[PATCH 0/6] drm: Explicit target vblank seqno for page flips

[Michel Dänzer]

On 04.08.2016 19:12, Daniel Stone wrote:
> On 4 August 2016 at 11:01, Michel Dänzer <michel at daenzer.net> wrote:
>> On 04.08.2016 18:51, Daniel Stone wrote:
>>> On 4 August 2016 at 04:39, Michel Dänzer <michel at daenzer.net> wrote:
>>>> Patch 6 extends the ioctl with new flags, which allow userspace to
>>>> explicitly specify the target vblank seqno. This can also avoid delaying
>>>> flips in some cases where we are already in the target vertical blank
>>>> period when the ioctl is called.
>>>
>>> Is there open userspace for this?
>>
>> Sure, referenced in patch 6:
>>
>> https://cgit.freedesktop.org/~daenzer/xf86-video-ati/commit/?id=fc884a8af25345c32bd4104c864ecfeb9bb3db9b
>>
>> https://cgit.freedesktop.org/~daenzer/xf86-video-amdgpu/commit/?id=b8631a9ba49c0d0ebe5dcd1dbfb68fcfe907296f

[...]

>> The only change compared to the existing ioctl is that userspace can ask
>> for a flip to take effect in the current vblank seqno. The code added by
>> the patch checks for target vblank seqno > current vblank seqno + 1 and
>> returns -EINVAL in that case. This is also documented in drm_mode.h.
> 
> Is there any particular benefit to having split absolute/relative
> modes in this case? Personally I'm struggling to see the use of
> relative.

See the userspace patches above. It allows userspace to set the target
to the current/next vblank seqno without a DRM_IOCTL_WAIT_VBLANK
round-trip (which could also result in the target getting delayed by one
frame compared to using a relative target directly).


>>> Is all this tested somewhere?
>>
>> Yes, I've been using it for a while on all my machines.
> 
> I mean in a test suite. :)

Not yet. Are you thinking of intel-gpu-tools, or something else?


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