https://bugs.winehq.org/show_bug.cgi?id=49764
Bug ID: 49764 Summary: Every program using virtual desktop crashes on startup after a recent Mesa snapshot added a second Vulkan device Product: Wine Version: 5.15 Hardware: x86-64 OS: Linux Status: UNCONFIRMED Severity: normal Priority: P2 Component: -unknown Assignee: wine-bugs@winehq.org Reporter: subgraph93@gmail.com Distribution: ---
Distribution: Ubuntu 20.04.1
Filing this report here because I think a Wine issue may be present. A specific Wine setting always triggers the issue, and other configurations (Wine without it or native programs) did not trigger the issue. I apologize for taking your time if it's a purely Mesa issue.
An update to a newer Mesa snapshot (Git acf756a -> Git a27823e) has introduced a new Vulkan device for vallium/llvmpipe (software renderer). As a result, I have two "GPUs", and after the Mesa update, I found that programs running in virtual desktop now crash at startup. I believe the update and the issue to be related.
I tested it on winecfg in a clean wineprefix. (I backed up the usual ~/.wine, ran `wine winecfg` to create the prefix, turned virtual desktop mode on by default, closed winecfg, then ran `wine winecfg` again.) These two lines seem to be always output: 0024:err:module:LdrInitializeThunk "comctl32.dll" failed to initialize, aborting 0024:err:module:LdrInitializeThunk Initializing dlls for L"C:\windows\system32\winecfg.exe" failed, status c0000005
This issue affects both 5.15 and 5.16, both devel and staging for me.
I know that more information is likely needed (probably logs with debug channels, or vulkaninfo?), not sure what I need to provide though. I have +relay,+seh logs of the tests with winecfg (as said above) for 5.15-devel and 5.16-devel.
#49631 may be related?
https://bugs.winehq.org/show_bug.cgi?id=49764
joaopa jeremielapuree@yahoo.fr changed:
What |Removed |Added ---------------------------------------------------------------------------- CC| |jeremielapuree@yahoo.fr
--- Comment #1 from joaopa jeremielapuree@yahoo.fr ---
A specific Wine setting always triggers the issue
What is the specific setting that triggers the bug? You need to be more explicit.
https://bugs.winehq.org/show_bug.cgi?id=49764
--- Comment #2 from subgraph93@gmail.com --- The specific setting is turning on virtual desktop emulation for any program. Sorry if that was not clear.
https://bugs.winehq.org/show_bug.cgi?id=49764
Zhiyi Zhang zzhang@codeweavers.com changed:
What |Removed |Added ---------------------------------------------------------------------------- CC| |zzhang@codeweavers.com
https://bugs.winehq.org/show_bug.cgi?id=49764
--- Comment #3 from Zhiyi Zhang zzhang@codeweavers.com --- Please provide a +vulkan,+x11drv,+xrandr,+seh,+module log
https://bugs.winehq.org/show_bug.cgi?id=49764
--- Comment #4 from subgraph93@gmail.com --- Created attachment 68070 --> https://bugs.winehq.org/attachment.cgi?id=68070 +vulkan,+x11drv,+xrandr,+seh,+module log for the winecfg test using 5.16-devel
Attached a compressed log file (the uncompressed file is ~2.2 Mbytes, the archive is 48 Kbytes)
https://bugs.winehq.org/show_bug.cgi?id=49764
--- Comment #6 from subgraph93@gmail.com --- The issue also happened in Wine 5.18 and 5.19. On the Mesa side, however, the device got renamed to "lavapipe" some time ago, and more importantly, after I installed a new snapshot today, it no longer shows for me in vulkaninfo, and the virtual desktop issue is also gone. Appears the lavapipe driver is somehow no longer installed (as opposed to still installed and not causing the issue), which I'm not sure affects this Wine issue. (At least it'll be difficult for me to test any Wine changes)
This potentially relevant commit got included in the update I performed today: https://gitlab.freedesktop.org/mesa/mesa/-/commit/335132cd4dd5fd04c789bacd1c...
I apologize for maybe posting too much about this, and also possibly for submitting a questionably valid report.
https://bugs.winehq.org/show_bug.cgi?id=49764
Andrey Gusev andrey.goosev@gmail.com changed:
What |Removed |Added ---------------------------------------------------------------------------- Status|UNCONFIRMED |NEW Keywords| |regression Ever confirmed|0 |1 Regression SHA1| |145cfce1135a7e59cc4c89cd05b | |572403f188161 Component|-unknown |winex11.drv
--- Comment #7 from Andrey Gusev andrey.goosev@gmail.com --- Caused by https://source.winehq.org/git/wine.git/commit/145cfce1135a7e59cc4c89cd05b572...
https://bugs.winehq.org/show_bug.cgi?id=49764
Zhiyi Zhang zzhang@codeweavers.com changed:
What |Removed |Added ---------------------------------------------------------------------------- CC| |loothelion@nvidia.com
--- Comment #8 from Zhiyi Zhang zzhang@codeweavers.com --- *** Bug 50542 has been marked as a duplicate of this bug. ***
https://bugs.winehq.org/show_bug.cgi?id=49764
Zhiyi Zhang zzhang@codeweavers.com changed:
What |Removed |Added ---------------------------------------------------------------------------- Resolution|--- |NOTOURBUG Status|NEW |RESOLVED
--- Comment #9 from Zhiyi Zhang zzhang@codeweavers.com --- Should be resolved by https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8551
https://bugs.winehq.org/show_bug.cgi?id=49764
Zhiyi Zhang zzhang@codeweavers.com changed:
What |Removed |Added ---------------------------------------------------------------------------- CC| |edpil02@orange.fr
--- Comment #10 from Zhiyi Zhang zzhang@codeweavers.com --- *** Bug 50481 has been marked as a duplicate of this bug. ***
https://bugs.winehq.org/show_bug.cgi?id=49764
--- Comment #11 from Zhiyi Zhang zzhang@codeweavers.com --- For those affected, you can also use the VK_ICD_FILENAMES workaround at https://gitlab.freedesktop.org/mesa/mesa/-/issues/4112 before the mesa fix land.
https://bugs.winehq.org/show_bug.cgi?id=49764
--- Comment #12 from subgraph93@gmail.com --- I'm still getting the issue with Wine 6.0 and Mesa 21.1.0-devel (git-34e3e16 2021-01-27 groovy-oibaf-ppa) that looks to me like it should contain the fix. (Created a fresh prefix, set notepad to use virtual desktop, started notepad, it crashed with the same error message.) The VK_ICD_FILENAMES variable is still a needed workaround for me. Is there something wrong with how my system is set up?
https://bugs.winehq.org/show_bug.cgi?id=49764
Zhiyi Zhang zzhang@codeweavers.com changed:
What |Removed |Added ---------------------------------------------------------------------------- CC| |googlemail@email-postfach.i | |nfo
--- Comment #13 from Zhiyi Zhang zzhang@codeweavers.com --- *** Bug 51405 has been marked as a duplicate of this bug. ***
https://bugs.winehq.org/show_bug.cgi?id=49764
--- Comment #14 from googlemail@email-postfach.info --- This caused me some confusion on current Debian Bullseye, which ships Mesa 20.3.4.
After glancing at the VK_ICD_FILENAMES workaround and the output of vulkaninfo ("WARNING: lavapipe is not a conformant vulkan implementation, testing use only."), it seems that /usr/share/vulkan/icd.d/lvp_icd.[i686|x86_64].json are causing more trouble than they're usefull, so I just renamed them.
Although somewhat dirty and invasive, this should be more convenient compared to messing around with VK_ICD_FILENAMES all the time? Using dpkg-divert instead might be a bit more Debian-y.
The affected Debian Bullseye packages are mesa-vulkan-drivers:amd64 and mesa-vulkan-drivers:i386, though I'm not sure how to report this issue in Debian.