On 10/15/21 6:22 AM, Pierre-Louis Bossart wrote: > If this patch causes issues outside of the trigger phase, then maybe we > could just change the BE nonatomic flag temporarily and restore it after > taking the lock, but IMHO something else is broken in other drivers. Now that I think of it, this wouldn't work, the type of locks for the BEs has to be set before the trigger: the DPCM flow is that we start from the FEs and find which BEs need to be triggered. That would prevent us from modifying a global BE property - each FE is independent.
