kernel/task_work.c

   1 #include <linux/spinlock.h>
   2 #include <linux/task_work.h>
   3 #include <linux/tracehook.h>
   4
   5 static struct callback_head work_exited; /* all we need is ->next == NULL */
   6
   7 /**
   8  * task_work_add - ask the @task to execute @work->func()
   9  * @task: the task which should run the callback
  10  * @work: the callback to run
  11  * @notify: send the notification if true
  12  *
  13  * Queue @work for task_work_run() below and notify the @task if @notify.
  14  * Fails if the @task is exiting/exited and thus it can't process this @work.
  15  * Otherwise @work->func() will be called when the @task returns from kernel
  16  * mode or exits.
  17  *
  18  * This is like the signal handler which runs in kernel mode, but it doesn't
  19  * try to wake up the @task.
  20  *
  21  * Note: there is no ordering guarantee on works queued here.
  22  *
  23  * RETURNS:
  24  * 0 if succeeds or -ESRCH.
  25  */
  26 int
  27 task_work_add(struct task_struct *task, struct callback_head *work, bool notify)
  28 {
  29         struct callback_head *head;
  30
  31         do {
  32                 head = READ_ONCE(task->task_works);
  33                 if (unlikely(head == &work_exited))
  34                         return -ESRCH;
  35                 work->next = head;
  36         } while (cmpxchg(&task->task_works, head, work) != head);
  37
  38         if (notify)
  39                 set_notify_resume(task);
  40         return 0;
  41 }
  42
  43 /**
  44  * task_work_cancel - cancel a pending work added by task_work_add()
  45  * @task: the task which should execute the work
  46  * @func: identifies the work to remove
  47  *
  48  * Find the last queued pending work with ->func == @func and remove
  49  * it from queue.
  50  *
  51  * RETURNS:
  52  * The found work or NULL if not found.
  53  */
  54 struct callback_head *
  55 task_work_cancel(struct task_struct *task, task_work_func_t func)
  56 {
  57         struct callback_head **pprev = &task->task_works;
  58         struct callback_head *work;
  59         unsigned long flags;
  60
  61         if (likely(!task->task_works))
  62                 return NULL;
  63         /*
  64          * If cmpxchg() fails we continue without updating pprev.
  65          * Either we raced with task_work_add() which added the
  66          * new entry before this work, we will find it again. Or
  67          * we raced with task_work_run(), *pprev == NULL/exited.
  68          */
  69         raw_spin_lock_irqsave(&task->pi_lock, flags);
  70         while ((work = READ_ONCE(*pprev))) {
  71                 if (work->func != func)
  72                         pprev = &work->next;
  73                 else if (cmpxchg(pprev, work, work->next) == work)
  74                         break;
  75         }
  76         raw_spin_unlock_irqrestore(&task->pi_lock, flags);
  77
  78         return work;
  79 }
  80
  81 /**
  82  * task_work_run - execute the works added by task_work_add()
  83  *
  84  * Flush the pending works. Should be used by the core kernel code.
  85  * Called before the task returns to the user-mode or stops, or when
  86  * it exits. In the latter case task_work_add() can no longer add the
  87  * new work after task_work_run() returns.
  88  */
  89 void task_work_run(void)
  90 {
  91         struct task_struct *task = current;
  92         struct callback_head *work, *head, *next;
  93
  94         for (;;) {
  95                 /*
  96                  * work->func() can do task_work_add(), do not set
  97                  * work_exited unless the list is empty.
  98                  */
  99                 do {
 100                         work = READ_ONCE(task->task_works);
 101                         head = !work && (task->flags & PF_EXITING) ?
 102                                 &work_exited : NULL;
 103                 } while (cmpxchg(&task->task_works, work, head) != work);
 104
 105                 if (!work)
 106                         break;
 107                 /*
 108                  * Synchronize with task_work_cancel(). It can't remove
 109                  * the first entry == work, cmpxchg(task_works) should
 110                  * fail, but it can play with *work and other entries.
 111                  */
 112                 raw_spin_unlock_wait(&task->pi_lock);
 113
 114                 do {
 115                         next = work->next;
 116                         work->func(work);
 117                         work = next;
 118                         cond_resched();
 119                 } while (work);
 120         }
 121 }
 122 EXPORT_SYMBOL_GPL(task_work_run);