__set_current_state

#define __set_current_state(state_value)                \
    current->state = (state_value)

/* task state bitmask, copied from include/linux/sched.h */
#define TASK_RUNNING        0
#define TASK_INTERRUPTIBLE    1
#define TASK_UNINTERRUPTIBLE    2
#define __TASK_STOPPED        4
#define __TASK_TRACED        8
/* in tsk->exit_state */
#define EXIT_DEAD        16
#define EXIT_ZOMBIE        32
#define EXIT_TRACE        (EXIT_ZOMBIE | EXIT_DEAD)
/* in tsk->state again */
#define TASK_DEAD        64
#define TASK_WAKEKILL        128
#define TASK_WAKING        256
#define TASK_PARKED        512

asmlinkage __visible void __sched schedule(void)
{
    struct task_struct *tsk = current;

    sched_submit_work(tsk);
    do {
        preempt_disable();
        __schedule(false);
        sched_preempt_enable_no_resched();
    } while (need_resched());
    sched_update_worker(tsk);
}
EXPORT_SYMBOL(schedule);

static void __sched notrace __schedule(bool preempt)
{
    struct task_struct *prev, *next;
    unsigned long *switch_count;
    struct rq_flags rf;
    struct rq *rq;
    int cpu;

    cpu = smp_processor_id();
    rq = cpu_rq(cpu);
    prev = rq->curr;

    schedule_debug(prev, preempt);

    if (sched_feat(HRTICK))
        hrtick_clear(rq);

    local_irq_disable();
    rcu_note_context_switch(preempt);

    /*
     * Make sure that signal_pending_state()->signal_pending() below
     * can't be reordered with __set_current_state(TASK_INTERRUPTIBLE)
     * done by the caller to avoid the race with signal_wake_up().
     *
     * The membarrier system call requires a full memory barrier
     * after coming from user-space, before storing to rq->curr.
     */
    rq_lock(rq, &rf);
    smp_mb__after_spinlock();

    /* Promote REQ to ACT */
    rq->clock_update_flags <<= 1;
    update_rq_clock(rq);

    switch_count = &prev->nivcsw;
    if (!preempt && prev->state) {
        if (signal_pending_state(prev->state, prev)) {
            prev->state = TASK_RUNNING;
        } else {
            deactivate_task(rq, prev, DEQUEUE_SLEEP | DEQUEUE_NOCLOCK);//cpu运行队列中删除任务

            if (prev->in_iowait) {
                atomic_inc(&rq->nr_iowait);
                delayacct_blkio_start();
            }
        }
        switch_count = &prev->nvcsw;
    }

    next = pick_next_task(rq, prev, &rf);
    clear_tsk_need_resched(prev);
    clear_preempt_need_resched();

    if (likely(prev != next)) {
        rq->nr_switches++;
        /*
         * RCU users of rcu_dereference(rq->curr) may not see
         * changes to task_struct made by pick_next_task().
         */
        RCU_INIT_POINTER(rq->curr, next);
        /*
         * The membarrier system call requires each architecture
         * to have a full memory barrier after updating
         * rq->curr, before returning to user-space.
         *
         * Here are the schemes providing that barrier on the
         * various architectures:
         * - mm ? switch_mm() : mmdrop() for x86, s390, sparc, PowerPC.
         *   switch_mm() rely on membarrier_arch_switch_mm() on PowerPC.
         * - finish_lock_switch() for weakly-ordered
         *   architectures where spin_unlock is a full barrier,
         * - switch_to() for arm64 (weakly-ordered, spin_unlock
         *   is a RELEASE barrier),
         */
        ++*switch_count;

        trace_sched_switch(preempt, prev, next);

        /* Also unlocks the rq: */
        rq = context_switch(rq, prev, next, &rf);
    } else {
        rq->clock_update_flags &= ~(RQCF_ACT_SKIP|RQCF_REQ_SKIP);
        rq_unlock_irq(rq, &rf);
    }

    balance_callback(rq);
}