1 /* 2 * Fast and scalable bitmaps. 3 * 4 * Copyright (C) 2016 Facebook 5 * Copyright (C) 2013-2014 Jens Axboe 6 * 7 * This program is free software; you can redistribute it and/or 8 * modify it under the terms of the GNU General Public 9 * License v2 as published by the Free Software Foundation. 10 * 11 * This program is distributed in the hope that it will be useful, 12 * but WITHOUT ANY WARRANTY; without even the implied warranty of 13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU 14 * General Public License for more details. 15 * 16 * You should have received a copy of the GNU General Public License 17 * along with this program. If not, see <https://www.gnu.org/licenses/>. 18 */ 19 20 #ifndef __LINUX_SCALE_BITMAP_H 21 #define __LINUX_SCALE_BITMAP_H 22 23 #include <linux/kernel.h> 24 #include <linux/slab.h> 25 26 /** 27 * struct sbitmap_word - Word in a &struct sbitmap. 28 */ 29 struct sbitmap_word { 30 /** 31 * @word: The bitmap word itself. 32 */ 33 unsigned long word; 34 35 /** 36 * @depth: Number of bits being used in @word. 37 */ 38 unsigned long depth; 39 } ____cacheline_aligned_in_smp; 40 41 /** 42 * struct sbitmap - Scalable bitmap. 43 * 44 * A &struct sbitmap is spread over multiple cachelines to avoid ping-pong. This 45 * trades off higher memory usage for better scalability. 46 */ 47 struct sbitmap { 48 /** 49 * @depth: Number of bits used in the whole bitmap. 50 */ 51 unsigned int depth; 52 53 /** 54 * @shift: log2(number of bits used per word) 55 */ 56 unsigned int shift; 57 58 /** 59 * @map_nr: Number of words (cachelines) being used for the bitmap. 60 */ 61 unsigned int map_nr; 62 63 /** 64 * @map: Allocated bitmap. 65 */ 66 struct sbitmap_word *map; 67 }; 68 69 #define SBQ_WAIT_QUEUES 8 70 #define SBQ_WAKE_BATCH 8 71 72 /** 73 * struct sbq_wait_state - Wait queue in a &struct sbitmap_queue. 74 */ 75 struct sbq_wait_state { 76 /** 77 * @wait_cnt: Number of frees remaining before we wake up. 78 */ 79 atomic_t wait_cnt; 80 81 /** 82 * @wait: Wait queue. 83 */ 84 wait_queue_head_t wait; 85 } ____cacheline_aligned_in_smp; 86 87 /** 88 * struct sbitmap_queue - Scalable bitmap with the added ability to wait on free 89 * bits. 90 * 91 * A &struct sbitmap_queue uses multiple wait queues and rolling wakeups to 92 * avoid contention on the wait queue spinlock. This ensures that we don't hit a 93 * scalability wall when we run out of free bits and have to start putting tasks 94 * to sleep. 95 */ 96 struct sbitmap_queue { 97 /** 98 * @sb: Scalable bitmap. 99 */ 100 struct sbitmap sb; 101 102 /* 103 * @alloc_hint: Cache of last successfully allocated or freed bit. 104 * 105 * This is per-cpu, which allows multiple users to stick to different 106 * cachelines until the map is exhausted. 107 */ 108 unsigned int __percpu *alloc_hint; 109 110 /** 111 * @wake_batch: Number of bits which must be freed before we wake up any 112 * waiters. 113 */ 114 unsigned int wake_batch; 115 116 /** 117 * @wake_index: Next wait queue in @ws to wake up. 118 */ 119 atomic_t wake_index; 120 121 /** 122 * @ws: Wait queues. 123 */ 124 struct sbq_wait_state *ws; 125 126 /** 127 * @round_robin: Allocate bits in strict round-robin order. 128 */ 129 bool round_robin; 130 131 /** 132 * @min_shallow_depth: The minimum shallow depth which may be passed to 133 * sbitmap_queue_get_shallow() or __sbitmap_queue_get_shallow(). 134 */ 135 unsigned int min_shallow_depth; 136 }; 137 138 /** 139 * sbitmap_init_node() - Initialize a &struct sbitmap on a specific memory node. 140 * @sb: Bitmap to initialize. 141 * @depth: Number of bits to allocate. 142 * @shift: Use 2^@shift bits per word in the bitmap; if a negative number if 143 * given, a good default is chosen. 144 * @flags: Allocation flags. 145 * @node: Memory node to allocate on. 146 * 147 * Return: Zero on success or negative errno on failure. 148 */ 149 int sbitmap_init_node(struct sbitmap *sb, unsigned int depth, int shift, 150 gfp_t flags, int node); 151 152 /** 153 * sbitmap_free() - Free memory used by a &struct sbitmap. 154 * @sb: Bitmap to free. 155 */ 156 static inline void sbitmap_free(struct sbitmap *sb) 157 { 158 kfree(sb->map); 159 sb->map = NULL; 160 } 161 162 /** 163 * sbitmap_resize() - Resize a &struct sbitmap. 164 * @sb: Bitmap to resize. 165 * @depth: New number of bits to resize to. 166 * 167 * Doesn't reallocate anything. It's up to the caller to ensure that the new 168 * depth doesn't exceed the depth that the sb was initialized with. 169 */ 170 void sbitmap_resize(struct sbitmap *sb, unsigned int depth); 171 172 /** 173 * sbitmap_get() - Try to allocate a free bit from a &struct sbitmap. 174 * @sb: Bitmap to allocate from. 175 * @alloc_hint: Hint for where to start searching for a free bit. 176 * @round_robin: If true, be stricter about allocation order; always allocate 177 * starting from the last allocated bit. This is less efficient 178 * than the default behavior (false). 179 * 180 * This operation provides acquire barrier semantics if it succeeds. 181 * 182 * Return: Non-negative allocated bit number if successful, -1 otherwise. 183 */ 184 int sbitmap_get(struct sbitmap *sb, unsigned int alloc_hint, bool round_robin); 185 186 /** 187 * sbitmap_get_shallow() - Try to allocate a free bit from a &struct sbitmap, 188 * limiting the depth used from each word. 189 * @sb: Bitmap to allocate from. 190 * @alloc_hint: Hint for where to start searching for a free bit. 191 * @shallow_depth: The maximum number of bits to allocate from a single word. 192 * 193 * This rather specific operation allows for having multiple users with 194 * different allocation limits. E.g., there can be a high-priority class that 195 * uses sbitmap_get() and a low-priority class that uses sbitmap_get_shallow() 196 * with a @shallow_depth of (1 << (@sb->shift - 1)). Then, the low-priority 197 * class can only allocate half of the total bits in the bitmap, preventing it 198 * from starving out the high-priority class. 199 * 200 * Return: Non-negative allocated bit number if successful, -1 otherwise. 201 */ 202 int sbitmap_get_shallow(struct sbitmap *sb, unsigned int alloc_hint, 203 unsigned long shallow_depth); 204 205 /** 206 * sbitmap_any_bit_set() - Check for a set bit in a &struct sbitmap. 207 * @sb: Bitmap to check. 208 * 209 * Return: true if any bit in the bitmap is set, false otherwise. 210 */ 211 bool sbitmap_any_bit_set(const struct sbitmap *sb); 212 213 /** 214 * sbitmap_any_bit_clear() - Check for an unset bit in a &struct 215 * sbitmap. 216 * @sb: Bitmap to check. 217 * 218 * Return: true if any bit in the bitmap is clear, false otherwise. 219 */ 220 bool sbitmap_any_bit_clear(const struct sbitmap *sb); 221 222 #define SB_NR_TO_INDEX(sb, bitnr) ((bitnr) >> (sb)->shift) 223 #define SB_NR_TO_BIT(sb, bitnr) ((bitnr) & ((1U << (sb)->shift) - 1U)) 224 225 typedef bool (*sb_for_each_fn)(struct sbitmap *, unsigned int, void *); 226 227 /** 228 * __sbitmap_for_each_set() - Iterate over each set bit in a &struct sbitmap. 229 * @start: Where to start the iteration. 230 * @sb: Bitmap to iterate over. 231 * @fn: Callback. Should return true to continue or false to break early. 232 * @data: Pointer to pass to callback. 233 * 234 * This is inline even though it's non-trivial so that the function calls to the 235 * callback will hopefully get optimized away. 236 */ 237 static inline void __sbitmap_for_each_set(struct sbitmap *sb, 238 unsigned int start, 239 sb_for_each_fn fn, void *data) 240 { 241 unsigned int index; 242 unsigned int nr; 243 unsigned int scanned = 0; 244 245 if (start >= sb->depth) 246 start = 0; 247 index = SB_NR_TO_INDEX(sb, start); 248 nr = SB_NR_TO_BIT(sb, start); 249 250 while (scanned < sb->depth) { 251 struct sbitmap_word *word = &sb->map[index]; 252 unsigned int depth = min_t(unsigned int, word->depth - nr, 253 sb->depth - scanned); 254 255 scanned += depth; 256 if (!word->word) 257 goto next; 258 259 /* 260 * On the first iteration of the outer loop, we need to add the 261 * bit offset back to the size of the word for find_next_bit(). 262 * On all other iterations, nr is zero, so this is a noop. 263 */ 264 depth += nr; 265 while (1) { 266 nr = find_next_bit(&word->word, depth, nr); 267 if (nr >= depth) 268 break; 269 if (!fn(sb, (index << sb->shift) + nr, data)) 270 return; 271 272 nr++; 273 } 274 next: 275 nr = 0; 276 if (++index >= sb->map_nr) 277 index = 0; 278 } 279 } 280 281 /** 282 * sbitmap_for_each_set() - Iterate over each set bit in a &struct sbitmap. 283 * @sb: Bitmap to iterate over. 284 * @fn: Callback. Should return true to continue or false to break early. 285 * @data: Pointer to pass to callback. 286 */ 287 static inline void sbitmap_for_each_set(struct sbitmap *sb, sb_for_each_fn fn, 288 void *data) 289 { 290 __sbitmap_for_each_set(sb, 0, fn, data); 291 } 292 293 static inline unsigned long *__sbitmap_word(struct sbitmap *sb, 294 unsigned int bitnr) 295 { 296 return &sb->map[SB_NR_TO_INDEX(sb, bitnr)].word; 297 } 298 299 /* Helpers equivalent to the operations in asm/bitops.h and linux/bitmap.h */ 300 301 static inline void sbitmap_set_bit(struct sbitmap *sb, unsigned int bitnr) 302 { 303 set_bit(SB_NR_TO_BIT(sb, bitnr), __sbitmap_word(sb, bitnr)); 304 } 305 306 static inline void sbitmap_clear_bit(struct sbitmap *sb, unsigned int bitnr) 307 { 308 clear_bit(SB_NR_TO_BIT(sb, bitnr), __sbitmap_word(sb, bitnr)); 309 } 310 311 static inline void sbitmap_clear_bit_unlock(struct sbitmap *sb, 312 unsigned int bitnr) 313 { 314 clear_bit_unlock(SB_NR_TO_BIT(sb, bitnr), __sbitmap_word(sb, bitnr)); 315 } 316 317 static inline int sbitmap_test_bit(struct sbitmap *sb, unsigned int bitnr) 318 { 319 return test_bit(SB_NR_TO_BIT(sb, bitnr), __sbitmap_word(sb, bitnr)); 320 } 321 322 unsigned int sbitmap_weight(const struct sbitmap *sb); 323 324 /** 325 * sbitmap_show() - Dump &struct sbitmap information to a &struct seq_file. 326 * @sb: Bitmap to show. 327 * @m: struct seq_file to write to. 328 * 329 * This is intended for debugging. The format may change at any time. 330 */ 331 void sbitmap_show(struct sbitmap *sb, struct seq_file *m); 332 333 /** 334 * sbitmap_bitmap_show() - Write a hex dump of a &struct sbitmap to a &struct 335 * seq_file. 336 * @sb: Bitmap to show. 337 * @m: struct seq_file to write to. 338 * 339 * This is intended for debugging. The output isn't guaranteed to be internally 340 * consistent. 341 */ 342 void sbitmap_bitmap_show(struct sbitmap *sb, struct seq_file *m); 343 344 /** 345 * sbitmap_queue_init_node() - Initialize a &struct sbitmap_queue on a specific 346 * memory node. 347 * @sbq: Bitmap queue to initialize. 348 * @depth: See sbitmap_init_node(). 349 * @shift: See sbitmap_init_node(). 350 * @round_robin: See sbitmap_get(). 351 * @flags: Allocation flags. 352 * @node: Memory node to allocate on. 353 * 354 * Return: Zero on success or negative errno on failure. 355 */ 356 int sbitmap_queue_init_node(struct sbitmap_queue *sbq, unsigned int depth, 357 int shift, bool round_robin, gfp_t flags, int node); 358 359 /** 360 * sbitmap_queue_free() - Free memory used by a &struct sbitmap_queue. 361 * 362 * @sbq: Bitmap queue to free. 363 */ 364 static inline void sbitmap_queue_free(struct sbitmap_queue *sbq) 365 { 366 kfree(sbq->ws); 367 free_percpu(sbq->alloc_hint); 368 sbitmap_free(&sbq->sb); 369 } 370 371 /** 372 * sbitmap_queue_resize() - Resize a &struct sbitmap_queue. 373 * @sbq: Bitmap queue to resize. 374 * @depth: New number of bits to resize to. 375 * 376 * Like sbitmap_resize(), this doesn't reallocate anything. It has to do 377 * some extra work on the &struct sbitmap_queue, so it's not safe to just 378 * resize the underlying &struct sbitmap. 379 */ 380 void sbitmap_queue_resize(struct sbitmap_queue *sbq, unsigned int depth); 381 382 /** 383 * __sbitmap_queue_get() - Try to allocate a free bit from a &struct 384 * sbitmap_queue with preemption already disabled. 385 * @sbq: Bitmap queue to allocate from. 386 * 387 * Return: Non-negative allocated bit number if successful, -1 otherwise. 388 */ 389 int __sbitmap_queue_get(struct sbitmap_queue *sbq); 390 391 /** 392 * __sbitmap_queue_get_shallow() - Try to allocate a free bit from a &struct 393 * sbitmap_queue, limiting the depth used from each word, with preemption 394 * already disabled. 395 * @sbq: Bitmap queue to allocate from. 396 * @shallow_depth: The maximum number of bits to allocate from a single word. 397 * See sbitmap_get_shallow(). 398 * 399 * If you call this, make sure to call sbitmap_queue_min_shallow_depth() after 400 * initializing @sbq. 401 * 402 * Return: Non-negative allocated bit number if successful, -1 otherwise. 403 */ 404 int __sbitmap_queue_get_shallow(struct sbitmap_queue *sbq, 405 unsigned int shallow_depth); 406 407 /** 408 * sbitmap_queue_get() - Try to allocate a free bit from a &struct 409 * sbitmap_queue. 410 * @sbq: Bitmap queue to allocate from. 411 * @cpu: Output parameter; will contain the CPU we ran on (e.g., to be passed to 412 * sbitmap_queue_clear()). 413 * 414 * Return: Non-negative allocated bit number if successful, -1 otherwise. 415 */ 416 static inline int sbitmap_queue_get(struct sbitmap_queue *sbq, 417 unsigned int *cpu) 418 { 419 int nr; 420 421 *cpu = get_cpu(); 422 nr = __sbitmap_queue_get(sbq); 423 put_cpu(); 424 return nr; 425 } 426 427 /** 428 * sbitmap_queue_get_shallow() - Try to allocate a free bit from a &struct 429 * sbitmap_queue, limiting the depth used from each word. 430 * @sbq: Bitmap queue to allocate from. 431 * @cpu: Output parameter; will contain the CPU we ran on (e.g., to be passed to 432 * sbitmap_queue_clear()). 433 * @shallow_depth: The maximum number of bits to allocate from a single word. 434 * See sbitmap_get_shallow(). 435 * 436 * If you call this, make sure to call sbitmap_queue_min_shallow_depth() after 437 * initializing @sbq. 438 * 439 * Return: Non-negative allocated bit number if successful, -1 otherwise. 440 */ 441 static inline int sbitmap_queue_get_shallow(struct sbitmap_queue *sbq, 442 unsigned int *cpu, 443 unsigned int shallow_depth) 444 { 445 int nr; 446 447 *cpu = get_cpu(); 448 nr = __sbitmap_queue_get_shallow(sbq, shallow_depth); 449 put_cpu(); 450 return nr; 451 } 452 453 /** 454 * sbitmap_queue_min_shallow_depth() - Inform a &struct sbitmap_queue of the 455 * minimum shallow depth that will be used. 456 * @sbq: Bitmap queue in question. 457 * @min_shallow_depth: The minimum shallow depth that will be passed to 458 * sbitmap_queue_get_shallow() or __sbitmap_queue_get_shallow(). 459 * 460 * sbitmap_queue_clear() batches wakeups as an optimization. The batch size 461 * depends on the depth of the bitmap. Since the shallow allocation functions 462 * effectively operate with a different depth, the shallow depth must be taken 463 * into account when calculating the batch size. This function must be called 464 * with the minimum shallow depth that will be used. Failure to do so can result 465 * in missed wakeups. 466 */ 467 void sbitmap_queue_min_shallow_depth(struct sbitmap_queue *sbq, 468 unsigned int min_shallow_depth); 469 470 /** 471 * sbitmap_queue_clear() - Free an allocated bit and wake up waiters on a 472 * &struct sbitmap_queue. 473 * @sbq: Bitmap to free from. 474 * @nr: Bit number to free. 475 * @cpu: CPU the bit was allocated on. 476 */ 477 void sbitmap_queue_clear(struct sbitmap_queue *sbq, unsigned int nr, 478 unsigned int cpu); 479 480 static inline int sbq_index_inc(int index) 481 { 482 return (index + 1) & (SBQ_WAIT_QUEUES - 1); 483 } 484 485 static inline void sbq_index_atomic_inc(atomic_t *index) 486 { 487 int old = atomic_read(index); 488 int new = sbq_index_inc(old); 489 atomic_cmpxchg(index, old, new); 490 } 491 492 /** 493 * sbq_wait_ptr() - Get the next wait queue to use for a &struct 494 * sbitmap_queue. 495 * @sbq: Bitmap queue to wait on. 496 * @wait_index: A counter per "user" of @sbq. 497 */ 498 static inline struct sbq_wait_state *sbq_wait_ptr(struct sbitmap_queue *sbq, 499 atomic_t *wait_index) 500 { 501 struct sbq_wait_state *ws; 502 503 ws = &sbq->ws[atomic_read(wait_index)]; 504 sbq_index_atomic_inc(wait_index); 505 return ws; 506 } 507 508 /** 509 * sbitmap_queue_wake_all() - Wake up everything waiting on a &struct 510 * sbitmap_queue. 511 * @sbq: Bitmap queue to wake up. 512 */ 513 void sbitmap_queue_wake_all(struct sbitmap_queue *sbq); 514 515 /** 516 * sbitmap_queue_wake_up() - Wake up some of waiters in one waitqueue 517 * on a &struct sbitmap_queue. 518 * @sbq: Bitmap queue to wake up. 519 */ 520 void sbitmap_queue_wake_up(struct sbitmap_queue *sbq); 521 522 /** 523 * sbitmap_queue_show() - Dump &struct sbitmap_queue information to a &struct 524 * seq_file. 525 * @sbq: Bitmap queue to show. 526 * @m: struct seq_file to write to. 527 * 528 * This is intended for debugging. The format may change at any time. 529 */ 530 void sbitmap_queue_show(struct sbitmap_queue *sbq, struct seq_file *m); 531 532 #endif /* __LINUX_SCALE_BITMAP_H */ 533