cgroup: make per-cgroup pressure stall tracking configurable

PSI accounts stalls for each cgroup separately and aggregates it at each
level of the hierarchy. This causes additional overhead with psi_avgs_work
being called for each cgroup in the hierarchy. psi_avgs_work has been
highly optimized, however on systems with large number of cgroups the
overhead becomes noticeable.
Systems which use PSI only at the system level could avoid this overhead
if PSI can be configured to skip per-cgroup stall accounting.
Add "cgroup_disable=pressure" kernel command-line option to allow
requesting system-wide only pressure stall accounting. When set, it
keeps system-wide accounting under /proc/pressure/ but skips accounting
for individual cgroups and does not expose PSI nodes in cgroup hierarchy.

Signed-off-by: Suren Baghdasaryan <[email protected]>
Acked-by: Peter Zijlstra (Intel) <[email protected]>
Acked-by: Johannes Weiner <[email protected]>
Signed-off-by: Tejun Heo <[email protected]>

Author: surenbaghdasaryan

Documentation/admin-guide/kernel-parameters.txt

@@ -497,16 +497,21 @@
 	ccw_timeout_log	[S390]
 			See Documentation/s390/common_io.rst for details.
 
-	cgroup_disable=	[KNL] Disable a particular controller
-			Format: {name of the controller(s) to disable}
+	cgroup_disable=	[KNL] Disable a particular controller or optional feature
+			Format: {name of the controller(s) or feature(s) to disable}
 			The effects of cgroup_disable=foo are:
 			- foo isn't auto-mounted if you mount all cgroups in
 			  a single hierarchy
 			- foo isn't visible as an individually mountable
 			  subsystem
+			- if foo is an optional feature then the feature is
+			  disabled and corresponding cgroup files are not
+			  created
 			{Currently only "memory" controller deal with this and
 			cut the overhead, others just disable the usage. So
 			only cgroup_disable=memory is actually worthy}
+			Specifying "pressure" disables per-cgroup pressure
+			stall information accounting feature
 
 	cgroup_no_v1=	[KNL] Disable cgroup controllers and named hierarchies in v1
 			Format: { { controller | "all" | "named" }

include/linux/cgroup-defs.h

@@ -113,6 +113,7 @@ enum {
 	CFTYPE_NO_PREFIX	= (1 << 3),	/* (DON'T USE FOR NEW FILES) no subsys prefix */
 	CFTYPE_WORLD_WRITABLE	= (1 << 4),	/* (DON'T USE FOR NEW FILES) S_IWUGO */
 	CFTYPE_DEBUG		= (1 << 5),	/* create when cgroup_debug */
+	CFTYPE_PRESSURE		= (1 << 6),	/* only if pressure feature is enabled */
 
 	/* internal flags, do not use outside cgroup core proper */
 	__CFTYPE_ONLY_ON_DFL	= (1 << 16),	/* only on default hierarchy */

include/linux/cgroup.h

@@ -676,6 +676,8 @@ static inline struct psi_group *cgroup_psi(struct cgroup *cgrp)
 	return &cgrp->psi;
 }
 
+bool cgroup_psi_enabled(void);
+
 static inline void cgroup_init_kthreadd(void)
 {
 	/*
@@ -735,6 +737,11 @@ static inline struct psi_group *cgroup_psi(struct cgroup *cgrp)
 	return NULL;
 }
 
+static inline bool cgroup_psi_enabled(void)
+{
+	return false;
+}
+
 static inline bool task_under_cgroup_hierarchy(struct task_struct *task,
 					       struct cgroup *ancestor)
 {

kernel/cgroup/cgroup.c

@@ -209,6 +209,22 @@ struct cgroup_namespace init_cgroup_ns = {
 static struct file_system_type cgroup2_fs_type;
 static struct cftype cgroup_base_files[];
 
+/* cgroup optional features */
+enum cgroup_opt_features {
+#ifdef CONFIG_PSI
+	OPT_FEATURE_PRESSURE,
+#endif
+	OPT_FEATURE_COUNT
+};
+
+static const char *cgroup_opt_feature_names[OPT_FEATURE_COUNT] = {
+#ifdef CONFIG_PSI
+	"pressure",
+#endif
+};
+
+static u16 cgroup_feature_disable_mask __read_mostly;
+
 static int cgroup_apply_control(struct cgroup *cgrp);
 static void cgroup_finalize_control(struct cgroup *cgrp, int ret);
 static void css_task_iter_skip(struct css_task_iter *it,
@@ -3631,6 +3647,18 @@ static void cgroup_pressure_release(struct kernfs_open_file *of)
 {
 	psi_trigger_replace(&of->priv, NULL);
 }
+
+bool cgroup_psi_enabled(void)
+{
+	return (cgroup_feature_disable_mask & (1 << OPT_FEATURE_PRESSURE)) == 0;
+}
+
+#else /* CONFIG_PSI */
+bool cgroup_psi_enabled(void)
+{
+	return false;
+}
+
 #endif /* CONFIG_PSI */
 
 static int cgroup_freeze_show(struct seq_file *seq, void *v)
@@ -3955,6 +3983,8 @@ static int cgroup_addrm_files(struct cgroup_subsys_state *css,
 restart:
 	for (cft = cfts; cft != cft_end && cft->name[0] != '\0'; cft++) {
 		/* does cft->flags tell us to skip this file on @cgrp? */
+		if ((cft->flags & CFTYPE_PRESSURE) && !cgroup_psi_enabled())
+			continue;
 		if ((cft->flags & __CFTYPE_ONLY_ON_DFL) && !cgroup_on_dfl(cgrp))
 			continue;
 		if ((cft->flags & __CFTYPE_NOT_ON_DFL) && cgroup_on_dfl(cgrp))
@@ -4032,6 +4062,9 @@ static int cgroup_init_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
 
 		WARN_ON(cft->ss || cft->kf_ops);
 
+		if ((cft->flags & CFTYPE_PRESSURE) && !cgroup_psi_enabled())
+			continue;
+
 		if (cft->seq_start)
 			kf_ops = &cgroup_kf_ops;
 		else
@@ -4945,20 +4978,23 @@ static struct cftype cgroup_base_files[] = {
 #ifdef CONFIG_PSI
 	{
 		.name = "io.pressure",
+		.flags = CFTYPE_PRESSURE,
 		.seq_show = cgroup_io_pressure_show,
 		.write = cgroup_io_pressure_write,
 		.poll = cgroup_pressure_poll,
 		.release = cgroup_pressure_release,
 	},
 	{
 		.name = "memory.pressure",
+		.flags = CFTYPE_PRESSURE,
 		.seq_show = cgroup_memory_pressure_show,
 		.write = cgroup_memory_pressure_write,
 		.poll = cgroup_pressure_poll,
 		.release = cgroup_pressure_release,
 	},
 	{
 		.name = "cpu.pressure",
+		.flags = CFTYPE_PRESSURE,
 		.seq_show = cgroup_cpu_pressure_show,
 		.write = cgroup_cpu_pressure_write,
 		.poll = cgroup_pressure_poll,
@@ -6313,6 +6349,15 @@ static int __init cgroup_disable(char *str)
 			pr_info("Disabling %s control group subsystem\n",
 				ss->name);
 		}
+
+		for (i = 0; i < OPT_FEATURE_COUNT; i++) {
+			if (strcmp(token, cgroup_opt_feature_names[i]))
+				continue;
+			cgroup_feature_disable_mask |= 1 << i;
+			pr_info("Disabling %s control group feature\n",
+				cgroup_opt_feature_names[i]);
+			break;
+		}
 	}
 	return 1;
 }
@@ -6611,6 +6656,9 @@ static ssize_t show_delegatable_files(struct cftype *files, char *buf,
 		if (!(cft->flags & CFTYPE_NS_DELEGATABLE))
 			continue;
 
+		if ((cft->flags & CFTYPE_PRESSURE) && !cgroup_psi_enabled())
+			continue;
+
 		if (prefix)
 			ret += snprintf(buf + ret, size - ret, "%s.", prefix);
 

kernel/sched/psi.c

@@ -148,6 +148,7 @@
 static int psi_bug __read_mostly;
 
 DEFINE_STATIC_KEY_FALSE(psi_disabled);
+DEFINE_STATIC_KEY_TRUE(psi_cgroups_enabled);
 
 #ifdef CONFIG_PSI_DEFAULT_DISABLED
 static bool psi_enable;
@@ -211,6 +212,9 @@ void __init psi_init(void)
 		return;
 	}
 
+	if (!cgroup_psi_enabled())
+		static_branch_disable(&psi_cgroups_enabled);
+
 	psi_period = jiffies_to_nsecs(PSI_FREQ);
 	group_init(&psi_system);
 }
@@ -744,23 +748,23 @@ static void psi_group_change(struct psi_group *group, int cpu,
 
 static struct psi_group *iterate_groups(struct task_struct *task, void **iter)
 {
+	if (*iter == &psi_system)
+		return NULL;
+
 #ifdef CONFIG_CGROUPS
-	struct cgroup *cgroup = NULL;
+	if (static_branch_likely(&psi_cgroups_enabled)) {
+		struct cgroup *cgroup = NULL;
 
-	if (!*iter)
-		cgroup = task->cgroups->dfl_cgrp;
-	else if (*iter == &psi_system)
-		return NULL;
-	else
-		cgroup = cgroup_parent(*iter);
+		if (!*iter)
+			cgroup = task->cgroups->dfl_cgrp;
+		else
+			cgroup = cgroup_parent(*iter);
 
-	if (cgroup && cgroup_parent(cgroup)) {
-		*iter = cgroup;
-		return cgroup_psi(cgroup);
+		if (cgroup && cgroup_parent(cgroup)) {
+			*iter = cgroup;
+			return cgroup_psi(cgroup);
+		}
 	}
-#else
-	if (*iter)
-		return NULL;
 #endif
 	*iter = &psi_system;
 	return &psi_system;