xref: /linux/fs/bcachefs/util.h (revision 79ac11393328fb1717d17c12e3c0eef0e9fa0647)
1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef _BCACHEFS_UTIL_H
3 #define _BCACHEFS_UTIL_H
4 
5 #include <linux/bio.h>
6 #include <linux/blkdev.h>
7 #include <linux/closure.h>
8 #include <linux/errno.h>
9 #include <linux/freezer.h>
10 #include <linux/kernel.h>
11 #include <linux/sched/clock.h>
12 #include <linux/llist.h>
13 #include <linux/log2.h>
14 #include <linux/percpu.h>
15 #include <linux/preempt.h>
16 #include <linux/ratelimit.h>
17 #include <linux/slab.h>
18 #include <linux/vmalloc.h>
19 #include <linux/workqueue.h>
20 
21 #include "mean_and_variance.h"
22 
23 #include "darray.h"
24 
25 struct closure;
26 
27 #ifdef CONFIG_BCACHEFS_DEBUG
28 #define EBUG_ON(cond)		BUG_ON(cond)
29 #else
30 #define EBUG_ON(cond)
31 #endif
32 
33 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
34 #define CPU_BIG_ENDIAN		0
35 #elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
36 #define CPU_BIG_ENDIAN		1
37 #endif
38 
39 /* type hackery */
40 
41 #define type_is_exact(_val, _type)					\
42 	__builtin_types_compatible_p(typeof(_val), _type)
43 
44 #define type_is(_val, _type)						\
45 	(__builtin_types_compatible_p(typeof(_val), _type) ||		\
46 	 __builtin_types_compatible_p(typeof(_val), const _type))
47 
48 /* Userspace doesn't align allocations as nicely as the kernel allocators: */
49 static inline size_t buf_pages(void *p, size_t len)
50 {
51 	return DIV_ROUND_UP(len +
52 			    ((unsigned long) p & (PAGE_SIZE - 1)),
53 			    PAGE_SIZE);
54 }
55 
56 static inline void vpfree(void *p, size_t size)
57 {
58 	if (is_vmalloc_addr(p))
59 		vfree(p);
60 	else
61 		free_pages((unsigned long) p, get_order(size));
62 }
63 
64 static inline void *vpmalloc(size_t size, gfp_t gfp_mask)
65 {
66 	return (void *) __get_free_pages(gfp_mask|__GFP_NOWARN,
67 					 get_order(size)) ?:
68 		__vmalloc(size, gfp_mask);
69 }
70 
71 static inline void kvpfree(void *p, size_t size)
72 {
73 	if (size < PAGE_SIZE)
74 		kfree(p);
75 	else
76 		vpfree(p, size);
77 }
78 
79 static inline void *kvpmalloc(size_t size, gfp_t gfp_mask)
80 {
81 	return size < PAGE_SIZE
82 		? kmalloc(size, gfp_mask)
83 		: vpmalloc(size, gfp_mask);
84 }
85 
86 int mempool_init_kvpmalloc_pool(mempool_t *, int, size_t);
87 
88 #define HEAP(type)							\
89 struct {								\
90 	size_t size, used;						\
91 	type *data;							\
92 }
93 
94 #define DECLARE_HEAP(type, name) HEAP(type) name
95 
96 #define init_heap(heap, _size, gfp)					\
97 ({									\
98 	(heap)->used = 0;						\
99 	(heap)->size = (_size);						\
100 	(heap)->data = kvpmalloc((heap)->size * sizeof((heap)->data[0]),\
101 				 (gfp));				\
102 })
103 
104 #define free_heap(heap)							\
105 do {									\
106 	kvpfree((heap)->data, (heap)->size * sizeof((heap)->data[0]));	\
107 	(heap)->data = NULL;						\
108 } while (0)
109 
110 #define heap_set_backpointer(h, i, _fn)					\
111 do {									\
112 	void (*fn)(typeof(h), size_t) = _fn;				\
113 	if (fn)								\
114 		fn(h, i);						\
115 } while (0)
116 
117 #define heap_swap(h, i, j, set_backpointer)				\
118 do {									\
119 	swap((h)->data[i], (h)->data[j]);				\
120 	heap_set_backpointer(h, i, set_backpointer);			\
121 	heap_set_backpointer(h, j, set_backpointer);			\
122 } while (0)
123 
124 #define heap_peek(h)							\
125 ({									\
126 	EBUG_ON(!(h)->used);						\
127 	(h)->data[0];							\
128 })
129 
130 #define heap_full(h)	((h)->used == (h)->size)
131 
132 #define heap_sift_down(h, i, cmp, set_backpointer)			\
133 do {									\
134 	size_t _c, _j = i;						\
135 									\
136 	for (; _j * 2 + 1 < (h)->used; _j = _c) {			\
137 		_c = _j * 2 + 1;					\
138 		if (_c + 1 < (h)->used &&				\
139 		    cmp(h, (h)->data[_c], (h)->data[_c + 1]) >= 0)	\
140 			_c++;						\
141 									\
142 		if (cmp(h, (h)->data[_c], (h)->data[_j]) >= 0)		\
143 			break;						\
144 		heap_swap(h, _c, _j, set_backpointer);			\
145 	}								\
146 } while (0)
147 
148 #define heap_sift_up(h, i, cmp, set_backpointer)			\
149 do {									\
150 	while (i) {							\
151 		size_t p = (i - 1) / 2;					\
152 		if (cmp(h, (h)->data[i], (h)->data[p]) >= 0)		\
153 			break;						\
154 		heap_swap(h, i, p, set_backpointer);			\
155 		i = p;							\
156 	}								\
157 } while (0)
158 
159 #define __heap_add(h, d, cmp, set_backpointer)				\
160 ({									\
161 	size_t _i = (h)->used++;					\
162 	(h)->data[_i] = d;						\
163 	heap_set_backpointer(h, _i, set_backpointer);			\
164 									\
165 	heap_sift_up(h, _i, cmp, set_backpointer);			\
166 	_i;								\
167 })
168 
169 #define heap_add(h, d, cmp, set_backpointer)				\
170 ({									\
171 	bool _r = !heap_full(h);					\
172 	if (_r)								\
173 		__heap_add(h, d, cmp, set_backpointer);			\
174 	_r;								\
175 })
176 
177 #define heap_add_or_replace(h, new, cmp, set_backpointer)		\
178 do {									\
179 	if (!heap_add(h, new, cmp, set_backpointer) &&			\
180 	    cmp(h, new, heap_peek(h)) >= 0) {				\
181 		(h)->data[0] = new;					\
182 		heap_set_backpointer(h, 0, set_backpointer);		\
183 		heap_sift_down(h, 0, cmp, set_backpointer);		\
184 	}								\
185 } while (0)
186 
187 #define heap_del(h, i, cmp, set_backpointer)				\
188 do {									\
189 	size_t _i = (i);						\
190 									\
191 	BUG_ON(_i >= (h)->used);					\
192 	(h)->used--;							\
193 	if ((_i) < (h)->used) {						\
194 		heap_swap(h, _i, (h)->used, set_backpointer);		\
195 		heap_sift_up(h, _i, cmp, set_backpointer);		\
196 		heap_sift_down(h, _i, cmp, set_backpointer);		\
197 	}								\
198 } while (0)
199 
200 #define heap_pop(h, d, cmp, set_backpointer)				\
201 ({									\
202 	bool _r = (h)->used;						\
203 	if (_r) {							\
204 		(d) = (h)->data[0];					\
205 		heap_del(h, 0, cmp, set_backpointer);			\
206 	}								\
207 	_r;								\
208 })
209 
210 #define heap_resort(heap, cmp, set_backpointer)				\
211 do {									\
212 	ssize_t _i;							\
213 	for (_i = (ssize_t) (heap)->used / 2 -  1; _i >= 0; --_i)	\
214 		heap_sift_down(heap, _i, cmp, set_backpointer);		\
215 } while (0)
216 
217 #define ANYSINT_MAX(t)							\
218 	((((t) 1 << (sizeof(t) * 8 - 2)) - (t) 1) * (t) 2 + (t) 1)
219 
220 #include "printbuf.h"
221 
222 #define prt_vprintf(_out, ...)		bch2_prt_vprintf(_out, __VA_ARGS__)
223 #define prt_printf(_out, ...)		bch2_prt_printf(_out, __VA_ARGS__)
224 #define printbuf_str(_buf)		bch2_printbuf_str(_buf)
225 #define printbuf_exit(_buf)		bch2_printbuf_exit(_buf)
226 
227 #define printbuf_tabstops_reset(_buf)	bch2_printbuf_tabstops_reset(_buf)
228 #define printbuf_tabstop_pop(_buf)	bch2_printbuf_tabstop_pop(_buf)
229 #define printbuf_tabstop_push(_buf, _n)	bch2_printbuf_tabstop_push(_buf, _n)
230 
231 #define printbuf_indent_add(_out, _n)	bch2_printbuf_indent_add(_out, _n)
232 #define printbuf_indent_sub(_out, _n)	bch2_printbuf_indent_sub(_out, _n)
233 
234 #define prt_newline(_out)		bch2_prt_newline(_out)
235 #define prt_tab(_out)			bch2_prt_tab(_out)
236 #define prt_tab_rjust(_out)		bch2_prt_tab_rjust(_out)
237 
238 #define prt_bytes_indented(...)		bch2_prt_bytes_indented(__VA_ARGS__)
239 #define prt_u64(_out, _v)		prt_printf(_out, "%llu", (u64) (_v))
240 #define prt_human_readable_u64(...)	bch2_prt_human_readable_u64(__VA_ARGS__)
241 #define prt_human_readable_s64(...)	bch2_prt_human_readable_s64(__VA_ARGS__)
242 #define prt_units_u64(...)		bch2_prt_units_u64(__VA_ARGS__)
243 #define prt_units_s64(...)		bch2_prt_units_s64(__VA_ARGS__)
244 #define prt_string_option(...)		bch2_prt_string_option(__VA_ARGS__)
245 #define prt_bitflags(...)		bch2_prt_bitflags(__VA_ARGS__)
246 
247 void bch2_pr_time_units(struct printbuf *, u64);
248 void bch2_prt_datetime(struct printbuf *, time64_t);
249 
250 #ifdef __KERNEL__
251 static inline void uuid_unparse_lower(u8 *uuid, char *out)
252 {
253 	sprintf(out, "%pUb", uuid);
254 }
255 #else
256 #include <uuid/uuid.h>
257 #endif
258 
259 static inline void pr_uuid(struct printbuf *out, u8 *uuid)
260 {
261 	char uuid_str[40];
262 
263 	uuid_unparse_lower(uuid, uuid_str);
264 	prt_printf(out, "%s", uuid_str);
265 }
266 
267 int bch2_strtoint_h(const char *, int *);
268 int bch2_strtouint_h(const char *, unsigned int *);
269 int bch2_strtoll_h(const char *, long long *);
270 int bch2_strtoull_h(const char *, unsigned long long *);
271 int bch2_strtou64_h(const char *, u64 *);
272 
273 static inline int bch2_strtol_h(const char *cp, long *res)
274 {
275 #if BITS_PER_LONG == 32
276 	return bch2_strtoint_h(cp, (int *) res);
277 #else
278 	return bch2_strtoll_h(cp, (long long *) res);
279 #endif
280 }
281 
282 static inline int bch2_strtoul_h(const char *cp, long *res)
283 {
284 #if BITS_PER_LONG == 32
285 	return bch2_strtouint_h(cp, (unsigned int *) res);
286 #else
287 	return bch2_strtoull_h(cp, (unsigned long long *) res);
288 #endif
289 }
290 
291 #define strtoi_h(cp, res)						\
292 	( type_is(*res, int)		? bch2_strtoint_h(cp, (void *) res)\
293 	: type_is(*res, long)		? bch2_strtol_h(cp, (void *) res)\
294 	: type_is(*res, long long)	? bch2_strtoll_h(cp, (void *) res)\
295 	: type_is(*res, unsigned)	? bch2_strtouint_h(cp, (void *) res)\
296 	: type_is(*res, unsigned long)	? bch2_strtoul_h(cp, (void *) res)\
297 	: type_is(*res, unsigned long long) ? bch2_strtoull_h(cp, (void *) res)\
298 	: -EINVAL)
299 
300 #define strtoul_safe(cp, var)						\
301 ({									\
302 	unsigned long _v;						\
303 	int _r = kstrtoul(cp, 10, &_v);					\
304 	if (!_r)							\
305 		var = _v;						\
306 	_r;								\
307 })
308 
309 #define strtoul_safe_clamp(cp, var, min, max)				\
310 ({									\
311 	unsigned long _v;						\
312 	int _r = kstrtoul(cp, 10, &_v);					\
313 	if (!_r)							\
314 		var = clamp_t(typeof(var), _v, min, max);		\
315 	_r;								\
316 })
317 
318 #define strtoul_safe_restrict(cp, var, min, max)			\
319 ({									\
320 	unsigned long _v;						\
321 	int _r = kstrtoul(cp, 10, &_v);					\
322 	if (!_r && _v >= min && _v <= max)				\
323 		var = _v;						\
324 	else								\
325 		_r = -EINVAL;						\
326 	_r;								\
327 })
328 
329 #define snprint(out, var)						\
330 	prt_printf(out,							\
331 		   type_is(var, int)		? "%i\n"		\
332 		 : type_is(var, unsigned)	? "%u\n"		\
333 		 : type_is(var, long)		? "%li\n"		\
334 		 : type_is(var, unsigned long)	? "%lu\n"		\
335 		 : type_is(var, s64)		? "%lli\n"		\
336 		 : type_is(var, u64)		? "%llu\n"		\
337 		 : type_is(var, char *)		? "%s\n"		\
338 		 : "%i\n", var)
339 
340 bool bch2_is_zero(const void *, size_t);
341 
342 u64 bch2_read_flag_list(char *, const char * const[]);
343 
344 void bch2_prt_u64_binary(struct printbuf *, u64, unsigned);
345 
346 void bch2_print_string_as_lines(const char *prefix, const char *lines);
347 
348 typedef DARRAY(unsigned long) bch_stacktrace;
349 int bch2_save_backtrace(bch_stacktrace *stack, struct task_struct *);
350 void bch2_prt_backtrace(struct printbuf *, bch_stacktrace *);
351 int bch2_prt_task_backtrace(struct printbuf *, struct task_struct *);
352 
353 #define NR_QUANTILES	15
354 #define QUANTILE_IDX(i)	inorder_to_eytzinger0(i, NR_QUANTILES)
355 #define QUANTILE_FIRST	eytzinger0_first(NR_QUANTILES)
356 #define QUANTILE_LAST	eytzinger0_last(NR_QUANTILES)
357 
358 struct bch2_quantiles {
359 	struct bch2_quantile_entry {
360 		u64	m;
361 		u64	step;
362 	}		entries[NR_QUANTILES];
363 };
364 
365 struct bch2_time_stat_buffer {
366 	unsigned	nr;
367 	struct bch2_time_stat_buffer_entry {
368 		u64	start;
369 		u64	end;
370 	}		entries[32];
371 };
372 
373 struct bch2_time_stats {
374 	spinlock_t	lock;
375 	/* all fields are in nanoseconds */
376 	u64		max_duration;
377 	u64             min_duration;
378 	u64             max_freq;
379 	u64             min_freq;
380 	u64		last_event;
381 	struct bch2_quantiles quantiles;
382 
383 	struct mean_and_variance	  duration_stats;
384 	struct mean_and_variance_weighted duration_stats_weighted;
385 	struct mean_and_variance	  freq_stats;
386 	struct mean_and_variance_weighted freq_stats_weighted;
387 	struct bch2_time_stat_buffer __percpu *buffer;
388 };
389 
390 #ifndef CONFIG_BCACHEFS_NO_LATENCY_ACCT
391 void __bch2_time_stats_update(struct bch2_time_stats *stats, u64, u64);
392 #else
393 static inline void __bch2_time_stats_update(struct bch2_time_stats *stats, u64 start, u64 end) {}
394 #endif
395 
396 static inline void bch2_time_stats_update(struct bch2_time_stats *stats, u64 start)
397 {
398 	__bch2_time_stats_update(stats, start, local_clock());
399 }
400 
401 void bch2_time_stats_to_text(struct printbuf *, struct bch2_time_stats *);
402 
403 void bch2_time_stats_exit(struct bch2_time_stats *);
404 void bch2_time_stats_init(struct bch2_time_stats *);
405 
406 #define ewma_add(ewma, val, weight)					\
407 ({									\
408 	typeof(ewma) _ewma = (ewma);					\
409 	typeof(weight) _weight = (weight);				\
410 									\
411 	(((_ewma << _weight) - _ewma) + (val)) >> _weight;		\
412 })
413 
414 struct bch_ratelimit {
415 	/* Next time we want to do some work, in nanoseconds */
416 	u64			next;
417 
418 	/*
419 	 * Rate at which we want to do work, in units per nanosecond
420 	 * The units here correspond to the units passed to
421 	 * bch2_ratelimit_increment()
422 	 */
423 	unsigned		rate;
424 };
425 
426 static inline void bch2_ratelimit_reset(struct bch_ratelimit *d)
427 {
428 	d->next = local_clock();
429 }
430 
431 u64 bch2_ratelimit_delay(struct bch_ratelimit *);
432 void bch2_ratelimit_increment(struct bch_ratelimit *, u64);
433 
434 struct bch_pd_controller {
435 	struct bch_ratelimit	rate;
436 	unsigned long		last_update;
437 
438 	s64			last_actual;
439 	s64			smoothed_derivative;
440 
441 	unsigned		p_term_inverse;
442 	unsigned		d_smooth;
443 	unsigned		d_term;
444 
445 	/* for exporting to sysfs (no effect on behavior) */
446 	s64			last_derivative;
447 	s64			last_proportional;
448 	s64			last_change;
449 	s64			last_target;
450 
451 	/*
452 	 * If true, the rate will not increase if bch2_ratelimit_delay()
453 	 * is not being called often enough.
454 	 */
455 	bool			backpressure;
456 };
457 
458 void bch2_pd_controller_update(struct bch_pd_controller *, s64, s64, int);
459 void bch2_pd_controller_init(struct bch_pd_controller *);
460 void bch2_pd_controller_debug_to_text(struct printbuf *, struct bch_pd_controller *);
461 
462 #define sysfs_pd_controller_attribute(name)				\
463 	rw_attribute(name##_rate);					\
464 	rw_attribute(name##_rate_bytes);				\
465 	rw_attribute(name##_rate_d_term);				\
466 	rw_attribute(name##_rate_p_term_inverse);			\
467 	read_attribute(name##_rate_debug)
468 
469 #define sysfs_pd_controller_files(name)					\
470 	&sysfs_##name##_rate,						\
471 	&sysfs_##name##_rate_bytes,					\
472 	&sysfs_##name##_rate_d_term,					\
473 	&sysfs_##name##_rate_p_term_inverse,				\
474 	&sysfs_##name##_rate_debug
475 
476 #define sysfs_pd_controller_show(name, var)				\
477 do {									\
478 	sysfs_hprint(name##_rate,		(var)->rate.rate);	\
479 	sysfs_print(name##_rate_bytes,		(var)->rate.rate);	\
480 	sysfs_print(name##_rate_d_term,		(var)->d_term);		\
481 	sysfs_print(name##_rate_p_term_inverse,	(var)->p_term_inverse);	\
482 									\
483 	if (attr == &sysfs_##name##_rate_debug)				\
484 		bch2_pd_controller_debug_to_text(out, var);		\
485 } while (0)
486 
487 #define sysfs_pd_controller_store(name, var)				\
488 do {									\
489 	sysfs_strtoul_clamp(name##_rate,				\
490 			    (var)->rate.rate, 1, UINT_MAX);		\
491 	sysfs_strtoul_clamp(name##_rate_bytes,				\
492 			    (var)->rate.rate, 1, UINT_MAX);		\
493 	sysfs_strtoul(name##_rate_d_term,	(var)->d_term);		\
494 	sysfs_strtoul_clamp(name##_rate_p_term_inverse,			\
495 			    (var)->p_term_inverse, 1, INT_MAX);		\
496 } while (0)
497 
498 #define container_of_or_null(ptr, type, member)				\
499 ({									\
500 	typeof(ptr) _ptr = ptr;						\
501 	_ptr ? container_of(_ptr, type, member) : NULL;			\
502 })
503 
504 /* Does linear interpolation between powers of two */
505 static inline unsigned fract_exp_two(unsigned x, unsigned fract_bits)
506 {
507 	unsigned fract = x & ~(~0 << fract_bits);
508 
509 	x >>= fract_bits;
510 	x   = 1 << x;
511 	x  += (x * fract) >> fract_bits;
512 
513 	return x;
514 }
515 
516 void bch2_bio_map(struct bio *bio, void *base, size_t);
517 int bch2_bio_alloc_pages(struct bio *, size_t, gfp_t);
518 
519 static inline sector_t bdev_sectors(struct block_device *bdev)
520 {
521 	return bdev->bd_inode->i_size >> 9;
522 }
523 
524 #define closure_bio_submit(bio, cl)					\
525 do {									\
526 	closure_get(cl);						\
527 	submit_bio(bio);						\
528 } while (0)
529 
530 #define kthread_wait(cond)						\
531 ({									\
532 	int _ret = 0;							\
533 									\
534 	while (1) {							\
535 		set_current_state(TASK_INTERRUPTIBLE);			\
536 		if (kthread_should_stop()) {				\
537 			_ret = -1;					\
538 			break;						\
539 		}							\
540 									\
541 		if (cond)						\
542 			break;						\
543 									\
544 		schedule();						\
545 	}								\
546 	set_current_state(TASK_RUNNING);				\
547 	_ret;								\
548 })
549 
550 #define kthread_wait_freezable(cond)					\
551 ({									\
552 	int _ret = 0;							\
553 	while (1) {							\
554 		set_current_state(TASK_INTERRUPTIBLE);			\
555 		if (kthread_should_stop()) {				\
556 			_ret = -1;					\
557 			break;						\
558 		}							\
559 									\
560 		if (cond)						\
561 			break;						\
562 									\
563 		schedule();						\
564 		try_to_freeze();					\
565 	}								\
566 	set_current_state(TASK_RUNNING);				\
567 	_ret;								\
568 })
569 
570 size_t bch2_rand_range(size_t);
571 
572 void memcpy_to_bio(struct bio *, struct bvec_iter, const void *);
573 void memcpy_from_bio(void *, struct bio *, struct bvec_iter);
574 
575 static inline void memcpy_u64s_small(void *dst, const void *src,
576 				     unsigned u64s)
577 {
578 	u64 *d = dst;
579 	const u64 *s = src;
580 
581 	while (u64s--)
582 		*d++ = *s++;
583 }
584 
585 static inline void __memcpy_u64s(void *dst, const void *src,
586 				 unsigned u64s)
587 {
588 #ifdef CONFIG_X86_64
589 	long d0, d1, d2;
590 
591 	asm volatile("rep ; movsq"
592 		     : "=&c" (d0), "=&D" (d1), "=&S" (d2)
593 		     : "0" (u64s), "1" (dst), "2" (src)
594 		     : "memory");
595 #else
596 	u64 *d = dst;
597 	const u64 *s = src;
598 
599 	while (u64s--)
600 		*d++ = *s++;
601 #endif
602 }
603 
604 static inline void memcpy_u64s(void *dst, const void *src,
605 			       unsigned u64s)
606 {
607 	EBUG_ON(!(dst >= src + u64s * sizeof(u64) ||
608 		 dst + u64s * sizeof(u64) <= src));
609 
610 	__memcpy_u64s(dst, src, u64s);
611 }
612 
613 static inline void __memmove_u64s_down(void *dst, const void *src,
614 				       unsigned u64s)
615 {
616 	__memcpy_u64s(dst, src, u64s);
617 }
618 
619 static inline void memmove_u64s_down(void *dst, const void *src,
620 				     unsigned u64s)
621 {
622 	EBUG_ON(dst > src);
623 
624 	__memmove_u64s_down(dst, src, u64s);
625 }
626 
627 static inline void __memmove_u64s_down_small(void *dst, const void *src,
628 				       unsigned u64s)
629 {
630 	memcpy_u64s_small(dst, src, u64s);
631 }
632 
633 static inline void memmove_u64s_down_small(void *dst, const void *src,
634 				     unsigned u64s)
635 {
636 	EBUG_ON(dst > src);
637 
638 	__memmove_u64s_down_small(dst, src, u64s);
639 }
640 
641 static inline void __memmove_u64s_up_small(void *_dst, const void *_src,
642 					   unsigned u64s)
643 {
644 	u64 *dst = (u64 *) _dst + u64s;
645 	u64 *src = (u64 *) _src + u64s;
646 
647 	while (u64s--)
648 		*--dst = *--src;
649 }
650 
651 static inline void memmove_u64s_up_small(void *dst, const void *src,
652 					 unsigned u64s)
653 {
654 	EBUG_ON(dst < src);
655 
656 	__memmove_u64s_up_small(dst, src, u64s);
657 }
658 
659 static inline void __memmove_u64s_up(void *_dst, const void *_src,
660 				     unsigned u64s)
661 {
662 	u64 *dst = (u64 *) _dst + u64s - 1;
663 	u64 *src = (u64 *) _src + u64s - 1;
664 
665 #ifdef CONFIG_X86_64
666 	long d0, d1, d2;
667 
668 	asm volatile("std ;\n"
669 		     "rep ; movsq\n"
670 		     "cld ;\n"
671 		     : "=&c" (d0), "=&D" (d1), "=&S" (d2)
672 		     : "0" (u64s), "1" (dst), "2" (src)
673 		     : "memory");
674 #else
675 	while (u64s--)
676 		*dst-- = *src--;
677 #endif
678 }
679 
680 static inline void memmove_u64s_up(void *dst, const void *src,
681 				   unsigned u64s)
682 {
683 	EBUG_ON(dst < src);
684 
685 	__memmove_u64s_up(dst, src, u64s);
686 }
687 
688 static inline void memmove_u64s(void *dst, const void *src,
689 				unsigned u64s)
690 {
691 	if (dst < src)
692 		__memmove_u64s_down(dst, src, u64s);
693 	else
694 		__memmove_u64s_up(dst, src, u64s);
695 }
696 
697 /* Set the last few bytes up to a u64 boundary given an offset into a buffer. */
698 static inline void memset_u64s_tail(void *s, int c, unsigned bytes)
699 {
700 	unsigned rem = round_up(bytes, sizeof(u64)) - bytes;
701 
702 	memset(s + bytes, c, rem);
703 }
704 
705 void sort_cmp_size(void *base, size_t num, size_t size,
706 	  int (*cmp_func)(const void *, const void *, size_t),
707 	  void (*swap_func)(void *, void *, size_t));
708 
709 /* just the memmove, doesn't update @_nr */
710 #define __array_insert_item(_array, _nr, _pos)				\
711 	memmove(&(_array)[(_pos) + 1],					\
712 		&(_array)[(_pos)],					\
713 		sizeof((_array)[0]) * ((_nr) - (_pos)))
714 
715 #define array_insert_item(_array, _nr, _pos, _new_item)			\
716 do {									\
717 	__array_insert_item(_array, _nr, _pos);				\
718 	(_nr)++;							\
719 	(_array)[(_pos)] = (_new_item);					\
720 } while (0)
721 
722 #define array_remove_items(_array, _nr, _pos, _nr_to_remove)		\
723 do {									\
724 	(_nr) -= (_nr_to_remove);					\
725 	memmove(&(_array)[(_pos)],					\
726 		&(_array)[(_pos) + (_nr_to_remove)],			\
727 		sizeof((_array)[0]) * ((_nr) - (_pos)));		\
728 } while (0)
729 
730 #define array_remove_item(_array, _nr, _pos)				\
731 	array_remove_items(_array, _nr, _pos, 1)
732 
733 static inline void __move_gap(void *array, size_t element_size,
734 			      size_t nr, size_t size,
735 			      size_t old_gap, size_t new_gap)
736 {
737 	size_t gap_end = old_gap + size - nr;
738 
739 	if (new_gap < old_gap) {
740 		size_t move = old_gap - new_gap;
741 
742 		memmove(array + element_size * (gap_end - move),
743 			array + element_size * (old_gap - move),
744 				element_size * move);
745 	} else if (new_gap > old_gap) {
746 		size_t move = new_gap - old_gap;
747 
748 		memmove(array + element_size * old_gap,
749 			array + element_size * gap_end,
750 				element_size * move);
751 	}
752 }
753 
754 /* Move the gap in a gap buffer: */
755 #define move_gap(_array, _nr, _size, _old_gap, _new_gap)	\
756 	__move_gap(_array, sizeof(_array[0]), _nr, _size, _old_gap, _new_gap)
757 
758 #define bubble_sort(_base, _nr, _cmp)					\
759 do {									\
760 	ssize_t _i, _last;						\
761 	bool _swapped = true;						\
762 									\
763 	for (_last= (ssize_t) (_nr) - 1; _last > 0 && _swapped; --_last) {\
764 		_swapped = false;					\
765 		for (_i = 0; _i < _last; _i++)				\
766 			if (_cmp((_base)[_i], (_base)[_i + 1]) > 0) {	\
767 				swap((_base)[_i], (_base)[_i + 1]);	\
768 				_swapped = true;			\
769 			}						\
770 	}								\
771 } while (0)
772 
773 static inline u64 percpu_u64_get(u64 __percpu *src)
774 {
775 	u64 ret = 0;
776 	int cpu;
777 
778 	for_each_possible_cpu(cpu)
779 		ret += *per_cpu_ptr(src, cpu);
780 	return ret;
781 }
782 
783 static inline void percpu_u64_set(u64 __percpu *dst, u64 src)
784 {
785 	int cpu;
786 
787 	for_each_possible_cpu(cpu)
788 		*per_cpu_ptr(dst, cpu) = 0;
789 	this_cpu_write(*dst, src);
790 }
791 
792 static inline void acc_u64s(u64 *acc, const u64 *src, unsigned nr)
793 {
794 	unsigned i;
795 
796 	for (i = 0; i < nr; i++)
797 		acc[i] += src[i];
798 }
799 
800 static inline void acc_u64s_percpu(u64 *acc, const u64 __percpu *src,
801 				   unsigned nr)
802 {
803 	int cpu;
804 
805 	for_each_possible_cpu(cpu)
806 		acc_u64s(acc, per_cpu_ptr(src, cpu), nr);
807 }
808 
809 static inline void percpu_memset(void __percpu *p, int c, size_t bytes)
810 {
811 	int cpu;
812 
813 	for_each_possible_cpu(cpu)
814 		memset(per_cpu_ptr(p, cpu), c, bytes);
815 }
816 
817 u64 *bch2_acc_percpu_u64s(u64 __percpu *, unsigned);
818 
819 #define cmp_int(l, r)		((l > r) - (l < r))
820 
821 static inline int u8_cmp(u8 l, u8 r)
822 {
823 	return cmp_int(l, r);
824 }
825 
826 static inline int cmp_le32(__le32 l, __le32 r)
827 {
828 	return cmp_int(le32_to_cpu(l), le32_to_cpu(r));
829 }
830 
831 #include <linux/uuid.h>
832 
833 #endif /* _BCACHEFS_UTIL_H */
834