Re: [RFC] [PATCH 1/1] hrtimers: Cache next hrtimer

[Ashwin Chaugule]

Thomas Gleixner wrote:

> Hmm. That's related to NOHZ I guess. Ok, that's a serious issue and we
> need to look at that.
>   
Yes. I'm running with NOHZ.

> > So, you suggest checking the ktime of the hrtimer thats about to expire and
> > compare it with expires_next ?
> >     
>
> What's wrong with that ?
>   
Nothing :) 

Just didn't know the following could have the same effect. (base->offset is confusing)

+	expires = ktime_sub(hrtimer_get_expires(timer), base->offset);
+	if (base->cpu_base->expires_next.tv64 == expires.tv64)



> Can you reevaluate against my patch please ?
>
>   

The avg. startup time with your patch came to: 26.4 sec (10 runs) as against 25.8 sec (my patch).

To calculate the hit ratio, I made some changes to your code as shown below.

diff --git a/kernel/hrtimer.c b/kernel/hrtimer.c
index 49da79a..91d099c 100644
--- a/kernel/hrtimer.c
+++ b/kernel/hrtimer.c
@@ -906,19 +906,30 @@ static void __remove_hrtimer(struct hrtimer *timer,
			     struct hrtimer_clock_base *base,
			     unsigned long newstate, int reprogram)
{
-	if (timer->state & HRTIMER_STATE_ENQUEUED) {
-		/*
-		 * Remove the timer from the rbtree and replace the
-		 * first entry pointer if necessary.
-		 */
-		if (base->first == &timer->node) {
-			base->first = rb_next(&timer->node);
-			/* Reprogram the clock event device. if enabled */
-			if (reprogram && hrtimer_hres_active())
-				hrtimer_force_reprogram(base->cpu_base);
-		}
-		rb_erase(&timer->node, &base->active);
-	}
+	ktime_t expires;
+
+	if (!(timer->state & HRTIMER_STATE_ENQUEUED))
+		goto out;
+
+	/*
+	 * Remove the timer from the rbtree and replace the first
+	 * entry pointer if necessary.
+	 */
+	rb_erase(&timer->node, &base->active);
+
+	if (base->first != &timer->node)
+		goto out;
+
+	base->first = rb_next(&timer->node);
+	/* Reprogram the clock event device. if enabled */
+	if (!reprogram || !hrtimer_hres_active())
+		goto out;

	else
		timer->total_calls++

+
+	expires = ktime_sub(hrtimer_get_expires(timer), base->offset);
+	if (base->cpu_base->expires_next.tv64 == expires.tv64)
		{

			timer->cache_hits++

			hrtimer_force_reprogram(base->cpu_base);
		}
+
+out:
	timer->state = newstate;
}



So basically, total_count is the number of times the reprogram would have been forced, cache_hit is number of times it is reduced to.
In my patch I had these counters as follows:


@@ -858,10 +858,18 @@ static void __remove_hrtimer(struct hrtimer *timer,
		 */
		if (base->first == &timer->node) {
			base->first = rb_next(&timer->node);
+#ifdef CONFIG_TIMER_STATS
+			if (reprogram && hrtimer_hres_active())
+				timer->total_calls++;
+#endif
			if (next_hrtimer == timer) {
				/* Reprogram the clock event device. if enabled */
-				if (reprogram && hrtimer_hres_active())
+				if (reprogram && hrtimer_hres_active()) {
					hrtimer_force_reprogram(base->cpu_base);
+#ifdef CONFIG_TIMER_STATS
+					timer->cache_hits++;
+#endif
+				}


Both counters are init'd to 0 in hrtimer_init_hres(timer).

Your patch looks perfect but, total_calls is always equal to cache_hits ?! IOW, the timer we're removing is always the next one to expire, hence we see no benefit.



Cheers,
Ashwin





--
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/