1 /* SPDX-License-Identifier: GPL-2.0 */ 2 /* 3 * DAMON api 4 * 5 * Author: SeongJae Park <[email protected]> 6 */ 7 8 #ifndef _DAMON_H_ 9 #define _DAMON_H_ 10 11 #include <linux/memcontrol.h> 12 #include <linux/mutex.h> 13 #include <linux/time64.h> 14 #include <linux/types.h> 15 #include <linux/random.h> 16 17 /* Minimal region size. Every damon_region is aligned by this. */ 18 #define DAMON_MIN_REGION PAGE_SIZE 19 /* Max priority score for DAMON-based operation schemes */ 20 #define DAMOS_MAX_SCORE (99) 21 22 /* Get a random number in [l, r) */ 23 static inline unsigned long damon_rand(unsigned long l, unsigned long r) 24 { 25 return l + get_random_u32_below(r - l); 26 } 27 28 /** 29 * struct damon_addr_range - Represents an address region of [@start, @end). 30 * @start: Start address of the region (inclusive). 31 * @end: End address of the region (exclusive). 32 */ 33 struct damon_addr_range { 34 unsigned long start; 35 unsigned long end; 36 }; 37 38 /** 39 * struct damon_region - Represents a monitoring target region. 40 * @ar: The address range of the region. 41 * @sampling_addr: Address of the sample for the next access check. 42 * @nr_accesses: Access frequency of this region. 43 * @nr_accesses_bp: @nr_accesses in basis point (0.01%) that updated for 44 * each sampling interval. 45 * @list: List head for siblings. 46 * @age: Age of this region. 47 * 48 * @nr_accesses is reset to zero for every &damon_attrs->aggr_interval and be 49 * increased for every &damon_attrs->sample_interval if an access to the region 50 * during the last sampling interval is found. The update of this field should 51 * not be done with direct access but with the helper function, 52 * damon_update_region_access_rate(). 53 * 54 * @nr_accesses_bp is another representation of @nr_accesses in basis point 55 * (1 in 10,000) that updated for every &damon_attrs->sample_interval in a 56 * manner similar to moving sum. By the algorithm, this value becomes 57 * @nr_accesses * 10000 for every &struct damon_attrs->aggr_interval. This can 58 * be used when the aggregation interval is too huge and therefore cannot wait 59 * for it before getting the access monitoring results. 60 * 61 * @age is initially zero, increased for each aggregation interval, and reset 62 * to zero again if the access frequency is significantly changed. If two 63 * regions are merged into a new region, both @nr_accesses and @age of the new 64 * region are set as region size-weighted average of those of the two regions. 65 */ 66 struct damon_region { 67 struct damon_addr_range ar; 68 unsigned long sampling_addr; 69 unsigned int nr_accesses; 70 unsigned int nr_accesses_bp; 71 struct list_head list; 72 73 unsigned int age; 74 /* private: Internal value for age calculation. */ 75 unsigned int last_nr_accesses; 76 }; 77 78 /** 79 * struct damon_target - Represents a monitoring target. 80 * @pid: The PID of the virtual address space to monitor. 81 * @nr_regions: Number of monitoring target regions of this target. 82 * @regions_list: Head of the monitoring target regions of this target. 83 * @list: List head for siblings. 84 * 85 * Each monitoring context could have multiple targets. For example, a context 86 * for virtual memory address spaces could have multiple target processes. The 87 * @pid should be set for appropriate &struct damon_operations including the 88 * virtual address spaces monitoring operations. 89 */ 90 struct damon_target { 91 struct pid *pid; 92 unsigned int nr_regions; 93 struct list_head regions_list; 94 struct list_head list; 95 }; 96 97 /** 98 * enum damos_action - Represents an action of a Data Access Monitoring-based 99 * Operation Scheme. 100 * 101 * @DAMOS_WILLNEED: Call ``madvise()`` for the region with MADV_WILLNEED. 102 * @DAMOS_COLD: Call ``madvise()`` for the region with MADV_COLD. 103 * @DAMOS_PAGEOUT: Call ``madvise()`` for the region with MADV_PAGEOUT. 104 * @DAMOS_HUGEPAGE: Call ``madvise()`` for the region with MADV_HUGEPAGE. 105 * @DAMOS_NOHUGEPAGE: Call ``madvise()`` for the region with MADV_NOHUGEPAGE. 106 * @DAMOS_LRU_PRIO: Prioritize the region on its LRU lists. 107 * @DAMOS_LRU_DEPRIO: Deprioritize the region on its LRU lists. 108 * @DAMOS_MIGRATE_HOT: Migrate the regions prioritizing warmer regions. 109 * @DAMOS_MIGRATE_COLD: Migrate the regions prioritizing colder regions. 110 * @DAMOS_STAT: Do nothing but count the stat. 111 * @NR_DAMOS_ACTIONS: Total number of DAMOS actions 112 * 113 * The support of each action is up to running &struct damon_operations. 114 * &enum DAMON_OPS_VADDR and &enum DAMON_OPS_FVADDR supports all actions except 115 * &enum DAMOS_LRU_PRIO and &enum DAMOS_LRU_DEPRIO. &enum DAMON_OPS_PADDR 116 * supports only &enum DAMOS_PAGEOUT, &enum DAMOS_LRU_PRIO, &enum 117 * DAMOS_LRU_DEPRIO, and &DAMOS_STAT. 118 */ 119 enum damos_action { 120 DAMOS_WILLNEED, 121 DAMOS_COLD, 122 DAMOS_PAGEOUT, 123 DAMOS_HUGEPAGE, 124 DAMOS_NOHUGEPAGE, 125 DAMOS_LRU_PRIO, 126 DAMOS_LRU_DEPRIO, 127 DAMOS_MIGRATE_HOT, 128 DAMOS_MIGRATE_COLD, 129 DAMOS_STAT, /* Do nothing but only record the stat */ 130 NR_DAMOS_ACTIONS, 131 }; 132 133 /** 134 * enum damos_quota_goal_metric - Represents the metric to be used as the goal 135 * 136 * @DAMOS_QUOTA_USER_INPUT: User-input value. 137 * @DAMOS_QUOTA_SOME_MEM_PSI_US: System level some memory PSI in us. 138 * @NR_DAMOS_QUOTA_GOAL_METRICS: Number of DAMOS quota goal metrics. 139 * 140 * Metrics equal to larger than @NR_DAMOS_QUOTA_GOAL_METRICS are unsupported. 141 */ 142 enum damos_quota_goal_metric { 143 DAMOS_QUOTA_USER_INPUT, 144 DAMOS_QUOTA_SOME_MEM_PSI_US, 145 NR_DAMOS_QUOTA_GOAL_METRICS, 146 }; 147 148 /** 149 * struct damos_quota_goal - DAMOS scheme quota auto-tuning goal. 150 * @metric: Metric to be used for representing the goal. 151 * @target_value: Target value of @metric to achieve with the tuning. 152 * @current_value: Current value of @metric. 153 * @last_psi_total: Last measured total PSI 154 * @list: List head for siblings. 155 * 156 * Data structure for getting the current score of the quota tuning goal. The 157 * score is calculated by how close @current_value and @target_value are. Then 158 * the score is entered to DAMON's internal feedback loop mechanism to get the 159 * auto-tuned quota. 160 * 161 * If @metric is DAMOS_QUOTA_USER_INPUT, @current_value should be manually 162 * entered by the user, probably inside the kdamond callbacks. Otherwise, 163 * DAMON sets @current_value with self-measured value of @metric. 164 */ 165 struct damos_quota_goal { 166 enum damos_quota_goal_metric metric; 167 unsigned long target_value; 168 unsigned long current_value; 169 /* metric-dependent fields */ 170 union { 171 u64 last_psi_total; 172 }; 173 struct list_head list; 174 }; 175 176 /** 177 * struct damos_quota - Controls the aggressiveness of the given scheme. 178 * @reset_interval: Charge reset interval in milliseconds. 179 * @ms: Maximum milliseconds that the scheme can use. 180 * @sz: Maximum bytes of memory that the action can be applied. 181 * @goals: Head of quota tuning goals (&damos_quota_goal) list. 182 * @esz: Effective size quota in bytes. 183 * 184 * @weight_sz: Weight of the region's size for prioritization. 185 * @weight_nr_accesses: Weight of the region's nr_accesses for prioritization. 186 * @weight_age: Weight of the region's age for prioritization. 187 * 188 * To avoid consuming too much CPU time or IO resources for applying the 189 * &struct damos->action to large memory, DAMON allows users to set time and/or 190 * size quotas. The quotas can be set by writing non-zero values to &ms and 191 * &sz, respectively. If the time quota is set, DAMON tries to use only up to 192 * &ms milliseconds within &reset_interval for applying the action. If the 193 * size quota is set, DAMON tries to apply the action only up to &sz bytes 194 * within &reset_interval. 195 * 196 * Internally, the time quota is transformed to a size quota using estimated 197 * throughput of the scheme's action. DAMON then compares it against &sz and 198 * uses smaller one as the effective quota. 199 * 200 * If @goals is not empt, DAMON calculates yet another size quota based on the 201 * goals using its internal feedback loop algorithm, for every @reset_interval. 202 * Then, if the new size quota is smaller than the effective quota, it uses the 203 * new size quota as the effective quota. 204 * 205 * The resulting effective size quota in bytes is set to @esz. 206 * 207 * For selecting regions within the quota, DAMON prioritizes current scheme's 208 * target memory regions using the &struct damon_operations->get_scheme_score. 209 * You could customize the prioritization logic by setting &weight_sz, 210 * &weight_nr_accesses, and &weight_age, because monitoring operations are 211 * encouraged to respect those. 212 */ 213 struct damos_quota { 214 unsigned long reset_interval; 215 unsigned long ms; 216 unsigned long sz; 217 struct list_head goals; 218 unsigned long esz; 219 220 unsigned int weight_sz; 221 unsigned int weight_nr_accesses; 222 unsigned int weight_age; 223 224 /* private: */ 225 /* For throughput estimation */ 226 unsigned long total_charged_sz; 227 unsigned long total_charged_ns; 228 229 /* For charging the quota */ 230 unsigned long charged_sz; 231 unsigned long charged_from; 232 struct damon_target *charge_target_from; 233 unsigned long charge_addr_from; 234 235 /* For prioritization */ 236 unsigned int min_score; 237 238 /* For feedback loop */ 239 unsigned long esz_bp; 240 }; 241 242 /** 243 * enum damos_wmark_metric - Represents the watermark metric. 244 * 245 * @DAMOS_WMARK_NONE: Ignore the watermarks of the given scheme. 246 * @DAMOS_WMARK_FREE_MEM_RATE: Free memory rate of the system in [0,1000]. 247 * @NR_DAMOS_WMARK_METRICS: Total number of DAMOS watermark metrics 248 */ 249 enum damos_wmark_metric { 250 DAMOS_WMARK_NONE, 251 DAMOS_WMARK_FREE_MEM_RATE, 252 NR_DAMOS_WMARK_METRICS, 253 }; 254 255 /** 256 * struct damos_watermarks - Controls when a given scheme should be activated. 257 * @metric: Metric for the watermarks. 258 * @interval: Watermarks check time interval in microseconds. 259 * @high: High watermark. 260 * @mid: Middle watermark. 261 * @low: Low watermark. 262 * 263 * If &metric is &DAMOS_WMARK_NONE, the scheme is always active. Being active 264 * means DAMON does monitoring and applying the action of the scheme to 265 * appropriate memory regions. Else, DAMON checks &metric of the system for at 266 * least every &interval microseconds and works as below. 267 * 268 * If &metric is higher than &high, the scheme is inactivated. If &metric is 269 * between &mid and &low, the scheme is activated. If &metric is lower than 270 * &low, the scheme is inactivated. 271 */ 272 struct damos_watermarks { 273 enum damos_wmark_metric metric; 274 unsigned long interval; 275 unsigned long high; 276 unsigned long mid; 277 unsigned long low; 278 279 /* private: */ 280 bool activated; 281 }; 282 283 /** 284 * struct damos_stat - Statistics on a given scheme. 285 * @nr_tried: Total number of regions that the scheme is tried to be applied. 286 * @sz_tried: Total size of regions that the scheme is tried to be applied. 287 * @nr_applied: Total number of regions that the scheme is applied. 288 * @sz_applied: Total size of regions that the scheme is applied. 289 * @sz_ops_filter_passed: 290 * Total bytes that passed ops layer-handled DAMOS filters. 291 * @qt_exceeds: Total number of times the quota of the scheme has exceeded. 292 * 293 * "Tried an action to a region" in this context means the DAMOS core logic 294 * determined the region as eligible to apply the action. The access pattern 295 * (&struct damos_access_pattern), quotas (&struct damos_quota), watermarks 296 * (&struct damos_watermarks) and filters (&struct damos_filter) that handled 297 * on core logic can affect this. The core logic asks the operation set 298 * (&struct damon_operations) to apply the action to the region. 299 * 300 * "Applied an action to a region" in this context means the operation set 301 * (&struct damon_operations) successfully applied the action to the region, at 302 * least to a part of the region. The filters (&struct damos_filter) that 303 * handled on operation set layer and type of the action and pages of the 304 * region can affect this. For example, if a filter is set to exclude 305 * anonymous pages and the region has only anonymous pages, the region will be 306 * failed at applying the action. If the action is &DAMOS_PAGEOUT and all 307 * pages of the region are already paged out, the region will be failed at 308 * applying the action. 309 */ 310 struct damos_stat { 311 unsigned long nr_tried; 312 unsigned long sz_tried; 313 unsigned long nr_applied; 314 unsigned long sz_applied; 315 unsigned long sz_ops_filter_passed; 316 unsigned long qt_exceeds; 317 }; 318 319 /** 320 * enum damos_filter_type - Type of memory for &struct damos_filter 321 * @DAMOS_FILTER_TYPE_ANON: Anonymous pages. 322 * @DAMOS_FILTER_TYPE_MEMCG: Specific memcg's pages. 323 * @DAMOS_FILTER_TYPE_YOUNG: Recently accessed pages. 324 * @DAMOS_FILTER_TYPE_ADDR: Address range. 325 * @DAMOS_FILTER_TYPE_TARGET: Data Access Monitoring target. 326 * @NR_DAMOS_FILTER_TYPES: Number of filter types. 327 * 328 * The anon pages type and memcg type filters are handled by underlying 329 * &struct damon_operations as a part of scheme action trying, and therefore 330 * accounted as 'tried'. In contrast, other types are handled by core layer 331 * before trying of the action and therefore not accounted as 'tried'. 332 * 333 * The support of the filters that handled by &struct damon_operations depend 334 * on the running &struct damon_operations. 335 * &enum DAMON_OPS_PADDR supports both anon pages type and memcg type filters, 336 * while &enum DAMON_OPS_VADDR and &enum DAMON_OPS_FVADDR don't support any of 337 * the two types. 338 */ 339 enum damos_filter_type { 340 DAMOS_FILTER_TYPE_ANON, 341 DAMOS_FILTER_TYPE_MEMCG, 342 DAMOS_FILTER_TYPE_YOUNG, 343 DAMOS_FILTER_TYPE_ADDR, 344 DAMOS_FILTER_TYPE_TARGET, 345 NR_DAMOS_FILTER_TYPES, 346 }; 347 348 /** 349 * struct damos_filter - DAMOS action target memory filter. 350 * @type: Type of the page. 351 * @matching: If the matching page should filtered out or in. 352 * @memcg_id: Memcg id of the question if @type is DAMOS_FILTER_MEMCG. 353 * @addr_range: Address range if @type is DAMOS_FILTER_TYPE_ADDR. 354 * @target_idx: Index of the &struct damon_target of 355 * &damon_ctx->adaptive_targets if @type is 356 * DAMOS_FILTER_TYPE_TARGET. 357 * @list: List head for siblings. 358 * 359 * Before applying the &damos->action to a memory region, DAMOS checks if each 360 * page of the region matches to this and avoid applying the action if so. 361 * Support of each filter type depends on the running &struct damon_operations 362 * and the type. Refer to &enum damos_filter_type for more detai. 363 */ 364 struct damos_filter { 365 enum damos_filter_type type; 366 bool matching; 367 union { 368 unsigned short memcg_id; 369 struct damon_addr_range addr_range; 370 int target_idx; 371 }; 372 struct list_head list; 373 }; 374 375 struct damon_ctx; 376 struct damos; 377 378 /** 379 * struct damos_walk_control - Control damos_walk(). 380 * 381 * @walk_fn: Function to be called back for each region. 382 * @data: Data that will be passed to walk functions. 383 * 384 * Control damos_walk(), which requests specific kdamond to invoke the given 385 * function to each region that eligible to apply actions of the kdamond's 386 * schemes. Refer to damos_walk() for more details. 387 */ 388 struct damos_walk_control { 389 void (*walk_fn)(void *data, struct damon_ctx *ctx, 390 struct damon_target *t, struct damon_region *r, 391 struct damos *s, unsigned long sz_filter_passed); 392 void *data; 393 /* private: internal use only */ 394 /* informs if the kdamond finished handling of the walk request */ 395 struct completion completion; 396 /* informs if the walk is canceled. */ 397 bool canceled; 398 }; 399 400 /** 401 * struct damos_access_pattern - Target access pattern of the given scheme. 402 * @min_sz_region: Minimum size of target regions. 403 * @max_sz_region: Maximum size of target regions. 404 * @min_nr_accesses: Minimum ``->nr_accesses`` of target regions. 405 * @max_nr_accesses: Maximum ``->nr_accesses`` of target regions. 406 * @min_age_region: Minimum age of target regions. 407 * @max_age_region: Maximum age of target regions. 408 */ 409 struct damos_access_pattern { 410 unsigned long min_sz_region; 411 unsigned long max_sz_region; 412 unsigned int min_nr_accesses; 413 unsigned int max_nr_accesses; 414 unsigned int min_age_region; 415 unsigned int max_age_region; 416 }; 417 418 /** 419 * struct damos - Represents a Data Access Monitoring-based Operation Scheme. 420 * @pattern: Access pattern of target regions. 421 * @action: &damo_action to be applied to the target regions. 422 * @apply_interval_us: The time between applying the @action. 423 * @quota: Control the aggressiveness of this scheme. 424 * @wmarks: Watermarks for automated (in)activation of this scheme. 425 * @target_nid: Destination node if @action is "migrate_{hot,cold}". 426 * @filters: Additional set of &struct damos_filter for &action. 427 * @stat: Statistics of this scheme. 428 * @list: List head for siblings. 429 * 430 * For each @apply_interval_us, DAMON finds regions which fit in the 431 * &pattern and applies &action to those. To avoid consuming too much 432 * CPU time or IO resources for the &action, "a is used. 433 * 434 * If @apply_interval_us is zero, &damon_attrs->aggr_interval is used instead. 435 * 436 * To do the work only when needed, schemes can be activated for specific 437 * system situations using &wmarks. If all schemes that registered to the 438 * monitoring context are inactive, DAMON stops monitoring either, and just 439 * repeatedly checks the watermarks. 440 * 441 * @target_nid is used to set the migration target node for migrate_hot or 442 * migrate_cold actions, which means it's only meaningful when @action is either 443 * "migrate_hot" or "migrate_cold". 444 * 445 * Before applying the &action to a memory region, &struct damon_operations 446 * implementation could check pages of the region and skip &action to respect 447 * &filters 448 * 449 * After applying the &action to each region, &stat_count and &stat_sz is 450 * updated to reflect the number of regions and total size of regions that the 451 * &action is applied. 452 */ 453 struct damos { 454 struct damos_access_pattern pattern; 455 enum damos_action action; 456 unsigned long apply_interval_us; 457 /* private: internal use only */ 458 /* 459 * number of sample intervals that should be passed before applying 460 * @action 461 */ 462 unsigned long next_apply_sis; 463 /* informs if ongoing DAMOS walk for this scheme is finished */ 464 bool walk_completed; 465 /* public: */ 466 struct damos_quota quota; 467 struct damos_watermarks wmarks; 468 union { 469 int target_nid; 470 }; 471 struct list_head filters; 472 struct damos_stat stat; 473 struct list_head list; 474 }; 475 476 /** 477 * enum damon_ops_id - Identifier for each monitoring operations implementation 478 * 479 * @DAMON_OPS_VADDR: Monitoring operations for virtual address spaces 480 * @DAMON_OPS_FVADDR: Monitoring operations for only fixed ranges of virtual 481 * address spaces 482 * @DAMON_OPS_PADDR: Monitoring operations for the physical address space 483 * @NR_DAMON_OPS: Number of monitoring operations implementations 484 */ 485 enum damon_ops_id { 486 DAMON_OPS_VADDR, 487 DAMON_OPS_FVADDR, 488 DAMON_OPS_PADDR, 489 NR_DAMON_OPS, 490 }; 491 492 /** 493 * struct damon_operations - Monitoring operations for given use cases. 494 * 495 * @id: Identifier of this operations set. 496 * @init: Initialize operations-related data structures. 497 * @update: Update operations-related data structures. 498 * @prepare_access_checks: Prepare next access check of target regions. 499 * @check_accesses: Check the accesses to target regions. 500 * @reset_aggregated: Reset aggregated accesses monitoring results. 501 * @get_scheme_score: Get the score of a region for a scheme. 502 * @apply_scheme: Apply a DAMON-based operation scheme. 503 * @target_valid: Determine if the target is valid. 504 * @cleanup: Clean up the context. 505 * 506 * DAMON can be extended for various address spaces and usages. For this, 507 * users should register the low level operations for their target address 508 * space and usecase via the &damon_ctx.ops. Then, the monitoring thread 509 * (&damon_ctx.kdamond) calls @init and @prepare_access_checks before starting 510 * the monitoring, @update after each &damon_attrs.ops_update_interval, and 511 * @check_accesses, @target_valid and @prepare_access_checks after each 512 * &damon_attrs.sample_interval. Finally, @reset_aggregated is called after 513 * each &damon_attrs.aggr_interval. 514 * 515 * Each &struct damon_operations instance having valid @id can be registered 516 * via damon_register_ops() and selected by damon_select_ops() later. 517 * @init should initialize operations-related data structures. For example, 518 * this could be used to construct proper monitoring target regions and link 519 * those to @damon_ctx.adaptive_targets. 520 * @update should update the operations-related data structures. For example, 521 * this could be used to update monitoring target regions for current status. 522 * @prepare_access_checks should manipulate the monitoring regions to be 523 * prepared for the next access check. 524 * @check_accesses should check the accesses to each region that made after the 525 * last preparation and update the number of observed accesses of each region. 526 * It should also return max number of observed accesses that made as a result 527 * of its update. The value will be used for regions adjustment threshold. 528 * @reset_aggregated should reset the access monitoring results that aggregated 529 * by @check_accesses. 530 * @get_scheme_score should return the priority score of a region for a scheme 531 * as an integer in [0, &DAMOS_MAX_SCORE]. 532 * @apply_scheme is called from @kdamond when a region for user provided 533 * DAMON-based operation scheme is found. It should apply the scheme's action 534 * to the region and return bytes of the region that the action is successfully 535 * applied. It should also report how many bytes of the region has passed 536 * filters (&struct damos_filter) that handled by itself. 537 * @target_valid should check whether the target is still valid for the 538 * monitoring. 539 * @cleanup is called from @kdamond just before its termination. 540 */ 541 struct damon_operations { 542 enum damon_ops_id id; 543 void (*init)(struct damon_ctx *context); 544 void (*update)(struct damon_ctx *context); 545 void (*prepare_access_checks)(struct damon_ctx *context); 546 unsigned int (*check_accesses)(struct damon_ctx *context); 547 void (*reset_aggregated)(struct damon_ctx *context); 548 int (*get_scheme_score)(struct damon_ctx *context, 549 struct damon_target *t, struct damon_region *r, 550 struct damos *scheme); 551 unsigned long (*apply_scheme)(struct damon_ctx *context, 552 struct damon_target *t, struct damon_region *r, 553 struct damos *scheme, unsigned long *sz_filter_passed); 554 bool (*target_valid)(struct damon_target *t); 555 void (*cleanup)(struct damon_ctx *context); 556 }; 557 558 /** 559 * struct damon_callback - Monitoring events notification callbacks. 560 * 561 * @before_start: Called before starting the monitoring. 562 * @after_wmarks_check: Called after each schemes' watermarks check. 563 * @after_sampling: Called after each sampling. 564 * @after_aggregation: Called after each aggregation. 565 * @before_damos_apply: Called before applying DAMOS action. 566 * @before_terminate: Called before terminating the monitoring. 567 * @private: User private data. 568 * 569 * The monitoring thread (&damon_ctx.kdamond) calls @before_start and 570 * @before_terminate just before starting and finishing the monitoring, 571 * respectively. Therefore, those are good places for installing and cleaning 572 * @private. 573 * 574 * The monitoring thread calls @after_wmarks_check after each DAMON-based 575 * operation schemes' watermarks check. If users need to make changes to the 576 * attributes of the monitoring context while it's deactivated due to the 577 * watermarks, this is the good place to do. 578 * 579 * The monitoring thread calls @after_sampling and @after_aggregation for each 580 * of the sampling intervals and aggregation intervals, respectively. 581 * Therefore, users can safely access the monitoring results without additional 582 * protection. For the reason, users are recommended to use these callback for 583 * the accesses to the results. 584 * 585 * If any callback returns non-zero, monitoring stops. 586 */ 587 struct damon_callback { 588 void *private; 589 590 int (*before_start)(struct damon_ctx *context); 591 int (*after_wmarks_check)(struct damon_ctx *context); 592 int (*after_sampling)(struct damon_ctx *context); 593 int (*after_aggregation)(struct damon_ctx *context); 594 int (*before_damos_apply)(struct damon_ctx *context, 595 struct damon_target *target, 596 struct damon_region *region, 597 struct damos *scheme); 598 void (*before_terminate)(struct damon_ctx *context); 599 }; 600 601 /* 602 * struct damon_call_control - Control damon_call(). 603 * 604 * @fn: Function to be called back. 605 * @data: Data that will be passed to @fn. 606 * @return_code: Return code from @fn invocation. 607 * 608 * Control damon_call(), which requests specific kdamond to invoke a given 609 * function. Refer to damon_call() for more details. 610 */ 611 struct damon_call_control { 612 int (*fn)(void *data); 613 void *data; 614 int return_code; 615 /* private: internal use only */ 616 /* informs if the kdamond finished handling of the request */ 617 struct completion completion; 618 /* informs if the kdamond canceled @fn infocation */ 619 bool canceled; 620 }; 621 622 /** 623 * struct damon_attrs - Monitoring attributes for accuracy/overhead control. 624 * 625 * @sample_interval: The time between access samplings. 626 * @aggr_interval: The time between monitor results aggregations. 627 * @ops_update_interval: The time between monitoring operations updates. 628 * @min_nr_regions: The minimum number of adaptive monitoring 629 * regions. 630 * @max_nr_regions: The maximum number of adaptive monitoring 631 * regions. 632 * 633 * For each @sample_interval, DAMON checks whether each region is accessed or 634 * not during the last @sample_interval. If such access is found, DAMON 635 * aggregates the information by increasing &damon_region->nr_accesses for 636 * @aggr_interval time. For each @aggr_interval, the count is reset. DAMON 637 * also checks whether the target memory regions need update (e.g., by 638 * ``mmap()`` calls from the application, in case of virtual memory monitoring) 639 * and applies the changes for each @ops_update_interval. All time intervals 640 * are in micro-seconds. Please refer to &struct damon_operations and &struct 641 * damon_callback for more detail. 642 */ 643 struct damon_attrs { 644 unsigned long sample_interval; 645 unsigned long aggr_interval; 646 unsigned long ops_update_interval; 647 unsigned long min_nr_regions; 648 unsigned long max_nr_regions; 649 }; 650 651 /** 652 * struct damon_ctx - Represents a context for each monitoring. This is the 653 * main interface that allows users to set the attributes and get the results 654 * of the monitoring. 655 * 656 * @attrs: Monitoring attributes for accuracy/overhead control. 657 * @kdamond: Kernel thread who does the monitoring. 658 * @kdamond_lock: Mutex for the synchronizations with @kdamond. 659 * 660 * For each monitoring context, one kernel thread for the monitoring is 661 * created. The pointer to the thread is stored in @kdamond. 662 * 663 * Once started, the monitoring thread runs until explicitly required to be 664 * terminated or every monitoring target is invalid. The validity of the 665 * targets is checked via the &damon_operations.target_valid of @ops. The 666 * termination can also be explicitly requested by calling damon_stop(). 667 * The thread sets @kdamond to NULL when it terminates. Therefore, users can 668 * know whether the monitoring is ongoing or terminated by reading @kdamond. 669 * Reads and writes to @kdamond from outside of the monitoring thread must 670 * be protected by @kdamond_lock. 671 * 672 * Note that the monitoring thread protects only @kdamond via @kdamond_lock. 673 * Accesses to other fields must be protected by themselves. 674 * 675 * @ops: Set of monitoring operations for given use cases. 676 * @callback: Set of callbacks for monitoring events notifications. 677 * 678 * @adaptive_targets: Head of monitoring targets (&damon_target) list. 679 * @schemes: Head of schemes (&damos) list. 680 */ 681 struct damon_ctx { 682 struct damon_attrs attrs; 683 684 /* private: internal use only */ 685 /* number of sample intervals that passed since this context started */ 686 unsigned long passed_sample_intervals; 687 /* 688 * number of sample intervals that should be passed before next 689 * aggregation 690 */ 691 unsigned long next_aggregation_sis; 692 /* 693 * number of sample intervals that should be passed before next ops 694 * update 695 */ 696 unsigned long next_ops_update_sis; 697 /* for waiting until the execution of the kdamond_fn is started */ 698 struct completion kdamond_started; 699 /* for scheme quotas prioritization */ 700 unsigned long *regions_score_histogram; 701 702 struct damon_call_control *call_control; 703 struct mutex call_control_lock; 704 705 struct damos_walk_control *walk_control; 706 struct mutex walk_control_lock; 707 708 /* public: */ 709 struct task_struct *kdamond; 710 struct mutex kdamond_lock; 711 712 struct damon_operations ops; 713 struct damon_callback callback; 714 715 struct list_head adaptive_targets; 716 struct list_head schemes; 717 }; 718 719 static inline struct damon_region *damon_next_region(struct damon_region *r) 720 { 721 return container_of(r->list.next, struct damon_region, list); 722 } 723 724 static inline struct damon_region *damon_prev_region(struct damon_region *r) 725 { 726 return container_of(r->list.prev, struct damon_region, list); 727 } 728 729 static inline struct damon_region *damon_last_region(struct damon_target *t) 730 { 731 return list_last_entry(&t->regions_list, struct damon_region, list); 732 } 733 734 static inline struct damon_region *damon_first_region(struct damon_target *t) 735 { 736 return list_first_entry(&t->regions_list, struct damon_region, list); 737 } 738 739 static inline unsigned long damon_sz_region(struct damon_region *r) 740 { 741 return r->ar.end - r->ar.start; 742 } 743 744 745 #define damon_for_each_region(r, t) \ 746 list_for_each_entry(r, &t->regions_list, list) 747 748 #define damon_for_each_region_from(r, t) \ 749 list_for_each_entry_from(r, &t->regions_list, list) 750 751 #define damon_for_each_region_safe(r, next, t) \ 752 list_for_each_entry_safe(r, next, &t->regions_list, list) 753 754 #define damon_for_each_target(t, ctx) \ 755 list_for_each_entry(t, &(ctx)->adaptive_targets, list) 756 757 #define damon_for_each_target_safe(t, next, ctx) \ 758 list_for_each_entry_safe(t, next, &(ctx)->adaptive_targets, list) 759 760 #define damon_for_each_scheme(s, ctx) \ 761 list_for_each_entry(s, &(ctx)->schemes, list) 762 763 #define damon_for_each_scheme_safe(s, next, ctx) \ 764 list_for_each_entry_safe(s, next, &(ctx)->schemes, list) 765 766 #define damos_for_each_quota_goal(goal, quota) \ 767 list_for_each_entry(goal, "a->goals, list) 768 769 #define damos_for_each_quota_goal_safe(goal, next, quota) \ 770 list_for_each_entry_safe(goal, next, &(quota)->goals, list) 771 772 #define damos_for_each_filter(f, scheme) \ 773 list_for_each_entry(f, &(scheme)->filters, list) 774 775 #define damos_for_each_filter_safe(f, next, scheme) \ 776 list_for_each_entry_safe(f, next, &(scheme)->filters, list) 777 778 #ifdef CONFIG_DAMON 779 780 struct damon_region *damon_new_region(unsigned long start, unsigned long end); 781 782 /* 783 * Add a region between two other regions 784 */ 785 static inline void damon_insert_region(struct damon_region *r, 786 struct damon_region *prev, struct damon_region *next, 787 struct damon_target *t) 788 { 789 __list_add(&r->list, &prev->list, &next->list); 790 t->nr_regions++; 791 } 792 793 void damon_add_region(struct damon_region *r, struct damon_target *t); 794 void damon_destroy_region(struct damon_region *r, struct damon_target *t); 795 int damon_set_regions(struct damon_target *t, struct damon_addr_range *ranges, 796 unsigned int nr_ranges); 797 void damon_update_region_access_rate(struct damon_region *r, bool accessed, 798 struct damon_attrs *attrs); 799 800 struct damos_filter *damos_new_filter(enum damos_filter_type type, 801 bool matching); 802 void damos_add_filter(struct damos *s, struct damos_filter *f); 803 void damos_destroy_filter(struct damos_filter *f); 804 805 struct damos_quota_goal *damos_new_quota_goal( 806 enum damos_quota_goal_metric metric, 807 unsigned long target_value); 808 void damos_add_quota_goal(struct damos_quota *q, struct damos_quota_goal *g); 809 void damos_destroy_quota_goal(struct damos_quota_goal *goal); 810 811 struct damos *damon_new_scheme(struct damos_access_pattern *pattern, 812 enum damos_action action, 813 unsigned long apply_interval_us, 814 struct damos_quota *quota, 815 struct damos_watermarks *wmarks, 816 int target_nid); 817 void damon_add_scheme(struct damon_ctx *ctx, struct damos *s); 818 void damon_destroy_scheme(struct damos *s); 819 int damos_commit_quota_goals(struct damos_quota *dst, struct damos_quota *src); 820 821 struct damon_target *damon_new_target(void); 822 void damon_add_target(struct damon_ctx *ctx, struct damon_target *t); 823 bool damon_targets_empty(struct damon_ctx *ctx); 824 void damon_free_target(struct damon_target *t); 825 void damon_destroy_target(struct damon_target *t); 826 unsigned int damon_nr_regions(struct damon_target *t); 827 828 struct damon_ctx *damon_new_ctx(void); 829 void damon_destroy_ctx(struct damon_ctx *ctx); 830 int damon_set_attrs(struct damon_ctx *ctx, struct damon_attrs *attrs); 831 void damon_set_schemes(struct damon_ctx *ctx, 832 struct damos **schemes, ssize_t nr_schemes); 833 int damon_commit_ctx(struct damon_ctx *old_ctx, struct damon_ctx *new_ctx); 834 int damon_nr_running_ctxs(void); 835 bool damon_is_registered_ops(enum damon_ops_id id); 836 int damon_register_ops(struct damon_operations *ops); 837 int damon_select_ops(struct damon_ctx *ctx, enum damon_ops_id id); 838 839 static inline bool damon_target_has_pid(const struct damon_ctx *ctx) 840 { 841 return ctx->ops.id == DAMON_OPS_VADDR || ctx->ops.id == DAMON_OPS_FVADDR; 842 } 843 844 static inline unsigned int damon_max_nr_accesses(const struct damon_attrs *attrs) 845 { 846 /* {aggr,sample}_interval are unsigned long, hence could overflow */ 847 return min(attrs->aggr_interval / attrs->sample_interval, 848 (unsigned long)UINT_MAX); 849 } 850 851 852 int damon_start(struct damon_ctx **ctxs, int nr_ctxs, bool exclusive); 853 int damon_stop(struct damon_ctx **ctxs, int nr_ctxs); 854 855 int damon_call(struct damon_ctx *ctx, struct damon_call_control *control); 856 int damos_walk(struct damon_ctx *ctx, struct damos_walk_control *control); 857 858 int damon_set_region_biggest_system_ram_default(struct damon_target *t, 859 unsigned long *start, unsigned long *end); 860 861 #endif /* CONFIG_DAMON */ 862 863 #endif /* _DAMON_H */ 864