Op 30-06-2020 om 14:31 schreef Tvrtko Ursulin: > > On 30/06/2020 12:52, Maarten Lankhorst wrote: >> Op 29-06-2020 om 17:08 schreef Tvrtko Ursulin: >>> >>> On 23/06/2020 15:28, Maarten Lankhorst wrote: >>>> This reverts commit 0f1dd02295f35dcdcbaafcbcbbec0753884ab974. >>>> This conflicts with the ww mutex handling, which needs to drop >>>> the references after gpu submission anyway, because otherwise we >>>> may risk unlocking a BO after first freeing it. >>> >>> What is the problem here? eb_vma_array_put in eb_move_to_gpu? If so, could you just move this put to later in the sequence? I am simply thinking how to avoid controversial reverts. Because on the other hand I did not figure out what 0f1dd02295f35dcdcbaafcbcbbec0753884ab974 fixed in a few minutes I spent staring at the patch. >> >> >> We need to unlock before we unref to prevent a use-after-free in unlock, so freeing and releasing in eb_move_to_gpu() is too early. This means we only end up with 1 path for unlock, so it's fine to revert. > > You are saying the reason 0f1dd02295f35dcdcbaafcbcbbec0753884ab974 was added for will not be there after your changes? > > Regards, > > Tvrtko Yes. :) _______________________________________________ Intel-gfx mailing list Intel-gfx@xxxxxxxxxxxxxxxxxxxxx https://lists.freedesktop.org/mailman/listinfo/intel-gfx
