[PATCH 12/12] drm/msm/adreno: Switch to chip-id for identifying GPU

[Rob Clark]

On Sat, Jul 15, 2023 at 6:38 AM Konrad Dybcio <konrad.dybcio at linaro.org> wrote:
>
> On 7.07.2023 18:08, Rob Clark wrote:
> > On Thu, Jul 6, 2023 at 5:25 PM Konrad Dybcio <konrad.dybcio at linaro.org> wrote:
> >>
> >> Apart from all these comments, I don't really see the point of this patch,
> >> other than trying to tie together Qualcomm's almost-meaningless chipids on
> >> a7xx into the picture..
> >>
> >> Since they can't even be read back from the hardware, I don't think trying
> >> to force them into the upstream kernel makes any sense.
> >
> > Sure, we _could_ pick our own arbitrary identifiers, we don't have to
> > align with kgsl.  But that would be a super huge PITA for mesa, which
> > has support for both kernels.
> Perhaps I'm biased towards keeping this kind of stuff out of the kernel,
> but I'd say that Linux should decide on one logical path.
>
> In between us starting this discussion, Qualcomm managed to introduce
> yet another way of deciding what GPU is present on the system in their
> downstream driver[1]. We're at like 5 now. Do we wanna keep up each time?
> New ID rework for A8xx when that comes out one day?

[snip]

> [1] they now read parts of socinfo from smem and decide the CHIPID and
>     speedbin based on that, but it's not available on most existing SoCs,
>     that was thrown in with SOCID v17

This is actually exactly what we want.. something that we can read
from hw/fw and that blob userspace also uses (so we don't have to
worry about qc forgetting to change the id, etc)

BR,
-R