Bug#504535: xserver-xorg-video-radeon: Freeze in 24bit and 16bit fullscreen (Radeon Mobility M6 LY)

[Alex Deucher]

On Fri, Feb 13, 2009 at 3:34 PM, Sylvain Beucler <beuc at beuc.net> wrote:
> On Sun, Feb 01, 2009 at 05:52:29PM +0100, Brice Goglin wrote:
>> Sylvain Beucler wrote:
>> >> Does
>> >>
>> >>    Option  "AGPMode" "1"
>> >>
>> >> (or "2") help? Please verify for your tests that the option value is
>> >> reflected in the log line above.
>> >>
>> >
>> > Apparently this helps, but the freezes are not as easily reproduceable
>> > as I thought, so I can't say for sure. I'll run some more tests.
>> >
>>
>> Any news about this?
>> Once you'll have found a AGPMode value that works better, we'll ask
>> upstream to add a quirk for your machine.
>
> Apparently 2x works best.
>
> 24bit - 4x
>
> (==) RADEON(0): Using AGP 4x
>
> Feb 13 20:26:57 kobo kernel: [ 5112.044659] agpgart: Found an AGP 2.0 compliant device at 0000:00:00.0.
> Feb 13 20:26:57 kobo kernel: [ 5112.044659] agpgart: Putting AGP V2 device at 0000:00:00.0 into 4x mode
> Feb 13 20:26:57 kobo kernel: [ 5112.044659] agpgart: Putting AGP V2 device at 0000:01:00.0 into 4x mode
> Feb 13 20:26:59 kobo kernel: [ 5113.556487] [drm] Setting GART location based on new memory map
> Feb 13 20:26:59 kobo kernel: [ 5113.556487] [drm] Loading R100 Microcode
> Feb 13 20:26:59 kobo kernel: [ 5113.695213] [drm] writeback test failed
>
> Sometimes very small horizontal lines all across the screen, sometimes
>  it will crash. It probably depends on temperature / how intensively
>  the card was used before. This is why the freezes aren't easy to
>  reproduce in all situations.
>
> torus-troopers windowed: FREEZE if the card was stressed
> torus-troopers fullscreen: FREEZE guaranted after a little while
> title screen fonts: OK
>
>
> 24bit - 1x
>
> (**) RADEON(0): Using AGP 1x
>
> Feb 13 20:43:55 kobo kernel: [  312.164131] agpgart: Found an AGP 2.0 compliant device at 0000:00:00.0.
> Feb 13 20:43:55 kobo kernel: [  312.164170] agpgart: Putting AGP V2 device at 0000:00:00.0 into 1x mode
> Feb 13 20:43:55 kobo kernel: [  312.164192] agpgart: Putting AGP V2 device at 0000:01:00.0 into 1x mode
> Feb 13 20:43:56 kobo kernel: [  313.673178] [drm] Setting GART location based on new memory map
> Feb 13 20:43:56 kobo kernel: [  313.673178] [drm] Loading R100 Microcode
> Feb 13 20:43:56 kobo kernel: [  313.812819] [drm] writeback test failed
>
> torus-troopers windowed: OK
> torus-troopers fullscreen: OK
> title screen fonts: MESSY
>
>
> 24bit - 2x
>
> (**) RADEON(0): Using AGP 2x
>
> Feb 13 21:23:53 kobo kernel: [ 1214.753137] agpgart: Found an AGP 2.0 compliant device at 0000:00:00.0.
> Feb 13 21:23:53 kobo kernel: [ 1214.753137] agpgart: Putting AGP V2 device at 0000:00:00.0 into 2x mode
> Feb 13 21:23:53 kobo kernel: [ 1214.753137] agpgart: Putting AGP V2 device at 0000:01:00.0 into 2x mode
> Feb 13 21:23:54 kobo kernel: [ 1216.260518] [drm] Setting GART location based on new memory map
> Feb 13 21:23:54 kobo kernel: [ 1216.260518] [drm] Loading R100 Microcode
> Feb 13 21:23:54 kobo kernel: [ 1216.260518] [drm] writeback test succeeded in 1 usecs
> ==> not sure what this failed/succeeded means
>
> torus-troopers windowed: OK
> torus-troopers fullscreen: OK
> title screen fonts: OK
>
> At first glance, no freeze.
>
>
> When the card freezes, SSH is still functional, but the only way to
> get the screen back is to reboot. The mouse may still be functional
> though.

Looks like we already have a quirk for this system.  It should be
handled properly in newer versions of the driver.

Alex