Re: radeon mobility x1600 broken since 3.4.0-rc6 (bisected)

[Hausensteiner Keg <skitching@apache.org>]

On 27/08/12 21:04, Hausensteiner Keg wrote:
> On 27/08/12 20:29, Jerome Glisse wrote:
>> On Sun, Aug 26, 2012 at 8:34 PM, Hausensteiner Keg 
>> <skitching@apache.org> wrote:
>>> ...
>> Does patch
>>
>> http://people.freedesktop.org/~glisse/0001-drm-radeon-extra-type-safe-for-fence-emission.patch 
>>
>>
>> Helps ?
>
> Thanks Jerome. Will try patch tonight or tomorrow morning.
>
> How do I check whether the radeon is on an AGP bus?
> The lspci output does not mention AGP; line for video controller is:
>   01:00.0 VGA compatible controller: Advanced Micro Devices [AMD] nee 
> ATI M56P [Radeon Mobility X1600]
> The "lshw" output also shows no signs of AGP.
>
> The dmesg output contains "[0.909546] Linux agpgart interface v0.103" 
> but otherwise no mention of "agp" or "AGP".
>
> By the way: tested the 3.5.3 release made today : still corrupt 
> graphics. And the v3.5 release (which was oddly working) also showed 
> the same corrupt graphics as soon as I forced a rebuild of the initrd 
> - maybe an older radeon module somehow ended up in the initrd built by 
> "make-kpkg --initrd".
>

Patch 0001-drm-radeon-extra-type-safe-for-fence-emission applies cleanly 
to both 3.5.3 and 3.6.0-rc3, but makes no difference in either case.

I've done some more bisecting, and the full sequence is:

* Prior to commit bb635567 ("convert fence to uint64_t v4") all works fine.
* Commit bb635567 causes system to freeze on Plymouth startup.
* The following commit (3b7a2b24, "rework fence handling, drop fence 
list v7") causes Plymouth to work again, but causes graphics corruption 
on greeter screen. Pressing ctrl-alt-f2 immediately displays a virtual 
console correctly, and the system is useable. However any attempt to use 
the mouse on the greeter screen triggers further graphics corruption, 
and system appears to hang.

This situation continues up to and including 3.5.3.

In the 3.6 dev cycle, the corruption issue continues up to 9c19415c. 
However the following commit, 876dc9f3 "remove radeon_fence_create" 
instead causes the screen to go black immediately after plymouth 
completes (ie when greeter screen is expected), and ctrl-alt-f2 has no 
effect.

FWIW, looks like two separate issues to me, with 876dc9f3 masking the 
earlier one..

Regards,
Simon