Am 28.09.23 um 15:37 schrieb Michel Dänzer:
On 9/28/23 14:59, Ray Strode wrote:
On Thu, Sep 28, 2023 at 5:43 AM Michel Dänzer
<michel.daen...@mailbox.org> wrote:
When it's really not desirable to account the CPU overhead to the
process initiating it then you probably rather want to use an non
blocking commit plus a dma_fence to wait for the work to end from userspace.
Well, first I don't think that's very convenient. You're talking about
a per-plane property, so there would need to be a separate file
descriptor allocated for every plane, right? and user-space would have
to block on all of them before proceeding?
OUT_FENCE_PTR is a per-CRTC property, not per-plane.
Okay, sure.
Also, at least in this particular case, a single sync file (not dma_fence) for
any CRTC might suffice.
I don't see how we could rely on that given the provided api and
multitude of drivers. It might work and then break randomly.
If it's supposed to work from the KMS API PoV, any bugs to the contrary should
be fixed.
I'm not really seeing the big difference between using a single fence or
multiple, anyway.
The big difference is that a standard modeset can take some time, e.g.
setting up power levels, waiting for PLLs to settle, waiting for a
vblank etc..
That this happens async in the background so that the frontend
application can still respond to other signals seems reasonable.
But in the case of turning thing off, what should we wait for? I think
we still support the out fence, but it doesn't really make sense to use.
I do wonder if there might be a time window where the out fences have
signalled, but the atomic commit ioctl will still fail with EBUSY. If there is
though, I'd expect it to affect the flip completion events as well.
I'm not deep enough into that code to confirm that. Daniel or other
display folks need to help here.
Regards,
Christian.
