1 | /* SPDX-License-Identifier: GPL-2.0 */ |
2 | |
3 | #ifndef _BCACHE_UTIL_H |
4 | #define _BCACHE_UTIL_H |
5 | |
6 | #include <linux/blkdev.h> |
7 | #include <linux/closure.h> |
8 | #include <linux/errno.h> |
9 | #include <linux/kernel.h> |
10 | #include <linux/sched/clock.h> |
11 | #include <linux/llist.h> |
12 | #include <linux/ratelimit.h> |
13 | #include <linux/vmalloc.h> |
14 | #include <linux/workqueue.h> |
15 | #include <linux/crc64.h> |
16 | |
17 | struct closure; |
18 | |
19 | #ifdef CONFIG_BCACHE_DEBUG |
20 | |
21 | #define EBUG_ON(cond) BUG_ON(cond) |
22 | #define atomic_dec_bug(v) BUG_ON(atomic_dec_return(v) < 0) |
23 | #define atomic_inc_bug(v, i) BUG_ON(atomic_inc_return(v) <= i) |
24 | |
25 | #else /* DEBUG */ |
26 | |
27 | #define EBUG_ON(cond) do { if (cond) do {} while (0); } while (0) |
28 | #define atomic_dec_bug(v) atomic_dec(v) |
29 | #define atomic_inc_bug(v, i) atomic_inc(v) |
30 | |
31 | #endif |
32 | |
33 | #define DECLARE_HEAP(type, name) \ |
34 | struct { \ |
35 | size_t size, used; \ |
36 | type *data; \ |
37 | } name |
38 | |
39 | #define init_heap(heap, _size, gfp) \ |
40 | ({ \ |
41 | size_t _bytes; \ |
42 | (heap)->used = 0; \ |
43 | (heap)->size = (_size); \ |
44 | _bytes = (heap)->size * sizeof(*(heap)->data); \ |
45 | (heap)->data = kvmalloc(_bytes, (gfp) & GFP_KERNEL); \ |
46 | (heap)->data; \ |
47 | }) |
48 | |
49 | #define free_heap(heap) \ |
50 | do { \ |
51 | kvfree((heap)->data); \ |
52 | (heap)->data = NULL; \ |
53 | } while (0) |
54 | |
55 | #define heap_swap(h, i, j) swap((h)->data[i], (h)->data[j]) |
56 | |
57 | #define heap_sift(h, i, cmp) \ |
58 | do { \ |
59 | size_t _r, _j = i; \ |
60 | \ |
61 | for (; _j * 2 + 1 < (h)->used; _j = _r) { \ |
62 | _r = _j * 2 + 1; \ |
63 | if (_r + 1 < (h)->used && \ |
64 | cmp((h)->data[_r], (h)->data[_r + 1])) \ |
65 | _r++; \ |
66 | \ |
67 | if (cmp((h)->data[_r], (h)->data[_j])) \ |
68 | break; \ |
69 | heap_swap(h, _r, _j); \ |
70 | } \ |
71 | } while (0) |
72 | |
73 | #define heap_sift_down(h, i, cmp) \ |
74 | do { \ |
75 | while (i) { \ |
76 | size_t p = (i - 1) / 2; \ |
77 | if (cmp((h)->data[i], (h)->data[p])) \ |
78 | break; \ |
79 | heap_swap(h, i, p); \ |
80 | i = p; \ |
81 | } \ |
82 | } while (0) |
83 | |
84 | #define heap_add(h, d, cmp) \ |
85 | ({ \ |
86 | bool _r = !heap_full(h); \ |
87 | if (_r) { \ |
88 | size_t _i = (h)->used++; \ |
89 | (h)->data[_i] = d; \ |
90 | \ |
91 | heap_sift_down(h, _i, cmp); \ |
92 | heap_sift(h, _i, cmp); \ |
93 | } \ |
94 | _r; \ |
95 | }) |
96 | |
97 | #define heap_pop(h, d, cmp) \ |
98 | ({ \ |
99 | bool _r = (h)->used; \ |
100 | if (_r) { \ |
101 | (d) = (h)->data[0]; \ |
102 | (h)->used--; \ |
103 | heap_swap(h, 0, (h)->used); \ |
104 | heap_sift(h, 0, cmp); \ |
105 | } \ |
106 | _r; \ |
107 | }) |
108 | |
109 | #define heap_peek(h) ((h)->used ? (h)->data[0] : NULL) |
110 | |
111 | #define heap_full(h) ((h)->used == (h)->size) |
112 | |
113 | #define DECLARE_FIFO(type, name) \ |
114 | struct { \ |
115 | size_t front, back, size, mask; \ |
116 | type *data; \ |
117 | } name |
118 | |
119 | #define fifo_for_each(c, fifo, iter) \ |
120 | for (iter = (fifo)->front; \ |
121 | c = (fifo)->data[iter], iter != (fifo)->back; \ |
122 | iter = (iter + 1) & (fifo)->mask) |
123 | |
124 | #define __init_fifo(fifo, gfp) \ |
125 | ({ \ |
126 | size_t _allocated_size, _bytes; \ |
127 | BUG_ON(!(fifo)->size); \ |
128 | \ |
129 | _allocated_size = roundup_pow_of_two((fifo)->size + 1); \ |
130 | _bytes = _allocated_size * sizeof(*(fifo)->data); \ |
131 | \ |
132 | (fifo)->mask = _allocated_size - 1; \ |
133 | (fifo)->front = (fifo)->back = 0; \ |
134 | \ |
135 | (fifo)->data = kvmalloc(_bytes, (gfp) & GFP_KERNEL); \ |
136 | (fifo)->data; \ |
137 | }) |
138 | |
139 | #define init_fifo_exact(fifo, _size, gfp) \ |
140 | ({ \ |
141 | (fifo)->size = (_size); \ |
142 | __init_fifo(fifo, gfp); \ |
143 | }) |
144 | |
145 | #define init_fifo(fifo, _size, gfp) \ |
146 | ({ \ |
147 | (fifo)->size = (_size); \ |
148 | if ((fifo)->size > 4) \ |
149 | (fifo)->size = roundup_pow_of_two((fifo)->size) - 1; \ |
150 | __init_fifo(fifo, gfp); \ |
151 | }) |
152 | |
153 | #define free_fifo(fifo) \ |
154 | do { \ |
155 | kvfree((fifo)->data); \ |
156 | (fifo)->data = NULL; \ |
157 | } while (0) |
158 | |
159 | #define fifo_used(fifo) (((fifo)->back - (fifo)->front) & (fifo)->mask) |
160 | #define fifo_free(fifo) ((fifo)->size - fifo_used(fifo)) |
161 | |
162 | #define fifo_empty(fifo) (!fifo_used(fifo)) |
163 | #define fifo_full(fifo) (!fifo_free(fifo)) |
164 | |
165 | #define fifo_front(fifo) ((fifo)->data[(fifo)->front]) |
166 | #define fifo_back(fifo) \ |
167 | ((fifo)->data[((fifo)->back - 1) & (fifo)->mask]) |
168 | |
169 | #define fifo_idx(fifo, p) (((p) - &fifo_front(fifo)) & (fifo)->mask) |
170 | |
171 | #define fifo_push_back(fifo, i) \ |
172 | ({ \ |
173 | bool _r = !fifo_full((fifo)); \ |
174 | if (_r) { \ |
175 | (fifo)->data[(fifo)->back++] = (i); \ |
176 | (fifo)->back &= (fifo)->mask; \ |
177 | } \ |
178 | _r; \ |
179 | }) |
180 | |
181 | #define fifo_pop_front(fifo, i) \ |
182 | ({ \ |
183 | bool _r = !fifo_empty((fifo)); \ |
184 | if (_r) { \ |
185 | (i) = (fifo)->data[(fifo)->front++]; \ |
186 | (fifo)->front &= (fifo)->mask; \ |
187 | } \ |
188 | _r; \ |
189 | }) |
190 | |
191 | #define fifo_push_front(fifo, i) \ |
192 | ({ \ |
193 | bool _r = !fifo_full((fifo)); \ |
194 | if (_r) { \ |
195 | --(fifo)->front; \ |
196 | (fifo)->front &= (fifo)->mask; \ |
197 | (fifo)->data[(fifo)->front] = (i); \ |
198 | } \ |
199 | _r; \ |
200 | }) |
201 | |
202 | #define fifo_pop_back(fifo, i) \ |
203 | ({ \ |
204 | bool _r = !fifo_empty((fifo)); \ |
205 | if (_r) { \ |
206 | --(fifo)->back; \ |
207 | (fifo)->back &= (fifo)->mask; \ |
208 | (i) = (fifo)->data[(fifo)->back] \ |
209 | } \ |
210 | _r; \ |
211 | }) |
212 | |
213 | #define fifo_push(fifo, i) fifo_push_back(fifo, (i)) |
214 | #define fifo_pop(fifo, i) fifo_pop_front(fifo, (i)) |
215 | |
216 | #define fifo_swap(l, r) \ |
217 | do { \ |
218 | swap((l)->front, (r)->front); \ |
219 | swap((l)->back, (r)->back); \ |
220 | swap((l)->size, (r)->size); \ |
221 | swap((l)->mask, (r)->mask); \ |
222 | swap((l)->data, (r)->data); \ |
223 | } while (0) |
224 | |
225 | #define fifo_move(dest, src) \ |
226 | do { \ |
227 | typeof(*((dest)->data)) _t; \ |
228 | while (!fifo_full(dest) && \ |
229 | fifo_pop(src, _t)) \ |
230 | fifo_push(dest, _t); \ |
231 | } while (0) |
232 | |
233 | /* |
234 | * Simple array based allocator - preallocates a number of elements and you can |
235 | * never allocate more than that, also has no locking. |
236 | * |
237 | * Handy because if you know you only need a fixed number of elements you don't |
238 | * have to worry about memory allocation failure, and sometimes a mempool isn't |
239 | * what you want. |
240 | * |
241 | * We treat the free elements as entries in a singly linked list, and the |
242 | * freelist as a stack - allocating and freeing push and pop off the freelist. |
243 | */ |
244 | |
245 | #define DECLARE_ARRAY_ALLOCATOR(type, name, size) \ |
246 | struct { \ |
247 | type *freelist; \ |
248 | type data[size]; \ |
249 | } name |
250 | |
251 | #define array_alloc(array) \ |
252 | ({ \ |
253 | typeof((array)->freelist) _ret = (array)->freelist; \ |
254 | \ |
255 | if (_ret) \ |
256 | (array)->freelist = *((typeof((array)->freelist) *) _ret);\ |
257 | \ |
258 | _ret; \ |
259 | }) |
260 | |
261 | #define array_free(array, ptr) \ |
262 | do { \ |
263 | typeof((array)->freelist) _ptr = ptr; \ |
264 | \ |
265 | *((typeof((array)->freelist) *) _ptr) = (array)->freelist; \ |
266 | (array)->freelist = _ptr; \ |
267 | } while (0) |
268 | |
269 | #define array_allocator_init(array) \ |
270 | do { \ |
271 | typeof((array)->freelist) _i; \ |
272 | \ |
273 | BUILD_BUG_ON(sizeof((array)->data[0]) < sizeof(void *)); \ |
274 | (array)->freelist = NULL; \ |
275 | \ |
276 | for (_i = (array)->data; \ |
277 | _i < (array)->data + ARRAY_SIZE((array)->data); \ |
278 | _i++) \ |
279 | array_free(array, _i); \ |
280 | } while (0) |
281 | |
282 | #define array_freelist_empty(array) ((array)->freelist == NULL) |
283 | |
284 | #define ANYSINT_MAX(t) \ |
285 | ((((t) 1 << (sizeof(t) * 8 - 2)) - (t) 1) * (t) 2 + (t) 1) |
286 | |
287 | int bch_strtoint_h(const char *cp, int *res); |
288 | int bch_strtouint_h(const char *cp, unsigned int *res); |
289 | int bch_strtoll_h(const char *cp, long long *res); |
290 | int bch_strtoull_h(const char *cp, unsigned long long *res); |
291 | |
292 | static inline int bch_strtol_h(const char *cp, long *res) |
293 | { |
294 | #if BITS_PER_LONG == 32 |
295 | return bch_strtoint_h(cp, (int *) res); |
296 | #else |
297 | return bch_strtoll_h(cp, res: (long long *) res); |
298 | #endif |
299 | } |
300 | |
301 | static inline int bch_strtoul_h(const char *cp, long *res) |
302 | { |
303 | #if BITS_PER_LONG == 32 |
304 | return bch_strtouint_h(cp, (unsigned int *) res); |
305 | #else |
306 | return bch_strtoull_h(cp, res: (unsigned long long *) res); |
307 | #endif |
308 | } |
309 | |
310 | #define strtoi_h(cp, res) \ |
311 | (__builtin_types_compatible_p(typeof(*res), int) \ |
312 | ? bch_strtoint_h(cp, (void *) res) \ |
313 | : __builtin_types_compatible_p(typeof(*res), long) \ |
314 | ? bch_strtol_h(cp, (void *) res) \ |
315 | : __builtin_types_compatible_p(typeof(*res), long long) \ |
316 | ? bch_strtoll_h(cp, (void *) res) \ |
317 | : __builtin_types_compatible_p(typeof(*res), unsigned int) \ |
318 | ? bch_strtouint_h(cp, (void *) res) \ |
319 | : __builtin_types_compatible_p(typeof(*res), unsigned long) \ |
320 | ? bch_strtoul_h(cp, (void *) res) \ |
321 | : __builtin_types_compatible_p(typeof(*res), unsigned long long)\ |
322 | ? bch_strtoull_h(cp, (void *) res) : -EINVAL) |
323 | |
324 | #define strtoul_safe(cp, var) \ |
325 | ({ \ |
326 | unsigned long _v; \ |
327 | int _r = kstrtoul(cp, 10, &_v); \ |
328 | if (!_r) \ |
329 | var = _v; \ |
330 | _r; \ |
331 | }) |
332 | |
333 | #define strtoul_safe_clamp(cp, var, min, max) \ |
334 | ({ \ |
335 | unsigned long _v; \ |
336 | int _r = kstrtoul(cp, 10, &_v); \ |
337 | if (!_r) \ |
338 | var = clamp_t(typeof(var), _v, min, max); \ |
339 | _r; \ |
340 | }) |
341 | |
342 | ssize_t bch_hprint(char *buf, int64_t v); |
343 | |
344 | bool bch_is_zero(const char *p, size_t n); |
345 | int bch_parse_uuid(const char *s, char *uuid); |
346 | |
347 | struct time_stats { |
348 | spinlock_t lock; |
349 | /* |
350 | * all fields are in nanoseconds, averages are ewmas stored left shifted |
351 | * by 8 |
352 | */ |
353 | uint64_t max_duration; |
354 | uint64_t average_duration; |
355 | uint64_t average_frequency; |
356 | uint64_t last; |
357 | }; |
358 | |
359 | void bch_time_stats_update(struct time_stats *stats, uint64_t time); |
360 | |
361 | static inline unsigned int local_clock_us(void) |
362 | { |
363 | return local_clock() >> 10; |
364 | } |
365 | |
366 | #define NSEC_PER_ns 1L |
367 | #define NSEC_PER_us NSEC_PER_USEC |
368 | #define NSEC_PER_ms NSEC_PER_MSEC |
369 | #define NSEC_PER_sec NSEC_PER_SEC |
370 | |
371 | #define __print_time_stat(stats, name, stat, units) \ |
372 | sysfs_print(name ## _ ## stat ## _ ## units, \ |
373 | div_u64((stats)->stat >> 8, NSEC_PER_ ## units)) |
374 | |
375 | #define sysfs_print_time_stats(stats, name, \ |
376 | frequency_units, \ |
377 | duration_units) \ |
378 | do { \ |
379 | __print_time_stat(stats, name, \ |
380 | average_frequency, frequency_units); \ |
381 | __print_time_stat(stats, name, \ |
382 | average_duration, duration_units); \ |
383 | sysfs_print(name ## _ ##max_duration ## _ ## duration_units, \ |
384 | div_u64((stats)->max_duration, \ |
385 | NSEC_PER_ ## duration_units)); \ |
386 | \ |
387 | sysfs_print(name ## _last_ ## frequency_units, (stats)->last \ |
388 | ? div_s64(local_clock() - (stats)->last, \ |
389 | NSEC_PER_ ## frequency_units) \ |
390 | : -1LL); \ |
391 | } while (0) |
392 | |
393 | #define sysfs_time_stats_attribute(name, \ |
394 | frequency_units, \ |
395 | duration_units) \ |
396 | read_attribute(name ## _average_frequency_ ## frequency_units); \ |
397 | read_attribute(name ## _average_duration_ ## duration_units); \ |
398 | read_attribute(name ## _max_duration_ ## duration_units); \ |
399 | read_attribute(name ## _last_ ## frequency_units) |
400 | |
401 | #define sysfs_time_stats_attribute_list(name, \ |
402 | frequency_units, \ |
403 | duration_units) \ |
404 | &sysfs_ ## name ## _average_frequency_ ## frequency_units, \ |
405 | &sysfs_ ## name ## _average_duration_ ## duration_units, \ |
406 | &sysfs_ ## name ## _max_duration_ ## duration_units, \ |
407 | &sysfs_ ## name ## _last_ ## frequency_units, |
408 | |
409 | #define ewma_add(ewma, val, weight, factor) \ |
410 | ({ \ |
411 | (ewma) *= (weight) - 1; \ |
412 | (ewma) += (val) << factor; \ |
413 | (ewma) /= (weight); \ |
414 | (ewma) >> factor; \ |
415 | }) |
416 | |
417 | struct bch_ratelimit { |
418 | /* Next time we want to do some work, in nanoseconds */ |
419 | uint64_t next; |
420 | |
421 | /* |
422 | * Rate at which we want to do work, in units per second |
423 | * The units here correspond to the units passed to bch_next_delay() |
424 | */ |
425 | atomic_long_t rate; |
426 | }; |
427 | |
428 | static inline void bch_ratelimit_reset(struct bch_ratelimit *d) |
429 | { |
430 | d->next = local_clock(); |
431 | } |
432 | |
433 | uint64_t bch_next_delay(struct bch_ratelimit *d, uint64_t done); |
434 | |
435 | #define __DIV_SAFE(n, d, zero) \ |
436 | ({ \ |
437 | typeof(n) _n = (n); \ |
438 | typeof(d) _d = (d); \ |
439 | _d ? _n / _d : zero; \ |
440 | }) |
441 | |
442 | #define DIV_SAFE(n, d) __DIV_SAFE(n, d, 0) |
443 | |
444 | #define container_of_or_null(ptr, type, member) \ |
445 | ({ \ |
446 | typeof(ptr) _ptr = ptr; \ |
447 | _ptr ? container_of(_ptr, type, member) : NULL; \ |
448 | }) |
449 | |
450 | #define RB_INSERT(root, new, member, cmp) \ |
451 | ({ \ |
452 | __label__ dup; \ |
453 | struct rb_node **n = &(root)->rb_node, *parent = NULL; \ |
454 | typeof(new) this; \ |
455 | int res, ret = -1; \ |
456 | \ |
457 | while (*n) { \ |
458 | parent = *n; \ |
459 | this = container_of(*n, typeof(*(new)), member); \ |
460 | res = cmp(new, this); \ |
461 | if (!res) \ |
462 | goto dup; \ |
463 | n = res < 0 \ |
464 | ? &(*n)->rb_left \ |
465 | : &(*n)->rb_right; \ |
466 | } \ |
467 | \ |
468 | rb_link_node(&(new)->member, parent, n); \ |
469 | rb_insert_color(&(new)->member, root); \ |
470 | ret = 0; \ |
471 | dup: \ |
472 | ret; \ |
473 | }) |
474 | |
475 | #define RB_SEARCH(root, search, member, cmp) \ |
476 | ({ \ |
477 | struct rb_node *n = (root)->rb_node; \ |
478 | typeof(&(search)) this, ret = NULL; \ |
479 | int res; \ |
480 | \ |
481 | while (n) { \ |
482 | this = container_of(n, typeof(search), member); \ |
483 | res = cmp(&(search), this); \ |
484 | if (!res) { \ |
485 | ret = this; \ |
486 | break; \ |
487 | } \ |
488 | n = res < 0 \ |
489 | ? n->rb_left \ |
490 | : n->rb_right; \ |
491 | } \ |
492 | ret; \ |
493 | }) |
494 | |
495 | #define RB_GREATER(root, search, member, cmp) \ |
496 | ({ \ |
497 | struct rb_node *n = (root)->rb_node; \ |
498 | typeof(&(search)) this, ret = NULL; \ |
499 | int res; \ |
500 | \ |
501 | while (n) { \ |
502 | this = container_of(n, typeof(search), member); \ |
503 | res = cmp(&(search), this); \ |
504 | if (res < 0) { \ |
505 | ret = this; \ |
506 | n = n->rb_left; \ |
507 | } else \ |
508 | n = n->rb_right; \ |
509 | } \ |
510 | ret; \ |
511 | }) |
512 | |
513 | #define RB_FIRST(root, type, member) \ |
514 | container_of_or_null(rb_first(root), type, member) |
515 | |
516 | #define RB_LAST(root, type, member) \ |
517 | container_of_or_null(rb_last(root), type, member) |
518 | |
519 | #define RB_NEXT(ptr, member) \ |
520 | container_of_or_null(rb_next(&(ptr)->member), typeof(*ptr), member) |
521 | |
522 | #define RB_PREV(ptr, member) \ |
523 | container_of_or_null(rb_prev(&(ptr)->member), typeof(*ptr), member) |
524 | |
525 | static inline uint64_t bch_crc64(const void *p, size_t len) |
526 | { |
527 | uint64_t crc = 0xffffffffffffffffULL; |
528 | |
529 | crc = crc64_be(crc, p, len); |
530 | return crc ^ 0xffffffffffffffffULL; |
531 | } |
532 | |
533 | /* |
534 | * A stepwise-linear pseudo-exponential. This returns 1 << (x >> |
535 | * frac_bits), with the less-significant bits filled in by linear |
536 | * interpolation. |
537 | * |
538 | * This can also be interpreted as a floating-point number format, |
539 | * where the low frac_bits are the mantissa (with implicit leading |
540 | * 1 bit), and the more significant bits are the exponent. |
541 | * The return value is 1.mantissa * 2^exponent. |
542 | * |
543 | * The way this is used, fract_bits is 6 and the largest possible |
544 | * input is CONGESTED_MAX-1 = 1023 (exponent 16, mantissa 0x1.fc), |
545 | * so the maximum output is 0x1fc00. |
546 | */ |
547 | static inline unsigned int fract_exp_two(unsigned int x, |
548 | unsigned int fract_bits) |
549 | { |
550 | unsigned int mantissa = 1 << fract_bits; /* Implicit bit */ |
551 | |
552 | mantissa += x & (mantissa - 1); |
553 | x >>= fract_bits; /* The exponent */ |
554 | /* Largest intermediate value 0x7f0000 */ |
555 | return mantissa << x >> fract_bits; |
556 | } |
557 | |
558 | void bch_bio_map(struct bio *bio, void *base); |
559 | int bch_bio_alloc_pages(struct bio *bio, gfp_t gfp_mask); |
560 | |
561 | #endif /* _BCACHE_UTIL_H */ |
562 | |