Ensure progress for dma_fence_array

Chia-I Wu olvaffe at gmail.com
Tue Nov 12 21:10:38 UTC 2024


On Tue, Nov 12, 2024 at 4:00 AM Christian König
<ckoenig.leichtzumerken at gmail.com> wrote:
>
> Am 09.11.24 um 01:32 schrieb Chia-I Wu:
> > On Fri, Nov 8, 2024 at 1:43 AM Christian König
> > <ckoenig.leichtzumerken at gmail.com> wrote:
> >> Hi guys,
> >>
> >> as pointed out by Chia-I userspace doesn't see any progress when
> >> signaling is not enabled and Boris noted that this is because
> >> dma_fence_array_signaled() never returns true in this case.
> >>
> >> Improve this by fixing the dma_fence_array_signaled() implementation to
> >> also return true even if signaling was never explicitely enabled.
> > Yeah, this fixes the timeout I was seeing on panvk.
>
> Any objections to add your Tested-by?
No.  v3 is also

  Tested-by: Chia-I Wu <olvaffe at gmail.com>

>
> >
> >> We should probably adjust the documentation as well that when the
> >> callback is implemented it should make progess visible even without
> >> enabling signaling.
> > That would be really nice.  Both dma_fence_is_signaled and
> > dma_fence_ops::signaled explicitly state otherwise at the moment.
>
> I have put that on my TODO list, but could take a while until I have
> time for it.
>
> If anybody wants to make suggestions on the wording feel free to make a
> patch.
>
> Thanks,
> Christian.
>
> >
> >
> >
> >> Please test and review,
> >> Christian.
> >>
> >>
>


More information about the dri-devel mailing list