[PATCH v2] arm64: dts: qcom: sc7180: Add A618 gpu dt blob

Doug Anderson dianders at chromium.org
Fri Jan 31 21:21:07 UTC 2020


Hi,

On Fri, Jan 31, 2020 at 1:18 PM Jordan Crouse <jcrouse at codeaurora.org> wrote:
>
> On Fri, Jan 31, 2020 at 08:08:09AM -0800, Doug Anderson wrote:
> > Hi,
> >
> > On Fri, Jan 31, 2020 at 4:16 AM <smasetty at codeaurora.org> wrote:
> > >
> > > >> +                       reg = <0 0x0506a000 0 0x31000>, <0 0x0b290000
> > > >> 0 0x10000>,
> > > >> +                               <0 0x0b490000 0 0x10000>;
> > > >> +                       reg-names = "gmu", "gmu_pdc", "gmu_pdc_seq";
> > > >> +                       interrupts = <GIC_SPI 304
> > > >> IRQ_TYPE_LEVEL_HIGH>,
> > > >> +                                  <GIC_SPI 305 IRQ_TYPE_LEVEL_HIGH>;
> > > >> +                       interrupt-names = "hfi", "gmu";
> > > >> +                       clocks = <&gpucc GPU_CC_CX_GMU_CLK>,
> > > >> +                              <&gpucc GPU_CC_CXO_CLK>,
> > > >> +                              <&gcc GCC_DDRSS_GPU_AXI_CLK>,
> > > >> +                              <&gcc GCC_GPU_MEMNOC_GFX_CLK>;
> > > >> +                       clock-names = "gmu", "cxo", "axi", "memnoc";
> > > >> +                       power-domains = <&gpucc CX_GDSC>;
> > > >
> > > > Bindings claim that you need both CX and GC.  Is sc7180 somehow
> > > > different?  Bindings also claim that you should be providing
> > > > power-domain-names.
> > > No this is still needed, We need the GX power domain for GPU recovery
> > > use cases where the shutdown was not successful.
> >
> > This almost sounds as if the bindings should mark the GX power domain
> > as optional?  The driver can function without it but doesn't get all
> > the features?  As the binding is written right now I think it is
> > "invalid" to not specify a a GX power domain and once the yaml
> > conversion is done then it will even be flagged as an error.  That's
> > going to make it harder to land the your patch...
>
> For GMU attached targets the GX power domain is mandatory assuming you want to
> recover successfully from a hard GMU hang, that is.

Sure.  I guess we can quibble about whether this means optional or
mandatory, but it won't gain much.  ;-)

...seems like for now (assuming it works) we should at least specify
it and put a <0>.  Then we should make it a relatively high priority
to get it hooked up more properly.

-Doug


More information about the dri-devel mailing list