at v6.17 23 kB view raw
1/* SPDX-License-Identifier: GPL-2.0 */ 2#ifndef _LINUX_WAIT_BIT_H 3#define _LINUX_WAIT_BIT_H 4 5/* 6 * Linux wait-bit related types and methods: 7 */ 8#include <linux/wait.h> 9 10struct wait_bit_key { 11 unsigned long *flags; 12 int bit_nr; 13 unsigned long timeout; 14}; 15 16struct wait_bit_queue_entry { 17 struct wait_bit_key key; 18 struct wait_queue_entry wq_entry; 19}; 20 21#define __WAIT_BIT_KEY_INITIALIZER(word, bit) \ 22 { .flags = word, .bit_nr = bit, } 23 24typedef int wait_bit_action_f(struct wait_bit_key *key, int mode); 25 26void __wake_up_bit(struct wait_queue_head *wq_head, unsigned long *word, int bit); 27int __wait_on_bit(struct wait_queue_head *wq_head, struct wait_bit_queue_entry *wbq_entry, wait_bit_action_f *action, unsigned int mode); 28int __wait_on_bit_lock(struct wait_queue_head *wq_head, struct wait_bit_queue_entry *wbq_entry, wait_bit_action_f *action, unsigned int mode); 29void wake_up_bit(unsigned long *word, int bit); 30int out_of_line_wait_on_bit(unsigned long *word, int, wait_bit_action_f *action, unsigned int mode); 31int out_of_line_wait_on_bit_timeout(unsigned long *word, int, wait_bit_action_f *action, unsigned int mode, unsigned long timeout); 32int out_of_line_wait_on_bit_lock(unsigned long *word, int, wait_bit_action_f *action, unsigned int mode); 33struct wait_queue_head *bit_waitqueue(unsigned long *word, int bit); 34extern void __init wait_bit_init(void); 35 36int wake_bit_function(struct wait_queue_entry *wq_entry, unsigned mode, int sync, void *key); 37 38#define DEFINE_WAIT_BIT(name, word, bit) \ 39 struct wait_bit_queue_entry name = { \ 40 .key = __WAIT_BIT_KEY_INITIALIZER(word, bit), \ 41 .wq_entry = { \ 42 .private = current, \ 43 .func = wake_bit_function, \ 44 .entry = \ 45 LIST_HEAD_INIT((name).wq_entry.entry), \ 46 }, \ 47 } 48 49extern int bit_wait(struct wait_bit_key *key, int mode); 50extern int bit_wait_io(struct wait_bit_key *key, int mode); 51extern int bit_wait_timeout(struct wait_bit_key *key, int mode); 52 53/** 54 * wait_on_bit - wait for a bit to be cleared 55 * @word: the address containing the bit being waited on 56 * @bit: the bit at that address being waited on 57 * @mode: the task state to sleep in 58 * 59 * Wait for the given bit in an unsigned long or bitmap (see DECLARE_BITMAP()) 60 * to be cleared. The clearing of the bit must be signalled with 61 * wake_up_bit(), often as clear_and_wake_up_bit(). 62 * 63 * The process will wait on a waitqueue selected by hash from a shared 64 * pool. It will only be woken on a wake_up for the target bit, even 65 * if other processes on the same queue are waiting for other bits. 66 * 67 * Returned value will be zero if the bit was cleared in which case the 68 * call has ACQUIRE semantics, or %-EINTR if the process received a 69 * signal and the mode permitted wake up on that signal. 70 */ 71static inline int 72wait_on_bit(unsigned long *word, int bit, unsigned mode) 73{ 74 might_sleep(); 75 if (!test_bit_acquire(bit, word)) 76 return 0; 77 return out_of_line_wait_on_bit(word, bit, 78 bit_wait, 79 mode); 80} 81 82/** 83 * wait_on_bit_io - wait for a bit to be cleared 84 * @word: the address containing the bit being waited on 85 * @bit: the bit at that address being waited on 86 * @mode: the task state to sleep in 87 * 88 * Wait for the given bit in an unsigned long or bitmap (see DECLARE_BITMAP()) 89 * to be cleared. The clearing of the bit must be signalled with 90 * wake_up_bit(), often as clear_and_wake_up_bit(). 91 * 92 * This is similar to wait_on_bit(), but calls io_schedule() instead of 93 * schedule() for the actual waiting. 94 * 95 * Returned value will be zero if the bit was cleared in which case the 96 * call has ACQUIRE semantics, or %-EINTR if the process received a 97 * signal and the mode permitted wake up on that signal. 98 */ 99static inline int 100wait_on_bit_io(unsigned long *word, int bit, unsigned mode) 101{ 102 might_sleep(); 103 if (!test_bit_acquire(bit, word)) 104 return 0; 105 return out_of_line_wait_on_bit(word, bit, 106 bit_wait_io, 107 mode); 108} 109 110/** 111 * wait_on_bit_timeout - wait for a bit to be cleared or a timeout to elapse 112 * @word: the address containing the bit being waited on 113 * @bit: the bit at that address being waited on 114 * @mode: the task state to sleep in 115 * @timeout: timeout, in jiffies 116 * 117 * Wait for the given bit in an unsigned long or bitmap (see 118 * DECLARE_BITMAP()) to be cleared, or for a timeout to expire. The 119 * clearing of the bit must be signalled with wake_up_bit(), often as 120 * clear_and_wake_up_bit(). 121 * 122 * This is similar to wait_on_bit(), except it also takes a timeout 123 * parameter. 124 * 125 * Returned value will be zero if the bit was cleared in which case the 126 * call has ACQUIRE semantics, or %-EINTR if the process received a 127 * signal and the mode permitted wake up on that signal, or %-EAGAIN if the 128 * timeout elapsed. 129 */ 130static inline int 131wait_on_bit_timeout(unsigned long *word, int bit, unsigned mode, 132 unsigned long timeout) 133{ 134 might_sleep(); 135 if (!test_bit_acquire(bit, word)) 136 return 0; 137 return out_of_line_wait_on_bit_timeout(word, bit, 138 bit_wait_timeout, 139 mode, timeout); 140} 141 142/** 143 * wait_on_bit_action - wait for a bit to be cleared 144 * @word: the address containing the bit waited on 145 * @bit: the bit at that address being waited on 146 * @action: the function used to sleep, which may take special actions 147 * @mode: the task state to sleep in 148 * 149 * Wait for the given bit in an unsigned long or bitmap (see DECLARE_BITMAP()) 150 * to be cleared. The clearing of the bit must be signalled with 151 * wake_up_bit(), often as clear_and_wake_up_bit(). 152 * 153 * This is similar to wait_on_bit(), but calls @action() instead of 154 * schedule() for the actual waiting. 155 * 156 * Returned value will be zero if the bit was cleared in which case the 157 * call has ACQUIRE semantics, or the error code returned by @action if 158 * that call returned non-zero. 159 */ 160static inline int 161wait_on_bit_action(unsigned long *word, int bit, wait_bit_action_f *action, 162 unsigned mode) 163{ 164 might_sleep(); 165 if (!test_bit_acquire(bit, word)) 166 return 0; 167 return out_of_line_wait_on_bit(word, bit, action, mode); 168} 169 170/** 171 * wait_on_bit_lock - wait for a bit to be cleared, then set it 172 * @word: the address containing the bit being waited on 173 * @bit: the bit of the word being waited on and set 174 * @mode: the task state to sleep in 175 * 176 * Wait for the given bit in an unsigned long or bitmap (see 177 * DECLARE_BITMAP()) to be cleared. The clearing of the bit must be 178 * signalled with wake_up_bit(), often as clear_and_wake_up_bit(). As 179 * soon as it is clear, atomically set it and return. 180 * 181 * This is similar to wait_on_bit(), but sets the bit before returning. 182 * 183 * Returned value will be zero if the bit was successfully set in which 184 * case the call has the same memory sequencing semantics as 185 * test_and_clear_bit(), or %-EINTR if the process received a signal and 186 * the mode permitted wake up on that signal. 187 */ 188static inline int 189wait_on_bit_lock(unsigned long *word, int bit, unsigned mode) 190{ 191 might_sleep(); 192 if (!test_and_set_bit(bit, word)) 193 return 0; 194 return out_of_line_wait_on_bit_lock(word, bit, bit_wait, mode); 195} 196 197/** 198 * wait_on_bit_lock_io - wait for a bit to be cleared, then set it 199 * @word: the address containing the bit being waited on 200 * @bit: the bit of the word being waited on and set 201 * @mode: the task state to sleep in 202 * 203 * Wait for the given bit in an unsigned long or bitmap (see 204 * DECLARE_BITMAP()) to be cleared. The clearing of the bit must be 205 * signalled with wake_up_bit(), often as clear_and_wake_up_bit(). As 206 * soon as it is clear, atomically set it and return. 207 * 208 * This is similar to wait_on_bit_lock(), but calls io_schedule() instead 209 * of schedule(). 210 * 211 * Returns zero if the bit was (eventually) found to be clear and was 212 * set. Returns non-zero if a signal was delivered to the process and 213 * the @mode allows that signal to wake the process. 214 */ 215static inline int 216wait_on_bit_lock_io(unsigned long *word, int bit, unsigned mode) 217{ 218 might_sleep(); 219 if (!test_and_set_bit(bit, word)) 220 return 0; 221 return out_of_line_wait_on_bit_lock(word, bit, bit_wait_io, mode); 222} 223 224/** 225 * wait_on_bit_lock_action - wait for a bit to be cleared, then set it 226 * @word: the address containing the bit being waited on 227 * @bit: the bit of the word being waited on and set 228 * @action: the function used to sleep, which may take special actions 229 * @mode: the task state to sleep in 230 * 231 * This is similar to wait_on_bit_lock(), but calls @action() instead of 232 * schedule() for the actual waiting. 233 * 234 * Returned value will be zero if the bit was successfully set in which 235 * case the call has the same memory sequencing semantics as 236 * test_and_clear_bit(), or the error code returned by @action if that 237 * call returned non-zero. 238 */ 239static inline int 240wait_on_bit_lock_action(unsigned long *word, int bit, wait_bit_action_f *action, 241 unsigned mode) 242{ 243 might_sleep(); 244 if (!test_and_set_bit(bit, word)) 245 return 0; 246 return out_of_line_wait_on_bit_lock(word, bit, action, mode); 247} 248 249extern void init_wait_var_entry(struct wait_bit_queue_entry *wbq_entry, void *var, int flags); 250extern void wake_up_var(void *var); 251extern wait_queue_head_t *__var_waitqueue(void *p); 252 253#define ___wait_var_event(var, condition, state, exclusive, ret, cmd) \ 254({ \ 255 __label__ __out; \ 256 struct wait_queue_head *__wq_head = __var_waitqueue(var); \ 257 struct wait_bit_queue_entry __wbq_entry; \ 258 long __ret = ret; /* explicit shadow */ \ 259 \ 260 init_wait_var_entry(&__wbq_entry, var, \ 261 exclusive ? WQ_FLAG_EXCLUSIVE : 0); \ 262 for (;;) { \ 263 long __int = prepare_to_wait_event(__wq_head, \ 264 &__wbq_entry.wq_entry, \ 265 state); \ 266 if (condition) \ 267 break; \ 268 \ 269 if (___wait_is_interruptible(state) && __int) { \ 270 __ret = __int; \ 271 goto __out; \ 272 } \ 273 \ 274 cmd; \ 275 } \ 276 finish_wait(__wq_head, &__wbq_entry.wq_entry); \ 277__out: __ret; \ 278}) 279 280#define __wait_var_event(var, condition) \ 281 ___wait_var_event(var, condition, TASK_UNINTERRUPTIBLE, 0, 0, \ 282 schedule()) 283#define __wait_var_event_io(var, condition) \ 284 ___wait_var_event(var, condition, TASK_UNINTERRUPTIBLE, 0, 0, \ 285 io_schedule()) 286 287/** 288 * wait_var_event - wait for a variable to be updated and notified 289 * @var: the address of variable being waited on 290 * @condition: the condition to wait for 291 * 292 * Wait for a @condition to be true, only re-checking when a wake up is 293 * received for the given @var (an arbitrary kernel address which need 294 * not be directly related to the given condition, but usually is). 295 * 296 * The process will wait on a waitqueue selected by hash from a shared 297 * pool. It will only be woken on a wake_up for the given address. 298 * 299 * The condition should normally use smp_load_acquire() or a similarly 300 * ordered access to ensure that any changes to memory made before the 301 * condition became true will be visible after the wait completes. 302 */ 303#define wait_var_event(var, condition) \ 304do { \ 305 might_sleep(); \ 306 if (condition) \ 307 break; \ 308 __wait_var_event(var, condition); \ 309} while (0) 310 311/** 312 * wait_var_event_io - wait for a variable to be updated and notified 313 * @var: the address of variable being waited on 314 * @condition: the condition to wait for 315 * 316 * Wait for an IO related @condition to be true, only re-checking when a 317 * wake up is received for the given @var (an arbitrary kernel address 318 * which need not be directly related to the given condition, but 319 * usually is). 320 * 321 * The process will wait on a waitqueue selected by hash from a shared 322 * pool. It will only be woken on a wake_up for the given address. 323 * 324 * This is similar to wait_var_event(), but calls io_schedule() instead 325 * of schedule(). 326 * 327 * The condition should normally use smp_load_acquire() or a similarly 328 * ordered access to ensure that any changes to memory made before the 329 * condition became true will be visible after the wait completes. 330 */ 331#define wait_var_event_io(var, condition) \ 332do { \ 333 might_sleep(); \ 334 if (condition) \ 335 break; \ 336 __wait_var_event_io(var, condition); \ 337} while (0) 338 339#define __wait_var_event_killable(var, condition) \ 340 ___wait_var_event(var, condition, TASK_KILLABLE, 0, 0, \ 341 schedule()) 342 343/** 344 * wait_var_event_killable - wait for a variable to be updated and notified 345 * @var: the address of variable being waited on 346 * @condition: the condition to wait for 347 * 348 * Wait for a @condition to be true or a fatal signal to be received, 349 * only re-checking the condition when a wake up is received for the given 350 * @var (an arbitrary kernel address which need not be directly related 351 * to the given condition, but usually is). 352 * 353 * This is similar to wait_var_event() but returns a value which is 354 * 0 if the condition became true, or %-ERESTARTSYS if a fatal signal 355 * was received. 356 * 357 * The condition should normally use smp_load_acquire() or a similarly 358 * ordered access to ensure that any changes to memory made before the 359 * condition became true will be visible after the wait completes. 360 */ 361#define wait_var_event_killable(var, condition) \ 362({ \ 363 int __ret = 0; \ 364 might_sleep(); \ 365 if (!(condition)) \ 366 __ret = __wait_var_event_killable(var, condition); \ 367 __ret; \ 368}) 369 370#define __wait_var_event_timeout(var, condition, timeout) \ 371 ___wait_var_event(var, ___wait_cond_timeout(condition), \ 372 TASK_UNINTERRUPTIBLE, 0, timeout, \ 373 __ret = schedule_timeout(__ret)) 374 375/** 376 * wait_var_event_timeout - wait for a variable to be updated or a timeout to expire 377 * @var: the address of variable being waited on 378 * @condition: the condition to wait for 379 * @timeout: maximum time to wait in jiffies 380 * 381 * Wait for a @condition to be true or a timeout to expire, only 382 * re-checking the condition when a wake up is received for the given 383 * @var (an arbitrary kernel address which need not be directly related 384 * to the given condition, but usually is). 385 * 386 * This is similar to wait_var_event() but returns a value which is 0 if 387 * the timeout expired and the condition was still false, or the 388 * remaining time left in the timeout (but at least 1) if the condition 389 * was found to be true. 390 * 391 * The condition should normally use smp_load_acquire() or a similarly 392 * ordered access to ensure that any changes to memory made before the 393 * condition became true will be visible after the wait completes. 394 */ 395#define wait_var_event_timeout(var, condition, timeout) \ 396({ \ 397 long __ret = timeout; \ 398 might_sleep(); \ 399 if (!___wait_cond_timeout(condition)) \ 400 __ret = __wait_var_event_timeout(var, condition, timeout); \ 401 __ret; \ 402}) 403 404#define __wait_var_event_interruptible(var, condition) \ 405 ___wait_var_event(var, condition, TASK_INTERRUPTIBLE, 0, 0, \ 406 schedule()) 407 408/** 409 * wait_var_event_killable - wait for a variable to be updated and notified 410 * @var: the address of variable being waited on 411 * @condition: the condition to wait for 412 * 413 * Wait for a @condition to be true or a signal to be received, only 414 * re-checking the condition when a wake up is received for the given 415 * @var (an arbitrary kernel address which need not be directly related 416 * to the given condition, but usually is). 417 * 418 * This is similar to wait_var_event() but returns a value which is 0 if 419 * the condition became true, or %-ERESTARTSYS if a signal was received. 420 * 421 * The condition should normally use smp_load_acquire() or a similarly 422 * ordered access to ensure that any changes to memory made before the 423 * condition became true will be visible after the wait completes. 424 */ 425#define wait_var_event_interruptible(var, condition) \ 426({ \ 427 int __ret = 0; \ 428 might_sleep(); \ 429 if (!(condition)) \ 430 __ret = __wait_var_event_interruptible(var, condition); \ 431 __ret; \ 432}) 433 434/** 435 * wait_var_event_any_lock - wait for a variable to be updated under a lock 436 * @var: the address of the variable being waited on 437 * @condition: condition to wait for 438 * @lock: the object that is locked to protect updates to the variable 439 * @type: prefix on lock and unlock operations 440 * @state: waiting state, %TASK_UNINTERRUPTIBLE etc. 441 * 442 * Wait for a condition which can only be reliably tested while holding 443 * a lock. The variables assessed in the condition will normal be updated 444 * under the same lock, and the wake up should be signalled with 445 * wake_up_var_locked() under the same lock. 446 * 447 * This is similar to wait_var_event(), but assumes a lock is held 448 * while calling this function and while updating the variable. 449 * 450 * This must be called while the given lock is held and the lock will be 451 * dropped when schedule() is called to wait for a wake up, and will be 452 * reclaimed before testing the condition again. The functions used to 453 * unlock and lock the object are constructed by appending _unlock and _lock 454 * to @type. 455 * 456 * Return %-ERESTARTSYS if a signal arrives which is allowed to interrupt 457 * the wait according to @state. 458 */ 459#define wait_var_event_any_lock(var, condition, lock, type, state) \ 460({ \ 461 int __ret = 0; \ 462 if (!(condition)) \ 463 __ret = ___wait_var_event(var, condition, state, 0, 0, \ 464 type ## _unlock(lock); \ 465 schedule(); \ 466 type ## _lock(lock)); \ 467 __ret; \ 468}) 469 470/** 471 * wait_var_event_spinlock - wait for a variable to be updated under a spinlock 472 * @var: the address of the variable being waited on 473 * @condition: condition to wait for 474 * @lock: the spinlock which protects updates to the variable 475 * 476 * Wait for a condition which can only be reliably tested while holding 477 * a spinlock. The variables assessed in the condition will normal be updated 478 * under the same spinlock, and the wake up should be signalled with 479 * wake_up_var_locked() under the same spinlock. 480 * 481 * This is similar to wait_var_event(), but assumes a spinlock is held 482 * while calling this function and while updating the variable. 483 * 484 * This must be called while the given lock is held and the lock will be 485 * dropped when schedule() is called to wait for a wake up, and will be 486 * reclaimed before testing the condition again. 487 */ 488#define wait_var_event_spinlock(var, condition, lock) \ 489 wait_var_event_any_lock(var, condition, lock, spin, TASK_UNINTERRUPTIBLE) 490 491/** 492 * wait_var_event_mutex - wait for a variable to be updated under a mutex 493 * @var: the address of the variable being waited on 494 * @condition: condition to wait for 495 * @mutex: the mutex which protects updates to the variable 496 * 497 * Wait for a condition which can only be reliably tested while holding 498 * a mutex. The variables assessed in the condition will normal be 499 * updated under the same mutex, and the wake up should be signalled 500 * with wake_up_var_locked() under the same mutex. 501 * 502 * This is similar to wait_var_event(), but assumes a mutex is held 503 * while calling this function and while updating the variable. 504 * 505 * This must be called while the given mutex is held and the mutex will be 506 * dropped when schedule() is called to wait for a wake up, and will be 507 * reclaimed before testing the condition again. 508 */ 509#define wait_var_event_mutex(var, condition, lock) \ 510 wait_var_event_any_lock(var, condition, lock, mutex, TASK_UNINTERRUPTIBLE) 511 512/** 513 * wake_up_var_protected - wake up waiters for a variable asserting that it is safe 514 * @var: the address of the variable being waited on 515 * @cond: the condition which afirms this is safe 516 * 517 * When waking waiters which use wait_var_event_any_lock() the waker must be 518 * holding the reelvant lock to avoid races. This version of wake_up_var() 519 * asserts that the relevant lock is held and so no barrier is needed. 520 * The @cond is only tested when CONFIG_LOCKDEP is enabled. 521 */ 522#define wake_up_var_protected(var, cond) \ 523do { \ 524 lockdep_assert(cond); \ 525 wake_up_var(var); \ 526} while (0) 527 528/** 529 * wake_up_var_locked - wake up waiters for a variable while holding a spinlock or mutex 530 * @var: the address of the variable being waited on 531 * @lock: The spinlock or mutex what protects the variable 532 * 533 * Send a wake up for the given variable which should be waited for with 534 * wait_var_event_spinlock() or wait_var_event_mutex(). Unlike wake_up_var(), 535 * no extra barriers are needed as the locking provides sufficient sequencing. 536 */ 537#define wake_up_var_locked(var, lock) \ 538 wake_up_var_protected(var, lockdep_is_held(lock)) 539 540/** 541 * clear_and_wake_up_bit - clear a bit and wake up anyone waiting on that bit 542 * @bit: the bit of the word being waited on 543 * @word: the address containing the bit being waited on 544 * 545 * The designated bit is cleared and any tasks waiting in wait_on_bit() 546 * or similar will be woken. This call has RELEASE semantics so that 547 * any changes to memory made before this call are guaranteed to be visible 548 * after the corresponding wait_on_bit() completes. 549 */ 550static inline void clear_and_wake_up_bit(int bit, unsigned long *word) 551{ 552 clear_bit_unlock(bit, word); 553 /* See wake_up_bit() for which memory barrier you need to use. */ 554 smp_mb__after_atomic(); 555 wake_up_bit(word, bit); 556} 557 558/** 559 * test_and_clear_wake_up_bit - clear a bit if it was set: wake up anyone waiting on that bit 560 * @bit: the bit of the word being waited on 561 * @word: the address of memory containing that bit 562 * 563 * If the bit is set and can be atomically cleared, any tasks waiting in 564 * wait_on_bit() or similar will be woken. This call has the same 565 * complete ordering semantics as test_and_clear_bit(). Any changes to 566 * memory made before this call are guaranteed to be visible after the 567 * corresponding wait_on_bit() completes. 568 * 569 * Returns %true if the bit was successfully set and the wake up was sent. 570 */ 571static inline bool test_and_clear_wake_up_bit(int bit, unsigned long *word) 572{ 573 if (!test_and_clear_bit(bit, word)) 574 return false; 575 /* no extra barrier required */ 576 wake_up_bit(word, bit); 577 return true; 578} 579 580/** 581 * atomic_dec_and_wake_up - decrement an atomic_t and if zero, wake up waiters 582 * @var: the variable to dec and test 583 * 584 * Decrements the atomic variable and if it reaches zero, send a wake_up to any 585 * processes waiting on the variable. 586 * 587 * This function has the same complete ordering semantics as atomic_dec_and_test. 588 * 589 * Returns %true is the variable reaches zero and the wake up was sent. 590 */ 591 592static inline bool atomic_dec_and_wake_up(atomic_t *var) 593{ 594 if (!atomic_dec_and_test(var)) 595 return false; 596 /* No extra barrier required */ 597 wake_up_var(var); 598 return true; 599} 600 601/** 602 * store_release_wake_up - update a variable and send a wake_up 603 * @var: the address of the variable to be updated and woken 604 * @val: the value to store in the variable. 605 * 606 * Store the given value in the variable send a wake up to any tasks 607 * waiting on the variable. All necessary barriers are included to ensure 608 * the task calling wait_var_event() sees the new value and all values 609 * written to memory before this call. 610 */ 611#define store_release_wake_up(var, val) \ 612do { \ 613 smp_store_release(var, val); \ 614 smp_mb(); \ 615 wake_up_var(var); \ 616} while (0) 617 618#endif /* _LINUX_WAIT_BIT_H */