sched: Remove avg_wakeup

Testing the load which led to this heuristic (nfs4 kbuild) shows that it has outlived it's usefullness. With intervening load balancing changes, I cannot see any difference with/without, so recover there fastpath cycles. Signed-off-by: Mike Galbraith <efault@gmx.de> Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl> LKML-Reference: <1268301062.6785.29.camel@marge.simson.net> Signed-off-by: Ingo Molnar <mingo@elte.hu>

sched: Remove avg_wakeup
Testing the load which led to this heuristic (nfs4 kbuild) shows that it has outlived it's usefullness. With intervening load balancing changes, I cannot see any difference with/without, so recover there fastpath cycles. Signed-off-by: Mike Galbraith <efault@gmx.de> Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl> LKML-Reference: <1268301062.6785.29.camel@marge.simson.net> Signed-off-by: Ingo Molnar <mingo@elte.hu>
Mike Galbraith · Ingo Molnar
1 parent 39c0cbe215
Showing 5 changed files with 4 additions and 63 deletions Side-by-side Diff
include/linux/sched.h
kernel/sched.c
kernel/sched_debug.c
kernel/sched_fair.c
kernel/sched_features.h
@@ -1185,9 +1185,6 @@
  
 	u64			nr_migrations;
  
-	u64			start_runtime;
-	u64			avg_wakeup;
-
 #ifdef CONFIG_SCHEDSTATS
 	struct sched_statistics statistics;
 #endif
@@ -1880,9 +1880,6 @@
 static void
 enqueue_task(struct rq *rq, struct task_struct *p, int wakeup, bool head)
 {
-	if (wakeup)
-		p->se.start_runtime = p->se.sum_exec_runtime;
-
 	sched_info_queued(p);
 	p->sched_class->enqueue_task(rq, p, wakeup, head);
 	p->se.on_rq = 1;
  
@@ -1890,17 +1887,11 @@
  
 static void dequeue_task(struct rq *rq, struct task_struct *p, int sleep)
 {
-	if (sleep) {
-		if (p->se.last_wakeup) {
-			update_avg(&p->se.avg_overlap,
-				p->se.sum_exec_runtime - p->se.last_wakeup);
-			p->se.last_wakeup = 0;
-		} else {
-			update_avg(&p->se.avg_wakeup,
-				sysctl_sched_wakeup_granularity);
-		}
+	if (sleep && p->se.last_wakeup) {
+		update_avg(&p->se.avg_overlap,
+			p->se.sum_exec_runtime - p->se.last_wakeup);
+		p->se.last_wakeup = 0;
 	}
-
 	sched_info_dequeued(p);
 	p->sched_class->dequeue_task(rq, p, sleep);
 	p->se.on_rq = 0;
  
@@ -2466,14 +2457,7 @@
 	 */
 	if (!in_interrupt()) {
 		struct sched_entity *se = &current->se;
-		u64 sample = se->sum_exec_runtime;
  
-		if (se->last_wakeup)
-			sample -= se->last_wakeup;
-		else
-			sample -= se->start_runtime;
-		update_avg(&se->avg_wakeup, sample);
-
 		se->last_wakeup = se->sum_exec_runtime;
 	}
  
@@ -2540,8 +2524,6 @@
 	p->se.nr_migrations		= 0;
 	p->se.last_wakeup		= 0;
 	p->se.avg_overlap		= 0;
-	p->se.start_runtime		= 0;
-	p->se.avg_wakeup		= sysctl_sched_wakeup_granularity;
  
 #ifdef CONFIG_SCHEDSTATS
 	memset(&p->se.statistics, 0, sizeof(p->se.statistics));
@@ -408,7 +408,6 @@
 	PN(se.vruntime);
 	PN(se.sum_exec_runtime);
 	PN(se.avg_overlap);
-	PN(se.avg_wakeup);
  
 	nr_switches = p->nvcsw + p->nivcsw;
  
@@ -1592,41 +1592,10 @@
 }
 #endif /* CONFIG_SMP */
  
-/*
- * Adaptive granularity
- *
- * se->avg_wakeup gives the average time a task runs until it does a wakeup,
- * with the limit of wakeup_gran -- when it never does a wakeup.
- *
- * So the smaller avg_wakeup is the faster we want this task to preempt,
- * but we don't want to treat the preemptee unfairly and therefore allow it
- * to run for at least the amount of time we'd like to run.
- *
- * NOTE: we use 2*avg_wakeup to increase the probability of actually doing one
- *
- * NOTE: we use *nr_running to scale with load, this nicely matches the
- *       degrading latency on load.
- */
 static unsigned long
-adaptive_gran(struct sched_entity *curr, struct sched_entity *se)
-{
-	u64 this_run = curr->sum_exec_runtime - curr->prev_sum_exec_runtime;
-	u64 expected_wakeup = 2*se->avg_wakeup * cfs_rq_of(se)->nr_running;
-	u64 gran = 0;
-
-	if (this_run < expected_wakeup)
-		gran = expected_wakeup - this_run;
-
-	return min_t(s64, gran, sysctl_sched_wakeup_granularity);
-}
-
-static unsigned long
 wakeup_gran(struct sched_entity *curr, struct sched_entity *se)
 {
 	unsigned long gran = sysctl_sched_wakeup_granularity;
-
-	if (cfs_rq_of(curr)->curr && sched_feat(ADAPTIVE_GRAN))
-		gran = adaptive_gran(curr, se);
  
 	/*
 	 * Since its curr running now, convert the gran from real-time
@@ -31,12 +31,6 @@
 SCHED_FEAT(WAKEUP_PREEMPT, 1)
  
 /*
- * Compute wakeup_gran based on task behaviour, clipped to
- *  [0, sched_wakeup_gran_ns]
- */
-SCHED_FEAT(ADAPTIVE_GRAN, 1)
-
-/*
  * When converting the wakeup granularity to virtual time, do it such
  * that heavier tasks preempting a lighter task have an edge.
  */
...	...	@@ -1185,9 +1185,6 @@
1185	1185
1186	1186	u64 nr_migrations;
1187	1187
1188		- u64 start_runtime;
1189		- u64 avg_wakeup;
1190		-
1191	1188	#ifdef CONFIG_SCHEDSTATS
1192	1189	struct sched_statistics statistics;
1193	1190	#endif
...	...	@@ -1880,9 +1880,6 @@
1880	1880	static void
1881	1881	enqueue_task(struct rq rq, struct task_struct p, int wakeup, bool head)
1882	1882	{
1883		- if (wakeup)
1884		- p->se.start_runtime = p->se.sum_exec_runtime;
1885		-
1886	1883	sched_info_queued(p);
1887	1884	p->sched_class->enqueue_task(rq, p, wakeup, head);
1888	1885	p->se.on_rq = 1;
1889	1886
...	...	@@ -1890,17 +1887,11 @@
1890	1887
1891	1888	static void dequeue_task(struct rq rq, struct task_struct p, int sleep)
1892	1889	{
1893		- if (sleep) {
1894		- if (p->se.last_wakeup) {
1895		- update_avg(&p->se.avg_overlap,
1896		- p->se.sum_exec_runtime - p->se.last_wakeup);
1897		- p->se.last_wakeup = 0;
1898		- } else {
1899		- update_avg(&p->se.avg_wakeup,
1900		- sysctl_sched_wakeup_granularity);
1901		- }
	1890	+ if (sleep && p->se.last_wakeup) {
	1891	+ update_avg(&p->se.avg_overlap,
	1892	+ p->se.sum_exec_runtime - p->se.last_wakeup);
	1893	+ p->se.last_wakeup = 0;
1902	1894	}
1903		-
1904	1895	sched_info_dequeued(p);
1905	1896	p->sched_class->dequeue_task(rq, p, sleep);
1906	1897	p->se.on_rq = 0;
1907	1898
...	...	@@ -2466,14 +2457,7 @@
2466	2457	*/
2467	2458	if (!in_interrupt()) {
2468	2459	struct sched_entity *se = &current->se;
2469		- u64 sample = se->sum_exec_runtime;
2470	2460
2471		- if (se->last_wakeup)
2472		- sample -= se->last_wakeup;
2473		- else
2474		- sample -= se->start_runtime;
2475		- update_avg(&se->avg_wakeup, sample);
2476		-
2477	2461	se->last_wakeup = se->sum_exec_runtime;
2478	2462	}
2479	2463
...	...	@@ -2540,8 +2524,6 @@
2540	2524	p->se.nr_migrations = 0;
2541	2525	p->se.last_wakeup = 0;
2542	2526	p->se.avg_overlap = 0;
2543		- p->se.start_runtime = 0;
2544		- p->se.avg_wakeup = sysctl_sched_wakeup_granularity;
2545	2527
2546	2528	#ifdef CONFIG_SCHEDSTATS
2547	2529	memset(&p->se.statistics, 0, sizeof(p->se.statistics));
...	...	@@ -408,7 +408,6 @@
408	408	PN(se.vruntime);
409	409	PN(se.sum_exec_runtime);
410	410	PN(se.avg_overlap);
411		- PN(se.avg_wakeup);
412	411
413	412	nr_switches = p->nvcsw + p->nivcsw;
414	413
...	...	@@ -1592,41 +1592,10 @@
1592	1592	}
1593	1593	#endif /* CONFIG_SMP */
1594	1594
1595		-/*
1596		- * Adaptive granularity
1597		- *
1598		- * se->avg_wakeup gives the average time a task runs until it does a wakeup,
1599		- * with the limit of wakeup_gran -- when it never does a wakeup.
1600		- *
1601		- * So the smaller avg_wakeup is the faster we want this task to preempt,
1602		- * but we don't want to treat the preemptee unfairly and therefore allow it
1603		- * to run for at least the amount of time we'd like to run.
1604		- *
1605		- * NOTE: we use 2*avg_wakeup to increase the probability of actually doing one
1606		- *
1607		- * NOTE: we use *nr_running to scale with load, this nicely matches the
1608		- * degrading latency on load.
1609		- */
1610	1595	static unsigned long
1611		-adaptive_gran(struct sched_entity curr, struct sched_entity se)
1612		-{
1613		- u64 this_run = curr->sum_exec_runtime - curr->prev_sum_exec_runtime;
1614		- u64 expected_wakeup = 2se->avg_wakeup cfs_rq_of(se)->nr_running;
1615		- u64 gran = 0;
1616		-
1617		- if (this_run < expected_wakeup)
1618		- gran = expected_wakeup - this_run;
1619		-
1620		- return min_t(s64, gran, sysctl_sched_wakeup_granularity);
1621		-}
1622		-
1623		-static unsigned long
1624	1596	wakeup_gran(struct sched_entity curr, struct sched_entity se)
1625	1597	{
1626	1598	unsigned long gran = sysctl_sched_wakeup_granularity;
1627		-
1628		- if (cfs_rq_of(curr)->curr && sched_feat(ADAPTIVE_GRAN))
1629		- gran = adaptive_gran(curr, se);
1630	1599
1631	1600	/*
1632	1601	* Since its curr running now, convert the gran from real-time
...	...	@@ -31,12 +31,6 @@
31	31	SCHED_FEAT(WAKEUP_PREEMPT, 1)
32	32
33	33	/*
34		- * Compute wakeup_gran based on task behaviour, clipped to
35		- * [0, sched_wakeup_gran_ns]
36		- */
37		-SCHED_FEAT(ADAPTIVE_GRAN, 1)
38		-
39		-/*
40	34	* When converting the wakeup granularity to virtual time, do it such
41	35	* that heavier tasks preempting a lighter task have an edge.
42	36	*/