Re: + drivers-gpu-drm-i915-intel_lvdsc-fix-locking-snafu.patch added to -mm tree

[Daniel Vetter <daniel@ffwll.ch>]

On Mon, Feb 02, 2009 at 10:44:48PM -0800, Andrew Morton wrote:
> > Enabling kms works flawlessly now, but when I fire up X, the screen blanks
> > (no more blinking cursors), then X hangs. vt-switchings doesn't work
> > anymore, otherwise the machine looked fine (ping on the network was fine,
> > couldn't check anything else for lack of a running sshd on the crashing
> > machine). Twice using SysRq-T (half a minute in between) showed that Xorg
> > was indeed stuck, both times with the exact same backtrace:
> > 
> > Xorg          D 00203246  6448  6049   6048
> >  f1c81df0 00203046 f6322720 00203246 f1c81de0 f83fb98d f6322720 f632297c
> >  00203046 f1d88444 00203046 f2f63cc0 f8388dae f1d88444 ffffffff f1d88408
> >  00203246 f1c81e2c c02e18ba f83fb98d 00000000 f6322720 f1d88430 f1d88444
> > Call Trace:
> >  [<f83fb98d>] ? intel_lvds_get_modes+0x69/0x94 [i915]
> >  [<f8388dae>] ? drm_mode_getconnector+0x54/0x31f [drm]
> >  [<c02e18ba>] mutex_lock_nested+0x158/0x254
> >  [<f83fb98d>] ? intel_lvds_get_modes+0x69/0x94 [i915]
> >  [<f83fb98d>] intel_lvds_get_modes+0x69/0x94 [i915]
> >  [<f838a170>] drm_helper_probe_single_connector_modes+0xb8/0x194 [drm]
> >  [<f8388e20>] drm_mode_getconnector+0xc6/0x31f [drm]
> >  [<c02e13bc>] ? mutex_unlock+0xd/0xf
> >  [<f837f71f>] drm_ioctl+0x1c1/0x23d [drm]
> >  [<f8388d5a>] ? drm_mode_getconnector+0x0/0x31f [drm]
> >  [<f837f55e>] ? drm_ioctl+0x0/0x23d [drm]
> >  [<c0191277>] vfs_ioctl+0x43/0x56
> >  [<c0191a34>] do_vfs_ioctl+0x49f/0x4e0
> >  [<c0186c42>] ? vfs_write+0xf5/0x131
> >  [<c0191aba>] sys_ioctl+0x45/0x5f
> >  [<c0102e91>] sysenter_do_call+0x12/0x31
> > 
> > This is on 2.6.29-rc3-00100-gf2257b7.
> > 
> 
> So I assume that it would make sense to track this as a post-2.6.28
> regression?

Nope. Like the previous issue it only happens when I run-time enable
kernel modesetting (with # modprobe i915 modeset=1), which is not the
default.

I'm gonna open a new bz entry with all the details (and all the people on cc
minus regression handlers). But first I'll check with CONFIG_LOCKDEP
whether it's really a locking goof-up.

-Daniel

-- 
Daniel Vetter
E-Mail: daniel.vetter@ffwll.ch
Tel.: +41 (0)79 365 57 48