the processor
and the implementation of L2 cache, the two share th e same
programming model. While much of this manual refers specifically
to MVME2600 hardware, the programming information is
applicable to both the MVME2600 and MVME2700 families.
❏ To reflect that a Universe II chip is available and that it includes
fixes for PCI reset problems which existed in the earlier version of
the Universe chip. For more details, see PCI Reset Problems Associated with the Initial Version of the Universe Chip on page
4-14.
Introduction
Maps
1
This manual provides programming infor mat ion for the MVME2600 and
MVME2700 Single Board Computers (SBCs). Extensive programming
information is pr ovided for several Appl ication-Specific I ntegrated Circuit
(ASIC) devices used on the boards. Reference information is included in
Appendix A for the Large Scale Integration (LSI) devices used on the
boards and sources for additional information are listed.
This chapter briefly describes the board level hardware features of the
MVME2600 series Single Board Computers. The chapter begins with a
board level overvi ew and feat ures lis t. Memo ry maps are n ext, and are the
major feature of this chapter.
1-1
Page 19
1
Board Description and Memory Maps
Programmable registers in the MVME2600/2700 series that reside in
ASICs are covered in the chapters on those ASICs. Chapter 2 covers the
Raven chip, Chapter 3 cover s the Falcon chip set, Chapter 4 covers the
Universe chip, and Chapter 5 covers certain programming features, such
as interrupts and exceptions. Appendix A lists all related documentation.
Manual Terminology
Throughout this manual, a convention is used which precedes data and
address parameters by a character identifying the numeric format as
follows:
$
dollar
%
percent
&
ampersand
For example, “12” is t he decimal numbe r twelve, and “$12” is the decimal
number eighteen.
Unless otherwise specified, all address references are in hexadecimal.
specifies a hexadecimal character
specifies a binary number
specifies a decimal number
1-2
An asterisk (*) following the signal name for signals which are level
significant denotes that the signal is true or valid when the signal is low.
An asterisk (*) following the signal name for signals which are edge
significant denotes that the acti ons initiat ed by that s ignal occur on h igh to
low transitio n.
NoteIn some places in this docum ent, an undersc ore (_) foll owing
the signal name is used to indicate an active low signal.
In this manual, assertion and negation are used to specify forcing a signal
to a particula r sta te. In pa rtic ular, asser ti on and a ssert ref er to a signal that
is active or true; negation and negate indicate a signal that is inactive or
false. These terms ar e used independently of the vo ltage level (high or l ow)
that they represent.
Data and address sizes for MPC60x chips are defined as follows:
Page 20
Overview
❏ A byte i s eight bits, numb ered 0 through 7, wit h bit 0 being the leas t
significant.
❏ A half-word is 16 b its, number ed 0 thr oug h 15, with bit 0 b eing th e
least significant.
❏ A word or single wor d is 32 bits , numbered 0 t hrough 31, wit h bit 0
being the least significant.
❏ A double word is 64 bits, numbered 0 through 63, with bit 0 being
the least significant.
Refer to Chapter 5 for Endian Issues, which covers which parts of the
MVME2600/2700 series use big-endian byte ordering, and which use small-endian byte ordering.
The terms control bit and status bit are used extensively in this document.
The term control b it is used to d escribe a bit in a register that can be set and
cleared under softwar e c ont rol . The te rm tr ue is used to indica te tha t a bit
is in the state that enables the function it controls. The term false is used to
indicate that the bit is in the state that disables the function it controls. In
all tables, th e terms 0 and 1 are used to describe the actual va lue that should
be written to the bit, or the value that it yields when read. The term status bit is us ed to describe a bit in a register that reflects a specific condition.
The status bit can be read by software to determine operational or
exception conditions.
1
Overview
The MVME2600/2700 series SBC families provide many standard
features required by a computer system: SCSI, Ethernet interface,
keyboard interface, mouse interface, sync and async serial ports, parallel
port, boot Flash, and up to 256MB of ECC DRAM.
1-3
Page 21
1
Board Description and Memory Maps
Feature Summary
There are many models based on the MVME2600/27 00 series architecture.
The following table summarizes the major features of the MVME2600
series:
Table 1-1. MVME2600 Series Features Summary
FeatureDescription
ProcessorsSingle
Supports BGA processors only:
MVME2600: MPC603, MPC604.
MVME2700: MPC750
Bus Clock Frequencies up to 66MHz
L2 CacheBuild-option for 256KB Look-aside L2 Cache
Flash4MB or 8MB (64-bit wide), with socketed 1MB (16-bit wide)
DRAM16MB to 256MB, ECC Protected (Single-bit Correction, Double-bit
The MVME2600/2700 series provides the 256KB look-aside external
cache option. The Falcon chip set controls the boot Flash and the ECC
DRAM. The Raven ASIC functions as the 64-bit PCI host bridge and the
MPIC interrupt contro ller. PCI devices include: SCSI, VME, Ethernet, and
one PMC slot. Standar d I/O functions are provided by the S uper I/O device
which resides on the ISA bus. The NVRAM/RTC and the optional
synchronous serial ports also reside on the ISA bus. The general system
block diagram for MVME2600/2700 series is shown below:
System Block Diagram
1
1-5
Page 23
1
Board Description and Memory Maps
CLOCK
GENERATOR
PHB & MPIC
RAVEN ASIC
64-BIT PMC SLOT
L2 CACHE
256K
PROCESSOR
MPC603/604
REGISTERS
ISA
FLASH
1MB
MEMORY CONTROLLER
66MHz MPC604 PROCESSOR BUS
33MHz 32/64-BIT PCI LOCAL BUS
PIB
W83C553
ISA BUS
ETHERNET
DEC21140
AUI/10BT/100BTX
RTC/NVRAM/WD
MK48T559
FALCON CHIPSET
MEMORY EXPANSION CONNECTORSDEBUG CONNECTOR
SCSI
53C825A
FLASH
4MB or 8MB
SYSTEM
REGISTERS
VME BRIDGE
UNIVERSE
BUFFERS
PCI EXPANSION
1-6
MOUSEKBDFLOPPY & LED
PMC FRONT I/O SLOT
SUPER I/O
PC87308
PARALLEL
FRONT PA N EL
ESCC
85230
SERIAL
712/761 P2 I/O OPTIONS
VME P2VME P1
CIO
Z8536
Figure 1-1. MVME2600 Series System Block Diagram
11536.00 9611
Page 24
Functional Description
Overview
The MVME2600/2700 series is a family of single-slot SBCs. The
MVME2600 family consists of t he MPC603/604 processor, the Ra ven PCI
Bridge & Interrupt Controller, the ECC Memory Controller Falcon
chipset, 5MB or 9MB of Flash memory, 16MB to 256MB of ECCprotected DRAM, and a rich set of features of I/O peripherals. The
MVME2700 family offers the same hardware features, but with the
MCP750 processor.
I/O peripheral devices on the PCI bus are: SCSI chip, Ethernet chip,
Universe VMEbus interface ASIC, and one PMC slot . Functions prov ided
from the ISA bus are: a P1 284/Parallel port, two asynchronous serial por ts,
two sync/async serial ports, a real-time clock, and counters/timers.
The MVME2600/2700 series board interfaces to the VMEbus via the P1
and P2 connectors, which use the new 5-row 160-pin connectors as
specified in the propos ed VME64 Exten sion Sta ndard. I t also draws +5V,
+12V, and -12V power from the VMEbus backplane through these two
connectors. 3.3V supply is regulated onboard from the +5V power.
Functional Description
1
Front panel connector s on the MVME2600 /2700 seri es board include: a 6 pin circular DIN conn ector for the keyboar d interface, a 6-pin circular DIN
connector for the mouse interface, and a 50-pin connector for the floppy
and status LEDs. All signals fo r the serial ports, the P1284/Pri nter port, the
SCSI interface, and the Ethernet interface are routed to P2.
There are two P2 I/O options supported by the MVME2600/2700 series:
the MVME712M mode and the MVME761 mode. The MVME761 mode
provides the enhanced P1284 parallel port interface and the full
synchronous support for Serial Ports 3 and 4. The MVME712M version
provides backward comp at ibi lity wit h pr evi ous SBCs. I n ei ther mode, 16bit SCSI capability can only be used by systems with 5-row DIN support
because the additional 8 bits of SCSI data lines reside on row Z of P2.
1-7
Page 25
1
Board Description and Memory Maps
The MVME2600/2700 series contains one IEEE1386.1 PCI Mezzanine
Card (PMC) slot. This PMC s lot i s 64-bit capable and su ppor ts bo th front
and rear I/O. Pins 1 through 30 of th e PMC connector J14 are routed to pins
D1 through D30 of the 5-row DIN P2 connector. J14 pin 31 is connected
to P2 pin Z29, and J14 pin 32 is connected to P2 pin Z31.
Additional PCI expansion is supported with a 114-pin Mictor connector.
This connection allows stacking of a carrier board to increase the I/O
capability, such as a dual-PMC carrier board.
Programming Model
Memory Maps
The following sections describe the memory maps for the
MVME2600/2700 series.
Processor Memory Maps
1-8
The Processor memory map is controlled by the Raven ASIC and the
Falcon chipset. The Raven ASIC and the Falcon chipset have flexible
programming Map Decoder registers to customize the system to fit many
different applications.
Page 26
Programming Model
Default Processor Memory Map
After a reset, the Raven ASIC and the Falcon chipset provide the default
processor memory map as shown in the following table.
FEF8 0000FEF8 FFFF64KFalcon Registers
FEF9 0000FEFE FFFF384KNot mapped
FEFF 0000FEFF FFFF64KRaven Registers
FF00 0000FFEF FFFF15MNot mapped
FFF0 0000FFFF FFFF1MROM/FLASH Bank A or Bank B2
SizeDefinition
Not mapped
- 640K
Notes:
1. This default map for PCI/ISA I/O space allows software to
determine if the system is MPC105-based or Falcon/Raven-based
by examining either the PHB Device ID or the CPU Type Register.
2. The first one Mbyte of ROM/FLASH Bank A appears at this range
after a reset if the rom_b_rv control bit is cleared. If the rom_b_rv
control bit is set t hen this address range maps t o ROM/FLASH Bank
B.
Notes
1-9
Page 27
1
Board Description and Memory Maps
Processor CHRP Memory Map
The following table shows a recommended CHRP memory map from the
point of view of the processor.
FD00 0000FDFF FFFF16MZero-Based PCI/ISA Memory Space
(mapped to 00000000 to 00FFFFFF)
FE00 0000FE7F FFFF8MZero-Based PCI/ISA I/O Space
(mapped to 00000000 to 007FFFFF)
FE80 0000FEF7 FFFF7.5MR eserved
FEF8 0000FEF8 FFFF64KFalcon Registers
FEF9 0000FEFE FFFF384KReserved
FEFF 0000FEFF FFFF64KRaven Registers9
FF00 0000FF7F FFFF8MROM/FLASH Bank A1,7
FF80 0000FF8F FFFF1MROM/FLASH Bank B1,7
FF50 0000FFEF FFFF6MReserved
Notes
3,4,8
3,8
3,5,8
FFF0 0000FFFF FFFF1MROM/FLASH Bank A or Bank B7
Notes:
1. Programmable via Falcon chipset.
2. To enable the “Processor-hole” area, program the Falcon chi pset to
3. Programmable via Raven ASIC.
1-10
ignore 0x000A0000 - 0x000BFFFF add ress range and progra m the
Raven to map this address range to PCI memory space.
Page 28
Programming Model
4. CHRP requires the starting address for t he PCI memory space to b e
256MB-aligned.
5. Programmable via Raven ASIC for either contiguous or spr ead-I/O
mode.
6. The actual size of each ROM/FLASH bank may vary.
7. The first one Mbyte of ROM/FLASH Bank A appears at this range
after a reset if the rom_b_rv control bit is cleared. If the rom_b_rv
control bit is set t hen this address range maps t o ROM/FLASH Bank
B.
8. This range can be mapped to the VMEbus by programming the
Universe ASIC accordingly. The map shown is the recommended
setting which uses the Sp ecial PCI Sl ave Ima ge and two of the f our
programmable PCI Slave Images.
9. The only method to generate a PCI Interrupt Acknowledge cycle
(8259 IACK) is to perform a read access to the Raven’s PIACK
register at 0xFEFF0030.
The following table shows the programmed values for the assoc- iated
Raven MPC registers for the processor CHRP memory map.
1
Table 1-4. Raven MPC Register Values for CHRP Memory Map
The Raven/Falcon chipset can be programmed for PREP-compatible
memory map. The following table shows the PREP memory map of the
MVME2600/2700 series from the point of view of the processor.
PCI Configuration acc esses ar e accomplis hed via t he CONFIG_ADD and
CONFIG_DAT registers. These two r egisters are implemented b y the
Raven ASIC. In the CHRP memory map example, the CONFIG_ADD and
CONFIG_DAT registers are located at 0xFE000CF8 and 0xFE000CFC,
respectively. With the PREP memory map, the CONFIG_ADD register
and the CONFIG_DAT register are located at 0x80000CF8 and
0x80000CFC, respectively.
1-13
Page 31
1
Board Description and Memory Maps
PCI Memory Maps
The PCI memory map is controlled by the Raven ASIC and the Universe
ASIC. The Raven ASIC and the Universe ASIC have flexible
programming Map Decoder registers to customize the system to fit many
different applications.
Default PCI Memory Map
After a reset, the Raven ASIC and the Univer se ASIC turn all the PCI slave
map decoders off. Software must program the appropriate map decoders
for a specific environment.
PCI CHRP Memory Map
The following table shows a PCI memory map of the MVME2600/2700
series that is CHRP-compatible fro m the point of view of the PCI local bu s.
The processor can access any address range in the VMEbus with the help
from the address translation capabilities of the Universe ASIC. The
recommended mapping is shown in the Processor Memory Map section.
The following figure illustrates how the VMEbus master mapping is
accomplished.
Programming Model
1
1-21
Page 39
1
Board Description and Memory Maps
ONBOARD
MEMORY
PCI MEMORY
SPACE
PCI/ISA
MEMORY SPACE
PCI
I/O SP AC E
NOTE 1
NOTE 1
PCI MEMORYPROCESSOR
NOTE 2
NOTE 3
VMEBUS
PROGRAMMABLE
SPACE
VME A24
VME A16
VME A24
VME A16
VME A24
VME A16
VME A24
VME A16
MPC
RESOURCES
Notes:
1. Programmable mapping done by the Raven ASIC.
1-22
11553.00 9609
Figure 1-2. VMEbus Master Mapping
Page 40
Programming Model
2. Programmable mapping via the four PCI Slave Images in the
Universe ASIC.
3. Programmable mapping via the Special Slave Image (SLSI) in the
Universe ASIC.
VMEbus Slave Map
The four programmable VME Slave Images in the Universe ASIC allow
other VMEbus masters to get to any devices on the MVME2600/2700
series. The combination of the four Universe VME Slave Images and the
four Raven PCI Slave Decoders offers a lot of flexibility for mapping the
system resources as seen from the VMEbus. In most applications, the
VMEbus only needs to s ee the syste m memory and , pe rhaps, the so ftware
interrupt registe rs (SI R1 and SIR2 regis ters) . An exampl e of the VMEbus
slave map is shown below:
1
1-23
Page 41
1
Board Description and Memory Maps
Processor
Onboard
Memory
ISA Space
Software INT
Registers
NOTE 2
NOTE3
PCI Memory
PCI I/O Space
VMEbus
NOTE 1
NOTE 1
1896 9609
Notes:
1. Programmable mapping via the four VME Slave Images in the
2. Programmable mapping via PCI Slave Images in the Raven ASIC.
3. Fixed mapping via the PIB device.
1-24
Figure 1-3. VMEbus Slave Mapping
Universe ASIC.
Page 42
Programming Model
The following table shows the programmed values for the associated
Universe registers for the VMEbus slave function.
Table 1-13. Universe PCI Register Values for VMEbus Slave Map Example
The above register values yield the following VMEbus slave map:
1-25
Page 43
1
Board Description and Memory Maps
Table 1-14. VMEbus Slave Map Example
VMEbus Address
RangeMode
4000 0000 4000 0FFF
1000 0000 1FFF FFFF
A32 U/S/P/D
D08/16/32
A32 U/S/P/D
D08/16/32/64
RMW
SizeCHRP MapPREP Map
4KP CI/IS A I/O Spac e:
0000 1000 - 0000 1FFF
256MPCI/ISA Memory Space
(On-board DRAM)
0000 0000 - 0FFF FFFF
PCI/ISA I/O Space:
0000 1000 - 0000 1FFF
PCI/ISA Memory Space
(On-board DRAM)
8000 0000 - 8FFF FFFF
1-26
Page 44
Falcon-Controlled System Registers
The Falcon chipset latches the states of the DRAM data lines onto the
PR_STAT1 and PR_STAT2 registers. The MVME2600/2700 series use
these status registers to provide the system configuration information. In
addition, the Falcon chipset performs the decode and control for an
external register port. This function is utilized by the MVME2600/2700
series to provide the system control registers.
Table 1-15. System Register Summary
0
BIT # ---->
FEF80400
FEF80404
FEF88000
FEF88300
123
System External Cache
Control Register
CPU C
4
5
6789101112131415161718192021222324252627282930
System Configuration Register (Upper Falcon’s PR_STAT1)
The following sub-sections describe these system registers in detail.
Programming Model
1
31
1-27
Page 45
1
Board Description and Memory Maps
System Configuration Register (SYSCR)
The states of the RD[0:31] DRAM data pins, which have weak internal
pull-ups, are l atched by the upper Falcon chip at a ris ing edge of th e powerup reset and stored in this System Configu ration Regist er to provid e some
information about t he system. Configuration is accomplished wit h external
pull-down resistors. This 32-bit read-only register is defined as follows:
REGSystem Configuration Register - $FEF80400
BIT
FIELD
OPER
RESET
012345678
SYSIDSYSCLKSYSXCP0STATP1STAT
$FEXXXX$F$F
SYSIDSystem Identification. This field specifies the type of the
SYSCLKSystem Clock Speed. This field relays the syst em clock
overall system configuration so that the software may
appropriately hand le any softwa re visible differences . For
the MVME2600/2700 series, this field returns a value of
$FE.
speed and the PCI clock speed information as follows:
31
SYSXCSystem External Cache S ize. This field reflec ts size of the
0b0000 to 0b1011Reserved
1-28
look-aside cache on the system bus.
SYSXC ValueExternal Look-aside Cache Size
Page 46
Programming Model
SYSXC ValueExternal Look-aside Cache Size
0b11001M
0b1101512K
0b1110256K
0b1111None
P0/1STAT Processor 0/1 Status. This field is encoded as follows:
1
P0/1STAT ValueProcessor 0/1 Present
0b0000 to 0b0011ReservedReserved
0b0100YES1M
0b0101YES512K
0b0110YES256K
0b0111YESNone
0b1000 to 0b1111NON/A
External In-line Cache
Size
1-29
Page 47
1
Board Description and Memory Maps
Memory Configuration Register (MEMCR)
The states of the RD[00:31] DRAM data pins, which have weak internal
pull-ups, are lat ched by the lowe r Falcon chip a t a rising edge of the powerup reset and s tored in this Memory Configuration Regis ter to provide some
information about the system memory. Configuration is accomplished
with external pull-down re sistors. This 32-bit read-onl y Register is defined
as follows :
These two bi ts reflect the co mbined status of the four
blocks of DRAM. Initialization software uses this
information to program the ram_spd0
and ram_spd1
control bits in the Falcon’s Chip Revision Register.
R_A/B_TYP[0:1]ROM/Flash Type. This field is encoded as
follows:
ROM_A/B_TYP[0:2]ROM/FLASH Type
0b000 to 0b101Reserved
1
0b110
0b111Unknown type (i.e. ROM/FLASH Sockets)
Intel 16-bit wide FLASH with 16K Bottom Boot
Block
Note: The device width i s dif ferent f rom the widt h of t he FLASH bank . If
the bank width is 64-bit and the device width is 16-bit then the FLASH
bank consists of four FLASH devices.
System External Cache Control Register (SXCCR)
The System Cache Control Register is accessed via the RD[32:39] data
lines of the upper Falcon device. This 8-bit register is defined as follows:
REGSystem External Cache Control Register - $FEF88000
BIT01234567
SXC_DIS_
FIELD
OPERR/W
SXC_RST_
SXC_MI_
SXC_FLSH_
RESET1111XXXX
SXC_DIS_ System External Cache Enable. When this bit is cleared,
it disables this cache from responding to any bus cycles.
1-31
Page 49
1
Board Description and Memory Maps
SXC_FLSH_System External Cache Flush. When this bit is pulsed
true for at least 8 clock periods, it causes the system
external cache to write back dirty cache lines out to
system memory and clears all the tag valid bits. This
operation causes the Glance pair to r eque st and hold the
MPC bus until it has completed the flush operation
(approximately 4100 clock cycles). This may be an issue
if other devices cannot wait that long to become MPC bus
master .
SXC_RST_System External Cache Reset. When this bit is cleared, it
invalidates all tags and holds the cache in a reset
condition. There is a bug in Glance - It really does not
hold the chip in a reset condition. The tag invalidate still
works okay though.
SXC_MI_ System External Cache Miss Inhibit. When this bit is
cleared, it prevents line fills on cache misses.
Software should never clear more than one of these bi ts at the
!
Warning
same time. If more than one is cleared at the same time, the
Glance pair behaves indeterminately.
1-32
Page 50
Programming Model
CPU Control Register
The CPU Control Register is accesse d via the RD[32 :39] dat a lines of the
upper Falcon device. This 8-bit register is defined as follows:
REGCPU Control Register - $FEF88300
BIT01234567
LEMODE
FIELD
OPERRRR/WR/WRRRR
RESETX011XXXX
P1_TBEN
P0_TBEN
LEMODE Little Endian Mode. This bit must be set in conjunction
with the LEND bit in the Raven for little-endian mode.
P0/1_TBENProcessor 0/1 T ime Base Enable . When this bit is clea red,
the TBEN pin of Processor 0/1 will be driven low.
1
1-33
Page 51
1
Board Description and Memory Maps
ISA Local Resource Bus
W83C553 PIB Registers
The PIB contains ISA Bridge I/O registers for various functions. These
registers are actually accessible from the PCI bus. Refer to the W83C553
Data Book for details.
PC87308VUL Super I/O (ISASIO) Strapping
The PC87308VUL Super I/O (ISASIO) provides the following functions
to the MVME2600/2700 series: a keyboard interface, a PS/2 mouse
interface, a PS/2 floppy port, two async serial ports and a parallel port.
Refer to the PC87308VUL Data Sheet for additional details and
programming information.
The following table shows the hardware strapping for the Super I/O
device:
Table 1-16. Strap Pins Configuration for the PC87308VUL
PinsReset Configuration
CFG00 - FDC, KBC and RTC wake up inactive.
CFG11 - Xbus Data Buffer (XDB) enabled.
CFG3, CFG200 - Clock source is 24MHz fed via X1 pin.
BADDR1, BADDR211 - PnP Motherboard, Wake in Config State. Index $002E.
SELCS1 - CS0# on CS0# pin.
NVRAM/RTC & Watchdog Timer Registers
The MK48T59/559 provides t he MVME2600/2700 s eries wit h 8K of non volatile SRAM, a time-of-day clock, and a watchdog timer. Accesses to
the MK48T59559 are accomplished via three registers: The
NVRAM/RTC Address Strobe 0 Register, the NVRAM/RTC Address
1-34
Page 52
ISA Local Resource Bus
Strobe 1 Register, and the NVRAM/RTC Data Port Register. The
NVRAM/RTC Address Strobe 0 Register latches the lower 8 bits of the
address and the NVRAM/RTC Ad dress Strobe 1 Register latches the upper
5 bits of the address.
The following sub sections describe these registers in detail.
CPU Configuration Register
The CPU Configuration Register is an 8-bit register located at ISA I/O
address x0800. This register is def ined for the MVME2600/ 2700 series to
provide some backward compatibility with older MVME1600 products.
The Base Module Status Register should be used to identify the base
module type and the System Configuration Register should be used to
obtain information about the overall system.
REGOld CPU Configuration Register - $FE000800
BITSD7SD6SD5SD4SD3SD2SD1SD0
FIELDCPUTYPE
OPERRR
RESET$E$F
CPUTYPE CPU Type. This field will always read as $E for the
MVME2600/2700 series. The System Configuration
Register should be used for additional information.
1-36
Page 54
ISA Local Resource Bus
Base Module Feature Register
The Base Module Feature Register is an 8-bit register providing the
configuration info rmation about th e Genesis Single Board Computer. This
read-only register is located at ISA I/O address x0802.
REGBase Module Feature Register - Offset $0802
BITSD7SD6SD5SD4SD3SD2SD1SD0
FIELD
OPERRRRRRRRR
RESETXXXXXXXX
SCCP_
PMC2P_
PMC1P_
VMEP_
GFXP_
LANP_
SCCP_Z85230 ESCC Present. If set, there is no on-board sync
serial support. If cleared, there is on-board support for
sync serial interface.
1
SCSIP_
PMC2P_PMC/PCIX Slot 2 Present. If set, there is no PMC/PCIX
device installed in the PMC/PCIX Slot 2. If cleared, the
PMC/PCIX Slot 2 contains a PCI Mezzanine Card or a
PCI device.
PMC1P_PMC Slot 1 Present. If set, there is no PCI Mezzanine
Card installed in the PMC Slot 1. If cleared, the PMC Slot
1 contains a PMC.
VMEP_VMEbus Present. If set, ther e i s no VM Ebus int er fac e. I f
cleared, VMEbus interface is supported.
GFXP_Graphics Present. If set, there is no on-board Graphics
interface. If cleared, there is an on-board graphics
capability.
LANP_Ethernet Present. If set, there is no Ethernet transceiver
interface. If cleared, there is on-board Ethernet su pport.
1-37
Page 55
1
Board Description and Memory Maps
SCSIP_SCSI Present. If set, there is no on-board SCSI interface.
If cleared, on-board SCSI is supported.
Base Module Status Register (BMSR)
The Base Module Status Register is an 8-bit read-only register located at
ISA I/O address x0803.
REGBase Module Status Register - Offset $0803
BITSD7SD6SD5SD4SD3SD2SD1SD0
FIELDBASE_TYPE
OPERR
RESETN/A
BASE_TYPEBase Module Type. This eight bit field is used to
provide the category of the base mo dule and is def ined as
follows:
1-38
BASE_TYPE ValueBase Mo dule Type
$0 to $F9Reserved
$FA
$FBMVME2600/2700 with MVME712M I/O
$FCMVME2600/2700 with MVME761 I/O
$FDMVME3600 with MVME712M I/O
$FEMVME3600 with MVME761 I/O
$FFMVME1600-001 or MVME1600-011
Reserved -- Special with MVME712M I/O and
100BaseT4 on row Z
Page 56
ISA Local Resource Bus
Seven-Segment Display Register
This 16-bit register allows data to be sent to the 4-digit hexadecimal
diagnostic display. The register al so a ll ows the data to be read
back.
REG7-Segment Display Register - Offset $08C0
SD15
SD14
SD13
BIT
FIELDDIG3[3:0]DIG2[3:0]DIG1[3:0]DIG0[3:0]
OPERR/W
RESETXXXXXXXXXXXXXXXX
SD12
SD10
SD11
SD9
SD8
SD7
SD6
SD5
SD4
SD3
SD2
SD1
DIG3[3:0] Hexadecimal value of the most significant digit.
DIG2[3:0] Hexadecimal value of the third significant digit.
DIG1[3:0] Hexadecimal value of the second significant digit.
1
SD0
DIG0[3:0] Hexadecimal value of the least significant digit.
VME Registers
The following registers provide the following functions for the VMEbus
interface: a software interrupt capability, a location monitor function, and
a geographical a ddress st atus. For these regist ers to b e accessi ble fro m the
VMEbus, the Universe ASIC must be programmed to map the VMEbus
Slave Image 0 into the appropriate PCI I/O address range. Refer to the
VMEbus Slave Map section for additional details. The following table
shows the registers provided for various VME functions:
1-39
Page 57
1
Board Description and Memory Maps
Table 1-19. VME Registers
PCI I/O AddressFunction
0000 1000SIG/LM Control Register
0000 1001SIG/LM Status Register
0000 1002VMEbus Location Monitor Upper Base Address
0000 1003VMEbus Location Monitor Lower Base Address
0000 1004VMEbus Semaphore Register 1
0000 1005VMEbus Semaphore Register 2
0000 1006VMEbus Geographical Address Status
These registers are described in the following sub-sections.
LM/SIG Control Register
The LM/SIG Control Register is an 8-bit register located at ISA I/O
address x1000. This register provides a method to generate software
interrupts. The Universe ASIC is programmed so that this register can be
accessed from the VMEbus to generate software interrupts to the
processor(s).
REGLM/SIG Control Register - Offset $1000
BITSD7SD6SD5SD4SD3SD2SD1SD0
FIELDSET
SIG1
OPERWRITE-ONLY
RESET00000000
SET
SIG0
SET
LM1
SET
LM0
CLR
SIG1
CLR
SIG0
CLR
LM1
CLR
LM0
SET_SIG1 Writing a 1 to this bit will set the SIG1 status bit.
SET_SIG0 Writing a 1 to this bit will set the SIG0 status bit.
1-40
Page 58
ISA Local Resource Bus
SET_LM1 Writing a 1 to this bit will set the LM1 status bit.
SET_LM0 Writing a 1 to this bit will set the LM0 status bit.
CLR_SIG1Writing a 1 to this bit will clear the SIG1 status bit.
CLR_SIG0Writing a 1 to this bit will clear the SIG0 status bit.
CLR_LM1 Writing a 1 to this bit will clear the LM1 status bit.
CLR_LM0 Writing a 1 to this bit will clear the LM0 status bit.
LM/SIG Status Register
The LM/SIG Status Registe r is an 8-bit regi ster locate d at ISA I/O address
x1001. This register, in conjunction with the LM/SIG Control Register,
provides a method to generate interrupts. The Universe ASIC is
programmed so that this register can be accessed from the VMEbus to
provide a capability to generate software interrupts to the onboard
processor(s) from the VMEbus.
REGLM/SIG Status Register - Offset $1001
1
BITSD7SD6SD5SD4SD3SD2SD1SD0
FIELDEN
SIG1
OPERR/WREAD-ONLY
RESET00000000
EN
SIG0
EN
LM1
EN
LM0
SIG1SIG0LM1LM0
EN_SIG1 When the EN_SIG1 bit is set, a LM/SIG Interrupt 1 is
generated if the SIG1 bit is asserted.
EN_SIG0 When the EN_SIG0 bit is set, a LM/SIG Interrupt 0 is
generated if the SIG0 bit is asserted.
EN_LM1When the EN_LM1 bit is set, a LM/SIG Interrup t 1 is
generated and the LM1 bit is asserted.
EN_LM0When the EN_LM0 bit is set, a LM/SIG Interrup t 0 is
generated and the LM0 bit is asserted.
1-41
Page 59
1
Board Description and Memory Maps
SIG1SIG1 status bit. This bit can only be set by the SET_LM1
control bit. It can onl y be cle ar ed by a reset or by writing
a 1 to the CLR_LM1 control bit.
SIG0SIG0 status bit. This bit can only be set by the SET_LM0
control bit. It can onl y be cle ar ed by a reset or by writing
a 1 to the CLR_LM0 control bit.
LM1LM1 status bit. This bit can be set by either the location
monitor function or the SET_LM1 control bit. LM1
correspond to offset 3 from the location monitor base
address. This bit can only be cleared by a reset or by
writing a 1 to the CLR_LM1 control bit.
LM0LM0 status bit. This bit can be set by either the location
monitor function or the SET_LM0 control bit. LM0
correspond to offset 1 from the location monitor base
address. This bit can only be cleared by a reset or by
writing a 1 to the CLR_LM0 control bit.
Location Monitor Upper Base Address Register
The Location Monitor Upper Base Address Register is an 8-bit register
located at ISA I/O address x1002. The Universe ASIC is programmed so
that this register can be accessed from the VMEbus to provide VMEbus
location monitor function.
REGLocation Monitor Upper Base Address Register - Offset $1002
BITSD7SD6SD5SD4SD3SD2SD1SD0
FIELDVA15VA14VA13VA12VA11VA10VA9VA8
OPERR/W
RESET00000000
VA[15:8]Upper Base Address for the location monitor function.
1-42
Page 60
ISA Local Resource Bus
Location Monitor Lower Base Address Register
The Location Monitor Lower Base Address Register is an 8-bit register
located at ISA I/O address x1003. The Universe ASIC is programmed so
that this register can be accessed from the VMEbus to provide VMEbus
location monitor function.
REGLocation Monitor Lower Base Address Register - Offset $1003
VA[7:4]Lower Base Address for the location monitor function.
LMENThis bi t must be set to enable the location monitor
function.
1
Semaphore Register 1
The Semaphore Register 1 is an 8-bit register located at ISA I/O address
x1004. The Universe ASIC is programmed so that this register can be
accessible from the VMEbus. This register can only be updated if bit 7 is
low or if the new value has the most significant bit cleared. When bit 7 is
high, this register will not latch in the new value if the new value has the
most significant bit set.
REGSemaphore Register 1 - Offset $1004
BITSD7SD6SD5SD4SD3SD2SD1SD0
FIELDSEM1
OPERR/W
RESET00000000
1-43
Page 61
1
Board Description and Memory Maps
Semaphore Register 2
The Semaphore Register 2 is an 8-bit register located at ISA I/O address
x1005. The Universe ASIC is programmed so that this register can be
accessible from the VMEbus. This register can only be updated if bit 7 is
low or if the new value has the most significant bit cleared. When bit 7 is
high, this register will not latch in the new value if the new value has the
most significant bit set.
REGSemaphore Register 2 - Offset $1005
BITSD7SD6SD5SD4SD3SD2SD1SD0
FIELDSEM2
OPERR/W
RESET00000000
VME Geographical Address Register (VGAR)
The VME Geographical Address Register is an 8-bit read-only register
located at ISA I/O address x1006. This register reflects the states of the
geographical address pins at the 5-row, 160-pin P1 connector.
The Z85230 ESCC is used to provide the two sync/async serial ports on
some MVME2600/2700 series models. The PCLK which can be used to
derived the baud rates, is 10 MHz. Refer to the SCC User’s Manual for
programming information on the Z85230 ESCC device.
1-44
Page 62
The Z8536 CIO is used to provide the modem control lines not provided
by the Z85230 ESCC and a method to inquire the module ID of the two
sync/async seria l port s that resi de o n the MVME76 1 module. Re fe r to th e
Z8536 Data Sheet for programming information.
Z8536/Z85230 Registers
Accesses to the Z8536 CIO and the Z85230 ESCC are accomplished via
Port Control and Port Data Registers. The PCLK to the Z8536 is 5 MHz.
Also, a Pseudo IACK Register is also de fined to retri eve int errup t vecto rs
from these devices. The Z8536 CIO has higher priority than the Z85230
ESCC in the interrupt daisy chain. The following list the registers
associated with accessing these two devices:
Table 1-20. Z8536/Z85230 Access Registers
PCI I/O AddressFunction
0000 0840Z85230: Port B (Serial Port 4) Control
0000 0841Z85230: Port B (Serial Port 4) Data
ISA Local Resource Bus
1
0000 0842Z85230: Port A (Serial Port 3) Control
0000 0843Z85230: Port A (Serial Port 3) Data
0000 0844Z853 6 CIO: Port C’s Data Regist er
0000 0845Z8536 CIO: Port B’s Data Register
0000 0846Z8536 CIO: Port A’s Data Register
0000 0847Z8536 CIO: Control Register
0000 084FZ85230/Z8536 Pseudo IACK
1-45
Page 63
1
Board Description and Memory Maps
Z8536 CIO Port Pins
The assignment for the Port pins of the Z8536 CIO is as follows:
Table 1-21. Z8536 CIO Port Pins Assignment
Port
Pin
PA0TM3_
PA1DSR3_
PA2RI3_InputPort 3 Ring Indicator
PA3LLB3_
PA4RLB3_OutputPort 3 Remote Loopback
PA5DTR3_OutputP ort 3 Data Terminal Ready
PA6BRDFAILOutputBoard Fail: When set will cause FAIL LED to be lit.
PA 7IDREQ_OutputModule ID Request - low true
PB0TM4_
PB1DSR4_
Signal
Name
MID0
MID1
MODSEL
MID2
MID3
DirectionDescriptions
InputPort 3 Test Mode when IDREQ_ = 1;
Module ID Bit 0 when IDREQ_ = 0.
InputPort 3 Data Set Ready when IDREQ_ = 1;
Module ID Bit 1 when IDREQ_ = 0.
OutputPort 3 Local Loopback (IDREQ_ = 1) or
Port Select (IDR EQ_ = 0):
IDREQ_ = 0 & MODSEL = 0 => Port 3 ID Select
IDREQ_ = 0 & MODSEL = 1 => Port 4 ID Select
InputPort 4 Test Mode when IDREQ_ = 1;
Module ID Bit 2 when IDREQ_ = 0.
InputPort 4 Data Set Ready when IDREQ_ = 1;
Module ID Bit 3 when IDREQ_ = 0.
PB2RI4_InputPort 4 Ring Indicator
PB3LLB4_OutputPort 4 Local Loopback
PB4RLB4_OutputPort 4 Remote Loopback
PB5DTR4_OutputPort 4 Data Terminal Ready
PB6FUSEInputFUSE = 1 means that at least one of the fuses or
PB7ABORT_InputStatus of ABORT# signal
1-46
polyswitches is open.
Page 64
ISA Local Resource Bus
Table 1-21. Z8536 CIO Port Pins Assignment (Continued)
NOTE: The direction and the polarity of th e Z8536’s port pins are software
programmable.
1-47
Page 65
1
Board Description and Memory Maps
The module ID signals, which are only valid when IDREQ_ is asserted,
indicate the type of the serial module that is installed on either Port 3 or
Port 4. The following table shows how to interpret the MID3-MID0
signals:
NoteBecause IDREQ_ and MID3-MID0 signals go through the
P2MX (P2 multiplexing) function used on
MVME2600/2700 series modules configured for the
MVME761-type transition module, software must wait for
the MID3-MID0 to become valid afte r asserting IDREQ_.
The waiting time should be about 4 microseconds because
the sampling rate is about 1.6 microsecond with a 10MHz
MXCLK clock.
1-48
Page 66
ISA DMA Channels
There are seven ISA DMA channels in the PIB. Channels 0 through 3
support only 8-bit DMA devices whi le Channels 5 through 7 suppor t only
16-bit DMA devices. These DMA channels are assigned as follows:
Table 1-23. PIB DMA Channel Assignments
ISA Local Resource Bus
1
PIB
Priority
HighestChannel 0DMA1Serial Port 3 Receiver (Z85230 Port A Rx)
LowestChannel 7Not Used
PIB LabelControllerDMA Assignment
Channel 1Serial Port 3 Transmitter (Z85230 Port A Tx)
Channel 2Floppy Driv e Control ler
Channel 3Paralle l Por t
Channel 4DMA2Not available - Cascaded from DMA1
Channel 5Serial Port 4 Receiver (Z85230 Port B Rx)
Channel 6Serial Port 4 Transmitter (Z85230 Port B Tx)
NoteBecause the Z85230 is an 8-bit device and Channels 5 and 6
are 16-bit DMA Channels, only every other byte (the even
bytes) from memory is valid .
1-49
Page 67
1
Board Description and Memory Maps
1-50
Page 68
2Raven PCI Host Bridge & Multi-
Processor Interrupt Controller Chip
Introduction
Overview
This document describes the architecture and usage of the Raven, a
PowerPC to PCI Local Bus Bridge ASIC. The Raven is intended to
provide MPC60x compliant devic es acce ss to device s resi ding on th e PCI
Local Bus in a very efficient manner. In the remainder of this chapter, the
MPC60x bus will be re ferr ed to as th e MPC b us and th e PCI Loc al Bus as
PCI.
No manufacturer currently has plans to support the MPC bus directly.
Therefore, some alternative I/O bus will be necessary in any PowerPC
product. This I/O bus must be robust and efficient enough to handle the
high bandwidth, burst oriented traffic required for Ethernet, SCSI,
graphics, and VMEbus interfaces.
PCI is a high performance 32-bit or 64-bit, burst mode, synchronous bus
capable of transfer rates of 132 MByte/sec in 32-bit mode or 264
MByte/sec in 64-bit mode. While the PCI spec ification is relatively new,
it has received overwhelming support among PC clone manufacturers.
Many peripheral device manufacturers have designed PCI Local Bus
compliant products. NCR tested a PCI SCSI con troller and a PCI Ether net
controller. Ethernet controllers from AMD and National Semiconductor
are available. Graphics controllers are available from Trident, S3, Oak
Technologies, Weitek, Chips and Technologies, Headland Technologies,
and NCR.
The Raven must provide a high throughput interface between multiple
MPC60x processors and 32/64-bit PCI local bus. It must be capable of
supporting up to two MPC60x processors and contain a multiprocessing
interrupt st ructur e to eff icie ntly di stri bute inter rupts d ynamical ly b etwe en
these processors.
Features
❏ MPC Bus Interface
– Direct interface to MPC601, MPC603 or MPC604 processors.
– 64-bit data bus, 32-bit address bus.
– Optional bus arbitr ation logic supporting up to three bus mast ers.
– Four independent software programmable slave map decoders.
– Multi-le vel write post FI FO for writes to PCI.
– Support for MPC bus clock speeds up to 66 MHz.
– Selectable big or little endian operation.
– Support for accesses to all four PCI address spaces.
– Single-level write posting buffers for writes to the MPC bus.
– Read-ahead buffer for reads from the MPC bus.
– Four independent software programmable slave map decoders.
❏ Interrupt Controller
– MPIC compliant.
– Support for 16 external interrupt sources and two processors.
– Multiprocessor i nterrupt control allowing any inter rupt source to
be directed to either processor.
– Multilevel cross processor interrupt control for multiprocessor
synchronization.
Page 70
Introduction
– Four 31 b it tick timers.
❏ Two 64-bit general purpose registers for cross-processor
The MPC Bus Interface is designed to be coupled directly to up to two
MPC601, MPC603, or MPC604 microprocessors as well as a
memory/cache subsystem. It uses a subset of the capabilities of the
MPC60x bus protocol.
MPC Arbiter
The MPC Arbiter is an optional feature in the Raven. The Raven MPC
Arbiter is enabled i f both CPUID pi ns are sampl ed high on t he rising edg e
of RST*. When this feature is enabled, the MARB bit in the General
Control/Status Register (GCSR) will be set. When this feature is not
enabled, the MARB bit will be cleared and the Raven will be configured
for external arbitration.
The MPC Arbiter function is responsible for determining address bus
ownership. The MPC Arbiter function does not provide data bus
arbitration. Determining data bus ownership is the responsibility of each
MPC master and follows the ownership ordering established on the
address bus.
2
The MPC Arbiter supports a total of four participants. One participant is
the Raven MPC master function, which represents PCI initiated MPC
transactions. The remaining three participants are external to the Raven,
and are represented as request/grant signal pairs.
The MPC Arbiter supports a mixture of Fixed Priority and Round-Robin
Priority arbitration schemes. PCI initiated transactions will always have
the highest priority over all external requests. The external requests will be
honored in a Round-Robin algorithm. This al gorithm enforces fairness fo r
bus ownership by assigning the lo west priority t o the most rec ently granted
bus requester. If a bus requester is not granted bus ownership during an
arbitration even t, the priority of that r equester will be increased f or the next
arbitration event.
bits in the MPC Arbiter Control Register (MARB). Following reset, all
requests will be enabled.
The MPC Arbiter supports optional bus parking in order to reduce
arbitration latency. If bus parking is enabled, then one of two modes may
be selected. The arbiter can be configured to grant the bus to the last
selected master, or it can be configured to grant the bus to one “default”
device. The parking enable , parking mode a nd defa ult mas ter inf ormatio n
is represented as control bits within the MARB register.
There is a special ‘Glance Mode’ incorporated into the MPC Arbiter
design. This mode was designed to compensate for an undesirable
characteristi c associate d with early r elease Glance look-aside cache chip s.
Under certain conditions, the Glance would not maintain single-level
pipelined operation when the Raven was MPC bus master. When Glance
Mode is enabled, the MPC Arbiter tr acks address and data tenure s and will
grant bus ownership to non-Rav en bus masters in a manner that guara ntees
single-level pipelined depth. This mode is controlled by the GLMD bit
within the MPC Arbiter Control register. The default state is to have this
mode disabled.
There is another special mode called ‘Benign Address Retry Mode’. This
mode was designed to compensate for an anomaly discovered when
running a PPC603 and a pair of early release Falcon memory controllers.
It is possible that conte ntion of the PPC603 single por t cache tag memory
causes the PPC603 to issue a false (benign) address retry. Under certain
circumstances, these benign address retry cycles would create problems
with Falcon. When the Benign Address Retry Mode is enabled, the MPC
Arbiter will be watching for benign address retry cycles. When one is
detected, the arbiter will hold-off all non-Raven bus masters for a short
period of time. This mode is controlled by the BAMD bit within the MPC
Arbiter Control register. The default state is to have this mode enabled.
Each external reque st can be individually ena bled or disabled using cont rol
A side benefit of the Benign Address Retry Mode is that it provides a
possible solution to a k nown live- lock con dition t hat can happen whe n the
603 is executing a tig ht loop out of cach e and another ma ster is atte mpting
to transfer a coheren t ca che line to or from memory. The Benign Addres s
Retry Mode effectively introduces jitter between the contesting
participants.
2-6
Page 74
Functional Description
MPC Map Decoders
The Raven address decoders have been designed to be as flexible as
possible to provide a wide range of addre ssing possi bilities . There are five
address map decoders in the Raven which determine the MPC bus
addresses to which the Raven will respond: the MPC Register File
Decoder, and four programmable decoders. Table 2-1 shows a typical
CHRP compliant memory map. (Another si milar map is shown i n Table 1-
$FEF90000-$FEF9FFFFFalcon 1 Registers (64K)
$FEFA000 0-$ F EFAFFFFFalcon 2 Registers (64K)
$FEFB0000-$FEFBFFFFFalcon 3 Registers (64K)
$FEFC0000-$FEFEFFFFreserved (192K)
$FEFF0000-$FEFFFFFFRaven Registers (64K) (EXT00 => 0)
$FF000000-$FFFFFFFFSystem ROM /Fl as h (16MB)
The MPC Register File decoder determines the address location of the
Raven’s MPC registers from the MPC bus. These registers may be
accessed using only 1-, 2- , 3-, 4-, or 8-b yte operat ions. The loc ation of t he
MPC register file is fixed beginning at MPC address $FEFE0000 or
$FEFF0000, depending on the state of the EXT01 bit at the time RST* is
file will start at address $FEFE0000. If the EXT01 pin is sampled in the
high state, the MPC register file will start at address $FEFF0000. All
references to the MPC register file within this specification will assume a
base address of FEFF0 000. All Raven re gisters are described in d etail later
in this chapter.
The Raven includes four programmable decoders which control accesses
from the MPC bus to the PCI bus. These dec ode rs pr ovi de a window into
the PCI bus from the MPC bus. The most significant 16 bits of the MPC
address are compared with the address range of each map decoder, and if
the address fal ls with in the s pecif ied range , the acc ess is passed on to PCI .
For each map, there is an associated set of attributes. These attributes are
used to enable read accesses, enable write accesses, enable write posting,
and define the PCI t ransfer characteri stics. Each map decoder also includes
a programmable 16-bit address offset. The offset is added to the 16 most
significant bits of the MPC address, and the result is used as the PCI
address. This offset allows PCI devices to reside at any PCI address,
independent of the MPC address map.
Care should be taken to assure that all programmable decoders decode
unique address ranges. Ove rlapp ing addr ess ra nges wil l lea d to undef ined
operation.
released. If the EXT01 pin is sampled in the low state, the MPC regist er
MPC Write Posting
The MPC writ e FIFO stores u p to eight data beats in any combination of
single- and burst trans actions. If writ e posting is e nabled, Raven st ores the
data necessary to complete an MPC write transfer to the PCI bus and
immediately ackno wledges the tra nsaction on th e MPC bus. This f rees the
MPC bus from waiting for the potent ially long PCI arbitration and transfer.
The MPC bus may be used for more useful work whil e the Raven manages
the completion of the write posted transaction on PCI.
If the write post FIFO is full, any other accesses to the Rav en ar e de lay ed
(AACK* will not be asserted) until there is room in the FIFO to store the
complete transaction.
2-8
Page 76
Functional Description
MPC Master
All write posted tra nsfers will be complete d before a non-write post ed read
or write is begun to assure that all transfers are completed in the order
issued. All write posted t ransfers will al so be completed befor e any access
to the Raven’s registers is begun.
Wherever possible, the MPC master will attempt to consolidate data
movement into a pair of burst transfers called couplets. If th ere is not
enough data movement to pe rform a couple t, the MPC master will attempt
singular burst transfer s. The MPC master will perform single beat tr ansfers
as required during all non-cache aligned writes and some non-cache
aligned reads. A 64-bit by 16 entry FIFO is used to hold data between the
PCI slave and the MPC master to ensure that optimum data throughput is
maintained. While the PCI slave is filling the FIFO with one cache line
worth of data, the MPC master can be moving another cache line worth
onto the MPC bus. This will allow the PCI slave to receive long block
transfers without stalling.
When programmed in “read ahead” mode (the RAEN bit in the PSATTx
register is set) and the PCI slave rec eives a Memory Read Li ne or Memory
Read Multiple command, the MPC mas ter will fetch data in bursts and
store it in the FI FO. The con tents of th e FIFO wil l th en be us ed to att empt
to satisfy the data requirements for the remainder of the PCI block
transaction. If the data requested is not in the FIFO, the MPC master will
read another cache lin e. The cont ent s of the FIFO are “invalidated” at the
end of each PCI block transaction.
2
Notes1. Read ahead mode sho uld not be used when data coherency
The MPC bus transfer types generated by the MPC master depend on the
PCI command code and the INV bit in the PSATTx registers.
may be a problem as there is no way to snoop all MPC bus
transactions and invalidate the contents of the FIFO.
2. Accesses near the top of local memory with read-ahead
mode enabled could cau se the MPC maste r to perf orm read s
beyond the top of local memory which could result in an
MPC bus timeout error.
Memory Read,
Memory Read
Multiple,
Memory Read Line
Memory Read,
Memory Read
Multiple,
Memory Read Line
Memory Write,
Memory Write and
Invalidate
Memory Write,
Memory Write and
Invalidate
The MPC master incorporate s an optional operating mode cal led Bus Hog.
When Bus Hog is enabled, the MPC master will continually request the
MPC bus for the entire dura tion of each PCI transfer. When Bus Hog is not
enabled, the MPC master will structure its bus request actions around its
desire to perform couplets. This means the bus request will be deasserted
between couplets. Caution sh ould be exercised when using this mode si nce
the over-generosity of bus ownership to the MPC master can be
detrimental to the host CPU’s performance. The Bus Hog mode can be
controlled by the BHOG bi t within the GCSR. T he default sta te for BHOG
is disabled.
MPC Transfer
Type
0ReadBurst/Single Beat01010
1Read With Intent to
Modify
xWrite with KillBurst00110
xWrite with FlushSingle Beat00010
MPC Transfer SizeTT0-TT4
Burst/Single Beat01110
MPC Bus Timer
The MPC bus timer allows the current bus master to reco ver from a lock up condition caused when no slave responds to the transfer request.
The time-out length of the bus timer i s determined by t he MBT field in th e
Global Control/Status Register.
2-10
Page 78
Functional Description
The bus timer starts ticking at the beginning of an address transfer (TS*
asserted), and if the address transfer is not terminated (AACK* asserted)
before the time -out period h as passed, the Raven will a ssert the M ATO bit
in the MPC Error Status Register, latch the MPC address in the MPC Error
Address Register, and then immediately assert AACK*.
The MATO bit may be configured to generate an interrupt or a machine
check through the MEREN register.
The timer is disabled if the transfer is intended for PCI. PCI bound
transfers will be timed by the PCI master.
PCI Interface
The Raven PCI Interface is desi gned to connect directly to a PCI Loca l Bus
compliant I/O bus.
The PCI interface may operate at any clock speed up to 33 MHz. The
PCLK input must be externally synchronized with the MCLK input, and
the frequency of the PCLK input must be exactly hal f the frequency of the
MCLK input.
PCI Map Decoders
2
The Raven contains four prog rammable decod ers which provide win dows
into the MPC bus from the PCI bus. The most significant 16 bits of the PCI
address is compare d with the address ra nge of each map decoder , and if the
address falls withi n the specified range, the acce ss is passed on to the MPC
bus. For each map, there is an independent set of attributes. These
attributes are used to enable read accesses, enable write accesses, enable
write posting, and define the MPC bus transfer characteristics. Each map
decoder also includes a programmable 16-bit address offset. The offset is
added to the 16 most significant bits of the PCI address, and the result is
used as the MPC address. This offse t allows devices to reside at any MPC
address, independent of the PCI address map.
All Raven address decoders are prioritized so that programming multiple
decoders to respond to the same address will not be a problem. When the
PCI address fall s into the range of more than one decoder, only the high est
priority one will respond. The decoders are prioritized as shown below.
The Raven does not have an IDSEL pin. An internal connection is made
within the Raven that logi cally assoc iates t he ass erti on of I DSEL with the
assertion of either AD30 or AD31. The exact association depends on the
state of the EXT02 pin when the RST* pin is released. If EXT02 is
sampled low, the Raven will associate AD30 with IDSEL. If EXT02 is
sampled high, the Raven will associate AD31 with IDSEL.
PCI Write Posting
If write posting is enabled, the Raven sto res the ta rge t addr es s, att ri but es ,
and up to 128 bytes of data from one PCI write transaction and
immediately acknowled ges the transaction on the PCI bus. This allows t he
slower PCI to continue to transfer data at its maximum bandwidth, and the
faster MPC bus to accept data in high performance cache-line burst
transfers.
Only one PCI transaction may be write posted at any given time. If the
Raven is busy processing a previous write posted transa ct ion when a new
PCI transaction begin s, t he next PCI transaction will be de la yed ( TRDY*
will not be asserte d) until the previou s transaction has compl eted. If during
a transaction the write post buffer gets full, subsequent PCI data transfers
will be delayed (TRDY* will not be asserted ) until the Raven has r emoved
some data from the FIFO. Under normal conditions, the Raven should be
able to empty the FIFO faster than the PCI bus can fill it.
PCI Configuration cy cles intended for internal Ra ven registers will also be
delayed if Raven is busy so that co ntrol bits which may affect writ e posting
do not change until all write posted transactions have completed.
2-12
Page 80
Functional Description
PCI Master
The PCI master, in conju nction with the cap abilities of the MPC s lave, will
attempt to move data in either singl e beat or burst transactions. All single
beat transactions will be subdivided into one or two 32-bit transfers,
depending on the alignment and size of the transaction. The PCI master
will attempt to transf er all burst t ransac tions in 64- bit mode. If at a ny time
during the transaction the PCI target indicates it can not support 64-bit
mode, the PCI mast er will c ontinue to transfer t he remai ning data i n 32-bit
mode.
The PCI Command Codes generated by the PCI master depend on the
MPC transfer type, TBST*, and the MEM field in the MSATTx registers.
Each programmable slave may be configured to generate PCI I/O or
memory accesses th rough the MEM and IOM fiel ds in its Attribute re gister
as shown below.
The IBM CHRP specification describes two approache s f or hand ling PCI
I/O addressing: contiguous or spread address modes. When the MEM bit
is cleared, the I OM bit is used to select between these two modes whene ver
a PCI I/O cycle is to be perform ed.
When MEM is clea r or IOM is c lear, the Ra ven will take the MPC address ,
apply the offset specified in the MSOFFx register, and map the result
directly to PCI.
When MEM is clear an d IOM is set, t he Raven will t ake the MP C address,
apply the offset specified in the MSOFFx register, and map the result to
PCI as shown in Figure 2-2.
.
MPC Address + Offset
3112 115 40
310
0 0 0 0 0 0 0
0000000
PCI Address
25 24
Figure 2-2. PCI Spread I/O Cycle Mapping
This CHRP compliant spread I/O mode allows each PCI device’s I/O
registers to res ide on a dif ferent MPC memor y page, so device drive rs can
be protected from each other using memory page protection.
2-14
54
1915 9610
Page 82
Functional Description
All I/O accesses must be performed within natural word boundaries. Any
I/O access that is not contai ned within a nat ural word boun dary will resul t
in unpredictable operation. For exampl e, an I/O transfer of 4 bytes start ing
at address $80000010 is considered a valid transfer. An I/O transfer of 4
bytes startin g at ad dress $8 0000011 is consi dered an inva lid tr ansfer since
it crosses the natural word boundary at address $80000013/$80000014.
Generating PCI Configuration Cycles
Mechanism one (as just described above) is utilized to generate
configuration cycle s. Two 32 bit PCI I/O po rts at $CF8 and $CFC are used
to access PCI configuration space. One of the four MPC Slave Address
Registers is used to ga in access to $ CF8 and $CFC. Note that MSADD3 is
initialized at reset to access PCI I/O space with an MPC address of
$80000000.
The resource at $CF8 is a 32 bi t config urati on addr ess p ort and i s ref erre d
to as the CONFIG_ADDRESS register. The resource at $CFC is a 32 bit
configuration data port and is referred to as the CONFIG_DATA register.
Accessing a PCI functions’s configuration port is a two step process;
❏ Write the bus number, physical device number, function number
and register number to the CONFIG_ADDRESS register.
2
❏ Perform an I/O read from or a write to the CONFIG_DATA register.
Generating PC I Special Cycles
To prime Raven to generate a speci al cyc le, the hos t proce ssor must write
a 32 bit value to the CONFIG_ADDRESS register. The contents of the
write are defined later in this chapter under the CONFIG_ADDRESS
register definition. After the write to $CF8 has been accomplished, the
next write to the CONFI G_DATA re giste r causes the Raven t o gene rate a
special cycle on the PCI bus. The write data is driven onto AD[31:0]
during the special cycle’s data phase.
Performing a read from the PIACK register will initiate a single PCI
Interrupt Acknowledge cycle. Any single byte or combination of bytes
may be read from, and the act ual byte enable patte rn u sed during the read
will be passed on to the PCI bus. Upon completion of the PCI interrupt
acknowledge cycle, the Raven will p resent the result ing vector inform ation
obtained from the PCI bus as read data.
Endian Conversion
The Raven supports both Big- and Little-Endian data formats. Since PCI
is inherently Little- Endian, convers ion is necessa ry if all MPC devices ar e
configured for Big-Endian operation. The Raven may be programmed to
perform the Endian conversion described below.
When MPC Devices are Big-Endian
When all MPC devices are op erating in Bi g-Endian mode, a ll data to/fr om
PCI must be swapped such th at PCI looks big endia n from the MPC bus ’s
perspective. This is shown in Figure 2-3.
2-16
Page 84
Functional Description
.
2
DH07-00
DH15-08
DH23-16
DH31-24DL07-00
D0D1D2D3D4D5D6D7
DL15-08
DL23-16
DL31-24
PPC Bus
D7D6D5D4D3D2D1D0
AD31-24
AD23-16
AD15-08
DL23-16
AD07-00
DL31-24
AD63-56
AD55-48
AD47-40
AD39-32
DH07-00
DH15-08
DH23-16
DH31-24
DL07-00
DL15-08
D0D1D2D3D4D5D6D7
D7D6D5D4
D3D2D1D0
AD31-24
AD23-16
AD15-08
32-bit PCI
AD07-00
Figure 2-3. Big to Little Endian Data Swap
64-bit PCI
PPC Bus
1916 9610
When MPC Devices are Little Endian
When all MPC devices are operating in little endian mode, the MPC
address must be modified to remove the exclusive-ORing applied by
MPC60x processors before being passed on to PCI. The three low order
processor bus address bits are exclusive-ORed with a three-bit value that
depends on the length of the operand, as shown in Table 2-4.
Table 2-4. Address Modification for Little Endian Transfers
Data Length
(bytes)
1XOR with 111
2XOR with 110
4XOR with 100
8no change
NoteThe only legal data lengths supported in little endian mode
are 1, 2, 4, or 8-byte aligned transfers.
Cycles Originating From PCI
For bus cycles initiated by PCI masters, the PCI address will be modified
the same way the MCP60x processor does in little endian mode. The
modification will be the same as that described in Section 3.7.2 above.
Since this method has some difficulties dealing with unaligned transfers,
the Raven will break up all unaligned PCI transfers into multiple aligned
transfers on the MPC bus.
Address Modification
Error Handling
The Raven will be capable of de tecti ng and re porti ng the f ollowi ng error s
to one or more MPC masters:
❏ MPC address bus time-out
❏ PCI master signalled master abort
❏ PCI master received target abort
❏ PCI parity er ror
❏ PCI system error
2-18
Page 86
Functional Description
Each of these error condi tions wil l cause an err or st atus bi t to be se t in th e
MPC Error Status Register. If a second error is detected while any of the
error bits is set, the OVFL bit is asserte d, but none of t he error bits ar e
changed. Each bit in the MPC Error Status Register may be cleared by
writing a 1 to it; writing a 0 to it has no effect. New error bits may be set
only when all previous error bits have been cleared.
When any bit in the MPC Error Status register is s et, the Raven will
attempt to latch as much information as possible about the error in the
MPC Error Address and Attribute Registers. Information is saved as
follows:
Error StatusError Address and Attributes
MATOFrom MPC bus
SMAFrom PCI bus
RTAFrom PCI bus
PERRInvalid
SERRInvalid
Each MERST error bit may be programmed to generate a machine check
and/or a standard in terrupt. The er ror response is programmed thro ugh the
MPC Error Enable Register on a sour ce by sourc e basis . When a machi ne
check is enabled, eit her the MID fiel d in the MPC Error Attribute Reg ister
or the DFLT bit in the MEREN Register determine the master to which the
machine check is directed. For errors in which the master who originated
the transaction can be det ermined, the MID field is used, provided th e MID
is%00 (processor 0), %01 (processor 1), or %10 (proce ssor 2). For errors
not associated with a particular MPC master, or associated with masters
other than processor 0,1 or 2, the DFLT bit is used. One example of an
error condition which cannot be associated with a particular MPC master
would be a PCI system error.
The Raven has a stall detection mechanism that detects when there is a
possible resource contention problem (i.e. deadlock) as a result of
overlapping MPC and PCI initiated transactions. The MPC Slave and the
PCI Slave functions contain the logic needed to implement this fe ature.
The PCI Slave function contributes to the stall detection mechanism by
issuing a stall signal to the MPC Slave function whenever it is currently
processing a t ransa ction t hat must have contr ol o f the MPC bus b efore the
transaction can be completed. The events that activate this signal are:
❏ PCI read cycle
❏ PCI non-posted write cycle
❏ PCI posted write cycle with flush-before-read (FLBRD) enabled
❏ PCI posted w rite cycle an d internal FIFO full
The MPC Slave function determines its future actions based on the stall
signal and the current MPC bus activity. If the MPC Slave function
determines there will be contention between a cycle completing on the
MPC bus and an incoming PCI cycle , the MPC Slave will issu e a retry fo r
the current MPC t ransaction. This re try will free up the MPC bus and al low
the PCI initiated transaction to complete. An idle MPC bus obviously gives
immediate access to the pending PCI initiated transaction.
2-20
If the MPC bus is currently supporting a read cycle of any type, the cycle
will be terminated wi th a retry. Note that i f the read cycle is ac ros s a mod4 address boundary (i.e. f rom address 0 x...02, 3 bytes) , it is poss ible that a
portion of the read could have been completed before the stall condition
was detected. The previously read data will be discarded and the current
transaction will be retried.
If the MPC bus is currently supporting a posted write transaction, the
transaction will be allowed to complete since this type of transaction is
guaranteed completion. If the MPC bus is currently supporting a nonposted write transaction, the transaction will be terminated with a retry.
Note that a mod-4 non-posted write transaction could be interrupted
Page 88
Functional Description
between write cycles, and thereby result in a partially completed write
cycle. It is recommended that write cycles to write-sensitive non- posted
locations be performed on mod-4 address boundaries.
The Raven has a programmable option to guarantee all PCI write posted
transactions are completed before an MPC initiated read transaction may
be allowed to comple te. Th is opt ion is contr olled by t he FLBRD b it i n the
GSCR register. If this bit is set, all MPC read transactions will be retried
until all posted PCI write tr ansact ions have c omplete d. It is re commended
that this option be disabled, and the FLBRD bit be left in the default
(disabled) state.
This chapter provides a detailed description of all Raven registers. These
registers are broken into two groups: the MPC Registers and the PCI
Configuration Registe rs. The MP C Regist ers are a cce ssible onl y from t he
MPC bus using any valid transfer size. The PCI Configuration Registers
reside in PCI configurat io n spa ce. They are accessible from the MPC bus
through the Raven. The MPC Registers are described first; the PCI
Configuration Registers are described next.
The following conventions are used in the Raven register charts:
❏ RRead O nly field.
❏ R/WRead/Write field.
❏ SWriting a ONE to this field sets this field.
❏ CWriting a ONE to this field clears this field.
VENIDVendor ID. This register identifies the manufacturer of
DEVIDDevice ID. This register identifies this particular device.
Revision ID Register
1011121314151617181920212223242526272829303
1
the device. This identifier is allocated by the PCI SIG to
ensure uniqueness. $1 057 has bee n assigne d to Motor ola.
This register is duplicated in the PCI Configuration
Registers.
The Raven will always return $4801. This register is
duplicated in the PCI Configuration Registers.
REVIDRevision ID. This register identifies the Raven revision
level. This register is dupli cated in t he PCI Configura tion
Registers.
2-24
Page 92
Registers
General Control-Status/Feature Registers
Address$FEFF0008
Bit
0 1 2 3 4 5 6 7 8 9
NameGCSRFEAT
Operation
Reset
LEND
R/WRRRR/W
00000000000000000
BHOG
FLBRD
R/W
MBT1
MBT0
R/W
R/WRRRRRRRRRRRRRRRRRRRRRRRR
1011121314151617181920212223242526272829303
P64
MARB
MPIC
MID1
MID0
EXT14
0/1
EXT13
0/1
EXT12
0/1
EXT11
0/1
EXT10
0/1
EXT09
0/1
EXT08
0/1
LENDEndian Select. If set, the MPC bus is operating in little
endian mode. The MPC address will be modified as
described in the section When MPC Devices are Little Endian. When LEND is clear, the MPC bus is operating
in big endian mode, and all data to/from PCI is swapped
as described in the section When MPC Devices are Big-Endian.
FLBRDFlush Before Read. If set, the Raven will guarantee that
all PCI initiated posted write transactions will be
completed before any MPC initiated read transactions will
be allowed to complet e. When FLBRD is clear , there wi ll
be no correlation between these transaction types and
their order of completion. Please refer to the section on
PCI/MPC Contention Handling for more information.
EXT07
0/1
EXT06
0/1
EXT05
0/1
EXT04
0/1
EXT03
0/1
EXT02
0/1
EXT01
0/1
2
1
EXT00
0/1
BHOGBus Hog. If set, the Raven MPC master will o perate in the
Bus Hog mode. Bus Hog mode means the MPC master
will continually request the MPC bus for the entire
duration of each PCI transfer. If Bus Hog is not enabled,
the MPC master will reques t t he bus i n a normal manner.
Please refer to the section on MPC Master for more
information.
time-out length. The time-out length is encoded as
follows:
MBTxMPC Bus Time-out. This field specifies the MPC bus
MBTTime Out Length
00256 µsec
0164 µsec
108 µsec
11disabled
P646 4-bit PCI Mode Enable. If set, the Raven is connected
to a 64-bit PCI bus. This bit is set if REQ64* is asserted
on the rising edge of RESET*.
MARBMPC Arbiter Enable. If set, the Raven internal MPC
Arbiter is enabled. This bit is set if CPUID is %111 on the
rising edge of RESET*.
MPICMulti-Processor Interrupt Controller Enable. If set,
the Raven internal MPIC interrupt controller is enabled.
This bit is set if EXT15 is high on the rising edge of
RESET*. If cleared, Raven detected errors wi ll be pa ssed
on to processor 0 INT pin.
2-26
MIDxMaster ID. This field is encoded as shown below to
indicate who is currently the MPC bus master. When the
internal MPC arbiter is enabl ed (MARB is set), these bi ts
are controlled by the internal arbiter. When the internal
arbiter is disabled (MARB is clear) these bits reflect the
status of the CPUID pins. In a multiprocessor
environment, these bits allow software to determine on
which processor it i s currently runni ng. The internal MPC
arbiter encodes this field as follows:
Page 94
Registers
.
MIDCurrent MPC Data Bus Master
00device on ABG0*
01device on ABG1*
10device on ABG2
11Raven
2
FEATFeature Register. Each bit in this register reflects the
state of one of the external interrupt input pins on the
rising edge of RESET*. This register may be used to
report hardware configuration parameters to system
software.
BREN2Bus Request 2 Enable. If set, the processor bus request
BREN1Bus Request 1 Enable. If set, the processor bus request
BREN0Bus Request 0 Enable. If set, the processor bus request
1011121314151617181920212223242526272829303
BREN2
BREN1
BREN0
PKEN
PKMD
GLMD
BAMD
RRRRR
000001110001001
R/W
R/W
R/W
R/W
R/W
R/W
R/WRR
signal ABR2* is enabled. If cleared, ABR2* is not
enabled, and ABG2* will never be asserted.
signal ABR1* is enabled. If cleared, ABR1* is not
enabled, and ABG1* will never be asserted.
signal ABR0* is enabled. If cleared, ABR0* is not
enabled, and ABG0* will never be asserted.
1
DEFM1
DEFM0
R/W
R/W
1
2-28
PKENBus Parking Enable. If set, the MPC arbiter will park an
MPC master on the bus when no b us requests are pen ding.
If cleared, no MPC master will be granted the bus witho ut
first asserting its ABRx*.
PKMDBus Parking Mode. When bus parking is enabled (PKEN
is set), this bit defines the method used to determine which
MPC master is parked on the bus. If set, the master
specified in the DEFM field is parked on the bus when no
bus requests are pending. If cleared, the last active MPC
master is parked on the bus when no bus requests are
pending. This field has no meaning when PKEN is clear.
Page 96
Registers
GLMDGlance Mode. If set, the MPC arb iter will operate in the
BAMDBenign Address Retry Mode. If set, the MPC arbiter wil l
DEFMxDefault Master. This field specifies which MPC bus
DEFM1DEFM0Parked Device
Prescaler Adjust Register
Glance mode. Please ref er to the section on the MPC
Arbiter for more information.
operate in the benign address retry mode. Please refer to
the section on the MPC Arbiter for more information.
master will be parked on the bus when default parking is
enabled. DEFM only has meaning when PRKEN and
PRKMD are both set.
00device on ABR0*
01device on ABR1*
10device on ABR2*
11 Raven
2
Address$FEFF0010
Bit
0 1 2 3 4 5 6 7 8 9
NamePADJ
OperationRRRR/W
Reset$00$00$00$B4
1011121314151617181920212223242526272829303
PADJPr escaler Adjust. This registe r is used to specify a scale
factor for the presca ler to ensure tha t the time base for the
bus timer is 1 MHz. The scale factor is calculated as
follows:
following table shows the scale factor s for some common
CLK frequencies.
where Clk is the frequency of the CLK inp ut in MHz. The
FrequencyPADJ
66$B4
50$CE
40$D8
33$DF
25$E7
MPC Error Enable Register
Address$FEFF0020
Bit
0 1 2 3 4 5 6 7 8 9
NameMEREN
1011121314151617181920212223242526272829303
DFLT
MAT OM
PERRM
SERRM
SMAM
RTAM
MAT OII
PERRI
SERRI
SMAI
RTAI
1
OperationRR
Reset$00$00
DFLTDefault MPC Master ID. This bit dete rmines which
MCHK* pin will be asserte d for error con ditions in which
the MPC master ID can not be determined or the Raven
was the MPC master. For example, in event of a PCI
parity error for a transaction in which the Raven’s PCI
master was not involved, the MPC master ID can not be
determined. When DFLT is set, MCHK1* is used. When
DFLT is clear, MCHK0* will be used.
2-30
R
R/W
R/WRR/W
000000000000000
R/W
R/W
R/WRR
R/WRR/W
R/W
R/W
R/W
0
Page 98
Registers
MATOMMPC Address Bus Time-out Machine Check Enable.
When this bit is set, the MA T O bit in the MERST register
will be used to assert the MCHK output to the current
address bus master. When this bit is clear , MCHK will not
be asserted.
PERRMPCI Parity Error Machine Check Enable. When this
bit is set, the PERR bit i n the MERST register wil l be used
to assert the MCHK output to bu s master 0. When thi s bit
is clear, MCHK will not be asserted.
SERRMPCI System Error Machine Check Enable. When this
bit is set, the SERR bit i n the MERST register wil l be used
to assert the MCHK output to bu s master 0. When thi s bit
is clear, MCHK will not be asserted.
When this bit is set, the SMA bit in the MERST register
will be used to asse rt t he MCHK ou tput t o the bus mast er
which initia ted the transac tion. When this bi t is clear,
MCHK will not be asserted.
RTAMPCI Master Received Target Abort Ma chine Check
Enable.When this bit is set, the RTA bit in the MERST
register will be used t o assert the MCHK ou tput to the bus
master which initiated the transaction. Whe n this bit is
clear, MCHK will not be asserted.
2
MATOIMPC Address Bus Time-out Interrupt Enable.When
this bit is set, the MA TO bit in the MERST register will be
used to assert an interrupt through the MPIC interrupt
controller. When this bit is clear, no interrupt will be
asserted.
PERRIPCI Parity Err or Interrupt Enabl e.When th is bit is s et,
the PERR bit in the MERST registe r will be u sed to asse rt
an interrupt throu gh the MP IC int erru pt co ntrol ler. When
this bit is clear, no interrupt will be asserted.
set, the PERR bit in the MERST register will be used to
assert an interrupt through the MPIC interrupt controller.
When this bit is clear, no interrupt will be asserted.
SMAIPCI Master Signalled Master Abort Interrupt
Enable.When this bit is set, the SMA bit in the MERST
register will be used to assert an interrupt through the
MPIC interrupt controller. When this bit is clear, no
interrupt will be asserted.
RTAIPCI Master Received Target Abort Interrupt
Enable.When this bit is set, the RTA bit in the MERST
register will be used to assert an interrupt through the
MPIC interrupt controller. When this bit is clear, no
interrupt will be asserted.
MPC Error Status Register
Address$FEFF0024
SERRIPCI System Error Interrupt Enable.When this bit is
Bit
0 1 2 3 4 5 6 7 8 9
NameMERST
1011121314151617181920212223242526272829303
OVF
MATO
PERR
SERR
SMA
RTA
1
OperationRRR
Reset$00$00$00
OVFError Status Overflow. This bit is set when any error is
detected and any of t he error status bits a re already set. It
may be cleared by writing a 1 to it; writing a 0 to it has no
effect.
MATOMPC Address Bus Time-out. This bit is set when the
MPC address bus timer times out. It may be cleared by
writing it to a 1; writing it to a 0 has no effect. When the
2-32
R/CRR/CRR/C
0000000
R/C
R/C
R/C
0
Page 100
Registers
MA T OM bit in the MEREN regi ster is set, the asser tion of
this bit will assert MCHK to the master designated by the
MID field in the MERAT register. When the MATOI bit
in the MEREN register is set, the assertion of this bit will
assert an interrupt through the MPIC interrupt controller.
PERRPCI Parity Error . This bit is set when the PCI PERR* pin
is asserted. It may be cleared by writing it to a 1; writing
it to a 0 has no effect. When the PERRM bit in the
MEREN register is set , the as serti on of this bit will as se rt
MCHK to the master designated by the DFLT bit in the
MERAT register. When the PERRI bit in the MEREN
register is set, the assertion of this bit will assert an
interrupt through the MPIC interrupt controller.
SERRPCI System Error. This bit is set when the PCI SERR*
pin is asserted. It may be cleared by writing it to a 1;
writing it to a 0 ha s no eff ect. When t he SERRM bit in the
MEREN register is set , the as serti on of this bit will as se rt
MCHK to the master designated by the DFLT bit in the
MERAT register. When the SERRI bit in the MEREN
register is set, the assertion of this bit will assert an
interrupt through the MPIC interrupt controller.
2
SMAPCI Master Signalled Master Abort. This bit is set
when the PCI master signals master abort to terminate a
PCI transaction. It may be cleared by writing it to a 1;
writing it to a 0 has no ef fect . When t he SMAM bi t in t he
MEREN register is set , the asser tion of t his b it wi ll as sert
MCHK to the master designated by the MID field in the
MERAT register. When the SMAI bit in the MEREN
register is set, the assertion of this bit will assert an
interrupt through the MPIC interrupt controller.
RT APCI Master Received Target Abort. This bit is set when
the PCI master receives target abort to terminate a PCI
transaction. It may be cl eared by writin g it to a 1; writi ng
it to a 0 has no effect. When the RT AM bit in the MEREN
register is set, the assertion of this bit will assert MCHK
to the master designated by the MID field in the MERAT
2-33
Loading...
+ hidden pages
You need points to download manuals.
1 point = 1 manual.
You can buy points or you can get point for every manual you upload.