On Fri, May 22, 2020 at 10:36:09AM -0400, Alan Stern wrote: > On Fri, May 22, 2020 at 03:56:59AM -0700, Paul E. McKenney wrote: > > On Fri, May 22, 2020 at 11:44:07AM +0200, Peter Zijlstra wrote: > > > On Thu, May 21, 2020 at 05:38:50PM -0700, Paul E. McKenney wrote: > > > > Hello! > > > > > > > > Just wanted to call your attention to some pretty cool and pretty serious > > > > litmus tests that Andrii did as part of his BPF ring-buffer work: > > > > > > > > https://lore.kernel.org/bpf/20200517195727.279322-3-andriin@xxxxxx/ > > > > > > > > Thoughts? > > > > > > I find: > > > > > > smp_wmb() > > > smp_store_release() > > > > > > a _very_ weird construct. What is that supposed to even do? > > > > Indeed, and I asked about that in my review of the patch containing the > > code. It -could- make sense if there is a prior read and a later store: > > > > r1 = READ_ONCE(a); > > WRITE_ONCE(b, 1); > > smp_wmb(); > > smp_store_release(&c, 1); > > WRITE_ONCE(d, 1); > > > > So a->c and b->c is smp_store_release() and b->d is smp_wmb(). But if > > there were only stores, the smp_wmb() would suffice. And if there wasn't > > the trailing store, smp_store_release() would suffice. > > But that wasn't the context in the litmus test. The context was: > > smp_wmb(); > smp_store_release(); > spin_unlock(); > smp_store_release(); > > That certainly looks like a lot more ordering than is really needed. I suspect that you are right. I asked him if there were other accesses in my response to his ringbuffer (as opposed to litmus-test) patch: https://lore.kernel.org/bpf/20200522002502.GF2869@paulmck-ThinkPad-P72/ If there are other accesses requiring both, the litmus tests might need to be updated. Thanx, Paul
