damon.h

  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_SZ	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_size_range - Represents size for filter to operate on [@min, @max].
 40 * @min:	Min size (inclusive).
 41 * @max:	Max size (inclusive).
 42 */
 43struct damon_size_range {
 44	unsigned long min;
 45	unsigned long max;
 46};
 47
 48/**
 49 * struct damon_region - Represents a monitoring target region.
 50 * @ar:			The address range of the region.
 51 * @sampling_addr:	Address of the sample for the next access check.
 52 * @nr_accesses:	Access frequency of this region.
 53 * @nr_accesses_bp:	@nr_accesses in basis point (0.01%) that updated for
 54 *			each sampling interval.
 55 * @list:		List head for siblings.
 56 * @age:		Age of this region.
 57 *
 58 * @nr_accesses is reset to zero for every &damon_attrs->aggr_interval and be
 59 * increased for every &damon_attrs->sample_interval if an access to the region
 60 * during the last sampling interval is found.  The update of this field should
 61 * not be done with direct access but with the helper function,
 62 * damon_update_region_access_rate().
 63 *
 64 * @nr_accesses_bp is another representation of @nr_accesses in basis point
 65 * (1 in 10,000) that updated for every &damon_attrs->sample_interval in a
 66 * manner similar to moving sum.  By the algorithm, this value becomes
 67 * @nr_accesses * 10000 for every &struct damon_attrs->aggr_interval.  This can
 68 * be used when the aggregation interval is too huge and therefore cannot wait
 69 * for it before getting the access monitoring results.
 70 *
 71 * @age is initially zero, increased for each aggregation interval, and reset
 72 * to zero again if the access frequency is significantly changed.  If two
 73 * regions are merged into a new region, both @nr_accesses and @age of the new
 74 * region are set as region size-weighted average of those of the two regions.
 75 */
 76struct damon_region {
 77	struct damon_addr_range ar;
 78	unsigned long sampling_addr;
 79	unsigned int nr_accesses;
 80	unsigned int nr_accesses_bp;
 81	struct list_head list;
 82
 83	unsigned int age;
 84/* private: Internal value for age calculation. */
 85	unsigned int last_nr_accesses;
 86};
 87
 88/**
 89 * struct damon_target - Represents a monitoring target.
 90 * @pid:		The PID of the virtual address space to monitor.
 91 * @nr_regions:		Number of monitoring target regions of this target.
 92 * @regions_list:	Head of the monitoring target regions of this target.
 93 * @list:		List head for siblings.
 94 * @obsolete:		Whether the commit destination target is obsolete.
 95 *
 96 * Each monitoring context could have multiple targets.  For example, a context
 97 * for virtual memory address spaces could have multiple target processes.  The
 98 * @pid should be set for appropriate &struct damon_operations including the
 99 * virtual address spaces monitoring operations.
100 *
101 * @obsolete is used only for damon_commit_targets() source targets, to specify
102 * the matching destination targets are obsolete.  Read damon_commit_targets()
103 * to see how it is handled.
104 */
105struct damon_target {
106	struct pid *pid;
107	unsigned int nr_regions;
108	struct list_head regions_list;
109	struct list_head list;
110	bool obsolete;
111};
112
113/**
114 * enum damos_action - Represents an action of a Data Access Monitoring-based
115 * Operation Scheme.
116 *
117 * @DAMOS_WILLNEED:	Call ``madvise()`` for the region with MADV_WILLNEED.
118 * @DAMOS_COLD:		Call ``madvise()`` for the region with MADV_COLD.
119 * @DAMOS_PAGEOUT:	Reclaim the region.
120 * @DAMOS_HUGEPAGE:	Call ``madvise()`` for the region with MADV_HUGEPAGE.
121 * @DAMOS_NOHUGEPAGE:	Call ``madvise()`` for the region with MADV_NOHUGEPAGE.
122 * @DAMOS_LRU_PRIO:	Prioritize the region on its LRU lists.
123 * @DAMOS_LRU_DEPRIO:	Deprioritize the region on its LRU lists.
124 * @DAMOS_MIGRATE_HOT:  Migrate the regions prioritizing warmer regions.
125 * @DAMOS_MIGRATE_COLD:	Migrate the regions prioritizing colder regions.
126 * @DAMOS_STAT:		Do nothing but count the stat.
127 * @NR_DAMOS_ACTIONS:	Total number of DAMOS actions
128 *
129 * The support of each action is up to running &struct damon_operations.
130 * Refer to 'Operation Action' section of Documentation/mm/damon/design.rst for
131 * status of the supports.
132 *
133 * Note that DAMOS_PAGEOUT doesn't trigger demotions.
134 */
135enum damos_action {
136	DAMOS_WILLNEED,
137	DAMOS_COLD,
138	DAMOS_PAGEOUT,
139	DAMOS_HUGEPAGE,
140	DAMOS_NOHUGEPAGE,
141	DAMOS_LRU_PRIO,
142	DAMOS_LRU_DEPRIO,
143	DAMOS_MIGRATE_HOT,
144	DAMOS_MIGRATE_COLD,
145	DAMOS_STAT,		/* Do nothing but only record the stat */
146	NR_DAMOS_ACTIONS,
147};
148
149/**
150 * enum damos_quota_goal_metric - Represents the metric to be used as the goal
151 *
152 * @DAMOS_QUOTA_USER_INPUT:	User-input value.
153 * @DAMOS_QUOTA_SOME_MEM_PSI_US:	System level some memory PSI in us.
154 * @DAMOS_QUOTA_NODE_MEM_USED_BP:	MemUsed ratio of a node.
155 * @DAMOS_QUOTA_NODE_MEM_FREE_BP:	MemFree ratio of a node.
156 * @DAMOS_QUOTA_NODE_MEMCG_USED_BP:	MemUsed ratio of a node for a cgroup.
157 * @DAMOS_QUOTA_NODE_MEMCG_FREE_BP:	MemFree ratio of a node for a cgroup.
158 * @NR_DAMOS_QUOTA_GOAL_METRICS:	Number of DAMOS quota goal metrics.
159 *
160 * Metrics equal to larger than @NR_DAMOS_QUOTA_GOAL_METRICS are unsupported.
161 */
162enum damos_quota_goal_metric {
163	DAMOS_QUOTA_USER_INPUT,
164	DAMOS_QUOTA_SOME_MEM_PSI_US,
165	DAMOS_QUOTA_NODE_MEM_USED_BP,
166	DAMOS_QUOTA_NODE_MEM_FREE_BP,
167	DAMOS_QUOTA_NODE_MEMCG_USED_BP,
168	DAMOS_QUOTA_NODE_MEMCG_FREE_BP,
169	NR_DAMOS_QUOTA_GOAL_METRICS,
170};
171
172/**
173 * struct damos_quota_goal - DAMOS scheme quota auto-tuning goal.
174 * @metric:		Metric to be used for representing the goal.
175 * @target_value:	Target value of @metric to achieve with the tuning.
176 * @current_value:	Current value of @metric.
177 * @last_psi_total:	Last measured total PSI
178 * @nid:		Node id.
179 * @memcg_id:		Memcg id.
180 * @list:		List head for siblings.
181 *
182 * Data structure for getting the current score of the quota tuning goal.  The
183 * score is calculated by how close @current_value and @target_value are.  Then
184 * the score is entered to DAMON's internal feedback loop mechanism to get the
185 * auto-tuned quota.
186 *
187 * If @metric is DAMOS_QUOTA_USER_INPUT, @current_value should be manually
188 * entered by the user, probably inside the kdamond callbacks.  Otherwise,
189 * DAMON sets @current_value with self-measured value of @metric.
190 *
191 * If @metric is DAMOS_QUOTA_NODE_MEM_{USED,FREE}_BP, @nid represents the node
192 * id of the target node to account the used/free memory.
193 *
194 * If @metric is DAMOS_QUOTA_NODE_MEMCG_{USED,FREE}_BP, @nid and @memcg_id
195 * represents the node id and the cgroup to account the used memory for.
196 */
197struct damos_quota_goal {
198	enum damos_quota_goal_metric metric;
199	unsigned long target_value;
200	unsigned long current_value;
201	/* metric-dependent fields */
202	union {
203		u64 last_psi_total;
204		struct {
205			int nid;
206			unsigned short memcg_id;
207		};
208	};
209	struct list_head list;
210};
211
212/**
213 * struct damos_quota - Controls the aggressiveness of the given scheme.
214 * @reset_interval:	Charge reset interval in milliseconds.
215 * @ms:			Maximum milliseconds that the scheme can use.
216 * @sz:			Maximum bytes of memory that the action can be applied.
217 * @goals:		Head of quota tuning goals (&damos_quota_goal) list.
218 * @esz:		Effective size quota in bytes.
219 *
220 * @weight_sz:		Weight of the region's size for prioritization.
221 * @weight_nr_accesses:	Weight of the region's nr_accesses for prioritization.
222 * @weight_age:		Weight of the region's age for prioritization.
223 *
224 * To avoid consuming too much CPU time or IO resources for applying the
225 * &struct damos->action to large memory, DAMON allows users to set time and/or
226 * size quotas.  The quotas can be set by writing non-zero values to &ms and
227 * &sz, respectively.  If the time quota is set, DAMON tries to use only up to
228 * &ms milliseconds within &reset_interval for applying the action.  If the
229 * size quota is set, DAMON tries to apply the action only up to &sz bytes
230 * within &reset_interval.
231 *
232 * To convince the different types of quotas and goals, DAMON internally
233 * converts those into one single size quota called "effective quota".  DAMON
234 * internally uses it as the only one real quota.  The conversion is made as
235 * follows.
236 *
237 * The time quota is transformed to a size quota using estimated throughput of
238 * the scheme's action.  DAMON then compares it against &sz and uses smaller
239 * one as the effective quota.
240 *
241 * If @goals is not empty, DAMON calculates yet another size quota based on the
242 * goals using its internal feedback loop algorithm, for every @reset_interval.
243 * Then, if the new size quota is smaller than the effective quota, it uses the
244 * new size quota as the effective quota.
245 *
246 * The resulting effective size quota in bytes is set to @esz.
247 *
248 * For selecting regions within the quota, DAMON prioritizes current scheme's
249 * target memory regions using the &struct damon_operations->get_scheme_score.
250 * You could customize the prioritization logic by setting &weight_sz,
251 * &weight_nr_accesses, and &weight_age, because monitoring operations are
252 * encouraged to respect those.
253 */
254struct damos_quota {
255	unsigned long reset_interval;
256	unsigned long ms;
257	unsigned long sz;
258	struct list_head goals;
259	unsigned long esz;
260
261	unsigned int weight_sz;
262	unsigned int weight_nr_accesses;
263	unsigned int weight_age;
264
265/* private: */
266	/* For throughput estimation */
267	unsigned long total_charged_sz;
268	unsigned long total_charged_ns;
269
270	/* For charging the quota */
271	unsigned long charged_sz;
272	unsigned long charged_from;
273	struct damon_target *charge_target_from;
274	unsigned long charge_addr_from;
275
276	/* For prioritization */
277	unsigned int min_score;
278
279	/* For feedback loop */
280	unsigned long esz_bp;
281};
282
283/**
284 * enum damos_wmark_metric - Represents the watermark metric.
285 *
286 * @DAMOS_WMARK_NONE:		Ignore the watermarks of the given scheme.
287 * @DAMOS_WMARK_FREE_MEM_RATE:	Free memory rate of the system in [0,1000].
288 * @NR_DAMOS_WMARK_METRICS:	Total number of DAMOS watermark metrics
289 */
290enum damos_wmark_metric {
291	DAMOS_WMARK_NONE,
292	DAMOS_WMARK_FREE_MEM_RATE,
293	NR_DAMOS_WMARK_METRICS,
294};
295
296/**
297 * struct damos_watermarks - Controls when a given scheme should be activated.
298 * @metric:	Metric for the watermarks.
299 * @interval:	Watermarks check time interval in microseconds.
300 * @high:	High watermark.
301 * @mid:	Middle watermark.
302 * @low:	Low watermark.
303 *
304 * If &metric is &DAMOS_WMARK_NONE, the scheme is always active.  Being active
305 * means DAMON does monitoring and applying the action of the scheme to
306 * appropriate memory regions.  Else, DAMON checks &metric of the system for at
307 * least every &interval microseconds and works as below.
308 *
309 * If &metric is higher than &high, the scheme is inactivated.  If &metric is
310 * between &mid and &low, the scheme is activated.  If &metric is lower than
311 * &low, the scheme is inactivated.
312 */
313struct damos_watermarks {
314	enum damos_wmark_metric metric;
315	unsigned long interval;
316	unsigned long high;
317	unsigned long mid;
318	unsigned long low;
319
320/* private: */
321	bool activated;
322};
323
324/**
325 * struct damos_stat - Statistics on a given scheme.
326 * @nr_tried:	Total number of regions that the scheme is tried to be applied.
327 * @sz_tried:	Total size of regions that the scheme is tried to be applied.
328 * @nr_applied:	Total number of regions that the scheme is applied.
329 * @sz_applied:	Total size of regions that the scheme is applied.
330 * @sz_ops_filter_passed:
331 *		Total bytes that passed ops layer-handled DAMOS filters.
332 * @qt_exceeds: Total number of times the quota of the scheme has exceeded.
333 *
334 * "Tried an action to a region" in this context means the DAMOS core logic
335 * determined the region as eligible to apply the action.  The access pattern
336 * (&struct damos_access_pattern), quotas (&struct damos_quota), watermarks
337 * (&struct damos_watermarks) and filters (&struct damos_filter) that handled
338 * on core logic can affect this.  The core logic asks the operation set
339 * (&struct damon_operations) to apply the action to the region.
340 *
341 * "Applied an action to a region" in this context means the operation set
342 * (&struct damon_operations) successfully applied the action to the region, at
343 * least to a part of the region.  The filters (&struct damos_filter) that
344 * handled on operation set layer and type of the action and pages of the
345 * region can affect this.  For example, if a filter is set to exclude
346 * anonymous pages and the region has only anonymous pages, the region will be
347 * failed at applying the action.  If the action is &DAMOS_PAGEOUT and all
348 * pages of the region are already paged out, the region will be failed at
349 * applying the action.
350 */
351struct damos_stat {
352	unsigned long nr_tried;
353	unsigned long sz_tried;
354	unsigned long nr_applied;
355	unsigned long sz_applied;
356	unsigned long sz_ops_filter_passed;
357	unsigned long qt_exceeds;
358};
359
360/**
361 * enum damos_filter_type - Type of memory for &struct damos_filter
362 * @DAMOS_FILTER_TYPE_ANON:	Anonymous pages.
363 * @DAMOS_FILTER_TYPE_ACTIVE:	Active pages.
364 * @DAMOS_FILTER_TYPE_MEMCG:	Specific memcg's pages.
365 * @DAMOS_FILTER_TYPE_YOUNG:	Recently accessed pages.
366 * @DAMOS_FILTER_TYPE_HUGEPAGE_SIZE:	Page is part of a hugepage.
367 * @DAMOS_FILTER_TYPE_UNMAPPED:	Unmapped pages.
368 * @DAMOS_FILTER_TYPE_ADDR:	Address range.
369 * @DAMOS_FILTER_TYPE_TARGET:	Data Access Monitoring target.
370 * @NR_DAMOS_FILTER_TYPES:	Number of filter types.
371 *
372 * The anon pages type and memcg type filters are handled by underlying
373 * &struct damon_operations as a part of scheme action trying, and therefore
374 * accounted as 'tried'.  In contrast, other types are handled by core layer
375 * before trying of the action and therefore not accounted as 'tried'.
376 *
377 * The support of the filters that handled by &struct damon_operations depend
378 * on the running &struct damon_operations.
379 * &enum DAMON_OPS_PADDR supports both anon pages type and memcg type filters,
380 * while &enum DAMON_OPS_VADDR and &enum DAMON_OPS_FVADDR don't support any of
381 * the two types.
382 */
383enum damos_filter_type {
384	DAMOS_FILTER_TYPE_ANON,
385	DAMOS_FILTER_TYPE_ACTIVE,
386	DAMOS_FILTER_TYPE_MEMCG,
387	DAMOS_FILTER_TYPE_YOUNG,
388	DAMOS_FILTER_TYPE_HUGEPAGE_SIZE,
389	DAMOS_FILTER_TYPE_UNMAPPED,
390	DAMOS_FILTER_TYPE_ADDR,
391	DAMOS_FILTER_TYPE_TARGET,
392	NR_DAMOS_FILTER_TYPES,
393};
394
395/**
396 * struct damos_filter - DAMOS action target memory filter.
397 * @type:	Type of the target memory.
398 * @matching:	Whether this is for @type-matching memory.
399 * @allow:	Whether to include or exclude the @matching memory.
400 * @memcg_id:	Memcg id of the question if @type is DAMOS_FILTER_MEMCG.
401 * @addr_range:	Address range if @type is DAMOS_FILTER_TYPE_ADDR.
402 * @target_idx:	Index of the &struct damon_target of
403 *		&damon_ctx->adaptive_targets if @type is
404 *		DAMOS_FILTER_TYPE_TARGET.
405 * @sz_range:	Size range if @type is DAMOS_FILTER_TYPE_HUGEPAGE_SIZE.
406 * @list:	List head for siblings.
407 *
408 * Before applying the &damos->action to a memory region, DAMOS checks if each
409 * byte of the region matches to this given condition and avoid applying the
410 * action if so.  Support of each filter type depends on the running &struct
411 * damon_operations and the type.  Refer to &enum damos_filter_type for more
412 * details.
413 */
414struct damos_filter {
415	enum damos_filter_type type;
416	bool matching;
417	bool allow;
418	union {
419		unsigned short memcg_id;
420		struct damon_addr_range addr_range;
421		int target_idx;
422		struct damon_size_range sz_range;
423	};
424	struct list_head list;
425};
426
427struct damon_ctx;
428struct damos;
429
430/**
431 * struct damos_walk_control - Control damos_walk().
432 *
433 * @walk_fn:	Function to be called back for each region.
434 * @data:	Data that will be passed to walk functions.
435 *
436 * Control damos_walk(), which requests specific kdamond to invoke the given
437 * function to each region that eligible to apply actions of the kdamond's
438 * schemes.  Refer to damos_walk() for more details.
439 */
440struct damos_walk_control {
441	void (*walk_fn)(void *data, struct damon_ctx *ctx,
442			struct damon_target *t, struct damon_region *r,
443			struct damos *s, unsigned long sz_filter_passed);
444	void *data;
445/* private: internal use only */
446	/* informs if the kdamond finished handling of the walk request */
447	struct completion completion;
448	/* informs if the walk is canceled. */
449	bool canceled;
450};
451
452/**
453 * struct damos_access_pattern - Target access pattern of the given scheme.
454 * @min_sz_region:	Minimum size of target regions.
455 * @max_sz_region:	Maximum size of target regions.
456 * @min_nr_accesses:	Minimum ``->nr_accesses`` of target regions.
457 * @max_nr_accesses:	Maximum ``->nr_accesses`` of target regions.
458 * @min_age_region:	Minimum age of target regions.
459 * @max_age_region:	Maximum age of target regions.
460 */
461struct damos_access_pattern {
462	unsigned long min_sz_region;
463	unsigned long max_sz_region;
464	unsigned int min_nr_accesses;
465	unsigned int max_nr_accesses;
466	unsigned int min_age_region;
467	unsigned int max_age_region;
468};
469
470/**
471 * struct damos_migrate_dests - Migration destination nodes and their weights.
472 * @node_id_arr:	Array of migration destination node ids.
473 * @weight_arr:		Array of migration weights for @node_id_arr.
474 * @nr_dests:		Length of the @node_id_arr and @weight_arr arrays.
475 *
476 * @node_id_arr is an array of the ids of migration destination nodes.
477 * @weight_arr is an array of the weights for those.  The weights in
478 * @weight_arr are for nodes in @node_id_arr of same array index.
479 */
480struct damos_migrate_dests {
481	unsigned int *node_id_arr;
482	unsigned int *weight_arr;
483	size_t nr_dests;
484};
485
486/**
487 * struct damos - Represents a Data Access Monitoring-based Operation Scheme.
488 * @pattern:		Access pattern of target regions.
489 * @action:		&damos_action to be applied to the target regions.
490 * @apply_interval_us:	The time between applying the @action.
491 * @quota:		Control the aggressiveness of this scheme.
492 * @wmarks:		Watermarks for automated (in)activation of this scheme.
493 * @migrate_dests:	Destination nodes if @action is "migrate_{hot,cold}".
494 * @target_nid:		Destination node if @action is "migrate_{hot,cold}".
495 * @core_filters:	Additional set of &struct damos_filter for &action.
496 * @ops_filters:	ops layer handling &struct damos_filter objects list.
497 * @last_applied:	Last @action applied ops-managing entity.
498 * @stat:		Statistics of this scheme.
499 * @list:		List head for siblings.
500 *
501 * For each @apply_interval_us, DAMON finds regions which fit in the
502 * &pattern and applies &action to those. To avoid consuming too much
503 * CPU time or IO resources for the &action, &quota is used.
504 *
505 * If @apply_interval_us is zero, &damon_attrs->aggr_interval is used instead.
506 *
507 * To do the work only when needed, schemes can be activated for specific
508 * system situations using &wmarks.  If all schemes that registered to the
509 * monitoring context are inactive, DAMON stops monitoring either, and just
510 * repeatedly checks the watermarks.
511 *
512 * @migrate_dests specifies multiple migration target nodes with different
513 * weights for migrate_hot or migrate_cold actions.  @target_nid is ignored if
514 * this is set.
515 *
516 * @target_nid is used to set the migration target node for migrate_hot or
517 * migrate_cold actions, and @migrate_dests is unset.
518 *
519 * Before applying the &action to a memory region, &struct damon_operations
520 * implementation could check pages of the region and skip &action to respect
521 * &core_filters
522 *
523 * The minimum entity that @action can be applied depends on the underlying
524 * &struct damon_operations.  Since it may not be aligned with the core layer
525 * abstract, namely &struct damon_region, &struct damon_operations could apply
526 * @action to same entity multiple times.  Large folios that underlying on
527 * multiple &struct damon region objects could be such examples.  The &struct
528 * damon_operations can use @last_applied to avoid that.  DAMOS core logic
529 * unsets @last_applied when each regions walking for applying the scheme is
530 * finished.
531 *
532 * After applying the &action to each region, &stat_count and &stat_sz is
533 * updated to reflect the number of regions and total size of regions that the
534 * &action is applied.
535 */
536struct damos {
537	struct damos_access_pattern pattern;
538	enum damos_action action;
539	unsigned long apply_interval_us;
540/* private: internal use only */
541	/*
542	 * number of sample intervals that should be passed before applying
543	 * @action
544	 */
545	unsigned long next_apply_sis;
546	/* informs if ongoing DAMOS walk for this scheme is finished */
547	bool walk_completed;
548	/*
549	 * If the current region in the filtering stage is allowed by core
550	 * layer-handled filters.  If true, operations layer allows it, too.
551	 */
552	bool core_filters_allowed;
553	/* whether to reject core/ops filters umatched regions */
554	bool core_filters_default_reject;
555	bool ops_filters_default_reject;
556/* public: */
557	struct damos_quota quota;
558	struct damos_watermarks wmarks;
559	union {
560		struct {
561			int target_nid;
562			struct damos_migrate_dests migrate_dests;
563		};
564	};
565	struct list_head core_filters;
566	struct list_head ops_filters;
567	void *last_applied;
568	struct damos_stat stat;
569	struct list_head list;
570};
571
572/**
573 * enum damon_ops_id - Identifier for each monitoring operations implementation
574 *
575 * @DAMON_OPS_VADDR:	Monitoring operations for virtual address spaces
576 * @DAMON_OPS_FVADDR:	Monitoring operations for only fixed ranges of virtual
577 *			address spaces
578 * @DAMON_OPS_PADDR:	Monitoring operations for the physical address space
579 * @NR_DAMON_OPS:	Number of monitoring operations implementations
580 */
581enum damon_ops_id {
582	DAMON_OPS_VADDR,
583	DAMON_OPS_FVADDR,
584	DAMON_OPS_PADDR,
585	NR_DAMON_OPS,
586};
587
588/**
589 * struct damon_operations - Monitoring operations for given use cases.
590 *
591 * @id:				Identifier of this operations set.
592 * @init:			Initialize operations-related data structures.
593 * @update:			Update operations-related data structures.
594 * @prepare_access_checks:	Prepare next access check of target regions.
595 * @check_accesses:		Check the accesses to target regions.
596 * @get_scheme_score:		Get the score of a region for a scheme.
597 * @apply_scheme:		Apply a DAMON-based operation scheme.
598 * @target_valid:		Determine if the target is valid.
599 * @cleanup_target:		Clean up each target before deallocation.
600 * @cleanup:			Clean up the context.
601 *
602 * DAMON can be extended for various address spaces and usages.  For this,
603 * users should register the low level operations for their target address
604 * space and usecase via the &damon_ctx.ops.  Then, the monitoring thread
605 * (&damon_ctx.kdamond) calls @init and @prepare_access_checks before starting
606 * the monitoring, @update after each &damon_attrs.ops_update_interval, and
607 * @check_accesses, @target_valid and @prepare_access_checks after each
608 * &damon_attrs.sample_interval.
609 *
610 * Each &struct damon_operations instance having valid @id can be registered
611 * via damon_register_ops() and selected by damon_select_ops() later.
612 * @init should initialize operations-related data structures.  For example,
613 * this could be used to construct proper monitoring target regions and link
614 * those to @damon_ctx.adaptive_targets.
615 * @update should update the operations-related data structures.  For example,
616 * this could be used to update monitoring target regions for current status.
617 * @prepare_access_checks should manipulate the monitoring regions to be
618 * prepared for the next access check.
619 * @check_accesses should check the accesses to each region that made after the
620 * last preparation and update the number of observed accesses of each region.
621 * It should also return max number of observed accesses that made as a result
622 * of its update.  The value will be used for regions adjustment threshold.
623 * @get_scheme_score should return the priority score of a region for a scheme
624 * as an integer in [0, &DAMOS_MAX_SCORE].
625 * @apply_scheme is called from @kdamond when a region for user provided
626 * DAMON-based operation scheme is found.  It should apply the scheme's action
627 * to the region and return bytes of the region that the action is successfully
628 * applied.  It should also report how many bytes of the region has passed
629 * filters (&struct damos_filter) that handled by itself.
630 * @target_valid should check whether the target is still valid for the
631 * monitoring.
632 * @cleanup_target is called before the target will be deallocated.
633 * @cleanup is called from @kdamond just before its termination.
634 */
635struct damon_operations {
636	enum damon_ops_id id;
637	void (*init)(struct damon_ctx *context);
638	void (*update)(struct damon_ctx *context);
639	void (*prepare_access_checks)(struct damon_ctx *context);
640	unsigned int (*check_accesses)(struct damon_ctx *context);
641	int (*get_scheme_score)(struct damon_ctx *context,
642			struct damon_target *t, struct damon_region *r,
643			struct damos *scheme);
644	unsigned long (*apply_scheme)(struct damon_ctx *context,
645			struct damon_target *t, struct damon_region *r,
646			struct damos *scheme, unsigned long *sz_filter_passed);
647	bool (*target_valid)(struct damon_target *t);
648	void (*cleanup_target)(struct damon_target *t);
649	void (*cleanup)(struct damon_ctx *context);
650};
651
652/*
653 * struct damon_call_control - Control damon_call().
654 *
655 * @fn:			Function to be called back.
656 * @data:		Data that will be passed to @fn.
657 * @repeat:		Repeat invocations.
658 * @return_code:	Return code from @fn invocation.
659 * @dealloc_on_cancel:	De-allocate when canceled.
660 *
661 * Control damon_call(), which requests specific kdamond to invoke a given
662 * function.  Refer to damon_call() for more details.
663 */
664struct damon_call_control {
665	int (*fn)(void *data);
666	void *data;
667	bool repeat;
668	int return_code;
669	bool dealloc_on_cancel;
670/* private: internal use only */
671	/* informs if the kdamond finished handling of the request */
672	struct completion completion;
673	/* informs if the kdamond canceled @fn infocation */
674	bool canceled;
675	/* List head for siblings. */
676	struct list_head list;
677};
678
679/**
680 * struct damon_intervals_goal - Monitoring intervals auto-tuning goal.
681 *
682 * @access_bp:		Access events observation ratio to achieve in bp.
683 * @aggrs:		Number of aggregations to achieve @access_bp within.
684 * @min_sample_us:	Minimum resulting sampling interval in microseconds.
685 * @max_sample_us:	Maximum resulting sampling interval in microseconds.
686 *
687 * DAMON automatically tunes &damon_attrs->sample_interval and
688 * &damon_attrs->aggr_interval aiming the ratio in bp (1/10,000) of
689 * DAMON-observed access events to theoretical maximum amount within @aggrs
690 * aggregations be same to @access_bp.  The logic increases
691 * &damon_attrs->aggr_interval and &damon_attrs->sampling_interval in same
692 * ratio if the current access events observation ratio is lower than the
693 * target for each @aggrs aggregations, and vice versa.
694 *
695 * If @aggrs is zero, the tuning is disabled and hence this struct is ignored.
696 */
697struct damon_intervals_goal {
698	unsigned long access_bp;
699	unsigned long aggrs;
700	unsigned long min_sample_us;
701	unsigned long max_sample_us;
702};
703
704/**
705 * struct damon_attrs - Monitoring attributes for accuracy/overhead control.
706 *
707 * @sample_interval:		The time between access samplings.
708 * @aggr_interval:		The time between monitor results aggregations.
709 * @ops_update_interval:	The time between monitoring operations updates.
710 * @intervals_goal:		Intervals auto-tuning goal.
711 * @min_nr_regions:		The minimum number of adaptive monitoring
712 *				regions.
713 * @max_nr_regions:		The maximum number of adaptive monitoring
714 *				regions.
715 *
716 * For each @sample_interval, DAMON checks whether each region is accessed or
717 * not during the last @sample_interval.  If such access is found, DAMON
718 * aggregates the information by increasing &damon_region->nr_accesses for
719 * @aggr_interval time.  For each @aggr_interval, the count is reset.  DAMON
720 * also checks whether the target memory regions need update (e.g., by
721 * ``mmap()`` calls from the application, in case of virtual memory monitoring)
722 * and applies the changes for each @ops_update_interval.  All time intervals
723 * are in micro-seconds.  Please refer to &struct damon_operations and &struct
724 * damon_call_control for more detail.
725 */
726struct damon_attrs {
727	unsigned long sample_interval;
728	unsigned long aggr_interval;
729	unsigned long ops_update_interval;
730	struct damon_intervals_goal intervals_goal;
731	unsigned long min_nr_regions;
732	unsigned long max_nr_regions;
733/* private: internal use only */
734	/*
735	 * @aggr_interval to @sample_interval ratio.
736	 * Core-external components call damon_set_attrs() with &damon_attrs
737	 * that this field is unset.  In the case, damon_set_attrs() sets this
738	 * field of resulting &damon_attrs.  Core-internal components such as
739	 * kdamond_tune_intervals() calls damon_set_attrs() with &damon_attrs
740	 * that this field is set.  In the case, damon_set_attrs() just keep
741	 * it.
742	 */
743	unsigned long aggr_samples;
744};
745
746/**
747 * struct damon_ctx - Represents a context for each monitoring.  This is the
748 * main interface that allows users to set the attributes and get the results
749 * of the monitoring.
750 *
751 * @attrs:		Monitoring attributes for accuracy/overhead control.
752 * @kdamond:		Kernel thread who does the monitoring.
753 * @kdamond_lock:	Mutex for the synchronizations with @kdamond.
754 *
755 * For each monitoring context, one kernel thread for the monitoring is
756 * created.  The pointer to the thread is stored in @kdamond.
757 *
758 * Once started, the monitoring thread runs until explicitly required to be
759 * terminated or every monitoring target is invalid.  The validity of the
760 * targets is checked via the &damon_operations.target_valid of @ops.  The
761 * termination can also be explicitly requested by calling damon_stop().
762 * The thread sets @kdamond to NULL when it terminates. Therefore, users can
763 * know whether the monitoring is ongoing or terminated by reading @kdamond.
764 * Reads and writes to @kdamond from outside of the monitoring thread must
765 * be protected by @kdamond_lock.
766 *
767 * Note that the monitoring thread protects only @kdamond via @kdamond_lock.
768 * Accesses to other fields must be protected by themselves.
769 *
770 * @ops:	Set of monitoring operations for given use cases.
771 * @addr_unit:	Scale factor for core to ops address conversion.
772 * @min_region_sz:	Minimum region size.
773 * @adaptive_targets:	Head of monitoring targets (&damon_target) list.
774 * @schemes:		Head of schemes (&damos) list.
775 */
776struct damon_ctx {
777	struct damon_attrs attrs;
778
779/* private: internal use only */
780	/* number of sample intervals that passed since this context started */
781	unsigned long passed_sample_intervals;
782	/*
783	 * number of sample intervals that should be passed before next
784	 * aggregation
785	 */
786	unsigned long next_aggregation_sis;
787	/*
788	 * number of sample intervals that should be passed before next ops
789	 * update
790	 */
791	unsigned long next_ops_update_sis;
792	/*
793	 * number of sample intervals that should be passed before next
794	 * intervals tuning
795	 */
796	unsigned long next_intervals_tune_sis;
797	/* for waiting until the execution of the kdamond_fn is started */
798	struct completion kdamond_started;
799	/* for scheme quotas prioritization */
800	unsigned long *regions_score_histogram;
801
802	/* lists of &struct damon_call_control */
803	struct list_head call_controls;
804	struct mutex call_controls_lock;
805
806	struct damos_walk_control *walk_control;
807	struct mutex walk_control_lock;
808
809	/*
810	 * indicate if this may be corrupted.  Currentonly this is set only for
811	 * damon_commit_ctx() failure.
812	 */
813	bool maybe_corrupted;
814
815/* public: */
816	/* Working thread of the given DAMON context */
817	struct task_struct *kdamond;
818	struct mutex kdamond_lock;
819
820	struct damon_operations ops;
821	unsigned long addr_unit;
822	unsigned long min_region_sz;
823
824	struct list_head adaptive_targets;
825	struct list_head schemes;
826};
827
828static inline struct damon_region *damon_next_region(struct damon_region *r)
829{
830	return container_of(r->list.next, struct damon_region, list);
831}
832
833static inline struct damon_region *damon_prev_region(struct damon_region *r)
834{
835	return container_of(r->list.prev, struct damon_region, list);
836}
837
838static inline struct damon_region *damon_last_region(struct damon_target *t)
839{
840	return list_last_entry(&t->regions_list, struct damon_region, list);
841}
842
843static inline struct damon_region *damon_first_region(struct damon_target *t)
844{
845	return list_first_entry(&t->regions_list, struct damon_region, list);
846}
847
848static inline unsigned long damon_sz_region(struct damon_region *r)
849{
850	return r->ar.end - r->ar.start;
851}
852
853
854#define damon_for_each_region(r, t) \
855	list_for_each_entry(r, &t->regions_list, list)
856
857#define damon_for_each_region_from(r, t) \
858	list_for_each_entry_from(r, &t->regions_list, list)
859
860#define damon_for_each_region_safe(r, next, t) \
861	list_for_each_entry_safe(r, next, &t->regions_list, list)
862
863#define damon_for_each_target(t, ctx) \
864	list_for_each_entry(t, &(ctx)->adaptive_targets, list)
865
866#define damon_for_each_target_safe(t, next, ctx)	\
867	list_for_each_entry_safe(t, next, &(ctx)->adaptive_targets, list)
868
869#define damon_for_each_scheme(s, ctx) \
870	list_for_each_entry(s, &(ctx)->schemes, list)
871
872#define damon_for_each_scheme_safe(s, next, ctx) \
873	list_for_each_entry_safe(s, next, &(ctx)->schemes, list)
874
875#define damos_for_each_quota_goal(goal, quota) \
876	list_for_each_entry(goal, &quota->goals, list)
877
878#define damos_for_each_quota_goal_safe(goal, next, quota) \
879	list_for_each_entry_safe(goal, next, &(quota)->goals, list)
880
881#define damos_for_each_core_filter(f, scheme) \
882	list_for_each_entry(f, &(scheme)->core_filters, list)
883
884#define damos_for_each_core_filter_safe(f, next, scheme) \
885	list_for_each_entry_safe(f, next, &(scheme)->core_filters, list)
886
887#define damos_for_each_ops_filter(f, scheme) \
888	list_for_each_entry(f, &(scheme)->ops_filters, list)
889
890#define damos_for_each_ops_filter_safe(f, next, scheme) \
891	list_for_each_entry_safe(f, next, &(scheme)->ops_filters, list)
892
893#ifdef CONFIG_DAMON
894
895struct damon_region *damon_new_region(unsigned long start, unsigned long end);
896
897/*
898 * Add a region between two other regions
899 */
900static inline void damon_insert_region(struct damon_region *r,
901		struct damon_region *prev, struct damon_region *next,
902		struct damon_target *t)
903{
904	__list_add(&r->list, &prev->list, &next->list);
905	t->nr_regions++;
906}
907
908void damon_add_region(struct damon_region *r, struct damon_target *t);
909void damon_destroy_region(struct damon_region *r, struct damon_target *t);
910int damon_set_regions(struct damon_target *t, struct damon_addr_range *ranges,
911		unsigned int nr_ranges, unsigned long min_region_sz);
912void damon_update_region_access_rate(struct damon_region *r, bool accessed,
913		struct damon_attrs *attrs);
914
915struct damos_filter *damos_new_filter(enum damos_filter_type type,
916		bool matching, bool allow);
917void damos_add_filter(struct damos *s, struct damos_filter *f);
918bool damos_filter_for_ops(enum damos_filter_type type);
919void damos_destroy_filter(struct damos_filter *f);
920
921struct damos_quota_goal *damos_new_quota_goal(
922		enum damos_quota_goal_metric metric,
923		unsigned long target_value);
924void damos_add_quota_goal(struct damos_quota *q, struct damos_quota_goal *g);
925void damos_destroy_quota_goal(struct damos_quota_goal *goal);
926
927struct damos *damon_new_scheme(struct damos_access_pattern *pattern,
928			enum damos_action action,
929			unsigned long apply_interval_us,
930			struct damos_quota *quota,
931			struct damos_watermarks *wmarks,
932			int target_nid);
933void damon_add_scheme(struct damon_ctx *ctx, struct damos *s);
934void damon_destroy_scheme(struct damos *s);
935int damos_commit_quota_goals(struct damos_quota *dst, struct damos_quota *src);
936
937struct damon_target *damon_new_target(void);
938void damon_add_target(struct damon_ctx *ctx, struct damon_target *t);
939bool damon_targets_empty(struct damon_ctx *ctx);
940void damon_free_target(struct damon_target *t);
941void damon_destroy_target(struct damon_target *t, struct damon_ctx *ctx);
942unsigned int damon_nr_regions(struct damon_target *t);
943
944struct damon_ctx *damon_new_ctx(void);
945void damon_destroy_ctx(struct damon_ctx *ctx);
946int damon_set_attrs(struct damon_ctx *ctx, struct damon_attrs *attrs);
947void damon_set_schemes(struct damon_ctx *ctx,
948			struct damos **schemes, ssize_t nr_schemes);
949int damon_commit_ctx(struct damon_ctx *old_ctx, struct damon_ctx *new_ctx);
950int damon_nr_running_ctxs(void);
951bool damon_is_registered_ops(enum damon_ops_id id);
952int damon_register_ops(struct damon_operations *ops);
953int damon_select_ops(struct damon_ctx *ctx, enum damon_ops_id id);
954
955static inline bool damon_target_has_pid(const struct damon_ctx *ctx)
956{
957	return ctx->ops.id == DAMON_OPS_VADDR || ctx->ops.id == DAMON_OPS_FVADDR;
958}
959
960static inline unsigned int damon_max_nr_accesses(const struct damon_attrs *attrs)
961{
962	/* {aggr,sample}_interval are unsigned long, hence could overflow */
963	return min(attrs->aggr_interval / attrs->sample_interval,
964			(unsigned long)UINT_MAX);
965}
966
967
968bool damon_initialized(void);
969int damon_start(struct damon_ctx **ctxs, int nr_ctxs, bool exclusive);
970int damon_stop(struct damon_ctx **ctxs, int nr_ctxs);
971bool damon_is_running(struct damon_ctx *ctx);
972
973int damon_call(struct damon_ctx *ctx, struct damon_call_control *control);
974int damos_walk(struct damon_ctx *ctx, struct damos_walk_control *control);
975
976int damon_set_region_biggest_system_ram_default(struct damon_target *t,
977				unsigned long *start, unsigned long *end,
978				unsigned long min_region_sz);
979
980#endif	/* CONFIG_DAMON */
981
982#endif	/* _DAMON_H */