On Sat, Dec 27, 2025 at 11:56 PM Krzysztof Kozlowski <[email protected]> wrote:
>
> On 27/12/2025 23:01, Rob Clark wrote:
> > On Sat, Dec 27, 2025 at 3:05 AM Krzysztof Kozlowski
> > <[email protected]> wrote:
> >>
> >> DTS files for qcom,adreno-610.0 and qcom,adreno-07000200 contain only one
> >> "reg" entry, not two, and the binding defines the second entry in
> >> "reg-names" differently than top-level part, so just simplify it and
> >> narrow to only one entry.
> >
> > I'll defer to Akhil about whether this is actually needed (vs just
> > incomplete gpu devcoredump support for certain GPUs).  In general
> > cx_dbgc is needed to capture state for gpu devcoredump state
> > snapshots, but not directly used in normal operations.  It seems
> > similar to the situation with mapping gpucc as part of gmu, ie. not
> > something the CPU normally deals with directly, but necessary to
> > capture crash state.
>
> I don't get why binding was added with cx_dbgc, but DTS not. Neither
> binding nor DTS depends on actual usage, so I assume someone
> intentionally did not want DTS to contain cx_dbgc and binding should
> follow. Otherwise we should make the DTS complete and make the binding
> strict (leading to warnings if DTS is not updated).

I'm not sure about the history.. but I can say that cx_dbgc is only
used for gpu state snapshot / devcoredump.  So it would be easy to not
notice if it were missing.

We have a similar slightly ugly thing where gpucc is included in the
gmu map.. only for devcoredump.  Maybe we need a different way to
handle these things that are only mapped for state capture?

BR,
-R

Reply via email to