alinux: sched: Fix regression caused by nr_uninterruptible

fix #27788368

per cgroup nr_uninterruptible tracking leads to huge performance regression
of hackbench. This patch delete nr_uninterruptible related code for now, to
address performance regression issue.

Fixes: 9410d314 ("alinux: cpuacct: Export nr_running & nr_uninterruptible")
Fixes: 36da4fe9 ("alinux: sched: Maintain "nr_uninterruptible" in runqueue")
Signed-off-by: NYihao Wu <wuyihao@linux.alibaba.com>
Acked-by: NShanpei Chen <shanpeic@linux.alibaba.com>

kernel/sched/core.c

@@ -744,28 +744,18 @@ static inline void dequeue_task(struct rq *rq, struct task_struct *p, int flags)
 	p->sched_class->dequeue_task(rq, p, flags);
 }
 
-static void update_nr_uninterruptible(struct task_struct *tsk, long inc)
-{
-	if (tsk->sched_class->update_nr_uninterruptible)
-		tsk->sched_class->update_nr_uninterruptible(tsk, inc);
-}
-
 void activate_task(struct rq *rq, struct task_struct *p, int flags)
 {
-	if (task_contributes_to_load(p)) {
-		update_nr_uninterruptible(p, -1);
+	if (task_contributes_to_load(p))
 		rq->nr_uninterruptible--;
-	}
 
 	enqueue_task(rq, p, flags);
 }
 
 void deactivate_task(struct rq *rq, struct task_struct *p, int flags)
 {
-	if (task_contributes_to_load(p)) {
-		update_nr_uninterruptible(p, 1);
+	if (task_contributes_to_load(p))
 		rq->nr_uninterruptible++;
-	}
 
 	dequeue_task(rq, p, flags);
 }
@@ -1700,10 +1690,8 @@ ttwu_do_activate(struct rq *rq, struct task_struct *p, int wake_flags,
 	lockdep_assert_held(&rq->lock);
 
 #ifdef CONFIG_SMP
-	if (p->sched_contributes_to_load) {
-		update_nr_uninterruptible(p, -1);
+	if (p->sched_contributes_to_load)
 		rq->nr_uninterruptible--;
-	}
 
 	if (wake_flags & WF_MIGRATED)
 		en_flags |= ENQUEUE_MIGRATED;
@@ -6394,18 +6382,8 @@ void sched_move_task(struct task_struct *tsk)
 	if (running)
 		put_prev_task(rq, tsk);
 
-	/* decrease old group */
-	if ((!queued && task_contributes_to_load(tsk)) ||
-	    (tsk->state == TASK_WAKING && tsk->sched_contributes_to_load))
-		update_nr_uninterruptible(tsk, -1);
-
 	sched_change_group(tsk, TASK_MOVE_GROUP);
 
-	/* increase new group after change */
-	if ((!queued && task_contributes_to_load(tsk)) ||
-	    (tsk->state == TASK_WAKING && tsk->sched_contributes_to_load))
-		update_nr_uninterruptible(tsk, 1);
-
 	if (queued)
 		enqueue_task(rq, tsk, queue_flags);
 	if (running)

kernel/sched/cpuacct.c

@@ -358,95 +358,6 @@ static inline struct task_group *cgroup_tg(struct cgroup *cgrp)
 				struct task_group, css);
 }
 
-static inline unsigned long nr_uninterruptible(void)
-{
-	unsigned long i, sum = 0;
-
-	for_each_possible_cpu(i)
-		sum += cpu_rq(i)->nr_uninterruptible;
-
-	/*
-	 * Since we read the counters lockless, it might be slightly
-	 * inaccurate. Do not allow it to go below zero though:
-	 */
-	if (unlikely((long)sum < 0))
-		sum = 0;
-
-	return sum;
-}
-
-#ifdef CONFIG_CFS_BANDWIDTH
-static inline bool tg_cfs_throttled(struct task_group *tg, int cpu)
-{
-	return tg->cfs_rq[cpu]->throttle_count;
-}
-#else
-static inline bool tg_cfs_throttled(struct task_group *tg, int cpu)
-{
-	return false;
-}
-#endif
-
-#ifdef CONFIG_RT_GROUP_SCHED
-static inline bool tg_rt_throttled(struct task_group *tg, int cpu)
-{
-	return tg->rt_rq[cpu]->rt_throttled && !tg->rt_rq[cpu]->rt_nr_boosted;
-}
-#endif
-
-static unsigned long ca_running(struct cpuacct *ca, int cpu)
-{
-	unsigned long nr_running = 0;
-	struct cgroup *cgrp = ca->css.cgroup;
-	struct task_group *tg;
-
-	/* Make sure it is only called for non-root cpuacct */
-	if (ca == &root_cpuacct)
-		return 0;
-
-	rcu_read_lock();
-	tg = cgroup_tg(cgrp);
-	if (unlikely(!tg))
-		goto out;
-
-	if (!tg_cfs_throttled(tg, cpu))
-		nr_running += tg->cfs_rq[cpu]->h_nr_running;
-#ifdef CONFIG_RT_GROUP_SCHED
-	if (!tg_rt_throttled(tg, cpu))
-		nr_running += tg->rt_rq[cpu]->rt_nr_running;
-#endif
-	/* SCHED_DEADLINE doesn't support cgroup yet */
-
-out:
-	rcu_read_unlock();
-	return nr_running;
-}
-
-static unsigned long ca_uninterruptible(struct cpuacct *ca, int cpu)
-{
-	unsigned long nr = 0;
-	struct cgroup *cgrp = ca->css.cgroup;
-	struct task_group *tg;
-
-	/* Make sure it is only called for non-root cpuacct */
-	if (ca == &root_cpuacct)
-		return nr;
-
-	rcu_read_lock();
-	tg = cgroup_tg(cgrp);
-	if (unlikely(!tg))
-		goto out_rcu_unlock;
-
-	nr = tg->cfs_rq[cpu]->nr_uninterruptible;
-#ifdef CONFIG_RT_GROUP_SCHED
-	nr += tg->rt_rq[cpu]->nr_uninterruptible;
-#endif
-
-out_rcu_unlock:
-	rcu_read_unlock();
-	return nr;
-}
-
 void cgroup_idle_start(struct sched_entity *se)
 {
 	unsigned long flags;
@@ -625,7 +536,6 @@ static int cpuacct_proc_stats_show(struct seq_file *sf, void *v)
 	u64 user, nice, system, idle, iowait, irq, softirq, steal, guest;
 	u64 nr_migrations = 0;
 	struct cpuacct_alistats *alistats;
-	unsigned long nr_run = 0, nr_uninter = 0;
 	int cpu;
 
 	user = nice = system = idle = iowait =
@@ -656,8 +566,6 @@ static int cpuacct_proc_stats_show(struct seq_file *sf, void *v)
 
 			alistats = per_cpu_ptr(ca->alistats, cpu);
 			nr_migrations += alistats->nr_migrations;
-			nr_run += ca_running(ca, cpu);
-			nr_uninter += ca_uninterruptible(ca, cpu);
 		}
 	} else {
 		struct kernel_cpustat *kcpustat;
@@ -677,9 +585,6 @@ static int cpuacct_proc_stats_show(struct seq_file *sf, void *v)
 			alistats = per_cpu_ptr(ca->alistats, cpu);
 			nr_migrations += alistats->nr_migrations;
 		}
-
-		nr_run = nr_running();
-		nr_uninter = nr_uninterruptible();
 	}
 
 	seq_printf(sf, "user %lld\n", nsec_to_clock_t(user));
@@ -692,10 +597,6 @@ static int cpuacct_proc_stats_show(struct seq_file *sf, void *v)
 	seq_printf(sf, "steal %lld\n", nsec_to_clock_t(steal));
 	seq_printf(sf, "guest %lld\n", nsec_to_clock_t(guest));
 
-	seq_printf(sf, "nr_running %lld\n", (u64)nr_run);
-	if ((long) nr_uninter < 0)
-		nr_uninter = 0;
-	seq_printf(sf, "nr_uninterruptible %lld\n", (u64)nr_uninter);
 	seq_printf(sf, "nr_migrations %lld\n", (u64)nr_migrations);
 
 	return 0;

kernel/sched/fair.c

@@ -10311,16 +10311,6 @@ static unsigned int get_rr_interval_fair(struct rq *rq, struct task_struct *task
 	return rr_interval;
 }
 
-#ifdef CONFIG_SCHED_SLI
-static void update_nr_uninterruptible_fair(struct task_struct *p, long inc)
-{
-	struct sched_entity *se = &p->se;
-
-	for_each_sched_entity(se)
-		cfs_rq_of(se)->nr_uninterruptible += inc;
-}
-#endif
-
 /*
  * All the scheduling class methods:
  */
@@ -10364,7 +10354,6 @@ const struct sched_class fair_sched_class = {
 #endif
 
 #ifdef CONFIG_SCHED_SLI
-	.update_nr_uninterruptible = update_nr_uninterruptible_fair,
 	.update_nr_iowait	= update_nr_iowait_fair,
 #endif
 };

kernel/sched/rt.c

@@ -2374,16 +2374,6 @@ static unsigned int get_rr_interval_rt(struct rq *rq, struct task_struct *task)
 		return 0;
 }
 
-#ifdef CONFIG_SCHED_SLI
-static void update_nr_uninterruptible_rt(struct task_struct *p, long inc)
-{
-	struct sched_rt_entity *se = &p->rt;
-
-	for_each_sched_rt_entity(se)
-		rt_rq_of_se(se)->nr_uninterruptible += inc;
-}
-#endif
-
 const struct sched_class rt_sched_class = {
 	.next			= &fair_sched_class,
 	.enqueue_task		= enqueue_task_rt,
@@ -2414,10 +2404,6 @@ const struct sched_class rt_sched_class = {
 	.switched_to		= switched_to_rt,
 
 	.update_curr		= update_curr_rt,
-
-#ifdef CONFIG_SCHED_SLI
-	.update_nr_uninterruptible = update_nr_uninterruptible_rt,
-#endif
 };
 
 #ifdef CONFIG_RT_GROUP_SCHED

kernel/sched/sched.h

@@ -575,8 +575,6 @@ struct cfs_rq {
 #endif /* CONFIG_CFS_BANDWIDTH */
 #endif /* CONFIG_FAIR_GROUP_SCHED */
 
-	unsigned long nr_uninterruptible;
-
 	ALI_HOTFIX_RESERVE(1)
 	ALI_HOTFIX_RESERVE(2)
 	ALI_HOTFIX_RESERVE(3)
@@ -627,8 +625,6 @@ struct rt_rq {
 	struct rq		*rq;
 	struct task_group	*tg;
 #endif
-
-	unsigned long nr_uninterruptible;
 };
 
 static inline bool rt_rq_is_runnable(struct rt_rq *rt_rq)
@@ -1675,8 +1671,6 @@ struct sched_class {
 #ifdef CONFIG_FAIR_GROUP_SCHED
 	void (*task_change_group)(struct task_struct *p, int type);
 #endif
-
-	void (*update_nr_uninterruptible)(struct task_struct *p, long inc);
 	void (*update_nr_iowait)(struct task_struct *p, long inc);
 };