[PATCH v4] trace,hardirq: No moar _rcuidle() tracing

[Ingo Molnar]

* Peter Zijlstra <peterz@xxxxxxxxxxxxx> wrote:

> On Tue, Jan 17, 2023 at 01:24:46PM +0900, Masami Hiramatsu wrote:
> > Hi Peter,
> > 
> > On Thu, 12 Jan 2023 20:43:49 +0100
> > Peter Zijlstra <peterz@xxxxxxxxxxxxx> wrote:
> > 
> > > Robot reported that trace_hardirqs_{on,off}() tickle the forbidden
> > > _rcuidle() tracepoint through local_irq_{en,dis}able().
> > > 
> > > For 'sane' configs, these calls will only happen with RCU enabled and
> > > as such can use the regular tracepoint. This also means it's possible
> > > to trace them from NMI context again.
> > > 
> > > Signed-off-by: Peter Zijlstra (Intel) <peterz@xxxxxxxxxxxxx>
> > 
> > The code looks good to me. I just have a question about comment.
> > 
> > > ---
> > >  kernel/trace/trace_preemptirq.c |   21 +++++++++++++--------
> > >  1 file changed, 13 insertions(+), 8 deletions(-)
> > > 
> > > --- a/kernel/trace/trace_preemptirq.c
> > > +++ b/kernel/trace/trace_preemptirq.c
> > > @@ -20,6 +20,15 @@
> > >  static DEFINE_PER_CPU(int, tracing_irq_cpu);
> > >  
> > >  /*
> > > + * ...
> > 
> > Is this intended? Wouldn't you leave any comment here?
> 
> I indeed forgot to write the comment before posting, my bad :/ Ingo fixed
> it up when he applied.

For completeness, I've attached the final commit, which has this comment 
included:

+/*
+ * Use regular trace points on architectures that implement noinstr
+ * tooling: these calls will only happen with RCU enabled, which can
+ * use a regular tracepoint.
+ *
+ * On older architectures, use the rcuidle tracing methods (which
+ * aren't NMI-safe - so exclude NMI contexts):
+ */

Thanks,

	Ingo

================>
From: Peter Zijlstra <peterz@xxxxxxxxxxxxx>
Date: Thu, 12 Jan 2023 20:43:49 +0100
Subject: [PATCH] tracing, hardirq: No moar _rcuidle() tracing

Robot reported that trace_hardirqs_{on,off}() tickle the forbidden
_rcuidle() tracepoint through local_irq_{en,dis}able().

For 'sane' configs, these calls will only happen with RCU enabled and
as such can use the regular tracepoint. This also means it's possible
to trace them from NMI context again.

Reported-by: kernel test robot <lkp@xxxxxxxxx>
Signed-off-by: Peter Zijlstra (Intel) <peterz@xxxxxxxxxxxxx>
Signed-off-by: Ingo Molnar <mingo@xxxxxxxxxx>
Link: https://lore.kernel.org/r/20230112195541.477416709@xxxxxxxxxxxxx
---
 kernel/trace/trace_preemptirq.c | 26 ++++++++++++++++++--------
 1 file changed, 18 insertions(+), 8 deletions(-)

diff --git a/kernel/trace/trace_preemptirq.c b/kernel/trace/trace_preemptirq.c
index 629f2854e12b..f992444a0b1f 100644
--- a/kernel/trace/trace_preemptirq.c
+++ b/kernel/trace/trace_preemptirq.c
@@ -19,6 +19,20 @@
 /* Per-cpu variable to prevent redundant calls when IRQs already off */
 static DEFINE_PER_CPU(int, tracing_irq_cpu);
 
+/*
+ * Use regular trace points on architectures that implement noinstr
+ * tooling: these calls will only happen with RCU enabled, which can
+ * use a regular tracepoint.
+ *
+ * On older architectures, use the rcuidle tracing methods (which
+ * aren't NMI-safe - so exclude NMI contexts):
+ */
+#ifdef CONFIG_ARCH_WANTS_NO_INSTR
+#define trace(point)	trace_##point
+#else
+#define trace(point)	if (!in_nmi()) trace_##point##_rcuidle
+#endif
+
 /*
  * Like trace_hardirqs_on() but without the lockdep invocation. This is
  * used in the low level entry code where the ordering vs. RCU is important
@@ -28,8 +42,7 @@ static DEFINE_PER_CPU(int, tracing_irq_cpu);
 void trace_hardirqs_on_prepare(void)
 {
 	if (this_cpu_read(tracing_irq_cpu)) {
-		if (!in_nmi())
-			trace_irq_enable(CALLER_ADDR0, CALLER_ADDR1);
+		trace(irq_enable)(CALLER_ADDR0, CALLER_ADDR1);
 		tracer_hardirqs_on(CALLER_ADDR0, CALLER_ADDR1);
 		this_cpu_write(tracing_irq_cpu, 0);
 	}
@@ -40,8 +53,7 @@ NOKPROBE_SYMBOL(trace_hardirqs_on_prepare);
 void trace_hardirqs_on(void)
 {
 	if (this_cpu_read(tracing_irq_cpu)) {
-		if (!in_nmi())
-			trace_irq_enable_rcuidle(CALLER_ADDR0, CALLER_ADDR1);
+		trace(irq_enable)(CALLER_ADDR0, CALLER_ADDR1);
 		tracer_hardirqs_on(CALLER_ADDR0, CALLER_ADDR1);
 		this_cpu_write(tracing_irq_cpu, 0);
 	}
@@ -63,8 +75,7 @@ void trace_hardirqs_off_finish(void)
 	if (!this_cpu_read(tracing_irq_cpu)) {
 		this_cpu_write(tracing_irq_cpu, 1);
 		tracer_hardirqs_off(CALLER_ADDR0, CALLER_ADDR1);
-		if (!in_nmi())
-			trace_irq_disable(CALLER_ADDR0, CALLER_ADDR1);
+		trace(irq_disable)(CALLER_ADDR0, CALLER_ADDR1);
 	}
 
 }
@@ -78,8 +89,7 @@ void trace_hardirqs_off(void)
 	if (!this_cpu_read(tracing_irq_cpu)) {
 		this_cpu_write(tracing_irq_cpu, 1);
 		tracer_hardirqs_off(CALLER_ADDR0, CALLER_ADDR1);
-		if (!in_nmi())
-			trace_irq_disable_rcuidle(CALLER_ADDR0, CALLER_ADDR1);
+		trace(irq_disable)(CALLER_ADDR0, CALLER_ADDR1);
 	}
 }
 EXPORT_SYMBOL(trace_hardirqs_off);