Commit beed33a8 authored Oct 11, 2006 by Nick Piggin Committed by Linus Torvalds Oct 11, 2006

[PATCH] sched: likely profiling



This likely profiling is pretty fun. I found a few possible problems
in sched.c.

This patch may be not measurable, but when I did measure long ago,
nooping (un)likely cost a couple of % on scheduler heavy benchmarks, so
it all adds up.

Tweak some branch hints:

- the 2nd 64 bits in the bitmask is likely to be populated, because it
  contains the first 28 bits (nearly 3/4) of the normal priorities.
  (ratio of 669669:691 ~= 1000:1).

- it isn't unlikely that context switching switches to another process. it
  might be very rapidly switching to and from the idle process (ratio of
  475815:419004 and 471330:423544). Let the branch predictor decide.

- preempt_enable seems to be very often called in a nested preempt_disable
  or with interrupts disabled (ratio of 3567760:87965 ~= 40:1)

Signed-off-by: Nick Piggin <npiggin@suse.de>
Acked-by: Ingo Molnar <mingo@elte.hu>
Cc: Daniel Walker <dwalker@mvista.com>
Cc: Hua Zhong <hzhong@gmail.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

parent f33d9bd5

include/asm-generic/bitops/sched.h

+1 −1

Original line number	Diff line number	Diff line
		@@ -15,7 +15,7 @@ static inline int sched_find_first_bit(const unsigned long *b)
		#if BITS_PER_LONG == 64
		if (unlikely(b[0]))
		return __ffs(b[0]);
		if (unlikely(b[1]))
		if (likely(b[1]))
		return __ffs(b[1]) + 64;
		return __ffs(b[2]) + 128;
		#elif BITS_PER_LONG == 32

kernel/sched.c

+3 −3

Original line number	Diff line number	Diff line
		@@ -1822,14 +1822,14 @@ context_switch(struct rq rq, struct task_struct prev,
		struct mm_struct *mm = next->mm;
		struct mm_struct *oldmm = prev->active_mm;

		if (unlikely(!mm)) {
		if (!mm) {
		next->active_mm = oldmm;
		atomic_inc(&oldmm->mm_count);
		enter_lazy_tlb(oldmm, next);
		} else
		switch_mm(oldmm, mm, next);

		if (unlikely(!prev->mm)) {
		if (!prev->mm) {
		prev->active_mm = NULL;
		WARN_ON(rq->prev_mm);
		rq->prev_mm = oldmm;
		@@ -3491,7 +3491,7 @@ asmlinkage void __sched preempt_schedule(void)
		* If there is a non-zero preempt_count or interrupts are disabled,
		* we do not want to preempt the current task. Just return..
		*/
		if (unlikely(ti->preempt_count \|\| irqs_disabled()))
		if (likely(ti->preempt_count \|\| irqs_disabled()))
		return;

		need_resched:

Admin message