Bug 460513

Created attachment 152874 [details]
Output of journalctl /usr/bin/plasmashell, including crashes

for completeness/readability:

Operating System: Arch Linux
KDE Plasma Version: 5.26.0
KDE Frameworks Version: 5.99.0
Qt Version: 5.15.6
Kernel Version: 6.0.1-arch2-1 (64-bit)
Graphics Platform: Wayland

quick question - when kwin_wayland crashes, does it just leave you with a black screen + movable cursor?

(In reply to Jason Playne from comment #3)
> quick question - when kwin_wayland crashes, does it just leave you with a
> black screen + movable cursor?

To clarify, kwin_wayland didn't crash - I killed it manually.
But yes, I just got a black screen + movable cursor. (testing this now after killing plasmashell, I get the same)

I managed to generate a backtrace by recompiling the plasma-workspace package with options=(debug !strip), and using
gdb -p $(pidof plasmashell) --command=debug-script

debug-script:
```set logging file plasmashell.gdb
set logging enabled

break checkWaylandError(wl_display*)
continue
# backtrace
bt

# generate coredump
gcore
```

(I identified the source of the message "The Wayland connection broke. Did the Wayland compositor die?" with `strings --print-file-name *.so* | grep -i 'compositor die?'`, traced it to libQt5WaylandClient.so, and downloaded+searched the source for that string. 
The function is https://github.com/qt/qtwayland/blob/2303ee38ead0a4eafa9f8af629fd8495c45d1442/src/client/qwaylanddisplay.cpp#L66
static void checkWaylandError(struct wl_display *display))


Thread 1 "plasmashell" hit Breakpoint 1, checkWaylandError (display=0x55b3ab934160) at /usr/src/debug/qtwayland/src/client/qwaylanddisplay.cpp:91
91      {
#0  checkWaylandError(wl_display*) (display=0x55b3ab934160) at /usr/src/debug/qtwayland/src/client/qwaylanddisplay.cpp:91
#1  0x00007fb211122b7c in QtWaylandClient::EventThread::readAndDispatchEvents() (this=<optimized out>)
    at /usr/src/debug/qtwayland/src/client/qwaylanddisplay.cpp:141
#2  QtWaylandClient::QWaylandDisplay::flushRequests() (this=<optimized out>) at /usr/src/debug/qtwayland/src/client/qwaylanddisplay.cpp:419
#3  0x00007fb2106b0520 in QObject::event(QEvent*) () at /usr/lib/libQt5Core.so.5
#4  0x00007fb211378b1c in QApplicationPrivate::notify_helper(QObject*, QEvent*) () at /usr/lib/libQt5Widgets.so.5
#5  0x00007fb21068cb88 in QCoreApplication::notifyInternal2(QObject*, QEvent*) () at /usr/lib/libQt5Core.so.5
#6  0x00007fb21068d693 in QCoreApplicationPrivate::sendPostedEvents(QObject*, int, QThreadData*) () at /usr/lib/libQt5Core.so.5
#7  0x00007fb2106d3728 in  () at /usr/lib/libQt5Core.so.5
#8  0x00007fb20ed1981b in g_main_context_dispatch () at /usr/lib/libglib-2.0.so.0
#9  0x00007fb20ed6fec9 in  () at /usr/lib/libglib-2.0.so.0
#10 0x00007fb20ed180d2 in g_main_context_iteration () at /usr/lib/libglib-2.0.so.0
#11 0x00007fb2106d750c in QEventDispatcherGlib::processEvents(QFlags<QEventLoop::ProcessEventsFlag>) () at /usr/lib/libQt5Core.so.5
#12 0x00007fb21068532c in QEventLoop::exec(QFlags<QEventLoop::ProcessEventsFlag>) () at /usr/lib/libQt5Core.so.5
#13 0x00007fb21068fe59 in QCoreApplication::exec() () at /usr/lib/libQt5Core.so.5
#14 0x000055b3ab3521a3 in main(int, char**) (argc=<optimized out>, argv=<optimized out>) at /usr/src/debug/plasma-workspace-5.26.0/shell/main.cpp:233
warning: Memory read failed for corefile section, 4096 bytes at 0xffffffffff600000.
Saved corefile core.158040

I'm not sure how useful this is.

I also recompiled+installed the package qt5-wayland, which may explain the source file in the backtrace
#0  checkWaylandError(wl_display*) (display=0x55b3ab934160) at /usr/src/debug/qtwayland/src/client/qwaylanddisplay.cpp:91
instead of a binary
#3  0x00007fb2106b0520 in QObject::event(QEvent*) () at /usr/lib/libQt5Core.so.5

I don't know which way is preferred for debugging.

*** Bug 460336 has been marked as a duplicate of this bug. ***

(In reply to Jason Playne from comment #6)
> *** Bug 460336 has been marked as a duplicate of this bug. ***

I'm not sure this is a duplicate. For me, my programs remained open -- only plasmashell crashed. They were only lost when I killed kwin_wayland. There were no segfaults in either kwin_wayland or plasmashell.

To reproduce the black screen with movable cursor:
- killall plasmashell    (the screen is now black, as the desktop background is handled by plasmashell, but all the programs are open. In the original bug, plasmashell failed to relaunch)
- ps $(pgrep kwin_wayland)
    PID TTY      STAT   TIME COMMAND
 220965 ?        Ssl    0:00 /usr/bin/kwin_wayland_wrapper --xwayland
 237394 ?        Sl     0:02 /usr/bin/kwin_wayland --wayland-fd 7 --socket wayland-0 --xwayland-fd 8 --xwayland-fd 9
- killall kwin_wayland    (the screen is black, and all the programs are gone, but I can still move my cursor and launch a terminal with ctrl-alt-t)
- ps $(pgrep kwin_wayland)
    PID TTY      STAT   TIME COMMAND
 220965 ?        Ssl    0:00 /usr/bin/kwin_wayland_wrapper --xwayland
 238586 ?        Sl     0:02 /usr/bin/kwin_wayland --wayland-fd 7 --socket wayland-0 --xwayland-fd 8 --xwayland-fd 9 
(kwin_wayland has automatically been restarted. Killing kwin_wayland_wrapper gets me to the login screen)
- kstart plasmashell   (I can now use my session again)

Sorry if the original bug description was misleading. "Output of journalctl /usr/bin/kwin_wayland, including crashes" should have been "including crashes of plasmashell".

(In reply to space9301 from comment #7)
> (In reply to Jason Playne from comment #6)
> > *** Bug 460336 has been marked as a duplicate of this bug. ***
> 
> I'm not sure this is a duplicate. For me, my programs remained open -- only
> plasmashell crashed. They were only lost when I killed kwin_wayland. There
> were no segfaults in either kwin_wayland or plasmashell.
> 
> To reproduce the black screen with movable cursor:
> - killall plasmashell    (the screen is now black, as the desktop background
> is handled by plasmashell, but all the programs are open. In the original
> bug, plasmashell failed to relaunch)
> - ps $(pgrep kwin_wayland)
>     PID TTY      STAT   TIME COMMAND
>  220965 ?        Ssl    0:00 /usr/bin/kwin_wayland_wrapper --xwayland
>  237394 ?        Sl     0:02 /usr/bin/kwin_wayland --wayland-fd 7 --socket
> wayland-0 --xwayland-fd 8 --xwayland-fd 9
> - killall kwin_wayland    (the screen is black, and all the programs are
> gone, but I can still move my cursor and launch a terminal with ctrl-alt-t)
> - ps $(pgrep kwin_wayland)
>     PID TTY      STAT   TIME COMMAND
>  220965 ?        Ssl    0:00 /usr/bin/kwin_wayland_wrapper --xwayland
>  238586 ?        Sl     0:02 /usr/bin/kwin_wayland --wayland-fd 7 --socket
> wayland-0 --xwayland-fd 8 --xwayland-fd 9 
> (kwin_wayland has automatically been restarted. Killing kwin_wayland_wrapper
> gets me to the login screen)
> - kstart plasmashell   (I can now use my session again)
> 
> Sorry if the original bug description was misleading. "Output of journalctl
> /usr/bin/kwin_wayland, including crashes" should have been "including
> crashes of plasmashell".

Your description describes what I am seeing (better than I am able to put into words)

The segfault's and back traces I have seen I am thinking are incidental.

I believe what I see is what you are describing.

I suddenly get a back screen with a movable cursor, kwin_wayland is still running - lots of complains from the apps about losing their connection to the wayland server.

I *think* the apps are still alive - I work around the problem with restarting sddm (the big hammer approach)

Yes, it does look similar, though the cause is different. A few things:
- you reported a segfault in kwin_wayland, i.e. it tried to access invalid memory, then crashed. This is not incidental to what you experienced, it is the bug.
- kwin_wayland is not still running, it restarted. Note the different process IDs, 237394 vs 238586
- maybe other programs are still running after the crash, but I don't see them either. You would have to take a process tree before the crash with something like `ps axjf` and log to a file, then compare with after.

I'm happy to try to help you to reproduce your bug, though I do have limited time to spend on this now. I'll move further discussion over there.

Adding in some notes (to see if my usage patterns jive with yours)

I have 3 virtual desktops. When I primarily use just one of them (#3) the issue does not manifest

When I am regularly switching between each the problem manifests (but not on change)

On each desktop I have
#1 Java XWayland (jetbrains ide) + Opera Web Browser (Wayland)
#2 Vivaldi (Wayland) and Teams (X11, Electron)
#3 Discord (X11) and Firefox (X11)

I am wondering if this might be a Chromium/Wayland caused thing?

Making as dup as of 
> I identified the source of the message "The Wayland connection broke. Did the Wayland compositor die?"

*** This bug has been marked as a duplicate of bug 392376 ***