Merge tag 'sched-core-2024-01-08' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip

Pull scheduler updates from Ingo Molnar:
 "Energy scheduling:

   - Consolidate how the max compute capacity is used in the scheduler
     and how we calculate the frequency for a level of utilization.

   - Rework interface between the scheduler and the schedutil governor

   - Simplify the util_est logic

  Deadline scheduler:

   - Work more towards reducing SCHED_DEADLINE starvation of low
     priority tasks (e.g., SCHED_OTHER) tasks when higher priority tasks
     monopolize CPU cycles, via the introduction of 'deadline servers'
     (nested/2-level scheduling).

     "Fair servers" to make use of this facility are not introduced yet.

  EEVDF:

   - Introduce O(1) fastpath for EEVDF task selection

  NUMA balancing:

   - Tune the NUMA-balancing vma scanning logic some more, to better
     distribute the probability of a particular vma getting scanned.

  Plus misc fixes, cleanups and updates"

* tag 'sched-core-2024-01-08' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip: (30 commits)
  sched/fair: Fix tg->load when offlining a CPU
  sched/fair: Remove unused 'next_buddy_marked' local variable in check_preempt_wakeup_fair()
  sched/fair: Use all little CPUs for CPU-bound workloads
  sched/fair: Simplify util_est
  sched/fair: Remove SCHED_FEAT(UTIL_EST_FASTUP, true)
  arm64/amu: Use capacity_ref_freq() to set AMU ratio
  cpufreq/cppc: Set the frequency used for computing the capacity
  cpufreq/cppc: Move and rename cppc_cpufreq_{perf_to_khz|khz_to_perf}()
  energy_model: Use a fixed reference frequency
  cpufreq/schedutil: Use a fixed reference frequency
  cpufreq: Use the fixed and coherent frequency for scaling capacity
  sched/topology: Add a new arch_scale_freq_ref() method
  freezer,sched: Clean saved_state when restoring it during thaw
  sched/fair: Update min_vruntime for reweight_entity() correctly
  sched/doc: Update documentation after renames and synchronize Chinese version
  sched/cpufreq: Rework iowait boost
  sched/cpufreq: Rework schedutil governor performance estimation
  sched/pelt: Avoid underestimation of task utilization
  sched/timers: Explain why idle task schedules out on remote timer enqueue
  sched/cpuidle: Comment about timers requirements VS idle handler
  ...

12 files changed, 797 insertions, 578 deletions

diff --git a/kernel/freezer.c b/kernel/freezer.c
index 759006a9a910..f57aaf96b829 100644
--- a/kernel/freezer.c
+++ b/kernel/freezer.c
@@ -187,6 +187,7 @@ static int __restore_freezer_state(struct task_struct *p, void *arg)
 
 	if (state != TASK_RUNNING) {
 		WRITE_ONCE(p->__state, state);
+		p->saved_state = TASK_RUNNING;
 		return 1;
 	}
 
diff --git a/kernel/sched/core.c b/kernel/sched/core.c
index a708d225c28e..db4be4921e7f 100644
--- a/kernel/sched/core.c
+++ b/kernel/sched/core.c
@@ -1131,6 +1131,28 @@ static void wake_up_idle_cpu(int cpu)
 	if (cpu == smp_processor_id())
 		return;
 
+	/*
+	 * Set TIF_NEED_RESCHED and send an IPI if in the non-polling
+	 * part of the idle loop. This forces an exit from the idle loop
+	 * and a round trip to schedule(). Now this could be optimized
+	 * because a simple new idle loop iteration is enough to
+	 * re-evaluate the next tick. Provided some re-ordering of tick
+	 * nohz functions that would need to follow TIF_NR_POLLING
+	 * clearing:
+	 *
+	 * - On most archs, a simple fetch_or on ti::flags with a
+	 *   "0" value would be enough to know if an IPI needs to be sent.
+	 *
+	 * - x86 needs to perform a last need_resched() check between
+	 *   monitor and mwait which doesn't take timers into account.
+	 *   There a dedicated TIF_TIMER flag would be required to
+	 *   fetch_or here and be checked along with TIF_NEED_RESCHED
+	 *   before mwait().
+	 *
+	 * However, remote timer enqueue is not such a frequent event
+	 * and testing of the above solutions didn't appear to report
+	 * much benefits.
+	 */
 	if (set_nr_and_not_polling(rq->idle))
 		smp_send_reschedule(cpu);
 	else
@@ -2124,12 +2146,14 @@ void activate_task(struct rq *rq, struct task_struct *p, int flags)
 
 	enqueue_task(rq, p, flags);
 
-	p->on_rq = TASK_ON_RQ_QUEUED;
+	WRITE_ONCE(p->on_rq, TASK_ON_RQ_QUEUED);
+	ASSERT_EXCLUSIVE_WRITER(p->on_rq);
 }
 
 void deactivate_task(struct rq *rq, struct task_struct *p, int flags)
 {
-	p->on_rq = (flags & DEQUEUE_SLEEP) ? 0 : TASK_ON_RQ_MIGRATING;
+	WRITE_ONCE(p->on_rq, (flags & DEQUEUE_SLEEP) ? 0 : TASK_ON_RQ_MIGRATING);
+	ASSERT_EXCLUSIVE_WRITER(p->on_rq);
 
 	dequeue_task(rq, p, flags);
 }
@@ -3795,6 +3819,8 @@ ttwu_do_activate(struct rq *rq, struct task_struct *p, int wake_flags,
 		rq->idle_stamp = 0;
 	}
 #endif
+
+	p->dl_server = NULL;
 }
 
 /*
@@ -4509,10 +4535,7 @@ static void __sched_fork(unsigned long clone_flags, struct task_struct *p)
 	memset(&p->stats, 0, sizeof(p->stats));
 #endif
 
-	RB_CLEAR_NODE(&p->dl.rb_node);
-	init_dl_task_timer(&p->dl);
-	init_dl_inactive_task_timer(&p->dl);
-	__dl_clear_params(p);
+	init_dl_entity(&p->dl);
 
 	INIT_LIST_HEAD(&p->rt.run_list);
 	p->rt.timeout		= 0;
@@ -6004,12 +6027,27 @@ __pick_next_task(struct rq *rq, struct task_struct *prev, struct rq_flags *rf)
 			p = pick_next_task_idle(rq);
 		}
 
+		/*
+		 * This is the fast path; it cannot be a DL server pick;
+		 * therefore even if @p == @prev, ->dl_server must be NULL.
+		 */
+		if (p->dl_server)
+			p->dl_server = NULL;
+
 		return p;
 	}
 
 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_class(class) {
 		p = class->pick_next_task(rq);
 		if (p)
@@ -7429,18 +7467,13 @@ int sched_core_idle_cpu(int cpu)
  * required to meet deadlines.
  */
 unsigned long effective_cpu_util(int cpu, unsigned long util_cfs,
-				 enum cpu_util_type type,
-				 struct task_struct *p)
+				 unsigned long *min,
+				 unsigned long *max)
 {
-	unsigned long dl_util, util, irq, max;
+	unsigned long util, irq, scale;
 	struct rq *rq = cpu_rq(cpu);
 
-	max = arch_scale_cpu_capacity(cpu);
-
-	if (!uclamp_is_used() &&
-	    type == FREQUENCY_UTIL && rt_rq_is_runnable(&rq->rt)) {
-		return max;
-	}
+	scale = arch_scale_cpu_capacity(cpu);
 
 	/*
 	 * Early check to see if IRQ/steal time saturates the CPU, can be
@@ -7448,45 +7481,49 @@ unsigned long effective_cpu_util(int cpu, unsigned long util_cfs,
 	 * update_irq_load_avg().
 	 */
 	irq = cpu_util_irq(rq);
-	if (unlikely(irq >= max))
-		return max;
+	if (unlikely(irq >= scale)) {
+		if (min)
+			*min = scale;
+		if (max)
+			*max = scale;
+		return scale;
+	}
+
+	if (min) {
+		/*
+		 * The minimum utilization returns the highest level between:
+		 * - the computed DL bandwidth needed with the IRQ pressure which
+		 *   steals time to the deadline task.
+		 * - The minimum performance requirement for CFS and/or RT.
+		 */
+		*min = max(irq + cpu_bw_dl(rq), uclamp_rq_get(rq, UCLAMP_MIN));
+
+		/*
+		 * When an RT task is runnable and uclamp is not used, we must
+		 * ensure that the task will run at maximum compute capacity.
+		 */
+		if (!uclamp_is_used() && rt_rq_is_runnable(&rq->rt))
+			*min = max(*min, scale);
+	}
 
 	/*
 	 * Because the time spend on RT/DL tasks is visible as 'lost' time to
 	 * CFS tasks and we use the same metric to track the effective
 	 * utilization (PELT windows are synchronized) we can directly add them
 	 * to obtain the CPU's actual utilization.
-	 *
-	 * CFS and RT utilization can be boosted or capped, depending on
-	 * utilization clamp constraints requested by currently RUNNABLE
-	 * tasks.
-	 * When there are no CFS RUNNABLE tasks, clamps are released and
-	 * frequency will be gracefully reduced with the utilization decay.
 	 */
 	util = util_cfs + cpu_util_rt(rq);
-	if (type == FREQUENCY_UTIL)
-		util = uclamp_rq_util_with(rq, util, p);
-
-	dl_util = cpu_util_dl(rq);
+	util += cpu_util_dl(rq);
 
 	/*
-	 * For frequency selection we do not make cpu_util_dl() a permanent part
-	 * of this sum because we want to use cpu_bw_dl() later on, but we need
-	 * to check if the CFS+RT+DL sum is saturated (ie. no idle time) such
-	 * that we select f_max when there is no idle time.
-	 *
-	 * NOTE: numerical errors or stop class might cause us to not quite hit
-	 * saturation when we should -- something for later.
+	 * The maximum hint is a soft bandwidth requirement, which can be lower
+	 * than the actual utilization because of uclamp_max requirements.
 	 */
-	if (util + dl_util >= max)
-		return max;
+	if (max)
+		*max = min(scale, uclamp_rq_get(rq, UCLAMP_MAX));
 
-	/*
-	 * OTOH, for energy computation we need the estimated running time, so
-	 * include util_dl and ignore dl_bw.
-	 */
-	if (type == ENERGY_UTIL)
-		util += dl_util;
+	if (util >= scale)
+		return scale;
 
 	/*
 	 * There is still idle time; further improve the number by using the
@@ -7497,28 +7534,15 @@ unsigned long effective_cpu_util(int cpu, unsigned long util_cfs,
 	 *   U' = irq + --------- * U
 	 *                 max
 	 */
-	util = scale_irq_capacity(util, irq, max);
+	util = scale_irq_capacity(util, irq, scale);
 	util += irq;
 
-	/*
-	 * Bandwidth required by DEADLINE must always be granted while, for
-	 * FAIR and RT, we use blocked utilization of IDLE CPUs as a mechanism
-	 * to gracefully reduce the frequency when no tasks show up for longer
-	 * periods of time.
-	 *
-	 * Ideally we would like to set bw_dl as min/guaranteed freq and util +
-	 * bw_dl as requested freq. However, cpufreq is not yet ready for such
-	 * an interface. So, we only do the latter for now.
-	 */
-	if (type == FREQUENCY_UTIL)
-		util += cpu_bw_dl(rq);
-
-	return min(max, util);
+	return min(scale, util);
 }
 
 unsigned long sched_cpu_util(int cpu)
 {
-	return effective_cpu_util(cpu, cpu_util_cfs(cpu), ENERGY_UTIL, NULL);
+	return effective_cpu_util(cpu, cpu_util_cfs(cpu), NULL, NULL);
 }
 #endif /* CONFIG_SMP */
 
diff --git a/kernel/sched/cpufreq_schedutil.c b/kernel/sched/cpufreq_schedutil.c
index 5888176354e2..95c3c097083e 100644
--- a/kernel/sched/cpufreq_schedutil.c
+++ b/kernel/sched/cpufreq_schedutil.c
@@ -47,7 +47,7 @@ struct sugov_cpu {
 	u64			last_update;
 
 	unsigned long		util;
-	unsigned long		bw_dl;
+	unsigned long		bw_min;
 
 	/* The field below is for single-CPU policies only: */
 #ifdef CONFIG_NO_HZ_COMMON
@@ -115,6 +115,28 @@ static void sugov_deferred_update(struct sugov_policy *sg_policy)
 }
 
 /**
+ * get_capacity_ref_freq - get the reference frequency that has been used to
+ * correlate frequency and compute capacity for a given cpufreq policy. We use
+ * the CPU managing it for the arch_scale_freq_ref() call in the function.
+ * @policy: the cpufreq policy of the CPU in question.
+ *
+ * Return: the reference CPU frequency to compute a capacity.
+ */
+static __always_inline
+unsigned long get_capacity_ref_freq(struct cpufreq_policy *policy)
+{
+	unsigned int freq = arch_scale_freq_ref(policy->cpu);
+
+	if (freq)
+		return freq;
+
+	if (arch_scale_freq_invariant())
+		return policy->cpuinfo.max_freq;
+
+	return policy->cur;
+}
+
+/**
  * get_next_freq - Compute a new frequency for a given cpufreq policy.
  * @sg_policy: schedutil policy object to compute the new frequency for.
  * @util: Current CPU utilization.
@@ -140,10 +162,9 @@ static unsigned int get_next_freq(struct sugov_policy *sg_policy,
 				  unsigned long util, unsigned long max)
 {
 	struct cpufreq_policy *policy = sg_policy->policy;
-	unsigned int freq = arch_scale_freq_invariant() ?
-				policy->cpuinfo.max_freq : policy->cur;
+	unsigned int freq;
 
-	util = map_util_perf(util);
+	freq = get_capacity_ref_freq(policy);
 	freq = map_util_freq(util, freq, max);
 
 	if (freq == sg_policy->cached_raw_freq && !sg_policy->need_freq_update)
@@ -153,14 +174,31 @@ static unsigned int get_next_freq(struct sugov_policy *sg_policy,
 	return cpufreq_driver_resolve_freq(policy, freq);
 }
 
-static void sugov_get_util(struct sugov_cpu *sg_cpu)
+unsigned long sugov_effective_cpu_perf(int cpu, unsigned long actual,
+				 unsigned long min,
+				 unsigned long max)
+{
+	/* Add dvfs headroom to actual utilization */
+	actual = map_util_perf(actual);
+	/* Actually we don't need to target the max performance */
+	if (actual < max)
+		max = actual;
+
+	/*
+	 * Ensure at least minimum performance while providing more compute
+	 * capacity when possible.
+	 */
+	return max(min, max);
+}
+
+static void sugov_get_util(struct sugov_cpu *sg_cpu, unsigned long boost)
 {
-	unsigned long util = cpu_util_cfs_boost(sg_cpu->cpu);
-	struct rq *rq = cpu_rq(sg_cpu->cpu);
+	unsigned long min, max, util = cpu_util_cfs_boost(sg_cpu->cpu);
 
-	sg_cpu->bw_dl = cpu_bw_dl(rq);
-	sg_cpu->util = effective_cpu_util(sg_cpu->cpu, util,
-					  FREQUENCY_UTIL, NULL);
+	util = effective_cpu_util(sg_cpu->cpu, util, &min, &max);
+	util = max(util, boost);
+	sg_cpu->bw_min = min;
+	sg_cpu->util = sugov_effective_cpu_perf(sg_cpu->cpu, util, min, max);
 }
 
 /**
@@ -251,18 +289,16 @@ static void sugov_iowait_boost(struct sugov_cpu *sg_cpu, u64 time,
  * This mechanism is designed to boost high frequently IO waiting tasks, while
  * being more conservative on tasks which does sporadic IO operations.
  */
-static void sugov_iowait_apply(struct sugov_cpu *sg_cpu, u64 time,
+static unsigned long sugov_iowait_apply(struct sugov_cpu *sg_cpu, u64 time,
 			       unsigned long max_cap)
 {
-	unsigned long boost;
-
 	/* No boost currently required */
 	if (!sg_cpu->iowait_boost)
-		return;
+		return 0;
 
 	/* Reset boost if the CPU appears to have been idle enough */
 	if (sugov_iowait_reset(sg_cpu, time, false))
-		return;
+		return 0;
 
 	if (!sg_cpu->iowait_boost_pending) {
 		/*
@@ -271,7 +307,7 @@ static void sugov_iowait_apply(struct sugov_cpu *sg_cpu, u64 time,
 		sg_cpu->iowait_boost >>= 1;
 		if (sg_cpu->iowait_boost < IOWAIT_BOOST_MIN) {
 			sg_cpu->iowait_boost = 0;
-			return;
+			return 0;
 		}
 	}
 
@@ -281,10 +317,7 @@ static void sugov_iowait_apply(struct sugov_cpu *sg_cpu, u64 time,
 	 * sg_cpu->util is already in capacity scale; convert iowait_boost
 	 * into the same scale so we can compare.
 	 */
-	boost = (sg_cpu->iowait_boost * max_cap) >> SCHED_CAPACITY_SHIFT;
-	boost = uclamp_rq_util_with(cpu_rq(sg_cpu->cpu), boost, NULL);
-	if (sg_cpu->util < boost)
-		sg_cpu->util = boost;
+	return (sg_cpu->iowait_boost * max_cap) >> SCHED_CAPACITY_SHIFT;
 }
 
 #ifdef CONFIG_NO_HZ_COMMON
@@ -306,7 +339,7 @@ static inline bool sugov_cpu_is_busy(struct sugov_cpu *sg_cpu) { return false; }
  */
 static inline void ignore_dl_rate_limit(struct sugov_cpu *sg_cpu)
 {
-	if (cpu_bw_dl(cpu_rq(sg_cpu->cpu)) > sg_cpu->bw_dl)
+	if (cpu_bw_dl(cpu_rq(sg_cpu->cpu)) > sg_cpu->bw_min)
 		sg_cpu->sg_policy->limits_changed = true;
 }
 
@@ -314,6 +347,8 @@ static inline bool sugov_update_single_common(struct sugov_cpu *sg_cpu,
 					      u64 time, unsigned long max_cap,
 					      unsigned int flags)
 {
+	unsigned long boost;
+
 	sugov_iowait_boost(sg_cpu, time, flags);
 	sg_cpu->last_update = time;
 
@@ -322,8 +357,8 @@ static inline bool sugov_update_single_common(struct sugov_cpu *sg_cpu,
 	if (!sugov_should_update_freq(sg_cpu->sg_policy, time))
 		return false;
 
-	sugov_get_util(sg_cpu);
-	sugov_iowait_apply(sg_cpu, time, max_cap);
+	boost = sugov_iowait_apply(sg_cpu, time, max_cap);
+	sugov_get_util(sg_cpu, boost);
 
 	return true;
 }
@@ -407,8 +442,8 @@ static void sugov_update_single_perf(struct update_util_data *hook, u64 time,
 	    sugov_cpu_is_busy(sg_cpu) && sg_cpu->util < prev_util)
 		sg_cpu->util = prev_util;
 
-	cpufreq_driver_adjust_perf(sg_cpu->cpu, map_util_perf(sg_cpu->bw_dl),
-				   map_util_perf(sg_cpu->util), max_cap);
+	cpufreq_driver_adjust_perf(sg_cpu->cpu, sg_cpu->bw_min,
+				   sg_cpu->util, max_cap);
 
 	sg_cpu->sg_policy->last_freq_update_time = time;
 }
@@ -424,9 +459,10 @@ static unsigned int sugov_next_freq_shared(struct sugov_cpu *sg_cpu, u64 time)
 
 	for_each_cpu(j, policy->cpus) {
 		struct sugov_cpu *j_sg_cpu = &per_cpu(sugov_cpu, j);
+		unsigned long boost;
 
-		sugov_get_util(j_sg_cpu);
-		sugov_iowait_apply(j_sg_cpu, time, max_cap);
+		boost = sugov_iowait_apply(j_sg_cpu, time, max_cap);
+		sugov_get_util(j_sg_cpu, boost);
 
 		util = max(j_sg_cpu->util, util);
 	}
diff --git a/kernel/sched/deadline.c b/kernel/sched/deadline.c
index b28114478b82..a04a436af8cc 100644
--- a/kernel/sched/deadline.c
+++ b/kernel/sched/deadline.c
@@ -54,8 +54,14 @@ static int __init sched_dl_sysctl_init(void)
 late_initcall(sched_dl_sysctl_init);
 #endif
 
+static bool dl_server(struct sched_dl_entity *dl_se)
+{
+	return dl_se->dl_server;
+}
+
 static inline struct task_struct *dl_task_of(struct sched_dl_entity *dl_se)
 {
+	BUG_ON(dl_server(dl_se));
 	return container_of(dl_se, struct task_struct, dl);
 }
 
@@ -64,12 +70,19 @@ static inline struct rq *rq_of_dl_rq(struct dl_rq *dl_rq)
 	return container_of(dl_rq, struct rq, dl);
 }
 
-static inline struct dl_rq *dl_rq_of_se(struct sched_dl_entity *dl_se)
+static inline struct rq *rq_of_dl_se(struct sched_dl_entity *dl_se)
 {
-	struct task_struct *p = dl_task_of(dl_se);
-	struct rq *rq = task_rq(p);
+	struct rq *rq = dl_se->rq;
+
+	if (!dl_server(dl_se))
+		rq = task_rq(dl_task_of(dl_se));
+
+	return rq;
+}
 
-	return &rq->dl;
+static inline struct dl_rq *dl_rq_of_se(struct sched_dl_entity *dl_se)
+{
+	return &rq_of_dl_se(dl_se)->dl;
 }
 
 static inline int on_dl_rq(struct sched_dl_entity *dl_se)
@@ -335,6 +348,8 @@ static void dl_change_utilization(struct task_struct *p, u64 new_bw)
 	__add_rq_bw(new_bw, &rq->dl);
 }
 
+static void __dl_clear_params(struct sched_dl_entity *dl_se);
+
 /*
  * The utilization of a task cannot be immediately removed from
  * the rq active utilization (running_bw) when the task blocks.
@@ -389,12 +404,11 @@ static void dl_change_utilization(struct task_struct *p, u64 new_bw)
  * up, and checks if the task is still in the "ACTIVE non contending"
  * state or not (in the second case, it updates running_bw).
  */
-static void task_non_contending(struct task_struct *p)
+static void task_non_contending(struct sched_dl_entity *dl_se)
 {
-	struct sched_dl_entity *dl_se = &p->dl;
 	struct hrtimer *timer = &dl_se->inactive_timer;
-	struct dl_rq *dl_rq = dl_rq_of_se(dl_se);
-	struct rq *rq = rq_of_dl_rq(dl_rq);
+	struct rq *rq = rq_of_dl_se(dl_se);
+	struct dl_rq *dl_rq = &rq->dl;
 	s64 zerolag_time;
 
 	/*
@@ -424,24 +438,33 @@ static void task_non_contending(struct task_struct *p)
 	 * utilization now, instead of starting a timer
 	 */
 	if ((zerolag_time < 0) || hrtimer_active(&dl_se->inactive_timer)) {
-		if (dl_task(p))
+		if (dl_server(dl_se)) {
 			sub_running_bw(dl_se, dl_rq);
-		if (!dl_task(p) || READ_ONCE(p->__state) == TASK_DEAD) {
-			struct dl_bw *dl_b = dl_bw_of(task_cpu(p));
-
-			if (READ_ONCE(p->__state) == TASK_DEAD)
-				sub_rq_bw(&p->dl, &rq->dl);
-			raw_spin_lock(&dl_b->lock);
-			__dl_sub(dl_b, p->dl.dl_bw, dl_bw_cpus(task_cpu(p)));
-			raw_spin_unlock(&dl_b->lock);
-			__dl_clear_params(p);
+		} else {
+			struct task_struct *p = dl_task_of(dl_se);
+
+			if (dl_task(p))
+				sub_running_bw(dl_se, dl_rq);
+
+			if (!dl_task(p) || READ_ONCE(p->__state) == TASK_DEAD) {
+				struct dl_bw *dl_b = dl_bw_of(task_cpu(p));
+
+				if (READ_ONCE(p->__state) == TASK_DEAD)
+					sub_rq_bw(dl_se, &rq->dl);
+				raw_spin_lock(&dl_b->lock);
+				__dl_sub(dl_b, dl_se->dl_bw, dl_bw_cpus(task_cpu(p)));
+				raw_spin_unlock(&dl_b->lock);
+				__dl_clear_params(dl_se);
+			}
 		}
 
 		return;
 	}
 
 	dl_se->dl_non_contending = 1;
-	get_task_struct(p);
+	if (!dl_server(dl_se))
+		get_task_struct(dl_task_of(dl_se));
+
 	hrtimer_start(timer, ns_to_ktime(zerolag_time), HRTIMER_MODE_REL_HARD);
 }
 
@@ -468,8 +491,10 @@ static void task_contending(struct sched_dl_entity *dl_se, int flags)
 		 * will not touch the rq's active utilization,
 		 * so we are still safe.
 		 */
-		if (hrtimer_try_to_cancel(&dl_se->inactive_timer) == 1)
-			put_task_struct(dl_task_of(dl_se));
+		if (hrtimer_try_to_cancel(&dl_se->inactive_timer) == 1) {
+			if (!dl_server(dl_se))
+				put_task_struct(dl_task_of(dl_se));
+		}
 	} else {
 		/*
 		 * Since "dl_non_contending" is not set, the
@@ -482,10 +507,8 @@ static void task_contending(struct sched_dl_entity *dl_se, int flags)
 	}
 }
 
-static inline int is_leftmost(struct task_struct *p, struct dl_rq *dl_rq)
+static inline int is_leftmost(struct sched_dl_entity *dl_se, struct dl_rq *dl_rq)
 {
-	struct sched_dl_entity *dl_se = &p->dl;
-
 	return rb_first_cached(&dl_rq->root) == &dl_se->rb_node;
 }
 
@@ -737,8 +760,10 @@ static inline void deadline_queue_pull_task(struct rq *rq)
 }
 #endif /* CONFIG_SMP */
 
+static void
+enqueue_dl_entity(struct sched_dl_entity *dl_se, int flags);
 static void enqueue_task_dl(struct rq *rq, struct task_struct *p, int flags);
-static void __dequeue_task_dl(struct rq *rq, struct task_struct *p, int flags);
+static void dequeue_dl_entity(struct sched_dl_entity *dl_se, int flags);
 static void wakeup_preempt_dl(struct rq *rq, struct task_struct *p, int flags);
 
 static inline void replenish_dl_new_period(struct sched_dl_entity *dl_se,
@@ -986,8 +1011,7 @@ static inline bool dl_is_implicit(struct sched_dl_entity *dl_se)
  */
 static void update_dl_entity(struct sched_dl_entity *dl_se)
 {
-	struct dl_rq *dl_rq = dl_rq_of_se(dl_se);
-	struct rq *rq = rq_of_dl_rq(dl_rq);
+	struct rq *rq = rq_of_dl_se(dl_se);
 
 	if (dl_time_before(dl_se->deadline, rq_clock(rq)) ||
 	    dl_entity_overflow(dl_se, rq_clock(rq))) {
@@ -1018,11 +1042,11 @@ static inline u64 dl_next_period(struct sched_dl_entity *dl_se)
  * actually started or not (i.e., the replenishment instant is in
  * the future or in the past).
  */
-static int start_dl_timer(struct task_struct *p)
+static int start_dl_timer(struct sched_dl_entity *dl_se)
 {
-	struct sched_dl_entity *dl_se = &p->dl;
 	struct hrtimer *timer = &dl_se->dl_timer;
-	struct rq *rq = task_rq(p);
+	struct dl_rq *dl_rq = dl_rq_of_se(dl_se);
+	struct rq *rq = rq_of_dl_rq(dl_rq);
 	ktime_t now, act;
 	s64 delta;
 
@@ -1056,13 +1080,33 @@ static int start_dl_timer(struct task_struct *p)
 	 * and observe our state.
 	 */
 	if (!hrtimer_is_queued(timer)) {
-		get_task_struct(p);
+		if (!dl_server(dl_se))
+			get_task_struct(dl_task_of(dl_se));
 		hrtimer_start(timer, act, HRTIMER_MODE_ABS_HARD);
 	}
 
 	return 1;
 }
 
+static void __push_dl_task(struct rq *rq, struct rq_flags *rf)
+{
+#ifdef CONFIG_SMP
+	/*
+	 * Queueing this task back might have overloaded rq, check if we need
+	 * to kick someone away.
+	 */
+	if (has_pushable_dl_tasks(rq)) {
+		/*
+		 * Nothing relies on rq->lock after this, so its safe to drop
+		 * rq->lock.
+		 */
+		rq_unpin_lock(rq, rf);
+		push_dl_task(rq);
+		rq_repin_lock(rq, rf);
+	}
+#endif
+}
+
 /*
  * 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
@@ -1081,10 +1125,34 @@ static enum hrtimer_restart dl_task_timer(struct hrtimer *timer)
 	struct sched_dl_entity *dl_se = container_of(timer,
 						     struct sched_dl_entity,
 						     dl_timer);
-	struct task_struct *p = dl_task_of(dl_se);
+	struct task_struct *p;
 	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;
+	}
+
+	p = dl_task_of(dl_se);
 	rq = task_rq_lock(p, &rf);
 
 	/*
@@ -1155,21 +1223,7 @@ static enum hrtimer_restart dl_task_timer(struct hrtimer *timer)
 	else
 		resched_curr(rq);
 
-#ifdef CONFIG_SMP
-	/*
-	 * Queueing this task back might have overloaded rq, check if we need
-	 * to kick someone away.
-	 */
-	if (has_pushable_dl_tasks(rq)) {
-		/*
-		 * Nothing relies on rq->lock after this, so its safe to drop
-		 * rq->lock.
-		 */
-		rq_unpin_lock(rq, &rf);
-		push_dl_task(rq);
-		rq_repin_lock(rq, &rf);
-	}
-#endif
+	__push_dl_task(rq, &rf);
 
 unlock:
 	task_rq_unlock(rq, p, &rf);
@@ -1183,7 +1237,7 @@ unlock:
 	return HRTIMER_NORESTART;
 }
 
-void init_dl_task_timer(struct sched_dl_entity *dl_se)
+static void init_dl_task_timer(struct sched_dl_entity *dl_se)
 {
 	struct hrtimer *timer = &dl_se->dl_timer;
 
@@ -1211,12 +1265,11 @@ void init_dl_task_timer(struct sched_dl_entity *dl_se)
  */
 static inline void dl_check_constrained_dl(struct sched_dl_entity *dl_se)
 {
-	struct task_struct *p = dl_task_of(dl_se);
-	struct rq *rq = rq_of_dl_rq(dl_rq_of_se(dl_se));
+	struct rq *rq = rq_of_dl_se(dl_se);
 
 	if (dl_time_before(dl_se->deadline, rq_clock(rq)) &&
 	    dl_time_before(rq_clock(rq), dl_next_period(dl_se))) {
-		if (unlikely(is_dl_boosted(dl_se) || !start_dl_timer(p)))
+		if (unlikely(is_dl_boosted(dl_se) || !start_dl_timer(dl_se)))
 			return;
 		dl_se->dl_throttled = 1;
 		if (dl_se->runtime > 0)
@@ -1267,44 +1320,19 @@ static u64 grub_reclaim(u64 delta, struct rq *rq, struct sched_dl_entity *dl_se)
 	return (delta * u_act) >> BW_SHIFT;
 }
 
-/*
- * Update the current task's runtime statistics (provided it is still
- * a -deadline task and has not been removed from the dl_rq).
- */
-static void update_curr_dl(struct rq *rq)
+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)
 {
-	struct task_struct *curr = rq->curr;
-	struct sched_dl_entity *dl_se = &curr->dl;
-	u64 delta_exec, scaled_delta_exec;
-	int cpu = cpu_of(rq);
-	u64 now;
-
-	if (!dl_task(curr) || !on_dl_rq(dl_se))
-		return;
+	s64 scaled_delta_exec;
 
-	/*
-	 * Consumed budget is computed considering the time as
-	 * observed by schedulable tasks (excluding time spent
-	 * in hardirq context, etc.). Deadlines are instead
-	 * computed using hard walltime. This seems to be the more
-	 * natural solution, but the full ramifications of this
-	 * approach need further study.
-	 */
-	now = rq_clock_task(rq);
-	delta_exec = now - curr->se.exec_start;
-	if (unlikely((s64)delta_exec <= 0)) {
+	if (unlikely(delta_exec <= 0)) {
 		if (unlikely(dl_se->dl_yielded))
 			goto throttle;
 		return;
 	}
 
-	schedstat_set(curr->stats.exec_max,
-		      max(curr->stats.exec_max, delta_exec));
-
-	trace_sched_stat_runtime(curr, delta_exec, 0);
-
-	update_current_exec_runtime(curr, now, delta_exec);
-
 	if (dl_entity_is_special(dl_se))
 		return;
 
@@ -1316,10 +1344,9 @@ static void update_curr_dl(struct rq *rq)
 	 * according to current frequency and CPU maximum capacity.
 	 */
 	if (unlikely(dl_se->flags & SCHED_FLAG_RECLAIM)) {
-		scaled_delta_exec = grub_reclaim(delta_exec,
-						 rq,
-						 &curr->dl);
+		scaled_delta_exec = grub_reclaim(delta_exec, rq, dl_se);
 	} else {
+		int cpu = cpu_of(rq);
 		unsigned long scale_freq = arch_scale_freq_capacity(cpu);
 		unsigned long scale_cpu = arch_scale_cpu_capacity(cpu);
 
@@ -1338,11 +1365,20 @@ throttle:
 		    (dl_se->flags & SCHED_FLAG_DL_OVERRUN))
 			dl_se->dl_overrun = 1;
 
-		__dequeue_task_dl(rq, curr, 0);
-		if (unlikely(is_dl_boosted(dl_se) || !start_dl_timer(curr)))
-			enqueue_task_dl(rq, curr, ENQUEUE_REPLENISH);
+		dequeue_dl_entity(dl_se, 0);
+		if (!dl_server(dl_se)) {
+			update_stats_dequeue_dl(&rq->dl, dl_se, 0);
+			dequeue_pushable_dl_task(rq, dl_task_of(dl_se));
+		}
+
+		if (unlikely(is_dl_boosted(dl_se) || !start_dl_timer(dl_se))) {
+			if (dl_server(dl_se))
+				enqueue_dl_entity(dl_se, ENQUEUE_REPLENISH);
+			else
+				enqueue_task_dl(rq, dl_task_of(dl_se), ENQUEUE_REPLENISH);
+		}
 
-		if (!is_leftmost(curr, &rq->dl))
+		if (!is_leftmost(dl_se, &rq->dl))
 			resched_curr(rq);
 	}
 
@@ -1372,20 +1408,82 @@ throttle:
 	}
 }
 
+void dl_server_update(struct sched_dl_entity *dl_se, s64 delta_exec)
+{
+	update_curr_dl_se(dl_se->rq, dl_se, delta_exec);
+}
+
+void dl_server_start(struct sched_dl_entity *dl_se)
+{
+	if (!dl_server(dl_se)) {
+		dl_se->dl_server = 1;
+		setup_new_dl_entity(dl_se);
+	}
+	enqueue_dl_entity(dl_se, ENQUEUE_WAKEUP);
+}
+
+void dl_server_stop(struct sched_dl_entity *dl_se)
+{
+	dequeue_dl_entity(dl_se, DEQUEUE_SLEEP);
+}
+
+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_se->rq = rq;
+	dl_se->server_has_tasks = has_tasks;
+	dl_se->server_pick = pick;
+}
+
+/*
+ * Update the current task's runtime statistics (provided it is still
+ * a -deadline task and has not been removed from the dl_rq).
+ */
+static void update_curr_dl(struct rq *rq)
+{
+	struct task_struct *curr = rq->curr;
+	struct sched_dl_entity *dl_se = &curr->dl;
+	s64 delta_exec;
+
+	if (!dl_task(curr) || !on_dl_rq(dl_se))
+		return;
+
+	/*
+	 * Consumed budget is computed considering the time as
+	 * observed by schedulable tasks (excluding time spent
+	 * in hardirq context, etc.). Deadlines are instead
+	 * computed using hard walltime. This seems to be the more
+	 * natural solution, but the full ramifications of this
+	 * approach need further study.
+	 */
+	delta_exec = update_curr_common(rq);
+	update_curr_dl_se(rq, dl_se, delta_exec);
+}
+
 static enum hrtimer_restart inactive_task_timer(struct hrtimer *timer)
 {
 	struct sched_dl_entity *dl_se = container_of(timer,
 						     struct sched_dl_entity,
 						     inactive_timer);
-	struct task_struct *p = dl_task_of(dl_se);
+	struct task_struct *p = NULL;
 	struct rq_flags rf;
 	struct rq *rq;
 
-	rq = task_rq_lock(p, &rf);
+	if (!dl_server(dl_se)) {
+		p = dl_task_of(dl_se);
+		rq = task_rq_lock(p, &rf);
+	} else {
+		rq = dl_se->rq;
+		rq_lock(rq, &rf);
+	}
 
 	sched_clock_tick();
 	update_rq_clock(rq);
 
+	if (dl_server(dl_se))
+		goto no_task;
+
 	if (!dl_task(p) || READ_ONCE(p->__state) == TASK_DEAD) {
 		struct dl_bw *dl_b = dl_bw_of(task_cpu(p));
 
@@ -1398,23 +1496,30 @@ static enum hrtimer_restart inactive_task_timer(struct hrtimer *timer)
 		raw_spin_lock(&dl_b->lock);
 		__dl_sub(dl_b, p->dl.dl_bw, dl_bw_cpus(task_cpu(p)));
 		raw_spin_unlock(&dl_b->lock);
-		__dl_clear_params(p);
+		__dl_clear_params(dl_se);
 
 		goto unlock;
 	}
+
+no_task:
 	if (dl_se->dl_non_contending == 0)
 		goto unlock;
 
 	sub_running_bw(dl_se, &rq->dl);
 	dl_se->dl_non_contending = 0;
 unlock:
-	task_rq_unlock(rq, p, &rf);
-	put_task_struct(p);
+
+	if (!dl_server(dl_se)) {
+		task_rq_unlock(rq, p, &rf);
+		put_task_struct(p);
+	} else {
+		rq_unlock(rq, &rf);
+	}
 
 	return HRTIMER_NORESTART;
 }
 
-void init_dl_inactive_task_timer(struct sched_dl_entity *dl_se)
+static void init_dl_inactive_task_timer(struct sched_dl_entity *dl_se)