[PATCH] drm/amd/display: Add fast path for cursor plane updates

[Michel Dänzer]

On 2018-12-05 8:59 p.m., Nicholas Kazlauskas wrote:
> [Why]
> Legacy cursor plane updates from drm helpers go through the full
> atomic codepath. A high volume of cursor updates through this slow
> code path can cause subsequent page-flips to skip vblank intervals
> since each individual update is slow.
> 
> This problem is particularly noticeable for the compton compositor.
> 
> [How]
> A fast path for cursor plane updates is added by using DRM asynchronous
> commit support provided by async_check and async_update. These don't do
> a full state/flip_done dependency stall and they don't block other
> commit work.
> 
> However, DC still expects itself to be single-threaded for anything
> that can issue register writes. Screen corruption or hangs can occur
> if write sequences overlap. Every call that potentially perform
> register writes needs to be guarded for asynchronous updates to work.
> The dc_lock mutex was added for this.
> 
> Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=106175
> 
> Cc: Leo Li <sunpeng.li at amd.com>
> Cc: Harry Wentland <harry.wentland at amd.com>
> Signed-off-by: Nicholas Kazlauskas <nicholas.kazlauskas at amd.com>

Looks like this change introduced (or at least exposed) a reference
counting bug resulting in use-after-free when Xorg shuts down[0]. See
the attached dmesg excerpt (note that I wrapped the !bo->pin_count check
in amdgpu_bo_unpin in WARN_ON_ONCE).


[0] Only with
https://gitlab.freedesktop.org/xorg/driver/xf86-video-amdgpu/commit/0d60233d26ec70d4e1faa343b438e33829c6d5e4
, i.e. alternating between two BOs for the HW cursor, instead of always
using the same one.


-- 
Earthling Michel Dänzer               |               http://www.amd.com
Libre software enthusiast             |             Mesa and X developer
-------------- next part --------------
A non-text attachment was scrubbed...
Name: kern.log
Type: text/x-log
Size: 41935 bytes
Desc: not available
URL: <https://lists.freedesktop.org/archives/amd-gfx/attachments/20181213/b1aa1372/attachment-0001.bin>