Re: [Patch 2/2] UefiCpuPkg/MpInitLib: Enhance waiting for AP initialization logic.

[Laszlo Ersek <lersek@redhat.com>]

CC Paolo

On 10/23/17 09:22, Eric Dong wrote:
> Current logic always waiting for a specific value to collect all APs
> count. This logic may caused some platforms cost too much time to
> wait for time out.
> This patch add new logic to collect APs count. It adds new variable
> NumApsExecuting to detect whether all APs have finished initialization.
> Each AP let NumApsExecuting++ when begin to initialize itself and let
> NumApsExecuting-- when it finish the initialization. BSP base on whether
> NumApsExecuting == 0  to finished the collect AP process.
> 
> Cc: Ruiyu Ni <ruiyu.ni@intel.com>
> Contributed-under: TianoCore Contribution Agreement 1.1
> Signed-off-by: Eric Dong <eric.dong@intel.com>
> ---
>  UefiCpuPkg/Library/MpInitLib/Ia32/MpEqu.inc    |  1 +
>  UefiCpuPkg/Library/MpInitLib/Ia32/MpFuncs.nasm |  6 ++++++
>  UefiCpuPkg/Library/MpInitLib/MpLib.c           | 20 ++++++++++++++------
>  UefiCpuPkg/Library/MpInitLib/MpLib.h           |  1 +
>  UefiCpuPkg/Library/MpInitLib/X64/MpEqu.inc     |  3 ++-
>  UefiCpuPkg/Library/MpInitLib/X64/MpFuncs.nasm  |  6 ++++++
>  6 files changed, 30 insertions(+), 7 deletions(-)

I was out of office yesterday, and did not get a chance to comment on
this patch.

In a virtualization guest, I see the following problem with the patch:

VCPUs (virtual CPUs) may not be scheduled by the hypervisor similarly to
how physical CPUs behave on a physical board. It is possible that a VCPU
starts up and even finishes its initialization routine before another
VCPU starts running at all.

Therefore the locked NumApsExecuting counter may hit zero, even multiple
times, before all APs have finished initializing.

In OVMF, we query QEMU about the exact number of virtual processors, in
PlatformPei. So OVMF configures the logical processor count in advance
that MpInitLib has to wait for. Correspondingly, we also set the timeout
to "infinity".

Please see the MaxCpuCountInitialization() function in following commit:

  https://github.com/tianocore/edk2/commit/45a70db3c3a59

In the past, we used to have AP initialization problems in OVMF due to
the VCPU scheduling artifacts I mention above. After commit
45a70db3c3a59, things have been stable; it would be nice to keep that
working.

Please note that simply testing this patch on my end is not sufficient.
The AP init problems we used to face were sporadic and also specific to
the virtualization host systems (i.e., dependent on the physical
hardware and the host kernel).

Furthermore:

> diff --git a/UefiCpuPkg/Library/MpInitLib/Ia32/MpEqu.inc b/UefiCpuPkg/Library/MpInitLib/Ia32/MpEqu.inc
> index 976af1f..bdfe0d3 100644
> --- a/UefiCpuPkg/Library/MpInitLib/Ia32/MpEqu.inc
> +++ b/UefiCpuPkg/Library/MpInitLib/Ia32/MpEqu.inc
> @@ -40,4 +40,5 @@ EnableExecuteDisableLocation  equ        LockLocation + 30h
>  Cr3Location                   equ        LockLocation + 34h
>  InitFlagLocation              equ        LockLocation + 38h
>  CpuInfoLocation               equ        LockLocation + 3Ch
> +NumApsExecutingLocation       equ        LockLocation + 40h
>  
> diff --git a/UefiCpuPkg/Library/MpInitLib/Ia32/MpFuncs.nasm b/UefiCpuPkg/Library/MpInitLib/Ia32/MpFuncs.nasm
> index 1b9c6a6..2b6c27d 100644
> --- a/UefiCpuPkg/Library/MpInitLib/Ia32/MpFuncs.nasm
> +++ b/UefiCpuPkg/Library/MpInitLib/Ia32/MpFuncs.nasm
> @@ -86,6 +86,12 @@ Flat32Start:                                   ; protected mode entry point
>  
>      mov        esi, ebx
>  
> +    ; Increment the number of APs executing here as early as possible
> +    ; This is decremented in C code when AP is finished executing
> +    mov        edi, esi
> +    add        edi, NumApsExecutingLocation
> +    lock inc   dword [edi]
> +
>      mov         edi, esi
>      add         edi, EnableExecuteDisableLocation
>      cmp         byte [edi], 0
> diff --git a/UefiCpuPkg/Library/MpInitLib/MpLib.c b/UefiCpuPkg/Library/MpInitLib/MpLib.c
> index db923c9..48f930b 100644
> --- a/UefiCpuPkg/Library/MpInitLib/MpLib.c
> +++ b/UefiCpuPkg/Library/MpInitLib/MpLib.c
> @@ -662,6 +662,7 @@ ApWakeupFunction (
>      // AP finished executing C code
>      //
>      InterlockedIncrement ((UINT32 *) &CpuMpData->FinishedCount);
> +    InterlockedDecrement ((UINT32 *) &CpuMpData->MpCpuExchangeInfo->NumApsExecuting);
>  
>      //
>      // Place AP is specified loop mode
> @@ -765,6 +766,7 @@ FillExchangeInfoData (
>  
>    ExchangeInfo->CFunction       = (UINTN) ApWakeupFunction;
>    ExchangeInfo->ApIndex         = 0;
> +  ExchangeInfo->NumApsExecuting = 0;
>    ExchangeInfo->InitFlag        = (UINTN) CpuMpData->InitFlag;
>    ExchangeInfo->CpuInfo         = (CPU_INFO_IN_HOB *) (UINTN) CpuMpData->CpuInfoInHob;
>    ExchangeInfo->CpuMpData       = CpuMpData;
> @@ -934,13 +936,19 @@ WakeUpAP (
>      }
>      if (CpuMpData->InitFlag == ApInitConfig) {
>        //
> -      // Wait for all potential APs waken up in one specified period
> +      // Wait for one potential AP waken up in one specified period
>        //
> -      TimedWaitForApFinish (
> -        CpuMpData,
> -        PcdGet32 (PcdCpuMaxLogicalProcessorNumber) - 1,
> -        PcdGet32 (PcdCpuApInitTimeOutInMicroSeconds)
> -        );
> +      if (CpuMpData->CpuCount == 0) {
> +        TimedWaitForApFinish (
> +          CpuMpData,
> +          PcdGet32 (PcdCpuMaxLogicalProcessorNumber) - 1,
> +          PcdGet32 (PcdCpuApInitTimeOutInMicroSeconds)
> +          );
> +      }

I don't understand this change. The new comment says,

  Wait for *one* potential AP waken up in one specified period

However, the second parameter of TimedWaitForApFinish(), namely
"FinishedApLimit", gets the same value as before:

  PcdGet32 (PcdCpuMaxLogicalProcessorNumber) - 1

It means that all of the (possible) APs are waited-for, just the same as
before.

In other words, I think the patch does not correctly implement what the
commit message says -- and for QEMU / OVMF, that's actually a good thing
at the moment, because a correct implementation of the description would
likely break on QEMU.

Thanks
Laszlo

> +
> +      while (CpuMpData->MpCpuExchangeInfo->NumApsExecuting != 0) {
> +        CpuPause();
> +      }
>      } else {
>        //
>        // Wait all APs waken up if this is not the 1st broadcast of SIPI
> diff --git a/UefiCpuPkg/Library/MpInitLib/MpLib.h b/UefiCpuPkg/Library/MpInitLib/MpLib.h
> index e41d2db..d13d5c0 100644
> --- a/UefiCpuPkg/Library/MpInitLib/MpLib.h
> +++ b/UefiCpuPkg/Library/MpInitLib/MpLib.h
> @@ -176,6 +176,7 @@ typedef struct {
>    UINTN                 Cr3;
>    UINTN                 InitFlag;
>    CPU_INFO_IN_HOB       *CpuInfo;
> +  UINTN                 NumApsExecuting;
>    CPU_MP_DATA           *CpuMpData;
>    UINTN                 InitializeFloatingPointUnitsAddress;
>  } MP_CPU_EXCHANGE_INFO;
> diff --git a/UefiCpuPkg/Library/MpInitLib/X64/MpEqu.inc b/UefiCpuPkg/Library/MpInitLib/X64/MpEqu.inc
> index 114f4e0..d255ca5 100644
> --- a/UefiCpuPkg/Library/MpInitLib/X64/MpEqu.inc
> +++ b/UefiCpuPkg/Library/MpInitLib/X64/MpEqu.inc
> @@ -40,5 +40,6 @@ EnableExecuteDisableLocation  equ        LockLocation + 5Ch
>  Cr3Location                   equ        LockLocation + 64h
>  InitFlagLocation              equ        LockLocation + 6Ch
>  CpuInfoLocation               equ        LockLocation + 74h
> -InitializeFloatingPointUnitsAddress equ  LockLocation + 84h
> +NumApsExecutingLocation       equ        LockLocation + 7Ch
> +InitializeFloatingPointUnitsAddress equ  LockLocation + 8Ch
>  
> diff --git a/UefiCpuPkg/Library/MpInitLib/X64/MpFuncs.nasm b/UefiCpuPkg/Library/MpInitLib/X64/MpFuncs.nasm
> index 4ada649..21d2786 100644
> --- a/UefiCpuPkg/Library/MpInitLib/X64/MpFuncs.nasm
> +++ b/UefiCpuPkg/Library/MpInitLib/X64/MpFuncs.nasm
> @@ -124,6 +124,12 @@ LongModeStart:
>      cmp        qword [edi], 1       ; ApInitConfig
>      jnz        GetApicId
>  
> +    ; Increment the number of APs executing here as early as possible
> +    ; This is decremented in C code when AP is finished executing
> +    mov        edi, esi
> +    add        edi, NumApsExecutingLocation
> +    lock inc   dword [edi]
> +
>      ; AP init
>      mov        edi, esi
>      add        edi, LockLocation
>