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