Skip to contentSkip to conversation

Timekeeping changes for 4.10 (take 2)

GIT PULL In review John Stultz 8 participants 29 messages Nov 28, 2016 – Dec 1, 2016

Subsystems (12)
Documentation x86 Kselftest Tracing +8 more subsystems
lore
Share & tools
lore
#0 lore raw

Hey Thomas, Ingo,
Just wanted to resend my patch queue for 4.10, including changes
from the last submission suggested by Ingo and some additional changes from Joel to add extra a boot trace clock.

Let me know if you have any comments or objections.

thanks
-john

Cc Richard Cochran <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc Prarit Bhargava <[email protected]>
Cc Thomas Gleixner <[email protected]>

The following changes since commit a909d3e636995ba7c349e2ca5dbb528154d4ac30:

Linux 4.9-rc3 (2016-10-29 13:52:02 -0700)

are available in the git repository at:

https://git.linaro.org/people/john.stultz/linux.git fortglx/4.10/time

for you to fetch changes up to 6ade3fa56aa5a0a3339caa8c1c268d19982101e3:

trace: Update documentation for mono, mono_raw and boot clock (2016-11-28 14:28:39 -0800)

----------------------------------------------------------------
Baolin Wang (1):
time: alarmtimer: Add the tracepoints for alarmtimer

Chen Yu (1):
timekeeping: Ignore the bogus sleep time if pm_trace is enabled

Chris Metcalf (1):
timekeeping: clocksource_cyc2ns: Document intended range limitation

Colin Ian King (1):
selftests/timers: Fix spelling mistake "Asyncrhonous" ->
"Asynchronous"

Joel Fernandes (3):
timekeeping: Add a fast and NMI safe boot clock
trace: Add an option for boot clock as trace clock
trace: Update documentation for mono, mono_raw and boot clock

 Documentation/trace/ftrace.txt                    | 20 +++++
 arch/x86/kernel/rtc.c                             |  9 +++
 drivers/base/power/trace.c                        | 26 +++++++
 drivers/rtc/rtc-cmos.c                            |  7 ++
 include/linux/clocksource.h                       |  5 +-
 include/linux/mc146818rtc.h                       |  1 +
 include/linux/pm-trace.h                          |  9 ++-
 include/linux/timekeeping.h                       |  1 +
 include/trace/events/alarmtimer.h                 | 92 +++++++++++++++++++++++
 kernel/time/alarmtimer.c                          | 16 +++-
 kernel/time/timekeeping.c                         | 29 +++++++
 kernel/trace/trace.c                              |  1 +
 tools/testing/selftests/timers/skew_consistency.c |  2 +-
 13 files changed, 213 insertions(+), 5 deletions(-)
 create mode 100644 include/trace/events/alarmtimer.h
Signature · 1 line
2.7.4
#1 lore raw

PATCH 1/7 selftests/timers: Fix spelling mistake "Asyncrhonous" -> "Asynchronous"

From: Colin Ian King <[email protected]>

Trivial fix to spelling mistake

Cc Thomas Gleixner <[email protected]>
Cc Richard Cochran <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc Shuah Khan <[email protected]>
Cc Prarit Bhargava <[email protected]>
---
 tools/testing/selftests/timers/skew_consistency.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
tools/testing/selftests/timers/skew_consistency.c +1 -1
index 5562f84..2a996e0 100644
@@ -57,7 +57,7 @@ int main(int argv, char **argc)
 	pid_t pid;
 
 
-	printf("Running Asyncrhonous Frequency Changing Tests...\n");
+	printf("Running Asynchronous Frequency Changing Tests...\n");
 
 	pid = fork();
 	if (!pid)
Signature · 1 line
2.7.4
#2 lore raw

[tip:timers/core] selftests/timers: Fix spelling mistake "Asyncrhonous" -> "Asynchronous"

selftests/timers: Fix spelling mistake "Asyncrhonous" -> "Asynchronous"
Commit-ID:  a109ded26c8c85c17872c8f61e8979c5d650004d
Gitweb:     http://git.kernel.org/tip/a109ded26c8c85c17872c8f61e8979c5d650004d
Author:     Colin Ian King <[email protected]>
AuthorDate: Mon, 28 Nov 2016 14:35:18 -0800
Committer:  Thomas Gleixner <[email protected]>
CommitDate: Tue, 29 Nov 2016 18:02:57 +0100

selftests/timers: Fix spelling mistake "Asyncrhonous" -> "Asynchronous"

Trivial fix to spelling mistake

---
 tools/testing/selftests/timers/skew_consistency.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
tools/testing/selftests/timers/skew_consistency.c +1 -1
index 5562f84..2a996e0 100644
@@ -57,7 +57,7 @@ int main(int argv, char **argc)
 	pid_t pid;
 
 
-	printf("Running Asyncrhonous Frequency Changing Tests...\n");
+	printf("Running Asynchronous Frequency Changing Tests...\n");
 
 	pid = fork();
 	if (!pid)
#3 lore raw

PATCH 2/7 timekeeping: Ignore the bogus sleep time if pm_trace is enabled

In reply to John Stultz

From: Chen Yu <[email protected]>

Previously we encountered some memory overflow issues due to the bogus sleep time brought by inconsistent rtc, which is triggered when pm_trace is enabled, and we have fixed it in recent kernel. However it's improper in the first place to call __timekeeping_inject_sleeptime() in case that pm_trace is enabled simply because that "hash" time value will wreckage the timekeeping subsystem.

This patch is originally written by Thomas, which would bypass the bogus rtc interval when pm_trace is enabled. Meanwhile, if system succeed to resume back with pm_trace set, the users are warned to adjust the bogus rtc either by 'ntpdate' or 'rdate', by resetting pm_trace_rtc_abused to false, otherwise above tools might not work as expected.

Originally-from: Thomas Gleixner <[email protected]>

Cc "Rafael J. Wysocki" <[email protected]>
Cc John Stultz <[email protected]>
Cc Xunlei Pang <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc Len Brown <[email protected]>
Cc "H. Peter Anvin" <[email protected]>
Cc Pavel Machek <[email protected]>
Cc Thomas Gleixner <[email protected]>
Cc Prarit Bhargava <[email protected]>
Cc Richard Cochran <[email protected]>

[jstultz: Fixed up pr_warn formating, and implemented suggestions
from Ingo]

---
v2: Added suggestions from Ingo

---
 arch/x86/kernel/rtc.c       |  9 +++++++++
 drivers/base/power/trace.c  | 26 ++++++++++++++++++++++++++
 drivers/rtc/rtc-cmos.c      |  7 +++++++
 include/linux/mc146818rtc.h |  1 +
 include/linux/pm-trace.h    |  9 ++++++++-
 5 files changed, 51 insertions(+), 1 deletion(-)
arch/x86/kernel/rtc.c +9 -0
index 79c6311c..3282a88 100644
@@ -64,6 +64,15 @@ void mach_get_cmos_time(struct timespec *now)
 	unsigned int status, year, mon, day, hour, min, sec, century = 0;
 	unsigned long flags;
 
+	/*
+	 * If pm_trace abused the RTC as storage set the timespec to 0
+	 * which tells the caller that this RTC value is bogus.
+	 */
+	if (!pm_trace_rtc_valid()) {
+		now->tv_sec = now->tv_nsec = 0;
+		return;
+	}
+
 	spin_lock_irqsave(&rtc_lock, flags);
 
 	/*
drivers/base/power/trace.c +26 -0
index efec10b..15563c4 100644
@@ -10,6 +10,7 @@
 #include <linux/pm-trace.h>
 #include <linux/export.h>
 #include <linux/rtc.h>
+#include <linux/suspend.h>
 
 #include <linux/mc146818rtc.h>
 
@@ -74,6 +75,9 @@
 
 #define DEVSEED (7919)
 
+bool pm_trace_rtc_abused __read_mostly;
+EXPORT_SYMBOL(pm_trace_rtc_abused);
+
 static unsigned int dev_hash_value;
 
 static int set_magic_time(unsigned int user, unsigned int file, unsigned int device)
@@ -104,6 +108,7 @@ static int set_magic_time(unsigned int user, unsigned int file, unsigned int dev
 	time.tm_min = (n % 20) * 3;
 	n /= 20;
 	mc146818_set_time(&time);
+	pm_trace_rtc_abused = true;
 	return n ? -1 : 0;
 }
 
@@ -238,10 +243,31 @@ int show_trace_dev_match(char *buf, size_t size)
 	device_pm_unlock();
 	return ret;
 }
+static int pm_trace_notify(struct notifier_block *nb,
+				unsigned long mode, void *_unused)
+{
+	switch (mode) {
+	case PM_POST_HIBERNATION:
+	case PM_POST_SUSPEND:
+		if (pm_trace_rtc_abused) {
+			pm_trace_rtc_abused = false;
+			pr_warn("Possible incorrect RTC due to pm_trace, please use 'ntpdate' or 'rdate' to reset.\n");
+		}
+		break;
+	default:
+		break;
+	}
+	return 0;
+}
+
+static struct notifier_block pm_trace_nb = {
+	.notifier_call = pm_trace_notify,
+};
 
 static int early_resume_init(void)
 {
 	hash_value_early_read = read_magic_time();
+	register_pm_notifier(&pm_trace_nb);
 	return 0;
 }
 
drivers/rtc/rtc-cmos.c +7 -0
index dd3d598..9cf06b7 100644
@@ -191,6 +191,13 @@ static inline void cmos_write_bank2(unsigned char val, unsigned char addr)
 
 static int cmos_read_time(struct device *dev, struct rtc_time *t)
 {
+	/*
+	 * If pm_trace abused the RTC for storage tell the caller that it is
+	 * unusable.
+	 */
+	if (!pm_trace_rtc_valid())
+		return -EIO;
+
 	/* REVISIT:  if the clock has a "century" register, use
 	 * that instead of the heuristic in mc146818_get_time().
 	 * That'll make Y3K compatility (year > 2070) easy!
include/linux/mc146818rtc.h +1 -0
index a585b4b..0661af1 100644
@@ -16,6 +16,7 @@
 #include <asm/mc146818rtc.h>		/* register access macros */
 #include <linux/bcd.h>
 #include <linux/delay.h>
+#include <linux/pm-trace.h>
 
 #ifdef __KERNEL__
 #include <linux/spinlock.h>		/* spinlock_t */
include/linux/pm-trace.h +8 -1
index ecbde7a..7b78793 100644
@@ -1,11 +1,17 @@
 #ifndef PM_TRACE_H
 #define PM_TRACE_H
 
+#include <linux/types.h>
 #ifdef CONFIG_PM_TRACE
 #include <asm/pm-trace.h>
-#include <linux/types.h>
 
 extern int pm_trace_enabled;
+extern bool pm_trace_rtc_abused;
+
+static inline bool pm_trace_rtc_valid(void)
+{
+	return !pm_trace_rtc_abused;
+}
 
 static inline int pm_trace_is_enabled(void)
 {
@@ -24,6 +30,7 @@ extern int show_trace_dev_match(char *buf, size_t size);
 
 #else
 
+static inline bool pm_trace_rtc_valid(void) { return true; }
 static inline int pm_trace_is_enabled(void) { return 0; }
 
 #define TRACE_DEVICE(dev) do { } while (0)
Signature · 1 line
2.7.4
#4 lore raw

Re: [PATCH 2/7] timekeeping: Ignore the bogus sleep time if pm_trace is enabled

Quoted 9 lines · John Stultz

From: Chen Yu <[email protected]>

Previously we encountered some memory overflow issues due to the bogus sleep time brought by inconsistent rtc, which is triggered when pm_trace is enabled, and we have fixed it in recent kernel. However it's improper in the first place to call __timekeeping_inject_sleeptime() in case that pm_trace is enabled simply because that "hash" time value will wreckage the timekeeping subsystem.

s/

Previously we encountered memory overflow issues due to bogus sleep time brought by an inconsistent RTC, which is triggered when pm_trace is enabled, and we have fixed it in recent kernels. However it's improper in the first place to call __timekeeping_inject_sleeptime() in case pm_trace is enabled simply because the "hash" time value will wreckage the timekeeping subsystem.

Half a dozen typos ...

Quoted 6 lines

This patch is originally written by Thomas, which would bypass the bogus rtc interval when pm_trace is enabled. Meanwhile, if system succeed to resume back with pm_trace set, the users are warned to adjust the bogus rtc either by 'ntpdate' or 'rdate', by resetting pm_trace_rtc_abused to false, otherwise above tools might not work as expected.

s/

This patch was originally written by Thomas, which would bypass the bogus RTC interval when pm_trace is enabled. Meanwhile, if the system succeeds to resume back with pm_trace set, users are warned to adjust the bogus RTC either by 'ntpdate' or 'rdate', by resetting pm_trace_rtc_abused to false, otherwise above tools might not work as expected.

arch/x86/kernel/rtc.c void mach_get_cmos_time(struct timespec *now) · line 70

Quoted 4 lines
arch/x86/kernel/rtc.c +4 -0
+	/*
+	 * If pm_trace abused the RTC as storage set the timespec to 0
+	 * which tells the caller that this RTC value is bogus.
+	 */

s/

	/*
	 * If pm_trace abused the RTC as storage, set the timespec to 0,
	 * which tells the caller that this RTC value is bogus.
	 */

drivers/base/power/trace.c line 79

Quoted 6 lines
drivers/base/power/trace.c +2 -0
@@ -74,6 +75,9 @@
 
 #define DEVSEED (7919)
 
+bool pm_trace_rtc_abused __read_mostly;
+EXPORT_SYMBOL(pm_trace_rtc_abused);

EXPORT_SYMBOL_GPL()

drivers/base/power/trace.c int show_trace_dev_match(char *buf, size_t size) · line 247

Quoted 2 lines
drivers/base/power/trace.c +2 -0
+static int pm_trace_notify(struct notifier_block *nb,
+				unsigned long mode, void *_unused)

Please no nonsensical linebreaks in the middle of an argument list.

drivers/base/power/trace.c int show_trace_dev_match(char *buf, size_t size) · line 254

Quoted 7 lines
drivers/base/power/trace.c +7 -0
+{
+	switch (mode) {
+	case PM_POST_HIBERNATION:
+	case PM_POST_SUSPEND:
+		if (pm_trace_rtc_abused) {
+			pm_trace_rtc_abused = false;
+			pr_warn("Possible incorrect RTC due to pm_trace, please use 'ntpdate' or 'rdate' to reset.\n");

s/to reset./to reset it.

Thanks,

Ingo

#5 lore raw

Re: [PATCH 2/7] timekeeping: Ignore the bogus sleep time if pm_trace is enabled

Quoted 77 lines · Ingo Molnar
Quoted 9 lines · John Stultz

From: Chen Yu <[email protected]>

Previously we encountered some memory overflow issues due to the bogus sleep time brought by inconsistent rtc, which is triggered when pm_trace is enabled, and we have fixed it in recent kernel. However it's improper in the first place to call __timekeeping_inject_sleeptime() in case that pm_trace is enabled simply because that "hash" time value will wreckage the timekeeping subsystem.

s/

Previously we encountered memory overflow issues due to bogus sleep time brought by an inconsistent RTC, which is triggered when pm_trace is enabled, and we have fixed it in recent kernels. However it's improper in the first place to call __timekeeping_inject_sleeptime() in case pm_trace is enabled simply because the "hash" time value will wreckage the timekeeping subsystem.

Half a dozen typos ...

Quoted 6 lines

This patch is originally written by Thomas, which would bypass the bogus rtc interval when pm_trace is enabled. Meanwhile, if system succeed to resume back with pm_trace set, the users are warned to adjust the bogus rtc either by 'ntpdate' or 'rdate', by resetting pm_trace_rtc_abused to false, otherwise above tools might not work as expected.

s/

This patch was originally written by Thomas, which would bypass the bogus RTC interval when pm_trace is enabled. Meanwhile, if the system succeeds to resume back with pm_trace set, users are warned to adjust the bogus RTC either by 'ntpdate' or 'rdate', by resetting pm_trace_rtc_abused to false, otherwise above tools might not work as expected.

arch/x86/kernel/rtc.c void mach_get_cmos_time(struct timespec *now) · line 70

Quoted 4 lines
arch/x86/kernel/rtc.c +4 -0
+	/*
+	 * If pm_trace abused the RTC as storage set the timespec to 0
+	 * which tells the caller that this RTC value is bogus.
+	 */

s/

	/*
	 * If pm_trace abused the RTC as storage, set the timespec to 0,
	 * which tells the caller that this RTC value is bogus.
	 */

drivers/base/power/trace.c line 79

Quoted 6 lines
drivers/base/power/trace.c +2 -0
@@ -74,6 +75,9 @@
 
 #define DEVSEED (7919)
 
+bool pm_trace_rtc_abused __read_mostly;
+EXPORT_SYMBOL(pm_trace_rtc_abused);

EXPORT_SYMBOL_GPL()

drivers/base/power/trace.c int show_trace_dev_match(char *buf, size_t size) · line 247

Quoted 2 lines
drivers/base/power/trace.c +2 -0
+static int pm_trace_notify(struct notifier_block *nb,
+				unsigned long mode, void *_unused)

Please no nonsensical linebreaks in the middle of an argument list.

drivers/base/power/trace.c int show_trace_dev_match(char *buf, size_t size) · line 254

Quoted 7 lines
drivers/base/power/trace.c +7 -0
+{
+	switch (mode) {
+	case PM_POST_HIBERNATION:
+	case PM_POST_SUSPEND:
+		if (pm_trace_rtc_abused) {
+			pm_trace_rtc_abused = false;
+			pr_warn("Possible incorrect RTC due to pm_trace, please use 'ntpdate' or 'rdate' to reset.\n");

s/to reset./to reset it.

Quoted 3 lines

Thanks,

Ingo

Thanks Ingo, I've sent out a new versin based on your comments. Yu

#6 lore raw

[tip:timers/core] timekeeping: Ignore the bogus sleep time if pm_trace is enabled

In reply to John Stultz
timekeeping: Ignore the bogus sleep time if pm_trace is enabled
Commit-ID:  ba58d1020a54933c6b087a3107661c8513556cb8
Gitweb:     http://git.kernel.org/tip/ba58d1020a54933c6b087a3107661c8513556cb8
Author:     Chen Yu <[email protected]>
AuthorDate: Mon, 28 Nov 2016 14:35:19 -0800
Committer:  Thomas Gleixner <[email protected]>
CommitDate: Tue, 29 Nov 2016 18:02:58 +0100

timekeeping: Ignore the bogus sleep time if pm_trace is enabled

Power management suspend/resume tracing (ab)uses the RTC to store suspend/resume information persistently. As a consequence the RTC value is clobbered when timekeeping is resumed and tries to inject the sleep time.

Commit a4f8f6667f09 ("timekeeping: Cap array access in timekeeping_debug") plugged a out of bounds array access in the timekeeping debug code which was caused by the clobbered RTC value, but we still use the clobbered RTC value for sleep time injection into kernel timekeeping, which will result in random adjustments depending on the stored "hash" value.

To prevent this keep track of the RTC clobbering and ignore the invalid RTC timestamp at resume. If the system resumed successfully clear the flag, which marks the RTC as unusable, warn the user about the RTC clobber and recommend to adjust the RTC with 'ntpdate' or 'rdate'.

[jstultz: Fixed up pr_warn formating, and implemented suggestions from Ingo] [ tglx: Rewrote changelog ]

Originally-from: Thomas Gleixner <[email protected]>

---
 arch/x86/kernel/rtc.c       |  9 +++++++++
 drivers/base/power/trace.c  | 27 +++++++++++++++++++++++++++
 drivers/rtc/rtc-cmos.c      |  7 +++++++
 include/linux/mc146818rtc.h |  1 +
 include/linux/pm-trace.h    |  9 ++++++++-
 5 files changed, 52 insertions(+), 1 deletion(-)
arch/x86/kernel/rtc.c +9 -0
index 79c6311c..5b21cb7 100644
@@ -64,6 +64,15 @@ void mach_get_cmos_time(struct timespec *now)
 	unsigned int status, year, mon, day, hour, min, sec, century = 0;
 	unsigned long flags;
 
+	/*
+	 * If pm_trace abused the RTC as storage, set the timespec to 0,
+	 * which tells the caller that this RTC value is unusable.
+	 */
+	if (!pm_trace_rtc_valid()) {
+		now->tv_sec = now->tv_nsec = 0;
+		return;
+	}
+
 	spin_lock_irqsave(&rtc_lock, flags);
 
 	/*
drivers/base/power/trace.c +27 -0
index efec10b..1cda505 100644
@@ -10,6 +10,7 @@
 #include <linux/pm-trace.h>
 #include <linux/export.h>
 #include <linux/rtc.h>
+#include <linux/suspend.h>
 
 #include <linux/mc146818rtc.h>
 
@@ -74,6 +75,9 @@
 
 #define DEVSEED (7919)
 
+bool pm_trace_rtc_abused __read_mostly;
+EXPORT_SYMBOL_GPL(pm_trace_rtc_abused);
+
 static unsigned int dev_hash_value;
 
 static int set_magic_time(unsigned int user, unsigned int file, unsigned int device)
@@ -104,6 +108,7 @@ static int set_magic_time(unsigned int user, unsigned int file, unsigned int dev
 	time.tm_min = (n % 20) * 3;
 	n /= 20;
 	mc146818_set_time(&time);
+	pm_trace_rtc_abused = true;
 	return n ? -1 : 0;
 }
 
@@ -239,9 +244,31 @@ int show_trace_dev_match(char *buf, size_t size)
 	return ret;
 }
 
+static int
+pm_trace_notify(struct notifier_block *nb, unsigned long mode, void *_unused)
+{
+	switch (mode) {
+	case PM_POST_HIBERNATION:
+	case PM_POST_SUSPEND:
+		if (pm_trace_rtc_abused) {
+			pm_trace_rtc_abused = false;
+			pr_warn("Possible incorrect RTC due to pm_trace, please use 'ntpdate' or 'rdate' to reset it.\n");
+		}
+		break;
+	default:
+		break;
+	}
+	return 0;
+}
+
+static struct notifier_block pm_trace_nb = {
+	.notifier_call = pm_trace_notify,
+};
+
 static int early_resume_init(void)
 {
 	hash_value_early_read = read_magic_time();
+	register_pm_notifier(&pm_trace_nb);
 	return 0;
 }
 
drivers/rtc/rtc-cmos.c +7 -0
index dd3d598..b27a189 100644
@@ -191,6 +191,13 @@ static inline void cmos_write_bank2(unsigned char val, unsigned char addr)
 
 static int cmos_read_time(struct device *dev, struct rtc_time *t)
 {
+	/*
+	 * If pm_trace abused the RTC for storage, set the timespec to 0,
+	 * which tells the caller that this RTC value is unusable.
+	 */
+	if (!pm_trace_rtc_valid())
+		return -EIO;
+
 	/* REVISIT:  if the clock has a "century" register, use
 	 * that instead of the heuristic in mc146818_get_time().
 	 * That'll make Y3K compatility (year > 2070) easy!
include/linux/mc146818rtc.h +1 -0
index a585b4b..0661af1 100644
@@ -16,6 +16,7 @@
 #include <asm/mc146818rtc.h>		/* register access macros */
 #include <linux/bcd.h>
 #include <linux/delay.h>
+#include <linux/pm-trace.h>
 
 #ifdef __KERNEL__
 #include <linux/spinlock.h>		/* spinlock_t */
include/linux/pm-trace.h +8 -1
index ecbde7a..7b78793 100644
@@ -1,11 +1,17 @@
 #ifndef PM_TRACE_H
 #define PM_TRACE_H
 
+#include <linux/types.h>
 #ifdef CONFIG_PM_TRACE
 #include <asm/pm-trace.h>
-#include <linux/types.h>
 
 extern int pm_trace_enabled;
+extern bool pm_trace_rtc_abused;
+
+static inline bool pm_trace_rtc_valid(void)
+{
+	return !pm_trace_rtc_abused;
+}
 
 static inline int pm_trace_is_enabled(void)
 {
@@ -24,6 +30,7 @@ extern int show_trace_dev_match(char *buf, size_t size);
 
 #else
 
+static inline bool pm_trace_rtc_valid(void) { return true; }
 static inline int pm_trace_is_enabled(void) { return 0; }
 
 #define TRACE_DEVICE(dev) do { } while (0)
#7 lore raw

PATCH 3/7 timekeeping: clocksource_cyc2ns: Document intended range limitation

In reply to John Stultz

From: Chris Metcalf <[email protected]>

The "cycles" argument should not be an absolute clocksource cycle value, as the implementation's arithmetic will overflow relatively easily with wide (64 bit) clocksource counters.

For performance, the implementation is simple and fast, since the function is intended for only relatively small delta values of clocksource cycles.

Cc Richard Cochran <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc Prarit Bhargava <[email protected]>
Cc Thomas Gleixner <[email protected]>

[jstultz: Fixed up to merge against HEAD & commit message tweaks,
also included rewording suggestion by Ingo]

---
v2: Reworded comments as suggested by Ingo

---
 include/linux/clocksource.h | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)
include/linux/clocksource.h +4 -1
index 0839818..65602d3 100644
@@ -169,7 +169,10 @@ static inline u32 clocksource_hz2mult(u32 hz, u32 shift_constant)
  * @mult:	cycle to nanosecond multiplier
  * @shift:	cycle to nanosecond divisor (power of two)
  *
- * Converts cycles to nanoseconds, using the given mult and shift.
+ * Converts clocksource cycles to nanoseconds, using the given @mult and @shift.
+ * The code is optimized for performance and is not intended to work
+ * with absolute clocksource cycles (as those will easily overflow),
+ * but is only intended to be used with relative (delta) clocksource cycles.
  *
  * XXX - This could use some mult_lxl_ll() asm optimization
  */
Signature · 1 line
2.7.4
#8 lore raw

[tip:timers/core] timekeeping/clocksource_cyc2ns: Document intended range limitation

timekeeping/clocksource_cyc2ns: Document intended range limitation
Commit-ID:  ec4101e8903e318b9fd4e3bbf72b1eaba53c64e1
Gitweb:     http://git.kernel.org/tip/ec4101e8903e318b9fd4e3bbf72b1eaba53c64e1
Author:     Chris Metcalf <[email protected]>
AuthorDate: Mon, 28 Nov 2016 14:35:20 -0800
Committer:  Thomas Gleixner <[email protected]>
CommitDate: Tue, 29 Nov 2016 18:02:58 +0100

timekeeping/clocksource_cyc2ns: Document intended range limitation

The "cycles" argument should not be an absolute clocksource cycle value, as the implementation's arithmetic will overflow relatively easily with wide (64 bit) clocksource counters.

For performance, the implementation is simple and fast, since the function is intended for only relatively small delta values of clocksource cycles.

[jstultz: Fixed up to merge against HEAD & commit message tweaks,
also included rewording suggestion by Ingo]

---
 include/linux/clocksource.h | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)
include/linux/clocksource.h +4 -1
index 0839818..65602d3 100644
@@ -169,7 +169,10 @@ static inline u32 clocksource_hz2mult(u32 hz, u32 shift_constant)
  * @mult:	cycle to nanosecond multiplier
  * @shift:	cycle to nanosecond divisor (power of two)
  *
- * Converts cycles to nanoseconds, using the given mult and shift.
+ * Converts clocksource cycles to nanoseconds, using the given @mult and @shift.
+ * The code is optimized for performance and is not intended to work
+ * with absolute clocksource cycles (as those will easily overflow),
+ * but is only intended to be used with relative (delta) clocksource cycles.
  *
  * XXX - This could use some mult_lxl_ll() asm optimization
  */
#9 lore raw

PATCH 4/7 time: alarmtimer: Add the tracepoints for alarmtimer

In reply to John Stultz

From: Baolin Wang <[email protected]>

For system debugging, we sometimes want to know who sets one alarm timer, the time of the timer, when the timer started and fired and so on. Thus adding tracepoints can help us trace the alarmtimer information.

For example, when we debug the system supend/resume, if the system is always resumed by RTC alarm, we can find out which process set the alarm timer to resume system by below trace log:

......

Binder:3292_2-3304 [000] d..2 149.981123: alarmtimer_cancel:
alarmtimer:ffffffc1319a7800 type:REALTIME expires:1325463120000000000 now:1325376810370370245

Binder:3292_2-3304 [000] d..2 149.981136: alarmtimer_start:
alarmtimer:ffffffc1319a7800 type:REALTIME expires:1325376840000000000 now:1325376810370384591

Binder:3292_9-3953 [000] d..2 150.212991: alarmtimer_cancel:
alarmtimer:ffffffc1319a5a00 type:BOOTTIME
expires:179552000000 now:150154008122

Binder:3292_9-3953 [000] d..2 150.213006: alarmtimer_start:
alarmtimer:ffffffc1319a5a00 type:BOOTTIME
expires:179551000000 now:150154025622

......

system_server-3000 [002] ...1 162.701940: alarmtimer_suspend:
alarmtimer type:REALTIME expires:1325376839802714584

......

Quoted 1 line

From the trace log, we can find out the 'Binder:3292_2' process

set one alarm timer which resumes the system.

Cc Thomas Gleixner <[email protected]>
Cc Richard Cochran <[email protected]>
Cc Prarit Bhargava <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc Steven Rostedt <[email protected]>

---
Changes since v4:
- Initialize 'type' to -1 and rename it in alarmtimer_suspend().
- Fix typo in subject line.

Changes since v3:
- Remove the "ALARM_" prefix in the string.
- Add the ACK by Steven Rostedt.

Changes since v2:
- Save time as s64 type.
- Remove 'process_name' parameter and add 'now' parameter.
- Rename the trace event name.
- Remove restart trace event.
- Other optimization.

---
 include/trace/events/alarmtimer.h | 92 +++++++++++++++++++++++++++++++++++++++
 kernel/time/alarmtimer.c          | 16 ++++++-
 2 files changed, 106 insertions(+), 2 deletions(-)
 create mode 100644 include/trace/events/alarmtimer.h
include/trace/events/alarmtimer.h new file +92 -0
new file mode 100644
index 0000000..61ea556
@@ -0,0 +1,92 @@
+#undef TRACE_SYSTEM
+#define TRACE_SYSTEM alarmtimer
+
+#if !defined(_TRACE_ALARMTIMER_H) || defined(TRACE_HEADER_MULTI_READ)
+#define _TRACE_ALARMTIMER_H
+
+#include <linux/alarmtimer.h>
+#include <linux/rtc.h>
+#include <linux/tracepoint.h>
+
+TRACE_DEFINE_ENUM(ALARM_REALTIME);
+TRACE_DEFINE_ENUM(ALARM_BOOTTIME);
+
+#define show_alarm_type(type)	__print_flags(type, " | ",	\
+	{ 1 << ALARM_REALTIME, "REALTIME" },			\
+	{ 1 << ALARM_BOOTTIME, "BOOTTIME" })
+
+TRACE_EVENT(alarmtimer_suspend,
+
+	TP_PROTO(ktime_t expires, int flag),
+
+	TP_ARGS(expires, flag),
+
+	TP_STRUCT__entry(
+		__field(s64, expires)
+		__field(unsigned char, alarm_type)
+	),
+
+	TP_fast_assign(
+		__entry->expires = expires.tv64;
+		__entry->alarm_type = flag;
+	),
+
+	TP_printk("alarmtimer type:%s expires:%llu",
+		  show_alarm_type((1 << __entry->alarm_type)),
+		  __entry->expires
+	)
+);
+
+DECLARE_EVENT_CLASS(alarm_class,
+
+	TP_PROTO(struct alarm *alarm, ktime_t now),
+
+	TP_ARGS(alarm, now),
+
+	TP_STRUCT__entry(
+		__field(void *,	alarm)
+		__field(unsigned char, alarm_type)
+		__field(s64, expires)
+		__field(s64, now)
+	),
+
+	TP_fast_assign(
+		__entry->alarm = alarm;
+		__entry->alarm_type = alarm->type;
+		__entry->expires = alarm->node.expires.tv64;
+		__entry->now = now.tv64;
+	),
+
+	TP_printk("alarmtimer:%p type:%s expires:%llu now:%llu",
+		  __entry->alarm,
+		  show_alarm_type((1 << __entry->alarm_type)),
+		  __entry->expires,
+		  __entry->now
+	)
+);
+
+DEFINE_EVENT(alarm_class, alarmtimer_fired,
+
+	TP_PROTO(struct alarm *alarm, ktime_t now),
+
+	TP_ARGS(alarm, now)
+);
+
+DEFINE_EVENT(alarm_class, alarmtimer_start,
+
+	TP_PROTO(struct alarm *alarm, ktime_t now),
+
+	TP_ARGS(alarm, now)
+);
+
+DEFINE_EVENT(alarm_class, alarmtimer_cancel,
+
+	TP_PROTO(struct alarm *alarm, ktime_t now),
+
+	TP_ARGS(alarm, now)
+);
+
+#endif /* _TRACE_ALARMTIMER_H */
+
+/* This part must be outside protection */
+#include <trace/define_trace.h>
kernel/time/alarmtimer.c +14 -2
index 12dd190..4590f23 100644
@@ -26,6 +26,9 @@
 #include <linux/workqueue.h>
 #include <linux/freezer.h>
 
+#define CREATE_TRACE_POINTS
+#include <trace/events/alarmtimer.h>
+
 /**
  * struct alarm_base - Alarm timer bases
  * @lock:		Lock for syncrhonized access to the base
@@ -194,6 +197,7 @@ static enum hrtimer_restart alarmtimer_fired(struct hrtimer *timer)
 	}
 	spin_unlock_irqrestore(&base->lock, flags);
 
+	trace_alarmtimer_fired(alarm, base->gettime());
 	return ret;
 
 }
@@ -222,7 +226,7 @@ static int alarmtimer_suspend(struct device *dev)
 	ktime_t min, now;
 	unsigned long flags;
 	struct rtc_device *rtc;
-	int i;
+	int i, min_type = -1;
 	int ret;
 
 	spin_lock_irqsave(&freezer_delta_lock, flags);
@@ -247,8 +251,10 @@ static int alarmtimer_suspend(struct device *dev)
 		if (!next)
 			continue;
 		delta = ktime_sub(next->expires, base->gettime());
-		if (!min.tv64 || (delta.tv64 < min.tv64))
+		if (!min.tv64 || (delta.tv64 < min.tv64)) {
 			min = delta;
+			min_type = i;
+		}
 	}
 	if (min.tv64 == 0)
 		return 0;
@@ -264,6 +270,8 @@ static int alarmtimer_suspend(struct device *dev)
 	now = rtc_tm_to_ktime(tm);
 	now = ktime_add(now, min);
 
+	trace_alarmtimer_suspend(now, min_type);
+
 	/* Set alarm, if in the past reject suspend briefly to handle */
 	ret = rtc_timer_start(rtc, &rtctimer, now, ktime_set(0, 0));
 	if (ret < 0)
@@ -342,6 +350,8 @@ void alarm_start(struct alarm *alarm, ktime_t start)
 	alarmtimer_enqueue(base, alarm);
 	hrtimer_start(&alarm->timer, alarm->node.expires, HRTIMER_MODE_ABS);
 	spin_unlock_irqrestore(&base->lock, flags);
+
+	trace_alarmtimer_start(alarm, base->gettime());
 }
 EXPORT_SYMBOL_GPL(alarm_start);
 
@@ -390,6 +400,8 @@ int alarm_try_to_cancel(struct alarm *alarm)
 	if (ret >= 0)
 		alarmtimer_dequeue(base, alarm);
 	spin_unlock_irqrestore(&base->lock, flags);
+
+	trace_alarmtimer_cancel(alarm, base->gettime());
 	return ret;
 }
 EXPORT_SYMBOL_GPL(alarm_try_to_cancel);
Signature · 1 line
2.7.4
#10 lore raw

Re: [PATCH 4/7] time: alarmtimer: Add the tracepoints for alarmtimer

Quoted 6 lines · John Stultz

From: Baolin Wang <[email protected]>

For system debugging, we sometimes want to know who sets one alarm timer, the time of the timer, when the timer started and fired and so on. Thus adding tracepoints can help us trace the alarmtimer information.

s/one alarm timer/an alarm timer

Quoted 3 lines

For example, when we debug the system supend/resume, if the system is always resumed by RTC alarm, we can find out which process set the alarm timer to resume system by below trace log:

s/when we debug the system/when we debug system
s/supend/suspend
s/resume system/resume the system
s/by below trace log/by the trace log below

Quoted 2 lines

From the trace log, we can find out the 'Binder:3292_2' process set one alarm timer which resumes the system.

s/set one alarm timer/set an alarm timer

Quoted 8 lines

Changes since v4:
- Initialize 'type' to -1 and rename it in alarmtimer_suspend().
- Fix typo in subject line.

Changes since v3:
- Remove the "ALARM_" prefix in the string.
- Add the ACK by Steven Rostedt.

Quoted 6 lines

Changes since v2:
- Save time as s64 type.
- Remove 'process_name' parameter and add 'now' parameter.
- Rename the trace event name.
- Remove restart trace event.
- Other optimization.

I find it really sad that a patch that has gone through 4 iterations still has so many typos and grammar errors in its changelog :-(

Thanks,

Ingo

#11 lore raw

Re: [PATCH 4/7] time: alarmtimer: Add the tracepoints for alarmtimer

Quoted 41 lines · Ingo Molnar
Quoted 6 lines · John Stultz

From: Baolin Wang <[email protected]>

For system debugging, we sometimes want to know who sets one alarm timer, the time of the timer, when the timer started and fired and so on. Thus adding tracepoints can help us trace the alarmtimer information.

s/one alarm timer/an alarm timer

Quoted 3 lines

For example, when we debug the system supend/resume, if the system is always resumed by RTC alarm, we can find out which process set the alarm timer to resume system by below trace log:

s/when we debug the system/when we debug system
s/supend/suspend
s/resume system/resume the system
s/by below trace log/by the trace log below

Quoted 2 lines

From the trace log, we can find out the 'Binder:3292_2' process set one alarm timer which resumes the system.

s/set one alarm timer/set an alarm timer

Quoted 8 lines

Changes since v4:
- Initialize 'type' to -1 and rename it in alarmtimer_suspend().
- Fix typo in subject line.

Changes since v3:
- Remove the "ALARM_" prefix in the string.
- Add the ACK by Steven Rostedt.

Quoted 6 lines

Changes since v2:
- Save time as s64 type.
- Remove 'process_name' parameter and add 'now' parameter.
- Rename the trace event name.
- Remove restart trace event.
- Other optimization.

Quoted 2 lines

I find it really sad that a patch that has gone through 4 iterations still has so many typos and grammar errors in its changelog :-(

Really sorry for these elementary errors, I will fix these errors in new patch. Sorry for troubles again.

Signature · 2 lines
Baolin.wang
Best Regards
#12 lore raw

Re: [PATCH 4/7] time: alarmtimer: Add the tracepoints for alarmtimer

Quoted 42 lines · Baolin Wang
Quoted 41 lines · Ingo Molnar
Quoted 6 lines · John Stultz

From: Baolin Wang <[email protected]>

For system debugging, we sometimes want to know who sets one alarm timer, the time of the timer, when the timer started and fired and so on. Thus adding tracepoints can help us trace the alarmtimer information.

s/one alarm timer/an alarm timer

Quoted 3 lines

For example, when we debug the system supend/resume, if the system is always resumed by RTC alarm, we can find out which process set the alarm timer to resume system by below trace log:

s/when we debug the system/when we debug system
s/supend/suspend
s/resume system/resume the system
s/by below trace log/by the trace log below

Quoted 2 lines

From the trace log, we can find out the 'Binder:3292_2' process set one alarm timer which resumes the system.

s/set one alarm timer/set an alarm timer

Quoted 14 lines

Changes since v4:
- Initialize 'type' to -1 and rename it in alarmtimer_suspend().
- Fix typo in subject line.

Changes since v3:
- Remove the "ALARM_" prefix in the string.
- Add the ACK by Steven Rostedt.

Changes since v2:
- Save time as s64 type.
- Remove 'process_name' parameter and add 'now' parameter.
- Rename the trace event name.
- Remove restart trace event.
- Other optimization.

Quoted 5 lines
Quoted 2 lines

I find it really sad that a patch that has gone through 4 iterations still has so many typos and grammar errors in its changelog :-(

Really sorry for these elementary errors, I will fix these errors in new patch. Sorry for troubles again.

No problem - the code looks fine to me, so we can fix this when applying the patches.

Thanks,

Ingo

#13 lore raw

Re: [PATCH 4/7] time: alarmtimer: Add the tracepoints for alarmtimer

Quoted 45 lines · Ingo Molnar
Quoted 42 lines · Baolin Wang
Quoted 41 lines · Ingo Molnar
Quoted 6 lines · John Stultz

From: Baolin Wang <[email protected]>

For system debugging, we sometimes want to know who sets one alarm timer, the time of the timer, when the timer started and fired and so on. Thus adding tracepoints can help us trace the alarmtimer information.

s/one alarm timer/an alarm timer

Quoted 3 lines

For example, when we debug the system supend/resume, if the system is always resumed by RTC alarm, we can find out which process set the alarm timer to resume system by below trace log:

s/when we debug the system/when we debug system
s/supend/suspend
s/resume system/resume the system
s/by below trace log/by the trace log below

Quoted 2 lines

From the trace log, we can find out the 'Binder:3292_2' process set one alarm timer which resumes the system.

s/set one alarm timer/set an alarm timer

Quoted 14 lines

Changes since v4:
- Initialize 'type' to -1 and rename it in alarmtimer_suspend().
- Fix typo in subject line.

Changes since v3:
- Remove the "ALARM_" prefix in the string.
- Add the ACK by Steven Rostedt.

Changes since v2:
- Save time as s64 type.
- Remove 'process_name' parameter and add 'now' parameter.
- Rename the trace event name.
- Remove restart trace event.
- Other optimization.

Quoted 8 lines
Quoted 5 lines
Quoted 2 lines

I find it really sad that a patch that has gone through 4 iterations still has so many typos and grammar errors in its changelog :-(

Really sorry for these elementary errors, I will fix these errors in new patch. Sorry for troubles again.

No problem - the code looks fine to me, so we can fix this when applying the patches.

Okay, thanks a lot.

Signature · 2 lines
Baolin.wang
Best Regards
Acked-by Steven Rostedt on v1 4/7
#14 lore raw

Re: [PATCH 4/7] time: alarmtimer: Add the tracepoints for alarmtimer

In reply to John Stultz
Quoted 29 lines · John Stultz

From: Baolin Wang <[email protected]>

For system debugging, we sometimes want to know who sets one alarm timer, the time of the timer, when the timer started and fired and so on. Thus adding tracepoints can help us trace the alarmtimer information.

For example, when we debug the system supend/resume, if the system is always resumed by RTC alarm, we can find out which process set the alarm timer to resume system by below trace log:

......

Binder:3292_2-3304 [000] d..2 149.981123: alarmtimer_cancel:
alarmtimer:ffffffc1319a7800 type:REALTIME expires:1325463120000000000 now:1325376810370370245

Binder:3292_2-3304 [000] d..2 149.981136: alarmtimer_start:
alarmtimer:ffffffc1319a7800 type:REALTIME expires:1325376840000000000 now:1325376810370384591

Binder:3292_9-3953 [000] d..2 150.212991: alarmtimer_cancel:
alarmtimer:ffffffc1319a5a00 type:BOOTTIME
expires:179552000000 now:150154008122

Binder:3292_9-3953 [000] d..2 150.213006: alarmtimer_start:
alarmtimer:ffffffc1319a5a00 type:BOOTTIME
expires:179551000000 now:150154025622

Quoted 8 lines

......

system_server-3000 [002] ...1 162.701940: alarmtimer_suspend:
alarmtimer type:REALTIME expires:1325376839802714584
......

Quoted 1 line

From the trace log, we can find out the 'Binder:3292_2' process

set one alarm timer which resumes the system.

To be honest. I have a hard time to figure out that Binder:3292_2 armed the timer.

	suspend: expires:1325376839802714584

Binder start: expires:1325376840000000000

That does not make any sense at all. And this is a result of how that suspend tracepoint got glued into alarmtimer_suspend():

Quoted 22 lines
(quoted diff) +4 -2
@@ -222,7 +226,7 @@ static int alarmtimer_suspend(struct device *dev)
 	ktime_t min, now;
 	unsigned long flags;
 	struct rtc_device *rtc;
-	int i;
+	int i, min_type = -1;
 	int ret;
 
 	spin_lock_irqsave(&freezer_delta_lock, flags);
@@ -247,8 +251,10 @@ static int alarmtimer_suspend(struct device *dev)
 		if (!next)
 			continue;
 		delta = ktime_sub(next->expires, base->gettime());
-		if (!min.tv64 || (delta.tv64 < min.tv64))
+		if (!min.tv64 || (delta.tv64 < min.tv64)) {
 			min = delta;
+			min_type = i;
+		}
 	}
 	if (min.tv64 == 0)
 		return 0;

Review static int alarmtimer_suspend(struct device *dev) · line 273

Quoted 4 lines
(quoted diff) +1 -0
@@ -264,6 +270,8 @@ static int alarmtimer_suspend(struct device *dev)
 	now = rtc_tm_to_ktime(tm);
 	now = ktime_add(now, min);
 
+	trace_alarmtimer_suspend(now, min_type);

So first we calculate the relative expiry time from now, add it to the RTC and then print out the sum. That's utter crap especially if the first expiring timer is based on CLOCK_BOOTTIME. Even with CLOCK_REALTIME you do not get matching expiry time as you can see above.

Now looking at the other possibility to arm the resume time: freezer_delta. That's even more undebuggable because the freezer time is stored as relative timeout without any type information. How is one supposed to map that back to an alarm timer start?

Find below a revamped version (untested) of this including a proper written changelog.

Thanks,

tglx

8<--------------
Subject: time/alarmtimer: Add tracepoints for alarmtimer
From: John Stultz <[email protected]>
Date: Mon, 28 Nov 2016 14:35:21 -0800

From: Baolin Wang <[email protected]>

Alarm timers are one of the mechanisms to wake up a system from suspend, but there exist no tracepoints to analyse which process/thread armed an alarmtimer.

Add tracepoints for start/cancel/expire of individual alarm timers and one for tracing the suspend time decision when to resume the system.

The following trace excerpt illustrates the new mechanism:

Binder:3292_2-3304 [000] d..2 149.981123: alarmtimer_cancel:
alarmtimer:ffffffc1319a7800 type:REALTIME expires:1325463120000000000 now:1325376810370370245

Binder:3292_2-3304 [000] d..2 149.981136: alarmtimer_start:
alarmtimer:ffffffc1319a7800 type:REALTIME expires:1325376840000000000 now:1325376810370384591

Binder:3292_9-3953 [000] d..2 150.212991: alarmtimer_cancel:
alarmtimer:ffffffc1319a5a00 type:BOOTTIME
expires:179552000000 now:150154008122

Binder:3292_9-3953 [000] d..2 150.213006: alarmtimer_start:
alarmtimer:ffffffc1319a5a00 type:BOOTTIME
expires:179551000000 now:150154025622

system_server-3000 [002] ...1 162.701940: alarmtimer_suspend:
alarmtimer type:REALTIME expires:1325376840000000000

The wakeup time which is selected at suspend time allows to map it back to the task arming the timer: Binder:3292_2.

[ tglx: Store real alarm timer expiry time instead of some useless RTC
relative information, add proper type information for wakeups which are handled via the clock_nanosleep/freezer and massage the changelog. ]

---
Changes since v4:
- Initialize 'type' to -1 and rename it in alarmtimer_suspend().
- Fix typo in subject line.

Changes since v3:
- Remove the "ALARM_" prefix in the string.
- Add the ACK by Steven Rostedt.

Changes since v2:
- Save time as s64 type.
- Remove 'process_name' parameter and add 'now' parameter.
- Rename the trace event name.
- Remove restart trace event.
- Other optimization.

---
 include/linux/alarmtimer.h        |    5 +
 include/trace/events/alarmtimer.h |   96 ++++++++++++++++++++++++++++++++++++++
 kernel/time/alarmtimer.c          |   53 +++++++++++++++++---
 3 files changed, 144 insertions(+), 10 deletions(-)
 create mode 100644 include/trace/events/alarmtimer.h
include/linux/alarmtimer.h +5 -0
@@ -10,7 +10,12 @@ enum alarmtimer_type {
 	ALARM_REALTIME,
 	ALARM_BOOTTIME,
 
+	/* Supported types end here */
 	ALARM_NUMTYPE,
+
+	/* Used for tracing information. No usable types. */
+	ALARM_REALTIME_FREEZER,
+	ALARM_BOOTTIME_FREEZER,
 };
 
 enum alarmtimer_restart {
include/trace/events/alarmtimer.h new file +96 -0

96 lines not on this page · raw message

kernel/time/alarmtimer.c +43 -10

118 lines not on this page · raw message

#15 lore raw

Re: [PATCH 4/7] time: alarmtimer: Add the tracepoints for alarmtimer

Quoted 81 lines · Thomas Gleixner
Quoted 29 lines · John Stultz

From: Baolin Wang <[email protected]>

For system debugging, we sometimes want to know who sets one alarm timer, the time of the timer, when the timer started and fired and so on. Thus adding tracepoints can help us trace the alarmtimer information.

For example, when we debug the system supend/resume, if the system is always resumed by RTC alarm, we can find out which process set the alarm timer to resume system by below trace log:

......

Binder:3292_2-3304 [000] d..2 149.981123: alarmtimer_cancel:
alarmtimer:ffffffc1319a7800 type:REALTIME expires:1325463120000000000 now:1325376810370370245

Binder:3292_2-3304 [000] d..2 149.981136: alarmtimer_start:
alarmtimer:ffffffc1319a7800 type:REALTIME expires:1325376840000000000 now:1325376810370384591

Binder:3292_9-3953 [000] d..2 150.212991: alarmtimer_cancel:
alarmtimer:ffffffc1319a5a00 type:BOOTTIME
expires:179552000000 now:150154008122

Binder:3292_9-3953 [000] d..2 150.213006: alarmtimer_start:
alarmtimer:ffffffc1319a5a00 type:BOOTTIME
expires:179551000000 now:150154025622

Quoted 8 lines

......

system_server-3000 [002] ...1 162.701940: alarmtimer_suspend:
alarmtimer type:REALTIME expires:1325376839802714584
......

Quoted 1 line

From the trace log, we can find out the 'Binder:3292_2' process

set one alarm timer which resumes the system.

To be honest. I have a hard time to figure out that Binder:3292_2 armed the timer.

        suspend: expires:1325376839802714584

Binder start: expires:1325376840000000000

That does not make any sense at all. And this is a result of how that suspend tracepoint got glued into alarmtimer_suspend():

Quoted 22 lines
kernel/time/alarmtimer.c +4 -2
@@ -222,7 +226,7 @@ static int alarmtimer_suspend(struct device *dev)
      ktime_t min, now;
      unsigned long flags;
      struct rtc_device *rtc;
-     int i;
+     int i, min_type = -1;
      int ret;
 
      spin_lock_irqsave(&freezer_delta_lock, flags);
@@ -247,8 +251,10 @@ static int alarmtimer_suspend(struct device *dev)
              if (!next)
                      continue;
              delta = ktime_sub(next->expires, base->gettime());
-             if (!min.tv64 || (delta.tv64 < min.tv64))
+             if (!min.tv64 || (delta.tv64 < min.tv64)) {
                      min = delta;
+                     min_type = i;
+             }
      }
      if (min.tv64 == 0)
              return 0;

kernel/time/alarmtimer.c static int alarmtimer_suspend(struct device *dev) · line 273

Quoted 4 lines
kernel/time/alarmtimer.c +1 -0
@@ -264,6 +270,8 @@ static int alarmtimer_suspend(struct device *dev)
      now = rtc_tm_to_ktime(tm);
      now = ktime_add(now, min);
 
+     trace_alarmtimer_suspend(now, min_type);

So first we calculate the relative expiry time from now, add it to the RTC and then print out the sum. That's utter crap especially if the first expiring timer is based on CLOCK_BOOTTIME. Even with CLOCK_REALTIME you do not get matching expiry time as you can see above.

Quoted 7 lines

Now looking at the other possibility to arm the resume time: freezer_delta. That's even more undebuggable because the freezer time is stored as relative timeout without any type information. How is one supposed to map that back to an alarm timer start?

Find below a revamped version (untested) of this including a proper written changelog.

Make sense. Thanks for your patch and I will help to test it.

Signature · 2 lines
Baolin.wang
Best Regards
Tested-by
#16 lore raw

Re: [PATCH 4/7] time: alarmtimer: Add the tracepoints for alarmtimer

In reply to Thomas Gleixner
Quoted 136 lines · Thomas Gleixner
Quoted 29 lines · John Stultz

From: Baolin Wang <[email protected]>

For system debugging, we sometimes want to know who sets one alarm timer, the time of the timer, when the timer started and fired and so on. Thus adding tracepoints can help us trace the alarmtimer information.

For example, when we debug the system supend/resume, if the system is always resumed by RTC alarm, we can find out which process set the alarm timer to resume system by below trace log:

......

Binder:3292_2-3304 [000] d..2 149.981123: alarmtimer_cancel:
alarmtimer:ffffffc1319a7800 type:REALTIME expires:1325463120000000000 now:1325376810370370245

Binder:3292_2-3304 [000] d..2 149.981136: alarmtimer_start:
alarmtimer:ffffffc1319a7800 type:REALTIME expires:1325376840000000000 now:1325376810370384591

Binder:3292_9-3953 [000] d..2 150.212991: alarmtimer_cancel:
alarmtimer:ffffffc1319a5a00 type:BOOTTIME
expires:179552000000 now:150154008122

Binder:3292_9-3953 [000] d..2 150.213006: alarmtimer_start:
alarmtimer:ffffffc1319a5a00 type:BOOTTIME
expires:179551000000 now:150154025622

Quoted 8 lines

......

system_server-3000 [002] ...1 162.701940: alarmtimer_suspend:
alarmtimer type:REALTIME expires:1325376839802714584
......

Quoted 1 line

From the trace log, we can find out the 'Binder:3292_2' process

set one alarm timer which resumes the system.

To be honest. I have a hard time to figure out that Binder:3292_2 armed the timer.

        suspend: expires:1325376839802714584

Binder start: expires:1325376840000000000

That does not make any sense at all. And this is a result of how that suspend tracepoint got glued into alarmtimer_suspend():

Quoted 22 lines
kernel/time/alarmtimer.c +4 -2
@@ -222,7 +226,7 @@ static int alarmtimer_suspend(struct device *dev)
      ktime_t min, now;
      unsigned long flags;
      struct rtc_device *rtc;
-     int i;
+     int i, min_type = -1;
      int ret;
 
      spin_lock_irqsave(&freezer_delta_lock, flags);
@@ -247,8 +251,10 @@ static int alarmtimer_suspend(struct device *dev)
              if (!next)
                      continue;
              delta = ktime_sub(next->expires, base->gettime());
-             if (!min.tv64 || (delta.tv64 < min.tv64))
+             if (!min.tv64 || (delta.tv64 < min.tv64)) {
                      min = delta;
+                     min_type = i;
+             }
      }
      if (min.tv64 == 0)
              return 0;

kernel/time/alarmtimer.c static int alarmtimer_suspend(struct device *dev) · line 273

Quoted 4 lines
kernel/time/alarmtimer.c +1 -0
@@ -264,6 +270,8 @@ static int alarmtimer_suspend(struct device *dev)
      now = rtc_tm_to_ktime(tm);
      now = ktime_add(now, min);
 
+     trace_alarmtimer_suspend(now, min_type);

So first we calculate the relative expiry time from now, add it to the RTC and then print out the sum. That's utter crap especially if the first expiring timer is based on CLOCK_BOOTTIME. Even with CLOCK_REALTIME you do not get matching expiry time as you can see above.

Now looking at the other possibility to arm the resume time: freezer_delta. That's even more undebuggable because the freezer time is stored as relative timeout without any type information. How is one supposed to map that back to an alarm timer start?

Find below a revamped version (untested) of this including a proper written changelog.

Thanks,

tglx

8<--------------
Subject: time/alarmtimer: Add tracepoints for alarmtimer
From: John Stultz <[email protected]>
Date: Mon, 28 Nov 2016 14:35:21 -0800

From: Baolin Wang <[email protected]>

Alarm timers are one of the mechanisms to wake up a system from suspend, but there exist no tracepoints to analyse which process/thread armed an alarmtimer.

Add tracepoints for start/cancel/expire of individual alarm timers and one for tracing the suspend time decision when to resume the system.

The following trace excerpt illustrates the new mechanism:

Binder:3292_2-3304 [000] d..2 149.981123: alarmtimer_cancel:
alarmtimer:ffffffc1319a7800 type:REALTIME expires:1325463120000000000 now:1325376810370370245

Binder:3292_2-3304 [000] d..2 149.981136: alarmtimer_start:
alarmtimer:ffffffc1319a7800 type:REALTIME expires:1325376840000000000 now:1325376810370384591

Binder:3292_9-3953 [000] d..2 150.212991: alarmtimer_cancel:
alarmtimer:ffffffc1319a5a00 type:BOOTTIME
expires:179552000000 now:150154008122

Binder:3292_9-3953 [000] d..2 150.213006: alarmtimer_start:
alarmtimer:ffffffc1319a5a00 type:BOOTTIME
expires:179551000000 now:150154025622

system_server-3000 [002] ...1 162.701940: alarmtimer_suspend:
alarmtimer type:REALTIME expires:1325376840000000000

The wakeup time which is selected at suspend time allows to map it back to the task arming the timer: Binder:3292_2.

[ tglx: Store real alarm timer expiry time instead of some useless RTC
relative information, add proper type information for wakeups which are handled via the clock_nanosleep/freezer and massage the changelog. ]

Quoted 7 lines
Signature · 2 lines
Baolin.wang
Best Regards
#17 lore raw

[tip:timers/core] alarmtimer: Add tracepoints for alarm timers

In reply to John Stultz
alarmtimer: Add tracepoints for alarm timers
Commit-ID:  4a057549d6044c2dea47e80f8369a76225ec9d90
Gitweb:     http://git.kernel.org/tip/4a057549d6044c2dea47e80f8369a76225ec9d90
Author:     Baolin Wang <[email protected]>
AuthorDate: Mon, 28 Nov 2016 14:35:21 -0800
Committer:  Thomas Gleixner <[email protected]>
CommitDate: Thu, 1 Dec 2016 14:45:08 +0100

alarmtimer: Add tracepoints for alarm timers

Alarm timers are one of the mechanisms to wake up a system from suspend, but there exist no tracepoints to analyse which process/thread armed an alarmtimer.

Add tracepoints for start/cancel/expire of individual alarm timers and one for tracing the suspend time decision when to resume the system.

The following trace excerpt illustrates the new mechanism:

Binder:3292_2-3304 [000] d..2 149.981123: alarmtimer_cancel:
alarmtimer:ffffffc1319a7800 type:REALTIME expires:1325463120000000000 now:1325376810370370245

Binder:3292_2-3304 [000] d..2 149.981136: alarmtimer_start:
alarmtimer:ffffffc1319a7800 type:REALTIME expires:1325376840000000000 now:1325376810370384591

Binder:3292_9-3953 [000] d..2 150.212991: alarmtimer_cancel:
alarmtimer:ffffffc1319a5a00 type:BOOTTIME
expires:179552000000 now:150154008122

Binder:3292_9-3953 [000] d..2 150.213006: alarmtimer_start:
alarmtimer:ffffffc1319a5a00 type:BOOTTIME
expires:179551000000 now:150154025622

system_server-3000 [002] ...1 162.701940: alarmtimer_suspend:
alarmtimer type:REALTIME expires:1325376840000000000

The wakeup time which is selected at suspend time allows to map it back to the task arming the timer: Binder:3292_2.

[ tglx: Store alarm timer expiry time instead of some useless RTC relative
information, add proper type information for wakeups which are handled via the clock_nanosleep/freezer and massage the changelog. ]

---
 include/linux/alarmtimer.h        |  5 ++
 include/trace/events/alarmtimer.h | 96 +++++++++++++++++++++++++++++++++++++++
 kernel/time/alarmtimer.c          | 53 +++++++++++++++++----
 3 files changed, 144 insertions(+), 10 deletions(-)
include/linux/alarmtimer.h +5 -0
index 9d80312..c70aac1 100644
@@ -10,7 +10,12 @@ enum alarmtimer_type {
 	ALARM_REALTIME,
 	ALARM_BOOTTIME,
 
+	/* Supported types end here */
 	ALARM_NUMTYPE,
+
+	/* Used for tracing information. No usable types. */
+	ALARM_REALTIME_FREEZER,
+	ALARM_BOOTTIME_FREEZER,
 };
 
 enum alarmtimer_restart {
include/trace/events/alarmtimer.h new file +96 -0

96 lines not on this page · raw message

kernel/time/alarmtimer.c +43 -10

118 lines not on this page · raw message

#18 lore raw

PATCH 5/7 timekeeping: Add a fast and NMI safe boot clock

In reply to John Stultz

From: Joel Fernandes <[email protected]>

This boot clock can be used as a tracing clock and will account for suspend time.

To keep it NMI safe since we're accessing from tracing, we're not using a separate timekeeper with updates to monotonic clock and boot offset protected with seqlocks. This has the following minor side effects:

(1) Its possible that a timestamp be taken after the boot offset is updated but before the timekeeper is updated. If this happens, the new boot offset is added to the old timekeeping making the clock appear to update slightly earlier:
CPU 0 CPU 1
timekeeping_inject_sleeptime64()

   __timekeeping_inject_sleeptime(tk, delta);
                                                timestamp();

timekeeping_update(tk, TK_CLEAR_NTP...);

(2) On 32-bit systems, the 64-bit boot offset (tk->offs_boot) may be partially updated. Since the tk->offs_boot update is a rare event, this should be a rare occurrence which postprocessing should be able to handle.

Cc Steven Rostedt <[email protected]>
Cc Thomas Gleixner <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc Richard Cochran <[email protected]>
Cc Prarit Bhargava <[email protected]>
---
 include/linux/timekeeping.h |  1 +
 kernel/time/timekeeping.c   | 29 +++++++++++++++++++++++++++++
 2 files changed, 30 insertions(+)
include/linux/timekeeping.h +1 -0
index 09168c5..361f8bf 100644
@@ -249,6 +249,7 @@ static inline u64 ktime_get_raw_ns(void)
 
 extern u64 ktime_get_mono_fast_ns(void);
 extern u64 ktime_get_raw_fast_ns(void);
+extern u64 ktime_get_boot_fast_ns(void);
 
 /*
  * Timespec interfaces utilizing the ktime based ones
kernel/time/timekeeping.c +29 -0
index 37dec7e..b2286e9 100644
@@ -425,6 +425,35 @@ u64 ktime_get_raw_fast_ns(void)
 }
 EXPORT_SYMBOL_GPL(ktime_get_raw_fast_ns);
 
+/**
+ * ktime_get_boot_fast_ns - NMI safe and fast access to boot clock.
+ *
+ * To keep it NMI safe since we're accessing from tracing, we're not using a
+ * separate timekeeper with updates to monotonic clock and boot offset
+ * protected with seqlocks. This has the following minor side effects:
+ *
+ * (1) Its possible that a timestamp be taken after the boot offset is updated
+ * but before the timekeeper is updated. If this happens, the new boot offset
+ * is added to the old timekeeping making the clock appear to update slightly
+ * earlier:
+ *    CPU 0                                        CPU 1
+ *    timekeeping_inject_sleeptime64()
+ *    __timekeeping_inject_sleeptime(tk, delta);
+ *                                                 timestamp();
+ *    timekeeping_update(tk, TK_CLEAR_NTP...);
+ *
+ * (2) On 32-bit systems, the 64-bit boot offset (tk->offs_boot) may be
+ * partially updated.  Since the tk->offs_boot update is a rare event, this
+ * should be a rare occurrence which postprocessing should be able to handle.
+ */
+u64 notrace ktime_get_boot_fast_ns(void)
+{
+	struct timekeeper *tk = &tk_core.timekeeper;
+
+	return (ktime_get_mono_fast_ns() + ktime_to_ns(tk->offs_boot));
+}
+EXPORT_SYMBOL_GPL(ktime_get_boot_fast_ns);
+
 /* Suspend-time cycles value for halted fast timekeeper. */
 static cycle_t cycles_at_suspend;
 
Signature · 1 line
2.7.4
#19 lore raw

[tip:timers/core] timekeeping: Add a fast and NMI safe boot clock

timekeeping: Add a fast and NMI safe boot clock
Commit-ID:  948a5312f41658f7b76a598a139ef1f4dea09ca9
Gitweb:     http://git.kernel.org/tip/948a5312f41658f7b76a598a139ef1f4dea09ca9
Author:     Joel Fernandes <[email protected]>
AuthorDate: Mon, 28 Nov 2016 14:35:22 -0800
Committer:  Thomas Gleixner <[email protected]>
CommitDate: Tue, 29 Nov 2016 18:02:59 +0100

timekeeping: Add a fast and NMI safe boot clock

This boot clock can be used as a tracing clock and will account for suspend time.

To keep it NMI safe since we're accessing from tracing, we're not using a separate timekeeper with updates to monotonic clock and boot offset protected with seqlocks. This has the following minor side effects:

(1) Its possible that a timestamp be taken after the boot offset is updated but before the timekeeper is updated. If this happens, the new boot offset is added to the old timekeeping making the clock appear to update slightly earlier:
CPU 0 CPU 1
timekeeping_inject_sleeptime64()

   __timekeeping_inject_sleeptime(tk, delta);
                                                timestamp();

timekeeping_update(tk, TK_CLEAR_NTP...);

(2) On 32-bit systems, the 64-bit boot offset (tk->offs_boot) may be partially updated. Since the tk->offs_boot update is a rare event, this should be a rare occurrence which postprocessing should be able to handle.

---
 include/linux/timekeeping.h |  1 +
 kernel/time/timekeeping.c   | 29 +++++++++++++++++++++++++++++
 2 files changed, 30 insertions(+)
include/linux/timekeeping.h +1 -0
index 09168c5..361f8bf 100644
@@ -249,6 +249,7 @@ static inline u64 ktime_get_raw_ns(void)
 
 extern u64 ktime_get_mono_fast_ns(void);
 extern u64 ktime_get_raw_fast_ns(void);
+extern u64 ktime_get_boot_fast_ns(void);
 
 /*
  * Timespec interfaces utilizing the ktime based ones
kernel/time/timekeeping.c +29 -0
index 37dec7e..b2286e9 100644
@@ -425,6 +425,35 @@ u64 ktime_get_raw_fast_ns(void)
 }
 EXPORT_SYMBOL_GPL(ktime_get_raw_fast_ns);
 
+/**
+ * ktime_get_boot_fast_ns - NMI safe and fast access to boot clock.
+ *
+ * To keep it NMI safe since we're accessing from tracing, we're not using a
+ * separate timekeeper with updates to monotonic clock and boot offset
+ * protected with seqlocks. This has the following minor side effects:
+ *
+ * (1) Its possible that a timestamp be taken after the boot offset is updated
+ * but before the timekeeper is updated. If this happens, the new boot offset
+ * is added to the old timekeeping making the clock appear to update slightly
+ * earlier:
+ *    CPU 0                                        CPU 1
+ *    timekeeping_inject_sleeptime64()
+ *    __timekeeping_inject_sleeptime(tk, delta);
+ *                                                 timestamp();
+ *    timekeeping_update(tk, TK_CLEAR_NTP...);
+ *
+ * (2) On 32-bit systems, the 64-bit boot offset (tk->offs_boot) may be
+ * partially updated.  Since the tk->offs_boot update is a rare event, this
+ * should be a rare occurrence which postprocessing should be able to handle.
+ */
+u64 notrace ktime_get_boot_fast_ns(void)
+{
+	struct timekeeper *tk = &tk_core.timekeeper;
+
+	return (ktime_get_mono_fast_ns() + ktime_to_ns(tk->offs_boot));
+}
+EXPORT_SYMBOL_GPL(ktime_get_boot_fast_ns);
+
 /* Suspend-time cycles value for halted fast timekeeper. */
 static cycle_t cycles_at_suspend;
 
#20 lore raw

PATCH 6/7 trace: Add an option for boot clock as trace clock

In reply to John Stultz

From: Joel Fernandes <[email protected]>

Unlike monotonic clock, boot clock as a trace clock will account for time spent in suspend useful for tracing suspend/resume. This uses earlier introduced infrastructure for using the fast boot clock.

Cc Steven Rostedt <[email protected]>
Cc Thomas Gleixner <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc Richard Cochran <[email protected]>
Cc Prarit Bhargava <[email protected]>
---
 kernel/trace/trace.c | 1 +
 1 file changed, 1 insertion(+)
kernel/trace/trace.c +1 -0
index 8696ce6..f7b64db 100644
@@ -1125,6 +1125,7 @@ static struct {
 	{ trace_clock,			"perf",		1 },
 	{ ktime_get_mono_fast_ns,	"mono",		1 },
 	{ ktime_get_raw_fast_ns,	"mono_raw",	1 },
+	{ ktime_get_boot_fast_ns,	"boot",		1 },
 	ARCH_TRACE_CLOCKS
 };
 
Signature · 1 line
2.7.4
maintainer Acked-by on v1 6/7
#21 lore raw

Re: [PATCH 6/7] trace: Add an option for boot clock as trace clock

Quoted 6 lines · On Mon, 28 Nov 2016 14:35:23 -0800 John Stultz <[email protected]> wrote:

From: Joel Fernandes <[email protected]>

Unlike monotonic clock, boot clock as a trace clock will account for time spent in suspend useful for tracing suspend/resume. This uses earlier introduced infrastructure for using the fast boot clock.

-- Steve

Quoted 24 lines
Cc Steven Rostedt <[email protected]>
Cc Thomas Gleixner <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc Richard Cochran <[email protected]>
Cc Prarit Bhargava <[email protected]>
---
 kernel/trace/trace.c | 1 +
 1 file changed, 1 insertion(+)
kernel/trace/trace.c +1 -0
index 8696ce6..f7b64db 100644
@@ -1125,6 +1125,7 @@ static struct {
 	{ trace_clock,			"perf",		1 },
 	{ ktime_get_mono_fast_ns,	"mono",		1 },
 	{ ktime_get_raw_fast_ns,	"mono_raw",	1 },
+	{ ktime_get_boot_fast_ns,	"boot",		1 },
 	ARCH_TRACE_CLOCKS
 };
 
#22 lore raw

[tip:timers/core] trace: Add an option for boot clock as trace clock

In reply to John Stultz
trace: Add an option for boot clock as trace clock
Commit-ID:  80ec3552107ac16a836dbff4cf3c23fdd3256ee3
Gitweb:     http://git.kernel.org/tip/80ec3552107ac16a836dbff4cf3c23fdd3256ee3
Author:     Joel Fernandes <[email protected]>
AuthorDate: Mon, 28 Nov 2016 14:35:23 -0800
Committer:  Thomas Gleixner <[email protected]>
CommitDate: Tue, 29 Nov 2016 18:02:59 +0100

trace: Add an option for boot clock as trace clock

Unlike monotonic clock, boot clock as a trace clock will account for time spent in suspend useful for tracing suspend/resume. This uses earlier introduced infrastructure for using the fast boot clock.

---
 kernel/trace/trace.c | 1 +
 1 file changed, 1 insertion(+)
kernel/trace/trace.c +1 -0
index 8696ce6..f7b64db 100644
@@ -1125,6 +1125,7 @@ static struct {
 	{ trace_clock,			"perf",		1 },
 	{ ktime_get_mono_fast_ns,	"mono",		1 },
 	{ ktime_get_raw_fast_ns,	"mono_raw",	1 },
+	{ ktime_get_boot_fast_ns,	"boot",		1 },
 	ARCH_TRACE_CLOCKS
 };
 
#23 lore raw

PATCH 7/7 trace: Update documentation for mono, mono_raw and boot clock

In reply to John Stultz

From: Joel Fernandes <[email protected]>

Documentation was missing for mono and mono_raw, add them and also for the boot clock introduced in this series.

Cc Steven Rostedt <[email protected]>
Cc Thomas Gleixner <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc Richard Cochran <[email protected]>
Cc Prarit Bhargava <[email protected]>
---
 Documentation/trace/ftrace.txt | 20 ++++++++++++++++++++
 1 file changed, 20 insertions(+)
Documentation/trace/ftrace.txt +20 -0
index 185c39f..5180b09 100644
@@ -362,6 +362,26 @@ of ftrace. Here is a list of some of the key files:
 		  to correlate events across hypervisor/guest if
 		  tb_offset is known.
 
+	  mono: This uses the fast monotonic clock (CLOCK_MONOTONIC)
+		which is monotonic and is subject to NTP rate adjustments.
+
+	  mono_raw:
+		This is the raw monotonic clock (CLOCK_MONOTONIC_RAW)
+		which is montonic but is not subject to any rate adjustments
+		and ticks at the same rate as the hardware clocksource.
+
+	  boot: This is the boot clock (CLOCK_BOOTTIME) and is based on the
+		fast monotonic clock, but also accounts for time spent in
+		suspend. Since the clock access is designed for use in
+		tracing in the suspend path, some side effects are possible
+		if clock is accessed after the suspend time is accounted before
+		the fast mono clock is updated. In this case, the clock update
+		appears to happen slightly sooner than it normally would have.
+		Also on 32-bit systems, its possible that the 64-bit boot offset
+		sees a partial update. These effects are rare and post
+		processing should be able to handle them. See comments on
+		ktime_get_boot_fast_ns function for more information.
+
 	To set a clock, simply echo the clock name into this file.
 
 	  echo global > trace_clock
Signature · 1 line
2.7.4
maintainer Acked-by on v1 7/7
#24 lore raw

Re: [PATCH 7/7] trace: Update documentation for mono, mono_raw and boot clock

Quoted 6 lines · On Mon, 28 Nov 2016 14:35:24 -0800 John Stultz <[email protected]> wrote:

From: Joel Fernandes <[email protected]>

Documentation was missing for mono and mono_raw, add them and also for the boot clock introduced in this series.

Cc Steven Rostedt <[email protected]>

-- Steve

Quoted 8 lines
Cc Thomas Gleixner <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc Richard Cochran <[email protected]>
Cc Prarit Bhargava <[email protected]>

---

#25 lore raw

Re: [PATCH 7/7] trace: Update documentation for mono, mono_raw and boot clock

In reply to John Stultz

Documentation/trace/ftrace.txt of ftrace. Here is a list of some of the key files: · line 383

Quoted 11 lines · John Stultz
Documentation/trace/ftrace.txt +11 -0
+	  boot: This is the boot clock (CLOCK_BOOTTIME) and is based on the
+		fast monotonic clock, but also accounts for time spent in
+		suspend. Since the clock access is designed for use in
+		tracing in the suspend path, some side effects are possible
+		if clock is accessed after the suspend time is accounted before
+		the fast mono clock is updated. In this case, the clock update
+		appears to happen slightly sooner than it normally would have.
+		Also on 32-bit systems, its possible that the 64-bit boot offset
+		sees a partial update. These effects are rare and post
+		processing should be able to handle them. See comments on
+		ktime_get_boot_fast_ns function for more information.

s/its possible/it's possible
s/comments on ktime_get_boost_fast_ns function/comments in the ktime_get_boost_fast_ns() function

Thanks,

Ingo

#26 lore raw

Re: [PATCH 7/7] trace: Update documentation for mono, mono_raw and boot clock

Quoted 15 lines · Ingo Molnar

Documentation/trace/ftrace.txt of ftrace. Here is a list of some of the key files: · line 383

Quoted 11 lines · John Stultz
Documentation/trace/ftrace.txt +11 -0
+       boot: This is the boot clock (CLOCK_BOOTTIME) and is based on the
+             fast monotonic clock, but also accounts for time spent in
+             suspend. Since the clock access is designed for use in
+             tracing in the suspend path, some side effects are possible
+             if clock is accessed after the suspend time is accounted before
+             the fast mono clock is updated. In this case, the clock update
+             appears to happen slightly sooner than it normally would have.
+             Also on 32-bit systems, its possible that the 64-bit boot offset
+             sees a partial update. These effects are rare and post
+             processing should be able to handle them. See comments on
+             ktime_get_boot_fast_ns function for more information.
Quoted 3 lines

s/its possible/it's possible
s/comments on ktime_get_boost_fast_ns function/comments in the ktime_get_boost_fast_ns() function

Thanks, I'll fix these up and repost.

Regards,
Joel

Quoted 3 lines

Thanks,

Ingo

#27 lore raw

Re: [PATCH 7/7] trace: Update documentation for mono, mono_raw and boot clock

Quoted 16 lines · Joel Fernandes
Quoted 15 lines · Ingo Molnar
Quoted 11 lines · John Stultz
Documentation/trace/ftrace.txt +11 -0
+       boot: This is the boot clock (CLOCK_BOOTTIME) and is based on the
+             fast monotonic clock, but also accounts for time spent in
+             suspend. Since the clock access is designed for use in
+             tracing in the suspend path, some side effects are possible
+             if clock is accessed after the suspend time is accounted before
+             the fast mono clock is updated. In this case, the clock update
+             appears to happen slightly sooner than it normally would have.
+             Also on 32-bit systems, its possible that the 64-bit boot offset
+             sees a partial update. These effects are rare and post
+             processing should be able to handle them. See comments on
+             ktime_get_boot_fast_ns function for more information.
Quoted 5 lines
Quoted 3 lines

s/its possible/it's possible
s/comments on ktime_get_boost_fast_ns function/comments in the ktime_get_boost_fast_ns() function

Thanks, I'll fix these up and repost.

Don't bother. I have fixed it up locally already.

Thanks,

tglx

#28 lore raw

[tip:timers/core] trace: Update documentation for mono, mono_raw and boot clock

In reply to John Stultz
trace: Update documentation for mono, mono_raw and boot clock
Commit-ID:  2924ecd441f3c12da59b297daade265f49d0831b
Gitweb:     http://git.kernel.org/tip/2924ecd441f3c12da59b297daade265f49d0831b
Author:     Joel Fernandes <[email protected]>
AuthorDate: Mon, 28 Nov 2016 14:35:24 -0800
Committer:  Thomas Gleixner <[email protected]>
CommitDate: Tue, 29 Nov 2016 18:03:00 +0100

trace: Update documentation for mono, mono_raw and boot clock

Documentation was missing for mono and mono_raw, add them and also for the boot clock introduced in this series.

---
 Documentation/trace/ftrace.txt | 20 ++++++++++++++++++++
 1 file changed, 20 insertions(+)
Documentation/trace/ftrace.txt +20 -0
index 185c39f..5596e2d 100644
@@ -362,6 +362,26 @@ of ftrace. Here is a list of some of the key files:
 		  to correlate events across hypervisor/guest if
 		  tb_offset is known.
 
+	  mono: This uses the fast monotonic clock (CLOCK_MONOTONIC)
+		which is monotonic and is subject to NTP rate adjustments.
+
+	  mono_raw:
+		This is the raw monotonic clock (CLOCK_MONOTONIC_RAW)
+		which is montonic but is not subject to any rate adjustments
+		and ticks at the same rate as the hardware clocksource.
+
+	  boot: This is the boot clock (CLOCK_BOOTTIME) and is based on the
+		fast monotonic clock, but also accounts for time spent in
+		suspend. Since the clock access is designed for use in
+		tracing in the suspend path, some side effects are possible
+		if clock is accessed after the suspend time is accounted before
+		the fast mono clock is updated. In this case, the clock update
+		appears to happen slightly sooner than it normally would have.
+		Also on 32-bit systems, it's possible that the 64-bit boot offset
+		sees a partial update. These effects are rare and post
+		processing should be able to handle them. See comments in the
+		ktime_get_boot_fast_ns() function for more information.
+
 	To set a clock, simply echo the clock name into this file.
 
 	  echo global > trace_clock