memcontrol.c revision c8dad2bb6307f5b00f804a686917105206a4d5c9
1/* memcontrol.c - Memory Controller 2 * 3 * Copyright IBM Corporation, 2007 4 * Author Balbir Singh <balbir@linux.vnet.ibm.com> 5 * 6 * Copyright 2007 OpenVZ SWsoft Inc 7 * Author: Pavel Emelianov <xemul@openvz.org> 8 * 9 * This program is free software; you can redistribute it and/or modify 10 * it under the terms of the GNU General Public License as published by 11 * the Free Software Foundation; either version 2 of the License, or 12 * (at your option) any later version. 13 * 14 * This program is distributed in the hope that it will be useful, 15 * but WITHOUT ANY WARRANTY; without even the implied warranty of 16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 17 * GNU General Public License for more details. 18 */ 19 20#include <linux/res_counter.h> 21#include <linux/memcontrol.h> 22#include <linux/cgroup.h> 23#include <linux/mm.h> 24#include <linux/smp.h> 25#include <linux/page-flags.h> 26#include <linux/backing-dev.h> 27#include <linux/bit_spinlock.h> 28#include <linux/rcupdate.h> 29#include <linux/slab.h> 30#include <linux/swap.h> 31#include <linux/spinlock.h> 32#include <linux/fs.h> 33#include <linux/seq_file.h> 34#include <linux/vmalloc.h> 35#include <linux/mm_inline.h> 36#include <linux/page_cgroup.h> 37 38#include <asm/uaccess.h> 39 40struct cgroup_subsys mem_cgroup_subsys __read_mostly; 41#define MEM_CGROUP_RECLAIM_RETRIES 5 42 43/* 44 * Statistics for memory cgroup. 45 */ 46enum mem_cgroup_stat_index { 47 /* 48 * For MEM_CONTAINER_TYPE_ALL, usage = pagecache + rss. 49 */ 50 MEM_CGROUP_STAT_CACHE, /* # of pages charged as cache */ 51 MEM_CGROUP_STAT_RSS, /* # of pages charged as rss */ 52 MEM_CGROUP_STAT_PGPGIN_COUNT, /* # of pages paged in */ 53 MEM_CGROUP_STAT_PGPGOUT_COUNT, /* # of pages paged out */ 54 55 MEM_CGROUP_STAT_NSTATS, 56}; 57 58struct mem_cgroup_stat_cpu { 59 s64 count[MEM_CGROUP_STAT_NSTATS]; 60} ____cacheline_aligned_in_smp; 61 62struct mem_cgroup_stat { 63 struct mem_cgroup_stat_cpu cpustat[0]; 64}; 65 66/* 67 * For accounting under irq disable, no need for increment preempt count. 68 */ 69static inline void __mem_cgroup_stat_add_safe(struct mem_cgroup_stat_cpu *stat, 70 enum mem_cgroup_stat_index idx, int val) 71{ 72 stat->count[idx] += val; 73} 74 75static s64 mem_cgroup_read_stat(struct mem_cgroup_stat *stat, 76 enum mem_cgroup_stat_index idx) 77{ 78 int cpu; 79 s64 ret = 0; 80 for_each_possible_cpu(cpu) 81 ret += stat->cpustat[cpu].count[idx]; 82 return ret; 83} 84 85/* 86 * per-zone information in memory controller. 87 */ 88struct mem_cgroup_per_zone { 89 /* 90 * spin_lock to protect the per cgroup LRU 91 */ 92 spinlock_t lru_lock; 93 struct list_head lists[NR_LRU_LISTS]; 94 unsigned long count[NR_LRU_LISTS]; 95}; 96/* Macro for accessing counter */ 97#define MEM_CGROUP_ZSTAT(mz, idx) ((mz)->count[(idx)]) 98 99struct mem_cgroup_per_node { 100 struct mem_cgroup_per_zone zoneinfo[MAX_NR_ZONES]; 101}; 102 103struct mem_cgroup_lru_info { 104 struct mem_cgroup_per_node *nodeinfo[MAX_NUMNODES]; 105}; 106 107/* 108 * The memory controller data structure. The memory controller controls both 109 * page cache and RSS per cgroup. We would eventually like to provide 110 * statistics based on the statistics developed by Rik Van Riel for clock-pro, 111 * to help the administrator determine what knobs to tune. 112 * 113 * TODO: Add a water mark for the memory controller. Reclaim will begin when 114 * we hit the water mark. May be even add a low water mark, such that 115 * no reclaim occurs from a cgroup at it's low water mark, this is 116 * a feature that will be implemented much later in the future. 117 */ 118struct mem_cgroup { 119 struct cgroup_subsys_state css; 120 /* 121 * the counter to account for memory usage 122 */ 123 struct res_counter res; 124 /* 125 * Per cgroup active and inactive list, similar to the 126 * per zone LRU lists. 127 */ 128 struct mem_cgroup_lru_info info; 129 130 int prev_priority; /* for recording reclaim priority */ 131 /* 132 * statistics. This must be placed at the end of memcg. 133 */ 134 struct mem_cgroup_stat stat; 135}; 136 137enum charge_type { 138 MEM_CGROUP_CHARGE_TYPE_CACHE = 0, 139 MEM_CGROUP_CHARGE_TYPE_MAPPED, 140 MEM_CGROUP_CHARGE_TYPE_SHMEM, /* used by page migration of shmem */ 141 MEM_CGROUP_CHARGE_TYPE_FORCE, /* used by force_empty */ 142 NR_CHARGE_TYPE, 143}; 144 145/* only for here (for easy reading.) */ 146#define PCGF_CACHE (1UL << PCG_CACHE) 147#define PCGF_USED (1UL << PCG_USED) 148#define PCGF_ACTIVE (1UL << PCG_ACTIVE) 149#define PCGF_LOCK (1UL << PCG_LOCK) 150#define PCGF_FILE (1UL << PCG_FILE) 151static const unsigned long 152pcg_default_flags[NR_CHARGE_TYPE] = { 153 PCGF_CACHE | PCGF_FILE | PCGF_USED | PCGF_LOCK, /* File Cache */ 154 PCGF_ACTIVE | PCGF_USED | PCGF_LOCK, /* Anon */ 155 PCGF_ACTIVE | PCGF_CACHE | PCGF_USED | PCGF_LOCK, /* Shmem */ 156 0, /* FORCE */ 157}; 158 159/* 160 * Always modified under lru lock. Then, not necessary to preempt_disable() 161 */ 162static void mem_cgroup_charge_statistics(struct mem_cgroup *mem, 163 struct page_cgroup *pc, 164 bool charge) 165{ 166 int val = (charge)? 1 : -1; 167 struct mem_cgroup_stat *stat = &mem->stat; 168 struct mem_cgroup_stat_cpu *cpustat; 169 170 VM_BUG_ON(!irqs_disabled()); 171 172 cpustat = &stat->cpustat[smp_processor_id()]; 173 if (PageCgroupCache(pc)) 174 __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_CACHE, val); 175 else 176 __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_RSS, val); 177 178 if (charge) 179 __mem_cgroup_stat_add_safe(cpustat, 180 MEM_CGROUP_STAT_PGPGIN_COUNT, 1); 181 else 182 __mem_cgroup_stat_add_safe(cpustat, 183 MEM_CGROUP_STAT_PGPGOUT_COUNT, 1); 184} 185 186static struct mem_cgroup_per_zone * 187mem_cgroup_zoneinfo(struct mem_cgroup *mem, int nid, int zid) 188{ 189 return &mem->info.nodeinfo[nid]->zoneinfo[zid]; 190} 191 192static struct mem_cgroup_per_zone * 193page_cgroup_zoneinfo(struct page_cgroup *pc) 194{ 195 struct mem_cgroup *mem = pc->mem_cgroup; 196 int nid = page_cgroup_nid(pc); 197 int zid = page_cgroup_zid(pc); 198 199 return mem_cgroup_zoneinfo(mem, nid, zid); 200} 201 202static unsigned long mem_cgroup_get_all_zonestat(struct mem_cgroup *mem, 203 enum lru_list idx) 204{ 205 int nid, zid; 206 struct mem_cgroup_per_zone *mz; 207 u64 total = 0; 208 209 for_each_online_node(nid) 210 for (zid = 0; zid < MAX_NR_ZONES; zid++) { 211 mz = mem_cgroup_zoneinfo(mem, nid, zid); 212 total += MEM_CGROUP_ZSTAT(mz, idx); 213 } 214 return total; 215} 216 217static struct mem_cgroup *mem_cgroup_from_cont(struct cgroup *cont) 218{ 219 return container_of(cgroup_subsys_state(cont, 220 mem_cgroup_subsys_id), struct mem_cgroup, 221 css); 222} 223 224struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) 225{ 226 /* 227 * mm_update_next_owner() may clear mm->owner to NULL 228 * if it races with swapoff, page migration, etc. 229 * So this can be called with p == NULL. 230 */ 231 if (unlikely(!p)) 232 return NULL; 233 234 return container_of(task_subsys_state(p, mem_cgroup_subsys_id), 235 struct mem_cgroup, css); 236} 237 238static void __mem_cgroup_remove_list(struct mem_cgroup_per_zone *mz, 239 struct page_cgroup *pc) 240{ 241 int lru = LRU_BASE; 242 243 if (PageCgroupUnevictable(pc)) 244 lru = LRU_UNEVICTABLE; 245 else { 246 if (PageCgroupActive(pc)) 247 lru += LRU_ACTIVE; 248 if (PageCgroupFile(pc)) 249 lru += LRU_FILE; 250 } 251 252 MEM_CGROUP_ZSTAT(mz, lru) -= 1; 253 254 mem_cgroup_charge_statistics(pc->mem_cgroup, pc, false); 255 list_del(&pc->lru); 256} 257 258static void __mem_cgroup_add_list(struct mem_cgroup_per_zone *mz, 259 struct page_cgroup *pc, bool hot) 260{ 261 int lru = LRU_BASE; 262 263 if (PageCgroupUnevictable(pc)) 264 lru = LRU_UNEVICTABLE; 265 else { 266 if (PageCgroupActive(pc)) 267 lru += LRU_ACTIVE; 268 if (PageCgroupFile(pc)) 269 lru += LRU_FILE; 270 } 271 272 MEM_CGROUP_ZSTAT(mz, lru) += 1; 273 if (hot) 274 list_add(&pc->lru, &mz->lists[lru]); 275 else 276 list_add_tail(&pc->lru, &mz->lists[lru]); 277 278 mem_cgroup_charge_statistics(pc->mem_cgroup, pc, true); 279} 280 281static void __mem_cgroup_move_lists(struct page_cgroup *pc, enum lru_list lru) 282{ 283 struct mem_cgroup_per_zone *mz = page_cgroup_zoneinfo(pc); 284 int active = PageCgroupActive(pc); 285 int file = PageCgroupFile(pc); 286 int unevictable = PageCgroupUnevictable(pc); 287 enum lru_list from = unevictable ? LRU_UNEVICTABLE : 288 (LRU_FILE * !!file + !!active); 289 290 if (lru == from) 291 return; 292 293 MEM_CGROUP_ZSTAT(mz, from) -= 1; 294 /* 295 * However this is done under mz->lru_lock, another flags, which 296 * are not related to LRU, will be modified from out-of-lock. 297 * We have to use atomic set/clear flags. 298 */ 299 if (is_unevictable_lru(lru)) { 300 ClearPageCgroupActive(pc); 301 SetPageCgroupUnevictable(pc); 302 } else { 303 if (is_active_lru(lru)) 304 SetPageCgroupActive(pc); 305 else 306 ClearPageCgroupActive(pc); 307 ClearPageCgroupUnevictable(pc); 308 } 309 310 MEM_CGROUP_ZSTAT(mz, lru) += 1; 311 list_move(&pc->lru, &mz->lists[lru]); 312} 313 314int task_in_mem_cgroup(struct task_struct *task, const struct mem_cgroup *mem) 315{ 316 int ret; 317 318 task_lock(task); 319 ret = task->mm && mm_match_cgroup(task->mm, mem); 320 task_unlock(task); 321 return ret; 322} 323 324/* 325 * This routine assumes that the appropriate zone's lru lock is already held 326 */ 327void mem_cgroup_move_lists(struct page *page, enum lru_list lru) 328{ 329 struct page_cgroup *pc; 330 struct mem_cgroup_per_zone *mz; 331 unsigned long flags; 332 333 if (mem_cgroup_subsys.disabled) 334 return; 335 336 /* 337 * We cannot lock_page_cgroup while holding zone's lru_lock, 338 * because other holders of lock_page_cgroup can be interrupted 339 * with an attempt to rotate_reclaimable_page. But we cannot 340 * safely get to page_cgroup without it, so just try_lock it: 341 * mem_cgroup_isolate_pages allows for page left on wrong list. 342 */ 343 pc = lookup_page_cgroup(page); 344 if (!trylock_page_cgroup(pc)) 345 return; 346 if (pc && PageCgroupUsed(pc)) { 347 mz = page_cgroup_zoneinfo(pc); 348 spin_lock_irqsave(&mz->lru_lock, flags); 349 __mem_cgroup_move_lists(pc, lru); 350 spin_unlock_irqrestore(&mz->lru_lock, flags); 351 } 352 unlock_page_cgroup(pc); 353} 354 355/* 356 * Calculate mapped_ratio under memory controller. This will be used in 357 * vmscan.c for deteremining we have to reclaim mapped pages. 358 */ 359int mem_cgroup_calc_mapped_ratio(struct mem_cgroup *mem) 360{ 361 long total, rss; 362 363 /* 364 * usage is recorded in bytes. But, here, we assume the number of 365 * physical pages can be represented by "long" on any arch. 366 */ 367 total = (long) (mem->res.usage >> PAGE_SHIFT) + 1L; 368 rss = (long)mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_RSS); 369 return (int)((rss * 100L) / total); 370} 371 372/* 373 * prev_priority control...this will be used in memory reclaim path. 374 */ 375int mem_cgroup_get_reclaim_priority(struct mem_cgroup *mem) 376{ 377 return mem->prev_priority; 378} 379 380void mem_cgroup_note_reclaim_priority(struct mem_cgroup *mem, int priority) 381{ 382 if (priority < mem->prev_priority) 383 mem->prev_priority = priority; 384} 385 386void mem_cgroup_record_reclaim_priority(struct mem_cgroup *mem, int priority) 387{ 388 mem->prev_priority = priority; 389} 390 391/* 392 * Calculate # of pages to be scanned in this priority/zone. 393 * See also vmscan.c 394 * 395 * priority starts from "DEF_PRIORITY" and decremented in each loop. 396 * (see include/linux/mmzone.h) 397 */ 398 399long mem_cgroup_calc_reclaim(struct mem_cgroup *mem, struct zone *zone, 400 int priority, enum lru_list lru) 401{ 402 long nr_pages; 403 int nid = zone->zone_pgdat->node_id; 404 int zid = zone_idx(zone); 405 struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(mem, nid, zid); 406 407 nr_pages = MEM_CGROUP_ZSTAT(mz, lru); 408 409 return (nr_pages >> priority); 410} 411 412unsigned long mem_cgroup_isolate_pages(unsigned long nr_to_scan, 413 struct list_head *dst, 414 unsigned long *scanned, int order, 415 int mode, struct zone *z, 416 struct mem_cgroup *mem_cont, 417 int active, int file) 418{ 419 unsigned long nr_taken = 0; 420 struct page *page; 421 unsigned long scan; 422 LIST_HEAD(pc_list); 423 struct list_head *src; 424 struct page_cgroup *pc, *tmp; 425 int nid = z->zone_pgdat->node_id; 426 int zid = zone_idx(z); 427 struct mem_cgroup_per_zone *mz; 428 int lru = LRU_FILE * !!file + !!active; 429 430 BUG_ON(!mem_cont); 431 mz = mem_cgroup_zoneinfo(mem_cont, nid, zid); 432 src = &mz->lists[lru]; 433 434 spin_lock(&mz->lru_lock); 435 scan = 0; 436 list_for_each_entry_safe_reverse(pc, tmp, src, lru) { 437 if (scan >= nr_to_scan) 438 break; 439 if (unlikely(!PageCgroupUsed(pc))) 440 continue; 441 page = pc->page; 442 443 if (unlikely(!PageLRU(page))) 444 continue; 445 446 /* 447 * TODO: play better with lumpy reclaim, grabbing anything. 448 */ 449 if (PageUnevictable(page) || 450 (PageActive(page) && !active) || 451 (!PageActive(page) && active)) { 452 __mem_cgroup_move_lists(pc, page_lru(page)); 453 continue; 454 } 455 456 scan++; 457 list_move(&pc->lru, &pc_list); 458 459 if (__isolate_lru_page(page, mode, file) == 0) { 460 list_move(&page->lru, dst); 461 nr_taken++; 462 } 463 } 464 465 list_splice(&pc_list, src); 466 spin_unlock(&mz->lru_lock); 467 468 *scanned = scan; 469 return nr_taken; 470} 471 472/* 473 * Unlike exported interface, "oom" parameter is added. if oom==true, 474 * oom-killer can be invoked. 475 */ 476static int __mem_cgroup_try_charge(struct mm_struct *mm, 477 gfp_t gfp_mask, struct mem_cgroup **memcg, bool oom) 478{ 479 struct mem_cgroup *mem; 480 int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; 481 /* 482 * We always charge the cgroup the mm_struct belongs to. 483 * The mm_struct's mem_cgroup changes on task migration if the 484 * thread group leader migrates. It's possible that mm is not 485 * set, if so charge the init_mm (happens for pagecache usage). 486 */ 487 if (likely(!*memcg)) { 488 rcu_read_lock(); 489 mem = mem_cgroup_from_task(rcu_dereference(mm->owner)); 490 if (unlikely(!mem)) { 491 rcu_read_unlock(); 492 return 0; 493 } 494 /* 495 * For every charge from the cgroup, increment reference count 496 */ 497 css_get(&mem->css); 498 *memcg = mem; 499 rcu_read_unlock(); 500 } else { 501 mem = *memcg; 502 css_get(&mem->css); 503 } 504 505 506 while (unlikely(res_counter_charge(&mem->res, PAGE_SIZE))) { 507 if (!(gfp_mask & __GFP_WAIT)) 508 goto nomem; 509 510 if (try_to_free_mem_cgroup_pages(mem, gfp_mask)) 511 continue; 512 513 /* 514 * try_to_free_mem_cgroup_pages() might not give us a full 515 * picture of reclaim. Some pages are reclaimed and might be 516 * moved to swap cache or just unmapped from the cgroup. 517 * Check the limit again to see if the reclaim reduced the 518 * current usage of the cgroup before giving up 519 */ 520 if (res_counter_check_under_limit(&mem->res)) 521 continue; 522 523 if (!nr_retries--) { 524 if (oom) 525 mem_cgroup_out_of_memory(mem, gfp_mask); 526 goto nomem; 527 } 528 } 529 return 0; 530nomem: 531 css_put(&mem->css); 532 return -ENOMEM; 533} 534 535/** 536 * mem_cgroup_try_charge - get charge of PAGE_SIZE. 537 * @mm: an mm_struct which is charged against. (when *memcg is NULL) 538 * @gfp_mask: gfp_mask for reclaim. 539 * @memcg: a pointer to memory cgroup which is charged against. 540 * 541 * charge against memory cgroup pointed by *memcg. if *memcg == NULL, estimated 542 * memory cgroup from @mm is got and stored in *memcg. 543 * 544 * Returns 0 if success. -ENOMEM at failure. 545 * This call can invoke OOM-Killer. 546 */ 547 548int mem_cgroup_try_charge(struct mm_struct *mm, 549 gfp_t mask, struct mem_cgroup **memcg) 550{ 551 return __mem_cgroup_try_charge(mm, mask, memcg, true); 552} 553 554/* 555 * commit a charge got by mem_cgroup_try_charge() and makes page_cgroup to be 556 * USED state. If already USED, uncharge and return. 557 */ 558 559static void __mem_cgroup_commit_charge(struct mem_cgroup *mem, 560 struct page_cgroup *pc, 561 enum charge_type ctype) 562{ 563 struct mem_cgroup_per_zone *mz; 564 unsigned long flags; 565 566 /* try_charge() can return NULL to *memcg, taking care of it. */ 567 if (!mem) 568 return; 569 570 lock_page_cgroup(pc); 571 if (unlikely(PageCgroupUsed(pc))) { 572 unlock_page_cgroup(pc); 573 res_counter_uncharge(&mem->res, PAGE_SIZE); 574 css_put(&mem->css); 575 return; 576 } 577 pc->mem_cgroup = mem; 578 /* 579 * If a page is accounted as a page cache, insert to inactive list. 580 * If anon, insert to active list. 581 */ 582 pc->flags = pcg_default_flags[ctype]; 583 584 mz = page_cgroup_zoneinfo(pc); 585 586 spin_lock_irqsave(&mz->lru_lock, flags); 587 __mem_cgroup_add_list(mz, pc, true); 588 spin_unlock_irqrestore(&mz->lru_lock, flags); 589 unlock_page_cgroup(pc); 590} 591 592/** 593 * mem_cgroup_move_account - move account of the page 594 * @pc: page_cgroup of the page. 595 * @from: mem_cgroup which the page is moved from. 596 * @to: mem_cgroup which the page is moved to. @from != @to. 597 * 598 * The caller must confirm following. 599 * 1. disable irq. 600 * 2. lru_lock of old mem_cgroup(@from) should be held. 601 * 602 * returns 0 at success, 603 * returns -EBUSY when lock is busy or "pc" is unstable. 604 * 605 * This function does "uncharge" from old cgroup but doesn't do "charge" to 606 * new cgroup. It should be done by a caller. 607 */ 608 609static int mem_cgroup_move_account(struct page_cgroup *pc, 610 struct mem_cgroup *from, struct mem_cgroup *to) 611{ 612 struct mem_cgroup_per_zone *from_mz, *to_mz; 613 int nid, zid; 614 int ret = -EBUSY; 615 616 VM_BUG_ON(!irqs_disabled()); 617 VM_BUG_ON(from == to); 618 619 nid = page_cgroup_nid(pc); 620 zid = page_cgroup_zid(pc); 621 from_mz = mem_cgroup_zoneinfo(from, nid, zid); 622 to_mz = mem_cgroup_zoneinfo(to, nid, zid); 623 624 625 if (!trylock_page_cgroup(pc)) 626 return ret; 627 628 if (!PageCgroupUsed(pc)) 629 goto out; 630 631 if (pc->mem_cgroup != from) 632 goto out; 633 634 if (spin_trylock(&to_mz->lru_lock)) { 635 __mem_cgroup_remove_list(from_mz, pc); 636 css_put(&from->css); 637 res_counter_uncharge(&from->res, PAGE_SIZE); 638 pc->mem_cgroup = to; 639 css_get(&to->css); 640 __mem_cgroup_add_list(to_mz, pc, false); 641 ret = 0; 642 spin_unlock(&to_mz->lru_lock); 643 } 644out: 645 unlock_page_cgroup(pc); 646 return ret; 647} 648 649/* 650 * move charges to its parent. 651 */ 652 653static int mem_cgroup_move_parent(struct page_cgroup *pc, 654 struct mem_cgroup *child, 655 gfp_t gfp_mask) 656{ 657 struct cgroup *cg = child->css.cgroup; 658 struct cgroup *pcg = cg->parent; 659 struct mem_cgroup *parent; 660 struct mem_cgroup_per_zone *mz; 661 unsigned long flags; 662 int ret; 663 664 /* Is ROOT ? */ 665 if (!pcg) 666 return -EINVAL; 667 668 parent = mem_cgroup_from_cont(pcg); 669 670 ret = __mem_cgroup_try_charge(NULL, gfp_mask, &parent, false); 671 if (ret) 672 return ret; 673 674 mz = mem_cgroup_zoneinfo(child, 675 page_cgroup_nid(pc), page_cgroup_zid(pc)); 676 677 spin_lock_irqsave(&mz->lru_lock, flags); 678 ret = mem_cgroup_move_account(pc, child, parent); 679 spin_unlock_irqrestore(&mz->lru_lock, flags); 680 681 /* drop extra refcnt */ 682 css_put(&parent->css); 683 /* uncharge if move fails */ 684 if (ret) 685 res_counter_uncharge(&parent->res, PAGE_SIZE); 686 687 return ret; 688} 689 690/* 691 * Charge the memory controller for page usage. 692 * Return 693 * 0 if the charge was successful 694 * < 0 if the cgroup is over its limit 695 */ 696static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm, 697 gfp_t gfp_mask, enum charge_type ctype, 698 struct mem_cgroup *memcg) 699{ 700 struct mem_cgroup *mem; 701 struct page_cgroup *pc; 702 int ret; 703 704 pc = lookup_page_cgroup(page); 705 /* can happen at boot */ 706 if (unlikely(!pc)) 707 return 0; 708 prefetchw(pc); 709 710 mem = memcg; 711 ret = __mem_cgroup_try_charge(mm, gfp_mask, &mem, true); 712 if (ret) 713 return ret; 714 715 __mem_cgroup_commit_charge(mem, pc, ctype); 716 return 0; 717} 718 719int mem_cgroup_newpage_charge(struct page *page, 720 struct mm_struct *mm, gfp_t gfp_mask) 721{ 722 if (mem_cgroup_subsys.disabled) 723 return 0; 724 if (PageCompound(page)) 725 return 0; 726 /* 727 * If already mapped, we don't have to account. 728 * If page cache, page->mapping has address_space. 729 * But page->mapping may have out-of-use anon_vma pointer, 730 * detecit it by PageAnon() check. newly-mapped-anon's page->mapping 731 * is NULL. 732 */ 733 if (page_mapped(page) || (page->mapping && !PageAnon(page))) 734 return 0; 735 if (unlikely(!mm)) 736 mm = &init_mm; 737 return mem_cgroup_charge_common(page, mm, gfp_mask, 738 MEM_CGROUP_CHARGE_TYPE_MAPPED, NULL); 739} 740 741int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm, 742 gfp_t gfp_mask) 743{ 744 if (mem_cgroup_subsys.disabled) 745 return 0; 746 if (PageCompound(page)) 747 return 0; 748 /* 749 * Corner case handling. This is called from add_to_page_cache() 750 * in usual. But some FS (shmem) precharges this page before calling it 751 * and call add_to_page_cache() with GFP_NOWAIT. 752 * 753 * For GFP_NOWAIT case, the page may be pre-charged before calling 754 * add_to_page_cache(). (See shmem.c) check it here and avoid to call 755 * charge twice. (It works but has to pay a bit larger cost.) 756 */ 757 if (!(gfp_mask & __GFP_WAIT)) { 758 struct page_cgroup *pc; 759 760 761 pc = lookup_page_cgroup(page); 762 if (!pc) 763 return 0; 764 lock_page_cgroup(pc); 765 if (PageCgroupUsed(pc)) { 766 unlock_page_cgroup(pc); 767 return 0; 768 } 769 unlock_page_cgroup(pc); 770 } 771 772 if (unlikely(!mm)) 773 mm = &init_mm; 774 775 if (page_is_file_cache(page)) 776 return mem_cgroup_charge_common(page, mm, gfp_mask, 777 MEM_CGROUP_CHARGE_TYPE_CACHE, NULL); 778 else 779 return mem_cgroup_charge_common(page, mm, gfp_mask, 780 MEM_CGROUP_CHARGE_TYPE_SHMEM, NULL); 781} 782 783void mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr) 784{ 785 struct page_cgroup *pc; 786 787 if (mem_cgroup_subsys.disabled) 788 return; 789 if (!ptr) 790 return; 791 pc = lookup_page_cgroup(page); 792 __mem_cgroup_commit_charge(ptr, pc, MEM_CGROUP_CHARGE_TYPE_MAPPED); 793} 794 795void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *mem) 796{ 797 if (mem_cgroup_subsys.disabled) 798 return; 799 if (!mem) 800 return; 801 res_counter_uncharge(&mem->res, PAGE_SIZE); 802 css_put(&mem->css); 803} 804 805 806/* 807 * uncharge if !page_mapped(page) 808 */ 809static void 810__mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype) 811{ 812 struct page_cgroup *pc; 813 struct mem_cgroup *mem; 814 struct mem_cgroup_per_zone *mz; 815 unsigned long flags; 816 817 if (mem_cgroup_subsys.disabled) 818 return; 819 820 /* 821 * Check if our page_cgroup is valid 822 */ 823 pc = lookup_page_cgroup(page); 824 if (unlikely(!pc || !PageCgroupUsed(pc))) 825 return; 826 827 lock_page_cgroup(pc); 828 if ((ctype == MEM_CGROUP_CHARGE_TYPE_MAPPED && page_mapped(page)) 829 || !PageCgroupUsed(pc)) { 830 /* This happens at race in zap_pte_range() and do_swap_page()*/ 831 unlock_page_cgroup(pc); 832 return; 833 } 834 ClearPageCgroupUsed(pc); 835 mem = pc->mem_cgroup; 836 837 mz = page_cgroup_zoneinfo(pc); 838 spin_lock_irqsave(&mz->lru_lock, flags); 839 __mem_cgroup_remove_list(mz, pc); 840 spin_unlock_irqrestore(&mz->lru_lock, flags); 841 unlock_page_cgroup(pc); 842 843 res_counter_uncharge(&mem->res, PAGE_SIZE); 844 css_put(&mem->css); 845 846 return; 847} 848 849void mem_cgroup_uncharge_page(struct page *page) 850{ 851 /* early check. */ 852 if (page_mapped(page)) 853 return; 854 if (page->mapping && !PageAnon(page)) 855 return; 856 __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_MAPPED); 857} 858 859void mem_cgroup_uncharge_cache_page(struct page *page) 860{ 861 VM_BUG_ON(page_mapped(page)); 862 VM_BUG_ON(page->mapping); 863 __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_CACHE); 864} 865 866/* 867 * Before starting migration, account PAGE_SIZE to mem_cgroup that the old 868 * page belongs to. 869 */ 870int mem_cgroup_prepare_migration(struct page *page, struct mem_cgroup **ptr) 871{ 872 struct page_cgroup *pc; 873 struct mem_cgroup *mem = NULL; 874 int ret = 0; 875 876 if (mem_cgroup_subsys.disabled) 877 return 0; 878 879 pc = lookup_page_cgroup(page); 880 lock_page_cgroup(pc); 881 if (PageCgroupUsed(pc)) { 882 mem = pc->mem_cgroup; 883 css_get(&mem->css); 884 } 885 unlock_page_cgroup(pc); 886 887 if (mem) { 888 ret = mem_cgroup_try_charge(NULL, GFP_HIGHUSER_MOVABLE, &mem); 889 css_put(&mem->css); 890 } 891 *ptr = mem; 892 return ret; 893} 894 895/* remove redundant charge if migration failed*/ 896void mem_cgroup_end_migration(struct mem_cgroup *mem, 897 struct page *oldpage, struct page *newpage) 898{ 899 struct page *target, *unused; 900 struct page_cgroup *pc; 901 enum charge_type ctype; 902 903 if (!mem) 904 return; 905 906 /* at migration success, oldpage->mapping is NULL. */ 907 if (oldpage->mapping) { 908 target = oldpage; 909 unused = NULL; 910 } else { 911 target = newpage; 912 unused = oldpage; 913 } 914 915 if (PageAnon(target)) 916 ctype = MEM_CGROUP_CHARGE_TYPE_MAPPED; 917 else if (page_is_file_cache(target)) 918 ctype = MEM_CGROUP_CHARGE_TYPE_CACHE; 919 else 920 ctype = MEM_CGROUP_CHARGE_TYPE_SHMEM; 921 922 /* unused page is not on radix-tree now. */ 923 if (unused && ctype != MEM_CGROUP_CHARGE_TYPE_MAPPED) 924 __mem_cgroup_uncharge_common(unused, ctype); 925 926 pc = lookup_page_cgroup(target); 927 /* 928 * __mem_cgroup_commit_charge() check PCG_USED bit of page_cgroup. 929 * So, double-counting is effectively avoided. 930 */ 931 __mem_cgroup_commit_charge(mem, pc, ctype); 932 933 /* 934 * Both of oldpage and newpage are still under lock_page(). 935 * Then, we don't have to care about race in radix-tree. 936 * But we have to be careful that this page is unmapped or not. 937 * 938 * There is a case for !page_mapped(). At the start of 939 * migration, oldpage was mapped. But now, it's zapped. 940 * But we know *target* page is not freed/reused under us. 941 * mem_cgroup_uncharge_page() does all necessary checks. 942 */ 943 if (ctype == MEM_CGROUP_CHARGE_TYPE_MAPPED) 944 mem_cgroup_uncharge_page(target); 945} 946 947/* 948 * A call to try to shrink memory usage under specified resource controller. 949 * This is typically used for page reclaiming for shmem for reducing side 950 * effect of page allocation from shmem, which is used by some mem_cgroup. 951 */ 952int mem_cgroup_shrink_usage(struct mm_struct *mm, gfp_t gfp_mask) 953{ 954 struct mem_cgroup *mem; 955 int progress = 0; 956 int retry = MEM_CGROUP_RECLAIM_RETRIES; 957 958 if (mem_cgroup_subsys.disabled) 959 return 0; 960 if (!mm) 961 return 0; 962 963 rcu_read_lock(); 964 mem = mem_cgroup_from_task(rcu_dereference(mm->owner)); 965 if (unlikely(!mem)) { 966 rcu_read_unlock(); 967 return 0; 968 } 969 css_get(&mem->css); 970 rcu_read_unlock(); 971 972 do { 973 progress = try_to_free_mem_cgroup_pages(mem, gfp_mask); 974 progress += res_counter_check_under_limit(&mem->res); 975 } while (!progress && --retry); 976 977 css_put(&mem->css); 978 if (!retry) 979 return -ENOMEM; 980 return 0; 981} 982 983static int mem_cgroup_resize_limit(struct mem_cgroup *memcg, 984 unsigned long long val) 985{ 986 987 int retry_count = MEM_CGROUP_RECLAIM_RETRIES; 988 int progress; 989 int ret = 0; 990 991 while (res_counter_set_limit(&memcg->res, val)) { 992 if (signal_pending(current)) { 993 ret = -EINTR; 994 break; 995 } 996 if (!retry_count) { 997 ret = -EBUSY; 998 break; 999 } 1000 progress = try_to_free_mem_cgroup_pages(memcg, 1001 GFP_HIGHUSER_MOVABLE); 1002 if (!progress) 1003 retry_count--; 1004 } 1005 return ret; 1006} 1007 1008 1009/* 1010 * This routine traverse page_cgroup in given list and drop them all. 1011 * *And* this routine doesn't reclaim page itself, just removes page_cgroup. 1012 */ 1013static int mem_cgroup_force_empty_list(struct mem_cgroup *mem, 1014 struct mem_cgroup_per_zone *mz, 1015 enum lru_list lru) 1016{ 1017 struct page_cgroup *pc, *busy; 1018 unsigned long flags; 1019 unsigned long loop; 1020 struct list_head *list; 1021 int ret = 0; 1022 1023 list = &mz->lists[lru]; 1024 1025 loop = MEM_CGROUP_ZSTAT(mz, lru); 1026 /* give some margin against EBUSY etc...*/ 1027 loop += 256; 1028 busy = NULL; 1029 while (loop--) { 1030 ret = 0; 1031 spin_lock_irqsave(&mz->lru_lock, flags); 1032 if (list_empty(list)) { 1033 spin_unlock_irqrestore(&mz->lru_lock, flags); 1034 break; 1035 } 1036 pc = list_entry(list->prev, struct page_cgroup, lru); 1037 if (busy == pc) { 1038 list_move(&pc->lru, list); 1039 busy = 0; 1040 spin_unlock_irqrestore(&mz->lru_lock, flags); 1041 continue; 1042 } 1043 spin_unlock_irqrestore(&mz->lru_lock, flags); 1044 1045 ret = mem_cgroup_move_parent(pc, mem, GFP_HIGHUSER_MOVABLE); 1046 if (ret == -ENOMEM) 1047 break; 1048 1049 if (ret == -EBUSY || ret == -EINVAL) { 1050 /* found lock contention or "pc" is obsolete. */ 1051 busy = pc; 1052 cond_resched(); 1053 } else 1054 busy = NULL; 1055 } 1056 if (!ret && !list_empty(list)) 1057 return -EBUSY; 1058 return ret; 1059} 1060 1061/* 1062 * make mem_cgroup's charge to be 0 if there is no task. 1063 * This enables deleting this mem_cgroup. 1064 */ 1065static int mem_cgroup_force_empty(struct mem_cgroup *mem) 1066{ 1067 int ret; 1068 int node, zid, shrink; 1069 int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; 1070 1071 css_get(&mem->css); 1072 1073 shrink = 0; 1074move_account: 1075 while (mem->res.usage > 0) { 1076 ret = -EBUSY; 1077 if (atomic_read(&mem->css.cgroup->count) > 0) 1078 goto out; 1079 1080 /* This is for making all *used* pages to be on LRU. */ 1081 lru_add_drain_all(); 1082 ret = 0; 1083 for_each_node_state(node, N_POSSIBLE) { 1084 for (zid = 0; !ret && zid < MAX_NR_ZONES; zid++) { 1085 struct mem_cgroup_per_zone *mz; 1086 enum lru_list l; 1087 mz = mem_cgroup_zoneinfo(mem, node, zid); 1088 for_each_lru(l) { 1089 ret = mem_cgroup_force_empty_list(mem, 1090 mz, l); 1091 if (ret) 1092 break; 1093 } 1094 } 1095 if (ret) 1096 break; 1097 } 1098 /* it seems parent cgroup doesn't have enough mem */ 1099 if (ret == -ENOMEM) 1100 goto try_to_free; 1101 cond_resched(); 1102 } 1103 ret = 0; 1104out: 1105 css_put(&mem->css); 1106 return ret; 1107 1108try_to_free: 1109 /* returns EBUSY if we come here twice. */ 1110 if (shrink) { 1111 ret = -EBUSY; 1112 goto out; 1113 } 1114 /* try to free all pages in this cgroup */ 1115 shrink = 1; 1116 while (nr_retries && mem->res.usage > 0) { 1117 int progress; 1118 progress = try_to_free_mem_cgroup_pages(mem, 1119 GFP_HIGHUSER_MOVABLE); 1120 if (!progress) 1121 nr_retries--; 1122 1123 } 1124 /* try move_account...there may be some *locked* pages. */ 1125 if (mem->res.usage) 1126 goto move_account; 1127 ret = 0; 1128 goto out; 1129} 1130 1131static u64 mem_cgroup_read(struct cgroup *cont, struct cftype *cft) 1132{ 1133 return res_counter_read_u64(&mem_cgroup_from_cont(cont)->res, 1134 cft->private); 1135} 1136/* 1137 * The user of this function is... 1138 * RES_LIMIT. 1139 */ 1140static int mem_cgroup_write(struct cgroup *cont, struct cftype *cft, 1141 const char *buffer) 1142{ 1143 struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); 1144 unsigned long long val; 1145 int ret; 1146 1147 switch (cft->private) { 1148 case RES_LIMIT: 1149 /* This function does all necessary parse...reuse it */ 1150 ret = res_counter_memparse_write_strategy(buffer, &val); 1151 if (!ret) 1152 ret = mem_cgroup_resize_limit(memcg, val); 1153 break; 1154 default: 1155 ret = -EINVAL; /* should be BUG() ? */ 1156 break; 1157 } 1158 return ret; 1159} 1160 1161static int mem_cgroup_reset(struct cgroup *cont, unsigned int event) 1162{ 1163 struct mem_cgroup *mem; 1164 1165 mem = mem_cgroup_from_cont(cont); 1166 switch (event) { 1167 case RES_MAX_USAGE: 1168 res_counter_reset_max(&mem->res); 1169 break; 1170 case RES_FAILCNT: 1171 res_counter_reset_failcnt(&mem->res); 1172 break; 1173 } 1174 return 0; 1175} 1176 1177static const struct mem_cgroup_stat_desc { 1178 const char *msg; 1179 u64 unit; 1180} mem_cgroup_stat_desc[] = { 1181 [MEM_CGROUP_STAT_CACHE] = { "cache", PAGE_SIZE, }, 1182 [MEM_CGROUP_STAT_RSS] = { "rss", PAGE_SIZE, }, 1183 [MEM_CGROUP_STAT_PGPGIN_COUNT] = {"pgpgin", 1, }, 1184 [MEM_CGROUP_STAT_PGPGOUT_COUNT] = {"pgpgout", 1, }, 1185}; 1186 1187static int mem_control_stat_show(struct cgroup *cont, struct cftype *cft, 1188 struct cgroup_map_cb *cb) 1189{ 1190 struct mem_cgroup *mem_cont = mem_cgroup_from_cont(cont); 1191 struct mem_cgroup_stat *stat = &mem_cont->stat; 1192 int i; 1193 1194 for (i = 0; i < ARRAY_SIZE(stat->cpustat[0].count); i++) { 1195 s64 val; 1196 1197 val = mem_cgroup_read_stat(stat, i); 1198 val *= mem_cgroup_stat_desc[i].unit; 1199 cb->fill(cb, mem_cgroup_stat_desc[i].msg, val); 1200 } 1201 /* showing # of active pages */ 1202 { 1203 unsigned long active_anon, inactive_anon; 1204 unsigned long active_file, inactive_file; 1205 unsigned long unevictable; 1206 1207 inactive_anon = mem_cgroup_get_all_zonestat(mem_cont, 1208 LRU_INACTIVE_ANON); 1209 active_anon = mem_cgroup_get_all_zonestat(mem_cont, 1210 LRU_ACTIVE_ANON); 1211 inactive_file = mem_cgroup_get_all_zonestat(mem_cont, 1212 LRU_INACTIVE_FILE); 1213 active_file = mem_cgroup_get_all_zonestat(mem_cont, 1214 LRU_ACTIVE_FILE); 1215 unevictable = mem_cgroup_get_all_zonestat(mem_cont, 1216 LRU_UNEVICTABLE); 1217 1218 cb->fill(cb, "active_anon", (active_anon) * PAGE_SIZE); 1219 cb->fill(cb, "inactive_anon", (inactive_anon) * PAGE_SIZE); 1220 cb->fill(cb, "active_file", (active_file) * PAGE_SIZE); 1221 cb->fill(cb, "inactive_file", (inactive_file) * PAGE_SIZE); 1222 cb->fill(cb, "unevictable", unevictable * PAGE_SIZE); 1223 1224 } 1225 return 0; 1226} 1227 1228static struct cftype mem_cgroup_files[] = { 1229 { 1230 .name = "usage_in_bytes", 1231 .private = RES_USAGE, 1232 .read_u64 = mem_cgroup_read, 1233 }, 1234 { 1235 .name = "max_usage_in_bytes", 1236 .private = RES_MAX_USAGE, 1237 .trigger = mem_cgroup_reset, 1238 .read_u64 = mem_cgroup_read, 1239 }, 1240 { 1241 .name = "limit_in_bytes", 1242 .private = RES_LIMIT, 1243 .write_string = mem_cgroup_write, 1244 .read_u64 = mem_cgroup_read, 1245 }, 1246 { 1247 .name = "failcnt", 1248 .private = RES_FAILCNT, 1249 .trigger = mem_cgroup_reset, 1250 .read_u64 = mem_cgroup_read, 1251 }, 1252 { 1253 .name = "stat", 1254 .read_map = mem_control_stat_show, 1255 }, 1256}; 1257 1258static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node) 1259{ 1260 struct mem_cgroup_per_node *pn; 1261 struct mem_cgroup_per_zone *mz; 1262 enum lru_list l; 1263 int zone, tmp = node; 1264 /* 1265 * This routine is called against possible nodes. 1266 * But it's BUG to call kmalloc() against offline node. 1267 * 1268 * TODO: this routine can waste much memory for nodes which will 1269 * never be onlined. It's better to use memory hotplug callback 1270 * function. 1271 */ 1272 if (!node_state(node, N_NORMAL_MEMORY)) 1273 tmp = -1; 1274 pn = kmalloc_node(sizeof(*pn), GFP_KERNEL, tmp); 1275 if (!pn) 1276 return 1; 1277 1278 mem->info.nodeinfo[node] = pn; 1279 memset(pn, 0, sizeof(*pn)); 1280 1281 for (zone = 0; zone < MAX_NR_ZONES; zone++) { 1282 mz = &pn->zoneinfo[zone]; 1283 spin_lock_init(&mz->lru_lock); 1284 for_each_lru(l) 1285 INIT_LIST_HEAD(&mz->lists[l]); 1286 } 1287 return 0; 1288} 1289 1290static void free_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node) 1291{ 1292 kfree(mem->info.nodeinfo[node]); 1293} 1294 1295static int mem_cgroup_size(void) 1296{ 1297 int cpustat_size = nr_cpu_ids * sizeof(struct mem_cgroup_stat_cpu); 1298 return sizeof(struct mem_cgroup) + cpustat_size; 1299} 1300 1301static struct mem_cgroup *mem_cgroup_alloc(void) 1302{ 1303 struct mem_cgroup *mem; 1304 int size = mem_cgroup_size(); 1305 1306 if (size < PAGE_SIZE) 1307 mem = kmalloc(size, GFP_KERNEL); 1308 else 1309 mem = vmalloc(size); 1310 1311 if (mem) 1312 memset(mem, 0, size); 1313 return mem; 1314} 1315 1316static void mem_cgroup_free(struct mem_cgroup *mem) 1317{ 1318 if (mem_cgroup_size() < PAGE_SIZE) 1319 kfree(mem); 1320 else 1321 vfree(mem); 1322} 1323 1324 1325static struct cgroup_subsys_state * 1326mem_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cont) 1327{ 1328 struct mem_cgroup *mem; 1329 int node; 1330 1331 mem = mem_cgroup_alloc(); 1332 if (!mem) 1333 return ERR_PTR(-ENOMEM); 1334 1335 res_counter_init(&mem->res); 1336 1337 for_each_node_state(node, N_POSSIBLE) 1338 if (alloc_mem_cgroup_per_zone_info(mem, node)) 1339 goto free_out; 1340 1341 return &mem->css; 1342free_out: 1343 for_each_node_state(node, N_POSSIBLE) 1344 free_mem_cgroup_per_zone_info(mem, node); 1345 mem_cgroup_free(mem); 1346 return ERR_PTR(-ENOMEM); 1347} 1348 1349static void mem_cgroup_pre_destroy(struct cgroup_subsys *ss, 1350 struct cgroup *cont) 1351{ 1352 struct mem_cgroup *mem = mem_cgroup_from_cont(cont); 1353 mem_cgroup_force_empty(mem); 1354} 1355 1356static void mem_cgroup_destroy(struct cgroup_subsys *ss, 1357 struct cgroup *cont) 1358{ 1359 int node; 1360 struct mem_cgroup *mem = mem_cgroup_from_cont(cont); 1361 1362 for_each_node_state(node, N_POSSIBLE) 1363 free_mem_cgroup_per_zone_info(mem, node); 1364 1365 mem_cgroup_free(mem_cgroup_from_cont(cont)); 1366} 1367 1368static int mem_cgroup_populate(struct cgroup_subsys *ss, 1369 struct cgroup *cont) 1370{ 1371 return cgroup_add_files(cont, ss, mem_cgroup_files, 1372 ARRAY_SIZE(mem_cgroup_files)); 1373} 1374 1375static void mem_cgroup_move_task(struct cgroup_subsys *ss, 1376 struct cgroup *cont, 1377 struct cgroup *old_cont, 1378 struct task_struct *p) 1379{ 1380 struct mm_struct *mm; 1381 struct mem_cgroup *mem, *old_mem; 1382 1383 mm = get_task_mm(p); 1384 if (mm == NULL) 1385 return; 1386 1387 mem = mem_cgroup_from_cont(cont); 1388 old_mem = mem_cgroup_from_cont(old_cont); 1389 1390 /* 1391 * Only thread group leaders are allowed to migrate, the mm_struct is 1392 * in effect owned by the leader 1393 */ 1394 if (!thread_group_leader(p)) 1395 goto out; 1396 1397out: 1398 mmput(mm); 1399} 1400 1401struct cgroup_subsys mem_cgroup_subsys = { 1402 .name = "memory", 1403 .subsys_id = mem_cgroup_subsys_id, 1404 .create = mem_cgroup_create, 1405 .pre_destroy = mem_cgroup_pre_destroy, 1406 .destroy = mem_cgroup_destroy, 1407 .populate = mem_cgroup_populate, 1408 .attach = mem_cgroup_move_task, 1409 .early_init = 0, 1410}; 1411