haproxy/include/common/time.h

602 lines
20 KiB
C
Raw Normal View History

/*
* include/common/time.h
* Time calculation functions and macros.
*
* Copyright (C) 2000-2011 Willy Tarreau - w@1wt.eu
*
* This library is free software; you can redistribute it and/or
* modify it under the terms of the GNU Lesser General Public
* License as published by the Free Software Foundation, version 2.1
* exclusively.
*
* This library is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
* Lesser General Public License for more details.
*
* You should have received a copy of the GNU Lesser General Public
* License along with this library; if not, write to the Free Software
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
*/
#ifndef _COMMON_TIME_H
#define _COMMON_TIME_H
#include <inttypes.h>
#include <stdlib.h>
#include <unistd.h>
#include <sys/time.h>
#include <common/config.h>
#include <common/standard.h>
/* eternity when exprimed in timeval */
#ifndef TV_ETERNITY
#define TV_ETERNITY (~0UL)
#endif
/* eternity when exprimed in ms */
#ifndef TV_ETERNITY_MS
#define TV_ETERNITY_MS (-1)
#endif
#define TIME_ETERNITY (TV_ETERNITY_MS)
/* we want to be able to detect time jumps. Fix the maximum wait time to a low
* value so that we know the time has changed if we wait longer.
*/
#define MAX_DELAY_MS 1000
/* returns the lowest delay amongst <old> and <new>, and respects TIME_ETERNITY */
#define MINTIME(old, new) (((new)<0)?(old):(((old)<0||(new)<(old))?(new):(old)))
#define SETNOW(a) (*a=now)
extern THREAD_LOCAL unsigned int curr_sec_ms; /* millisecond of current second (0..999) */
extern THREAD_LOCAL unsigned int ms_left_scaled; /* milliseconds left for current second (0..2^32-1) */
extern THREAD_LOCAL unsigned int curr_sec_ms_scaled; /* millisecond of current second (0..2^32-1) */
extern THREAD_LOCAL unsigned int now_ms; /* internal date in milliseconds (may wrap) */
extern THREAD_LOCAL unsigned int samp_time; /* total elapsed time over current sample */
extern THREAD_LOCAL unsigned int idle_time; /* total idle time over current sample */
extern THREAD_LOCAL unsigned int idle_pct; /* idle to total ratio over last sample (percent) */
extern THREAD_LOCAL struct timeval now; /* internal date is a monotonic function of real clock */
extern THREAD_LOCAL struct timeval date; /* the real current date */
extern struct timeval start_date; /* the process's start date */
extern THREAD_LOCAL struct timeval before_poll; /* system date before calling poll() */
extern THREAD_LOCAL struct timeval after_poll; /* system date after leaving poll() */
MEDIUM: time: measure the time stolen by other threads The purpose is to detect if threads or processes are competing for the same CPU. This can happen when threads are incorrectly bound, or after a reload if the previous process still has an important activity. With threads this situation is problematic because a preempted thread holding a lock will block other ones waiting for this lock to be released. A first attempt consisted in measuring the cumulated lost time more precisely but the system's scheduler is smart enough to try to limit the thread preemption rate by mostly context switching during poll()'s blank periods, so most of the time lost is not seen. In essence this is good because it means a thread is not preempted with a lock held, and even regarding the rendez-vous point it cannot prevent the other ones from making progress. But still it happens tens to hundreds of times per second that a thread might be preempted, so it's still possible to detect that the situation is happening, thus it's interesting to measure and report its frequency. Each time we enter the poller, we check the CPU time spent working and see if we've lost time doing something else. To limit false positives, we're only interested in losses of 500 microseconds or more (i.e. half a clock tick on a 1 kHz system). If so, it indicates that some time was stolen by another thread or process. Note that we purposely store some sub-millisecond counters so that under heavy traffic with a 1 kHz clock, it's still possible to measure something without being subject to the risk of rounding errors (i.e. if exactly 1 ms is stolen it's possible that the time difference could often be slightly lower). This counter of lost CPU time slots time is reported in "show activity" in numbers of milliseconds of CPU lost per second, per 15s, and total over the process' life. By definition, the per-second counter cannot report values larger than 1000 per thread per second and the 15s one will be limited to 15000/s in the worst case, but it's possible that peak values exceed such thresholds after long pauses.
2018-10-17 17:01:24 +00:00
extern THREAD_LOCAL uint64_t prev_cpu_time; /* previous per thread CPU time */
extern THREAD_LOCAL uint64_t prev_mono_time; /* previous system wide monotonic time */
/**** exported functions *************************************************/
/*
* adds <ms> ms to <from>, set the result to <tv> and returns a pointer <tv>
*/
REGPRM3 struct timeval *tv_ms_add(struct timeval *tv, const struct timeval *from, int ms);
/*
* compares <tv1> and <tv2> modulo 1ms: returns 0 if equal, -1 if tv1 < tv2, 1 if tv1 > tv2
* Must not be used when either argument is eternity. Use tv_ms_cmp2() for that.
*/
REGPRM2 int tv_ms_cmp(const struct timeval *tv1, const struct timeval *tv2);
/*
* compares <tv1> and <tv2> modulo 1 ms: returns 0 if equal, -1 if tv1 < tv2, 1 if tv1 > tv2,
* assuming that TV_ETERNITY is greater than everything.
*/
REGPRM2 int tv_ms_cmp2(const struct timeval *tv1, const struct timeval *tv2);
/**** general purpose functions and macros *******************************/
/* tv_now: sets <tv> to the current time */
REGPRM1 static inline struct timeval *tv_now(struct timeval *tv)
{
gettimeofday(tv, NULL);
return tv;
}
/* tv_udpate_date: sets <date> to system time, and sets <now> to something as
* close as possible to real time, following a monotonic function. The main
* principle consists in detecting backwards and forwards time jumps and adjust
* an offset to correct them. This function should be called only once after
* each poll. The poll's timeout should be passed in <max_wait>, and the return
* value in <interrupted> (a non-zero value means that we have not expired the
* timeout).
*/
REGPRM2 void tv_update_date(int max_wait, int interrupted);
/*
* sets a struct timeval to its highest value so that it can never happen
* note that only tv_usec is necessary to detect it since a tv_usec > 999999
* is normally not possible.
*/
REGPRM1 static inline struct timeval *tv_eternity(struct timeval *tv)
{
tv->tv_sec = (typeof(tv->tv_sec))TV_ETERNITY;
tv->tv_usec = (typeof(tv->tv_usec))TV_ETERNITY;
return tv;
}
/*
* sets a struct timeval to 0
*
*/
REGPRM1 static inline struct timeval *tv_zero(struct timeval *tv) {
tv->tv_sec = tv->tv_usec = 0;
return tv;
}
/*
* returns non null if tv is [eternity], otherwise 0.
*/
#define tv_iseternity(tv) ((tv)->tv_usec == (typeof((tv)->tv_usec))TV_ETERNITY)
/*
* returns 0 if tv is [eternity], otherwise non-zero.
*/
#define tv_isset(tv) ((tv)->tv_usec != (typeof((tv)->tv_usec))TV_ETERNITY)
/*
* returns non null if tv is [0], otherwise 0.
*/
#define tv_iszero(tv) (((tv)->tv_sec | (tv)->tv_usec) == 0)
/*
* Converts a struct timeval to a number of milliseconds.
*/
REGPRM1 static inline unsigned long __tv_to_ms(const struct timeval *tv)
{
unsigned long ret;
ret = tv->tv_sec * 1000;
ret += tv->tv_usec / 1000;
return ret;
}
/*
* Converts a struct timeval to a number of milliseconds.
*/
REGPRM2 static inline struct timeval * __tv_from_ms(struct timeval *tv, unsigned long ms)
{
tv->tv_sec = ms / 1000;
tv->tv_usec = (ms % 1000) * 1000;
return tv;
}
/* Return a number of 1024Hz ticks between 0 and 1023 for input number of
* usecs between 0 and 999999. This function has been optimized to remove
* any divide and multiply, as it is completely optimized away by the compiler
* on CPUs which don't have a fast multiply. Its avg error rate is 305 ppm,
* which is almost twice as low as a direct usec to ms conversion. This version
* also has the benefit of returning 1024 for 1000000.
*/
REGPRM1 static inline unsigned int __usec_to_1024th(unsigned int usec)
{
return (usec * 1073 + 742516) >> 20;
}
/**** comparison functions and macros ***********************************/
/* tv_cmp: compares <tv1> and <tv2> : returns 0 if equal, -1 if tv1 < tv2, 1 if tv1 > tv2. */
REGPRM2 static inline int __tv_cmp(const struct timeval *tv1, const struct timeval *tv2)
{
if ((unsigned)tv1->tv_sec < (unsigned)tv2->tv_sec)
return -1;
else if ((unsigned)tv1->tv_sec > (unsigned)tv2->tv_sec)
return 1;
else if ((unsigned)tv1->tv_usec < (unsigned)tv2->tv_usec)
return -1;
else if ((unsigned)tv1->tv_usec > (unsigned)tv2->tv_usec)
return 1;
else
return 0;
}
/* tv_iseq: compares <tv1> and <tv2> : returns 1 if tv1 == tv2, otherwise 0 */
#define tv_iseq __tv_iseq
REGPRM2 static inline int __tv_iseq(const struct timeval *tv1, const struct timeval *tv2)
{
return ((unsigned)tv1->tv_sec == (unsigned)tv2->tv_sec) &&
((unsigned)tv1->tv_usec == (unsigned)tv2->tv_usec);
}
/* tv_isgt: compares <tv1> and <tv2> : returns 1 if tv1 > tv2, otherwise 0 */
#define tv_isgt _tv_isgt
REGPRM2 int _tv_isgt(const struct timeval *tv1, const struct timeval *tv2);
REGPRM2 static inline int __tv_isgt(const struct timeval *tv1, const struct timeval *tv2)
{
return
((unsigned)tv1->tv_sec == (unsigned)tv2->tv_sec) ?
((unsigned)tv1->tv_usec > (unsigned)tv2->tv_usec) :
((unsigned)tv1->tv_sec > (unsigned)tv2->tv_sec);
}
/* tv_isge: compares <tv1> and <tv2> : returns 1 if tv1 >= tv2, otherwise 0 */
#define tv_isge __tv_isge
REGPRM2 static inline int __tv_isge(const struct timeval *tv1, const struct timeval *tv2)
{
return
((unsigned)tv1->tv_sec == (unsigned)tv2->tv_sec) ?
((unsigned)tv1->tv_usec >= (unsigned)tv2->tv_usec) :
((unsigned)tv1->tv_sec > (unsigned)tv2->tv_sec);
}
/* tv_islt: compares <tv1> and <tv2> : returns 1 if tv1 < tv2, otherwise 0 */
#define tv_islt __tv_islt
REGPRM2 static inline int __tv_islt(const struct timeval *tv1, const struct timeval *tv2)
{
return
((unsigned)tv1->tv_sec == (unsigned)tv2->tv_sec) ?
((unsigned)tv1->tv_usec < (unsigned)tv2->tv_usec) :
((unsigned)tv1->tv_sec < (unsigned)tv2->tv_sec);
}
/* tv_isle: compares <tv1> and <tv2> : returns 1 if tv1 <= tv2, otherwise 0 */
#define tv_isle _tv_isle
REGPRM2 int _tv_isle(const struct timeval *tv1, const struct timeval *tv2);
REGPRM2 static inline int __tv_isle(const struct timeval *tv1, const struct timeval *tv2)
{
return
((unsigned)tv1->tv_sec == (unsigned)tv2->tv_sec) ?
((unsigned)tv1->tv_usec <= (unsigned)tv2->tv_usec) :
((unsigned)tv1->tv_sec < (unsigned)tv2->tv_sec);
}
/*
* compares <tv1> and <tv2> modulo 1ms: returns 0 if equal, -1 if tv1 < tv2, 1 if tv1 > tv2
* Must not be used when either argument is eternity. Use tv_ms_cmp2() for that.
*/
#define tv_ms_cmp _tv_ms_cmp
REGPRM2 int _tv_ms_cmp(const struct timeval *tv1, const struct timeval *tv2);
REGPRM2 static inline int __tv_ms_cmp(const struct timeval *tv1, const struct timeval *tv2)
{
if ((unsigned)tv1->tv_sec == (unsigned)tv2->tv_sec) {
if ((unsigned)tv2->tv_usec >= (unsigned)tv1->tv_usec + 1000)
return -1;
else if ((unsigned)tv1->tv_usec >= (unsigned)tv2->tv_usec + 1000)
return 1;
else
return 0;
}
else if (((unsigned)tv2->tv_sec > (unsigned)tv1->tv_sec + 1) ||
(((unsigned)tv2->tv_sec == (unsigned)tv1->tv_sec + 1) &&
((unsigned)tv2->tv_usec + 1000000 >= (unsigned)tv1->tv_usec + 1000)))
return -1;
else if (((unsigned)tv1->tv_sec > (unsigned)tv2->tv_sec + 1) ||
(((unsigned)tv1->tv_sec == (unsigned)tv2->tv_sec + 1) &&
((unsigned)tv1->tv_usec + 1000000 >= (unsigned)tv2->tv_usec + 1000)))
return 1;
else
return 0;
}
/*
* compares <tv1> and <tv2> modulo 1 ms: returns 0 if equal, -1 if tv1 < tv2, 1 if tv1 > tv2,
* assuming that TV_ETERNITY is greater than everything.
*/
#define tv_ms_cmp2 _tv_ms_cmp2
REGPRM2 int _tv_ms_cmp2(const struct timeval *tv1, const struct timeval *tv2);
REGPRM2 static inline int __tv_ms_cmp2(const struct timeval *tv1, const struct timeval *tv2)
{
if (tv_iseternity(tv1))
if (tv_iseternity(tv2))
return 0; /* same */
else
return 1; /* tv1 later than tv2 */
else if (tv_iseternity(tv2))
return -1; /* tv2 later than tv1 */
return tv_ms_cmp(tv1, tv2);
}
/*
* compares <tv1> and <tv2> modulo 1 ms: returns 1 if tv1 <= tv2, 0 if tv1 > tv2,
* assuming that TV_ETERNITY is greater than everything. Returns 0 if tv1 is
* TV_ETERNITY, and always assumes that tv2 != TV_ETERNITY. Designed to replace
* occurrences of (tv_ms_cmp2(tv,now) <= 0).
*/
#define tv_ms_le2 _tv_ms_le2
REGPRM2 int _tv_ms_le2(const struct timeval *tv1, const struct timeval *tv2);
REGPRM2 static inline int __tv_ms_le2(const struct timeval *tv1, const struct timeval *tv2)
{
if (likely((unsigned)tv1->tv_sec > (unsigned)tv2->tv_sec + 1))
return 0;
if (likely((unsigned)tv1->tv_sec < (unsigned)tv2->tv_sec))
return 1;
if (likely((unsigned)tv1->tv_sec == (unsigned)tv2->tv_sec)) {
if ((unsigned)tv2->tv_usec >= (unsigned)tv1->tv_usec + 1000)
return 1;
else
return 0;
}
if (unlikely(((unsigned)tv1->tv_sec == (unsigned)tv2->tv_sec + 1) &&
((unsigned)tv1->tv_usec + 1000000 >= (unsigned)tv2->tv_usec + 1000)))
return 0;
else
return 1;
}
/**** operators **********************************************************/
/*
* Returns the time in ms elapsed between tv1 and tv2, assuming that tv1<=tv2.
* Must not be used when either argument is eternity.
*/
#define tv_ms_elapsed __tv_ms_elapsed
REGPRM2 unsigned long _tv_ms_elapsed(const struct timeval *tv1, const struct timeval *tv2);
REGPRM2 static inline unsigned long __tv_ms_elapsed(const struct timeval *tv1, const struct timeval *tv2)
{
unsigned long ret;
ret = ((signed long)(tv2->tv_sec - tv1->tv_sec)) * 1000;
ret += ((signed long)(tv2->tv_usec - tv1->tv_usec)) / 1000;
return ret;
}
/*
* returns the remaining time between tv1=now and event=tv2
* if tv2 is passed, 0 is returned.
* Must not be used when either argument is eternity.
*/
#define tv_ms_remain __tv_ms_remain
REGPRM2 unsigned long _tv_ms_remain(const struct timeval *tv1, const struct timeval *tv2);
REGPRM2 static inline unsigned long __tv_ms_remain(const struct timeval *tv1, const struct timeval *tv2)
{
if (tv_ms_cmp(tv1, tv2) >= 0)
return 0; /* event elapsed */
return __tv_ms_elapsed(tv1, tv2);
}
/*
* returns the remaining time between tv1=now and event=tv2
* if tv2 is passed, 0 is returned.
* Returns TIME_ETERNITY if tv2 is eternity.
*/
#define tv_ms_remain2 _tv_ms_remain2
REGPRM2 unsigned long _tv_ms_remain2(const struct timeval *tv1, const struct timeval *tv2);
REGPRM2 static inline unsigned long __tv_ms_remain2(const struct timeval *tv1, const struct timeval *tv2)
{
if (tv_iseternity(tv2))
return TIME_ETERNITY;
return tv_ms_remain(tv1, tv2);
}
/*
* adds <inc> to <from>, set the result to <tv> and returns a pointer <tv>
*/
#define tv_add _tv_add
REGPRM3 struct timeval *_tv_add(struct timeval *tv, const struct timeval *from, const struct timeval *inc);
REGPRM3 static inline struct timeval *__tv_add(struct timeval *tv, const struct timeval *from, const struct timeval *inc)
{
tv->tv_usec = from->tv_usec + inc->tv_usec;
tv->tv_sec = from->tv_sec + inc->tv_sec;
if (tv->tv_usec >= 1000000) {
tv->tv_usec -= 1000000;
tv->tv_sec++;
}
return tv;
}
/*
* If <inc> is set, then add it to <from> and set the result to <tv>, then
* return 1, otherwise return 0. It is meant to be used in if conditions.
*/
#define tv_add_ifset _tv_add_ifset
REGPRM3 int _tv_add_ifset(struct timeval *tv, const struct timeval *from, const struct timeval *inc);
REGPRM3 static inline int __tv_add_ifset(struct timeval *tv, const struct timeval *from, const struct timeval *inc)
{
if (tv_iseternity(inc))
return 0;
tv->tv_usec = from->tv_usec + inc->tv_usec;
tv->tv_sec = from->tv_sec + inc->tv_sec;
if (tv->tv_usec >= 1000000) {
tv->tv_usec -= 1000000;
tv->tv_sec++;
}
return 1;
}
/*
* adds <inc> to <tv> and returns a pointer <tv>
*/
REGPRM2 static inline struct timeval *__tv_add2(struct timeval *tv, const struct timeval *inc)
{
tv->tv_usec += inc->tv_usec;
tv->tv_sec += inc->tv_sec;
if (tv->tv_usec >= 1000000) {
tv->tv_usec -= 1000000;
tv->tv_sec++;
}
return tv;
}
/*
* Computes the remaining time between tv1=now and event=tv2. if tv2 is passed,
* 0 is returned. The result is stored into tv.
*/
#define tv_remain _tv_remain
REGPRM3 struct timeval *_tv_remain(const struct timeval *tv1, const struct timeval *tv2, struct timeval *tv);
REGPRM3 static inline struct timeval *__tv_remain(const struct timeval *tv1, const struct timeval *tv2, struct timeval *tv)
{
tv->tv_usec = tv2->tv_usec - tv1->tv_usec;
tv->tv_sec = tv2->tv_sec - tv1->tv_sec;
if ((signed)tv->tv_sec > 0) {
if ((signed)tv->tv_usec < 0) {
tv->tv_usec += 1000000;
tv->tv_sec--;
}
} else if (tv->tv_sec == 0) {
if ((signed)tv->tv_usec < 0)
tv->tv_usec = 0;
} else {
tv->tv_sec = 0;
tv->tv_usec = 0;
}
return tv;
}
/*
* Computes the remaining time between tv1=now and event=tv2. if tv2 is passed,
* 0 is returned. The result is stored into tv. Returns ETERNITY if tv2 is
* eternity.
*/
#define tv_remain2 _tv_remain2
REGPRM3 struct timeval *_tv_remain2(const struct timeval *tv1, const struct timeval *tv2, struct timeval *tv);
REGPRM3 static inline struct timeval *__tv_remain2(const struct timeval *tv1, const struct timeval *tv2, struct timeval *tv)
{
if (tv_iseternity(tv2))
return tv_eternity(tv);
return __tv_remain(tv1, tv2, tv);
}
/*
* adds <ms> ms to <from>, set the result to <tv> and returns a pointer <tv>
*/
#define tv_ms_add _tv_ms_add
REGPRM3 struct timeval *_tv_ms_add(struct timeval *tv, const struct timeval *from, int ms);
REGPRM3 static inline struct timeval *__tv_ms_add(struct timeval *tv, const struct timeval *from, int ms)
{
tv->tv_usec = from->tv_usec + (ms % 1000) * 1000;
tv->tv_sec = from->tv_sec + (ms / 1000);
while (tv->tv_usec >= 1000000) {
tv->tv_usec -= 1000000;
tv->tv_sec++;
}
return tv;
}
/*
* compares <tv1> and <tv2> : returns 1 if <tv1> is before <tv2>, otherwise 0.
* This should be very fast because it's used in schedulers.
* It has been optimized to return 1 (so call it in a loop which continues
* as long as tv1<=tv2)
*/
#define tv_isbefore(tv1, tv2) \
(unlikely((unsigned)(tv1)->tv_sec < (unsigned)(tv2)->tv_sec) ? 1 : \
(unlikely((unsigned)(tv1)->tv_sec > (unsigned)(tv2)->tv_sec) ? 0 : \
unlikely((unsigned)(tv1)->tv_usec < (unsigned)(tv2)->tv_usec)))
/*
* returns the first event between <tv1> and <tv2> into <tvmin>.
* a zero tv is ignored. <tvmin> is returned. If <tvmin> is known
* to be the same as <tv1> or <tv2>, it is recommended to use
* tv_bound instead.
*/
#define tv_min(tvmin, tv1, tv2) ({ \
if (tv_isbefore(tv1, tv2)) { \
*tvmin = *tv1; \
} \
else { \
*tvmin = *tv2; \
} \
tvmin; \
})
/*
* returns the first event between <tv1> and <tv2> into <tvmin>.
* a zero tv is ignored. <tvmin> is returned. This function has been
* optimized to be called as tv_min(a,a,b) or tv_min(b,a,b).
*/
#define tv_bound(tv1, tv2) ({ \
if (tv_isbefore(tv2, tv1)) \
*tv1 = *tv2; \
tv1; \
})
/* returns the system's monotonic time in nanoseconds if supported, otherwise zero */
static inline uint64_t now_mono_time()
{
#if defined(_POSIX_TIMERS) && defined(_POSIX_MONOTONIC_CLOCK)
struct timespec ts;
clock_gettime(CLOCK_MONOTONIC, &ts);
return ts.tv_sec * 1000000000ULL + ts.tv_nsec;
#else
return 0;
#endif
}
/* returns the current thread's cumulated CPU time in nanoseconds if supported, otherwise zero */
static inline uint64_t now_cpu_time()
{
#if defined(_POSIX_TIMERS) && defined(_POSIX_THREAD_CPUTIME)
struct timespec ts;
clock_gettime(CLOCK_THREAD_CPUTIME_ID, &ts);
return ts.tv_sec * 1000000000ULL + ts.tv_nsec;
#else
return 0;
#endif
}
/* Update the idle time value twice a second, to be called after
* tv_update_date() when called after poll(). It relies on <before_poll> to be
* updated to the system time before calling poll().
*/
static inline void measure_idle()
{
/* Let's compute the idle to work ratio. We worked between after_poll
* and before_poll, and slept between before_poll and date. The idle_pct
* is updated at most twice every second. Note that the current second
* rarely changes so we avoid a multiply when not needed.
*/
int delta;
if ((delta = date.tv_sec - before_poll.tv_sec))
delta *= 1000000;
idle_time += delta + (date.tv_usec - before_poll.tv_usec);
if ((delta = date.tv_sec - after_poll.tv_sec))
delta *= 1000000;
samp_time += delta + (date.tv_usec - after_poll.tv_usec);
after_poll.tv_sec = date.tv_sec; after_poll.tv_usec = date.tv_usec;
if (samp_time < 500000)
return;
idle_pct = (100 * idle_time + samp_time / 2) / samp_time;
idle_time = samp_time = 0;
}
/* Collect date and time information before calling poll(). This will be used
* to count the run time of the past loop and the sleep time of the next poll.
*/
static inline void tv_entering_poll()
{
gettimeofday(&before_poll, NULL);
}
/* Collect date and time information after leaving poll(). <timeout> must be
* set to the maximum sleep time passed to poll (in milliseconds), and
* <interrupted> must be zero if the poller reached the timeout or non-zero
* otherwise, which generally is provided by the poller's return value.
*/
static inline void tv_leaving_poll(int timeout, int interrupted)
{
measure_idle();
MEDIUM: time: measure the time stolen by other threads The purpose is to detect if threads or processes are competing for the same CPU. This can happen when threads are incorrectly bound, or after a reload if the previous process still has an important activity. With threads this situation is problematic because a preempted thread holding a lock will block other ones waiting for this lock to be released. A first attempt consisted in measuring the cumulated lost time more precisely but the system's scheduler is smart enough to try to limit the thread preemption rate by mostly context switching during poll()'s blank periods, so most of the time lost is not seen. In essence this is good because it means a thread is not preempted with a lock held, and even regarding the rendez-vous point it cannot prevent the other ones from making progress. But still it happens tens to hundreds of times per second that a thread might be preempted, so it's still possible to detect that the situation is happening, thus it's interesting to measure and report its frequency. Each time we enter the poller, we check the CPU time spent working and see if we've lost time doing something else. To limit false positives, we're only interested in losses of 500 microseconds or more (i.e. half a clock tick on a 1 kHz system). If so, it indicates that some time was stolen by another thread or process. Note that we purposely store some sub-millisecond counters so that under heavy traffic with a 1 kHz clock, it's still possible to measure something without being subject to the risk of rounding errors (i.e. if exactly 1 ms is stolen it's possible that the time difference could often be slightly lower). This counter of lost CPU time slots time is reported in "show activity" in numbers of milliseconds of CPU lost per second, per 15s, and total over the process' life. By definition, the per-second counter cannot report values larger than 1000 per thread per second and the 15s one will be limited to 15000/s in the worst case, but it's possible that peak values exceed such thresholds after long pauses.
2018-10-17 17:01:24 +00:00
prev_cpu_time = now_cpu_time();
prev_mono_time = now_mono_time();
}
#endif /* _COMMON_TIME_H */
/*
* Local variables:
* c-indent-level: 8
* c-basic-offset: 8
* End:
*/