| 1 | // SPDX-License-Identifier: GPL-2.0-only |
| 2 | /* Copyright (c) 2016 Facebook |
| 3 | */ |
| 4 | #include <linux/cpumask.h> |
| 5 | #include <linux/spinlock.h> |
| 6 | #include <linux/percpu.h> |
| 7 | |
| 8 | #include "bpf_lru_list.h" |
| 9 | |
| 10 | #define LOCAL_FREE_TARGET (128) |
| 11 | #define LOCAL_NR_SCANS LOCAL_FREE_TARGET |
| 12 | |
| 13 | #define PERCPU_FREE_TARGET (4) |
| 14 | #define PERCPU_NR_SCANS PERCPU_FREE_TARGET |
| 15 | |
| 16 | /* Helpers to get the local list index */ |
| 17 | #define LOCAL_LIST_IDX(t) ((t) - BPF_LOCAL_LIST_T_OFFSET) |
| 18 | #define LOCAL_FREE_LIST_IDX LOCAL_LIST_IDX(BPF_LRU_LOCAL_LIST_T_FREE) |
| 19 | #define LOCAL_PENDING_LIST_IDX LOCAL_LIST_IDX(BPF_LRU_LOCAL_LIST_T_PENDING) |
| 20 | #define IS_LOCAL_LIST_TYPE(t) ((t) >= BPF_LOCAL_LIST_T_OFFSET) |
| 21 | |
| 22 | /* Local list helpers */ |
| 23 | static struct list_head *local_free_list(struct bpf_lru_locallist *loc_l) |
| 24 | { |
| 25 | return &loc_l->lists[LOCAL_FREE_LIST_IDX]; |
| 26 | } |
| 27 | |
| 28 | static struct list_head *local_pending_list(struct bpf_lru_locallist *loc_l) |
| 29 | { |
| 30 | return &loc_l->lists[LOCAL_PENDING_LIST_IDX]; |
| 31 | } |
| 32 | |
| 33 | /* bpf_lru_node helpers */ |
| 34 | static bool bpf_lru_node_is_ref(const struct bpf_lru_node *node) |
| 35 | { |
| 36 | return READ_ONCE(node->ref); |
| 37 | } |
| 38 | |
| 39 | static void bpf_lru_node_clear_ref(struct bpf_lru_node *node) |
| 40 | { |
| 41 | WRITE_ONCE(node->ref, 0); |
| 42 | } |
| 43 | |
| 44 | static void bpf_lru_list_count_inc(struct bpf_lru_list *l, |
| 45 | enum bpf_lru_list_type type) |
| 46 | { |
| 47 | if (type < NR_BPF_LRU_LIST_COUNT) |
| 48 | l->counts[type]++; |
| 49 | } |
| 50 | |
| 51 | static void bpf_lru_list_count_dec(struct bpf_lru_list *l, |
| 52 | enum bpf_lru_list_type type) |
| 53 | { |
| 54 | if (type < NR_BPF_LRU_LIST_COUNT) |
| 55 | l->counts[type]--; |
| 56 | } |
| 57 | |
| 58 | static void __bpf_lru_node_move_to_free(struct bpf_lru_list *l, |
| 59 | struct bpf_lru_node *node, |
| 60 | struct list_head *free_list, |
| 61 | enum bpf_lru_list_type tgt_free_type) |
| 62 | { |
| 63 | if (WARN_ON_ONCE(IS_LOCAL_LIST_TYPE(node->type))) |
| 64 | return; |
| 65 | |
| 66 | /* If the removing node is the next_inactive_rotation candidate, |
| 67 | * move the next_inactive_rotation pointer also. |
| 68 | */ |
| 69 | if (&node->list == l->next_inactive_rotation) |
| 70 | l->next_inactive_rotation = l->next_inactive_rotation->prev; |
| 71 | |
| 72 | bpf_lru_list_count_dec(l, type: node->type); |
| 73 | |
| 74 | node->type = tgt_free_type; |
| 75 | list_move(list: &node->list, head: free_list); |
| 76 | } |
| 77 | |
| 78 | /* Move nodes from local list to the LRU list */ |
| 79 | static void __bpf_lru_node_move_in(struct bpf_lru_list *l, |
| 80 | struct bpf_lru_node *node, |
| 81 | enum bpf_lru_list_type tgt_type) |
| 82 | { |
| 83 | if (WARN_ON_ONCE(!IS_LOCAL_LIST_TYPE(node->type)) || |
| 84 | WARN_ON_ONCE(IS_LOCAL_LIST_TYPE(tgt_type))) |
| 85 | return; |
| 86 | |
| 87 | bpf_lru_list_count_inc(l, type: tgt_type); |
| 88 | node->type = tgt_type; |
| 89 | bpf_lru_node_clear_ref(node); |
| 90 | list_move(list: &node->list, head: &l->lists[tgt_type]); |
| 91 | } |
| 92 | |
| 93 | /* Move nodes between or within active and inactive list (like |
| 94 | * active to inactive, inactive to active or tail of active back to |
| 95 | * the head of active). |
| 96 | */ |
| 97 | static void __bpf_lru_node_move(struct bpf_lru_list *l, |
| 98 | struct bpf_lru_node *node, |
| 99 | enum bpf_lru_list_type tgt_type) |
| 100 | { |
| 101 | if (WARN_ON_ONCE(IS_LOCAL_LIST_TYPE(node->type)) || |
| 102 | WARN_ON_ONCE(IS_LOCAL_LIST_TYPE(tgt_type))) |
| 103 | return; |
| 104 | |
| 105 | if (node->type != tgt_type) { |
| 106 | bpf_lru_list_count_dec(l, type: node->type); |
| 107 | bpf_lru_list_count_inc(l, type: tgt_type); |
| 108 | node->type = tgt_type; |
| 109 | } |
| 110 | bpf_lru_node_clear_ref(node); |
| 111 | |
| 112 | /* If the moving node is the next_inactive_rotation candidate, |
| 113 | * move the next_inactive_rotation pointer also. |
| 114 | */ |
| 115 | if (&node->list == l->next_inactive_rotation) |
| 116 | l->next_inactive_rotation = l->next_inactive_rotation->prev; |
| 117 | |
| 118 | list_move(list: &node->list, head: &l->lists[tgt_type]); |
| 119 | } |
| 120 | |
| 121 | static bool bpf_lru_list_inactive_low(const struct bpf_lru_list *l) |
| 122 | { |
| 123 | return l->counts[BPF_LRU_LIST_T_INACTIVE] < |
| 124 | l->counts[BPF_LRU_LIST_T_ACTIVE]; |
| 125 | } |
| 126 | |
| 127 | /* Rotate the active list: |
| 128 | * 1. Start from tail |
| 129 | * 2. If the node has the ref bit set, it will be rotated |
| 130 | * back to the head of active list with the ref bit cleared. |
| 131 | * Give this node one more chance to survive in the active list. |
| 132 | * 3. If the ref bit is not set, move it to the head of the |
| 133 | * inactive list. |
| 134 | * 4. It will at most scan nr_scans nodes |
| 135 | */ |
| 136 | static void __bpf_lru_list_rotate_active(struct bpf_lru *lru, |
| 137 | struct bpf_lru_list *l) |
| 138 | { |
| 139 | struct list_head *active = &l->lists[BPF_LRU_LIST_T_ACTIVE]; |
| 140 | struct bpf_lru_node *node, *tmp_node, *first_node; |
| 141 | unsigned int i = 0; |
| 142 | |
| 143 | first_node = list_first_entry(active, struct bpf_lru_node, list); |
| 144 | list_for_each_entry_safe_reverse(node, tmp_node, active, list) { |
| 145 | if (bpf_lru_node_is_ref(node)) |
| 146 | __bpf_lru_node_move(l, node, tgt_type: BPF_LRU_LIST_T_ACTIVE); |
| 147 | else |
| 148 | __bpf_lru_node_move(l, node, tgt_type: BPF_LRU_LIST_T_INACTIVE); |
| 149 | |
| 150 | if (++i == lru->nr_scans || node == first_node) |
| 151 | break; |
| 152 | } |
| 153 | } |
| 154 | |
| 155 | /* Rotate the inactive list. It starts from the next_inactive_rotation |
| 156 | * 1. If the node has ref bit set, it will be moved to the head |
| 157 | * of active list with the ref bit cleared. |
| 158 | * 2. If the node does not have ref bit set, it will leave it |
| 159 | * at its current location (i.e. do nothing) so that it can |
| 160 | * be considered during the next inactive_shrink. |
| 161 | * 3. It will at most scan nr_scans nodes |
| 162 | */ |
| 163 | static void __bpf_lru_list_rotate_inactive(struct bpf_lru *lru, |
| 164 | struct bpf_lru_list *l) |
| 165 | { |
| 166 | struct list_head *inactive = &l->lists[BPF_LRU_LIST_T_INACTIVE]; |
| 167 | struct list_head *cur, *last, *next = inactive; |
| 168 | struct bpf_lru_node *node; |
| 169 | unsigned int i = 0; |
| 170 | |
| 171 | if (list_empty(head: inactive)) |
| 172 | return; |
| 173 | |
| 174 | last = l->next_inactive_rotation->next; |
| 175 | if (last == inactive) |
| 176 | last = last->next; |
| 177 | |
| 178 | cur = l->next_inactive_rotation; |
| 179 | while (i < lru->nr_scans) { |
| 180 | if (cur == inactive) { |
| 181 | cur = cur->prev; |
| 182 | continue; |
| 183 | } |
| 184 | |
| 185 | node = list_entry(cur, struct bpf_lru_node, list); |
| 186 | next = cur->prev; |
| 187 | if (bpf_lru_node_is_ref(node)) |
| 188 | __bpf_lru_node_move(l, node, tgt_type: BPF_LRU_LIST_T_ACTIVE); |
| 189 | if (cur == last) |
| 190 | break; |
| 191 | cur = next; |
| 192 | i++; |
| 193 | } |
| 194 | |
| 195 | l->next_inactive_rotation = next; |
| 196 | } |
| 197 | |
| 198 | /* Shrink the inactive list. It starts from the tail of the |
| 199 | * inactive list and only move the nodes without the ref bit |
| 200 | * set to the designated free list. |
| 201 | */ |
| 202 | static unsigned int |
| 203 | __bpf_lru_list_shrink_inactive(struct bpf_lru *lru, |
| 204 | struct bpf_lru_list *l, |
| 205 | unsigned int tgt_nshrink, |
| 206 | struct list_head *free_list, |
| 207 | enum bpf_lru_list_type tgt_free_type) |
| 208 | { |
| 209 | struct list_head *inactive = &l->lists[BPF_LRU_LIST_T_INACTIVE]; |
| 210 | struct bpf_lru_node *node, *tmp_node; |
| 211 | unsigned int nshrinked = 0; |
| 212 | unsigned int i = 0; |
| 213 | |
| 214 | list_for_each_entry_safe_reverse(node, tmp_node, inactive, list) { |
| 215 | if (bpf_lru_node_is_ref(node)) { |
| 216 | __bpf_lru_node_move(l, node, tgt_type: BPF_LRU_LIST_T_ACTIVE); |
| 217 | } else if (lru->del_from_htab(lru->del_arg, node)) { |
| 218 | __bpf_lru_node_move_to_free(l, node, free_list, |
| 219 | tgt_free_type); |
| 220 | if (++nshrinked == tgt_nshrink) |
| 221 | break; |
| 222 | } |
| 223 | |
| 224 | if (++i == lru->nr_scans) |
| 225 | break; |
| 226 | } |
| 227 | |
| 228 | return nshrinked; |
| 229 | } |
| 230 | |
| 231 | /* 1. Rotate the active list (if needed) |
| 232 | * 2. Always rotate the inactive list |
| 233 | */ |
| 234 | static void __bpf_lru_list_rotate(struct bpf_lru *lru, struct bpf_lru_list *l) |
| 235 | { |
| 236 | if (bpf_lru_list_inactive_low(l)) |
| 237 | __bpf_lru_list_rotate_active(lru, l); |
| 238 | |
| 239 | __bpf_lru_list_rotate_inactive(lru, l); |
| 240 | } |
| 241 | |
| 242 | /* Calls __bpf_lru_list_shrink_inactive() to shrink some |
| 243 | * ref-bit-cleared nodes and move them to the designated |
| 244 | * free list. |
| 245 | * |
| 246 | * If it cannot get a free node after calling |
| 247 | * __bpf_lru_list_shrink_inactive(). It will just remove |
| 248 | * one node from either inactive or active list without |
| 249 | * honoring the ref-bit. It prefers inactive list to active |
| 250 | * list in this situation. |
| 251 | */ |
| 252 | static unsigned int __bpf_lru_list_shrink(struct bpf_lru *lru, |
| 253 | struct bpf_lru_list *l, |
| 254 | unsigned int tgt_nshrink, |
| 255 | struct list_head *free_list, |
| 256 | enum bpf_lru_list_type tgt_free_type) |
| 257 | |
| 258 | { |
| 259 | struct bpf_lru_node *node, *tmp_node; |
| 260 | struct list_head *force_shrink_list; |
| 261 | unsigned int nshrinked; |
| 262 | |
| 263 | nshrinked = __bpf_lru_list_shrink_inactive(lru, l, tgt_nshrink, |
| 264 | free_list, tgt_free_type); |
| 265 | if (nshrinked) |
| 266 | return nshrinked; |
| 267 | |
| 268 | /* Do a force shrink by ignoring the reference bit */ |
| 269 | if (!list_empty(head: &l->lists[BPF_LRU_LIST_T_INACTIVE])) |
| 270 | force_shrink_list = &l->lists[BPF_LRU_LIST_T_INACTIVE]; |
| 271 | else |
| 272 | force_shrink_list = &l->lists[BPF_LRU_LIST_T_ACTIVE]; |
| 273 | |
| 274 | list_for_each_entry_safe_reverse(node, tmp_node, force_shrink_list, |
| 275 | list) { |
| 276 | if (lru->del_from_htab(lru->del_arg, node)) { |
| 277 | __bpf_lru_node_move_to_free(l, node, free_list, |
| 278 | tgt_free_type); |
| 279 | return 1; |
| 280 | } |
| 281 | } |
| 282 | |
| 283 | return 0; |
| 284 | } |
| 285 | |
| 286 | /* Flush the nodes from the local pending list to the LRU list */ |
| 287 | static void __local_list_flush(struct bpf_lru_list *l, |
| 288 | struct bpf_lru_locallist *loc_l) |
| 289 | { |
| 290 | struct bpf_lru_node *node, *tmp_node; |
| 291 | |
| 292 | list_for_each_entry_safe_reverse(node, tmp_node, |
| 293 | local_pending_list(loc_l), list) { |
| 294 | if (bpf_lru_node_is_ref(node)) |
| 295 | __bpf_lru_node_move_in(l, node, tgt_type: BPF_LRU_LIST_T_ACTIVE); |
| 296 | else |
| 297 | __bpf_lru_node_move_in(l, node, |
| 298 | tgt_type: BPF_LRU_LIST_T_INACTIVE); |
| 299 | } |
| 300 | } |
| 301 | |
| 302 | static void bpf_lru_list_push_free(struct bpf_lru_list *l, |
| 303 | struct bpf_lru_node *node) |
| 304 | { |
| 305 | unsigned long flags; |
| 306 | |
| 307 | if (WARN_ON_ONCE(IS_LOCAL_LIST_TYPE(node->type))) |
| 308 | return; |
| 309 | |
| 310 | raw_spin_lock_irqsave(&l->lock, flags); |
| 311 | __bpf_lru_node_move(l, node, tgt_type: BPF_LRU_LIST_T_FREE); |
| 312 | raw_spin_unlock_irqrestore(&l->lock, flags); |
| 313 | } |
| 314 | |
| 315 | static void bpf_lru_list_pop_free_to_local(struct bpf_lru *lru, |
| 316 | struct bpf_lru_locallist *loc_l) |
| 317 | { |
| 318 | struct bpf_lru_list *l = &lru->common_lru.lru_list; |
| 319 | struct bpf_lru_node *node, *tmp_node; |
| 320 | unsigned int nfree = 0; |
| 321 | |
| 322 | raw_spin_lock(&l->lock); |
| 323 | |
| 324 | __local_list_flush(l, loc_l); |
| 325 | |
| 326 | __bpf_lru_list_rotate(lru, l); |
| 327 | |
| 328 | list_for_each_entry_safe(node, tmp_node, &l->lists[BPF_LRU_LIST_T_FREE], |
| 329 | list) { |
| 330 | __bpf_lru_node_move_to_free(l, node, free_list: local_free_list(loc_l), |
| 331 | tgt_free_type: BPF_LRU_LOCAL_LIST_T_FREE); |
| 332 | if (++nfree == lru->target_free) |
| 333 | break; |
| 334 | } |
| 335 | |
| 336 | if (nfree < lru->target_free) |
| 337 | __bpf_lru_list_shrink(lru, l, tgt_nshrink: lru->target_free - nfree, |
| 338 | free_list: local_free_list(loc_l), |
| 339 | tgt_free_type: BPF_LRU_LOCAL_LIST_T_FREE); |
| 340 | |
| 341 | raw_spin_unlock(&l->lock); |
| 342 | } |
| 343 | |
| 344 | static void __local_list_add_pending(struct bpf_lru *lru, |
| 345 | struct bpf_lru_locallist *loc_l, |
| 346 | int cpu, |
| 347 | struct bpf_lru_node *node, |
| 348 | u32 hash) |
| 349 | { |
| 350 | *(u32 *)((void *)node + lru->hash_offset) = hash; |
| 351 | node->cpu = cpu; |
| 352 | node->type = BPF_LRU_LOCAL_LIST_T_PENDING; |
| 353 | bpf_lru_node_clear_ref(node); |
| 354 | list_add(new: &node->list, head: local_pending_list(loc_l)); |
| 355 | } |
| 356 | |
| 357 | static struct bpf_lru_node * |
| 358 | __local_list_pop_free(struct bpf_lru_locallist *loc_l) |
| 359 | { |
| 360 | struct bpf_lru_node *node; |
| 361 | |
| 362 | node = list_first_entry_or_null(local_free_list(loc_l), |
| 363 | struct bpf_lru_node, |
| 364 | list); |
| 365 | if (node) |
| 366 | list_del(entry: &node->list); |
| 367 | |
| 368 | return node; |
| 369 | } |
| 370 | |
| 371 | static struct bpf_lru_node * |
| 372 | __local_list_pop_pending(struct bpf_lru *lru, struct bpf_lru_locallist *loc_l) |
| 373 | { |
| 374 | struct bpf_lru_node *node; |
| 375 | bool force = false; |
| 376 | |
| 377 | ignore_ref: |
| 378 | /* Get from the tail (i.e. older element) of the pending list. */ |
| 379 | list_for_each_entry_reverse(node, local_pending_list(loc_l), |
| 380 | list) { |
| 381 | if ((!bpf_lru_node_is_ref(node) || force) && |
| 382 | lru->del_from_htab(lru->del_arg, node)) { |
| 383 | list_del(entry: &node->list); |
| 384 | return node; |
| 385 | } |
| 386 | } |
| 387 | |
| 388 | if (!force) { |
| 389 | force = true; |
| 390 | goto ignore_ref; |
| 391 | } |
| 392 | |
| 393 | return NULL; |
| 394 | } |
| 395 | |
| 396 | static struct bpf_lru_node *bpf_percpu_lru_pop_free(struct bpf_lru *lru, |
| 397 | u32 hash) |
| 398 | { |
| 399 | struct list_head *free_list; |
| 400 | struct bpf_lru_node *node = NULL; |
| 401 | struct bpf_lru_list *l; |
| 402 | unsigned long flags; |
| 403 | int cpu = raw_smp_processor_id(); |
| 404 | |
| 405 | l = per_cpu_ptr(lru->percpu_lru, cpu); |
| 406 | |
| 407 | raw_spin_lock_irqsave(&l->lock, flags); |
| 408 | |
| 409 | __bpf_lru_list_rotate(lru, l); |
| 410 | |
| 411 | free_list = &l->lists[BPF_LRU_LIST_T_FREE]; |
| 412 | if (list_empty(head: free_list)) |
| 413 | __bpf_lru_list_shrink(lru, l, PERCPU_FREE_TARGET, free_list, |
| 414 | tgt_free_type: BPF_LRU_LIST_T_FREE); |
| 415 | |
| 416 | if (!list_empty(head: free_list)) { |
| 417 | node = list_first_entry(free_list, struct bpf_lru_node, list); |
| 418 | *(u32 *)((void *)node + lru->hash_offset) = hash; |
| 419 | bpf_lru_node_clear_ref(node); |
| 420 | __bpf_lru_node_move(l, node, tgt_type: BPF_LRU_LIST_T_INACTIVE); |
| 421 | } |
| 422 | |
| 423 | raw_spin_unlock_irqrestore(&l->lock, flags); |
| 424 | |
| 425 | return node; |
| 426 | } |
| 427 | |
| 428 | static struct bpf_lru_node *bpf_common_lru_pop_free(struct bpf_lru *lru, |
| 429 | u32 hash) |
| 430 | { |
| 431 | struct bpf_lru_locallist *loc_l, *steal_loc_l; |
| 432 | struct bpf_common_lru *clru = &lru->common_lru; |
| 433 | struct bpf_lru_node *node; |
| 434 | int steal, first_steal; |
| 435 | unsigned long flags; |
| 436 | int cpu = raw_smp_processor_id(); |
| 437 | |
| 438 | loc_l = per_cpu_ptr(clru->local_list, cpu); |
| 439 | |
| 440 | raw_spin_lock_irqsave(&loc_l->lock, flags); |
| 441 | |
| 442 | node = __local_list_pop_free(loc_l); |
| 443 | if (!node) { |
| 444 | bpf_lru_list_pop_free_to_local(lru, loc_l); |
| 445 | node = __local_list_pop_free(loc_l); |
| 446 | } |
| 447 | |
| 448 | if (node) |
| 449 | __local_list_add_pending(lru, loc_l, cpu, node, hash); |
| 450 | |
| 451 | raw_spin_unlock_irqrestore(&loc_l->lock, flags); |
| 452 | |
| 453 | if (node) |
| 454 | return node; |
| 455 | |
| 456 | /* No free nodes found from the local free list and |
| 457 | * the global LRU list. |
| 458 | * |
| 459 | * Steal from the local free/pending list of the |
| 460 | * current CPU and remote CPU in RR. It starts |
| 461 | * with the loc_l->next_steal CPU. |
| 462 | */ |
| 463 | |
| 464 | first_steal = loc_l->next_steal; |
| 465 | steal = first_steal; |
| 466 | do { |
| 467 | steal_loc_l = per_cpu_ptr(clru->local_list, steal); |
| 468 | |
| 469 | raw_spin_lock_irqsave(&steal_loc_l->lock, flags); |
| 470 | |
| 471 | node = __local_list_pop_free(loc_l: steal_loc_l); |
| 472 | if (!node) |
| 473 | node = __local_list_pop_pending(lru, loc_l: steal_loc_l); |
| 474 | |
| 475 | raw_spin_unlock_irqrestore(&steal_loc_l->lock, flags); |
| 476 | |
| 477 | steal = cpumask_next_wrap(n: steal, cpu_possible_mask); |
| 478 | } while (!node && steal != first_steal); |
| 479 | |
| 480 | loc_l->next_steal = steal; |
| 481 | |
| 482 | if (node) { |
| 483 | raw_spin_lock_irqsave(&loc_l->lock, flags); |
| 484 | __local_list_add_pending(lru, loc_l, cpu, node, hash); |
| 485 | raw_spin_unlock_irqrestore(&loc_l->lock, flags); |
| 486 | } |
| 487 | |
| 488 | return node; |
| 489 | } |
| 490 | |
| 491 | struct bpf_lru_node *bpf_lru_pop_free(struct bpf_lru *lru, u32 hash) |
| 492 | { |
| 493 | if (lru->percpu) |
| 494 | return bpf_percpu_lru_pop_free(lru, hash); |
| 495 | else |
| 496 | return bpf_common_lru_pop_free(lru, hash); |
| 497 | } |
| 498 | |
| 499 | static void bpf_common_lru_push_free(struct bpf_lru *lru, |
| 500 | struct bpf_lru_node *node) |
| 501 | { |
| 502 | u8 node_type = READ_ONCE(node->type); |
| 503 | unsigned long flags; |
| 504 | |
| 505 | if (WARN_ON_ONCE(node_type == BPF_LRU_LIST_T_FREE) || |
| 506 | WARN_ON_ONCE(node_type == BPF_LRU_LOCAL_LIST_T_FREE)) |
| 507 | return; |
| 508 | |
| 509 | if (node_type == BPF_LRU_LOCAL_LIST_T_PENDING) { |
| 510 | struct bpf_lru_locallist *loc_l; |
| 511 | |
| 512 | loc_l = per_cpu_ptr(lru->common_lru.local_list, node->cpu); |
| 513 | |
| 514 | raw_spin_lock_irqsave(&loc_l->lock, flags); |
| 515 | |
| 516 | if (unlikely(node->type != BPF_LRU_LOCAL_LIST_T_PENDING)) { |
| 517 | raw_spin_unlock_irqrestore(&loc_l->lock, flags); |
| 518 | goto check_lru_list; |
| 519 | } |
| 520 | |
| 521 | node->type = BPF_LRU_LOCAL_LIST_T_FREE; |
| 522 | bpf_lru_node_clear_ref(node); |
| 523 | list_move(list: &node->list, head: local_free_list(loc_l)); |
| 524 | |
| 525 | raw_spin_unlock_irqrestore(&loc_l->lock, flags); |
| 526 | return; |
| 527 | } |
| 528 | |
| 529 | check_lru_list: |
| 530 | bpf_lru_list_push_free(l: &lru->common_lru.lru_list, node); |
| 531 | } |
| 532 | |
| 533 | static void bpf_percpu_lru_push_free(struct bpf_lru *lru, |
| 534 | struct bpf_lru_node *node) |
| 535 | { |
| 536 | struct bpf_lru_list *l; |
| 537 | unsigned long flags; |
| 538 | |
| 539 | l = per_cpu_ptr(lru->percpu_lru, node->cpu); |
| 540 | |
| 541 | raw_spin_lock_irqsave(&l->lock, flags); |
| 542 | |
| 543 | __bpf_lru_node_move(l, node, tgt_type: BPF_LRU_LIST_T_FREE); |
| 544 | |
| 545 | raw_spin_unlock_irqrestore(&l->lock, flags); |
| 546 | } |
| 547 | |
| 548 | void bpf_lru_push_free(struct bpf_lru *lru, struct bpf_lru_node *node) |
| 549 | { |
| 550 | if (lru->percpu) |
| 551 | bpf_percpu_lru_push_free(lru, node); |
| 552 | else |
| 553 | bpf_common_lru_push_free(lru, node); |
| 554 | } |
| 555 | |
| 556 | static void bpf_common_lru_populate(struct bpf_lru *lru, void *buf, |
| 557 | u32 node_offset, u32 elem_size, |
| 558 | u32 nr_elems) |
| 559 | { |
| 560 | struct bpf_lru_list *l = &lru->common_lru.lru_list; |
| 561 | u32 i; |
| 562 | |
| 563 | for (i = 0; i < nr_elems; i++) { |
| 564 | struct bpf_lru_node *node; |
| 565 | |
| 566 | node = (struct bpf_lru_node *)(buf + node_offset); |
| 567 | node->type = BPF_LRU_LIST_T_FREE; |
| 568 | bpf_lru_node_clear_ref(node); |
| 569 | list_add(new: &node->list, head: &l->lists[BPF_LRU_LIST_T_FREE]); |
| 570 | buf += elem_size; |
| 571 | } |
| 572 | |
| 573 | lru->target_free = clamp((nr_elems / num_possible_cpus()) / 2, |
| 574 | 1, LOCAL_FREE_TARGET); |
| 575 | } |
| 576 | |
| 577 | static void bpf_percpu_lru_populate(struct bpf_lru *lru, void *buf, |
| 578 | u32 node_offset, u32 elem_size, |
| 579 | u32 nr_elems) |
| 580 | { |
| 581 | u32 i, pcpu_entries; |
| 582 | int cpu; |
| 583 | struct bpf_lru_list *l; |
| 584 | |
| 585 | pcpu_entries = nr_elems / num_possible_cpus(); |
| 586 | |
| 587 | i = 0; |
| 588 | |
| 589 | for_each_possible_cpu(cpu) { |
| 590 | struct bpf_lru_node *node; |
| 591 | |
| 592 | l = per_cpu_ptr(lru->percpu_lru, cpu); |
| 593 | again: |
| 594 | node = (struct bpf_lru_node *)(buf + node_offset); |
| 595 | node->cpu = cpu; |
| 596 | node->type = BPF_LRU_LIST_T_FREE; |
| 597 | bpf_lru_node_clear_ref(node); |
| 598 | list_add(new: &node->list, head: &l->lists[BPF_LRU_LIST_T_FREE]); |
| 599 | i++; |
| 600 | buf += elem_size; |
| 601 | if (i == nr_elems) |
| 602 | break; |
| 603 | if (i % pcpu_entries) |
| 604 | goto again; |
| 605 | } |
| 606 | } |
| 607 | |
| 608 | void bpf_lru_populate(struct bpf_lru *lru, void *buf, u32 node_offset, |
| 609 | u32 elem_size, u32 nr_elems) |
| 610 | { |
| 611 | if (lru->percpu) |
| 612 | bpf_percpu_lru_populate(lru, buf, node_offset, elem_size, |
| 613 | nr_elems); |
| 614 | else |
| 615 | bpf_common_lru_populate(lru, buf, node_offset, elem_size, |
| 616 | nr_elems); |
| 617 | } |
| 618 | |
| 619 | static void bpf_lru_locallist_init(struct bpf_lru_locallist *loc_l, int cpu) |
| 620 | { |
| 621 | int i; |
| 622 | |
| 623 | for (i = 0; i < NR_BPF_LRU_LOCAL_LIST_T; i++) |
| 624 | INIT_LIST_HEAD(list: &loc_l->lists[i]); |
| 625 | |
| 626 | loc_l->next_steal = cpu; |
| 627 | |
| 628 | raw_spin_lock_init(&loc_l->lock); |
| 629 | } |
| 630 | |
| 631 | static void bpf_lru_list_init(struct bpf_lru_list *l) |
| 632 | { |
| 633 | int i; |
| 634 | |
| 635 | for (i = 0; i < NR_BPF_LRU_LIST_T; i++) |
| 636 | INIT_LIST_HEAD(list: &l->lists[i]); |
| 637 | |
| 638 | for (i = 0; i < NR_BPF_LRU_LIST_COUNT; i++) |
| 639 | l->counts[i] = 0; |
| 640 | |
| 641 | l->next_inactive_rotation = &l->lists[BPF_LRU_LIST_T_INACTIVE]; |
| 642 | |
| 643 | raw_spin_lock_init(&l->lock); |
| 644 | } |
| 645 | |
| 646 | int bpf_lru_init(struct bpf_lru *lru, bool percpu, u32 hash_offset, |
| 647 | del_from_htab_func del_from_htab, void *del_arg) |
| 648 | { |
| 649 | int cpu; |
| 650 | |
| 651 | if (percpu) { |
| 652 | lru->percpu_lru = alloc_percpu(struct bpf_lru_list); |
| 653 | if (!lru->percpu_lru) |
| 654 | return -ENOMEM; |
| 655 | |
| 656 | for_each_possible_cpu(cpu) { |
| 657 | struct bpf_lru_list *l; |
| 658 | |
| 659 | l = per_cpu_ptr(lru->percpu_lru, cpu); |
| 660 | bpf_lru_list_init(l); |
| 661 | } |
| 662 | lru->nr_scans = PERCPU_NR_SCANS; |
| 663 | } else { |
| 664 | struct bpf_common_lru *clru = &lru->common_lru; |
| 665 | |
| 666 | clru->local_list = alloc_percpu(struct bpf_lru_locallist); |
| 667 | if (!clru->local_list) |
| 668 | return -ENOMEM; |
| 669 | |
| 670 | for_each_possible_cpu(cpu) { |
| 671 | struct bpf_lru_locallist *loc_l; |
| 672 | |
| 673 | loc_l = per_cpu_ptr(clru->local_list, cpu); |
| 674 | bpf_lru_locallist_init(loc_l, cpu); |
| 675 | } |
| 676 | |
| 677 | bpf_lru_list_init(l: &clru->lru_list); |
| 678 | lru->nr_scans = LOCAL_NR_SCANS; |
| 679 | } |
| 680 | |
| 681 | lru->percpu = percpu; |
| 682 | lru->del_from_htab = del_from_htab; |
| 683 | lru->del_arg = del_arg; |
| 684 | lru->hash_offset = hash_offset; |
| 685 | |
| 686 | return 0; |
| 687 | } |
| 688 | |
| 689 | void bpf_lru_destroy(struct bpf_lru *lru) |
| 690 | { |
| 691 | if (lru->percpu) |
| 692 | free_percpu(pdata: lru->percpu_lru); |
| 693 | else |
| 694 | free_percpu(pdata: lru->common_lru.local_list); |
| 695 | } |
| 696 | |