From a8b62fd0850503cf1e557d7e5a98d3f1f5c25eef Mon Sep 17 00:00:00 2001
From: Peter Zijlstra <peterz@infradead.org>
Date: Mon, 21 Sep 2020 12:58:17 +0200
Subject: stop_machine: Add function and caller debug info

Crashes in stop-machine are hard to connect to the calling code, add a
little something to help with that.

Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Valentin Schneider <valentin.schneider@arm.com>
Reviewed-by: Daniel Bristot de Oliveira <bristot@redhat.com>
Link: https://lkml.kernel.org/r/20201023102346.116513635@infradead.org
---
 include/linux/stop_machine.h | 5 +++++
 1 file changed, 5 insertions(+)

(limited to 'include/linux')

diff --git a/include/linux/stop_machine.h b/include/linux/stop_machine.h
index 76d8b09384a7..30577c3aecf8 100644
--- a/include/linux/stop_machine.h
+++ b/include/linux/stop_machine.h
@@ -24,6 +24,7 @@ typedef int (*cpu_stop_fn_t)(void *arg);
 struct cpu_stop_work {
 	struct list_head	list;		/* cpu_stopper->works */
 	cpu_stop_fn_t		fn;
+	unsigned long		caller;
 	void			*arg;
 	struct cpu_stop_done	*done;
 };
@@ -36,6 +37,8 @@ void stop_machine_park(int cpu);
 void stop_machine_unpark(int cpu);
 void stop_machine_yield(const struct cpumask *cpumask);
 
+extern void print_stop_info(const char *log_lvl, struct task_struct *task);
+
 #else	/* CONFIG_SMP */
 
 #include <linux/workqueue.h>
@@ -80,6 +83,8 @@ static inline bool stop_one_cpu_nowait(unsigned int cpu,
 	return false;
 }
 
+static inline void print_stop_info(const char *log_lvl, struct task_struct *task) { }
+
 #endif	/* CONFIG_SMP */
 
 /*
-- 
cgit v1.2.3-71-gd317


From 1cf12e08bc4d50a76b80c42a3109c53d8794a0c9 Mon Sep 17 00:00:00 2001
From: Thomas Gleixner <tglx@linutronix.de>
Date: Wed, 16 Sep 2020 09:27:18 +0200
Subject: sched/hotplug: Consolidate task migration on CPU unplug

With the new mechanism which kicks tasks off the outgoing CPU at the end of
schedule() the situation on an outgoing CPU right before the stopper thread
brings it down completely is:

 - All user tasks and all unbound kernel threads have either been migrated
   away or are not running and the next wakeup will move them to a online CPU.

 - All per CPU kernel threads, except cpu hotplug thread and the stopper
   thread have either been unbound or parked by the responsible CPU hotplug
   callback.

That means that at the last step before the stopper thread is invoked the
cpu hotplug thread is the last legitimate running task on the outgoing
CPU.

Add a final wait step right before the stopper thread is kicked which
ensures that any still running tasks on the way to park or on the way to
kick themself of the CPU are either sleeping or gone.

This allows to remove the migrate_tasks() crutch in sched_cpu_dying(). If
sched_cpu_dying() detects that there is still another running task aside of
the stopper thread then it will explode with the appropriate fireworks.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Valentin Schneider <valentin.schneider@arm.com>
Reviewed-by: Daniel Bristot de Oliveira <bristot@redhat.com>
Link: https://lkml.kernel.org/r/20201023102346.547163969@infradead.org
---
 include/linux/cpuhotplug.h    |   1 +
 include/linux/sched/hotplug.h |   2 +
 kernel/cpu.c                  |   9 ++-
 kernel/sched/core.c           | 154 ++++++++++--------------------------------
 4 files changed, 46 insertions(+), 120 deletions(-)

(limited to 'include/linux')

diff --git a/include/linux/cpuhotplug.h b/include/linux/cpuhotplug.h
index bc56287a1ed1..0042ef362511 100644
--- a/include/linux/cpuhotplug.h
+++ b/include/linux/cpuhotplug.h
@@ -152,6 +152,7 @@ enum cpuhp_state {
 	CPUHP_AP_ONLINE,
 	CPUHP_TEARDOWN_CPU,
 	CPUHP_AP_ONLINE_IDLE,
+	CPUHP_AP_SCHED_WAIT_EMPTY,
 	CPUHP_AP_SMPBOOT_THREADS,
 	CPUHP_AP_X86_VDSO_VMA_ONLINE,
 	CPUHP_AP_IRQ_AFFINITY_ONLINE,
diff --git a/include/linux/sched/hotplug.h b/include/linux/sched/hotplug.h
index 9a62ffdd296f..412cdaba33eb 100644
--- a/include/linux/sched/hotplug.h
+++ b/include/linux/sched/hotplug.h
@@ -11,8 +11,10 @@ extern int sched_cpu_activate(unsigned int cpu);
 extern int sched_cpu_deactivate(unsigned int cpu);
 
 #ifdef CONFIG_HOTPLUG_CPU
+extern int sched_cpu_wait_empty(unsigned int cpu);
 extern int sched_cpu_dying(unsigned int cpu);
 #else
+# define sched_cpu_wait_empty	NULL
 # define sched_cpu_dying	NULL
 #endif
 
diff --git a/kernel/cpu.c b/kernel/cpu.c
index 6ff2578ecf17..fa535eaa4826 100644
--- a/kernel/cpu.c
+++ b/kernel/cpu.c
@@ -1602,7 +1602,7 @@ static struct cpuhp_step cpuhp_hp_states[] = {
 		.name			= "ap:online",
 	},
 	/*
-	 * Handled on controll processor until the plugged processor manages
+	 * Handled on control processor until the plugged processor manages
 	 * this itself.
 	 */
 	[CPUHP_TEARDOWN_CPU] = {
@@ -1611,6 +1611,13 @@ static struct cpuhp_step cpuhp_hp_states[] = {
 		.teardown.single	= takedown_cpu,
 		.cant_stop		= true,
 	},
+
+	[CPUHP_AP_SCHED_WAIT_EMPTY] = {
+		.name			= "sched:waitempty",
+		.startup.single		= NULL,
+		.teardown.single	= sched_cpu_wait_empty,
+	},
+
 	/* Handle smpboot threads park/unpark */
 	[CPUHP_AP_SMPBOOT_THREADS] = {
 		.name			= "smpboot/threads:online",
diff --git a/kernel/sched/core.c b/kernel/sched/core.c
index e1093c443ff9..6c89806c834b 100644
--- a/kernel/sched/core.c
+++ b/kernel/sched/core.c
@@ -6741,120 +6741,6 @@ void idle_task_exit(void)
 	/* finish_cpu(), as ran on the BP, will clean up the active_mm state */
 }
 
-/*
- * Since this CPU is going 'away' for a while, fold any nr_active delta
- * we might have. Assumes we're called after migrate_tasks() so that the
- * nr_active count is stable. We need to take the teardown thread which
- * is calling this into account, so we hand in adjust = 1 to the load
- * calculation.
- *
- * Also see the comment "Global load-average calculations".
- */
-static void calc_load_migrate(struct rq *rq)
-{
-	long delta = calc_load_fold_active(rq, 1);
-	if (delta)
-		atomic_long_add(delta, &calc_load_tasks);
-}
-
-static struct task_struct *__pick_migrate_task(struct rq *rq)
-{
-	const struct sched_class *class;
-	struct task_struct *next;
-
-	for_each_class(class) {
-		next = class->pick_next_task(rq);
-		if (next) {
-			next->sched_class->put_prev_task(rq, next);
-			return next;
-		}
-	}
-
-	/* The idle class should always have a runnable task */
-	BUG();
-}
-
-/*
- * Migrate all tasks from the rq, sleeping tasks will be migrated by
- * try_to_wake_up()->select_task_rq().
- *
- * Called with rq->lock held even though we'er in stop_machine() and
- * there's no concurrency possible, we hold the required locks anyway
- * because of lock validation efforts.
- */
-static void migrate_tasks(struct rq *dead_rq, struct rq_flags *rf)
-{
-	struct rq *rq = dead_rq;
-	struct task_struct *next, *stop = rq->stop;
-	struct rq_flags orf = *rf;
-	int dest_cpu;
-
-	/*
-	 * Fudge the rq selection such that the below task selection loop
-	 * doesn't get stuck on the currently eligible stop task.
-	 *
-	 * We're currently inside stop_machine() and the rq is either stuck
-	 * in the stop_machine_cpu_stop() loop, or we're executing this code,
-	 * either way we should never end up calling schedule() until we're
-	 * done here.
-	 */
-	rq->stop = NULL;
-
-	/*
-	 * put_prev_task() and pick_next_task() sched
-	 * class method both need to have an up-to-date
-	 * value of rq->clock[_task]
-	 */
-	update_rq_clock(rq);
-
-	for (;;) {
-		/*
-		 * There's this thread running, bail when that's the only
-		 * remaining thread:
-		 */
-		if (rq->nr_running == 1)
-			break;
-
-		next = __pick_migrate_task(rq);
-
-		/*
-		 * Rules for changing task_struct::cpus_mask are holding
-		 * both pi_lock and rq->lock, such that holding either
-		 * stabilizes the mask.
-		 *
-		 * Drop rq->lock is not quite as disastrous as it usually is
-		 * because !cpu_active at this point, which means load-balance
-		 * will not interfere. Also, stop-machine.
-		 */
-		rq_unlock(rq, rf);
-		raw_spin_lock(&next->pi_lock);
-		rq_relock(rq, rf);
-
-		/*
-		 * Since we're inside stop-machine, _nothing_ should have
-		 * changed the task, WARN if weird stuff happened, because in
-		 * that case the above rq->lock drop is a fail too.
-		 */
-		if (WARN_ON(task_rq(next) != rq || !task_on_rq_queued(next))) {
-			raw_spin_unlock(&next->pi_lock);
-			continue;
-		}
-
-		/* Find suitable destination for @next, with force if needed. */
-		dest_cpu = select_fallback_rq(dead_rq->cpu, next);
-		rq = __migrate_task(rq, rf, next, dest_cpu);
-		if (rq != dead_rq) {
-			rq_unlock(rq, rf);
-			rq = dead_rq;
-			*rf = orf;
-			rq_relock(rq, rf);
-		}
-		raw_spin_unlock(&next->pi_lock);
-	}
-
-	rq->stop = stop;
-}
-
 static int __balance_push_cpu_stop(void *arg)
 {
 	struct task_struct *p = arg;
@@ -7123,10 +7009,6 @@ int sched_cpu_deactivate(unsigned int cpu)
 		return ret;
 	}
 	sched_domains_numa_masks_clear(cpu);
-
-	/* Wait for all non per CPU kernel threads to vanish. */
-	balance_hotplug_wait();
-
 	return 0;
 }
 
@@ -7146,6 +7028,41 @@ int sched_cpu_starting(unsigned int cpu)
 }
 
 #ifdef CONFIG_HOTPLUG_CPU
+
+/*
+ * Invoked immediately before the stopper thread is invoked to bring the
+ * CPU down completely. At this point all per CPU kthreads except the
+ * hotplug thread (current) and the stopper thread (inactive) have been
+ * either parked or have been unbound from the outgoing CPU. Ensure that
+ * any of those which might be on the way out are gone.
+ *
+ * If after this point a bound task is being woken on this CPU then the
+ * responsible hotplug callback has failed to do it's job.
+ * sched_cpu_dying() will catch it with the appropriate fireworks.
+ */
+int sched_cpu_wait_empty(unsigned int cpu)
+{
+	balance_hotplug_wait();
+	return 0;
+}
+
+/*
+ * Since this CPU is going 'away' for a while, fold any nr_active delta we
+ * might have. Called from the CPU stopper task after ensuring that the
+ * stopper is the last running task on the CPU, so nr_active count is
+ * stable. We need to take the teardown thread which is calling this into
+ * account, so we hand in adjust = 1 to the load calculation.
+ *
+ * Also see the comment "Global load-average calculations".
+ */
+static void calc_load_migrate(struct rq *rq)
+{
+	long delta = calc_load_fold_active(rq, 1);
+
+	if (delta)
+		atomic_long_add(delta, &calc_load_tasks);
+}
+
 int sched_cpu_dying(unsigned int cpu)
 {
 	struct rq *rq = cpu_rq(cpu);
@@ -7159,7 +7076,6 @@ int sched_cpu_dying(unsigned int cpu)
 		BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span));
 		set_rq_offline(rq);
 	}
-	migrate_tasks(rq, &rf);
 	BUG_ON(rq->nr_running != 1);
 	rq_unlock_irqrestore(rq, &rf);
 
-- 
cgit v1.2.3-71-gd317


From af449901b84c98cbd84a0113223ba3bcfcb12a26 Mon Sep 17 00:00:00 2001
From: Peter Zijlstra <peterz@infradead.org>
Date: Thu, 17 Sep 2020 10:38:30 +0200
Subject: sched: Add migrate_disable()

Add the base migrate_disable() support (under protest).

While migrate_disable() is (currently) required for PREEMPT_RT, it is
also one of the biggest flaws in the system.

Notably this is just the base implementation, it is broken vs
sched_setaffinity() and hotplug, both solved in additional patches for
ease of review.

Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Valentin Schneider <valentin.schneider@arm.com>
Reviewed-by: Daniel Bristot de Oliveira <bristot@redhat.com>
Link: https://lkml.kernel.org/r/20201023102346.818170844@infradead.org
---
 include/linux/preempt.h |  65 ++++++++++++++++++++++++++++
 include/linux/sched.h   |   3 ++
 kernel/sched/core.c     | 112 +++++++++++++++++++++++++++++++++++++++++++++---
 kernel/sched/sched.h    |   6 ++-
 lib/smp_processor_id.c  |   5 +++
 5 files changed, 183 insertions(+), 8 deletions(-)

(limited to 'include/linux')

diff --git a/include/linux/preempt.h b/include/linux/preempt.h
index 7d9c1c0e149c..97ba7c920653 100644
--- a/include/linux/preempt.h
+++ b/include/linux/preempt.h
@@ -322,6 +322,69 @@ static inline void preempt_notifier_init(struct preempt_notifier *notifier,
 
 #endif
 
+#if defined(CONFIG_SMP) && defined(CONFIG_PREEMPT_RT)
+
+/*
+ * Migrate-Disable and why it is (strongly) undesired.
+ *
+ * The premise of the Real-Time schedulers we have on Linux
+ * (SCHED_FIFO/SCHED_DEADLINE) is that M CPUs can/will run M tasks
+ * concurrently, provided there are sufficient runnable tasks, also known as
+ * work-conserving. For instance SCHED_DEADLINE tries to schedule the M
+ * earliest deadline threads, and SCHED_FIFO the M highest priority threads.
+ *
+ * The correctness of various scheduling models depends on this, but is it
+ * broken by migrate_disable() that doesn't imply preempt_disable(). Where
+ * preempt_disable() implies an immediate priority ceiling, preemptible
+ * migrate_disable() allows nesting.
+ *
+ * The worst case is that all tasks preempt one another in a migrate_disable()
+ * region and stack on a single CPU. This then reduces the available bandwidth
+ * to a single CPU. And since Real-Time schedulability theory considers the
+ * Worst-Case only, all Real-Time analysis shall revert to single-CPU
+ * (instantly solving the SMP analysis problem).
+ *
+ *
+ * The reason we have it anyway.
+ *
+ * PREEMPT_RT breaks a number of assumptions traditionally held. By forcing a
+ * number of primitives into becoming preemptible, they would also allow
+ * migration. This turns out to break a bunch of per-cpu usage. To this end,
+ * all these primitives employ migirate_disable() to restore this implicit
+ * assumption.
+ *
+ * This is a 'temporary' work-around at best. The correct solution is getting
+ * rid of the above assumptions and reworking the code to employ explicit
+ * per-cpu locking or short preempt-disable regions.
+ *
+ * The end goal must be to get rid of migrate_disable(), alternatively we need
+ * a schedulability theory that does not depend on abritrary migration.
+ *
+ *
+ * Notes on the implementation.
+ *
+ * The implementation is particularly tricky since existing code patterns
+ * dictate neither migrate_disable() nor migrate_enable() is allowed to block.
+ * This means that it cannot use cpus_read_lock() to serialize against hotplug,
+ * nor can it easily migrate itself into a pending affinity mask change on
+ * migrate_enable().
+ *
+ *
+ * Note: even non-work-conserving schedulers like semi-partitioned depends on
+ *       migration, so migrate_disable() is not only a problem for
+ *       work-conserving schedulers.
+ *
+ */
+extern void migrate_disable(void);
+extern void migrate_enable(void);
+
+#elif defined(CONFIG_PREEMPT_RT)
+
+static inline void migrate_disable(void) { }
+static inline void migrate_enable(void) { }
+
+#else /* !CONFIG_PREEMPT_RT */
+
 /**
  * migrate_disable - Prevent migration of the current task
  *
@@ -352,4 +415,6 @@ static __always_inline void migrate_enable(void)
 	preempt_enable();
 }
 
+#endif /* CONFIG_SMP && CONFIG_PREEMPT_RT */
+
 #endif /* __LINUX_PREEMPT_H */
diff --git a/include/linux/sched.h b/include/linux/sched.h
index 063cd120b459..0732356c0eca 100644
--- a/include/linux/sched.h
+++ b/include/linux/sched.h
@@ -714,6 +714,9 @@ struct task_struct {
 	int				nr_cpus_allowed;
 	const cpumask_t			*cpus_ptr;
 	cpumask_t			cpus_mask;
+#if defined(CONFIG_SMP) && defined(CONFIG_PREEMPT_RT)
+	int				migration_disabled;
+#endif
 
 #ifdef CONFIG_PREEMPT_RCU
 	int				rcu_read_lock_nesting;
diff --git a/kernel/sched/core.c b/kernel/sched/core.c
index 396accb1d69c..6a3f1c2e185b 100644
--- a/kernel/sched/core.c
+++ b/kernel/sched/core.c
@@ -1696,6 +1696,61 @@ void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags)
 
 #ifdef CONFIG_SMP
 
+#ifdef CONFIG_PREEMPT_RT
+
+static void
+__do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask, u32 flags);
+
+static int __set_cpus_allowed_ptr(struct task_struct *p,
+				  const struct cpumask *new_mask,
+				  u32 flags);
+
+static void migrate_disable_switch(struct rq *rq, struct task_struct *p)
+{
+	if (likely(!p->migration_disabled))
+		return;
+
+	if (p->cpus_ptr != &p->cpus_mask)
+		return;
+
+	/*
+	 * Violates locking rules! see comment in __do_set_cpus_allowed().
+	 */
+	__do_set_cpus_allowed(p, cpumask_of(rq->cpu), SCA_MIGRATE_DISABLE);
+}
+
+void migrate_disable(void)
+{
+	if (current->migration_disabled++)
+		return;
+
+	barrier();
+}
+EXPORT_SYMBOL_GPL(migrate_disable);
+
+void migrate_enable(void)
+{
+	struct task_struct *p = current;
+
+	if (--p->migration_disabled)
+		return;
+
+	barrier();
+
+	if (p->cpus_ptr == &p->cpus_mask)
+		return;
+
+	__set_cpus_allowed_ptr(p, &p->cpus_mask, SCA_MIGRATE_ENABLE);
+}
+EXPORT_SYMBOL_GPL(migrate_enable);
+
+static inline bool is_migration_disabled(struct task_struct *p)
+{
+	return p->migration_disabled;
+}
+
+#endif
+
 /*
  * Per-CPU kthreads are allowed to run on !active && online CPUs, see
  * __set_cpus_allowed_ptr() and select_fallback_rq().
@@ -1705,7 +1760,7 @@ static inline bool is_cpu_allowed(struct task_struct *p, int cpu)
 	if (!cpumask_test_cpu(cpu, p->cpus_ptr))
 		return false;
 
-	if (is_per_cpu_kthread(p))
+	if (is_per_cpu_kthread(p) || is_migration_disabled(p))
 		return cpu_online(cpu);
 
 	return cpu_active(cpu);
@@ -1826,6 +1881,11 @@ static int migration_cpu_stop(void *data)
  */
 void set_cpus_allowed_common(struct task_struct *p, const struct cpumask *new_mask, u32 flags)
 {
+	if (flags & (SCA_MIGRATE_ENABLE | SCA_MIGRATE_DISABLE)) {
+		p->cpus_ptr = new_mask;
+		return;
+	}
+
 	cpumask_copy(&p->cpus_mask, new_mask);
 	p->nr_cpus_allowed = cpumask_weight(new_mask);
 }
@@ -1836,7 +1896,22 @@ __do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask, u32
 	struct rq *rq = task_rq(p);
 	bool queued, running;
 
-	lockdep_assert_held(&p->pi_lock);
+	/*
+	 * This here violates the locking rules for affinity, since we're only
+	 * supposed to change these variables while holding both rq->lock and
+	 * p->pi_lock.
+	 *
+	 * HOWEVER, it magically works, because ttwu() is the only code that
+	 * accesses these variables under p->pi_lock and only does so after
+	 * smp_cond_load_acquire(&p->on_cpu, !VAL), and we're in __schedule()
+	 * before finish_task().
+	 *
+	 * XXX do further audits, this smells like something putrid.
+	 */
+	if (flags & SCA_MIGRATE_DISABLE)
+		SCHED_WARN_ON(!p->on_cpu);
+	else
+		lockdep_assert_held(&p->pi_lock);
 
 	queued = task_on_rq_queued(p);
 	running = task_current(rq, p);
@@ -1887,9 +1962,14 @@ static int __set_cpus_allowed_ptr(struct task_struct *p,
 	rq = task_rq_lock(p, &rf);
 	update_rq_clock(rq);
 
-	if (p->flags & PF_KTHREAD) {
+	if (p->flags & PF_KTHREAD || is_migration_disabled(p)) {
 		/*
-		 * Kernel threads are allowed on online && !active CPUs
+		 * Kernel threads are allowed on online && !active CPUs.
+		 *
+		 * Specifically, migration_disabled() tasks must not fail the
+		 * cpumask_any_and_distribute() pick below, esp. so on
+		 * SCA_MIGRATE_ENABLE, otherwise we'll not call
+		 * set_cpus_allowed_common() and actually reset p->cpus_ptr.
 		 */
 		cpu_valid_mask = cpu_online_mask;
 	}
@@ -1903,7 +1983,7 @@ static int __set_cpus_allowed_ptr(struct task_struct *p,
 		goto out;
 	}
 
-	if (cpumask_equal(&p->cpus_mask, new_mask))
+	if (!(flags & SCA_MIGRATE_ENABLE) && cpumask_equal(&p->cpus_mask, new_mask))
 		goto out;
 
 	/*
@@ -1995,6 +2075,8 @@ void set_task_cpu(struct task_struct *p, unsigned int new_cpu)
 	 * Clearly, migrating tasks to offline CPUs is a fairly daft thing.
 	 */
 	WARN_ON_ONCE(!cpu_online(new_cpu));
+
+	WARN_ON_ONCE(is_migration_disabled(p));
 #endif
 
 	trace_sched_migrate_task(p, new_cpu);
@@ -2325,6 +2407,12 @@ static int select_fallback_rq(int cpu, struct task_struct *p)
 			}
 			fallthrough;
 		case possible:
+			/*
+			 * XXX When called from select_task_rq() we only
+			 * hold p->pi_lock and again violate locking order.
+			 *
+			 * More yuck to audit.
+			 */
 			do_set_cpus_allowed(p, cpu_possible_mask);
 			state = fail;
 			break;
@@ -2359,7 +2447,7 @@ int select_task_rq(struct task_struct *p, int cpu, int sd_flags, int wake_flags)
 {
 	lockdep_assert_held(&p->pi_lock);
 
-	if (p->nr_cpus_allowed > 1)
+	if (p->nr_cpus_allowed > 1 && !is_migration_disabled(p))
 		cpu = p->sched_class->select_task_rq(p, cpu, sd_flags, wake_flags);
 	else
 		cpu = cpumask_any(p->cpus_ptr);
@@ -2421,6 +2509,17 @@ static inline int __set_cpus_allowed_ptr(struct task_struct *p,
 
 #endif /* CONFIG_SMP */
 
+#if !defined(CONFIG_SMP) || !defined(CONFIG_PREEMPT_RT)
+
+static inline void migrate_disable_switch(struct rq *rq, struct task_struct *p) { }
+
+static inline bool is_migration_disabled(struct task_struct *p)
+{
+	return false;
+}
+
+#endif
+
 static void
 ttwu_stat(struct task_struct *p, int cpu, int wake_flags)
 {
@@ -4570,6 +4669,7 @@ static void __sched notrace __schedule(bool preempt)
 		 */
 		++*switch_count;
 
+		migrate_disable_switch(rq, prev);
 		psi_sched_switch(prev, next, !task_on_rq_queued(prev));
 
 		trace_sched_switch(preempt, prev, next);
diff --git a/kernel/sched/sched.h b/kernel/sched/sched.h
index 0420d80fb250..72d8e47cf0bb 100644
--- a/kernel/sched/sched.h
+++ b/kernel/sched/sched.h
@@ -1902,14 +1902,16 @@ static inline bool sched_fair_runnable(struct rq *rq)
 extern struct task_struct *pick_next_task_fair(struct rq *rq, struct task_struct *prev, struct rq_flags *rf);
 extern struct task_struct *pick_next_task_idle(struct rq *rq);
 
+#define SCA_CHECK		0x01
+#define SCA_MIGRATE_DISABLE	0x02
+#define SCA_MIGRATE_ENABLE	0x04
+
 #ifdef CONFIG_SMP
 
 extern void update_group_capacity(struct sched_domain *sd, int cpu);
 
 extern void trigger_load_balance(struct rq *rq);
 
-#define SCA_CHECK		0x01
-
 extern void set_cpus_allowed_common(struct task_struct *p, const struct cpumask *new_mask, u32 flags);
 
 #endif
diff --git a/lib/smp_processor_id.c b/lib/smp_processor_id.c
index 525222e4f409..faaa927ac2c8 100644
--- a/lib/smp_processor_id.c
+++ b/lib/smp_processor_id.c
@@ -26,6 +26,11 @@ unsigned int check_preemption_disabled(const char *what1, const char *what2)
 	if (current->nr_cpus_allowed == 1)
 		goto out;
 
+#if defined(CONFIG_SMP) && defined(CONFIG_PREEMPT_RT)
+	if (current->migration_disabled)
+		goto out;
+#endif
+
 	/*
 	 * It is valid to assume CPU-locality during early bootup:
 	 */
-- 
cgit v1.2.3-71-gd317


From 6d337eab041d56bb8f0e7794f39906c21054c512 Mon Sep 17 00:00:00 2001
From: Peter Zijlstra <peterz@infradead.org>
Date: Fri, 18 Sep 2020 17:24:31 +0200
Subject: sched: Fix migrate_disable() vs set_cpus_allowed_ptr()

Concurrent migrate_disable() and set_cpus_allowed_ptr() has
interesting features. We rely on set_cpus_allowed_ptr() to not return
until the task runs inside the provided mask. This expectation is
exported to userspace.

This means that any set_cpus_allowed_ptr() caller must wait until
migrate_enable() allows migrations.

At the same time, we don't want migrate_enable() to schedule, due to
patterns like:

	preempt_disable();
	migrate_disable();
	...
	migrate_enable();
	preempt_enable();

And:

	raw_spin_lock(&B);
	spin_unlock(&A);

this means that when migrate_enable() must restore the affinity
mask, it cannot wait for completion thereof. Luck will have it that
that is exactly the case where there is a pending
set_cpus_allowed_ptr(), so let that provide storage for the async stop
machine.

Much thanks to Valentin who used TLA+ most effective and found lots of
'interesting' cases.

Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Valentin Schneider <valentin.schneider@arm.com>
Reviewed-by: Daniel Bristot de Oliveira <bristot@redhat.com>
Link: https://lkml.kernel.org/r/20201023102346.921768277@infradead.org
---
 include/linux/sched.h |   1 +
 kernel/sched/core.c   | 236 +++++++++++++++++++++++++++++++++++++++++++-------
 2 files changed, 207 insertions(+), 30 deletions(-)

(limited to 'include/linux')

diff --git a/include/linux/sched.h b/include/linux/sched.h
index 0732356c0eca..90a0c92741d7 100644
--- a/include/linux/sched.h
+++ b/include/linux/sched.h
@@ -714,6 +714,7 @@ struct task_struct {
 	int				nr_cpus_allowed;
 	const cpumask_t			*cpus_ptr;
 	cpumask_t			cpus_mask;
+	void				*migration_pending;
 #if defined(CONFIG_SMP) && defined(CONFIG_PREEMPT_RT)
 	int				migration_disabled;
 #endif
diff --git a/kernel/sched/core.c b/kernel/sched/core.c
index 6a3f1c2e185b..0efc1e41bb60 100644
--- a/kernel/sched/core.c
+++ b/kernel/sched/core.c
@@ -1732,15 +1732,26 @@ void migrate_enable(void)
 {
 	struct task_struct *p = current;
 
-	if (--p->migration_disabled)
+	if (p->migration_disabled > 1) {
+		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().
+	 */
+	preempt_disable();
+	if (p->cpus_ptr != &p->cpus_mask)
+		__set_cpus_allowed_ptr(p, &p->cpus_mask, SCA_MIGRATE_ENABLE);
+	/*
+	 * Mustn't clear migration_disabled() until cpus_ptr points back at the
+	 * regular cpus_mask, otherwise things that race (eg.
+	 * select_fallback_rq) get confused.
+	 */
 	barrier();
-
-	if (p->cpus_ptr == &p->cpus_mask)
-		return;
-
-	__set_cpus_allowed_ptr(p, &p->cpus_mask, SCA_MIGRATE_ENABLE);
+	p->migration_disabled = 0;
+	preempt_enable();
 }
 EXPORT_SYMBOL_GPL(migrate_enable);
 
@@ -1805,8 +1816,16 @@ static struct rq *move_queued_task(struct rq *rq, struct rq_flags *rf,
 }
 
 struct migration_arg {
-	struct task_struct *task;
-	int dest_cpu;
+	struct task_struct		*task;
+	int				dest_cpu;
+	struct set_affinity_pending	*pending;
+};
+
+struct set_affinity_pending {
+	refcount_t		refs;
+	struct completion	done;
+	struct cpu_stop_work	stop_work;
+	struct migration_arg	arg;
 };
 
 /*
@@ -1838,16 +1857,19 @@ static struct rq *__migrate_task(struct rq *rq, struct rq_flags *rf,
  */
 static int migration_cpu_stop(void *data)
 {
+	struct set_affinity_pending *pending;
 	struct migration_arg *arg = data;
 	struct task_struct *p = arg->task;
+	int dest_cpu = arg->dest_cpu;
 	struct rq *rq = this_rq();
+	bool complete = false;
 	struct rq_flags rf;
 
 	/*
 	 * The original target CPU might have gone down and we might
 	 * be on another CPU but it doesn't matter.
 	 */
-	local_irq_disable();
+	local_irq_save(rf.flags);
 	/*
 	 * We need to explicitly wake pending tasks before running
 	 * __migrate_task() such that we will not miss enforcing cpus_ptr
@@ -1857,21 +1879,83 @@ static int migration_cpu_stop(void *data)
 
 	raw_spin_lock(&p->pi_lock);
 	rq_lock(rq, &rf);
+
+	pending = p->migration_pending;
 	/*
 	 * If task_rq(p) != rq, it cannot be migrated here, because we're
 	 * holding rq->lock, if p->on_rq == 0 it cannot get enqueued because
 	 * we're holding p->pi_lock.
 	 */
 	if (task_rq(p) == rq) {
+		if (is_migration_disabled(p))
+			goto out;
+
+		if (pending) {
+			p->migration_pending = NULL;
+			complete = true;
+		}
+
+		/* migrate_enable() --  we must not race against SCA */
+		if (dest_cpu < 0) {
+			/*
+			 * When this was migrate_enable() but we no longer
+			 * have a @pending, a concurrent SCA 'fixed' things
+			 * and we should be valid again. Nothing to do.
+			 */
+			if (!pending) {
+				WARN_ON_ONCE(!is_cpu_allowed(p, cpu_of(rq)));
+				goto out;
+			}
+
+			dest_cpu = cpumask_any_distribute(&p->cpus_mask);
+		}
+
 		if (task_on_rq_queued(p))
-			rq = __migrate_task(rq, &rf, p, arg->dest_cpu);
+			rq = __migrate_task(rq, &rf, p, dest_cpu);
 		else
-			p->wake_cpu = arg->dest_cpu;
+			p->wake_cpu = dest_cpu;
+
+	} else if (dest_cpu < 0) {
+		/*
+		 * This happens when we get migrated between migrate_enable()'s
+		 * preempt_enable() and scheduling the stopper task. At that
+		 * point we're a regular task again and not current anymore.
+		 *
+		 * A !PREEMPT kernel has a giant hole here, which makes it far
+		 * more likely.
+		 */
+
+		/*
+		 * When this was migrate_enable() but we no longer have an
+		 * @pending, a concurrent SCA 'fixed' things and we should be
+		 * valid again. Nothing to do.
+		 */
+		if (!pending) {
+			WARN_ON_ONCE(!is_cpu_allowed(p, cpu_of(rq)));
+			goto out;
+		}
+
+		/*
+		 * When migrate_enable() hits a rq mis-match we can't reliably
+		 * determine is_migration_disabled() and so have to chase after
+		 * it.
+		 */
+		task_rq_unlock(rq, p, &rf);
+		stop_one_cpu_nowait(task_cpu(p), migration_cpu_stop,
+				    &pending->arg, &pending->stop_work);
+		return 0;
 	}
-	rq_unlock(rq, &rf);
-	raw_spin_unlock(&p->pi_lock);
+out:
+	task_rq_unlock(rq, p, &rf);
+
+	if (complete)
+		complete_all(&pending->done);
+
+	/* For pending->{arg,stop_work} */
+	pending = arg->pending;
+	if (pending && refcount_dec_and_test(&pending->refs))
+		wake_up_var(&pending->refs);
 
-	local_irq_enable();
 	return 0;
 }
 
@@ -1940,6 +2024,112 @@ void do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask)
 	__do_set_cpus_allowed(p, new_mask, 0);
 }
 
+/*
+ * This function is wildly self concurrent, consider at least 3 times.
+ */
+static int affine_move_task(struct rq *rq, struct task_struct *p, struct rq_flags *rf,
+			    int dest_cpu, unsigned int flags)
+{
+	struct set_affinity_pending my_pending = { }, *pending = NULL;
+	struct migration_arg arg = {
+		.task = p,
+		.dest_cpu = dest_cpu,
+	};
+	bool complete = false;
+
+	/* Can the task run on the task's current CPU? If so, we're done */
+	if (cpumask_test_cpu(task_cpu(p), &p->cpus_mask)) {
+		pending = p->migration_pending;
+		if (pending) {
+			refcount_inc(&pending->refs);
+			p->migration_pending = NULL;
+			complete = true;
+		}
+		task_rq_unlock(rq, p, rf);
+
+		if (complete)
+			goto do_complete;
+
+		return 0;
+	}
+
+	if (!(flags & SCA_MIGRATE_ENABLE)) {
+		/* serialized by p->pi_lock */
+		if (!p->migration_pending) {
+			refcount_set(&my_pending.refs, 1);
+			init_completion(&my_pending.done);
+			p->migration_pending = &my_pending;
+		} else {
+			pending = p->migration_pending;
+			refcount_inc(&pending->refs);
+		}
+	}
+	pending = p->migration_pending;
+	/*
+	 * - !MIGRATE_ENABLE:
+	 *   we'll have installed a pending if there wasn't one already.
+	 *
+	 * - MIGRATE_ENABLE:
+	 *   we're here because the current CPU isn't matching anymore,
+	 *   the only way that can happen is because of a concurrent
+	 *   set_cpus_allowed_ptr() call, which should then still be
+	 *   pending completion.
+	 *
+	 * Either way, we really should have a @pending here.
+	 */
+	if (WARN_ON_ONCE(!pending)) {
+		task_rq_unlock(rq, p, rf);
+		return -EINVAL;
+	}
+
+	if (flags & SCA_MIGRATE_ENABLE) {
+
+		refcount_inc(&pending->refs); /* pending->{arg,stop_work} */
+		task_rq_unlock(rq, p, rf);
+
+		pending->arg = (struct migration_arg) {
+			.task = p,
+			.dest_cpu = -1,
+			.pending = pending,
+		};
+
+		stop_one_cpu_nowait(cpu_of(rq), migration_cpu_stop,
+				    &pending->arg, &pending->stop_work);
+
+		return 0;
+	}
+
+	if (task_running(rq, p) || p->state == TASK_WAKING) {
+
+		task_rq_unlock(rq, p, rf);
+		stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg);
+
+	} else {
+
+		if (!is_migration_disabled(p)) {
+			if (task_on_rq_queued(p))
+				rq = move_queued_task(rq, rf, p, dest_cpu);
+
+			p->migration_pending = NULL;
+			complete = true;
+		}
+		task_rq_unlock(rq, p, rf);
+
+do_complete:
+		if (complete)
+			complete_all(&pending->done);
+	}
+
+	wait_for_completion(&pending->done);
+
+	if (refcount_dec_and_test(&pending->refs))
+		wake_up_var(&pending->refs);
+
+	wait_var_event(&my_pending.refs, !refcount_read(&my_pending.refs));
+
+	return 0;
+}
+
 /*
  * Change a given task's CPU affinity. Migrate the thread to a
  * proper CPU and schedule it away if the CPU it's executing on
@@ -2009,23 +2199,8 @@ static int __set_cpus_allowed_ptr(struct task_struct *p,
 			p->nr_cpus_allowed != 1);
 	}
 
-	/* Can the task run on the task's current CPU? If so, we're done */
-	if (cpumask_test_cpu(task_cpu(p), new_mask))
-		goto out;
+	return affine_move_task(rq, p, &rf, dest_cpu, flags);
 
-	if (task_running(rq, p) || p->state == TASK_WAKING) {
-		struct migration_arg arg = { p, dest_cpu };
-		/* Need help from migration thread: drop lock and wait. */
-		task_rq_unlock(rq, p, &rf);
-		stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg);
-		return 0;
-	} else if (task_on_rq_queued(p)) {
-		/*
-		 * OK, since we're going to drop the lock immediately
-		 * afterwards anyway.
-		 */
-		rq = move_queued_task(rq, &rf, p, dest_cpu);
-	}
 out:
 	task_rq_unlock(rq, p, &rf);
 
@@ -3205,6 +3380,7 @@ static void __sched_fork(unsigned long clone_flags, struct task_struct *p)
 	init_numa_balancing(clone_flags, p);
 #ifdef CONFIG_SMP
 	p->wake_entry.u_flags = CSD_TYPE_TTWU;
+	p->migration_pending = NULL;
 #endif
 }
 
-- 
cgit v1.2.3-71-gd317


From 14e292f8d45380c519a83d9b0f37089a17eedcdf Mon Sep 17 00:00:00 2001
From: Peter Zijlstra <peterz@infradead.org>
Date: Thu, 1 Oct 2020 15:54:14 +0200
Subject: sched,rt: Use cpumask_any*_distribute()

Replace a bunch of cpumask_any*() instances with
cpumask_any*_distribute(), by injecting this little bit of random in
cpu selection, we reduce the chance two competing balance operations
working off the same lowest_mask pick the same CPU.

Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Valentin Schneider <valentin.schneider@arm.com>
Reviewed-by: Daniel Bristot de Oliveira <bristot@redhat.com>
Link: https://lkml.kernel.org/r/20201023102347.190759694@infradead.org
---
 include/linux/cpumask.h |  6 ++++++
 kernel/sched/deadline.c |  6 +++---
 kernel/sched/rt.c       |  6 +++---
 lib/cpumask.c           | 18 ++++++++++++++++++
 4 files changed, 30 insertions(+), 6 deletions(-)

(limited to 'include/linux')

diff --git a/include/linux/cpumask.h b/include/linux/cpumask.h
index f0d895d6ac39..383684e30f12 100644
--- a/include/linux/cpumask.h
+++ b/include/linux/cpumask.h
@@ -199,6 +199,11 @@ static inline int cpumask_any_and_distribute(const struct cpumask *src1p,
 	return cpumask_next_and(-1, src1p, src2p);
 }
 
+static inline int cpumask_any_distribute(const struct cpumask *srcp)
+{
+	return cpumask_first(srcp);
+}
+
 #define for_each_cpu(cpu, mask)			\
 	for ((cpu) = 0; (cpu) < 1; (cpu)++, (void)mask)
 #define for_each_cpu_not(cpu, mask)		\
@@ -252,6 +257,7 @@ int cpumask_any_but(const struct cpumask *mask, unsigned int cpu);
 unsigned int cpumask_local_spread(unsigned int i, int node);
 int cpumask_any_and_distribute(const struct cpumask *src1p,
 			       const struct cpumask *src2p);
+int cpumask_any_distribute(const struct cpumask *srcp);
 
 /**
  * for_each_cpu - iterate over every cpu in a mask
diff --git a/kernel/sched/deadline.c b/kernel/sched/deadline.c
index e97c7c2708bc..206a0703fcbc 100644
--- a/kernel/sched/deadline.c
+++ b/kernel/sched/deadline.c
@@ -2002,8 +2002,8 @@ static int find_later_rq(struct task_struct *task)
 				return this_cpu;
 			}
 
-			best_cpu = cpumask_first_and(later_mask,
-							sched_domain_span(sd));
+			best_cpu = cpumask_any_and_distribute(later_mask,
+							      sched_domain_span(sd));
 			/*
 			 * Last chance: if a CPU being in both later_mask
 			 * and current sd span is valid, that becomes our
@@ -2025,7 +2025,7 @@ static int find_later_rq(struct task_struct *task)
 	if (this_cpu != -1)
 		return this_cpu;
 
-	cpu = cpumask_any(later_mask);
+	cpu = cpumask_any_distribute(later_mask);
 	if (cpu < nr_cpu_ids)
 		return cpu;
 
diff --git a/kernel/sched/rt.c b/kernel/sched/rt.c
index 40a46639f78a..2525a1beed26 100644
--- a/kernel/sched/rt.c
+++ b/kernel/sched/rt.c
@@ -1752,8 +1752,8 @@ static int find_lowest_rq(struct task_struct *task)
 				return this_cpu;
 			}
 
-			best_cpu = cpumask_first_and(lowest_mask,
-						     sched_domain_span(sd));
+			best_cpu = cpumask_any_and_distribute(lowest_mask,
+							      sched_domain_span(sd));
 			if (best_cpu < nr_cpu_ids) {
 				rcu_read_unlock();
 				return best_cpu;
@@ -1770,7 +1770,7 @@ static int find_lowest_rq(struct task_struct *task)
 	if (this_cpu != -1)
 		return this_cpu;
 
-	cpu = cpumask_any(lowest_mask);
+	cpu = cpumask_any_distribute(lowest_mask);
 	if (cpu < nr_cpu_ids)
 		return cpu;
 
diff --git a/lib/cpumask.c b/lib/cpumask.c
index 85da6ab4fbb5..35924025097b 100644
--- a/lib/cpumask.c
+++ b/lib/cpumask.c
@@ -267,3 +267,21 @@ int cpumask_any_and_distribute(const struct cpumask *src1p,
 	return next;
 }
 EXPORT_SYMBOL(cpumask_any_and_distribute);
+
+int cpumask_any_distribute(const struct cpumask *srcp)
+{
+	int next, prev;
+
+	/* NOTE: our first selection will skip 0. */
+	prev = __this_cpu_read(distribute_cpu_mask_prev);
+
+	next = cpumask_next(prev, srcp);
+	if (next >= nr_cpu_ids)
+		next = cpumask_first(srcp);
+
+	if (next < nr_cpu_ids)
+		__this_cpu_write(distribute_cpu_mask_prev, next);
+
+	return next;
+}
+EXPORT_SYMBOL(cpumask_any_distribute);
-- 
cgit v1.2.3-71-gd317


From a7c81556ec4d341dfdbf2cc478ead89d73e474a7 Mon Sep 17 00:00:00 2001
From: Peter Zijlstra <peterz@infradead.org>
Date: Mon, 28 Sep 2020 17:06:07 +0200
Subject: sched: Fix migrate_disable() vs rt/dl balancing

In order to minimize the interference of migrate_disable() on lower
priority tasks, which can be deprived of runtime due to being stuck
below a higher priority task. Teach the RT/DL balancers to push away
these higher priority tasks when a lower priority task gets selected
to run on a freshly demoted CPU (pull).

This adds migration interference to the higher priority task, but
restores bandwidth to system that would otherwise be irrevocably lost.
Without this it would be possible to have all tasks on the system
stuck on a single CPU, each task preempted in a migrate_disable()
section with a single high priority task running.

This way we can still approximate running the M highest priority tasks
on the system.

Migrating the top task away is (ofcourse) still subject to
migrate_disable() too, which means the lower task is subject to an
interference equivalent to the worst case migrate_disable() section.

Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Daniel Bristot de Oliveira <bristot@redhat.com>
Link: https://lkml.kernel.org/r/20201023102347.499155098@infradead.org
---
 include/linux/preempt.h | 40 ++++++++++++++++-------------
 include/linux/sched.h   |  3 ++-
 kernel/sched/core.c     | 67 +++++++++++++++++++++++++++++++++++++++++--------
 kernel/sched/deadline.c | 29 +++++++++++++++------
 kernel/sched/rt.c       | 63 +++++++++++++++++++++++++++++++++++++---------
 kernel/sched/sched.h    | 32 +++++++++++++++++++++++
 6 files changed, 186 insertions(+), 48 deletions(-)

(limited to 'include/linux')

diff --git a/include/linux/preempt.h b/include/linux/preempt.h
index 97ba7c920653..8b43922e65df 100644
--- a/include/linux/preempt.h
+++ b/include/linux/preempt.h
@@ -325,24 +325,28 @@ static inline void preempt_notifier_init(struct preempt_notifier *notifier,
 #if defined(CONFIG_SMP) && defined(CONFIG_PREEMPT_RT)
 
 /*
- * Migrate-Disable and why it is (strongly) undesired.
- *
- * The premise of the Real-Time schedulers we have on Linux
- * (SCHED_FIFO/SCHED_DEADLINE) is that M CPUs can/will run M tasks
- * concurrently, provided there are sufficient runnable tasks, also known as
- * work-conserving. For instance SCHED_DEADLINE tries to schedule the M
- * earliest deadline threads, and SCHED_FIFO the M highest priority threads.
- *
- * The correctness of various scheduling models depends on this, but is it
- * broken by migrate_disable() that doesn't imply preempt_disable(). Where
- * preempt_disable() implies an immediate priority ceiling, preemptible
- * migrate_disable() allows nesting.
- *
- * The worst case is that all tasks preempt one another in a migrate_disable()
- * region and stack on a single CPU. This then reduces the available bandwidth
- * to a single CPU. And since Real-Time schedulability theory considers the
- * Worst-Case only, all Real-Time analysis shall revert to single-CPU
- * (instantly solving the SMP analysis problem).
+ * Migrate-Disable and why it is undesired.
+ *
+ * When a preempted task becomes elegible to run under the ideal model (IOW it
+ * becomes one of the M highest priority tasks), it might still have to wait
+ * for the preemptee's migrate_disable() section to complete. Thereby suffering
+ * a reduction in bandwidth in the exact duration of the migrate_disable()
+ * section.
+ *
+ * Per this argument, the change from preempt_disable() to migrate_disable()
+ * gets us:
+ *
+ * - a higher priority tasks gains reduced wake-up latency; with preempt_disable()
+ *   it would have had to wait for the lower priority task.
+ *
+ * - a lower priority tasks; which under preempt_disable() could've instantly
+ *   migrated away when another CPU becomes available, is now constrained
+ *   by the ability to push the higher priority task away, which might itself be
+ *   in a migrate_disable() section, reducing it's available bandwidth.
+ *
+ * IOW it trades latency / moves the interference term, but it stays in the
+ * system, and as long as it remains unbounded, the system is not fully
+ * deterministic.
  *
  *
  * The reason we have it anyway.
diff --git a/include/linux/sched.h b/include/linux/sched.h
index 90a0c92741d7..3af9d52fe093 100644
--- a/include/linux/sched.h
+++ b/include/linux/sched.h
@@ -716,8 +716,9 @@ struct task_struct {
 	cpumask_t			cpus_mask;
 	void				*migration_pending;
 #if defined(CONFIG_SMP) && defined(CONFIG_PREEMPT_RT)
-	int				migration_disabled;
+	unsigned short			migration_disabled;
 #endif
+	unsigned short			migration_flags;
 
 #ifdef CONFIG_PREEMPT_RCU
 	int				rcu_read_lock_nesting;
diff --git a/kernel/sched/core.c b/kernel/sched/core.c
index 9ce2fc7d3d51..e92d7853057c 100644
--- a/kernel/sched/core.c
+++ b/kernel/sched/core.c
@@ -1763,11 +1763,6 @@ void migrate_enable(void)
 }
 EXPORT_SYMBOL_GPL(migrate_enable);
 
-static inline bool is_migration_disabled(struct task_struct *p)
-{
-	return p->migration_disabled;
-}
-
 static inline bool rq_has_pinned_tasks(struct rq *rq)
 {
 	return rq->nr_pinned;
@@ -1972,6 +1967,49 @@ out:
 	return 0;
 }
 
+int push_cpu_stop(void *arg)
+{
+	struct rq *lowest_rq = NULL, *rq = this_rq();
+	struct task_struct *p = arg;
+
+	raw_spin_lock_irq(&p->pi_lock);
+	raw_spin_lock(&rq->lock);
+
+	if (task_rq(p) != rq)
+		goto out_unlock;
+
+	if (is_migration_disabled(p)) {
+		p->migration_flags |= MDF_PUSH;
+		goto out_unlock;
+	}
+
+	p->migration_flags &= ~MDF_PUSH;
+
+	if (p->sched_class->find_lock_rq)
+		lowest_rq = p->sched_class->find_lock_rq(p, rq);
+
+	if (!lowest_rq)
+		goto out_unlock;
+
+	// XXX validate p is still the highest prio task
+	if (task_rq(p) == rq) {
+		deactivate_task(rq, p, 0);
+		set_task_cpu(p, lowest_rq->cpu);
+		activate_task(lowest_rq, p, 0);
+		resched_curr(lowest_rq);
+	}
+
+	double_unlock_balance(rq, lowest_rq);
+
+out_unlock:
+	rq->push_busy = false;
+	raw_spin_unlock(&rq->lock);
+	raw_spin_unlock_irq(&p->pi_lock);
+
+	put_task_struct(p);
+	return 0;
+}
+
 /*
  * sched_class::set_cpus_allowed must do the below, but is not required to
  * actually call this function.
@@ -2052,6 +2090,14 @@ static int affine_move_task(struct rq *rq, struct task_struct *p, struct rq_flag
 
 	/* Can the task run on the task's current CPU? If so, we're done */
 	if (cpumask_test_cpu(task_cpu(p), &p->cpus_mask)) {
+		struct task_struct *push_task = NULL;
+
+		if ((flags & SCA_MIGRATE_ENABLE) &&
+		    (p->migration_flags & MDF_PUSH) && !rq->push_busy) {
+			rq->push_busy = true;
+			push_task = get_task_struct(p);
+		}
+
 		pending = p->migration_pending;
 		if (pending) {
 			refcount_inc(&pending->refs);
@@ -2060,6 +2106,11 @@ static int affine_move_task(struct rq *rq, struct task_struct *p, struct rq_flag
 		}
 		task_rq_unlock(rq, p, rf);
 
+		if (push_task) {
+			stop_one_cpu_nowait(rq->cpu, push_cpu_stop,
+					    p, &rq->push_work);
+		}
+
 		if (complete)
 			goto do_complete;
 
@@ -2098,6 +2149,7 @@ static int affine_move_task(struct rq *rq, struct task_struct *p, struct rq_flag
 	if (flags & SCA_MIGRATE_ENABLE) {
 
 		refcount_inc(&pending->refs); /* pending->{arg,stop_work} */
+		p->migration_flags &= ~MDF_PUSH;
 		task_rq_unlock(rq, p, rf);
 
 		pending->arg = (struct migration_arg) {
@@ -2716,11 +2768,6 @@ static inline int __set_cpus_allowed_ptr(struct task_struct *p,
 
 static inline void migrate_disable_switch(struct rq *rq, struct task_struct *p) { }
 
-static inline bool is_migration_disabled(struct task_struct *p)
-{
-	return false;
-}
-
 static inline bool rq_has_pinned_tasks(struct rq *rq)
 {
 	return false;
diff --git a/kernel/sched/deadline.c b/kernel/sched/deadline.c
index 3d3fd8370447..eed2e449b313 100644
--- a/kernel/sched/deadline.c
+++ b/kernel/sched/deadline.c
@@ -2129,6 +2129,9 @@ static int push_dl_task(struct rq *rq)
 		return 0;
 
 retry:
+	if (is_migration_disabled(next_task))
+		return 0;
+
 	if (WARN_ON(next_task == rq->curr))
 		return 0;
 
@@ -2206,7 +2209,7 @@ static void push_dl_tasks(struct rq *rq)
 static void pull_dl_task(struct rq *this_rq)
 {
 	int this_cpu = this_rq->cpu, cpu;
-	struct task_struct *p;
+	struct task_struct *p, *push_task;
 	bool resched = false;
 	struct rq *src_rq;
 	u64 dmin = LONG_MAX;
@@ -2236,6 +2239,7 @@ static void pull_dl_task(struct rq *this_rq)
 			continue;
 
 		/* Might drop this_rq->lock */
+		push_task = NULL;
 		double_lock_balance(this_rq, src_rq);
 
 		/*
@@ -2267,17 +2271,27 @@ static void pull_dl_task(struct rq *this_rq)
 					   src_rq->curr->dl.deadline))
 				goto skip;
 
-			resched = true;
-
-			deactivate_task(src_rq, p, 0);
-			set_task_cpu(p, this_cpu);
-			activate_task(this_rq, p, 0);
-			dmin = p->dl.deadline;
+			if (is_migration_disabled(p)) {
+				push_task = get_push_task(src_rq);
+			} else {
+				deactivate_task(src_rq, p, 0);
+				set_task_cpu(p, this_cpu);
+				activate_task(this_rq, p, 0);
+				dmin = p->dl.deadline;
+				resched = true;
+			}
 
 			/* Is there any other task even earlier? */
 		}
 skip:
 		double_unlock_balance(this_rq, src_rq);
+
+		if (push_task) {
+			raw_spin_unlock(&this_rq->lock);
+			stop_one_cpu_nowait(src_rq->cpu, push_cpu_stop,
+					    push_task, &src_rq->push_work);
+			raw_spin_lock(&this_rq->lock);
+		}
 	}
 
 	if (resched)
@@ -2524,6 +2538,7 @@ const struct sched_class dl_sched_class
 	.rq_online              = rq_online_dl,
 	.rq_offline             = rq_offline_dl,
 	.task_woken		= task_woken_dl,
+	.find_lock_rq		= find_lock_later_rq,
 #endif
 
 	.task_tick		= task_tick_dl,
diff --git a/kernel/sched/rt.c b/kernel/sched/rt.c
index cf63346a07e4..c592e47cafed 100644
--- a/kernel/sched/rt.c
+++ b/kernel/sched/rt.c
@@ -1859,7 +1859,7 @@ static struct task_struct *pick_next_pushable_task(struct rq *rq)
  * running task can migrate over to a CPU that is running a task
  * of lesser priority.
  */
-static int push_rt_task(struct rq *rq)
+static int push_rt_task(struct rq *rq, bool pull)
 {
 	struct task_struct *next_task;
 	struct rq *lowest_rq;
@@ -1873,6 +1873,34 @@ static int push_rt_task(struct rq *rq)
 		return 0;
 
 retry:
+	if (is_migration_disabled(next_task)) {
+		struct task_struct *push_task = NULL;
+		int cpu;
+
+		if (!pull || rq->push_busy)
+			return 0;
+
+		cpu = find_lowest_rq(rq->curr);
+		if (cpu == -1 || cpu == rq->cpu)
+			return 0;
+
+		/*
+		 * Given we found a CPU with lower priority than @next_task,
+		 * therefore it should be running. However we cannot migrate it
+		 * to this other CPU, instead attempt to push the current
+		 * running task on this CPU away.
+		 */
+		push_task = get_push_task(rq);
+		if (push_task) {
+			raw_spin_unlock(&rq->lock);
+			stop_one_cpu_nowait(rq->cpu, push_cpu_stop,
+					    push_task, &rq->push_work);
+			raw_spin_lock(&rq->lock);
+		}
+
+		return 0;
+	}
+
 	if (WARN_ON(next_task == rq->curr))
 		return 0;
 
@@ -1927,12 +1955,10 @@ retry:
 	deactivate_task(rq, next_task, 0);
 	set_task_cpu(next_task, lowest_rq->cpu);
 	activate_task(lowest_rq, next_task, 0);
-	ret = 1;
-
 	resched_curr(lowest_rq);
+	ret = 1;
 
 	double_unlock_balance(rq, lowest_rq);
-
 out:
 	put_task_struct(next_task);
 
@@ -1942,7 +1968,7 @@ out:
 static void push_rt_tasks(struct rq *rq)
 {
 	/* push_rt_task will return true if it moved an RT */
-	while (push_rt_task(rq))
+	while (push_rt_task(rq, false))
 		;
 }
 
@@ -2095,7 +2121,8 @@ void rto_push_irq_work_func(struct irq_work *work)
 	 */
 	if (has_pushable_tasks(rq)) {
 		raw_spin_lock(&rq->lock);
-		push_rt_tasks(rq);
+		while (push_rt_task(rq, true))
+			;
 		raw_spin_unlock(&rq->lock);
 	}
 
@@ -2120,7 +2147,7 @@ static void pull_rt_task(struct rq *this_rq)
 {
 	int this_cpu = this_rq->cpu, cpu;
 	bool resched = false;
-	struct task_struct *p;
+	struct task_struct *p, *push_task;
 	struct rq *src_rq;
 	int rt_overload_count = rt_overloaded(this_rq);
 
@@ -2167,6 +2194,7 @@ static void pull_rt_task(struct rq *this_rq)
 		 * double_lock_balance, and another CPU could
 		 * alter this_rq
 		 */
+		push_task = NULL;
 		double_lock_balance(this_rq, src_rq);
 
 		/*
@@ -2194,11 +2222,14 @@ static void pull_rt_task(struct rq *this_rq)
 			if (p->prio < src_rq->curr->prio)
 				goto skip;
 
-			resched = true;
-
-			deactivate_task(src_rq, p, 0);
-			set_task_cpu(p, this_cpu);
-			activate_task(this_rq, p, 0);
+			if (is_migration_disabled(p)) {
+				push_task = get_push_task(src_rq);
+			} else {
+				deactivate_task(src_rq, p, 0);
+				set_task_cpu(p, this_cpu);
+				activate_task(this_rq, p, 0);
+				resched = true;
+			}
 			/*
 			 * We continue with the search, just in
 			 * case there's an even higher prio task
@@ -2208,6 +2239,13 @@ static void pull_rt_task(struct rq *this_rq)
 		}
 skip:
 		double_unlock_balance(this_rq, src_rq);
+
+		if (push_task) {
+			raw_spin_unlock(&this_rq->lock);
+			stop_one_cpu_nowait(src_rq->cpu, push_cpu_stop,
+					    push_task, &src_rq->push_work);
+			raw_spin_lock(&this_rq->lock);
+		}
 	}
 
 	if (resched)
@@ -2449,6 +2487,7 @@ const struct sched_class rt_sched_class
 	.rq_offline             = rq_offline_rt,
 	.task_woken		= task_woken_rt,
 	.switched_from		= switched_from_rt,
+	.find_lock_rq		= find_lock_lowest_rq,
 #endif
 
 	.task_tick		= task_tick_rt,
diff --git a/kernel/sched/sched.h b/kernel/sched/sched.h
index 42de1406c0dc..56992aaca48e 100644
--- a/kernel/sched/sched.h
+++ b/kernel/sched/sched.h
@@ -1057,6 +1057,8 @@ struct rq {
 #if defined(CONFIG_PREEMPT_RT) && defined(CONFIG_SMP)
 	unsigned int		nr_pinned;
 #endif
+	unsigned int		push_busy;
+	struct cpu_stop_work	push_work;
 };
 
 #ifdef CONFIG_FAIR_GROUP_SCHED
@@ -1084,6 +1086,16 @@ static inline int cpu_of(struct rq *rq)
 #endif
 }
 
+#define MDF_PUSH	0x01
+
+static inline bool is_migration_disabled(struct task_struct *p)
+{
+#if defined(CONFIG_SMP) && defined(CONFIG_PREEMPT_RT)
+	return p->migration_disabled;
+#else
+	return false;
+#endif
+}
 
 #ifdef CONFIG_SCHED_SMT
 extern void __update_idle_core(struct rq *rq);
@@ -1823,6 +1835,8 @@ struct sched_class {
 
 	void (*rq_online)(struct rq *rq);
 	void (*rq_offline)(struct rq *rq);
+
+	struct rq *(*find_lock_rq)(struct task_struct *p, struct rq *rq);
 #endif
 
 	void (*task_tick)(struct rq *rq, struct task_struct *p, int queued);
@@ -1918,6 +1932,24 @@ extern void trigger_load_balance(struct rq *rq);
 
 extern void set_cpus_allowed_common(struct task_struct *p, const struct cpumask *new_mask, u32 flags);
 
+static inline struct task_struct *get_push_task(struct rq *rq)
+{
+	struct task_struct *p = rq->curr;
+
+	lockdep_assert_held(&rq->lock);
+
+	if (rq->push_busy)
+		return NULL;
+
+	if (p->nr_cpus_allowed == 1)
+		return NULL;
+
+	rq->push_busy = true;
+	return get_task_struct(p);
+}
+
+extern int push_cpu_stop(void *arg);
+
 #endif
 
 #ifdef CONFIG_CPU_IDLE
-- 
cgit v1.2.3-71-gd317