Re: [PATCH v2] MdeModulePkg/NonDiscoverablePciDeviceDxe: Allow partial FreeBuffer

["Jeff Brasen" <jbrasen@nvidia.com>]

> -----Original Message-----
> From: Ard Biesheuvel <ardb@kernel.org>
> Sent: Tuesday, August 2, 2022 10:51 AM
> To: Jeff Brasen <jbrasen@nvidia.com>
> Cc: devel@edk2.groups.io; hao.a.wu@intel.com; ray.ni@intel.com;
> quic_llindhol@quicinc.com; ardb+tianocore@kernel.org
> Subject: Re: [PATCH v2] MdeModulePkg/NonDiscoverablePciDeviceDxe:
> Allow partial FreeBuffer
> 
> External email: Use caution opening links or attachments
> 
> 
> On Tue, 2 Aug 2022 at 17:32, Jeff Brasen <jbrasen@nvidia.com> wrote:
> >
> >
> >
> > > -----Original Message-----
> > > From: Ard Biesheuvel <ardb@kernel.org>
> > > Sent: Friday, July 29, 2022 9:48 AM
> > > To: Jeff Brasen <jbrasen@nvidia.com>
> > > Cc: devel@edk2.groups.io; hao.a.wu@intel.com; ray.ni@intel.com;
> > > quic_llindhol@quicinc.com; ardb+tianocore@kernel.org
> > > Subject: Re: [PATCH v2] MdeModulePkg/NonDiscoverablePciDeviceDxe:
> > > Allow partial FreeBuffer
> > >
> > > External email: Use caution opening links or attachments
> > >
> > >
> > > On Thu, 28 Jul 2022 at 13:25, Jeff Brasen <jbrasen@nvidia.com> wrote:
> > > >
> > > >
> > > > Adding Leif/Ard to CC incase they have any comments on this patch.
> > > >
> > >
> > > This generally looks ok to me. I just wonder if it wouldn't be
> > > simpler to reuse the existing allocation descriptor if it is not
> > > being freed entirely. Given the [presumably] the most common case is
> > > to allocate and then free some pages at the end, lowering the page
> > > count on the existing descriptor would cover most cases, and we'd
> > > only need to allocate new ones if pages are being freed at the start or in
> the middle.
> >
> > There is often freeing at the beginning as well as this is being used to create
> a 64K aligned section of memory in the case. So it over allocates and the
> free's some at the beginning and the end. I could probably make it detect
> and use that but figured this code would support all cases and required less
> case specific detection.
> >
> 
> Ah interesting. Would it help if the allocate routine aligned allocations to their
> size?

The PciIo->AllocateBuffer function doesn't support passing the request in so we would need to know that info beforehand. The current calling in the XHCI driver does a free at the beginning and then the end of the buffer so we could the existing allocation tracker but figured it would be better to correct the function just in case someone called it to free in the middle.