On Wed, Oct 11, 2017 at 05:02:28PM +0800, zhouchengming wrote: > On 2017/10/11 15:45, Peter Zijlstra wrote:
> > + if (!(flags& PERF_EF_START)) > > + p_event->hw.state = PERF_HES_STOPPED; > > Don't we need to check the flags for ftrace perf_event? > So if we should put this outside the if (!tp_event->class->reg()) ? Oh, right you are, that's independent of the hlist nonsense. > > + list = this_cpu_ptr(pcpu_list); > > + hlist_add_head_rcu(&p_event->hlist_entry, list); > > + } > > Now we don't add perf_event to the pcpu_list, so we also can avoid > to alloc pcpu_list for function tp_event in perf_trace_event_reg(). Yeah that allocation is superfluous, Not entirely sure how to cleanly get rid of that though; that's another web to untangle.
