"kernel BUG" when stopping xdm? (only with ati-drivers)

[EricHsu]

guys

I've been experiencing the same problem with kernel > 2.6.14.

I got it solved the day before yesterday by searching through the forums.

By doing some tracing job, I guess this problem is caused by the using the fglrx's internal agpgart support. it seems that fglrx's internal agpgart dislike kernel > 2.6.14, it crashes the system and causes the pageattr.c error we've been seeing in the system log.

so, here is the solution - use external agpgart support.

steps:

1. compile your kernel with a proper agpgart support according to your chipset:

# make menuconfig

Code: Select all

Device Drivers -> Character devices -> /dev/agpgart (AGP support) 

please DO choose the right chipset support, I used to thought my ati card should be using "ATI chipset support", but, I was wrong, my chipset is Intel 855PM, so I have to choose

Code: Select all

 <M>   Intel 440LX/BX/GX, I8xx and E7x05 chipset support 

you can find out your chipset by /usr/sbin/lspci. then:

recompile your kernel and reboot

Code: Select all

# make && make modules_install

2. config your xorg.conf to use the external agpgart support rather than fglrx's built-in agpgart support:

add the UserInternalAGPGART option to xorg.conf

Code: Select all

Section "Device"
        Identifier  "ATI Graphics Adapter 0"
        Driver      "fglrx"
        Option "UseInternalAGPGART" "no"
...

let me know if this works for you : )

note: using external agpgart does not decrease x's rendering performance, at least I'm having the same glxgears output w/o fglrx's internal agpgart support - yes I know glxgears is not for performance testing...

refs:

- http://forums.gentoo.org/viewtopic-t-42 ... ttr+c.html
- http://forums.gentoo.org/viewtopic-t-44 ... eattr.html

edit:

I've done some testing again, if you have built the right external agpgart module, no matter commenting out the "UseInternal..." line OR setting it to "yes", fglrx will use the external agpgart module, dmesg:

Code: Select all

[fglrx] Internal AGP support requested, but kernel AGP support active.
[fglrx] Have to use kernel AGP support to avoid conflicts.

if you do not build any agpgart chipset support or build a wrong one as I did, fglrx will use its internal agpgart support which will crash your x as mentioned.

so just build the right agpgart chipset support while compiling kernel. then, you may add "UseInternal..." line in the xorg.conf to tell x to use external agpgart or just let fglrx auto detect it for you

[EricHsu]

Here's a question for you guys. If you get the kernel crash when you reboot/poweroff, how are you rebooting/poweroffing?

use the magical SysRq key

1. compile your kernel with sysrq key support: make menuconfig, go to "Kernel hacking":

Code: Select all

[*] Magic SysRq key

2. when system crashes

Code: Select all

 - Alt + SysRq + S: "S"ync data from mem to disk
 - Alt + SysRq + U: "U"mount partitions
 - Alt + SysRq + B: re"B"oot the system

note: if you're using laptop, maybe you'll have to press one more key "Fn" like me: "Fn + Alt + SysRq + <command>" - since my SysRq key is the same key with "Delete" in the keyboard.

HTH

[-fenice-]

This Option "UseInternalAGPGART" "no" thing correct the bug of logout crash... But it causes a new problem on my laptop : OpenGL doesn't work anymore...

[EricHsu]

This Option "UseInternalAGPGART" "no" thing correct the bug of logout crash... But it causes a new problem on my laptop : OpenGL doesn't work anymore...

did you mean the dri not working anymore?

$ glxinfo
name of display: :0.0
display: :0 screen: 0
direct rendering: No

could you post your lspci result & the kernel config of the agpgart

Code: Select all

Device Drivers -> Character devices -> /dev/agpgart (AGP support)
(paste the chipset support you've selected)

I experienced that too, since I built-in a wrong chipset support, then dri didn't work if set "UseInternal..." to "no".

[Spiffster]

I'm experiencing this on a laptop with an Radeon 9700 Mobility (as I wrote before), and the hiroki and pmouse seem to have a Mobility 9600. So at least it isn't one specific model, but it seems (from all who have postet which cards they use) that it affects the 9x00 series.

I can confirm this problem also exists on my laptop which has a Radeon 9700 Mobility. I know the 9700 mobility is based on the Radeon 9600 core (not 9700), so perhaps the problem is related to the 9600 core?

For me it means I cannot log out of X, but if i choose to shutdown, the system shuts down cleanly, although the screen turns black when X shuts down. Also, it still reacts to ctrl-alt-del or the power button (acpid shuts it down cleanly). But i cannot make changes to my X configuration and restart X, or even log out and log in again.

[Spiffster]

As suggested in this thread, I have tried to switch back to the kernel AGP driver (compiled Intel 440LX/BX/GX as module and set "UseInternalAGPGART" "no" in /etc/X11/xorg.conf).

This has partly solved the problem. I can now select shutdown from the Gnome logout menu, and actually get some console output while shutting down (instead of a black screen). Switching to VT1 while X is running (and switching back to X) works, but this has also worked previously. However, I still get the kernel panic if i try to run

Code: Select all

/etc/init.d/xdm stop

from the console. This also happens if i try to shut down the machine from the console (as opposed to from the Gnome logout menu).

I use xorg-x11-6.8.2-r7, kernel compiled from gentoo-sources-2.6.16-r7 and ati-drivers-8.24.8. The machine is an Acer Travelmate 8004LMi which has an ATI Radeon 9700 Mobility. Any suggestions for further things I could try?

[mathfeel]

Has any one fixed this problem yet?

Requiring gdm to restart X and/or switching on/off internalAGP for fglrx does not solve the problem for me. I still can't switch between tty and X and kernel panic whenever /etc/init.d/xdm stop

gentoo-sources-2.6.16-r17
ati-drivers-8.24.8
xorg-x11-7.0-r1

[ponzio]

also for me, nothing solves the problem...
and i experience a kernel panic also using mplayer!!!
ati-drivers-8.24.8
kernel 2.6.16-gentoo-r1

[Jkay]

Switching to internal kernel AGP, and altering Xorg.conf, and then of course re-installing the ATI proprietary drivers worked perfectly for me.

Make sure you are 100% certain that you chose the correct mobo chipset in the AGP drivers section of the kernel menuconfig

[mathfeel]

Switching to internal kernel AGP, and altering Xorg.conf, and then of course re-installing the ATI proprietary drivers worked perfectly for me.

Make sure you are 100% certain that you chose the correct mobo chipset in the AGP drivers section of the kernel menuconfig

This is the confusing point. Is the setting to enable internal kernel AGP

Code: Select all

Option "UseInternalAGPGART" "no" 

This is my current setting.

I am using intel_agp, pretty sure that's correct. I have an IBM ThinkPad T42.

[BlueShark]

Hey guys,

I had this problem for a long time and finally I found this thread. I followed the describtions in this thread to change my xorg.conf, my kernel config, etc. in the hope that all problems caused by the ati-drivers would disapear. But I received still a kernel panic by closing X. However I realized that the kernel panic appears no longer this often. By taking a closer look to that, I made a really weird notice. Now I am 90% sure, that X crashes during shutdown on my system, when I do not close all programms by hand, which I started by hand.

For example I start evolution and close evolution. When I do a ps -ax then I see that evolution-data-server and evolution alarm-modify is still running. Without killing these procceses X will crash. Same thing if I start any init script with '/etc/init.d/any start' after xdm start has started. Then I have to stop that before X is stopped.

Since I began following these rules I've never had those problems. But I have no clue why.

[ninjaboy13]

I had the same problem, the only way I could fix it was to drop down my kernel to 2.6.14-r5 and then reinstall my ati-drivers. I am using 8.25. Hope that helps.

[BlueShark]

After a new driver release from ATI I installed the new ones ( version 8.25.18 ). Since I installed those drivers, all my problems with the ati-drivers seems to be gone.