On Tue, Apr 07, 2009 at 09:12:17AM +0200, Martin Olsson wrote: > Bryce Harrington wrote: > > hang / freeze / lockup are all typically used as synonyms. I think I'd > > prefer 'freeze' over hang. "lockup" would imply a GPU lockup - which is > > a common bug but not all freezes are the GPU. > > > Usually I can make _some_ useful progress on hang bugs where X.org goes > into a CPU spin (using gdb) but I find that much harder for GPU lockups where > X.org is basically just stuck blocking on some ioctl() waiting for the > GPU to finish something that is never going to finish. I have no idea how > to debug the gpu hangs really and if there is also no clear repro steps > and thing in xorg.log / dmesg usuallt I just curse a little bit and don't > even upstream the bug because at least right now upstream have many unfixed > bugs _with_ repros steps and/or good error messages. However, I also think > that upstream devs have done some work to improve this situation, more > specifically they added a new /proc/dri/0/ entry and then they wrote a > script that allows you to dump out the whole ring buffers on the card as > oppsed to just the GPU registers which we already have a script for. > So hopefully when the karmic kernel lands we can get a better way to > debug gpu hang bugs.
That sounds like good news. If we had a good way to debug these gpu lockups we could really make some good progress towards stabilizing things. Wish we had that stuff in there now. Bryce -- Ubuntu-x mailing list [email protected] Modify settings or unsubscribe at: https://lists.ubuntu.com/mailman/listinfo/ubuntu-x
