Re: [patch] Real-Time Preemption, -RT-2.6.9-rc4-mm1-U8

[Ingo Molnar]

* Florian Schmidt <mista.tapas@xxxxxxx> wrote:

> i just wanted to let you know that with U8 i still experience the
> "pauses" i reported on U6, too. I would guess that it's some scheduler
> thing as jackd running SCHED_FIFO and all its clients (at least the
> audio threads running SCHED_FIFO) are not affected by the pauses (i
> don't see any xruns from jackd and audio processing happily goes along
> without audible dropouts).

ok.

> Also it seems that /proc/sys/kernel/trace_enabled == 1 is not the only
> thing being able to trigger the pauses. With U6 i also experienced
> them with trace_enabled == 0. I have to add though that it took quite
> a while for them to kick in (hours) after setting trace_enabled to 0.
> So my conclusion is that trace_enabled == 1 just increases the
> probability of such pauses by several magnitudes (with 1 i get about
> one of these pauses per 2-10 minutes, with 0 it took several hours for
> the first pause to occur and then they stayed less frequent than with
> 1).

i dont think it's caused by trace_enabled - the trace you sent last time
clearly showed erratic behavior. There's one piece of code i suspect in
particular - could you try the patch below ontop of -U8? (i have
compile- and boot- tested it)

	Ingo

--- linux/kernel/sched.c.orig
+++ linux/kernel/sched.c
@@ -2764,6 +2764,8 @@ need_resched:
 		else
 			deactivate_task(prev, rq);
 	}
+	if (preempt_count() & PREEMPT_ACTIVE)
+		sub_preempt_count(PREEMPT_ACTIVE);
 	if (unlikely(prev->flags & PF_DEAD)) {
 		BUG_ON(prev->state != TASK_RUNNING);
 		prev->state = __TASK_DEAD;
@@ -2940,6 +2942,7 @@ asmlinkage void __sched preempt_schedule
 		return;
 
 need_resched:
+	local_irq_disable();
 	add_preempt_count(PREEMPT_ACTIVE);
 	/*
 	 * We keep the big kernel semaphore locked, but we
@@ -2950,11 +2953,10 @@ need_resched:
 	saved_lock_depth = task->lock_depth;
 	task->lock_depth = -1;
 #endif
-	schedule();
+	__schedule();
 #ifdef CONFIG_PREEMPT_BKL
 	task->lock_depth = saved_lock_depth;
 #endif
-	sub_preempt_count(PREEMPT_ACTIVE);
 
 	/* we could miss a preemption opportunity between schedule and now */
 	barrier();
@@ -3002,7 +3004,6 @@ need_resched:
 #ifdef CONFIG_PREEMPT_BKL
 	task->lock_depth = saved_lock_depth;
 #endif
-	sub_preempt_count(PREEMPT_ACTIVE);
 
 	/* we could miss a preemption opportunity between schedule and now */
 	barrier();
@@ -3842,9 +3843,9 @@ static inline void __cond_resched(void)
 	if (preempt_count() & PREEMPT_ACTIVE)
 		return;
 	do {
+		local_irq_disable();
 		add_preempt_count(PREEMPT_ACTIVE);
-		schedule();
-		sub_preempt_count(PREEMPT_ACTIVE);
+		__schedule();
 	} while (need_resched());
 }
 
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/