[PATCH] drm/atomic: Add target_vblank support in atomic helpers (v2)

[Daniel Vetter]

On Thu, Jan 12, 2017 at 9:28 AM, Michel Dänzer <michel at daenzer.net> wrote:
> On 12/01/17 12:48 AM, Grodzovsky, Andrey wrote:
>>> From: Michel Dänzer [mailto:michel at daenzer.net]
>>> On 09/01/17 06:59 PM, Daniel Vetter wrote:
>>>> On Fri, Jan 06, 2017 at 03:39:40PM -0500, Andrey Grodzovsky wrote:
>>>>> Allows usage of the new page_flip_target hook for drivers
>>>>> implementing the atomic path.
>>>>> Provides default atomic helper for the new hook.
>>>>>
>>>>> v2:
>>>>> Update code sharing logic between exsiting and the new flip hooks.
>>>>> Improve kerneldoc.
>>>>>
>>>>> Signed-off-by: Andrey Grodzovsky <Andrey.Grodzovsky at amd.com>
>>>>
>>>> Looks all reasonable, I think an ack from Alex that the amd side is in
>>>> shape too, and I'll pull this into drm-misc.
>>>
>>> Andrey, is there an updated patch 2 adapted to current patch 1? Other than
>>> that and some questionable indentation of parameters in function
>>> signatures, looks good to me FWIW.
>>
>> We are unable to use the atomic helpers both for page_flip and page_flip_target
>> At their current form mostly due to DRM_MODE_PAGE_FLIP_ASYNC flag rejection they do.
>> I discussed this with Daniel Vetter on IRC and suggested
>> to remove the rejection but he said the precise semantics of
>> atomic async flip is not clear yet and it's better to leave that out for now
>> until there is a  userspace asking for it.
>> So I tested it by just hacking  the helper to remove the rejection.
>> Until that settled the original change [PATCH 2/2] drm/amd/dal: Switch to page_flip_target hook in DAL
>> Is what we plan to use in DAL
>
> IIRC Daniel suggested (on IRC?) to use the helper for non-async flips
> and the current DC code for async flips. Is that feasible?

We do have some async flip flag reserved for atomic, so we could route
it through. But since atm there's no one asking for async flips on the
atomic ioctl I'm a bit wary for fear of ending up with ill-defined
semantics. But I guess if we do this for legacy pageflips only, and
make sure we do still reject async flips submitted through the atomic
ioctl that should be all fine. And it should allow amdgpu to entirely
get rid of the legacy flip path, which would be nice.

Once we have that we could even use it for cursor plane updates
(through the legacy ioctl, for drivers with universal planes), for
those drivers that support it.
-Daniel
-- 
Daniel Vetter
Software Engineer, Intel Corporation
+41 (0) 79 365 57 48 - http://blog.ffwll.ch