Re: [PATCH v2 01/12] dt-bindings: media: qcom,venus: Add context bank subnodes to common schema

[Vikash Garodia]

On 8/6/2026 4:44 PM, Dmitry Baryshkov wrote:
> On Thu, Aug 06, 2026 at 11:50:01AM +0100, Bryan O'Donoghue wrote:
> > On 06/08/2026 11:32, Krzysztof Kozlowski wrote:
> > > > Yes -master.
> > > >
> > > > My issue is how long this will take. Several kernel cycles and the
> > > > debate is ongoing.
> > > >
> > > > Daneil's fix has not been applied which means the replicable system bug
> > > > is out there in the wild, not addressed.
> > > >
> > > > Daniel's fix could have been applied and rolled out to all users in
> > > > mainline and consequently stable.
> > > >
> > > > Instead another architectural level fix is still under discussion, which
> > > > is of no value to end users whose systems can reset in the meantime.
> > > >
> > > > Its just not responsible behaviour. So I feel both of these drivers need
> > > > to be marked as BROKEN unless/until a fix lands.
> > > >
> > > > I'm hoping the vidc team will come up with something else.
> > > I did not object taking a known fix for the problem for the mainline
> > > kernel, regardless if the fix is optimal or not. I only objected the
> > > reason of taking patches because of something in stable kernels or
> > > making backports easier or whatever other semi-expressed reason, which
> > > is not the mainline-kernel reason.
> > >
> > > Best regards,
> > > Krzysztof
> >
> > I'm all in favour of rolling out Daniel's fix.
>
> The biggest problem with it is that it looks like a hotfix or a hack
> rather than a proper description / solution. From this PoV we will land
> the fixes for all the boards in mainline and then fix the same issue
> _again_.

I would call it as un-optimal than a hack, and certainly i can say it 
would *not* have to be patched again. Its simple enough that it applies 
one reserve region to all, instead of applying it selectively to 
specific stream.

> > I don't want to propose BROKEN but at the same time the timeline on this
> > demands action:
>
> I think this might be a solution, which everybody dislikes, but in the
> end it is true (for both drivers).
>
> > - Sub-nodes proposed around December ?
> >    Which means the system resetting bug was known
> >    and understood but not promoted as a bug requiring
> >    resolution as it should have been.
> >
> >    Fine people make mistakes, that's life, lessons learned.
> >
> > - Daniel's fix posted.
> >    Not endorsed or rolled out. Its not "perfect" but
> >    it is certainly "good enough".
> >
> > - Now more debate on sub-nodes
> >    I'm all in favour of that but NOT at the expense
> >    of replicable bugs from real users.
> >
> > TBH, when I start to lay out the timeline on this, I get annoyed. It seems
> > to me qcom has understood this bug for nearly six months more ? and mainline
> > is still waiting on a resolution.
> >
> > So I'll leave it up to the vidc people to decide. But some kind of deadline
> > needs to be put on this.

let me post Daniel's patch as fixes, and then the sub node patches can 
continue separately and drop the fixes tag from sub node patches. I 
would be patching both venus and iris.

> > Close of business GMT Friday - either something mergable lands for sub-nodes
> > or we replicate Daniel's fix to all platforms. Failing both, I will have no
> > choice but to set both drivers to BROKEN.
> >
> > Stability has to come first.
> >
> > ---
> > bod

Regards,
Vikash