[RFC PATCH v7 4/5] tracing: Measure delayed hrtimer offset latency

[Binoy Jayan]

Measure latencies caused due to delayed timer offsets in nanoseconds.
It shows the latency captured due to a delayed timer expire event. It
happens for example when a timer misses its deadline due to disabled
interrupts. A process if scheduled as a result of the timer expiration
suffers this latency. It is used to calculate the total wakeup latency
of a process which is the sum of the delayed timer offset and the
wakeup latency.

[
Initial work and idea by Carsten
Link: https://git.kernel.org/cgit/linux/kernel/git/rt/linux-stable-rt.git/commit/?h=v3.14-rt-rebase&id=56d50cc34943bbba12b8c5942ee1ae3b29f73acb
]

Cc: Carsten Emde <C.Emde@xxxxxxxxx>
Signed-off-by: Binoy Jayan <binoy.jayan@xxxxxxxxxx>
---
 include/linux/hrtimer.h |  4 ++++
 include/linux/sched.h   |  3 +++
 kernel/time/Kconfig     |  8 ++++++++
 kernel/time/hrtimer.c   | 47 +++++++++++++++++++++++++++++++++++++++++++++++
 4 files changed, 62 insertions(+)

diff --git a/include/linux/hrtimer.h b/include/linux/hrtimer.h
index 5e00f80..05d8086 100644
--- a/include/linux/hrtimer.h
+++ b/include/linux/hrtimer.h
@@ -90,6 +90,7 @@ enum hrtimer_restart {
  * @is_rel:	Set if the timer was armed relative
  * @start_pid:  timer statistics field to store the pid of the task which
  *		started the timer
+ * @tim_expiry: hrtimer expiry time or 0 in case already expired
  * @start_site:	timer statistics field to store the site where the timer
  *		was started
  * @start_comm: timer statistics field to store the name of the process which
@@ -104,6 +105,9 @@ struct hrtimer {
 	struct hrtimer_clock_base	*base;
 	u8				state;
 	u8				is_rel;
+#ifdef CONFIG_TRACE_DELAYED_TIMER_OFFSETS
+	ktime_t				tim_expiry;
+#endif
 #ifdef CONFIG_TIMER_STATS
 	int				start_pid;
 	void				*start_site;
diff --git a/include/linux/sched.h b/include/linux/sched.h
index 62c68e5..7bf67f8 100644
--- a/include/linux/sched.h
+++ b/include/linux/sched.h
@@ -1891,6 +1891,9 @@ struct task_struct {
 	/* bitmask and counter of trace recursion */
 	unsigned long trace_recursion;
 #endif /* CONFIG_TRACING */
+#ifdef CONFIG_TRACE_DELAYED_TIMER_OFFSETS
+	long timer_offset;
+#endif /* CONFIG_TRACE_DELAYED_TIMER_OFFSETS */
 #ifdef CONFIG_KCOV
 	/* Coverage collection mode enabled for this task (0 if disabled). */
 	enum kcov_mode kcov_mode;
diff --git a/kernel/time/Kconfig b/kernel/time/Kconfig
index 4008d9f..de4793c 100644
--- a/kernel/time/Kconfig
+++ b/kernel/time/Kconfig
@@ -193,5 +193,13 @@ config HIGH_RES_TIMERS
 	  hardware is not capable then this option only increases
 	  the size of the kernel image.
 
+config TRACE_DELAYED_TIMER_OFFSETS
+	depends on HIGH_RES_TIMERS
+	select GENERIC_TRACER
+	bool "Delayed Timer Offsets"
+	help
+	  Capture offsets of delayed hrtimer in nanoseconds. It is used
+	  to construct wakeup latency histogram.
+
 endmenu
 endif
diff --git a/kernel/time/hrtimer.c b/kernel/time/hrtimer.c
index 9ba7c82..7048f86 100644
--- a/kernel/time/hrtimer.c
+++ b/kernel/time/hrtimer.c
@@ -56,6 +56,8 @@
 
 #include "tick-internal.h"
 
+static enum hrtimer_restart hrtimer_wakeup(struct hrtimer *timer);
+
 /*
  * The timer bases:
  *
@@ -960,6 +962,47 @@ static inline ktime_t hrtimer_update_lowres(struct hrtimer *timer, ktime_t tim,
 	return tim;
 }
 
+#ifdef CONFIG_TRACE_DELAYED_TIMER_OFFSETS
+static inline void latency_hrtimer_timing_start(struct hrtimer *timer,
+					 struct hrtimer_clock_base *new_base,
+					 ktime_t tim)
+{
+	ktime_t now = new_base->get_time();
+
+	if (ktime_to_ns(tim) < ktime_to_ns(now))
+		timer->tim_expiry = now;
+	else
+		timer->tim_expiry = ktime_set(0, 0);
+}
+
+static inline void latency_hrtimer_timing_stop(struct hrtimer *timer,
+						ktime_t basenow)
+{
+	long latency;
+	struct task_struct *task;
+
+	latency = ktime_to_ns(ktime_sub(basenow,
+			      ktime_to_ns(timer->tim_expiry) ?
+			      timer->tim_expiry : hrtimer_get_expires(timer)));
+	task = timer->function == hrtimer_wakeup ?
+			container_of(timer, struct hrtimer_sleeper,
+				     timer)->task : NULL;
+	if (task && latency > 0)
+		task->timer_offset = latency;
+}
+#else
+static inline void latency_hrtimer_timing_start(struct hrtimer *timer,
+					 struct hrtimer_clock_base *new_base,
+					 ktime_t tim)
+{
+}
+static inline void latency_hrtimer_timing_stop(struct hrtimer *timer,
+						ktime_t basenow)
+{
+}
+
+#endif
+
 /**
  * hrtimer_start_range_ns - (re)start an hrtimer on the current CPU
  * @timer:	the timer to be added
@@ -992,6 +1035,8 @@ void hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim,
 
 	timer_stats_hrtimer_set_start_info(timer);
 
+	latency_hrtimer_timing_start(timer, new_base, tim);
+
 	leftmost = enqueue_hrtimer(timer, new_base);
 	if (!leftmost)
 		goto unlock;
@@ -1284,6 +1329,8 @@ static void __hrtimer_run_queues(struct hrtimer_cpu_base *cpu_base, ktime_t now)
 
 			timer = container_of(node, struct hrtimer, node);
 
+			latency_hrtimer_timing_stop(timer, basenow);
+
 			/*
 			 * The immediate goal for using the softexpires is
 			 * minimizing wakeups, not running timers at the
-- 
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum,
a Linux Foundation Collaborative Project