On Wed, Sep 22, 2021 at 09:30:25AM +0200, o...@omarpolo.com wrote: > >Synopsis: drm errors in dmesg and X freezes > >Category: system > >Environment: > System : OpenBSD 7.0 > Details : OpenBSD 7.0 (GENERIC.MP) #221: Sat Sep 18 11:37:50 MDT > 2021 > > dera...@amd64.openbsd.org:/usr/src/sys/arch/amd64/compile/GENERIC.MP > > Architecture: OpenBSD.amd64 > Machine : amd64 > >Description: > > I guess this is nothing new, but I have some X freezes with > amdgpu(4).
Yes, this problem has persisted across multiple major drm versions. It can be triggered by running piglit from ports but isn't related to a particular test. Running piglit with -1 to disable concurrent tests does not trigger it. > > The system is still responsive (i.e. audio works and I can ssh > into the machine), but X stays frozen. Sometimes I can move > the mouse, not always, and when happens the shape of it doesn't > change. I'm unable to change tty. Restarting xenodm fails. > > The timing feels random. It can never happen in a day, or even > in a streak (i.e. freeze, reboot, freezes again after N > minutes, reboot, ...) but it's rarer. > > This seems to be slightly related to some drm errors logged to > the dmesg. I've seen these message reported without freeze, > and freeze without these messages, but most of the time they > happen when there's a freeze. > > I've just had a freeze with this logged to dmesg: > > [drm] *ERROR* ring sdma0 timeout, signaled seq=975, emitted seq=975 > [drm] *ERROR* Process information: process pid 0 thread pid 0 > [drm] *ERROR* ring sdma0 timeout, signaled seq=1649, emitted seq=1649 > [drm] *ERROR* Process information: process pid 0 thread pid 0 > > Yesterday I got (yep, without a newline before the second > ``[drm]'', dunno if it's important) > > [drm] *ERROR* Waiting for fences timed out![drm] *ERROR* ring sdma0 timeout > signaled seq=2733, emitted seq=2733 > [drm] *ERROR* Process information: process pid 0 thread pid 0 > > and from what I recall the message is always the timeout (with > different seq number every time ofc, but the signaled seq > number and the emitted seq is always the same) > > >How-To-Repeat: > > I don't know. If feels random. I usually have a bunch of > emacs frames and xterm windows open in cwm, with a firefox or > chrome window, gajim and mpv playing music. Sometimes I have a > vm running, but it doesn't seem to trigger/avoid the freeze. > > I've had the *impression* that some GTK applications have more > chances of triggering these freezes. I've switched from dino > to gajim because I've had several freezes when interacting > with its UI (in particular the emoji picker). NB: this > paragraph is complete speculation, I don't have exact data to > back up my statement, it just *feels* that way. > > It seems agostic wrt system load. I've had some freeze while > building some ports, and other when the system was "almost" > idle. > > >Fix: > > No idea, sorry. Restarting xenodm from ssh result in a > failure, and the screen stays frozen. I can shudown(8) > cleanly thought.