gnome-settings-daemon / check_gl_texture_size deadlock because g-s-d has server grab

[Chris Coulson]

I've noticed for the past week or so that every time I open my laptop lid or press the display key, my laptop appears to lock up entirely. Initially, I thought this might be a compiz or X bug, as the only obvious way to recover is to kill my session and log in again. However, I did a bit of debugging this morning and this is due to the recent gnome-desktop change to work around bug 824099.

What happens is:

1) gnome-settings-daemon takes a server grab
2) It then runs check_gl_texture_size
3) check_gl_texture_size hangs in XOpenDisplay because g-s-d has a grab
4) g-s-d hangs because it waits for check_gl_texture_size to exit
5) The entire desktop appears to be locked up, because g-s-d never lets go of its grab

This is repeatable for me - it happens every single time I open my laptop lid, which is a bit of an inconvenience tbh (my laptop seems to trigger a fake Fn+F7 keypress when this happens, which is another matter)

If you're curious, this is where g-s-d hangs (sorry, I didn't have gnome-desktop symbols installed for this one, but the missing symbol in frame 9 is screen_update):

#0 0x00007f93e7b69c3d in __libc_waitpid (pid=<optimized out>, stat_loc=<optimized out>, options=<optimized out>) at ../sysdeps/unix/sysv/linux/waitpid.c:41
#1 0x00007f93e77435fa in ?? () from /usr/lib/libgnome-desktop-3.so.2
#2 0x00007f93e774477b in gnome_rr_config_applicable () from /usr/lib/libgnome-desktop-3.so.2
#3 0x00007f93d7dfa19d in auto_configure_outputs (timestamp=1506448, manager=0xb64190) at gsd-xrandr-manager.c:1564
#4 on_randr_event (screen=<optimized out>, data=<optimized out>) at gsd-xrandr-manager.c:1681
#5 0x00007f93e68db0a4 in g_closure_invoke (closure=0xb6c890, return_value=0x0, n_param_values=1, param_values=0xd68660, invocation_hint=<optimized out>)
    at /build/buildd/glib2.0-2.30.0/./gobject/gclosure.c:774
#6 0x00007f93e68ed02a in signal_emit_unlocked_R (node=<optimized out>, detail=0, instance=0xb5d180, emission_return=0x0, instance_and_params=0xd68660)
    at /build/buildd/glib2.0-2.30.0/./gobject/gsignal.c:3272
#7 0x00007f93e68f66b1 in g_signal_emit_valist (instance=<optimized out>, signal_id=<optimized out>, detail=<optimized out>, var_args=<optimized out>)
    at /build/buildd/glib2.0-2.30.0/./gobject/gsignal.c:3003
#8 0x00007f93e68f6852 in g_signal_emit (instance=<optimized out>, signal_id=<optimized out>, detail=<optimized out>) at /build/buildd/glib2.0-2.30.0/./gobject/gsignal.c:3060
#9 0x00007f93e77412b4 in ?? () from /usr/lib/libgnome-desktop-3.so.2
#10 0x00007f93e7741390 in gnome_rr_screen_refresh () from /usr/lib/libgnome-desktop-3.so.2
#11 0x00007f93d7df9a4a in handle_fn_f7 (timestamp=1505379, mgr=0xb64190) at gsd-xrandr-manager.c:1185
#12 gsd_xrandr_manager_2_video_mode_switch (timestamp=1505379, manager=0xb64190, error=<optimized out>) at gsd-xrandr-manager.c:677
#13 handle_method_call (connection=<optimized out>, sender=<optimized out>, object_path=<optimized out>, interface_name=<optimized out>, method_name=<optimized out>,
    parameters=<optimized out>, invocation=0xc482a0, user_data=0xb64190) at gsd-xrandr-manager.c:2047
#14 0x00007f93e6bcb620 in call_in_idle_cb (user_data=<optimized out>) at /build/buildd/glib2.0-2.30.0/./gio/gdbusconnection.c:4446
#15 0x00007f93e6211a5d in g_main_dispatch (context=0xac1600) at /build/buildd/glib2.0-2.30.0/./glib/gmain.c:2441
#16 g_main_context_dispatch (context=0xac1600) at /build/buildd/glib2.0-2.30.0/./glib/gmain.c:3011
#17 0x00007f93e6212258 in g_main_context_iterate (context=0xac1600, block=<optimized out>, dispatch=1, self=<optimized out>) at /build/buildd/glib2.0-2.30.0/./glib/gmain.c:3089
#18 0x00007f93e6212792 in g_main_loop_run (loop=0x7f93d8016810) at /build/buildd/glib2.0-2.30.0/./glib/gmain.c:3297
#19 0x00007f93e722ce1d in gtk_main () at /build/buildd/gtk+3.0-3.2.0/./gtk/gtkmain.c:1367
#20 0x0000000000403caf in main (argc=1, argv=0x7fffd483a4f8) at main.c:466

And check_gl_texture_size:

#0 0x00007f7ad6716738 in __GI___poll (fds=0x7fff42d8f530, nfds=1, timeout=-1) at ../sysdeps/unix/sysv/linux/poll.c:83
#1 0x00007f7ad642eecd in ?? () from /usr/lib/x86_64-linux-gnu/libxcb.so.1
#2 0x00007f7ad642d369 in xcb_connect_to_fd () from /usr/lib/x86_64-linux-gnu/libxcb.so.1
#3 0x00007f7ad64301d8 in xcb_connect_to_display_with_auth_info () from /usr/lib/x86_64-linux-gnu/libxcb.so.1
#4 0x00007f7ad6c7ba02 in _XConnectXCB (dpy=0x1b90010, display=0x0, screenp=0x7fff42d8f87c) at ../../src/xcb_disp.c:78
#5 0x00007f7ad6c6b741 in XOpenDisplay (display=0x0) at ../../src/OpenDis.c:129
#6 0x0000000000400872 in main (argc=<optimized out>, argv=<optimized out>) at check_gl_texture_size.c:12

[Chris Halse Rogers (raof)]

I'm surprised that this happens each time you open your lid - check_gl_texture_size should be run exactly once per g-s-d process, since it caches the result.

I'll see what's happening with the server grab.

[Robert Roth (evfool)]

Does this have anything to do with bug #862161?

[Chris Halse Rogers (raof)]

I don't believe so, no.

[Martin Pitt (pitti)]

Hello Chris, or anyone else affected,

Accepted gnome-desktop3 into oneiric-proposed, the package will build now and be available in a few hours. Please test and give feedback here. See https://wiki.ubuntu.com/Testing/EnableProposed for documentation how to enable and use -proposed. Thank you in advance!

[Chris Coulson (chrisccoulson)]

The actual deadlock is fixed, although I still end up with a completely blank screen when I undock my laptop. However, it's recoverable by pressing the Fn+F7 key again, so I suspect it's not related to this and is another symptom of my laptop faking this keypress when I undock it.

[Launchpad Janitor (janitor)]

This bug was fixed in the package gnome-desktop3 - 3.2.0-0ubuntu4.1

---------------
gnome-desktop3 (3.2.0-0ubuntu4.1) oneiric-proposed; urgency=low

  * debian/patches/02_refuse_to_break_GL_compositors.patch:
    - Add extra paranoid checks to fail gracefully in the case of errors in
      check_gl_texture_size. Eliminates (otherwise harmless) crashes.
      (LP: #862161, LP: #862998, LP: #861804, LP: #861548)
    - Wait for check_gl_texture_size no longer than 500msec. Ensures that
      this check will not block the desktop, as in (LP: #871188).
  * Remove debian-changes patch which accidentally reverted the change meant
    to be applied in 3.2.0-0ubuntu4.
 -- Christopher James Halse Rogers <email address hidden> Tue, 11 Oct 2011 10:30:22 +1100