[Powerpc] Sam460ex Canyonlands issue -Kernel 4.4.6-rc1

[Michel Dänzer]

On 31.03.2016 19:00, Julian Margetson wrote:
> On 3/31/2016 2:50 AM, Michel Dänzer wrote:
>> On 30.03.2016 19:36, Julian Margetson wrote:
>>> On 3/29/2016 11:49 PM, Michel Dänzer wrote:
>>>> On 29.03.2016 18:55, Julian Margetson wrote:
>>>>> On 3/28/2016 11:15 PM, Michel Dänzer wrote:
>>>>>> On 29.03.2016 08:47, Julian Margetson wrote:
>>>>>>> Seeing the following when booting kernel 4.6-rc1 on Acube Sam460ex
>>>>>>> Canyonlands board.
>>>>>>> This loops for a few times then the kernel boots.
>>>>>>> No problem with the 4.6-rc1 with an A-eon Tabor Freescale e500v2
>>>>>>> board.
>>>>>>>
>>>>>>>     [    2.197839] ------------[ cut here ]------------
>>>>>>>     [    2.197850] WARNING: CPU: 0 PID: 1 at
>>>>>>> drivers/gpu/drm/drm_irq.c:1368 drm_vblank_off+0x2c/0x1e0
>>>>>> That's
>>>>>>
>>>>>>       if (WARN_ON(pipe >= dev->num_crtcs))
>>>>>>
>>>>>> My best guess is that drm_vblank_off is called before
>>>>>> drm_vblank_init, so dev->num_crtcs is still 0.
>>>>>>
>>>>>>
>>>>>> Please provide the full dmesg output corresponding to the problem.
>>>>> 
>>>>> [drm] radeon: irq initialized.
>>>>> [drm:r600_ring_test] *ERROR* radeon: ring 0 test failed
>>>>> (scratch(0x850C)=0xCAFEDEAD)
>>>>> radeon 0001:81:00.0: disabling GPU acceleration
>>>> Okay, so the problem is that acceleration fails to initialize, in which
>>>> case the driver calls drm_vblank_cleanup.
>>>>
>>>> I can see two basic options for a solution: Either don't call
>>>> radeon_irq_kms_fini/drm_vblank_cleanup if acceleration fails to
>>>> initialize, or check if acceleration is enabled before calling
>>>> drm_vblank_on/off. Any preferences?
>>>>
>>>>
>>> Thanks
>>>
>>> No preferences .
>> Does the attached patch fix the problem?
>>
>>
>>> I would be interested in a fix for the acceleration problem if
>>> possible .
>> Beware that while the kernel side of this might be relatively easy to
>> fix, making the userspace radeonsi driver work on big endian hosts would
>> likely require substantial effort.
> 
> Thanks
> 
> The patch solved the problem .
> Tested-by: Julian Margetson <runaway at candw.ms>

Thanks, patch submitted for inclusion.


> I  would indeed like to see the kernel side fix for the the acceleration
> problem. (Been hoping for it for almost 3 years now)
> http://marc.info/?l=dri-devel&m=137881893518240&w=2
> I also have Northern Islands cards  and  acceleration with these does
> work with my other powerpc board, an e500v2 .
> It also requires the patch in the linked post by Hans Verkuil for Radeon
> NI and SI cards to work .

Is Hans' patch enough for SI or not? Anyway, somebody should split Hans'
patch up between ttm and radeon, clean up the latter (use PRIx64 or a
phys_addr_t specific format sequence instead of %llx, apply the same fix
in radeon_doorbell_init) and submit them as properly Git formatted patches.


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