Quoting Martin Peres (2017-06-07 12:05:14)
> On 07/06/17 13:34, Chris Wilson wrote:
> > This reverts commit 7c8703fb02b248c2bcf9756bba8812bcfe7ed5d3.
> >
> > If we expect it to fail until we find a solution, let the hw fail and
> > continue to track the known failure in CI/bugs.
> >
> > Cc: Martin Peres
>
> I agree, the kernel/HW is broken.
>
> However, this fix is not necessarily wrong, as in, the real kernel
> fix/workaround may actually be worse than the problem it is trying to
> fix, because it would increase the power usage while only fixing a use
> case that has not been seen in the wild (or has it?).
Until we know the actual effect we are countering, it's hard to tell. If
it's kthread placement, we can try different wq. If it is just scheduler
latency, we raise it with the scheduler guys where a last resort may be
a tunable that compositors can use just to give themselves higher
responsiveness. It's all in the knowing. I presume that low latency
applications are already looking at making the display stack RT...
> We thus need to prove that this is the case and document this issue
> extensively in both IGT and the kernel. Until then, let's merge the
> revert to keep it in the task list:
>
> Acked-by: Martin Peres
Pushed until we find a third way.
-Chris
___
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
