linux 时间相关的一些总结

仅作为内核代码中时间管理模块的笔记,3.10内核,很乱,不喜勿喷。

先有time,后有timer。

常用的time结构有哪些?除了大名鼎鼎的jiffies和jiffies64之外,还有常用的一些结构如下:

ktime_t 经常用在timer中,
union ktime {
    s64    tv64;
#if BITS_PER_LONG != 64 && !defined(CONFIG_KTIME_SCALAR)
    struct {
# ifdef __BIG_ENDIAN
    s32    sec, nsec;
# else
    s32    nsec, sec;
# endif
    } tv;
#endif
};

typedef union ktime ktime_t;        /* Kill this */

经常用在fs中的timespec,低一点精度的timeval,以及时区结构timezone。主要用来做时间戳等。

struct timespec {
    __kernel_time_t    tv_sec;            /* seconds */
    long        tv_nsec;        /* nanoseconds */
};
struct timeval {
    __kernel_time_t     tv_sec;     /* seconds */
    __kernel_suseconds_t    tv_usec;    /* microseconds */
};

struct timezone {
    int tz_minuteswest; /* minutes west of Greenwich */
    int tz_dsttime; /* type of dst correction */
};
 

这些结构之间的常用转换函数:

/* convert a timespec to ktime_t format: */
static inline ktime_t timespec_to_ktime(struct timespec ts)
{
    return ktime_set(ts.tv_sec, ts.tv_nsec);
}

/* convert a timespec64 to ktime_t format: */
static inline ktime_t timespec64_to_ktime(struct timespec64 ts)
{
    return ktime_set(ts.tv_sec, ts.tv_nsec);
}

/* convert a timeval to ktime_t format: */
static inline ktime_t timeval_to_ktime(struct timeval tv)
{
    return ktime_set(tv.tv_sec, tv.tv_usec * NSEC_PER_USEC);
}

/* Map the ktime_t to timespec conversion to ns_to_timespec function */
#define ktime_to_timespec(kt)        ns_to_timespec((kt).tv64)

/* Map the ktime_t to timespec conversion to ns_to_timespec function */
#define ktime_to_timespec64(kt)        ns_to_timespec64((kt).tv64)

/* Map the ktime_t to timeval conversion to ns_to_timeval function */
#define ktime_to_timeval(kt)        ns_to_timeval((kt).tv64)

/* Convert ktime_t to nanoseconds - NOP in the scalar storage format: */
#define ktime_to_ns(kt)            ((kt).tv64)

比如有时候自己不想那么高精度的时间戳怎么办呢?内核还提供了这个函数,取到秒级,最方便的是这个函数还被导出了,很好用。

unsigned long get_seconds(void)
{
    struct timekeeper *tk = &timekeeper;

    return tk->xtime_sec;
}
EXPORT_SYMBOL(get_seconds);

还有个有趣的问题是,这个时间的维护,精度要更高的话,就需要用顺序锁去读取 timekeeper 变量。

struct timespec current_kernel_time(void)
{
    struct timekeeper *tk = &timekeeper;
    struct timespec64 now;
    unsigned long seq;

    do {
        seq = read_seqcount_begin(&timekeeper_seq);

        now = tk_xtime(tk);
    } while (read_seqcount_retry(&timekeeper_seq, seq));

    return timespec64_to_timespec(now);
}
EXPORT_SYMBOL(current_kernel_time);

好了,time除了用来做时间戳之前,另外一个大的应用就是timer的超时时间了。在描述timer之前,有必要描述linux 关于时间管理的几个大的概念,

低精度的timer定义:

crash> tvec_base
struct tvec_base {
    spinlock_t lock;
    struct timer_list *running_timer;
    unsigned long timer_jiffies;
    unsigned long next_timer;
    unsigned long active_timers;
    struct tvec_root tv1;
    struct tvec tv2;
    struct tvec tv3;
    struct tvec tv4;
    struct tvec tv5;
    unsigned long all_timers;
}

 低精度定时器结构:

struct timer_list {---------------------低精度定时器结构,
    struct list_head entry;-------------用这个挂入到时间轮的链表中,与高精度的rb_node类比
    unsigned long expires;--------------超期时间
    struct tvec_base *base;-------------指向某个cpu的 tvec_base
    void (*function)(unsigned long);----回调
    unsigned long data;
    int slack;
    int start_pid;
    void *start_site;
    char start_comm[16];
}

常用的配套函数有:add_timer,mod_timer,add_timer_on(指定cpu添加timer),del_timer,DEFINE_TIMER,setup_timer等,这些在协议栈代码里面非常常见,一般用来等待超时。既然是超时,那么对时间精度要求就不那么高了,所以实现的时候,用了著名的定时器轮。

add_timer的流程和mod_timer的流程差不多,先判断该timer是不是pending,pending的意思就是从定时器轮已经摘取了,可能正在执行中,它的特征就是 该timer的 entry的next是否为NULL

static inline int timer_pending(const struct timer_list * timer)
{
    return timer->entry.next != NULL;
}

一句话总结:正等待被调度执行的定时器对象就是pending的。如果一个定时器不是pending的,那么肯定在定时器轮上。

接下来,自然要先从原来的位置摘除,

static inline void detach_timer(struct timer_list *timer, bool clear_pending)
{
    struct list_head *entry = &timer->entry;

    debug_deactivate(timer);

    __list_del(entry->prev, entry->next);-----如果timer以前没加入在定时器轮中,则这个啥都不做。
    if (clear_pending)
        entry->next = NULL;
    entry->prev = LIST_POISON2;
}

然后根据这个定时器的超时时间,加入到定时器轮中对应的vec中,主要改动两个,一个是timer的base,还有一个是timer的entry的所处的位置。

crash> p tvec_bases:0
per_cpu(tvec_bases, 0) = $30 = (struct tvec_base *) 0xffffffff81ea71c0 <boot_tvec_bases>
crash> tvec_bases
PER-CPU DATA TYPE:
  struct tvec_base *tvec_bases;
PER-CPU ADDRESSES:
  [0]: ffff8827dca13948
  [1]: ffff8827dca53948
  [2]: ffff8827dca93948
  [3]: ffff8827dcad3948
  [4]: ffff8827dcb13948
  [5]: ffff8827dcb53948
  [6]: ffff8827dcb93948
。。。。

 这里还有一个细节,就是timer的base,由于这个是一个指针,所以至少是4字节对齐的,也就是后面两位肯定为0,被用来做标记了,当从timer中取这个base指针的时候,就需要将这两

位处理掉,不能直接用来解引用,否则会出现访问错误。

由于低精度的定时器是以jiffies来作为最低精度的,所以精度有限制,但随着硬件以及多媒体发展的实时性较高的要求,后来,又引入了高精度定时器。它是以纳秒为精度的。高精度定时器结构如下:

crash> hrtimer
struct hrtimer {
    struct timerqueue_node node;---------------------------用来插入到红黑树中
    ktime_t _softexpires;----------------------------------超期的时间
    enum hrtimer_restart (*function)(struct hrtimer *);----回调函数,肯定都有,不过它的返回值只有两个
    struct hrtimer_clock_base *base;-----------------------和低精度定时器类似,也有指向一个percpu的base的一个指针,不过base结构与低精度定时器time_list不同
    unsigned long state;
    int start_pid;
    void *start_site;
    char start_comm[16];
}
SIZE: 96

它指向的base是percpu的 hrtimer_bases,注意和低精度定时器的base相区别,因为低精度的base是percpu的 tvec_base

 而高精度定时器的索引,也不是低精度那个vec管理,而是红黑树来管理的。

crash> timerqueue_head
struct timerqueue_head {
    struct rb_root head;
    struct timerqueue_node *next;
}
SIZE: 16
crash> hrtimer_clock_base
struct hrtimer_clock_base {
    struct hrtimer_cpu_base *cpu_base;
    int index;
    clockid_t clockid;
    struct timerqueue_head active;------------管理同类型的hrtimer的红黑树封装
    ktime_t resolution;
    ktime_t (*get_time)(void);
    ktime_t rh_reserved_softirq_time;
    ktime_t offset;
}

crash> hrtimer_cpu_base
struct hrtimer_cpu_base {
    raw_spinlock_t lock;
    unsigned int active_bases;
    unsigned int clock_was_set;
    ktime_t expires_next;
    int hres_active;
    int hang_detected;
    unsigned long nr_events;
    unsigned long nr_retries;
    unsigned long nr_hangs;
    ktime_t max_hang_time;
    struct hrtimer_clock_base clock_base[4];-------------它的地位,和时间轮的vec相当,是用来管理timer的,通过clockid来分类

int cpu;
    int in_hrtirq; 
}

相应的percpu管理结构,与低精度的tvec_base相对比:

crash> hrtimer_bases------------整个hrtimer_interrupt都是以这个变量为基础
PER-CPU DATA TYPE:
  struct hrtimer_cpu_base hrtimer_bases;
PER-CPU ADDRESSES:
  [0]: ffff8827dca13960
  [1]: ffff8827dca53960
  [2]: ffff8827dca93960
  [3]: ffff8827dcad3960
  [4]: ffff8827dcb13960
  [5]: ffff8827dcb53960
  [6]: ffff8827dcb93960
  [7]: ffff8827dcbd3960
  [8]: ffff8827dcc13960
  [9]: ffff8827dcc53960
  [10]: ffff8827dcc93960
。。。。
 

crash> p hrtimer_bases:0
per_cpu(hrtimer_bases, 0) = $16 = {
  lock = {
    raw_lock = {
      val = {
        counter = 0
      }
    }
  },
  active_bases = 3,
  clock_was_set = 6,
  expires_next = {
    tv64 = 558945095814132
  },
  hres_active = 1,
  hang_detected = 0,
  nr_events = 2303159495,
  nr_retries = 5938805,
  nr_hangs = 5,
  max_hang_time = {
    tv64 = 21681
  },
  clock_base = {{
      cpu_base = 0xffff8827dca13960,
      index = 0,
      clockid = 1,
      active = {
        head = {
          rb_node = 0xffff881677e57e88
        },
        next = 0xffffe8d01d20f220
      },
      resolution = {
        tv64 = 1
      },
      get_time = 0xffffffff810f0670 <ktime_get>,
      rh_reserved_softirq_time = {
        tv64 = 0
      },
      offset = {
        tv64 = 0
      }
    }, {
      cpu_base = 0xffff8827dca13960,
      index = 1,
      clockid = 0,
      active = {
        head = {
          rb_node = 0xffff881c433fbd38
        },
        next = 0xffff884a744a7d38
      },
      resolution = {
        tv64 = 1
      },
      get_time = 0xffffffff810f0ad0 <ktime_get_real>,
      rh_reserved_softirq_time = {
        tv64 = 0
      },
      offset = {
        tv64 = 1540819482621868102
      }
    }, {
      cpu_base = 0xffff8827dca13960,
      index = 2,
      clockid = 7,
      active = {
        head = {
          rb_node = 0x0
        },
        next = 0x0
      },
      resolution = {
        tv64 = 1
      },
      get_time = 0xffffffff810f0c40 <ktime_get_boottime>,
      rh_reserved_softirq_time = {
        tv64 = 0
      },
      offset = {
        tv64 = 0
      }
    }, {
      cpu_base = 0xffff8827dca13960,
      index = 3,
      clockid = 11,
      active = {
        head = {
          rb_node = 0x0
        },
        next = 0x0
      },
      resolution = {
        tv64 = 1
      },
      get_time = 0xffffffff810f08f0 <ktime_get_clocktai>,
      rh_reserved_softirq_time = {
        tv64 = 0
      },
      offset = {
        tv64 = 1540819482621868102
      }
    }},
  cpu = 0,
  in_hrtirq = 0
}
 

两类定时器模块的初始化,在start_kernel中,

asmlinkage void __init start_kernel(void)
{
。。。。    
    init_timers();//定时器模块初始化
    hrtimers_init();//高精度定时器模块初始化
。。。。
}

对比了两类定时器的定义,从定时器的执行再来对比一下,会加深印象。

对于低精度来说,

void __init init_timers(void)
{
    int err;

    /* ensure there are enough low bits for flags in timer->base pointer */
    BUILD_BUG_ON(__alignof__(struct tvec_base) & TIMER_FLAG_MASK);

    err = timer_cpu_notify(&timers_nb, (unsigned long)CPU_UP_PREPARE,
                   (void *)(long)smp_processor_id());
    init_timer_stats();

    BUG_ON(err != NOTIFY_OK);
    register_cpu_notifier(&timers_nb);
    open_softirq(TIMER_SOFTIRQ, run_timer_softirq);
}
在收到TIMER_SOFTIRQ 之后,run_timer_softirq-->__run_timers,这个函数会执行所有到期的定时的回调函数。执行回调的时候都持有 base->lock 这把自旋锁,所以也要求执行函数不能耗时太多。
 
对于高精度定时器来说,由于有两种模式,所以需要单独说明调用流程,

如果是处于低分辨率模式,则会在周期性的 update_process_times-->run_local_timers-->hrtimer_run_queues-->__hrtimer_run_queues 来把这些高精度定时器回调来执行;

update_process_times 调用 run_local_timers 来触发TIMER_SOFTIRQ软中断,run_timer_softirq负责调用__run_timers处理 TIMER_SOFTIRQ软中断。

 run_local_timers  除了触发软中断,还调用   hrtimer_run_queues();看能否从低分辨率定时器切换到高分辨率。
run_local_timers | -->hrtimer_run_queues 负责分辨率切换--->hrtimer_switch_to_hres-->tick_setup_sched_timer
                           |-->raise_softirq(TIMER_SOFTIRQ)

如果是处于高精度模式,则虽然周期性的 update_process_times-->run_local_timers-->hrtimer_run_queues 会执行,但不会调用 __hrtimer_run_queues ,而是在 hrtimer_interrupt

函数中调用 __hrtimer_run_queues->__run_hrtimer 来完成定时器的调用。

调用链如下:

hrtimer_interrupt-->__hrtimer_run_queues-->__run_hrtimer-->执行回调。

这点和网上的不一致,因为网上大多是2.6的内核描述,其实在哪处理不是很关键,主要是理解数据结构和调用。

 总结一下:

  • 每个cpu有一个tvec_base结构;

  • tvec_base结构管理着5个不同超时时间的数组,它采用的基准时间是jiffies。

  • 加入时间轮的时候,通过timer_list的超时时间,来指定它vec,

  • 时间轮,按到期时间进行处理,第一轮vec处理完毕,会在第二轮中取一个数组元素填充第一轮的256个到底的元素,

  • 它通过__run_timers来执行所有到期的低精度定时器

  • 每个cpu有一个hrtimer_cpu_base结构;
  • hrtimer_cpu_base结构管理着3种不同的时间基准系统的hrtimer,分别是:实时时间,启动时间和单调时间;它的基准时间是纳秒。
  • 每种时间基准系统通过它的active字段(timerqueue_head结构指针),指向它们各自的红黑树;
  • 红黑树上,按到期时间进行排序,最先到期的hrtimer位于最左下的节点,并被记录在active.next字段中;
  • 3中时间基准的最先到期时间可能不同,所以,它们之中最先到期的时间被记录在hrtimer_cpu_base的expires_next字段中。
 

有一点需要注意,高精度定时器要生效,意味着我们要有高精度的时钟源,那么当没有这么高精度的时钟源的时候,高精度定时器的运转,则精度会降低。

说到时钟源:在我的机器上,3.10的内核,封装了一个结构,叫clocksource如下:

crash> list clocksource.list -H clocksource_list
ffffffff81a273c0
ffffffff81a2bb40
ffffffff81aebb80
ffffffff81eb5980
ffffffff81a52c40
crash> clocksource ffffffff81a273c0
struct clocksource {
  read = 0xffffffff81032e20 <read_tsc>,-----------这个成员的位置放到第一个,因为它最频繁使用,和2.6.18系列版本不一样,大家定义结构的时候把最常使用的放前面,便于cache命中
  cycle_last = 2592996216546832,
  mask = 18446744073709551615,
  mult = 4194304,
  shift = 23,
  max_idle_ns = 428122390528,
  maxadj = 461373,
  archdata = {
    vclock_mode = 1
  },
  name = 0xffffffff819217b4 "tsc",
  list = {
    next = 0xffffffff81a2bb78 <clocksource_hpet+56>,
    prev = 0xffffffff81a52c30 <clocksource_list>
  },
  rating = 300,----------------精度最高
  enable = 0x0,
  disable = 0x0,
  flags = 35,
  suspend = 0x0,
  resume = 0x0,
  owner = 0x0
}
crash> clocksource ffffffff81a2bb40
struct clocksource {
  read = 0xffffffff81062430 <read_hpet>,
  cycle_last = 103666886,
  mask = 4294967295,
  mult = 2796202783,
  shift = 26,
  max_idle_ns = 69681373356,
  maxadj = 307582306,
  archdata = {
    vclock_mode = 2
  },
  name = 0xffffffff818ff927 "hpet",
  list = {
    next = 0xffffffff81aebbb8 <clocksource_acpi_pm+56>,
    prev = 0xffffffff81a273f8 <clocksource_tsc+56>
  },
  rating = 250,
  enable = 0x0,
  disable = 0x0,
  flags = 33,
  suspend = 0x0,
  resume = 0xffffffff810619e0 <hpet_resume_counter>,
  owner = 0x0
}
crash> clocksource ffffffff81aebb80
struct clocksource {
  read = 0xffffffff8153cb10 <acpi_pm_read>,
  cycle_last = 0,
  mask = 16777215,
  mult = 2343484437,
  shift = 23,
  max_idle_ns = 3649976793,
  maxadj = 257783288,
  archdata = {
    vclock_mode = 0
  },
  name = 0xffffffff8191e0b6 "acpi_pm",
  list = {
    next = 0xffffffff81eb59b8 <refined_jiffies+56>,
    prev = 0xffffffff81a2bb78 <clocksource_hpet+56>
  },
  rating = 200,
  enable = 0x0,
  disable = 0x0,
  flags = 33,
  suspend = 0x0,
  resume = 0x0,
  owner = 0x0
}
crash> clocksource ffffffff81eb5980
struct clocksource {
  read = 0xffffffff810f3290 <jiffies_read>,
  cycle_last = 0,
  mask = 4294967295,
  mult = 255961088,
  shift = 8,
  max_idle_ns = 3344197395684985,
  maxadj = 28155719,
  archdata = {
    vclock_mode = 0
  },
  name = 0xffffffff8191e0fb "refined-jiffies",
  list = {
    next = 0xffffffff81a52c78 <clocksource_jiffies+56>,
    prev = 0xffffffff81aebbb8 <clocksource_acpi_pm+56>
  },
  rating = 2,------------------------精度最低
  enable = 0x0,
  disable = 0x0,
  flags = 0,
  suspend = 0x0,
  resume = 0x0,
  owner = 0x0
}
crash> clocksource ffffffff81a52c40
struct clocksource {
  read = 0xffffffff810f3290 <jiffies_read>,
  cycle_last = 4294669298,
  mask = 4294967295,
  mult = 256000000,
  shift = 8,
  max_idle_ns = 3344705780981250,
  maxadj = 28160000,
  archdata = {
    vclock_mode = 0
  },
  name = 0xffffffff8191e103 "jiffies",
  list = {
    next = 0xffffffff81a52c30 <clocksource_list>,
    prev = 0xffffffff81eb59b8 <refined_jiffies+56>
  },
  rating = 1,
  enable = 0x0,
  disable = 0x0,
  flags = 0,
  suspend = 0x0,
  resume = 0x0,
  owner = 0x0
}

 用户可以通过 手工来切换clocksource,比如我的环境上有tsc,hpet,acpi_pm三个可用的clocksource(这个比crash中列的少一些)

cat /sys/devices/system/clocksource/clocksource0/available_clocksource
tsc hpet acpi_pm
[root@localhost ~]# cat /sys/devices/system/clocksource/clocksource0/current_clocksource
tsc
[root@localhost ~]# cat /sys/devices/system/clocksource/clocksource0/unbind_clocksource
cat: /sys/devices/system/clocksource/clocksource0/unbind_clocksource: Permission denied
[root@localhost ~]# cat /sys/devices/system/clocksource/clocksource0/current_clocksource
tsc
[root@localhost ~]# ls -alrt /sys/devices/system/clocksource/clocksource0/current_clocksource
-rw-r--r-- 1 root root 4096 Oct 30 09:52 /sys/devices/system/clocksource/clocksource0/current_clocksource
[root@localhost ~]# echo hpet > /sys/devices/system/clocksource/clocksource0/current_clocksource
[root@localhost ~]# cat /sys/devices/system/clocksource/clocksource0/current_clocksource
hpet
[root@localhost ~]# echo tsc > /sys/devices/system/clocksource/clocksource0/current_clocksource
切换之后会有打印,有时候也可以在message中看到内核自动切换的打印。

[44890.290544] Switched to clocksource hpet
[44902.121090] Switched to clocksource tsc

介绍完时钟源的定义和使用,有必要介绍下一个重要概念,时钟事件设备。

时间事件设备允许注册一个事件,在未来一个指定的时间点上发生,但与定时器实现相比,它只能存储一个事件。

举一个clock_event_device 的例子:

clock_event_device ffff8827dcf11140
struct clock_event_device {
  event_handler = 0xffffffff810b9890 <hrtimer_interrupt>,
  set_next_event = 0xffffffff81053df0 <lapic_next_deadline>,
  set_next_ktime = 0x0,
  next_event = {
    tv64 = 62900678796701
  },
  max_delta_ns = 2199023255551,
  min_delta_ns = 1000,
  mult = 8388608,
  shift = 27,
  mode = CLOCK_EVT_MODE_ONESHOT,
  features = 2,-------------------------------------------属性,为2说明是oneshot模式
  retries = 19117,
  broadcast = 0xffffffff81053e30 <lapic_timer_broadcast>,
  set_mode = 0xffffffff81054620 <lapic_timer_setup>,
  suspend = 0x0,
  resume = 0x0,
  min_delta_ticks = 15,
  max_delta_ticks = 18446744073709551615,
  name = 0xffffffff818fefdd "lapic",------------------事件设备的名称
  rating = 150,
  irq = -1,
  bound_on = 0,
  cpumask = 0xffffffff816e7c60 <cpu_bit_bitmap+26240>,
  list = {
    next = 0xffff8857bc2d11d8,
    prev = 0xffff8827dcf511d8
  },
  owner = 0x0
}
/*
 * Clock event features
 */
#define CLOCK_EVT_FEAT_PERIODIC        0x000001
#define CLOCK_EVT_FEAT_ONESHOT        0x000002
#define CLOCK_EVT_FEAT_KTIME        0x000004
/*
 * x86(64) specific misfeatures:
 *
 * - Clockevent source stops in C3 State and needs broadcast support.
 * - Local APIC timer is used as a dummy device.
 */
#define CLOCK_EVT_FEAT_C3STOP        0x000008
#define CLOCK_EVT_FEAT_DUMMY        0x000010

/*
 * Core shall set the interrupt affinity dynamically in broadcast mode
 */
#define CLOCK_EVT_FEAT_DYNIRQ        0x000020

/*
 * Clockevent device is based on a hrtimer for broadcast
 */
#define CLOCK_EVT_FEAT_HRTIMER        0x000080

每个时钟硬件设备注册一个时钟设备tick_device 和一个时钟事件设备。

struct tick_device {
    struct clock_event_device *evtdev;
    enum tick_device_mode mode;
};
可以看出,时钟设备就是时钟事件设备的简单封装。
 

为了精度,系统兼容了两套定时器,一套是时间轮的低精度定时器,一种是高精度的hrtimer。定时器软中断调用 hrtimer_run_queues 来处理高分辨率定时器队列,哪怕底层时钟事件设备只提供了低分辨率,也是如此。这使得可以使用现存的框架,而无需关注时钟的分辨率。

为了节能,系统又引入了tickless模型,也就是nohz模型,其实就是将原来周期性的tick,变为按需触发,对于需要模拟tick的周期性函数,则由相应的cpu来完成,其他cpu如果没事可以

休息。 

nohz_mode目前包含三种模式,一种是未开启nohz,一种是系统工作于低分辨率模式下的动态时钟,一种是系统工作于高精度模式下的动态时钟。

struct tick_sched {
    struct hrtimer            sched_timer;---用于高分辨率模式下,模拟周期时钟的一个timer
    unsigned long            check_clocks;
    enum tick_nohz_mode        nohz_mode;---包含三种模式,
    ktime_t                last_tick;
    ktime_t                next_tick;
    int                inidle;
    int                tick_stopped;
    unsigned long            idle_jiffies;
    unsigned long            idle_calls;
    unsigned long            idle_sleeps;
    int                idle_active;
    ktime_t                idle_entrytime;
    ktime_t                idle_waketime;
    ktime_t                idle_exittime;
    ktime_t                idle_sleeptime;
    ktime_t                iowait_sleeptime;
    ktime_t                sleep_length;
    unsigned long            last_jiffies;
    u64                next_timer;
    ktime_t                idle_expires;
    int                do_timer_last;
};
 
crash> tick_sched
struct tick_sched {
    struct hrtimer sched_timer;
    unsigned long check_clocks;
    enum tick_nohz_mode nohz_mode;
    ktime_t last_tick;
    ktime_t next_tick;
    int inidle;
    int tick_stopped;
    unsigned long idle_jiffies;
    unsigned long idle_calls;
    unsigned long idle_sleeps;
    int idle_active;
    ktime_t idle_entrytime;
    ktime_t idle_waketime;
    ktime_t idle_exittime;
    ktime_t idle_sleeptime;
    ktime_t iowait_sleeptime;
    ktime_t sleep_length;
    unsigned long last_jiffies;
    u64 next_timer;
    ktime_t idle_expires;
    int do_timer_last;
}

tick_sched 中收集的统计信息通过/proc/timer_list 导出到用户层。

crash> tick_cpu_sched
PER-CPU DATA TYPE:
  struct tick_sched tick_cpu_sched;
PER-CPU ADDRESSES:
  [0]: ffff8827dca13f20
  [1]: ffff8827dca53f20
  [2]: ffff8827dca93f20
  [3]: ffff8827dcad3f20
。。。。。。。。。。。。。

crash> p tick_cpu_sched:0
per_cpu(tick_cpu_sched, 0) = $18 = {
  sched_timer = {
    node = {
      node = {
        __rb_parent_color = 18446612303169076872,
        rb_right = 0x0,
        rb_left = 0x0
      },
      expires = {
        tv64 = 579956705000000
      }
    },
    _softexpires = {
      tv64 = 579956705000000
    },
    function = 0xffffffff810f9170 <tick_sched_timer>,
    base = 0xffff8827dca139a0,
    state = 1,
    start_pid = 0,
    start_site = 0xffffffff810f95c2 <tick_nohz_stop_sched_tick+690>,
    start_comm = "swapper/0\000\000\000\000\000\000"
  },
  check_clocks = 1,
  nohz_mode = NOHZ_MODE_HIGHRES,
  last_tick = {
    tv64 = 579956549000000
  },
  next_tick = {
    tv64 = 579956705000000
  },
  inidle = 1,
  tick_stopped = 1,
  idle_jiffies = 4874623845,
  idle_calls = 2616662184,
  idle_sleeps = 2409217702,
  idle_active = 1,
  idle_entrytime = {
    tv64 = 579956548218826
  },
  idle_waketime = {
    tv64 = 579956548210360
  },
  idle_exittime = {
    tv64 = 579956548213511
  },
  idle_sleeptime = {
    tv64 = 548323643001010
  },
  iowait_sleeptime = {
    tv64 = 53588522970
  },
  sleep_length = {
    tv64 = 515114
  },
  last_jiffies = 4874623845,
  next_timer = 579956705000000,
  idle_expires = {
    tv64 = 579956705000000
  },
  do_timer_last = 0
}

对时钟的禁用是按cpu指定的,一般来说,所有cpu都空闲的概率还是比较低的。

crash> p tick_next_period
tick_next_period = $19 = {
  tv64 = 580792669000000
}
crash> p tick_next_period
tick_next_period = $20 = {
  tv64 = 580794124000000
}
crash> p tick_next_period
tick_next_period = $21 = {
  tv64 = 580795263000000
}
crash> p tick_next_period
tick_next_period = $22 = {
  tv64 = 580796247000000
}
crash> p last_jiffies_update
last_jiffies_update = $23 = {
  tv64 = 580801981000000
}
crash> p last_jiffies_update
last_jiffies_update = $24 = {
  tv64 = 580802792000000
}
crash> p last_jiffies_update
last_jiffies_update = $25 = {
  tv64 = 580803530000000
}

 时间相关系统调用及外部设置:

   adjtimex 系统调用,NTP设置,

内核的工作模式:

  1. 没有动态时钟的低分辨率系统,总是用周期时钟。这时不会支持单触发模式

  2. 启用了动态时钟的低分辨率系统,将以单触发模式是用时钟设备

  3. 高分辨率系统总是用单触发模式,无论是否启用了动态时钟特性

  4. 高分辨率时钟系统,每个cpu会使用一个hrtimer来模拟周期时钟,提供tick,毕竟精度高的要模拟精度低的比较容易,同时又能纳入自己的高分辨率框架。模拟的函数为:tick_sched_timer

非广播时最终的处理函数:

高分辨率动态时钟:hrtimer_interrupt

高分辨率周期时钟:hrtimer_interrupt

低分辨率动态时钟:tick_nohz_handler

低分辨率周期时钟:tick_handle_periodic

广播时最终的处理函数:

高分辨率动态时钟:tick_handle_oneshot_broadcast

高分辨率周期时钟:tick_handle_oneshot_broadcast

低分辨率动态时钟:tick_handle_oneshot_broadcast

低分辨率周期时钟:tick_handle_periodic_broadcast

参考资料:

linux 3.10内核源码
原文:https://blog.csdn.net/goodluckwhh/article/details/9048565 

 

猜你喜欢

转载自www.cnblogs.com/10087622blog/p/9854619.html