Since rcu_is_watching() is notrace (and needs to be, as it can be
called from the tracers), make sure everything it in turn calls is
notrace too.
To that effect, mark rcu_dynticks_curr_cpu_in_eqs() inline, which
implies notrace, as the function is tiny.
Signed-off-by: Peter Zijlstra (Intel) <[email protected]>
---
kernel/rcu/tree.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
--- a/kernel/rcu/tree.c
+++ b/kernel/rcu/tree.c
@@ -294,7 +294,7 @@ static void rcu_dynticks_eqs_online(void
*
* No ordering, as we are sampling CPU-local information.
*/
-static bool rcu_dynticks_curr_cpu_in_eqs(void)
+static inline bool rcu_dynticks_curr_cpu_in_eqs(void)
{
struct rcu_data *rdp = this_cpu_ptr(&rcu_data);
On Wed, Feb 12, 2020 at 10:01:41PM +0100, Peter Zijlstra wrote:
> Since rcu_is_watching() is notrace (and needs to be, as it can be
> called from the tracers), make sure everything it in turn calls is
> notrace too.
>
> To that effect, mark rcu_dynticks_curr_cpu_in_eqs() inline, which
> implies notrace, as the function is tiny.
>
> Signed-off-by: Peter Zijlstra (Intel) <[email protected]>
> ---
> kernel/rcu/tree.c | 2 +-
> 1 file changed, 1 insertion(+), 1 deletion(-)
>
> --- a/kernel/rcu/tree.c
> +++ b/kernel/rcu/tree.c
> @@ -294,7 +294,7 @@ static void rcu_dynticks_eqs_online(void
> *
> * No ordering, as we are sampling CPU-local information.
> */
> -static bool rcu_dynticks_curr_cpu_in_eqs(void)
> +static inline bool rcu_dynticks_curr_cpu_in_eqs(void)
I think there are ways to turn off function inlining, such as gcc's:
-fkeep-inline-functions
And just to be sure weird compilers (clang *cough*) don't screw this up,
could we make it static inline notrace?
Build tested it on the tip tree on top of your patch:
---8<-----------------------
diff --git a/kernel/rcu/tree.c b/kernel/rcu/tree.c
index f3cb824fe5bbf..078d56951c8e7 100644
--- a/kernel/rcu/tree.c
+++ b/kernel/rcu/tree.c
@@ -294,7 +294,7 @@ static void rcu_dynticks_eqs_online(void)
*
* No ordering, as we are sampling CPU-local information.
*/
-static inline bool rcu_dynticks_curr_cpu_in_eqs(void)
+static inline notrace bool rcu_dynticks_curr_cpu_in_eqs(void)
{
struct rcu_data *rdp = this_cpu_ptr(&rcu_data);
On Wed, 12 Feb 2020 17:38:18 -0500
Joel Fernandes <[email protected]> wrote:
> I think there are ways to turn off function inlining, such as gcc's:
> -fkeep-inline-functions
>
> And just to be sure weird compilers (clang *cough*) don't screw this up,
> could we make it static inline notrace?
inline is defined as notrace, so not needed.
I did that because of surprises when functions marked as inline
suddenly became non inlined and traced, which caused issues with
function tracing (before I finally got recursion protection working).
But even then, I figured, if something is inlined and gcc actually
inlines it, it wont be traced. For consistency, if something is marked
inline, it should not be traced.
-- Steve
On Wed, Feb 12, 2020 at 8:41 PM Steven Rostedt <[email protected]> wrote:
>
> On Wed, 12 Feb 2020 17:38:18 -0500
> Joel Fernandes <[email protected]> wrote:
>
> > I think there are ways to turn off function inlining, such as gcc's:
> > -fkeep-inline-functions
> >
> > And just to be sure weird compilers (clang *cough*) don't screw this up,
> > could we make it static inline notrace?
>
> inline is defined as notrace, so not needed.
>
> I did that because of surprises when functions marked as inline
> suddenly became non inlined and traced, which caused issues with
> function tracing (before I finally got recursion protection working).
> But even then, I figured, if something is inlined and gcc actually
> inlines it, it wont be traced. For consistency, if something is marked
> inline, it should not be traced.
Ah I see it, thanks for the clarification Steve! That looks like a
good idea. I withdraw my previous comment.
- Joel