gdm/xdm/kdm crash if running two card setup

[David Klasinc]

Binary package hint: xorg

Until now I had only Nvidia 8800GTS graphics card. I used two monitors with Xinerama enabled and one desktop stretched across both screens. Now I decided to add one more monitor. First I installed Nvidia 7800GT in the second PCIe slot. Both cards worked fine.

Two monitors are connected to 7800GT and one to 8800GTS.

I've created a xorg.conf with nvidia-settings. When restarting X I saw Nvidia logo on the main monitor connected to 8800GT and mouse cursor briefly appeared. All three monitors came to life from power savings mode, the other two just didn't display any picture yet. Then gdm simlpy crashed. The process repeaded itself for six times until I was thrown back in console with a notice that gdm won't start.

I started digging up and playing with various configurations. Gdm (or kdm or xdm, I tried all three) will die only if there are two screens on separate graphics cards. There was nothing in Xorg error logs that would give any idea what could be the problem. Then I decided and I ran Xorg -config /etc/X11/xorg.conf and I got xorg running on all three screens. Mouse was there and I was able to move it across all three screens. Xorg by itself was running as it should. I tried running Xorg with the same options as xdm to see if it fails, but it didn't. Only when Xorg is ran by xdm it will die, probably because xdm crashes first. Running xdm in debug nodaemon mode didn't help. I have attached my Xorg.conf and xdm logs when running in daemon mode.

Any ideas on what might be wrong will be much appreciated. Also if there are any working setups of similar configurations, I'd like to know about them.

Versions:
Ubuntu 8.10 (with regular updates)

X.Org X Server 1.5.2
Release Date: 10 October 2008
X Protocol Version 11, Revision 0
Build Operating System: Linux 2.6.24-19-server i686 Ubuntu
Current Operating System: Linux thefish 2.6.27-9-generic #1 SMP Thu Nov 20 21:57:00 UTC 2008 i686
Build Date: 24 October 2008 08:00:16AM

Nvidia Drivers 177.80

[David Klasinc (bigwhale)]

I am not sure that this is really a duplicate. My Xorg starts if I start it from console and it initializes both adapters and all three monitors. It fails only when gdm is trying to start it.

[David Klasinc (bigwhale)]

I also tried to run gdm in -nodaemon mode. This is what I got back:

Xlib: extension "RANDR" missing on display ":0.0".
gdm[14634]: Gdk-ERROR: unable to find a usable depth
aborting...

gdm[14568]: WARNING: failsafe dialog failed (inhibitions: 0 0)
Xlib: extension "RANDR" missing on display ":0.0".
gdm[14635]: Gdk-ERROR: unable to find a usable depth
aborting...

As I said before, Xorg starts, but dies because of gdm aborting.

[eko (mail4eko)]

I am also having the same issue/behaviour here.
With 2 monitors on the same GPU everything is working perfectly, as soon as you try to have any input "involving" the other GPU everything crashes.

my lspci
01:00.0 VGA compatible controller: nVidia Corporation G70 [GeForce 7800 GTX] (rev a1)
02:00.0 VGA compatible controller: nVidia Corporation GeForce 9500 GT (rev a1)

I also do not think that this a duplicate of the other bug.. the other one is just related to the fact that xorg is not starting at all. A lot of people there is happz about the fact that adding the specific PCI whatever line in the xorg.conf everything is working.... that's not our case here!
As it can be seen from the xorg.conf file the line is already there...

[eko (mail4eko)]

my Xorg.0.log

[Mark Shuttleworth (sabdfl)]

@eko, you may then want to file a different bug if you don't think it's
a dup. In the bug report, say you don't think it's a dup and say why.

Mark

[eko (mail4eko)]

I do think that this bug (314566) it has not to be considered as a duplicate of number 267241.

Because the bug 267241 is about the fact that xorg is not capable to understand on which card it should run when two gpus are installed.
This is something that I have experienced too, but it is resolved adding the Busid spcification inside the device section and the everything it is working.

/----- example starts here
    Section "Device"
 Identifier "Configured Video Device"
 Busid "PCI:1:0:0"
 Driver "nvidia"
    EndSection
/---- example ends here

What David Klansic and myself are reporting is that, when we are trying to enable xorg to run on more than gpu it crashes without any explicit motivitation.

I have tried to:
run two monitorns with one gpu->ok
rutn two monitors with other gput (after swithcing the card position on the motherboard->ok
two monitors on different gpu->crashes
three monitors on different gpus->crashes

So that's why I am not considering this bug as a dup. The issue here is about having output utilizing at the same tiem more than one gpu. The other one (bug 267241) was just about the fact that xorg was not capable to run at all, due to the missing Busid specification.

LMK.
eko

[Mal (mal-cybersanitarium)]

Hi. I was running a three-monitor setup with two nVidia cards (7950GT and 7300GS) and it worked okay, but when I upgraded to a 8600GT and swapped out the 7300GS with the 7950GT, I started having problems.

After playing around with it a bit, X actually does start but starting any X program makes it crash (and the program segfaults). xeyes and xedit still work, but anything else I've tried to start makes everything crash.

I believe I'm having the same problem here. Is this perhaps a problem when combining two separate series nVidia GPUs? (ie: 7000 series with 8000 or 9000, etc)

I'm using the Hardy version of Xorg right now because it works just fine, for some reason, while the Intrepid version does not. If needed, I can upgrade back for the purpose of providing log files from the crash.

Also, I don't think this is a duplicate of #267241. I've been through the process of setting the busid and had it working previously with two same-series nVidia GPUs.

[David Klasinc (bigwhale)]

I've noticed the same thing that Mal did. Running Xorg by itself works with no problems. Running xinit or other programs will always segfault.

I am assuming that this is because two different nvidia cards.

[David Klasinc (bigwhale)]

https://bugs.launchpad.net/ubuntu/+source/xorg/+bug/337172

Here's the new bug report.

[Timo Aaltonen (tjaalton)]

the correct action would've been to undupe this one instead of filing a new bug. Marked the new one as a dupe of this.

[David Klasinc (bigwhale)]

Mark suggested to file a new bug report thats why it was filed.

[Bryce Harrington (bryce)]

Hi bigwhale,

Could you please collect a full backtrace from when this crash occurs? Directions on collecting full backtraces are at https://wiki.ubuntu.com/X/Backtracing

[This is an automated message. Apologies if it has reached you inappropriately; please just reply to this message indicating so.]

[eko (mail4eko)]

Hi,
there is a bunch of stuff there:
   1. Debug symbol information
   2. Log in remotely
   3. Backtrace with gdb
   4. Untrap signals
   5. Debugging Error Exits
   6. Debugging Hangs / Freezes / Lockups
   7. DRI / drm problems
   8. Xorg Memory Usage
   9. Backtracing Using LiveCD
  10. Using Screen to get backtraces for Suspend/Resume crashes
  11. Obtaining the video BIOS
  12. More information

Which point should I follow in your opinion to provide the best info?

tnx,
e

[David Klasinc (bigwhale)]

Ok, I've been doing some testing today. Here are my findings:

Xorg doesn't actually crash. Other programs that require Xorg will crash. I've installed debug libraries and did some testing.

First I ran Xorg -config xorg.config.tripple X started up as it should, all three screens turned on and mouse was working. From a remote terminal I executed xeyes. Success. Eyes were following my mouse and all was well. Then I tried to run xterm. Xterm segfaulted and all three screens became dark and it seemed that Xorg crashed. It didn't, all screens came back on like nothing has happened.

 /var/log/messages indicated that xterm segfaulted:

xterm[7337]: segfault at 4 ip b7dcdf87 sp bf8d27dc error 4 in libX11.so.6.2.0[b7da6000+eb000]

I attached a file with full backtrace of xterms crash. Xterm crashed in libX11.

Then I tried running couple more programs. Xclock died horribly with a segfault in XftColorAllocValue () from /usr/lib/libXft.so.2. Backtrace included. In the end I just tried running firefox. It died with a message:

Xlib: extension "RANDR" missing on display ":0.0".

Gdk-ERROR **: unable to find a usable depth
aborting...

Program received signal SIGABRT, Aborted.

This is the same with -180 and -177 nvidia drivers. Programs simply segfault when you try running them. My setup and config files and logs are already attached.

It must me an issue with using two different nvidia cards. From a different series. I'll try to get few more cards to see if I can find a combination that would work.

[eko (mail4eko)]

David, which motherboard are you using?

[eko (mail4eko)]

I have done some bugtracking using the following command:
sudo bash -c "/usr/share/bug/xserver-xorg-core/script 3>&1" > file

as inidicate here:
http://wiki.debian.org/XStrikeForce/XserverDebugging

I tried twice:
1) with the non-working xorg.conf (2 nvidia devices-3 screens configuration)
2) with thw workking xorg.conf (1 nvidia device-2 screens configuration)

[David Klasinc (bigwhale)]

Two 8800GTS cards work as they should. No crashes here, I'm running three screens right now. So it must be some problem with combining 7000 and 8000 series cards. I couldn't get any 9000 series.

[X3K6A2 (me-x3k6a2)]

I'm runnning a 9800 GX2 (PCI-X) and a 7100 (onboard) in a three monitor setup. Two are connected to the 9800 and one to the 7100.

Same problem as above. Same things can be replicated at will (crash on xterm start and so on...). Problems started at the same time (ubuntu 8.04->8.10 update) downgrading X solved the problem.

[eko (mail4eko)]

I can confirm that downgrading to ubuntu 8.04 xorg version is fixing everything.
I followed the steps indicated here:
https://bugs.launchpad.net/ubuntu/+bug/296167/comments/50

So...this is cleary another bug of the xorg version coming with Intrepid...

[Bryce Harrington (bryce)]

Maybe check if this is a dupe of 324465, which is a common crash on -nvidia with xinerama enabled.

[David Klasinc (bigwhale)]

No, this has nothing to do with keyboard repeat. Xapps crash without any keys being pressed. No need. ;)

[eko (mail4eko)]

even worse... xorg is not starting at all!

[Bryce Harrington (bryce)]

[This is an automated message]

In Jaunty (9.04), we have just updated to the latest
nvidia-graphics-drivers-180 package from nVidia, version 180.44.

This package provides fixes for a large number of bugs, and we need your
assistance in testing if it fixes the issue you reported.

To do this, please do the following:

 a. Update to the 180.44 version of -nvidia using your favorite update
     method

 b. Attempt to reproduce your bug

 c. If your bug still remains, please simply reply to this email
     indicating so.

 d. If your bug is now solved, you can help us by setting your bug
     report to Fix Released:
     * In launchpad, go to your bug report
     * Click on the downward pointing arrow under Status
     * Set the Status field to 'Fix Released'
     * Comment on the change, such as, 'Verified fixed in 180.44'
     * Click 'Save Changes'

 e. If the original problem is solved but there are now other problems,
     please close the original bug and open new ones for those issues.

Thank you!

For details on the changes in this version of -nvidia, please see:

   http://www.nvidia.com/object/linux_display_ia32_180.44.html

[eko (mail4eko)]

I have just updated to Jaunty and... there we go again. I am crashing at start-up phase. I am starting the xorg downgrade procesdure again nad let's see.

[jmvidal (jmvidal-gmail)]

I have the same problem. Attached is my xorg.conf. When I use as is it works: I get 3 monitors each running a separate X sessions (I can move the mouse across them but I cannot move windows across screens, as, I think, is expected).

But, if I change the xorg.conf to have

Option "Xinerama" "on"

instead of "off" then I get the same "Gdk-ERROR: unable to find a usable depth
aborting..." when I try to "startx". I also noticed that if I just run "X" that it runs fine with no crash.

Does anyone have a work around?

[Bryce Harrington (bryce)]

I've posted a new version of the -nvidia driver to our xorg-edgers PPA,
would you mind testing it either on Jaunty or Karmic and see if it
resolves this bug?

Get nvidia-graphics-drivers-180 - 185.18.14 here:

  https://edge.launchpad.net/~xorg-edgers/+archive/ppa

[Bryce Harrington (bryce)]

We're closing this bug since it is has been some time with no response from the original reporter. However, if the issue still exists please feel free to reopen with the requested information. Also, if you could, please test against the latest development version of Ubuntu, since this confirms the bug is one we may be able to pass upstream for help.

[Lars Kneschke (lkneschke)]

I have the still the same problem with Karmic and 185.18.36.

Startx is working, while kdm is crashing. Using the nv driver everything is working.