]>
Commit | Line | Data |
---|---|---|
8cdea7c0 BS |
1 | /* memcontrol.c - Memory Controller |
2 | * | |
3 | * Copyright IBM Corporation, 2007 | |
4 | * Author Balbir Singh <balbir@linux.vnet.ibm.com> | |
5 | * | |
78fb7466 PE |
6 | * Copyright 2007 OpenVZ SWsoft Inc |
7 | * Author: Pavel Emelianov <xemul@openvz.org> | |
8 | * | |
8cdea7c0 BS |
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> | |
78fb7466 | 23 | #include <linux/mm.h> |
8a9f3ccd | 24 | #include <linux/page-flags.h> |
66e1707b | 25 | #include <linux/backing-dev.h> |
8a9f3ccd BS |
26 | #include <linux/bit_spinlock.h> |
27 | #include <linux/rcupdate.h> | |
66e1707b BS |
28 | #include <linux/swap.h> |
29 | #include <linux/spinlock.h> | |
30 | #include <linux/fs.h> | |
8cdea7c0 | 31 | |
8697d331 BS |
32 | #include <asm/uaccess.h> |
33 | ||
8cdea7c0 | 34 | struct cgroup_subsys mem_cgroup_subsys; |
66e1707b | 35 | static const int MEM_CGROUP_RECLAIM_RETRIES = 5; |
8cdea7c0 BS |
36 | |
37 | /* | |
38 | * The memory controller data structure. The memory controller controls both | |
39 | * page cache and RSS per cgroup. We would eventually like to provide | |
40 | * statistics based on the statistics developed by Rik Van Riel for clock-pro, | |
41 | * to help the administrator determine what knobs to tune. | |
42 | * | |
43 | * TODO: Add a water mark for the memory controller. Reclaim will begin when | |
8a9f3ccd BS |
44 | * we hit the water mark. May be even add a low water mark, such that |
45 | * no reclaim occurs from a cgroup at it's low water mark, this is | |
46 | * a feature that will be implemented much later in the future. | |
8cdea7c0 BS |
47 | */ |
48 | struct mem_cgroup { | |
49 | struct cgroup_subsys_state css; | |
50 | /* | |
51 | * the counter to account for memory usage | |
52 | */ | |
53 | struct res_counter res; | |
78fb7466 PE |
54 | /* |
55 | * Per cgroup active and inactive list, similar to the | |
56 | * per zone LRU lists. | |
57 | * TODO: Consider making these lists per zone | |
58 | */ | |
59 | struct list_head active_list; | |
60 | struct list_head inactive_list; | |
66e1707b BS |
61 | /* |
62 | * spin_lock to protect the per cgroup LRU | |
63 | */ | |
64 | spinlock_t lru_lock; | |
8697d331 | 65 | unsigned long control_type; /* control RSS or RSS+Pagecache */ |
8cdea7c0 BS |
66 | }; |
67 | ||
8a9f3ccd BS |
68 | /* |
69 | * We use the lower bit of the page->page_cgroup pointer as a bit spin | |
70 | * lock. We need to ensure that page->page_cgroup is atleast two | |
71 | * byte aligned (based on comments from Nick Piggin) | |
72 | */ | |
73 | #define PAGE_CGROUP_LOCK_BIT 0x0 | |
74 | #define PAGE_CGROUP_LOCK (1 << PAGE_CGROUP_LOCK_BIT) | |
75 | ||
8cdea7c0 BS |
76 | /* |
77 | * A page_cgroup page is associated with every page descriptor. The | |
78 | * page_cgroup helps us identify information about the cgroup | |
79 | */ | |
80 | struct page_cgroup { | |
81 | struct list_head lru; /* per cgroup LRU list */ | |
82 | struct page *page; | |
83 | struct mem_cgroup *mem_cgroup; | |
8a9f3ccd BS |
84 | atomic_t ref_cnt; /* Helpful when pages move b/w */ |
85 | /* mapped and cached states */ | |
217bc319 | 86 | int flags; |
8cdea7c0 | 87 | }; |
217bc319 | 88 | #define PAGE_CGROUP_FLAG_CACHE (0x1) /* charged as cache */ |
8cdea7c0 | 89 | |
8697d331 BS |
90 | enum { |
91 | MEM_CGROUP_TYPE_UNSPEC = 0, | |
92 | MEM_CGROUP_TYPE_MAPPED, | |
93 | MEM_CGROUP_TYPE_CACHED, | |
94 | MEM_CGROUP_TYPE_ALL, | |
95 | MEM_CGROUP_TYPE_MAX, | |
96 | }; | |
97 | ||
217bc319 KH |
98 | enum charge_type { |
99 | MEM_CGROUP_CHARGE_TYPE_CACHE = 0, | |
100 | MEM_CGROUP_CHARGE_TYPE_MAPPED, | |
101 | }; | |
102 | ||
8697d331 | 103 | static struct mem_cgroup init_mem_cgroup; |
8cdea7c0 BS |
104 | |
105 | static inline | |
106 | struct mem_cgroup *mem_cgroup_from_cont(struct cgroup *cont) | |
107 | { | |
108 | return container_of(cgroup_subsys_state(cont, | |
109 | mem_cgroup_subsys_id), struct mem_cgroup, | |
110 | css); | |
111 | } | |
112 | ||
78fb7466 PE |
113 | static inline |
114 | struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) | |
115 | { | |
116 | return container_of(task_subsys_state(p, mem_cgroup_subsys_id), | |
117 | struct mem_cgroup, css); | |
118 | } | |
119 | ||
120 | void mm_init_cgroup(struct mm_struct *mm, struct task_struct *p) | |
121 | { | |
122 | struct mem_cgroup *mem; | |
123 | ||
124 | mem = mem_cgroup_from_task(p); | |
125 | css_get(&mem->css); | |
126 | mm->mem_cgroup = mem; | |
127 | } | |
128 | ||
129 | void mm_free_cgroup(struct mm_struct *mm) | |
130 | { | |
131 | css_put(&mm->mem_cgroup->css); | |
132 | } | |
133 | ||
8a9f3ccd BS |
134 | static inline int page_cgroup_locked(struct page *page) |
135 | { | |
136 | return bit_spin_is_locked(PAGE_CGROUP_LOCK_BIT, | |
137 | &page->page_cgroup); | |
138 | } | |
139 | ||
78fb7466 PE |
140 | void page_assign_page_cgroup(struct page *page, struct page_cgroup *pc) |
141 | { | |
8a9f3ccd BS |
142 | int locked; |
143 | ||
144 | /* | |
145 | * While resetting the page_cgroup we might not hold the | |
146 | * page_cgroup lock. free_hot_cold_page() is an example | |
147 | * of such a scenario | |
148 | */ | |
149 | if (pc) | |
150 | VM_BUG_ON(!page_cgroup_locked(page)); | |
151 | locked = (page->page_cgroup & PAGE_CGROUP_LOCK); | |
152 | page->page_cgroup = ((unsigned long)pc | locked); | |
78fb7466 PE |
153 | } |
154 | ||
155 | struct page_cgroup *page_get_page_cgroup(struct page *page) | |
156 | { | |
8a9f3ccd BS |
157 | return (struct page_cgroup *) |
158 | (page->page_cgroup & ~PAGE_CGROUP_LOCK); | |
159 | } | |
160 | ||
8697d331 | 161 | static void __always_inline lock_page_cgroup(struct page *page) |
8a9f3ccd BS |
162 | { |
163 | bit_spin_lock(PAGE_CGROUP_LOCK_BIT, &page->page_cgroup); | |
164 | VM_BUG_ON(!page_cgroup_locked(page)); | |
165 | } | |
166 | ||
8697d331 | 167 | static void __always_inline unlock_page_cgroup(struct page *page) |
8a9f3ccd BS |
168 | { |
169 | bit_spin_unlock(PAGE_CGROUP_LOCK_BIT, &page->page_cgroup); | |
170 | } | |
171 | ||
9175e031 KH |
172 | /* |
173 | * Tie new page_cgroup to struct page under lock_page_cgroup() | |
174 | * This can fail if the page has been tied to a page_cgroup. | |
175 | * If success, returns 0. | |
176 | */ | |
177 | static inline int | |
178 | page_cgroup_assign_new_page_cgroup(struct page *page, struct page_cgroup *pc) | |
179 | { | |
180 | int ret = 0; | |
181 | ||
182 | lock_page_cgroup(page); | |
183 | if (!page_get_page_cgroup(page)) | |
184 | page_assign_page_cgroup(page, pc); | |
185 | else /* A page is tied to other pc. */ | |
186 | ret = 1; | |
187 | unlock_page_cgroup(page); | |
188 | return ret; | |
189 | } | |
190 | ||
191 | /* | |
192 | * Clear page->page_cgroup member under lock_page_cgroup(). | |
193 | * If given "pc" value is different from one page->page_cgroup, | |
194 | * page->cgroup is not cleared. | |
195 | * Returns a value of page->page_cgroup at lock taken. | |
196 | * A can can detect failure of clearing by following | |
197 | * clear_page_cgroup(page, pc) == pc | |
198 | */ | |
199 | ||
200 | static inline struct page_cgroup * | |
201 | clear_page_cgroup(struct page *page, struct page_cgroup *pc) | |
202 | { | |
203 | struct page_cgroup *ret; | |
204 | /* lock and clear */ | |
205 | lock_page_cgroup(page); | |
206 | ret = page_get_page_cgroup(page); | |
207 | if (likely(ret == pc)) | |
208 | page_assign_page_cgroup(page, NULL); | |
209 | unlock_page_cgroup(page); | |
210 | return ret; | |
211 | } | |
212 | ||
213 | ||
8697d331 | 214 | static void __mem_cgroup_move_lists(struct page_cgroup *pc, bool active) |
66e1707b BS |
215 | { |
216 | if (active) | |
217 | list_move(&pc->lru, &pc->mem_cgroup->active_list); | |
218 | else | |
219 | list_move(&pc->lru, &pc->mem_cgroup->inactive_list); | |
220 | } | |
221 | ||
4c4a2214 DR |
222 | int task_in_mem_cgroup(struct task_struct *task, const struct mem_cgroup *mem) |
223 | { | |
224 | int ret; | |
225 | ||
226 | task_lock(task); | |
227 | ret = task->mm && mm_cgroup(task->mm) == mem; | |
228 | task_unlock(task); | |
229 | return ret; | |
230 | } | |
231 | ||
66e1707b BS |
232 | /* |
233 | * This routine assumes that the appropriate zone's lru lock is already held | |
234 | */ | |
235 | void mem_cgroup_move_lists(struct page_cgroup *pc, bool active) | |
236 | { | |
237 | struct mem_cgroup *mem; | |
238 | if (!pc) | |
239 | return; | |
240 | ||
241 | mem = pc->mem_cgroup; | |
242 | ||
243 | spin_lock(&mem->lru_lock); | |
244 | __mem_cgroup_move_lists(pc, active); | |
245 | spin_unlock(&mem->lru_lock); | |
246 | } | |
247 | ||
248 | unsigned long mem_cgroup_isolate_pages(unsigned long nr_to_scan, | |
249 | struct list_head *dst, | |
250 | unsigned long *scanned, int order, | |
251 | int mode, struct zone *z, | |
252 | struct mem_cgroup *mem_cont, | |
253 | int active) | |
254 | { | |
255 | unsigned long nr_taken = 0; | |
256 | struct page *page; | |
257 | unsigned long scan; | |
258 | LIST_HEAD(pc_list); | |
259 | struct list_head *src; | |
ff7283fa | 260 | struct page_cgroup *pc, *tmp; |
66e1707b BS |
261 | |
262 | if (active) | |
263 | src = &mem_cont->active_list; | |
264 | else | |
265 | src = &mem_cont->inactive_list; | |
266 | ||
267 | spin_lock(&mem_cont->lru_lock); | |
ff7283fa KH |
268 | scan = 0; |
269 | list_for_each_entry_safe_reverse(pc, tmp, src, lru) { | |
436c6541 | 270 | if (scan >= nr_to_scan) |
ff7283fa | 271 | break; |
66e1707b BS |
272 | page = pc->page; |
273 | VM_BUG_ON(!pc); | |
274 | ||
436c6541 | 275 | if (unlikely(!PageLRU(page))) |
ff7283fa | 276 | continue; |
ff7283fa | 277 | |
66e1707b BS |
278 | if (PageActive(page) && !active) { |
279 | __mem_cgroup_move_lists(pc, true); | |
66e1707b BS |
280 | continue; |
281 | } | |
282 | if (!PageActive(page) && active) { | |
283 | __mem_cgroup_move_lists(pc, false); | |
66e1707b BS |
284 | continue; |
285 | } | |
286 | ||
287 | /* | |
288 | * Reclaim, per zone | |
289 | * TODO: make the active/inactive lists per zone | |
290 | */ | |
291 | if (page_zone(page) != z) | |
292 | continue; | |
293 | ||
436c6541 HD |
294 | scan++; |
295 | list_move(&pc->lru, &pc_list); | |
66e1707b BS |
296 | |
297 | if (__isolate_lru_page(page, mode) == 0) { | |
298 | list_move(&page->lru, dst); | |
299 | nr_taken++; | |
300 | } | |
301 | } | |
302 | ||
303 | list_splice(&pc_list, src); | |
304 | spin_unlock(&mem_cont->lru_lock); | |
305 | ||
306 | *scanned = scan; | |
307 | return nr_taken; | |
308 | } | |
309 | ||
8a9f3ccd BS |
310 | /* |
311 | * Charge the memory controller for page usage. | |
312 | * Return | |
313 | * 0 if the charge was successful | |
314 | * < 0 if the cgroup is over its limit | |
315 | */ | |
217bc319 KH |
316 | static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm, |
317 | gfp_t gfp_mask, enum charge_type ctype) | |
8a9f3ccd BS |
318 | { |
319 | struct mem_cgroup *mem; | |
9175e031 | 320 | struct page_cgroup *pc; |
66e1707b BS |
321 | unsigned long flags; |
322 | unsigned long nr_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
8a9f3ccd BS |
323 | |
324 | /* | |
325 | * Should page_cgroup's go to their own slab? | |
326 | * One could optimize the performance of the charging routine | |
327 | * by saving a bit in the page_flags and using it as a lock | |
328 | * to see if the cgroup page already has a page_cgroup associated | |
329 | * with it | |
330 | */ | |
66e1707b | 331 | retry: |
8a9f3ccd BS |
332 | lock_page_cgroup(page); |
333 | pc = page_get_page_cgroup(page); | |
334 | /* | |
335 | * The page_cgroup exists and the page has already been accounted | |
336 | */ | |
337 | if (pc) { | |
66e1707b BS |
338 | if (unlikely(!atomic_inc_not_zero(&pc->ref_cnt))) { |
339 | /* this page is under being uncharged ? */ | |
340 | unlock_page_cgroup(page); | |
341 | cpu_relax(); | |
342 | goto retry; | |
9175e031 KH |
343 | } else { |
344 | unlock_page_cgroup(page); | |
66e1707b | 345 | goto done; |
9175e031 | 346 | } |
8a9f3ccd BS |
347 | } |
348 | ||
349 | unlock_page_cgroup(page); | |
350 | ||
e1a1cd59 | 351 | pc = kzalloc(sizeof(struct page_cgroup), gfp_mask); |
8a9f3ccd BS |
352 | if (pc == NULL) |
353 | goto err; | |
354 | ||
355 | rcu_read_lock(); | |
356 | /* | |
357 | * We always charge the cgroup the mm_struct belongs to | |
358 | * the mm_struct's mem_cgroup changes on task migration if the | |
359 | * thread group leader migrates. It's possible that mm is not | |
360 | * set, if so charge the init_mm (happens for pagecache usage). | |
361 | */ | |
362 | if (!mm) | |
363 | mm = &init_mm; | |
364 | ||
365 | mem = rcu_dereference(mm->mem_cgroup); | |
366 | /* | |
367 | * For every charge from the cgroup, increment reference | |
368 | * count | |
369 | */ | |
370 | css_get(&mem->css); | |
371 | rcu_read_unlock(); | |
372 | ||
373 | /* | |
374 | * If we created the page_cgroup, we should free it on exceeding | |
375 | * the cgroup limit. | |
376 | */ | |
0eea1030 | 377 | while (res_counter_charge(&mem->res, PAGE_SIZE)) { |
e1a1cd59 BS |
378 | bool is_atomic = gfp_mask & GFP_ATOMIC; |
379 | /* | |
380 | * We cannot reclaim under GFP_ATOMIC, fail the charge | |
381 | */ | |
382 | if (is_atomic) | |
383 | goto noreclaim; | |
384 | ||
385 | if (try_to_free_mem_cgroup_pages(mem, gfp_mask)) | |
66e1707b BS |
386 | continue; |
387 | ||
388 | /* | |
389 | * try_to_free_mem_cgroup_pages() might not give us a full | |
390 | * picture of reclaim. Some pages are reclaimed and might be | |
391 | * moved to swap cache or just unmapped from the cgroup. | |
392 | * Check the limit again to see if the reclaim reduced the | |
393 | * current usage of the cgroup before giving up | |
394 | */ | |
395 | if (res_counter_check_under_limit(&mem->res)) | |
396 | continue; | |
397 | /* | |
398 | * Since we control both RSS and cache, we end up with a | |
399 | * very interesting scenario where we end up reclaiming | |
400 | * memory (essentially RSS), since the memory is pushed | |
401 | * to swap cache, we eventually end up adding those | |
402 | * pages back to our list. Hence we give ourselves a | |
403 | * few chances before we fail | |
404 | */ | |
405 | else if (nr_retries--) { | |
406 | congestion_wait(WRITE, HZ/10); | |
407 | continue; | |
408 | } | |
e1a1cd59 | 409 | noreclaim: |
8a9f3ccd | 410 | css_put(&mem->css); |
e1a1cd59 BS |
411 | if (!is_atomic) |
412 | mem_cgroup_out_of_memory(mem, GFP_KERNEL); | |
8a9f3ccd BS |
413 | goto free_pc; |
414 | } | |
415 | ||
8a9f3ccd BS |
416 | atomic_set(&pc->ref_cnt, 1); |
417 | pc->mem_cgroup = mem; | |
418 | pc->page = page; | |
217bc319 KH |
419 | pc->flags = 0; |
420 | if (ctype == MEM_CGROUP_CHARGE_TYPE_CACHE) | |
421 | pc->flags |= PAGE_CGROUP_FLAG_CACHE; | |
9175e031 KH |
422 | if (page_cgroup_assign_new_page_cgroup(page, pc)) { |
423 | /* | |
424 | * an another charge is added to this page already. | |
425 | * we do take lock_page_cgroup(page) again and read | |
426 | * page->cgroup, increment refcnt.... just retry is OK. | |
427 | */ | |
428 | res_counter_uncharge(&mem->res, PAGE_SIZE); | |
429 | css_put(&mem->css); | |
430 | kfree(pc); | |
431 | goto retry; | |
432 | } | |
8a9f3ccd | 433 | |
66e1707b BS |
434 | spin_lock_irqsave(&mem->lru_lock, flags); |
435 | list_add(&pc->lru, &mem->active_list); | |
436 | spin_unlock_irqrestore(&mem->lru_lock, flags); | |
437 | ||
8a9f3ccd | 438 | done: |
8a9f3ccd BS |
439 | return 0; |
440 | free_pc: | |
441 | kfree(pc); | |
8a9f3ccd | 442 | err: |
8a9f3ccd BS |
443 | return -ENOMEM; |
444 | } | |
445 | ||
217bc319 KH |
446 | int mem_cgroup_charge(struct page *page, struct mm_struct *mm, |
447 | gfp_t gfp_mask) | |
448 | { | |
449 | return mem_cgroup_charge_common(page, mm, gfp_mask, | |
450 | MEM_CGROUP_CHARGE_TYPE_MAPPED); | |
451 | } | |
452 | ||
8697d331 BS |
453 | /* |
454 | * See if the cached pages should be charged at all? | |
455 | */ | |
e1a1cd59 BS |
456 | int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm, |
457 | gfp_t gfp_mask) | |
8697d331 | 458 | { |
ac44d354 | 459 | int ret = 0; |
8697d331 BS |
460 | struct mem_cgroup *mem; |
461 | if (!mm) | |
462 | mm = &init_mm; | |
463 | ||
ac44d354 | 464 | rcu_read_lock(); |
8697d331 | 465 | mem = rcu_dereference(mm->mem_cgroup); |
ac44d354 BS |
466 | css_get(&mem->css); |
467 | rcu_read_unlock(); | |
8697d331 | 468 | if (mem->control_type == MEM_CGROUP_TYPE_ALL) |
ac44d354 | 469 | ret = mem_cgroup_charge_common(page, mm, gfp_mask, |
217bc319 | 470 | MEM_CGROUP_CHARGE_TYPE_CACHE); |
ac44d354 BS |
471 | css_put(&mem->css); |
472 | return ret; | |
8697d331 BS |
473 | } |
474 | ||
8a9f3ccd BS |
475 | /* |
476 | * Uncharging is always a welcome operation, we never complain, simply | |
477 | * uncharge. | |
478 | */ | |
479 | void mem_cgroup_uncharge(struct page_cgroup *pc) | |
480 | { | |
481 | struct mem_cgroup *mem; | |
482 | struct page *page; | |
66e1707b | 483 | unsigned long flags; |
8a9f3ccd | 484 | |
8697d331 BS |
485 | /* |
486 | * This can handle cases when a page is not charged at all and we | |
487 | * are switching between handling the control_type. | |
488 | */ | |
8a9f3ccd BS |
489 | if (!pc) |
490 | return; | |
491 | ||
492 | if (atomic_dec_and_test(&pc->ref_cnt)) { | |
493 | page = pc->page; | |
9175e031 KH |
494 | /* |
495 | * get page->cgroup and clear it under lock. | |
cc847582 | 496 | * force_empty can drop page->cgroup without checking refcnt. |
9175e031 KH |
497 | */ |
498 | if (clear_page_cgroup(page, pc) == pc) { | |
499 | mem = pc->mem_cgroup; | |
500 | css_put(&mem->css); | |
501 | res_counter_uncharge(&mem->res, PAGE_SIZE); | |
502 | spin_lock_irqsave(&mem->lru_lock, flags); | |
503 | list_del_init(&pc->lru); | |
504 | spin_unlock_irqrestore(&mem->lru_lock, flags); | |
505 | kfree(pc); | |
9175e031 | 506 | } |
8a9f3ccd | 507 | } |
78fb7466 | 508 | } |
ae41be37 KH |
509 | /* |
510 | * Returns non-zero if a page (under migration) has valid page_cgroup member. | |
511 | * Refcnt of page_cgroup is incremented. | |
512 | */ | |
513 | ||
514 | int mem_cgroup_prepare_migration(struct page *page) | |
515 | { | |
516 | struct page_cgroup *pc; | |
517 | int ret = 0; | |
518 | lock_page_cgroup(page); | |
519 | pc = page_get_page_cgroup(page); | |
520 | if (pc && atomic_inc_not_zero(&pc->ref_cnt)) | |
521 | ret = 1; | |
522 | unlock_page_cgroup(page); | |
523 | return ret; | |
524 | } | |
525 | ||
526 | void mem_cgroup_end_migration(struct page *page) | |
527 | { | |
528 | struct page_cgroup *pc = page_get_page_cgroup(page); | |
529 | mem_cgroup_uncharge(pc); | |
530 | } | |
531 | /* | |
532 | * We know both *page* and *newpage* are now not-on-LRU and Pg_locked. | |
533 | * And no race with uncharge() routines because page_cgroup for *page* | |
534 | * has extra one reference by mem_cgroup_prepare_migration. | |
535 | */ | |
536 | ||
537 | void mem_cgroup_page_migration(struct page *page, struct page *newpage) | |
538 | { | |
539 | struct page_cgroup *pc; | |
540 | retry: | |
541 | pc = page_get_page_cgroup(page); | |
542 | if (!pc) | |
543 | return; | |
544 | if (clear_page_cgroup(page, pc) != pc) | |
545 | goto retry; | |
546 | pc->page = newpage; | |
547 | lock_page_cgroup(newpage); | |
548 | page_assign_page_cgroup(newpage, pc); | |
549 | unlock_page_cgroup(newpage); | |
550 | return; | |
551 | } | |
78fb7466 | 552 | |
cc847582 KH |
553 | /* |
554 | * This routine traverse page_cgroup in given list and drop them all. | |
555 | * This routine ignores page_cgroup->ref_cnt. | |
556 | * *And* this routine doesn't reclaim page itself, just removes page_cgroup. | |
557 | */ | |
558 | #define FORCE_UNCHARGE_BATCH (128) | |
559 | static void | |
560 | mem_cgroup_force_empty_list(struct mem_cgroup *mem, struct list_head *list) | |
561 | { | |
562 | struct page_cgroup *pc; | |
563 | struct page *page; | |
564 | int count; | |
565 | unsigned long flags; | |
566 | ||
567 | retry: | |
568 | count = FORCE_UNCHARGE_BATCH; | |
569 | spin_lock_irqsave(&mem->lru_lock, flags); | |
570 | ||
571 | while (--count && !list_empty(list)) { | |
572 | pc = list_entry(list->prev, struct page_cgroup, lru); | |
573 | page = pc->page; | |
574 | /* Avoid race with charge */ | |
575 | atomic_set(&pc->ref_cnt, 0); | |
576 | if (clear_page_cgroup(page, pc) == pc) { | |
577 | css_put(&mem->css); | |
578 | res_counter_uncharge(&mem->res, PAGE_SIZE); | |
579 | list_del_init(&pc->lru); | |
580 | kfree(pc); | |
581 | } else /* being uncharged ? ...do relax */ | |
582 | break; | |
583 | } | |
584 | spin_unlock_irqrestore(&mem->lru_lock, flags); | |
585 | if (!list_empty(list)) { | |
586 | cond_resched(); | |
587 | goto retry; | |
588 | } | |
589 | return; | |
590 | } | |
591 | ||
592 | /* | |
593 | * make mem_cgroup's charge to be 0 if there is no task. | |
594 | * This enables deleting this mem_cgroup. | |
595 | */ | |
596 | ||
597 | int mem_cgroup_force_empty(struct mem_cgroup *mem) | |
598 | { | |
599 | int ret = -EBUSY; | |
600 | css_get(&mem->css); | |
601 | /* | |
602 | * page reclaim code (kswapd etc..) will move pages between | |
603 | ` * active_list <-> inactive_list while we don't take a lock. | |
604 | * So, we have to do loop here until all lists are empty. | |
605 | */ | |
606 | while (!(list_empty(&mem->active_list) && | |
607 | list_empty(&mem->inactive_list))) { | |
608 | if (atomic_read(&mem->css.cgroup->count) > 0) | |
609 | goto out; | |
610 | /* drop all page_cgroup in active_list */ | |
611 | mem_cgroup_force_empty_list(mem, &mem->active_list); | |
612 | /* drop all page_cgroup in inactive_list */ | |
613 | mem_cgroup_force_empty_list(mem, &mem->inactive_list); | |
614 | } | |
615 | ret = 0; | |
616 | out: | |
617 | css_put(&mem->css); | |
618 | return ret; | |
619 | } | |
620 | ||
621 | ||
622 | ||
0eea1030 BS |
623 | int mem_cgroup_write_strategy(char *buf, unsigned long long *tmp) |
624 | { | |
625 | *tmp = memparse(buf, &buf); | |
626 | if (*buf != '\0') | |
627 | return -EINVAL; | |
628 | ||
629 | /* | |
630 | * Round up the value to the closest page size | |
631 | */ | |
632 | *tmp = ((*tmp + PAGE_SIZE - 1) >> PAGE_SHIFT) << PAGE_SHIFT; | |
633 | return 0; | |
634 | } | |
635 | ||
636 | static ssize_t mem_cgroup_read(struct cgroup *cont, | |
637 | struct cftype *cft, struct file *file, | |
638 | char __user *userbuf, size_t nbytes, loff_t *ppos) | |
8cdea7c0 BS |
639 | { |
640 | return res_counter_read(&mem_cgroup_from_cont(cont)->res, | |
0eea1030 BS |
641 | cft->private, userbuf, nbytes, ppos, |
642 | NULL); | |
8cdea7c0 BS |
643 | } |
644 | ||
645 | static ssize_t mem_cgroup_write(struct cgroup *cont, struct cftype *cft, | |
646 | struct file *file, const char __user *userbuf, | |
647 | size_t nbytes, loff_t *ppos) | |
648 | { | |
649 | return res_counter_write(&mem_cgroup_from_cont(cont)->res, | |
0eea1030 BS |
650 | cft->private, userbuf, nbytes, ppos, |
651 | mem_cgroup_write_strategy); | |
8cdea7c0 BS |
652 | } |
653 | ||
8697d331 BS |
654 | static ssize_t mem_control_type_write(struct cgroup *cont, |
655 | struct cftype *cft, struct file *file, | |
656 | const char __user *userbuf, | |
657 | size_t nbytes, loff_t *pos) | |
658 | { | |
659 | int ret; | |
660 | char *buf, *end; | |
661 | unsigned long tmp; | |
662 | struct mem_cgroup *mem; | |
663 | ||
664 | mem = mem_cgroup_from_cont(cont); | |
665 | buf = kmalloc(nbytes + 1, GFP_KERNEL); | |
666 | ret = -ENOMEM; | |
667 | if (buf == NULL) | |
668 | goto out; | |
669 | ||
670 | buf[nbytes] = 0; | |
671 | ret = -EFAULT; | |
672 | if (copy_from_user(buf, userbuf, nbytes)) | |
673 | goto out_free; | |
674 | ||
675 | ret = -EINVAL; | |
676 | tmp = simple_strtoul(buf, &end, 10); | |
677 | if (*end != '\0') | |
678 | goto out_free; | |
679 | ||
680 | if (tmp <= MEM_CGROUP_TYPE_UNSPEC || tmp >= MEM_CGROUP_TYPE_MAX) | |
681 | goto out_free; | |
682 | ||
683 | mem->control_type = tmp; | |
684 | ret = nbytes; | |
685 | out_free: | |
686 | kfree(buf); | |
687 | out: | |
688 | return ret; | |
689 | } | |
690 | ||
691 | static ssize_t mem_control_type_read(struct cgroup *cont, | |
692 | struct cftype *cft, | |
693 | struct file *file, char __user *userbuf, | |
694 | size_t nbytes, loff_t *ppos) | |
695 | { | |
696 | unsigned long val; | |
697 | char buf[64], *s; | |
698 | struct mem_cgroup *mem; | |
699 | ||
700 | mem = mem_cgroup_from_cont(cont); | |
701 | s = buf; | |
702 | val = mem->control_type; | |
703 | s += sprintf(s, "%lu\n", val); | |
704 | return simple_read_from_buffer((void __user *)userbuf, nbytes, | |
705 | ppos, buf, s - buf); | |
706 | } | |
707 | ||
cc847582 KH |
708 | |
709 | static ssize_t mem_force_empty_write(struct cgroup *cont, | |
710 | struct cftype *cft, struct file *file, | |
711 | const char __user *userbuf, | |
712 | size_t nbytes, loff_t *ppos) | |
713 | { | |
714 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); | |
715 | int ret; | |
716 | ret = mem_cgroup_force_empty(mem); | |
717 | if (!ret) | |
718 | ret = nbytes; | |
719 | return ret; | |
720 | } | |
721 | ||
722 | /* | |
723 | * Note: This should be removed if cgroup supports write-only file. | |
724 | */ | |
725 | ||
726 | static ssize_t mem_force_empty_read(struct cgroup *cont, | |
727 | struct cftype *cft, | |
728 | struct file *file, char __user *userbuf, | |
729 | size_t nbytes, loff_t *ppos) | |
730 | { | |
731 | return -EINVAL; | |
732 | } | |
733 | ||
734 | ||
8cdea7c0 BS |
735 | static struct cftype mem_cgroup_files[] = { |
736 | { | |
0eea1030 | 737 | .name = "usage_in_bytes", |
8cdea7c0 BS |
738 | .private = RES_USAGE, |
739 | .read = mem_cgroup_read, | |
740 | }, | |
741 | { | |
0eea1030 | 742 | .name = "limit_in_bytes", |
8cdea7c0 BS |
743 | .private = RES_LIMIT, |
744 | .write = mem_cgroup_write, | |
745 | .read = mem_cgroup_read, | |
746 | }, | |
747 | { | |
748 | .name = "failcnt", | |
749 | .private = RES_FAILCNT, | |
750 | .read = mem_cgroup_read, | |
751 | }, | |
8697d331 BS |
752 | { |
753 | .name = "control_type", | |
754 | .write = mem_control_type_write, | |
755 | .read = mem_control_type_read, | |
756 | }, | |
cc847582 KH |
757 | { |
758 | .name = "force_empty", | |
759 | .write = mem_force_empty_write, | |
760 | .read = mem_force_empty_read, | |
761 | }, | |
8cdea7c0 BS |
762 | }; |
763 | ||
78fb7466 PE |
764 | static struct mem_cgroup init_mem_cgroup; |
765 | ||
8cdea7c0 BS |
766 | static struct cgroup_subsys_state * |
767 | mem_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cont) | |
768 | { | |
769 | struct mem_cgroup *mem; | |
770 | ||
78fb7466 PE |
771 | if (unlikely((cont->parent) == NULL)) { |
772 | mem = &init_mem_cgroup; | |
773 | init_mm.mem_cgroup = mem; | |
774 | } else | |
775 | mem = kzalloc(sizeof(struct mem_cgroup), GFP_KERNEL); | |
776 | ||
777 | if (mem == NULL) | |
778 | return NULL; | |
8cdea7c0 BS |
779 | |
780 | res_counter_init(&mem->res); | |
8a9f3ccd BS |
781 | INIT_LIST_HEAD(&mem->active_list); |
782 | INIT_LIST_HEAD(&mem->inactive_list); | |
66e1707b | 783 | spin_lock_init(&mem->lru_lock); |
8697d331 | 784 | mem->control_type = MEM_CGROUP_TYPE_ALL; |
8cdea7c0 BS |
785 | return &mem->css; |
786 | } | |
787 | ||
788 | static void mem_cgroup_destroy(struct cgroup_subsys *ss, | |
789 | struct cgroup *cont) | |
790 | { | |
791 | kfree(mem_cgroup_from_cont(cont)); | |
792 | } | |
793 | ||
794 | static int mem_cgroup_populate(struct cgroup_subsys *ss, | |
795 | struct cgroup *cont) | |
796 | { | |
797 | return cgroup_add_files(cont, ss, mem_cgroup_files, | |
798 | ARRAY_SIZE(mem_cgroup_files)); | |
799 | } | |
800 | ||
67e465a7 BS |
801 | static void mem_cgroup_move_task(struct cgroup_subsys *ss, |
802 | struct cgroup *cont, | |
803 | struct cgroup *old_cont, | |
804 | struct task_struct *p) | |
805 | { | |
806 | struct mm_struct *mm; | |
807 | struct mem_cgroup *mem, *old_mem; | |
808 | ||
809 | mm = get_task_mm(p); | |
810 | if (mm == NULL) | |
811 | return; | |
812 | ||
813 | mem = mem_cgroup_from_cont(cont); | |
814 | old_mem = mem_cgroup_from_cont(old_cont); | |
815 | ||
816 | if (mem == old_mem) | |
817 | goto out; | |
818 | ||
819 | /* | |
820 | * Only thread group leaders are allowed to migrate, the mm_struct is | |
821 | * in effect owned by the leader | |
822 | */ | |
823 | if (p->tgid != p->pid) | |
824 | goto out; | |
825 | ||
826 | css_get(&mem->css); | |
827 | rcu_assign_pointer(mm->mem_cgroup, mem); | |
828 | css_put(&old_mem->css); | |
829 | ||
830 | out: | |
831 | mmput(mm); | |
832 | return; | |
833 | } | |
834 | ||
8cdea7c0 BS |
835 | struct cgroup_subsys mem_cgroup_subsys = { |
836 | .name = "memory", | |
837 | .subsys_id = mem_cgroup_subsys_id, | |
838 | .create = mem_cgroup_create, | |
839 | .destroy = mem_cgroup_destroy, | |
840 | .populate = mem_cgroup_populate, | |
67e465a7 | 841 | .attach = mem_cgroup_move_task, |
78fb7466 | 842 | .early_init = 1, |
8cdea7c0 | 843 | }; |