Le Wed, May 25, 2022 at 11:36:44AM +0200, Landry Breuil a écrit : > Le Wed, May 25, 2022 at 09:47:53AM +0200, Landry Breuil a écrit : > > Le Tue, May 24, 2022 at 06:04:27PM -0400, George Koehler a écrit : > > > Hi, > > > > > > This mail is about the OpenBSD package of firefox-esr-91.9.1, on my > > > amd64 with amdgpu. Some tabs crashed. My reproducer stopped crashing > > > after I unveiled /dev/dri in the content process, > > > > Thanks for the details bugreport, i'll have to undust the amdgpu laptop > > to test, but some observations/questions first: > > - does the same problem also affect firefox 100 ? > > ive been able to reproduce the crash on firefox-esr on the T495s, and > can confirm that unveiling /dev/dri for content process works around the > crash.
looking at mesa code, the only occurence i've found of /dev/dri that wasnt /dev/dri/card0 is https://github.com/openbsd/xenocara/blob/e26c45de6dcbcdbd24a9f4c0b565fc5b1fb86b1e/lib/mesa/src/gallium/drivers/radeonsi/glsl_tests/amdgcn_glslc.c#L75 - does it work for you with /dev/dri/renderD128 unveiled (in addition to card0, without completely unveiling /dev/dri) ? I would also have expected a "ERROR: Couldn't open /dev/dri/renderD128" message on stderr. I've commited the /dev/dri line to unveil.content for now, but i want to make sure that the problem really comes from mesa. I havent spotted differences between unveil configs in firefox & firefox-esr, so i dont understand why firefox doesnt crash the same. Maybe different codepaths taken in a more recent codebase ? Landry
