| 1 | |
| 2 | * Copyright (C) 2016 Facebook
|
| 3 | * Copyright (C) 2013-2014 Jens Axboe
|
| 4 | *
|
| 5 | * This program is free software; you can redistribute it and/or
|
| 6 | * modify it under the terms of the GNU General Public
|
| 7 | * License v2 as published by the Free Software Foundation.
|
| 8 | *
|
| 9 | * This program is distributed in the hope that it will be useful,
|
| 10 | * but WITHOUT ANY WARRANTY; without even the implied warranty of
|
| 11 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
|
| 12 | * General Public License for more details.
|
| 13 | *
|
| 14 | * You should have received a copy of the GNU General Public License
|
| 15 | * along with this program. If not, see <https://www.gnu.org/licenses/>.
|
| 16 |
|
| 17 |
|
| 18 | #include <linux/sched.h>
|
| 19 | #include <linux/random.h>
|
| 20 | #include <linux/sbitmap.h>
|
| 21 | #include <linux/seq_file.h>
|
| 22 |
|
| 23 | |
| 24 | * See if we have deferred clears that we can batch move
|
| 25 |
|
| 26 | static inline bool sbitmap_deferred_clear(struct sbitmap *sb, int index)
|
| 27 | {
|
| 28 | unsigned long mask, val;
|
| 29 | bool ret = false;
|
| 30 | unsigned long flags;
|
| 31 |
|
| 32 | spin_lock_irqsave(&sb->map[index].swap_lock, flags);
|
| 33 |
|
| 34 | if (!sb->map[index].cleared)
|
| 35 | goto out_unlock;
|
| 36 |
|
| 37 | |
| 38 | * First get a stable cleared mask, setting the old mask to 0.
|
| 39 |
|
| 40 | do {
|
| 41 | mask = sb->map[index].cleared;
|
| 42 | } while (cmpxchg(&sb->map[index].cleared, mask, 0) != mask);
|
| 43 |
|
| 44 | |
| 45 | * Now clear the masked bits in our free word
|
| 46 |
|
| 47 | do {
|
| 48 | val = sb->map[index].word;
|
| 49 | } while (cmpxchg(&sb->map[index].word, val, val & ~mask) != val);
|
| 50 |
|
| 51 | ret = true;
|
| 52 | out_unlock:
|
| 53 | spin_unlock_irqrestore(&sb->map[index].swap_lock, flags);
|
| 54 | return ret;
|
| 55 | }
|
| 56 |
|
| 57 | int sbitmap_init_node(struct sbitmap *sb, unsigned int depth, int shift,
|
| 58 | gfp_t flags, int node)
|
| 59 | {
|
| 60 | unsigned int bits_per_word;
|
| 61 | unsigned int i;
|
| 62 |
|
| 63 | if (shift < 0) {
|
| 64 | shift = ilog2(BITS_PER_LONG);
|
| 65 | |
| 66 | * If the bitmap is small, shrink the number of bits per word so
|
| 67 | * we spread over a few cachelines, at least. If less than 4
|
| 68 | * bits, just forget about it, it's not going to work optimally
|
| 69 | * anyway.
|
| 70 |
|
| 71 | if (depth >= 4) {
|
| 72 | while ((4U << shift) > depth)
|
| 73 | shift--;
|
| 74 | }
|
| 75 | }
|
| 76 | bits_per_word = 1U << shift;
|
| 77 | if (bits_per_word > BITS_PER_LONG)
|
| 78 | return -EINVAL;
|
| 79 |
|
| 80 | sb->shift = shift;
|
| 81 | sb->depth = depth;
|
| 82 | sb->map_nr = DIV_ROUND_UP(sb->depth, bits_per_word);
|
| 83 |
|
| 84 | if (depth == 0) {
|
| 85 | sb->map = NULL;
|
| 86 | return 0;
|
| 87 | }
|
| 88 |
|
| 89 | sb->map = kcalloc_node(sb->map_nr, sizeof(*sb->map), flags, node);
|
| 90 | if (!sb->map)
|
| 91 | return -ENOMEM;
|
| 92 |
|
| 93 | for (i = 0; i < sb->map_nr; i++) {
|
| 94 | sb->map[i].depth = min(depth, bits_per_word);
|
| 95 | depth -= sb->map[i].depth;
|
| 96 | spin_lock_init(&sb->map[i].swap_lock);
|
| 97 | }
|
| 98 | return 0;
|
| 99 | }
|
| 100 | EXPORT_SYMBOL_GPL(sbitmap_init_node);
|
| 101 |
|
| 102 | void sbitmap_resize(struct sbitmap *sb, unsigned int depth)
|
| 103 | {
|
| 104 | unsigned int bits_per_word = 1U << sb->shift;
|
| 105 | unsigned int i;
|
| 106 |
|
| 107 | for (i = 0; i < sb->map_nr; i++)
|
| 108 | sbitmap_deferred_clear(sb, i);
|
| 109 |
|
| 110 | sb->depth = depth;
|
| 111 | sb->map_nr = DIV_ROUND_UP(sb->depth, bits_per_word);
|
| 112 |
|
| 113 | for (i = 0; i < sb->map_nr; i++) {
|
| 114 | sb->map[i].depth = min(depth, bits_per_word);
|
| 115 | depth -= sb->map[i].depth;
|
| 116 | }
|
| 117 | }
|
| 118 | EXPORT_SYMBOL_GPL(sbitmap_resize);
|
| 119 |
|
| 120 | static int __sbitmap_get_word(unsigned long *word, unsigned long depth,
|
| 121 | unsigned int hint, bool wrap)
|
| 122 | {
|
| 123 | unsigned int orig_hint = hint;
|
| 124 | int nr;
|
| 125 |
|
| 126 | while (1) {
|
| 127 | nr = find_next_zero_bit(word, depth, hint);
|
| 128 | if (unlikely(nr >= depth)) {
|
| 129 | |
| 130 | * We started with an offset, and we didn't reset the
|
| 131 | * offset to 0 in a failure case, so start from 0 to
|
| 132 | * exhaust the map.
|
| 133 |
|
| 134 | if (orig_hint && hint && wrap) {
|
| 135 | hint = orig_hint = 0;
|
| 136 | continue;
|
| 137 | }
|
| 138 | return -1;
|
| 139 | }
|
| 140 |
|
| 141 | if (!test_and_set_bit_lock(nr, word))
|
| 142 | break;
|
| 143 |
|
| 144 | hint = nr + 1;
|
| 145 | if (hint >= depth - 1)
|
| 146 | hint = 0;
|
| 147 | }
|
| 148 |
|
| 149 | return nr;
|
| 150 | }
|
| 151 |
|
| 152 | static int sbitmap_find_bit_in_index(struct sbitmap *sb, int index,
|
| 153 | unsigned int alloc_hint, bool round_robin)
|
| 154 | {
|
| 155 | int nr;
|
| 156 |
|
| 157 | do {
|
| 158 | nr = __sbitmap_get_word(&sb->map[index].word,
|
| 159 | sb->map[index].depth, alloc_hint,
|
| 160 | !round_robin);
|
| 161 | if (nr != -1)
|
| 162 | break;
|
| 163 | if (!sbitmap_deferred_clear(sb, index))
|
| 164 | break;
|
| 165 | } while (1);
|
| 166 |
|
| 167 | return nr;
|
| 168 | }
|
| 169 |
|
| 170 | int sbitmap_get(struct sbitmap *sb, unsigned int alloc_hint, bool round_robin)
|
| 171 | {
|
| 172 | unsigned int i, index;
|
| 173 | int nr = -1;
|
| 174 |
|
| 175 | index = SB_NR_TO_INDEX(sb, alloc_hint);
|
| 176 |
|
| 177 | |
| 178 | * Unless we're doing round robin tag allocation, just use the
|
| 179 | * alloc_hint to find the right word index. No point in looping
|
| 180 | * twice in find_next_zero_bit() for that case.
|
| 181 |
|
| 182 | if (round_robin)
|
| 183 | alloc_hint = SB_NR_TO_BIT(sb, alloc_hint);
|
| 184 | else
|
| 185 | alloc_hint = 0;
|
| 186 |
|
| 187 | for (i = 0; i < sb->map_nr; i++) {
|
| 188 | nr = sbitmap_find_bit_in_index(sb, index, alloc_hint,
|
| 189 | round_robin);
|
| 190 | if (nr != -1) {
|
| 191 | nr += index << sb->shift;
|
| 192 | break;
|
| 193 | }
|
| 194 |
|
| 195 |
|
| 196 | alloc_hint = 0;
|
| 197 | if (++index >= sb->map_nr)
|
| 198 | index = 0;
|
| 199 | }
|
| 200 |
|
| 201 | return nr;
|
| 202 | }
|
| 203 | EXPORT_SYMBOL_GPL(sbitmap_get);
|
| 204 |
|
| 205 | int sbitmap_get_shallow(struct sbitmap *sb, unsigned int alloc_hint,
|
| 206 | unsigned long shallow_depth)
|
| 207 | {
|
| 208 | unsigned int i, index;
|
| 209 | int nr = -1;
|
| 210 |
|
| 211 | index = SB_NR_TO_INDEX(sb, alloc_hint);
|
| 212 |
|
| 213 | for (i = 0; i < sb->map_nr; i++) {
|
| 214 | again:
|
| 215 | nr = __sbitmap_get_word(&sb->map[index].word,
|
| 216 | min(sb->map[index].depth, shallow_depth),
|
| 217 | SB_NR_TO_BIT(sb, alloc_hint), true);
|
| 218 | if (nr != -1) {
|
| 219 | nr += index << sb->shift;
|
| 220 | break;
|
| 221 | }
|
| 222 |
|
| 223 | if (sbitmap_deferred_clear(sb, index))
|
| 224 | goto again;
|
| 225 |
|
| 226 |
|
| 227 | index++;
|
| 228 | alloc_hint = index << sb->shift;
|
| 229 |
|
| 230 | if (index >= sb->map_nr) {
|
| 231 | index = 0;
|
| 232 | alloc_hint = 0;
|
| 233 | }
|
| 234 | }
|
| 235 |
|
| 236 | return nr;
|
| 237 | }
|
| 238 | EXPORT_SYMBOL_GPL(sbitmap_get_shallow);
|
| 239 |
|
| 240 | bool sbitmap_any_bit_set(const struct sbitmap *sb)
|
| 241 | {
|
| 242 | unsigned int i;
|
| 243 |
|
| 244 | for (i = 0; i < sb->map_nr; i++) {
|
| 245 | if (sb->map[i].word & ~sb->map[i].cleared)
|
| 246 | return true;
|
| 247 | }
|
| 248 | return false;
|
| 249 | }
|
| 250 | EXPORT_SYMBOL_GPL(sbitmap_any_bit_set);
|
| 251 |
|
| 252 | bool sbitmap_any_bit_clear(const struct sbitmap *sb)
|
| 253 | {
|
| 254 | unsigned int i;
|
| 255 |
|
| 256 | for (i = 0; i < sb->map_nr; i++) {
|
| 257 | const struct sbitmap_word *word = &sb->map[i];
|
| 258 | unsigned long mask = word->word & ~word->cleared;
|
| 259 | unsigned long ret;
|
| 260 |
|
| 261 | ret = find_first_zero_bit(&mask, word->depth);
|
| 262 | if (ret < word->depth)
|
| 263 | return true;
|
| 264 | }
|
| 265 | return false;
|
| 266 | }
|
| 267 | EXPORT_SYMBOL_GPL(sbitmap_any_bit_clear);
|
| 268 |
|
| 269 | static unsigned int __sbitmap_weight(const struct sbitmap *sb, bool set)
|
| 270 | {
|
| 271 | unsigned int i, weight = 0;
|
| 272 |
|
| 273 | for (i = 0; i < sb->map_nr; i++) {
|
| 274 | const struct sbitmap_word *word = &sb->map[i];
|
| 275 |
|
| 276 | if (set)
|
| 277 | weight += bitmap_weight(&word->word, word->depth);
|
| 278 | else
|
| 279 | weight += bitmap_weight(&word->cleared, word->depth);
|
| 280 | }
|
| 281 | return weight;
|
| 282 | }
|
| 283 |
|
| 284 | static unsigned int sbitmap_weight(const struct sbitmap *sb)
|
| 285 | {
|
| 286 | return __sbitmap_weight(sb, true);
|
| 287 | }
|
| 288 |
|
| 289 | static unsigned int sbitmap_cleared(const struct sbitmap *sb)
|
| 290 | {
|
| 291 | return __sbitmap_weight(sb, false);
|
| 292 | }
|
| 293 |
|
| 294 | void sbitmap_show(struct sbitmap *sb, struct seq_file *m)
|
| 295 | {
|
| 296 | seq_printf(m, "depth=%u\n", sb->depth);
|
| 297 | seq_printf(m, "busy=%u\n", sbitmap_weight(sb) - sbitmap_cleared(sb));
|
| 298 | seq_printf(m, "cleared=%u\n", sbitmap_cleared(sb));
|
| 299 | seq_printf(m, "bits_per_word=%u\n", 1U << sb->shift);
|
| 300 | seq_printf(m, "map_nr=%u\n", sb->map_nr);
|
| 301 | }
|
| 302 | EXPORT_SYMBOL_GPL(sbitmap_show);
|
| 303 |
|
| 304 | static inline void emit_byte(struct seq_file *m, unsigned int offset, u8 byte)
|
| 305 | {
|
| 306 | if ((offset & 0xf) == 0) {
|
| 307 | if (offset != 0)
|
| 308 | seq_putc(m, '\n');
|
| 309 | seq_printf(m, "%08x:", offset);
|
| 310 | }
|
| 311 | if ((offset & 0x1) == 0)
|
| 312 | seq_putc(m, ' ');
|
| 313 | seq_printf(m, "%02x", byte);
|
| 314 | }
|
| 315 |
|
| 316 | void sbitmap_bitmap_show(struct sbitmap *sb, struct seq_file *m)
|
| 317 | {
|
| 318 | u8 byte = 0;
|
| 319 | unsigned int byte_bits = 0;
|
| 320 | unsigned int offset = 0;
|
| 321 | int i;
|
| 322 |
|
| 323 | for (i = 0; i < sb->map_nr; i++) {
|
| 324 | unsigned long word = READ_ONCE(sb->map[i].word);
|
| 325 | unsigned int word_bits = READ_ONCE(sb->map[i].depth);
|
| 326 |
|
| 327 | while (word_bits > 0) {
|
| 328 | unsigned int bits = min(8 - byte_bits, word_bits);
|
| 329 |
|
| 330 | byte |= (word & (BIT(bits) - 1)) << byte_bits;
|
| 331 | byte_bits += bits;
|
| 332 | if (byte_bits == 8) {
|
| 333 | emit_byte(m, offset, byte);
|
| 334 | byte = 0;
|
| 335 | byte_bits = 0;
|
| 336 | offset++;
|
| 337 | }
|
| 338 | word >>= bits;
|
| 339 | word_bits -= bits;
|
| 340 | }
|
| 341 | }
|
| 342 | if (byte_bits) {
|
| 343 | emit_byte(m, offset, byte);
|
| 344 | offset++;
|
| 345 | }
|
| 346 | if (offset)
|
| 347 | seq_putc(m, '\n');
|
| 348 | }
|
| 349 | EXPORT_SYMBOL_GPL(sbitmap_bitmap_show);
|
| 350 |
|
| 351 | static unsigned int sbq_calc_wake_batch(struct sbitmap_queue *sbq,
|
| 352 | unsigned int depth)
|
| 353 | {
|
| 354 | unsigned int wake_batch;
|
| 355 | unsigned int shallow_depth;
|
| 356 |
|
| 357 | |
| 358 | * For each batch, we wake up one queue. We need to make sure that our
|
| 359 | * batch size is small enough that the full depth of the bitmap,
|
| 360 | * potentially limited by a shallow depth, is enough to wake up all of
|
| 361 | * the queues.
|
| 362 | *
|
| 363 | * Each full word of the bitmap has bits_per_word bits, and there might
|
| 364 | * be a partial word. There are depth / bits_per_word full words and
|
| 365 | * depth % bits_per_word bits left over. In bitwise arithmetic:
|
| 366 | *
|
| 367 | * bits_per_word = 1 << shift
|
| 368 | * depth / bits_per_word = depth >> shift
|
| 369 | * depth % bits_per_word = depth & ((1 << shift) - 1)
|
| 370 | *
|
| 371 | * Each word can be limited to sbq->min_shallow_depth bits.
|
| 372 |
|
| 373 | shallow_depth = min(1U << sbq->sb.shift, sbq->min_shallow_depth);
|
| 374 | depth = ((depth >> sbq->sb.shift) * shallow_depth +
|
| 375 | min(depth & ((1U << sbq->sb.shift) - 1), shallow_depth));
|
| 376 | wake_batch = clamp_t(unsigned int, depth / SBQ_WAIT_QUEUES, 1,
|
| 377 | SBQ_WAKE_BATCH);
|
| 378 |
|
| 379 | return wake_batch;
|
| 380 | }
|
| 381 |
|
| 382 | int sbitmap_queue_init_node(struct sbitmap_queue *sbq, unsigned int depth,
|
| 383 | int shift, bool round_robin, gfp_t flags, int node)
|
| 384 | {
|
| 385 | int ret;
|
| 386 | int i;
|
| 387 |
|
| 388 | ret = sbitmap_init_node(&sbq->sb, depth, shift, flags, node);
|
| 389 | if (ret)
|
| 390 | return ret;
|
| 391 |
|
| 392 | sbq->alloc_hint = alloc_percpu_gfp(unsigned int, flags);
|
| 393 | if (!sbq->alloc_hint) {
|
| 394 | sbitmap_free(&sbq->sb);
|
| 395 | return -ENOMEM;
|
| 396 | }
|
| 397 |
|
| 398 | if (depth && !round_robin) {
|
| 399 | for_each_possible_cpu(i)
|
| 400 | *per_cpu_ptr(sbq->alloc_hint, i) = prandom_u32() % depth;
|
| 401 | }
|
| 402 |
|
| 403 | sbq->min_shallow_depth = UINT_MAX;
|
| 404 | sbq->wake_batch = sbq_calc_wake_batch(sbq, depth);
|
| 405 | atomic_set(&sbq->wake_index, 0);
|
| 406 | atomic_set(&sbq->ws_active, 0);
|
| 407 |
|
| 408 | sbq->ws = kzalloc_node(SBQ_WAIT_QUEUES * sizeof(*sbq->ws), flags, node);
|
| 409 | if (!sbq->ws) {
|
| 410 | free_percpu(sbq->alloc_hint);
|
| 411 | sbitmap_free(&sbq->sb);
|
| 412 | return -ENOMEM;
|
| 413 | }
|
| 414 |
|
| 415 | for (i = 0; i < SBQ_WAIT_QUEUES; i++) {
|
| 416 | init_waitqueue_head(&sbq->ws[i].wait);
|
| 417 | atomic_set(&sbq->ws[i].wait_cnt, sbq->wake_batch);
|
| 418 | }
|
| 419 |
|
| 420 | sbq->round_robin = round_robin;
|
| 421 | return 0;
|
| 422 | }
|
| 423 | EXPORT_SYMBOL_GPL(sbitmap_queue_init_node);
|
| 424 |
|
| 425 | static void sbitmap_queue_update_wake_batch(struct sbitmap_queue *sbq,
|
| 426 | unsigned int depth)
|
| 427 | {
|
| 428 | unsigned int wake_batch = sbq_calc_wake_batch(sbq, depth);
|
| 429 | int i;
|
| 430 |
|
| 431 | if (sbq->wake_batch != wake_batch) {
|
| 432 | WRITE_ONCE(sbq->wake_batch, wake_batch);
|
| 433 | |
| 434 | * Pairs with the memory barrier in sbitmap_queue_wake_up()
|
| 435 | * to ensure that the batch size is updated before the wait
|
| 436 | * counts.
|
| 437 |
|
| 438 | smp_mb__before_atomic();
|
| 439 | for (i = 0; i < SBQ_WAIT_QUEUES; i++)
|
| 440 | atomic_set(&sbq->ws[i].wait_cnt, 1);
|
| 441 | }
|
| 442 | }
|
| 443 |
|
| 444 | void sbitmap_queue_resize(struct sbitmap_queue *sbq, unsigned int depth)
|
| 445 | {
|
| 446 | sbitmap_queue_update_wake_batch(sbq, depth);
|
| 447 | sbitmap_resize(&sbq->sb, depth);
|
| 448 | }
|
| 449 | EXPORT_SYMBOL_GPL(sbitmap_queue_resize);
|
| 450 |
|
| 451 | int __sbitmap_queue_get(struct sbitmap_queue *sbq)
|
| 452 | {
|
| 453 | unsigned int hint, depth;
|
| 454 | int nr;
|
| 455 |
|
| 456 | hint = this_cpu_read(*sbq->alloc_hint);
|
| 457 | depth = READ_ONCE(sbq->sb.depth);
|
| 458 | if (unlikely(hint >= depth)) {
|
| 459 | hint = depth ? prandom_u32() % depth : 0;
|
| 460 | this_cpu_write(*sbq->alloc_hint, hint);
|
| 461 | }
|
| 462 | nr = sbitmap_get(&sbq->sb, hint, sbq->round_robin);
|
| 463 |
|
| 464 | if (nr == -1) {
|
| 465 |
|
| 466 | this_cpu_write(*sbq->alloc_hint, 0);
|
| 467 | } else if (nr == hint || unlikely(sbq->round_robin)) {
|
| 468 |
|
| 469 | hint = nr + 1;
|
| 470 | if (hint >= depth - 1)
|
| 471 | hint = 0;
|
| 472 | this_cpu_write(*sbq->alloc_hint, hint);
|
| 473 | }
|
| 474 |
|
| 475 | return nr;
|
| 476 | }
|
| 477 | EXPORT_SYMBOL_GPL(__sbitmap_queue_get);
|
| 478 |
|
| 479 | int __sbitmap_queue_get_shallow(struct sbitmap_queue *sbq,
|
| 480 | unsigned int shallow_depth)
|
| 481 | {
|
| 482 | unsigned int hint, depth;
|
| 483 | int nr;
|
| 484 |
|
| 485 | WARN_ON_ONCE(shallow_depth < sbq->min_shallow_depth);
|
| 486 |
|
| 487 | hint = this_cpu_read(*sbq->alloc_hint);
|
| 488 | depth = READ_ONCE(sbq->sb.depth);
|
| 489 | if (unlikely(hint >= depth)) {
|
| 490 | hint = depth ? prandom_u32() % depth : 0;
|
| 491 | this_cpu_write(*sbq->alloc_hint, hint);
|
| 492 | }
|
| 493 | nr = sbitmap_get_shallow(&sbq->sb, hint, shallow_depth);
|
| 494 |
|
| 495 | if (nr == -1) {
|
| 496 |
|
| 497 | this_cpu_write(*sbq->alloc_hint, 0);
|
| 498 | } else if (nr == hint || unlikely(sbq->round_robin)) {
|
| 499 |
|
| 500 | hint = nr + 1;
|
| 501 | if (hint >= depth - 1)
|
| 502 | hint = 0;
|
| 503 | this_cpu_write(*sbq->alloc_hint, hint);
|
| 504 | }
|
| 505 |
|
| 506 | return nr;
|
| 507 | }
|
| 508 | EXPORT_SYMBOL_GPL(__sbitmap_queue_get_shallow);
|
| 509 |
|
| 510 | void sbitmap_queue_min_shallow_depth(struct sbitmap_queue *sbq,
|
| 511 | unsigned int min_shallow_depth)
|
| 512 | {
|
| 513 | sbq->min_shallow_depth = min_shallow_depth;
|
| 514 | sbitmap_queue_update_wake_batch(sbq, sbq->sb.depth);
|
| 515 | }
|
| 516 | EXPORT_SYMBOL_GPL(sbitmap_queue_min_shallow_depth);
|
| 517 |
|
| 518 | static struct sbq_wait_state *sbq_wake_ptr(struct sbitmap_queue *sbq)
|
| 519 | {
|
| 520 | int i, wake_index;
|
| 521 |
|
| 522 | if (!atomic_read(&sbq->ws_active))
|
| 523 | return NULL;
|
| 524 |
|
| 525 | wake_index = atomic_read(&sbq->wake_index);
|
| 526 | for (i = 0; i < SBQ_WAIT_QUEUES; i++) {
|
| 527 | struct sbq_wait_state *ws = &sbq->ws[wake_index];
|
| 528 |
|
| 529 | if (waitqueue_active(&ws->wait)) {
|
| 530 | int o = atomic_read(&sbq->wake_index);
|
| 531 |
|
| 532 | if (wake_index != o)
|
| 533 | atomic_cmpxchg(&sbq->wake_index, o, wake_index);
|
| 534 | return ws;
|
| 535 | }
|
| 536 |
|
| 537 | wake_index = sbq_index_inc(wake_index);
|
| 538 | }
|
| 539 |
|
| 540 | return NULL;
|
| 541 | }
|
| 542 |
|
| 543 | static bool __sbq_wake_up(struct sbitmap_queue *sbq)
|
| 544 | {
|
| 545 | struct sbq_wait_state *ws;
|
| 546 | unsigned int wake_batch;
|
| 547 | int wait_cnt;
|
| 548 |
|
| 549 | ws = sbq_wake_ptr(sbq);
|
| 550 | if (!ws)
|
| 551 | return false;
|
| 552 |
|
| 553 | wait_cnt = atomic_dec_return(&ws->wait_cnt);
|
| 554 | if (wait_cnt <= 0) {
|
| 555 | int ret;
|
| 556 |
|
| 557 | wake_batch = READ_ONCE(sbq->wake_batch);
|
| 558 |
|
| 559 | |
| 560 | * Pairs with the memory barrier in sbitmap_queue_resize() to
|
| 561 | * ensure that we see the batch size update before the wait
|
| 562 | * count is reset.
|
| 563 |
|
| 564 | smp_mb__before_atomic();
|
| 565 |
|
| 566 | |
| 567 | * For concurrent callers of this, the one that failed the
|
| 568 | * atomic_cmpxhcg() race should call this function again
|
| 569 | * to wakeup a new batch on a different 'ws'.
|
| 570 |
|
| 571 | ret = atomic_cmpxchg(&ws->wait_cnt, wait_cnt, wake_batch);
|
| 572 | if (ret == wait_cnt) {
|
| 573 | sbq_index_atomic_inc(&sbq->wake_index);
|
| 574 | wake_up_nr(&ws->wait, wake_batch);
|
| 575 | return false;
|
| 576 | }
|
| 577 |
|
| 578 | return true;
|
| 579 | }
|
| 580 |
|
| 581 | return false;
|
| 582 | }
|
| 583 |
|
| 584 | void sbitmap_queue_wake_up(struct sbitmap_queue *sbq)
|
| 585 | {
|
| 586 | while (__sbq_wake_up(sbq))
|
| 587 | ;
|
| 588 | }
|
| 589 | EXPORT_SYMBOL_GPL(sbitmap_queue_wake_up);
|
| 590 |
|
| 591 | void sbitmap_queue_clear(struct sbitmap_queue *sbq, unsigned int nr,
|
| 592 | unsigned int cpu)
|
| 593 | {
|
| 594 | |
| 595 | * Once the clear bit is set, the bit may be allocated out.
|
| 596 | *
|
| 597 | * Orders READ/WRITE on the asssociated instance(such as request
|
| 598 | * of blk_mq) by this bit for avoiding race with re-allocation,
|
| 599 | * and its pair is the memory barrier implied in __sbitmap_get_word.
|
| 600 | *
|
| 601 | * One invariant is that the clear bit has to be zero when the bit
|
| 602 | * is in use.
|
| 603 |
|
| 604 | smp_mb__before_atomic();
|
| 605 | sbitmap_deferred_clear_bit(&sbq->sb, nr);
|
| 606 |
|
| 607 | |
| 608 | * Pairs with the memory barrier in set_current_state() to ensure the
|
| 609 | * proper ordering of clear_bit_unlock()/waitqueue_active() in the waker
|
| 610 | * and test_and_set_bit_lock()/prepare_to_wait()/finish_wait() in the
|
| 611 | * waiter. See the comment on waitqueue_active().
|
| 612 |
|
| 613 | smp_mb__after_atomic();
|
| 614 | sbitmap_queue_wake_up(sbq);
|
| 615 |
|
| 616 | if (likely(!sbq->round_robin && nr < sbq->sb.depth))
|
| 617 | *per_cpu_ptr(sbq->alloc_hint, cpu) = nr;
|
| 618 | }
|
| 619 | EXPORT_SYMBOL_GPL(sbitmap_queue_clear);
|
| 620 |
|
| 621 | void sbitmap_queue_wake_all(struct sbitmap_queue *sbq)
|
| 622 | {
|
| 623 | int i, wake_index;
|
| 624 |
|
| 625 | |
| 626 | * Pairs with the memory barrier in set_current_state() like in
|
| 627 | * sbitmap_queue_wake_up().
|
| 628 |
|
| 629 | smp_mb();
|
| 630 | wake_index = atomic_read(&sbq->wake_index);
|
| 631 | for (i = 0; i < SBQ_WAIT_QUEUES; i++) {
|
| 632 | struct sbq_wait_state *ws = &sbq->ws[wake_index];
|
| 633 |
|
| 634 | if (waitqueue_active(&ws->wait))
|
| 635 | wake_up(&ws->wait);
|
| 636 |
|
| 637 | wake_index = sbq_index_inc(wake_index);
|
| 638 | }
|
| 639 | }
|
| 640 | EXPORT_SYMBOL_GPL(sbitmap_queue_wake_all);
|
| 641 |
|
| 642 | void sbitmap_queue_show(struct sbitmap_queue *sbq, struct seq_file *m)
|
| 643 | {
|
| 644 | bool first;
|
| 645 | int i;
|
| 646 |
|
| 647 | sbitmap_show(&sbq->sb, m);
|
| 648 |
|
| 649 | seq_puts(m, "alloc_hint={");
|
| 650 | first = true;
|
| 651 | for_each_possible_cpu(i) {
|
| 652 | if (!first)
|
| 653 | seq_puts(m, ", ");
|
| 654 | first = false;
|
| 655 | seq_printf(m, "%u", *per_cpu_ptr(sbq->alloc_hint, i));
|
| 656 | }
|
| 657 | seq_puts(m, "}\n");
|
| 658 |
|
| 659 | seq_printf(m, "wake_batch=%u\n", sbq->wake_batch);
|
| 660 | seq_printf(m, "wake_index=%d\n", atomic_read(&sbq->wake_index));
|
| 661 | seq_printf(m, "ws_active=%d\n", atomic_read(&sbq->ws_active));
|
| 662 |
|
| 663 | seq_puts(m, "ws={\n");
|
| 664 | for (i = 0; i < SBQ_WAIT_QUEUES; i++) {
|
| 665 | struct sbq_wait_state *ws = &sbq->ws[i];
|
| 666 |
|
| 667 | seq_printf(m, "\t{.wait_cnt=%d, .wait=%s},\n",
|
| 668 | atomic_read(&ws->wait_cnt),
|
| 669 | waitqueue_active(&ws->wait) ? "active" : "inactive");
|
| 670 | }
|
| 671 | seq_puts(m, "}\n");
|
| 672 |
|
| 673 | seq_printf(m, "round_robin=%d\n", sbq->round_robin);
|
| 674 | seq_printf(m, "min_shallow_depth=%u\n", sbq->min_shallow_depth);
|
| 675 | }
|
| 676 | EXPORT_SYMBOL_GPL(sbitmap_queue_show);
|
| 677 |
|
| 678 | void sbitmap_add_wait_queue(struct sbitmap_queue *sbq,
|
| 679 | struct sbq_wait_state *ws,
|
| 680 | struct sbq_wait *sbq_wait)
|
| 681 | {
|
| 682 | if (!sbq_wait->sbq) {
|
| 683 | sbq_wait->sbq = sbq;
|
| 684 | atomic_inc(&sbq->ws_active);
|
| 685 | }
|
| 686 | add_wait_queue(&ws->wait, &sbq_wait->wait);
|
| 687 | }
|
| 688 | EXPORT_SYMBOL_GPL(sbitmap_add_wait_queue);
|
| 689 |
|
| 690 | void sbitmap_del_wait_queue(struct sbq_wait *sbq_wait)
|
| 691 | {
|
| 692 | list_del_init(&sbq_wait->wait.entry);
|
| 693 | if (sbq_wait->sbq) {
|
| 694 | atomic_dec(&sbq_wait->sbq->ws_active);
|
| 695 | sbq_wait->sbq = NULL;
|
| 696 | }
|
| 697 | }
|
| 698 | EXPORT_SYMBOL_GPL(sbitmap_del_wait_queue);
|
| 699 |
|
| 700 | void sbitmap_prepare_to_wait(struct sbitmap_queue *sbq,
|
| 701 | struct sbq_wait_state *ws,
|
| 702 | struct sbq_wait *sbq_wait, int state)
|
| 703 | {
|
| 704 | if (!sbq_wait->sbq) {
|
| 705 | atomic_inc(&sbq->ws_active);
|
| 706 | sbq_wait->sbq = sbq;
|
| 707 | }
|
| 708 | prepare_to_wait_exclusive(&ws->wait, &sbq_wait->wait, state);
|
| 709 | }
|
| 710 | EXPORT_SYMBOL_GPL(sbitmap_prepare_to_wait);
|
| 711 |
|
| 712 | void sbitmap_finish_wait(struct sbitmap_queue *sbq, struct sbq_wait_state *ws,
|
| 713 | struct sbq_wait *sbq_wait)
|
| 714 | {
|
| 715 | finish_wait(&ws->wait, &sbq_wait->wait);
|
| 716 | if (sbq_wait->sbq) {
|
| 717 | atomic_dec(&sbq->ws_active);
|
| 718 | sbq_wait->sbq = NULL;
|
| 719 | }
|
| 720 | }
|
| 721 | EXPORT_SYMBOL_GPL(sbitmap_finish_wait);
|