xref: /openbmc/linux/kernel/task_work.c (revision e9b7b8b3)
1 // SPDX-License-Identifier: GPL-2.0
2 #include <linux/spinlock.h>
3 #include <linux/task_work.h>
4 #include <linux/resume_user_mode.h>
5 
6 static struct callback_head work_exited; /* all we need is ->next == NULL */
7 
8 /**
9  * task_work_add - ask the @task to execute @work->func()
10  * @task: the task which should run the callback
11  * @work: the callback to run
12  * @notify: how to notify the targeted task
13  *
14  * Queue @work for task_work_run() below and notify the @task if @notify
15  * is @TWA_RESUME, @TWA_SIGNAL, or @TWA_SIGNAL_NO_IPI.
16  *
17  * @TWA_SIGNAL works like signals, in that the it will interrupt the targeted
18  * task and run the task_work, regardless of whether the task is currently
19  * running in the kernel or userspace.
20  * @TWA_SIGNAL_NO_IPI works like @TWA_SIGNAL, except it doesn't send a
21  * reschedule IPI to force the targeted task to reschedule and run task_work.
22  * This can be advantageous if there's no strict requirement that the
23  * task_work be run as soon as possible, just whenever the task enters the
24  * kernel anyway.
25  * @TWA_RESUME work is run only when the task exits the kernel and returns to
26  * user mode, or before entering guest mode.
27  *
28  * Fails if the @task is exiting/exited and thus it can't process this @work.
29  * Otherwise @work->func() will be called when the @task goes through one of
30  * the aforementioned transitions, or exits.
31  *
32  * If the targeted task is exiting, then an error is returned and the work item
33  * is not queued. It's up to the caller to arrange for an alternative mechanism
34  * in that case.
35  *
36  * Note: there is no ordering guarantee on works queued here. The task_work
37  * list is LIFO.
38  *
39  * RETURNS:
40  * 0 if succeeds or -ESRCH.
41  */
42 int task_work_add(struct task_struct *task, struct callback_head *work,
43 		  enum task_work_notify_mode notify)
44 {
45 	struct callback_head *head;
46 
47 	/* record the work call stack in order to print it in KASAN reports */
48 	kasan_record_aux_stack(work);
49 
50 	head = READ_ONCE(task->task_works);
51 	do {
52 		if (unlikely(head == &work_exited))
53 			return -ESRCH;
54 		work->next = head;
55 	} while (!try_cmpxchg(&task->task_works, &head, work));
56 
57 	switch (notify) {
58 	case TWA_NONE:
59 		break;
60 	case TWA_RESUME:
61 		set_notify_resume(task);
62 		break;
63 	case TWA_SIGNAL:
64 		set_notify_signal(task);
65 		break;
66 	case TWA_SIGNAL_NO_IPI:
67 		__set_notify_signal(task);
68 		break;
69 	default:
70 		WARN_ON_ONCE(1);
71 		break;
72 	}
73 
74 	return 0;
75 }
76 
77 /**
78  * task_work_cancel_match - cancel a pending work added by task_work_add()
79  * @task: the task which should execute the work
80  * @match: match function to call
81  * @data: data to be passed in to match function
82  *
83  * RETURNS:
84  * The found work or NULL if not found.
85  */
86 struct callback_head *
87 task_work_cancel_match(struct task_struct *task,
88 		       bool (*match)(struct callback_head *, void *data),
89 		       void *data)
90 {
91 	struct callback_head **pprev = &task->task_works;
92 	struct callback_head *work;
93 	unsigned long flags;
94 
95 	if (likely(!task_work_pending(task)))
96 		return NULL;
97 	/*
98 	 * If cmpxchg() fails we continue without updating pprev.
99 	 * Either we raced with task_work_add() which added the
100 	 * new entry before this work, we will find it again. Or
101 	 * we raced with task_work_run(), *pprev == NULL/exited.
102 	 */
103 	raw_spin_lock_irqsave(&task->pi_lock, flags);
104 	work = READ_ONCE(*pprev);
105 	while (work) {
106 		if (!match(work, data)) {
107 			pprev = &work->next;
108 			work = READ_ONCE(*pprev);
109 		} else if (try_cmpxchg(pprev, &work, work->next))
110 			break;
111 	}
112 	raw_spin_unlock_irqrestore(&task->pi_lock, flags);
113 
114 	return work;
115 }
116 
117 static bool task_work_func_match(struct callback_head *cb, void *data)
118 {
119 	return cb->func == data;
120 }
121 
122 /**
123  * task_work_cancel - cancel a pending work added by task_work_add()
124  * @task: the task which should execute the work
125  * @func: identifies the work to remove
126  *
127  * Find the last queued pending work with ->func == @func and remove
128  * it from queue.
129  *
130  * RETURNS:
131  * The found work or NULL if not found.
132  */
133 struct callback_head *
134 task_work_cancel(struct task_struct *task, task_work_func_t func)
135 {
136 	return task_work_cancel_match(task, task_work_func_match, func);
137 }
138 
139 /**
140  * task_work_run - execute the works added by task_work_add()
141  *
142  * Flush the pending works. Should be used by the core kernel code.
143  * Called before the task returns to the user-mode or stops, or when
144  * it exits. In the latter case task_work_add() can no longer add the
145  * new work after task_work_run() returns.
146  */
147 void task_work_run(void)
148 {
149 	struct task_struct *task = current;
150 	struct callback_head *work, *head, *next;
151 
152 	for (;;) {
153 		/*
154 		 * work->func() can do task_work_add(), do not set
155 		 * work_exited unless the list is empty.
156 		 */
157 		work = READ_ONCE(task->task_works);
158 		do {
159 			head = NULL;
160 			if (!work) {
161 				if (task->flags & PF_EXITING)
162 					head = &work_exited;
163 				else
164 					break;
165 			}
166 		} while (!try_cmpxchg(&task->task_works, &work, head));
167 
168 		if (!work)
169 			break;
170 		/*
171 		 * Synchronize with task_work_cancel(). It can not remove
172 		 * the first entry == work, cmpxchg(task_works) must fail.
173 		 * But it can remove another entry from the ->next list.
174 		 */
175 		raw_spin_lock_irq(&task->pi_lock);
176 		raw_spin_unlock_irq(&task->pi_lock);
177 
178 		do {
179 			next = work->next;
180 			work->func(work);
181 			work = next;
182 			cond_resched();
183 		} while (work);
184 	}
185 }
186