The tracepoint interface will stop providing regular RCU context; make
sure we do it ourselves, since perf makes use of regular RCU protected
data.
Suggested-by: Steven Rostedt (VMware) <[email protected]>
Suggested-by: Mathieu Desnoyers <[email protected]>
Signed-off-by: Peter Zijlstra (Intel) <[email protected]>
---
kernel/events/core.c | 5 +++++
1 file changed, 5 insertions(+)
--- a/kernel/events/core.c
+++ b/kernel/events/core.c
@@ -8950,6 +8950,7 @@ void perf_tp_event(u16 event_type, u64 c
{
struct perf_sample_data data;
struct perf_event *event;
+ unsigned long rcu_flags;
struct perf_raw_record raw = {
.frag = {
@@ -8961,6 +8962,8 @@ void perf_tp_event(u16 event_type, u64 c
perf_sample_data_init(&data, 0, 0);
data.raw = &raw;
+ rcu_flags = trace_rcu_enter();
+
perf_trace_buf_update(record, event_type);
hlist_for_each_entry_rcu(event, head, hlist_entry) {
@@ -8996,6 +8999,8 @@ void perf_tp_event(u16 event_type, u64 c
}
perf_swevent_put_recursion_context(rctx);
+
+ trace_rcu_exit(rcu_flags);
}
EXPORT_SYMBOL_GPL(perf_tp_event);
On Wed, Feb 12, 2020 at 10:01:45PM +0100, Peter Zijlstra wrote:
> The tracepoint interface will stop providing regular RCU context; make
> sure we do it ourselves, since perf makes use of regular RCU protected
> data.
>
> Suggested-by: Steven Rostedt (VMware) <[email protected]>
> Suggested-by: Mathieu Desnoyers <[email protected]>
> Signed-off-by: Peter Zijlstra (Intel) <[email protected]>
> ---
> kernel/events/core.c | 5 +++++
> 1 file changed, 5 insertions(+)
>
> --- a/kernel/events/core.c
> +++ b/kernel/events/core.c
> @@ -8950,6 +8950,7 @@ void perf_tp_event(u16 event_type, u64 c
> {
> struct perf_sample_data data;
> struct perf_event *event;
> + unsigned long rcu_flags;
The flags are not needed I guess, if you agree on not using in_nmi() in
trace_rcu_enter().
thanks,
- Joel
> struct perf_raw_record raw = {
> .frag = {
> @@ -8961,6 +8962,8 @@ void perf_tp_event(u16 event_type, u64 c
> perf_sample_data_init(&data, 0, 0);
> data.raw = &raw;
>
> + rcu_flags = trace_rcu_enter();
> +
> perf_trace_buf_update(record, event_type);
>
> hlist_for_each_entry_rcu(event, head, hlist_entry) {
> @@ -8996,6 +8999,8 @@ void perf_tp_event(u16 event_type, u64 c
> }
>
> perf_swevent_put_recursion_context(rctx);
> +
> + trace_rcu_exit(rcu_flags);
> }
> EXPORT_SYMBOL_GPL(perf_tp_event);
>
>
>
On Wed, Feb 12, 2020 at 06:28:30PM -0500, Joel Fernandes wrote:
> On Wed, Feb 12, 2020 at 10:01:45PM +0100, Peter Zijlstra wrote:
> > The tracepoint interface will stop providing regular RCU context; make
> > sure we do it ourselves, since perf makes use of regular RCU protected
> > data.
> >
> > Suggested-by: Steven Rostedt (VMware) <[email protected]>
> > Suggested-by: Mathieu Desnoyers <[email protected]>
> > Signed-off-by: Peter Zijlstra (Intel) <[email protected]>
> > ---
> > kernel/events/core.c | 5 +++++
> > 1 file changed, 5 insertions(+)
> >
> > --- a/kernel/events/core.c
> > +++ b/kernel/events/core.c
> > @@ -8950,6 +8950,7 @@ void perf_tp_event(u16 event_type, u64 c
> > {
> > struct perf_sample_data data;
> > struct perf_event *event;
> > + unsigned long rcu_flags;
>
> The flags are not needed I guess, if you agree on not using in_nmi() in
> trace_rcu_enter().
Even then we need to store the state: 'didn't do nothing' vs 'did call
rcu_needs_to_wake_up_and_pay_attention_noaw'. That is, we only need to
do something (expensive!) when !rcu_is_watching().
On Thu, Feb 13, 2020 at 09:29:51AM +0100, Peter Zijlstra wrote:
> On Wed, Feb 12, 2020 at 06:28:30PM -0500, Joel Fernandes wrote:
> > On Wed, Feb 12, 2020 at 10:01:45PM +0100, Peter Zijlstra wrote:
> > > The tracepoint interface will stop providing regular RCU context; make
> > > sure we do it ourselves, since perf makes use of regular RCU protected
> > > data.
> > >
> > > Suggested-by: Steven Rostedt (VMware) <[email protected]>
> > > Suggested-by: Mathieu Desnoyers <[email protected]>
> > > Signed-off-by: Peter Zijlstra (Intel) <[email protected]>
> > > ---
> > > kernel/events/core.c | 5 +++++
> > > 1 file changed, 5 insertions(+)
> > >
> > > --- a/kernel/events/core.c
> > > +++ b/kernel/events/core.c
> > > @@ -8950,6 +8950,7 @@ void perf_tp_event(u16 event_type, u64 c
> > > {
> > > struct perf_sample_data data;
> > > struct perf_event *event;
> > > + unsigned long rcu_flags;
> >
> > The flags are not needed I guess, if you agree on not using in_nmi() in
> > trace_rcu_enter().
>
> Even then we need to store the state: 'didn't do nothing' vs 'did call
> rcu_needs_to_wake_up_and_pay_attention_noaw'. That is, we only need to
> do something (expensive!) when !rcu_is_watching().
You are right, that sounds good. I was talking to Paul and we chatted that if
in_nmi() is safe (which I believe it is as we are not calling RCU before you
update the preempt counts), then in RCU we can replace the @irq with
!in_nmi() and simplify that code. Then we can simplify this bit as well
(keep rcu_flags but only call rcu_irq_enter_irqsave() instead of
rcu_nmi_enter(). May be you can do the RCU internal bits in your v3 or should
those be separate patches? Whatever Paul and you want to do.
thanks,
- Joel
- Joel