Make hrtimer granularity and minimum hrtimeout configurable in sysctl…

…. Set default granularity to 100us and min timeout to 500us.

kernel/sysctl.c

@@ -128,7 +128,9 @@ extern int sched_interactive;
 extern int sched_iso_cpu;
 extern int sched_yield_type;
 #endif
-#ifdef CONFIG_PRINTK
+extern int hrtimer_granularity_us;
+extern int hrtimeout_min_us;
+#if defined(CONFIG_PRINTK) || defined(CONFIG_SCHED_MUQSS)
 static int ten_thousand = 10000;
 #endif
 #ifdef CONFIG_PERF_EVENTS
@@ -1901,7 +1903,24 @@ static struct ctl_table kern_table[] = {
 	},
 #endif /* CONFIG_SMP && CONFIG_SCHEDSTATS */
 #endif /* CONFIG_SCHED_MUQSS */
-
+	{
+		.procname	= "hrtimer_granularity_us",
+		.data		= &hrtimer_granularity_us,
+		.maxlen		= sizeof(int),
+		.mode		= 0644,
+		.proc_handler	= &proc_dointvec_minmax,
+		.extra1		= &one,
+		.extra2		= &ten_thousand,
+	},
+	{
+		.procname	= "hrtimeout_min_us",
+		.data		= &hrtimeout_min_us,
+		.maxlen		= sizeof(int),
+		.mode		= 0644,
+		.proc_handler	= &proc_dointvec_minmax,
+		.extra1		= &one,
+		.extra2		= &ten_thousand,
+	},
 #if defined(CONFIG_ENERGY_MODEL) && defined(CONFIG_CPU_FREQ_GOV_SCHEDUTIL)
 	{
 		.procname	= "sched_energy_aware",

kernel/time/clockevents.c

@@ -190,13 +190,9 @@ int clockevents_tick_resume(struct clock_event_device *dev)
 
 #ifdef CONFIG_GENERIC_CLOCKEVENTS_MIN_ADJUST
 
-#ifdef CONFIG_SCHED_MUQSS
+int __read_mostly hrtimer_granularity_us = 100;
 /* Limit min_delta to 100us */
-#define MIN_DELTA_LIMIT		(NSEC_PER_SEC / 10000)
-#else
-/* Limit min_delta to a jiffie */
-#define MIN_DELTA_LIMIT		(NSEC_PER_SEC / HZ)
-#endif
+#define MIN_DELTA_LIMIT		(hrtimer_granularity_us * NSEC_PER_USEC)
 
 /**
  * clockevents_increase_min_delta - raise minimum delta of a clock event device

kernel/time/hrtimer.c

@@ -2244,7 +2244,7 @@ EXPORT_SYMBOL_GPL(schedule_hrtimeout);
 long __sched schedule_msec_hrtimeout(long timeout)
 {
 	struct hrtimer_sleeper t;
-	int delta, secs, jiffs;
+	int delta, jiffs;
 	ktime_t expires;
 
 	if (!timeout) {
@@ -2261,9 +2261,8 @@ long __sched schedule_msec_hrtimeout(long timeout)
 	if (jiffs > 4 || hrtimer_resolution >= NSEC_PER_SEC / HZ || pm_freezing)
 		return schedule_timeout(jiffs);
 
-	secs = timeout / 1000;
 	delta = (timeout % 1000) * NSEC_PER_MSEC;
-	expires = ktime_set(secs, delta);
+	expires = ktime_set(0, delta);
 
 	hrtimer_init_sleeper_on_stack(&t, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
 	hrtimer_set_expires_range_ns(&t.timer, expires, delta);
@@ -2285,9 +2284,51 @@ long __sched schedule_msec_hrtimeout(long timeout)
 
 EXPORT_SYMBOL(schedule_msec_hrtimeout);
 
+#define USECS_PER_SEC 1000000
+extern int hrtimer_granularity_us;
+
+static inline long schedule_usec_hrtimeout(long timeout)
+{
+	struct hrtimer_sleeper t;
+	ktime_t expires;
+	int delta;
+
+	if (!timeout) {
+		__set_current_state(TASK_RUNNING);
+		return 0;
+	}
+
+	if (hrtimer_resolution >= NSEC_PER_SEC / HZ)
+		return schedule_timeout(usecs_to_jiffies(timeout));
+
+	if (timeout < hrtimer_granularity_us)
+		timeout = hrtimer_granularity_us;
+	delta = (timeout % USECS_PER_SEC) * NSEC_PER_USEC;
+	expires = ktime_set(0, delta);
+
+	hrtimer_init_sleeper_on_stack(&t, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
+	hrtimer_set_expires_range_ns(&t.timer, expires, delta);
+
+	hrtimer_sleeper_start_expires(&t, HRTIMER_MODE_REL);
+
+	if (likely(t.task))
+		schedule();
+
+	hrtimer_cancel(&t.timer);
+	destroy_hrtimer_on_stack(&t.timer);
+
+	__set_current_state(TASK_RUNNING);
+
+	expires = hrtimer_expires_remaining(&t.timer);
+	timeout = ktime_to_us(expires);
+	return timeout < 0 ? 0 : timeout;
+}
+
+int __read_mostly hrtimeout_min_us = 500;
+
 long __sched schedule_min_hrtimeout(void)
 {
-	return schedule_msec_hrtimeout(1);
+	return usecs_to_jiffies(schedule_usec_hrtimeout(hrtimeout_min_us));
 }
 
 EXPORT_SYMBOL(schedule_min_hrtimeout);