[Devel] [PATCH vz9 07/27] sched/stat: account ctxsw per task group

[Nikita Yushchenko]

From: Vladimir Davydov <vdavydov at parallels.com>

This is a backport of diff-sched-account-ctxsw-per-task-group:

 Subject: sched: account ctxsw per task group
 Date: Fri, 28 Dec 2012 15:09:45 +0400

* [sched] the number of context switches should be reported correctly
        inside a CT in /proc/stat (PSBM-18113)

For /proc/stat:ctxt to be correct inside containers.

https://jira.sw.ru/browse/PSBM-18113

Signed-off-by: Vladimir Davydov <vdavydov at parallels.com>

(cherry picked from vz7 commit d388f0bf64adb74cd62c4deff58e181bd63d62ac)
Signed-off-by: Konstantin Khorenko <khorenko at virtuozzo.com>
Reviewed-by: Andrey Ryabinin <aryabinin at virtuozzo.com>

+++
sched: Port diff-sched-clear-prev-entity-if-curr-is-dequeued

Author: Vladimir Davydov
Email: vdavydov at parallels.com
Subject: sched: clear prev entity if curr is dequeued
Date: Fri, 20 Sep 2013 16:55:23 +0400

cfs_rq->prev is used for ctxsw accounting: on put_prev_entity()
cfs_rq->prev is set to curr if curr is on rq, and on set_next_entity()
nr_switches is increased if curr != prev. However, If the current task
goes to sleep, cfs_rq->prev won't be updated, which will lead to a ctxsw
not accounted. The patch fixes this by clearing cfs_rq->prev if current
is not on rq.

Signed-off-by: Vladimir Davydov <vdavydov at parallels.com>

=============================================================================

Related to https://jira.sw.ru/browse/PSBM-33642

Signed-off-by: Vladimir Davydov <vdavydov at parallels.com>

https://jira.sw.ru/browse/PSBM-127780
(cherry-picked from vz7 commit 409c82642d4f ("sched: Port
diff-sched-clear-prev-entity-if-curr-is-dequeued"))
mFixes: dc03756f0c397 ("sched/stat: account ctxsw per task group")

Signed-off-by: Pavel Tikhomirov <ptikhomirov at virtuozzo.com>

(cherry-picked from vz8 commit 5345b3e5cd7b ("sched/stat: account
ctxsw per task group"))

Signed-off-by: Nikita Yushchenko <nikita.yushchenko at virtuozzo.com>
---
 kernel/sched/cpuacct.c |  4 +++-
 kernel/sched/fair.c    | 17 ++++++++++++++---
 kernel/sched/sched.h   |  3 +++
 3 files changed, 20 insertions(+), 4 deletions(-)

diff --git a/kernel/sched/cpuacct.c b/kernel/sched/cpuacct.c
index 871b6f8ccb0d..a409c441aba3 100644
--- a/kernel/sched/cpuacct.c
+++ b/kernel/sched/cpuacct.c
@@ -649,6 +649,7 @@ int cpu_cgroup_proc_stat(struct cgroup_subsys_state *cpu_css,
 	struct kernel_cpustat *kcpustat;
 	unsigned long tg_nr_running = 0;
 	unsigned long tg_nr_iowait = 0;
+	unsigned long long tg_nr_switches = 0;
 
 	time_ns = ve_get_time_ns(get_exec_env());
 	if (time_ns) {
@@ -672,6 +673,7 @@ int cpu_cgroup_proc_stat(struct cgroup_subsys_state *cpu_css,
 #ifdef CONFIG_FAIR_GROUP_SCHED
 		tg_nr_running += tg->cfs_rq[i]->h_nr_running;
 		tg_nr_iowait  += tg->cfs_rq[i]->nr_iowait;
+		tg_nr_switches += tg->cfs_rq[i]->nr_switches;
 #endif
 #ifdef CONFIG_RT_GROUP_SCHED
 		tg_nr_running += tg->rt_rq[i]->rt_nr_running;
@@ -745,7 +747,7 @@ int cpu_cgroup_proc_stat(struct cgroup_subsys_state *cpu_css,
 		   "processes %lu\n"
 		   "procs_running %lu\n"
 		   "procs_blocked %lu\n",
-		   nr_context_switches(),
+		   tg_nr_switches,
 		   (unsigned long long)boottime.tv_sec,
 		   total_forks,
 		   tg_nr_running,
diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
index c0bf49db868d..b8de26a5629d 100644
--- a/kernel/sched/fair.c
+++ b/kernel/sched/fair.c
@@ -4488,6 +4488,9 @@ dequeue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int flags)
 
 	clear_buddies(cfs_rq, se);
 
+	if (cfs_rq->prev == se)
+		cfs_rq->prev = NULL;
+
 	if (se != cfs_rq->curr)
 		__dequeue_entity(cfs_rq, se);
 	se->on_rq = 0;
@@ -4502,8 +4505,12 @@ dequeue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int flags)
 	if (!(flags & DEQUEUE_SLEEP))
 		se->vruntime -= cfs_rq->min_vruntime;
 
-	/* return excess runtime on last dequeue */
-	return_cfs_rq_runtime(cfs_rq);
+	if (!cfs_rq->nr_running) {
+		/* return excess runtime on last dequeue */
+		return_cfs_rq_runtime(cfs_rq);
+		/* account switch to idle task */
+		cfs_rq->nr_switches++;
+	}
 
 	update_cfs_group(se);
 
@@ -4579,6 +4586,8 @@ set_next_entity(struct cfs_rq *cfs_rq, struct sched_entity *se)
 
 	update_stats_curr_start(cfs_rq, se);
 	cfs_rq->curr = se;
+	if (cfs_rq->prev != se)
+		cfs_rq->nr_switches++;
 
 	/*
 	 * Track our maximum slice length, if the CPU's load is at
@@ -4676,7 +4685,9 @@ static void put_prev_entity(struct cfs_rq *cfs_rq, struct sched_entity *prev)
 		__enqueue_entity(cfs_rq, prev);
 		/* in !on_rq case, update occurred at dequeue */
 		update_load_avg(cfs_rq, prev, 0);
-	}
+		cfs_rq->prev = prev;
+	} else
+		cfs_rq->prev = NULL;
 	cfs_rq->curr = NULL;
 }
 
diff --git a/kernel/sched/sched.h b/kernel/sched/sched.h
index 13c1941d0593..5042f45bf887 100644
--- a/kernel/sched/sched.h
+++ b/kernel/sched/sched.h
@@ -579,6 +579,9 @@ struct cfs_rq {
 	struct sched_entity	*next;
 	struct sched_entity	*last;
 	struct sched_entity	*skip;
+	struct sched_entity	*prev;
+
+	u64 nr_switches;
 
 #ifdef	CONFIG_SCHED_DEBUG
 	unsigned int		nr_spread_over;
-- 
2.30.2