VBoxMouse segfaults when starting up the X Server => fixed in SVN

[Miguel]

When I start KDM, the X Server segfaults and exits. The stack trace shows an entry in vboxmouse_drv.so; see the attached log file for details.

[Miguel]

X Server log

[Michael Thayer]

Which guest is this precisely?

[Miguel]

KUbuntu Intrepid, up-to-date with the package repository as of yesterday.

[Steve Dodd]

I too have just noticed this using the Ubuntu Intrepid beta (​http://www.ubuntu.com/testing/intrepid/beta). Incidentally, I had to add vboxvideo and vboxmouse to the xorg.conf by hand, as the Guest Additions installation didn't do it.

[Michael Thayer]

I will check this. By the way, vboxmouse should load automatically without being in xorg.conf. As of version 2.0.4 this will be true for vboxvideo as well.

[kingofmonopoli]

Hello Miguel, it's kind of off-topic, but: I noticed from your xorg.log that you have working DRI and AIGLX within a Linux guest on a Windows host. I am stuck with getting this to work for a Gentoo 2.6 guest under Windows XP host. Can you point me to the trick?

[Miguel]

Replying to kingofmonopoli:

Hello Miguel, it's kind of off-topic, but: I noticed from your xorg.log that you have working DRI and AIGLX within a Linux guest on a Windows host. I am stuck with getting this to work for a Gentoo 2.6 guest under Windows XP host. Can you point me to the trick?

I didn't do anything in particular aimed at getting that working, so I'm not sure if there is a trick I can tell you about. Actually, are you sure it's working for me? The log says:

(II) AIGLX: Screen 0 is not DRI capable

[kingofmonopoli]

Thanks, meanwhile I figured out, that gentoo uses an old versione of X and mesa in the stable tree, which does not feature software rendering. Switched X to the new version, it works. Sorry for the noise.

[Miguel]

Just FYI, this still happens in 2.0.4.

[Michael Thayer]

Sorry for not responding to this any further. Could you try installing the debug symbols for the Xorg package to get a more detailed backtrace? The Ubuntu debug repository is

deb http://ddebs.ubuntu.com/ intrepid main restricted universe multiverse

[Miguel]

I've just spent half an hour trying to get those backtraces, to no avail.

The problem is that I can only reproduce the bug when X is started from kdm (but not from the command line), and the bug crashes X instantly after it's started. I haven't managed to configure kdm to start the X server under gdb, and I can't attach gdb to the running X process from another terminal because it dies far too quickly.

Any ideas?

[Michael Thayer]

Did you install the debug symbols package? I hope that the backtrace in the Xorg log file will contain more information when it is installed.

[Miguel]

I did install the debug symbols packages and I couldn't see any more information in the log files. I'll re-check later today just to be sure.

[Miguel]

Yep, no change in the backtrace. I think that backtrace is generated by some code in X that presumably doesn't know about the symbols installed by the symbols packages. That's why I wanted to use gdb.

[Miguel]

Still seeing the problem in 2.0.6.

[Miguel]

And still seeing it unchanged in 2.1.0. By the way, as I said in an earlier post I only experience the problem when starting KDE from kdm, so I've been working around it by starting KDE from the command line instead.

[Michael Thayer]

Have you looked at the script on this page:

​http://www.x.org/wiki/Development/Documentation/ServerDebugging

That script has helped me quite a few times when I needed to get backtraces from the X server.

[Miguel]

Now I'm hitting the gdb issue described in ​ticket 2977, so no luck. I'll try again when that issue is fixed.

[Michael Thayer]

Perhaps it will work if you enable hardware virtualisation for that VM?

[Miguel]

Yes, enabling hardware virtualisation worked, and I managed to get a backtrace from gdb. Here it is:

#0  0x0809db3c in GetPointerEvents (events=0x8953758, pDev=0x8954010, type=6, 
    buttons=0, flags=<value optimized out>, first_valuator=0, num_valuators=2, 
    valuators=0x81e0480) at ../../dix/getevents.c:681
#1  0x080da8bc in xf86PostMotionEventP (device=0x8954010, is_absolute=1, 
    first_valuator=0, num_valuators=2, valuators=0x81e0480)
    at ../../../../hw/xfree86/common/xf86Xinput.c:591
#2  0x080daa48 in xf86PostMotionEvent (device=0x8954010, is_absolute=1, 
    first_valuator=0, num_valuators=2)
    at ../../../../hw/xfree86/common/xf86Xinput.c:542
#3  0xb7a03177 in ?? () from /usr/lib/xorg/modules/input//vboxmouse_drv.so
#4  0x080c3207 in xf86SigioReadInput (fd=12, closure=0x8953968)
    at ../../../../hw/xfree86/common/xf86Events.c:517
#5  0x080b2d7c in xf86SIGIO (sig=29)
    at ../../../../../hw/xfree86/os-support/linux/../shared/sigio.c:114
#6  <signal handler called>
#7  0xb7f99424 in __kernel_vsyscall ()
#8  0xb7c2adc0 in nanosleep () from /lib/tls/i686/cmov/libc.so.6
#9  0xb7c6859c in usleep () from /lib/tls/i686/cmov/libc.so.6
#10 0xb79f6846 in ps2SendPacket (pInfo=0x8953a48, 
    bytes=0xbfb99c6a "ÿ\bô\237\237·\003", len=1) at ../../src/pnp.c:604
#11 0xb79f6910 in ps2Reset (pInfo=0x8953a48) at ../../src/pnp.c:684
#12 0xb79f1e38 in SetupMouse (pInfo=0x8953a48) at ../../src/mouse.c:2915
#13 0xb79f64b2 in MouseProc (device=0x8954170, what=1)
    at ../../src/mouse.c:1771
#14 0x0808564d in EnableDevice (dev=0x8954170) at ../../dix/devices.c:233
#15 0x080857bb in InitAndStartDevices () at ../../dix/devices.c:525
#16 0x08071bf0 in main (argc=8, argv=0xbfb99fc4, envp=Cannot access memory at address 0x8
) at ../../dix/main.c:385

[Michael Thayer]

Thanks for the backtrace, that gives me a bit more of an idea where the issue might be. I've also been able to reproduce this myself now by explicitly setting VBoxMouse as the core pointer in xorg.conf.

[Michael Thayer]

This should now be fixed/worked around in the development sources. It seems to be a bug in the X server which causes a crash if a driver for an absolute pointing device is loaded from xorg.conf (unusual with server version 1.5 which supports device autodetection) and the device is moved immediately. Just for your information, you should not need to load vboxmouse from xorg.conf either - it should be autoloaded.

[Miguel]

Thanks. I had added Option "AutoAddDevices" "No" to my xorg.conf a while ago to work around some other problem (can't remember which any more), so I had to load VBoxMouse explicitly. I've now removed the AutoAddDevices and the explicit VBoxMouse load and everything seems to work fine.

[Miguel]

Oh, I found the problem I was working around now. If I don't disable AutoAddDevices, my British keyboard layout isn't recognised, and I get what seems like an American layout instead.