1239962 - SIGBUS crash in libgtk-3.so.0.1400.5@0x360303

Status: RESOLVED DUPLICATE of bug 1205199

(Core :: Widget: Gtk, defect)

Description

[Calixte Denizet (:calixte)]

This bug was filed from the Socorro interface and is 
report bp-48367756-4568-42d1-8874-dd42d2160115.
=============================================================

1) I report this bug for Sylvestre Ledru just to try Crash Stats.
2) The fx is stayed opened during two days without any activitiy and when Sylvestre came back fx crashed
3) Only few crashes for the moment.

Comment 1

[Calixte Denizet (:calixte)]

As we are planning to ship with GTK 3 this year, I don't think we should be shipping with the topcrash.

Comment 2

[Sylvestre Ledru [:Sylvestre]]

Tracking for the reason Calixte mentioned.

Comment 3

[Sylvestre Ledru [:Sylvestre]]

Karl, Firefox 45 with gtk3 is crashing a lot on this bug.
It is one of the most important source of crashes for Firefox 45 in aurora.
Could you help here? Thanks

Comment 4

[Mike Hommey [:glandium]]

Note that since we don't have symbols for those system libs, there may well be several different crashes, but we can't know.

Comment 5

[Karl Tomlinson (:karlt)]

All the SIGBUS crash reports I've seen with the signatures linked here have
similar short stacks.  The only comments I've found so far are:

"just closed the window while it was playing a youtube music video"

"Spotify Web crashed Firefox Nightly"
https://crash-stats.mozilla.com/report/index/c1ae1c25-2dc7-4991-b991-2b25d2160122

The only SIGSEGV report with signature linked here that seems to be related is
at 0x18 in 
https://crash-stats.mozilla.com/report/index/343e57a5-fd08-47cd-b116-b8d882160129
That stack is the same as a SIGBUS at 0x0 in
https://crash-stats.mozilla.com/report/index/96ff491b-a615-46ac-99bb-01dec2160129

Many of the signatures linked here have quite different looking SIGSEGV crash stacks and no SIGBUS reports.  I don't have a reason to believe they are related.

Taking libgtk-3-0_3.16.7-0ubuntu3_amd64.deb of Ubuntu Wily,

% /home/karl/moz/dev/obj/toolkit/crashreporter/google-breakpad/src/tools/linux/dump_syms/dump_syms usr/lib/x86_64-linux-gnu/libgtk-3.so.0.1600.7 2>/dev/null | grep MODULE
MODULE Linux x86_64 A70A00B4F80CF8FB855888B4E8EA0B660 libgtk-3.so.0.1600.7
% /home/karl/moz/dev/obj/toolkit/crashreporter/google-breakpad/src/tools/linux/dump_syms/dump_syms usr/lib/x86_64-linux-gnu/libgdk-3.so.0.1600.7 2>/dev/null | grep MODULE 
MODULE Linux x86_64 520E1B50883A2309F0C809D1931623CB0 libgdk-3.so.0.1600.7

https://crash-stats.mozilla.com/report/index/ecb01470-90e5-469d-bb7e-489952160115
is a typical SIGBUS crash report similar to that in comment 0, and has module
debug identifiers matching those above.

% addr2line -if -e usr/lib/debug/usr/lib/x86_64-linux-gnu/libgtk-3.so.0.1600.7 0x386a00
gtk_socket_filter_func
/build/gtk+3.0-pIzeMy/gtk+3.0-3.16.7/./gtk/gtksocket.c:1371
~/downloads/ubuntu/wily% addr2line -if -e usr/lib/debug/usr/lib/x86_64-linux-gnu/libgdk-3.so.0.1600.7 0x58290
gdk_event_apply_filters
/build/gtk+3.0-pIzeMy/gtk+3.0-3.16.7/./gdk/x11/gdkeventsource.c:81
~/downloads/ubuntu/wily% addr2line -if -e usr/lib/debug/usr/lib/x86_64-linux-gnu/libgdk-3.so.0.1600.7 0x5856c
gdk_event_source_translate_event
/build/gtk+3.0-pIzeMy/gtk+3.0-3.16.7/./gdk/x11/gdkeventsource.c:202
_gdk_x11_display_queue_events
/build/gtk+3.0-pIzeMy/gtk+3.0-3.16.7/./gdk/x11/gdkeventsource.c:338

This looks like an issue with the browser code hosting windowed plugins.
I'll see if I can work out more.

Comment 6

[Martin Stránský [:stransky] (ni? me)]

I'm going to disable Mozilla crash reports in Fedora Firefox to catch those crashes by Fedora ABRT tool which is able to get debuginfo from Gtk3 and other local backages.

Comment 7

[Karl Tomlinson (:karlt)]

I'm removing signatures that look unrelated to comment 0 (bug 1205199).
If any of the other signatures are top crashers then we'll need to resolve
them separately, which would be better tracked in a different bug report.

(In reply to Martin Stránský from comment #6)
> I'm going to disable Mozilla crash reports in Fedora Firefox to catch those
> crashes by Fedora ABRT tool which is able to get debuginfo from Gtk3 and
> other local backages.

Thanks, Martin.  That makes a lot of sense while the topcrashers are crashing
with system libraries on the stack.

Comment 8

[Liz Henry (:lizzard) (relman/hg->git project)]

Tracking for 46+ since this is a topcrash.

Comment 9

[Mike Hommey [:glandium]]

We should add symbols to soccoro for Gtk+3 from common versions on e.g. Ubuntu. Ted, can we do that easily?

Comment 10

[(not currently active) Ted Mielczarek]

(In reply to Mike Hommey [:glandium] from comment #9)
> We should add symbols to soccoro for Gtk+3 from common versions on e.g.
> Ubuntu. Ted, can we do that easily?

I have scripts for doing this for Ubuntu packages:
https://github.com/luser/linux-symbol-scraping/

It wouldn't be hard to tweak them to fetch just specific packages. I assume it wouldn't be that hard to scrape symbols from Fedora RPMs either, it's just a matter of locating them, unpacking them, running dump_syms.

Comment 11

[Sylvestre Ledru [:Sylvestre]]

We disabled GTK 3 in 45, blocking 46 now.

Comment 12

[Martin Stránský [:stransky] (ni? me)]

Fedora backtrace(s) for this one seems to be this one:

https://retrace.fedoraproject.org/faf/problems/bthash/?bth=1ce2b5eddf03a259a06facde825e3ed6e228d140&bth=2f33a0d0869a696c7b03799c7bd0a16698b62567&bth=685f02eb466c4e9db73f99094b32b8246239b830&bth=a80aa58d97747038183b1b3cb134cd2adff8514d&bth=e2c60a0233161e44d3ab4ef834b6c6b3a906e772

1 gtk_socket_filter_func
2 gdk_event_apply_filters
3 _gdk_x11_display_queue_events
4 gdk_display_get_event
5 gdk_event_source_dispatch

Comment 13

[Martin Stránský [:stransky] (ni? me)]

This one looks interesting:

https://retrace.fedoraproject.org/faf/reports/1000357/

[...]
5 RefPtr<nsWindow>::AddRefTraits<nsWindow>::AddRef(nsWindow*) 
6 nsWindow::SetNonXEmbedPluginFocus() 
7 plugin_window_filter_func
8 gdk_event_apply_filters
9 _gdk_x11_display_queue_events

Comment 14

[Milan Sreckovic [:milan] (needinfo for best results)]

We "just" have the crash reports and no way to reproduce this locally?

Comment 15

[Karl Tomlinson (:karlt)]

The reports in comment 0, 5, and 12 should be fixed by the patch in bug 1205199, thanks.

I don't know what is happening in comment 11, but PluginType_NONXEMBED is rare, which acroread being the most notable, but that is disabled by default.  Anyway, that seems to be quite a different bug.

Comment 16

[Karl Tomlinson (:karlt)]

(In reply to Karl Tomlinson (ni?:karlt) from comment #15)
> I don't know what is happening in comment 11, but PluginType_NONXEMBED is

Comment 13, I mean.

Comment 17

[Liz Henry (:lizzard) (relman/hg->git project)]

Tracking in bug 1205199 instead since this is now marked as a duplicate.