xref: /linux/include/linux/damon.h (revision 9c5968db9e625019a0ee5226c7eebef5519d366a)
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) */
23 static 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  */
33 struct 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  */
66 struct 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  */
90 struct 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_MIGRATE_HOT:  Migrate the regions prioritizing warmer regions.
109  * @DAMOS_MIGRATE_COLD:	Migrate the regions prioritizing colder regions.
110  * @DAMOS_STAT:		Do nothing but count the stat.
111  * @NR_DAMOS_ACTIONS:	Total number of DAMOS actions
112  *
113  * The support of each action is up to running &struct damon_operations.
114  * &enum DAMON_OPS_VADDR and &enum DAMON_OPS_FVADDR supports all actions except
115  * &enum DAMOS_LRU_PRIO and &enum DAMOS_LRU_DEPRIO.  &enum DAMON_OPS_PADDR
116  * supports only &enum DAMOS_PAGEOUT, &enum DAMOS_LRU_PRIO, &enum
117  * DAMOS_LRU_DEPRIO, and &DAMOS_STAT.
118  */
119 enum damos_action {
120 	DAMOS_WILLNEED,
121 	DAMOS_COLD,
122 	DAMOS_PAGEOUT,
123 	DAMOS_HUGEPAGE,
124 	DAMOS_NOHUGEPAGE,
125 	DAMOS_LRU_PRIO,
126 	DAMOS_LRU_DEPRIO,
127 	DAMOS_MIGRATE_HOT,
128 	DAMOS_MIGRATE_COLD,
129 	DAMOS_STAT,		/* Do nothing but only record the stat */
130 	NR_DAMOS_ACTIONS,
131 };
132 
133 /**
134  * enum damos_quota_goal_metric - Represents the metric to be used as the goal
135  *
136  * @DAMOS_QUOTA_USER_INPUT:	User-input value.
137  * @DAMOS_QUOTA_SOME_MEM_PSI_US:	System level some memory PSI in us.
138  * @NR_DAMOS_QUOTA_GOAL_METRICS:	Number of DAMOS quota goal metrics.
139  *
140  * Metrics equal to larger than @NR_DAMOS_QUOTA_GOAL_METRICS are unsupported.
141  */
142 enum damos_quota_goal_metric {
143 	DAMOS_QUOTA_USER_INPUT,
144 	DAMOS_QUOTA_SOME_MEM_PSI_US,
145 	NR_DAMOS_QUOTA_GOAL_METRICS,
146 };
147 
148 /**
149  * struct damos_quota_goal - DAMOS scheme quota auto-tuning goal.
150  * @metric:		Metric to be used for representing the goal.
151  * @target_value:	Target value of @metric to achieve with the tuning.
152  * @current_value:	Current value of @metric.
153  * @last_psi_total:	Last measured total PSI
154  * @list:		List head for siblings.
155  *
156  * Data structure for getting the current score of the quota tuning goal.  The
157  * score is calculated by how close @current_value and @target_value are.  Then
158  * the score is entered to DAMON's internal feedback loop mechanism to get the
159  * auto-tuned quota.
160  *
161  * If @metric is DAMOS_QUOTA_USER_INPUT, @current_value should be manually
162  * entered by the user, probably inside the kdamond callbacks.  Otherwise,
163  * DAMON sets @current_value with self-measured value of @metric.
164  */
165 struct damos_quota_goal {
166 	enum damos_quota_goal_metric metric;
167 	unsigned long target_value;
168 	unsigned long current_value;
169 	/* metric-dependent fields */
170 	union {
171 		u64 last_psi_total;
172 	};
173 	struct list_head list;
174 };
175 
176 /**
177  * struct damos_quota - Controls the aggressiveness of the given scheme.
178  * @reset_interval:	Charge reset interval in milliseconds.
179  * @ms:			Maximum milliseconds that the scheme can use.
180  * @sz:			Maximum bytes of memory that the action can be applied.
181  * @goals:		Head of quota tuning goals (&damos_quota_goal) list.
182  * @esz:		Effective size quota in bytes.
183  *
184  * @weight_sz:		Weight of the region's size for prioritization.
185  * @weight_nr_accesses:	Weight of the region's nr_accesses for prioritization.
186  * @weight_age:		Weight of the region's age for prioritization.
187  *
188  * To avoid consuming too much CPU time or IO resources for applying the
189  * &struct damos->action to large memory, DAMON allows users to set time and/or
190  * size quotas.  The quotas can be set by writing non-zero values to &ms and
191  * &sz, respectively.  If the time quota is set, DAMON tries to use only up to
192  * &ms milliseconds within &reset_interval for applying the action.  If the
193  * size quota is set, DAMON tries to apply the action only up to &sz bytes
194  * within &reset_interval.
195  *
196  * To convince the different types of quotas and goals, DAMON internally
197  * converts those into one single size quota called "effective quota".  DAMON
198  * internally uses it as the only one real quota.  The conversion is made as
199  * follows.
200  *
201  * The time quota is transformed to a size quota using estimated throughput of
202  * the scheme's action.  DAMON then compares it against &sz and uses smaller
203  * one as the effective quota.
204  *
205  * If @goals is not empty, DAMON calculates yet another size quota based on the
206  * goals using its internal feedback loop algorithm, for every @reset_interval.
207  * Then, if the new size quota is smaller than the effective quota, it uses the
208  * new size quota as the effective quota.
209  *
210  * The resulting effective size quota in bytes is set to @esz.
211  *
212  * For selecting regions within the quota, DAMON prioritizes current scheme's
213  * target memory regions using the &struct damon_operations->get_scheme_score.
214  * You could customize the prioritization logic by setting &weight_sz,
215  * &weight_nr_accesses, and &weight_age, because monitoring operations are
216  * encouraged to respect those.
217  */
218 struct damos_quota {
219 	unsigned long reset_interval;
220 	unsigned long ms;
221 	unsigned long sz;
222 	struct list_head goals;
223 	unsigned long esz;
224 
225 	unsigned int weight_sz;
226 	unsigned int weight_nr_accesses;
227 	unsigned int weight_age;
228 
229 /* private: */
230 	/* For throughput estimation */
231 	unsigned long total_charged_sz;
232 	unsigned long total_charged_ns;
233 
234 	/* For charging the quota */
235 	unsigned long charged_sz;
236 	unsigned long charged_from;
237 	struct damon_target *charge_target_from;
238 	unsigned long charge_addr_from;
239 
240 	/* For prioritization */
241 	unsigned int min_score;
242 
243 	/* For feedback loop */
244 	unsigned long esz_bp;
245 };
246 
247 /**
248  * enum damos_wmark_metric - Represents the watermark metric.
249  *
250  * @DAMOS_WMARK_NONE:		Ignore the watermarks of the given scheme.
251  * @DAMOS_WMARK_FREE_MEM_RATE:	Free memory rate of the system in [0,1000].
252  * @NR_DAMOS_WMARK_METRICS:	Total number of DAMOS watermark metrics
253  */
254 enum damos_wmark_metric {
255 	DAMOS_WMARK_NONE,
256 	DAMOS_WMARK_FREE_MEM_RATE,
257 	NR_DAMOS_WMARK_METRICS,
258 };
259 
260 /**
261  * struct damos_watermarks - Controls when a given scheme should be activated.
262  * @metric:	Metric for the watermarks.
263  * @interval:	Watermarks check time interval in microseconds.
264  * @high:	High watermark.
265  * @mid:	Middle watermark.
266  * @low:	Low watermark.
267  *
268  * If &metric is &DAMOS_WMARK_NONE, the scheme is always active.  Being active
269  * means DAMON does monitoring and applying the action of the scheme to
270  * appropriate memory regions.  Else, DAMON checks &metric of the system for at
271  * least every &interval microseconds and works as below.
272  *
273  * If &metric is higher than &high, the scheme is inactivated.  If &metric is
274  * between &mid and &low, the scheme is activated.  If &metric is lower than
275  * &low, the scheme is inactivated.
276  */
277 struct damos_watermarks {
278 	enum damos_wmark_metric metric;
279 	unsigned long interval;
280 	unsigned long high;
281 	unsigned long mid;
282 	unsigned long low;
283 
284 /* private: */
285 	bool activated;
286 };
287 
288 /**
289  * struct damos_stat - Statistics on a given scheme.
290  * @nr_tried:	Total number of regions that the scheme is tried to be applied.
291  * @sz_tried:	Total size of regions that the scheme is tried to be applied.
292  * @nr_applied:	Total number of regions that the scheme is applied.
293  * @sz_applied:	Total size of regions that the scheme is applied.
294  * @sz_ops_filter_passed:
295  *		Total bytes that passed ops layer-handled DAMOS filters.
296  * @qt_exceeds: Total number of times the quota of the scheme has exceeded.
297  *
298  * "Tried an action to a region" in this context means the DAMOS core logic
299  * determined the region as eligible to apply the action.  The access pattern
300  * (&struct damos_access_pattern), quotas (&struct damos_quota), watermarks
301  * (&struct damos_watermarks) and filters (&struct damos_filter) that handled
302  * on core logic can affect this.  The core logic asks the operation set
303  * (&struct damon_operations) to apply the action to the region.
304  *
305  * "Applied an action to a region" in this context means the operation set
306  * (&struct damon_operations) successfully applied the action to the region, at
307  * least to a part of the region.  The filters (&struct damos_filter) that
308  * handled on operation set layer and type of the action and pages of the
309  * region can affect this.  For example, if a filter is set to exclude
310  * anonymous pages and the region has only anonymous pages, the region will be
311  * failed at applying the action.  If the action is &DAMOS_PAGEOUT and all
312  * pages of the region are already paged out, the region will be failed at
313  * applying the action.
314  */
315 struct damos_stat {
316 	unsigned long nr_tried;
317 	unsigned long sz_tried;
318 	unsigned long nr_applied;
319 	unsigned long sz_applied;
320 	unsigned long sz_ops_filter_passed;
321 	unsigned long qt_exceeds;
322 };
323 
324 /**
325  * enum damos_filter_type - Type of memory for &struct damos_filter
326  * @DAMOS_FILTER_TYPE_ANON:	Anonymous pages.
327  * @DAMOS_FILTER_TYPE_MEMCG:	Specific memcg's pages.
328  * @DAMOS_FILTER_TYPE_YOUNG:	Recently accessed pages.
329  * @DAMOS_FILTER_TYPE_ADDR:	Address range.
330  * @DAMOS_FILTER_TYPE_TARGET:	Data Access Monitoring target.
331  * @NR_DAMOS_FILTER_TYPES:	Number of filter types.
332  *
333  * The anon pages type and memcg type filters are handled by underlying
334  * &struct damon_operations as a part of scheme action trying, and therefore
335  * accounted as 'tried'.  In contrast, other types are handled by core layer
336  * before trying of the action and therefore not accounted as 'tried'.
337  *
338  * The support of the filters that handled by &struct damon_operations depend
339  * on the running &struct damon_operations.
340  * &enum DAMON_OPS_PADDR supports both anon pages type and memcg type filters,
341  * while &enum DAMON_OPS_VADDR and &enum DAMON_OPS_FVADDR don't support any of
342  * the two types.
343  */
344 enum damos_filter_type {
345 	DAMOS_FILTER_TYPE_ANON,
346 	DAMOS_FILTER_TYPE_MEMCG,
347 	DAMOS_FILTER_TYPE_YOUNG,
348 	DAMOS_FILTER_TYPE_ADDR,
349 	DAMOS_FILTER_TYPE_TARGET,
350 	NR_DAMOS_FILTER_TYPES,
351 };
352 
353 /**
354  * struct damos_filter - DAMOS action target memory filter.
355  * @type:	Type of the target memory.
356  * @matching:	Whether this is for @type-matching memory.
357  * @allow:	Whether to include or exclude the @matching memory.
358  * @memcg_id:	Memcg id of the question if @type is DAMOS_FILTER_MEMCG.
359  * @addr_range:	Address range if @type is DAMOS_FILTER_TYPE_ADDR.
360  * @target_idx:	Index of the &struct damon_target of
361  *		&damon_ctx->adaptive_targets if @type is
362  *		DAMOS_FILTER_TYPE_TARGET.
363  * @list:	List head for siblings.
364  *
365  * Before applying the &damos->action to a memory region, DAMOS checks if each
366  * byte of the region matches to this given condition and avoid applying the
367  * action if so.  Support of each filter type depends on the running &struct
368  * damon_operations and the type.  Refer to &enum damos_filter_type for more
369  * details.
370  */
371 struct damos_filter {
372 	enum damos_filter_type type;
373 	bool matching;
374 	bool allow;
375 	union {
376 		unsigned short memcg_id;
377 		struct damon_addr_range addr_range;
378 		int target_idx;
379 	};
380 	struct list_head list;
381 };
382 
383 struct damon_ctx;
384 struct damos;
385 
386 /**
387  * struct damos_walk_control - Control damos_walk().
388  *
389  * @walk_fn:	Function to be called back for each region.
390  * @data:	Data that will be passed to walk functions.
391  *
392  * Control damos_walk(), which requests specific kdamond to invoke the given
393  * function to each region that eligible to apply actions of the kdamond's
394  * schemes.  Refer to damos_walk() for more details.
395  */
396 struct damos_walk_control {
397 	void (*walk_fn)(void *data, struct damon_ctx *ctx,
398 			struct damon_target *t, struct damon_region *r,
399 			struct damos *s, unsigned long sz_filter_passed);
400 	void *data;
401 /* private: internal use only */
402 	/* informs if the kdamond finished handling of the walk request */
403 	struct completion completion;
404 	/* informs if the walk is canceled. */
405 	bool canceled;
406 };
407 
408 /**
409  * struct damos_access_pattern - Target access pattern of the given scheme.
410  * @min_sz_region:	Minimum size of target regions.
411  * @max_sz_region:	Maximum size of target regions.
412  * @min_nr_accesses:	Minimum ``->nr_accesses`` of target regions.
413  * @max_nr_accesses:	Maximum ``->nr_accesses`` of target regions.
414  * @min_age_region:	Minimum age of target regions.
415  * @max_age_region:	Maximum age of target regions.
416  */
417 struct damos_access_pattern {
418 	unsigned long min_sz_region;
419 	unsigned long max_sz_region;
420 	unsigned int min_nr_accesses;
421 	unsigned int max_nr_accesses;
422 	unsigned int min_age_region;
423 	unsigned int max_age_region;
424 };
425 
426 /**
427  * struct damos - Represents a Data Access Monitoring-based Operation Scheme.
428  * @pattern:		Access pattern of target regions.
429  * @action:		&damo_action to be applied to the target regions.
430  * @apply_interval_us:	The time between applying the @action.
431  * @quota:		Control the aggressiveness of this scheme.
432  * @wmarks:		Watermarks for automated (in)activation of this scheme.
433  * @target_nid:		Destination node if @action is "migrate_{hot,cold}".
434  * @filters:		Additional set of &struct damos_filter for &action.
435  * @stat:		Statistics of this scheme.
436  * @list:		List head for siblings.
437  *
438  * For each @apply_interval_us, DAMON finds regions which fit in the
439  * &pattern and applies &action to those. To avoid consuming too much
440  * CPU time or IO resources for the &action, &quota is used.
441  *
442  * If @apply_interval_us is zero, &damon_attrs->aggr_interval is used instead.
443  *
444  * To do the work only when needed, schemes can be activated for specific
445  * system situations using &wmarks.  If all schemes that registered to the
446  * monitoring context are inactive, DAMON stops monitoring either, and just
447  * repeatedly checks the watermarks.
448  *
449  * @target_nid is used to set the migration target node for migrate_hot or
450  * migrate_cold actions, which means it's only meaningful when @action is either
451  * "migrate_hot" or "migrate_cold".
452  *
453  * Before applying the &action to a memory region, &struct damon_operations
454  * implementation could check pages of the region and skip &action to respect
455  * &filters
456  *
457  * After applying the &action to each region, &stat_count and &stat_sz is
458  * updated to reflect the number of regions and total size of regions that the
459  * &action is applied.
460  */
461 struct damos {
462 	struct damos_access_pattern pattern;
463 	enum damos_action action;
464 	unsigned long apply_interval_us;
465 /* private: internal use only */
466 	/*
467 	 * number of sample intervals that should be passed before applying
468 	 * @action
469 	 */
470 	unsigned long next_apply_sis;
471 	/* informs if ongoing DAMOS walk for this scheme is finished */
472 	bool walk_completed;
473 /* public: */
474 	struct damos_quota quota;
475 	struct damos_watermarks wmarks;
476 	union {
477 		int target_nid;
478 	};
479 	struct list_head filters;
480 	struct damos_stat stat;
481 	struct list_head list;
482 };
483 
484 /**
485  * enum damon_ops_id - Identifier for each monitoring operations implementation
486  *
487  * @DAMON_OPS_VADDR:	Monitoring operations for virtual address spaces
488  * @DAMON_OPS_FVADDR:	Monitoring operations for only fixed ranges of virtual
489  *			address spaces
490  * @DAMON_OPS_PADDR:	Monitoring operations for the physical address space
491  * @NR_DAMON_OPS:	Number of monitoring operations implementations
492  */
493 enum damon_ops_id {
494 	DAMON_OPS_VADDR,
495 	DAMON_OPS_FVADDR,
496 	DAMON_OPS_PADDR,
497 	NR_DAMON_OPS,
498 };
499 
500 /**
501  * struct damon_operations - Monitoring operations for given use cases.
502  *
503  * @id:				Identifier of this operations set.
504  * @init:			Initialize operations-related data structures.
505  * @update:			Update operations-related data structures.
506  * @prepare_access_checks:	Prepare next access check of target regions.
507  * @check_accesses:		Check the accesses to target regions.
508  * @reset_aggregated:		Reset aggregated accesses monitoring results.
509  * @get_scheme_score:		Get the score of a region for a scheme.
510  * @apply_scheme:		Apply a DAMON-based operation scheme.
511  * @target_valid:		Determine if the target is valid.
512  * @cleanup:			Clean up the context.
513  *
514  * DAMON can be extended for various address spaces and usages.  For this,
515  * users should register the low level operations for their target address
516  * space and usecase via the &damon_ctx.ops.  Then, the monitoring thread
517  * (&damon_ctx.kdamond) calls @init and @prepare_access_checks before starting
518  * the monitoring, @update after each &damon_attrs.ops_update_interval, and
519  * @check_accesses, @target_valid and @prepare_access_checks after each
520  * &damon_attrs.sample_interval.  Finally, @reset_aggregated is called after
521  * each &damon_attrs.aggr_interval.
522  *
523  * Each &struct damon_operations instance having valid @id can be registered
524  * via damon_register_ops() and selected by damon_select_ops() later.
525  * @init should initialize operations-related data structures.  For example,
526  * this could be used to construct proper monitoring target regions and link
527  * those to @damon_ctx.adaptive_targets.
528  * @update should update the operations-related data structures.  For example,
529  * this could be used to update monitoring target regions for current status.
530  * @prepare_access_checks should manipulate the monitoring regions to be
531  * prepared for the next access check.
532  * @check_accesses should check the accesses to each region that made after the
533  * last preparation and update the number of observed accesses of each region.
534  * It should also return max number of observed accesses that made as a result
535  * of its update.  The value will be used for regions adjustment threshold.
536  * @reset_aggregated should reset the access monitoring results that aggregated
537  * by @check_accesses.
538  * @get_scheme_score should return the priority score of a region for a scheme
539  * as an integer in [0, &DAMOS_MAX_SCORE].
540  * @apply_scheme is called from @kdamond when a region for user provided
541  * DAMON-based operation scheme is found.  It should apply the scheme's action
542  * to the region and return bytes of the region that the action is successfully
543  * applied.  It should also report how many bytes of the region has passed
544  * filters (&struct damos_filter) that handled by itself.
545  * @target_valid should check whether the target is still valid for the
546  * monitoring.
547  * @cleanup is called from @kdamond just before its termination.
548  */
549 struct damon_operations {
550 	enum damon_ops_id id;
551 	void (*init)(struct damon_ctx *context);
552 	void (*update)(struct damon_ctx *context);
553 	void (*prepare_access_checks)(struct damon_ctx *context);
554 	unsigned int (*check_accesses)(struct damon_ctx *context);
555 	void (*reset_aggregated)(struct damon_ctx *context);
556 	int (*get_scheme_score)(struct damon_ctx *context,
557 			struct damon_target *t, struct damon_region *r,
558 			struct damos *scheme);
559 	unsigned long (*apply_scheme)(struct damon_ctx *context,
560 			struct damon_target *t, struct damon_region *r,
561 			struct damos *scheme, unsigned long *sz_filter_passed);
562 	bool (*target_valid)(struct damon_target *t);
563 	void (*cleanup)(struct damon_ctx *context);
564 };
565 
566 /**
567  * struct damon_callback - Monitoring events notification callbacks.
568  *
569  * @before_start:	Called before starting the monitoring.
570  * @after_wmarks_check:	Called after each schemes' watermarks check.
571  * @after_sampling:	Called after each sampling.
572  * @after_aggregation:	Called after each aggregation.
573  * @before_damos_apply:	Called before applying DAMOS action.
574  * @before_terminate:	Called before terminating the monitoring.
575  * @private:		User private data.
576  *
577  * The monitoring thread (&damon_ctx.kdamond) calls @before_start and
578  * @before_terminate just before starting and finishing the monitoring,
579  * respectively.  Therefore, those are good places for installing and cleaning
580  * @private.
581  *
582  * The monitoring thread calls @after_wmarks_check after each DAMON-based
583  * operation schemes' watermarks check.  If users need to make changes to the
584  * attributes of the monitoring context while it's deactivated due to the
585  * watermarks, this is the good place to do.
586  *
587  * The monitoring thread calls @after_sampling and @after_aggregation for each
588  * of the sampling intervals and aggregation intervals, respectively.
589  * Therefore, users can safely access the monitoring results without additional
590  * protection.  For the reason, users are recommended to use these callback for
591  * the accesses to the results.
592  *
593  * If any callback returns non-zero, monitoring stops.
594  */
595 struct damon_callback {
596 	void *private;
597 
598 	int (*before_start)(struct damon_ctx *context);
599 	int (*after_wmarks_check)(struct damon_ctx *context);
600 	int (*after_sampling)(struct damon_ctx *context);
601 	int (*after_aggregation)(struct damon_ctx *context);
602 	int (*before_damos_apply)(struct damon_ctx *context,
603 			struct damon_target *target,
604 			struct damon_region *region,
605 			struct damos *scheme);
606 	void (*before_terminate)(struct damon_ctx *context);
607 };
608 
609 /*
610  * struct damon_call_control - Control damon_call().
611  *
612  * @fn:			Function to be called back.
613  * @data:		Data that will be passed to @fn.
614  * @return_code:	Return code from @fn invocation.
615  *
616  * Control damon_call(), which requests specific kdamond to invoke a given
617  * function.  Refer to damon_call() for more details.
618  */
619 struct damon_call_control {
620 	int (*fn)(void *data);
621 	void *data;
622 	int return_code;
623 /* private: internal use only */
624 	/* informs if the kdamond finished handling of the request */
625 	struct completion completion;
626 	/* informs if the kdamond canceled @fn infocation */
627 	bool canceled;
628 };
629 
630 /**
631  * struct damon_attrs - Monitoring attributes for accuracy/overhead control.
632  *
633  * @sample_interval:		The time between access samplings.
634  * @aggr_interval:		The time between monitor results aggregations.
635  * @ops_update_interval:	The time between monitoring operations updates.
636  * @min_nr_regions:		The minimum number of adaptive monitoring
637  *				regions.
638  * @max_nr_regions:		The maximum number of adaptive monitoring
639  *				regions.
640  *
641  * For each @sample_interval, DAMON checks whether each region is accessed or
642  * not during the last @sample_interval.  If such access is found, DAMON
643  * aggregates the information by increasing &damon_region->nr_accesses for
644  * @aggr_interval time.  For each @aggr_interval, the count is reset.  DAMON
645  * also checks whether the target memory regions need update (e.g., by
646  * ``mmap()`` calls from the application, in case of virtual memory monitoring)
647  * and applies the changes for each @ops_update_interval.  All time intervals
648  * are in micro-seconds.  Please refer to &struct damon_operations and &struct
649  * damon_callback for more detail.
650  */
651 struct damon_attrs {
652 	unsigned long sample_interval;
653 	unsigned long aggr_interval;
654 	unsigned long ops_update_interval;
655 	unsigned long min_nr_regions;
656 	unsigned long max_nr_regions;
657 };
658 
659 /**
660  * struct damon_ctx - Represents a context for each monitoring.  This is the
661  * main interface that allows users to set the attributes and get the results
662  * of the monitoring.
663  *
664  * @attrs:		Monitoring attributes for accuracy/overhead control.
665  * @kdamond:		Kernel thread who does the monitoring.
666  * @kdamond_lock:	Mutex for the synchronizations with @kdamond.
667  *
668  * For each monitoring context, one kernel thread for the monitoring is
669  * created.  The pointer to the thread is stored in @kdamond.
670  *
671  * Once started, the monitoring thread runs until explicitly required to be
672  * terminated or every monitoring target is invalid.  The validity of the
673  * targets is checked via the &damon_operations.target_valid of @ops.  The
674  * termination can also be explicitly requested by calling damon_stop().
675  * The thread sets @kdamond to NULL when it terminates. Therefore, users can
676  * know whether the monitoring is ongoing or terminated by reading @kdamond.
677  * Reads and writes to @kdamond from outside of the monitoring thread must
678  * be protected by @kdamond_lock.
679  *
680  * Note that the monitoring thread protects only @kdamond via @kdamond_lock.
681  * Accesses to other fields must be protected by themselves.
682  *
683  * @ops:	Set of monitoring operations for given use cases.
684  * @callback:	Set of callbacks for monitoring events notifications.
685  *
686  * @adaptive_targets:	Head of monitoring targets (&damon_target) list.
687  * @schemes:		Head of schemes (&damos) list.
688  */
689 struct damon_ctx {
690 	struct damon_attrs attrs;
691 
692 /* private: internal use only */
693 	/* number of sample intervals that passed since this context started */
694 	unsigned long passed_sample_intervals;
695 	/*
696 	 * number of sample intervals that should be passed before next
697 	 * aggregation
698 	 */
699 	unsigned long next_aggregation_sis;
700 	/*
701 	 * number of sample intervals that should be passed before next ops
702 	 * update
703 	 */
704 	unsigned long next_ops_update_sis;
705 	/* for waiting until the execution of the kdamond_fn is started */
706 	struct completion kdamond_started;
707 	/* for scheme quotas prioritization */
708 	unsigned long *regions_score_histogram;
709 
710 	struct damon_call_control *call_control;
711 	struct mutex call_control_lock;
712 
713 	struct damos_walk_control *walk_control;
714 	struct mutex walk_control_lock;
715 
716 /* public: */
717 	struct task_struct *kdamond;
718 	struct mutex kdamond_lock;
719 
720 	struct damon_operations ops;
721 	struct damon_callback callback;
722 
723 	struct list_head adaptive_targets;
724 	struct list_head schemes;
725 };
726 
727 static inline struct damon_region *damon_next_region(struct damon_region *r)
728 {
729 	return container_of(r->list.next, struct damon_region, list);
730 }
731 
732 static inline struct damon_region *damon_prev_region(struct damon_region *r)
733 {
734 	return container_of(r->list.prev, struct damon_region, list);
735 }
736 
737 static inline struct damon_region *damon_last_region(struct damon_target *t)
738 {
739 	return list_last_entry(&t->regions_list, struct damon_region, list);
740 }
741 
742 static inline struct damon_region *damon_first_region(struct damon_target *t)
743 {
744 	return list_first_entry(&t->regions_list, struct damon_region, list);
745 }
746 
747 static inline unsigned long damon_sz_region(struct damon_region *r)
748 {
749 	return r->ar.end - r->ar.start;
750 }
751 
752 
753 #define damon_for_each_region(r, t) \
754 	list_for_each_entry(r, &t->regions_list, list)
755 
756 #define damon_for_each_region_from(r, t) \
757 	list_for_each_entry_from(r, &t->regions_list, list)
758 
759 #define damon_for_each_region_safe(r, next, t) \
760 	list_for_each_entry_safe(r, next, &t->regions_list, list)
761 
762 #define damon_for_each_target(t, ctx) \
763 	list_for_each_entry(t, &(ctx)->adaptive_targets, list)
764 
765 #define damon_for_each_target_safe(t, next, ctx)	\
766 	list_for_each_entry_safe(t, next, &(ctx)->adaptive_targets, list)
767 
768 #define damon_for_each_scheme(s, ctx) \
769 	list_for_each_entry(s, &(ctx)->schemes, list)
770 
771 #define damon_for_each_scheme_safe(s, next, ctx) \
772 	list_for_each_entry_safe(s, next, &(ctx)->schemes, list)
773 
774 #define damos_for_each_quota_goal(goal, quota) \
775 	list_for_each_entry(goal, &quota->goals, list)
776 
777 #define damos_for_each_quota_goal_safe(goal, next, quota) \
778 	list_for_each_entry_safe(goal, next, &(quota)->goals, list)
779 
780 #define damos_for_each_filter(f, scheme) \
781 	list_for_each_entry(f, &(scheme)->filters, list)
782 
783 #define damos_for_each_filter_safe(f, next, scheme) \
784 	list_for_each_entry_safe(f, next, &(scheme)->filters, list)
785 
786 #ifdef CONFIG_DAMON
787 
788 struct damon_region *damon_new_region(unsigned long start, unsigned long end);
789 
790 /*
791  * Add a region between two other regions
792  */
793 static inline void damon_insert_region(struct damon_region *r,
794 		struct damon_region *prev, struct damon_region *next,
795 		struct damon_target *t)
796 {
797 	__list_add(&r->list, &prev->list, &next->list);
798 	t->nr_regions++;
799 }
800 
801 void damon_add_region(struct damon_region *r, struct damon_target *t);
802 void damon_destroy_region(struct damon_region *r, struct damon_target *t);
803 int damon_set_regions(struct damon_target *t, struct damon_addr_range *ranges,
804 		unsigned int nr_ranges);
805 void damon_update_region_access_rate(struct damon_region *r, bool accessed,
806 		struct damon_attrs *attrs);
807 
808 struct damos_filter *damos_new_filter(enum damos_filter_type type,
809 		bool matching, bool allow);
810 void damos_add_filter(struct damos *s, struct damos_filter *f);
811 void damos_destroy_filter(struct damos_filter *f);
812 
813 struct damos_quota_goal *damos_new_quota_goal(
814 		enum damos_quota_goal_metric metric,
815 		unsigned long target_value);
816 void damos_add_quota_goal(struct damos_quota *q, struct damos_quota_goal *g);
817 void damos_destroy_quota_goal(struct damos_quota_goal *goal);
818 
819 struct damos *damon_new_scheme(struct damos_access_pattern *pattern,
820 			enum damos_action action,
821 			unsigned long apply_interval_us,
822 			struct damos_quota *quota,
823 			struct damos_watermarks *wmarks,
824 			int target_nid);
825 void damon_add_scheme(struct damon_ctx *ctx, struct damos *s);
826 void damon_destroy_scheme(struct damos *s);
827 int damos_commit_quota_goals(struct damos_quota *dst, struct damos_quota *src);
828 
829 struct damon_target *damon_new_target(void);
830 void damon_add_target(struct damon_ctx *ctx, struct damon_target *t);
831 bool damon_targets_empty(struct damon_ctx *ctx);
832 void damon_free_target(struct damon_target *t);
833 void damon_destroy_target(struct damon_target *t);
834 unsigned int damon_nr_regions(struct damon_target *t);
835 
836 struct damon_ctx *damon_new_ctx(void);
837 void damon_destroy_ctx(struct damon_ctx *ctx);
838 int damon_set_attrs(struct damon_ctx *ctx, struct damon_attrs *attrs);
839 void damon_set_schemes(struct damon_ctx *ctx,
840 			struct damos **schemes, ssize_t nr_schemes);
841 int damon_commit_ctx(struct damon_ctx *old_ctx, struct damon_ctx *new_ctx);
842 int damon_nr_running_ctxs(void);
843 bool damon_is_registered_ops(enum damon_ops_id id);
844 int damon_register_ops(struct damon_operations *ops);
845 int damon_select_ops(struct damon_ctx *ctx, enum damon_ops_id id);
846 
847 static inline bool damon_target_has_pid(const struct damon_ctx *ctx)
848 {
849 	return ctx->ops.id == DAMON_OPS_VADDR || ctx->ops.id == DAMON_OPS_FVADDR;
850 }
851 
852 static inline unsigned int damon_max_nr_accesses(const struct damon_attrs *attrs)
853 {
854 	/* {aggr,sample}_interval are unsigned long, hence could overflow */
855 	return min(attrs->aggr_interval / attrs->sample_interval,
856 			(unsigned long)UINT_MAX);
857 }
858 
859 
860 int damon_start(struct damon_ctx **ctxs, int nr_ctxs, bool exclusive);
861 int damon_stop(struct damon_ctx **ctxs, int nr_ctxs);
862 
863 int damon_call(struct damon_ctx *ctx, struct damon_call_control *control);
864 int damos_walk(struct damon_ctx *ctx, struct damos_walk_control *control);
865 
866 int damon_set_region_biggest_system_ram_default(struct damon_target *t,
867 				unsigned long *start, unsigned long *end);
868 
869 #endif	/* CONFIG_DAMON */
870 
871 #endif	/* _DAMON_H */
872