Re: [PATCH 0/2] UefiCpuPkg/MpInitLib: recover from a lost Start-up IPI #VSTOR-146639

["Denis V. Lunev" <den@virtuozzo.com>]

On 9/23/26 15:30, Andrey Drobyshev wrote:
> On 9/21/26 6:44 PM, Denis V. Lunev wrote:
>> WaitApWakeup() spins until an AP clears its wakeup signal, with no timeout
>> and no way out. A hypervisor can lose the Start-up IPI: KVM latches INIT
>> while the target vCPU is in SMM but discards the SIPI outright. The AP is
>> left in wait-for-SIPI, the BSP spins forever, and the guest never boots.
>>
>> Analysis of the live hung guest is in the bug.
>>
>> Patch 1 exports SendStartupIpi() from LocalApicLib, so one AP can be poked
>> without an INIT that would reset an AP already running the procedure.
>> Patch 2 makes WaitApWakeup() resend rather than spin.
>>
>> Verified by dropping Start-up IPIs in KVM with a kprobe on
>> kvm_apic_send_ipi(); both firmware binaries are unmodified. Killing one
>> INIT-SIPI-SIPI sequence hangs the current firmware at all 8 drop points
>> tried, leaving the APs at CR0=0x00000010. The patched firmware boots at
>> all 8, issuing exactly one directed Start-up IPI per stranded AP. Without
>> injection, 80 boots patched against stock show no regression.
>>
>> Submitted upstream as https://github.com/tianocore/edk2/pull/13164
>>
>> Denis V. Lunev (2):
>>   UefiCpuPkg: add SendStartupIpi to LocalApicLib #VSTOR-146639
>>   UefiCpuPkg/MpInitLib: resend the Start-up IPI when an AP misses it
>>     #VSTOR-146639
>>
>>  UefiCpuPkg/Include/Library/LocalApicLib.h     | 19 +++++
>>  .../Library/BaseXApicLib/BaseXApicLib.c       | 32 +++++--
>>  .../BaseXApicX2ApicLib/BaseXApicX2ApicLib.c   | 32 +++++--
>>  UefiCpuPkg/Library/MpInitLib/MpLib.c          | 85 +++++++++++++++++--
>>  UefiCpuPkg/Library/MpInitLib/MpLib.h          |  2 +
>>  5 files changed, 153 insertions(+), 17 deletions(-)
>
> Please put the component name in the subject, we have multiple
> components in development in svt-core@.  Same for the branch name.  We
> have FC for hci-8.0 now - should this series go into 8.0, or 8.1, or both?
>
> Andrey
both