Messages in this thread Patch in this message | | | Date | Tue, 21 Jul 2015 13:07:00 +0200 | From | Peter Zijlstra <> | Subject | Re: [tip:sched/core] sched/preempt: Optimize preemption operations on __schedule() callers |
| |
On Mon, Jul 20, 2015 at 06:14:27PM -0400, Sasha Levin wrote: > On 05/19/2015 03:17 AM, tip-bot for Frederic Weisbecker wrote: > > Commit-ID: b30f0e3ffedfa52b1d67a302ae5860c49998e5e2
> > sched/preempt: Optimize preemption operations on __schedule() callers > >
> I was seeing RCU stall warnings that appeared to have different backtrace > each time, but would reliably trigger when fuzzing and looked something like this:
> [4394886.426627] preempt_schedule_irq (./arch/x86/include/asm/paravirt.h:807 kernel/sched/core.c:3218)
The only 'obvious' difference there seems to be something like the below. Prior to the patch preempt_schedule_irq() used __preempt_count_{add,sub}() which are the !tracing versions.
Does this make it go again? If not, can you provide a .config?
--- include/linux/preempt.h | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)
diff --git a/include/linux/preempt.h b/include/linux/preempt.h index 84991f185173..235a9f2d76fe 100644 --- a/include/linux/preempt.h +++ b/include/linux/preempt.h @@ -139,14 +139,14 @@ extern void preempt_count_sub(int val); #define preempt_active_enter() \ do { \ - preempt_count_add(PREEMPT_ACTIVE + PREEMPT_DISABLE_OFFSET); \ + __preempt_count_add(PREEMPT_ACTIVE + PREEMPT_DISABLE_OFFSET); \ barrier(); \ } while (0) #define preempt_active_exit() \ do { \ barrier(); \ - preempt_count_sub(PREEMPT_ACTIVE + PREEMPT_DISABLE_OFFSET); \ + __preempt_count_sub(PREEMPT_ACTIVE + PREEMPT_DISABLE_OFFSET); \ } while (0) #ifdef CONFIG_PREEMPT_COUNT
| |