Re: [PATCH 1/2] EmbeddedPkg/NonCoherentDmaLib: implement support for DMA range limits

["Leif Lindholm" <leif.lindholm@linaro.org>]

On Mon, Nov 25, 2019 at 14:13:19 +0100, Ard Biesheuvel wrote:
> On Mon, 25 Nov 2019 at 13:58, Leif Lindholm <leif.lindholm@linaro.org> wrote:
> >
> > On Mon, Nov 25, 2019 at 13:52:24 +0100, Ard Biesheuvel wrote:
> > > > > -  if (MemoryType == EfiBootServicesData) {
> > > > > -    Allocation = AllocateAlignedPages (Pages, Alignment);
> > > > > -  } else if (MemoryType == EfiRuntimeServicesData) {
> > > > > -    Allocation = AllocateAlignedRuntimePages (Pages, Alignment);
> > > > > +  if (MemoryType == EfiBootServicesData ||
> > > > > +      MemoryType == EfiRuntimeServicesData) {
> > > > > +    Allocation = InternalAllocateAlignedPages (MemoryType, Pages, Alignment);
> > > > >    } else {
> > > > >      return EFI_INVALID_PARAMETER;
> > > > >    }
> > > > > @@ -479,6 +596,15 @@ NonCoherentDmaLibConstructor (
> > > > >  {
> > > > >    InitializeListHead (&UncachedAllocationList);
> > > > >
> > > > > +  //
> > > > > +  // Ensure that the combination of DMA addressing offset and limit produces
> > > > > +  // a sane value.
> > > > > +  //
> > > > > +  ASSERT (PcdGet64 (PcdDmaDeviceLimit) > PcdGet64 (PcdDmaDeviceOffset));
> > > >
> > > > Is this worth turning into a hard conditional and error return rather
> > > > than an assert? The following statement will end up wrapping downwards
> > > > if this condition is not true.
> > >
> > >
> > > Constructor return values are only used in ASSERT_EFI_ERROR() calls in
> > > the ProcessLibraryConstructorList() routine that gets generated by the
> > > build tools (in AutoGen.c), so returning an error here would
> > > essentially be dead code, given that we would only make it to this
> > > point on builds that are known to ignore it.
> >
> > Hmm, ack.
> > Print error message?
> 
> How? Using DEBUG()? That would be dead code as well, since it is only
> active in DEBUG builds, which means the ASSERT() would fire first and
> prevent the DEBUG() from ever being reached.

Clearly not.

> So that leaves printing to the console using boot services, which is
> not something we should be doing from a DXE_DRIVER type module.

Yeah, maybe that's too urgh.

> Given that these are both platform specific values defined at platform
> creation time, I think policing them only in a DEBUG build is quite
> appropriate, tbh.

I dunno, it just feels to me like it's following the pattern of the
ArmPlatformGetVirtualMemoryMap() implementations (which do the
same). This is a hard error detectable at compile time.

My preferred way to handle it would really be a compilation error, but
that brings us back to last week's discussion on runtime
vs. preprocessor...

/
    Leif