| 1 | // SPDX-License-Identifier: GPL-2.0-only |
| 2 | /* |
| 3 | * Copyright (C) 2016 Facebook |
| 4 | * Copyright (C) 2013-2014 Jens Axboe |
| 5 | */ |
| 6 | |
| 7 | #include <linux/sched.h> |
| 8 | #include <linux/random.h> |
| 9 | #include <linux/sbitmap.h> |
| 10 | #include <linux/seq_file.h> |
| 11 | |
| 12 | static int init_alloc_hint(struct sbitmap *sb, gfp_t flags) |
| 13 | { |
| 14 | unsigned depth = sb->depth; |
| 15 | |
| 16 | sb->alloc_hint = alloc_percpu_gfp(unsigned int, flags); |
| 17 | if (!sb->alloc_hint) |
| 18 | return -ENOMEM; |
| 19 | |
| 20 | if (depth && !sb->round_robin) { |
| 21 | int i; |
| 22 | |
| 23 | for_each_possible_cpu(i) |
| 24 | *per_cpu_ptr(sb->alloc_hint, i) = get_random_u32_below(depth); |
| 25 | } |
| 26 | return 0; |
| 27 | } |
| 28 | |
| 29 | static inline unsigned update_alloc_hint_before_get(struct sbitmap *sb, |
| 30 | unsigned int depth) |
| 31 | { |
| 32 | unsigned hint; |
| 33 | |
| 34 | hint = this_cpu_read(*sb->alloc_hint); |
| 35 | if (unlikely(hint >= depth)) { |
| 36 | hint = depth ? get_random_u32_below(depth) : 0; |
| 37 | this_cpu_write(*sb->alloc_hint, hint); |
| 38 | } |
| 39 | |
| 40 | return hint; |
| 41 | } |
| 42 | |
| 43 | static inline void update_alloc_hint_after_get(struct sbitmap *sb, |
| 44 | unsigned int depth, |
| 45 | unsigned int hint, |
| 46 | unsigned int nr) |
| 47 | { |
| 48 | if (nr == -1) { |
| 49 | /* If the map is full, a hint won't do us much good. */ |
| 50 | this_cpu_write(*sb->alloc_hint, 0); |
| 51 | } else if (nr == hint || unlikely(sb->round_robin)) { |
| 52 | /* Only update the hint if we used it. */ |
| 53 | hint = nr + 1; |
| 54 | if (hint >= depth - 1) |
| 55 | hint = 0; |
| 56 | this_cpu_write(*sb->alloc_hint, hint); |
| 57 | } |
| 58 | } |
| 59 | |
| 60 | /* |
| 61 | * See if we have deferred clears that we can batch move |
| 62 | */ |
| 63 | static inline bool sbitmap_deferred_clear(struct sbitmap_word *map, |
| 64 | unsigned int depth, unsigned int alloc_hint, bool wrap) |
| 65 | { |
| 66 | unsigned long mask, word_mask; |
| 67 | |
| 68 | guard(raw_spinlock_irqsave)(&map->swap_lock); |
| 69 | |
| 70 | if (!map->cleared) { |
| 71 | if (depth == 0) |
| 72 | return false; |
| 73 | |
| 74 | word_mask = (~0UL) >> (BITS_PER_LONG - depth); |
| 75 | /* |
| 76 | * The current behavior is to always retry after moving |
| 77 | * ->cleared to word, and we change it to retry in case |
| 78 | * of any free bits. To avoid an infinite loop, we need |
| 79 | * to take wrap & alloc_hint into account, otherwise a |
| 80 | * soft lockup may occur. |
| 81 | */ |
| 82 | if (!wrap && alloc_hint) |
| 83 | word_mask &= ~((1UL << alloc_hint) - 1); |
| 84 | |
| 85 | return (READ_ONCE(map->word) & word_mask) != word_mask; |
| 86 | } |
| 87 | |
| 88 | /* |
| 89 | * First get a stable cleared mask, setting the old mask to 0. |
| 90 | */ |
| 91 | mask = xchg(&map->cleared, 0); |
| 92 | |
| 93 | /* |
| 94 | * Now clear the masked bits in our free word |
| 95 | */ |
| 96 | atomic_long_andnot(mask, (atomic_long_t *)&map->word); |
| 97 | BUILD_BUG_ON(sizeof(atomic_long_t) != sizeof(map->word)); |
| 98 | return true; |
| 99 | } |
| 100 | |
| 101 | int sbitmap_init_node(struct sbitmap *sb, unsigned int depth, int shift, |
| 102 | gfp_t flags, int node, bool round_robin, |
| 103 | bool alloc_hint) |
| 104 | { |
| 105 | unsigned int bits_per_word; |
| 106 | int i; |
| 107 | |
| 108 | if (shift < 0) |
| 109 | shift = sbitmap_calculate_shift(depth); |
| 110 | |
| 111 | bits_per_word = 1U << shift; |
| 112 | if (bits_per_word > BITS_PER_LONG) |
| 113 | return -EINVAL; |
| 114 | |
| 115 | sb->shift = shift; |
| 116 | sb->depth = depth; |
| 117 | sb->map_nr = DIV_ROUND_UP(sb->depth, bits_per_word); |
| 118 | sb->round_robin = round_robin; |
| 119 | |
| 120 | if (depth == 0) { |
| 121 | sb->map = NULL; |
| 122 | return 0; |
| 123 | } |
| 124 | |
| 125 | if (alloc_hint) { |
| 126 | if (init_alloc_hint(sb, flags)) |
| 127 | return -ENOMEM; |
| 128 | } else { |
| 129 | sb->alloc_hint = NULL; |
| 130 | } |
| 131 | |
| 132 | sb->map = kvzalloc_node(sb->map_nr * sizeof(*sb->map), flags, node); |
| 133 | if (!sb->map) { |
| 134 | free_percpu(sb->alloc_hint); |
| 135 | return -ENOMEM; |
| 136 | } |
| 137 | |
| 138 | for (i = 0; i < sb->map_nr; i++) |
| 139 | raw_spin_lock_init(&sb->map[i].swap_lock); |
| 140 | |
| 141 | return 0; |
| 142 | } |
| 143 | EXPORT_SYMBOL_GPL(sbitmap_init_node); |
| 144 | |
| 145 | void sbitmap_resize(struct sbitmap *sb, unsigned int depth) |
| 146 | { |
| 147 | unsigned int bits_per_word = 1U << sb->shift; |
| 148 | unsigned int i; |
| 149 | |
| 150 | for (i = 0; i < sb->map_nr; i++) |
| 151 | sbitmap_deferred_clear(&sb->map[i], 0, 0, 0); |
| 152 | |
| 153 | sb->depth = depth; |
| 154 | sb->map_nr = DIV_ROUND_UP(sb->depth, bits_per_word); |
| 155 | } |
| 156 | EXPORT_SYMBOL_GPL(sbitmap_resize); |
| 157 | |
| 158 | static int __sbitmap_get_word(unsigned long *word, unsigned long depth, |
| 159 | unsigned int hint, bool wrap) |
| 160 | { |
| 161 | int nr; |
| 162 | |
| 163 | /* don't wrap if starting from 0 */ |
| 164 | wrap = wrap && hint; |
| 165 | |
| 166 | while (1) { |
| 167 | nr = find_next_zero_bit(word, depth, hint); |
| 168 | if (unlikely(nr >= depth)) { |
| 169 | /* |
| 170 | * We started with an offset, and we didn't reset the |
| 171 | * offset to 0 in a failure case, so start from 0 to |
| 172 | * exhaust the map. |
| 173 | */ |
| 174 | if (hint && wrap) { |
| 175 | hint = 0; |
| 176 | continue; |
| 177 | } |
| 178 | return -1; |
| 179 | } |
| 180 | |
| 181 | if (!test_and_set_bit_lock(nr, word)) |
| 182 | break; |
| 183 | |
| 184 | hint = nr + 1; |
| 185 | if (hint >= depth - 1) |
| 186 | hint = 0; |
| 187 | } |
| 188 | |
| 189 | return nr; |
| 190 | } |
| 191 | |
| 192 | static int sbitmap_find_bit_in_word(struct sbitmap_word *map, |
| 193 | unsigned int depth, |
| 194 | unsigned int alloc_hint, |
| 195 | bool wrap) |
| 196 | { |
| 197 | int nr; |
| 198 | |
| 199 | do { |
| 200 | nr = __sbitmap_get_word(&map->word, depth, |
| 201 | alloc_hint, wrap); |
| 202 | if (nr != -1) |
| 203 | break; |
| 204 | if (!sbitmap_deferred_clear(map, depth, alloc_hint, wrap)) |
| 205 | break; |
| 206 | } while (1); |
| 207 | |
| 208 | return nr; |
| 209 | } |
| 210 | |
| 211 | static int sbitmap_find_bit(struct sbitmap *sb, |
| 212 | unsigned int depth, |
| 213 | unsigned int index, |
| 214 | unsigned int alloc_hint, |
| 215 | bool wrap) |
| 216 | { |
| 217 | unsigned int i; |
| 218 | int nr = -1; |
| 219 | |
| 220 | for (i = 0; i < sb->map_nr; i++) { |
| 221 | nr = sbitmap_find_bit_in_word(&sb->map[index], |
| 222 | min_t(unsigned int, |
| 223 | __map_depth(sb, index), |
| 224 | depth), |
| 225 | alloc_hint, wrap); |
| 226 | |
| 227 | if (nr != -1) { |
| 228 | nr += index << sb->shift; |
| 229 | break; |
| 230 | } |
| 231 | |
| 232 | /* Jump to next index. */ |
| 233 | alloc_hint = 0; |
| 234 | if (++index >= sb->map_nr) |
| 235 | index = 0; |
| 236 | } |
| 237 | |
| 238 | return nr; |
| 239 | } |
| 240 | |
| 241 | static int __sbitmap_get(struct sbitmap *sb, unsigned int alloc_hint) |
| 242 | { |
| 243 | unsigned int index; |
| 244 | |
| 245 | index = SB_NR_TO_INDEX(sb, alloc_hint); |
| 246 | |
| 247 | /* |
| 248 | * Unless we're doing round robin tag allocation, just use the |
| 249 | * alloc_hint to find the right word index. No point in looping |
| 250 | * twice in find_next_zero_bit() for that case. |
| 251 | */ |
| 252 | if (sb->round_robin) |
| 253 | alloc_hint = SB_NR_TO_BIT(sb, alloc_hint); |
| 254 | else |
| 255 | alloc_hint = 0; |
| 256 | |
| 257 | return sbitmap_find_bit(sb, UINT_MAX, index, alloc_hint, |
| 258 | !sb->round_robin); |
| 259 | } |
| 260 | |
| 261 | int sbitmap_get(struct sbitmap *sb) |
| 262 | { |
| 263 | int nr; |
| 264 | unsigned int hint, depth; |
| 265 | |
| 266 | if (WARN_ON_ONCE(unlikely(!sb->alloc_hint))) |
| 267 | return -1; |
| 268 | |
| 269 | depth = READ_ONCE(sb->depth); |
| 270 | hint = update_alloc_hint_before_get(sb, depth); |
| 271 | nr = __sbitmap_get(sb, hint); |
| 272 | update_alloc_hint_after_get(sb, depth, hint, nr); |
| 273 | |
| 274 | return nr; |
| 275 | } |
| 276 | EXPORT_SYMBOL_GPL(sbitmap_get); |
| 277 | |
| 278 | static int __sbitmap_get_shallow(struct sbitmap *sb, |
| 279 | unsigned int alloc_hint, |
| 280 | unsigned long shallow_depth) |
| 281 | { |
| 282 | unsigned int index; |
| 283 | |
| 284 | index = SB_NR_TO_INDEX(sb, alloc_hint); |
| 285 | alloc_hint = SB_NR_TO_BIT(sb, alloc_hint); |
| 286 | |
| 287 | return sbitmap_find_bit(sb, shallow_depth, index, alloc_hint, true); |
| 288 | } |
| 289 | |
| 290 | int sbitmap_get_shallow(struct sbitmap *sb, unsigned long shallow_depth) |
| 291 | { |
| 292 | int nr; |
| 293 | unsigned int hint, depth; |
| 294 | |
| 295 | if (WARN_ON_ONCE(unlikely(!sb->alloc_hint))) |
| 296 | return -1; |
| 297 | |
| 298 | depth = READ_ONCE(sb->depth); |
| 299 | hint = update_alloc_hint_before_get(sb, depth); |
| 300 | nr = __sbitmap_get_shallow(sb, hint, shallow_depth); |
| 301 | update_alloc_hint_after_get(sb, depth, hint, nr); |
| 302 | |
| 303 | return nr; |
| 304 | } |
| 305 | EXPORT_SYMBOL_GPL(sbitmap_get_shallow); |
| 306 | |
| 307 | bool sbitmap_any_bit_set(const struct sbitmap *sb) |
| 308 | { |
| 309 | unsigned int i; |
| 310 | |
| 311 | for (i = 0; i < sb->map_nr; i++) { |
| 312 | if (sb->map[i].word & ~sb->map[i].cleared) |
| 313 | return true; |
| 314 | } |
| 315 | return false; |
| 316 | } |
| 317 | EXPORT_SYMBOL_GPL(sbitmap_any_bit_set); |
| 318 | |
| 319 | static unsigned int __sbitmap_weight(const struct sbitmap *sb, bool set) |
| 320 | { |
| 321 | unsigned int i, weight = 0; |
| 322 | |
| 323 | for (i = 0; i < sb->map_nr; i++) { |
| 324 | const struct sbitmap_word *word = &sb->map[i]; |
| 325 | unsigned int word_depth = __map_depth(sb, i); |
| 326 | |
| 327 | if (set) |
| 328 | weight += bitmap_weight(&word->word, word_depth); |
| 329 | else |
| 330 | weight += bitmap_weight(&word->cleared, word_depth); |
| 331 | } |
| 332 | return weight; |
| 333 | } |
| 334 | |
| 335 | static unsigned int sbitmap_cleared(const struct sbitmap *sb) |
| 336 | { |
| 337 | return __sbitmap_weight(sb, false); |
| 338 | } |
| 339 | |
| 340 | unsigned int sbitmap_weight(const struct sbitmap *sb) |
| 341 | { |
| 342 | return __sbitmap_weight(sb, true) - sbitmap_cleared(sb); |
| 343 | } |
| 344 | EXPORT_SYMBOL_GPL(sbitmap_weight); |
| 345 | |
| 346 | void sbitmap_show(struct sbitmap *sb, struct seq_file *m) |
| 347 | { |
| 348 | seq_printf(m, "depth=%u\n", sb->depth); |
| 349 | seq_printf(m, "busy=%u\n", sbitmap_weight(sb)); |
| 350 | seq_printf(m, "cleared=%u\n", sbitmap_cleared(sb)); |
| 351 | seq_printf(m, "bits_per_word=%u\n", 1U << sb->shift); |
| 352 | seq_printf(m, "map_nr=%u\n", sb->map_nr); |
| 353 | } |
| 354 | EXPORT_SYMBOL_GPL(sbitmap_show); |
| 355 | |
| 356 | static inline void emit_byte(struct seq_file *m, unsigned int offset, u8 byte) |
| 357 | { |
| 358 | if ((offset & 0xf) == 0) { |
| 359 | if (offset != 0) |
| 360 | seq_putc(m, '\n'); |
| 361 | seq_printf(m, "%08x:", offset); |
| 362 | } |
| 363 | if ((offset & 0x1) == 0) |
| 364 | seq_putc(m, ' '); |
| 365 | seq_printf(m, "%02x", byte); |
| 366 | } |
| 367 | |
| 368 | void sbitmap_bitmap_show(struct sbitmap *sb, struct seq_file *m) |
| 369 | { |
| 370 | u8 byte = 0; |
| 371 | unsigned int byte_bits = 0; |
| 372 | unsigned int offset = 0; |
| 373 | int i; |
| 374 | |
| 375 | for (i = 0; i < sb->map_nr; i++) { |
| 376 | unsigned long word = READ_ONCE(sb->map[i].word); |
| 377 | unsigned long cleared = READ_ONCE(sb->map[i].cleared); |
| 378 | unsigned int word_bits = __map_depth(sb, i); |
| 379 | |
| 380 | word &= ~cleared; |
| 381 | |
| 382 | while (word_bits > 0) { |
| 383 | unsigned int bits = min(8 - byte_bits, word_bits); |
| 384 | |
| 385 | byte |= (word & (BIT(bits) - 1)) << byte_bits; |
| 386 | byte_bits += bits; |
| 387 | if (byte_bits == 8) { |
| 388 | emit_byte(m, offset, byte); |
| 389 | byte = 0; |
| 390 | byte_bits = 0; |
| 391 | offset++; |
| 392 | } |
| 393 | word >>= bits; |
| 394 | word_bits -= bits; |
| 395 | } |
| 396 | } |
| 397 | if (byte_bits) { |
| 398 | emit_byte(m, offset, byte); |
| 399 | offset++; |
| 400 | } |
| 401 | if (offset) |
| 402 | seq_putc(m, '\n'); |
| 403 | } |
| 404 | EXPORT_SYMBOL_GPL(sbitmap_bitmap_show); |
| 405 | |
| 406 | static unsigned int sbq_calc_wake_batch(struct sbitmap_queue *sbq, |
| 407 | unsigned int depth) |
| 408 | { |
| 409 | unsigned int wake_batch; |
| 410 | unsigned int shallow_depth; |
| 411 | |
| 412 | /* |
| 413 | * Each full word of the bitmap has bits_per_word bits, and there might |
| 414 | * be a partial word. There are depth / bits_per_word full words and |
| 415 | * depth % bits_per_word bits left over. In bitwise arithmetic: |
| 416 | * |
| 417 | * bits_per_word = 1 << shift |
| 418 | * depth / bits_per_word = depth >> shift |
| 419 | * depth % bits_per_word = depth & ((1 << shift) - 1) |
| 420 | * |
| 421 | * Each word can be limited to sbq->min_shallow_depth bits. |
| 422 | */ |
| 423 | shallow_depth = min(1U << sbq->sb.shift, sbq->min_shallow_depth); |
| 424 | depth = ((depth >> sbq->sb.shift) * shallow_depth + |
| 425 | min(depth & ((1U << sbq->sb.shift) - 1), shallow_depth)); |
| 426 | wake_batch = clamp_t(unsigned int, depth / SBQ_WAIT_QUEUES, 1, |
| 427 | SBQ_WAKE_BATCH); |
| 428 | |
| 429 | return wake_batch; |
| 430 | } |
| 431 | |
| 432 | int sbitmap_queue_init_node(struct sbitmap_queue *sbq, unsigned int depth, |
| 433 | int shift, bool round_robin, gfp_t flags, int node) |
| 434 | { |
| 435 | int ret; |
| 436 | int i; |
| 437 | |
| 438 | ret = sbitmap_init_node(&sbq->sb, depth, shift, flags, node, |
| 439 | round_robin, true); |
| 440 | if (ret) |
| 441 | return ret; |
| 442 | |
| 443 | sbq->min_shallow_depth = UINT_MAX; |
| 444 | sbq->wake_batch = sbq_calc_wake_batch(sbq, depth); |
| 445 | atomic_set(&sbq->wake_index, 0); |
| 446 | atomic_set(&sbq->ws_active, 0); |
| 447 | atomic_set(&sbq->completion_cnt, 0); |
| 448 | atomic_set(&sbq->wakeup_cnt, 0); |
| 449 | |
| 450 | sbq->ws = kzalloc_node(SBQ_WAIT_QUEUES * sizeof(*sbq->ws), flags, node); |
| 451 | if (!sbq->ws) { |
| 452 | sbitmap_free(&sbq->sb); |
| 453 | return -ENOMEM; |
| 454 | } |
| 455 | |
| 456 | for (i = 0; i < SBQ_WAIT_QUEUES; i++) |
| 457 | init_waitqueue_head(&sbq->ws[i].wait); |
| 458 | |
| 459 | return 0; |
| 460 | } |
| 461 | EXPORT_SYMBOL_GPL(sbitmap_queue_init_node); |
| 462 | |
| 463 | static void sbitmap_queue_update_wake_batch(struct sbitmap_queue *sbq, |
| 464 | unsigned int depth) |
| 465 | { |
| 466 | unsigned int wake_batch; |
| 467 | |
| 468 | wake_batch = sbq_calc_wake_batch(sbq, depth); |
| 469 | if (sbq->wake_batch != wake_batch) |
| 470 | WRITE_ONCE(sbq->wake_batch, wake_batch); |
| 471 | } |
| 472 | |
| 473 | void sbitmap_queue_recalculate_wake_batch(struct sbitmap_queue *sbq, |
| 474 | unsigned int users) |
| 475 | { |
| 476 | unsigned int wake_batch; |
| 477 | unsigned int depth = (sbq->sb.depth + users - 1) / users; |
| 478 | |
| 479 | wake_batch = clamp_val(depth / SBQ_WAIT_QUEUES, |
| 480 | 1, SBQ_WAKE_BATCH); |
| 481 | |
| 482 | WRITE_ONCE(sbq->wake_batch, wake_batch); |
| 483 | } |
| 484 | EXPORT_SYMBOL_GPL(sbitmap_queue_recalculate_wake_batch); |
| 485 | |
| 486 | void sbitmap_queue_resize(struct sbitmap_queue *sbq, unsigned int depth) |
| 487 | { |
| 488 | sbitmap_queue_update_wake_batch(sbq, depth); |
| 489 | sbitmap_resize(&sbq->sb, depth); |
| 490 | } |
| 491 | EXPORT_SYMBOL_GPL(sbitmap_queue_resize); |
| 492 | |
| 493 | int __sbitmap_queue_get(struct sbitmap_queue *sbq) |
| 494 | { |
| 495 | return sbitmap_get(&sbq->sb); |
| 496 | } |
| 497 | EXPORT_SYMBOL_GPL(__sbitmap_queue_get); |
| 498 | |
| 499 | unsigned long __sbitmap_queue_get_batch(struct sbitmap_queue *sbq, int nr_tags, |
| 500 | unsigned int *offset) |
| 501 | { |
| 502 | struct sbitmap *sb = &sbq->sb; |
| 503 | unsigned int hint, depth; |
| 504 | unsigned long index, nr; |
| 505 | int i; |
| 506 | |
| 507 | if (unlikely(sb->round_robin)) |
| 508 | return 0; |
| 509 | |
| 510 | depth = READ_ONCE(sb->depth); |
| 511 | hint = update_alloc_hint_before_get(sb, depth); |
| 512 | |
| 513 | index = SB_NR_TO_INDEX(sb, hint); |
| 514 | |
| 515 | for (i = 0; i < sb->map_nr; i++) { |
| 516 | struct sbitmap_word *map = &sb->map[index]; |
| 517 | unsigned long get_mask; |
| 518 | unsigned int map_depth = __map_depth(sb, index); |
| 519 | unsigned long val; |
| 520 | |
| 521 | sbitmap_deferred_clear(map, 0, 0, 0); |
| 522 | val = READ_ONCE(map->word); |
| 523 | if (val == (1UL << (map_depth - 1)) - 1) |
| 524 | goto next; |
| 525 | |
| 526 | nr = find_first_zero_bit(&val, map_depth); |
| 527 | if (nr + nr_tags <= map_depth) { |
| 528 | atomic_long_t *ptr = (atomic_long_t *) &map->word; |
| 529 | |
| 530 | get_mask = ((1UL << nr_tags) - 1) << nr; |
| 531 | while (!atomic_long_try_cmpxchg(ptr, &val, |
| 532 | get_mask | val)) |
| 533 | ; |
| 534 | get_mask = (get_mask & ~val) >> nr; |
| 535 | if (get_mask) { |
| 536 | *offset = nr + (index << sb->shift); |
| 537 | update_alloc_hint_after_get(sb, depth, hint, |
| 538 | *offset + nr_tags - 1); |
| 539 | return get_mask; |
| 540 | } |
| 541 | } |
| 542 | next: |
| 543 | /* Jump to next index. */ |
| 544 | if (++index >= sb->map_nr) |
| 545 | index = 0; |
| 546 | } |
| 547 | |
| 548 | return 0; |
| 549 | } |
| 550 | |
| 551 | int sbitmap_queue_get_shallow(struct sbitmap_queue *sbq, |
| 552 | unsigned int shallow_depth) |
| 553 | { |
| 554 | WARN_ON_ONCE(shallow_depth < sbq->min_shallow_depth); |
| 555 | |
| 556 | return sbitmap_get_shallow(&sbq->sb, shallow_depth); |
| 557 | } |
| 558 | EXPORT_SYMBOL_GPL(sbitmap_queue_get_shallow); |
| 559 | |
| 560 | void sbitmap_queue_min_shallow_depth(struct sbitmap_queue *sbq, |
| 561 | unsigned int min_shallow_depth) |
| 562 | { |
| 563 | sbq->min_shallow_depth = min_shallow_depth; |
| 564 | sbitmap_queue_update_wake_batch(sbq, sbq->sb.depth); |
| 565 | } |
| 566 | EXPORT_SYMBOL_GPL(sbitmap_queue_min_shallow_depth); |
| 567 | |
| 568 | static void __sbitmap_queue_wake_up(struct sbitmap_queue *sbq, int nr) |
| 569 | { |
| 570 | int i, wake_index, woken; |
| 571 | |
| 572 | if (!atomic_read(&sbq->ws_active)) |
| 573 | return; |
| 574 | |
| 575 | wake_index = atomic_read(&sbq->wake_index); |
| 576 | for (i = 0; i < SBQ_WAIT_QUEUES; i++) { |
| 577 | struct sbq_wait_state *ws = &sbq->ws[wake_index]; |
| 578 | |
| 579 | /* |
| 580 | * Advance the index before checking the current queue. |
| 581 | * It improves fairness, by ensuring the queue doesn't |
| 582 | * need to be fully emptied before trying to wake up |
| 583 | * from the next one. |
| 584 | */ |
| 585 | wake_index = sbq_index_inc(wake_index); |
| 586 | |
| 587 | if (waitqueue_active(&ws->wait)) { |
| 588 | woken = wake_up_nr(&ws->wait, nr); |
| 589 | if (woken == nr) |
| 590 | break; |
| 591 | nr -= woken; |
| 592 | } |
| 593 | } |
| 594 | |
| 595 | if (wake_index != atomic_read(&sbq->wake_index)) |
| 596 | atomic_set(&sbq->wake_index, wake_index); |
| 597 | } |
| 598 | |
| 599 | void sbitmap_queue_wake_up(struct sbitmap_queue *sbq, int nr) |
| 600 | { |
| 601 | unsigned int wake_batch = READ_ONCE(sbq->wake_batch); |
| 602 | unsigned int wakeups; |
| 603 | |
| 604 | if (!atomic_read(&sbq->ws_active)) |
| 605 | return; |
| 606 | |
| 607 | atomic_add(nr, &sbq->completion_cnt); |
| 608 | wakeups = atomic_read(&sbq->wakeup_cnt); |
| 609 | |
| 610 | do { |
| 611 | if (atomic_read(&sbq->completion_cnt) - wakeups < wake_batch) |
| 612 | return; |
| 613 | } while (!atomic_try_cmpxchg(&sbq->wakeup_cnt, |
| 614 | &wakeups, wakeups + wake_batch)); |
| 615 | |
| 616 | __sbitmap_queue_wake_up(sbq, wake_batch); |
| 617 | } |
| 618 | EXPORT_SYMBOL_GPL(sbitmap_queue_wake_up); |
| 619 | |
| 620 | static inline void sbitmap_update_cpu_hint(struct sbitmap *sb, int cpu, int tag) |
| 621 | { |
| 622 | if (likely(!sb->round_robin && tag < sb->depth)) |
| 623 | data_race(*per_cpu_ptr(sb->alloc_hint, cpu) = tag); |
| 624 | } |
| 625 | |
| 626 | void sbitmap_queue_clear_batch(struct sbitmap_queue *sbq, int offset, |
| 627 | int *tags, int nr_tags) |
| 628 | { |
| 629 | struct sbitmap *sb = &sbq->sb; |
| 630 | unsigned long *addr = NULL; |
| 631 | unsigned long mask = 0; |
| 632 | int i; |
| 633 | |
| 634 | smp_mb__before_atomic(); |
| 635 | for (i = 0; i < nr_tags; i++) { |
| 636 | const int tag = tags[i] - offset; |
| 637 | unsigned long *this_addr; |
| 638 | |
| 639 | /* since we're clearing a batch, skip the deferred map */ |
| 640 | this_addr = &sb->map[SB_NR_TO_INDEX(sb, tag)].word; |
| 641 | if (!addr) { |
| 642 | addr = this_addr; |
| 643 | } else if (addr != this_addr) { |
| 644 | atomic_long_andnot(mask, (atomic_long_t *) addr); |
| 645 | mask = 0; |
| 646 | addr = this_addr; |
| 647 | } |
| 648 | mask |= (1UL << SB_NR_TO_BIT(sb, tag)); |
| 649 | } |
| 650 | |
| 651 | if (mask) |
| 652 | atomic_long_andnot(mask, (atomic_long_t *) addr); |
| 653 | |
| 654 | smp_mb__after_atomic(); |
| 655 | sbitmap_queue_wake_up(sbq, nr_tags); |
| 656 | sbitmap_update_cpu_hint(&sbq->sb, raw_smp_processor_id(), |
| 657 | tags[nr_tags - 1] - offset); |
| 658 | } |
| 659 | |
| 660 | void sbitmap_queue_clear(struct sbitmap_queue *sbq, unsigned int nr, |
| 661 | unsigned int cpu) |
| 662 | { |
| 663 | /* |
| 664 | * Once the clear bit is set, the bit may be allocated out. |
| 665 | * |
| 666 | * Orders READ/WRITE on the associated instance(such as request |
| 667 | * of blk_mq) by this bit for avoiding race with re-allocation, |
| 668 | * and its pair is the memory barrier implied in __sbitmap_get_word. |
| 669 | * |
| 670 | * One invariant is that the clear bit has to be zero when the bit |
| 671 | * is in use. |
| 672 | */ |
| 673 | smp_mb__before_atomic(); |
| 674 | sbitmap_deferred_clear_bit(&sbq->sb, nr); |
| 675 | |
| 676 | /* |
| 677 | * Pairs with the memory barrier in set_current_state() to ensure the |
| 678 | * proper ordering of clear_bit_unlock()/waitqueue_active() in the waker |
| 679 | * and test_and_set_bit_lock()/prepare_to_wait()/finish_wait() in the |
| 680 | * waiter. See the comment on waitqueue_active(). |
| 681 | */ |
| 682 | smp_mb__after_atomic(); |
| 683 | sbitmap_queue_wake_up(sbq, 1); |
| 684 | sbitmap_update_cpu_hint(&sbq->sb, cpu, nr); |
| 685 | } |
| 686 | EXPORT_SYMBOL_GPL(sbitmap_queue_clear); |
| 687 | |
| 688 | void sbitmap_queue_wake_all(struct sbitmap_queue *sbq) |
| 689 | { |
| 690 | int i, wake_index; |
| 691 | |
| 692 | /* |
| 693 | * Pairs with the memory barrier in set_current_state() like in |
| 694 | * sbitmap_queue_wake_up(). |
| 695 | */ |
| 696 | smp_mb(); |
| 697 | wake_index = atomic_read(&sbq->wake_index); |
| 698 | for (i = 0; i < SBQ_WAIT_QUEUES; i++) { |
| 699 | struct sbq_wait_state *ws = &sbq->ws[wake_index]; |
| 700 | |
| 701 | if (waitqueue_active(&ws->wait)) |
| 702 | wake_up(&ws->wait); |
| 703 | |
| 704 | wake_index = sbq_index_inc(wake_index); |
| 705 | } |
| 706 | } |
| 707 | EXPORT_SYMBOL_GPL(sbitmap_queue_wake_all); |
| 708 | |
| 709 | void sbitmap_queue_show(struct sbitmap_queue *sbq, struct seq_file *m) |
| 710 | { |
| 711 | bool first; |
| 712 | int i; |
| 713 | |
| 714 | sbitmap_show(&sbq->sb, m); |
| 715 | |
| 716 | seq_puts(m, "alloc_hint={"); |
| 717 | first = true; |
| 718 | for_each_possible_cpu(i) { |
| 719 | if (!first) |
| 720 | seq_puts(m, ", "); |
| 721 | first = false; |
| 722 | seq_printf(m, "%u", *per_cpu_ptr(sbq->sb.alloc_hint, i)); |
| 723 | } |
| 724 | seq_puts(m, "}\n"); |
| 725 | |
| 726 | seq_printf(m, "wake_batch=%u\n", sbq->wake_batch); |
| 727 | seq_printf(m, "wake_index=%d\n", atomic_read(&sbq->wake_index)); |
| 728 | seq_printf(m, "ws_active=%d\n", atomic_read(&sbq->ws_active)); |
| 729 | |
| 730 | seq_puts(m, "ws={\n"); |
| 731 | for (i = 0; i < SBQ_WAIT_QUEUES; i++) { |
| 732 | struct sbq_wait_state *ws = &sbq->ws[i]; |
| 733 | seq_printf(m, "\t{.wait=%s},\n", |
| 734 | waitqueue_active(&ws->wait) ? "active" : "inactive"); |
| 735 | } |
| 736 | seq_puts(m, "}\n"); |
| 737 | |
| 738 | seq_printf(m, "round_robin=%d\n", sbq->sb.round_robin); |
| 739 | seq_printf(m, "min_shallow_depth=%u\n", sbq->min_shallow_depth); |
| 740 | } |
| 741 | EXPORT_SYMBOL_GPL(sbitmap_queue_show); |
| 742 | |
| 743 | void sbitmap_add_wait_queue(struct sbitmap_queue *sbq, |
| 744 | struct sbq_wait_state *ws, |
| 745 | struct sbq_wait *sbq_wait) |
| 746 | { |
| 747 | if (!sbq_wait->sbq) { |
| 748 | sbq_wait->sbq = sbq; |
| 749 | atomic_inc(&sbq->ws_active); |
| 750 | add_wait_queue(&ws->wait, &sbq_wait->wait); |
| 751 | } |
| 752 | } |
| 753 | EXPORT_SYMBOL_GPL(sbitmap_add_wait_queue); |
| 754 | |
| 755 | void sbitmap_del_wait_queue(struct sbq_wait *sbq_wait) |
| 756 | { |
| 757 | list_del_init(&sbq_wait->wait.entry); |
| 758 | if (sbq_wait->sbq) { |
| 759 | atomic_dec(&sbq_wait->sbq->ws_active); |
| 760 | sbq_wait->sbq = NULL; |
| 761 | } |
| 762 | } |
| 763 | EXPORT_SYMBOL_GPL(sbitmap_del_wait_queue); |
| 764 | |
| 765 | void sbitmap_prepare_to_wait(struct sbitmap_queue *sbq, |
| 766 | struct sbq_wait_state *ws, |
| 767 | struct sbq_wait *sbq_wait, int state) |
| 768 | { |
| 769 | if (!sbq_wait->sbq) { |
| 770 | atomic_inc(&sbq->ws_active); |
| 771 | sbq_wait->sbq = sbq; |
| 772 | } |
| 773 | prepare_to_wait_exclusive(&ws->wait, &sbq_wait->wait, state); |
| 774 | } |
| 775 | EXPORT_SYMBOL_GPL(sbitmap_prepare_to_wait); |
| 776 | |
| 777 | void sbitmap_finish_wait(struct sbitmap_queue *sbq, struct sbq_wait_state *ws, |
| 778 | struct sbq_wait *sbq_wait) |
| 779 | { |
| 780 | finish_wait(&ws->wait, &sbq_wait->wait); |
| 781 | if (sbq_wait->sbq) { |
| 782 | atomic_dec(&sbq->ws_active); |
| 783 | sbq_wait->sbq = NULL; |
| 784 | } |
| 785 | } |
| 786 | EXPORT_SYMBOL_GPL(sbitmap_finish_wait); |