[drm] BUG: spinlock lockup suspected on CPU#0, swapper/0/1

[David Herrmann]

Hi

On Tue, Jan 14, 2014 at 2:45 PM, Fengguang Wu <fengguang.wu at intel.com> wrote:
> Hi David,
>
> I'm not sure if this is a good bisect because the errors are noisy,
> however it's good to inform you of this possibility.
>
> First bad commit may be a3483353c ("drm: check for !kdev in drm_unplug_minor()")

To be honest, I doubt that it's somehow related to this commit. We
replace "!device_is_registered(minor->kdev)" with a "!minor->kdev".
Both of these expressions have no side-effects (especially the latter
one) and they only evaluate to true during DRM unplugging. Having said
that, your log also contains this:

[   12.058502] [drm] Initialized drm 1.1.0 20060810
[   12.059918] [drm] radeon kernel modesetting enabled.
[   12.061791] [drm:cirrus_vram_init] *ERROR* can't reserve VRAM
[   12.063409] cirrus 0000:00:02.0: Fatal error during GPU init: -6
[   12.064914] Trying to free nonexistent resource <00000000febf0000-000000=
00febf0fff>

The DRM cirrus driver fails to load and given the error message I
suspect the error-path is not clean. Dave Airlie should know more
about cirrus, but I really think this is related to cirrus not
correctly cleaning up it's resources. Dave, any ideas? Disbling
CONFIG_CIRRUS should be a good workaround, in case anyone cares.

The commit in question highly reduces the required stack-space during
DRM unload (which includes the cirrus error-path) so I think the
bisect was alright, but the commit is not the actual culprit. It just
reveals some stack-corruption, I guess.

Thanks
David