On Wed, Dec 5, 2018 at 12:07 PM H.J. Lu wrote: > > On Wed, Dec 5, 2018 at 11:57 AM Florian Weimer wrote: > > > > * H. J. Lu: > > > > > I updated the patch with a very simple test. > > > > The simple test has a race condition unfortunately. > > > > I think we could fold the test into the existing affinity tests. > > Sure, I will do that. Like this. > > The main question is whether getcpu is fast enough. It does not perform > > a system call, but in my tests a while back, it was still rather slow. > > It is fast enough for NUMA spinlock. > > > The new rseq-based getcpu does not support NUMA nodes, it seems. Maybe > > it's still possible to fix that. > > That will be great. We can take advantage it when it is available. > > -- > H.J. -- H.J.