Re: Visual artifacts in Firefox after bus_space changes

[David Brownlee <abs%absd.org@localhost>]

On Mon, 30 Dec 2019 at 01:42, Taylor R Campbell
<campbell+netbsd-port-amd64%mumble.net@localhost> wrote:
>
> > Date: Thu, 26 Dec 2019 22:39:58 +0000
> > From: David Brownlee <abs%absd.org@localhost>
> >
> > I'm seeing visual artifacts in Firefox - I believe after
> > http://releng.netbsd.org/cgi-bin/req-9.cgi?show=566 was pulled up to
> > netbsd-9, and from a quick test with a current kernel its as if a
> > partial screen update can happen, when drawing a new character when
> > typing (eg into gmail or slack), highlighting some text or even
> > updating a screen element on hover. Its quickly replaced with the
> > correct content.
>
> Interesting.  I would be curious to see a stack trace for when
> bus_space_barrier gets called during this event -- perhaps with
>
> dtrace -n 'fbt::bus_space_barrier:entry { @[stack()] = count() }'
>
> and running the test (then hit ^C when done), or by modifying
> bus_space_barrier to print a stack trace and consulting dmesg.
>
> I don't see any obvious way for it to be called by any of the drmkms
> drivers themselves, but perhaps it's coming from another driver.
> Either way, my best guess so far is that the impact on timing of
> issuing a fence is large enough that it makes drawing activity miss
> the deadline for a vblank event.

This would match with me possibly seeing the issue prior to the
bus_space change (rare and very fleeting), while with the bus_space
change it happens very frequently and is much more visible when it
triggers.

David