[PATCH] drm/radeon: Fix negative cursor position

[Michel Dänzer]

On 26/09/16 09:09 PM, Takashi Iwai wrote:
> On Mon, 26 Sep 2016 11:42:16 +0200, Takashi Iwai wrote:
>> On Mon, 26 Sep 2016 10:57:50 +0200, Michel D4nzer wrote:
>>> On 23/09/16 10:06 PM, Takashi Iwai wrote:
>>>> radeon_cursor_move_unlock() contains a workaround for AVIVO chips that
>>>> are older than DCE6 when the cursor ends on 128 pixel boundary.  It
>>>> decreases the position when the calculated end position is on 128
>>>> pixel boundary.  However, it hits also the condition where x=-1 and
>>>> width=1 are passed, since cursor_end is 0 (which is aligned with 128,
>>>> too).  Then the value gets decreased and it hits WARN_ON_ONCE()
>>>> below, eventually screws up the GPU.
>>>>
>>>> The fix is simply to skip the workaround when x is already zero.
>>>> Also, remove the superfluous WARN_ON_ON() for the negative value check
>>>> that wouldn't happen any longer after this change.
>>>>
>>>> Bugzilla: https://bugzilla.suse.com/show_bug.cgi?id=1000433
>>>> Cc: <stable at vger.kernel.org>
>>>> Signed-off-by: Takashi Iwai <tiwai at suse.de>
>>>> ---
>>>>  drivers/gpu/drm/radeon/radeon_cursor.c | 4 +---
>>>>  1 file changed, 1 insertion(+), 3 deletions(-)
>>>>
>>>> diff --git a/drivers/gpu/drm/radeon/radeon_cursor.c b/drivers/gpu/drm/radeon/radeon_cursor.c
>>>> index 2a10e24b34b1..4e436eb49a56 100644
>>>> --- a/drivers/gpu/drm/radeon/radeon_cursor.c
>>>> +++ b/drivers/gpu/drm/radeon/radeon_cursor.c
>>>> @@ -193,10 +193,8 @@ static int radeon_cursor_move_locked(struct drm_crtc *crtc, int x, int y)
>>>>  			if (w <= 0) {
>>>>  				w = 1;
>>>>  				cursor_end = x - xorigin + w;
>>>> -				if (!(cursor_end & 0x7f)) {
>>>> +				if (x > 0 && !(cursor_end & 0x7f))
>>>>  					x--;
>>>> -					WARN_ON_ONCE(x < 0);
>>>> -				}
>>>>  			}
>>>
>>> The problem with this change is that the horizontal cursor end position
>>> can again and up as a multiple of 128, which this code is trying to
>>> avoid, because it can break the cursor with some versions of AVIVO / DCE
>>> display engines.
>>>
>>> Attached is a proof-of-concept (only compile tested) alternative fix
>>> which hides the cursor while it's out of bounds. I'll test it, clean it
>>> up and split it up into multiple patches when I get a chance, but maybe
>>> you or the reporter of the referenced bug can test it in the meantime.
>>
>> Sure, I'll ask the reporter to test with your tentative fix.
> 
> Through a quick test by Tomas (added to Cc), the new patch seems
> working.

Thanks for testing, Tomáš! I sent an updated patch series to the amd-gfx
mailing list and put you guys in Cc.


-- 
Earthling Michel Dänzer               |               http://www.amd.com
Libre software enthusiast             |             Mesa and X developer