On Thu, Nov 20, 2025 at 3:45 AM Laurent Pinchart <[email protected]> wrote:
> > But I digress. I think initially we should just look for a clean fix for > > the platforms affected: > > > > - Add the implementation into the drivers? > > - Add helpers to the core? > > - Add a flag of some kind so the core can do the right thing? > > drm_atomic_helper_commit_modeset_enables_crtc_early() would be more > cumbersome to implement manually in drivers as most of the functions it > calls are not exported. drm_atomic_helper_commit_tail_crtc_early_late() > shouldn't be difficult to implement in the MCDE driver. But the second patch patch adding drm_atomic_helper_commit_modeset_disables_crtc_late() would be needed for symmetry. > > I made a quick test with the flag approach, below. It's not many lines, > > but... Ugh, it does feel like a hack. > > Without seeing the code I can already imagine how this would feel like a > hack, so I agree not to go that way. It seems we cannot reach consensus on a regression fix that doesn't require large re-architecturing, so I'll go ahead and propose a revert instead, it's the only sensible thing. I hope nothing breaks from the revert... Yours, Linus Walleij
