On Tue, Jun 10, 2025 at 12:39:18PM -1000, Tejun Heo wrote: > Hello, > > On Tue, Jun 10, 2025 at 03:31:03PM -0700, Shakeel Butt wrote: > ... > > Couple of lines above I have llist_on_list(&rstatc->lnode) check which > > should be as cheap as data_race(css_rstat_cpu(css, cpu)->updated_next). > > Ah, I missed that. > > > So, I can add lnode for nmi and non-nmi contexts (with irqs disabled) > > but I think that is not needed. Actually I ran the netperf benchmark (36 > > parallel instances) and I see no significant differences with and > > without the patch. > > Yeah, as long as the hot path doesn't hit the extra cmpxchg, I think it > should be fine. Can you fortify the comments a bit that the synchronization > is against the stacking contexts on the same CPU. The use of cmpxchg for > something like this is a bit unusual and it'd be nice to have explanation on > why it's done this way and why the overhead doesn't matter. I was actually thinking of using this_cpu_cmpxchg but then I need to also check for CONFIG_ARCH_HAS_NMI_SAFE_THIS_CPU_OPS. However if you prefer that, I can try this_cpu_cmpxchg in the next version. I will also fix the comment with additional information about stacking context.
