[PATCH 2/2] sched/psi: iterate through cgroups directly

[Kairui Song <ryncsn-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org>]

From: Kairui Song <kasong-1Nz4purKYjRBDgjK7y7TUQ@public.gmane.org>

psi_group->parent has the same hierarchy as the cgroup it's in.
So just iterate through cgroup instead.

By adjusting the iteration logic, save some space in psi_group
struct, and the performance is actually better. I see a measurable
performance gain using mmtests/perfpipe:

(AVG of 100 test, ops/sec, the higher the better)
KVM guest on a i7-9700:
        psi=0         root cgroup   5 levels of cgroup
Before: 59221         55352         47821
After:  60100         56036         50884

KVM guest on a Ryzen 9 5900HX:
        psi=0         root cgroup   5 levels of cgroup
Before: 144566        138919        128888
After:  145812        139580        133514

Signed-off-by: Kairui Song <kasong-1Nz4purKYjRBDgjK7y7TUQ@public.gmane.org>
Signed-off-by: Kairui Song <ryncsn-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org>
---
 include/linux/psi_types.h |  1 -
 kernel/sched/psi.c        | 47 ++++++++++++++++++++++++++++-----------
 2 files changed, 34 insertions(+), 14 deletions(-)

diff --git a/include/linux/psi_types.h b/include/linux/psi_types.h
index 1e0a0d7ace3a..4066b846ce4a 100644
--- a/include/linux/psi_types.h
+++ b/include/linux/psi_types.h
@@ -154,7 +154,6 @@ struct psi_trigger {
 };
 
 struct psi_group {
-	struct psi_group *parent;
 	bool enabled;
 
 	/* Protects data used by the aggregator */
diff --git a/kernel/sched/psi.c b/kernel/sched/psi.c
index 8ac8b81bfee6..c74f8ce46f81 100644
--- a/kernel/sched/psi.c
+++ b/kernel/sched/psi.c
@@ -858,15 +858,34 @@ static void psi_group_change(struct psi_group *group, int cpu,
 		schedule_delayed_work(&group->avgs_work, PSI_FREQ);
 }
 
-static inline struct psi_group *task_psi_group(struct task_struct *task)
+static inline struct psi_group *psi_iter_first(struct task_struct *task, void **iter)
 {
 #ifdef CONFIG_CGROUPS
-	if (static_branch_likely(&psi_cgroups_enabled))
-		return cgroup_psi(task_dfl_cgroup(task));
+	if (static_branch_likely(&psi_cgroups_enabled)) {
+		struct cgroup *cgroup = task_dfl_cgroup(task);
+
+		*iter = cgroup_parent(cgroup);
+		return cgroup_psi(cgroup);
+	}
 #endif
 	return &psi_system;
 }
 
+static inline struct psi_group *psi_iter_next(void **iter)
+{
+#ifdef CONFIG_CGROUPS
+	if (static_branch_likely(&psi_cgroups_enabled)) {
+		struct cgroup *cgroup = *iter;
+
+		if (cgroup) {
+			*iter = cgroup_parent(cgroup);
+			return cgroup_psi(cgroup);
+		}
+	}
+#endif
+	return NULL;
+}
+
 static void psi_flags_change(struct task_struct *task, int clear, int set)
 {
 	if (((task->psi_flags & set) ||
@@ -886,6 +905,7 @@ void psi_task_change(struct task_struct *task, int clear, int set)
 {
 	int cpu = task_cpu(task);
 	struct psi_group *group;
+	void *iter;
 	u64 now;
 
 	if (!task->pid)
@@ -895,16 +915,17 @@ void psi_task_change(struct task_struct *task, int clear, int set)
 
 	now = cpu_clock(cpu);
 
-	group = task_psi_group(task);
+	group = psi_iter_first(task, &iter);
 	do {
 		psi_group_change(group, cpu, clear, set, now, true);
-	} while ((group = group->parent));
+	} while ((group = psi_iter_next(&iter)));
 }
 
 void psi_task_switch(struct task_struct *prev, struct task_struct *next,
 		     bool sleep)
 {
 	struct psi_group *group, *common = NULL;
+	void *iter;
 	int cpu = task_cpu(prev);
 	u64 now = cpu_clock(cpu);
 
@@ -915,7 +936,7 @@ void psi_task_switch(struct task_struct *prev, struct task_struct *next,
 		 * ancestors with @prev, those will already have @prev's
 		 * TSK_ONCPU bit set, and we can stop the iteration there.
 		 */
-		group = task_psi_group(next);
+		group = psi_iter_first(prev, &iter);
 		do {
 			if (per_cpu_ptr(group->pcpu, cpu)->state_mask &
 			    PSI_ONCPU) {
@@ -924,7 +945,7 @@ void psi_task_switch(struct task_struct *prev, struct task_struct *next,
 			}
 
 			psi_group_change(group, cpu, 0, TSK_ONCPU, now, true);
-		} while ((group = group->parent));
+		} while ((group = psi_iter_next(&iter)));
 	}
 
 	if (prev->pid) {
@@ -957,12 +978,12 @@ void psi_task_switch(struct task_struct *prev, struct task_struct *next,
 
 		psi_flags_change(prev, clear, set);
 
-		group = task_psi_group(prev);
+		group = psi_iter_first(prev, &iter);
 		do {
 			if (group == common)
 				break;
 			psi_group_change(group, cpu, clear, set, now, wake_clock);
-		} while ((group = group->parent));
+		} while ((group = psi_iter_next(&iter)));
 
 		/*
 		 * TSK_ONCPU is handled up to the common ancestor. If there are
@@ -972,7 +993,7 @@ void psi_task_switch(struct task_struct *prev, struct task_struct *next,
 		 */
 		if ((prev->psi_flags ^ next->psi_flags) & ~TSK_ONCPU) {
 			clear &= ~TSK_ONCPU;
-			for (; group; group = group->parent)
+			for (; group; group = psi_iter_next(&iter))
 				psi_group_change(group, cpu, clear, set, now, wake_clock);
 		}
 	}
@@ -983,6 +1004,7 @@ void psi_account_irqtime(struct task_struct *task, u32 delta)
 {
 	int cpu = task_cpu(task);
 	struct psi_group *group;
+	void *iter;
 	struct psi_group_cpu *groupc;
 	u64 now;
 
@@ -991,7 +1013,7 @@ void psi_account_irqtime(struct task_struct *task, u32 delta)
 
 	now = cpu_clock(cpu);
 
-	group = task_psi_group(task);
+	group = psi_iter_first(task, &iter);
 	do {
 		if (!group->enabled)
 			continue;
@@ -1007,7 +1029,7 @@ void psi_account_irqtime(struct task_struct *task, u32 delta)
 
 		if (group->poll_states & (1 << PSI_IRQ_FULL))
 			psi_schedule_poll_work(group, 1, false);
-	} while ((group = group->parent));
+	} while ((group = psi_iter_next(&iter)));
 }
 #endif
 
@@ -1089,7 +1111,6 @@ int psi_cgroup_alloc(struct cgroup *cgroup)
 		return -ENOMEM;
 	}
 	group_init(cgroup->psi);
-	cgroup->psi->parent = cgroup_psi(cgroup_parent(cgroup));
 	return 0;
 }
 
-- 
2.39.1