| 1 | #ifndef __LINUX_SEQLOCK_H
|
| 2 | #define __LINUX_SEQLOCK_H
|
| 3 | |
| 4 | * Reader/writer consistent mechanism without starving writers. This type of
|
| 5 | * lock for data where the reader wants a consistent set of information
|
| 6 | * and is willing to retry if the information changes. There are two types
|
| 7 | * of readers:
|
| 8 | * 1. Sequence readers which never block a writer but they may have to retry
|
| 9 | * if a writer is in progress by detecting change in sequence number.
|
| 10 | * Writers do not wait for a sequence reader.
|
| 11 | * 2. Locking readers which will wait if a writer or another locking reader
|
| 12 | * is in progress. A locking reader in progress will also block a writer
|
| 13 | * from going forward. Unlike the regular rwlock, the read lock here is
|
| 14 | * exclusive so that only one locking reader can get it.
|
| 15 | *
|
| 16 | * This is not as cache friendly as brlock. Also, this may not work well
|
| 17 | * for data that contains pointers, because any writer could
|
| 18 | * invalidate a pointer that a reader was following.
|
| 19 | *
|
| 20 | * Expected non-blocking reader usage:
|
| 21 | * do {
|
| 22 | * seq = read_seqbegin(&foo);
|
| 23 | * ...
|
| 24 | * } while (read_seqretry(&foo, seq));
|
| 25 | *
|
| 26 | *
|
| 27 | * On non-SMP the spin locks disappear but the writer still needs
|
| 28 | * to increment the sequence variables because an interrupt routine could
|
| 29 | * change the state of the data.
|
| 30 | *
|
| 31 | * Based on x86_64 vsyscall gettimeofday
|
| 32 | * by Keith Owens and Andrea Arcangeli
|
| 33 |
|
| 34 |
|
| 35 | #include <linux/spinlock.h>
|
| 36 | #include <linux/preempt.h>
|
| 37 | #include <linux/lockdep.h>
|
| 38 | #include <asm/processor.h>
|
| 39 |
|
| 40 | |
| 41 | * Version using sequence counter only.
|
| 42 | * This can be used when code has its own mutex protecting the
|
| 43 | * updating starting before the write_seqcountbeqin() and ending
|
| 44 | * after the write_seqcount_end().
|
| 45 |
|
| 46 | typedef struct seqcount {
|
| 47 | unsigned sequence;
|
| 48 | #ifdef CONFIG_DEBUG_LOCK_ALLOC
|
| 49 | struct lockdep_map dep_map;
|
| 50 | #endif
|
| 51 | } seqcount_t;
|
| 52 |
|
| 53 | static inline void __seqcount_init(seqcount_t *s, const char *name,
|
| 54 | struct lock_class_key *key)
|
| 55 | {
|
| 56 | |
| 57 | * Make sure we are not reinitializing a held lock:
|
| 58 |
|
| 59 | lockdep_init_map(&s->dep_map, name, key, 0);
|
| 60 | s->sequence = 0;
|
| 61 | }
|
| 62 |
|
| 63 | #ifdef CONFIG_DEBUG_LOCK_ALLOC
|
| 64 | # define SEQCOUNT_DEP_MAP_INIT(lockname) \
|
| 65 | .dep_map = { .name = #lockname } \
|
| 66 |
|
| 67 | # define seqcount_init(s) \
|
| 68 | do { \
|
| 69 | static struct lock_class_key __key; \
|
| 70 | __seqcount_init((s), #s, &__key); \
|
| 71 | } while (0)
|
| 72 |
|
| 73 | static inline void seqcount_lockdep_reader_access(const seqcount_t *s)
|
| 74 | {
|
| 75 | seqcount_t *l = (seqcount_t *)s;
|
| 76 | unsigned long flags;
|
| 77 |
|
| 78 | local_irq_save(flags);
|
| 79 | seqcount_acquire_read(&l->dep_map, 0, 0, _RET_IP_);
|
| 80 | seqcount_release(&l->dep_map, 1, _RET_IP_);
|
| 81 | local_irq_restore(flags);
|
| 82 | }
|
| 83 |
|
| 84 | #else
|
| 85 | # define SEQCOUNT_DEP_MAP_INIT(lockname)
|
| 86 | # define seqcount_init(s) __seqcount_init(s, NULL, NULL)
|
| 87 | # define seqcount_lockdep_reader_access(x)
|
| 88 | #endif
|
| 89 |
|
| 90 | #define SEQCNT_ZERO(lockname) { .sequence = 0, SEQCOUNT_DEP_MAP_INIT(lockname)}
|
| 91 |
|
| 92 |
|
| 93 | |
| 94 | * __read_seqcount_begin - begin a seq-read critical section (without barrier)
|
| 95 | * @s: pointer to seqcount_t
|
| 96 | * Returns: count to be passed to read_seqcount_retry
|
| 97 | *
|
| 98 | * __read_seqcount_begin is like read_seqcount_begin, but has no smp_rmb()
|
| 99 | * barrier. Callers should ensure that smp_rmb() or equivalent ordering is
|
| 100 | * provided before actually loading any of the variables that are to be
|
| 101 | * protected in this critical section.
|
| 102 | *
|
| 103 | * Use carefully, only in critical code, and comment how the barrier is
|
| 104 | * provided.
|
| 105 |
|
| 106 | static inline unsigned __read_seqcount_begin(const seqcount_t *s)
|
| 107 | {
|
| 108 | unsigned ret;
|
| 109 |
|
| 110 | repeat:
|
| 111 | ret = READ_ONCE(s->sequence);
|
| 112 | if (unlikely(ret & 1)) {
|
| 113 | cpu_relax();
|
| 114 | goto repeat;
|
| 115 | }
|
| 116 | return ret;
|
| 117 | }
|
| 118 |
|
| 119 | |
| 120 | * raw_read_seqcount - Read the raw seqcount
|
| 121 | * @s: pointer to seqcount_t
|
| 122 | * Returns: count to be passed to read_seqcount_retry
|
| 123 | *
|
| 124 | * raw_read_seqcount opens a read critical section of the given
|
| 125 | * seqcount without any lockdep checking and without checking or
|
| 126 | * masking the LSB. Calling code is responsible for handling that.
|
| 127 |
|
| 128 | static inline unsigned raw_read_seqcount(const seqcount_t *s)
|
| 129 | {
|
| 130 | unsigned ret = READ_ONCE(s->sequence);
|
| 131 | smp_rmb();
|
| 132 | return ret;
|
| 133 | }
|
| 134 |
|
| 135 | |
| 136 | * raw_read_seqcount_begin - start seq-read critical section w/o lockdep
|
| 137 | * @s: pointer to seqcount_t
|
| 138 | * Returns: count to be passed to read_seqcount_retry
|
| 139 | *
|
| 140 | * raw_read_seqcount_begin opens a read critical section of the given
|
| 141 | * seqcount, but without any lockdep checking. Validity of the critical
|
| 142 | * section is tested by checking read_seqcount_retry function.
|
| 143 |
|
| 144 | static inline unsigned raw_read_seqcount_begin(const seqcount_t *s)
|
| 145 | {
|
| 146 | unsigned ret = __read_seqcount_begin(s);
|
| 147 | smp_rmb();
|
| 148 | return ret;
|
| 149 | }
|
| 150 |
|
| 151 | |
| 152 | * read_seqcount_begin - begin a seq-read critical section
|
| 153 | * @s: pointer to seqcount_t
|
| 154 | * Returns: count to be passed to read_seqcount_retry
|
| 155 | *
|
| 156 | * read_seqcount_begin opens a read critical section of the given seqcount.
|
| 157 | * Validity of the critical section is tested by checking read_seqcount_retry
|
| 158 | * function.
|
| 159 |
|
| 160 | static inline unsigned read_seqcount_begin(const seqcount_t *s)
|
| 161 | {
|
| 162 | seqcount_lockdep_reader_access(s);
|
| 163 | return raw_read_seqcount_begin(s);
|
| 164 | }
|
| 165 |
|
| 166 | |
| 167 | * raw_seqcount_begin - begin a seq-read critical section
|
| 168 | * @s: pointer to seqcount_t
|
| 169 | * Returns: count to be passed to read_seqcount_retry
|
| 170 | *
|
| 171 | * raw_seqcount_begin opens a read critical section of the given seqcount.
|
| 172 | * Validity of the critical section is tested by checking read_seqcount_retry
|
| 173 | * function.
|
| 174 | *
|
| 175 | * Unlike read_seqcount_begin(), this function will not wait for the count
|
| 176 | * to stabilize. If a writer is active when we begin, we will fail the
|
| 177 | * read_seqcount_retry() instead of stabilizing at the beginning of the
|
| 178 | * critical section.
|
| 179 |
|
| 180 | static inline unsigned raw_seqcount_begin(const seqcount_t *s)
|
| 181 | {
|
| 182 | unsigned ret = READ_ONCE(s->sequence);
|
| 183 | smp_rmb();
|
| 184 | return ret & ~1;
|
| 185 | }
|
| 186 |
|
| 187 | |
| 188 | * __read_seqcount_retry - end a seq-read critical section (without barrier)
|
| 189 | * @s: pointer to seqcount_t
|
| 190 | * @start: count, from read_seqcount_begin
|
| 191 | * Returns: 1 if retry is required, else 0
|
| 192 | *
|
| 193 | * __read_seqcount_retry is like read_seqcount_retry, but has no smp_rmb()
|
| 194 | * barrier. Callers should ensure that smp_rmb() or equivalent ordering is
|
| 195 | * provided before actually loading any of the variables that are to be
|
| 196 | * protected in this critical section.
|
| 197 | *
|
| 198 | * Use carefully, only in critical code, and comment how the barrier is
|
| 199 | * provided.
|
| 200 |
|
| 201 | static inline int __read_seqcount_retry(const seqcount_t *s, unsigned start)
|
| 202 | {
|
| 203 | return unlikely(s->sequence != start);
|
| 204 | }
|
| 205 |
|
| 206 | |
| 207 | * read_seqcount_retry - end a seq-read critical section
|
| 208 | * @s: pointer to seqcount_t
|
| 209 | * @start: count, from read_seqcount_begin
|
| 210 | * Returns: 1 if retry is required, else 0
|
| 211 | *
|
| 212 | * read_seqcount_retry closes a read critical section of the given seqcount.
|
| 213 | * If the critical section was invalid, it must be ignored (and typically
|
| 214 | * retried).
|
| 215 |
|
| 216 | static inline int read_seqcount_retry(const seqcount_t *s, unsigned start)
|
| 217 | {
|
| 218 | smp_rmb();
|
| 219 | return __read_seqcount_retry(s, start);
|
| 220 | }
|
| 221 |
|
| 222 |
|
| 223 |
|
| 224 | static inline void raw_write_seqcount_begin(seqcount_t *s)
|
| 225 | {
|
| 226 | s->sequence++;
|
| 227 | smp_wmb();
|
| 228 | }
|
| 229 |
|
| 230 | static inline void raw_write_seqcount_end(seqcount_t *s)
|
| 231 | {
|
| 232 | smp_wmb();
|
| 233 | s->sequence++;
|
| 234 | }
|
| 235 |
|
| 236 | |
| 237 | * raw_write_seqcount_latch - redirect readers to even/odd copy
|
| 238 | * @s: pointer to seqcount_t
|
| 239 |
|
| 240 | static inline void raw_write_seqcount_latch(seqcount_t *s)
|
| 241 | {
|
| 242 | smp_wmb();
|
| 243 | s->sequence++;
|
| 244 | smp_wmb();
|
| 245 | }
|
| 246 |
|
| 247 | |
| 248 | * Sequence counter only version assumes that callers are using their
|
| 249 | * own mutexing.
|
| 250 |
|
| 251 | static inline void write_seqcount_begin_nested(seqcount_t *s, int subclass)
|
| 252 | {
|
| 253 | raw_write_seqcount_begin(s);
|
| 254 | seqcount_acquire(&s->dep_map, subclass, 0, _RET_IP_);
|
| 255 | }
|
| 256 |
|
| 257 | static inline void write_seqcount_begin(seqcount_t *s)
|
| 258 | {
|
| 259 | write_seqcount_begin_nested(s, 0);
|
| 260 | }
|
| 261 |
|
| 262 | static inline void write_seqcount_end(seqcount_t *s)
|
| 263 | {
|
| 264 | seqcount_release(&s->dep_map, 1, _RET_IP_);
|
| 265 | raw_write_seqcount_end(s);
|
| 266 | }
|
| 267 |
|
| 268 | |
| 269 | * write_seqcount_barrier - invalidate in-progress read-side seq operations
|
| 270 | * @s: pointer to seqcount_t
|
| 271 | *
|
| 272 | * After write_seqcount_barrier, no read-side seq operations will complete
|
| 273 | * successfully and see data older than this.
|
| 274 |
|
| 275 | static inline void write_seqcount_barrier(seqcount_t *s)
|
| 276 | {
|
| 277 | smp_wmb();
|
| 278 | s->sequence+=2;
|
| 279 | }
|
| 280 |
|
| 281 | typedef struct {
|
| 282 | struct seqcount seqcount;
|
| 283 | spinlock_t lock;
|
| 284 | } seqlock_t;
|
| 285 |
|
| 286 | |
| 287 | * These macros triggered gcc-3.x compile-time problems. We think these are
|
| 288 | * OK now. Be cautious.
|
| 289 |
|
| 290 | #define __SEQLOCK_UNLOCKED(lockname) \
|
| 291 | { \
|
| 292 | .seqcount = SEQCNT_ZERO(lockname), \
|
| 293 | .lock = __SPIN_LOCK_UNLOCKED(lockname) \
|
| 294 | }
|
| 295 |
|
| 296 | #define seqlock_init(x) \
|
| 297 | do { \
|
| 298 | seqcount_init(&(x)->seqcount); \
|
| 299 | spin_lock_init(&(x)->lock); \
|
| 300 | } while (0)
|
| 301 |
|
| 302 | #define DEFINE_SEQLOCK(x) \
|
| 303 | seqlock_t x = __SEQLOCK_UNLOCKED(x)
|
| 304 |
|
| 305 | |
| 306 | * Read side functions for starting and finalizing a read side section.
|
| 307 |
|
| 308 | static inline unsigned read_seqbegin(const seqlock_t *sl)
|
| 309 | {
|
| 310 | return read_seqcount_begin(&sl->seqcount);
|
| 311 | }
|
| 312 |
|
| 313 | static inline unsigned read_seqretry(const seqlock_t *sl, unsigned start)
|
| 314 | {
|
| 315 | return read_seqcount_retry(&sl->seqcount, start);
|
| 316 | }
|
| 317 |
|
| 318 | |
| 319 | * Lock out other writers and update the count.
|
| 320 | * Acts like a normal spin_lock/unlock.
|
| 321 | * Don't need preempt_disable() because that is in the spin_lock already.
|
| 322 |
|
| 323 | static inline void write_seqlock(seqlock_t *sl)
|
| 324 | {
|
| 325 | spin_lock(&sl->lock);
|
| 326 | write_seqcount_begin(&sl->seqcount);
|
| 327 | }
|
| 328 |
|
| 329 | static inline void write_sequnlock(seqlock_t *sl)
|
| 330 | {
|
| 331 | write_seqcount_end(&sl->seqcount);
|
| 332 | spin_unlock(&sl->lock);
|
| 333 | }
|
| 334 |
|
| 335 | static inline void write_seqlock_bh(seqlock_t *sl)
|
| 336 | {
|
| 337 | spin_lock_bh(&sl->lock);
|
| 338 | write_seqcount_begin(&sl->seqcount);
|
| 339 | }
|
| 340 |
|
| 341 | static inline void write_sequnlock_bh(seqlock_t *sl)
|
| 342 | {
|
| 343 | write_seqcount_end(&sl->seqcount);
|
| 344 | spin_unlock_bh(&sl->lock);
|
| 345 | }
|
| 346 |
|
| 347 | static inline void write_seqlock_irq(seqlock_t *sl)
|
| 348 | {
|
| 349 | spin_lock_irq(&sl->lock);
|
| 350 | write_seqcount_begin(&sl->seqcount);
|
| 351 | }
|
| 352 |
|
| 353 | static inline void write_sequnlock_irq(seqlock_t *sl)
|
| 354 | {
|
| 355 | write_seqcount_end(&sl->seqcount);
|
| 356 | spin_unlock_irq(&sl->lock);
|
| 357 | }
|
| 358 |
|
| 359 | static inline unsigned long __write_seqlock_irqsave(seqlock_t *sl)
|
| 360 | {
|
| 361 | unsigned long flags;
|
| 362 |
|
| 363 | spin_lock_irqsave(&sl->lock, flags);
|
| 364 | write_seqcount_begin(&sl->seqcount);
|
| 365 | return flags;
|
| 366 | }
|
| 367 |
|
| 368 | #define write_seqlock_irqsave(lock, flags) \
|
| 369 | do { flags = __write_seqlock_irqsave(lock); } while (0)
|
| 370 |
|
| 371 | static inline void
|
| 372 | write_sequnlock_irqrestore(seqlock_t *sl, unsigned long flags)
|
| 373 | {
|
| 374 | write_seqcount_end(&sl->seqcount);
|
| 375 | spin_unlock_irqrestore(&sl->lock, flags);
|
| 376 | }
|
| 377 |
|
| 378 | |
| 379 | * A locking reader exclusively locks out other writers and locking readers,
|
| 380 | * but doesn't update the sequence number. Acts like a normal spin_lock/unlock.
|
| 381 | * Don't need preempt_disable() because that is in the spin_lock already.
|
| 382 |
|
| 383 | static inline void read_seqlock_excl(seqlock_t *sl)
|
| 384 | {
|
| 385 | spin_lock(&sl->lock);
|
| 386 | }
|
| 387 |
|
| 388 | static inline void read_sequnlock_excl(seqlock_t *sl)
|
| 389 | {
|
| 390 | spin_unlock(&sl->lock);
|
| 391 | }
|
| 392 |
|
| 393 | |
| 394 | * read_seqbegin_or_lock - begin a sequence number check or locking block
|
| 395 | * @lock: sequence lock
|
| 396 | * @seq : sequence number to be checked
|
| 397 | *
|
| 398 | * First try it once optimistically without taking the lock. If that fails,
|
| 399 | * take the lock. The sequence number is also used as a marker for deciding
|
| 400 | * whether to be a reader (even) or writer (odd).
|
| 401 | * N.B. seq must be initialized to an even number to begin with.
|
| 402 |
|
| 403 | static inline void read_seqbegin_or_lock(seqlock_t *lock, int *seq)
|
| 404 | {
|
| 405 | if (!(*seq & 1))
|
| 406 | *seq = read_seqbegin(lock);
|
| 407 | else
|
| 408 | read_seqlock_excl(lock);
|
| 409 | }
|
| 410 |
|
| 411 | static inline int need_seqretry(seqlock_t *lock, int seq)
|
| 412 | {
|
| 413 | return !(seq & 1) && read_seqretry(lock, seq);
|
| 414 | }
|
| 415 |
|
| 416 | static inline void done_seqretry(seqlock_t *lock, int seq)
|
| 417 | {
|
| 418 | if (seq & 1)
|
| 419 | read_sequnlock_excl(lock);
|
| 420 | }
|
| 421 |
|
| 422 | static inline void read_seqlock_excl_bh(seqlock_t *sl)
|
| 423 | {
|
| 424 | spin_lock_bh(&sl->lock);
|
| 425 | }
|
| 426 |
|
| 427 | static inline void read_sequnlock_excl_bh(seqlock_t *sl)
|
| 428 | {
|
| 429 | spin_unlock_bh(&sl->lock);
|
| 430 | }
|
| 431 |
|
| 432 | static inline void read_seqlock_excl_irq(seqlock_t *sl)
|
| 433 | {
|
| 434 | spin_lock_irq(&sl->lock);
|
| 435 | }
|
| 436 |
|
| 437 | static inline void read_sequnlock_excl_irq(seqlock_t *sl)
|
| 438 | {
|
| 439 | spin_unlock_irq(&sl->lock);
|
| 440 | }
|
| 441 |
|
| 442 | static inline unsigned long __read_seqlock_excl_irqsave(seqlock_t *sl)
|
| 443 | {
|
| 444 | unsigned long flags;
|
| 445 |
|
| 446 | spin_lock_irqsave(&sl->lock, flags);
|
| 447 | return flags;
|
| 448 | }
|
| 449 |
|
| 450 | #define read_seqlock_excl_irqsave(lock, flags) \
|
| 451 | do { flags = __read_seqlock_excl_irqsave(lock); } while (0)
|
| 452 |
|
| 453 | static inline void
|
| 454 | read_sequnlock_excl_irqrestore(seqlock_t *sl, unsigned long flags)
|
| 455 | {
|
| 456 | spin_unlock_irqrestore(&sl->lock, flags);
|
| 457 | }
|
| 458 |
|
| 459 | static inline unsigned long
|
| 460 | read_seqbegin_or_lock_irqsave(seqlock_t *lock, int *seq)
|
| 461 | {
|
| 462 | unsigned long flags = 0;
|
| 463 |
|
| 464 | if (!(*seq & 1))
|
| 465 | *seq = read_seqbegin(lock);
|
| 466 | else
|
| 467 | read_seqlock_excl_irqsave(lock, flags);
|
| 468 |
|
| 469 | return flags;
|
| 470 | }
|
| 471 |
|
| 472 | static inline void
|
| 473 | done_seqretry_irqrestore(seqlock_t *lock, int seq, unsigned long flags)
|
| 474 | {
|
| 475 | if (seq & 1)
|
| 476 | read_sequnlock_excl_irqrestore(lock, flags);
|
| 477 | }
|
| 478 | #endif
|