Re: [Patch 2/2] UefiCpuPkg/MpInitLib: Remove redundant parameter.

["Dong, Eric" <eric.dong@intel.com>]

Hi Laszlo,

> -----Original Message-----
> From: Laszlo Ersek [mailto:lersek@redhat.com]
> Sent: Thursday, June 28, 2018 11:37 PM
> To: Dong, Eric <eric.dong@intel.com>; edk2-devel@lists.01.org
> Cc: Ni, Ruiyu <ruiyu.ni@intel.com>
> Subject: Re: [edk2] [Patch 2/2] UefiCpuPkg/MpInitLib: Remove redundant
> parameter.
> 
> Hi Eric,
> 
> On 06/28/18 13:29, Eric Dong wrote:
> > Parameter StartCount duplicates with RunningCount. After this change,
> > RunningCount means the running AP count.
> >
> > Done Tests:
> > 1.PI SCT Test
> > 2.Boot OS / S3
> >
> > Cc: Ruiyu Ni <ruiyu.ni@intel.com>
> > Cc: Jeff Fan <vanjeff_919@hotmail.com>
> > Cc: Laszlo Ersek <lersek@redhat.com>
> > Contributed-under: TianoCore Contribution Agreement 1.1
> > Signed-off-by: Eric Dong <eric.dong@intel.com>
> > ---
> >  UefiCpuPkg/Library/MpInitLib/MpLib.c | 11 +++++------
> > UefiCpuPkg/Library/MpInitLib/MpLib.h |  1 -
> >  2 files changed, 5 insertions(+), 7 deletions(-)
> >
> > diff --git a/UefiCpuPkg/Library/MpInitLib/MpLib.c
> > b/UefiCpuPkg/Library/MpInitLib/MpLib.c
> > index 3945771764..52c9679099 100644
> > --- a/UefiCpuPkg/Library/MpInitLib/MpLib.c
> > +++ b/UefiCpuPkg/Library/MpInitLib/MpLib.c
> > @@ -1400,7 +1400,7 @@ CheckAllAPs (
> >      // value of state after setting the it to CpuStateFinished, so BSP can safely
> make use of its value.
> >      //
> >      if (GetApState(CpuData) != CpuStateBusy) {
> > -      CpuMpData->RunningCount ++;
> > +      CpuMpData->RunningCount --;
> >        CpuMpData->CpuData[ProcessorNumber].Waiting = FALSE;
> >
> >        //
> > @@ -1425,7 +1425,7 @@ CheckAllAPs (
> >    //
> >    // If all APs finish, return EFI_SUCCESS.
> >    //
> > -  if (CpuMpData->RunningCount == CpuMpData->StartCount) {
> > +  if (CpuMpData->RunningCount == 0) {
> >      return EFI_SUCCESS;
> >    }
> >
> > @@ -1442,7 +1442,7 @@ CheckAllAPs (
> >      //
> >      if (CpuMpData->FailedCpuList != NULL) {
> >        *CpuMpData->FailedCpuList =
> > -         AllocatePool ((CpuMpData->StartCount - CpuMpData->FinishedCount
> + 1) * sizeof (UINTN));
> > +         AllocatePool ((CpuMpData->RunningCount + 1) * sizeof
> > + (UINTN));
> >        ASSERT (*CpuMpData->FailedCpuList != NULL);
> >      }
> >      ListIndex = 0;
> > @@ -2121,7 +2121,7 @@ StartupAllAPsWorker (
> >      return EFI_NOT_STARTED;
> >    }
> >
> > -  CpuMpData->StartCount = 0;
> > +  CpuMpData->RunningCount = 0;
> >    for (ProcessorNumber = 0; ProcessorNumber < ProcessorCount;
> ProcessorNumber++) {
> >      CpuData = &CpuMpData->CpuData[ProcessorNumber];
> >      CpuData->Waiting = FALSE;
> > @@ -2131,7 +2131,7 @@ StartupAllAPsWorker (
> >          // Mark this processor as responsible for current calling.
> >          //
> >          CpuData->Waiting = TRUE;
> > -        CpuMpData->StartCount++;
> > +        CpuMpData->RunningCount++;
> >        }
> >      }
> >    }
> > @@ -2140,7 +2140,6 @@ StartupAllAPsWorker (
> >    CpuMpData->ProcArguments = ProcedureArgument;
> >    CpuMpData->SingleThread  = SingleThread;
> >    CpuMpData->FinishedCount = 0;
> > -  CpuMpData->RunningCount  = 0;
> >    CpuMpData->FailedCpuList = FailedCpuList;
> >    CpuMpData->ExpectedTime  = CalculateTimeout (
> >                                 TimeoutInMicroseconds, diff --git
> > a/UefiCpuPkg/Library/MpInitLib/MpLib.h
> > b/UefiCpuPkg/Library/MpInitLib/MpLib.h
> > index 90c09fb8fb..4166734687 100644
> > --- a/UefiCpuPkg/Library/MpInitLib/MpLib.h
> > +++ b/UefiCpuPkg/Library/MpInitLib/MpLib.h
> > @@ -210,7 +210,6 @@ struct _CPU_MP_DATA {
> >    UINTN                          BackupBuffer;
> >    UINTN                          BackupBufferSize;
> >
> > -  volatile UINT32                StartCount;
> >    volatile UINT32                FinishedCount;
> >    volatile UINT32                RunningCount;
> >    BOOLEAN                        SingleThread;
> >
> 
> I'm currently missing a good understanding of how these counters are
> modified. They are all qualified "volatile", which suggests they are accessed
> from multiple processors. Is that correct?

No, actually only FinishedCount is changed by BSP and Aps, other two are only changed by BSP. StartCount stands for the AP count which will do the task. It is calculated by BSP before Aps start the task. RunningCount stands for the AP count which have finished the task. It also detected and changed by BSP.

> 
> Here's my concern: assume we have 3 APs, and we maintain how many of
> them are running concurrently at any given point. Assume we start them up
> all at once, and then later they all finish. The modified "RunningCount" value
> might advance like this:
> 
>   AP#1  AP#2  AP#3  RunningCount  comment
>   ----  ----  ----  ------------  ----------------
>                     0             no AP running yet
>    v                1             AP#1 starts
>    |     v          2             AP#2 starts
>    |     |     v    3             AP#3 starts
>    |     |     |    3             all APs working
>    |     ^     |    2             AP#2 finishes
>    |           ^    1             AP#3 finishes
>    ^                0             AP#1 finishes, done
> 
> However, the following could happen as well:
> 
>   AP#1  AP#2  AP#3  RunningCount  comment
>   ----  ----  ----  ------------  ----------------
>                     0             no AP running yet
>    v                1             AP#1 starts
>    ^                0             AP#1 finishes
>          v          1             AP#2 starts
>          |     v    2             AP#3 starts
>          |     ^    1             AP#3 finishes
>          ^          0             AP#2 finishes, done
> 
> In the second scheduling, we get RunningCount=0 when AP#1 finishes, even
> though AP#2 and AP#3 are not done (they haven't even started yet).
> 
> Is this a realistic concern, or is the above scenario impossible?

For this patch, RunningCount is calculated by BSP before it start the task for Aps, so how the Aps start the task will not impact the logic.  The update of RunningCount also done by BSP.

I will update the patch to remove the volatile for RunningCount.

> 
> (I'd like to test the series after understanding this.)
> 
> Actually... if the problem scenario is possible, I think it could affect the current
> (pre-patch) code as well. I hope I'm wrong!
> 
> Thanks,
> Laszlo