Re: 2.6.14 X spinning in the kernel

[Dave Airlie <airlied@linux.ie>]

> > Yes invariably the GPU has crashed and isn't responding to anything.
>
> But radeon_do_wait_for_idle() and radeon_do_wait_for_fifo() have timeouts.
> Should Badari have waited longer?

They timeout, and X usually goes straight back in there again, I can't
remember the codepath exactly at this stage but it just goes round and
round until you kick the machine..

in theory X should probably deal with the situation better... I think it
might be able to at least gracefully die or reset the chip...

> > Also what X was doing etc at the time is invalulable info..
> >
>
> And whether a particualr kernel version introduced this behaviour.

Yes, usually if a kernel introduced it it is because I've done something
really dumb (shouldn't happen too often and with radeons we usually catch
that before stable releases), or because the X server wasn't using DRI
before due to a too old DRM and suddenly the new DRM appears in the kernel
and it uses it ...

There is one known issue with some later version of X on radeons crashing
on PCI GART setups, benh was cooking a patch for X, it isn't something we
can fix in the kernel..

Dave.

-- 
David Airlie, Software Engineer
http://www.skynet.ie/~airlied / airlied at skynet.ie
Linux kernel - DRI, VAX / pam_smb / ILUG