nouveau X lockup on 36-rc (regression from 2.6.35)

[Johannes Berg]

Since 2.6.36-rc kernels (both rc1 and rc2) I've had X lockups where X
sits in a loop just using CPU.

The only kernel message I got this time was this one:

[22290.792075] [drm] nouveau 0000:02:00.0: PFIFO_DMA_PUSHER - Ch 2

stracing X just repeats, in a loop,

ioctl(9, 0x40086482, 0x7fff6a873af0)    = ? ERESTARTSYS (To be restarted)
--- SIGALRM (Alarm clock) @ 0 (0) ---
rt_sigreturn(0xe)                       = -1 EINTR (Interrupted system call)


so apparently that ioctl never returns.

This has never happened on 2.6.35, and I've gone back now. Large screen
updates seem to trigger it more likely, but I haven't found a way to
really reproduce it. FWIW, I'm running with two screens connected.

johannes