[PATCH] Removed unnecessary use of local variable in v3.12.

[PATCH] Removed unnecessary use of local variable in v3.12.

[Meenakshi Aggarwal]

This patch remove use of temporary variable 'ip' which optimizes the code and results in
reduction of assembly instructions and hence improves latency.

Benefit of 18 lines of assembly instructions is observed on e6500 core.

Signed-off-by: Meenakshi Aggarwal <meenakshi.aggarwal@freescale.com>
---
 kernel/sched/core.c |   13 ++++++++-----
 1 file changed, 8 insertions(+), 5 deletions(-)

diff --git a/kernel/sched/core.c b/kernel/sched/core.c
index 8749d20..cdf43f8 100644
--- a/kernel/sched/core.c
+++ b/kernel/sched/core.c
@@ -2296,13 +2296,16 @@ void __kprobes add_preempt_count(int val)
 	DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >=
 				PREEMPT_MASK - 10);
 #endif
-	if (preempt_count() == val) {
-		unsigned long ip = get_parent_ip(CALLER_ADDR1);
+
+	if (preempt_count() == val)
 #ifdef CONFIG_DEBUG_PREEMPT
-		current->preempt_disable_ip = ip;
+		trace_preempt_off(CALLER_ADDR0,
+			current->preempt_disable_ip =
+			get_parent_ip(CALLER_ADDR1));
+#else
+		trace_preempt_off(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1));
 #endif
-		trace_preempt_off(CALLER_ADDR0, ip);
-	}
+
 }
 EXPORT_SYMBOL(add_preempt_count);
 
-- 
1.7.10.4

Re: [PATCH] Removed unnecessary use of local variable in v3.12.

[Sebastian Andrzej Siewior]

* Meenakshi Aggarwal | 2014-05-15 16:08:54 [+0530]:

>--- a/kernel/sched/core.c
>+++ b/kernel/sched/core.c
>@@ -2296,13 +2296,16 @@ void __kprobes add_preempt_count(int val)
> 	DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >=
> 				PREEMPT_MASK - 10);
> #endif
>-	if (preempt_count() == val) {
>-		unsigned long ip = get_parent_ip(CALLER_ADDR1);
>+
>+	if (preempt_count() == val)
> #ifdef CONFIG_DEBUG_PREEMPT
>-		current->preempt_disable_ip = ip;
>+		trace_preempt_off(CALLER_ADDR0,
>+			current->preempt_disable_ip =
>+			get_parent_ip(CALLER_ADDR1));

So if I apply this, I can imagine tglx will run to his armoury looking
for a fine axe and then throwing it at me. So sorry, but I can't apply
this.
I suggest you look into gcc and figure out why this hackery of yours
saves 18 instructions. It does not make sense to me because all you do
is to inline replace ip with "get_parent_ip(CALLER_ADDR1)". It might be
that the compiler unrolls the stack each time ip is accessed but this is
clearly a compiler short comming.

Sebastian