Re: [edk2-devel] [PATCH v2 1/2] UefiCpuPkg/PiSmmCpuDxeSmm: Remove dependence between APs

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

Hi Ray,

> -----Original Message-----
> From: Ni, Ray
> Sent: Monday, December 23, 2019 3:38 PM
> To: devel@edk2.groups.io; Dong, Eric <eric.dong@intel.com>
> Cc: Laszlo Ersek <lersek@redhat.com>
> Subject: RE: [edk2-devel] [PATCH v2 1/2] UefiCpuPkg/PiSmmCpuDxeSmm:
> Remove dependence between APs
> 
> >
> > +  WaitForSemaphore (&Token->RunningApCount);
> > +
> > +  if (Token->RunningApCount == 0) {
> > +    ReleaseSpinLock (Token->SpinLock);
> >    }
> 
> 1. if (InterlockedDecrement (&Token->RunningApCount) == 0) {
>       ReleaseSpinLock (Token->SpinLock);
>     }
> 
> We should avoid checking RunningApCount directly because it's possible
> that AP#1 decrease the Count to 1 and before AP#1 checks the value against
> 0
> the Count is decreased by AP#2 to 0. So that causes AP#1 and AP#2 call
> ReleaseSpinLock() on the same SpinLock.
> 
[[Eric]] good comments, will update it in next version.

> >
> > +      // Decrease the count to mark this AP as finished.
> 
> 2. BSP is also handled here. So this comment is mis-leading.
[[Eric]] will enhance the comments in next version.

> 
> >
> > +      //
> >
> > +      if (Token != NULL) {
> > +        WaitForSemaphore (&ProcToken->RunningApCount);
> 
> 3. The code is written correctly but improperly IMO.
> Token is checked but ProcToken is deferenced.
> I suggest you check ProcToken directly.
[[Eric]] The other place in this function all check the Token status then update code. 
So this code consistent with other place.  I will keep this code to keep consistent.

Thanks,
Eric