On Tue, 29 Sep 2020 13:25:41 +0200 Peter Zijlstra <peterz@xxxxxxxxxxxxx> wrote: > On Tue, May 19, 2020 at 07:52:33PM -0000, tip-bot2 for Thomas Gleixner wrote: > > @@ -979,7 +988,7 @@ static void rcu_disable_urgency_upon_qs(struct rcu_data *rdp) > > * if the current CPU is not in its idle loop or is in an interrupt or > > * NMI handler, return true. > > */ > > -bool notrace rcu_is_watching(void) > > +bool rcu_is_watching(void) > > { > > bool ret; > > > > This ^.. > > it is required because __ftrace_ops_list_func() / > ftrace_ops_assist_func() call it outside of ftrace recursion, but only > when FL_RCU, and perf happens to be the only user of that. > > another morning wasted... :/ Those are fun to debug. :-p (I'll use this in another email). Anyway, you bring up a good point. I should have this: diff --git a/kernel/trace/ftrace.c b/kernel/trace/ftrace.c index 84f32dbc7be8..2d76eaaad4a7 100644 --- a/kernel/trace/ftrace.c +++ b/kernel/trace/ftrace.c @@ -6993,16 +6993,14 @@ static void ftrace_ops_assist_func(unsigned long ip, unsigned long parent_ip, { int bit; - if ((op->flags & FTRACE_OPS_FL_RCU) && !rcu_is_watching()) - return; - bit = trace_test_and_set_recursion(TRACE_LIST_START, TRACE_LIST_MAX); if (bit < 0) return; preempt_disable_notrace(); - op->func(ip, parent_ip, op, regs); + if (!(op->flags & FTRACE_OPS_FL_RCU) || rcu_is_watching()) + op->func(ip, parent_ip, op, regs); preempt_enable_notrace(); trace_clear_recursion(bit); -- Steve
![]() |