1. can we directly call AcquireSpinLock()? *OrFail() can be removed IMO.
2. It's a patch to change the behavior of SmmStartupThisAP(). So that to reduce
the potential bugs in caller's code. Patch title is a bit mis-leading.
> -----Original Message-----
> From: Nikodem, Damian
> Sent: Tuesday, September 3, 2019 7:58 AM
> To: devel@edk2.groups.io
> Cc: Nikodem, Damian <damian.niko...@intel.com>; Dong, Eric
> <eric.d...@intel.com>; Ni, Ray <ray...@intel.com>; You,
> Benjamin <benjamin....@intel.com>; Laszlo Ersek <ler...@redhat.com>; Rusocki,
> Krzysztof <krzysztof.ruso...@intel.com>
> Subject: [PATCH] UefiCpuPkg/PiSmmCpuDxeSmm: Race condition between
> APHandler's release of Busy spinlock and user-
> triggered SmmStartupThisAP's
>
> Race condition between APHandler's release of Busy spinlock and
> user-triggered SmmStartupThisAP's acquisition attempt of the Busy spinlock
> (non-blocking mode).
>
> UserProc is the user's procedure to execute on an AP.
> UserProcCompletion is the user procedure's completion spinlock.
> All other variables are from EDK2.
>
> BSP AP
> =====================================================================================
>
> APHandler ()
>
> WaitForSemaphore (Run)
>
> << initial
> state >>
>
> AcquireSpinLock (UserProcCompletion)
> SmmStartupThisAp (Procedure)
> AcquireSpinLockOrFail (Busy)
> ReleaseSemaphore (Run)
>
> UserProc ()
> DoStuff() DoSomeOtherStuff ()
>
> AcquireSpinLockOrFail (UserProcCompletion) AcquireSpinLockOrFail
> (UserProcCompletion)
>
> ^^ waiting in a loop for user procedure's
> completion == these fail
> ReleaseSpinLock (UserProcCompletion) AcquireSpinLockOrFail
> (UserProcCompletion)
>
> ^^ this succeeds
>
> ReleaseSpinLock (UserProcCompletion)
>
> << return control to the caller and
> reenter the flow >>>
>
> AcquireSpinLock (UserProcCompletion)
> SmmStartupThisAp (Procedure)
> AcquireSpinLockOrFail (Busy)
> ^^ this wins the race with AP's
> ReleaseSpinLock and fails;
>
> ReleaseSpinLock (Busy)
> return EFI_INVALID_PARAMETER;
>
> To remedy, if AcquireSpinLockOrFail (of the Busy spinlock) fails, perform
> regular AcquireSpinLock -- this eliminates the race
> condition.
>
> Signed-off-by: Damian Nikodem <damian.niko...@intel.com>
> Cc: Eric Dong <eric.d...@intel.com>
> Cc: Ray Ni <ray...@intel.com>
> Cc: Benjamin You <benjamin....@intel.com>
> Cc: Laszlo Ersek <ler...@redhat.com>
> Cc: Krzysztof Rusocki <krzysztof.ruso...@intel.com>
>
> diff --git a/UefiCpuPkg/PiSmmCpuDxeSmm/MpService.c
> b/UefiCpuPkg/PiSmmCpuDxeSmm/MpService.c
> index d8d2b6f444..206e196a76 100644
> --- a/UefiCpuPkg/PiSmmCpuDxeSmm/MpService.c
> +++ b/UefiCpuPkg/PiSmmCpuDxeSmm/MpService.c
> @@ -1239,8 +1239,16 @@ InternalSmmStartupThisAp (
> AcquireSpinLock (mSmmMpSyncData->CpuData[CpuIndex].Busy);
> } else {
> if (!AcquireSpinLockOrFail (mSmmMpSyncData->CpuData[CpuIndex].Busy)) {
> - DEBUG((DEBUG_ERROR, "Can't acquire
> mSmmMpSyncData->CpuData[%d].Busy\n", CpuIndex));
> - return EFI_NOT_READY;
> + DEBUG ((DEBUG_INFO, "BSP[%d] finds AP[%d] busy at proc 0x%llX (param
> 0x%llX), ",
> + mSmmMpSyncData->BspIndex,
> + CpuIndex,
> + *mSmmMpSyncData->CpuData[CpuIndex].Procedure,
> + (VOID*)mSmmMpSyncData->CpuData[CpuIndex].Parameter));
> + DEBUG ((DEBUG_INFO, "new proc 0x%llX (param 0x%llX). Waiting for the
> previous AP procedure to complete...\n",
> + Procedure,
> + ProcArguments));
> +
> + AcquireSpinLock (mSmmMpSyncData->CpuData[CpuIndex].Busy);
> }
>
> *Token = (MM_COMPLETION) CreateToken ();
-=-=-=-=-=-=-=-=-=-=-=-
Groups.io Links: You receive all messages sent to this group.
View/Reply Online (#46714): https://edk2.groups.io/g/devel/message/46714
Mute This Topic: https://groups.io/mt/33128825/21656
Group Owner: devel+ow...@edk2.groups.io
Unsubscribe: https://edk2.groups.io/g/devel/unsub [arch...@mail-archive.com]
-=-=-=-=-=-=-=-=-=-=-=-
