sched/fair: Add trivial fair server

Use deadline servers to service fair tasks.

This patch adds a fair_server deadline entity which acts as a container
for fair entities and can be used to fix starvation when higher priority
(wrt fair) tasks are monopolizing CPU(s).

Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Daniel Bristot de Oliveira <bristot@kernel.org>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Tested-by: Juri Lelli <juri.lelli@redhat.com>
Link: https://lore.kernel.org/r/b6b0bcefaf25391bcf5b6ecdb9f1218de402d42e.1716811044.git.bristot@kernel.org

kernel/sched/core.c

@@ -8336,6 +8336,7 @@ void __init sched_init(void)
 #endif /* CONFIG_SMP */
 		hrtick_rq_init(rq);
 		atomic_set(&rq->nr_iowait, 0);
+		fair_server_init(rq);
 
 #ifdef CONFIG_SCHED_CORE
 		rq->core = rq;

kernel/sched/deadline.c

@@ -1381,6 +1381,13 @@ throttle:
 			resched_curr(rq);
 	}
 
+	/*
+	 * The fair server (sole dl_server) does not account for real-time
+	 * workload because it is running fair work.
+	 */
+	if (dl_se == &rq->fair_server)
+		return;
+
 	/*
 	 * Because -- for now -- we share the rt bandwidth, we need to
 	 * account our runtime there too, otherwise actual rt tasks
@@ -1414,15 +1421,31 @@ void dl_server_update(struct sched_dl_entity *dl_se, s64 delta_exec)
 
 void dl_server_start(struct sched_dl_entity *dl_se)
 {
+	struct rq *rq = dl_se->rq;
+
 	if (!dl_server(dl_se)) {
+		/* Disabled */
+		dl_se->dl_runtime = 0;
+		dl_se->dl_deadline = 1000 * NSEC_PER_MSEC;
+		dl_se->dl_period = 1000 * NSEC_PER_MSEC;
+
 		dl_se->dl_server = 1;
 		setup_new_dl_entity(dl_se);
 	}
+
+	if (!dl_se->dl_runtime)
+		return;
+
 	enqueue_dl_entity(dl_se, ENQUEUE_WAKEUP);
+	if (!dl_task(dl_se->rq->curr) || dl_entity_preempt(dl_se, &rq->curr->dl))
+		resched_curr(dl_se->rq);
 }
 
 void dl_server_stop(struct sched_dl_entity *dl_se)
 {
+	if (!dl_se->dl_runtime)
+		return;
+
 	dequeue_dl_entity(dl_se, DEQUEUE_SLEEP);
 }
 

kernel/sched/fair.c

@@ -5765,6 +5765,7 @@ static bool throttle_cfs_rq(struct cfs_rq *cfs_rq)
 	struct cfs_bandwidth *cfs_b = tg_cfs_bandwidth(cfs_rq->tg);
 	struct sched_entity *se;
 	long task_delta, idle_task_delta, dequeue = 1;
+	long rq_h_nr_running = rq->cfs.h_nr_running;
 
 	raw_spin_lock(&cfs_b->lock);
 	/* This will start the period timer if necessary */
@@ -5837,6 +5838,9 @@ static bool throttle_cfs_rq(struct cfs_rq *cfs_rq)
 	sub_nr_running(rq, task_delta);
 
 done:
+	/* Stop the fair server if throttling resulted in no runnable tasks */
+	if (rq_h_nr_running && !rq->cfs.h_nr_running)
+		dl_server_stop(&rq->fair_server);
 	/*
 	 * Note: distribution will already see us throttled via the
 	 * throttled-list.  rq->lock protects completion.
@@ -5854,6 +5858,7 @@ void unthrottle_cfs_rq(struct cfs_rq *cfs_rq)
 	struct cfs_bandwidth *cfs_b = tg_cfs_bandwidth(cfs_rq->tg);
 	struct sched_entity *se;
 	long task_delta, idle_task_delta;
+	long rq_h_nr_running = rq->cfs.h_nr_running;
 
 	se = cfs_rq->tg->se[cpu_of(rq)];
 
@@ -5929,6 +5934,10 @@ void unthrottle_cfs_rq(struct cfs_rq *cfs_rq)
 unthrottle_throttle:
 	assert_list_leaf_cfs_rq(rq);
 
+	/* Start the fair server if un-throttling resulted in new runnable tasks */
+	if (!rq_h_nr_running && rq->cfs.h_nr_running)
+		dl_server_start(&rq->fair_server);
+
 	/* Determine whether we need to wake up potentially idle CPU: */
 	if (rq->curr == rq->idle && rq->cfs.nr_running)
 		resched_curr(rq);
@@ -6759,6 +6768,9 @@ enqueue_task_fair(struct rq *rq, struct task_struct *p, int flags)
 	 */
 	util_est_enqueue(&rq->cfs, p);
 
+	if (!throttled_hierarchy(task_cfs_rq(p)) && !rq->cfs.h_nr_running)
+		dl_server_start(&rq->fair_server);
+
 	/*
 	 * If in_iowait is set, the code below may not trigger any cpufreq
 	 * utilization updates, so do it here explicitly with the IOWAIT flag
@@ -6903,6 +6915,9 @@ static void dequeue_task_fair(struct rq *rq, struct task_struct *p, int flags)
 		rq->next_balance = jiffies;
 
 dequeue_throttle:
+	if (!throttled_hierarchy(task_cfs_rq(p)) && !rq->cfs.h_nr_running)
+		dl_server_stop(&rq->fair_server);
+
 	util_est_update(&rq->cfs, p, task_sleep);
 	hrtick_update(rq);
 }
@@ -8602,6 +8617,25 @@ static struct task_struct *__pick_next_task_fair(struct rq *rq)
 	return pick_next_task_fair(rq, NULL, NULL);
 }
 
+static bool fair_server_has_tasks(struct sched_dl_entity *dl_se)
+{
+	return !!dl_se->rq->cfs.nr_running;
+}
+
+static struct task_struct *fair_server_pick(struct sched_dl_entity *dl_se)
+{
+	return pick_next_task_fair(dl_se->rq, NULL, NULL);
+}
+
+void fair_server_init(struct rq *rq)
+{
+	struct sched_dl_entity *dl_se = &rq->fair_server;
+
+	init_dl_entity(dl_se);
+
+	dl_server_init(dl_se, rq, fair_server_has_tasks, fair_server_pick);
+}
+
 /*
  * Account for a descheduled task:
  */

kernel/sched/sched.h

@@ -363,6 +363,8 @@ extern void dl_server_init(struct sched_dl_entity *dl_se, struct rq *rq,
 		    dl_server_has_tasks_f has_tasks,
 		    dl_server_pick_f pick);
 
+extern void fair_server_init(struct rq *rq);
+
 #ifdef CONFIG_CGROUP_SCHED
 
 extern struct list_head task_groups;
@@ -1039,6 +1041,8 @@ struct rq {
 	struct rt_rq		rt;
 	struct dl_rq		dl;
 
+	struct sched_dl_entity	fair_server;
+
 #ifdef CONFIG_FAIR_GROUP_SCHED
 	/* list of leaf cfs_rq on this CPU: */
 	struct list_head	leaf_cfs_rq_list;