| Message ID | 20240125032328.2522472-54-xiaoyao.li@intel.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | QEMU Guest memfd + QEMU TDX support | expand |
Xiaoyao Li <xiaoyao.li@intel.com> writes: > Integrate TDX's TDX_REPORT_FATAL_ERROR into QEMU GuestPanic facility > > Originated-from: Isaku Yamahata <isaku.yamahata@intel.com> > Signed-off-by: Xiaoyao Li <xiaoyao.li@intel.com> > --- > Changes in v4: > - refine the documentation; (Markus) > > Changes in v3: > - Add docmentation of new type and struct; (Daniel) > - refine the error message handling; (Daniel) > --- > qapi/run-state.json | 28 ++++++++++++++++++++-- > system/runstate.c | 54 +++++++++++++++++++++++++++++++++++++++++++ > target/i386/kvm/tdx.c | 24 ++++++++++++++++++- > 3 files changed, 103 insertions(+), 3 deletions(-) > > diff --git a/qapi/run-state.json b/qapi/run-state.json > index 08bc99cb8561..5429116679e3 100644 > --- a/qapi/run-state.json > +++ b/qapi/run-state.json > @@ -485,10 +485,12 @@ > # > # @s390: s390 guest panic information type (Since: 2.12) > # > +# @tdx: tdx guest panic information type (Since: 8.2) > +# > # Since: 2.9 > ## > { 'enum': 'GuestPanicInformationType', > - 'data': [ 'hyper-v', 's390' ] } > + 'data': [ 'hyper-v', 's390', 'tdx' ] } > > ## > # @GuestPanicInformation: > @@ -503,7 +505,8 @@ > 'base': {'type': 'GuestPanicInformationType'}, > 'discriminator': 'type', > 'data': {'hyper-v': 'GuestPanicInformationHyperV', > - 's390': 'GuestPanicInformationS390'}} > + 's390': 'GuestPanicInformationS390', > + 'tdx' : 'GuestPanicInformationTdx'}} > > ## > # @GuestPanicInformationHyperV: > @@ -566,6 +569,27 @@ > 'psw-addr': 'uint64', > 'reason': 'S390CrashReason'}} > > +## > +# @GuestPanicInformationTdx: > +# > +# TDX Guest panic information specific to TDX GCHI > +# TDG.VP.VMCALL<ReportFatalError>. > +# > +# @error-code: TD-specific error code Where could a user find information on these error codes? > +# > +# @gpa: guest-physical address of a page that contains additional > +# error data, in forms of zero-terminated string. "in the form of a zero-terminated string" > +# > +# @message: Human-readable error message provided by the guest. Not > +# to be trusted. How is this message related to the one pointed to by @gpa? > +# > +# Since: 9.0 > +## > +{'struct': 'GuestPanicInformationTdx', > + 'data': {'error-code': 'uint64', > + 'gpa': 'uint64', > + 'message': 'str'}} > + > ## > # @MEMORY_FAILURE: > # [...]
On 2/19/2024 8:53 PM, Markus Armbruster wrote: > Xiaoyao Li <xiaoyao.li@intel.com> writes: > >> Integrate TDX's TDX_REPORT_FATAL_ERROR into QEMU GuestPanic facility >> >> Originated-from: Isaku Yamahata <isaku.yamahata@intel.com> >> Signed-off-by: Xiaoyao Li <xiaoyao.li@intel.com> >> --- >> Changes in v4: >> - refine the documentation; (Markus) >> >> Changes in v3: >> - Add docmentation of new type and struct; (Daniel) >> - refine the error message handling; (Daniel) >> --- >> qapi/run-state.json | 28 ++++++++++++++++++++-- >> system/runstate.c | 54 +++++++++++++++++++++++++++++++++++++++++++ >> target/i386/kvm/tdx.c | 24 ++++++++++++++++++- >> 3 files changed, 103 insertions(+), 3 deletions(-) >> >> diff --git a/qapi/run-state.json b/qapi/run-state.json >> index 08bc99cb8561..5429116679e3 100644 >> --- a/qapi/run-state.json >> +++ b/qapi/run-state.json >> @@ -485,10 +485,12 @@ >> # >> # @s390: s390 guest panic information type (Since: 2.12) >> # >> +# @tdx: tdx guest panic information type (Since: 8.2) >> +# >> # Since: 2.9 >> ## >> { 'enum': 'GuestPanicInformationType', >> - 'data': [ 'hyper-v', 's390' ] } >> + 'data': [ 'hyper-v', 's390', 'tdx' ] } >> >> ## >> # @GuestPanicInformation: >> @@ -503,7 +505,8 @@ >> 'base': {'type': 'GuestPanicInformationType'}, >> 'discriminator': 'type', >> 'data': {'hyper-v': 'GuestPanicInformationHyperV', >> - 's390': 'GuestPanicInformationS390'}} >> + 's390': 'GuestPanicInformationS390', >> + 'tdx' : 'GuestPanicInformationTdx'}} >> >> ## >> # @GuestPanicInformationHyperV: >> @@ -566,6 +569,27 @@ >> 'psw-addr': 'uint64', >> 'reason': 'S390CrashReason'}} >> >> +## >> +# @GuestPanicInformationTdx: >> +# >> +# TDX Guest panic information specific to TDX GCHI >> +# TDG.VP.VMCALL<ReportFatalError>. >> +# >> +# @error-code: TD-specific error code > > Where could a user find information on these error codes? TDX GHCI (Guset-host-communication-Interface)spec. It defines all the TDVMCALL leaves. 0: panic; 0x1 - 0xffffffff: reserved. >> +# >> +# @gpa: guest-physical address of a page that contains additional >> +# error data, in forms of zero-terminated string. > > "in the form of a zero-terminated string" fixed. >> +# >> +# @message: Human-readable error message provided by the guest. Not >> +# to be trusted. > > How is this message related to the one pointed to by @gpa? In general, @message contains a brief message of the error. While @gpa (when valid) contains a verbose message. The reason why we need both is because sometime when TD guest hits a fatal error, its memory may get corrupted so we cannot pass information via @gpa. Information in @message is passed through GPRs. >> +# >> +# Since: 9.0 >> +## >> +{'struct': 'GuestPanicInformationTdx', >> + 'data': {'error-code': 'uint64', >> + 'gpa': 'uint64', >> + 'message': 'str'}} >> + >> ## >> # @MEMORY_FAILURE: >> # > > [...] >
Xiaoyao Li <xiaoyao.li@intel.com> writes: > On 2/19/2024 8:53 PM, Markus Armbruster wrote: >> Xiaoyao Li <xiaoyao.li@intel.com> writes: >> >>> Integrate TDX's TDX_REPORT_FATAL_ERROR into QEMU GuestPanic facility >>> >>> Originated-from: Isaku Yamahata <isaku.yamahata@intel.com> >>> Signed-off-by: Xiaoyao Li <xiaoyao.li@intel.com> >>> --- >>> Changes in v4: >>> - refine the documentation; (Markus) >>> >>> Changes in v3: >>> - Add docmentation of new type and struct; (Daniel) >>> - refine the error message handling; (Daniel) >>> --- >>> qapi/run-state.json | 28 ++++++++++++++++++++-- >>> system/runstate.c | 54 +++++++++++++++++++++++++++++++++++++++++++ >>> target/i386/kvm/tdx.c | 24 ++++++++++++++++++- >>> 3 files changed, 103 insertions(+), 3 deletions(-) >>> >>> diff --git a/qapi/run-state.json b/qapi/run-state.json >>> index 08bc99cb8561..5429116679e3 100644 >>> --- a/qapi/run-state.json >>> +++ b/qapi/run-state.json >>> @@ -485,10 +485,12 @@ >>> # >>> # @s390: s390 guest panic information type (Since: 2.12) >>> # >>> +# @tdx: tdx guest panic information type (Since: 8.2) >>> +# >>> # Since: 2.9 >>> ## >>> { 'enum': 'GuestPanicInformationType', >>> - 'data': [ 'hyper-v', 's390' ] } >>> + 'data': [ 'hyper-v', 's390', 'tdx' ] } >>> >>> ## >>> # @GuestPanicInformation: >>> @@ -503,7 +505,8 @@ >>> 'base': {'type': 'GuestPanicInformationType'}, >>> 'discriminator': 'type', >>> 'data': {'hyper-v': 'GuestPanicInformationHyperV', >>> - 's390': 'GuestPanicInformationS390'}} >>> + 's390': 'GuestPanicInformationS390', >>> + 'tdx' : 'GuestPanicInformationTdx'}} >>> >>> ## >>> # @GuestPanicInformationHyperV: >>> @@ -566,6 +569,27 @@ >>> 'psw-addr': 'uint64', >>> 'reason': 'S390CrashReason'}} >>> >>> +## >>> +# @GuestPanicInformationTdx: >>> +# >>> +# TDX Guest panic information specific to TDX GCHI >>> +# TDG.VP.VMCALL<ReportFatalError>. >>> +# >>> +# @error-code: TD-specific error code >> >> Where could a user find information on these error codes? > > TDX GHCI (Guset-host-communication-Interface)spec. It defines all the > TDVMCALL leaves. > > 0: panic; > 0x1 - 0xffffffff: reserved. Would it make sense to add a reference? >>> +# >>> +# @gpa: guest-physical address of a page that contains additional >>> +# error data, in forms of zero-terminated string. >> >> "in the form of a zero-terminated string" > > fixed. > >>> +# >>> +# @message: Human-readable error message provided by the guest. Not >>> +# to be trusted. >> >> How is this message related to the one pointed to by @gpa? > > In general, @message contains a brief message of the error. While @gpa > (when valid) contains a verbose message. > > The reason why we need both is because sometime when TD guest hits a > fatal error, its memory may get corrupted so we cannot pass information > via @gpa. Information in @message is passed through GPRs. Well, we do pass information via @gpa, always. I guess it page's contents can be corrupted. Perhaps something like # @message: Human-readable error message provided by the guest. Not # to be trusted. # # @gpa: guest-physical address of a page that contains more verbose # error information, as zero-terminated string. Note that guest # memory corruption can corrupt the page's contents. >>> +# >>> +# Since: 9.0 >>> +## >>> +{'struct': 'GuestPanicInformationTdx', >>> + 'data': {'error-code': 'uint64', >>> + 'gpa': 'uint64', >>> + 'message': 'str'}} Note that my proposed doc string has the members in a different order. Recommend to use the same order here. >>> + >>> ## >>> # @MEMORY_FAILURE: >>> # >> >> [...] >>
On 2/27/2024 7:51 PM, Markus Armbruster wrote: > Xiaoyao Li <xiaoyao.li@intel.com> writes: > >> On 2/19/2024 8:53 PM, Markus Armbruster wrote: >>> Xiaoyao Li <xiaoyao.li@intel.com> writes: >>> >>>> Integrate TDX's TDX_REPORT_FATAL_ERROR into QEMU GuestPanic facility >>>> >>>> Originated-from: Isaku Yamahata <isaku.yamahata@intel.com> >>>> Signed-off-by: Xiaoyao Li <xiaoyao.li@intel.com> >>>> --- >>>> Changes in v4: >>>> - refine the documentation; (Markus) >>>> >>>> Changes in v3: >>>> - Add docmentation of new type and struct; (Daniel) >>>> - refine the error message handling; (Daniel) >>>> --- >>>> qapi/run-state.json | 28 ++++++++++++++++++++-- >>>> system/runstate.c | 54 +++++++++++++++++++++++++++++++++++++++++++ >>>> target/i386/kvm/tdx.c | 24 ++++++++++++++++++- >>>> 3 files changed, 103 insertions(+), 3 deletions(-) >>>> >>>> diff --git a/qapi/run-state.json b/qapi/run-state.json >>>> index 08bc99cb8561..5429116679e3 100644 >>>> --- a/qapi/run-state.json >>>> +++ b/qapi/run-state.json >>>> @@ -485,10 +485,12 @@ >>>> # >>>> # @s390: s390 guest panic information type (Since: 2.12) >>>> # >>>> +# @tdx: tdx guest panic information type (Since: 8.2) >>>> +# >>>> # Since: 2.9 >>>> ## >>>> { 'enum': 'GuestPanicInformationType', >>>> - 'data': [ 'hyper-v', 's390' ] } >>>> + 'data': [ 'hyper-v', 's390', 'tdx' ] } >>>> >>>> ## >>>> # @GuestPanicInformation: >>>> @@ -503,7 +505,8 @@ >>>> 'base': {'type': 'GuestPanicInformationType'}, >>>> 'discriminator': 'type', >>>> 'data': {'hyper-v': 'GuestPanicInformationHyperV', >>>> - 's390': 'GuestPanicInformationS390'}} >>>> + 's390': 'GuestPanicInformationS390', >>>> + 'tdx' : 'GuestPanicInformationTdx'}} >>>> >>>> ## >>>> # @GuestPanicInformationHyperV: >>>> @@ -566,6 +569,27 @@ >>>> 'psw-addr': 'uint64', >>>> 'reason': 'S390CrashReason'}} >>>> >>>> +## >>>> +# @GuestPanicInformationTdx: >>>> +# >>>> +# TDX Guest panic information specific to TDX GCHI >>>> +# TDG.VP.VMCALL<ReportFatalError>. >>>> +# >>>> +# @error-code: TD-specific error code >>> >>> Where could a user find information on these error codes? >> >> TDX GHCI (Guset-host-communication-Interface)spec. It defines all the >> TDVMCALL leaves. >> >> 0: panic; >> 0x1 - 0xffffffff: reserved. > > Would it make sense to add a reference? https://cdrdv2.intel.com/v1/dl/getContent/726792 >>>> +# >>>> +# @gpa: guest-physical address of a page that contains additional >>>> +# error data, in forms of zero-terminated string. >>> >>> "in the form of a zero-terminated string" >> >> fixed. >> >>>> +# >>>> +# @message: Human-readable error message provided by the guest. Not >>>> +# to be trusted. >>> >>> How is this message related to the one pointed to by @gpa? >> >> In general, @message contains a brief message of the error. While @gpa >> (when valid) contains a verbose message. >> >> The reason why we need both is because sometime when TD guest hits a >> fatal error, its memory may get corrupted so we cannot pass information >> via @gpa. Information in @message is passed through GPRs. > > Well, we do pass information via @gpa, always. I guess it page's > contents can be corrupted. No. It's not always. the bit 63 of the error code is "GPA valid" bit. @gpa is valid only when bit 63 of error code is 1. And current Linux TD guest implementation doesn't use @gpa at all. https://github.com/torvalds/linux/blob/45ec2f5f6ed3ec3a79ba1329ad585497cdcbe663/arch/x86/coco/tdx/tdx.c#L131 > Perhaps something like > > # @message: Human-readable error message provided by the guest. Not > # to be trusted. > # > # @gpa: guest-physical address of a page that contains more verbose > # error information, as zero-terminated string. Note that guest > # memory corruption can corrupt the page's contents. > >>>> +# >>>> +# Since: 9.0 >>>> +## >>>> +{'struct': 'GuestPanicInformationTdx', >>>> + 'data': {'error-code': 'uint64', >>>> + 'gpa': 'uint64', >>>> + 'message': 'str'}} > > Note that my proposed doc string has the members in a different order. > Recommend to use the same order here. > >>>> + >>>> ## >>>> # @MEMORY_FAILURE: >>>> # >>> >>> [...] >>> >
Xiaoyao Li <xiaoyao.li@intel.com> writes: > On 2/27/2024 7:51 PM, Markus Armbruster wrote: >> Xiaoyao Li <xiaoyao.li@intel.com> writes: >> >>> On 2/19/2024 8:53 PM, Markus Armbruster wrote: >>>> Xiaoyao Li <xiaoyao.li@intel.com> writes: >>>> >>>>> Integrate TDX's TDX_REPORT_FATAL_ERROR into QEMU GuestPanic facility >>>>> >>>>> Originated-from: Isaku Yamahata <isaku.yamahata@intel.com> >>>>> Signed-off-by: Xiaoyao Li <xiaoyao.li@intel.com> >>>>> --- >>>>> Changes in v4: >>>>> - refine the documentation; (Markus) >>>>> >>>>> Changes in v3: >>>>> - Add docmentation of new type and struct; (Daniel) >>>>> - refine the error message handling; (Daniel) >>>>> --- >>>>> qapi/run-state.json | 28 ++++++++++++++++++++-- >>>>> system/runstate.c | 54 +++++++++++++++++++++++++++++++++++++++++++ >>>>> target/i386/kvm/tdx.c | 24 ++++++++++++++++++- >>>>> 3 files changed, 103 insertions(+), 3 deletions(-) >>>>> >>>>> diff --git a/qapi/run-state.json b/qapi/run-state.json >>>>> index 08bc99cb8561..5429116679e3 100644 >>>>> --- a/qapi/run-state.json >>>>> +++ b/qapi/run-state.json [...] >>>>> @@ -566,6 +569,27 @@ >>>>> 'psw-addr': 'uint64', >>>>> 'reason': 'S390CrashReason'}} >>>>> >>>>> +## >>>>> +# @GuestPanicInformationTdx: >>>>> +# >>>>> +# TDX Guest panic information specific to TDX GCHI >>>>> +# TDG.VP.VMCALL<ReportFatalError>. >>>>> +# >>>>> +# @error-code: TD-specific error code >>>> >>>> Where could a user find information on these error codes? >>> >>> TDX GHCI (Guset-host-communication-Interface)spec. It defines all the >>> TDVMCALL leaves. >>> >>> 0: panic; >>> 0x1 - 0xffffffff: reserved. >> >> Would it make sense to add a reference? > > https://cdrdv2.intel.com/v1/dl/getContent/726792 URLs have this annoying tendency to rot. What about # @error-code: Error code as defined in "Guest-Hypervisor Communication # Interface (GHCI) Specification for Intel TDX 1.5" >>>>> +# >>>>> +# @gpa: guest-physical address of a page that contains additional >>>>> +# error data, in forms of zero-terminated string. >>>> >>>> "in the form of a zero-terminated string" >>> >>> fixed. >>> >>>>> +# >>>>> +# @message: Human-readable error message provided by the guest. Not >>>>> +# to be trusted. >>>> >>>> How is this message related to the one pointed to by @gpa? >>> >>> In general, @message contains a brief message of the error. While @gpa >>> (when valid) contains a verbose message. >>> >>> The reason why we need both is because sometime when TD guest hits a >>> fatal error, its memory may get corrupted so we cannot pass information >>> via @gpa. Information in @message is passed through GPRs. >> >> Well, we do pass information via @gpa, always. I guess it page's >> contents can be corrupted. > > No. It's not always. the bit 63 of the error code is "GPA valid" bit. > @gpa is valid only when bit 63 of error code is 1. > > And current Linux TD guest implementation doesn't use @gpa at all. > https://github.com/torvalds/linux/blob/45ec2f5f6ed3ec3a79ba1329ad585497cdcbe663/arch/x86/coco/tdx/tdx.c#L131 Aha! Why would we want to include @gpa when the "GPA valid" bit is off? If we do want it, then # @gpa: guest-physical address of a page that contains more verbose # error information, as zero-terminated string. Valid when the # "GPA valid" bit is set in @error-code. If we don't, then make @gpa optional, present when valid, and document it like this: # @gpa: guest-physical address of a page that contains more verbose # error information, as zero-terminated string. Present when the # "GPA valid" bit is set in @error-code. >> Perhaps something like >> >> # @message: Human-readable error message provided by the guest. Not >> # to be trusted. >> # >> # @gpa: guest-physical address of a page that contains more verbose >> # error information, as zero-terminated string. Note that guest >> # memory corruption can corrupt the page's contents. >> >>>>> +# >>>>> +# Since: 9.0 >>>>> +## >>>>> +{'struct': 'GuestPanicInformationTdx', >>>>> + 'data': {'error-code': 'uint64', >>>>> + 'gpa': 'uint64', >>>>> + 'message': 'str'}} >> >> Note that my proposed doc string has the members in a different order. >> Recommend to use the same order here. >> >>>>> + >>>>> ## >>>>> # @MEMORY_FAILURE: >>>>> # >>>> >>>> [...] >>>> >>
On 2/27/2024 9:09 PM, Markus Armbruster wrote: [...] >>>>>> @@ -566,6 +569,27 @@ >>>>>> 'psw-addr': 'uint64', >>>>>> 'reason': 'S390CrashReason'}} >>>>>> >>>>>> +## >>>>>> +# @GuestPanicInformationTdx: >>>>>> +# >>>>>> +# TDX Guest panic information specific to TDX GCHI >>>>>> +# TDG.VP.VMCALL<ReportFatalError>. >>>>>> +# >>>>>> +# @error-code: TD-specific error code >>>>> >>>>> Where could a user find information on these error codes? >>>> >>>> TDX GHCI (Guset-host-communication-Interface)spec. It defines all the >>>> TDVMCALL leaves. >>>> >>>> 0: panic; >>>> 0x1 - 0xffffffff: reserved. >>> >>> Would it make sense to add a reference? >> >> https://cdrdv2.intel.com/v1/dl/getContent/726792 > > URLs have this annoying tendency to rot. > > What about > > # @error-code: Error code as defined in "Guest-Hypervisor Communication > # Interface (GHCI) Specification for Intel TDX 1.5" I think it gets mentioned at the beginning of @GuestPanicInformationTdx TDX Guest panic information specific to TDX GHCI TDG.VP.VMCALL<ReportFatalError>. Do we still to mention it in every single member? >>>>>> +# >>>>>> +# @gpa: guest-physical address of a page that contains additional >>>>>> +# error data, in forms of zero-terminated string. >>>>> >>>>> "in the form of a zero-terminated string" >>>> >>>> fixed. >>>> >>>>>> +# >>>>>> +# @message: Human-readable error message provided by the guest. Not >>>>>> +# to be trusted. >>>>> >>>>> How is this message related to the one pointed to by @gpa? >>>> >>>> In general, @message contains a brief message of the error. While @gpa >>>> (when valid) contains a verbose message. >>>> >>>> The reason why we need both is because sometime when TD guest hits a >>>> fatal error, its memory may get corrupted so we cannot pass information >>>> via @gpa. Information in @message is passed through GPRs. >>> >>> Well, we do pass information via @gpa, always. I guess it page's >>> contents can be corrupted. >> >> No. It's not always. the bit 63 of the error code is "GPA valid" bit. >> @gpa is valid only when bit 63 of error code is 1. >> >> And current Linux TD guest implementation doesn't use @gpa at all. >> https://github.com/torvalds/linux/blob/45ec2f5f6ed3ec3a79ba1329ad585497cdcbe663/arch/x86/coco/tdx/tdx.c#L131 > > Aha! > > Why would we want to include @gpa when the "GPA valid" bit is off? > > If we do want it, then > > # @gpa: guest-physical address of a page that contains more verbose > # error information, as zero-terminated string. Valid when the > # "GPA valid" bit is set in @error-code. > > If we don't, then make @gpa optional, present when valid, and document > it like this: > > # @gpa: guest-physical address of a page that contains more verbose > # error information, as zero-terminated string. Present when the > # "GPA valid" bit is set in @error-code. I will go this direction. thanks! >>> Perhaps something like >>> >>> # @message: Human-readable error message provided by the guest. Not >>> # to be trusted. >>> # >>> # @gpa: guest-physical address of a page that contains more verbose >>> # error information, as zero-terminated string. Note that guest >>> # memory corruption can corrupt the page's contents. >>> >>>>>> +# >>>>>> +# Since: 9.0 >>>>>> +## >>>>>> +{'struct': 'GuestPanicInformationTdx', >>>>>> + 'data': {'error-code': 'uint64', >>>>>> + 'gpa': 'uint64', >>>>>> + 'message': 'str'}} >>> >>> Note that my proposed doc string has the members in a different order. >>> Recommend to use the same order here. >>> >>>>>> + >>>>>> ## >>>>>> # @MEMORY_FAILURE: >>>>>> # >>>>> >>>>> [...] >>>>> >>> > >
Xiaoyao Li <xiaoyao.li@intel.com> writes: > On 2/27/2024 9:09 PM, Markus Armbruster wrote: > [...] >>>>>>> @@ -566,6 +569,27 @@ >>>>>>> 'psw-addr': 'uint64', >>>>>>> 'reason': 'S390CrashReason'}} >>>>>>> >>>>>>> +## >>>>>>> +# @GuestPanicInformationTdx: >>>>>>> +# >>>>>>> +# TDX Guest panic information specific to TDX GCHI >>>>>>> +# TDG.VP.VMCALL<ReportFatalError>. >>>>>>> +# >>>>>>> +# @error-code: TD-specific error code >>>>>> >>>>>> Where could a user find information on these error codes? >>>>> >>>>> TDX GHCI (Guset-host-communication-Interface)spec. It defines all the >>>>> TDVMCALL leaves. >>>>> >>>>> 0: panic; >>>>> 0x1 - 0xffffffff: reserved. >>>> >>>> Would it make sense to add a reference? >>> >>> https://cdrdv2.intel.com/v1/dl/getContent/726792 >> >> URLs have this annoying tendency to rot. >> >> What about >> >> # @error-code: Error code as defined in "Guest-Hypervisor Communication >> # Interface (GHCI) Specification for Intel TDX 1.5" > > I think it gets mentioned at the beginning of @GuestPanicInformationTdx > > TDX Guest panic information specific to TDX GHCI > TDG.VP.VMCALL<ReportFatalError>. > > Do we still to mention it in every single member? No, I didn't recognize the alphabet soup there as a reference :) Let me try again: ## # @GuestPanicInformationTdx: # # TDX Guest panic information specific to TDX, as specified in the # "Guest-Hypervisor Communication Interface (GHCI) Specification", # section TDG.VP.VMCALL<ReportFatalError>. # # @error-code: TD-specific error code # [...] # # Since: 9.0 ## >>>>>>> +# >>>>>>> +# @gpa: guest-physical address of a page that contains additional >>>>>>> +# error data, in forms of zero-terminated string. >>>>>> >>>>>> "in the form of a zero-terminated string" >>>>> >>>>> fixed. >>>>> >>>>>>> +# >>>>>>> +# @message: Human-readable error message provided by the guest. Not >>>>>>> +# to be trusted. >>>>>> >>>>>> How is this message related to the one pointed to by @gpa? >>>>> >>>>> In general, @message contains a brief message of the error. While @gpa >>>>> (when valid) contains a verbose message. >>>>> >>>>> The reason why we need both is because sometime when TD guest hits a >>>>> fatal error, its memory may get corrupted so we cannot pass information >>>>> via @gpa. Information in @message is passed through GPRs. >>>> >>>> Well, we do pass information via @gpa, always. I guess it page's >>>> contents can be corrupted. >>> >>> No. It's not always. the bit 63 of the error code is "GPA valid" bit. >>> @gpa is valid only when bit 63 of error code is 1. >>> >>> And current Linux TD guest implementation doesn't use @gpa at all. >>> https://github.com/torvalds/linux/blob/45ec2f5f6ed3ec3a79ba1329ad585497cdcbe663/arch/x86/coco/tdx/tdx.c#L131 >> >> Aha! >> >> Why would we want to include @gpa when the "GPA valid" bit is off? >> >> If we do want it, then >> >> # @gpa: guest-physical address of a page that contains more verbose >> # error information, as zero-terminated string. Valid when the >> # "GPA valid" bit is set in @error-code. >> >> If we don't, then make @gpa optional, present when valid, and document >> it like this: >> >> # @gpa: guest-physical address of a page that contains more verbose >> # error information, as zero-terminated string. Present when the >> # "GPA valid" bit is set in @error-code. > > I will go this direction. > > thanks! You're welcome! [...]
diff --git a/qapi/run-state.json b/qapi/run-state.json index 08bc99cb8561..5429116679e3 100644 --- a/qapi/run-state.json +++ b/qapi/run-state.json @@ -485,10 +485,12 @@ # # @s390: s390 guest panic information type (Since: 2.12) # +# @tdx: tdx guest panic information type (Since: 8.2) +# # Since: 2.9 ## { 'enum': 'GuestPanicInformationType', - 'data': [ 'hyper-v', 's390' ] } + 'data': [ 'hyper-v', 's390', 'tdx' ] } ## # @GuestPanicInformation: @@ -503,7 +505,8 @@ 'base': {'type': 'GuestPanicInformationType'}, 'discriminator': 'type', 'data': {'hyper-v': 'GuestPanicInformationHyperV', - 's390': 'GuestPanicInformationS390'}} + 's390': 'GuestPanicInformationS390', + 'tdx' : 'GuestPanicInformationTdx'}} ## # @GuestPanicInformationHyperV: @@ -566,6 +569,27 @@ 'psw-addr': 'uint64', 'reason': 'S390CrashReason'}} +## +# @GuestPanicInformationTdx: +# +# TDX Guest panic information specific to TDX GCHI +# TDG.VP.VMCALL<ReportFatalError>. +# +# @error-code: TD-specific error code +# +# @gpa: guest-physical address of a page that contains additional +# error data, in forms of zero-terminated string. +# +# @message: Human-readable error message provided by the guest. Not +# to be trusted. +# +# Since: 9.0 +## +{'struct': 'GuestPanicInformationTdx', + 'data': {'error-code': 'uint64', + 'gpa': 'uint64', + 'message': 'str'}} + ## # @MEMORY_FAILURE: # diff --git a/system/runstate.c b/system/runstate.c index d6ab860ecaa7..1ae85ea2c345 100644 --- a/system/runstate.c +++ b/system/runstate.c @@ -519,6 +519,52 @@ static void qemu_system_wakeup(void) } } +static char* tdx_parse_panic_message(char *message) +{ + bool printable = false; + char *buf = NULL; + int len = 0, i; + + /* + * Although message is defined as a json string, we shouldn't + * unconditionally treat it as is because the guest generated it and + * it's not necessarily trustable. + */ + if (message) { + /* The caller guarantees the NUL-terminated string. */ + len = strlen(message); + + printable = len > 0; + for (i = 0; i < len; i++) { + if (!(0x20 <= message[i] && message[i] <= 0x7e)) { + printable = false; + break; + } + } + } + + if (!printable && len) { + /* 3 = length of "%02x " */ + buf = g_malloc(len * 3); + for (i = 0; i < len; i++) { + if (message[i] == '\0') { + break; + } else { + sprintf(buf + 3 * i, "%02x ", message[i]); + } + } + if (i > 0) + /* replace the last ' '(space) to NUL */ + buf[i * 3 - 1] = '\0'; + else + buf[0] = '\0'; + + return buf; + } + + return message; +} + void qemu_system_guest_panicked(GuestPanicInformation *info) { qemu_log_mask(LOG_GUEST_ERROR, "Guest crashed"); @@ -560,7 +606,15 @@ void qemu_system_guest_panicked(GuestPanicInformation *info) S390CrashReason_str(info->u.s390.reason), info->u.s390.psw_mask, info->u.s390.psw_addr); + } else if (info->type == GUEST_PANIC_INFORMATION_TYPE_TDX) { + qemu_log_mask(LOG_GUEST_ERROR, + " TDX guest reports fatal error:\"%s\"" + " error code: 0x%016" PRIx64 " gpa page: 0x%016" PRIx64 "\n", + tdx_parse_panic_message(info->u.tdx.message), + info->u.tdx.error_code, + info->u.tdx.gpa); } + qapi_free_GuestPanicInformation(info); } } diff --git a/target/i386/kvm/tdx.c b/target/i386/kvm/tdx.c index 1c79032ca262..4fbb18135951 100644 --- a/target/i386/kvm/tdx.c +++ b/target/i386/kvm/tdx.c @@ -20,6 +20,7 @@ #include "qom/object_interfaces.h" #include "standard-headers/asm-x86/kvm_para.h" #include "sysemu/kvm.h" +#include "sysemu/runstate.h" #include "sysemu/sysemu.h" #include "exec/address-spaces.h" #include "exec/ramblock.h" @@ -1078,11 +1079,26 @@ static int tdx_handle_get_quote(X86CPU *cpu, struct kvm_tdx_vmcall *vmcall) return 0; } +static void tdx_panicked_on_fatal_error(X86CPU *cpu, uint64_t error_code, + uint64_t gpa, char *message) +{ + GuestPanicInformation *panic_info; + + panic_info = g_new0(GuestPanicInformation, 1); + panic_info->type = GUEST_PANIC_INFORMATION_TYPE_TDX; + panic_info->u.tdx.error_code = error_code; + panic_info->u.tdx.gpa = gpa; + panic_info->u.tdx.message = message; + + qemu_system_guest_panicked(panic_info); +} + static int tdx_handle_report_fatal_error(X86CPU *cpu, struct kvm_tdx_vmcall *vmcall) { uint64_t error_code = vmcall->in_r12; char *message = NULL; + uint64_t gpa = -1ull; if (error_code & 0xffff) { error_report("TDX: REPORT_FATAL_ERROR: invalid error code: " @@ -1111,7 +1127,13 @@ static int tdx_handle_report_fatal_error(X86CPU *cpu, assert((char *)tmp == message + GUEST_PANIC_INFO_TDX_MESSAGE_MAX); } - error_report("TD guest reports fatal error. %s\n", message ? : ""); +#define TDX_REPORT_FATAL_ERROR_GPA_VALID BIT_ULL(63) + if (error_code & TDX_REPORT_FATAL_ERROR_GPA_VALID) { + gpa = vmcall->in_r13; + } + + tdx_panicked_on_fatal_error(cpu, error_code, gpa, message); + return -1; }
Integrate TDX's TDX_REPORT_FATAL_ERROR into QEMU GuestPanic facility Originated-from: Isaku Yamahata <isaku.yamahata@intel.com> Signed-off-by: Xiaoyao Li <xiaoyao.li@intel.com> --- Changes in v4: - refine the documentation; (Markus) Changes in v3: - Add docmentation of new type and struct; (Daniel) - refine the error message handling; (Daniel) --- qapi/run-state.json | 28 ++++++++++++++++++++-- system/runstate.c | 54 +++++++++++++++++++++++++++++++++++++++++++ target/i386/kvm/tdx.c | 24 ++++++++++++++++++- 3 files changed, 103 insertions(+), 3 deletions(-)