xref: /openbmc/linux/kernel/task_work.c (revision c67ce71d)
1 // SPDX-License-Identifier: GPL-2.0
2 #include <linux/irq_work.h>
3 #include <linux/spinlock.h>
4 #include <linux/task_work.h>
5 #include <linux/resume_user_mode.h>
6 
7 static struct callback_head work_exited; /* all we need is ->next == NULL */
8 
9 #ifdef CONFIG_IRQ_WORK
10 static void task_work_set_notify_irq(struct irq_work *entry)
11 {
12 	test_and_set_tsk_thread_flag(current, TIF_NOTIFY_RESUME);
13 }
14 static DEFINE_PER_CPU(struct irq_work, irq_work_NMI_resume) =
15 	IRQ_WORK_INIT_HARD(task_work_set_notify_irq);
16 #endif
17 
18 /**
19  * task_work_add - ask the @task to execute @work->func()
20  * @task: the task which should run the callback
21  * @work: the callback to run
22  * @notify: how to notify the targeted task
23  *
24  * Queue @work for task_work_run() below and notify the @task if @notify
25  * is @TWA_RESUME, @TWA_SIGNAL, @TWA_SIGNAL_NO_IPI or @TWA_NMI_CURRENT.
26  *
27  * @TWA_SIGNAL works like signals, in that the it will interrupt the targeted
28  * task and run the task_work, regardless of whether the task is currently
29  * running in the kernel or userspace.
30  * @TWA_SIGNAL_NO_IPI works like @TWA_SIGNAL, except it doesn't send a
31  * reschedule IPI to force the targeted task to reschedule and run task_work.
32  * This can be advantageous if there's no strict requirement that the
33  * task_work be run as soon as possible, just whenever the task enters the
34  * kernel anyway.
35  * @TWA_RESUME work is run only when the task exits the kernel and returns to
36  * user mode, or before entering guest mode.
37  * @TWA_NMI_CURRENT works like @TWA_RESUME, except it can only be used for the
38  * current @task and if the current context is NMI.
39  *
40  * Fails if the @task is exiting/exited and thus it can't process this @work.
41  * Otherwise @work->func() will be called when the @task goes through one of
42  * the aforementioned transitions, or exits.
43  *
44  * If the targeted task is exiting, then an error is returned and the work item
45  * is not queued. It's up to the caller to arrange for an alternative mechanism
46  * in that case.
47  *
48  * Note: there is no ordering guarantee on works queued here. The task_work
49  * list is LIFO.
50  *
51  * RETURNS:
52  * 0 if succeeds or -ESRCH.
53  */
54 int task_work_add(struct task_struct *task, struct callback_head *work,
55 		  enum task_work_notify_mode notify)
56 {
57 	struct callback_head *head;
58 	int flags = notify & TWA_FLAGS;
59 
60 	notify &= ~TWA_FLAGS;
61 	if (notify == TWA_NMI_CURRENT) {
62 		if (WARN_ON_ONCE(task != current))
63 			return -EINVAL;
64 		if (!IS_ENABLED(CONFIG_IRQ_WORK))
65 			return -EINVAL;
66 	} else {
67 		/*
68 		 * Record the work call stack in order to print it in KASAN
69 		 * reports.
70 		 *
71 		 * Note that stack allocation can fail if TWAF_NO_ALLOC flag
72 		 * is set and new page is needed to expand the stack buffer.
73 		 */
74 		if (flags & TWAF_NO_ALLOC)
75 			kasan_record_aux_stack_noalloc(work);
76 		else
77 			kasan_record_aux_stack(work);
78 	}
79 
80 	head = READ_ONCE(task->task_works);
81 	do {
82 		if (unlikely(head == &work_exited))
83 			return -ESRCH;
84 		work->next = head;
85 	} while (!try_cmpxchg(&task->task_works, &head, work));
86 
87 	switch (notify) {
88 	case TWA_NONE:
89 		break;
90 	case TWA_RESUME:
91 		set_notify_resume(task);
92 		break;
93 	case TWA_SIGNAL:
94 		set_notify_signal(task);
95 		break;
96 	case TWA_SIGNAL_NO_IPI:
97 		__set_notify_signal(task);
98 		break;
99 #ifdef CONFIG_IRQ_WORK
100 	case TWA_NMI_CURRENT:
101 		irq_work_queue(this_cpu_ptr(&irq_work_NMI_resume));
102 		break;
103 #endif
104 	default:
105 		WARN_ON_ONCE(1);
106 		break;
107 	}
108 
109 	return 0;
110 }
111 
112 /**
113  * task_work_cancel_match - cancel a pending work added by task_work_add()
114  * @task: the task which should execute the work
115  * @match: match function to call
116  * @data: data to be passed in to match function
117  *
118  * RETURNS:
119  * The found work or NULL if not found.
120  */
121 struct callback_head *
122 task_work_cancel_match(struct task_struct *task,
123 		       bool (*match)(struct callback_head *, void *data),
124 		       void *data)
125 {
126 	struct callback_head **pprev = &task->task_works;
127 	struct callback_head *work;
128 	unsigned long flags;
129 
130 	if (likely(!task_work_pending(task)))
131 		return NULL;
132 	/*
133 	 * If cmpxchg() fails we continue without updating pprev.
134 	 * Either we raced with task_work_add() which added the
135 	 * new entry before this work, we will find it again. Or
136 	 * we raced with task_work_run(), *pprev == NULL/exited.
137 	 */
138 	raw_spin_lock_irqsave(&task->pi_lock, flags);
139 	work = READ_ONCE(*pprev);
140 	while (work) {
141 		if (!match(work, data)) {
142 			pprev = &work->next;
143 			work = READ_ONCE(*pprev);
144 		} else if (try_cmpxchg(pprev, &work, work->next))
145 			break;
146 	}
147 	raw_spin_unlock_irqrestore(&task->pi_lock, flags);
148 
149 	return work;
150 }
151 
152 static bool task_work_func_match(struct callback_head *cb, void *data)
153 {
154 	return cb->func == data;
155 }
156 
157 /**
158  * task_work_cancel_func - cancel a pending work matching a function added by task_work_add()
159  * @task: the task which should execute the func's work
160  * @func: identifies the func to match with a work to remove
161  *
162  * Find the last queued pending work with ->func == @func and remove
163  * it from queue.
164  *
165  * RETURNS:
166  * The found work or NULL if not found.
167  */
168 struct callback_head *
169 task_work_cancel_func(struct task_struct *task, task_work_func_t func)
170 {
171 	return task_work_cancel_match(task, task_work_func_match, func);
172 }
173 
174 static bool task_work_match(struct callback_head *cb, void *data)
175 {
176 	return cb == data;
177 }
178 
179 /**
180  * task_work_cancel - cancel a pending work added by task_work_add()
181  * @task: the task which should execute the work
182  * @cb: the callback to remove if queued
183  *
184  * Remove a callback from a task's queue if queued.
185  *
186  * RETURNS:
187  * True if the callback was queued and got cancelled, false otherwise.
188  */
189 bool task_work_cancel(struct task_struct *task, struct callback_head *cb)
190 {
191 	struct callback_head *ret;
192 
193 	ret = task_work_cancel_match(task, task_work_match, cb);
194 
195 	return ret == cb;
196 }
197 
198 /**
199  * task_work_run - execute the works added by task_work_add()
200  *
201  * Flush the pending works. Should be used by the core kernel code.
202  * Called before the task returns to the user-mode or stops, or when
203  * it exits. In the latter case task_work_add() can no longer add the
204  * new work after task_work_run() returns.
205  */
206 void task_work_run(void)
207 {
208 	struct task_struct *task = current;
209 	struct callback_head *work, *head, *next;
210 
211 	for (;;) {
212 		/*
213 		 * work->func() can do task_work_add(), do not set
214 		 * work_exited unless the list is empty.
215 		 */
216 		work = READ_ONCE(task->task_works);
217 		do {
218 			head = NULL;
219 			if (!work) {
220 				if (task->flags & PF_EXITING)
221 					head = &work_exited;
222 				else
223 					break;
224 			}
225 		} while (!try_cmpxchg(&task->task_works, &work, head));
226 
227 		if (!work)
228 			break;
229 		/*
230 		 * Synchronize with task_work_cancel_match(). It can not remove
231 		 * the first entry == work, cmpxchg(task_works) must fail.
232 		 * But it can remove another entry from the ->next list.
233 		 */
234 		raw_spin_lock_irq(&task->pi_lock);
235 		raw_spin_unlock_irq(&task->pi_lock);
236 
237 		do {
238 			next = work->next;
239 			work->func(work);
240 			work = next;
241 			cond_resched();
242 		} while (work);
243 	}
244 }
245