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
  ...

1 files changed, 82 insertions, 58 deletions

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 */