Re: [PATCH] UefiCpuPkg/MpInitLib: AP uses memory preceding IDT to store CpuMpData

[Laszlo Ersek <lersek@redhat.com>]

Hi Ray,

On 06/26/18 09:50, Ni, Ruiyu wrote:
> On 6/26/2018 1:01 AM, Laszlo Ersek wrote:
>> On 06/25/18 18:01, Laszlo Ersek wrote:
>>> Hello Ray,
>>>
>>> On 06/25/18 04:54, Ruiyu Ni wrote:
>>>> Today's MpInitLib PEI implementation directly calls
>>>> PeiServices->GetHobList() from AP which may cause racing issue.
>>>>
>>>> This patch fixes this issue by storing the CpuMpData to memory
>>>> preceding IDT. Pointer to PeiServices pointer is stored there,
>>>> so after AP procedure returns, the PeiServices pointer should be
>>>> restored.
>>>>
>>>> Contributed-under: TianoCore Contribution Agreement 1.1
>>>> Signed-off-by: Ruiyu Ni <ruiyu.ni@intel.com>
>>>> Cc: Jeff Fan <vanjeff_919@hotmail.com>
>>>> Cc: Eric Dong <eric.dong@intel.com>
>>>> Cc: Jiewen Yao <jiewen.yao@intel.com>
>>>> Cc: Fish Andrew <afish@apple.com>
>>>> ---
>>>>   UefiCpuPkg/Library/MpInitLib/DxeMpLib.c | 33 ++++++++++++++++++-
>>>>   UefiCpuPkg/Library/MpInitLib/MpLib.c    |  8 +++++
>>>>   UefiCpuPkg/Library/MpInitLib/MpLib.h    | 27 +++++++++++++++-
>>>>   UefiCpuPkg/Library/MpInitLib/PeiMpLib.c | 56
>>>> +++++++++++++++++++++++++++++++--
>>>>   4 files changed, 119 insertions(+), 5 deletions(-)
>>>>
>>>> diff --git a/UefiCpuPkg/Library/MpInitLib/DxeMpLib.c
>>>> b/UefiCpuPkg/Library/MpInitLib/DxeMpLib.c
>>>> index e7ed21c6cd..26fead2c66 100644
>>>> --- a/UefiCpuPkg/Library/MpInitLib/DxeMpLib.c
>>>> +++ b/UefiCpuPkg/Library/MpInitLib/DxeMpLib.c
>>>
>>> [snip]
>>>
>>>> +/**
>>>> +  Pop the CpuMpData.
>>>> +
>>>> +  @param[in] Pointer  The pointer value which was stored in where
>>>> the CPU MP Data is pushed.
>>>> +  @param[in] Context  The context of push/pop operation.
>>>> +**/
>>>> +VOID
>>>> +PopCpuMpData (
>>>> +  IN VOID           *Pointer,
>>>> +  IN VOID           *Context
>>>> +  )
>>>> +{
>>>
>>> (1) When I applied this patch for regression-testing, "git am"
>>> complained that the above line added two trailing space characters. Can
>>> you please remove them?
>>>
>>> (2) Other than that, I tested this patch with OVMF -- the patch breaks
>>> OVMF. The last part of the log is:
>>>
>>>> Loading PEIM [CpuMpPei]
>>>> Loading PEIM at 0x000BFE81000 EntryPoint=0x000BFE86DC6 CpuMpPei.efi
>>>> AP Loop Mode is 1
>>>> WakeupBufferStart = 9F000, WakeupBufferSize = 1000
>>>> TimedWaitForApFinish: reached FinishedApLimit=7 in 121 microseconds
>>>> APIC MODE is 1
>>>> MpInitLib: Find 8 processors in system.
>>>> Does not find any stored CPU BIST information from PPI!
>>>>    APICID - 0x00000000, BIST - 0x00000000
>>>>    APICID - 0x00000001, BIST - 0x00000000
>>>>    APICID - 0x00000002, BIST - 0x00000000
>>>>    APICID - 0x00000003, BIST - 0x00000000
>>>>    APICID - 0x00000004, BIST - 0x00000000
>>>>    APICID - 0x00000005, BIST - 0x00000000
>>>>    APICID - 0x00000006, BIST - 0x00000000
>>>>    APICID - 0x00000007, BIST - 0x00000000
>>>> Install PPI: [EfiSecPlatformInformation2Ppi]
>>>> Install PPI: [EfiPeiMpServicesPpi]
>>>> Notify: PPI Guid: [EfiPeiMpServicesPpi], Peim notify entry point:
>>>> 8524F8
>>>> PlatformPei: OnMpServicesAvailable
>>
>> Sorry, I failed to mention: after the last line is printed, the boot
>> process is stuck, and all CPUs appear spinning.
> 
> I tried to directly boot OVMF in my Win10. The OnMpServicesAvailable()
> is not called because the PPI notification is not installed.
> So I tried to modify InstallFeatureControlCallback() in
> PlatformPei/FeatureControl.c to unconditionally call
> "PeiServicesNotifyPpi (&mMpServicesNotify);".
> The new image can boot to Shell and I can find "OnMpServicesAvailable"
> debug message in the log.

When you run QEMU on Windows, that means that QEMU uses "TCG" (Tiny Code
Generator), which is a Just In Time compiler that runs the guest code
with software emulation. That's not the production use of QEMU.

For using QEMU with hardware virtualization (Intel vmx / AMD svm),
please run QEMU on Linux, with KVM. This is documented in the Wiki
article I referenced earlier -- the article specifically targets
developers with a Windows background. You only need one Linux server
(not desktop) for testing; for the interactive parts, you can use your
usual Windows workstation.

Thanks
Laszlo