Skip to contentSkip to conversation

optimized ktime_get[_ts] for GENERIC_TIME=y

PATCH In review Martin Schwidefsky 4 participants 9 messages Jul 6, 2009 – Jul 7, 2009 Timekeeping, clocksource core, NTP, alarmtimer

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 optimized ktime_get[_ts] for GENERIC_TIME=y

From: Martin Schwidefsky <[email protected]>

The generic ktime_get function defined in kernel/hrtimer.c is suboptimial for GENERIC_TIME=y:

 0)               |  ktime_get() {
 0)               |    ktime_get_ts() {
 0)               |      getnstimeofday() {
 0)               |        read_tod_clock() {
 0)   0.601 us    |        }
 0)   1.938 us    |      }
 0)               |      set_normalized_timespec() {
 0)   0.602 us    |      }
 0)   4.375 us    |    }
 0)   5.523 us    |  }

Overall there are two read_seqbegin/read_seqretry loops and a lot of unnecessary struct timespec calculations. ktime_get returns a nano second value which is the sum of xtime, wall_to_monotonic and the nano second delta from the clock source.

ktime_get can be optimized for GENERIC_TIME=y. The new version only calls clocksource_read:

 0)               |  ktime_get() {
 0)               |    read_tod_clock() {
 0)   0.610 us    |    }
 0)   1.977 us    |  }

It uses a single read_seqbegin/readseqretry loop and just adds everthing to a nano second value.

ktime_get_ts is optimized in a similar fashion.

Cc Ingo Molnar <[email protected]>
Cc Thomas Gleixner <[email protected]>
Cc john stultz <[email protected]>
---
 kernel/hrtimer.c          |    4 ++
 kernel/time/timekeeping.c |   63 ++++++++++++++++++++++++++++++++++++++++++++++
 2 files changed, 67 insertions(+)
kernel/hrtimer.c +4 -0
diff -urpN linux-2.5/kernel/hrtimer.c linux-2.5-patched/kernel/hrtimer.c
@@ -46,6 +46,7 @@
 
 #include <asm/uaccess.h>
 
+#ifndef CONFIG_GENERIC_TIME
 /**
  * ktime_get - get the monotonic time in ktime_t format
  *
@@ -60,6 +61,7 @@ ktime_t ktime_get(void)
 	return timespec_to_ktime(now);
 }
 EXPORT_SYMBOL_GPL(ktime_get);
+#endif
 
 /**
  * ktime_get_real - get the real (wall-) time in ktime_t format
@@ -104,6 +106,7 @@ DEFINE_PER_CPU(struct hrtimer_cpu_base, 
 	}
 };
 
+#ifndef CONFIG_GENERIC_TIME
 /**
  * ktime_get_ts - get the monotonic clock in timespec format
  * @ts:		pointer to timespec variable
@@ -128,6 +131,7 @@ void ktime_get_ts(struct timespec *ts)
 				ts->tv_nsec + tomono.tv_nsec);
 }
 EXPORT_SYMBOL_GPL(ktime_get_ts);
+#endif
 
 /*
  * Get the coarse grained time at the softirq based on xtime and
kernel/time/timekeeping.c +63 -0

69 lines not on this page · raw message

Signature · 4 lines
blue skies,
   Martin.

"Reality continues to ruin my life." - Calvin.
#1 lore raw
Quoted 85 lines · Martin Schwidefsky

From: Martin Schwidefsky <[email protected]>

The generic ktime_get function defined in kernel/hrtimer.c is suboptimial for GENERIC_TIME=y:

 0)               |  ktime_get() {
 0)               |    ktime_get_ts() {
 0)               |      getnstimeofday() {
 0)               |        read_tod_clock() {
 0)   0.601 us    |        }
 0)   1.938 us    |      }
 0)               |      set_normalized_timespec() {
 0)   0.602 us    |      }
 0)   4.375 us    |    }
 0)   5.523 us    |  }

Overall there are two read_seqbegin/read_seqretry loops and a lot of unnecessary struct timespec calculations. ktime_get returns a nano second value which is the sum of xtime, wall_to_monotonic and the nano second delta from the clock source.

ktime_get can be optimized for GENERIC_TIME=y. The new version only calls clocksource_read:

 0)               |  ktime_get() {
 0)               |    read_tod_clock() {
 0)   0.610 us    |    }
 0)   1.977 us    |  }

It uses a single read_seqbegin/readseqretry loop and just adds everthing to a nano second value.

ktime_get_ts is optimized in a similar fashion.

Cc Ingo Molnar <[email protected]>
Cc Thomas Gleixner <[email protected]>
Cc john stultz <[email protected]>
---
 kernel/hrtimer.c          |    4 ++
 kernel/time/timekeeping.c |   63 ++++++++++++++++++++++++++++++++++++++++++++++
 2 files changed, 67 insertions(+)

diff -urpN linux-2.5/kernel/hrtimer.c linux-2.5-patched/kernel/hrtimer.c

kernel/hrtimer.c +4 -0
@@ -46,6 +46,7 @@
 
 #include <asm/uaccess.h>
 
+#ifndef CONFIG_GENERIC_TIME
 /**
  * ktime_get - get the monotonic time in ktime_t format
  *
@@ -60,6 +61,7 @@ ktime_t ktime_get(void)
 	return timespec_to_ktime(now);
 }
 EXPORT_SYMBOL_GPL(ktime_get);
+#endif
 
 /**
  * ktime_get_real - get the real (wall-) time in ktime_t format
@@ -104,6 +106,7 @@ DEFINE_PER_CPU(struct hrtimer_cpu_base, 
 	}
 };
 
+#ifndef CONFIG_GENERIC_TIME
 /**
  * ktime_get_ts - get the monotonic clock in timespec format
  * @ts:		pointer to timespec variable
@@ -128,6 +131,7 @@ void ktime_get_ts(struct timespec *ts)
 				ts->tv_nsec + tomono.tv_nsec);
 }
 EXPORT_SYMBOL_GPL(ktime_get_ts);
+#endif
 
 /*
  * Get the coarse grained time at the softirq based on xtime and

diff -urpN linux-2.5/kernel/time/timekeeping.c linux-2.5-patched/kernel/time/timekeeping.c

kernel/time/timekeeping.c +0 -0
@@ -118,6 +118,69 @@ void getnstimeofday(struct timespec *ts)
 
 EXPORT_SYMBOL(getnstimeofday);

kernel/time/timekeeping.c void getnstimeofday(struct timespec *ts) · line 129

Quoted 10 lines
kernel/time/timekeeping.c +9 -0
 
+ktime_t ktime_get(void)
+{
+	cycle_t cycle_now, cycle_delta;
+	struct timespec time;
+	unsigned long seq;
+	s64 nsecs;
+
+	do {
+		seq = read_seqbegin(&xtime_lock);

minor nit : read_seqbegin() returns an "unsigned int", not an "unsigned long"

But apparently, most callers store the value into an unsigned long... oh well...

#2 lore raw
Quoted 7 lines · On Mon, 06 Jul 2009 16:07:22 +0200 Eric Dumazet <[email protected]> wrote:
Quoted 7 lines

diff -urpN linux-2.5/kernel/time/timekeeping.c linux-2.5-patched/kernel/time/timekeeping.c

kernel/time/timekeeping.c +0 -0
@@ -118,6 +118,69 @@ void getnstimeofday(struct timespec *ts)
 
 EXPORT_SYMBOL(getnstimeofday);
 
Quoted 11 lines

kernel/time/timekeeping.c void getnstimeofday(struct timespec *ts) · line 129

Quoted 9 lines
kernel/time/timekeeping.c +9 -0
+ktime_t ktime_get(void)
+{
+	cycle_t cycle_now, cycle_delta;
+	struct timespec time;
+	unsigned long seq;
+	s64 nsecs;
+
+	do {
+		seq = read_seqbegin(&xtime_lock);  

minor nit : read_seqbegin() returns an "unsigned int", not an "unsigned long"

Hmm, just cut-copy-pasted the seq-loop. Seems like a job for the janitor .. But I will update my patch.

Signature · 4 lines
blue skies,
   Martin.

"Reality continues to ruin my life." - Calvin.
Acked-by on v1 1/1
#3 lore raw
In reply to Martin Schwidefsky
Quoted 32 lines · Martin Schwidefsky

From: Martin Schwidefsky <[email protected]>

The generic ktime_get function defined in kernel/hrtimer.c is suboptimial for GENERIC_TIME=y:

 0)               |  ktime_get() {
 0)               |    ktime_get_ts() {
 0)               |      getnstimeofday() {
 0)               |        read_tod_clock() {
 0)   0.601 us    |        }
 0)   1.938 us    |      }
 0)               |      set_normalized_timespec() {
 0)   0.602 us    |      }
 0)   4.375 us    |    }
 0)   5.523 us    |  }

Overall there are two read_seqbegin/read_seqretry loops and a lot of unnecessary struct timespec calculations. ktime_get returns a nano second value which is the sum of xtime, wall_to_monotonic and the nano second delta from the clock source.

ktime_get can be optimized for GENERIC_TIME=y. The new version only calls clocksource_read:

 0)               |  ktime_get() {
 0)               |    read_tod_clock() {
 0)   0.610 us    |    }
 0)   1.977 us    |  }

It uses a single read_seqbegin/readseqretry loop and just adds everthing to a nano second value.

Quoted 6 lines

ktime_get_ts is optimized in a similar fashion.

Cc Ingo Molnar <[email protected]>
Cc Thomas Gleixner <[email protected]>
Cc john stultz <[email protected]>

Looks good to me.

Quoted 116 lines
---
 kernel/hrtimer.c          |    4 ++
 kernel/time/timekeeping.c |   63 ++++++++++++++++++++++++++++++++++++++++++++++
 2 files changed, 67 insertions(+)

diff -urpN linux-2.5/kernel/hrtimer.c linux-2.5-patched/kernel/hrtimer.c

kernel/hrtimer.c +4 -0
@@ -46,6 +46,7 @@
 
 #include <asm/uaccess.h>
 
+#ifndef CONFIG_GENERIC_TIME
 /**
  * ktime_get - get the monotonic time in ktime_t format
  *
@@ -60,6 +61,7 @@ ktime_t ktime_get(void)
 	return timespec_to_ktime(now);
 }
 EXPORT_SYMBOL_GPL(ktime_get);
+#endif
 
 /**
  * ktime_get_real - get the real (wall-) time in ktime_t format
@@ -104,6 +106,7 @@ DEFINE_PER_CPU(struct hrtimer_cpu_base, 
 	}
 };
 
+#ifndef CONFIG_GENERIC_TIME
 /**
  * ktime_get_ts - get the monotonic clock in timespec format
  * @ts:		pointer to timespec variable
@@ -128,6 +131,7 @@ void ktime_get_ts(struct timespec *ts)
 				ts->tv_nsec + tomono.tv_nsec);
 }
 EXPORT_SYMBOL_GPL(ktime_get_ts);
+#endif
 
 /*
  * Get the coarse grained time at the softirq based on xtime and

diff -urpN linux-2.5/kernel/time/timekeeping.c linux-2.5-patched/kernel/time/timekeeping.c

kernel/time/timekeeping.c +63 -0
@@ -118,6 +118,69 @@ void getnstimeofday(struct timespec *ts)
 
 EXPORT_SYMBOL(getnstimeofday);
 
+ktime_t ktime_get(void)
+{
+	cycle_t cycle_now, cycle_delta;
+	struct timespec time;
+	unsigned long seq;
+	s64 nsecs;
+
+	do {
+		seq = read_seqbegin(&xtime_lock);
+		time.tv_sec = xtime.tv_sec + wall_to_monotonic.tv_sec;
+		time.tv_nsec = xtime.tv_nsec + wall_to_monotonic.tv_nsec;
+
+		/* read clocksource: */
+		cycle_now = clocksource_read(clock);
+
+		/* calculate the delta since the last update_wall_time: */
+		cycle_delta = (cycle_now - clock->cycle_last) & clock->mask;
+
+		/* convert to nanoseconds: */
+		nsecs = cyc2ns(clock, cycle_delta);
+
+	} while (read_seqretry(&xtime_lock, seq));
+	nsecs += time.tv_sec * 1000000000ULL + time.tv_nsec;
+	return (ktime_t) { .tv64 = nsecs };
+}
+EXPORT_SYMBOL_GPL(ktime_get);
+
+/**
+ * ktime_get_ts - get the monotonic clock in timespec format
+ * @ts:		pointer to timespec variable
+ *
+ * The function calculates the monotonic clock from the realtime
+ * clock and the wall_to_monotonic offset and stores the result
+ * in normalized timespec format in the variable pointed to by @ts.
+ */
+void ktime_get_ts(struct timespec *ts)
+{
+	cycle_t cycle_now, cycle_delta;
+	struct timespec tomono;
+	unsigned long seq;
+	s64 nsecs;
+
+	do {
+		seq = read_seqbegin(&xtime_lock);
+		*ts = xtime;
+		tomono = wall_to_monotonic;
+
+		/* read clocksource: */
+		cycle_now = clocksource_read(clock);
+
+		/* calculate the delta since the last update_wall_time: */
+		cycle_delta = (cycle_now - clock->cycle_last) & clock->mask;
+
+		/* convert to nanoseconds: */
+		nsecs = cyc2ns(clock, cycle_delta);
+
+	} while (read_seqretry(&xtime_lock, seq));
+
+	set_normalized_timespec(ts, ts->tv_sec + tomono.tv_sec,
+				ts->tv_nsec + tomono.tv_nsec + nsecs);
+}
+EXPORT_SYMBOL_GPL(ktime_get_ts);
+
 /**
  * do_gettimeofday - Returns the time of day in a timeval
  * @tv:		pointer to the timeval to be set
#4 lore raw
In reply to Martin Schwidefsky

kernel/time/timekeeping.c void getnstimeofday(struct timespec *ts) · line 131

Quoted 11 lines · Martin Schwidefsky
kernel/time/timekeeping.c +11 -0
+ktime_t ktime_get(void)
+{
+	cycle_t cycle_now, cycle_delta;
+	struct timespec time;
+	unsigned long seq;
+	s64 nsecs;
+
+	do {
+		seq = read_seqbegin(&xtime_lock);
+		time.tv_sec = xtime.tv_sec + wall_to_monotonic.tv_sec;
+		time.tv_nsec = xtime.tv_nsec + wall_to_monotonic.tv_nsec;

That's actually a violation of the timespec semantics. tv_nsec can end up greater than (10^9 - 1). Please use separate sec and nsec variables.

  	        secs = xtime.tv_sec + wall_to_monotonic.tv_sec;
  	     	nsecs = xtime.tv_nsec + wall_to_monotonic.tv_nsec;

kernel/time/timekeeping.c void getnstimeofday(struct timespec *ts) · line 140

Quoted 8 lines
kernel/time/timekeeping.c +8 -0
+		/* read clocksource: */
+		cycle_now = clocksource_read(clock);
+
+		/* calculate the delta since the last update_wall_time: */
+		cycle_delta = (cycle_now - clock->cycle_last) & clock->mask;
+
+		/* convert to nanoseconds: */
+		nsecs = cyc2ns(clock, cycle_delta);

So that needs to be changed to:

		nsecs += cyc2ns(clock, cycle_delta);

kernel/time/timekeeping.c void getnstimeofday(struct timespec *ts) · line 144

Quoted 4 lines
kernel/time/timekeeping.c +4 -0
+
+	} while (read_seqretry(&xtime_lock, seq));
+	nsecs += time.tv_sec * 1000000000ULL + time.tv_nsec;
+	return (ktime_t) { .tv64 = nsecs };

This will break all 32bit architectures which do not have CONFIG_KTIME_SCALAR set.

With the above changes applied:

  return ktime_add_ns(ktime_set(secs, 0), nsecs);

will do the trick. Might need some comments though :)

Thanks,

tglx

#5 lore raw
Quoted 15 lines · On Mon, 6 Jul 2009 22:31:39 +0200 (CEST) Thomas Gleixner <[email protected]>…

kernel/time/timekeeping.c void getnstimeofday(struct timespec *ts) · line 131

Quoted 11 lines · Martin Schwidefsky
kernel/time/timekeeping.c +11 -0
+ktime_t ktime_get(void)
+{
+	cycle_t cycle_now, cycle_delta;
+	struct timespec time;
+	unsigned long seq;
+	s64 nsecs;
+
+	do {
+		seq = read_seqbegin(&xtime_lock);
+		time.tv_sec = xtime.tv_sec + wall_to_monotonic.tv_sec;
+		time.tv_nsec = xtime.tv_nsec + wall_to_monotonic.tv_nsec;

That's actually a violation of the timespec semantics. tv_nsec can end up greater than (10^9 - 1). Please use separate sec and nsec variables.

Well the tv_sec/tv_nsec fields of a timespec are long values. But its no problem to switch to local variables.

Quoted 14 lines
  	        secs = xtime.tv_sec + wall_to_monotonic.tv_sec;
  	     	nsecs = xtime.tv_nsec + wall_to_monotonic.tv_nsec;

kernel/time/timekeeping.c void getnstimeofday(struct timespec *ts) · line 140

Quoted 8 lines
kernel/time/timekeeping.c +8 -0
+		/* read clocksource: */
+		cycle_now = clocksource_read(clock);
+
+		/* calculate the delta since the last update_wall_time: */
+		cycle_delta = (cycle_now - clock->cycle_last) & clock->mask;
+
+		/* convert to nanoseconds: */
+		nsecs = cyc2ns(clock, cycle_delta);

So that needs to be changed to:

Quoted 1 line
		nsecs += cyc2ns(clock, cycle_delta);

Ok.

Quoted 7 lines

kernel/time/timekeeping.c void getnstimeofday(struct timespec *ts) · line 144

Quoted 4 lines
kernel/time/timekeeping.c +4 -0
+
+	} while (read_seqretry(&xtime_lock, seq));
+	nsecs += time.tv_sec * 1000000000ULL + time.tv_nsec;
+	return (ktime_t) { .tv64 = nsecs };

This will break all 32bit architectures which do not have CONFIG_KTIME_SCALAR set.

Yep, that is a bug.

Quoted 5 lines

With the above changes applied:

  return ktime_add_ns(ktime_set(secs, 0), nsecs);

will do the trick. Might need some comments though :)

Ok, will update the patch. Thanks for the review.

Signature · 4 lines
blue skies,
   Martin.

"Reality continues to ruin my life." - Calvin.
#6 lore raw

Martin,

Quoted 19 lines · Martin Schwidefsky
Quoted 15 lines · On Mon, 6 Jul 2009 22:31:39 +0200 (CEST) Thomas Gleixner <[email protected]>…

kernel/time/timekeeping.c void getnstimeofday(struct timespec *ts) · line 131

Quoted 11 lines · Martin Schwidefsky
kernel/time/timekeeping.c +11 -0
+ktime_t ktime_get(void)
+{
+	cycle_t cycle_now, cycle_delta;
+	struct timespec time;
+	unsigned long seq;
+	s64 nsecs;
+
+	do {
+		seq = read_seqbegin(&xtime_lock);
+		time.tv_sec = xtime.tv_sec + wall_to_monotonic.tv_sec;
+		time.tv_nsec = xtime.tv_nsec + wall_to_monotonic.tv_nsec;

That's actually a violation of the timespec semantics. tv_nsec can end up greater than (10^9 - 1). Please use separate sec and nsec variables.

Quoted 2 lines

Well the tv_sec/tv_nsec fields of a timespec are long values. But its no problem to switch to local variables.

Right. I'm not worried about an overflow of the variable. I was about to suggest using timespec_to_ktime() to fix the !SCALAR problem when I noticed that the timespec is possibly not normalized. Not using a timespec makes the code more obvious I think.

Thanks,

tglx

author Acked-by john stultz
#7 lore raw
Quoted 22 lines · On Tue, 7 Jul 2009 10:08:56 +0200 (CEST) Thomas Gleixner <[email protected]>…

Martin,

Quoted 19 lines · Martin Schwidefsky
Quoted 15 lines · On Mon, 6 Jul 2009 22:31:39 +0200 (CEST) Thomas Gleixner <[email protected]>…

Review

Quoted 11 lines · Martin Schwidefsky
(quoted diff) +11 -0
+ktime_t ktime_get(void)
+{
+	cycle_t cycle_now, cycle_delta;
+	struct timespec time;
+	unsigned long seq;
+	s64 nsecs;
+
+	do {
+		seq = read_seqbegin(&xtime_lock);
+		time.tv_sec = xtime.tv_sec + wall_to_monotonic.tv_sec;
+		time.tv_nsec = xtime.tv_nsec + wall_to_monotonic.tv_nsec;

That's actually a violation of the timespec semantics. tv_nsec can end up greater than (10^9 - 1). Please use separate sec and nsec variables.

Quoted 7 lines
Quoted 2 lines

Well the tv_sec/tv_nsec fields of a timespec are long values. But its no problem to switch to local variables.

Right. I'm not worried about an overflow of the variable. I was about to suggest using timespec_to_ktime() to fix the !SCALAR problem when I noticed that the timespec is possibly not normalized. Not using a timespec makes the code more obvious I think.

That makes sense. New patch:

--
Subject: [PATCH] optimized ktime_get[_ts] for GENERIC_TIME=y

From: Martin Schwidefsky <[email protected]>

The generic ktime_get function defined in kernel/hrtimer.c is suboptimial for GENERIC_TIME=y:

 0)               |  ktime_get() {
 0)               |    ktime_get_ts() {
 0)               |      getnstimeofday() {
 0)               |        read_tod_clock() {
 0)   0.601 us    |        }
 0)   1.938 us    |      }
 0)               |      set_normalized_timespec() {
 0)   0.602 us    |      }
 0)   4.375 us    |    }
 0)   5.523 us    |  }

Overall there are two read_seqbegin/read_seqretry loops and a lot of unnecessary struct timespec calculations. ktime_get returns a nano second value which is the sum of xtime, wall_to_monotonic and the nano second delta from the clock source.

ktime_get can be optimized for GENERIC_TIME=y. The new version only calls clocksource_read:

 0)               |  ktime_get() {
 0)               |    read_tod_clock() {
 0)   0.610 us    |    }
 0)   1.977 us    |  }

It uses a single read_seqbegin/readseqretry loop and just adds everthing to a nano second value.

ktime_get_ts is optimized in a similar fashion.

---

 kernel/hrtimer.c          |    4 ++
 kernel/time/timekeeping.c |   65 ++++++++++++++++++++++++++++++++++++++++++++++
 2 files changed, 69 insertions(+)
kernel/hrtimer.c +4 -0
diff -urpN linux-2.6/kernel/hrtimer.c linux-2.6-patched/kernel/hrtimer.c
@@ -48,6 +48,7 @@
 
 #include <asm/uaccess.h>
 
+#ifndef CONFIG_GENERIC_TIME
 /**
  * ktime_get - get the monotonic time in ktime_t format
  *
@@ -62,6 +63,7 @@ ktime_t ktime_get(void)
 	return timespec_to_ktime(now);
 }
 EXPORT_SYMBOL_GPL(ktime_get);
+#endif
 
 /**
  * ktime_get_real - get the real (wall-) time in ktime_t format
@@ -106,6 +108,7 @@ DEFINE_PER_CPU(struct hrtimer_cpu_base, 
 	}
 };
 
+#ifndef CONFIG_GENERIC_TIME
 /**
  * ktime_get_ts - get the monotonic clock in timespec format
  * @ts:		pointer to timespec variable
@@ -130,6 +133,7 @@ void ktime_get_ts(struct timespec *ts)
 				ts->tv_nsec + tomono.tv_nsec);
 }
 EXPORT_SYMBOL_GPL(ktime_get_ts);
+#endif
 
 /*
  * Get the coarse grained time at the softirq based on xtime and
kernel/time/timekeeping.c +65 -0

71 lines not on this page · raw message

Signature · 4 lines
blue skies,
   Martin.

"Reality continues to ruin my life." - Calvin.
#8 lore raw

[tip:timers/core] timekeeping: optimized ktime_get[_ts] for GENERIC_TIME=y

Commit-ID:  951ed4d36b77ba9fe1ea08fc3c59d8bb6c9bda32
Gitweb:     http://git.kernel.org/tip/951ed4d36b77ba9fe1ea08fc3c59d8bb6c9bda32
Author:     Martin Schwidefsky <[email protected]>
AuthorDate: Tue, 7 Jul 2009 11:27:28 +0200
Committer:  Thomas Gleixner <[email protected]>
CommitDate: Tue, 7 Jul 2009 12:47:33 +0200

timekeeping: optimized ktime_get[_ts] for GENERIC_TIME=y

The generic ktime_get function defined in kernel/hrtimer.c is suboptimial for GENERIC_TIME=y:

 0)               |  ktime_get() {
 0)               |    ktime_get_ts() {
 0)               |      getnstimeofday() {
 0)               |        read_tod_clock() {
 0)   0.601 us    |        }
 0)   1.938 us    |      }
 0)               |      set_normalized_timespec() {
 0)   0.602 us    |      }
 0)   4.375 us    |    }
 0)   5.523 us    |  }

Overall there are two read_seqbegin/read_seqretry loops and a lot of unnecessary struct timespec calculations. ktime_get returns a nano second value which is the sum of xtime, wall_to_monotonic and the nano second delta from the clock source.

ktime_get can be optimized for GENERIC_TIME=y. The new version only calls clocksource_read:

 0)               |  ktime_get() {
 0)               |    read_tod_clock() {
 0)   0.610 us    |    }
 0)   1.977 us    |  }

It uses a single read_seqbegin/readseqretry loop and just adds everthing to a nano second value.

ktime_get_ts is optimized in a similar fashion.

[ tglx: added WARN_ON(timekeeping_suspended) as in getnstimeofday() ]

LKML-Reference: <20090707112728.3005244d@skybase>

---
 kernel/hrtimer.c          |    4 ++
 kernel/time/timekeeping.c |   69 +++++++++++++++++++++++++++++++++++++++++++++
 2 files changed, 73 insertions(+), 0 deletions(-)
kernel/hrtimer.c +4 -0
index 9002958..829e066 100644
@@ -48,6 +48,7 @@
 
 #include <asm/uaccess.h>
 
+#ifndef CONFIG_GENERIC_TIME
 /**
  * ktime_get - get the monotonic time in ktime_t format
  *
@@ -62,6 +63,7 @@ ktime_t ktime_get(void)
 	return timespec_to_ktime(now);
 }
 EXPORT_SYMBOL_GPL(ktime_get);
+#endif
 
 /**
  * ktime_get_real - get the real (wall-) time in ktime_t format
@@ -106,6 +108,7 @@ DEFINE_PER_CPU(struct hrtimer_cpu_base, hrtimer_bases) =
 	}
 };
 
+#ifndef CONFIG_GENERIC_TIME
 /**
  * ktime_get_ts - get the monotonic clock in timespec format
  * @ts:		pointer to timespec variable
@@ -130,6 +133,7 @@ void ktime_get_ts(struct timespec *ts)
 				ts->tv_nsec + tomono.tv_nsec);
 }
 EXPORT_SYMBOL_GPL(ktime_get_ts);
+#endif
 
 /*
  * Get the coarse grained time at the softirq based on xtime and
kernel/time/timekeeping.c +69 -0

75 lines not on this page · raw message