Op 13-01-15 om 10:22 schreef Christian König: > Am 13.01.2015 um 10:04 schrieb Maarten Lankhorst: >> Op 13-01-15 om 10:36 schreef Jammy Zhou: >>> When the timeout value passed to reservation_object_wait_timeout_rcu >>> is zero, no wait should be done if the fences are not signaled. >>> >>> Return '1' for idle and '0' for busy if the specified timeout is '0' >>> to keep consistent with the case of non-zero timeout. >>> >>> v2: call fence_put if not signaled in the case of timeout==0 >>> >>> >> Looks more sane, but where do you need this? > > The rational is that we want a single function to call from the driver no > matter what timeout we have and get a zero if the call timed out and a non > zero value otherwise. > > When we called reservation_object_wait_timeout_rcu with a timeout of zero we > got a return value of zero as well independent of whether or not the fences > were signaled. That behavior looked inconsistent and Jammy is trying to fix > this with this patch. > > We could of course do the check in the calling code as well, but to me it > rather looked like the calling convention of > reservation_object_wait_timeout_rcu should be thought over one more time. > > Either calling it with timeout=0 is invalid and the driver needs to call > reservation_object_test_signaled_rcu directly instead or we apply this patch > or something similar to get an useful behavior in the case of timeout=0. I think it would be best to leave timeout=0 returning 0. Not handling it differently gives the same semantics as used by fence_wait_time and wait_event_timeout. Are there really many cases in which you don't know if timeout = 0 before or not?
~Maarten
