Merge "sched: warn/panic upon excessive scheduling latency"

Documentation/scheduler/sched-stats.txt

@@ -147,6 +147,15 @@ this file correlating for that process to:
      2) time spent waiting on a runqueue
      3) # of timeslices run on this cpu
 
+/proc/sys/kernel/{sched_latency_warn_threshold_us,sched_latency_panic_threshold_us}
+----------------
+schedstats provides procfs nodes /proc/sys/kernel/sched_latency_warn_threshold_us
+and /proc/sys/kernel/sched_latency_panic_threshold_us.  These can be configured
+to detect unreasonably high scheduling latency.
+Set sched_latency_warn_threshold_us or sched_latency_panic_threshold_us with
+non-zero threshold to warn or panic system when scheduling latency higher than
+configured threshold is detected.  Default is 0 (disabled) for both.
+
 A program could be easily written to make use of these extra fields to
 report on how well a particular process or set of processes is faring
 under the scheduler's policies.  A simple version of such a program is

include/linux/sched/sysctl.h

@@ -140,6 +140,11 @@ extern unsigned int sysctl_sched_cfs_bandwidth_slice;
 extern unsigned int sysctl_sched_autogroup_enabled;
 #endif
 
+#ifdef CONFIG_SCHEDSTATS
+extern unsigned int sysctl_sched_latency_panic_threshold;
+extern unsigned int sysctl_sched_latency_warn_threshold;
+#endif
+
 extern int sched_rr_timeslice;
 
 extern int sched_rr_handler(struct ctl_table *table, int write,

kernel/sched/fair.c

@@ -29,6 +29,7 @@
 #include <linux/mempolicy.h>
 #include <linux/migrate.h>
 #include <linux/task_work.h>
+#include <linux/ratelimit.h>
 
 #include <trace/events/sched.h>
 
@@ -121,6 +122,11 @@ unsigned int __read_mostly sysctl_sched_shares_window = 10000000UL;
 unsigned int sysctl_sched_cfs_bandwidth_slice = 5000UL;
 #endif
 
+#ifdef CONFIG_SCHEDSTATS
+unsigned int sysctl_sched_latency_panic_threshold;
+unsigned int sysctl_sched_latency_warn_threshold;
+#endif /* CONFIG_SCHEDSTATS */
+
 /*
  * Increase the granularity value when there are more CPUs,
  * because with more CPUs the 'effective latency' as visible
@@ -751,6 +757,31 @@ static void update_stats_enqueue(struct cfs_rq *cfs_rq, struct sched_entity *se,
 		update_stats_wait_start(cfs_rq, se, migrating);
 }
 
+#ifdef CONFIG_SCHEDSTATS
+static inline void check_for_high_latency(struct task_struct *p, u64 latency_us)
+{
+	int do_warn, do_panic;
+	const char *fmt = "excessive latency comm=%s pid=%d latency=%llu(us)\n";
+	static DEFINE_RATELIMIT_STATE(rs, DEFAULT_RATELIMIT_INTERVAL,
+				      DEFAULT_RATELIMIT_BURST);
+
+	do_warn = (sysctl_sched_latency_warn_threshold &&
+		   latency_us > sysctl_sched_latency_warn_threshold);
+	do_panic = (sysctl_sched_latency_panic_threshold &&
+		    latency_us > sysctl_sched_latency_panic_threshold);
+	if (unlikely(do_panic || (do_warn && __ratelimit(&rs)))) {
+		if (do_panic)
+			panic(fmt, p->comm, p->pid, latency_us);
+		else
+			printk_sched(fmt, p->comm, p->pid, latency_us);
+	}
+}
+#else
+static inline void check_for_high_latency(struct task_struct *p, u64 latency)
+{
+}
+#endif
+
 static void
 update_stats_wait_end(struct cfs_rq *cfs_rq, struct sched_entity *se,
 		      bool migrating)
@@ -768,8 +799,13 @@ update_stats_wait_end(struct cfs_rq *cfs_rq, struct sched_entity *se,
 			rq_of(cfs_rq)->clock - se->statistics.wait_start);
 #ifdef CONFIG_SCHEDSTATS
 	if (entity_is_task(se)) {
-		trace_sched_stat_wait(task_of(se),
-			rq_of(cfs_rq)->clock - se->statistics.wait_start);
+		u64 delta;
+
+		delta = rq_of(cfs_rq)->clock - se->statistics.wait_start;
+		trace_sched_stat_wait(task_of(se), delta);
+
+		delta = delta >> 10;
+		check_for_high_latency(task_of(se), delta);
 	}
 #endif
 	schedstat_set(se->statistics.wait_start, 0);

kernel/sysctl.c

@@ -603,6 +603,22 @@ static struct ctl_table kern_table[] = {
 		.extra1		= &one,
 	},
 #endif
+#ifdef CONFIG_SCHEDSTATS
+	{
+		.procname	= "sched_latency_panic_threshold_us",
+		.data		= &sysctl_sched_latency_panic_threshold,
+		.maxlen		= sizeof(unsigned int),
+		.mode		= 0644,
+		.proc_handler	= proc_dointvec_minmax,
+	},
+	{
+		.procname	= "sched_latency_warn_threshold_us",
+		.data		= &sysctl_sched_latency_warn_threshold,
+		.maxlen		= sizeof(unsigned int),
+		.mode		= 0644,
+		.proc_handler	= proc_dointvec_minmax,
+	},
+#endif
 #ifdef CONFIG_PROVE_LOCKING
 	{
 		.procname	= "prove_locking",