mesa: 21.0.1 -> 21.0.3

[primeos]

Motivation for this change

Builds fine but still causes regressions, see #118753 (comment). This time I've tested it using the r600 driver and Sway doesn't launch either. Given that this wasn't fixed since 20.0.2 it seems like a packaging / Nixpkgs specific problem (I suspected the shader cache could be causing it but removing ~/.cache/mesa_shader_cache doesn't help either). I can hopefully take a look at it later but help is welcome.

Unfortunately I'm not getting much output (even with sway --debug, LIBGL_DEBUG=verbose, and MESA_DEBUG=1) so I might have to analyze the stack trace:

00:00:00.000 [INFO] [sway/main.c:346] Sway version 1.6
00:00:00.004 [INFO] [sway/main.c:154] Linux quorra 5.10.27 #1-NixOS SMP Tue Mar 30 12:32:09 UTC 2021 x86_64 GNU/Linux
00:00:00.005 [INFO] [sway/main.c:170] Contents of /etc/os-release:
00:00:00.005 [INFO] [sway/main.c:154] NAME=NixOS
00:00:00.005 [INFO] [sway/main.c:154] ID=nixos
00:00:00.005 [INFO] [sway/main.c:154] VERSION="21.05.git.d235056d6d6 (Okapi)"
00:00:00.005 [INFO] [sway/main.c:154] VERSION_CODENAME=okapi
00:00:00.005 [INFO] [sway/main.c:154] VERSION_ID="21.05.git.d235056d6d6"
00:00:00.005 [INFO] [sway/main.c:154] PRETTY_NAME="NixOS 21.05 (Okapi)"
00:00:00.005 [INFO] [sway/main.c:154] LOGO="nix-snowflake"
00:00:00.005 [INFO] [sway/main.c:154] HOME_URL="https://nixos.org/"
00:00:00.005 [INFO] [sway/main.c:154] DOCUMENTATION_URL="https://nixos.org/learn.html"
00:00:00.005 [INFO] [sway/main.c:154] SUPPORT_URL="https://nixos.org/community.html"
00:00:00.005 [INFO] [sway/main.c:154] BUG_REPORT_URL="https://github.com/NixOS/nixpkgs/issues"
00:00:00.005 [INFO] [sway/main.c:142] LD_LIBRARY_PATH=
00:00:00.005 [INFO] [sway/main.c:142] LD_PRELOAD=
00:00:00.005 [INFO] [sway/main.c:142] PATH=/run/wrappers/bin:/home/michael/.nix-profile/bin:/etc/profiles/per-user/michael/bin:/nix/var/nix/profiles/default/bin:/run/current-system/sw/bin
00:00:00.005 [INFO] [sway/main.c:142] SWAYSOCK=
00:00:00.005 [DEBUG] [sway/server.c:49] Preparing Wayland server initialization
00:00:00.005 [INFO] [wlr] [backend/session/logind.c:572] Selecting session from XDG_SESSION_ID: 8
00:00:00.030 [INFO] [wlr] [backend/session/logind.c:706] Successfully loaded logind session
00:00:00.032 [INFO] [wlr] [backend/backend.c:167] Found 1 GPUs
00:00:00.032 [INFO] [wlr] [backend/drm/backend.c:155] Initializing DRM backend for /dev/dri/card0 (radeon)
00:00:00.032 [DEBUG] [wlr] [backend/drm/drm.c:65] Atomic modesetting unsupported, using legacy DRM interface
00:00:00.032 [DEBUG] [wlr] [backend/drm/drm.c:82] ADDFB2 modifiers unsupported
00:00:00.032 [INFO] [wlr] [backend/drm/drm.c:244] Found 6 DRM CRTCs
00:00:00.032 [INFO] [wlr] [backend/drm/drm.c:171] Found 6 DRM planes
libGL: Can't open configuration file /etc/drirc: No such file or directory.
libGL: Can't open configuration file /home/michael/.drirc: No such file or directory.
[...]
[dmesg:]
[ 2690.144483] sway[68675]: segfault at a0 ip 00007f6dfcd665c6 sp 00007ffc7ecf7d90 error 4 in r600_dri.so[7f6dfc843000+d9c000]
[ 2690.144491] Code: 64 48 8b 04 25 28 00 00 00 48 89 44 24 58 31 c0 48 85 c9 74 03 48 8b 19 b8 01 00 00 00 89 f1 48 c7 44 24 30 01 00 00 00 d3 e0 <23> 85 a0 00 00 00 48 c7 44 24 38 00 00 00 00 c7 44 24 44 01 00 00

Edit: GDB backtrace:

(gdb) run                                                                                                                       |
Starting program: /nix/store/sv9ijg1zzqj7z35fjkf1s6jsk3dqgp9i-sway-unwrapped-1.6/bin/sway                                       |
[Thread debugging using libthread_db enabled]                                                                                   |
Using host libthread_db library "/nix/store/1jn6apz0fa9h9x7rl3v6vwiymwnjznwv-glibc-2.32-40/lib/libthread_db.so.1".              |
[Detaching after vfork from child process 13323]                                                                                |
[New Thread 0x7fffef24a640 (LWP 13324)]
[New Thread 0x7fffee8d3640 (LWP 13325)]
[New Thread 0x7fffee0d2640 (LWP 13326)]
[New Thread 0x7fffed8d1640 (LWP 13327)]
[New Thread 0x7fffed0d0640 (LWP 13328)]

Thread 1 "sway" received signal SIGSEGV, Segmentation fault.
0x00007ffff58e65c6 in driCreateContextAttribs () from /run/opengl-driver/lib/dri/r600_dri.so
(gdb) bt
#0  0x00007ffff58e65c6 in driCreateContextAttribs () from /run/opengl-driver/lib/dri/r600_dri.so
#1  0x00007ffff58e6a6e in driCreateNewContext () from /run/opengl-driver/lib/dri/r600_dri.so
#2  0x00007fffec8560c6 in dri2_setup_extensions ()
   from /nix/store/83q47kb6kvls9vkd4yg87c50rhkgfw5y-mesa-21.0.3-drivers/lib/libEGL_mesa.so.0
#3  0x00007fffec85bca6 in dri2_initialize_drm ()
   from /nix/store/83q47kb6kvls9vkd4yg87c50rhkgfw5y-mesa-21.0.3-drivers/lib/libEGL_mesa.so.0
#4  0x00007fffec854258 in dri2_initialize ()
   from /nix/store/83q47kb6kvls9vkd4yg87c50rhkgfw5y-mesa-21.0.3-drivers/lib/libEGL_mesa.so.0
#5  0x00007fffec84cb0c in eglInitialize ()
   from /nix/store/83q47kb6kvls9vkd4yg87c50rhkgfw5y-mesa-21.0.3-drivers/lib/libEGL_mesa.so.0
#6  0x00007ffff78d1167 in wlr_egl_create () from /nix/store/jvkgicprzd0sq9akf6kjx9fgcyfljibz-wlroots-0.13.0/lib/libwlroots.so.8
#7  0x00007ffff78d4241 in wlr_renderer_autocreate_with_drm_fd ()
   from /nix/store/jvkgicprzd0sq9akf6kjx9fgcyfljibz-wlroots-0.13.0/lib/libwlroots.so.8
#8  0x00007ffff78dd00c in init_drm_renderer ()
   from /nix/store/jvkgicprzd0sq9akf6kjx9fgcyfljibz-wlroots-0.13.0/lib/libwlroots.so.8
#9  0x00007ffff78d87a2 in wlr_drm_backend_create ()
   from /nix/store/jvkgicprzd0sq9akf6kjx9fgcyfljibz-wlroots-0.13.0/lib/libwlroots.so.8
#10 0x00007ffff78d70a2 in attempt_drm_backend ()
   from /nix/store/jvkgicprzd0sq9akf6kjx9fgcyfljibz-wlroots-0.13.0/lib/libwlroots.so.8
#11 0x00007ffff78d78b8 in wlr_backend_autocreate ()
   from /nix/store/jvkgicprzd0sq9akf6kjx9fgcyfljibz-wlroots-0.13.0/lib/libwlroots.so.8
#12 0x000000000041a5c2 in server_privileged_prepare (server=server@entry=0x47b0c0 <server>) at ../sway/server.c:52
#13 0x000000000041a371 in main (argc=1, argv=0x7fffffffbfa8) at ../sway/main.c:378

Things done

• Tested using sandboxing (nix.useSandbox on NixOS, or option sandbox in nix.conf on non-NixOS linux)
• Built on platform(s)
  • NixOS
  • macOS
  • other Linux distributions
• Tested via one or more NixOS test(s) if existing and applicable for the change (look inside nixos/tests)
• Tested compilation of all pkgs that depend on this change using nix-shell -p nixpkgs-review --run "nixpkgs-review wip"
• Tested execution of all binary files (usually in ./result/bin/)
• Determined the impact on package closure size (by running nix path-info -S before and after)
• Ensured that relevant documentation is up to date
• Fits CONTRIBUTING.md.

[primeos]

FWIW the llvmpipe driver interestingly seems to be fine (noticed that wile working on a VM test for Sway that should also be useful for testing Mesa):

I also wondered if it's related to one of those but didn't have time for a closer look so far:

• https://gitlab.freedesktop.org/mesa/mesa/-/commit/9a439ebcac0ec27ee8ed9b90e266ac2df5333918
• https://gitlab.freedesktop.org/mesa/mesa/-/commit/f1ec9335a8b39d4c9437c4f5e134deadad9da563
• https://gitlab.freedesktop.org/mesa/mesa/-/commit/4e4962b4649e09b66b6b7862a5b6bfcb95a26e38

[primeos]

Ok, I can confirm that #119558 (comment) applies here as well.

So when using a single Nixpkgs revision for the whole system this will work as expected but at least hardware.opengl.package no longer works for testing Mesa updates and I guess this will have additional implications (like not being able to use a Sway built against an older Nixpkgs revision via another channel, Nix Flakes, etc.).

Problems like this where already caused by glibc updates (#95808) and it seems like we can now add Mesa to that list (which would be bad as we update Mesa much more often). Unfortunately I don't know what caused this regression, if it's avoidable, and if it only affects wlroots-based Wayland compositors or also other Wayland compositors and X11.

Not sure what to do about this PR. Should we merge it or not and do we even have a choice?

@GrahamcOfBorg eval
@GrahamcOfBorg test sway

[primeos]

Not sure what to do about this PR. Should we merge it or not and do we even have a choice?

Ok, so my current plan is the following: Wait for the NixOS 21.05 branch-off on May 21st then merge this (likely 21.0.4 by then) into master and merge #119558 a few days/weeks after that (also depending on how it goes).
(Alternatively we could already merge this and revert it for NixOS 21.05 but that likely only makes testing more difficult.)

[primeos]

I've updated the commit message to make this more discoverable and to briefly summarize the issues/implications.
I'll merge this into staging now (instead of master as previously planed) to have some cached builds (simplifies a few tests a bit) and we can still merge #119558 into master (I could take care of the resulting merge conflict).

Edit: Forgot to mention: Launching Sway also works with the following configuration:

  system.replaceRuntimeDependencies = [
    ({ original = pkgs.mesa; replacement = (import /srv/nixpkgs-test { }).pkgs.mesa; })
    ({ original = pkgs.mesa.drivers; replacement = (import /srv/nixpkgs-test { }).pkgs.mesa.drivers; })
  ];
  # Or instead of mesa.drivers: hardware.opengl.package = (import /srv/nixpkgs-test { }).pkgs.mesa.drivers;