Home » AMD Manuals » Processors » AMD 3200 » Manual Viewer

AMD 3200 Revision History - Page 33

Table, Probes, Cause, System

UPC - 730143241144

Add to My Manuals
Save this manual to your list of manuals

Page 33 highlights

48063 Rev. 3.18 October 2012 Revision Guide for AMD Family 15h Models 00h-0Fh Processors 540 GART Table Walk Probes May Cause System Hang Description Probes that are generated for GART table walks may overflow internal queues and lead to a deadlock. Potential Effect on System System hang. Suggested Workaround System software that initializes the GART Table Base Address (D18F3x98) should set GART Aperture Control[DisGartTblWlkPrb] (D18F3x90[6]) = 1b. The GART tables should be in UC DRAM or be updated only using strongly-ordered uncacheable writes. System software should not set HWCR[ForceRdWrSzPrb] (MSRC001_0015[23]) if GART is enabled. Fix Planned No fix planned Product Errata 33

Section	Page
List of Figures	3
List of Tables	4
Revision History	5
Overview	6
Revision Guide Policy	6
Conventions	7
Numbering	7
Register References and Mnemonics	7
Arithmetic and Logical Operators	8
Processor Identification	9
Revision Determination	9
D18F4x164 Fixed Errata Register	10
Mixed Processor Revision Support	11
Programming and Displaying the Processor Name String	11
Constructing the Processor Name String	11
Operating System Visible Workarounds	13
MSRC001_0140 OS Visible Work-around MSR0 (OSVW_ID_Length)	13
MSRC001_0141 OS Visible Work-around MSR1 (OSVW_Status)	13
Product Errata	14
Cross-Reference of Errata to Package Type	18
Cross-Reference of Errata to Processor Segments	21
361 Breakpoint Due to an Instruction That Has an Interrupt Shadow May Be Lost	24
503 APIC Task-Priority Register May Be Incorrect	25
504 Corrected L3 Errors May Lead to System Hang	26
505 Scrub Rate Control Register Address Depends on DctCfgSel	27
520 Some Lightweight Profiling Counters Stop Counting When Instruction-Based Sampling is Enabled	28
535 Lightweight Profiling May Not Indicate Fused Branch	29
536 Performance Counter for Instruction Cache Misses Does Not Increment for Sequential Prefetches	30
537 Performance Counter for Ineffective Software Prefetches Does Not Count for L2 Hits	31
538 Performance Counter Does Not Count for Some Retired Micro-Ops	32
540 GART Table Walk Probes May Cause System Hang	33
550 Latency Performance Counters Are Not Accurate	34
585 Incorrect Memory Controller Operation Due to a WrDatGrossDly Setting of 3.5 MEMCLKs	35
586 A Far Control Transfer Changing Processor Operating Mode May Generate a False Machine Check	36
592 VPEXTRQ and VPINSRQ May Not Signal Invalid-Opcode Exception	37
593 Last-Branch Record Enabled May Cause Machine Check and Incorrect LastBranchToIp	38
600 HyperTransport™ Link Retry Due to Partial CRC Error May Cause System Hang	39
602 HyperTransport™ Link Frequency Changes May Cause a System Hang	40
608 P-state Limit Changes May Not Generate Interrupts	41
619 Non-Posted Reads May Block Write Dependent on Probe Responses	42
623 Small Code Segment Limits May Cause Incorrect Limit Faults	43
624 SB-RMI Processor State Accesses May Persistently Timeout if Interrupted by a Warm Reset	44
625 SB-RMI Writes May Not Be Observed by Processor	45
636 Instruction Addresses Near Canonical Address Limit May Cause #GP Exception	46
637 Processor Does Not Report the Correct DRAM Address for MCA Errors Within the CC6 Save Area	47
657 MC1_STATUS Enable Bit Not Set When Logging Corrected Errors	49
658 CPUID Incorrectly Reports Large Page Support in L2 Instruction TLB	50
659 VMCB Interrupt Shadow Status May Be Incorrect	51
660 APERF May Increase Unpredictably	52
661 P-State Limit and Stop Clock Assertion May Cause System Hang	53
663 Local Interrupts LINT0/LINT1 May Occur While APIC is Software Disabled	54
667 Processor May Generate Incorrect P-state Limit Interrupts	55
668 Load Operation May Receive Incorrect Data After Floating-point Exception	56
671 Debug Breakpoint on Misaligned Store May Cause System Hang	57
672 SVM Guest Performance Counters May Be Inaccurate Due to SMI	58
673 Misaligned Page Crossing String Operations May Cause System Hang	59
674 Processor May Cache Prefetched Data from Remapped Memory Region	60
675 Instructions Performing Read-Modify-Write May Alter Architectural State Before #PF	61
685 Some Processor Cores May Have Inaccurate Instruction Cache Fetch Performance Counter	62
689 AM3r2 Six Core Processor May Limit P-State When Core C6 State Is Disabled	63
690 Northbridge FIFO Read/Write Pointer Overlap May Cause Hang or Protocol Error Machine Check	64
691 Processors Using 1 MB L3 Subcaches May Execute a Write-Back Invalidate Operation Incorrectly	65
693 Performance Counter May Incorrectly Count MXCSR Loads	66
694 IBS Sampling of Instruction Fetches May Be Uneven	67
695 Processor May Interpret FCW Incorrectly after FNSAVE/FSAVE Limit Fault	68
699 Processor May Generate Illegal Access in VMLOAD or VMSAVE Instruction	69
704 Processor May Report Incorrect Instruction Pointer	70
707 Performance Counter for Locked Operations May Count Cycles from Non-Locked Operations	71
708 Initial Value of Time Stamp Counter May Include an Offset Error	72
709 Processor May Be Limited to Minimum P-state After a P-state Limit Change	73
714 Processor May Check DRAM Address Maps While Using L2 Cache as General Storage during Boot	74
717 Instruction-Based Sampling May Be Inaccurate	75
718 Instruction-Based Sampling May Be Inaccurate	76
719 Instruction-Based Sampling Fetch Counter Always Starts at Maximum Value	77
720 Processor May Not Respect Interrupt Shadow	78
724 Unintercepted Halt Instruction May Cause Protocol Machine Check or Unpredictable System Behavior	79
725 Incorrect APIC Remote Read Behavior	80
726 Processor May Report Incorrect MCA Address for Loads that Cross Address Boundaries	81
727 Processor Core May Hang During CC6 Resume	82
734 Processor May Incorrectly Store VMCB Data	83
737 Processor Does Not Check 128-bit Canonical Address Boundary Case on Logical Address	84
739 Processor May Read Branch Status Register With Inconsistent Parity Bit	85
740 Lightweight Profiling May Cause System Hang with Concurrent Stop Clock	86
742 DRAM Scrub Request During Register Write May Cause Unpredictable Behavior	87
744 Processor CC6 May Not Restore Trap Registers	88
745 Processor May Incorrectly Report Cache Sharing Property in CPUID Topology	89
759 One Core May Observe a Time Stamp Counter Skew	90

Match case Limit results 1 per page

540

GART

Table

Walk

Probes

May

Cause

System

Hang

Description

Probes

that

are

generated

for

GART

table

walks

may

overflow

internal

queues

and

lead

deadlock

Potential

Effect

System

hang

Suggested

Workaround

System

software

that

initializes

the

GART

Table

Base

Address

(

should

set

GART

Aperture

Control

[

DisGartTblWlkPrb

] (

90[6] = 1

The

GART

tables

should

DRAM

updated

only

using

strongly-ordered

uncacheable

writes

System

software

should

not

set

HWCR

[

ForceRdWrSzPrb

] (

MSRC

001_0015[23]

GART

enabled

Fix

Planned

fix

planned

48063

Rev

. 3.18

October

2012

Revision

Guide

for

AMD

Family

Models

Processors

Product

Errata