[PATCH] rcu: Do not trace rcu_is_watching() functions

[PATCH] rcu: Do not trace rcu_is_watching() functions

[Steven Rostedt]

Paul,

As I added the rcu_is_watching() to the ftrace code to prevent perf
from tracing functions where RCU is not watching, I noticed that the
rcu_is_watching() functions can themselves be traced. I would say this
is a helper function and not something that we need to trace, as it can
cause a slowdown in function tracing (any recursion is detected, so it
only affects performance).

I marked the functions with notrace. You can take this patch and push
it for 3.13 or I can take it if you give me an "Acked-by". As it only
affects performance and does not crash the code it's not a must have in
my tree.

If you take it, I can write up a better change log too.

-- Steve

Signed-off-by: Steven Rostedt <rostedt@goodmis.org>

diff --git a/kernel/rcutiny.c b/kernel/rcutiny.c
index 0fa061d..590c8a8 100644
--- a/kernel/rcutiny.c
+++ b/kernel/rcutiny.c
@@ -179,7 +179,7 @@ EXPORT_SYMBOL_GPL(rcu_irq_enter);
 /*
  * Test whether RCU thinks that the current CPU is idle.
  */
-bool __rcu_is_watching(void)
+bool notrace __rcu_is_watching(void)
 {
 	return rcu_dynticks_nesting;
 }
diff --git a/kernel/rcutree.c b/kernel/rcutree.c
index 981d0c1..499bb2b 100644
--- a/kernel/rcutree.c
+++ b/kernel/rcutree.c
@@ -662,7 +662,7 @@ void rcu_nmi_exit(void)
  * rcu_is_watching(), the caller of __rcu_is_watching() must have at
  * least disabled preemption.
  */
-bool __rcu_is_watching(void)
+bool notrace __rcu_is_watching(void)
 {
 	return atomic_read(this_cpu_ptr(&rcu_dynticks.dynticks)) & 0x1;
 }
@@ -673,7 +673,7 @@ bool __rcu_is_watching(void)
  * If the current CPU is in its idle loop and is neither in an interrupt
  * or NMI handler, return true.
  */
-bool rcu_is_watching(void)
+bool notrace rcu_is_watching(void)
 {
 	int ret;
 

Re: [PATCH] rcu: Do not trace rcu_is_watching() functions

[Paul E. McKenney]

On Mon, Nov 04, 2013 at 08:27:36PM -0500, Steven Rostedt wrote:
> 
> Paul,
> 
> As I added the rcu_is_watching() to the ftrace code to prevent perf
> from tracing functions where RCU is not watching, I noticed that the
> rcu_is_watching() functions can themselves be traced. I would say this
> is a helper function and not something that we need to trace, as it can
> cause a slowdown in function tracing (any recursion is detected, so it
> only affects performance).
> 
> I marked the functions with notrace. You can take this patch and push
> it for 3.13 or I can take it if you give me an "Acked-by". As it only
> affects performance and does not crash the code it's not a must have in
> my tree.
> 
> If you take it, I can write up a better change log too.
> 
> -- Steve
> 
> Signed-off-by: Steven Rostedt <rostedt@goodmis.org>

Given an improved commit log:

Acked-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>

> diff --git a/kernel/rcutiny.c b/kernel/rcutiny.c
> index 0fa061d..590c8a8 100644
> --- a/kernel/rcutiny.c
> +++ b/kernel/rcutiny.c
> @@ -179,7 +179,7 @@ EXPORT_SYMBOL_GPL(rcu_irq_enter);
>  /*
>   * Test whether RCU thinks that the current CPU is idle.
>   */
> -bool __rcu_is_watching(void)
> +bool notrace __rcu_is_watching(void)
>  {
>  	return rcu_dynticks_nesting;
>  }
> diff --git a/kernel/rcutree.c b/kernel/rcutree.c
> index 981d0c1..499bb2b 100644
> --- a/kernel/rcutree.c
> +++ b/kernel/rcutree.c
> @@ -662,7 +662,7 @@ void rcu_nmi_exit(void)
>   * rcu_is_watching(), the caller of __rcu_is_watching() must have at
>   * least disabled preemption.
>   */
> -bool __rcu_is_watching(void)
> +bool notrace __rcu_is_watching(void)
>  {
>  	return atomic_read(this_cpu_ptr(&rcu_dynticks.dynticks)) & 0x1;
>  }
> @@ -673,7 +673,7 @@ bool __rcu_is_watching(void)
>   * If the current CPU is in its idle loop and is neither in an interrupt
>   * or NMI handler, return true.
>   */
> -bool rcu_is_watching(void)
> +bool notrace rcu_is_watching(void)
>  {
>  	int ret;
> 
>