Created attachment 44527 [details]
dmesg log, 2.6.38 with commit 73412c3854c877e5f37ad944ee8977addde4d35a reverted
There's a regression introduced in kernel >=2.6.38-rc8. It's in nouveau/master (commit 7972f91c24bb66643d76cf635faea78eb184e6b6) as well. Dmesg shows: "GPU lockup - switching to software fbcon". I've done a git bisect. The problematic commit is: 73412c3854c877e5f37ad944ee8977addde4d35a "drm/nouveau: allocate kernel's notifier object at end of block". I reverted that commit on top of the 2.6.38 to confirm the issue.
Created attachment 44528 [details]
dmesg log, 2.6.38-rc7, no GPU lockup information
Created attachment 44529 [details]
dmesg log, 2.6.38-rc8, GPU lockup
Created attachment 44589 [details]
As suggested, dmesg log 2.6.38 with drm.debug=14 3
I can confirm a similiar bug. But my system completely locks up, that's why I can't provide you with a proper dmesg log (I could make a picture though).
I also did a bisect and ended up with the same commit 73412c3854c877e5f37ad944ee8977addde4d35a. I got an AGP GeForce 7600 GT (NV40 - not NV50!).
Do you require further information?
Nevermind, this bug has been fixed by a18d89ca026140eb8ac4459bf70a01c571dd9a32 (2.6.39-rc2+).
*** Bug 36248 has been marked as a duplicate of this bug. ***
*** Bug 36330 has been marked as a duplicate of this bug. ***