1 | /* SPDX-License-Identifier: GPL-2.0 */ |
2 | #ifndef _LINUX_SWAIT_H |
3 | #define _LINUX_SWAIT_H |
4 | |
5 | #include <linux/list.h> |
6 | #include <linux/stddef.h> |
7 | #include <linux/spinlock.h> |
8 | #include <linux/wait.h> |
9 | #include <asm/current.h> |
10 | |
11 | /* |
12 | * Simple waitqueues are semantically very different to regular wait queues |
13 | * (wait.h). The most important difference is that the simple waitqueue allows |
14 | * for deterministic behaviour -- IOW it has strictly bounded IRQ and lock hold |
15 | * times. |
16 | * |
17 | * Mainly, this is accomplished by two things. Firstly not allowing swake_up_all |
18 | * from IRQ disabled, and dropping the lock upon every wakeup, giving a higher |
19 | * priority task a chance to run. |
20 | * |
21 | * Secondly, we had to drop a fair number of features of the other waitqueue |
22 | * code; notably: |
23 | * |
24 | * - mixing INTERRUPTIBLE and UNINTERRUPTIBLE sleeps on the same waitqueue; |
25 | * all wakeups are TASK_NORMAL in order to avoid O(n) lookups for the right |
26 | * sleeper state. |
27 | * |
28 | * - the !exclusive mode; because that leads to O(n) wakeups, everything is |
29 | * exclusive. As such swake_up_one will only ever awake _one_ waiter. |
30 | * |
31 | * - custom wake callback functions; because you cannot give any guarantees |
32 | * about random code. This also allows swait to be used in RT, such that |
33 | * raw spinlock can be used for the swait queue head. |
34 | * |
35 | * As a side effect of these; the data structures are slimmer albeit more ad-hoc. |
36 | * For all the above, note that simple wait queues should _only_ be used under |
37 | * very specific realtime constraints -- it is best to stick with the regular |
38 | * wait queues in most cases. |
39 | */ |
40 | |
41 | struct task_struct; |
42 | |
43 | struct swait_queue_head { |
44 | raw_spinlock_t lock; |
45 | struct list_head task_list; |
46 | }; |
47 | |
48 | struct swait_queue { |
49 | struct task_struct *task; |
50 | struct list_head task_list; |
51 | }; |
52 | |
53 | #define __SWAITQUEUE_INITIALIZER(name) { \ |
54 | .task = current, \ |
55 | .task_list = LIST_HEAD_INIT((name).task_list), \ |
56 | } |
57 | |
58 | #define DECLARE_SWAITQUEUE(name) \ |
59 | struct swait_queue name = __SWAITQUEUE_INITIALIZER(name) |
60 | |
61 | #define __SWAIT_QUEUE_HEAD_INITIALIZER(name) { \ |
62 | .lock = __RAW_SPIN_LOCK_UNLOCKED(name.lock), \ |
63 | .task_list = LIST_HEAD_INIT((name).task_list), \ |
64 | } |
65 | |
66 | #define DECLARE_SWAIT_QUEUE_HEAD(name) \ |
67 | struct swait_queue_head name = __SWAIT_QUEUE_HEAD_INITIALIZER(name) |
68 | |
69 | extern void __init_swait_queue_head(struct swait_queue_head *q, const char *name, |
70 | struct lock_class_key *key); |
71 | |
72 | #define init_swait_queue_head(q) \ |
73 | do { \ |
74 | static struct lock_class_key __key; \ |
75 | __init_swait_queue_head((q), #q, &__key); \ |
76 | } while (0) |
77 | |
78 | #ifdef CONFIG_LOCKDEP |
79 | # define __SWAIT_QUEUE_HEAD_INIT_ONSTACK(name) \ |
80 | ({ init_swait_queue_head(&name); name; }) |
81 | # define DECLARE_SWAIT_QUEUE_HEAD_ONSTACK(name) \ |
82 | struct swait_queue_head name = __SWAIT_QUEUE_HEAD_INIT_ONSTACK(name) |
83 | #else |
84 | # define DECLARE_SWAIT_QUEUE_HEAD_ONSTACK(name) \ |
85 | DECLARE_SWAIT_QUEUE_HEAD(name) |
86 | #endif |
87 | |
88 | /** |
89 | * swait_active -- locklessly test for waiters on the queue |
90 | * @wq: the waitqueue to test for waiters |
91 | * |
92 | * returns true if the wait list is not empty |
93 | * |
94 | * NOTE: this function is lockless and requires care, incorrect usage _will_ |
95 | * lead to sporadic and non-obvious failure. |
96 | * |
97 | * NOTE2: this function has the same above implications as regular waitqueues. |
98 | * |
99 | * Use either while holding swait_queue_head::lock or when used for wakeups |
100 | * with an extra smp_mb() like: |
101 | * |
102 | * CPU0 - waker CPU1 - waiter |
103 | * |
104 | * for (;;) { |
105 | * @cond = true; prepare_to_swait_exclusive(&wq_head, &wait, state); |
106 | * smp_mb(); // smp_mb() from set_current_state() |
107 | * if (swait_active(wq_head)) if (@cond) |
108 | * wake_up(wq_head); break; |
109 | * schedule(); |
110 | * } |
111 | * finish_swait(&wq_head, &wait); |
112 | * |
113 | * Because without the explicit smp_mb() it's possible for the |
114 | * swait_active() load to get hoisted over the @cond store such that we'll |
115 | * observe an empty wait list while the waiter might not observe @cond. |
116 | * This, in turn, can trigger missing wakeups. |
117 | * |
118 | * Also note that this 'optimization' trades a spin_lock() for an smp_mb(), |
119 | * which (when the lock is uncontended) are of roughly equal cost. |
120 | */ |
121 | static inline int swait_active(struct swait_queue_head *wq) |
122 | { |
123 | return !list_empty(head: &wq->task_list); |
124 | } |
125 | |
126 | /** |
127 | * swq_has_sleeper - check if there are any waiting processes |
128 | * @wq: the waitqueue to test for waiters |
129 | * |
130 | * Returns true if @wq has waiting processes |
131 | * |
132 | * Please refer to the comment for swait_active. |
133 | */ |
134 | static inline bool swq_has_sleeper(struct swait_queue_head *wq) |
135 | { |
136 | /* |
137 | * We need to be sure we are in sync with the list_add() |
138 | * modifications to the wait queue (task_list). |
139 | * |
140 | * This memory barrier should be paired with one on the |
141 | * waiting side. |
142 | */ |
143 | smp_mb(); |
144 | return swait_active(wq); |
145 | } |
146 | |
147 | extern void swake_up_one(struct swait_queue_head *q); |
148 | extern void swake_up_all(struct swait_queue_head *q); |
149 | extern void swake_up_locked(struct swait_queue_head *q, int wake_flags); |
150 | |
151 | extern void prepare_to_swait_exclusive(struct swait_queue_head *q, struct swait_queue *wait, int state); |
152 | extern long prepare_to_swait_event(struct swait_queue_head *q, struct swait_queue *wait, int state); |
153 | |
154 | extern void __finish_swait(struct swait_queue_head *q, struct swait_queue *wait); |
155 | extern void finish_swait(struct swait_queue_head *q, struct swait_queue *wait); |
156 | |
157 | /* as per ___wait_event() but for swait, therefore "exclusive == 1" */ |
158 | #define ___swait_event(wq, condition, state, ret, cmd) \ |
159 | ({ \ |
160 | __label__ __out; \ |
161 | struct swait_queue __wait; \ |
162 | long __ret = ret; \ |
163 | \ |
164 | INIT_LIST_HEAD(&__wait.task_list); \ |
165 | for (;;) { \ |
166 | long __int = prepare_to_swait_event(&wq, &__wait, state);\ |
167 | \ |
168 | if (condition) \ |
169 | break; \ |
170 | \ |
171 | if (___wait_is_interruptible(state) && __int) { \ |
172 | __ret = __int; \ |
173 | goto __out; \ |
174 | } \ |
175 | \ |
176 | cmd; \ |
177 | } \ |
178 | finish_swait(&wq, &__wait); \ |
179 | __out: __ret; \ |
180 | }) |
181 | |
182 | #define __swait_event(wq, condition) \ |
183 | (void)___swait_event(wq, condition, TASK_UNINTERRUPTIBLE, 0, \ |
184 | schedule()) |
185 | |
186 | #define swait_event_exclusive(wq, condition) \ |
187 | do { \ |
188 | if (condition) \ |
189 | break; \ |
190 | __swait_event(wq, condition); \ |
191 | } while (0) |
192 | |
193 | #define __swait_event_timeout(wq, condition, timeout) \ |
194 | ___swait_event(wq, ___wait_cond_timeout(condition), \ |
195 | TASK_UNINTERRUPTIBLE, timeout, \ |
196 | __ret = schedule_timeout(__ret)) |
197 | |
198 | #define swait_event_timeout_exclusive(wq, condition, timeout) \ |
199 | ({ \ |
200 | long __ret = timeout; \ |
201 | if (!___wait_cond_timeout(condition)) \ |
202 | __ret = __swait_event_timeout(wq, condition, timeout); \ |
203 | __ret; \ |
204 | }) |
205 | |
206 | #define __swait_event_interruptible(wq, condition) \ |
207 | ___swait_event(wq, condition, TASK_INTERRUPTIBLE, 0, \ |
208 | schedule()) |
209 | |
210 | #define swait_event_interruptible_exclusive(wq, condition) \ |
211 | ({ \ |
212 | int __ret = 0; \ |
213 | if (!(condition)) \ |
214 | __ret = __swait_event_interruptible(wq, condition); \ |
215 | __ret; \ |
216 | }) |
217 | |
218 | #define __swait_event_interruptible_timeout(wq, condition, timeout) \ |
219 | ___swait_event(wq, ___wait_cond_timeout(condition), \ |
220 | TASK_INTERRUPTIBLE, timeout, \ |
221 | __ret = schedule_timeout(__ret)) |
222 | |
223 | #define swait_event_interruptible_timeout_exclusive(wq, condition, timeout)\ |
224 | ({ \ |
225 | long __ret = timeout; \ |
226 | if (!___wait_cond_timeout(condition)) \ |
227 | __ret = __swait_event_interruptible_timeout(wq, \ |
228 | condition, timeout); \ |
229 | __ret; \ |
230 | }) |
231 | |
232 | #define __swait_event_idle(wq, condition) \ |
233 | (void)___swait_event(wq, condition, TASK_IDLE, 0, schedule()) |
234 | |
235 | /** |
236 | * swait_event_idle_exclusive - wait without system load contribution |
237 | * @wq: the waitqueue to wait on |
238 | * @condition: a C expression for the event to wait for |
239 | * |
240 | * The process is put to sleep (TASK_IDLE) until the @condition evaluates to |
241 | * true. The @condition is checked each time the waitqueue @wq is woken up. |
242 | * |
243 | * This function is mostly used when a kthread or workqueue waits for some |
244 | * condition and doesn't want to contribute to system load. Signals are |
245 | * ignored. |
246 | */ |
247 | #define swait_event_idle_exclusive(wq, condition) \ |
248 | do { \ |
249 | if (condition) \ |
250 | break; \ |
251 | __swait_event_idle(wq, condition); \ |
252 | } while (0) |
253 | |
254 | #define __swait_event_idle_timeout(wq, condition, timeout) \ |
255 | ___swait_event(wq, ___wait_cond_timeout(condition), \ |
256 | TASK_IDLE, timeout, \ |
257 | __ret = schedule_timeout(__ret)) |
258 | |
259 | /** |
260 | * swait_event_idle_timeout_exclusive - wait up to timeout without load contribution |
261 | * @wq: the waitqueue to wait on |
262 | * @condition: a C expression for the event to wait for |
263 | * @timeout: timeout at which we'll give up in jiffies |
264 | * |
265 | * The process is put to sleep (TASK_IDLE) until the @condition evaluates to |
266 | * true. The @condition is checked each time the waitqueue @wq is woken up. |
267 | * |
268 | * This function is mostly used when a kthread or workqueue waits for some |
269 | * condition and doesn't want to contribute to system load. Signals are |
270 | * ignored. |
271 | * |
272 | * Returns: |
273 | * 0 if the @condition evaluated to %false after the @timeout elapsed, |
274 | * 1 if the @condition evaluated to %true after the @timeout elapsed, |
275 | * or the remaining jiffies (at least 1) if the @condition evaluated |
276 | * to %true before the @timeout elapsed. |
277 | */ |
278 | #define swait_event_idle_timeout_exclusive(wq, condition, timeout) \ |
279 | ({ \ |
280 | long __ret = timeout; \ |
281 | if (!___wait_cond_timeout(condition)) \ |
282 | __ret = __swait_event_idle_timeout(wq, \ |
283 | condition, timeout); \ |
284 | __ret; \ |
285 | }) |
286 | |
287 | #endif /* _LINUX_SWAIT_H */ |
288 | |