[PATCH 0/2] Introduce CLOCK_BOOTTIME (v2)

[PATCH 0/2] Introduce CLOCK_BOOTTIME (v2)

[John Stultz]

After some discussions with Jamie Lokier about some of the 
drawbacks of CLOCK_MONOTONIC not incrementing during suspend
(see http://www.spinics.net/lists/linux-fsdevel/msg40272.html),
I wanted to see if we couldn't provide a new clockid that would 
allow applications that wanted to be aware of time passing during
suspend without having to deal with the inconsistencies of 
CLOCK_REALTIME caused by calls to settimeofday.

So this patchset introduces CLOCK_BOOTTIME, which is identical
to CLOCK_MONOTONIC, but includes any time spent in suspend.

This second version of the patch uses a array table for the
clock_id->hrtimer_base convesion instead of the switch.

Thomas: I've not been able to do much comparision at the asm
level for ppc and arm (while I have cross tools for compiling,
I don't have objdump for those arches on my build box) between
this and the previous switch code. But on x86, size breakdown
looks like:

Original:
   text	   data	    bss	    dec	    hex	filename
15517512	1899836	 891200	18308548	1175dc4	vmlinux

Switch:
   text	   data	    bss	    dec	    hex	filename
15517716	1899804	 891200	18308720	1175e70	vmlinux

Table:
   text	   data	    bss	    dec	    hex	filename
15517873	1899868	 891200	18308941	1175f4d	vmlinux

So that's not as great, but remvoing the WARN_ON added in 
hrtimer_clockid_to_base drops it down quite a bit:
Table-nowarn:
   text	   data	    bss	    dec	    hex	filename
15517617	1899804	 891200	18308621	1175e0d	vmlinux

Looking at the asm, the change from the original is limited
(as expected)to: hrtimer_get_res, hrtimer_init, 
__hrtimer_start_range_ns,  where the difference is a few extra
movs, and and hrtimers_init, where there's the extra table
initialization work.

If you have any tips for more specific comprisions please let
me know. Diffing objdump output doesn't always make it super
clear as to what changed.

Jamie: On platforms that don't implement read_persistent_clock,
your issue would still be present, but fixing that is on my list.
Other then that issue, does this seem to address your concern?

Arve: I believe CLOCK_BOOTTIME would be sufficient for what
Android is using as elapsedRealtime() or 
ANDROID_ALARM_ELAPSED_REALTIME. If not please let me know why.

thanks
-john


CC: Jamie Lokier <jamie@shareable.org>
CC: Thomas Gleixner <tglx@linutronix.de>
CC: Alexander Shishkin <virtuoso@slind.org>
CC: Arve Hjnnevg <arve@android.com>

John Stultz (2):
  [RFC] hrtimers: extend hrtimer base code to handle more then 2
    clockids
  [RFC] hrtimers: Add CLOCK_BOOTTIME clockid, hrtimerbase and posix
    interface

 include/linux/hrtimer.h   |    8 ++++-
 include/linux/time.h      |    4 ++
 kernel/hrtimer.c          |   63 +++++++++++++++++++++++++++--------
 kernel/posix-timers.c     |   16 ++++++++-
 kernel/time/timekeeping.c |   79 ++++++++++++++++++++++++++++++++++++++++++++-
 5 files changed, 152 insertions(+), 18 deletions(-)

-- 
1.7.3.2.146.gca209

[PATCH 1/2] [RFC] hrtimers: extend hrtimer base code to handle more then 2 clockids

[John Stultz]

The hrtimer code is written mainly with CLOCK_REALTIME and CLOCK_MONOTONIC
in mind. These are clockids 0 and 1 resepctively. However, if we are
to introduce any new hrtimer bases, using new clockids, we have to skip
the cputimers (clockids 2,3) as well as other clockids that may not impelement
timers.

This patch adds a little bit of indirection between the clockid and
the base, so that we can extend the base by one when we add
a new clockid at number 7 or so.

CC: Jamie Lokier <jamie@shareable.org>
CC: Thomas Gleixner <tglx@linutronix.de>
CC: Alexander Shishkin <virtuoso@slind.org>
CC: Arve Hjønnevåg <arve@android.com>
Signed-off-by: John Stultz <john.stultz@linaro.org>
---
 include/linux/hrtimer.h |    6 ++++-
 kernel/hrtimer.c        |   48 ++++++++++++++++++++++++++++++++++------------
 2 files changed, 40 insertions(+), 14 deletions(-)

diff --git a/include/linux/hrtimer.h b/include/linux/hrtimer.h
index 330586f..ebf6838 100644
--- a/include/linux/hrtimer.h
+++ b/include/linux/hrtimer.h
@@ -149,7 +149,11 @@ struct hrtimer_clock_base {
 #endif
 };
 
-#define HRTIMER_MAX_CLOCK_BASES 2
+enum  hrtimer_base_type {
+	HRTIMER_BASE_REALTIME,
+	HRTIMER_BASE_MONOTONIC,
+	HRTIMER_MAX_CLOCK_BASES,
+};
 
 /*
  * struct hrtimer_cpu_base - the per cpu clock bases
diff --git a/kernel/hrtimer.c b/kernel/hrtimer.c
index 7a7a206..d5c7091 100644
--- a/kernel/hrtimer.c
+++ b/kernel/hrtimer.c
@@ -53,11 +53,10 @@
 /*
  * The timer bases:
  *
- * Note: If we want to add new timer bases, we have to skip the two
- * clock ids captured by the cpu-timers. We do this by holding empty
- * entries rather than doing math adjustment of the clock ids.
- * This ensures that we capture erroneous accesses to these clock ids
- * rather than moving them into the range of valid clock id's.
+ * There are more clockids then hrtimer bases. Thus, we index
+ * into the timer bases by the hrtimer_base_type enum. When trying
+ * to reach a base using a clockid, hrtimer_clockid_to_base()
+ * is used to convert from clockid to the proper hrtimer_base_type.
  */
 DEFINE_PER_CPU(struct hrtimer_cpu_base, hrtimer_bases) =
 {
@@ -77,6 +76,17 @@ DEFINE_PER_CPU(struct hrtimer_cpu_base, hrtimer_bases) =
 	}
 };
 
+static int hrtimer_clock_to_base_table[MAX_CLOCKS];
+
+static inline int hrtimer_clockid_to_base(clockid_t clock_id)
+{
+	int ret = hrtimer_clock_to_base_table[clock_id];
+
+	WARN_ON(ret == -1);
+	return ret;
+}
+
+
 /*
  * Get the coarse grained time at the softirq based on xtime and
  * wall_to_monotonic.
@@ -95,8 +105,8 @@ static void hrtimer_get_softirq_time(struct hrtimer_cpu_base *base)
 
 	xtim = timespec_to_ktime(xts);
 	tomono = timespec_to_ktime(tom);
-	base->clock_base[CLOCK_REALTIME].softirq_time = xtim;
-	base->clock_base[CLOCK_MONOTONIC].softirq_time =
+	base->clock_base[HRTIMER_BASE_REALTIME].softirq_time = xtim;
+	base->clock_base[HRTIMER_BASE_MONOTONIC].softirq_time =
 		ktime_add(xtim, tomono);
 }
 
@@ -184,10 +194,11 @@ switch_hrtimer_base(struct hrtimer *timer, struct hrtimer_clock_base *base,
 	struct hrtimer_cpu_base *new_cpu_base;
 	int this_cpu = smp_processor_id();
 	int cpu = hrtimer_get_target(this_cpu, pinned);
+	int basenum = hrtimer_clockid_to_base(base->index);
 
 again:
 	new_cpu_base = &per_cpu(hrtimer_bases, cpu);
-	new_base = &new_cpu_base->clock_base[base->index];
+	new_base = &new_cpu_base->clock_base[basenum];
 
 	if (base != new_base) {
 		/*
@@ -627,7 +638,7 @@ static void retrigger_next_event(void *arg)
 
 	/* Adjust CLOCK_REALTIME offset */
 	raw_spin_lock(&base->lock);
-	base->clock_base[CLOCK_REALTIME].offset =
+	base->clock_base[HRTIMER_BASE_REALTIME].offset =
 		timespec_to_ktime(realtime_offset);
 
 	hrtimer_force_reprogram(base, 0);
@@ -725,8 +736,8 @@ static int hrtimer_switch_to_hres(void)
 		return 0;
 	}
 	base->hres_active = 1;
-	base->clock_base[CLOCK_REALTIME].resolution = KTIME_HIGH_RES;
-	base->clock_base[CLOCK_MONOTONIC].resolution = KTIME_HIGH_RES;
+	base->clock_base[HRTIMER_BASE_REALTIME].resolution = KTIME_HIGH_RES;
+	base->clock_base[HRTIMER_BASE_MONOTONIC].resolution = KTIME_HIGH_RES;
 
 	tick_setup_sched_timer();
 
@@ -1116,6 +1127,7 @@ static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
 			   enum hrtimer_mode mode)
 {
 	struct hrtimer_cpu_base *cpu_base;
+	int base;
 
 	memset(timer, 0, sizeof(struct hrtimer));
 
@@ -1124,7 +1136,8 @@ static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
 	if (clock_id == CLOCK_REALTIME && mode != HRTIMER_MODE_ABS)
 		clock_id = CLOCK_MONOTONIC;
 
-	timer->base = &cpu_base->clock_base[clock_id];
+	base = hrtimer_clockid_to_base(clock_id);
+	timer->base = &cpu_base->clock_base[base];
 	hrtimer_init_timer_hres(timer);
 	timerqueue_init(&timer->node);
 
@@ -1160,9 +1173,10 @@ EXPORT_SYMBOL_GPL(hrtimer_init);
 int hrtimer_get_res(const clockid_t which_clock, struct timespec *tp)
 {
 	struct hrtimer_cpu_base *cpu_base;
+	int base = hrtimer_clockid_to_base(which_clock);
 
 	cpu_base = &__raw_get_cpu_var(hrtimer_bases);
-	*tp = ktime_to_timespec(cpu_base->clock_base[which_clock].resolution);
+	*tp = ktime_to_timespec(cpu_base->clock_base[base].resolution);
 
 	return 0;
 }
@@ -1706,6 +1720,14 @@ static struct notifier_block __cpuinitdata hrtimers_nb = {
 
 void __init hrtimers_init(void)
 {
+	int i;
+
+	/* Init the clockid -> base mapping */
+	for (i = 0; i < MAX_CLOCKS; i++)
+		hrtimer_clock_to_base_table[i] = -1;
+	hrtimer_clock_to_base_table[CLOCK_REALTIME] = HRTIMER_BASE_REALTIME;
+	hrtimer_clock_to_base_table[CLOCK_MONOTONIC] = HRTIMER_BASE_MONOTONIC;
+
 	hrtimer_cpu_notify(&hrtimers_nb, (unsigned long)CPU_UP_PREPARE,
 			  (void *)(long)smp_processor_id());
 	register_cpu_notifier(&hrtimers_nb);
-- 
1.7.3.2.146.gca209

[PATCH 2/2] [RFC] hrtimers: Add CLOCK_BOOTTIME clockid, hrtimerbase and posix interface

[John Stultz]

CLOCK_MONOTONIC stops while the system is in suspend. This is because
to applications system suspend is invisible. However, there is a
growing set of applications that are wanting to be suspend-aware,
but do not want to deal with the complicatoins of CLOCK_REALTIME
(which might jump around if settimeofday is called).

For these applications, I propose a new clockid: CLOCK_BOOTTIME.
CLOCK_BOOTTIME is idential to CLOCK_MONOTONIC, except it also
includes any time spent in suspend.

This patch adds the new CLOCK_BOOTTIME clockid, as well as the
infrastructure needed to support hrtimers against it, and the
wiring to expose it out via the posix interface.

CC: Jamie Lokier <jamie@shareable.org>
CC: Thomas Gleixner <tglx@linutronix.de>
CC: Alexander Shishkin <virtuoso@slind.org>
CC: Arve Hjønnevåg <arve@android.com>
Signed-off-by: John Stultz <john.stultz@linaro.org>
---
 include/linux/hrtimer.h   |    2 +
 include/linux/time.h      |    4 ++
 kernel/hrtimer.c          |   15 +++++++-
 kernel/posix-timers.c     |   16 ++++++++-
 kernel/time/timekeeping.c |   79 ++++++++++++++++++++++++++++++++++++++++++++-
 5 files changed, 112 insertions(+), 4 deletions(-)

diff --git a/include/linux/hrtimer.h b/include/linux/hrtimer.h
index ebf6838..b75e93f 100644
--- a/include/linux/hrtimer.h
+++ b/include/linux/hrtimer.h
@@ -152,6 +152,7 @@ struct hrtimer_clock_base {
 enum  hrtimer_base_type {
 	HRTIMER_BASE_REALTIME,
 	HRTIMER_BASE_MONOTONIC,
+	HRTIMER_BASE_BOOTTIME,
 	HRTIMER_MAX_CLOCK_BASES,
 };
 
@@ -313,6 +314,7 @@ static inline int hrtimer_is_hres_active(struct hrtimer *timer)
 
 extern ktime_t ktime_get(void);
 extern ktime_t ktime_get_real(void);
+extern ktime_t ktime_get_boottime(void);
 
 
 DECLARE_PER_CPU(struct tick_device, tick_cpu_device);
diff --git a/include/linux/time.h b/include/linux/time.h
index 9f15ac7..4ed031a 100644
--- a/include/linux/time.h
+++ b/include/linux/time.h
@@ -126,6 +126,8 @@ unsigned long get_seconds(void);
 struct timespec current_kernel_time(void);
 struct timespec __current_kernel_time(void); /* does not take xtime_lock */
 struct timespec __get_wall_to_monotonic(void); /* does not take xtime_lock */
+extern struct timespec __get_sleep_time(void); /* does not take xtime_lock */
+
 struct timespec get_monotonic_coarse(void);
 
 #define CURRENT_TIME		(current_kernel_time())
@@ -160,6 +162,7 @@ extern void getnstimeofday(struct timespec *tv);
 extern void getrawmonotonic(struct timespec *ts);
 extern void getboottime(struct timespec *ts);
 extern void monotonic_to_bootbased(struct timespec *ts);
+extern void get_monotonic_boottime(struct timespec *ts);
 
 extern struct timespec timespec_trunc(struct timespec t, unsigned gran);
 extern int timekeeping_valid_for_hres(void);
@@ -290,6 +293,7 @@ struct itimerval {
 #define CLOCK_MONOTONIC_RAW		4
 #define CLOCK_REALTIME_COARSE		5
 #define CLOCK_MONOTONIC_COARSE		6
+#define CLOCK_BOOTTIME			7
 
 /*
  * The IDs of various hardware clocks:
diff --git a/kernel/hrtimer.c b/kernel/hrtimer.c
index d5c7091..3179657 100644
--- a/kernel/hrtimer.c
+++ b/kernel/hrtimer.c
@@ -73,6 +73,11 @@ DEFINE_PER_CPU(struct hrtimer_cpu_base, hrtimer_bases) =
 			.get_time = &ktime_get,
 			.resolution = KTIME_LOW_RES,
 		},
+		{
+			.index = CLOCK_BOOTTIME,
+			.get_time = &ktime_get_boottime,
+			.resolution = KTIME_LOW_RES,
+		},
 	}
 };
 
@@ -93,21 +98,26 @@ static inline int hrtimer_clockid_to_base(clockid_t clock_id)
  */
 static void hrtimer_get_softirq_time(struct hrtimer_cpu_base *base)
 {
-	ktime_t xtim, tomono;
-	struct timespec xts, tom;
+	ktime_t xtim, tomono, sleep;
+	struct timespec xts, tom, slp;
 	unsigned long seq;
 
 	do {
 		seq = read_seqbegin(&xtime_lock);
 		xts = __current_kernel_time();
 		tom = __get_wall_to_monotonic();
+		slp = __get_sleep_time();
 	} while (read_seqretry(&xtime_lock, seq));
 
 	xtim = timespec_to_ktime(xts);
 	tomono = timespec_to_ktime(tom);
+	sleep = timespec_to_ktime(slp);
 	base->clock_base[HRTIMER_BASE_REALTIME].softirq_time = xtim;
 	base->clock_base[HRTIMER_BASE_MONOTONIC].softirq_time =
 		ktime_add(xtim, tomono);
+	base->clock_base[HRTIMER_BASE_BOOTTIME].softirq_time =
+		ktime_add(base->clock_base[HRTIMER_BASE_MONOTONIC].softirq_time,
+				sleep);
 }
 
 /*
@@ -1727,6 +1737,7 @@ void __init hrtimers_init(void)
 		hrtimer_clock_to_base_table[i] = -1;
 	hrtimer_clock_to_base_table[CLOCK_REALTIME] = HRTIMER_BASE_REALTIME;
 	hrtimer_clock_to_base_table[CLOCK_MONOTONIC] = HRTIMER_BASE_MONOTONIC;
+	hrtimer_clock_to_base_table[CLOCK_BOOTTIME] = HRTIMER_BASE_BOOTTIME;
 
 	hrtimer_cpu_notify(&hrtimers_nb, (unsigned long)CPU_UP_PREPARE,
 			  (void *)(long)smp_processor_id());
diff --git a/kernel/posix-timers.c b/kernel/posix-timers.c
index 93bd2eb..389d5d5 100644
--- a/kernel/posix-timers.c
+++ b/kernel/posix-timers.c
@@ -240,7 +240,7 @@ static int posix_ktime_get_ts(clockid_t which_clock, struct timespec *tp)
 }
 
 /*
- * Get monotonic time for posix timers
+ * Get monotonic-raw time for posix timers
  */
 static int posix_get_monotonic_raw(clockid_t which_clock, struct timespec *tp)
 {
@@ -267,6 +267,14 @@ static int posix_get_coarse_res(const clockid_t which_clock, struct timespec *tp
 	*tp = ktime_to_timespec(KTIME_LOW_RES);
 	return 0;
 }
+
+static int posix_get_boottime(const clockid_t which_clock, struct timespec *tp)
+{
+	get_monotonic_boottime(tp);
+	return 0;
+}
+
+
 /*
  * Initialize everything, well, just everything in Posix clocks/timers ;)
  */
@@ -301,12 +309,18 @@ static __init int init_posix_timers(void)
 		.timer_create = no_timer_create,
 		.nsleep = no_nsleep,
 	};
+	struct k_clock clock_boottime = {
+		.clock_getres = hrtimer_get_res,
+		.clock_get = posix_get_boottime,
+		.clock_set = do_posix_clock_nosettime,
+	};
 
 	register_posix_clock(CLOCK_REALTIME, &clock_realtime);
 	register_posix_clock(CLOCK_MONOTONIC, &clock_monotonic);
 	register_posix_clock(CLOCK_MONOTONIC_RAW, &clock_monotonic_raw);
 	register_posix_clock(CLOCK_REALTIME_COARSE, &clock_realtime_coarse);
 	register_posix_clock(CLOCK_MONOTONIC_COARSE, &clock_monotonic_coarse);
+	register_posix_clock(CLOCK_BOOTTIME, &clock_boottime);
 
 	posix_timers_cache = kmem_cache_create("posix_timers_cache",
 					sizeof (struct k_itimer), 0, SLAB_PANIC,
diff --git a/kernel/time/timekeeping.c b/kernel/time/timekeeping.c
index 5bb86da..e12ec9f 100644
--- a/kernel/time/timekeeping.c
+++ b/kernel/time/timekeeping.c
@@ -828,7 +828,7 @@ void update_wall_time(void)
  * getboottime - Return the real time of system boot.
  * @ts:		pointer to the timespec to be set
  *
- * Returns the time of day in a timespec.
+ * Returns the wall-time of boot in a timespec.
  *
  * This is based on the wall_to_monotonic offset and the total suspend
  * time. Calls to settimeofday will affect the value returned (which
@@ -846,6 +846,83 @@ void getboottime(struct timespec *ts)
 }
 EXPORT_SYMBOL_GPL(getboottime);
 
+
+/**
+ * get_monotonic_boottime - Returns monotonic time since boot
+ * @ts:		pointer to the timespec to be set
+ *
+ * Returns the monotonic time since boot in a timespec.
+ *
+ * This is similar to CLOCK_MONTONIC/ktime_get_ts, but also
+ * includes the time spent in suspend.
+ */
+void get_monotonic_boottime(struct timespec *ts)
+{
+	struct timespec tomono, sleep;
+	unsigned int seq;
+	s64 nsecs;
+
+	WARN_ON(timekeeping_suspended);
+
+	do {
+		seq = read_seqbegin(&xtime_lock);
+		*ts = xtime;
+		tomono = wall_to_monotonic;
+		sleep = total_sleep_time;
+		nsecs = timekeeping_get_ns();
+
+	} while (read_seqretry(&xtime_lock, seq));
+
+	set_normalized_timespec(ts, ts->tv_sec + tomono.tv_sec + sleep.tv_sec,
+			ts->tv_nsec + tomono.tv_nsec + sleep.tv_nsec + nsecs);
+}
+EXPORT_SYMBOL_GPL(get_monotonic_boottime);
+
+/**
+ * ktime_get_boottime - Returns monotonic time since boot in a ktime
+ *
+ * Returns the monotonic time since boot in a ktime
+ *
+ * This is similar to CLOCK_MONTONIC/ktime_get, but also
+ * includes the time spent in suspend.
+ */
+ktime_t ktime_get_boottime(void)
+{
+	unsigned int seq;
+	s64 secs, nsecs;
+
+	WARN_ON(timekeeping_suspended);
+
+	do {
+		seq = read_seqbegin(&xtime_lock);
+		secs = xtime.tv_sec;
+		secs += wall_to_monotonic.tv_sec;
+		secs += total_sleep_time.tv_sec;
+		nsecs = xtime.tv_nsec;
+		nsecs += wall_to_monotonic.tv_nsec;
+		nsecs += total_sleep_time.tv_nsec;
+		nsecs += timekeeping_get_ns();
+
+	} while (read_seqretry(&xtime_lock, seq));
+	/*
+	 * Use ktime_set/ktime_add_ns to create a proper ktime on
+	 * 32-bit architectures without CONFIG_KTIME_SCALAR.
+	 */
+	return ktime_add_ns(ktime_set(secs, 0), nsecs);
+}
+EXPORT_SYMBOL_GPL(ktime_get_boottime);
+
+/**
+ * __get_sleep_time - returns total_sleep_time
+ *
+ * Returns total time spent in suspend.
+ * Requires the xtime lock be held
+ */
+struct timespec __get_sleep_time(void)
+{
+	return total_sleep_time;
+}
+
 /**
  * monotonic_to_bootbased - Convert the monotonic time to boot based.
  * @ts:		pointer to the timespec to be converted
-- 
1.7.3.2.146.gca209