Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 1 | /* memcontrol.c - Memory Controller |
| 2 | * |
| 3 | * Copyright IBM Corporation, 2007 |
| 4 | * Author Balbir Singh <balbir@linux.vnet.ibm.com> |
| 5 | * |
Pavel Emelianov | 78fb746 | 2008-02-07 00:13:51 -0800 | [diff] [blame] | 6 | * Copyright 2007 OpenVZ SWsoft Inc |
| 7 | * Author: Pavel Emelianov <xemul@openvz.org> |
| 8 | * |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 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> |
Pavel Emelianov | 78fb746 | 2008-02-07 00:13:51 -0800 | [diff] [blame] | 23 | #include <linux/mm.h> |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 24 | #include <linux/page-flags.h> |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 25 | #include <linux/backing-dev.h> |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 26 | #include <linux/bit_spinlock.h> |
| 27 | #include <linux/rcupdate.h> |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 28 | #include <linux/swap.h> |
| 29 | #include <linux/spinlock.h> |
| 30 | #include <linux/fs.h> |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 31 | |
Balbir Singh | 8697d33 | 2008-02-07 00:13:59 -0800 | [diff] [blame] | 32 | #include <asm/uaccess.h> |
| 33 | |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 34 | struct cgroup_subsys mem_cgroup_subsys; |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 35 | static const int MEM_CGROUP_RECLAIM_RETRIES = 5; |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 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 |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 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. |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 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; |
Pavel Emelianov | 78fb746 | 2008-02-07 00:13:51 -0800 | [diff] [blame] | 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; |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 61 | /* |
| 62 | * spin_lock to protect the per cgroup LRU |
| 63 | */ |
| 64 | spinlock_t lru_lock; |
Balbir Singh | 8697d33 | 2008-02-07 00:13:59 -0800 | [diff] [blame] | 65 | unsigned long control_type; /* control RSS or RSS+Pagecache */ |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 66 | }; |
| 67 | |
| 68 | /* |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 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 | |
| 76 | /* |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 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; |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 84 | atomic_t ref_cnt; /* Helpful when pages move b/w */ |
| 85 | /* mapped and cached states */ |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 86 | }; |
| 87 | |
Balbir Singh | 8697d33 | 2008-02-07 00:13:59 -0800 | [diff] [blame] | 88 | enum { |
| 89 | MEM_CGROUP_TYPE_UNSPEC = 0, |
| 90 | MEM_CGROUP_TYPE_MAPPED, |
| 91 | MEM_CGROUP_TYPE_CACHED, |
| 92 | MEM_CGROUP_TYPE_ALL, |
| 93 | MEM_CGROUP_TYPE_MAX, |
| 94 | }; |
| 95 | |
| 96 | static struct mem_cgroup init_mem_cgroup; |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 97 | |
| 98 | static inline |
| 99 | struct mem_cgroup *mem_cgroup_from_cont(struct cgroup *cont) |
| 100 | { |
| 101 | return container_of(cgroup_subsys_state(cont, |
| 102 | mem_cgroup_subsys_id), struct mem_cgroup, |
| 103 | css); |
| 104 | } |
| 105 | |
Pavel Emelianov | 78fb746 | 2008-02-07 00:13:51 -0800 | [diff] [blame] | 106 | static inline |
| 107 | struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) |
| 108 | { |
| 109 | return container_of(task_subsys_state(p, mem_cgroup_subsys_id), |
| 110 | struct mem_cgroup, css); |
| 111 | } |
| 112 | |
Balbir Singh | bed7161 | 2008-02-07 00:14:01 -0800 | [diff] [blame] | 113 | inline struct mem_cgroup *mm_cgroup(struct mm_struct *mm) |
| 114 | { |
| 115 | return rcu_dereference(mm->mem_cgroup); |
| 116 | } |
| 117 | |
Pavel Emelianov | 78fb746 | 2008-02-07 00:13:51 -0800 | [diff] [blame] | 118 | void mm_init_cgroup(struct mm_struct *mm, struct task_struct *p) |
| 119 | { |
| 120 | struct mem_cgroup *mem; |
| 121 | |
| 122 | mem = mem_cgroup_from_task(p); |
| 123 | css_get(&mem->css); |
| 124 | mm->mem_cgroup = mem; |
| 125 | } |
| 126 | |
| 127 | void mm_free_cgroup(struct mm_struct *mm) |
| 128 | { |
| 129 | css_put(&mm->mem_cgroup->css); |
| 130 | } |
| 131 | |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 132 | static inline int page_cgroup_locked(struct page *page) |
| 133 | { |
| 134 | return bit_spin_is_locked(PAGE_CGROUP_LOCK_BIT, |
| 135 | &page->page_cgroup); |
| 136 | } |
| 137 | |
Pavel Emelianov | 78fb746 | 2008-02-07 00:13:51 -0800 | [diff] [blame] | 138 | void page_assign_page_cgroup(struct page *page, struct page_cgroup *pc) |
| 139 | { |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 140 | int locked; |
| 141 | |
| 142 | /* |
| 143 | * While resetting the page_cgroup we might not hold the |
| 144 | * page_cgroup lock. free_hot_cold_page() is an example |
| 145 | * of such a scenario |
| 146 | */ |
| 147 | if (pc) |
| 148 | VM_BUG_ON(!page_cgroup_locked(page)); |
| 149 | locked = (page->page_cgroup & PAGE_CGROUP_LOCK); |
| 150 | page->page_cgroup = ((unsigned long)pc | locked); |
Pavel Emelianov | 78fb746 | 2008-02-07 00:13:51 -0800 | [diff] [blame] | 151 | } |
| 152 | |
| 153 | struct page_cgroup *page_get_page_cgroup(struct page *page) |
| 154 | { |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 155 | return (struct page_cgroup *) |
| 156 | (page->page_cgroup & ~PAGE_CGROUP_LOCK); |
| 157 | } |
| 158 | |
Balbir Singh | 8697d33 | 2008-02-07 00:13:59 -0800 | [diff] [blame] | 159 | static void __always_inline lock_page_cgroup(struct page *page) |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 160 | { |
| 161 | bit_spin_lock(PAGE_CGROUP_LOCK_BIT, &page->page_cgroup); |
| 162 | VM_BUG_ON(!page_cgroup_locked(page)); |
| 163 | } |
| 164 | |
Balbir Singh | 8697d33 | 2008-02-07 00:13:59 -0800 | [diff] [blame] | 165 | static void __always_inline unlock_page_cgroup(struct page *page) |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 166 | { |
| 167 | bit_spin_unlock(PAGE_CGROUP_LOCK_BIT, &page->page_cgroup); |
| 168 | } |
| 169 | |
Balbir Singh | 8697d33 | 2008-02-07 00:13:59 -0800 | [diff] [blame] | 170 | static void __mem_cgroup_move_lists(struct page_cgroup *pc, bool active) |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 171 | { |
| 172 | if (active) |
| 173 | list_move(&pc->lru, &pc->mem_cgroup->active_list); |
| 174 | else |
| 175 | list_move(&pc->lru, &pc->mem_cgroup->inactive_list); |
| 176 | } |
| 177 | |
| 178 | /* |
| 179 | * This routine assumes that the appropriate zone's lru lock is already held |
| 180 | */ |
| 181 | void mem_cgroup_move_lists(struct page_cgroup *pc, bool active) |
| 182 | { |
| 183 | struct mem_cgroup *mem; |
| 184 | if (!pc) |
| 185 | return; |
| 186 | |
| 187 | mem = pc->mem_cgroup; |
| 188 | |
| 189 | spin_lock(&mem->lru_lock); |
| 190 | __mem_cgroup_move_lists(pc, active); |
| 191 | spin_unlock(&mem->lru_lock); |
| 192 | } |
| 193 | |
| 194 | unsigned long mem_cgroup_isolate_pages(unsigned long nr_to_scan, |
| 195 | struct list_head *dst, |
| 196 | unsigned long *scanned, int order, |
| 197 | int mode, struct zone *z, |
| 198 | struct mem_cgroup *mem_cont, |
| 199 | int active) |
| 200 | { |
| 201 | unsigned long nr_taken = 0; |
| 202 | struct page *page; |
| 203 | unsigned long scan; |
| 204 | LIST_HEAD(pc_list); |
| 205 | struct list_head *src; |
| 206 | struct page_cgroup *pc; |
| 207 | |
| 208 | if (active) |
| 209 | src = &mem_cont->active_list; |
| 210 | else |
| 211 | src = &mem_cont->inactive_list; |
| 212 | |
| 213 | spin_lock(&mem_cont->lru_lock); |
| 214 | for (scan = 0; scan < nr_to_scan && !list_empty(src); scan++) { |
| 215 | pc = list_entry(src->prev, struct page_cgroup, lru); |
| 216 | page = pc->page; |
| 217 | VM_BUG_ON(!pc); |
| 218 | |
| 219 | if (PageActive(page) && !active) { |
| 220 | __mem_cgroup_move_lists(pc, true); |
| 221 | scan--; |
| 222 | continue; |
| 223 | } |
| 224 | if (!PageActive(page) && active) { |
| 225 | __mem_cgroup_move_lists(pc, false); |
| 226 | scan--; |
| 227 | continue; |
| 228 | } |
| 229 | |
| 230 | /* |
| 231 | * Reclaim, per zone |
| 232 | * TODO: make the active/inactive lists per zone |
| 233 | */ |
| 234 | if (page_zone(page) != z) |
| 235 | continue; |
| 236 | |
| 237 | /* |
| 238 | * Check if the meta page went away from under us |
| 239 | */ |
| 240 | if (!list_empty(&pc->lru)) |
| 241 | list_move(&pc->lru, &pc_list); |
| 242 | else |
| 243 | continue; |
| 244 | |
| 245 | if (__isolate_lru_page(page, mode) == 0) { |
| 246 | list_move(&page->lru, dst); |
| 247 | nr_taken++; |
| 248 | } |
| 249 | } |
| 250 | |
| 251 | list_splice(&pc_list, src); |
| 252 | spin_unlock(&mem_cont->lru_lock); |
| 253 | |
| 254 | *scanned = scan; |
| 255 | return nr_taken; |
| 256 | } |
| 257 | |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 258 | /* |
| 259 | * Charge the memory controller for page usage. |
| 260 | * Return |
| 261 | * 0 if the charge was successful |
| 262 | * < 0 if the cgroup is over its limit |
| 263 | */ |
Balbir Singh | e1a1cd5 | 2008-02-07 00:14:02 -0800 | [diff] [blame^] | 264 | int mem_cgroup_charge(struct page *page, struct mm_struct *mm, |
| 265 | gfp_t gfp_mask) |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 266 | { |
| 267 | struct mem_cgroup *mem; |
| 268 | struct page_cgroup *pc, *race_pc; |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 269 | unsigned long flags; |
| 270 | unsigned long nr_retries = MEM_CGROUP_RECLAIM_RETRIES; |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 271 | |
| 272 | /* |
| 273 | * Should page_cgroup's go to their own slab? |
| 274 | * One could optimize the performance of the charging routine |
| 275 | * by saving a bit in the page_flags and using it as a lock |
| 276 | * to see if the cgroup page already has a page_cgroup associated |
| 277 | * with it |
| 278 | */ |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 279 | retry: |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 280 | lock_page_cgroup(page); |
| 281 | pc = page_get_page_cgroup(page); |
| 282 | /* |
| 283 | * The page_cgroup exists and the page has already been accounted |
| 284 | */ |
| 285 | if (pc) { |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 286 | if (unlikely(!atomic_inc_not_zero(&pc->ref_cnt))) { |
| 287 | /* this page is under being uncharged ? */ |
| 288 | unlock_page_cgroup(page); |
| 289 | cpu_relax(); |
| 290 | goto retry; |
| 291 | } else |
| 292 | goto done; |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 293 | } |
| 294 | |
| 295 | unlock_page_cgroup(page); |
| 296 | |
Balbir Singh | e1a1cd5 | 2008-02-07 00:14:02 -0800 | [diff] [blame^] | 297 | pc = kzalloc(sizeof(struct page_cgroup), gfp_mask); |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 298 | if (pc == NULL) |
| 299 | goto err; |
| 300 | |
| 301 | rcu_read_lock(); |
| 302 | /* |
| 303 | * We always charge the cgroup the mm_struct belongs to |
| 304 | * the mm_struct's mem_cgroup changes on task migration if the |
| 305 | * thread group leader migrates. It's possible that mm is not |
| 306 | * set, if so charge the init_mm (happens for pagecache usage). |
| 307 | */ |
| 308 | if (!mm) |
| 309 | mm = &init_mm; |
| 310 | |
| 311 | mem = rcu_dereference(mm->mem_cgroup); |
| 312 | /* |
| 313 | * For every charge from the cgroup, increment reference |
| 314 | * count |
| 315 | */ |
| 316 | css_get(&mem->css); |
| 317 | rcu_read_unlock(); |
| 318 | |
| 319 | /* |
| 320 | * If we created the page_cgroup, we should free it on exceeding |
| 321 | * the cgroup limit. |
| 322 | */ |
Balbir Singh | 0eea103 | 2008-02-07 00:13:57 -0800 | [diff] [blame] | 323 | while (res_counter_charge(&mem->res, PAGE_SIZE)) { |
Balbir Singh | e1a1cd5 | 2008-02-07 00:14:02 -0800 | [diff] [blame^] | 324 | bool is_atomic = gfp_mask & GFP_ATOMIC; |
| 325 | /* |
| 326 | * We cannot reclaim under GFP_ATOMIC, fail the charge |
| 327 | */ |
| 328 | if (is_atomic) |
| 329 | goto noreclaim; |
| 330 | |
| 331 | if (try_to_free_mem_cgroup_pages(mem, gfp_mask)) |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 332 | continue; |
| 333 | |
| 334 | /* |
| 335 | * try_to_free_mem_cgroup_pages() might not give us a full |
| 336 | * picture of reclaim. Some pages are reclaimed and might be |
| 337 | * moved to swap cache or just unmapped from the cgroup. |
| 338 | * Check the limit again to see if the reclaim reduced the |
| 339 | * current usage of the cgroup before giving up |
| 340 | */ |
| 341 | if (res_counter_check_under_limit(&mem->res)) |
| 342 | continue; |
| 343 | /* |
| 344 | * Since we control both RSS and cache, we end up with a |
| 345 | * very interesting scenario where we end up reclaiming |
| 346 | * memory (essentially RSS), since the memory is pushed |
| 347 | * to swap cache, we eventually end up adding those |
| 348 | * pages back to our list. Hence we give ourselves a |
| 349 | * few chances before we fail |
| 350 | */ |
| 351 | else if (nr_retries--) { |
| 352 | congestion_wait(WRITE, HZ/10); |
| 353 | continue; |
| 354 | } |
Balbir Singh | e1a1cd5 | 2008-02-07 00:14:02 -0800 | [diff] [blame^] | 355 | noreclaim: |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 356 | css_put(&mem->css); |
Balbir Singh | e1a1cd5 | 2008-02-07 00:14:02 -0800 | [diff] [blame^] | 357 | if (!is_atomic) |
| 358 | mem_cgroup_out_of_memory(mem, GFP_KERNEL); |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 359 | goto free_pc; |
| 360 | } |
| 361 | |
| 362 | lock_page_cgroup(page); |
| 363 | /* |
| 364 | * Check if somebody else beat us to allocating the page_cgroup |
| 365 | */ |
| 366 | race_pc = page_get_page_cgroup(page); |
| 367 | if (race_pc) { |
| 368 | kfree(pc); |
| 369 | pc = race_pc; |
| 370 | atomic_inc(&pc->ref_cnt); |
Balbir Singh | 0eea103 | 2008-02-07 00:13:57 -0800 | [diff] [blame] | 371 | res_counter_uncharge(&mem->res, PAGE_SIZE); |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 372 | css_put(&mem->css); |
| 373 | goto done; |
| 374 | } |
| 375 | |
| 376 | atomic_set(&pc->ref_cnt, 1); |
| 377 | pc->mem_cgroup = mem; |
| 378 | pc->page = page; |
| 379 | page_assign_page_cgroup(page, pc); |
| 380 | |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 381 | spin_lock_irqsave(&mem->lru_lock, flags); |
| 382 | list_add(&pc->lru, &mem->active_list); |
| 383 | spin_unlock_irqrestore(&mem->lru_lock, flags); |
| 384 | |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 385 | done: |
| 386 | unlock_page_cgroup(page); |
| 387 | return 0; |
| 388 | free_pc: |
| 389 | kfree(pc); |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 390 | err: |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 391 | return -ENOMEM; |
| 392 | } |
| 393 | |
| 394 | /* |
Balbir Singh | 8697d33 | 2008-02-07 00:13:59 -0800 | [diff] [blame] | 395 | * See if the cached pages should be charged at all? |
| 396 | */ |
Balbir Singh | e1a1cd5 | 2008-02-07 00:14:02 -0800 | [diff] [blame^] | 397 | int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm, |
| 398 | gfp_t gfp_mask) |
Balbir Singh | 8697d33 | 2008-02-07 00:13:59 -0800 | [diff] [blame] | 399 | { |
| 400 | struct mem_cgroup *mem; |
| 401 | if (!mm) |
| 402 | mm = &init_mm; |
| 403 | |
| 404 | mem = rcu_dereference(mm->mem_cgroup); |
| 405 | if (mem->control_type == MEM_CGROUP_TYPE_ALL) |
Balbir Singh | e1a1cd5 | 2008-02-07 00:14:02 -0800 | [diff] [blame^] | 406 | return mem_cgroup_charge(page, mm, gfp_mask); |
Balbir Singh | 8697d33 | 2008-02-07 00:13:59 -0800 | [diff] [blame] | 407 | else |
| 408 | return 0; |
| 409 | } |
| 410 | |
| 411 | /* |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 412 | * Uncharging is always a welcome operation, we never complain, simply |
| 413 | * uncharge. |
| 414 | */ |
| 415 | void mem_cgroup_uncharge(struct page_cgroup *pc) |
| 416 | { |
| 417 | struct mem_cgroup *mem; |
| 418 | struct page *page; |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 419 | unsigned long flags; |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 420 | |
Balbir Singh | 8697d33 | 2008-02-07 00:13:59 -0800 | [diff] [blame] | 421 | /* |
| 422 | * This can handle cases when a page is not charged at all and we |
| 423 | * are switching between handling the control_type. |
| 424 | */ |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 425 | if (!pc) |
| 426 | return; |
| 427 | |
| 428 | if (atomic_dec_and_test(&pc->ref_cnt)) { |
| 429 | page = pc->page; |
| 430 | lock_page_cgroup(page); |
| 431 | mem = pc->mem_cgroup; |
| 432 | css_put(&mem->css); |
| 433 | page_assign_page_cgroup(page, NULL); |
| 434 | unlock_page_cgroup(page); |
Balbir Singh | 0eea103 | 2008-02-07 00:13:57 -0800 | [diff] [blame] | 435 | res_counter_uncharge(&mem->res, PAGE_SIZE); |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 436 | |
| 437 | spin_lock_irqsave(&mem->lru_lock, flags); |
| 438 | list_del_init(&pc->lru); |
| 439 | spin_unlock_irqrestore(&mem->lru_lock, flags); |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 440 | kfree(pc); |
| 441 | } |
Pavel Emelianov | 78fb746 | 2008-02-07 00:13:51 -0800 | [diff] [blame] | 442 | } |
| 443 | |
Balbir Singh | 0eea103 | 2008-02-07 00:13:57 -0800 | [diff] [blame] | 444 | int mem_cgroup_write_strategy(char *buf, unsigned long long *tmp) |
| 445 | { |
| 446 | *tmp = memparse(buf, &buf); |
| 447 | if (*buf != '\0') |
| 448 | return -EINVAL; |
| 449 | |
| 450 | /* |
| 451 | * Round up the value to the closest page size |
| 452 | */ |
| 453 | *tmp = ((*tmp + PAGE_SIZE - 1) >> PAGE_SHIFT) << PAGE_SHIFT; |
| 454 | return 0; |
| 455 | } |
| 456 | |
| 457 | static ssize_t mem_cgroup_read(struct cgroup *cont, |
| 458 | struct cftype *cft, struct file *file, |
| 459 | char __user *userbuf, size_t nbytes, loff_t *ppos) |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 460 | { |
| 461 | return res_counter_read(&mem_cgroup_from_cont(cont)->res, |
Balbir Singh | 0eea103 | 2008-02-07 00:13:57 -0800 | [diff] [blame] | 462 | cft->private, userbuf, nbytes, ppos, |
| 463 | NULL); |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 464 | } |
| 465 | |
| 466 | static ssize_t mem_cgroup_write(struct cgroup *cont, struct cftype *cft, |
| 467 | struct file *file, const char __user *userbuf, |
| 468 | size_t nbytes, loff_t *ppos) |
| 469 | { |
| 470 | return res_counter_write(&mem_cgroup_from_cont(cont)->res, |
Balbir Singh | 0eea103 | 2008-02-07 00:13:57 -0800 | [diff] [blame] | 471 | cft->private, userbuf, nbytes, ppos, |
| 472 | mem_cgroup_write_strategy); |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 473 | } |
| 474 | |
Balbir Singh | 8697d33 | 2008-02-07 00:13:59 -0800 | [diff] [blame] | 475 | static ssize_t mem_control_type_write(struct cgroup *cont, |
| 476 | struct cftype *cft, struct file *file, |
| 477 | const char __user *userbuf, |
| 478 | size_t nbytes, loff_t *pos) |
| 479 | { |
| 480 | int ret; |
| 481 | char *buf, *end; |
| 482 | unsigned long tmp; |
| 483 | struct mem_cgroup *mem; |
| 484 | |
| 485 | mem = mem_cgroup_from_cont(cont); |
| 486 | buf = kmalloc(nbytes + 1, GFP_KERNEL); |
| 487 | ret = -ENOMEM; |
| 488 | if (buf == NULL) |
| 489 | goto out; |
| 490 | |
| 491 | buf[nbytes] = 0; |
| 492 | ret = -EFAULT; |
| 493 | if (copy_from_user(buf, userbuf, nbytes)) |
| 494 | goto out_free; |
| 495 | |
| 496 | ret = -EINVAL; |
| 497 | tmp = simple_strtoul(buf, &end, 10); |
| 498 | if (*end != '\0') |
| 499 | goto out_free; |
| 500 | |
| 501 | if (tmp <= MEM_CGROUP_TYPE_UNSPEC || tmp >= MEM_CGROUP_TYPE_MAX) |
| 502 | goto out_free; |
| 503 | |
| 504 | mem->control_type = tmp; |
| 505 | ret = nbytes; |
| 506 | out_free: |
| 507 | kfree(buf); |
| 508 | out: |
| 509 | return ret; |
| 510 | } |
| 511 | |
| 512 | static ssize_t mem_control_type_read(struct cgroup *cont, |
| 513 | struct cftype *cft, |
| 514 | struct file *file, char __user *userbuf, |
| 515 | size_t nbytes, loff_t *ppos) |
| 516 | { |
| 517 | unsigned long val; |
| 518 | char buf[64], *s; |
| 519 | struct mem_cgroup *mem; |
| 520 | |
| 521 | mem = mem_cgroup_from_cont(cont); |
| 522 | s = buf; |
| 523 | val = mem->control_type; |
| 524 | s += sprintf(s, "%lu\n", val); |
| 525 | return simple_read_from_buffer((void __user *)userbuf, nbytes, |
| 526 | ppos, buf, s - buf); |
| 527 | } |
| 528 | |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 529 | static struct cftype mem_cgroup_files[] = { |
| 530 | { |
Balbir Singh | 0eea103 | 2008-02-07 00:13:57 -0800 | [diff] [blame] | 531 | .name = "usage_in_bytes", |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 532 | .private = RES_USAGE, |
| 533 | .read = mem_cgroup_read, |
| 534 | }, |
| 535 | { |
Balbir Singh | 0eea103 | 2008-02-07 00:13:57 -0800 | [diff] [blame] | 536 | .name = "limit_in_bytes", |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 537 | .private = RES_LIMIT, |
| 538 | .write = mem_cgroup_write, |
| 539 | .read = mem_cgroup_read, |
| 540 | }, |
| 541 | { |
| 542 | .name = "failcnt", |
| 543 | .private = RES_FAILCNT, |
| 544 | .read = mem_cgroup_read, |
| 545 | }, |
Balbir Singh | 8697d33 | 2008-02-07 00:13:59 -0800 | [diff] [blame] | 546 | { |
| 547 | .name = "control_type", |
| 548 | .write = mem_control_type_write, |
| 549 | .read = mem_control_type_read, |
| 550 | }, |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 551 | }; |
| 552 | |
Pavel Emelianov | 78fb746 | 2008-02-07 00:13:51 -0800 | [diff] [blame] | 553 | static struct mem_cgroup init_mem_cgroup; |
| 554 | |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 555 | static struct cgroup_subsys_state * |
| 556 | mem_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cont) |
| 557 | { |
| 558 | struct mem_cgroup *mem; |
| 559 | |
Pavel Emelianov | 78fb746 | 2008-02-07 00:13:51 -0800 | [diff] [blame] | 560 | if (unlikely((cont->parent) == NULL)) { |
| 561 | mem = &init_mem_cgroup; |
| 562 | init_mm.mem_cgroup = mem; |
| 563 | } else |
| 564 | mem = kzalloc(sizeof(struct mem_cgroup), GFP_KERNEL); |
| 565 | |
| 566 | if (mem == NULL) |
| 567 | return NULL; |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 568 | |
| 569 | res_counter_init(&mem->res); |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 570 | INIT_LIST_HEAD(&mem->active_list); |
| 571 | INIT_LIST_HEAD(&mem->inactive_list); |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 572 | spin_lock_init(&mem->lru_lock); |
Balbir Singh | 8697d33 | 2008-02-07 00:13:59 -0800 | [diff] [blame] | 573 | mem->control_type = MEM_CGROUP_TYPE_ALL; |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 574 | return &mem->css; |
| 575 | } |
| 576 | |
| 577 | static void mem_cgroup_destroy(struct cgroup_subsys *ss, |
| 578 | struct cgroup *cont) |
| 579 | { |
| 580 | kfree(mem_cgroup_from_cont(cont)); |
| 581 | } |
| 582 | |
| 583 | static int mem_cgroup_populate(struct cgroup_subsys *ss, |
| 584 | struct cgroup *cont) |
| 585 | { |
| 586 | return cgroup_add_files(cont, ss, mem_cgroup_files, |
| 587 | ARRAY_SIZE(mem_cgroup_files)); |
| 588 | } |
| 589 | |
Balbir Singh | 67e465a | 2008-02-07 00:13:54 -0800 | [diff] [blame] | 590 | static void mem_cgroup_move_task(struct cgroup_subsys *ss, |
| 591 | struct cgroup *cont, |
| 592 | struct cgroup *old_cont, |
| 593 | struct task_struct *p) |
| 594 | { |
| 595 | struct mm_struct *mm; |
| 596 | struct mem_cgroup *mem, *old_mem; |
| 597 | |
| 598 | mm = get_task_mm(p); |
| 599 | if (mm == NULL) |
| 600 | return; |
| 601 | |
| 602 | mem = mem_cgroup_from_cont(cont); |
| 603 | old_mem = mem_cgroup_from_cont(old_cont); |
| 604 | |
| 605 | if (mem == old_mem) |
| 606 | goto out; |
| 607 | |
| 608 | /* |
| 609 | * Only thread group leaders are allowed to migrate, the mm_struct is |
| 610 | * in effect owned by the leader |
| 611 | */ |
| 612 | if (p->tgid != p->pid) |
| 613 | goto out; |
| 614 | |
| 615 | css_get(&mem->css); |
| 616 | rcu_assign_pointer(mm->mem_cgroup, mem); |
| 617 | css_put(&old_mem->css); |
| 618 | |
| 619 | out: |
| 620 | mmput(mm); |
| 621 | return; |
| 622 | } |
| 623 | |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 624 | struct cgroup_subsys mem_cgroup_subsys = { |
| 625 | .name = "memory", |
| 626 | .subsys_id = mem_cgroup_subsys_id, |
| 627 | .create = mem_cgroup_create, |
| 628 | .destroy = mem_cgroup_destroy, |
| 629 | .populate = mem_cgroup_populate, |
Balbir Singh | 67e465a | 2008-02-07 00:13:54 -0800 | [diff] [blame] | 630 | .attach = mem_cgroup_move_task, |
Pavel Emelianov | 78fb746 | 2008-02-07 00:13:51 -0800 | [diff] [blame] | 631 | .early_init = 1, |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 632 | }; |