On 5/28/25 6:50 PM, Rob Clark wrote: > On Tue, May 27, 2025 at 11:18 AM Dmitry Baryshkov > <dmitry.barysh...@oss.qualcomm.com> wrote: >> >> On Tue, May 27, 2025 at 12:55:06PM +0200, Konrad Dybcio wrote: >>> On 5/26/25 5:28 PM, Rob Clark wrote: >>>> On Mon, May 26, 2025 at 1:36 AM Dmitry Baryshkov >>>> <dmitry.barysh...@oss.qualcomm.com> wrote: >>>>> >>>>> On Sun, May 25, 2025 at 09:43:36PM +0200, Aleksandrs Vinarskis wrote: >>>>>> On Sun, 25 May 2025 at 15:33, Dmitry Baryshkov >>>>>> <dmitry.barysh...@oss.qualcomm.com> wrote: >>>>>>> >>> >>> [...] >>> >>>>> Yes, please. Don't set okay status for the GPU until it gets enabled. >>>> >>>> Drive-by: Shouldn't the dtb describe the hw and not the state of the >>>> linux kernel's support for the hw? Ie. if bad things happen if we >>>> describe hw which is missing driver support, shouldn't we fix that in >>>> the driver. >>>> >>>> (In the case of the GPU there is the slight wrinkle that we don't have >>>> a gpu-id yet so there is no compatible in the dtb yet.) >>> >>> My two cents are that it's okay to enable it, at least in this case.. >> >> That would most likely make display unavailable as DPU driver uses GPU >> as one of the components. > > Hmm, perhaps we should allow the gpu to probe, but just fail to create > priv->gpu, similarly to how we handle missing firmware?
Yes, let's stop killing the entire combo-stack in all scenarios.. Konrad
