Bug 469572

The *entire* X session exits ?

There is a few multihead fixes for OOo we're not carrying in that version which I've since added in, but for the entire session to crash suggests more a more fundamental X level bogusity.

caolanm->mcepl: Can you see this problem ?

Reporter, there are two things:

a) if you are switching between non-packaged version of nvidia binary-only driver and nv open source driver (BTW, have you tried nouveau?) there are some things which nvidia driver changes outside of the driver itself. So, please, make sure, that your packages are not changed (rpm -Va; or look for more detailed instructions at https://fedoraproject.org/wiki/Xorg/3rd_Party_Video_Drivers).
b) Please, attach /var/log/Xorg.*.log, /var/log/dmesg, /var/log/messages and /etc/X11/xorg.conf (if there is any) as separate uncompressed attachments to this log.

> The *entire* X session exits ?

As in X restarts, and I have to log back in, yes.

> There is a few multihead fixes for OOo we're not carrying in that
> version which I've since added in

Where can I find those?  I update from rawhide daily.

> if you are switching between non-packaged version of nvidia
> binary-only driver and nv open source driver ...

First of all, I installed the binary only AFTER reproducing this error in the first place.  Second, I uninstalled the binary, rebooted, and checked to make sure that Xorg.0.log had no references to any nvidia provided code.  That said ...

> please, make sure, that your packages are not changed (rpm -Va)

OK (but, again, I reproduced this BEFORE ever installing the nvidia driver)

> Please, attach /var/log/Xorg.*.log, /var/log/dmesg,
> /var/log/messages and /etc/X11/xorg.conf

Should be able to do that sometime this week.

Created attachment 323070 [details]
/var/log/dmesg

Created attachment 323071 [details]
/var/log/messages

Created attachment 323072 [details]
/var/log/Xorg.0.log

Created attachment 323073 [details]
/var/log/Xorg.0.log.old

Created attachment 323074 [details]
/var/log/Xorg.1.log

Created attachment 323075 [details]
/var/log/Xorg.1.log.old

Comment on attachment 323074 [details]
/var/log/Xorg.1.log

From when I was uninstalling the nvidia drivers.

Comment on attachment 323075 [details]
/var/log/Xorg.1.log.old

From when I was uninstalling the nvidia drivers.

> b) Please, attach /var/log/Xorg.*.log, /var/log/dmesg, /var/log/messages and
> /etc/X11/xorg.conf (if there is any) as separate uncompressed attachments to
> this log.

#  ls -l messages dmesg Xorg.*
-rw-r--r-- 1 root root  50071 2008-11-10 08:26 dmesg
-rw------- 1 root root 238629 2008-11-10 08:48 messages
-rw-r--r-- 1 root root 162748 2008-11-10 08:37 Xorg.0.log
-rw-r--r-- 1 root root 226796 2008-11-10 08:34 Xorg.0.log.old
-rw-r--r-- 1 root root   4842 2008-11-05 22:24 Xorg.1.log
-rw-r--r-- 1 root root   4842 2008-11-05 22:24 Xorg.1.log.old
-rw-r--r-- 1 root root   4842 2008-11-05 22:24 Xorg.2.log
-rw-r--r-- 1 root root   4842 2008-11-05 22:24 Xorg.2.log.old
-rw-r--r-- 1 root root   4842 2008-11-05 22:24 Xorg.3.log
-rw-r--r-- 1 root root   4842 2008-11-05 22:24 Xorg.3.log.old
-rw-r--r-- 1 root root   4842 2008-11-05 22:24 Xorg.4.log
-rw-r--r-- 1 root root   4842 2008-11-05 22:24 Xorg.4.log.old
-rw-r--r-- 1 root root   4842 2008-11-05 22:24 Xorg.5.log
-rw-r--r-- 1 root root   4842 2008-11-05 22:24 Xorg.5.log.old

I attached /var/log/{dmesg, messages, Xorg.0.log*}.  Everything from today is in the Xorg.0.* files; older ones date back to when I uninstalled the nvidia driver, and made sure to force a reinstall of the stock code as outlined in the documents you had referenced.  There is no xorg.conf file.

Perhaps you are looking for this:

Backtrace:
0: /usr/bin/Xorg(xorg_backtrace+0x26) [0x4e7746]
1: /usr/bin/Xorg(xf86SigHandler+0x39) [0x47d3a9]
2: /lib64/libc.so.6 [0x3a38a33100]
3: /usr/bin/Xorg(XkbStringText+0x1c) [0x566b7c]
4: /usr/bin/Xorg(XkbWriteXKBGeometry+0x15a) [0x5581fa]
5: /usr/bin/Xorg(XkbWriteXKBKeymapForNames+0x5bb) [0x55777b]
6: /usr/bin/Xorg(XkbDDXLoadKeymapByNames+0x18a) [0x56171a]
7: /usr/bin/Xorg(ProcXkbGetKbdByName+0x3ab) [0x53f58b]
8: /usr/bin/Xorg(Dispatch+0x364) [0x446894]
9: /usr/bin/Xorg(main+0x45d) [0x42ccdd]
10: /lib64/libc.so.6(__libc_start_main+0xe6) [0x3a38a1e546]
11: /usr/bin/Xorg [0x42c0b9]

Fatal server error:
Caught signal 11.  Server aborting

Peter, isn't there something you want to tell us? :)

wow.

what's the output of setxkbmap -print before the crash happens?

Can you trigger the crash by running setxkbmap -layout us, or another combination of setxkbmap -layout .. -option ... etc?

The reason I ask is that setxkbmap also uses ProcXkbGetKbdByName so maybe we can find a good testcase here.

atropine:/tmp% eu-addr2line -e xorg-x11-server-debuginfo-1.5.2-12.fc10.x86_64/usr/lib/debug/usr/bin/Xorg.debug
0x566b7c
/usr/src/debug/xorg-server-1.5.2/xkb/xkbtext.c:526

Which is the for loop at the top of XkbStringText.  Which, uh.  Not really sure how that could fault...

*** IMPORTANT *** See the bottom of this report for critical information.

> what's the output of setxkbmap -print before the crash happens?

$  setxkbmap -print
xkb_keymap {
	xkb_keycodes  { include "evdev+aliases(qwerty)"	};
	xkb_types     { include "complete"	};
	xkb_compat    { include "complete"	};
	xkb_symbols   { include "pc+us+inet(evdev)"	};
	xkb_geometry  { include "pc(pc104)"	};
};

> Can you trigger the crash by running setxkbmap -layout us

[noel@noel-fedora10 ~]$  setxkbmap -layout us
[noel@noel-fedora10 ~]$ 

So, no.

> or another combination of setxkbmap -layout .. -option ... etc

Give me something concrete to try, and I'll be happy to do so.  :-)

*** HOWEVER *** we now have a new and different problem: AFTER TODAY'S UPDATES from rawhide, Impress no longer realizes that I have multiple monitors!  If I go into the Slide Show, anyway, Impress splits the slides across the expanded, virtual, display.  Impress does not go full-screen incorrectly, it just handles the Slide Show incorrectly.  I have tried rebooting with the second display connected and with it disconnected.  I have tried toggling it off and back on.  So far, I cannot get Impress to recognize that the second display exists.

I will attach the current Xorg.0.log to show today's situation.  Once we get THAT fixed, we can get back to this problem.  Do you want for me to open a new bug report?

Created attachment 323283 [details]
/var/log/Xorg.0.log

Two monitors configured, but Impress Slide Show Settings don't know about multiple monitors.

This bug appears to have been reported against 'rawhide' during the Fedora 10 development cycle.
Changing version to '10'.

More information and reason for this action is here:
http://fedoraproject.org/wiki/BugZappers/HouseKeeping

OK, we are back to reproducing this defect.

As noted in Bug 474719, we must use the nvidia driver because for whatever reason, OOov3 on Fedora 10 no longer recognizes multiple monitors when using nv or when using nvidia in Twinview mode (as noted in Bug 474719, I also tried the nouveau driver, but nouveau refuses to do anything other than mirror the displays).  The only way to get Impress to recognize multiple monitors on Fedora 10 appears to be using nvidia with separate X screens.

I am attaching my xorg.conf file and the Xorg log file.

Created attachment 325879 [details]
xorg.conf for Fedora 10 using Xinerama

Created attachment 325892 [details]
Shows the Xorg session crash

This is the crash with the Fedora 10 generated xorg.conf using nvidia and Xinerama and lacking the "Files" section (see comments in bug report).

OK, I finally have it working.  I had to use the nvidia driver -- nv and nouveau are not working -- with xinerama disabled.

It is very easy to crash X on Fedora 10, as noted in the logs, even with Fedora 10 generated xorg.conf files.  And it is not consistent in that the same section whose absence causes it to crash causes it not to crash depending on what else is in xorg.conf.  I have not tried to do an exhaustive test, just noted differences in content and behavior.  I've tried the xorg.conf file that was initially laid down by the RPM, as well as those generated by nvidia-settings on Fedora 10 (discounting those that I pulled over from Ubuntu when I was having a problem with nvidia-settings on Fedora).

Ideally, Bug 474719 will be resolved, and we won't have to deal with some of the xorg.conf issues.

Found this report after getting a very similar stacktrace:

Backtrace:
0: /usr/bin/Xorg(xorg_backtrace+0x3b) [0x812bc5b]
1: /usr/bin/Xorg(xf86SigHandler+0x51) [0x80bcc81]
2: [0x110400]
3: /usr/bin/Xorg(XkbStringText+0x2d) [0x81b0d0d]
4: /usr/bin/Xorg(XkbWriteXKBGeometry+0x1bf) [0x81a10ff]
5: /usr/bin/Xorg(XkbWriteXKBKeymapForNames+0x610) [0x81a0400]
6: /usr/bin/Xorg(XkbDDXLoadKeymapByNames+0x1d7) [0x81ab4c7]
7: /usr/bin/Xorg(ProcXkbGetKbdByName+0x40e) [0x81875be]
8: /usr/bin/Xorg [0x818c808]
9: /usr/bin/Xorg(Dispatch+0x34f) [0x8085e9f]
10: /usr/bin/Xorg(main+0x47d) [0x806b71d]
11: /lib/libc.so.6(__libc_start_main+0xe5) [0xa886d5]
12: /usr/bin/Xorg [0x806ab01]


Have seen it with OO3, but also getting without OO3 anywhere in site. I actually hit it today seemingly when using PackageKit (firefox, went to site which needed a new plugin)

Various apps seem to cause it, not found a reproducible pattern yet.

Language is en_GB, T60p laptop. Similr config to F9 where the issue did not occur.

My video driver is "radeonhd", also with F10, and all xorg* updates from the updates channel as of 11 Dec.

I do use a second screen (although it's not connected now) when at work, so I have a "Virtual   3200 1600" in xorg.conf which appeared to be needed for radeonhd (but not the ati driver)


Also found similar report at https://bugs.launchpad.net/ubuntu/+source/xorg/+bug/297442 for Ubuntu 8.10

Can provide more details if it helps (or raise seperate defect if required)

Also seems related to a new report https://bugzilla.redhat.com/show_bug.cgi?id=475746

*** Bug 475746 has been marked as a duplicate of this bug. ***

Agree with Nigel, that it's not related to OO3 impress, however I had OO3 Writer always just opened (did nothing).
P.S. intel video driver with no single external monitor working (just external VGA, LVDS is off) (dual-head stuff is not working properly anyway)

Back to being able to reproduce this, so here is the stack trace:

Backtrace:
0: /usr/bin/Xorg(xorg_backtrace+0x26) [0x4e7a26]
1: /usr/bin/Xorg(xf86SigHandler+0x39) [0x47a679]
2: /lib64/libc.so.6 [0x39a0832f90]
3: /usr/bin/Xorg(XkbStringText+0x1c) [0x56711c]
4: /usr/bin/Xorg(XkbWriteXKBGeometry+0x191) [0x5587d1]
5: /usr/bin/Xorg(XkbWriteXKBKeymapForNames+0x5bb) [0x557d1b]
6: /usr/bin/Xorg(XkbDDXLoadKeymapByNames+0x18a) [0x561cba]
7: /usr/bin/Xorg(ProcXkbGetKbdByName+0x3ab) [0x53fb2b]
8: /usr/bin/Xorg(Dispatch+0x364) [0x4468d4]
9: /usr/bin/Xorg(main+0x45d) [0x42cd1d]
10: /lib64/libc.so.6(__libc_start_main+0xe6) [0x39a081e576]
11: /usr/bin/Xorg [0x42c0f9]

Fatal server error:
Caught signal 11.  Server aborting

Reassigning to me, this seems to be uninitialised memory or so in xkb. A scratch build for testing is available (see below) which should avoid the issue.

http://koji.fedoraproject.org/scratch/whot/task_1041247/

To narrow this down further:
if you run xkbcomp -xkb :0 -, is anyone affected by this bug NOT getting a BadAlloc error? Anyone NOT on x86_64?

Peter, I have fully updated Fedora 10 from updates-testing, including OO3, and no longer see this problem.

The bug title is misleading, but I guess my bug is the same. I'm not using Fedora/Redhat, but since this bug is in xkb (and, actually, similarly looking xkb bugs are all over the bugtrackers), I hope I can contribute with a comment or two.

I can now easily reproduce it by letting this run for a minute:
  while;do setxkbmap -layout us;sleep 1;done
and using X normally (running firefox/terminal/openoffice).
Noel's backtrace has more symbols, thus I am not attaching mine, but it seems to die on ProcXkbGetKbdByName.

Additionaly, X log includes this (and a memory map):
[xkb] BOGUS LENGTH in write keyboard desc, expected 5548, got 5564
*** glibc detected *** /usr/bin/X: double free or corruption (!prev): 0x09a62df8 ***

I have reproduced it with xorg-server 1.5.3 and 1.6.0 RC1.

Peter:
I am running x86 and have no xorg.conf and no custom HAL policies, and after starting X, `xkbcomp -xkb :0 -` yields the BadAlloc error, but the error disappears after running setxkbmap.

Is the scratch xkb build still available?

I'm unable to get to http://koji.fedoraproject.org/scratch/whot/task_1041247/

I am getting this bug frequently. 

x86 (PAE kernel)

bash-3.2$ xkbcomp -xkb :0
Warning:          Could not load keyboard geometry for :0
                  BadAlloc (insufficient resources for operation)
                  Resulting keymap file will not describe geometry
Error:            Cannot open "server-0.xkb" to write keyboard description
                  Exiting
bash-3.2$

In comment #30, Noel mentioned he'd applied all updates-testing updates to F10 and no longer saw the issue.

I too have all updates-testing updates, but sadly *do* still see the issue :-(

I've not tried rawhide. Generally I'd probably pop it on sometime between beta and preview. I know alpha is soon.. a little reluctant to go down that route, but if there's any indication it might be fixed there then I would give it a whirl.

ANy thoughts in terms of
 - xserver config
 - OOO3 config

that might remove/reduce the risk of the problem occuring (other than not using Ooo3+dual screens)
?

*** Bug 479122 has been marked as a duplicate of this bug. ***

Updating bug title.

With current F10 X.org server (xorg-x11-server-Xorg-1.5.3-11.fc10.x86_64) I see lot of following errors:

(EE) Error loading keymap /var/lib/xkb/server-0.xkm

around of detecting my Trust mouse (see the bug marked as duplicate in comment #34) -- btw you marked the bug as duplicate, but isn't the main problem of that bug, that the mouse is identified as a keyboard?

Changed 2 things about 5 days ago -- and no crashes since

1. In keyboard preferences *deselected* option "Seperate layout for each window"
2. Disabled screensaver 

I've susbsequently applied F11 alpha/rawhide and will likely reintroduce the above settings to see if the issue comes back, but good news so far.

(In reply to comment #36)
> around of detecting my Trust mouse (see the bug marked as duplicate in comment
> #34) -- btw you marked the bug as duplicate, but isn't the main problem of that
> bug, that the mouse is identified as a keyboard?

No, a lot of mice these days are keyboards too.

Created attachment 331592 [details]
backtrace from gdb hooked to X after crash

I can recreate this at will, however it involves a USB Belkin KVM switch. 

I can recreate on x86_64 and i386.  This is a backtrace from gdb hooked up to X which I captured this morning. 

System setup:
x86_64 F10 machine with it's own monitor and all updates
i386 F10 machine with it's own monitor and all updates
USB Mouse & Keyboard hooked to Belkin USB KVM 2-port switch to control which machine has Mouse & Keyboard control

Steps to reproduce:
1.  Log into both machines
2.  Open OpenOffice (any one will do)
3.  Use KVM switch to change keyboard / mouse ownership to other machine
4.  If crash doesn't occur, switch back and it will crash.

I can reproduce 95% of the time. I can reproduce on either machine (i386 or x86_64).  Both systems have NVIDIA Quadros, but different models requiring different nvidia drivers -- but I have the latest for each.

...the xorg-x11-server-Xorg-1.5.3-11 has *not* disabled writing the geometry, I guess? Today after more than month without any problems when living with the build you pointed in [1], I've been hit again, probably because of the recent update to -11:

Backtrace:
0: /usr/bin/X(xorg_backtrace+0x26) [0x4e7b66]
1: /usr/bin/X(xf86SigHandler+0x39) [0x47a679]
2: /lib64/libc.so.6 [0x3d7aa32f90]
3: /usr/bin/X(XkbWriteXKBGeometry+0x14c) [0x55880c]
4: /usr/bin/X(XkbWriteXKBKeymapForNames+0x5bb) [0x557d9b]
5: /usr/bin/X(XkbDDXLoadKeymapByNames+0x18a) [0x561d3a]
6: /usr/bin/X(ProcXkbGetKbdByName+0x3ab) [0x53fbab]
7: /usr/bin/X(Dispatch+0x364) [0x4468d4]
8: /usr/bin/X(main+0x45d) [0x42cd1d]
9: /lib64/libc.so.6(__libc_start_main+0xe6) [0x3d7aa1e576]
10: /usr/bin/X [0x42c0f9]

Unfortunately the link to the koji build from [1] shows nothing now and I also couldn't find the RPMs from the given Koji task -- could you please post your packages with disabled geometry writing somewhere for the meantime:)? Thanks in advance.

[1] https://bugzilla.redhat.com/show_bug.cgi?id=479122#c5

Upstream bug: http://bugs.freedesktop.org/show_bug.cgi?id=20081

I think I may have found the issue. Can you please try the following package:
http://koji.fedoraproject.org/scratch/whot/task_1123333/


If it doesn't work, here's the same patch as in Comment #29 on top of 1.5.3-11:
http://koji.fedoraproject.org/scratch/whot/task_1122850/

*** Bug 483272 has been marked as a duplicate of this bug. ***

For some reason bodhi didn't add the comment. This bug has been fixed in xorg-x11-server-1.5.3-13.fc10 and is available from the testing repository.
 
https://admin.fedoraproject.org/updates/F10/FEDORA-2009-1763

xorg-x11-server-1.5.3-13.fc10 has been pushed to the Fedora 10 stable repository.  If problems still persist, please make note of it in this bug report.