1 #ifndef _BCACHEFS_UTIL_H
2 #define _BCACHEFS_UTIL_H
5 #include <linux/blkdev.h>
6 #include <linux/closure.h>
7 #include <linux/errno.h>
8 #include <linux/freezer.h>
9 #include <linux/kernel.h>
10 #include <linux/sched/clock.h>
11 #include <linux/llist.h>
12 #include <linux/log2.h>
13 #include <linux/percpu.h>
14 #include <linux/preempt.h>
15 #include <linux/ratelimit.h>
16 #include <linux/slab.h>
17 #include <linux/vmalloc.h>
18 #include <linux/workqueue.h>
20 #define PAGE_SECTOR_SHIFT (PAGE_SHIFT - 9)
21 #define PAGE_SECTORS (1UL << PAGE_SECTOR_SHIFT)
25 #ifdef CONFIG_BCACHEFS_DEBUG
27 #define EBUG_ON(cond) BUG_ON(cond)
28 #define atomic_dec_bug(v) BUG_ON(atomic_dec_return(v) < 0)
29 #define atomic_inc_bug(v, i) BUG_ON(atomic_inc_return(v) <= i)
30 #define atomic_sub_bug(i, v) BUG_ON(atomic_sub_return(i, v) < 0)
31 #define atomic_add_bug(i, v) BUG_ON(atomic_add_return(i, v) < 0)
32 #define atomic_long_dec_bug(v) BUG_ON(atomic_long_dec_return(v) < 0)
33 #define atomic_long_sub_bug(i, v) BUG_ON(atomic_long_sub_return(i, v) < 0)
34 #define atomic64_dec_bug(v) BUG_ON(atomic64_dec_return(v) < 0)
35 #define atomic64_inc_bug(v, i) BUG_ON(atomic64_inc_return(v) <= i)
36 #define atomic64_sub_bug(i, v) BUG_ON(atomic64_sub_return(i, v) < 0)
37 #define atomic64_add_bug(i, v) BUG_ON(atomic64_add_return(i, v) < 0)
39 #define memcpy(dst, src, len) \
42 const void *_src = (src); \
43 size_t _len = (len); \
45 BUG_ON(!((void *) (_dst) >= (void *) (_src) + (_len) || \
46 (void *) (_dst) + (_len) <= (void *) (_src))); \
47 memcpy(_dst, _src, _len); \
53 #define atomic_dec_bug(v) atomic_dec(v)
54 #define atomic_inc_bug(v, i) atomic_inc(v)
55 #define atomic_sub_bug(i, v) atomic_sub(i, v)
56 #define atomic_add_bug(i, v) atomic_add(i, v)
57 #define atomic_long_dec_bug(v) atomic_long_dec(v)
58 #define atomic_long_sub_bug(i, v) atomic_long_sub(i, v)
59 #define atomic64_dec_bug(v) atomic64_dec(v)
60 #define atomic64_inc_bug(v, i) atomic64_inc(v)
61 #define atomic64_sub_bug(i, v) atomic64_sub(i, v)
62 #define atomic64_add_bug(i, v) atomic64_add(i, v)
66 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
67 #define CPU_BIG_ENDIAN 0
68 #elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
69 #define CPU_BIG_ENDIAN 1
74 #define type_is_exact(_val, _type) \
75 __builtin_types_compatible_p(typeof(_val), _type)
77 #define type_is(_val, _type) \
78 (__builtin_types_compatible_p(typeof(_val), _type) || \
79 __builtin_types_compatible_p(typeof(_val), const _type))
81 /* Userspace doesn't align allocations as nicely as the kernel allocators: */
82 static inline size_t buf_pages(void *p, size_t len)
84 return DIV_ROUND_UP(len +
85 ((unsigned long) p & (PAGE_SIZE - 1)),
89 static inline void vpfree(void *p, size_t size)
91 if (is_vmalloc_addr(p))
94 free_pages((unsigned long) p, get_order(size));
97 static inline void *vpmalloc(size_t size, gfp_t gfp_mask)
99 return (void *) __get_free_pages(gfp_mask|__GFP_NOWARN,
101 __vmalloc(size, gfp_mask, PAGE_KERNEL);
104 static inline void kvpfree(void *p, size_t size)
106 if (size < PAGE_SIZE)
112 static inline void *kvpmalloc(size_t size, gfp_t gfp_mask)
114 return size < PAGE_SIZE
115 ? kmalloc(size, gfp_mask)
116 : vpmalloc(size, gfp_mask);
119 int mempool_init_kvpmalloc_pool(mempool_t *, int, size_t);
127 #define DECLARE_HEAP(type, name) HEAP(type) name
129 #define init_heap(heap, _size, gfp) \
132 (heap)->size = (_size); \
133 (heap)->data = kvpmalloc((heap)->size * sizeof((heap)->data[0]),\
137 #define free_heap(heap) \
139 kvpfree((heap)->data, (heap)->size * sizeof((heap)->data[0])); \
140 (heap)->data = NULL; \
143 #define heap_set_backpointer(h, i, _fn) \
145 void (*fn)(typeof(h), size_t) = _fn; \
150 #define heap_swap(h, i, j, set_backpointer) \
152 swap((h)->data[i], (h)->data[j]); \
153 heap_set_backpointer(h, i, set_backpointer); \
154 heap_set_backpointer(h, j, set_backpointer); \
157 #define heap_peek(h) \
159 EBUG_ON(!(h)->used); \
163 #define heap_full(h) ((h)->used == (h)->size)
165 #define heap_sift_down(h, i, cmp, set_backpointer) \
169 for (; _j * 2 + 1 < (h)->used; _j = _c) { \
171 if (_c + 1 < (h)->used && \
172 cmp(h, (h)->data[_c], (h)->data[_c + 1]) >= 0) \
175 if (cmp(h, (h)->data[_c], (h)->data[_j]) >= 0) \
177 heap_swap(h, _c, _j, set_backpointer); \
181 #define heap_sift_up(h, i, cmp, set_backpointer) \
184 size_t p = (i - 1) / 2; \
185 if (cmp(h, (h)->data[i], (h)->data[p]) >= 0) \
187 heap_swap(h, i, p, set_backpointer); \
192 #define __heap_add(h, d, cmp, set_backpointer) \
194 size_t _i = (h)->used++; \
196 heap_set_backpointer(h, _i, set_backpointer); \
198 heap_sift_up(h, _i, cmp, set_backpointer); \
202 #define heap_add(h, d, cmp, set_backpointer) \
204 bool _r = !heap_full(h); \
206 __heap_add(h, d, cmp, set_backpointer); \
210 #define heap_add_or_replace(h, new, cmp, set_backpointer) \
212 if (!heap_add(h, new, cmp, set_backpointer) && \
213 cmp(h, new, heap_peek(h)) >= 0) { \
214 (h)->data[0] = new; \
215 heap_set_backpointer(h, 0, set_backpointer); \
216 heap_sift_down(h, 0, cmp, set_backpointer); \
220 #define heap_del(h, i, cmp, set_backpointer) \
224 BUG_ON(_i >= (h)->used); \
226 heap_swap(h, _i, (h)->used, set_backpointer); \
227 heap_sift_up(h, _i, cmp, set_backpointer); \
228 heap_sift_down(h, _i, cmp, set_backpointer); \
231 #define heap_pop(h, d, cmp, set_backpointer) \
233 bool _r = (h)->used; \
235 (d) = (h)->data[0]; \
236 heap_del(h, 0, cmp, set_backpointer); \
241 #define heap_resort(heap, cmp, set_backpointer) \
244 for (_i = (ssize_t) (heap)->used / 2 - 1; _i >= 0; --_i) \
245 heap_sift_down(heap, _i, cmp, set_backpointer); \
248 #define ANYSINT_MAX(t) \
249 ((((t) 1 << (sizeof(t) * 8 - 2)) - (t) 1) * (t) 2 + (t) 1)
256 static inline size_t printbuf_remaining(struct printbuf *buf)
258 return buf->end - buf->pos;
261 #define _PBUF(_buf, _len) \
262 ((struct printbuf) { \
264 .end = _buf + _len, \
267 #define PBUF(_buf) _PBUF(_buf, sizeof(_buf))
269 #define pr_buf(_out, ...) \
271 (_out)->pos += scnprintf((_out)->pos, printbuf_remaining(_out), \
275 void bch_scnmemcpy(struct printbuf *, const char *, size_t);
277 int bch2_strtoint_h(const char *, int *);
278 int bch2_strtouint_h(const char *, unsigned int *);
279 int bch2_strtoll_h(const char *, long long *);
280 int bch2_strtoull_h(const char *, unsigned long long *);
281 int bch2_strtou64_h(const char *, u64 *);
283 static inline int bch2_strtol_h(const char *cp, long *res)
285 #if BITS_PER_LONG == 32
286 return bch2_strtoint_h(cp, (int *) res);
288 return bch2_strtoll_h(cp, (long long *) res);
292 static inline int bch2_strtoul_h(const char *cp, long *res)
294 #if BITS_PER_LONG == 32
295 return bch2_strtouint_h(cp, (unsigned int *) res);
297 return bch2_strtoull_h(cp, (unsigned long long *) res);
301 #define strtoi_h(cp, res) \
302 ( type_is(*res, int) ? bch2_strtoint_h(cp, (void *) res)\
303 : type_is(*res, long) ? bch2_strtol_h(cp, (void *) res)\
304 : type_is(*res, long long) ? bch2_strtoll_h(cp, (void *) res)\
305 : type_is(*res, unsigned) ? bch2_strtouint_h(cp, (void *) res)\
306 : type_is(*res, unsigned long) ? bch2_strtoul_h(cp, (void *) res)\
307 : type_is(*res, unsigned long long) ? bch2_strtoull_h(cp, (void *) res)\
310 #define strtoul_safe(cp, var) \
313 int _r = kstrtoul(cp, 10, &_v); \
319 #define strtoul_safe_clamp(cp, var, min, max) \
322 int _r = kstrtoul(cp, 10, &_v); \
324 var = clamp_t(typeof(var), _v, min, max); \
328 #define strtoul_safe_restrict(cp, var, min, max) \
331 int _r = kstrtoul(cp, 10, &_v); \
332 if (!_r && _v >= min && _v <= max) \
339 #define snprint(buf, size, var) \
340 snprintf(buf, size, \
341 type_is(var, int) ? "%i\n" \
342 : type_is(var, unsigned) ? "%u\n" \
343 : type_is(var, long) ? "%li\n" \
344 : type_is(var, unsigned long) ? "%lu\n" \
345 : type_is(var, s64) ? "%lli\n" \
346 : type_is(var, u64) ? "%llu\n" \
347 : type_is(var, char *) ? "%s\n" \
350 void bch2_hprint(struct printbuf *, s64);
352 bool bch2_is_zero(const void *, size_t);
354 void bch2_string_opt_to_text(struct printbuf *,
355 const char * const [], size_t);
357 void bch2_flags_to_text(struct printbuf *, const char * const[], u64);
358 u64 bch2_read_flag_list(char *, const char * const[]);
360 #define NR_QUANTILES 15
361 #define QUANTILE_IDX(i) inorder_to_eytzinger0(i, NR_QUANTILES)
362 #define QUANTILE_FIRST eytzinger0_first(NR_QUANTILES)
363 #define QUANTILE_LAST eytzinger0_last(NR_QUANTILES)
366 struct quantile_entry {
369 } entries[NR_QUANTILES];
372 struct time_stat_buffer {
374 struct time_stat_buffer_entry {
383 /* all fields are in nanoseconds */
384 u64 average_duration;
385 u64 average_frequency;
388 struct quantiles quantiles;
390 struct time_stat_buffer __percpu *buffer;
393 void __bch2_time_stats_update(struct time_stats *stats, u64, u64);
395 static inline void bch2_time_stats_update(struct time_stats *stats, u64 start)
397 __bch2_time_stats_update(stats, start, local_clock());
400 size_t bch2_time_stats_print(struct time_stats *, char *, size_t);
402 void bch2_time_stats_exit(struct time_stats *);
403 void bch2_time_stats_init(struct time_stats *);
405 #define ewma_add(ewma, val, weight) \
407 typeof(ewma) _ewma = (ewma); \
408 typeof(weight) _weight = (weight); \
410 (((_ewma << _weight) - _ewma) + (val)) >> _weight; \
413 struct bch_ratelimit {
414 /* Next time we want to do some work, in nanoseconds */
418 * Rate at which we want to do work, in units per nanosecond
419 * The units here correspond to the units passed to
420 * bch2_ratelimit_increment()
425 static inline void bch2_ratelimit_reset(struct bch_ratelimit *d)
427 d->next = local_clock();
430 u64 bch2_ratelimit_delay(struct bch_ratelimit *);
431 void bch2_ratelimit_increment(struct bch_ratelimit *, u64);
433 struct bch_pd_controller {
434 struct bch_ratelimit rate;
435 unsigned long last_update;
438 s64 smoothed_derivative;
440 unsigned p_term_inverse;
444 /* for exporting to sysfs (no effect on behavior) */
446 s64 last_proportional;
450 /* If true, the rate will not increase if bch2_ratelimit_delay()
451 * is not being called often enough. */
455 void bch2_pd_controller_update(struct bch_pd_controller *, s64, s64, int);
456 void bch2_pd_controller_init(struct bch_pd_controller *);
457 size_t bch2_pd_controller_print_debug(struct bch_pd_controller *, char *);
459 #define sysfs_pd_controller_attribute(name) \
460 rw_attribute(name##_rate); \
461 rw_attribute(name##_rate_bytes); \
462 rw_attribute(name##_rate_d_term); \
463 rw_attribute(name##_rate_p_term_inverse); \
464 read_attribute(name##_rate_debug)
466 #define sysfs_pd_controller_files(name) \
467 &sysfs_##name##_rate, \
468 &sysfs_##name##_rate_bytes, \
469 &sysfs_##name##_rate_d_term, \
470 &sysfs_##name##_rate_p_term_inverse, \
471 &sysfs_##name##_rate_debug
473 #define sysfs_pd_controller_show(name, var) \
475 sysfs_hprint(name##_rate, (var)->rate.rate); \
476 sysfs_print(name##_rate_bytes, (var)->rate.rate); \
477 sysfs_print(name##_rate_d_term, (var)->d_term); \
478 sysfs_print(name##_rate_p_term_inverse, (var)->p_term_inverse); \
480 if (attr == &sysfs_##name##_rate_debug) \
481 return bch2_pd_controller_print_debug(var, buf); \
484 #define sysfs_pd_controller_store(name, var) \
486 sysfs_strtoul_clamp(name##_rate, \
487 (var)->rate.rate, 1, UINT_MAX); \
488 sysfs_strtoul_clamp(name##_rate_bytes, \
489 (var)->rate.rate, 1, UINT_MAX); \
490 sysfs_strtoul(name##_rate_d_term, (var)->d_term); \
491 sysfs_strtoul_clamp(name##_rate_p_term_inverse, \
492 (var)->p_term_inverse, 1, INT_MAX); \
495 #define container_of_or_null(ptr, type, member) \
497 typeof(ptr) _ptr = ptr; \
498 _ptr ? container_of(_ptr, type, member) : NULL; \
501 /* Does linear interpolation between powers of two */
502 static inline unsigned fract_exp_two(unsigned x, unsigned fract_bits)
504 unsigned fract = x & ~(~0 << fract_bits);
508 x += (x * fract) >> fract_bits;
513 void bch2_bio_map(struct bio *bio, void *base);
514 int bch2_bio_alloc_pages(struct bio *bio, gfp_t gfp_mask);
516 static inline sector_t bdev_sectors(struct block_device *bdev)
518 return bdev->bd_inode->i_size >> 9;
521 #define closure_bio_submit(bio, cl) \
527 #define kthread_wait_freezable(cond) \
531 set_current_state(TASK_INTERRUPTIBLE); \
532 if (kthread_should_stop()) { \
543 set_current_state(TASK_RUNNING); \
547 size_t bch2_rand_range(size_t);
549 void memcpy_to_bio(struct bio *, struct bvec_iter, void *);
550 void memcpy_from_bio(void *, struct bio *, struct bvec_iter);
552 static inline void __memcpy_u64s(void *dst, const void *src,
557 asm volatile("rep ; movsq"
558 : "=&c" (d0), "=&D" (d1), "=&S" (d2)
559 : "0" (u64s), "1" (dst), "2" (src)
570 static inline void memcpy_u64s(void *dst, const void *src,
573 EBUG_ON(!(dst >= src + u64s * sizeof(u64) ||
574 dst + u64s * sizeof(u64) <= src));
576 __memcpy_u64s(dst, src, u64s);
579 static inline void __memmove_u64s_down(void *dst, const void *src,
582 __memcpy_u64s(dst, src, u64s);
585 static inline void memmove_u64s_down(void *dst, const void *src,
590 __memmove_u64s_down(dst, src, u64s);
593 static inline void __memmove_u64s_up(void *_dst, const void *_src,
596 u64 *dst = (u64 *) _dst + u64s - 1;
597 u64 *src = (u64 *) _src + u64s - 1;
601 asm volatile("std ;\n"
604 : "=&c" (d0), "=&D" (d1), "=&S" (d2)
605 : "0" (u64s), "1" (dst), "2" (src)
613 static inline void memmove_u64s_up(void *dst, const void *src,
618 __memmove_u64s_up(dst, src, u64s);
621 static inline void memmove_u64s(void *dst, const void *src,
625 __memmove_u64s_down(dst, src, u64s);
627 __memmove_u64s_up(dst, src, u64s);
630 static inline struct bio_vec next_contig_bvec(struct bio *bio,
631 struct bvec_iter *iter)
633 struct bio_vec bv = bio_iter_iovec(bio, *iter);
635 bio_advance_iter(bio, iter, bv.bv_len);
636 #ifndef CONFIG_HIGHMEM
637 while (iter->bi_size) {
638 struct bio_vec next = bio_iter_iovec(bio, *iter);
640 if (page_address(bv.bv_page) + bv.bv_offset + bv.bv_len !=
641 page_address(next.bv_page) + next.bv_offset)
644 bv.bv_len += next.bv_len;
645 bio_advance_iter(bio, iter, next.bv_len);
651 #define __bio_for_each_contig_segment(bv, bio, iter, start) \
652 for (iter = (start); \
654 ((bv = next_contig_bvec((bio), &(iter))), 1);)
656 #define bio_for_each_contig_segment(bv, bio, iter) \
657 __bio_for_each_contig_segment(bv, bio, iter, (bio)->bi_iter)
659 void sort_cmp_size(void *base, size_t num, size_t size,
660 int (*cmp_func)(const void *, const void *, size_t),
661 void (*swap_func)(void *, void *, size_t));
663 /* just the memmove, doesn't update @_nr */
664 #define __array_insert_item(_array, _nr, _pos) \
665 memmove(&(_array)[(_pos) + 1], \
667 sizeof((_array)[0]) * ((_nr) - (_pos)))
669 #define array_insert_item(_array, _nr, _pos, _new_item) \
671 __array_insert_item(_array, _nr, _pos); \
673 (_array)[(_pos)] = (_new_item); \
676 #define array_remove_items(_array, _nr, _pos, _nr_to_remove) \
678 (_nr) -= (_nr_to_remove); \
679 memmove(&(_array)[(_pos)], \
680 &(_array)[(_pos) + (_nr_to_remove)], \
681 sizeof((_array)[0]) * ((_nr) - (_pos))); \
684 #define array_remove_item(_array, _nr, _pos) \
685 array_remove_items(_array, _nr, _pos, 1)
687 #define bubble_sort(_base, _nr, _cmp) \
690 bool _swapped = true; \
692 for (_end = (ssize_t) (_nr) - 1; _end > 0 && _swapped; --_end) {\
694 for (_i = 0; _i < _end; _i++) \
695 if (_cmp((_base)[_i], (_base)[_i + 1]) > 0) { \
696 swap((_base)[_i], (_base)[_i + 1]); \
702 static inline u64 percpu_u64_get(u64 __percpu *src)
707 for_each_possible_cpu(cpu)
708 ret += *per_cpu_ptr(src, cpu);
712 static inline void percpu_u64_set(u64 __percpu *dst, u64 src)
716 for_each_possible_cpu(cpu)
717 *per_cpu_ptr(dst, cpu) = 0;
720 *this_cpu_ptr(dst) = src;
724 static inline void acc_u64s(u64 *acc, const u64 *src, unsigned nr)
728 for (i = 0; i < nr; i++)
732 static inline void acc_u64s_percpu(u64 *acc, const u64 __percpu *src,
737 for_each_possible_cpu(cpu)
738 acc_u64s(acc, per_cpu_ptr(src, cpu), nr);
741 u64 *bch2_acc_percpu_u64s(u64 __percpu *, unsigned);
743 #endif /* _BCACHEFS_UTIL_H */