core: refactor and improve surface commit

[ikalco]

Describe your PR, what does it fix/add?

part 2 of #9774

relevant commits:
cleanup SSurfaceState and surface pending commit tracking
move surface code from DRMSyncobj, and move acquire to SSurfaceState
use queue for comitted pending surface states like proto says
drop, not release, prev buffer if 2nd buffer wl_surface.attach is sent

Is there anything you want to mention? (unchecked code, possible bugs, found problems, breaking compatibility, etc.)

based on part 1

Is it ready for merging, or does it need work?

should be gtg now
nvm lol nvidia borkde again (╥_╥)
yes finally

[gulafaran]

currently discord typing is occasionally delayed, not updating. minecraft flickers, cs2 flickers.

[gulafaran]

and obs -> right click preview -> open windowed projector should die with

[ 653985.111] {Display Queue} wl_display#1.error(wp_linux_drm_syncobj_surface_v1#77, 4, "Missing acquire timeline")
wp_linux_drm_syncobj_surface_v1#77: error 4: Missing acquire timeline

as it does on -git, currently doesnt.

[ikalco]

and obs -> right click preview -> open windowed projector should die with

[ 653985.111] {Display Queue} wl_display#1.error(wp_linux_drm_syncobj_surface_v1#77, 4, "Missing acquire timeline")
wp_linux_drm_syncobj_surface_v1#77: error 4: Missing acquire timeline

as it does on -git, currently doesnt.

fixed

[ikalco]

works on nvidia now, no flicker or delayed frames :)
I had to dig up my old 1660 nvidia card (thank god I could repro lol), and this pr works on that now

[PlasmaPower]

Did you test this with the vkcube FIFO mode Tom mentioned was broken? That might've been from another PR but still worth testing here

[ikalco]

tested with explicit_sync both on and off, all work on nvidia

• minecraft, vsync and unlimited, windowed and ds
• csgo, windowed and ds
• obs recording
• vkcube --wsi wayland --present_mode {number here} with 0, 1, and 2

I'm also daily running this pr on an old Intel laptop with no issues (explicit enabled)

[ikalco]

@vaxerski do you mind if we just merge this as one pr lol, cause I don't really wanna go through all the combinations of testing all over again for #9804 if I know this works

maybe we can do something like
git squash the buffer ref stuff into one commit
git squash the surface stuff into another
and merge those 2 kernel style like you said before

or just merge the entire pr into one commit like normal

p.s. the other two prs aren't required for this, still drafts until I get the time to clean/fix them up, discuss, and test

[PlasmaPower]

I've been running this PR for a couple days and it seems to be working well for my usecases on NVIDIA.

[ikalco]

rebased to main

also should be good for final review/merge
PlasmaPower, gulafaran, and I, dont have any glitching bugs

[vaxerski]

my question is what is this supposed to fix?

[ikalco]

my question is what is this supposed to fix?

there isn't a specific bug this pr is fixing, but it is building towards more well defined surface commit behavior
so that in something like my proper buffer release pr (#9807) I know what gets called where,
and can hopefully fix the last of those pesky nvidia flickering bugs (#9640)

this pr also follows wayland surface protocol closer, like in the last 2 commits
and while I was in there changing all of that, I also decided to clean up SSurfaceState by

• making updated easier to use
• setting the right initial values based on the protocol text
• grouping functions and adding some clarifying comments

---

p.s.

here are some exmaples of previously not well defined behavior

• previously, buffer refs were tracked using a SP<CHLBufferReference> which has 2 ref counting methods, the SP and buffer->nLocks... that's confusing and unneeded, especially for having proper buffer releases

• release and acquire points were held in the CHLBufferReference, which is wrong
  this pr moves the acquire point to SurfaceState,
  and the release will be done properly in the later release stuff pr

• some commitPendingState calls were obscured through a events.precommit.emit() and done in DRMSyncobj protocol
  which again adds some confusion as to how and where state is committed
  and sometimes the DRMSyncobj protocol committed state incorrectly (some state in pending isn't committed...)

  if (!surface->pending.buffer && !PENDING_HAS_NEW_BUFFER && surface->current.buffer) {
          surface->current.bufferDamage.clear();
          surface->current.damage.clear();
          surface->commitPendingState(surface->current);
          return; // no new buffer, but we still have current around and a commit happend, commit current again.
    }

• previously, the pending surface state after a commit was being reset in multiple places in different ways, now we just have SSurfaceState::reset() to clearly define that new pending state

• simlarly to right above, the SSurfaceState::updateFrom just copied everything when a buffer commit happened, which is wrong. Now we only copy committed things

  if (ref.updated & SURFACE_UPDATED_BUFFER) {
          ref.updated &= ~SURFACE_UPDATED_BUFFER;
          *this = ref;
          ref.damage.clear();
          ref.bufferDamage.clear();
          ref.buffer = {};
  }

[vaxerski]

cs works fine on my end so yea