[HACK][RFC] sysfs, hrtimer: Add sysfs knob to defer non-rt timers to softirq context

[Gratian Crisan]

Current hrtimer implementation processes all expired timers marked as
irqsafe in the timer irq context. This includes timers created by non-rt
tasks and as a result it can create unbounded latencies for real-time tasks
waiting on an hrtimer to expire.

Provide /sys/kernel/hrtimer_defer_nonrt configuration knob which when set
results in only marking as irqsafe timers created by a rt task. The rest
will get deferred to the HRTIMER_SOFTIRQ for handling via the existing
mechanism in __hrtimer_run_queues/hrtimer_rt_defer.

Signed-off-by: Gratian Crisan <gratian.crisan@xxxxxx>
---
 kernel/ksysfs.c       | 18 ++++++++++++++++++
 kernel/time/hrtimer.c |  4 +++-
 2 files changed, 21 insertions(+), 1 deletion(-)

diff --git a/kernel/ksysfs.c b/kernel/ksysfs.c
index ddef079..9fd4f50 100644
--- a/kernel/ksysfs.c
+++ b/kernel/ksysfs.c
@@ -189,6 +189,23 @@ static ssize_t rcu_normal_store(struct kobject *kobj,
 KERNEL_ATTR_RW(rcu_normal);
 #endif /* #ifndef CONFIG_TINY_RCU */
 
+extern int hrtimer_defer_nonrt;
+static ssize_t hrtimer_defer_nonrt_show(struct kobject *kobj,
+					struct kobj_attribute *attr, char *buf)
+{
+	return sprintf(buf, "%d\n", hrtimer_defer_nonrt);
+}
+static ssize_t hrtimer_defer_nonrt_store(struct kobject *kobj,
+					struct kobj_attribute *attr,
+					const char *buf, size_t count)
+{
+	if (kstrtoint(buf, 0, &hrtimer_defer_nonrt))
+		return -EINVAL;
+
+	return count;
+}
+KERNEL_ATTR_RW(hrtimer_defer_nonrt);
+
 /*
  * Make /sys/kernel/notes give the raw contents of our kernel .notes section.
  */
@@ -237,6 +254,7 @@ static struct attribute * kernel_attrs[] = {
 #ifdef CONFIG_PREEMPT_RT_FULL
 	&realtime_attr.attr,
 #endif
+	&hrtimer_defer_nonrt_attr.attr,
 	NULL
 };
 
diff --git a/kernel/time/hrtimer.c b/kernel/time/hrtimer.c
index d85f638..df5783a 100644
--- a/kernel/time/hrtimer.c
+++ b/kernel/time/hrtimer.c
@@ -106,6 +106,8 @@ static inline int hrtimer_clockid_to_base(clockid_t clock_id)
 	return hrtimer_clock_to_base_table[clock_id];
 }
 
+int hrtimer_defer_nonrt;
+
 /*
  * Functions and macros which are different for UP/SMP systems are kept in a
  * single place
@@ -1644,7 +1646,7 @@ static enum hrtimer_restart hrtimer_wakeup(struct hrtimer *timer)
 void hrtimer_init_sleeper(struct hrtimer_sleeper *sl, struct task_struct *task)
 {
 	sl->timer.function = hrtimer_wakeup;
-	sl->timer.irqsafe = 1;
+	sl->timer.irqsafe = (hrtimer_defer_nonrt) ? rt_task(task) : 1;
 	sl->task = task;
 }
 EXPORT_SYMBOL_GPL(hrtimer_init_sleeper);
-- 
2.10.2