On Fri, Aug 11, 2023 at 9:31 AM Konrad Dybcio <konrad.dybcio@xxxxxxxxxx> wrote:
>
> On 11.08.2023 18:21, Rob Clark wrote:
> > On Fri, Aug 11, 2023 at 9:11 AM Konrad Dybcio <konrad.dybcio@xxxxxxxxxx> wrote:
> >>
> >> On 11.08.2023 18:09, Rob Clark wrote:
> >>> On Fri, Aug 11, 2023 at 9:05 AM Rob Clark <robdclark@xxxxxxxxx> wrote:
> >>>>
> >>>> From: Rob Clark <robdclark@xxxxxxxxxxxx>
> >>>>
> >>>> There isn't actually a a690_gmu.bin. But it appears that the normal
> >>>> a660_gmu.bin works fine. Normally all the devices within a sub-
> >>>> generation (or "family") will use the same fw, and a690 is in the a660
> >>>> family.
> >>>>
> >>>
> >>> possibly this could be considered as:
> >>>
> >>> Fixes: 5e7665b5e484 ("drm/msm/adreno: Add Adreno A690 support")
> >> For a lack of a better response, "meh"
> >
> > It would help route the change back to 6.4.y so we can stop explaining
> > to folks that they should create a symlink ;-)
> Ok fair!
>
> >
> >> Other than that:
> >>
> >> Reviewed-by: Konrad Dybcio <konrad.dybcio@xxxxxxxxxx>
> >>
> >> Perhaps we could take this further and do something like
> >>
> >> if (failed to load gmu fw)
> >> try loading "gmu.bin"
> >
> > that (loading random fw) sounds like a bad idea.. plus gmu.bin doesn't exist
> I guess we still should hear from Qualcomm on this one, but I
> do believe that "aXYZ_gmu.bin" is really "gmu firmware, tagged
> at aXYZ release". They're even versioned and adreno-gpulist.h
> downstream contains the minimum version for each SKU.
We definitely can't use a660_gmu.bin as is on a630 family devices, for
example. Maybe that is just down to sw <-> fw interface changes but
it means linux-firmware would still need to keep separate a630_gmu.bin
and a660_gmu.bin, otherwise new l-f with old kernel would explode.
That doesn't mean they are unrelated.. I think they probably just
branch the fw for each family.
BR,
-R
> Konrad
