Quoting Ville Syrjala (2020-01-22 20:43:29) > From: Ville Syrjälä <[email protected]> > > We've added more internal things that use modeset locks and > thus we need to be prepared for intel_atomic_check() grabbing > more locks than what our initial drm_modeset_lock_all_ctx() > took. So we're missing the backoff handling here. > > Also drm_atomic_helper_duplicate_state() works against us > by clearing state->acquire_ctx in anticipation of > drm_atomic_helper_commit_duplicated_state() being used to > commit the state. > > We could probably just reset acquire_ctx back, but instead > let's just rewrite the whole thing without using either of > those "helpers". There's also no need to add any connectors > to the state here since we just want the new watermarks > which don't depend on connectors. > > Cc: Chris Wilson <[email protected]> > Signed-off-by: Ville Syrjälä <[email protected]>
Some amount of staring later, aiui the new flow is correct. Reviewed-by: Chris Wilson <[email protected]> -Chris _______________________________________________ Intel-gfx mailing list [email protected] https://lists.freedesktop.org/mailman/listinfo/intel-gfx
