Comment # 5 on bug 1126480 from Takashi Iwai

OK, that's good to hear.  Yes, nouveau spews too much scaring errors.  I'm
going to suppress the kernel stack trace in that timeout error, which is simply
annoying.

The problem seems like that nouveau driver can't deal with the runtime PM
properly for the hybrid GPU (maybe only some models).  In the earlier code, it
tries to get from the loop via timeout reading the GPU timer register, but it
didn't work because the GPU itself stalled so the GPU timer wasn't updated.

That said, the stall should have been worked around.  But it's still more or
less broken.

I'm still interested in whether this problem appears with the latest upstream. 
I guess yes.  Could you give it a try?