1 // SPDX-License-Identifier: GPL-2.0
3 * random utiility code, for bcache but in theory not specific to bcache
5 * Copyright 2010, 2011 Kent Overstreet <kent.overstreet@gmail.com>
6 * Copyright 2012 Google, Inc.
10 #include <linux/blkdev.h>
11 #include <linux/ctype.h>
12 #include <linux/debugfs.h>
13 #include <linux/freezer.h>
14 #include <linux/kthread.h>
15 #include <linux/log2.h>
16 #include <linux/math64.h>
17 #include <linux/percpu.h>
18 #include <linux/preempt.h>
19 #include <linux/random.h>
20 #include <linux/seq_file.h>
21 #include <linux/string.h>
22 #include <linux/types.h>
23 #include <linux/sched/clock.h>
25 #include "eytzinger.h"
28 #define simple_strtoint(c, end, base) simple_strtol(c, end, base)
29 #define simple_strtouint(c, end, base) simple_strtoul(c, end, base)
31 static const char si_units[] = "?kMGTPEZY";
33 static int __bch2_strtoh(const char *cp, u64 *res,
34 u64 t_max, bool t_signed)
36 bool positive = *cp != '-';
40 if (*cp == '+' || *cp == '-')
50 if (v > U64_MAX - (*cp - '0'))
54 } while (isdigit(*cp));
56 for (u = 1; u < strlen(si_units); u++)
57 if (*cp == si_units[u]) {
68 if (fls64(v) + u * 10 > 64)
89 #define STRTO_H(name, type) \
90 int bch2_ ## name ## _h(const char *cp, type *res) \
93 int ret = __bch2_strtoh(cp, &v, ANYSINT_MAX(type), \
94 ANYSINT_MAX(type) != ((type) ~0ULL)); \
99 STRTO_H(strtoint, int)
100 STRTO_H(strtouint, unsigned int)
101 STRTO_H(strtoll, long long)
102 STRTO_H(strtoull, unsigned long long)
104 ssize_t bch2_hprint(char *buf, s64 v)
109 for (u = 0; v >= 1024 || v <= -1024; u++) {
110 t = v & ~(~0U << 10);
115 return sprintf(buf, "%lli", v);
118 * 103 is magic: t is in the range [-1023, 1023] and we want
119 * to turn it into [-9, 9]
121 if (v < 100 && v > -100)
122 scnprintf(dec, sizeof(dec), ".%i", t / 103);
124 return sprintf(buf, "%lli%s%c", v, dec, si_units[u]);
127 ssize_t bch2_scnprint_string_list(char *buf, size_t size,
128 const char * const list[],
137 for (i = 0; list[i]; i++)
138 out += scnprintf(out, buf + size - out,
139 i == selected ? "[%s] " : "%s ", list[i]);
147 ssize_t bch2_scnprint_flag_list(char *buf, size_t size,
148 const char * const list[], u64 flags)
150 char *out = buf, *end = buf + size;
151 unsigned bit, nr = 0;
159 while (flags && (bit = __ffs(flags)) < nr) {
160 out += scnprintf(out, end - out, "%s,", list[bit]);
170 u64 bch2_read_flag_list(char *opt, const char * const list[])
173 char *p, *s, *d = kstrndup(opt, PAGE_SIZE - 1, GFP_KERNEL);
180 while ((p = strsep(&s, ","))) {
181 int flag = match_string(list, -1, p);
195 bool bch2_is_zero(const void *_p, size_t n)
200 for (i = 0; i < n; i++)
208 #ifndef CONFIG_BCACHEFS_NO_LATENCY_ACCT
209 static void bch2_quantiles_update(struct bch2_quantiles *q, u64 v)
213 while (i < ARRAY_SIZE(q->entries)) {
214 struct bch2_quantile_entry *e = q->entries + i;
216 if (unlikely(!e->step)) {
218 e->step = max_t(unsigned, v / 2, 1024);
219 } else if (e->m > v) {
220 e->m = e->m >= e->step
223 } else if (e->m < v) {
224 e->m = e->m + e->step > e->m
229 if ((e->m > v ? e->m - v : v - e->m) < e->step)
230 e->step = max_t(unsigned, e->step / 2, 1);
235 i = eytzinger0_child(i, v > e->m);
239 static void bch2_time_stats_update_one(struct bch2_time_stats *stats,
244 duration = time_after64(end, start)
246 freq = time_after64(end, stats->last_event)
247 ? end - stats->last_event : 0;
251 stats->average_duration = stats->average_duration
252 ? ewma_add(stats->average_duration, duration, 6)
255 stats->average_frequency = stats->average_frequency
256 ? ewma_add(stats->average_frequency, freq, 6)
259 stats->max_duration = max(stats->max_duration, duration);
261 stats->last_event = end;
263 bch2_quantiles_update(&stats->quantiles, duration);
266 void __bch2_time_stats_update(struct bch2_time_stats *stats, u64 start, u64 end)
270 if (!stats->buffer) {
271 spin_lock_irqsave(&stats->lock, flags);
272 bch2_time_stats_update_one(stats, start, end);
274 if (stats->average_frequency < 32 &&
277 alloc_percpu_gfp(struct bch2_time_stat_buffer,
279 spin_unlock_irqrestore(&stats->lock, flags);
281 struct bch2_time_stat_buffer_entry *i;
282 struct bch2_time_stat_buffer *b;
285 b = this_cpu_ptr(stats->buffer);
287 BUG_ON(b->nr >= ARRAY_SIZE(b->entries));
288 b->entries[b->nr++] = (struct bch2_time_stat_buffer_entry) {
293 if (b->nr == ARRAY_SIZE(b->entries)) {
294 spin_lock_irqsave(&stats->lock, flags);
296 i < b->entries + ARRAY_SIZE(b->entries);
298 bch2_time_stats_update_one(stats, i->start, i->end);
299 spin_unlock_irqrestore(&stats->lock, flags);
309 static const struct time_unit {
314 { "us", NSEC_PER_USEC },
315 { "ms", NSEC_PER_MSEC },
316 { "sec", NSEC_PER_SEC },
319 static const struct time_unit *pick_time_units(u64 ns)
321 const struct time_unit *u;
324 u + 1 < time_units + ARRAY_SIZE(time_units) &&
325 ns >= u[1].nsecs << 1;
332 static size_t pr_time_units(char *buf, size_t len, u64 ns)
334 const struct time_unit *u = pick_time_units(ns);
336 return scnprintf(buf, len, "%llu %s", div_u64(ns, u->nsecs), u->name);
339 size_t bch2_time_stats_print(struct bch2_time_stats *stats, char *buf, size_t len)
341 char *out = buf, *end = buf + len;
342 const struct time_unit *u;
343 u64 freq = READ_ONCE(stats->average_frequency);
347 out += scnprintf(out, end - out, "count:\t\t%llu\n",
349 out += scnprintf(out, end - out, "rate:\t\t%llu/sec\n",
350 freq ? div64_u64(NSEC_PER_SEC, freq) : 0);
352 out += scnprintf(out, end - out, "frequency:\t");
353 out += pr_time_units(out, end - out, freq);
355 out += scnprintf(out, end - out, "\navg duration:\t");
356 out += pr_time_units(out, end - out, stats->average_duration);
358 out += scnprintf(out, end - out, "\nmax duration:\t");
359 out += pr_time_units(out, end - out, stats->max_duration);
361 i = eytzinger0_first(NR_QUANTILES);
362 u = pick_time_units(stats->quantiles.entries[i].m);
364 out += scnprintf(out, end - out, "\nquantiles (%s):\t", u->name);
365 eytzinger0_for_each(i, NR_QUANTILES) {
366 bool is_last = eytzinger0_next(i, NR_QUANTILES) == -1;
368 q = max(stats->quantiles.entries[i].m, last_q);
369 out += scnprintf(out, end - out, "%llu%s",
370 div_u64(q, u->nsecs),
371 is_last ? "\n" : " ");
378 void bch2_time_stats_exit(struct bch2_time_stats *stats)
380 free_percpu(stats->buffer);
383 void bch2_time_stats_init(struct bch2_time_stats *stats)
385 memset(stats, 0, sizeof(*stats));
386 spin_lock_init(&stats->lock);
392 * bch2_ratelimit_delay() - return how long to delay until the next time to do
395 * @d - the struct bch_ratelimit to update
397 * Returns the amount of time to delay by, in jiffies
399 u64 bch2_ratelimit_delay(struct bch_ratelimit *d)
401 u64 now = local_clock();
403 return time_after64(d->next, now)
404 ? nsecs_to_jiffies(d->next - now)
409 * bch2_ratelimit_increment() - increment @d by the amount of work done
411 * @d - the struct bch_ratelimit to update
412 * @done - the amount of work done, in arbitrary units
414 void bch2_ratelimit_increment(struct bch_ratelimit *d, u64 done)
416 u64 now = local_clock();
418 d->next += div_u64(done * NSEC_PER_SEC, d->rate);
420 if (time_before64(now + NSEC_PER_SEC, d->next))
421 d->next = now + NSEC_PER_SEC;
423 if (time_after64(now - NSEC_PER_SEC * 2, d->next))
424 d->next = now - NSEC_PER_SEC * 2;
427 int bch2_ratelimit_wait_freezable_stoppable(struct bch_ratelimit *d)
429 bool kthread = (current->flags & PF_KTHREAD) != 0;
432 u64 delay = bch2_ratelimit_delay(d);
435 set_current_state(TASK_INTERRUPTIBLE);
437 if (kthread && kthread_should_stop())
443 schedule_timeout(delay);
451 * Updates pd_controller. Attempts to scale inputed values to units per second.
452 * @target: desired value
453 * @actual: current value
455 * @sign: 1 or -1; 1 if increasing the rate makes actual go up, -1 if increasing
456 * it makes actual go down.
458 void bch2_pd_controller_update(struct bch_pd_controller *pd,
459 s64 target, s64 actual, int sign)
461 s64 proportional, derivative, change;
463 unsigned long seconds_since_update = (jiffies - pd->last_update) / HZ;
465 if (seconds_since_update == 0)
468 pd->last_update = jiffies;
470 proportional = actual - target;
471 proportional *= seconds_since_update;
472 proportional = div_s64(proportional, pd->p_term_inverse);
474 derivative = actual - pd->last_actual;
475 derivative = div_s64(derivative, seconds_since_update);
476 derivative = ewma_add(pd->smoothed_derivative, derivative,
477 (pd->d_term / seconds_since_update) ?: 1);
478 derivative = derivative * pd->d_term;
479 derivative = div_s64(derivative, pd->p_term_inverse);
481 change = proportional + derivative;
483 /* Don't increase rate if not keeping up */
486 time_after64(local_clock(),
487 pd->rate.next + NSEC_PER_MSEC))
490 change *= (sign * -1);
492 pd->rate.rate = clamp_t(s64, (s64) pd->rate.rate + change,
495 pd->last_actual = actual;
496 pd->last_derivative = derivative;
497 pd->last_proportional = proportional;
498 pd->last_change = change;
499 pd->last_target = target;
502 void bch2_pd_controller_init(struct bch_pd_controller *pd)
504 pd->rate.rate = 1024;
505 pd->last_update = jiffies;
506 pd->p_term_inverse = 6000;
508 pd->d_smooth = pd->d_term;
509 pd->backpressure = 1;
512 size_t bch2_pd_controller_print_debug(struct bch_pd_controller *pd, char *buf)
514 /* 2^64 - 1 is 20 digits, plus null byte */
518 char proportional[21];
523 bch2_hprint(rate, pd->rate.rate);
524 bch2_hprint(actual, pd->last_actual);
525 bch2_hprint(target, pd->last_target);
526 bch2_hprint(proportional, pd->last_proportional);
527 bch2_hprint(derivative, pd->last_derivative);
528 bch2_hprint(change, pd->last_change);
530 next_io = div64_s64(pd->rate.next - local_clock(), NSEC_PER_MSEC);
536 "proportional:\t%s\n"
538 "change:\t\t%s/sec\n"
539 "next io:\t%llims\n",
540 rate, target, actual, proportional,
541 derivative, change, next_io);
546 void bch2_bio_map(struct bio *bio, void *base)
548 size_t size = bio->bi_iter.bi_size;
549 struct bio_vec *bv = bio->bi_io_vec;
551 BUG_ON(!bio->bi_iter.bi_size);
552 BUG_ON(bio->bi_vcnt);
554 bv->bv_offset = base ? offset_in_page(base) : 0;
557 for (; size; bio->bi_vcnt++, bv++) {
559 start: bv->bv_len = min_t(size_t, PAGE_SIZE - bv->bv_offset,
561 BUG_ON(bio->bi_vcnt >= bio->bi_max_vecs);
563 bv->bv_page = is_vmalloc_addr(base)
564 ? vmalloc_to_page(base)
565 : virt_to_page(base);
574 int bch2_bio_alloc_pages(struct bio *bio, size_t size, gfp_t gfp_mask)
577 struct page *page = alloc_pages(gfp_mask, 0);
578 unsigned len = min_t(size_t, PAGE_SIZE, size);
583 if (unlikely(!bio_add_page(bio, page, len, 0))) {
594 size_t bch2_rand_range(size_t max)
602 rand = get_random_long();
603 rand &= roundup_pow_of_two(max) - 1;
604 } while (rand >= max);
609 void memcpy_to_bio(struct bio *dst, struct bvec_iter dst_iter, void *src)
612 struct bvec_iter iter;
614 __bio_for_each_segment(bv, dst, iter, dst_iter) {
615 void *dstp = kmap_atomic(bv.bv_page);
616 memcpy(dstp + bv.bv_offset, src, bv.bv_len);
623 void memcpy_from_bio(void *dst, struct bio *src, struct bvec_iter src_iter)
626 struct bvec_iter iter;
628 __bio_for_each_segment(bv, src, iter, src_iter) {
629 void *srcp = kmap_atomic(bv.bv_page);
630 memcpy(dst, srcp + bv.bv_offset, bv.bv_len);
637 size_t bch_scnmemcpy(char *buf, size_t size, const char *src, size_t len)
644 n = min(size - 1, len);
651 #include "eytzinger.h"
653 static int alignment_ok(const void *base, size_t align)
655 return IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) ||
656 ((unsigned long)base & (align - 1)) == 0;
659 static void u32_swap(void *a, void *b, size_t size)
662 *(u32 *)a = *(u32 *)b;
666 static void u64_swap(void *a, void *b, size_t size)
669 *(u64 *)a = *(u64 *)b;
673 static void generic_swap(void *a, void *b, size_t size)
679 *(char *)a++ = *(char *)b;
681 } while (--size > 0);
684 static inline int do_cmp(void *base, size_t n, size_t size,
685 int (*cmp_func)(const void *, const void *, size_t),
688 return cmp_func(base + inorder_to_eytzinger0(l, n) * size,
689 base + inorder_to_eytzinger0(r, n) * size,
693 static inline void do_swap(void *base, size_t n, size_t size,
694 void (*swap_func)(void *, void *, size_t),
697 swap_func(base + inorder_to_eytzinger0(l, n) * size,
698 base + inorder_to_eytzinger0(r, n) * size,
702 void eytzinger0_sort(void *base, size_t n, size_t size,
703 int (*cmp_func)(const void *, const void *, size_t),
704 void (*swap_func)(void *, void *, size_t))
709 if (size == 4 && alignment_ok(base, 4))
710 swap_func = u32_swap;
711 else if (size == 8 && alignment_ok(base, 8))
712 swap_func = u64_swap;
714 swap_func = generic_swap;
718 for (i = n / 2 - 1; i >= 0; --i) {
719 for (r = i; r * 2 + 1 < n; r = c) {
723 do_cmp(base, n, size, cmp_func, c, c + 1) < 0)
726 if (do_cmp(base, n, size, cmp_func, r, c) >= 0)
729 do_swap(base, n, size, swap_func, r, c);
734 for (i = n - 1; i > 0; --i) {
735 do_swap(base, n, size, swap_func, 0, i);
737 for (r = 0; r * 2 + 1 < i; r = c) {
741 do_cmp(base, n, size, cmp_func, c, c + 1) < 0)
744 if (do_cmp(base, n, size, cmp_func, r, c) >= 0)
747 do_swap(base, n, size, swap_func, r, c);
752 void sort_cmp_size(void *base, size_t num, size_t size,
753 int (*cmp_func)(const void *, const void *, size_t),
754 void (*swap_func)(void *, void *, size_t size))
756 /* pre-scale counters for performance */
757 int i = (num/2 - 1) * size, n = num * size, c, r;
760 if (size == 4 && alignment_ok(base, 4))
761 swap_func = u32_swap;
762 else if (size == 8 && alignment_ok(base, 8))
763 swap_func = u64_swap;
765 swap_func = generic_swap;
769 for ( ; i >= 0; i -= size) {
770 for (r = i; r * 2 + size < n; r = c) {
773 cmp_func(base + c, base + c + size, size) < 0)
775 if (cmp_func(base + r, base + c, size) >= 0)
777 swap_func(base + r, base + c, size);
782 for (i = n - size; i > 0; i -= size) {
783 swap_func(base, base + i, size);
784 for (r = 0; r * 2 + size < i; r = c) {
787 cmp_func(base + c, base + c + size, size) < 0)
789 if (cmp_func(base + r, base + c, size) >= 0)
791 swap_func(base + r, base + c, size);
796 static void mempool_free_vp(void *element, void *pool_data)
798 size_t size = (size_t) pool_data;
800 vpfree(element, size);
803 static void *mempool_alloc_vp(gfp_t gfp_mask, void *pool_data)
805 size_t size = (size_t) pool_data;
807 return vpmalloc(size, gfp_mask);
810 int mempool_init_kvpmalloc_pool(mempool_t *pool, int min_nr, size_t size)
812 return size < PAGE_SIZE
813 ? mempool_init_kmalloc_pool(pool, min_nr, size)
814 : mempool_init(pool, min_nr, mempool_alloc_vp,
815 mempool_free_vp, (void *) size);
819 void eytzinger1_test(void)
821 unsigned inorder, eytz, size;
823 pr_info("1 based eytzinger test:");
828 unsigned extra = eytzinger1_extra(size);
831 pr_info("tree size %u", size);
833 BUG_ON(eytzinger1_prev(0, size) != eytzinger1_last(size));
834 BUG_ON(eytzinger1_next(0, size) != eytzinger1_first(size));
836 BUG_ON(eytzinger1_prev(eytzinger1_first(size), size) != 0);
837 BUG_ON(eytzinger1_next(eytzinger1_last(size), size) != 0);
840 eytzinger1_for_each(eytz, size) {
841 BUG_ON(__inorder_to_eytzinger1(inorder, size, extra) != eytz);
842 BUG_ON(__eytzinger1_to_inorder(eytz, size, extra) != inorder);
843 BUG_ON(eytz != eytzinger1_last(size) &&
844 eytzinger1_prev(eytzinger1_next(eytz, size), size) != eytz);
851 void eytzinger0_test(void)
854 unsigned inorder, eytz, size;
856 pr_info("0 based eytzinger test:");
861 unsigned extra = eytzinger0_extra(size);
864 pr_info("tree size %u", size);
866 BUG_ON(eytzinger0_prev(-1, size) != eytzinger0_last(size));
867 BUG_ON(eytzinger0_next(-1, size) != eytzinger0_first(size));
869 BUG_ON(eytzinger0_prev(eytzinger0_first(size), size) != -1);
870 BUG_ON(eytzinger0_next(eytzinger0_last(size), size) != -1);
873 eytzinger0_for_each(eytz, size) {
874 BUG_ON(__inorder_to_eytzinger0(inorder, size, extra) != eytz);
875 BUG_ON(__eytzinger0_to_inorder(eytz, size, extra) != inorder);
876 BUG_ON(eytz != eytzinger0_last(size) &&
877 eytzinger0_prev(eytzinger0_next(eytz, size), size) != eytz);
884 static inline int cmp_u16(const void *_l, const void *_r, size_t size)
886 const u16 *l = _l, *r = _r;
888 return (*l > *r) - (*r - *l);
891 static void eytzinger0_find_test_val(u16 *test_array, unsigned nr, u16 search)
893 int i, c1 = -1, c2 = -1;
896 r = eytzinger0_find_le(test_array, nr,
897 sizeof(test_array[0]),
902 for (i = 0; i < nr; i++)
903 if (test_array[i] <= search && test_array[i] > c2)
907 eytzinger0_for_each(i, nr)
908 pr_info("[%3u] = %12u", i, test_array[i]);
909 pr_info("find_le(%2u) -> [%2zi] = %2i should be %2i",
914 void eytzinger0_find_test(void)
916 unsigned i, nr, allocated = 1 << 12;
917 u16 *test_array = kmalloc_array(allocated, sizeof(test_array[0]), GFP_KERNEL);
919 for (nr = 1; nr < allocated; nr++) {
920 pr_info("testing %u elems", nr);
922 get_random_bytes(test_array, nr * sizeof(test_array[0]));
923 eytzinger0_sort(test_array, nr, sizeof(test_array[0]), cmp_u16, NULL);
925 /* verify array is sorted correctly: */
926 eytzinger0_for_each(i, nr)
927 BUG_ON(i != eytzinger0_last(nr) &&
928 test_array[i] > test_array[eytzinger0_next(i, nr)]);
930 for (i = 0; i < U16_MAX; i += 1 << 12)
931 eytzinger0_find_test_val(test_array, nr, i);
933 for (i = 0; i < nr; i++) {
934 eytzinger0_find_test_val(test_array, nr, test_array[i] - 1);
935 eytzinger0_find_test_val(test_array, nr, test_array[i]);
936 eytzinger0_find_test_val(test_array, nr, test_array[i] + 1);