Line data Source code
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 0 : static int init_alloc_hint(struct sbitmap *sb, gfp_t flags)
13 : {
14 0 : unsigned depth = sb->depth;
15 :
16 0 : sb->alloc_hint = alloc_percpu_gfp(unsigned int, flags);
17 0 : if (!sb->alloc_hint)
18 : return -ENOMEM;
19 :
20 0 : if (depth && !sb->round_robin) {
21 : int i;
22 :
23 0 : for_each_possible_cpu(i)
24 0 : *per_cpu_ptr(sb->alloc_hint, i) = prandom_u32() % depth;
25 : }
26 : return 0;
27 : }
28 :
29 0 : static inline unsigned update_alloc_hint_before_get(struct sbitmap *sb,
30 : unsigned int depth)
31 : {
32 : unsigned hint;
33 :
34 0 : hint = this_cpu_read(*sb->alloc_hint);
35 0 : if (unlikely(hint >= depth)) {
36 0 : hint = depth ? prandom_u32() % depth : 0;
37 0 : this_cpu_write(*sb->alloc_hint, hint);
38 : }
39 :
40 0 : return hint;
41 : }
42 :
43 0 : 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 0 : if (nr == -1) {
49 : /* If the map is full, a hint won't do us much good. */
50 0 : this_cpu_write(*sb->alloc_hint, 0);
51 0 : } else if (nr == hint || unlikely(sb->round_robin)) {
52 : /* Only update the hint if we used it. */
53 0 : hint = nr + 1;
54 0 : if (hint >= depth - 1)
55 0 : hint = 0;
56 0 : this_cpu_write(*sb->alloc_hint, hint);
57 : }
58 0 : }
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 : {
65 : unsigned long mask;
66 :
67 0 : if (!READ_ONCE(map->cleared))
68 : return false;
69 :
70 : /*
71 : * First get a stable cleared mask, setting the old mask to 0.
72 : */
73 0 : mask = xchg(&map->cleared, 0);
74 :
75 : /*
76 : * Now clear the masked bits in our free word
77 : */
78 0 : atomic_long_andnot(mask, (atomic_long_t *)&map->word);
79 : BUILD_BUG_ON(sizeof(atomic_long_t) != sizeof(map->word));
80 : return true;
81 : }
82 :
83 0 : int sbitmap_init_node(struct sbitmap *sb, unsigned int depth, int shift,
84 : gfp_t flags, int node, bool round_robin,
85 : bool alloc_hint)
86 : {
87 : unsigned int bits_per_word;
88 :
89 0 : if (shift < 0)
90 : shift = sbitmap_calculate_shift(depth);
91 :
92 0 : bits_per_word = 1U << shift;
93 0 : if (bits_per_word > BITS_PER_LONG)
94 : return -EINVAL;
95 :
96 0 : sb->shift = shift;
97 0 : sb->depth = depth;
98 0 : sb->map_nr = DIV_ROUND_UP(sb->depth, bits_per_word);
99 0 : sb->round_robin = round_robin;
100 :
101 0 : if (depth == 0) {
102 0 : sb->map = NULL;
103 0 : return 0;
104 : }
105 :
106 0 : if (alloc_hint) {
107 0 : if (init_alloc_hint(sb, flags))
108 : return -ENOMEM;
109 : } else {
110 0 : sb->alloc_hint = NULL;
111 : }
112 :
113 0 : sb->map = kvzalloc_node(sb->map_nr * sizeof(*sb->map), flags, node);
114 0 : if (!sb->map) {
115 0 : free_percpu(sb->alloc_hint);
116 0 : return -ENOMEM;
117 : }
118 :
119 : return 0;
120 : }
121 : EXPORT_SYMBOL_GPL(sbitmap_init_node);
122 :
123 0 : void sbitmap_resize(struct sbitmap *sb, unsigned int depth)
124 : {
125 0 : unsigned int bits_per_word = 1U << sb->shift;
126 : unsigned int i;
127 :
128 0 : for (i = 0; i < sb->map_nr; i++)
129 0 : sbitmap_deferred_clear(&sb->map[i]);
130 :
131 0 : sb->depth = depth;
132 0 : sb->map_nr = DIV_ROUND_UP(sb->depth, bits_per_word);
133 0 : }
134 : EXPORT_SYMBOL_GPL(sbitmap_resize);
135 :
136 0 : static int __sbitmap_get_word(unsigned long *word, unsigned long depth,
137 : unsigned int hint, bool wrap)
138 : {
139 : int nr;
140 :
141 : /* don't wrap if starting from 0 */
142 0 : wrap = wrap && hint;
143 :
144 : while (1) {
145 0 : nr = find_next_zero_bit(word, depth, hint);
146 0 : if (unlikely(nr >= depth)) {
147 : /*
148 : * We started with an offset, and we didn't reset the
149 : * offset to 0 in a failure case, so start from 0 to
150 : * exhaust the map.
151 : */
152 0 : if (hint && wrap) {
153 0 : hint = 0;
154 0 : continue;
155 : }
156 : return -1;
157 : }
158 :
159 0 : if (!test_and_set_bit_lock(nr, word))
160 : break;
161 :
162 0 : hint = nr + 1;
163 0 : if (hint >= depth - 1)
164 0 : hint = 0;
165 : }
166 :
167 : return nr;
168 : }
169 :
170 0 : static int sbitmap_find_bit_in_index(struct sbitmap *sb, int index,
171 : unsigned int alloc_hint)
172 : {
173 0 : struct sbitmap_word *map = &sb->map[index];
174 : int nr;
175 :
176 : do {
177 0 : nr = __sbitmap_get_word(&map->word, __map_depth(sb, index),
178 0 : alloc_hint, !sb->round_robin);
179 0 : if (nr != -1)
180 : break;
181 0 : if (!sbitmap_deferred_clear(map))
182 : break;
183 : } while (1);
184 :
185 0 : return nr;
186 : }
187 :
188 0 : static int __sbitmap_get(struct sbitmap *sb, unsigned int alloc_hint)
189 : {
190 : unsigned int i, index;
191 0 : int nr = -1;
192 :
193 0 : index = SB_NR_TO_INDEX(sb, alloc_hint);
194 :
195 : /*
196 : * Unless we're doing round robin tag allocation, just use the
197 : * alloc_hint to find the right word index. No point in looping
198 : * twice in find_next_zero_bit() for that case.
199 : */
200 0 : if (sb->round_robin)
201 0 : alloc_hint = SB_NR_TO_BIT(sb, alloc_hint);
202 : else
203 : alloc_hint = 0;
204 :
205 0 : for (i = 0; i < sb->map_nr; i++) {
206 0 : nr = sbitmap_find_bit_in_index(sb, index, alloc_hint);
207 0 : if (nr != -1) {
208 0 : nr += index << sb->shift;
209 0 : break;
210 : }
211 :
212 : /* Jump to next index. */
213 0 : alloc_hint = 0;
214 0 : if (++index >= sb->map_nr)
215 0 : index = 0;
216 : }
217 :
218 0 : return nr;
219 : }
220 :
221 0 : int sbitmap_get(struct sbitmap *sb)
222 : {
223 : int nr;
224 : unsigned int hint, depth;
225 :
226 0 : if (WARN_ON_ONCE(unlikely(!sb->alloc_hint)))
227 : return -1;
228 :
229 0 : depth = READ_ONCE(sb->depth);
230 0 : hint = update_alloc_hint_before_get(sb, depth);
231 0 : nr = __sbitmap_get(sb, hint);
232 0 : update_alloc_hint_after_get(sb, depth, hint, nr);
233 :
234 0 : return nr;
235 : }
236 : EXPORT_SYMBOL_GPL(sbitmap_get);
237 :
238 0 : static int __sbitmap_get_shallow(struct sbitmap *sb,
239 : unsigned int alloc_hint,
240 : unsigned long shallow_depth)
241 : {
242 : unsigned int i, index;
243 0 : int nr = -1;
244 :
245 0 : index = SB_NR_TO_INDEX(sb, alloc_hint);
246 :
247 0 : for (i = 0; i < sb->map_nr; i++) {
248 : again:
249 0 : nr = __sbitmap_get_word(&sb->map[index].word,
250 0 : min_t(unsigned int,
251 : __map_depth(sb, index),
252 : shallow_depth),
253 0 : SB_NR_TO_BIT(sb, alloc_hint), true);
254 0 : if (nr != -1) {
255 0 : nr += index << sb->shift;
256 0 : break;
257 : }
258 :
259 0 : if (sbitmap_deferred_clear(&sb->map[index]))
260 : goto again;
261 :
262 : /* Jump to next index. */
263 0 : index++;
264 0 : alloc_hint = index << sb->shift;
265 :
266 0 : if (index >= sb->map_nr) {
267 0 : index = 0;
268 0 : alloc_hint = 0;
269 : }
270 : }
271 :
272 0 : return nr;
273 : }
274 :
275 0 : int sbitmap_get_shallow(struct sbitmap *sb, unsigned long shallow_depth)
276 : {
277 : int nr;
278 : unsigned int hint, depth;
279 :
280 0 : if (WARN_ON_ONCE(unlikely(!sb->alloc_hint)))
281 : return -1;
282 :
283 0 : depth = READ_ONCE(sb->depth);
284 0 : hint = update_alloc_hint_before_get(sb, depth);
285 0 : nr = __sbitmap_get_shallow(sb, hint, shallow_depth);
286 0 : update_alloc_hint_after_get(sb, depth, hint, nr);
287 :
288 0 : return nr;
289 : }
290 : EXPORT_SYMBOL_GPL(sbitmap_get_shallow);
291 :
292 0 : bool sbitmap_any_bit_set(const struct sbitmap *sb)
293 : {
294 : unsigned int i;
295 :
296 0 : for (i = 0; i < sb->map_nr; i++) {
297 0 : if (sb->map[i].word & ~sb->map[i].cleared)
298 : return true;
299 : }
300 : return false;
301 : }
302 : EXPORT_SYMBOL_GPL(sbitmap_any_bit_set);
303 :
304 0 : static unsigned int __sbitmap_weight(const struct sbitmap *sb, bool set)
305 : {
306 0 : unsigned int i, weight = 0;
307 :
308 0 : for (i = 0; i < sb->map_nr; i++) {
309 0 : const struct sbitmap_word *word = &sb->map[i];
310 0 : unsigned int word_depth = __map_depth(sb, i);
311 :
312 0 : if (set)
313 0 : weight += bitmap_weight(&word->word, word_depth);
314 : else
315 0 : weight += bitmap_weight(&word->cleared, word_depth);
316 : }
317 0 : return weight;
318 : }
319 :
320 : static unsigned int sbitmap_cleared(const struct sbitmap *sb)
321 : {
322 0 : return __sbitmap_weight(sb, false);
323 : }
324 :
325 0 : unsigned int sbitmap_weight(const struct sbitmap *sb)
326 : {
327 0 : return __sbitmap_weight(sb, true) - sbitmap_cleared(sb);
328 : }
329 : EXPORT_SYMBOL_GPL(sbitmap_weight);
330 :
331 0 : void sbitmap_show(struct sbitmap *sb, struct seq_file *m)
332 : {
333 0 : seq_printf(m, "depth=%u\n", sb->depth);
334 0 : seq_printf(m, "busy=%u\n", sbitmap_weight(sb));
335 0 : seq_printf(m, "cleared=%u\n", sbitmap_cleared(sb));
336 0 : seq_printf(m, "bits_per_word=%u\n", 1U << sb->shift);
337 0 : seq_printf(m, "map_nr=%u\n", sb->map_nr);
338 0 : }
339 : EXPORT_SYMBOL_GPL(sbitmap_show);
340 :
341 0 : static inline void emit_byte(struct seq_file *m, unsigned int offset, u8 byte)
342 : {
343 0 : if ((offset & 0xf) == 0) {
344 0 : if (offset != 0)
345 0 : seq_putc(m, '\n');
346 0 : seq_printf(m, "%08x:", offset);
347 : }
348 0 : if ((offset & 0x1) == 0)
349 0 : seq_putc(m, ' ');
350 0 : seq_printf(m, "%02x", byte);
351 0 : }
352 :
353 0 : void sbitmap_bitmap_show(struct sbitmap *sb, struct seq_file *m)
354 : {
355 0 : u8 byte = 0;
356 0 : unsigned int byte_bits = 0;
357 0 : unsigned int offset = 0;
358 : int i;
359 :
360 0 : for (i = 0; i < sb->map_nr; i++) {
361 0 : unsigned long word = READ_ONCE(sb->map[i].word);
362 0 : unsigned long cleared = READ_ONCE(sb->map[i].cleared);
363 0 : unsigned int word_bits = __map_depth(sb, i);
364 :
365 0 : word &= ~cleared;
366 :
367 0 : while (word_bits > 0) {
368 0 : unsigned int bits = min(8 - byte_bits, word_bits);
369 :
370 0 : byte |= (word & (BIT(bits) - 1)) << byte_bits;
371 0 : byte_bits += bits;
372 0 : if (byte_bits == 8) {
373 0 : emit_byte(m, offset, byte);
374 0 : byte = 0;
375 0 : byte_bits = 0;
376 0 : offset++;
377 : }
378 0 : word >>= bits;
379 0 : word_bits -= bits;
380 : }
381 : }
382 0 : if (byte_bits) {
383 0 : emit_byte(m, offset, byte);
384 0 : offset++;
385 : }
386 0 : if (offset)
387 0 : seq_putc(m, '\n');
388 0 : }
389 : EXPORT_SYMBOL_GPL(sbitmap_bitmap_show);
390 :
391 : static unsigned int sbq_calc_wake_batch(struct sbitmap_queue *sbq,
392 : unsigned int depth)
393 : {
394 : unsigned int wake_batch;
395 : unsigned int shallow_depth;
396 :
397 : /*
398 : * For each batch, we wake up one queue. We need to make sure that our
399 : * batch size is small enough that the full depth of the bitmap,
400 : * potentially limited by a shallow depth, is enough to wake up all of
401 : * the queues.
402 : *
403 : * Each full word of the bitmap has bits_per_word bits, and there might
404 : * be a partial word. There are depth / bits_per_word full words and
405 : * depth % bits_per_word bits left over. In bitwise arithmetic:
406 : *
407 : * bits_per_word = 1 << shift
408 : * depth / bits_per_word = depth >> shift
409 : * depth % bits_per_word = depth & ((1 << shift) - 1)
410 : *
411 : * Each word can be limited to sbq->min_shallow_depth bits.
412 : */
413 0 : shallow_depth = min(1U << sbq->sb.shift, sbq->min_shallow_depth);
414 0 : depth = ((depth >> sbq->sb.shift) * shallow_depth +
415 0 : min(depth & ((1U << sbq->sb.shift) - 1), shallow_depth));
416 0 : wake_batch = clamp_t(unsigned int, depth / SBQ_WAIT_QUEUES, 1,
417 : SBQ_WAKE_BATCH);
418 :
419 : return wake_batch;
420 : }
421 :
422 0 : int sbitmap_queue_init_node(struct sbitmap_queue *sbq, unsigned int depth,
423 : int shift, bool round_robin, gfp_t flags, int node)
424 : {
425 : int ret;
426 : int i;
427 :
428 0 : ret = sbitmap_init_node(&sbq->sb, depth, shift, flags, node,
429 : round_robin, true);
430 0 : if (ret)
431 : return ret;
432 :
433 0 : sbq->min_shallow_depth = UINT_MAX;
434 0 : sbq->wake_batch = sbq_calc_wake_batch(sbq, depth);
435 0 : atomic_set(&sbq->wake_index, 0);
436 0 : atomic_set(&sbq->ws_active, 0);
437 :
438 0 : sbq->ws = kzalloc_node(SBQ_WAIT_QUEUES * sizeof(*sbq->ws), flags, node);
439 0 : if (!sbq->ws) {
440 0 : sbitmap_free(&sbq->sb);
441 0 : return -ENOMEM;
442 : }
443 :
444 0 : for (i = 0; i < SBQ_WAIT_QUEUES; i++) {
445 0 : init_waitqueue_head(&sbq->ws[i].wait);
446 0 : atomic_set(&sbq->ws[i].wait_cnt, sbq->wake_batch);
447 : }
448 :
449 : return 0;
450 : }
451 : EXPORT_SYMBOL_GPL(sbitmap_queue_init_node);
452 :
453 : static inline void __sbitmap_queue_update_wake_batch(struct sbitmap_queue *sbq,
454 : unsigned int wake_batch)
455 : {
456 : int i;
457 :
458 0 : if (sbq->wake_batch != wake_batch) {
459 0 : WRITE_ONCE(sbq->wake_batch, wake_batch);
460 : /*
461 : * Pairs with the memory barrier in sbitmap_queue_wake_up()
462 : * to ensure that the batch size is updated before the wait
463 : * counts.
464 : */
465 0 : smp_mb();
466 0 : for (i = 0; i < SBQ_WAIT_QUEUES; i++)
467 0 : atomic_set(&sbq->ws[i].wait_cnt, 1);
468 : }
469 : }
470 :
471 0 : static void sbitmap_queue_update_wake_batch(struct sbitmap_queue *sbq,
472 : unsigned int depth)
473 : {
474 : unsigned int wake_batch;
475 :
476 0 : wake_batch = sbq_calc_wake_batch(sbq, depth);
477 0 : __sbitmap_queue_update_wake_batch(sbq, wake_batch);
478 0 : }
479 :
480 0 : void sbitmap_queue_recalculate_wake_batch(struct sbitmap_queue *sbq,
481 : unsigned int users)
482 : {
483 : unsigned int wake_batch;
484 : unsigned int min_batch;
485 0 : unsigned int depth = (sbq->sb.depth + users - 1) / users;
486 :
487 0 : min_batch = sbq->sb.depth >= (4 * SBQ_WAIT_QUEUES) ? 4 : 1;
488 :
489 0 : wake_batch = clamp_val(depth / SBQ_WAIT_QUEUES,
490 : min_batch, SBQ_WAKE_BATCH);
491 0 : __sbitmap_queue_update_wake_batch(sbq, wake_batch);
492 0 : }
493 : EXPORT_SYMBOL_GPL(sbitmap_queue_recalculate_wake_batch);
494 :
495 0 : void sbitmap_queue_resize(struct sbitmap_queue *sbq, unsigned int depth)
496 : {
497 0 : sbitmap_queue_update_wake_batch(sbq, depth);
498 0 : sbitmap_resize(&sbq->sb, depth);
499 0 : }
500 : EXPORT_SYMBOL_GPL(sbitmap_queue_resize);
501 :
502 0 : int __sbitmap_queue_get(struct sbitmap_queue *sbq)
503 : {
504 0 : return sbitmap_get(&sbq->sb);
505 : }
506 : EXPORT_SYMBOL_GPL(__sbitmap_queue_get);
507 :
508 0 : unsigned long __sbitmap_queue_get_batch(struct sbitmap_queue *sbq, int nr_tags,
509 : unsigned int *offset)
510 : {
511 0 : struct sbitmap *sb = &sbq->sb;
512 : unsigned int hint, depth;
513 : unsigned long index, nr;
514 : int i;
515 :
516 0 : if (unlikely(sb->round_robin))
517 : return 0;
518 :
519 0 : depth = READ_ONCE(sb->depth);
520 0 : hint = update_alloc_hint_before_get(sb, depth);
521 :
522 0 : index = SB_NR_TO_INDEX(sb, hint);
523 :
524 0 : for (i = 0; i < sb->map_nr; i++) {
525 0 : struct sbitmap_word *map = &sb->map[index];
526 : unsigned long get_mask;
527 0 : unsigned int map_depth = __map_depth(sb, index);
528 :
529 0 : sbitmap_deferred_clear(map);
530 0 : if (map->word == (1UL << (map_depth - 1)) - 1)
531 0 : continue;
532 :
533 0 : nr = find_first_zero_bit(&map->word, map_depth);
534 0 : if (nr + nr_tags <= map_depth) {
535 0 : atomic_long_t *ptr = (atomic_long_t *) &map->word;
536 0 : int map_tags = min_t(int, nr_tags, map_depth);
537 : unsigned long val, ret;
538 :
539 0 : get_mask = ((1UL << map_tags) - 1) << nr;
540 : do {
541 0 : val = READ_ONCE(map->word);
542 0 : ret = atomic_long_cmpxchg(ptr, val, get_mask | val);
543 0 : } while (ret != val);
544 0 : get_mask = (get_mask & ~ret) >> nr;
545 0 : if (get_mask) {
546 0 : *offset = nr + (index << sb->shift);
547 0 : update_alloc_hint_after_get(sb, depth, hint,
548 0 : *offset + map_tags - 1);
549 0 : return get_mask;
550 : }
551 : }
552 : /* Jump to next index. */
553 0 : if (++index >= sb->map_nr)
554 0 : index = 0;
555 : }
556 :
557 : return 0;
558 : }
559 :
560 0 : int sbitmap_queue_get_shallow(struct sbitmap_queue *sbq,
561 : unsigned int shallow_depth)
562 : {
563 0 : WARN_ON_ONCE(shallow_depth < sbq->min_shallow_depth);
564 :
565 0 : return sbitmap_get_shallow(&sbq->sb, shallow_depth);
566 : }
567 : EXPORT_SYMBOL_GPL(sbitmap_queue_get_shallow);
568 :
569 0 : void sbitmap_queue_min_shallow_depth(struct sbitmap_queue *sbq,
570 : unsigned int min_shallow_depth)
571 : {
572 0 : sbq->min_shallow_depth = min_shallow_depth;
573 0 : sbitmap_queue_update_wake_batch(sbq, sbq->sb.depth);
574 0 : }
575 : EXPORT_SYMBOL_GPL(sbitmap_queue_min_shallow_depth);
576 :
577 : static struct sbq_wait_state *sbq_wake_ptr(struct sbitmap_queue *sbq)
578 : {
579 : int i, wake_index;
580 :
581 0 : if (!atomic_read(&sbq->ws_active))
582 : return NULL;
583 :
584 0 : wake_index = atomic_read(&sbq->wake_index);
585 0 : for (i = 0; i < SBQ_WAIT_QUEUES; i++) {
586 0 : struct sbq_wait_state *ws = &sbq->ws[wake_index];
587 :
588 0 : if (waitqueue_active(&ws->wait)) {
589 0 : if (wake_index != atomic_read(&sbq->wake_index))
590 0 : atomic_set(&sbq->wake_index, wake_index);
591 : return ws;
592 : }
593 :
594 0 : wake_index = sbq_index_inc(wake_index);
595 : }
596 :
597 : return NULL;
598 : }
599 :
600 0 : static bool __sbq_wake_up(struct sbitmap_queue *sbq)
601 : {
602 : struct sbq_wait_state *ws;
603 : unsigned int wake_batch;
604 : int wait_cnt;
605 :
606 0 : ws = sbq_wake_ptr(sbq);
607 0 : if (!ws)
608 : return false;
609 :
610 0 : wait_cnt = atomic_dec_return(&ws->wait_cnt);
611 0 : if (wait_cnt <= 0) {
612 : int ret;
613 :
614 0 : wake_batch = READ_ONCE(sbq->wake_batch);
615 :
616 : /*
617 : * Pairs with the memory barrier in sbitmap_queue_resize() to
618 : * ensure that we see the batch size update before the wait
619 : * count is reset.
620 : */
621 0 : smp_mb__before_atomic();
622 :
623 : /*
624 : * For concurrent callers of this, the one that failed the
625 : * atomic_cmpxhcg() race should call this function again
626 : * to wakeup a new batch on a different 'ws'.
627 : */
628 0 : ret = atomic_cmpxchg(&ws->wait_cnt, wait_cnt, wake_batch);
629 0 : if (ret == wait_cnt) {
630 0 : sbq_index_atomic_inc(&sbq->wake_index);
631 0 : wake_up_nr(&ws->wait, wake_batch);
632 0 : return false;
633 : }
634 :
635 : return true;
636 : }
637 :
638 : return false;
639 : }
640 :
641 0 : void sbitmap_queue_wake_up(struct sbitmap_queue *sbq)
642 : {
643 0 : while (__sbq_wake_up(sbq))
644 : ;
645 0 : }
646 : EXPORT_SYMBOL_GPL(sbitmap_queue_wake_up);
647 :
648 : static inline void sbitmap_update_cpu_hint(struct sbitmap *sb, int cpu, int tag)
649 : {
650 0 : if (likely(!sb->round_robin && tag < sb->depth))
651 0 : data_race(*per_cpu_ptr(sb->alloc_hint, cpu) = tag);
652 : }
653 :
654 0 : void sbitmap_queue_clear_batch(struct sbitmap_queue *sbq, int offset,
655 : int *tags, int nr_tags)
656 : {
657 0 : struct sbitmap *sb = &sbq->sb;
658 0 : unsigned long *addr = NULL;
659 0 : unsigned long mask = 0;
660 : int i;
661 :
662 0 : smp_mb__before_atomic();
663 0 : for (i = 0; i < nr_tags; i++) {
664 0 : const int tag = tags[i] - offset;
665 : unsigned long *this_addr;
666 :
667 : /* since we're clearing a batch, skip the deferred map */
668 0 : this_addr = &sb->map[SB_NR_TO_INDEX(sb, tag)].word;
669 0 : if (!addr) {
670 : addr = this_addr;
671 0 : } else if (addr != this_addr) {
672 0 : atomic_long_andnot(mask, (atomic_long_t *) addr);
673 0 : mask = 0;
674 0 : addr = this_addr;
675 : }
676 0 : mask |= (1UL << SB_NR_TO_BIT(sb, tag));
677 : }
678 :
679 0 : if (mask)
680 0 : atomic_long_andnot(mask, (atomic_long_t *) addr);
681 :
682 0 : smp_mb__after_atomic();
683 0 : sbitmap_queue_wake_up(sbq);
684 0 : sbitmap_update_cpu_hint(&sbq->sb, raw_smp_processor_id(),
685 0 : tags[nr_tags - 1] - offset);
686 0 : }
687 :
688 0 : void sbitmap_queue_clear(struct sbitmap_queue *sbq, unsigned int nr,
689 : unsigned int cpu)
690 : {
691 : /*
692 : * Once the clear bit is set, the bit may be allocated out.
693 : *
694 : * Orders READ/WRITE on the associated instance(such as request
695 : * of blk_mq) by this bit for avoiding race with re-allocation,
696 : * and its pair is the memory barrier implied in __sbitmap_get_word.
697 : *
698 : * One invariant is that the clear bit has to be zero when the bit
699 : * is in use.
700 : */
701 0 : smp_mb__before_atomic();
702 0 : sbitmap_deferred_clear_bit(&sbq->sb, nr);
703 :
704 : /*
705 : * Pairs with the memory barrier in set_current_state() to ensure the
706 : * proper ordering of clear_bit_unlock()/waitqueue_active() in the waker
707 : * and test_and_set_bit_lock()/prepare_to_wait()/finish_wait() in the
708 : * waiter. See the comment on waitqueue_active().
709 : */
710 0 : smp_mb__after_atomic();
711 0 : sbitmap_queue_wake_up(sbq);
712 0 : sbitmap_update_cpu_hint(&sbq->sb, cpu, nr);
713 0 : }
714 : EXPORT_SYMBOL_GPL(sbitmap_queue_clear);
715 :
716 0 : void sbitmap_queue_wake_all(struct sbitmap_queue *sbq)
717 : {
718 : int i, wake_index;
719 :
720 : /*
721 : * Pairs with the memory barrier in set_current_state() like in
722 : * sbitmap_queue_wake_up().
723 : */
724 0 : smp_mb();
725 0 : wake_index = atomic_read(&sbq->wake_index);
726 0 : for (i = 0; i < SBQ_WAIT_QUEUES; i++) {
727 0 : struct sbq_wait_state *ws = &sbq->ws[wake_index];
728 :
729 0 : if (waitqueue_active(&ws->wait))
730 0 : wake_up(&ws->wait);
731 :
732 0 : wake_index = sbq_index_inc(wake_index);
733 : }
734 0 : }
735 : EXPORT_SYMBOL_GPL(sbitmap_queue_wake_all);
736 :
737 0 : void sbitmap_queue_show(struct sbitmap_queue *sbq, struct seq_file *m)
738 : {
739 : bool first;
740 : int i;
741 :
742 0 : sbitmap_show(&sbq->sb, m);
743 :
744 0 : seq_puts(m, "alloc_hint={");
745 0 : first = true;
746 0 : for_each_possible_cpu(i) {
747 0 : if (!first)
748 0 : seq_puts(m, ", ");
749 0 : first = false;
750 0 : seq_printf(m, "%u", *per_cpu_ptr(sbq->sb.alloc_hint, i));
751 : }
752 0 : seq_puts(m, "}\n");
753 :
754 0 : seq_printf(m, "wake_batch=%u\n", sbq->wake_batch);
755 0 : seq_printf(m, "wake_index=%d\n", atomic_read(&sbq->wake_index));
756 0 : seq_printf(m, "ws_active=%d\n", atomic_read(&sbq->ws_active));
757 :
758 0 : seq_puts(m, "ws={\n");
759 0 : for (i = 0; i < SBQ_WAIT_QUEUES; i++) {
760 0 : struct sbq_wait_state *ws = &sbq->ws[i];
761 :
762 0 : seq_printf(m, "\t{.wait_cnt=%d, .wait=%s},\n",
763 0 : atomic_read(&ws->wait_cnt),
764 0 : waitqueue_active(&ws->wait) ? "active" : "inactive");
765 : }
766 0 : seq_puts(m, "}\n");
767 :
768 0 : seq_printf(m, "round_robin=%d\n", sbq->sb.round_robin);
769 0 : seq_printf(m, "min_shallow_depth=%u\n", sbq->min_shallow_depth);
770 0 : }
771 : EXPORT_SYMBOL_GPL(sbitmap_queue_show);
772 :
773 0 : void sbitmap_add_wait_queue(struct sbitmap_queue *sbq,
774 : struct sbq_wait_state *ws,
775 : struct sbq_wait *sbq_wait)
776 : {
777 0 : if (!sbq_wait->sbq) {
778 0 : sbq_wait->sbq = sbq;
779 0 : atomic_inc(&sbq->ws_active);
780 0 : add_wait_queue(&ws->wait, &sbq_wait->wait);
781 : }
782 0 : }
783 : EXPORT_SYMBOL_GPL(sbitmap_add_wait_queue);
784 :
785 0 : void sbitmap_del_wait_queue(struct sbq_wait *sbq_wait)
786 : {
787 0 : list_del_init(&sbq_wait->wait.entry);
788 0 : if (sbq_wait->sbq) {
789 0 : atomic_dec(&sbq_wait->sbq->ws_active);
790 0 : sbq_wait->sbq = NULL;
791 : }
792 0 : }
793 : EXPORT_SYMBOL_GPL(sbitmap_del_wait_queue);
794 :
795 0 : void sbitmap_prepare_to_wait(struct sbitmap_queue *sbq,
796 : struct sbq_wait_state *ws,
797 : struct sbq_wait *sbq_wait, int state)
798 : {
799 0 : if (!sbq_wait->sbq) {
800 0 : atomic_inc(&sbq->ws_active);
801 0 : sbq_wait->sbq = sbq;
802 : }
803 0 : prepare_to_wait_exclusive(&ws->wait, &sbq_wait->wait, state);
804 0 : }
805 : EXPORT_SYMBOL_GPL(sbitmap_prepare_to_wait);
806 :
807 0 : void sbitmap_finish_wait(struct sbitmap_queue *sbq, struct sbq_wait_state *ws,
808 : struct sbq_wait *sbq_wait)
809 : {
810 0 : finish_wait(&ws->wait, &sbq_wait->wait);
811 0 : if (sbq_wait->sbq) {
812 0 : atomic_dec(&sbq->ws_active);
813 0 : sbq_wait->sbq = NULL;
814 : }
815 0 : }
816 : EXPORT_SYMBOL_GPL(sbitmap_finish_wait);
|
