f1024564 breaks s0ix suspend on AMD Renoir when using Wayland
Affected version
Fedora 35, mutter-41.1-1.fc35.x86_64, Wayland
Bug summary
We are, together with AMD, trying to solve some remaining s0ix bugs on Renoir. Please refer to https://gitlab.freedesktop.org/drm/amd/-/issues/1712 for the details. The failure rate on Wayland went from around 1% to 100% after I upgraded to Fedora 35. I bisected it down to commit f1024564:
$ git bisect log
# bad: [12c46e64a85964ab14f8e5917138592f07d27d7a] Bump version to 41.beta
# good: [3f2c8f5f3b8f05bb09daeb7f12217394c9d8dd1c] Bump version to 40.3
git bisect start '41.beta' '40.3'
# good: [faf4240c74024d04f88986f95f65364ca8121ba4] Tag release 40.1
git bisect good faf4240c74024d04f88986f95f65364ca8121ba4
# bad: [fddc63159946737964cf7b6e39c4512d14d9b500] context/test: Enable 'scale-monitor-framebuffer' backend setting
git bisect bad fddc63159946737964cf7b6e39c4512d14d9b500
# good: [ee8c252a8c755a7164f70f43e2b8b43309d6a306] native: Add device pool for tracking device file descriptors
git bisect good ee8c252a8c755a7164f70f43e2b8b43309d6a306
# good: [2d939754b1e7230da6afbac642e11cfd0c098dbf] crtc-mode-info: Add vblank duration field
git bisect good 2d939754b1e7230da6afbac642e11cfd0c098dbf
# bad: [5e8c808cfb3b1edbc023741f6d0d5f112c825614] ci: Add job for pushing coverity reports
git bisect bad 5e8c808cfb3b1edbc023741f6d0d5f112c825614
# bad: [8c4a91ddd62823f8116487722e214db919f63944] clutter: Add swap time and GPU rendering duration to FrameInfo
git bisect bad 8c4a91ddd62823f8116487722e214db919f63944
# good: [fbe6740df121db60e1dbe0fbc6b01a72a6401461] cogl: Add GPU timestamp querying utilities
git bisect good fbe6740df121db60e1dbe0fbc6b01a72a6401461
# bad: [f1024564a2651dda8beee5f39682369eb32dfade] cogl: Store CPU and GPU rendering timestamps in frame info
git bisect bad f1024564a2651dda8beee5f39682369eb32dfade
# good: [8c258d1de1c1b544b170ee971dde7037ec3f4346] cogl: Add CPU swap time and GPU rendering query to CoglFrameInfo
git bisect good 8c258d1de1c1b544b170ee971dde7037ec3f4346
# first bad commit: [f1024564a2651dda8beee5f39682369eb32dfade] cogl: Store CPU and GPU rendering timestamps in frame info
Most of the commits were tested with gnome-shell@7f7b515b as with the following commit gnome-shell would not start.
Steps to reproduce
- Get an affected AMD Renoir system
- Attempt to suspend
What happened
System did not recover from resume.
What did you expect to happen
System resumes normally, same as it does with X.org or prior to commit f1024564.
Relevant logs, screenshots, screencasts etc.
There is a plethora of logs in https://gitlab.freedesktop.org/drm/amd/-/issues/1712. I can provide logs here as well, please let me know which ones would be relevant.