浅析linux内核中timer定时器的生成和sofirq软中断调用流程(转自http://blog.chinaunix.net/uid-20564848-id-73480.html)

浅析linux内核中timer定时器的生成和sofirq软中断调用流程

  mod_timer添加的定时器timer在内核的软中断中发生调用,__run_timers会spin_lock_irq(&base->lock);禁止cpu中断,所以我们的timer回调处理函数handler工作在irq关闭的环境中,所以需要作很多考虑,比如在handler中尽量不要执行会引起pending的函数调用,比如kmalloc之类可能引起pending的操作,否则会使kernel永远停在我们的handler中不能返回,这样kernel将因为我们ko设计上的失败而死机[luther.gliethttp]!
  我们可以使用如下几行语句,向我们的ko驱动添加一个timer定时器,来处理时间事件:
struct __wlanwlan_check_tx_flow_timer
{
    struct timer_list timer;
    int timer_freq;
} wlan_check_tx_flow_timer = {
        .timer_freq = 8*1000,
};
static void wlan_check_tx_flow_timer_handler(unsigned long data)
{
    ...
    //重新启动timer定时器 
    mod_timer(&wlan_check_tx_flow_timer.timer, jiffies + msecs_to_jiffies(wlan_check_tx_flow_timer.timer_freq));
    ...
}
//设置定时器
setup_timer(&wlan_check_tx_flow_timer.timer, wlan_check_tx_flow_timer_handler, (unsigned long)&wlan_check_tx_flow_timer);
//添加定时器
mod_timer(&wlan_check_tx_flow_timer.timer, jiffies + msecs_to_jiffies(wlan_check_tx_flow_timer.timer_freq));

那么这个wlan_check_tx_flow_timer_handler处理函数在什么时候被调用的呢？那么我们追入内核中,看看kernel对定时器的具体管理.
首先kernel在启动的最前面注册TIMER_SOFTIRQ的处理函数[luther.gliethttp],
start_kernel
=>init_timers
=>open_softirq(TIMER_SOFTIRQ, run_timer_softirq, NULL);
那么由谁来调用raise_softirq(TIMER_SOFTIRQ);触发TIMER_SOFTIRQ软中断呢,这就和平台相关了,对于pxa935处理器来说[luther.gliethttp],
MACHINE_START(LUTHER, "luther")
    .phys_io = 0x40000000,
    .boot_params = 0xa0000100,
    .io_pg_offst = (io_p2v(0x40000000) >> 18) & 0xfffc,
    .map_io = pxa_map_io,
    .init_irq = pxa3xx_init_irq,
    .timer = &pxa_timer,
    .init_machine = luther_init,
MACHINE_END
=>pxa_timer_init//平台对应的定时器初始化
==>pxa_timer_irq.dev_id = &ckevt_32ktimer;
==>setup_irq(IRQ_OST_4_11, &pxa_timer_irq); //32768的rtc
==>clockevents_register_device(&ckevt_32ktimer);

pxa_timer_interrupt中断处理函数
=>c->event_handler(c);也就是tick_handle_periodic系统时钟函数
=>tick_handle_periodic
=>update_process_times
=>run_local_timers
=>raise_softirq(TIMER_SOFTIRQ);
这里仅仅是触发了TIMER_SOFTIRQ软中断,那么在什么地方处理我们mod_timer添加的timer定时器处理函数wlan_check_tx_flow_timer_handler呢[luther.gliethttp]？
__irq_svc://内核中发生的中断
__irq_usr://用户空间时发生的中断
=>asm_do_IRQ
=>irq_exit
=>do_softirq
=>__do_softirq
=>调用上面注册的run_timer_softirq软中断处理函数
=>run_timer_softirq
=>__run_timers
static inline void __run_timers(struct tvec_base *base)
{
    struct timer_list *timer;

    spin_lock_irq(&base->lock);//禁止中断
    while (time_after_eq(jiffies, base->timer_jiffies)) {
        ...
        if (时间到了) {
        ...
        fn = timer->function;
        data = timer->data;
        fn(data);//这就是我们上面添加的static void wlan_check_tx_flow_timer_handler(unsigned long data);定时器处理函数了.
        ...
        }
        ...
    }
    set_running_timer(base, NULL);
    spin_unlock_irq(&base->lock);//打开中断
}
//================
include/asm/hardirq.h
typedef struct {
    unsigned int __softirq_pending;
    unsigned int local_timer_irqs;
} ____cacheline_aligned irq_cpustat_t;
//================
kernel/softirq.c|45| irq_cpustat_t irq_stat[NR_CPUS] ____cacheline_aligned;
#ifndef __ARCH_IRQ_STAT
irq_cpustat_t irq_stat[NR_CPUS] ____cacheline_aligned;//在这里定义irq_stat存储空间
EXPORT_SYMBOL(irq_stat);
#endif
//================
include/linux/irq_cpustat.h
#ifndef __ARCH_IRQ_STAT
//引用的就是上面的irq_cpustat_t irq_stat[NR_CPUS] ____cacheline_aligned;
extern irq_cpustat_t irq_stat[];        /* defined in asm/hardirq.h */
#define __IRQ_STAT(cpu, member)    (irq_stat[cpu].member)
#endif
//================
arch/arm/kernel/entry-armv.S|331| .word    irq_stat

#ifdef CONFIG_PREEMPT
svc_preempt:
    teq    r8, #0                @ was preempt count = 0
    ldreq    r6, .LCirq_stat //操作
    movne    pc, lr                @ no
    ldr    r0, [r6, #4]            @ local_irq_count
    ldr    r1, [r6, #8]            @ local_bh_count
    adds    r0, r0, r1
    movne    pc, lr
    mov    r7, #0                @ preempt_schedule_irq
    str    r7, [tsk, #TI_PREEMPT]        @ expects preempt_count == 0
1:    bl    preempt_schedule_irq        @ irq en/disable is done inside
    ldr    r0, [tsk, #TI_FLAGS]        @ get new tasks TI_FLAGS
    tst    r0, #_TIF_NEED_RESCHED
    beq    preempt_return            @ go again
    b    1b
#endif

.LCirq_stat:
    .word    irq_stat //引用irq_cpustat_t irq_stat[NR_CPUS] ____cacheline_aligned;地址
#endif

  /* arch independent irq_stat fields */
#define local_softirq_pending() \
    __IRQ_STAT(smp_processor_id(), __softirq_pending)

#define __ARCH_IRQ_EXIT_IRQS_DISABLED    1
#ifdef __ARCH_IRQ_EXIT_IRQS_DISABLED
# define invoke_softirq()    __do_softirq() //是这个
#else
# define invoke_softirq()    do_softirq()
#endif

#ifndef __ARCH_SET_SOFTIRQ_PENDING
#define set_softirq_pending(x) (local_softirq_pending() = (x))
#define or_softirq_pending(x) (local_softirq_pending() |= (x))
#endif
#define __raise_softirq_irqoff(nr) do { or_softirq_pending(1UL << (nr)); } while (0)

inline void raise_softirq_irqoff(unsigned int nr)
{
    __raise_softirq_irqoff(nr);
        if (!in_interrupt())
        wakeup_softirqd();
}
void raise_softirq(unsigned int nr)
{
    unsigned long flags;

    local_irq_save(flags);
    raise_softirq_irqoff(nr);
    local_irq_restore(flags);
}


=>s3c2410_timer_interrupt
=>timer_tick

=>pxa_timer_init
==>pxa_timer_irq.dev_id = &ckevt_32ktimer;
==>setup_irq(IRQ_OST_4_11, &pxa_timer_irq); //32768的rtc
==>clockevents_register_device(&ckevt_32ktimer);
=>clockevents_register_device
=>clockevents_do_notify
=>raw_notifier_call_chain(&clockevents_chain, reason, dev);
=>__raw_notifier_call_chain
=>notifier_call_chain(&nh->head, val, v, nr_to_call, nr_calls);
=>nb->notifier_call(nb, val, v);就是tick_notify

start_kernel
=>tick_init

static struct notifier_block tick_notifier = {
    .notifier_call = tick_notify,
};
void __init tick_init(void)
{
    clockevents_register_notifier(&tick_notifier);
}
clockevents_register_notifier
=>raw_notifier_chain_register(&clockevents_chain, nb);
=>notifier_chain_register将tick_notifier添加到clockevents_chain这个单向链表中[luther.gliethttp]
static int tick_notify(struct notifier_block *nb, unsigned long reason,
             void *dev)
{
    switch (reason) {

    case CLOCK_EVT_NOTIFY_ADD:
        return tick_check_new_device(dev);
    ...
    return NOTIFY_OK;
}

=>tick_notify
=>tick_check_new_device
=>tick_setup_device(td, newdev, cpu, cpumask);
static void tick_setup_device(struct tick_device *td,
             struct clock_event_device *newdev, int cpu,
             cpumask_t cpumask)
{
    ktime_t next_event;
    void (*handler)(struct clock_event_device *) = NULL;

    /*
     * First device setup ?
     */
    if (!td->evtdev) {
        /*
         * If no cpu took the do_timer update, assign it to
         * this cpu:
         */
        if (tick_do_timer_cpu == -1) {
            tick_do_timer_cpu = cpu;
            tick_next_period = ktime_get();
            tick_period = ktime_set(0, NSEC_PER_SEC / HZ);
        }

        /*
         * Startup in periodic mode first.
         */
        td->mode = TICKDEV_MODE_PERIODIC;//设置第1个tick设备为TICKDEV_MODE_PERIODIC模式
    } else {
        handler = td->evtdev->event_handler;
        next_event = td->evtdev->next_event;
    }

    td->evtdev = newdev;
    ...
    if (td->mode == TICKDEV_MODE_PERIODIC)
        tick_setup_periodic(newdev, 0);
    else
        tick_setup_oneshot(newdev, handler, next_event);
}

void tick_setup_periodic(struct clock_event_device *dev, int broadcast)
{
    tick_set_periodic_handler(dev, broadcast);//设置event_handler处理函数为dev->event_handler = tick_handle_periodic;

    /* Broadcast setup ? */
    if (!tick_device_is_functional(dev))
        return;

    if (dev->features & CLOCK_EVT_FEAT_PERIODIC) {
        clockevents_set_mode(dev, CLOCK_EVT_MODE_PERIODIC);
    } else {
        unsigned long seq;
        ktime_t next;

        do {
            seq = read_seqbegin(&xtime_lock);
            next = tick_next_period;
        } while (read_seqretry(&xtime_lock, seq));

        clockevents_set_mode(dev, CLOCK_EVT_MODE_ONESHOT);

        for (;;) {
            if (!clockevents_program_event(dev, next, ktime_get()))
                return;
            next = ktime_add(next, tick_period);
        }
    }
}

void tick_set_periodic_handler(struct clock_event_device *dev, int broadcast)
{
    if (!broadcast)
        dev->event_handler = tick_handle_periodic;
    else
        dev->event_handler = tick_handle_periodic_broadcast;
}

=>pxa_timer_interrupt
{
    ...
    if (OSSR & OST_C4) {
        OIER &= ~OST_C4;
        OSSR = OST_C4;
        if (timer32k_enabled)
            c->event_handler(c);//调用tick_handle_periodic处理函数,作为
    }
    ... 
}

void tick_handle_periodic(struct clock_event_device *dev)
{
    int cpu = smp_processor_id();
    ktime_t next;

    tick_periodic(cpu);//调用do_timer(1);将jiffies_64加1

    if (dev->mode != CLOCK_EVT_MODE_ONESHOT)
        return;
    /*
     * Setup the next period for devices, which do not have
     * periodic mode:
     */
    next = ktime_add(dev->next_event, tick_period);
    for (;;) {
        if (!clockevents_program_event(dev, next, ktime_get()))
            return;
        tick_periodic(cpu);
        next = ktime_add(next, tick_period);
    }
}

static void tick_periodic(int cpu)
{
    if (tick_do_timer_cpu == cpu) {
        write_seqlock(&xtime_lock);

        /* Keep track of the next tick event */
        tick_next_period = ktime_add(tick_next_period, tick_period);

        do_timer(1);
        write_sequnlock(&xtime_lock);
    }

    update_process_times(user_mode(get_irq_regs()));
    profile_tick(CPU_PROFILING);
}

arch/arm/kernel/time.c|332| update_process_times(user_mode(get_irq_regs()));
=>update_process_times
=>run_local_timers
=>raise_softirq(TIMER_SOFTIRQ);//触发软中断,当irq_exit时调用__do_softirq来处理


=>run_timer_softirq
=>__run_timers
=>
fn = timer->function;//执行
data = timer->data;
fn(data);
//================
include/asm/arch-pxa/entry-macro.S|22| .macro    get_irqnr_and_base, irqnr, irqstat, base, tmp
//pxa获取irq中断号函数
//================
arch/arm/kernel/entry-armv.S|37| bne    asm_do_IRQ
    .macro    irq_handler
    get_irqnr_preamble r5, lr
1:    get_irqnr_and_base r0, r6, r5, lr //获取irq中断号,存储到r0寄存器中,作为参数传递给asm_do_IRQ
    movne    r1, sp
    @
    @ routine called with r0 = irq number, r1 = struct pt_regs *
    @
    adrne    lr, 1b
    bne    asm_do_IRQ
    ...
//================
    .align    5
__irq_svc://内核中发生的中断
    svc_entry
    ...
    irq_handler
    ...
//================
    .align    5
__irq_usr://用户空间时发生的中断
    usr_entry
    ...
    irq_handler
    ...
//================
    .macro    vector_stub, name, mode, correction=0
    .align    5

vector_\name:
    .if \correction
    sub    lr, lr, #\correction
    .endif

    @
    @ Save r0, lr_<exception> (parent PC) and spsr_<exception>
    @ (parent CPSR)
    @
    stmia    sp, {r0, lr}        @ save r0, lr
    mrs    lr, spsr
    str    lr, [sp, #8]        @ save spsr

    @
    @ Prepare for SVC32 mode. IRQs remain disabled.
    @
    mrs    r0, cpsr
    eor    r0, r0, #(\mode ^ SVC_MODE)
    msr    spsr_cxsf, r0

    @
    @ the branch table must immediately follow this code
    @
    and    lr, lr, #0x0f //lr存储了spsr,所以一共有16种cpu模式
    mov    r0, sp //传参
    ldr    lr, [pc, lr, lsl #2]//取出相应模式下的处理函数指针,比如__irq_usr或者__irq_svc
    movs    pc, lr            @ branch to handler in SVC mode
    .endm
//================
    .globl    __stubs_start
__stubs_start:
/*
 * Interrupt dispatcher
 */
    vector_stub    irq, IRQ_MODE, 4

    .long    __irq_usr            @ 0 (USR_26 / USR_32)
    .long    __irq_invalid        @ 1 (FIQ_26 / FIQ_32)
    .long    __irq_invalid        @ 2 (IRQ_26 / IRQ_32)
    .long    __irq_svc            @ 3 (SVC_26 / SVC_32)
    .long    __irq_invalid            @ 4
    .long    __irq_invalid            @ 5
    .long    __irq_invalid            @ 6
    .long    __irq_invalid            @ 7
    .long    __irq_invalid            @ 8
    .long    __irq_invalid            @ 9
    .long    __irq_invalid            @ a
    .long    __irq_invalid            @ b
    .long    __irq_invalid            @ c
    .long    __irq_invalid            @ d
    .long    __irq_invalid            @ e
    .long    __irq_invalid            @ f
//================
    .globl    __vectors_start
__vectors_start:
    swi    SYS_ERROR0
    b    vector_und + stubs_offset
    ldr    pc, .LCvswi + stubs_offset
    b    vector_pabt + stubs_offset
    b    vector_dabt + stubs_offset
    b    vector_addrexcptn + stubs_offset
    b    vector_irq + stubs_offset
    b    vector_fiq + stubs_offset
//================
asm_do_IRQ(unsigned int irq, struct pt_regs *regs)
=>desc_handle_irq(irq, desc);//
static inline void desc_handle_irq(unsigned int irq, struct irq_desc *desc)
{
    desc->handle_irq(irq, desc);//调用中断号irq对应的handler回调处理函数[luther.gliethttp]
}
__irq_svc://内核中发生的中断
__irq_usr://用户空间时发生的中断
=>asm_do_IRQ
=>irq_exit
=>do_softirq
=>__do_softirq
=>
{ 
    ...
    h = softirq_vec;//执行软中断函数

    do {
        if (pending & 1) {
            h->action(h);
//如果32768的时间到达,那asm_do_IRQ中将触发raise_softirq(TIMER_SOFTIRQ);
//在这里将执行管理系统tick的run_timer_softirq软中断[luther.gliethttp]
            rcu_bh_qsctr_inc(cpu);
        }
        h++;
        pending >>= 1;
    } while (pending);
    ...
}

start_kernel
=>init_timers
=>open_softirq(TIMER_SOFTIRQ, run_timer_softirq, NULL);
void open_softirq(int nr, void (*action)(struct softirq_action*), void *data)
{
    softirq_vec[nr].data = data;
    softirq_vec[nr].action = action;
}

static void run_timer_softirq(struct softirq_action *h)
{
    struct tvec_base *base = __get_cpu_var(tvec_bases);//获得time时间根

    hrtimer_run_pending();

    if (time_after_eq(jiffies, base->timer_jiffies))
        __run_timers(base);
}
//执行软中断
=>run_timer_softirq
=>__run_timers
=>
fn = timer->function;
data = timer->data;
fn(data);

static inline void __run_timers(struct tvec_base *base)
{
    ...
    spin_lock_irq(&base->lock);//禁止中断
    ...
    fn = timer->function;
    data = timer->data;
    fn(data);
    ...
    set_running_timer(base, NULL);
    spin_unlock_irq(&base->lock);//打开中断
}

mod_timer
=>__mod_timer
int __mod_timer(struct timer_list *timer, unsigned long expires)
{
    struct tvec_base *base, *new_base;
    unsigned long flags;
    int ret = 0;

    timer_stats_timer_set_start_info(timer);
    BUG_ON(!timer->function);

    base = lock_timer_base(timer, &flags);

    if (timer_pending(timer)) {
        detach_timer(timer, 0);
        ret = 1;
    }

    new_base = __get_cpu_var(tvec_bases);//获得time时间根

    if (base != new_base) {
        /*
         * We are trying to schedule the timer on the local CPU.
         * However we can't change timer's base while it is running,
         * otherwise del_timer_sync() can't detect that the timer's
         * handler yet has not finished. This also guarantees that
         * the timer is serialized wrt itself.
         */
        if (likely(base->running_timer != timer)) {
            /* See the comment in lock_timer_base() */
            timer_set_base(timer, NULL);
            spin_unlock(&base->lock);
            base = new_base;
            spin_lock(&base->lock);
            timer_set_base(timer, base);
        }
    }

    timer->expires = expires;
    internal_add_timer(base, timer);
//添加到链表上,这样当timer超时到达时,run_timer_softirq=>__run_timers软中断中将会回调该处理函数[luther.gliethttp].
    spin_unlock_irqrestore(&base->lock, flags);

    return ret;
}