Merge branch 'sched/core' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip into for-6.12

Pull tip/sched/core to resolve the following four conflicts. While 2-4 are
simple context conflicts, 1 is a bit subtle and easy to resolve incorrectly.

1. 2c8d046d5d51 ("sched: Add normal_policy()")
   vs.
   faa42d29419d ("sched/fair: Make SCHED_IDLE entity be preempted in strict hierarchy")

The former converts direct test on p->policy to use the helper
normal_policy(). The latter moves the p->policy test to a different
location. Resolve by converting the test on p->plicy in the new location to
use normal_policy().

2. a7a9fc549293 ("sched_ext: Add boilerplate for extensible scheduler class")
   vs.
   a110a81c52a9 ("sched/deadline: Deferrable dl server")

Both add calls to put_prev_task_idle() and set_next_task_idle(). Simple
context conflict. Resolve by taking changes from both.

3. a7a9fc549293 ("sched_ext: Add boilerplate for extensible scheduler class")
   vs.
   c245910049d0 ("sched/core: Add clearing of ->dl_server in put_prev_task_balance()")

The former changes for_each_class() itertion to use for_each_active_class().
The latter moves away the adjacent dl_server handling code. Simple context
conflict. Resolve by taking changes from both.

4. 60c27fb59f6c ("sched_ext: Implement sched_ext_ops.cpu_online/offline()")
   vs.
   31b164e2e4af ("sched/smt: Introduce sched_smt_present_inc/dec() helper")
   2f027354122f ("sched/core: Introduce sched_set_rq_on/offline() helper")

The former adds scx_rq_deactivate() call. The latter two change code around
it. Simple context conflict. Resolve by taking changes from both.

Signed-off-by: Tejun Heo <tj@kernel.org>

10 files changed, 879 insertions, 281 deletions

diff --git a/kernel/sched/core.c b/kernel/sched/core.c
index 22f86d5e9231..ec67b30f87c4 100644
--- a/kernel/sched/core.c
+++ b/kernel/sched/core.c
@@ -163,6 +163,9 @@ static inline int __task_prio(const struct task_struct *p)
 	if (p->sched_class == &stop_sched_class) /* trumps deadline */
 		return -2;
 
+	if (p->dl_server)
+		return -1; /* deadline */
+
 	if (rt_prio(p->prio)) /* includes deadline */
 		return p->prio; /* [-1, 99] */
 
@@ -195,8 +198,24 @@ static inline bool prio_less(const struct task_struct *a,
 	if (-pb < -pa)
 		return false;
 
-	if (pa == -1) /* dl_prio() doesn't work because of stop_class above */
-		return !dl_time_before(a->dl.deadline, b->dl.deadline);
+	if (pa == -1) { /* dl_prio() doesn't work because of stop_class above */
+		const struct sched_dl_entity *a_dl, *b_dl;
+
+		a_dl = &a->dl;
+		/*
+		 * Since,'a' and 'b' can be CFS tasks served by DL server,
+		 * __task_prio() can return -1 (for DL) even for those. In that
+		 * case, get to the dl_server's DL entity.
+		 */
+		if (a->dl_server)
+			a_dl = a->dl_server;
+
+		b_dl = &b->dl;
+		if (b->dl_server)
+			b_dl = b->dl_server;
+
+		return !dl_time_before(a_dl->deadline, b_dl->deadline);
+	}
 
 	if (pa == MAX_RT_PRIO + MAX_NICE)	/* fair */
 		return cfs_prio_less(a, b, in_fi);
@@ -1280,7 +1299,7 @@ bool sched_can_stop_tick(struct rq *rq)
 	 * dequeued by migrating while the constrained task continues to run.
 	 * E.g. going from 2->1 without going through pick_next_task().
 	 */
-	if (sched_feat(HZ_BW) && __need_bw_check(rq, rq->curr)) {
+	if (__need_bw_check(rq, rq->curr)) {
 		if (cfs_task_bw_constrained(rq->curr))
 			return false;
 	}
@@ -2255,6 +2274,12 @@ void migrate_disable(void)
 	struct task_struct *p = current;
 
 	if (p->migration_disabled) {
+#ifdef CONFIG_DEBUG_PREEMPT
+		/*
+		 *Warn about overflow half-way through the range.
+		 */
+		WARN_ON_ONCE((s16)p->migration_disabled < 0);
+#endif
 		p->migration_disabled++;
 		return;
 	}
@@ -2273,14 +2298,20 @@ void migrate_enable(void)
 		.flags     = SCA_MIGRATE_ENABLE,
 	};
 
+#ifdef CONFIG_DEBUG_PREEMPT
+	/*
+	 * Check both overflow from migrate_disable() and superfluous
+	 * migrate_enable().
+	 */
+	if (WARN_ON_ONCE((s16)p->migration_disabled <= 0))
+		return;
+#endif
+
 	if (p->migration_disabled > 1) {
 		p->migration_disabled--;
 		return;
 	}
 
-	if (WARN_ON_ONCE(!p->migration_disabled))
-		return;
-
 	/*
 	 * Ensure stop_task runs either before or after this, and that
 	 * __set_cpus_allowed_ptr(SCA_MIGRATE_ENABLE) doesn't schedule().
@@ -4737,7 +4768,7 @@ void wake_up_new_task(struct task_struct *p)
 	update_rq_clock(rq);
 	post_init_entity_util_avg(p);
 
-	activate_task(rq, p, ENQUEUE_NOCLOCK);
+	activate_task(rq, p, ENQUEUE_NOCLOCK | ENQUEUE_INITIAL);
 	trace_sched_wakeup_new(p);
 	wakeup_preempt(rq, p, WF_FORK);
 #ifdef CONFIG_SMP
@@ -5855,6 +5886,14 @@ static void put_prev_task_balance(struct rq *rq, struct task_struct *prev,
 #endif
 
 	put_prev_task(rq, prev);
+
+	/*
+	 * We've updated @prev and no longer need the server link, clear it.
+	 * Must be done before ->pick_next_task() because that can (re)set
+	 * ->dl_server.
+	 */
+	if (prev->dl_server)
+		prev->dl_server = NULL;
 }
 
 /*
@@ -5889,6 +5928,13 @@ __pick_next_task(struct rq *rq, struct task_struct *prev, struct rq_flags *rf)
 		}
 
 		/*
+		 * This is a normal CFS pick, but the previous could be a DL pick.
+		 * Clear it as previous is no longer picked.
+		 */
+		if (prev->dl_server)
+			prev->dl_server = NULL;
+
+		/*
 		 * This is the fast path; it cannot be a DL server pick;
 		 * therefore even if @p == @prev, ->dl_server must be NULL.
 		 */
@@ -5901,14 +5947,6 @@ __pick_next_task(struct rq *rq, struct task_struct *prev, struct rq_flags *rf)
 restart:
 	put_prev_task_balance(rq, prev, rf);
 
-	/*
-	 * We've updated @prev and no longer need the server link, clear it.
-	 * Must be done before ->pick_next_task() because that can (re)set
-	 * ->dl_server.
-	 */
-	if (prev->dl_server)
-		prev->dl_server = NULL;
-
 	for_each_active_class(class) {
 		p = class->pick_next_task(rq);
 		if (p) {
@@ -7925,6 +7963,30 @@ void set_rq_offline(struct rq *rq)
 	}
 }
 
+static inline void sched_set_rq_online(struct rq *rq, int cpu)
+{
+	struct rq_flags rf;
+
+	rq_lock_irqsave(rq, &rf);
+	if (rq->rd) {
+		BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span));
+		set_rq_online(rq);
+	}
+	rq_unlock_irqrestore(rq, &rf);
+}
+
+static inline void sched_set_rq_offline(struct rq *rq, int cpu)
+{
+	struct rq_flags rf;
+
+	rq_lock_irqsave(rq, &rf);
+	if (rq->rd) {
+		BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span));
+		set_rq_offline(rq);
+	}
+	rq_unlock_irqrestore(rq, &rf);
+}
+
 /*
  * used to mark begin/end of suspend/resume:
  */
@@ -7975,10 +8037,25 @@ static int cpuset_cpu_inactive(unsigned int cpu)
 	return 0;
 }
 
+static inline void sched_smt_present_inc(int cpu)
+{
+#ifdef CONFIG_SCHED_SMT
+	if (cpumask_weight(cpu_smt_mask(cpu)) == 2)
+		static_branch_inc_cpuslocked(&sched_smt_present);
+#endif
+}
+
+static inline void sched_smt_present_dec(int cpu)
+{
+#ifdef CONFIG_SCHED_SMT
+	if (cpumask_weight(cpu_smt_mask(cpu)) == 2)
+		static_branch_dec_cpuslocked(&sched_smt_present);
+#endif
+}
+
 int sched_cpu_activate(unsigned int cpu)
 {
 	struct rq *rq = cpu_rq(cpu);
-	struct rq_flags rf;
 
 	/*
 	 * Clear the balance_push callback and prepare to schedule
@@ -7986,13 +8063,10 @@ int sched_cpu_activate(unsigned int cpu)
 	 */
 	balance_push_set(cpu, false);
 
-#ifdef CONFIG_SCHED_SMT
 	/*
 	 * When going up, increment the number of cores with SMT present.
 	 */
-	if (cpumask_weight(cpu_smt_mask(cpu)) == 2)
-		static_branch_inc_cpuslocked(&sched_smt_present);
-#endif
+	sched_smt_present_inc(cpu);
 	set_cpu_active(cpu, true);
 
 	if (sched_smp_initialized) {
@@ -8012,12 +8086,7 @@ int sched_cpu_activate(unsigned int cpu)
 	 * 2) At runtime, if cpuset_cpu_active() fails to rebuild the
 	 *    domains.
 	 */
-	rq_lock_irqsave(rq, &rf);
-	if (rq->rd) {
-		BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span));
-		set_rq_online(rq);
-	}
-	rq_unlock_irqrestore(rq, &rf);
+	sched_set_rq_online(rq, cpu);
 
 	return 0;
 }
@@ -8025,7 +8094,6 @@ int sched_cpu_activate(unsigned int cpu)
 int sched_cpu_deactivate(unsigned int cpu)
 {
 	struct rq *rq = cpu_rq(cpu);
-	struct rq_flags rf;
 	int ret;
 
 	/*
@@ -8056,22 +8124,16 @@ int sched_cpu_deactivate(unsigned int cpu)
 	 */
 	synchronize_rcu();
 
-	rq_lock_irqsave(rq, &rf);
-	if (rq->rd) {
-		BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span));
-		set_rq_offline(rq);
-	}
-	rq_unlock_irqrestore(rq, &rf);
+	sched_set_rq_offline(rq, cpu);
 
 	scx_rq_deactivate(rq);
 
-#ifdef CONFIG_SCHED_SMT
 	/*
 	 * When going down, decrement the number of cores with SMT present.
 	 */
-	if (cpumask_weight(cpu_smt_mask(cpu)) == 2)
-		static_branch_dec_cpuslocked(&sched_smt_present);
+	sched_smt_present_dec(cpu);
 
+#ifdef CONFIG_SCHED_SMT
 	sched_core_cpu_deactivate(cpu);
 #endif
 
@@ -8081,6 +8143,8 @@ int sched_cpu_deactivate(unsigned int cpu)
 	sched_update_numa(cpu, false);
 	ret = cpuset_cpu_inactive(cpu);
 	if (ret) {
+		sched_smt_present_inc(cpu);
+		sched_set_rq_online(rq, cpu);
 		balance_push_set(cpu, false);
 		set_cpu_active(cpu, true);
 		sched_update_numa(cpu, true);
@@ -8290,8 +8354,6 @@ void __init sched_init(void)
 #endif /* CONFIG_RT_GROUP_SCHED */
 	}
 
-	init_rt_bandwidth(&def_rt_bandwidth, global_rt_period(), global_rt_runtime());
-
 #ifdef CONFIG_SMP
 	init_defrootdomain();
 #endif
@@ -8346,8 +8408,13 @@ void __init sched_init(void)
 		init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, NULL);
 #endif /* CONFIG_FAIR_GROUP_SCHED */
 
-		rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime;
 #ifdef CONFIG_RT_GROUP_SCHED
+		/*
+		 * This is required for init cpu because rt.c:__enable_runtime()
+		 * starts working after scheduler_running, which is not the case
+		 * yet.
+		 */
+		rq->rt.rt_runtime = global_rt_runtime();
 		init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, NULL);
 #endif
 #ifdef CONFIG_SMP
@@ -8379,6 +8446,7 @@ 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;
diff --git a/kernel/sched/cputime.c b/kernel/sched/cputime.c
index a5e00293ae43..0bed0fa1acd9 100644
--- a/kernel/sched/cputime.c
+++ b/kernel/sched/cputime.c
@@ -582,6 +582,12 @@ void cputime_adjust(struct task_cputime *curr, struct prev_cputime *prev,
 	}
 
 	stime = mul_u64_u64_div_u64(stime, rtime, stime + utime);
+	/*
+	 * Because mul_u64_u64_div_u64() can approximate on some
+	 * achitectures; enforce the constraint that: a*b/(b+c) <= a.
+	 */
+	if (unlikely(stime > rtime))
+		stime = rtime;
 
 update:
 	/*
diff --git a/kernel/sched/deadline.c b/kernel/sched/deadline.c
index f59e5c19d944..c5f1cc753a31 100644
--- a/kernel/sched/deadline.c
+++ b/kernel/sched/deadline.c
@@ -320,19 +320,12 @@ void sub_running_bw(struct sched_dl_entity *dl_se, struct dl_rq *dl_rq)
 		__sub_running_bw(dl_se->dl_bw, dl_rq);
 }
 
-static void dl_change_utilization(struct task_struct *p, u64 new_bw)
+static void dl_rq_change_utilization(struct rq *rq, struct sched_dl_entity *dl_se, u64 new_bw)
 {
-	struct rq *rq;
-
-	WARN_ON_ONCE(p->dl.flags & SCHED_FLAG_SUGOV);
-
-	if (task_on_rq_queued(p))
-		return;
+	if (dl_se->dl_non_contending) {
+		sub_running_bw(dl_se, &rq->dl);
+		dl_se->dl_non_contending = 0;
 
-	rq = task_rq(p);
-	if (p->dl.dl_non_contending) {
-		sub_running_bw(&p->dl, &rq->dl);
-		p->dl.dl_non_contending = 0;
 		/*
 		 * If the timer handler is currently running and the
 		 * timer cannot be canceled, inactive_task_timer()
@@ -340,13 +333,25 @@ static void dl_change_utilization(struct task_struct *p, u64 new_bw)
 		 * will not touch the rq's active utilization,
 		 * so we are still safe.
 		 */
-		if (hrtimer_try_to_cancel(&p->dl.inactive_timer) == 1)
-			put_task_struct(p);
+		if (hrtimer_try_to_cancel(&dl_se->inactive_timer) == 1) {
+			if (!dl_server(dl_se))
+				put_task_struct(dl_task_of(dl_se));
+		}
 	}
-	__sub_rq_bw(p->dl.dl_bw, &rq->dl);
+	__sub_rq_bw(dl_se->dl_bw, &rq->dl);
 	__add_rq_bw(new_bw, &rq->dl);
 }
 
+static void dl_change_utilization(struct task_struct *p, u64 new_bw)
+{
+	WARN_ON_ONCE(p->dl.flags & SCHED_FLAG_SUGOV);
+
+	if (task_on_rq_queued(p))
+		return;
+
+	dl_rq_change_utilization(task_rq(p), &p->dl, new_bw);
+}
+
 static void __dl_clear_params(struct sched_dl_entity *dl_se);
 
 /*
@@ -771,6 +776,15 @@ static inline void replenish_dl_new_period(struct sched_dl_entity *dl_se,
 	/* for non-boosted task, pi_of(dl_se) == dl_se */
 	dl_se->deadline = rq_clock(rq) + pi_of(dl_se)->dl_deadline;
 	dl_se->runtime = pi_of(dl_se)->dl_runtime;
+
+	/*
+	 * If it is a deferred reservation, and the server
+	 * is not handling an starvation case, defer it.
+	 */
+	if (dl_se->dl_defer & !dl_se->dl_defer_running) {
+		dl_se->dl_throttled = 1;
+		dl_se->dl_defer_armed = 1;
+	}
 }
 
 /*
@@ -809,6 +823,9 @@ static inline void setup_new_dl_entity(struct sched_dl_entity *dl_se)
 	replenish_dl_new_period(dl_se, rq);
 }
 
+static int start_dl_timer(struct sched_dl_entity *dl_se);
+static bool dl_entity_overflow(struct sched_dl_entity *dl_se, u64 t);
+
 /*
  * Pure Earliest Deadline First (EDF) scheduling does not deal with the
  * possibility of a entity lasting more than what it declared, and thus
@@ -837,9 +854,18 @@ static void replenish_dl_entity(struct sched_dl_entity *dl_se)
 	/*
 	 * This could be the case for a !-dl task that is boosted.
 	 * Just go with full inherited parameters.
+	 *
+	 * Or, it could be the case of a deferred reservation that
+	 * was not able to consume its runtime in background and
+	 * reached this point with current u > U.
+	 *
+	 * In both cases, set a new period.
 	 */
-	if (dl_se->dl_deadline == 0)
-		replenish_dl_new_period(dl_se, rq);
+	if (dl_se->dl_deadline == 0 ||
+	    (dl_se->dl_defer_armed && dl_entity_overflow(dl_se, rq_clock(rq)))) {
+		dl_se->deadline = rq_clock(rq) + pi_of(dl_se)->dl_deadline;
+		dl_se->runtime = pi_of(dl_se)->dl_runtime;
+	}
 
 	if (dl_se->dl_yielded && dl_se->runtime > 0)
 		dl_se->runtime = 0;
@@ -873,6 +899,44 @@ static void replenish_dl_entity(struct sched_dl_entity *dl_se)
 		dl_se->dl_yielded = 0;
 	if (dl_se->dl_throttled)
 		dl_se->dl_throttled = 0;
+
+	/*
+	 * If this is the replenishment of a deferred reservation,
+	 * clear the flag and return.
+	 */
+	if (dl_se->dl_defer_armed) {
+		dl_se->dl_defer_armed = 0;
+		return;
+	}
+
+	/*
+	 * A this point, if the deferred server is not armed, and the deadline
+	 * is in the future, if it is not running already, throttle the server
+	 * and arm the defer timer.
+	 */
+	if (dl_se->dl_defer && !dl_se->dl_defer_running &&
+	    dl_time_before(rq_clock(dl_se->rq), dl_se->deadline - dl_se->runtime)) {
+		if (!is_dl_boosted(dl_se) && dl_se->server_has_tasks(dl_se)) {
+
+			/*
+			 * Set dl_se->dl_defer_armed and dl_throttled variables to
+			 * inform the start_dl_timer() that this is a deferred
+			 * activation.
+			 */
+			dl_se->dl_defer_armed = 1;
+			dl_se->dl_throttled = 1;
+			if (!start_dl_timer(dl_se)) {
+				/*
+				 * If for whatever reason (delays), a previous timer was
+				 * queued but not serviced, cancel it and clean the
+				 * deferrable server variables intended for start_dl_timer().
+				 */
+				hrtimer_try_to_cancel(&dl_se->dl_timer);
+				dl_se->dl_defer_armed = 0;
+				dl_se->dl_throttled = 0;
+			}
+		}
+	}
 }
 
 /*
@@ -1023,6 +1087,15 @@ static void update_dl_entity(struct sched_dl_entity *dl_se)
 		}
 
 		replenish_dl_new_period(dl_se, rq);
+	} else if (dl_server(dl_se) && dl_se->dl_defer) {
+		/*
+		 * The server can still use its previous deadline, so check if
+		 * it left the dl_defer_running state.
+		 */
+		if (!dl_se->dl_defer_running) {
+			dl_se->dl_defer_armed = 1;
+			dl_se->dl_throttled = 1;
+		}
 	}
 }
 
@@ -1055,8 +1128,21 @@ static int start_dl_timer(struct sched_dl_entity *dl_se)
 	 * We want the timer to fire at the deadline, but considering
 	 * that it is actually coming from rq->clock and not from
 	 * hrtimer's time base reading.
+	 *
+	 * The deferred reservation will have its timer set to
+	 * (deadline - runtime). At that point, the CBS rule will decide
+	 * if the current deadline can be used, or if a replenishment is
+	 * required to avoid add too much pressure on the system
+	 * (current u > U).
 	 */
-	act = ns_to_ktime(dl_next_period(dl_se));
+	if (dl_se->dl_defer_armed) {
+		WARN_ON_ONCE(!dl_se->dl_throttled);
+		act = ns_to_ktime(dl_se->deadline - dl_se->runtime);
+	} else {
+		/* act = deadline - rel-deadline + period */
+		act = ns_to_ktime(dl_next_period(dl_se));
+	}
+
 	now = hrtimer_cb_get_time(timer);
 	delta = ktime_to_ns(now) - rq_clock(rq);
 	act = ktime_add_ns(act, delta);
@@ -1106,6 +1192,62 @@ static void __push_dl_task(struct rq *rq, struct rq_flags *rf)
 #endif
 }
 
+/* a defer timer will not be reset if the runtime consumed was < dl_server_min_res */
+static const u64 dl_server_min_res = 1 * NSEC_PER_MSEC;
+
+static enum hrtimer_restart dl_server_timer(struct hrtimer *timer, struct sched_dl_entity *dl_se)
+{
+	struct rq *rq = rq_of_dl_se(dl_se);
+	u64 fw;
+
+	scoped_guard (rq_lock, rq) {
+		struct rq_flags *rf = &scope.rf;
+
+		if (!dl_se->dl_throttled || !dl_se->dl_runtime)
+			return HRTIMER_NORESTART;
+
+		sched_clock_tick();
+		update_rq_clock(rq);
+
+		if (!dl_se->dl_runtime)
+			return HRTIMER_NORESTART;
+
+		if (!dl_se->server_has_tasks(dl_se)) {
+			replenish_dl_entity(dl_se);
+			return HRTIMER_NORESTART;
+		}
+
+		if (dl_se->dl_defer_armed) {
+			/*
+			 * First check if the server could consume runtime in background.
+			 * If so, it is possible to push the defer timer for this amount
+			 * of time. The dl_server_min_res serves as a limit to avoid
+			 * forwarding the timer for a too small amount of time.
+			 */
+			if (dl_time_before(rq_clock(dl_se->rq),
+					   (dl_se->deadline - dl_se->runtime - dl_server_min_res))) {
+
+				/* reset the defer timer */
+				fw = dl_se->deadline - rq_clock(dl_se->rq) - dl_se->runtime;
+
+				hrtimer_forward_now(timer, ns_to_ktime(fw));
+				return HRTIMER_RESTART;
+			}
+
+			dl_se->dl_defer_running = 1;
+		}
+
+		enqueue_dl_entity(dl_se, ENQUEUE_REPLENISH);
+
+		if (!dl_task(dl_se->rq->curr) || dl_entity_preempt(dl_se, &dl_se->rq->curr->dl))
+			resched_curr(rq);
+
+		__push_dl_task(rq, rf);
+	}
+
+	return HRTIMER_NORESTART;
+}
+
 /*
  * This is the bandwidth enforcement timer callback. If here, we know
  * a task is not on its dl_rq, since the fact that the timer was running
@@ -1128,28 +1270,8 @@ static enum hrtimer_restart dl_task_timer(struct hrtimer *timer)
 	struct rq_flags rf;
 	struct rq *rq;
 
-	if (dl_server(dl_se)) {
-		struct rq *rq = rq_of_dl_se(dl_se);
-		struct rq_flags rf;
-
-		rq_lock(rq, &rf);
-		if (dl_se->dl_throttled) {
-			sched_clock_tick();
-			update_rq_clock(rq);
-
-			if (dl_se->server_has_tasks(dl_se)) {
-				enqueue_dl_entity(dl_se, ENQUEUE_REPLENISH);
-				resched_curr(rq);
-				__push_dl_task(rq, &rf);
-			} else {
-				replenish_dl_entity(dl_se);
-			}
-
-		}
-		rq_unlock(rq, &rf);
-
-		return HRTIMER_NORESTART;
-	}
+	if (dl_server(dl_se))
+		return dl_server_timer(timer, dl_se);
 
 	p = dl_task_of(dl_se);
 	rq = task_rq_lock(p, &rf);
@@ -1319,22 +1441,10 @@ static u64 grub_reclaim(u64 delta, struct rq *rq, struct sched_dl_entity *dl_se)
 	return (delta * u_act) >> BW_SHIFT;
 }
 
-static inline void
-update_stats_dequeue_dl(struct dl_rq *dl_rq, struct sched_dl_entity *dl_se,
-                        int flags);
-static void update_curr_dl_se(struct rq *rq, struct sched_dl_entity *dl_se, s64 delta_exec)
+s64 dl_scaled_delta_exec(struct rq *rq, struct sched_dl_entity *dl_se, s64 delta_exec)
 {
 	s64 scaled_delta_exec;
 
-	if (unlikely(delta_exec <= 0)) {
-		if (unlikely(dl_se->dl_yielded))
-			goto throttle;
-		return;
-	}
-
-	if (dl_entity_is_special(dl_se))
-		return;
-
 	/*
 	 * For tasks that participate in GRUB, we implement GRUB-PA: the
 	 * spare reclaimed bandwidth is used to clock down frequency.
@@ -1353,8 +1463,64 @@ static void update_curr_dl_se(struct rq *rq, struct sched_dl_entity *dl_se, s64
 		scaled_delta_exec = cap_scale(scaled_delta_exec, scale_cpu);
 	}
 
+	return scaled_delta_exec;
+}
+
+static inline void
+update_stats_dequeue_dl(struct dl_rq *dl_rq, struct sched_dl_entity *dl_se,
+			int flags);
+static void update_curr_dl_se(struct rq *rq, struct sched_dl_entity *dl_se, s64 delta_exec)
+{
+	s64 scaled_delta_exec;
+
+	if (unlikely(delta_exec <= 0)) {
+		if (unlikely(dl_se->dl_yielded))
+			goto throttle;
+		return;
+	}
+
+	if (dl_server(dl_se) && dl_se->dl_throttled && !dl_se->dl_defer)
+		return;
+
+	if (dl_entity_is_special(dl_se))
+		return;
+
+	scaled_delta_exec = dl_scaled_delta_exec(rq, dl_se, delta_exec);
+
 	dl_se->runtime -= scaled_delta_exec;
 
+	/*
+	 * The fair server can consume its runtime while throttled (not queued/
+	 * running as regular CFS).
+	 *
+	 * If the server consumes its entire runtime in this state. The server
+	 * is not required for the current period. Thus, reset the server by
+	 * starting a new period, pushing the activation.
+	 */
+	if (dl_se->dl_defer && dl_se->dl_throttled && dl_runtime_exceeded(dl_se)) {
+		/*
+		 * If the server was previously activated - the starving condition
+		 * took place, it this point it went away because the fair scheduler
+		 * was able to get runtime in background. So return to the initial
+		 * state.
+		 */
+		dl_se->dl_defer_running = 0;
+
+		hrtimer_try_to_cancel(&dl_se->dl_timer);
+
+		replenish_dl_new_period(dl_se, dl_se->rq);
+
+		/*
+		 * Not being able to start the timer seems problematic. If it could not
+		 * be started for whatever reason, we need to "unthrottle" the DL server
+		 * and queue right away. Otherwise nothing might queue it. That's similar
+		 * to what enqueue_dl_entity() does on start_dl_timer==0. For now, just warn.
+		 */
+		WARN_ON_ONCE(!start_dl_timer(dl_se));
+
+		return;
+	}
+
 throttle:
 	if (dl_runtime_exceeded(dl_se) || dl_se->dl_yielded) {
 		dl_se->dl_throttled = 1;
@@ -1382,6 +1548,14 @@ throttle:
 	}
 
 	/*
+	 * 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;
+
+#ifdef CONFIG_RT_GROUP_SCHED
+	/*
 	 * Because -- for now -- we share the rt bandwidth, we need to
 	 * account our runtime there too, otherwise actual rt tasks
 	 * would be able to exceed the shared quota.
@@ -1405,34 +1579,157 @@ throttle:
 			rt_rq->rt_time += delta_exec;
 		raw_spin_unlock(&rt_rq->rt_runtime_lock);
 	}
+#endif
+}
+
+/*
+ * In the non-defer mode, the idle time is not accounted, as the
+ * server provides a guarantee.
+ *
+ * If the dl_server is in defer mode, the idle time is also considered
+ * as time available for the fair server, avoiding a penalty for the
+ * rt scheduler that did not consumed that time.
+ */
+void dl_server_update_idle_time(struct rq *rq, struct task_struct *p)
+{
+	s64 delta_exec, scaled_delta_exec;
+
+	if (!rq->fair_server.dl_defer)
+		return;
+
+	/* no need to discount more */
+	if (rq->fair_server.runtime < 0)
+		return;
+
+	delta_exec = rq_clock_task(rq) - p->se.exec_start;
+	if (delta_exec < 0)
+		return;
+
+	scaled_delta_exec = dl_scaled_delta_exec(rq, &rq->fair_server, delta_exec);
+
+	rq->fair_server.runtime -= scaled_delta_exec;
+
+	if (rq->fair_server.runtime < 0) {
+		rq->fair_server.dl_defer_running = 0;
+		rq->fair_server.runtime = 0;
+	}
+
+	p->se.exec_start = rq_clock_task(rq);
 }
 
 void dl_server_update(struct sched_dl_entity *dl_se, s64 delta_exec)
 {
-	update_curr_dl_se(dl_se->rq, dl_se, delta_exec);
+	/* 0 runtime = fair server disabled */
+	if (dl_se->dl_runtime)
+		update_curr_dl_se(dl_se->rq, dl_se, delta_exec);
 }
 
 void dl_server_start(struct sched_dl_entity *dl_se)
 {
+	struct rq *rq = dl_se->rq;
+
+	/*
+	 * XXX: the apply do not work fine at the init phase for the
+	 * fair server because things are not yet set. We need to improve
+	 * this before getting generic.
+	 */
 	if (!dl_server(dl_se)) {
+		u64 runtime =  50 * NSEC_PER_MSEC;
+		u64 period = 1000 * NSEC_PER_MSEC;
+
+		dl_server_apply_params(dl_se, runtime, period, 1);
+
 		dl_se->dl_server = 1;
+		dl_se->dl_defer = 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);
+	hrtimer_try_to_cancel(&dl_se->dl_timer);
+	dl_se->dl_defer_armed = 0;
+	dl_se->dl_throttled = 0;
 }
 
 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)
+		    dl_server_pick_f pick_next,
+		    dl_server_pick_f pick_task)
 {
 	dl_se->rq = rq;
 	dl_se->server_has_tasks = has_tasks;
-	dl_se->server_pick = pick;
+	dl_se->server_pick_next = pick_next;
+	dl_se->server_pick_task = pick_task;
+}
+
+void __dl_server_attach_root(struct sched_dl_entity *dl_se, struct rq *rq)
+{
+	u64 new_bw = dl_se->dl_bw;
+	int cpu = cpu_of(rq);
+	struct dl_bw *dl_b;
+
+	dl_b = dl_bw_of(cpu_of(rq));
+	guard(raw_spinlock)(&dl_b->lock);
+
+	if (!dl_bw_cpus(cpu))
+		return;
+
+	__dl_add(dl_b, new_bw, dl_bw_cpus(cpu));
+}
+
+int dl_server_apply_params(struct sched_dl_entity *dl_se, u64 runtime, u64 period, bool init)
+{
+	u64 old_bw = init ? 0 : to_ratio(dl_se->dl_period, dl_se->dl_runtime);
+	u64 new_bw = to_ratio(period, runtime);
+	struct rq *rq = dl_se->rq;
+	int cpu = cpu_of(rq);
+	struct dl_bw *dl_b;
+	unsigned long cap;
+	int retval = 0;
+	int cpus;
+
+	dl_b = dl_bw_of(cpu);
+	guard(raw_spinlock)(&dl_b->lock);
+
+	cpus = dl_bw_cpus(cpu);
+	cap = dl_bw_capacity(cpu);
+
+	if (__dl_overflow(dl_b, cap, old_bw, new_bw))
+		return -EBUSY;
+
+	if (init) {
+		__add_rq_bw(new_bw, &rq->dl);
+		__dl_add(dl_b, new_bw, cpus);
+	} else {
+		__dl_sub(dl_b, dl_se->dl_bw, cpus);
+		__dl_add(dl_b, new_bw, cpus);
+
+		dl_rq_change_utilization(rq, dl_se, new_bw);
+	}
+
+	dl_se->dl_runtime = runtime;
+	dl_se->dl_deadline = period;
+	dl_se->dl_period = period;
+
+	dl_se->runtime = 0;
+	dl_se->deadline = 0;
+
+	dl_se->dl_bw = to_ratio(dl_se->dl_period, dl_se->dl_runtime);
+	dl_se->dl_density = to_ratio(dl_se->dl_deadline, dl_se->dl_runtime);
+
+	return retval;
 }
 
 /*
@@ -1735,7 +2032,7 @@ enqueue_dl_entity(struct sched_dl_entity *dl_se, int flags)
 	 * be counted in the active utilization; hence, we need to call
 	 * add_running_bw().
 	 */
-	if (dl_se->dl_throttled && !(flags & ENQUEUE_REPLENISH)) {
+	if (!dl_se->dl_defer && dl_se->dl_throttled && !(flags & ENQUEUE_REPLENISH)) {
 		if (flags & ENQUEUE_WAKEUP)
 			task_contending(dl_se, flags);
 
@@ -1757,6 +2054,25 @@ enqueue_dl_entity(struct sched_dl_entity *dl_se, int flags)
 		setup_new_dl_entity(dl_se);
 	}
 
+	/*
+	 * If the reservation is still throttled, e.g., it got replenished but is a
+	 * deferred task and still got to wait, don't enqueue.
+	 */
+	if (dl_se->dl_throttled && start_dl_timer(dl_se))
+		return;
+
+	/*
+	 * We're about to enqueue, make sure we're not ->dl_throttled!
+	 * In case the timer was not started, say because the defer time
+	 * has passed, mark as not throttled and mark unarmed.
+	 * Also cancel earlier timers, since letting those run is pointless.
+	 */
+	if (dl_se->dl_throttled) {
+		hrtimer_try_to_cancel(&dl_se->dl_timer);
+		dl_se->dl_defer_armed = 0;
+		dl_se->dl_throttled = 0;
+	}
+
 	__enqueue_dl_entity(dl_se);
 }
 
@@ -2086,7 +2402,12 @@ static struct sched_dl_entity *pick_next_dl_entity(struct dl_rq *dl_rq)
 	return __node_2_dle(left);
 }
 
-static struct task_struct *pick_task_dl(struct rq *rq)
+/*
+ * __pick_next_task_dl - Helper to pick the next -deadline task to run.
+ * @rq: The runqueue to pick the next task from.
+ * @peek: If true, just peek at the next task. Only relevant for dlserver.
+ */
+static struct task_struct *__pick_next_task_dl(struct rq *rq, bool peek)
 {
 	struct sched_dl_entity *dl_se;
 	struct dl_rq *dl_rq = &rq->dl;
@@ -2100,7 +2421,10 @@ again:
 	WARN_ON_ONCE(!dl_se);
 
 	if (dl_server(dl_se)) {
-		p = dl_se->server_pick(dl_se);
+		if (IS_ENABLED(CONFIG_SMP) && peek)
+			p = dl_se->server_pick_task(dl_se);
+		else
+			p = dl_se->server_pick_next(dl_se);
 		if (!p) {
 			WARN_ON_ONCE(1);
 			dl_se->dl_yielded = 1;
@@ -2115,11 +2439,18 @@ again:
 	return p;
 }
 
+#ifdef CONFIG_SMP
+static struct task_struct *pick_task_dl(struct rq *rq)
+{
+	return __pick_next_task_dl(rq, true);
+}
+#endif
+
 static struct task_struct *pick_next_task_dl(struct rq *rq)
 {
 	struct task_struct *p;
 
-	p = pick_task_dl(rq);
+	p = __pick_next_task_dl(rq, false);
 	if (!p)
 		return p;
 
diff --git a/kernel/sched/debug.c b/kernel/sched/debug.c
index c057ef46c5f8..0148bc65d39c 100644
--- a/kernel/sched/debug.c
+++ b/kernel/sched/debug.c
@@ -333,8 +333,165 @@ static const struct file_operations sched_debug_fops = {
 	.release	= seq_release,
 };
 
+enum dl_param {
+	DL_RUNTIME = 0,
+	DL_PERIOD,
+};
+
+static unsigned long fair_server_period_max = (1 << 22) * NSEC_PER_USEC; /* ~4 seconds */
+static unsigned long fair_server_period_min = (100) * NSEC_PER_USEC;     /* 100 us */
+
+static ssize_t sched_fair_server_write(struct file *filp, const char __user *ubuf,
+				       size_t cnt, loff_t *ppos, enum dl_param param)
+{
+	long cpu = (long) ((struct seq_file *) filp->private_data)->private;
+	struct rq *rq = cpu_rq(cpu);
+	u64 runtime, period;
+	size_t err;
+	int retval;
+	u64 value;
+
+	err = kstrtoull_from_user(ubuf, cnt, 10, &value);
+	if (err)
+		return err;
+
+	scoped_guard (rq_lock_irqsave, rq) {
+		runtime  = rq->fair_server.dl_runtime;
+		period = rq->fair_server.dl_period;
+
+		switch (param) {
+		case DL_RUNTIME:
+			if (runtime == value)
+				break;
+			runtime = value;
+			break;
+		case DL_PERIOD:
+			if (value == period)
+				break;
+			period = value;
+			break;
+		}
+
+		if (runtime > period ||
+		    period > fair_server_period_max ||
+		    period < fair_server_period_min) {
+			return  -EINVAL;
+		}
+
+		if (rq->cfs.h_nr_running) {
+			update_rq_clock(rq);
+			dl_server_stop(&rq->fair_server);
+		}
+
+		retval = dl_server_apply_params(&rq->fair_server, runtime, period, 0);
+		if (retval)
+			cnt = retval;
+
+		if (!runtime)
+			printk_deferred("Fair s