Bug 88892

Summary:	loading cms-colord.so causes "terminate" to fail
Product:	Wayland	Reporter:	James <james>
Component:	weston	Assignee:	Wayland bug list <wayland-bugs>
Status:	RESOLVED MOVED	QA Contact:
Severity:	normal
Priority:	medium
Version:	unspecified
Hardware:	x86-64 (AMD64)
OS:	Linux (All)
Whiteboard:
i915 platform:		i915 features:
Attachments:	annotated weston log showing Xwayland crash

Description James 2015-02-01 01:00:01 UTC

On Arch Linux, wayland 1.6.1-1, weston 1.6.1-1.

Starting from a VT with just "weston", running the drm-backend.so, loading cms-colord.so in ".config/weston.ini", with "modules=xwayland.so,cms-colord.so", as described in man 5 weston.ini, causes "terminate" to fail, using "keymap_options=terminate:ctrl_alt_bksp".  This is on a two-headed display.

After attempting to terminate, one screen shows a virtual terminal, though not the VT from which weston was started, the other screen continuing to show the weston display, the keyboard becomes "locked", and weston continues to run. From a remote terminal, "ps" shows weston sleeping, and weston-keyboard and weston-desktop-shell as zombies.  "kill -2" does nothing.  "kill -9" terminates weston, and the display and keyboard return to normal operation.

For what it might be worth, running as "WAYLAND_DEBUG=1 weston" ends with:

Killed70.967]  -> wl_subcompositor@5.destroy()rface@21) 0)0, 5)ttach(10592907wl_buffer@37)_buffer@)_buffer@, 2774551, )300)@6000014)-motion button

Of course, simply not including cms-colord.so in "modules" avoids the problem.

Comment 1 James 2015-02-01 01:16:52 UTC

I see that this is the same action as in Bug 81526, but here, I am seeing a direct relationship to loading, or not loading, cms-colord.so.  The terminate failure is not at all random, but consistently reproducible.

Comment 2 James 2015-02-01 19:57:48 UTC

Created attachment 113024 [details]
annotated weston log showing Xwayland crash

Comment 3 James 2015-02-01 20:01:11 UTC

I have attached output from "weston > westonlog 2>&1", with the following stages annotated:

1       start weston and weston-terminal
2       start dillo
3       switch to Xorg
4       switch back to weston
5       mouse-over dillo and crash Xwayland, weston-terminal continues normally
6       access the panel and crash weston

An alternative sequence, switching to Xorg and back _before_ starting dillo, and _then_ starting dillo after the switch and back, "ps" will say that dillo and Xwayland are running, but no dillo window will be seen.  In contrast, at the same time, weston-terminal can still be started and presents a normal window.

Often, but not always, the first attempt at "Ctrl-Alt-F1" to switch to xorg will fail.  This "probabilistic" control response means something is wrong.

Often, but not always, accessing the panel after Xwayland has crashed, will also crash weston.

When Xwayland has _not_ crashed, and then terminating weston, Xwayland does not seem to terminate "gracefully", showing:

(EE)
Fatal server error:
(EE) failed to dispatch Wayland events: Broken pipe
(EE) 
Can't open display: :1

but I suppose that that is a separate issue.

Please see the attached log.

Comment 4 James 2015-02-01 20:04:40 UTC

Oops - I just added a comment and an attachment to the wrong bug report.  Grr...

Comment 5 James 2015-02-16 04:17:35 UTC

I tested this on weston 1.7.0 and wayland 1.7.0, and it is still a problem.

I do notice that if I start weston, switch VTs, then switch back to weston, then enter Ctrl-Alt-Backspace, weston will terminate properly, though the display that would otherwise continue to show the weston desktop when weston fails to terminate seems to take a longish time, several seconds, to redisplay the VT.  Without cms-colord.so loaded, both screens redisplay the VT instantly after weston terminates.

Weston fails to terminate only if I start weston, including cms-colord.so, and then try to terminate without first switching VTs.

Also, I'm not sure that "two-headed" is the correct terminology here.  I have a single video card - an AMD/ATI Redwood PRO Radeon HD 5570 - and two display monitors.

Comment 6 GitLab Migration User 2018-06-08 23:53:46 UTC

-- GitLab Migration Automatic Message --

This bug has been migrated to freedesktop.org's GitLab instance and has been closed from further activity.

You can subscribe and participate further through the new bug through this link to our GitLab instance: https://gitlab.freedesktop.org/wayland/weston/issues/63.

Use of freedesktop.org services, including Bugzilla, is subject to our Code of Conduct. How we collect and use information is described in our Privacy Policy.