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