Sorry for the noise... and currently this is fine. But if we want to add the wq_has_sleeper() checks into fs/pipe.c then pipe_poll() needs smp_mb() after it calls poll_wait(). Agreed? On 12/29, Oleg Nesterov wrote: > > On 12/29, Manfred Spraul wrote: > > > > >I think that your patch (and the original patch from WangYuli) has the same > > >proble with pipe_poll()->poll_wait()->__pollwait(). > > > > What is the memory barrier for pipe_poll()? > > > > There is poll_wait()->__pollwait()->add_wait_queue()->spin_unlock(). thus > > only store_release. > > > > And then READ_ONCE(), i.e. no memory barrier. > > > > Thus the CPU would be free to load pipe->head and pipe->tail before adding > > the entry to the poll table. > > > > Correct? > > Yes, this was my thinking. > > See also my initial reply to WangYuli > https://lore.kernel.org/all/20241226160007.GA11118@xxxxxxxxxx/ > > Do you agree? > > Oleg.
