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/ratelimit.h>
14 #include <linux/slab.h>
15 #include <linux/vmalloc.h>
16 #include <linux/workqueue.h>
18 #define PAGE_SECTOR_SHIFT (PAGE_SHIFT - 9)
19 #define PAGE_SECTORS (1UL << PAGE_SECTOR_SHIFT)
23 #ifdef CONFIG_BCACHEFS_DEBUG
25 #define EBUG_ON(cond) BUG_ON(cond)
26 #define atomic_dec_bug(v) BUG_ON(atomic_dec_return(v) < 0)
27 #define atomic_inc_bug(v, i) BUG_ON(atomic_inc_return(v) <= i)
28 #define atomic_sub_bug(i, v) BUG_ON(atomic_sub_return(i, v) < 0)
29 #define atomic_add_bug(i, v) BUG_ON(atomic_add_return(i, v) < 0)
30 #define atomic_long_dec_bug(v) BUG_ON(atomic_long_dec_return(v) < 0)
31 #define atomic_long_sub_bug(i, v) BUG_ON(atomic_long_sub_return(i, v) < 0)
32 #define atomic64_dec_bug(v) BUG_ON(atomic64_dec_return(v) < 0)
33 #define atomic64_inc_bug(v, i) BUG_ON(atomic64_inc_return(v) <= i)
34 #define atomic64_sub_bug(i, v) BUG_ON(atomic64_sub_return(i, v) < 0)
35 #define atomic64_add_bug(i, v) BUG_ON(atomic64_add_return(i, v) < 0)
37 #define memcpy(dst, src, len) \
40 const void *_src = (src); \
41 size_t _len = (len); \
43 BUG_ON(!((void *) (_dst) >= (void *) (_src) + (_len) || \
44 (void *) (_dst) + (_len) <= (void *) (_src))); \
45 memcpy(_dst, _src, _len); \
51 #define atomic_dec_bug(v) atomic_dec(v)
52 #define atomic_inc_bug(v, i) atomic_inc(v)
53 #define atomic_sub_bug(i, v) atomic_sub(i, v)
54 #define atomic_add_bug(i, v) atomic_add(i, v)
55 #define atomic_long_dec_bug(v) atomic_long_dec(v)
56 #define atomic_long_sub_bug(i, v) atomic_long_sub(i, v)
57 #define atomic64_dec_bug(v) atomic64_dec(v)
58 #define atomic64_inc_bug(v, i) atomic64_inc(v)
59 #define atomic64_sub_bug(i, v) atomic64_sub(i, v)
60 #define atomic64_add_bug(i, v) atomic64_add(i, v)
65 #define __flatten __attribute__((flatten))
67 /* sparse doesn't know about attribute((flatten)) */
71 #ifdef __LITTLE_ENDIAN
72 #define CPU_BIG_ENDIAN 0
74 #define CPU_BIG_ENDIAN 1
79 #define type_is_exact(_val, _type) \
80 __builtin_types_compatible_p(typeof(_val), _type)
82 #define type_is(_val, _type) \
83 (__builtin_types_compatible_p(typeof(_val), _type) || \
84 __builtin_types_compatible_p(typeof(_val), const _type))
86 static inline void vpfree(void *p, size_t size)
88 if (is_vmalloc_addr(p))
91 free_pages((unsigned long) p, get_order(size));
94 static inline void *vpmalloc(size_t size, gfp_t gfp_mask)
96 return (void *) __get_free_pages(gfp_mask|__GFP_NOWARN,
98 __vmalloc(size, gfp_mask, PAGE_KERNEL);
101 static inline void kvpfree(void *p, size_t size)
103 if (size < PAGE_SIZE)
109 static inline void *kvpmalloc(size_t size, gfp_t gfp_mask)
111 return size < PAGE_SIZE
112 ? kmalloc(size, gfp_mask)
113 : vpmalloc(size, gfp_mask);
116 void mempool_free_vp(void *element, void *pool_data);
117 void *mempool_alloc_vp(gfp_t gfp_mask, void *pool_data);
119 static inline int mempool_init_vp_pool(mempool_t *pool, int min_nr, size_t size)
121 return mempool_init(pool, min_nr, mempool_alloc_vp,
122 mempool_free_vp, (void *) size);
131 #define DECLARE_HEAP(type, name) HEAP(type) name
133 #define init_heap(heap, _size, gfp) \
136 (heap)->size = (_size); \
137 (heap)->data = kvpmalloc((heap)->size * sizeof((heap)->data[0]),\
141 #define free_heap(heap) \
143 kvpfree((heap)->data, (heap)->size * sizeof((heap)->data[0])); \
144 (heap)->data = NULL; \
147 #define heap_swap(h, i, j) swap((h)->data[i], (h)->data[j])
149 #define heap_peek(h) \
151 EBUG_ON(!(h)->used); \
155 #define heap_full(h) ((h)->used == (h)->size)
157 #define heap_sift_down(h, i, cmp) \
161 for (; _j * 2 + 1 < (h)->used; _j = _c) { \
163 if (_c + 1 < (h)->used && \
164 cmp(h, (h)->data[_c], (h)->data[_c + 1]) >= 0) \
167 if (cmp(h, (h)->data[_c], (h)->data[_j]) >= 0) \
169 heap_swap(h, _c, _j); \
173 #define heap_sift_up(h, i, cmp) \
176 size_t p = (i - 1) / 2; \
177 if (cmp(h, (h)->data[i], (h)->data[p]) >= 0) \
179 heap_swap(h, i, p); \
184 #define __heap_add(h, d, cmp) \
186 size_t _i = (h)->used++; \
189 heap_sift_up(h, _i, cmp); \
192 #define heap_add(h, d, cmp) \
194 bool _r = !heap_full(h); \
196 __heap_add(h, d, cmp); \
200 #define heap_add_or_replace(h, new, cmp) \
202 if (!heap_add(h, new, cmp) && \
203 cmp(h, new, heap_peek(h)) >= 0) { \
204 (h)->data[0] = new; \
205 heap_sift_down(h, 0, cmp); \
209 #define heap_del(h, i, cmp) \
213 BUG_ON(_i >= (h)->used); \
215 heap_swap(h, _i, (h)->used); \
216 heap_sift_up(h, _i, cmp); \
217 heap_sift_down(h, _i, cmp); \
220 #define heap_pop(h, d, cmp) \
222 bool _r = (h)->used; \
224 (d) = (h)->data[0]; \
225 heap_del(h, 0, cmp); \
230 #define heap_resort(heap, cmp) \
233 for (_i = (ssize_t) (heap)->used / 2 - 1; _i >= 0; --_i) \
234 heap_sift_down(heap, _i, cmp); \
238 * Simple array based allocator - preallocates a number of elements and you can
239 * never allocate more than that, also has no locking.
241 * Handy because if you know you only need a fixed number of elements you don't
242 * have to worry about memory allocation failure, and sometimes a mempool isn't
245 * We treat the free elements as entries in a singly linked list, and the
246 * freelist as a stack - allocating and freeing push and pop off the freelist.
249 #define DECLARE_ARRAY_ALLOCATOR(type, name, size) \
255 #define array_alloc(array) \
257 typeof((array)->freelist) _ret = (array)->freelist; \
260 (array)->freelist = *((typeof((array)->freelist) *) _ret);\
265 #define array_free(array, ptr) \
267 typeof((array)->freelist) _ptr = ptr; \
269 *((typeof((array)->freelist) *) _ptr) = (array)->freelist; \
270 (array)->freelist = _ptr; \
273 #define array_allocator_init(array) \
275 typeof((array)->freelist) _i; \
277 BUILD_BUG_ON(sizeof((array)->data[0]) < sizeof(void *)); \
278 (array)->freelist = NULL; \
280 for (_i = (array)->data; \
281 _i < (array)->data + ARRAY_SIZE((array)->data); \
283 array_free(array, _i); \
286 #define array_freelist_empty(array) ((array)->freelist == NULL)
288 #define ANYSINT_MAX(t) \
289 ((((t) 1 << (sizeof(t) * 8 - 2)) - (t) 1) * (t) 2 + (t) 1)
291 int bch2_strtoint_h(const char *, int *);
292 int bch2_strtouint_h(const char *, unsigned int *);
293 int bch2_strtoll_h(const char *, long long *);
294 int bch2_strtoull_h(const char *, unsigned long long *);
296 static inline int bch2_strtol_h(const char *cp, long *res)
298 #if BITS_PER_LONG == 32
299 return bch2_strtoint_h(cp, (int *) res);
301 return bch2_strtoll_h(cp, (long long *) res);
305 static inline int bch2_strtoul_h(const char *cp, long *res)
307 #if BITS_PER_LONG == 32
308 return bch2_strtouint_h(cp, (unsigned int *) res);
310 return bch2_strtoull_h(cp, (unsigned long long *) res);
314 #define strtoi_h(cp, res) \
315 ( type_is(*res, int) ? bch2_strtoint_h(cp, (void *) res)\
316 : type_is(*res, long) ? bch2_strtol_h(cp, (void *) res)\
317 : type_is(*res, long long) ? bch2_strtoll_h(cp, (void *) res)\
318 : type_is(*res, unsigned) ? bch2_strtouint_h(cp, (void *) res)\
319 : type_is(*res, unsigned long) ? bch2_strtoul_h(cp, (void *) res)\
320 : type_is(*res, unsigned long long) ? bch2_strtoull_h(cp, (void *) res)\
323 #define strtoul_safe(cp, var) \
326 int _r = kstrtoul(cp, 10, &_v); \
332 #define strtoul_safe_clamp(cp, var, min, max) \
335 int _r = kstrtoul(cp, 10, &_v); \
337 var = clamp_t(typeof(var), _v, min, max); \
341 #define strtoul_safe_restrict(cp, var, min, max) \
344 int _r = kstrtoul(cp, 10, &_v); \
345 if (!_r && _v >= min && _v <= max) \
352 #define snprint(buf, size, var) \
353 snprintf(buf, size, \
354 type_is(var, int) ? "%i\n" \
355 : type_is(var, unsigned) ? "%u\n" \
356 : type_is(var, long) ? "%li\n" \
357 : type_is(var, unsigned long) ? "%lu\n" \
358 : type_is(var, s64) ? "%lli\n" \
359 : type_is(var, u64) ? "%llu\n" \
360 : type_is(var, char *) ? "%s\n" \
363 ssize_t bch2_hprint(char *buf, s64 v);
365 bool bch2_is_zero(const void *, size_t);
367 ssize_t bch2_scnprint_string_list(char *, size_t, const char * const[], size_t);
369 ssize_t bch2_read_string_list(const char *, const char * const[]);
371 ssize_t bch2_scnprint_flag_list(char *, size_t, const char * const[], u64);
372 u64 bch2_read_flag_list(char *, const char * const[]);
378 * all fields are in nanoseconds, averages are ewmas stored left shifted
383 u64 average_duration;
384 u64 average_frequency;
388 void bch2_time_stats_clear(struct time_stats *stats);
389 void __bch2_time_stats_update(struct time_stats *stats, u64 time);
390 void bch2_time_stats_update(struct time_stats *stats, u64 time);
392 static inline unsigned local_clock_us(void)
394 return local_clock() >> 10;
397 #define NSEC_PER_ns 1L
398 #define NSEC_PER_us NSEC_PER_USEC
399 #define NSEC_PER_ms NSEC_PER_MSEC
400 #define NSEC_PER_sec NSEC_PER_SEC
402 #define __print_time_stat(stats, name, stat, units) \
403 sysfs_print(name ## _ ## stat ## _ ## units, \
404 div_u64((stats)->stat >> 8, NSEC_PER_ ## units))
406 #define sysfs_print_time_stats(stats, name, \
410 __print_time_stat(stats, name, \
411 average_frequency, frequency_units); \
412 __print_time_stat(stats, name, \
413 average_duration, duration_units); \
414 sysfs_print(name ## _ ##count, (stats)->count); \
415 sysfs_print(name ## _ ##last_duration ## _ ## duration_units, \
416 div_u64((stats)->last_duration, \
417 NSEC_PER_ ## duration_units)); \
418 sysfs_print(name ## _ ##max_duration ## _ ## duration_units, \
419 div_u64((stats)->max_duration, \
420 NSEC_PER_ ## duration_units)); \
422 sysfs_print(name ## _last_ ## frequency_units, (stats)->last \
423 ? div_s64(local_clock() - (stats)->last, \
424 NSEC_PER_ ## frequency_units) \
428 #define sysfs_clear_time_stats(stats, name) \
430 if (attr == &sysfs_ ## name ## _clear) \
431 bch2_time_stats_clear(stats); \
434 #define sysfs_time_stats_attribute(name, \
437 write_attribute(name ## _clear); \
438 read_attribute(name ## _count); \
439 read_attribute(name ## _average_frequency_ ## frequency_units); \
440 read_attribute(name ## _average_duration_ ## duration_units); \
441 read_attribute(name ## _last_duration_ ## duration_units); \
442 read_attribute(name ## _max_duration_ ## duration_units); \
443 read_attribute(name ## _last_ ## frequency_units)
445 #define sysfs_time_stats_attribute_list(name, \
448 &sysfs_ ## name ## _clear, \
449 &sysfs_ ## name ## _count, \
450 &sysfs_ ## name ## _average_frequency_ ## frequency_units, \
451 &sysfs_ ## name ## _average_duration_ ## duration_units, \
452 &sysfs_ ## name ## _last_duration_ ## duration_units, \
453 &sysfs_ ## name ## _max_duration_ ## duration_units, \
454 &sysfs_ ## name ## _last_ ## frequency_units,
456 #define ewma_add(ewma, val, weight) \
458 typeof(ewma) _ewma = (ewma); \
459 typeof(weight) _weight = (weight); \
461 (((_ewma << _weight) - _ewma) + (val)) >> _weight; \
464 struct bch_ratelimit {
465 /* Next time we want to do some work, in nanoseconds */
469 * Rate at which we want to do work, in units per nanosecond
470 * The units here correspond to the units passed to
471 * bch2_ratelimit_increment()
476 static inline void bch2_ratelimit_reset(struct bch_ratelimit *d)
478 d->next = local_clock();
481 u64 bch2_ratelimit_delay(struct bch_ratelimit *);
482 void bch2_ratelimit_increment(struct bch_ratelimit *, u64);
483 int bch2_ratelimit_wait_freezable_stoppable(struct bch_ratelimit *);
485 struct bch_pd_controller {
486 struct bch_ratelimit rate;
487 unsigned long last_update;
490 s64 smoothed_derivative;
492 unsigned p_term_inverse;
496 /* for exporting to sysfs (no effect on behavior) */
498 s64 last_proportional;
502 /* If true, the rate will not increase if bch2_ratelimit_delay()
503 * is not being called often enough. */
507 void bch2_pd_controller_update(struct bch_pd_controller *, s64, s64, int);
508 void bch2_pd_controller_init(struct bch_pd_controller *);
509 size_t bch2_pd_controller_print_debug(struct bch_pd_controller *, char *);
511 #define sysfs_pd_controller_attribute(name) \
512 rw_attribute(name##_rate); \
513 rw_attribute(name##_rate_bytes); \
514 rw_attribute(name##_rate_d_term); \
515 rw_attribute(name##_rate_p_term_inverse); \
516 read_attribute(name##_rate_debug)
518 #define sysfs_pd_controller_files(name) \
519 &sysfs_##name##_rate, \
520 &sysfs_##name##_rate_bytes, \
521 &sysfs_##name##_rate_d_term, \
522 &sysfs_##name##_rate_p_term_inverse, \
523 &sysfs_##name##_rate_debug
525 #define sysfs_pd_controller_show(name, var) \
527 sysfs_hprint(name##_rate, (var)->rate.rate); \
528 sysfs_print(name##_rate_bytes, (var)->rate.rate); \
529 sysfs_print(name##_rate_d_term, (var)->d_term); \
530 sysfs_print(name##_rate_p_term_inverse, (var)->p_term_inverse); \
532 if (attr == &sysfs_##name##_rate_debug) \
533 return bch2_pd_controller_print_debug(var, buf); \
536 #define sysfs_pd_controller_store(name, var) \
538 sysfs_strtoul_clamp(name##_rate, \
539 (var)->rate.rate, 1, UINT_MAX); \
540 sysfs_strtoul_clamp(name##_rate_bytes, \
541 (var)->rate.rate, 1, UINT_MAX); \
542 sysfs_strtoul(name##_rate_d_term, (var)->d_term); \
543 sysfs_strtoul_clamp(name##_rate_p_term_inverse, \
544 (var)->p_term_inverse, 1, INT_MAX); \
547 #define __DIV_SAFE(n, d, zero) \
549 typeof(n) _n = (n); \
550 typeof(d) _d = (d); \
551 _d ? _n / _d : zero; \
554 #define DIV_SAFE(n, d) __DIV_SAFE(n, d, 0)
556 #define container_of_or_null(ptr, type, member) \
558 typeof(ptr) _ptr = ptr; \
559 _ptr ? container_of(_ptr, type, member) : NULL; \
562 #define RB_INSERT(root, new, member, cmp) \
565 struct rb_node **n = &(root)->rb_node, *parent = NULL; \
571 this = container_of(*n, typeof(*(new)), member); \
572 res = cmp(new, this); \
580 rb_link_node(&(new)->member, parent, n); \
581 rb_insert_color(&(new)->member, root); \
587 #define RB_SEARCH(root, search, member, cmp) \
589 struct rb_node *n = (root)->rb_node; \
590 typeof(&(search)) this, ret = NULL; \
594 this = container_of(n, typeof(search), member); \
595 res = cmp(&(search), this); \
607 #define RB_GREATER(root, search, member, cmp) \
609 struct rb_node *n = (root)->rb_node; \
610 typeof(&(search)) this, ret = NULL; \
614 this = container_of(n, typeof(search), member); \
615 res = cmp(&(search), this); \
625 #define RB_FIRST(root, type, member) \
626 container_of_or_null(rb_first(root), type, member)
628 #define RB_LAST(root, type, member) \
629 container_of_or_null(rb_last(root), type, member)
631 #define RB_NEXT(ptr, member) \
632 container_of_or_null(rb_next(&(ptr)->member), typeof(*ptr), member)
634 #define RB_PREV(ptr, member) \
635 container_of_or_null(rb_prev(&(ptr)->member), typeof(*ptr), member)
637 /* Does linear interpolation between powers of two */
638 static inline unsigned fract_exp_two(unsigned x, unsigned fract_bits)
640 unsigned fract = x & ~(~0 << fract_bits);
644 x += (x * fract) >> fract_bits;
649 void bch2_bio_map(struct bio *bio, void *base);
651 static inline sector_t bdev_sectors(struct block_device *bdev)
653 return bdev->bd_inode->i_size >> 9;
656 #define closure_bio_submit(bio, cl) \
662 #define kthread_wait_freezable(cond) \
666 set_current_state(TASK_INTERRUPTIBLE); \
667 if (kthread_should_stop()) { \
678 set_current_state(TASK_RUNNING); \
682 size_t bch2_rand_range(size_t);
684 void memcpy_to_bio(struct bio *, struct bvec_iter, void *);
685 void memcpy_from_bio(void *, struct bio *, struct bvec_iter);
687 static inline void __memcpy_u64s(void *dst, const void *src,
692 asm volatile("rep ; movsq"
693 : "=&c" (d0), "=&D" (d1), "=&S" (d2)
694 : "0" (u64s), "1" (dst), "2" (src)
705 static inline void memcpy_u64s(void *dst, const void *src,
708 EBUG_ON(!(dst >= src + u64s * sizeof(u64) ||
709 dst + u64s * sizeof(u64) <= src));
711 __memcpy_u64s(dst, src, u64s);
714 static inline void __memmove_u64s_down(void *dst, const void *src,
717 __memcpy_u64s(dst, src, u64s);
720 static inline void memmove_u64s_down(void *dst, const void *src,
725 __memmove_u64s_down(dst, src, u64s);
728 static inline void __memmove_u64s_up(void *_dst, const void *_src,
731 u64 *dst = (u64 *) _dst + u64s - 1;
732 u64 *src = (u64 *) _src + u64s - 1;
736 asm volatile("std ;\n"
739 : "=&c" (d0), "=&D" (d1), "=&S" (d2)
740 : "0" (u64s), "1" (dst), "2" (src)
748 static inline void memmove_u64s_up(void *dst, const void *src,
753 __memmove_u64s_up(dst, src, u64s);
756 static inline void memmove_u64s(void *dst, const void *src,
760 __memmove_u64s_down(dst, src, u64s);
762 __memmove_u64s_up(dst, src, u64s);
765 static inline struct bio_vec next_contig_bvec(struct bio *bio,
766 struct bvec_iter *iter)
768 struct bio_vec bv = bio_iter_iovec(bio, *iter);
770 bio_advance_iter(bio, iter, bv.bv_len);
771 #ifndef CONFIG_HIGHMEM
772 while (iter->bi_size) {
773 struct bio_vec next = bio_iter_iovec(bio, *iter);
775 if (page_address(bv.bv_page) + bv.bv_offset + bv.bv_len !=
776 page_address(next.bv_page) + next.bv_offset)
779 bv.bv_len += next.bv_len;
780 bio_advance_iter(bio, iter, next.bv_len);
786 #define __bio_for_each_contig_segment(bv, bio, iter, start) \
787 for (iter = (start); \
789 ((bv = next_contig_bvec((bio), &(iter))), 1);)
791 #define bio_for_each_contig_segment(bv, bio, iter) \
792 __bio_for_each_contig_segment(bv, bio, iter, (bio)->bi_iter)
794 size_t bch_scnmemcpy(char *, size_t, const char *, size_t);
796 void sort_cmp_size(void *base, size_t num, size_t size,
797 int (*cmp_func)(const void *, const void *, size_t),
798 void (*swap_func)(void *, void *, size_t));
800 /* just the memmove, doesn't update @_nr */
801 #define __array_insert_item(_array, _nr, _pos) \
802 memmove(&(_array)[(_pos) + 1], \
804 sizeof((_array)[0]) * ((_nr) - (_pos)))
806 #define array_insert_item(_array, _nr, _pos, _new_item) \
808 __array_insert_item(_array, _nr, _pos); \
810 (_array)[(_pos)] = (_new_item); \
813 #define array_remove_items(_array, _nr, _pos, _nr_to_remove) \
815 (_nr) -= (_nr_to_remove); \
816 memmove(&(_array)[(_pos)], \
817 &(_array)[(_pos) + (_nr_to_remove)], \
818 sizeof((_array)[0]) * ((_nr) - (_pos))); \
821 #define array_remove_item(_array, _nr, _pos) \
822 array_remove_items(_array, _nr, _pos, 1)
824 #define bubble_sort(_base, _nr, _cmp) \
827 bool _swapped = true; \
829 for (_end = (ssize_t) (_nr) - 1; _end > 0 && _swapped; --_end) {\
831 for (_i = 0; _i < _end; _i++) \
832 if (_cmp((_base)[_i], (_base)[_i + 1]) > 0) { \
833 swap((_base)[_i], (_base)[_i + 1]); \
839 #endif /* _BCACHEFS_UTIL_H */