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.

Suren Baghdasaryan committed 5 years ago

Commit 3958e2d1 parent 2ca11b05 files · +80 −15Browse files

Unified · Split

5 files

Documentation/admin-guide/kernel-parameters.txt +7 −2
include/linux/cgroup-defs.h +1 −0
include/linux/cgroup.h +7 −0
kernel/cgroup/cgroup.c +48 −0
kernel/sched/psi.c +17 −13

Documentation/admin-guide/kernel-parameters.txt+7−2

 	ccw_timeout_log	[S390]
 			See Documentation/s390/common_io.rst for details.
-	cgroup_disable=	[KNL] Disable a particular controller
+	cgroup_disable=	[KNL] Disable a particular controller or optional feature
-			Format: {name of the controller(s) to disable}
+			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+1−0

@@ -113,6 +113,7 @@ enum { · 112 unmodified lines
113	CFTYPE_NO_PREFIX = (1 << 3), /* (DON'T USE FOR NEW FILES) no subsys prefix */	113	CFTYPE_NO_PREFIX = (1 << 3), /* (DON'T USE FOR NEW FILES) no subsys prefix */
114	CFTYPE_WORLD_WRITABLE = (1 << 4), /* (DON'T USE FOR NEW FILES) S_IWUGO */	114	CFTYPE_WORLD_WRITABLE = (1 << 4), /* (DON'T USE FOR NEW FILES) S_IWUGO */
115	CFTYPE_DEBUG = (1 << 5), /* create when cgroup_debug */	115	CFTYPE_DEBUG = (1 << 5), /* create when cgroup_debug */
		116	CFTYPE_PRESSURE = (1 << 6), /* only if pressure feature is enabled */
116		117
117	/* internal flags, do not use outside cgroup core proper */	118	/* internal flags, do not use outside cgroup core proper */
118	__CFTYPE_ONLY_ON_DFL = (1 << 16), /* only on default hierarchy */	119	__CFTYPE_ONLY_ON_DFL = (1 << 16), /* only on default hierarchy */

include/linux/cgroup.h+7−0

 	return &cgrp->psi;
 }
+bool cgroup_psi_enabled(void);
 static inline void cgroup_init_kthreadd(void)
 {
 	/*
 	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+48−0

 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,
 {
 	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)
 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))
 		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
 #ifdef CONFIG_PSI
 	{
 		.name = "io.pressure",
+		.flags = CFTYPE_PRESSURE,
 		.seq_show = cgroup_io_pressure_show,
 		.write = cgroup_io_pressure_write,
 		.poll = cgroup_pressure_poll,
 	},
 	{
 		.name = "memory.pressure",
+		.flags = CFTYPE_PRESSURE,
 		.seq_show = cgroup_memory_pressure_show,
 		.write = cgroup_memory_pressure_write,
 		.poll = cgroup_pressure_poll,
 	},
 	{
 		.name = "cpu.pressure",
+		.flags = CFTYPE_PRESSURE,
 		.seq_show = cgroup_cpu_pressure_show,
 		.write = cgroup_cpu_pressure_write,
 		.poll = cgroup_pressure_poll,
 			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;
 }
 		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+17−13

 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;
 		return;
 	}
+	if (!cgroup_psi_enabled())
+		static_branch_disable(&psi_cgroups_enabled);
 	psi_period = jiffies_to_nsecs(PSI_FREQ);
 	group_init(&psi_system);
 }
 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)
+		if (!*iter)
-		cgroup = task->cgroups->dfl_cgrp;
+			cgroup = task->cgroups->dfl_cgrp;
-	else if (*iter == &psi_system)
+		else
-		return NULL;
+			cgroup = cgroup_parent(*iter);
-	else
-		cgroup = cgroup_parent(*iter);
-	if (cgroup && cgroup_parent(cgroup)) {
+		if (cgroup && cgroup_parent(cgroup)) {
-		*iter = cgroup;
+			*iter = cgroup;
-		return cgroup_psi(cgroup);
+			return cgroup_psi(cgroup);
+		}
 	}
-#else
-	if (*iter)
-		return NULL;
 #endif
 	*iter = &psi_system;
 	return &psi_system;

cgroup: make per-cgroup pressure stall tracking configurable

Keyboard shortcuts

General

Issues and pull requests

Issues, pull requests and Inbox

Repository code