Disconnecting USB -> Graphic Lock

[Ren H?ek]

Hi,
I have some strange behaviour/bug with my graphic card and hope someone can give a hint...
The specs:
Mobility Radeon 9700 running on
Ubuntu 10.04 with the default
radeon driver (no Errors in /var/log/Xorg.0.log).

As long as I have a USB Mouse plugged in, everything goes well.
A few moments after unplugging, the xserver freezes.
In the background, the system keeps running, it is possible to switch to console.

The syslog says:

May  3 21:42:34 ***** kernel: [  257.824099] usb 2-1: USB disconnect, address 2
May  3 21:42:43 ***** kernel: [  266.376083] [drm:radeon_fence_wait] *ERROR* fence(f2da32e0:0x000013D7) 504ms timeout going to reset GPU
May  3 21:42:43 ***** kernel: [  266.377489] [drm] CP reset succeed (RBBM_STATUS=0x00000140)
May  3 21:42:43 ***** kernel: [  266.377498] [drm] radeon: cp idle (0x10000000)
May  3 21:42:43 ***** kernel: [  266.377545] [drm] radeon: ring at 0x00000000E0000000
May  3 21:42:43 ***** kernel: [  266.544429] [drm:r100_ring_test] *ERROR* radeon: ring test failed (sracth(0x15E4)=0xCAFEDEAD)
May  3 21:42:43 ***** kernel: [  266.544439] [drm:r100_cp_init] *ERROR* radeon: cp isn't working (-22).
May  3 21:42:43 ***** kernel: [  266.544449] [drm:r300_gpu_reset] *ERROR* Failed to reset GPU (RBBM_STATUS=0x80010140)
May  3 21:42:43 ***** kernel: [  266.544461] [drm:radeon_fence_wait] *ERROR* fence(f2da32e0:0x000013D7) 680ms timeout
May  3 21:42:43 ***** kernel: [  266.544470] [drm:radeon_fence_wait] *ERROR* last signaled fence(0x000013D7)
May  3 21:42:43 ***** kernel: [  266.546160] [drm:radeon_ib_schedule] *EMay  3 21:42:34 ***** kernel: [  257.824099] usb 2-1: USB disconnect, address 2
May  3 21:42:43 ***** kernel: [  266.376083] [drm:radeon_fence_wait] *ERROR* fence(f2da32e0:0x000013D7) 504ms timeout going to reset GPU
May  3 21:42:43 ***** kernel: [  266.377489] [drm] CP reset succeed (RBBM_STATUS=0x00000140)
May  3 21:42:43 ***** kernel: [  266.377498] [drm] radeon: cp idle (0x10000000)
May  3 21:42:43 ***** kernel: [  266.377545] [drm] radeon: ring at 0x00000000E0000000
May  3 21:42:43 ***** kernel: [  266.544429] [drm:r100_ring_test] *ERROR* radeon: ring test failed (sracth(0x15E4)=0xCAFEDEAD)
May  3 21:42:43 ***** kernel: [  266.544439] [drm:r100_cp_init] *ERROR* radeon: cp isn't working (-22).
May  3 21:42:43 ***** kernel: [  266.544449] [drm:r300_gpu_reset] *ERROR* Failed to reset GPU (RBBM_STATUS=0x80010140)
May  3 21:42:43 ***** kernel: [  266.544461] [drm:radeon_fence_wait] *ERROR* fence(f2da32e0:0x000013D7) 680ms timeout
May  3 21:42:43 ***** kernel: [  266.544470] [drm:radeon_fence_wait] *ERROR* last signaled fence(0x000013D7)
May  3 21:42:43 ***** kernel: [  266.546160] [drm:radeon_ib_schedule] *ERROR* radeon: couldn't schedule IB(6).
May  3 21:42:43 ***** kernel: [  266.546163] [drm:radeon_cs_ioctl] *ERROR* Faild to schedule IB !RROR* radeon: couldn't schedule IB(6).
May  3 21:42:43 ***** kernel: [  266.546163] [drm:radeon_cs_ioctl] *ERROR* Faild to schedule IB !
[..]

and after a while keeps going with the "Faild to schedule IB" messages.
More detailed syslog:

http://nopaste.info/7b559c4cb3.html

Is there something I could try?

[agd5f]

This is a non-drm kernel issue:

https://bugs.freedesktop.org/show_bug.cgi?id=12934

Try booting with nohz=off or irqpoll

[FunkyRider]

but isn't nohz=off turns off tickless scheduler hence worsen power consumption? not a long term solution

[agd5f]

You'll have to ask the kernel folks, perhaps not all motherboards supports tickless properly.

[Ren H?ek]

This is a non-drm kernel issue:

https://bugs.freedesktop.org/show_bug.cgi?id=12934

Try booting with nohz=off or irqpoll

Thank you.

nohz=off helps, no frozen desktop any more...
...maybe not optimal, as the system is a laptop, but referring to
"The Impact Of A Tickless Kernel"
it doesn't seem to make such a big difference in power consumption.

irqpoll doesn't solve the problem.