On 11/19/2012 03:03 PM, Maarten Lankhorst wrote: > Op 19-11-12 14:26, Thomas Hellstrom schreef: >> Hi, >> >> On 11/12/2012 03:00 PM, Maarten Lankhorst wrote: >>> move to release_list instead >> Can you describe why this change is made? cleanup? reorder locks in a later >> patch? >> Also please describe why you need move_notify and ttm unbind / destroy to be >> outside of >> reservation, because that's the main change in this patch and it's not even >> mentioned in the >> commit message. > Ok is a reword enough? In that case I'll resend. > > I moved all the destruction to happen when release_list refcount drops to 0. > This removes the special handling of ttm_bo_cleanup_memtype_use, and > makes it part of the normal bo destruction instead. > > It also meant that move_notify and unbind/destroy was without reservation, > simply > because it was done during normal destruction instead. At that point you may > no longer > hold a reservation, but you can already be sure you're the only one touching > it.
That's true, but even if we are the only users it would be good to keep reservation when calling move_notify and the unbind stuff, simply because drivers can easily detect reservation bugs if these functions are always called reserved. if we kan keep the reservation here without too much trouble, I'd like us to do so. Thanks, Thomas
