Intel T8300 Specifications - Page 45

A VM Exit Occurring in IA-32e Mode May Not Produce a VMX Abort

Page 45 highlights

Errata AZ57. A VM Exit Due to a Fault While Delivering a Software Interrupt May Save Incorrect Data into the VMCS Problem: If a fault occurs during delivery of a software interrupt (INTn) in virtual-8086 mode when virtual mode extensions are in effect and that fault causes a VM exit, incorrect data may be saved into the VMCS. Specifically, information about the software interrupt may not be reported in the IDT-vectoring information field. In addition, the interruptibility-state field may indicate blocking by STI or by MOV SS if such blocking were in effect before execution of the INTn instruction or before execution of the VM-entry instruction that injected the software interrupt. Implication: In general, VMM software that follows the guidelines given in the section "Handling VM Exits Due to Exceptions" of Intel® 64 and IA-32 Architectures Software Developer's Manual Volume 3B: System Programming Guide should not be affected. If the erratum improperly causes indication of blocking by STI or by MOV SS, the ability of a VMM to inject an interrupt may be delayed by one instruction. Workaround: VMM software should follow the guidelines given in the section "Handling VM Exits Due to Exceptions" of Intel® 64 and IA-32 Architectures Software Developer's Manual Volume 3B: System Programming Guide. Status: For the steppings affected, see the Summary Tables of Changes. AZ58. A VM Exit Occurring in IA-32e Mode May Not Produce a VMX Abort When Expected Problem: If a VM exit occurs while the processor is in IA-32e mode and the "host address-space size" VM-exit control is 0, a VMX abort should occur. Due to this erratum, the expected VMX aborts may not occur and instead the VM Exit will occur normally. The conditions required to observe this erratum are a VM entry that returns from SMM with the "IA-32e guest" VM-entry control set to 1 in the SMM VMCS and the "host address-space size" VMexit control cleared to 0 in the executive VMCS. Implication: A VM Exit will occur when a VMX Abort was expected. Workaround: An SMM VMM should always set the "IA-32e guest" VM-entry control in the SMM VMCS to be the value that was in the LMA bit (IA32_EFER.LMA.LMA[bit 10]) in the IA32_EFER MSR (C0000080H) at the time of the last SMM VM exit. If this guideline is followed, that value will be 1 only if the "host address-space size" VM-exit control is 1 in the executive VMCS. Status: For the steppings affected, see the Summary Tables of Changes. Specification Update 45

  • 1
  • 2
  • 3
  • 4
  • 5
  • 6
  • 7
  • 8
  • 9
  • 10
  • 11
  • 12
  • 13
  • 14
  • 15
  • 16
  • 17
  • 18
  • 19
  • 20
  • 21
  • 22
  • 23
  • 24
  • 25
  • 26
  • 27
  • 28
  • 29
  • 30
  • 31
  • 32
  • 33
  • 34
  • 35
  • 36
  • 37
  • 38
  • 39
  • 40
  • 41
  • 42
  • 43
  • 44
  • 45
  • 46
  • 47
  • 48
  • 49
  • 50
  • 51
  • 52
  • 53
  • 54
  • 55
  • 56

Errata
Specification Update
45
AZ57.
A VM Exit Due to a Fault While Delivering a Software Interrupt May
Save Incorrect Data into the VMCS
Problem:
If a fault occurs during delivery of a software interrupt (INT
n
) in virtual-8086 mode when
virtual mode extensions are in effect and that fault causes a VM exit, incorrect data may
be saved into the VMCS. Specifically, information about the software interrupt may not
be reported in the IDT-vectoring information field. In addition, the interruptibility-state
field may indicate blocking by STI or by MOV SS if such blocking were in effect before
execution of the INT
n
instruction or before execution of the VM-entry instruction that
injected the software interrupt.
Implication:
In general, VMM software that follows the guidelines given in the section “Handling VM
Exits Due to Exceptions” of
Intel® 64 and IA-
32 Architectures Software Developer’s
Manual Volume 3B: System Programming Guide
should not be affected. If the erratum
improperly causes indication of blocking by STI or by MOV SS, the ability of a VMM to
inject an interrupt may be delayed by one instruction.
Workaround:
VMM s
oftware should follow the guidelines given in the section “Handl
ing VM Exits Due to
Exceptions” of
Intel® 64 and IA-
32 Architectures Software Developer’s Manual Volume
3B: System Programming Guide
.
Status:
For the steppings affected, see the Summary Tables of Changes.
AZ58.
A VM Exit Occurring in IA-32e Mode May Not Produce a VMX Abort
When Expected
Problem:
If a VM exit occurs while the processor is in IA-
32e mode and the “host address
-space
size” VM
-exit control is 0, a VMX abort should occur. Due to this erratum, the expected
VMX aborts may not occur and instead the VM Exit will occur normally. The conditions
required to observe this erratum are a VM entry that returns from SMM with the “IA
-32e
guest” VM
-
entry control set to 1 in the SMM VMCS and the “host address
-
space size” VM
-
exit control cleared to 0 in the executive VMCS.
Implication:
A VM Exit will occur when a VMX Abort was expected.
Workaround:
An SMM VMM should always set the “IA
-
32e guest” VM
-entry control in the SMM VMCS to
be the value that was in the LMA bit (IA32_EFER.LMA.LMA[bit 10]) in the IA32_EFER
MSR (C0000080H) at the time of the last SMM VM exit. If this guideline is followed, that
value will be 1 only if the “host address
-
space size” VM
-exit control is 1 in the executive
VMCS.
Status:
For the steppings affected, see the Summary Tables of Changes.