Re: [Intel-gfx] Picture stutter on i915 Graphics P630]

Re: [Intel-gfx] Picture stutter on i915 Graphics P630]

[Sebastian Andrzej Siewior]

On 2021-09-29 18:07:00 [+0200], Anton Lundin wrote:
> Hi.
Hi,

> While I was chasing some other bugs I tested 5.15-rc3-rt5 and noted your
> code e446c2038b58 ("drm/i915/gt: Use spin_lock_irq() instead of
> local_irq_disable() + spin_lock()") and noted that this patch created a
> lot of visual glitches, freeze-frames for about 0.5s every couple of
> seconds.
> 
> The card identifies itself as: (if that matters)
> Intel Corporation CoffeeLake-S GT2 [UHD Graphics P630] [8086:3e96]
> 
> Reverting that patch resolved quite a lot of the glitches but there are
> still a couple of minor ones left, that I'm trying to chase down.
> 
> 
> None of those glitches are there on the same workload on a regular
> 5.15-rc3 kernel.

You are talking about that commit and 5.15-rc3-rt5 with PREEMPT_RT
enabled?

> 
> //Anton

Sebastian

Re: [Intel-gfx] Picture stutter on i915 Graphics P630]

[Anton Lundin]

On 30 September, 2021 - Sebastian Andrzej Siewior wrote:

> On 2021-09-29 18:07:00 [+0200], Anton Lundin wrote:
> > Hi.
> Hi,
> 
> > While I was chasing some other bugs I tested 5.15-rc3-rt5 and noted your
> > code e446c2038b58 ("drm/i915/gt: Use spin_lock_irq() instead of
> > local_irq_disable() + spin_lock()") and noted that this patch created a
> > lot of visual glitches, freeze-frames for about 0.5s every couple of
> > seconds.
> > 
> > The card identifies itself as: (if that matters)
> > Intel Corporation CoffeeLake-S GT2 [UHD Graphics P630] [8086:3e96]
> > 
> > Reverting that patch resolved quite a lot of the glitches but there are
> > still a couple of minor ones left, that I'm trying to chase down.
> > 
> > 
> > None of those glitches are there on the same workload on a regular
> > 5.15-rc3 kernel.
> 
> You are talking about that commit and 5.15-rc3-rt5 with PREEMPT_RT
> enabled?

Yes, with that commit and a PREEMPT_RT, there's a lot if visual
glitches. With that commit reverted, there are still minor glitches
which comes from somewhere else in the PREEMPT_RT codepaths,
that I haven't found yet, but with that patch reverted the i915 output
is almost usable.


//Anton