Kernel Debugging: "Parity Error: Bad Instruction Pointer"

ID: Q149015


The information in this article applies to:
  • Microsoft Windows NT Workstation versions 3.5, 3.51
  • Microsoft Windows NT Server versions 3.5, 3.51


SUMMARY

This article describes a parity error that is caused by a bad processor cache or by a faulty processor.


MORE INFORMATION

Article Contents


   - Stack Trace
   - Trap Frame
   - Unassemble of IoAllocateMdl function and EIP
   - Parity Error
   - Bad Instruction Pointer
   - Breakdown of the instruction at 801144dd and Opcode
   - Breakdown of the instruction at 801144dc and Opcode 


Stack Trace

The following is the stack trace from the failed system:


kv
ChildEBP RetAddr  Args to Child
8014068c 801144dc ff067010 ff06705c 00000103
NT!_KiTrap0E+0x252 (FPO: [0,0] TrapFrame @ 8014068c)

80140710 fc903dc3 ff177358 00000002 00000000
NT!_IoAllocateMdl+0x44 (FPO: [EBP 0xff177358] [5,1,4])

80140738 fc93102c 80140768 8014076c 00000000
NDIS!_NdisCopyBuffer+0x26 (FPO: [EBP 0xff067010] [6,0,4])

8014076c fc932071 ff07300c ff066028 fc9320ba
sfmatalk!_AtalkNdisSendPacket+0xa4 (FPO: [EBP 0x801407e8] [4,2,4])

801407e8 fc931950 ff07300c 80140838 ff9f673c
sfmatalk!_AtalkDdpTransmit+0x60b

8014083c 801144fc ff191800 fc942a80 ff11960c
sfmatalk!_AtalkDdpSend+0x71 (FPO: [EBP 0x00000000] [9,7,4])

80140884 fc94299f ff9f6720 ff065c20 00000246
NT!_IoAllocateMdl+0x64 (FPO: [EBP 0x0000001b] [5,1,4])

801408c0 fc943f5e ff11966c ff982390 ff17732a
sfmatalk!atalkAtpSendReqComplete (FPO: [0,0,1])

80140910 fc943e8b 00000000 ff11799c ff068854
sfmatalk!atalkAspPostWriteContinue+0x9c (FPO: [EBP 0x80140930] [0,1,4])

80140930 fc9421db 00000000 ff1199cc ff11799c
sfmatalk!_atalkAspSssXHandler+0x1f5

80140990 fc930c64 ff07300c
ff11960c ff066cf0 sfmatalk!_AtalkAtpPacketIn+0x5a9

801409dc fc90068b ff07300c ffdff420 ff070007
sfmatalk!_AtalkReceiveComplete+0xdc

801409f4 fc910f88 ff07c508 ff07c628 00000001
NDIS!_EthFilterDprIndicateReceiveComplete+0x33
(FPO: [EBP 0xffdff420] [1,0,4])

80140a04 fc8f04eb ff07c508 ff07c654 8013e820
NDIS!_EthFilterIndicateReceiveComplete+0x21
(FPO: [1,0,2]) 
In this TrapFrame, note that the instruction pointer is running what is at memory address 801144dc. The unassemble shows that the instruction that should have been called is 801144dd, rather than what is currently in EIP.

Trap Frame


kd> !trap 8014068c
!trap 8014068c
eax=ff0f6488 ebx=00000020 ecx=801440e0 edx=ff971b50 esi=00000000
edi=00000358
eip=801144dc esp=80140700 ebp=ff177358 iopl=0         nv up ei ng nz na pe
cy
cs=0008  ss=0010  ds=0023  es=0023  fs=0030  gs=0000
efl=00010283
ErrCode = 00000002
801144dc 0083f810760b     add     [ebx+0xb7610f8],al 

Unassemble of IoAllocateMdl Function and EIP

Correct Unassemble of IoAlloateMdl from the Beginning of the Function:


kd> u NT!_IoAllocateMdl NT!_IoAllocateMdl+0x50
u NT!_IoAllocateMdl NT!_IoAllocateMdl+0x50
NT!_IoAllocateMdl:
80114498 83ec04           sub  esp,0x4
8011449b f644240f80       test byte ptr [esp+0xf],0x80
801144a0 53               push ebx
801144a1 56               push esi
801144a2 57               push edi
801144a3 55               push ebp
801144a4 7407             jz   NT!_IoAllocateMdl+0x15 (801144ad)
801144a6 33c0             xor  eax,eax
801144a8 e99e010000       jmp  NT!_IoAllocateMdl+0x1b3 (8011464b)
801144ad 8b7c2418         mov  edi,[esp+0x18]
801144b1 8b44241c         mov  eax,[esp+0x1c]
801144b5 81e7ff0f0000     and  edi,0xfff
801144bb 33f6             xor  esi,esi
801144bd 03c7             add  eax,edi
801144bf 05ff0f0000       add  eax,0xfff
801144c4 c1e80c           shr  eax,0xc
801144c7 8d1c851c000000   lea  ebx,[0000001c+eax*4]
801144ce 81fbffff0000     cmp  ebx,0xffff
801144d4 7607             jbe  NT!_IoAllocateMdl+0x45 (801144dd)
801144d6 33c0             xor  eax,eax
801144d8 e96e010000       jmp  NT!_IoAllocateMdl+0x1b3 (8011464b)
801144dd 83f810           cmp  eax,0x10   <----------------- Position 1 **
801144e0 760b             jbe  NT!_IoAllocateMdl+0x55 (801144ed)
801144e2 ff0514411480     inc  dword ptr [NT!_IopMdlSizeMiss(80144114)] 


Unassemble of the Function Using the Current Instruction Pointer That is the Offset Into the Function When the Trap

Occurred


kd> u NT!_IoAllocateMdl+0x44
u NT!_IoAllocateMdl+0x44
NT!_IoAllocateMdl+0x44:
801144dc 0083f810760b     add  [ebx+0xb7610f8],al <---------- Position 2**
801144e2 ff0514411480     inc  dword ptr [NT!_IopMdlSizeMiss (80144114)]
801144e8 e99d000000       jmp  NT!_IoAllocateMdl+0xf2 (8011458a)
801144ed ba30e21380       mov  edx,0x8013e230
801144f2 b9e0401480       mov  ecx,0x801440e0
801144f7 e804370200       call NT!ExfInterlockedPopEntryList (80137c00)
801144fc 8bf0             mov  esi,eax
801144fe 85f6             test esi,esi 


The function is unassembled from the beginning, in order to show which instructions should have been run. The unassemble of the function starting at Position 2 should match the unassemble of the function at Position 1.

Parity Error

The difference between 801144dc (what was in EIP) and 801144dd (what should have been in EIP) is one bit, as illustrated below.

801144dc = 10000000000100010100010011011100
801144dd = 10000000000100010100010011011101

This difference of one bit is a parity error that could have been caused by a faulty cache or by faulty memory.

Bad Instruction Pointer

If you break down the Opcode, you will notice that the instruction at address 801144dd is 83f810. The operand is an instruction to compare the sign extended byte to a dword. The processor should have completed a compare of the EAX register to 0X10.

The instruction pointer in EIP was altered by one bit. The system now carries out instructions beginning at 801144dd, which causes the system to fail. The new instruction 0083f810760b is an add-byte register to memory byte.

Breakdown of the Instruction at 801144dd and Opcode


                       Opcode
                         |
                         |

Instruction = 801144dd 83f810  cmp     eax,0x10

Opcode breakdown

     83 = CMP,m32,imm8
     f8 = EAX
     10 = 0x10
 

Breakdown of the Instruction at 801144dc and Opcode


                       Opcode
                         |
                         |

Instruction = 801144dc 0083f810760b  add  [ebx+0xb7610f8],al

Opcode breakdown

     00 = ADD,m32,imm8
     83 = disp32(displace 32 bits)

b7610f8 = address used in ADD 


REFERENCES

The computer in this example failed because the invalid instruction accessed memory that had not been initialized. Once the motherboard was replaced, the computer functioned normally.

For more information about how to debug the kernel, query on the following word here in the Microsoft Knowledge Base:


   debugref 

Additional query words: prodnt debugref

Keywords : ntstop
Version : 3.50 3.51
Platform : WINDOWS
Issue type :


Last Reviewed: February 12, 1999
© 2000 Microsoft Corporation. All rights reserved. Terms of Use.