1 /* SPDX-License-Identifier: GPL-2.0 */ 1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef _LINUX_SWAIT_H 2 #ifndef _LINUX_SWAIT_H
3 #define _LINUX_SWAIT_H 3 #define _LINUX_SWAIT_H
4 4
5 #include <linux/list.h> 5 #include <linux/list.h>
6 #include <linux/stddef.h> 6 #include <linux/stddef.h>
7 #include <linux/spinlock.h> 7 #include <linux/spinlock.h>
8 #include <linux/wait.h> 8 #include <linux/wait.h>
9 #include <asm/current.h> 9 #include <asm/current.h>
10 10
11 /* 11 /*
12 * Simple waitqueues are semantically very dif !! 12 * BROKEN wait-queues.
13 * (wait.h). The most important difference is !! 13 *
14 * for deterministic behaviour -- IOW it has s !! 14 * These "simple" wait-queues are broken garbage, and should never be
15 * times. !! 15 * used. The comments below claim that they are "similar" to regular
>> 16 * wait-queues, but the semantics are actually completely different, and
>> 17 * every single user we have ever had has been buggy (or pointless).
>> 18 *
>> 19 * A "swake_up_one()" only wakes up _one_ waiter, which is not at all what
>> 20 * "wake_up()" does, and has led to problems. In other cases, it has
>> 21 * been fine, because there's only ever one waiter (kvm), but in that
>> 22 * case gthe whole "simple" wait-queue is just pointless to begin with,
>> 23 * since there is no "queue". Use "wake_up_process()" with a direct
>> 24 * pointer instead.
>> 25 *
>> 26 * While these are very similar to regular wait queues (wait.h) the most
>> 27 * important difference is that the simple waitqueue allows for deterministic
>> 28 * behaviour -- IOW it has strictly bounded IRQ and lock hold times.
16 * 29 *
17 * Mainly, this is accomplished by two things. 30 * Mainly, this is accomplished by two things. Firstly not allowing swake_up_all
18 * from IRQ disabled, and dropping the lock up 31 * from IRQ disabled, and dropping the lock upon every wakeup, giving a higher
19 * priority task a chance to run. 32 * priority task a chance to run.
20 * 33 *
21 * Secondly, we had to drop a fair number of f 34 * Secondly, we had to drop a fair number of features of the other waitqueue
22 * code; notably: 35 * code; notably:
23 * 36 *
24 * - mixing INTERRUPTIBLE and UNINTERRUPTIBLE 37 * - mixing INTERRUPTIBLE and UNINTERRUPTIBLE sleeps on the same waitqueue;
25 * all wakeups are TASK_NORMAL in order to 38 * all wakeups are TASK_NORMAL in order to avoid O(n) lookups for the right
26 * sleeper state. 39 * sleeper state.
27 * 40 *
28 * - the !exclusive mode; because that leads 41 * - the !exclusive mode; because that leads to O(n) wakeups, everything is
29 * exclusive. As such swake_up_one will onl !! 42 * exclusive.
30 * 43 *
31 * - custom wake callback functions; because 44 * - custom wake callback functions; because you cannot give any guarantees
32 * about random code. This also allows swai 45 * about random code. This also allows swait to be used in RT, such that
33 * raw spinlock can be used for the swait q 46 * raw spinlock can be used for the swait queue head.
34 * 47 *
35 * As a side effect of these; the data structu 48 * As a side effect of these; the data structures are slimmer albeit more ad-hoc.
36 * For all the above, note that simple wait qu 49 * For all the above, note that simple wait queues should _only_ be used under
37 * very specific realtime constraints -- it is 50 * very specific realtime constraints -- it is best to stick with the regular
38 * wait queues in most cases. 51 * wait queues in most cases.
39 */ 52 */
40 53
41 struct task_struct; 54 struct task_struct;
42 55
43 struct swait_queue_head { 56 struct swait_queue_head {
44 raw_spinlock_t lock; 57 raw_spinlock_t lock;
45 struct list_head task_list; 58 struct list_head task_list;
46 }; 59 };
47 60
48 struct swait_queue { 61 struct swait_queue {
49 struct task_struct *task; 62 struct task_struct *task;
50 struct list_head task_list; 63 struct list_head task_list;
51 }; 64 };
52 65
53 #define __SWAITQUEUE_INITIALIZER(name) { 66 #define __SWAITQUEUE_INITIALIZER(name) { \
54 .task = current, 67 .task = current, \
55 .task_list = LIST_HEAD_INIT((name 68 .task_list = LIST_HEAD_INIT((name).task_list), \
56 } 69 }
57 70
58 #define DECLARE_SWAITQUEUE(name) 71 #define DECLARE_SWAITQUEUE(name) \
59 struct swait_queue name = __SWAITQUEUE 72 struct swait_queue name = __SWAITQUEUE_INITIALIZER(name)
60 73
61 #define __SWAIT_QUEUE_HEAD_INITIALIZER(name) { 74 #define __SWAIT_QUEUE_HEAD_INITIALIZER(name) { \
62 .lock = __RAW_SPIN_LOCK_UNLO 75 .lock = __RAW_SPIN_LOCK_UNLOCKED(name.lock), \
63 .task_list = LIST_HEAD_INIT((name 76 .task_list = LIST_HEAD_INIT((name).task_list), \
64 } 77 }
65 78
66 #define DECLARE_SWAIT_QUEUE_HEAD(name) 79 #define DECLARE_SWAIT_QUEUE_HEAD(name) \
67 struct swait_queue_head name = __SWAIT 80 struct swait_queue_head name = __SWAIT_QUEUE_HEAD_INITIALIZER(name)
68 81
69 extern void __init_swait_queue_head(struct swa 82 extern void __init_swait_queue_head(struct swait_queue_head *q, const char *name,
70 struct loc 83 struct lock_class_key *key);
71 84
72 #define init_swait_queue_head(q) 85 #define init_swait_queue_head(q) \
73 do { 86 do { \
74 static struct lock_class_key _ 87 static struct lock_class_key __key; \
75 __init_swait_queue_head((q), # 88 __init_swait_queue_head((q), #q, &__key); \
76 } while (0) 89 } while (0)
77 90
78 #ifdef CONFIG_LOCKDEP 91 #ifdef CONFIG_LOCKDEP
79 # define __SWAIT_QUEUE_HEAD_INIT_ONSTACK(name) 92 # define __SWAIT_QUEUE_HEAD_INIT_ONSTACK(name) \
80 ({ init_swait_queue_head(&name); name; 93 ({ init_swait_queue_head(&name); name; })
81 # define DECLARE_SWAIT_QUEUE_HEAD_ONSTACK(name 94 # define DECLARE_SWAIT_QUEUE_HEAD_ONSTACK(name) \
82 struct swait_queue_head name = __SWAIT 95 struct swait_queue_head name = __SWAIT_QUEUE_HEAD_INIT_ONSTACK(name)
83 #else 96 #else
84 # define DECLARE_SWAIT_QUEUE_HEAD_ONSTACK(name 97 # define DECLARE_SWAIT_QUEUE_HEAD_ONSTACK(name) \
85 DECLARE_SWAIT_QUEUE_HEAD(name) 98 DECLARE_SWAIT_QUEUE_HEAD(name)
86 #endif 99 #endif
87 100
88 /** 101 /**
89 * swait_active -- locklessly test for waiters 102 * swait_active -- locklessly test for waiters on the queue
90 * @wq: the waitqueue to test for waiters 103 * @wq: the waitqueue to test for waiters
91 * 104 *
92 * returns true if the wait list is not empty 105 * returns true if the wait list is not empty
93 * 106 *
94 * NOTE: this function is lockless and require 107 * NOTE: this function is lockless and requires care, incorrect usage _will_
95 * lead to sporadic and non-obvious failure. 108 * lead to sporadic and non-obvious failure.
96 * 109 *
97 * NOTE2: this function has the same above imp 110 * NOTE2: this function has the same above implications as regular waitqueues.
98 * 111 *
99 * Use either while holding swait_queue_head:: 112 * Use either while holding swait_queue_head::lock or when used for wakeups
100 * with an extra smp_mb() like: 113 * with an extra smp_mb() like:
101 * 114 *
102 * CPU0 - waker CPU1 - 115 * CPU0 - waker CPU1 - waiter
103 * 116 *
104 * for (; 117 * for (;;) {
105 * @cond = true; prep 118 * @cond = true; prepare_to_swait_exclusive(&wq_head, &wait, state);
106 * smp_mb(); // s 119 * smp_mb(); // smp_mb() from set_current_state()
107 * if (swait_active(wq_head)) if ( 120 * if (swait_active(wq_head)) if (@cond)
108 * wake_up(wq_head); 121 * wake_up(wq_head); break;
109 * sche 122 * schedule();
110 * } 123 * }
111 * finish 124 * finish_swait(&wq_head, &wait);
112 * 125 *
113 * Because without the explicit smp_mb() it's 126 * Because without the explicit smp_mb() it's possible for the
114 * swait_active() load to get hoisted over the 127 * swait_active() load to get hoisted over the @cond store such that we'll
115 * observe an empty wait list while the waiter 128 * observe an empty wait list while the waiter might not observe @cond.
116 * This, in turn, can trigger missing wakeups. 129 * This, in turn, can trigger missing wakeups.
117 * 130 *
118 * Also note that this 'optimization' trades a 131 * Also note that this 'optimization' trades a spin_lock() for an smp_mb(),
119 * which (when the lock is uncontended) are of 132 * which (when the lock is uncontended) are of roughly equal cost.
120 */ 133 */
121 static inline int swait_active(struct swait_qu 134 static inline int swait_active(struct swait_queue_head *wq)
122 { 135 {
123 return !list_empty(&wq->task_list); 136 return !list_empty(&wq->task_list);
124 } 137 }
125 138
126 /** 139 /**
127 * swq_has_sleeper - check if there are any wa 140 * swq_has_sleeper - check if there are any waiting processes
128 * @wq: the waitqueue to test for waiters 141 * @wq: the waitqueue to test for waiters
129 * 142 *
130 * Returns true if @wq has waiting processes 143 * Returns true if @wq has waiting processes
131 * 144 *
132 * Please refer to the comment for swait_activ 145 * Please refer to the comment for swait_active.
133 */ 146 */
134 static inline bool swq_has_sleeper(struct swai 147 static inline bool swq_has_sleeper(struct swait_queue_head *wq)
135 { 148 {
136 /* 149 /*
137 * We need to be sure we are in sync w 150 * We need to be sure we are in sync with the list_add()
138 * modifications to the wait queue (ta 151 * modifications to the wait queue (task_list).
139 * 152 *
140 * This memory barrier should be paire 153 * This memory barrier should be paired with one on the
141 * waiting side. 154 * waiting side.
142 */ 155 */
143 smp_mb(); 156 smp_mb();
144 return swait_active(wq); 157 return swait_active(wq);
145 } 158 }
146 159
147 extern void swake_up_one(struct swait_queue_he 160 extern void swake_up_one(struct swait_queue_head *q);
148 extern void swake_up_all(struct swait_queue_he 161 extern void swake_up_all(struct swait_queue_head *q);
149 extern void swake_up_locked(struct swait_queue !! 162 extern void swake_up_locked(struct swait_queue_head *q);
150 163
151 extern void prepare_to_swait_exclusive(struct 164 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 swai 165 extern long prepare_to_swait_event(struct swait_queue_head *q, struct swait_queue *wait, int state);
153 166
154 extern void __finish_swait(struct swait_queue_ 167 extern void __finish_swait(struct swait_queue_head *q, struct swait_queue *wait);
155 extern void finish_swait(struct swait_queue_he 168 extern void finish_swait(struct swait_queue_head *q, struct swait_queue *wait);
156 169
157 /* as per ___wait_event() but for swait, there 170 /* as per ___wait_event() but for swait, therefore "exclusive == 1" */
158 #define ___swait_event(wq, condition, state, r 171 #define ___swait_event(wq, condition, state, ret, cmd) \
159 ({ 172 ({ \
160 __label__ __out; 173 __label__ __out; \
161 struct swait_queue __wait; 174 struct swait_queue __wait; \
162 long __ret = ret; 175 long __ret = ret; \
163 176 \
164 INIT_LIST_HEAD(&__wait.task_list); 177 INIT_LIST_HEAD(&__wait.task_list); \
165 for (;;) { 178 for (;;) { \
166 long __int = prepare_to_swait_ 179 long __int = prepare_to_swait_event(&wq, &__wait, state);\
167 180 \
168 if (condition) 181 if (condition) \
169 break; 182 break; \
170 183 \
171 if (___wait_is_interruptible(s 184 if (___wait_is_interruptible(state) && __int) { \
172 __ret = __int; 185 __ret = __int; \
173 goto __out; 186 goto __out; \
174 } 187 } \
175 188 \
176 cmd; 189 cmd; \
177 } 190 } \
178 finish_swait(&wq, &__wait); 191 finish_swait(&wq, &__wait); \
179 __out: __ret; 192 __out: __ret; \
180 }) 193 })
181 194
182 #define __swait_event(wq, condition) 195 #define __swait_event(wq, condition) \
183 (void)___swait_event(wq, condition, TA 196 (void)___swait_event(wq, condition, TASK_UNINTERRUPTIBLE, 0, \
184 schedule()) 197 schedule())
185 198
186 #define swait_event_exclusive(wq, condition) 199 #define swait_event_exclusive(wq, condition) \
187 do { 200 do { \
188 if (condition) 201 if (condition) \
189 break; 202 break; \
190 __swait_event(wq, condition); 203 __swait_event(wq, condition); \
191 } while (0) 204 } while (0)
192 205
193 #define __swait_event_timeout(wq, condition, t 206 #define __swait_event_timeout(wq, condition, timeout) \
194 ___swait_event(wq, ___wait_cond_timeou 207 ___swait_event(wq, ___wait_cond_timeout(condition), \
195 TASK_UNINTERRUPTIBLE, ti 208 TASK_UNINTERRUPTIBLE, timeout, \
196 __ret = schedule_timeout 209 __ret = schedule_timeout(__ret))
197 210
198 #define swait_event_timeout_exclusive(wq, cond 211 #define swait_event_timeout_exclusive(wq, condition, timeout) \
199 ({ 212 ({ \
200 long __ret = timeout; 213 long __ret = timeout; \
201 if (!___wait_cond_timeout(condition)) 214 if (!___wait_cond_timeout(condition)) \
202 __ret = __swait_event_timeout( 215 __ret = __swait_event_timeout(wq, condition, timeout); \
203 __ret; 216 __ret; \
204 }) 217 })
205 218
206 #define __swait_event_interruptible(wq, condit 219 #define __swait_event_interruptible(wq, condition) \
207 ___swait_event(wq, condition, TASK_INT 220 ___swait_event(wq, condition, TASK_INTERRUPTIBLE, 0, \
208 schedule()) 221 schedule())
209 222
210 #define swait_event_interruptible_exclusive(wq 223 #define swait_event_interruptible_exclusive(wq, condition) \
211 ({ 224 ({ \
212 int __ret = 0; 225 int __ret = 0; \
213 if (!(condition)) 226 if (!(condition)) \
214 __ret = __swait_event_interrup 227 __ret = __swait_event_interruptible(wq, condition); \
215 __ret; 228 __ret; \
216 }) 229 })
217 230
218 #define __swait_event_interruptible_timeout(wq 231 #define __swait_event_interruptible_timeout(wq, condition, timeout) \
219 ___swait_event(wq, ___wait_cond_timeou 232 ___swait_event(wq, ___wait_cond_timeout(condition), \
220 TASK_INTERRUPTIBLE, time 233 TASK_INTERRUPTIBLE, timeout, \
221 __ret = schedule_timeout 234 __ret = schedule_timeout(__ret))
222 235
223 #define swait_event_interruptible_timeout_excl 236 #define swait_event_interruptible_timeout_exclusive(wq, condition, timeout)\
224 ({ 237 ({ \
225 long __ret = timeout; 238 long __ret = timeout; \
226 if (!___wait_cond_timeout(condition)) 239 if (!___wait_cond_timeout(condition)) \
227 __ret = __swait_event_interrup 240 __ret = __swait_event_interruptible_timeout(wq, \
228 241 condition, timeout); \
229 __ret; 242 __ret; \
230 }) 243 })
231 244
232 #define __swait_event_idle(wq, condition) 245 #define __swait_event_idle(wq, condition) \
233 (void)___swait_event(wq, condition, TA 246 (void)___swait_event(wq, condition, TASK_IDLE, 0, schedule())
234 247
235 /** 248 /**
236 * swait_event_idle_exclusive - wait without s 249 * swait_event_idle_exclusive - wait without system load contribution
237 * @wq: the waitqueue to wait on 250 * @wq: the waitqueue to wait on
238 * @condition: a C expression for the event to 251 * @condition: a C expression for the event to wait for
239 * 252 *
240 * The process is put to sleep (TASK_IDLE) unt 253 * The process is put to sleep (TASK_IDLE) until the @condition evaluates to
241 * true. The @condition is checked each time t 254 * true. The @condition is checked each time the waitqueue @wq is woken up.
242 * 255 *
243 * This function is mostly used when a kthread 256 * This function is mostly used when a kthread or workqueue waits for some
244 * condition and doesn't want to contribute to 257 * condition and doesn't want to contribute to system load. Signals are
245 * ignored. 258 * ignored.
246 */ 259 */
247 #define swait_event_idle_exclusive(wq, conditi 260 #define swait_event_idle_exclusive(wq, condition) \
248 do { 261 do { \
249 if (condition) 262 if (condition) \
250 break; 263 break; \
251 __swait_event_idle(wq, condition); 264 __swait_event_idle(wq, condition); \
252 } while (0) 265 } while (0)
253 266
254 #define __swait_event_idle_timeout(wq, conditi 267 #define __swait_event_idle_timeout(wq, condition, timeout) \
255 ___swait_event(wq, ___wait_cond_timeou 268 ___swait_event(wq, ___wait_cond_timeout(condition), \
256 TASK_IDLE, timeout, 269 TASK_IDLE, timeout, \
257 __ret = schedule_timeou 270 __ret = schedule_timeout(__ret))
258 271
259 /** 272 /**
260 * swait_event_idle_timeout_exclusive - wait u 273 * swait_event_idle_timeout_exclusive - wait up to timeout without load contribution
261 * @wq: the waitqueue to wait on 274 * @wq: the waitqueue to wait on
262 * @condition: a C expression for the event to 275 * @condition: a C expression for the event to wait for
263 * @timeout: timeout at which we'll give up in 276 * @timeout: timeout at which we'll give up in jiffies
264 * 277 *
265 * The process is put to sleep (TASK_IDLE) unt 278 * The process is put to sleep (TASK_IDLE) until the @condition evaluates to
266 * true. The @condition is checked each time t 279 * true. The @condition is checked each time the waitqueue @wq is woken up.
267 * 280 *
268 * This function is mostly used when a kthread 281 * This function is mostly used when a kthread or workqueue waits for some
269 * condition and doesn't want to contribute to 282 * condition and doesn't want to contribute to system load. Signals are
270 * ignored. 283 * ignored.
271 * 284 *
272 * Returns: 285 * Returns:
273 * 0 if the @condition evaluated to %false aft 286 * 0 if the @condition evaluated to %false after the @timeout elapsed,
274 * 1 if the @condition evaluated to %true afte 287 * 1 if the @condition evaluated to %true after the @timeout elapsed,
275 * or the remaining jiffies (at least 1) if th 288 * or the remaining jiffies (at least 1) if the @condition evaluated
276 * to %true before the @timeout elapsed. 289 * to %true before the @timeout elapsed.
277 */ 290 */
278 #define swait_event_idle_timeout_exclusive(wq, 291 #define swait_event_idle_timeout_exclusive(wq, condition, timeout) \
279 ({ 292 ({ \
280 long __ret = timeout; 293 long __ret = timeout; \
281 if (!___wait_cond_timeout(condition)) 294 if (!___wait_cond_timeout(condition)) \
282 __ret = __swait_event_idle_tim 295 __ret = __swait_event_idle_timeout(wq, \
283 296 condition, timeout); \
284 __ret; 297 __ret; \
285 }) 298 })
286 299
287 #endif /* _LINUX_SWAIT_H */ 300 #endif /* _LINUX_SWAIT_H */
288 301
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.