On Wed, 21 Jan 2009, Linus Torvalds wrote: > > I was going to say that I've been VT switching and it has worked, but I > decided to double-check. And yes, I can reproduce it that way. And no, no > kernel messages that are visible that way either. Ahhah! No kernel messages, because the kernel isn't actually dead. Just X is. I verified that by doing a while : ; do echo t > /proc/sys/sysrq-trigger; sleep 5; done while doing these things. I'm travelling, it's 7AM in Hobart now, and I don't have another machine, so when X dies in graphics mode, there's not a whole lot else I can do to debug it. No other machine to use to ssh into the machine from etc.. Also, the reason I at one point thought that VT switching works is that quite often it _does_ work. I seem to need some timing or interrupt thing happening to trigger the problem, because when I wasn't doing anything else, I could VT switch back and forth something like 10 times without problems. That probably also explains why it then happens all the time on suspend/resume: other things are most definitely happening. So I did another shell loop (that also gives an audible clue that things are working, even if the machine appears dead): while : ; do cat /bin/echo ; sleep 5; done > /dev/audio and the VT switch failed the very next time I tried it. And since I had the sysrq loop running, and since it still made progress, I can now say that X seems to hang here: Xorg S f52a3b98 6088 2252 2251 f68db340 00003082 c06ea2f8 f52a3b98 c07bb73c c07bec40 c07bec40 c07bec4 c07bec40 f4a58980 f4a58be8 c2010c40 00000000 00000000 bfc1f838 0000020 bfc1f838 f4a58be8 fffe9b5a 000a037f 00003282 c022d7ff f514fe98 f4a58c5 Call Trace: [<c022d7ff>] lock_timer_base+0x19/0x35 [<c022d983>] __mod_timer+0xba/0xc3 [<c05a65ef>] schedule_timeout+0x9f/0xbb [<c022d522>] process_timeout+0x0/0x5 [<c05a65ea>] schedule_timeout+0x9a/0xbb [<c038c494>] drm_wait_vblank+0x2d8/0x396 [<c0221c4c>] default_wake_function+0x0/0x8 [<c038a5ab>] drm_ioctl+0x1a6/0x21e [<c038c1bc>] drm_wait_vblank+0x0/0x396 [<c0288d5e>] vfs_ioctl+0x49/0x5f [<c0289527>] do_vfs_ioctl+0x464/0x49f [<c02895a3>] sys_ioctl+0x41/0x58 [<c0202d01>] sysenter_do_call+0x12/0x21 ie it seems to be some vblank race. Looks like your commit dc1336ff4fe08ae7cfe8301bfd7f0b2cfd31d20a wasn't a complete solution? Because it's definitely something I have, and I have compiz, and I see the compiz hangs at VT switch that it claims to have fixed. Added the other people that the git logs say worked on the vblank code to the Cc. What an odd way to debug X lockups. What do you guys usually do? Just always make sure to have another machine handy? Linus -- To unsubscribe from this list: send the line "unsubscribe linux-pci" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
