Skip to contentSkip to conversation

fix smt nice lock contention and optimization

PATCH In review Chen, Kenneth W 6 participants 11 messages Jun 3, 2006

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 fix smt nice lock contention and optimization

OK, final rolled up patch with everyone's changes. I fixed one bug introduced by Con's earlier patch that there is an unpaired spin_trylock/spin_unlock in the for loop of dependent_sleeper(). Chris, Con, Nick - please review and provide your signed-off-by line. Andrew - please consider for -mm inclusion. Thanks.

[patch] fix smt nice lock contention and optimization

Initial report and lock contention fix from Chris Mason:

Recent benchmarks showed some performance regressions between 2.6.16 and 2.6.5. We tracked down one of the regressions to lock contention in schedule heavy workloads (~70,000 context switches per second)

kernel/sched.c:dependent_sleeper() was responsible for most of the lock contention, hammering on the run queue locks. The patch below is more of a discussion point than a suggested fix (although it does reduce lock contention significantly). The dependent_sleeper code looks very expensive to me, especially for using a spinlock to bounce control between two different siblings in the same cpu.

It is further optimized:

* perform dependent_sleeper check after next task is determined
* convert wake_sleeping_dependent to use trylock
* skip smt runqueue check if trylock fails
* optimize double_rq_lock now that smt nice is converted to trylock
* early exit in searching first SD_SHARE_CPUPOWER domain
* speedup fast path of dependent_sleeper

---

 sched.c |  168 ++++++++++++++++++----------------------------------------------
 1 files changed, 48 insertions(+), 120 deletions(-)
kernel/sched.c +48 -120
diff -Nurp 2.6.17-rc5-mm2/kernel/sched.c ken/kernel/sched.c
@@ -248,7 +248,6 @@ struct runqueue {
 
 	task_t *migration_thread;
 	struct list_head migration_queue;
-	int cpu;
 #endif
 
 #ifdef CONFIG_SCHEDSTATS
@@ -1887,9 +1886,6 @@ unsigned long nr_active(void)
 /*
  * double_rq_lock - safely lock two runqueues
  *
- * We must take them in cpu order to match code in
- * dependent_sleeper and wake_dependent_sleeper.
- *
  * Note this does not disable interrupts like task_rq_lock,
  * you need to do so manually before calling.
  */
@@ -1901,7 +1897,7 @@ static void double_rq_lock(runqueue_t *r
 		spin_lock(&rq1->lock);
 		__acquire(rq2->lock);	/* Fake it out ;) */
 	} else {
-		if (rq1->cpu < rq2->cpu) {
+		if (rq1 < rq2) {
 			spin_lock(&rq1->lock);
 			spin_lock(&rq2->lock);
 		} else {
@@ -1937,7 +1933,7 @@ static void double_lock_balance(runqueue
 	__acquires(this_rq->lock)
 {
 	if (unlikely(!spin_trylock(&busiest->lock))) {
-		if (busiest->cpu < this_rq->cpu) {
+		if (busiest < this_rq) {
 			spin_unlock_non_nested(&this_rq->lock);
 			spin_lock(&busiest->lock);
 			spin_lock(&this_rq->lock);
@@ -2969,48 +2965,33 @@ static inline void wakeup_busy_runqueue(
 		resched_task(rq->idle);
 }
 
-static void wake_sleeping_dependent(int this_cpu, runqueue_t *this_rq)
+/*
+ * Called with interrupt disabled and this_rq's runqueue locked.
+ */
+static void wake_sleeping_dependent(int this_cpu)
 {
 	struct sched_domain *tmp, *sd = NULL;
-	cpumask_t sibling_map;
 	int i;
 
 	for_each_domain(this_cpu, tmp)
-		if (tmp->flags & SD_SHARE_CPUPOWER)
+		if (tmp->flags & SD_SHARE_CPUPOWER) {
 			sd = tmp;
-
+			break;
+		}
 	if (!sd)
 		return;
 
-	/*
-	 * Unlock the current runqueue because we have to lock in
-	 * CPU order to avoid deadlocks. Caller knows that we might
-	 * unlock. We keep IRQs disabled.
-	 */
-	spin_unlock(&this_rq->lock);
-
-	sibling_map = sd->span;
-
-	for_each_cpu_mask(i, sibling_map)
-		spin_lock(&cpu_rq(i)->lock);
-	/*
-	 * We clear this CPU from the mask. This both simplifies the
-	 * inner loop and keps this_rq locked when we exit:
-	 */
-	cpu_clear(this_cpu, sibling_map);
-
-	for_each_cpu_mask(i, sibling_map) {
+	for_each_cpu_mask(i, sd->span) {
 		runqueue_t *smt_rq = cpu_rq(i);
 
+		if (i == this_cpu)
+			continue;
+		if (unlikely(!spin_trylock(&smt_rq->lock)))
+			continue;
+
 		wakeup_busy_runqueue(smt_rq);
+		spin_unlock(&smt_rq->lock);
 	}
-
-	for_each_cpu_mask(i, sibling_map)
-		spin_unlock_non_nested(&cpu_rq(i)->lock);
-	/*
-	 * We exit with this_cpu's rq still held and IRQs
-	 * still disabled:
-	 */
 }
 
 /*
@@ -3023,52 +3004,44 @@ static inline unsigned long smt_slice(ta
 	return p->time_slice * (100 - sd->per_cpu_gain) / 100;
 }
 
-static int dependent_sleeper(int this_cpu, runqueue_t *this_rq)
+/*
+ * To minimise lock contention and not have to drop this_rq's runlock we only
+ * trylock the sibling runqueues and bypass those runqueues if we fail to
+ * acquire their lock. As we only trylock the normal locking order does not
+ * need to be obeyed.
+ */
+static int dependent_sleeper(int this_cpu, runqueue_t *this_rq, task_t *p)
 {
 	struct sched_domain *tmp, *sd = NULL;
-	cpumask_t sibling_map;
-	prio_array_t *array;
 	int ret = 0, i;
-	task_t *p;
+
+	/* kernel/rt threads do not participate in dependent sleeping */
+	if (!p->mm || rt_task(p))
+		return 0;
 
 	for_each_domain(this_cpu, tmp)
-		if (tmp->flags & SD_SHARE_CPUPOWER)
+		if (tmp->flags & SD_SHARE_CPUPOWER) {
 			sd = tmp;
-
+			break;
+		}
 	if (!sd)
 		return 0;
 
-	/*
-	 * The same locking rules and details apply as for
-	 * wake_sleeping_dependent():
-	 */
-	spin_unlock_non_nested(&this_rq->lock);
-	sibling_map = sd->span;
-	for_each_cpu_mask(i, sibling_map)
-		spin_lock(&cpu_rq(i)->lock);
-	cpu_clear(this_cpu, sibling_map);
+	for_each_cpu_mask(i, sd->span) {
+		runqueue_t *smt_rq;
+		task_t *smt_curr;
 
-	/*
-	 * Establish next task to be run - it might have gone away because
-	 * we released the runqueue lock above:
-	 */
-	if (!this_rq->nr_running)
-		goto out_unlock;
-	array = this_rq->active;
-	if (!array->nr_active)
-		array = this_rq->expired;
-	BUG_ON(!array->nr_active);
+		if (i == this_cpu)
+			continue;
 
-	p = list_entry(array->queue[sched_find_first_bit(array->bitmap)].next,
-		task_t, run_list);
+		smt_rq = cpu_rq(i);
+		if (unlikely(!spin_trylock(&smt_rq->lock)))
+			continue;
 
-	for_each_cpu_mask(i, sibling_map) {
-		runqueue_t *smt_rq = cpu_rq(i);
-		task_t *smt_curr = smt_rq->curr;
+		smt_curr = smt_rq->curr;
 
-		/* Kernel threads do not participate in dependent sleeping */
-		if (!p->mm || !smt_curr->mm || rt_task(p))
-			goto check_smt_task;
+		if (!smt_curr->mm)
+			goto unlock;
 
 		/*
 		 * If a user task with lower static priority than the
@@ -3091,44 +3064,17 @@ static int dependent_sleeper(int this_cp
 				!TASK_PREEMPTS_CURR(p, smt_rq) &&
 				smt_slice(smt_curr, sd) > task_timeslice(p))
 					ret = 1;
-
-check_smt_task:
-		if ((!smt_curr->mm && smt_curr != smt_rq->idle) ||
-			rt_task(smt_curr))
-				continue;
-		if (!p->mm) {
-			wakeup_busy_runqueue(smt_rq);
-			continue;
-		}
-
-		/*
-		 * Reschedule a lower priority task on the SMT sibling for
-		 * it to be put to sleep, or wake it up if it has been put to
-		 * sleep for priority reasons to see if it should run now.
-		 */
-		if (rt_task(p)) {
-			if ((jiffies % DEF_TIMESLICE) >
-				(sd->per_cpu_gain * DEF_TIMESLICE / 100))
-					resched_task(smt_curr);
-		} else {
-			if (TASK_PREEMPTS_CURR(p, smt_rq) &&
-				smt_slice(p, sd) > task_timeslice(smt_curr))
-					resched_task(smt_curr);
-			else
-				wakeup_busy_runqueue(smt_rq);
-		}
+unlock:
+		spin_unlock(&smt_rq->lock);
 	}
-out_unlock:
-	for_each_cpu_mask(i, sibling_map)
-		spin_unlock_non_nested(&cpu_rq(i)->lock);
 	return ret;
 }
 #else
-static inline void wake_sleeping_dependent(int this_cpu, runqueue_t *this_rq)
+static inline void wake_sleeping_dependent(int this_cpu)
 {
 }
 
-static inline int dependent_sleeper(int this_cpu, runqueue_t *this_rq)
+static inline int dependent_sleeper(int this_cpu, runqueue_t *this_rq, task_t *p)
 {
 	return 0;
 }
@@ -3255,32 +3201,13 @@ need_resched_nonpreemptible:
 
 	cpu = smp_processor_id();
 	if (unlikely(!rq->nr_running)) {
-go_idle:
 		idle_balance(cpu, rq);
 		if (!rq->nr_running) {
 			next = rq->idle;
 			rq->expired_timestamp = 0;
-			wake_sleeping_dependent(cpu, rq);
-			/*
-			 * wake_sleeping_dependent() might have released
-			 * the runqueue, so break out if we got new
-			 * tasks meanwhile:
-			 */
-			if (!rq->nr_running)
-				goto switch_tasks;
-		}
-	} else {
-		if (dependent_sleeper(cpu, rq)) {
-			next = rq->idle;
+			wake_sleeping_dependent(cpu);
 			goto switch_tasks;
 		}
-		/*
-		 * dependent_sleeper() releases and reacquires the runqueue
-		 * lock, hence go into the idle loop if the rq went
-		 * empty meanwhile:
-		 */
-		if (unlikely(!rq->nr_running))
-			goto go_idle;
 	}
 
 	array = rq->active;
@@ -3318,6 +3245,8 @@ go_idle:
 		}
 	}
 	next->sleep_type = SLEEP_NORMAL;
+	if (dependent_sleeper(cpu, rq, next))
+		next = rq->idle;
 switch_tasks:
 	if (next == rq->idle)
 		schedstat_inc(rq, sched_goidle);
@@ -6666,7 +6595,6 @@ void __init sched_init(void)
 		rq->push_cpu = 0;
 		rq->migration_thread = NULL;
 		INIT_LIST_HEAD(&rq->migration_queue);
-		rq->cpu = i;
 #endif
 		atomic_set(&rq->nr_iowait, 0);
 
#2 lore raw
Quoted 6 lines · On Sat, 3 Jun 2006 09:49:20 +0200 Ingo Molnar <[email protected]> wrote:
Quoted 3 lines · Chen, Kenneth W

---

Quoted 9 lines
Quoted 2 lines
 sched.c |  168 ++++++++++++++++++----------------------------------------------
 1 files changed, 48 insertions(+), 120 deletions(-)

looks really good now to me.

lets try it in -mm?

Yup. I redid Ken's patch against mainline and them mangled lock-validator-special-locking-schedc.patch to suit.

#3 lore raw

Andrew Morton wrote on Saturday, June 03, 2006 1:11 AM

Quoted 10 lines
Quoted 6 lines · Ingo Molnar

looks really good now to me.

lets try it in -mm?

Yup. I redid Ken's patch against mainline and them mangled lock-validator-special-locking-schedc.patch to suit.

Hmm, wish I knew this beforehand, so that I won't spend extra 1/2 hour to port the patch to -mm and only to have you convert it back to mainline. I could just post the version I originally had against the mainline.

- Ken

#4 lore raw
Quoted 9 lines · Kenneth W

Andrew Morton wrote on Saturday, June 03, 2006 1:11 AM

Quoted 8 lines
Quoted 6 lines · Ingo Molnar

looks really good now to me.

lets try it in -mm?

Quoted 8 lines
Quoted 2 lines

Yup. I redid Ken's patch against mainline and them mangled lock-validator-special-locking-schedc.patch to suit.

Hmm, wish I knew this beforehand, so that I won't spend extra 1/2 hour to port the patch to -mm and only to have you convert it back to mainline. I could just post the version I originally had against the mainline.

Unless you're specifically working against code which is only in -mm (or a subsystem tree), please work against mainline.

#5 lore raw
In reply to Chen, Kenneth W
Quoted 5 lines · Kenneth W

OK, final rolled up patch with everyone's changes. I fixed one bug introduced by Con's earlier patch that there is an unpaired spin_trylock/spin_unlock in the for loop of dependent_sleeper(). Chris, Con, Nick - please review and provide your signed-off-by line. Andrew - please consider for -mm inclusion. Thanks.

Looks good. Just one style nitpick.

kernel/sched.c static inline void wakeup_busy_runqueue( · line 2980

Quoted 7 lines
kernel/sched.c +3 -2
 	for_each_domain(this_cpu, tmp)
-		if (tmp->flags & SD_SHARE_CPUPOWER)
+		if (tmp->flags & SD_SHARE_CPUPOWER) {
 			sd = tmp;
-
+			break;
+		}

Could we make this neater with extra braces such as:

 	for_each_domain(this_cpu, tmp) {
		if (tmp->flags & SD_SHARE_CPUPOWER) {
 			sd = tmp;
			break;
		}
	}

and same for the other uses of for_each ? I know it's redundant but it's neater IMO when there are multiple lines of code below it.

Signature · 1 line
-ck
#6 lore raw
Quoted 11 lines · Con Kolivas

Could we make this neater with extra braces such as:

 	for_each_domain(this_cpu, tmp) {
		if (tmp->flags & SD_SHARE_CPUPOWER) {
 			sd = tmp;
			break;
		}
	}

and same for the other uses of for_each ? I know it's redundant but it's neater IMO when there are multiple lines of code below it.

yep, that's the preferred style when there are multiple lines below a loop.

Ingo

#7 lore raw

Ingo Molnar wrote on Saturday, June 03, 2006 12:58 AM

Quoted 11 lines · Con Kolivas
Quoted 10 lines

Could we make this neater with extra braces such as:

 	for_each_domain(this_cpu, tmp) {
		if (tmp->flags & SD_SHARE_CPUPOWER) {
 			sd = tmp;
			break;
		}
	}
Quoted 5 lines
Quoted 2 lines

and same for the other uses of for_each ? I know it's redundant but it's neater IMO when there are multiple lines of code below it.

yep, that's the preferred style when there are multiple lines below a loop.

OK, thanks for the tips. Here is an incremental coding-style fix:

kernel/sched.c renamed from kernel/sched.c.orig +4 -2
@@ -2973,11 +2973,12 @@
 	struct sched_domain *tmp, *sd = NULL;
 	int i;
 
-	for_each_domain(this_cpu, tmp)
+	for_each_domain(this_cpu, tmp) {
 		if (tmp->flags & SD_SHARE_CPUPOWER) {
 			sd = tmp;
 			break;
 		}
+	}
 	if (!sd)
 		return;
 
@@ -3019,11 +3020,12 @@
 	if (!p->mm || rt_task(p))
 		return 0;
 
-	for_each_domain(this_cpu, tmp)
+	for_each_domain(this_cpu, tmp) {
 		if (tmp->flags & SD_SHARE_CPUPOWER) {
 			sd = tmp;
 			break;
 		}
+	}
 	if (!sd)
 		return 0;
 
#8 lore raw
Quoted 12 lines · Kenneth W

Ingo Molnar wrote on Saturday, June 03, 2006 12:58 AM

Quoted 10 lines · Con Kolivas
Quoted 9 lines

Could we make this neater with extra braces such as:

 	for_each_domain(this_cpu, tmp) {
		if (tmp->flags & SD_SHARE_CPUPOWER) {
 			sd = tmp;
			break;
		}
	}
Quoted 7 lines
Quoted 5 lines
Quoted 2 lines

and same for the other uses of for_each ? I know it's redundant but it's neater IMO when there are multiple lines of code below it.

yep, that's the preferred style when there are multiple lines below a loop.

OK, thanks for the tips. Here is an incremental coding-style fix:

Great!

Thanks Chris, Nick and Ken for your input.

for both

Signature · 1 line
-ck
#9 lore raw
In reply to Chen, Kenneth W
Quoted 5 lines · Kenneth W

OK, final rolled up patch with everyone's changes. I fixed one bug introduced by Con's earlier patch that there is an unpaired spin_trylock/spin_unlock in the for loop of dependent_sleeper(). Chris, Con, Nick - please review and provide your signed-off-by line. Andrew - please consider for -mm inclusion. Thanks.

Thanks Ken, you can add a Signed-off-by: Nick Piggin <[email protected]> for my part.

Signature · 2 lines
SUSE Labs, Novell Inc.
Send instant messages to your online friends http://au.messenger.yahoo.com 
#10 lore raw
In reply to Chen, Kenneth W
Quoted 5 lines · Kenneth W

OK, final rolled up patch with everyone's changes. I fixed one bug introduced by Con's earlier patch that there is an unpaired spin_trylock/spin_unlock in the for loop of dependent_sleeper(). Chris, Con, Nick - please review and provide your signed-off-by line. Andrew - please consider for -mm inclusion. Thanks.

Thanks for turning my half baked code into something nice. acked-by from me as well.

-chris