- moving default_timer_slack_ns from per-task field to global variable controlled via sysctl() in [0..1000000] range (usecs) provides some more system-wide control beyond per-task prctl();
- task_time_slack() should be used to get default timer slack if no special requirements exists.
Signed-off-by: Dmitry Antipov dmitry.antipov@linaro.org --- include/linux/hrtimer.h | 1 + include/linux/sched.h | 11 ++++++++--- kernel/fork.c | 3 --- kernel/futex.c | 4 ++-- kernel/hrtimer.c | 10 +++++++--- kernel/sys.c | 3 ++- kernel/sysctl.c | 10 ++++++++++ 7 files changed, 30 insertions(+), 12 deletions(-)
diff --git a/include/linux/hrtimer.h b/include/linux/hrtimer.h index fd0dc30..855d45e 100644 --- a/include/linux/hrtimer.h +++ b/include/linux/hrtimer.h @@ -323,6 +323,7 @@ extern ktime_t ktime_get_monotonic_offset(void);
DECLARE_PER_CPU(struct tick_device, tick_cpu_device);
+extern int default_timer_slack_ns;
/* Exported timer functions: */
diff --git a/include/linux/sched.h b/include/linux/sched.h index 2234985..4f475aa 100644 --- a/include/linux/sched.h +++ b/include/linux/sched.h @@ -1551,11 +1551,11 @@ struct task_struct { struct latency_record latency_record[LT_SAVECOUNT]; #endif /* - * time slack values; these are used to round up poll() and - * select() etc timeout values. These are in nanoseconds. + * High-resolution timer slack value, in nanoseconds. + * Used to round up poll()/select(), nanosleep, futex + * waiting, etc. timeout values for non-realtime tasks. */ unsigned long timer_slack_ns; - unsigned long default_timer_slack_ns;
struct list_head *scm_work_list; #ifdef CONFIG_FUNCTION_GRAPH_TRACER @@ -2622,6 +2622,11 @@ static inline int spin_needbreak(spinlock_t *lock) #endif }
+static inline unsigned long task_timer_slack(struct task_struct *tsk) +{ + return rt_task(tsk) ? 0 : tsk->timer_slack_ns; +} + /* * Thread group CPU time accounting. */ diff --git a/kernel/fork.c b/kernel/fork.c index 051f090..b0e8f63 100644 --- a/kernel/fork.c +++ b/kernel/fork.c @@ -1144,9 +1144,6 @@ static struct task_struct *copy_process(unsigned long clone_flags, #if defined(SPLIT_RSS_COUNTING) memset(&p->rss_stat, 0, sizeof(p->rss_stat)); #endif - - p->default_timer_slack_ns = current->timer_slack_ns; - task_io_accounting_init(&p->ioac); acct_clear_integrals(p);
diff --git a/kernel/futex.c b/kernel/futex.c index 1614be2..a0d302d 100644 --- a/kernel/futex.c +++ b/kernel/futex.c @@ -1887,7 +1887,7 @@ static int futex_wait(u32 __user *uaddr, unsigned int flags, u32 val, HRTIMER_MODE_ABS); hrtimer_init_sleeper(to, current); hrtimer_set_expires_range_ns(&to->timer, *abs_time, - current->timer_slack_ns); + task_timer_slack(current)); }
retry: @@ -2281,7 +2281,7 @@ static int futex_wait_requeue_pi(u32 __user *uaddr, unsigned int flags, HRTIMER_MODE_ABS); hrtimer_init_sleeper(to, current); hrtimer_set_expires_range_ns(&to->timer, *abs_time, - current->timer_slack_ns); + task_timer_slack(current)); }
/* diff --git a/kernel/hrtimer.c b/kernel/hrtimer.c index ae34bf5..0704559 100644 --- a/kernel/hrtimer.c +++ b/kernel/hrtimer.c @@ -51,6 +51,12 @@ #include <trace/events/timer.h>
/* + * Default hrtimer slack value, in nanoseconds. May be + * changed via sysctl within [0..1000000] range. + */ +int default_timer_slack_ns = 50000; + +/* * The timer bases: * * There are more clockids then hrtimer bases. Thus, we index @@ -1564,9 +1570,7 @@ long hrtimer_nanosleep(struct timespec *rqtp, struct timespec __user *rmtp, int ret = 0; unsigned long slack;
- slack = current->timer_slack_ns; - if (rt_task(current)) - slack = 0; + slack = task_timer_slack(current);
hrtimer_init_on_stack(&t.timer, clockid, mode); hrtimer_set_expires_range_ns(&t.timer, timespec_to_ktime(*rqtp), slack); diff --git a/kernel/sys.c b/kernel/sys.c index 4070153..db36543 100644 --- a/kernel/sys.c +++ b/kernel/sys.c @@ -22,6 +22,7 @@ #include <linux/device.h> #include <linux/key.h> #include <linux/times.h> +#include <linux/hrtimer.h> #include <linux/posix-timers.h> #include <linux/security.h> #include <linux/dcookies.h> @@ -1919,7 +1920,7 @@ SYSCALL_DEFINE5(prctl, int, option, unsigned long, arg2, unsigned long, arg3, case PR_SET_TIMERSLACK: if (arg2 <= 0) current->timer_slack_ns = - current->default_timer_slack_ns; + default_timer_slack_ns; else current->timer_slack_ns = arg2; error = 0; diff --git a/kernel/sysctl.c b/kernel/sysctl.c index f487f25..c9a731b 100644 --- a/kernel/sysctl.c +++ b/kernel/sysctl.c @@ -122,6 +122,7 @@ static int __maybe_unused two = 2; static int __maybe_unused three = 3; static unsigned long one_ul = 1; static int one_hundred = 100; +static int million = 1000000; #ifdef CONFIG_PRINTK static int ten_thousand = 10000; #endif @@ -1004,6 +1005,15 @@ static struct ctl_table kern_table[] = { .proc_handler = proc_dointvec, }, #endif + { + .procname = "timer_slack", + .data = &default_timer_slack_ns, + .maxlen = sizeof(int), + .mode = 0644, + .proc_handler = proc_dointvec_minmax, + .extra1 = &zero, + .extra2 = &million, + }, { } };
Dmitry,
On Fri, Jan 27, 2012 at 12:30 PM, Dmitry Antipov dmitry.antipov@linaro.org wrote:
- moving default_timer_slack_ns from per-task field to global variable controlled via sysctl() in [0..1000000] range (usecs) provides some more system-wide control beyond per-task prctl();
What benefits do I get from this change? In other words, don't just describe the change, please describe why it is desirable.
Have you done any tests where this improves things - reduction in wakeups, increase in throughput? Those results will also help in getting more attention on the patch.
- task_time_slack() should be used to get default timer slack if no special requirements exists.
Signed-off-by: Dmitry Antipov dmitry.antipov@linaro.org
include/linux/hrtimer.h | 1 + include/linux/sched.h | 11 ++++++++--- kernel/fork.c | 3 --- kernel/futex.c | 4 ++-- kernel/hrtimer.c | 10 +++++++--- kernel/sys.c | 3 ++- kernel/sysctl.c | 10 ++++++++++ 7 files changed, 30 insertions(+), 12 deletions(-)
diff --git a/include/linux/hrtimer.h b/include/linux/hrtimer.h index fd0dc30..855d45e 100644 --- a/include/linux/hrtimer.h +++ b/include/linux/hrtimer.h @@ -323,6 +323,7 @@ extern ktime_t ktime_get_monotonic_offset(void);
DECLARE_PER_CPU(struct tick_device, tick_cpu_device);
+extern int default_timer_slack_ns;
/* Exported timer functions: */
diff --git a/include/linux/sched.h b/include/linux/sched.h index 2234985..4f475aa 100644 --- a/include/linux/sched.h +++ b/include/linux/sched.h @@ -1551,11 +1551,11 @@ struct task_struct { struct latency_record latency_record[LT_SAVECOUNT]; #endif /*
- * time slack values; these are used to round up poll() and
- * select() etc timeout values. These are in nanoseconds.
- * High-resolution timer slack value, in nanoseconds.
- * Used to round up poll()/select(), nanosleep, futex
- * waiting, etc. timeout values for non-realtime tasks.
*/ unsigned long timer_slack_ns;
- unsigned long default_timer_slack_ns;
struct list_head *scm_work_list; #ifdef CONFIG_FUNCTION_GRAPH_TRACER @@ -2622,6 +2622,11 @@ static inline int spin_needbreak(spinlock_t *lock) #endif }
+static inline unsigned long task_timer_slack(struct task_struct *tsk) +{
- return rt_task(tsk) ? 0 : tsk->timer_slack_ns;
+}
/* * Thread group CPU time accounting. */ diff --git a/kernel/fork.c b/kernel/fork.c index 051f090..b0e8f63 100644 --- a/kernel/fork.c +++ b/kernel/fork.c @@ -1144,9 +1144,6 @@ static struct task_struct *copy_process(unsigned long clone_flags, #if defined(SPLIT_RSS_COUNTING) memset(&p->rss_stat, 0, sizeof(p->rss_stat)); #endif
- p->default_timer_slack_ns = current->timer_slack_ns;
task_io_accounting_init(&p->ioac); acct_clear_integrals(p);
diff --git a/kernel/futex.c b/kernel/futex.c index 1614be2..a0d302d 100644 --- a/kernel/futex.c +++ b/kernel/futex.c @@ -1887,7 +1887,7 @@ static int futex_wait(u32 __user *uaddr, unsigned int flags, u32 val, HRTIMER_MODE_ABS); hrtimer_init_sleeper(to, current); hrtimer_set_expires_range_ns(&to->timer, *abs_time,
- current->timer_slack_ns);
- task_timer_slack(current));
}
retry: @@ -2281,7 +2281,7 @@ static int futex_wait_requeue_pi(u32 __user *uaddr, unsigned int flags, HRTIMER_MODE_ABS); hrtimer_init_sleeper(to, current); hrtimer_set_expires_range_ns(&to->timer, *abs_time,
- current->timer_slack_ns);
- task_timer_slack(current));
}
/* diff --git a/kernel/hrtimer.c b/kernel/hrtimer.c index ae34bf5..0704559 100644 --- a/kernel/hrtimer.c +++ b/kernel/hrtimer.c @@ -51,6 +51,12 @@ #include <trace/events/timer.h>
/*
- Default hrtimer slack value, in nanoseconds. May be
- changed via sysctl within [0..1000000] range.
- */
+int default_timer_slack_ns = 50000;
+/* * The timer bases: * * There are more clockids then hrtimer bases. Thus, we index @@ -1564,9 +1570,7 @@ long hrtimer_nanosleep(struct timespec *rqtp, struct timespec __user *rmtp, int ret = 0; unsigned long slack;
- slack = current->timer_slack_ns;
- if (rt_task(current))
- slack = 0;
- slack = task_timer_slack(current);
hrtimer_init_on_stack(&t.timer, clockid, mode); hrtimer_set_expires_range_ns(&t.timer, timespec_to_ktime(*rqtp), slack); diff --git a/kernel/sys.c b/kernel/sys.c index 4070153..db36543 100644 --- a/kernel/sys.c +++ b/kernel/sys.c @@ -22,6 +22,7 @@ #include <linux/device.h> #include <linux/key.h> #include <linux/times.h> +#include <linux/hrtimer.h> #include <linux/posix-timers.h> #include <linux/security.h> #include <linux/dcookies.h> @@ -1919,7 +1920,7 @@ SYSCALL_DEFINE5(prctl, int, option, unsigned long, arg2, unsigned long, arg3, case PR_SET_TIMERSLACK: if (arg2 <= 0) current->timer_slack_ns =
- current->default_timer_slack_ns;
- default_timer_slack_ns;
else current->timer_slack_ns = arg2; error = 0; diff --git a/kernel/sysctl.c b/kernel/sysctl.c index f487f25..c9a731b 100644 --- a/kernel/sysctl.c +++ b/kernel/sysctl.c @@ -122,6 +122,7 @@ static int __maybe_unused two = 2; static int __maybe_unused three = 3; static unsigned long one_ul = 1; static int one_hundred = 100; +static int million = 1000000; #ifdef CONFIG_PRINTK static int ten_thousand = 10000; #endif @@ -1004,6 +1005,15 @@ static struct ctl_table kern_table[] = { .proc_handler = proc_dointvec, }, #endif
- {
- .procname = "timer_slack",
- .data = &default_timer_slack_ns,
- .maxlen = sizeof(int),
- .mode = 0644,
- .proc_handler = proc_dointvec_minmax,
- .extra1 = &zero,
- .extra2 = &million,
- },
{ } };
-- 1.7.7.5
-- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
-
Amit Kucheria -
Dmitry Antipov