On Mon, Nov 16, 2026 at 06:29:38PM +0100, Erwin Springer wrote:
> 
> Am 16.11.2026 um 10:11 schrieb Ville Syrjala:
> > From: Ville Syrjälä <[email protected]>
> >
> > The weird uapi semantics of the atomic async flip support make it
> > really hard for drivers to implement. Try to fix that by filtering
> > out the no-op junk from the commit that userspace likes to put there.
> >
> > Also a lot of people are apparently confusing the
> > .atomic_async_{update,check}() hack with async flips. The two
> > are completely orthogonal, so try to clear up that mess by
> > renaming the hack to .atomic_bypass_{update,check}().
> >
> > TODO: atomic+async is in serious need to proper tests, somehow
> >        all that seems to have gotten skipped when the feature
> >        was added :(
> 
> thanks, this fixes the KWin case for me, so my patch is no longer
> needed..
> 
> tested on Raptor Lake-U (Iris Xe) with unmodified KWin 6.8.2 and a
> fullscreen client that requests tearing through wp_tearing_control_v1,
> without my patch. I applied the series to the CachyOS 7.3.4 kernel,
> the only conflicts were the renames in amdgpu and msm.
> 
> Before, none of about 350 frames per run was presented without vsync.
> With the series 3330 of 3336 frames tear over four runs (with and
> without CCS modifiers), and the "async flip not supported" message is
> gone. What is still rejected is the first flip after the modifier or
> format changed, once per run, as intended, and in three of the runs
> one commit that touches the cursor plane, which can't do async flips:
> 
>    [PLANE:146:cursor A] does not support async flips
> No warnings and no flip_done timeouts in the log.
> 
> If you think it fits, feel free to add:
> 
> Reported-by: Erwin Springer <[email protected]>
> Tested-by: Erwin Springer <[email protected]>

Thanks. There are quite a few unresolved issues left here
though, so we might want you patch as a short term fix.
I'll have to go read it with some actual thought...

-- 
Ville Syrjälä
Intel

Reply via email to