1/* SPDX-License-Identifier: GPL-2.0 */
2/*
3 * DAMON api
4 *
5 * Author: SeongJae Park <sj@kernel.org>
6 */
7
8#ifndef _DAMON_H_
9#define _DAMON_H_
10
11#include <linux/memcontrol.h>
12#include <linux/mutex.h>
13#include <linux/time64.h>
14#include <linux/types.h>
15#include <linux/random.h>
16
17/* Minimal region size.  Every damon_region is aligned by this. */
18#define DAMON_MIN_REGION	PAGE_SIZE
19/* Max priority score for DAMON-based operation schemes */
20#define DAMOS_MAX_SCORE		(99)
21
22/* Get a random number in [l, r) */
23static inline unsigned long damon_rand(unsigned long l, unsigned long r)
24{
25	return l + get_random_u32_below(r - l);
26}
27
28/**
29 * struct damon_addr_range - Represents an address region of [@start, @end).
30 * @start:	Start address of the region (inclusive).
31 * @end:	End address of the region (exclusive).
32 */
33struct damon_addr_range {
34	unsigned long start;
35	unsigned long end;
36};
37
38/**
39 * struct damon_region - Represents a monitoring target region.
40 * @ar:			The address range of the region.
41 * @sampling_addr:	Address of the sample for the next access check.
42 * @nr_accesses:	Access frequency of this region.
43 * @nr_accesses_bp:	@nr_accesses in basis point (0.01%) that updated for
44 *			each sampling interval.
45 * @list:		List head for siblings.
46 * @age:		Age of this region.
47 *
48 * @nr_accesses is reset to zero for every &damon_attrs->aggr_interval and be
49 * increased for every &damon_attrs->sample_interval if an access to the region
50 * during the last sampling interval is found.  The update of this field should
51 * not be done with direct access but with the helper function,
52 * damon_update_region_access_rate().
53 *
54 * @nr_accesses_bp is another representation of @nr_accesses in basis point
55 * (1 in 10,000) that updated for every &damon_attrs->sample_interval in a
56 * manner similar to moving sum.  By the algorithm, this value becomes
57 * @nr_accesses * 10000 for every &struct damon_attrs->aggr_interval.  This can
58 * be used when the aggregation interval is too huge and therefore cannot wait
59 * for it before getting the access monitoring results.
60 *
61 * @age is initially zero, increased for each aggregation interval, and reset
62 * to zero again if the access frequency is significantly changed.  If two
63 * regions are merged into a new region, both @nr_accesses and @age of the new
64 * region are set as region size-weighted average of those of the two regions.
65 */
66struct damon_region {
67	struct damon_addr_range ar;
68	unsigned long sampling_addr;
69	unsigned int nr_accesses;
70	unsigned int nr_accesses_bp;
71	struct list_head list;
72
73	unsigned int age;
74/* private: Internal value for age calculation. */
75	unsigned int last_nr_accesses;
76};
77
78/**
79 * struct damon_target - Represents a monitoring target.
80 * @pid:		The PID of the virtual address space to monitor.
81 * @nr_regions:		Number of monitoring target regions of this target.
82 * @regions_list:	Head of the monitoring target regions of this target.
83 * @list:		List head for siblings.
84 *
85 * Each monitoring context could have multiple targets.  For example, a context
86 * for virtual memory address spaces could have multiple target processes.  The
87 * @pid should be set for appropriate &struct damon_operations including the
88 * virtual address spaces monitoring operations.
89 */
90struct damon_target {
91	struct pid *pid;
92	unsigned int nr_regions;
93	struct list_head regions_list;
94	struct list_head list;
95};
96
97/**
98 * enum damos_action - Represents an action of a Data Access Monitoring-based
99 * Operation Scheme.
100 *
101 * @DAMOS_WILLNEED:	Call ``madvise()`` for the region with MADV_WILLNEED.
102 * @DAMOS_COLD:		Call ``madvise()`` for the region with MADV_COLD.
103 * @DAMOS_PAGEOUT:	Call ``madvise()`` for the region with MADV_PAGEOUT.
104 * @DAMOS_HUGEPAGE:	Call ``madvise()`` for the region with MADV_HUGEPAGE.
105 * @DAMOS_NOHUGEPAGE:	Call ``madvise()`` for the region with MADV_NOHUGEPAGE.
106 * @DAMOS_LRU_PRIO:	Prioritize the region on its LRU lists.
107 * @DAMOS_LRU_DEPRIO:	Deprioritize the region on its LRU lists.
108 * @DAMOS_STAT:		Do nothing but count the stat.
109 * @NR_DAMOS_ACTIONS:	Total number of DAMOS actions
110 *
111 * The support of each action is up to running &struct damon_operations.
112 * &enum DAMON_OPS_VADDR and &enum DAMON_OPS_FVADDR supports all actions except
113 * &enum DAMOS_LRU_PRIO and &enum DAMOS_LRU_DEPRIO.  &enum DAMON_OPS_PADDR
114 * supports only &enum DAMOS_PAGEOUT, &enum DAMOS_LRU_PRIO, &enum
115 * DAMOS_LRU_DEPRIO, and &DAMOS_STAT.
116 */
117enum damos_action {
118	DAMOS_WILLNEED,
119	DAMOS_COLD,
120	DAMOS_PAGEOUT,
121	DAMOS_HUGEPAGE,
122	DAMOS_NOHUGEPAGE,
123	DAMOS_LRU_PRIO,
124	DAMOS_LRU_DEPRIO,
125	DAMOS_STAT,		/* Do nothing but only record the stat */
126	NR_DAMOS_ACTIONS,
127};
128
129/**
130 * enum damos_quota_goal_metric - Represents the metric to be used as the goal
131 *
132 * @DAMOS_QUOTA_USER_INPUT:	User-input value.
133 * @DAMOS_QUOTA_SOME_MEM_PSI_US:	System level some memory PSI in us.
134 * @NR_DAMOS_QUOTA_GOAL_METRICS:	Number of DAMOS quota goal metrics.
135 *
136 * Metrics equal to larger than @NR_DAMOS_QUOTA_GOAL_METRICS are unsupported.
137 */
138enum damos_quota_goal_metric {
139	DAMOS_QUOTA_USER_INPUT,
140	DAMOS_QUOTA_SOME_MEM_PSI_US,
141	NR_DAMOS_QUOTA_GOAL_METRICS,
142};
143
144/**
145 * struct damos_quota_goal - DAMOS scheme quota auto-tuning goal.
146 * @metric:		Metric to be used for representing the goal.
147 * @target_value:	Target value of @metric to achieve with the tuning.
148 * @current_value:	Current value of @metric.
149 * @last_psi_total:	Last measured total PSI
150 * @list:		List head for siblings.
151 *
152 * Data structure for getting the current score of the quota tuning goal.  The
153 * score is calculated by how close @current_value and @target_value are.  Then
154 * the score is entered to DAMON's internal feedback loop mechanism to get the
155 * auto-tuned quota.
156 *
157 * If @metric is DAMOS_QUOTA_USER_INPUT, @current_value should be manually
158 * entered by the user, probably inside the kdamond callbacks.  Otherwise,
159 * DAMON sets @current_value with self-measured value of @metric.
160 */
161struct damos_quota_goal {
162	enum damos_quota_goal_metric metric;
163	unsigned long target_value;
164	unsigned long current_value;
165	/* metric-dependent fields */
166	union {
167		u64 last_psi_total;
168	};
169	struct list_head list;
170};
171
172/**
173 * struct damos_quota - Controls the aggressiveness of the given scheme.
174 * @reset_interval:	Charge reset interval in milliseconds.
175 * @ms:			Maximum milliseconds that the scheme can use.
176 * @sz:			Maximum bytes of memory that the action can be applied.
177 * @goals:		Head of quota tuning goals (&damos_quota_goal) list.
178 * @esz:		Effective size quota in bytes.
179 *
180 * @weight_sz:		Weight of the region's size for prioritization.
181 * @weight_nr_accesses:	Weight of the region's nr_accesses for prioritization.
182 * @weight_age:		Weight of the region's age for prioritization.
183 *
184 * To avoid consuming too much CPU time or IO resources for applying the
185 * &struct damos->action to large memory, DAMON allows users to set time and/or
186 * size quotas.  The quotas can be set by writing non-zero values to &ms and
187 * &sz, respectively.  If the time quota is set, DAMON tries to use only up to
188 * &ms milliseconds within &reset_interval for applying the action.  If the
189 * size quota is set, DAMON tries to apply the action only up to &sz bytes
190 * within &reset_interval.
191 *
192 * Internally, the time quota is transformed to a size quota using estimated
193 * throughput of the scheme's action.  DAMON then compares it against &sz and
194 * uses smaller one as the effective quota.
195 *
196 * If @goals is not empt, DAMON calculates yet another size quota based on the
197 * goals using its internal feedback loop algorithm, for every @reset_interval.
198 * Then, if the new size quota is smaller than the effective quota, it uses the
199 * new size quota as the effective quota.
200 *
201 * The resulting effective size quota in bytes is set to @esz.
202 *
203 * For selecting regions within the quota, DAMON prioritizes current scheme's
204 * target memory regions using the &struct damon_operations->get_scheme_score.
205 * You could customize the prioritization logic by setting &weight_sz,
206 * &weight_nr_accesses, and &weight_age, because monitoring operations are
207 * encouraged to respect those.
208 */
209struct damos_quota {
210	unsigned long reset_interval;
211	unsigned long ms;
212	unsigned long sz;
213	struct list_head goals;
214	unsigned long esz;
215
216	unsigned int weight_sz;
217	unsigned int weight_nr_accesses;
218	unsigned int weight_age;
219
220/* private: */
221	/* For throughput estimation */
222	unsigned long total_charged_sz;
223	unsigned long total_charged_ns;
224
225	/* For charging the quota */
226	unsigned long charged_sz;
227	unsigned long charged_from;
228	struct damon_target *charge_target_from;
229	unsigned long charge_addr_from;
230
231	/* For prioritization */
232	unsigned long histogram[DAMOS_MAX_SCORE + 1];
233	unsigned int min_score;
234
235	/* For feedback loop */
236	unsigned long esz_bp;
237};
238
239/**
240 * enum damos_wmark_metric - Represents the watermark metric.
241 *
242 * @DAMOS_WMARK_NONE:		Ignore the watermarks of the given scheme.
243 * @DAMOS_WMARK_FREE_MEM_RATE:	Free memory rate of the system in [0,1000].
244 * @NR_DAMOS_WMARK_METRICS:	Total number of DAMOS watermark metrics
245 */
246enum damos_wmark_metric {
247	DAMOS_WMARK_NONE,
248	DAMOS_WMARK_FREE_MEM_RATE,
249	NR_DAMOS_WMARK_METRICS,
250};
251
252/**
253 * struct damos_watermarks - Controls when a given scheme should be activated.
254 * @metric:	Metric for the watermarks.
255 * @interval:	Watermarks check time interval in microseconds.
256 * @high:	High watermark.
257 * @mid:	Middle watermark.
258 * @low:	Low watermark.
259 *
260 * If &metric is &DAMOS_WMARK_NONE, the scheme is always active.  Being active
261 * means DAMON does monitoring and applying the action of the scheme to
262 * appropriate memory regions.  Else, DAMON checks &metric of the system for at
263 * least every &interval microseconds and works as below.
264 *
265 * If &metric is higher than &high, the scheme is inactivated.  If &metric is
266 * between &mid and &low, the scheme is activated.  If &metric is lower than
267 * &low, the scheme is inactivated.
268 */
269struct damos_watermarks {
270	enum damos_wmark_metric metric;
271	unsigned long interval;
272	unsigned long high;
273	unsigned long mid;
274	unsigned long low;
275
276/* private: */
277	bool activated;
278};
279
280/**
281 * struct damos_stat - Statistics on a given scheme.
282 * @nr_tried:	Total number of regions that the scheme is tried to be applied.
283 * @sz_tried:	Total size of regions that the scheme is tried to be applied.
284 * @nr_applied:	Total number of regions that the scheme is applied.
285 * @sz_applied:	Total size of regions that the scheme is applied.
286 * @qt_exceeds: Total number of times the quota of the scheme has exceeded.
287 */
288struct damos_stat {
289	unsigned long nr_tried;
290	unsigned long sz_tried;
291	unsigned long nr_applied;
292	unsigned long sz_applied;
293	unsigned long qt_exceeds;
294};
295
296/**
297 * enum damos_filter_type - Type of memory for &struct damos_filter
298 * @DAMOS_FILTER_TYPE_ANON:	Anonymous pages.
299 * @DAMOS_FILTER_TYPE_MEMCG:	Specific memcg's pages.
300 * @DAMOS_FILTER_TYPE_ADDR:	Address range.
301 * @DAMOS_FILTER_TYPE_TARGET:	Data Access Monitoring target.
302 * @NR_DAMOS_FILTER_TYPES:	Number of filter types.
303 *
304 * The anon pages type and memcg type filters are handled by underlying
305 * &struct damon_operations as a part of scheme action trying, and therefore
306 * accounted as 'tried'.  In contrast, other types are handled by core layer
307 * before trying of the action and therefore not accounted as 'tried'.
308 *
309 * The support of the filters that handled by &struct damon_operations depend
310 * on the running &struct damon_operations.
311 * &enum DAMON_OPS_PADDR supports both anon pages type and memcg type filters,
312 * while &enum DAMON_OPS_VADDR and &enum DAMON_OPS_FVADDR don't support any of
313 * the two types.
314 */
315enum damos_filter_type {
316	DAMOS_FILTER_TYPE_ANON,
317	DAMOS_FILTER_TYPE_MEMCG,
318	DAMOS_FILTER_TYPE_ADDR,
319	DAMOS_FILTER_TYPE_TARGET,
320	NR_DAMOS_FILTER_TYPES,
321};
322
323/**
324 * struct damos_filter - DAMOS action target memory filter.
325 * @type:	Type of the page.
326 * @matching:	If the matching page should filtered out or in.
327 * @memcg_id:	Memcg id of the question if @type is DAMOS_FILTER_MEMCG.
328 * @addr_range:	Address range if @type is DAMOS_FILTER_TYPE_ADDR.
329 * @target_idx:	Index of the &struct damon_target of
330 *		&damon_ctx->adaptive_targets if @type is
331 *		DAMOS_FILTER_TYPE_TARGET.
332 * @list:	List head for siblings.
333 *
334 * Before applying the &damos->action to a memory region, DAMOS checks if each
335 * page of the region matches to this and avoid applying the action if so.
336 * Support of each filter type depends on the running &struct damon_operations
337 * and the type.  Refer to &enum damos_filter_type for more detai.
338 */
339struct damos_filter {
340	enum damos_filter_type type;
341	bool matching;
342	union {
343		unsigned short memcg_id;
344		struct damon_addr_range addr_range;
345		int target_idx;
346	};
347	struct list_head list;
348};
349
350/**
351 * struct damos_access_pattern - Target access pattern of the given scheme.
352 * @min_sz_region:	Minimum size of target regions.
353 * @max_sz_region:	Maximum size of target regions.
354 * @min_nr_accesses:	Minimum ``->nr_accesses`` of target regions.
355 * @max_nr_accesses:	Maximum ``->nr_accesses`` of target regions.
356 * @min_age_region:	Minimum age of target regions.
357 * @max_age_region:	Maximum age of target regions.
358 */
359struct damos_access_pattern {
360	unsigned long min_sz_region;
361	unsigned long max_sz_region;
362	unsigned int min_nr_accesses;
363	unsigned int max_nr_accesses;
364	unsigned int min_age_region;
365	unsigned int max_age_region;
366};
367
368/**
369 * struct damos - Represents a Data Access Monitoring-based Operation Scheme.
370 * @pattern:		Access pattern of target regions.
371 * @action:		&damo_action to be applied to the target regions.
372 * @apply_interval_us:	The time between applying the @action.
373 * @quota:		Control the aggressiveness of this scheme.
374 * @wmarks:		Watermarks for automated (in)activation of this scheme.
375 * @filters:		Additional set of &struct damos_filter for &action.
376 * @stat:		Statistics of this scheme.
377 * @list:		List head for siblings.
378 *
379 * For each @apply_interval_us, DAMON finds regions which fit in the
380 * &pattern and applies &action to those. To avoid consuming too much
381 * CPU time or IO resources for the &action, &quota is used.
382 *
383 * If @apply_interval_us is zero, &damon_attrs->aggr_interval is used instead.
384 *
385 * To do the work only when needed, schemes can be activated for specific
386 * system situations using &wmarks.  If all schemes that registered to the
387 * monitoring context are inactive, DAMON stops monitoring either, and just
388 * repeatedly checks the watermarks.
389 *
390 * Before applying the &action to a memory region, &struct damon_operations
391 * implementation could check pages of the region and skip &action to respect
392 * &filters
393 *
394 * After applying the &action to each region, &stat_count and &stat_sz is
395 * updated to reflect the number of regions and total size of regions that the
396 * &action is applied.
397 */
398struct damos {
399	struct damos_access_pattern pattern;
400	enum damos_action action;
401	unsigned long apply_interval_us;
402/* private: internal use only */
403	/*
404	 * number of sample intervals that should be passed before applying
405	 * @action
406	 */
407	unsigned long next_apply_sis;
408/* public: */
409	struct damos_quota quota;
410	struct damos_watermarks wmarks;
411	struct list_head filters;
412	struct damos_stat stat;
413	struct list_head list;
414};
415
416/**
417 * enum damon_ops_id - Identifier for each monitoring operations implementation
418 *
419 * @DAMON_OPS_VADDR:	Monitoring operations for virtual address spaces
420 * @DAMON_OPS_FVADDR:	Monitoring operations for only fixed ranges of virtual
421 *			address spaces
422 * @DAMON_OPS_PADDR:	Monitoring operations for the physical address space
423 * @NR_DAMON_OPS:	Number of monitoring operations implementations
424 */
425enum damon_ops_id {
426	DAMON_OPS_VADDR,
427	DAMON_OPS_FVADDR,
428	DAMON_OPS_PADDR,
429	NR_DAMON_OPS,
430};
431
432struct damon_ctx;
433
434/**
435 * struct damon_operations - Monitoring operations for given use cases.
436 *
437 * @id:				Identifier of this operations set.
438 * @init:			Initialize operations-related data structures.
439 * @update:			Update operations-related data structures.
440 * @prepare_access_checks:	Prepare next access check of target regions.
441 * @check_accesses:		Check the accesses to target regions.
442 * @reset_aggregated:		Reset aggregated accesses monitoring results.
443 * @get_scheme_score:		Get the score of a region for a scheme.
444 * @apply_scheme:		Apply a DAMON-based operation scheme.
445 * @target_valid:		Determine if the target is valid.
446 * @cleanup:			Clean up the context.
447 *
448 * DAMON can be extended for various address spaces and usages.  For this,
449 * users should register the low level operations for their target address
450 * space and usecase via the &damon_ctx.ops.  Then, the monitoring thread
451 * (&damon_ctx.kdamond) calls @init and @prepare_access_checks before starting
452 * the monitoring, @update after each &damon_attrs.ops_update_interval, and
453 * @check_accesses, @target_valid and @prepare_access_checks after each
454 * &damon_attrs.sample_interval.  Finally, @reset_aggregated is called after
455 * each &damon_attrs.aggr_interval.
456 *
457 * Each &struct damon_operations instance having valid @id can be registered
458 * via damon_register_ops() and selected by damon_select_ops() later.
459 * @init should initialize operations-related data structures.  For example,
460 * this could be used to construct proper monitoring target regions and link
461 * those to @damon_ctx.adaptive_targets.
462 * @update should update the operations-related data structures.  For example,
463 * this could be used to update monitoring target regions for current status.
464 * @prepare_access_checks should manipulate the monitoring regions to be
465 * prepared for the next access check.
466 * @check_accesses should check the accesses to each region that made after the
467 * last preparation and update the number of observed accesses of each region.
468 * It should also return max number of observed accesses that made as a result
469 * of its update.  The value will be used for regions adjustment threshold.
470 * @reset_aggregated should reset the access monitoring results that aggregated
471 * by @check_accesses.
472 * @get_scheme_score should return the priority score of a region for a scheme
473 * as an integer in [0, &DAMOS_MAX_SCORE].
474 * @apply_scheme is called from @kdamond when a region for user provided
475 * DAMON-based operation scheme is found.  It should apply the scheme's action
476 * to the region and return bytes of the region that the action is successfully
477 * applied.
478 * @target_valid should check whether the target is still valid for the
479 * monitoring.
480 * @cleanup is called from @kdamond just before its termination.
481 */
482struct damon_operations {
483	enum damon_ops_id id;
484	void (*init)(struct damon_ctx *context);
485	void (*update)(struct damon_ctx *context);
486	void (*prepare_access_checks)(struct damon_ctx *context);
487	unsigned int (*check_accesses)(struct damon_ctx *context);
488	void (*reset_aggregated)(struct damon_ctx *context);
489	int (*get_scheme_score)(struct damon_ctx *context,
490			struct damon_target *t, struct damon_region *r,
491			struct damos *scheme);
492	unsigned long (*apply_scheme)(struct damon_ctx *context,
493			struct damon_target *t, struct damon_region *r,
494			struct damos *scheme);
495	bool (*target_valid)(struct damon_target *t);
496	void (*cleanup)(struct damon_ctx *context);
497};
498
499/**
500 * struct damon_callback - Monitoring events notification callbacks.
501 *
502 * @before_start:	Called before starting the monitoring.
503 * @after_wmarks_check:	Called after each schemes' watermarks check.
504 * @after_sampling:	Called after each sampling.
505 * @after_aggregation:	Called after each aggregation.
506 * @before_damos_apply:	Called before applying DAMOS action.
507 * @before_terminate:	Called before terminating the monitoring.
508 * @private:		User private data.
509 *
510 * The monitoring thread (&damon_ctx.kdamond) calls @before_start and
511 * @before_terminate just before starting and finishing the monitoring,
512 * respectively.  Therefore, those are good places for installing and cleaning
513 * @private.
514 *
515 * The monitoring thread calls @after_wmarks_check after each DAMON-based
516 * operation schemes' watermarks check.  If users need to make changes to the
517 * attributes of the monitoring context while it's deactivated due to the
518 * watermarks, this is the good place to do.
519 *
520 * The monitoring thread calls @after_sampling and @after_aggregation for each
521 * of the sampling intervals and aggregation intervals, respectively.
522 * Therefore, users can safely access the monitoring results without additional
523 * protection.  For the reason, users are recommended to use these callback for
524 * the accesses to the results.
525 *
526 * If any callback returns non-zero, monitoring stops.
527 */
528struct damon_callback {
529	void *private;
530
531	int (*before_start)(struct damon_ctx *context);
532	int (*after_wmarks_check)(struct damon_ctx *context);
533	int (*after_sampling)(struct damon_ctx *context);
534	int (*after_aggregation)(struct damon_ctx *context);
535	int (*before_damos_apply)(struct damon_ctx *context,
536			struct damon_target *target,
537			struct damon_region *region,
538			struct damos *scheme);
539	void (*before_terminate)(struct damon_ctx *context);
540};
541
542/**
543 * struct damon_attrs - Monitoring attributes for accuracy/overhead control.
544 *
545 * @sample_interval:		The time between access samplings.
546 * @aggr_interval:		The time between monitor results aggregations.
547 * @ops_update_interval:	The time between monitoring operations updates.
548 * @min_nr_regions:		The minimum number of adaptive monitoring
549 *				regions.
550 * @max_nr_regions:		The maximum number of adaptive monitoring
551 *				regions.
552 *
553 * For each @sample_interval, DAMON checks whether each region is accessed or
554 * not during the last @sample_interval.  If such access is found, DAMON
555 * aggregates the information by increasing &damon_region->nr_accesses for
556 * @aggr_interval time.  For each @aggr_interval, the count is reset.  DAMON
557 * also checks whether the target memory regions need update (e.g., by
558 * ``mmap()`` calls from the application, in case of virtual memory monitoring)
559 * and applies the changes for each @ops_update_interval.  All time intervals
560 * are in micro-seconds.  Please refer to &struct damon_operations and &struct
561 * damon_callback for more detail.
562 */
563struct damon_attrs {
564	unsigned long sample_interval;
565	unsigned long aggr_interval;
566	unsigned long ops_update_interval;
567	unsigned long min_nr_regions;
568	unsigned long max_nr_regions;
569};
570
571/**
572 * struct damon_ctx - Represents a context for each monitoring.  This is the
573 * main interface that allows users to set the attributes and get the results
574 * of the monitoring.
575 *
576 * @attrs:		Monitoring attributes for accuracy/overhead control.
577 * @kdamond:		Kernel thread who does the monitoring.
578 * @kdamond_lock:	Mutex for the synchronizations with @kdamond.
579 *
580 * For each monitoring context, one kernel thread for the monitoring is
581 * created.  The pointer to the thread is stored in @kdamond.
582 *
583 * Once started, the monitoring thread runs until explicitly required to be
584 * terminated or every monitoring target is invalid.  The validity of the
585 * targets is checked via the &damon_operations.target_valid of @ops.  The
586 * termination can also be explicitly requested by calling damon_stop().
587 * The thread sets @kdamond to NULL when it terminates. Therefore, users can
588 * know whether the monitoring is ongoing or terminated by reading @kdamond.
589 * Reads and writes to @kdamond from outside of the monitoring thread must
590 * be protected by @kdamond_lock.
591 *
592 * Note that the monitoring thread protects only @kdamond via @kdamond_lock.
593 * Accesses to other fields must be protected by themselves.
594 *
595 * @ops:	Set of monitoring operations for given use cases.
596 * @callback:	Set of callbacks for monitoring events notifications.
597 *
598 * @adaptive_targets:	Head of monitoring targets (&damon_target) list.
599 * @schemes:		Head of schemes (&damos) list.
600 */
601struct damon_ctx {
602	struct damon_attrs attrs;
603
604/* private: internal use only */
605	/* number of sample intervals that passed since this context started */
606	unsigned long passed_sample_intervals;
607	/*
608	 * number of sample intervals that should be passed before next
609	 * aggregation
610	 */
611	unsigned long next_aggregation_sis;
612	/*
613	 * number of sample intervals that should be passed before next ops
614	 * update
615	 */
616	unsigned long next_ops_update_sis;
617	/* for waiting until the execution of the kdamond_fn is started */
618	struct completion kdamond_started;
619
620/* public: */
621	struct task_struct *kdamond;
622	struct mutex kdamond_lock;
623
624	struct damon_operations ops;
625	struct damon_callback callback;
626
627	struct list_head adaptive_targets;
628	struct list_head schemes;
629};
630
631static inline struct damon_region *damon_next_region(struct damon_region *r)
632{
633	return container_of(r->list.next, struct damon_region, list);
634}
635
636static inline struct damon_region *damon_prev_region(struct damon_region *r)
637{
638	return container_of(r->list.prev, struct damon_region, list);
639}
640
641static inline struct damon_region *damon_last_region(struct damon_target *t)
642{
643	return list_last_entry(&t->regions_list, struct damon_region, list);
644}
645
646static inline struct damon_region *damon_first_region(struct damon_target *t)
647{
648	return list_first_entry(&t->regions_list, struct damon_region, list);
649}
650
651static inline unsigned long damon_sz_region(struct damon_region *r)
652{
653	return r->ar.end - r->ar.start;
654}
655
656
657#define damon_for_each_region(r, t) \
658	list_for_each_entry(r, &t->regions_list, list)
659
660#define damon_for_each_region_from(r, t) \
661	list_for_each_entry_from(r, &t->regions_list, list)
662
663#define damon_for_each_region_safe(r, next, t) \
664	list_for_each_entry_safe(r, next, &t->regions_list, list)
665
666#define damon_for_each_target(t, ctx) \
667	list_for_each_entry(t, &(ctx)->adaptive_targets, list)
668
669#define damon_for_each_target_safe(t, next, ctx)	\
670	list_for_each_entry_safe(t, next, &(ctx)->adaptive_targets, list)
671
672#define damon_for_each_scheme(s, ctx) \
673	list_for_each_entry(s, &(ctx)->schemes, list)
674
675#define damon_for_each_scheme_safe(s, next, ctx) \
676	list_for_each_entry_safe(s, next, &(ctx)->schemes, list)
677
678#define damos_for_each_quota_goal(goal, quota) \
679	list_for_each_entry(goal, &quota->goals, list)
680
681#define damos_for_each_quota_goal_safe(goal, next, quota) \
682	list_for_each_entry_safe(goal, next, &(quota)->goals, list)
683
684#define damos_for_each_filter(f, scheme) \
685	list_for_each_entry(f, &(scheme)->filters, list)
686
687#define damos_for_each_filter_safe(f, next, scheme) \
688	list_for_each_entry_safe(f, next, &(scheme)->filters, list)
689
690#ifdef CONFIG_DAMON
691
692struct damon_region *damon_new_region(unsigned long start, unsigned long end);
693
694/*
695 * Add a region between two other regions
696 */
697static inline void damon_insert_region(struct damon_region *r,
698		struct damon_region *prev, struct damon_region *next,
699		struct damon_target *t)
700{
701	__list_add(&r->list, &prev->list, &next->list);
702	t->nr_regions++;
703}
704
705void damon_add_region(struct damon_region *r, struct damon_target *t);
706void damon_destroy_region(struct damon_region *r, struct damon_target *t);
707int damon_set_regions(struct damon_target *t, struct damon_addr_range *ranges,
708		unsigned int nr_ranges);
709void damon_update_region_access_rate(struct damon_region *r, bool accessed,
710		struct damon_attrs *attrs);
711
712struct damos_filter *damos_new_filter(enum damos_filter_type type,
713		bool matching);
714void damos_add_filter(struct damos *s, struct damos_filter *f);
715void damos_destroy_filter(struct damos_filter *f);
716
717struct damos_quota_goal *damos_new_quota_goal(
718		enum damos_quota_goal_metric metric,
719		unsigned long target_value);
720void damos_add_quota_goal(struct damos_quota *q, struct damos_quota_goal *g);
721void damos_destroy_quota_goal(struct damos_quota_goal *goal);
722
723struct damos *damon_new_scheme(struct damos_access_pattern *pattern,
724			enum damos_action action,
725			unsigned long apply_interval_us,
726			struct damos_quota *quota,
727			struct damos_watermarks *wmarks);
728void damon_add_scheme(struct damon_ctx *ctx, struct damos *s);
729void damon_destroy_scheme(struct damos *s);
730
731struct damon_target *damon_new_target(void);
732void damon_add_target(struct damon_ctx *ctx, struct damon_target *t);
733bool damon_targets_empty(struct damon_ctx *ctx);
734void damon_free_target(struct damon_target *t);
735void damon_destroy_target(struct damon_target *t);
736unsigned int damon_nr_regions(struct damon_target *t);
737
738struct damon_ctx *damon_new_ctx(void);
739void damon_destroy_ctx(struct damon_ctx *ctx);
740int damon_set_attrs(struct damon_ctx *ctx, struct damon_attrs *attrs);
741void damon_set_schemes(struct damon_ctx *ctx,
742			struct damos **schemes, ssize_t nr_schemes);
743int damon_nr_running_ctxs(void);
744bool damon_is_registered_ops(enum damon_ops_id id);
745int damon_register_ops(struct damon_operations *ops);
746int damon_select_ops(struct damon_ctx *ctx, enum damon_ops_id id);
747
748static inline bool damon_target_has_pid(const struct damon_ctx *ctx)
749{
750	return ctx->ops.id == DAMON_OPS_VADDR || ctx->ops.id == DAMON_OPS_FVADDR;
751}
752
753static inline unsigned int damon_max_nr_accesses(const struct damon_attrs *attrs)
754{
755	/* {aggr,sample}_interval are unsigned long, hence could overflow */
756	return min(attrs->aggr_interval / attrs->sample_interval,
757			(unsigned long)UINT_MAX);
758}
759
760
761int damon_start(struct damon_ctx **ctxs, int nr_ctxs, bool exclusive);
762int damon_stop(struct damon_ctx **ctxs, int nr_ctxs);
763
764int damon_set_region_biggest_system_ram_default(struct damon_target *t,
765				unsigned long *start, unsigned long *end);
766
767#endif	/* CONFIG_DAMON */
768
769#endif	/* _DAMON_H */
770