Merge tag 'sched-psi-2022-10-14' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip

Pull PSI updates from Ingo Molnar:

 - Various performance optimizations, resulting in a 4%-9% speedup in
   the mmtests/config-scheduler-perfpipe micro-benchmark.

 - New interface to turn PSI on/off on a per cgroup level.

* tag 'sched-psi-2022-10-14' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  sched/psi: Per-cgroup PSI accounting disable/re-enable interface
  sched/psi: Cache parent psi_group to speed up group iteration
  sched/psi: Consolidate cgroup_psi()
  sched/psi: Add PSI_IRQ to track IRQ/SOFTIRQ pressure
  sched/psi: Remove NR_ONCPU task accounting
  sched/psi: Optimize task switch inside shared cgroups again
  sched/psi: Move private helpers to sched/stats.h
  sched/psi: Save percpu memory when !psi_cgroups_enabled
  sched/psi: Don't create cgroup PSI files when psi_disabled
  sched/psi: Fix periodic aggregation shut off

3 files changed, 213 insertions, 74 deletions

diff --git a/kernel/sched/core.c b/kernel/sched/core.c
index e4ce124ec701..5800b0623ff3 100644
--- a/kernel/sched/core.c
+++ b/kernel/sched/core.c
@@ -701,6 +701,7 @@ static void update_rq_clock_task(struct rq *rq, s64 delta)
 
 	rq->prev_irq_time += irq_delta;
 	delta -= irq_delta;
+	psi_account_irqtime(rq->curr, irq_delta);
 #endif
 #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
 	if (static_key_false((&paravirt_steal_rq_enabled))) {
diff --git a/kernel/sched/psi.c b/kernel/sched/psi.c
index 7f6030091aee..ee2ecc081422 100644
--- a/kernel/sched/psi.c
+++ b/kernel/sched/psi.c
@@ -181,6 +181,7 @@ static void group_init(struct psi_group *group)
 {
 	int cpu;
 
+	group->enabled = true;
 	for_each_possible_cpu(cpu)
 		seqcount_init(&per_cpu_ptr(group->pcpu, cpu)->seq);
 	group->avg_last_update = sched_clock();
@@ -201,6 +202,7 @@ void __init psi_init(void)
 {
 	if (!psi_enable) {
 		static_branch_enable(&psi_disabled);
+		static_branch_disable(&psi_cgroups_enabled);
 		return;
 	}
 
@@ -211,7 +213,7 @@ void __init psi_init(void)
 	group_init(&psi_system);
 }
 
-static bool test_state(unsigned int *tasks, enum psi_states state)
+static bool test_state(unsigned int *tasks, enum psi_states state, bool oncpu)
 {
 	switch (state) {
 	case PSI_IO_SOME:
@@ -224,9 +226,9 @@ static bool test_state(unsigned int *tasks, enum psi_states state)
 		return unlikely(tasks[NR_MEMSTALL] &&
 			tasks[NR_RUNNING] == tasks[NR_MEMSTALL_RUNNING]);
 	case PSI_CPU_SOME:
-		return unlikely(tasks[NR_RUNNING] > tasks[NR_ONCPU]);
+		return unlikely(tasks[NR_RUNNING] > oncpu);
 	case PSI_CPU_FULL:
-		return unlikely(tasks[NR_RUNNING] && !tasks[NR_ONCPU]);
+		return unlikely(tasks[NR_RUNNING] && !oncpu);
 	case PSI_NONIDLE:
 		return tasks[NR_IOWAIT] || tasks[NR_MEMSTALL] ||
 			tasks[NR_RUNNING];
@@ -688,35 +690,53 @@ static void psi_group_change(struct psi_group *group, int cpu,
 			     bool wake_clock)
 {
 	struct psi_group_cpu *groupc;
-	u32 state_mask = 0;
 	unsigned int t, m;
 	enum psi_states s;
+	u32 state_mask;
 
 	groupc = per_cpu_ptr(group->pcpu, cpu);
 
 	/*
-	 * First we assess the aggregate resource states this CPU's
-	 * tasks have been in since the last change, and account any
-	 * SOME and FULL time these may have resulted in.
-	 *
-	 * Then we update the task counts according to the state
+	 * First we update the task counts according to the state
 	 * change requested through the @clear and @set bits.
+	 *
+	 * Then if the cgroup PSI stats accounting enabled, we
+	 * assess the aggregate resource states this CPU's tasks
+	 * have been in since the last change, and account any
+	 * SOME and FULL time these may have resulted in.
 	 */
 	write_seqcount_begin(&groupc->seq);
 
-	record_times(groupc, now);
+	/*
+	 * Start with TSK_ONCPU, which doesn't have a corresponding
+	 * task count - it's just a boolean flag directly encoded in
+	 * the state mask. Clear, set, or carry the current state if
+	 * no changes are requested.
+	 */
+	if (unlikely(clear & TSK_ONCPU)) {
+		state_mask = 0;
+		clear &= ~TSK_ONCPU;
+	} else if (unlikely(set & TSK_ONCPU)) {
+		state_mask = PSI_ONCPU;
+		set &= ~TSK_ONCPU;
+	} else {
+		state_mask = groupc->state_mask & PSI_ONCPU;
+	}
 
+	/*
+	 * The rest of the state mask is calculated based on the task
+	 * counts. Update those first, then construct the mask.
+	 */
 	for (t = 0, m = clear; m; m &= ~(1 << t), t++) {
 		if (!(m & (1 << t)))
 			continue;
 		if (groupc->tasks[t]) {
 			groupc->tasks[t]--;
 		} else if (!psi_bug) {
-			printk_deferred(KERN_ERR "psi: task underflow! cpu=%d t=%d tasks=[%u %u %u %u %u] clear=%x set=%x\n",
+			printk_deferred(KERN_ERR "psi: task underflow! cpu=%d t=%d tasks=[%u %u %u %u] clear=%x set=%x\n",
 					cpu, t, groupc->tasks[0],
 					groupc->tasks[1], groupc->tasks[2],
-					groupc->tasks[3], groupc->tasks[4],
-					clear, set);
+					groupc->tasks[3], clear, set);
 			psi_bug = 1;
 		}
 	}
@@ -725,9 +745,25 @@ static void psi_group_change(struct psi_group *group, int cpu,
 		if (set & (1 << t))
 			groupc->tasks[t]++;
 
-	/* Calculate state mask representing active states */
+	if (!group->enabled) {
+		/*
+		 * On the first group change after disabling PSI, conclude
+		 * the current state and flush its time. This is unlikely
+		 * to matter to the user, but aggregation (get_recent_times)
+		 * may have already incorporated the live state into times_prev;
+		 * avoid a delta sample underflow when PSI is later re-enabled.
+		 */
+		if (unlikely(groupc->state_mask & (1 << PSI_NONIDLE)))
+			record_times(groupc, now);
+
+		groupc->state_mask = state_mask;
+
+		write_seqcount_end(&groupc->seq);
+		return;
+	}
+
 	for (s = 0; s < NR_PSI_STATES; s++) {
-		if (test_state(groupc->tasks, s))
+		if (test_state(groupc->tasks, s, state_mask & PSI_ONCPU))
 			state_mask |= (1 << s);
 	}
 
@@ -739,9 +775,11 @@ static void psi_group_change(struct psi_group *group, int cpu,
 	 * task in a cgroup is in_memstall, the corresponding groupc
 	 * on that cpu is in PSI_MEM_FULL state.
 	 */
-	if (unlikely(groupc->tasks[NR_ONCPU] && cpu_curr(cpu)->in_memstall))
+	if (unlikely((state_mask & PSI_ONCPU) && cpu_curr(cpu)->in_memstall))
 		state_mask |= (1 << PSI_MEM_FULL);
 
+	record_times(groupc, now);
+
 	groupc->state_mask = state_mask;
 
 	write_seqcount_end(&groupc->seq);
@@ -753,27 +791,12 @@ static void psi_group_change(struct psi_group *group, int cpu,
 		schedule_delayed_work(&group->avgs_work, PSI_FREQ);
 }
 
-static struct psi_group *iterate_groups(struct task_struct *task, void **iter)
+static inline struct psi_group *task_psi_group(struct task_struct *task)
 {
-	if (*iter == &psi_system)
-		return NULL;
-
 #ifdef CONFIG_CGROUPS
-	if (static_branch_likely(&psi_cgroups_enabled)) {
-		struct cgroup *cgroup = NULL;
-
-		if (!*iter)
-			cgroup = task->cgroups->dfl_cgrp;
-		else
-			cgroup = cgroup_parent(*iter);
-
-		if (cgroup && cgroup_parent(cgroup)) {
-			*iter = cgroup;
-			return cgroup_psi(cgroup);
-		}
-	}
+	if (static_branch_likely(&psi_cgroups_enabled))
+		return cgroup_psi(task_dfl_cgroup(task));
 #endif
-	*iter = &psi_system;
 	return &psi_system;
 }
 
@@ -796,8 +819,6 @@ void psi_task_change(struct task_struct *task, int clear, int set)
 {
 	int cpu = task_cpu(task);
 	struct psi_group *group;
-	bool wake_clock = true;
-	void *iter = NULL;
 	u64 now;
 
 	if (!task->pid)
@@ -806,19 +827,11 @@ void psi_task_change(struct task_struct *task, int clear, int set)
 	psi_flags_change(task, clear, set);
 
 	now = cpu_clock(cpu);
-	/*
-	 * Periodic aggregation shuts off if there is a period of no
-	 * task changes, so we wake it back up if necessary. However,
-	 * don't do this if the task change is the aggregation worker
-	 * itself going to sleep, or we'll ping-pong forever.
-	 */
-	if (unlikely((clear & TSK_RUNNING) &&
-		     (task->flags & PF_WQ_WORKER) &&
-		     wq_worker_last_func(task) == psi_avgs_work))
-		wake_clock = false;
 
-	while ((group = iterate_groups(task, &iter)))
-		psi_group_change(group, cpu, clear, set, now, wake_clock);
+	group = task_psi_group(task);
+	do {
+		psi_group_change(group, cpu, clear, set, now, true);
+	} while ((group = group->parent));
 }
 
 void psi_task_switch(struct task_struct *prev, struct task_struct *next,
@@ -826,34 +839,30 @@ void psi_task_switch(struct task_struct *prev, struct task_struct *next,
 {
 	struct psi_group *group, *common = NULL;
 	int cpu = task_cpu(prev);
-	void *iter;
 	u64 now = cpu_clock(cpu);
 
 	if (next->pid) {
-		bool identical_state;
-
 		psi_flags_change(next, 0, TSK_ONCPU);
 		/*
-		 * When switching between tasks that have an identical
-		 * runtime state, the cgroup that contains both tasks
-		 * we reach the first common ancestor. Iterate @next's
-		 * ancestors only until we encounter @prev's ONCPU.
+		 * Set TSK_ONCPU on @next's cgroups. If @next shares any
+		 * ancestors with @prev, those will already have @prev's
+		 * TSK_ONCPU bit set, and we can stop the iteration there.
 		 */
-		identical_state = prev->psi_flags == next->psi_flags;
-		iter = NULL;
-		while ((group = iterate_groups(next, &iter))) {
-			if (identical_state &&
-			    per_cpu_ptr(group->pcpu, cpu)->tasks[NR_ONCPU]) {
+		group = task_psi_group(next);
+		do {
+			if (per_cpu_ptr(group->pcpu, cpu)->state_mask &
+			    PSI_ONCPU) {
 				common = group;
 				break;
 			}
 
 			psi_group_change(group, cpu, 0, TSK_ONCPU, now, true);
-		}
+		} while ((group = group->parent));
 	}
 
 	if (prev->pid) {
 		int clear = TSK_ONCPU, set = 0;
+		bool wake_clock = true;
 
 		/*
 		 * When we're going to sleep, psi_dequeue() lets us
@@ -867,26 +876,74 @@ void psi_task_switch(struct task_struct *prev, struct task_struct *next,
 				clear |= TSK_MEMSTALL_RUNNING;
 			if (prev->in_iowait)
 				set |= TSK_IOWAIT;
+
+			/*
+			 * Periodic aggregation shuts off if there is a period of no
+			 * task changes, so we wake it back up if necessary. However,
+			 * don't do this if the task change is the aggregation worker
+			 * itself going to sleep, or we'll ping-pong forever.
+			 */
+			if (unlikely((prev->flags & PF_WQ_WORKER) &&
+				     wq_worker_last_func(prev) == psi_avgs_work))
+				wake_clock = false;
 		}
 
 		psi_flags_change(prev, clear, set);
 
-		iter = NULL;
-		while ((group = iterate_groups(prev, &iter)) && group != common)
-			psi_group_change(group, cpu, clear, set, now, true);
+		group = task_psi_group(prev);
+		do {
+			if (group == common)
+				break;
+			psi_group_change(group, cpu, clear, set, now, wake_clock);
+		} while ((group = group->parent));
 
 		/*
-		 * TSK_ONCPU is handled up to the common ancestor. If we're tasked
-		 * with dequeuing too, finish that for the rest of the hierarchy.
+		 * TSK_ONCPU is handled up to the common ancestor. If there are
+		 * any other differences between the two tasks (e.g. prev goes
+		 * to sleep, or only one task is memstall), finish propagating
+		 * those differences all the way up to the root.
 		 */
-		if (sleep) {
+		if ((prev->psi_flags ^ next->psi_flags) & ~TSK_ONCPU) {
 			clear &= ~TSK_ONCPU;
-			for (; group; group = iterate_groups(prev, &iter))
-				psi_group_change(group, cpu, clear, set, now, true);
+			for (; group; group = group->parent)
+				psi_group_change(group, cpu, clear, set, now, wake_clock);
 		}
 	}
 }
 
+#ifdef CONFIG_IRQ_TIME_ACCOUNTING
+void psi_account_irqtime(struct task_struct *task, u32 delta)
+{
+	int cpu = task_cpu(task);
+	struct psi_group *group;
+	struct psi_group_cpu *groupc;
+	u64 now;
+
+	if (!task->pid)
+		return;
+
+	now = cpu_clock(cpu);
+
+	group = task_psi_group(task);
+	do {
+		if (!group->enabled)
+			continue;
+
+		groupc = per_cpu_ptr(group->pcpu, cpu);
+
+		write_seqcount_begin(&groupc->seq);
+
+		record_times(groupc, now);
+		groupc->times[PSI_IRQ_FULL] += delta;
+
+		write_seqcount_end(&groupc->seq);
+
+		if (group->poll_states & (1 << PSI_IRQ_FULL))
+			psi_schedule_poll_work(group, 1);
+	} while ((group = group->parent));
+}
+#endif
+
 /**
  * psi_memstall_enter - mark the beginning of a memory stall section
  * @flags: flags to handle nested sections
@@ -952,7 +1009,7 @@ EXPORT_SYMBOL_GPL(psi_memstall_leave);
 #ifdef CONFIG_CGROUPS
 int psi_cgroup_alloc(struct cgroup *cgroup)
 {
-	if (static_branch_likely(&psi_disabled))
+	if (!static_branch_likely(&psi_cgroups_enabled))
 		return 0;
 
 	cgroup->psi = kzalloc(sizeof(struct psi_group), GFP_KERNEL);
@@ -965,12 +1022,13 @@ int psi_cgroup_alloc(struct cgroup *cgroup)
 		return -ENOMEM;
 	}
 	group_init(cgroup->psi);
+	cgroup->psi->parent = cgroup_psi(cgroup_parent(cgroup));
 	return 0;
 }
 
 void psi_cgroup_free(struct cgroup *cgroup)
 {
-	if (static_branch_likely(&psi_disabled))
+	if (!static_branch_likely(&psi_cgroups_enabled))
 		return;
 
 	cancel_delayed_work_sync(&cgroup->psi->avgs_work);
@@ -998,7 +1056,7 @@ void cgroup_move_task(struct task_struct *task, struct css_set *to)
 	struct rq_flags rf;
 	struct rq *rq;
 
-	if (static_branch_likely(&psi_disabled)) {
+	if (!static_branch_likely(&psi_cgroups_enabled)) {
 		/*
 		 * Lame to do this here, but the scheduler cannot be locked
 		 * from the outside, so we move cgroups from inside sched/.
@@ -1046,10 +1104,45 @@ void cgroup_move_task(struct task_struct *task, struct css_set *to)
 
 	task_rq_unlock(rq, task, &rf);
 }
+
+void psi_cgroup_restart(struct psi_group *group)
+{
+	int cpu;
+
+	/*
+	 * After we disable psi_group->enabled, we don't actually
+	 * stop percpu tasks accounting in each psi_group_cpu,
+	 * instead only stop test_state() loop, record_times()
+	 * and averaging worker, see psi_group_change() for details.
+	 *
+	 * When disable cgroup PSI, this function has nothing to sync
+	 * since cgroup pressure files are hidden and percpu psi_group_cpu
+	 * would see !psi_group->enabled and only do task accounting.
+	 *
+	 * When re-enable cgroup PSI, this function use psi_group_change()
+	 * to get correct state mask from test_state() loop on tasks[],
+	 * and restart groupc->state_start from now, use .clear = .set = 0
+	 * here since no task status really changed.
+	 */
+	if (!group->enabled)
+		return;
+
+	for_each_possible_cpu(cpu) {
+		struct rq *rq = cpu_rq(cpu);
+		struct rq_flags rf;
+		u64 now;
+
+		rq_lock_irq(rq, &rf);
+		now = cpu_clock(cpu);
+		psi_group_change(group, cpu, 0, 0, now, true);
+		rq_unlock_irq(rq, &rf);
+	}
+}
 #endif /* CONFIG_CGROUPS */
 
 int psi_show(struct seq_file *m, struct psi_group *group, enum psi_res res)
 {
+	bool only_full = false;
 	int full;
 	u64 now;
 
@@ -1064,7 +1157,11 @@ int psi_show(struct seq_file *m, struct psi_group *group, enum psi_res res)
 		group->avg_next_update = update_averages(group, now);
 	mutex_unlock(&group->avgs_lock);
 
-	for (full = 0; full < 2; full++) {
+#ifdef CONFIG_IRQ_TIME_ACCOUNTING
+	only_full = res == PSI_IRQ;
+#endif
+
+	for (full = 0; full < 2 - only_full; full++) {
 		unsigned long avg[3] = { 0, };
 		u64 total = 0;
 		int w;
@@ -1078,7 +1175,7 @@ int psi_show(struct seq_file *m, struct psi_group *group, enum psi_res res)
 		}
 
 		seq_printf(m, "%s avg10=%lu.%02lu avg60=%lu.%02lu avg300=%lu.%02lu total=%llu\n",
-			   full ? "full" : "some",
+			   full || only_full ? "full" : "some",
 			   LOAD_INT(avg[0]), LOAD_FRAC(avg[0]),
 			   LOAD_INT(avg[1]), LOAD_FRAC(avg[1]),
 			   LOAD_INT(avg[2]), LOAD_FRAC(avg[2]),
@@ -1106,6 +1203,11 @@ struct psi_trigger *psi_trigger_create(struct psi_group *group,
 	else
 		return ERR_PTR(-EINVAL);
 
+#ifdef CONFIG_IRQ_TIME_ACCOUNTING
+	if (res == PSI_IRQ && --state != PSI_IRQ_FULL)
+		return ERR_PTR(-EINVAL);
+#endif
+
 	if (state >= PSI_NONIDLE)
 		return ERR_PTR(-EINVAL);
 
@@ -1390,6 +1492,33 @@ static const struct proc_ops psi_cpu_proc_ops = {
 	.proc_release	= psi_fop_release,
 };
 
+#ifdef CONFIG_IRQ_TIME_ACCOUNTING
+static int psi_irq_show(struct seq_file *m, void *v)
+{
+	return psi_show(m, &psi_system, PSI_IRQ);
+}
+
+static int psi_irq_open(struct inode *inode, struct file *file)
+{
+	return psi_open(file, psi_irq_show);
+}
+
+static ssize_t psi_irq_write(struct file *file, const char __user *user_buf,
+			     size_t nbytes, loff_t *ppos)
+{
+	return psi_write(file, user_buf, nbytes, PSI_IRQ);
+}
+
+static const struct proc_ops psi_irq_proc_ops = {
+	.proc_open	= psi_irq_open,
+	.proc_read	= seq_read,
+	.proc_lseek	= seq_lseek,
+	.proc_write	= psi_irq_write,
+	.proc_poll	= psi_fop_poll,
+	.proc_release	= psi_fop_release,
+};
+#endif
+
 static int __init psi_proc_init(void)
 {
 	if (psi_enable) {
@@ -1397,6 +1526,9 @@ static int __init psi_proc_init(void)
 		proc_create("pressure/io", 0666, NULL, &psi_io_proc_ops);
 		proc_create("pressure/memory", 0666, NULL, &psi_memory_proc_ops);
 		proc_create("pressure/cpu", 0666, NULL, &psi_cpu_proc_ops);
+#ifdef CONFIG_IRQ_TIME_ACCOUNTING
+		proc_create("pressure/irq", 0666, NULL, &psi_irq_proc_ops);
+#endif
 	}
 	return 0;
 }
diff --git a/kernel/sched/stats.h b/kernel/sched/stats.h
index baa839c1ba96..84a188913cc9 100644
--- a/kernel/sched/stats.h
+++ b/kernel/sched/stats.h
@@ -107,6 +107,11 @@ __schedstats_from_se(struct sched_entity *se)
 }
 
 #ifdef CONFIG_PSI
+void psi_task_change(struct task_struct *task, int clear, int set);
+void psi_task_switch(struct task_struct *prev, struct task_struct *next,
+		     bool sleep);
+void psi_account_irqtime(struct task_struct *task, u32 delta);
+
 /*
  * PSI tracks state that persists across sleeps, such as iowaits and
  * memory stalls. As a result, it has to distinguish between sleeps,
@@ -201,6 +206,7 @@ static inline void psi_ttwu_dequeue(struct task_struct *p) {}
 static inline void psi_sched_switch(struct task_struct *prev,
 				    struct task_struct *next,
 				    bool sleep) {}
+static inline void psi_account_irqtime(struct task_struct *task, u32 delta) {}
 #endif /* CONFIG_PSI */
 
 #ifdef CONFIG_SCHED_INFO