xref: /openbmc/linux/mm/damon/core.c (revision 1c582c6d)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Data Access Monitor
4  *
5  * Author: SeongJae Park <sjpark@amazon.de>
6  */
7 
8 #define pr_fmt(fmt) "damon: " fmt
9 
10 #include <linux/damon.h>
11 #include <linux/delay.h>
12 #include <linux/kthread.h>
13 #include <linux/mm.h>
14 #include <linux/random.h>
15 #include <linux/slab.h>
16 #include <linux/string.h>
17 
18 #define CREATE_TRACE_POINTS
19 #include <trace/events/damon.h>
20 
21 #ifdef CONFIG_DAMON_KUNIT_TEST
22 #undef DAMON_MIN_REGION
23 #define DAMON_MIN_REGION 1
24 #endif
25 
26 /* Get a random number in [l, r) */
27 #define damon_rand(l, r) (l + prandom_u32_max(r - l))
28 
29 static DEFINE_MUTEX(damon_lock);
30 static int nr_running_ctxs;
31 
32 /*
33  * Construct a damon_region struct
34  *
35  * Returns the pointer to the new struct if success, or NULL otherwise
36  */
37 struct damon_region *damon_new_region(unsigned long start, unsigned long end)
38 {
39 	struct damon_region *region;
40 
41 	region = kmalloc(sizeof(*region), GFP_KERNEL);
42 	if (!region)
43 		return NULL;
44 
45 	region->ar.start = start;
46 	region->ar.end = end;
47 	region->nr_accesses = 0;
48 	INIT_LIST_HEAD(&region->list);
49 
50 	region->age = 0;
51 	region->last_nr_accesses = 0;
52 
53 	return region;
54 }
55 
56 /*
57  * Add a region between two other regions
58  */
59 inline void damon_insert_region(struct damon_region *r,
60 		struct damon_region *prev, struct damon_region *next,
61 		struct damon_target *t)
62 {
63 	__list_add(&r->list, &prev->list, &next->list);
64 	t->nr_regions++;
65 }
66 
67 void damon_add_region(struct damon_region *r, struct damon_target *t)
68 {
69 	list_add_tail(&r->list, &t->regions_list);
70 	t->nr_regions++;
71 }
72 
73 static void damon_del_region(struct damon_region *r, struct damon_target *t)
74 {
75 	list_del(&r->list);
76 	t->nr_regions--;
77 }
78 
79 static void damon_free_region(struct damon_region *r)
80 {
81 	kfree(r);
82 }
83 
84 void damon_destroy_region(struct damon_region *r, struct damon_target *t)
85 {
86 	damon_del_region(r, t);
87 	damon_free_region(r);
88 }
89 
90 struct damos *damon_new_scheme(
91 		unsigned long min_sz_region, unsigned long max_sz_region,
92 		unsigned int min_nr_accesses, unsigned int max_nr_accesses,
93 		unsigned int min_age_region, unsigned int max_age_region,
94 		enum damos_action action, struct damos_quota *quota,
95 		struct damos_watermarks *wmarks)
96 {
97 	struct damos *scheme;
98 
99 	scheme = kmalloc(sizeof(*scheme), GFP_KERNEL);
100 	if (!scheme)
101 		return NULL;
102 	scheme->min_sz_region = min_sz_region;
103 	scheme->max_sz_region = max_sz_region;
104 	scheme->min_nr_accesses = min_nr_accesses;
105 	scheme->max_nr_accesses = max_nr_accesses;
106 	scheme->min_age_region = min_age_region;
107 	scheme->max_age_region = max_age_region;
108 	scheme->action = action;
109 	scheme->stat_count = 0;
110 	scheme->stat_sz = 0;
111 	INIT_LIST_HEAD(&scheme->list);
112 
113 	scheme->quota.ms = quota->ms;
114 	scheme->quota.sz = quota->sz;
115 	scheme->quota.reset_interval = quota->reset_interval;
116 	scheme->quota.weight_sz = quota->weight_sz;
117 	scheme->quota.weight_nr_accesses = quota->weight_nr_accesses;
118 	scheme->quota.weight_age = quota->weight_age;
119 	scheme->quota.total_charged_sz = 0;
120 	scheme->quota.total_charged_ns = 0;
121 	scheme->quota.esz = 0;
122 	scheme->quota.charged_sz = 0;
123 	scheme->quota.charged_from = 0;
124 	scheme->quota.charge_target_from = NULL;
125 	scheme->quota.charge_addr_from = 0;
126 
127 	scheme->wmarks.metric = wmarks->metric;
128 	scheme->wmarks.interval = wmarks->interval;
129 	scheme->wmarks.high = wmarks->high;
130 	scheme->wmarks.mid = wmarks->mid;
131 	scheme->wmarks.low = wmarks->low;
132 	scheme->wmarks.activated = true;
133 
134 	return scheme;
135 }
136 
137 void damon_add_scheme(struct damon_ctx *ctx, struct damos *s)
138 {
139 	list_add_tail(&s->list, &ctx->schemes);
140 }
141 
142 static void damon_del_scheme(struct damos *s)
143 {
144 	list_del(&s->list);
145 }
146 
147 static void damon_free_scheme(struct damos *s)
148 {
149 	kfree(s);
150 }
151 
152 void damon_destroy_scheme(struct damos *s)
153 {
154 	damon_del_scheme(s);
155 	damon_free_scheme(s);
156 }
157 
158 /*
159  * Construct a damon_target struct
160  *
161  * Returns the pointer to the new struct if success, or NULL otherwise
162  */
163 struct damon_target *damon_new_target(unsigned long id)
164 {
165 	struct damon_target *t;
166 
167 	t = kmalloc(sizeof(*t), GFP_KERNEL);
168 	if (!t)
169 		return NULL;
170 
171 	t->id = id;
172 	t->nr_regions = 0;
173 	INIT_LIST_HEAD(&t->regions_list);
174 
175 	return t;
176 }
177 
178 void damon_add_target(struct damon_ctx *ctx, struct damon_target *t)
179 {
180 	list_add_tail(&t->list, &ctx->adaptive_targets);
181 }
182 
183 bool damon_targets_empty(struct damon_ctx *ctx)
184 {
185 	return list_empty(&ctx->adaptive_targets);
186 }
187 
188 static void damon_del_target(struct damon_target *t)
189 {
190 	list_del(&t->list);
191 }
192 
193 void damon_free_target(struct damon_target *t)
194 {
195 	struct damon_region *r, *next;
196 
197 	damon_for_each_region_safe(r, next, t)
198 		damon_free_region(r);
199 	kfree(t);
200 }
201 
202 void damon_destroy_target(struct damon_target *t)
203 {
204 	damon_del_target(t);
205 	damon_free_target(t);
206 }
207 
208 unsigned int damon_nr_regions(struct damon_target *t)
209 {
210 	return t->nr_regions;
211 }
212 
213 struct damon_ctx *damon_new_ctx(void)
214 {
215 	struct damon_ctx *ctx;
216 
217 	ctx = kzalloc(sizeof(*ctx), GFP_KERNEL);
218 	if (!ctx)
219 		return NULL;
220 
221 	ctx->sample_interval = 5 * 1000;
222 	ctx->aggr_interval = 100 * 1000;
223 	ctx->primitive_update_interval = 60 * 1000 * 1000;
224 
225 	ktime_get_coarse_ts64(&ctx->last_aggregation);
226 	ctx->last_primitive_update = ctx->last_aggregation;
227 
228 	mutex_init(&ctx->kdamond_lock);
229 
230 	ctx->min_nr_regions = 10;
231 	ctx->max_nr_regions = 1000;
232 
233 	INIT_LIST_HEAD(&ctx->adaptive_targets);
234 	INIT_LIST_HEAD(&ctx->schemes);
235 
236 	return ctx;
237 }
238 
239 static void damon_destroy_targets(struct damon_ctx *ctx)
240 {
241 	struct damon_target *t, *next_t;
242 
243 	if (ctx->primitive.cleanup) {
244 		ctx->primitive.cleanup(ctx);
245 		return;
246 	}
247 
248 	damon_for_each_target_safe(t, next_t, ctx)
249 		damon_destroy_target(t);
250 }
251 
252 void damon_destroy_ctx(struct damon_ctx *ctx)
253 {
254 	struct damos *s, *next_s;
255 
256 	damon_destroy_targets(ctx);
257 
258 	damon_for_each_scheme_safe(s, next_s, ctx)
259 		damon_destroy_scheme(s);
260 
261 	kfree(ctx);
262 }
263 
264 /**
265  * damon_set_targets() - Set monitoring targets.
266  * @ctx:	monitoring context
267  * @ids:	array of target ids
268  * @nr_ids:	number of entries in @ids
269  *
270  * This function should not be called while the kdamond is running.
271  *
272  * Return: 0 on success, negative error code otherwise.
273  */
274 int damon_set_targets(struct damon_ctx *ctx,
275 		      unsigned long *ids, ssize_t nr_ids)
276 {
277 	ssize_t i;
278 	struct damon_target *t, *next;
279 
280 	damon_destroy_targets(ctx);
281 
282 	for (i = 0; i < nr_ids; i++) {
283 		t = damon_new_target(ids[i]);
284 		if (!t) {
285 			/* The caller should do cleanup of the ids itself */
286 			damon_for_each_target_safe(t, next, ctx)
287 				damon_destroy_target(t);
288 			return -ENOMEM;
289 		}
290 		damon_add_target(ctx, t);
291 	}
292 
293 	return 0;
294 }
295 
296 /**
297  * damon_set_attrs() - Set attributes for the monitoring.
298  * @ctx:		monitoring context
299  * @sample_int:		time interval between samplings
300  * @aggr_int:		time interval between aggregations
301  * @primitive_upd_int:	time interval between monitoring primitive updates
302  * @min_nr_reg:		minimal number of regions
303  * @max_nr_reg:		maximum number of regions
304  *
305  * This function should not be called while the kdamond is running.
306  * Every time interval is in micro-seconds.
307  *
308  * Return: 0 on success, negative error code otherwise.
309  */
310 int damon_set_attrs(struct damon_ctx *ctx, unsigned long sample_int,
311 		    unsigned long aggr_int, unsigned long primitive_upd_int,
312 		    unsigned long min_nr_reg, unsigned long max_nr_reg)
313 {
314 	if (min_nr_reg < 3)
315 		return -EINVAL;
316 	if (min_nr_reg > max_nr_reg)
317 		return -EINVAL;
318 
319 	ctx->sample_interval = sample_int;
320 	ctx->aggr_interval = aggr_int;
321 	ctx->primitive_update_interval = primitive_upd_int;
322 	ctx->min_nr_regions = min_nr_reg;
323 	ctx->max_nr_regions = max_nr_reg;
324 
325 	return 0;
326 }
327 
328 /**
329  * damon_set_schemes() - Set data access monitoring based operation schemes.
330  * @ctx:	monitoring context
331  * @schemes:	array of the schemes
332  * @nr_schemes:	number of entries in @schemes
333  *
334  * This function should not be called while the kdamond of the context is
335  * running.
336  *
337  * Return: 0 if success, or negative error code otherwise.
338  */
339 int damon_set_schemes(struct damon_ctx *ctx, struct damos **schemes,
340 			ssize_t nr_schemes)
341 {
342 	struct damos *s, *next;
343 	ssize_t i;
344 
345 	damon_for_each_scheme_safe(s, next, ctx)
346 		damon_destroy_scheme(s);
347 	for (i = 0; i < nr_schemes; i++)
348 		damon_add_scheme(ctx, schemes[i]);
349 	return 0;
350 }
351 
352 /**
353  * damon_nr_running_ctxs() - Return number of currently running contexts.
354  */
355 int damon_nr_running_ctxs(void)
356 {
357 	int nr_ctxs;
358 
359 	mutex_lock(&damon_lock);
360 	nr_ctxs = nr_running_ctxs;
361 	mutex_unlock(&damon_lock);
362 
363 	return nr_ctxs;
364 }
365 
366 /* Returns the size upper limit for each monitoring region */
367 static unsigned long damon_region_sz_limit(struct damon_ctx *ctx)
368 {
369 	struct damon_target *t;
370 	struct damon_region *r;
371 	unsigned long sz = 0;
372 
373 	damon_for_each_target(t, ctx) {
374 		damon_for_each_region(r, t)
375 			sz += r->ar.end - r->ar.start;
376 	}
377 
378 	if (ctx->min_nr_regions)
379 		sz /= ctx->min_nr_regions;
380 	if (sz < DAMON_MIN_REGION)
381 		sz = DAMON_MIN_REGION;
382 
383 	return sz;
384 }
385 
386 static int kdamond_fn(void *data);
387 
388 /*
389  * __damon_start() - Starts monitoring with given context.
390  * @ctx:	monitoring context
391  *
392  * This function should be called while damon_lock is hold.
393  *
394  * Return: 0 on success, negative error code otherwise.
395  */
396 static int __damon_start(struct damon_ctx *ctx)
397 {
398 	int err = -EBUSY;
399 
400 	mutex_lock(&ctx->kdamond_lock);
401 	if (!ctx->kdamond) {
402 		err = 0;
403 		ctx->kdamond = kthread_run(kdamond_fn, ctx, "kdamond.%d",
404 				nr_running_ctxs);
405 		if (IS_ERR(ctx->kdamond)) {
406 			err = PTR_ERR(ctx->kdamond);
407 			ctx->kdamond = NULL;
408 		}
409 	}
410 	mutex_unlock(&ctx->kdamond_lock);
411 
412 	return err;
413 }
414 
415 /**
416  * damon_start() - Starts the monitorings for a given group of contexts.
417  * @ctxs:	an array of the pointers for contexts to start monitoring
418  * @nr_ctxs:	size of @ctxs
419  *
420  * This function starts a group of monitoring threads for a group of monitoring
421  * contexts.  One thread per each context is created and run in parallel.  The
422  * caller should handle synchronization between the threads by itself.  If a
423  * group of threads that created by other 'damon_start()' call is currently
424  * running, this function does nothing but returns -EBUSY.
425  *
426  * Return: 0 on success, negative error code otherwise.
427  */
428 int damon_start(struct damon_ctx **ctxs, int nr_ctxs)
429 {
430 	int i;
431 	int err = 0;
432 
433 	mutex_lock(&damon_lock);
434 	if (nr_running_ctxs) {
435 		mutex_unlock(&damon_lock);
436 		return -EBUSY;
437 	}
438 
439 	for (i = 0; i < nr_ctxs; i++) {
440 		err = __damon_start(ctxs[i]);
441 		if (err)
442 			break;
443 		nr_running_ctxs++;
444 	}
445 	mutex_unlock(&damon_lock);
446 
447 	return err;
448 }
449 
450 /*
451  * __damon_stop() - Stops monitoring of given context.
452  * @ctx:	monitoring context
453  *
454  * Return: 0 on success, negative error code otherwise.
455  */
456 static int __damon_stop(struct damon_ctx *ctx)
457 {
458 	struct task_struct *tsk;
459 
460 	mutex_lock(&ctx->kdamond_lock);
461 	tsk = ctx->kdamond;
462 	if (tsk) {
463 		get_task_struct(tsk);
464 		mutex_unlock(&ctx->kdamond_lock);
465 		kthread_stop(tsk);
466 		put_task_struct(tsk);
467 		return 0;
468 	}
469 	mutex_unlock(&ctx->kdamond_lock);
470 
471 	return -EPERM;
472 }
473 
474 /**
475  * damon_stop() - Stops the monitorings for a given group of contexts.
476  * @ctxs:	an array of the pointers for contexts to stop monitoring
477  * @nr_ctxs:	size of @ctxs
478  *
479  * Return: 0 on success, negative error code otherwise.
480  */
481 int damon_stop(struct damon_ctx **ctxs, int nr_ctxs)
482 {
483 	int i, err = 0;
484 
485 	for (i = 0; i < nr_ctxs; i++) {
486 		/* nr_running_ctxs is decremented in kdamond_fn */
487 		err = __damon_stop(ctxs[i]);
488 		if (err)
489 			return err;
490 	}
491 
492 	return err;
493 }
494 
495 /*
496  * damon_check_reset_time_interval() - Check if a time interval is elapsed.
497  * @baseline:	the time to check whether the interval has elapsed since
498  * @interval:	the time interval (microseconds)
499  *
500  * See whether the given time interval has passed since the given baseline
501  * time.  If so, it also updates the baseline to current time for next check.
502  *
503  * Return:	true if the time interval has passed, or false otherwise.
504  */
505 static bool damon_check_reset_time_interval(struct timespec64 *baseline,
506 		unsigned long interval)
507 {
508 	struct timespec64 now;
509 
510 	ktime_get_coarse_ts64(&now);
511 	if ((timespec64_to_ns(&now) - timespec64_to_ns(baseline)) <
512 			interval * 1000)
513 		return false;
514 	*baseline = now;
515 	return true;
516 }
517 
518 /*
519  * Check whether it is time to flush the aggregated information
520  */
521 static bool kdamond_aggregate_interval_passed(struct damon_ctx *ctx)
522 {
523 	return damon_check_reset_time_interval(&ctx->last_aggregation,
524 			ctx->aggr_interval);
525 }
526 
527 /*
528  * Reset the aggregated monitoring results ('nr_accesses' of each region).
529  */
530 static void kdamond_reset_aggregated(struct damon_ctx *c)
531 {
532 	struct damon_target *t;
533 
534 	damon_for_each_target(t, c) {
535 		struct damon_region *r;
536 
537 		damon_for_each_region(r, t) {
538 			trace_damon_aggregated(t, r, damon_nr_regions(t));
539 			r->last_nr_accesses = r->nr_accesses;
540 			r->nr_accesses = 0;
541 		}
542 	}
543 }
544 
545 static void damon_split_region_at(struct damon_ctx *ctx,
546 		struct damon_target *t, struct damon_region *r,
547 		unsigned long sz_r);
548 
549 static bool __damos_valid_target(struct damon_region *r, struct damos *s)
550 {
551 	unsigned long sz;
552 
553 	sz = r->ar.end - r->ar.start;
554 	return s->min_sz_region <= sz && sz <= s->max_sz_region &&
555 		s->min_nr_accesses <= r->nr_accesses &&
556 		r->nr_accesses <= s->max_nr_accesses &&
557 		s->min_age_region <= r->age && r->age <= s->max_age_region;
558 }
559 
560 static bool damos_valid_target(struct damon_ctx *c, struct damon_target *t,
561 		struct damon_region *r, struct damos *s)
562 {
563 	bool ret = __damos_valid_target(r, s);
564 
565 	if (!ret || !s->quota.esz || !c->primitive.get_scheme_score)
566 		return ret;
567 
568 	return c->primitive.get_scheme_score(c, t, r, s) >= s->quota.min_score;
569 }
570 
571 static void damon_do_apply_schemes(struct damon_ctx *c,
572 				   struct damon_target *t,
573 				   struct damon_region *r)
574 {
575 	struct damos *s;
576 
577 	damon_for_each_scheme(s, c) {
578 		struct damos_quota *quota = &s->quota;
579 		unsigned long sz = r->ar.end - r->ar.start;
580 		struct timespec64 begin, end;
581 
582 		if (!s->wmarks.activated)
583 			continue;
584 
585 		/* Check the quota */
586 		if (quota->esz && quota->charged_sz >= quota->esz)
587 			continue;
588 
589 		/* Skip previously charged regions */
590 		if (quota->charge_target_from) {
591 			if (t != quota->charge_target_from)
592 				continue;
593 			if (r == damon_last_region(t)) {
594 				quota->charge_target_from = NULL;
595 				quota->charge_addr_from = 0;
596 				continue;
597 			}
598 			if (quota->charge_addr_from &&
599 					r->ar.end <= quota->charge_addr_from)
600 				continue;
601 
602 			if (quota->charge_addr_from && r->ar.start <
603 					quota->charge_addr_from) {
604 				sz = ALIGN_DOWN(quota->charge_addr_from -
605 						r->ar.start, DAMON_MIN_REGION);
606 				if (!sz) {
607 					if (r->ar.end - r->ar.start <=
608 							DAMON_MIN_REGION)
609 						continue;
610 					sz = DAMON_MIN_REGION;
611 				}
612 				damon_split_region_at(c, t, r, sz);
613 				r = damon_next_region(r);
614 				sz = r->ar.end - r->ar.start;
615 			}
616 			quota->charge_target_from = NULL;
617 			quota->charge_addr_from = 0;
618 		}
619 
620 		if (!damos_valid_target(c, t, r, s))
621 			continue;
622 
623 		/* Apply the scheme */
624 		if (c->primitive.apply_scheme) {
625 			if (quota->esz &&
626 					quota->charged_sz + sz > quota->esz) {
627 				sz = ALIGN_DOWN(quota->esz - quota->charged_sz,
628 						DAMON_MIN_REGION);
629 				if (!sz)
630 					goto update_stat;
631 				damon_split_region_at(c, t, r, sz);
632 			}
633 			ktime_get_coarse_ts64(&begin);
634 			c->primitive.apply_scheme(c, t, r, s);
635 			ktime_get_coarse_ts64(&end);
636 			quota->total_charged_ns += timespec64_to_ns(&end) -
637 				timespec64_to_ns(&begin);
638 			quota->charged_sz += sz;
639 			if (quota->esz && quota->charged_sz >= quota->esz) {
640 				quota->charge_target_from = t;
641 				quota->charge_addr_from = r->ar.end + 1;
642 			}
643 		}
644 		if (s->action != DAMOS_STAT)
645 			r->age = 0;
646 
647 update_stat:
648 		s->stat_count++;
649 		s->stat_sz += sz;
650 	}
651 }
652 
653 /* Shouldn't be called if quota->ms and quota->sz are zero */
654 static void damos_set_effective_quota(struct damos_quota *quota)
655 {
656 	unsigned long throughput;
657 	unsigned long esz;
658 
659 	if (!quota->ms) {
660 		quota->esz = quota->sz;
661 		return;
662 	}
663 
664 	if (quota->total_charged_ns)
665 		throughput = quota->total_charged_sz * 1000000 /
666 			quota->total_charged_ns;
667 	else
668 		throughput = PAGE_SIZE * 1024;
669 	esz = throughput * quota->ms;
670 
671 	if (quota->sz && quota->sz < esz)
672 		esz = quota->sz;
673 	quota->esz = esz;
674 }
675 
676 static void kdamond_apply_schemes(struct damon_ctx *c)
677 {
678 	struct damon_target *t;
679 	struct damon_region *r, *next_r;
680 	struct damos *s;
681 
682 	damon_for_each_scheme(s, c) {
683 		struct damos_quota *quota = &s->quota;
684 		unsigned long cumulated_sz;
685 		unsigned int score, max_score = 0;
686 
687 		if (!s->wmarks.activated)
688 			continue;
689 
690 		if (!quota->ms && !quota->sz)
691 			continue;
692 
693 		/* New charge window starts */
694 		if (time_after_eq(jiffies, quota->charged_from +
695 					msecs_to_jiffies(
696 						quota->reset_interval))) {
697 			quota->total_charged_sz += quota->charged_sz;
698 			quota->charged_from = jiffies;
699 			quota->charged_sz = 0;
700 			damos_set_effective_quota(quota);
701 		}
702 
703 		if (!c->primitive.get_scheme_score)
704 			continue;
705 
706 		/* Fill up the score histogram */
707 		memset(quota->histogram, 0, sizeof(quota->histogram));
708 		damon_for_each_target(t, c) {
709 			damon_for_each_region(r, t) {
710 				if (!__damos_valid_target(r, s))
711 					continue;
712 				score = c->primitive.get_scheme_score(
713 						c, t, r, s);
714 				quota->histogram[score] +=
715 					r->ar.end - r->ar.start;
716 				if (score > max_score)
717 					max_score = score;
718 			}
719 		}
720 
721 		/* Set the min score limit */
722 		for (cumulated_sz = 0, score = max_score; ; score--) {
723 			cumulated_sz += quota->histogram[score];
724 			if (cumulated_sz >= quota->esz || !score)
725 				break;
726 		}
727 		quota->min_score = score;
728 	}
729 
730 	damon_for_each_target(t, c) {
731 		damon_for_each_region_safe(r, next_r, t)
732 			damon_do_apply_schemes(c, t, r);
733 	}
734 }
735 
736 #define sz_damon_region(r) (r->ar.end - r->ar.start)
737 
738 /*
739  * Merge two adjacent regions into one region
740  */
741 static void damon_merge_two_regions(struct damon_target *t,
742 		struct damon_region *l, struct damon_region *r)
743 {
744 	unsigned long sz_l = sz_damon_region(l), sz_r = sz_damon_region(r);
745 
746 	l->nr_accesses = (l->nr_accesses * sz_l + r->nr_accesses * sz_r) /
747 			(sz_l + sz_r);
748 	l->age = (l->age * sz_l + r->age * sz_r) / (sz_l + sz_r);
749 	l->ar.end = r->ar.end;
750 	damon_destroy_region(r, t);
751 }
752 
753 #define diff_of(a, b) (a > b ? a - b : b - a)
754 
755 /*
756  * Merge adjacent regions having similar access frequencies
757  *
758  * t		target affected by this merge operation
759  * thres	'->nr_accesses' diff threshold for the merge
760  * sz_limit	size upper limit of each region
761  */
762 static void damon_merge_regions_of(struct damon_target *t, unsigned int thres,
763 				   unsigned long sz_limit)
764 {
765 	struct damon_region *r, *prev = NULL, *next;
766 
767 	damon_for_each_region_safe(r, next, t) {
768 		if (diff_of(r->nr_accesses, r->last_nr_accesses) > thres)
769 			r->age = 0;
770 		else
771 			r->age++;
772 
773 		if (prev && prev->ar.end == r->ar.start &&
774 		    diff_of(prev->nr_accesses, r->nr_accesses) <= thres &&
775 		    sz_damon_region(prev) + sz_damon_region(r) <= sz_limit)
776 			damon_merge_two_regions(t, prev, r);
777 		else
778 			prev = r;
779 	}
780 }
781 
782 /*
783  * Merge adjacent regions having similar access frequencies
784  *
785  * threshold	'->nr_accesses' diff threshold for the merge
786  * sz_limit	size upper limit of each region
787  *
788  * This function merges monitoring target regions which are adjacent and their
789  * access frequencies are similar.  This is for minimizing the monitoring
790  * overhead under the dynamically changeable access pattern.  If a merge was
791  * unnecessarily made, later 'kdamond_split_regions()' will revert it.
792  */
793 static void kdamond_merge_regions(struct damon_ctx *c, unsigned int threshold,
794 				  unsigned long sz_limit)
795 {
796 	struct damon_target *t;
797 
798 	damon_for_each_target(t, c)
799 		damon_merge_regions_of(t, threshold, sz_limit);
800 }
801 
802 /*
803  * Split a region in two
804  *
805  * r		the region to be split
806  * sz_r		size of the first sub-region that will be made
807  */
808 static void damon_split_region_at(struct damon_ctx *ctx,
809 		struct damon_target *t, struct damon_region *r,
810 		unsigned long sz_r)
811 {
812 	struct damon_region *new;
813 
814 	new = damon_new_region(r->ar.start + sz_r, r->ar.end);
815 	if (!new)
816 		return;
817 
818 	r->ar.end = new->ar.start;
819 
820 	new->age = r->age;
821 	new->last_nr_accesses = r->last_nr_accesses;
822 
823 	damon_insert_region(new, r, damon_next_region(r), t);
824 }
825 
826 /* Split every region in the given target into 'nr_subs' regions */
827 static void damon_split_regions_of(struct damon_ctx *ctx,
828 				     struct damon_target *t, int nr_subs)
829 {
830 	struct damon_region *r, *next;
831 	unsigned long sz_region, sz_sub = 0;
832 	int i;
833 
834 	damon_for_each_region_safe(r, next, t) {
835 		sz_region = r->ar.end - r->ar.start;
836 
837 		for (i = 0; i < nr_subs - 1 &&
838 				sz_region > 2 * DAMON_MIN_REGION; i++) {
839 			/*
840 			 * Randomly select size of left sub-region to be at
841 			 * least 10 percent and at most 90% of original region
842 			 */
843 			sz_sub = ALIGN_DOWN(damon_rand(1, 10) *
844 					sz_region / 10, DAMON_MIN_REGION);
845 			/* Do not allow blank region */
846 			if (sz_sub == 0 || sz_sub >= sz_region)
847 				continue;
848 
849 			damon_split_region_at(ctx, t, r, sz_sub);
850 			sz_region = sz_sub;
851 		}
852 	}
853 }
854 
855 /*
856  * Split every target region into randomly-sized small regions
857  *
858  * This function splits every target region into random-sized small regions if
859  * current total number of the regions is equal or smaller than half of the
860  * user-specified maximum number of regions.  This is for maximizing the
861  * monitoring accuracy under the dynamically changeable access patterns.  If a
862  * split was unnecessarily made, later 'kdamond_merge_regions()' will revert
863  * it.
864  */
865 static void kdamond_split_regions(struct damon_ctx *ctx)
866 {
867 	struct damon_target *t;
868 	unsigned int nr_regions = 0;
869 	static unsigned int last_nr_regions;
870 	int nr_subregions = 2;
871 
872 	damon_for_each_target(t, ctx)
873 		nr_regions += damon_nr_regions(t);
874 
875 	if (nr_regions > ctx->max_nr_regions / 2)
876 		return;
877 
878 	/* Maybe the middle of the region has different access frequency */
879 	if (last_nr_regions == nr_regions &&
880 			nr_regions < ctx->max_nr_regions / 3)
881 		nr_subregions = 3;
882 
883 	damon_for_each_target(t, ctx)
884 		damon_split_regions_of(ctx, t, nr_subregions);
885 
886 	last_nr_regions = nr_regions;
887 }
888 
889 /*
890  * Check whether it is time to check and apply the target monitoring regions
891  *
892  * Returns true if it is.
893  */
894 static bool kdamond_need_update_primitive(struct damon_ctx *ctx)
895 {
896 	return damon_check_reset_time_interval(&ctx->last_primitive_update,
897 			ctx->primitive_update_interval);
898 }
899 
900 /*
901  * Check whether current monitoring should be stopped
902  *
903  * The monitoring is stopped when either the user requested to stop, or all
904  * monitoring targets are invalid.
905  *
906  * Returns true if need to stop current monitoring.
907  */
908 static bool kdamond_need_stop(struct damon_ctx *ctx)
909 {
910 	struct damon_target *t;
911 
912 	if (kthread_should_stop())
913 		return true;
914 
915 	if (!ctx->primitive.target_valid)
916 		return false;
917 
918 	damon_for_each_target(t, ctx) {
919 		if (ctx->primitive.target_valid(t))
920 			return false;
921 	}
922 
923 	return true;
924 }
925 
926 static unsigned long damos_wmark_metric_value(enum damos_wmark_metric metric)
927 {
928 	struct sysinfo i;
929 
930 	switch (metric) {
931 	case DAMOS_WMARK_FREE_MEM_RATE:
932 		si_meminfo(&i);
933 		return i.freeram * 1000 / i.totalram;
934 	default:
935 		break;
936 	}
937 	return -EINVAL;
938 }
939 
940 /*
941  * Returns zero if the scheme is active.  Else, returns time to wait for next
942  * watermark check in micro-seconds.
943  */
944 static unsigned long damos_wmark_wait_us(struct damos *scheme)
945 {
946 	unsigned long metric;
947 
948 	if (scheme->wmarks.metric == DAMOS_WMARK_NONE)
949 		return 0;
950 
951 	metric = damos_wmark_metric_value(scheme->wmarks.metric);
952 	/* higher than high watermark or lower than low watermark */
953 	if (metric > scheme->wmarks.high || scheme->wmarks.low > metric) {
954 		if (scheme->wmarks.activated)
955 			pr_debug("deactivate a scheme (%d) for %s wmark\n",
956 					scheme->action,
957 					metric > scheme->wmarks.high ?
958 					"high" : "low");
959 		scheme->wmarks.activated = false;
960 		return scheme->wmarks.interval;
961 	}
962 
963 	/* inactive and higher than middle watermark */
964 	if ((scheme->wmarks.high >= metric && metric >= scheme->wmarks.mid) &&
965 			!scheme->wmarks.activated)
966 		return scheme->wmarks.interval;
967 
968 	if (!scheme->wmarks.activated)
969 		pr_debug("activate a scheme (%d)\n", scheme->action);
970 	scheme->wmarks.activated = true;
971 	return 0;
972 }
973 
974 static void kdamond_usleep(unsigned long usecs)
975 {
976 	/* See Documentation/timers/timers-howto.rst for the thresholds */
977 	if (usecs > 20 * USEC_PER_MSEC)
978 		schedule_timeout_idle(usecs_to_jiffies(usecs));
979 	else
980 		usleep_idle_range(usecs, usecs + 1);
981 }
982 
983 /* Returns negative error code if it's not activated but should return */
984 static int kdamond_wait_activation(struct damon_ctx *ctx)
985 {
986 	struct damos *s;
987 	unsigned long wait_time;
988 	unsigned long min_wait_time = 0;
989 
990 	while (!kdamond_need_stop(ctx)) {
991 		damon_for_each_scheme(s, ctx) {
992 			wait_time = damos_wmark_wait_us(s);
993 			if (!min_wait_time || wait_time < min_wait_time)
994 				min_wait_time = wait_time;
995 		}
996 		if (!min_wait_time)
997 			return 0;
998 
999 		kdamond_usleep(min_wait_time);
1000 	}
1001 	return -EBUSY;
1002 }
1003 
1004 /*
1005  * The monitoring daemon that runs as a kernel thread
1006  */
1007 static int kdamond_fn(void *data)
1008 {
1009 	struct damon_ctx *ctx = (struct damon_ctx *)data;
1010 	struct damon_target *t;
1011 	struct damon_region *r, *next;
1012 	unsigned int max_nr_accesses = 0;
1013 	unsigned long sz_limit = 0;
1014 	bool done = false;
1015 
1016 	pr_debug("kdamond (%d) starts\n", current->pid);
1017 
1018 	if (ctx->primitive.init)
1019 		ctx->primitive.init(ctx);
1020 	if (ctx->callback.before_start && ctx->callback.before_start(ctx))
1021 		done = true;
1022 
1023 	sz_limit = damon_region_sz_limit(ctx);
1024 
1025 	while (!kdamond_need_stop(ctx) && !done) {
1026 		if (kdamond_wait_activation(ctx))
1027 			continue;
1028 
1029 		if (ctx->primitive.prepare_access_checks)
1030 			ctx->primitive.prepare_access_checks(ctx);
1031 		if (ctx->callback.after_sampling &&
1032 				ctx->callback.after_sampling(ctx))
1033 			done = true;
1034 
1035 		kdamond_usleep(ctx->sample_interval);
1036 
1037 		if (ctx->primitive.check_accesses)
1038 			max_nr_accesses = ctx->primitive.check_accesses(ctx);
1039 
1040 		if (kdamond_aggregate_interval_passed(ctx)) {
1041 			kdamond_merge_regions(ctx,
1042 					max_nr_accesses / 10,
1043 					sz_limit);
1044 			if (ctx->callback.after_aggregation &&
1045 					ctx->callback.after_aggregation(ctx))
1046 				done = true;
1047 			kdamond_apply_schemes(ctx);
1048 			kdamond_reset_aggregated(ctx);
1049 			kdamond_split_regions(ctx);
1050 			if (ctx->primitive.reset_aggregated)
1051 				ctx->primitive.reset_aggregated(ctx);
1052 		}
1053 
1054 		if (kdamond_need_update_primitive(ctx)) {
1055 			if (ctx->primitive.update)
1056 				ctx->primitive.update(ctx);
1057 			sz_limit = damon_region_sz_limit(ctx);
1058 		}
1059 	}
1060 	damon_for_each_target(t, ctx) {
1061 		damon_for_each_region_safe(r, next, t)
1062 			damon_destroy_region(r, t);
1063 	}
1064 
1065 	if (ctx->callback.before_terminate)
1066 		ctx->callback.before_terminate(ctx);
1067 	if (ctx->primitive.cleanup)
1068 		ctx->primitive.cleanup(ctx);
1069 
1070 	pr_debug("kdamond (%d) finishes\n", current->pid);
1071 	mutex_lock(&ctx->kdamond_lock);
1072 	ctx->kdamond = NULL;
1073 	mutex_unlock(&ctx->kdamond_lock);
1074 
1075 	mutex_lock(&damon_lock);
1076 	nr_running_ctxs--;
1077 	mutex_unlock(&damon_lock);
1078 
1079 	return 0;
1080 }
1081 
1082 #include "core-test.h"
1083