Commit | Line | Data |
---|---|---|
52d4b9ac KH |
1 | #include <linux/mm.h> |
2 | #include <linux/mmzone.h> | |
3 | #include <linux/bootmem.h> | |
4 | #include <linux/bit_spinlock.h> | |
5 | #include <linux/page_cgroup.h> | |
6 | #include <linux/hash.h> | |
94b6da5a | 7 | #include <linux/slab.h> |
52d4b9ac | 8 | #include <linux/memory.h> |
4c821042 | 9 | #include <linux/vmalloc.h> |
94b6da5a | 10 | #include <linux/cgroup.h> |
27a7faa0 | 11 | #include <linux/swapops.h> |
52d4b9ac KH |
12 | |
13 | static void __meminit | |
14 | __init_page_cgroup(struct page_cgroup *pc, unsigned long pfn) | |
15 | { | |
16 | pc->flags = 0; | |
17 | pc->mem_cgroup = NULL; | |
18 | pc->page = pfn_to_page(pfn); | |
08e552c6 | 19 | INIT_LIST_HEAD(&pc->lru); |
52d4b9ac KH |
20 | } |
21 | static unsigned long total_usage; | |
22 | ||
23 | #if !defined(CONFIG_SPARSEMEM) | |
24 | ||
25 | ||
31168481 | 26 | void __meminit pgdat_page_cgroup_init(struct pglist_data *pgdat) |
52d4b9ac KH |
27 | { |
28 | pgdat->node_page_cgroup = NULL; | |
29 | } | |
30 | ||
31 | struct page_cgroup *lookup_page_cgroup(struct page *page) | |
32 | { | |
33 | unsigned long pfn = page_to_pfn(page); | |
34 | unsigned long offset; | |
35 | struct page_cgroup *base; | |
36 | ||
37 | base = NODE_DATA(page_to_nid(page))->node_page_cgroup; | |
38 | if (unlikely(!base)) | |
39 | return NULL; | |
40 | ||
41 | offset = pfn - NODE_DATA(page_to_nid(page))->node_start_pfn; | |
42 | return base + offset; | |
43 | } | |
44 | ||
45 | static int __init alloc_node_page_cgroup(int nid) | |
46 | { | |
47 | struct page_cgroup *base, *pc; | |
48 | unsigned long table_size; | |
49 | unsigned long start_pfn, nr_pages, index; | |
50 | ||
51 | start_pfn = NODE_DATA(nid)->node_start_pfn; | |
52 | nr_pages = NODE_DATA(nid)->node_spanned_pages; | |
53 | ||
653d22c0 KH |
54 | if (!nr_pages) |
55 | return 0; | |
56 | ||
52d4b9ac KH |
57 | table_size = sizeof(struct page_cgroup) * nr_pages; |
58 | ||
59 | base = __alloc_bootmem_node_nopanic(NODE_DATA(nid), | |
60 | table_size, PAGE_SIZE, __pa(MAX_DMA_ADDRESS)); | |
61 | if (!base) | |
62 | return -ENOMEM; | |
63 | for (index = 0; index < nr_pages; index++) { | |
64 | pc = base + index; | |
65 | __init_page_cgroup(pc, start_pfn + index); | |
66 | } | |
67 | NODE_DATA(nid)->node_page_cgroup = base; | |
68 | total_usage += table_size; | |
69 | return 0; | |
70 | } | |
71 | ||
72 | void __init page_cgroup_init(void) | |
73 | { | |
74 | ||
75 | int nid, fail; | |
76 | ||
f8d66542 | 77 | if (mem_cgroup_disabled()) |
94b6da5a KH |
78 | return; |
79 | ||
52d4b9ac KH |
80 | for_each_online_node(nid) { |
81 | fail = alloc_node_page_cgroup(nid); | |
82 | if (fail) | |
83 | goto fail; | |
84 | } | |
85 | printk(KERN_INFO "allocated %ld bytes of page_cgroup\n", total_usage); | |
86 | printk(KERN_INFO "please try cgroup_disable=memory option if you" | |
87 | " don't want\n"); | |
88 | return; | |
89 | fail: | |
90 | printk(KERN_CRIT "allocation of page_cgroup was failed.\n"); | |
91 | printk(KERN_CRIT "please try cgroup_disable=memory boot option\n"); | |
92 | panic("Out of memory"); | |
93 | } | |
94 | ||
95 | #else /* CONFIG_FLAT_NODE_MEM_MAP */ | |
96 | ||
97 | struct page_cgroup *lookup_page_cgroup(struct page *page) | |
98 | { | |
99 | unsigned long pfn = page_to_pfn(page); | |
100 | struct mem_section *section = __pfn_to_section(pfn); | |
101 | ||
102 | return section->page_cgroup + pfn; | |
103 | } | |
104 | ||
31168481 | 105 | /* __alloc_bootmem...() is protected by !slab_available() */ |
feb16694 | 106 | static int __init_refok init_section_page_cgroup(unsigned long pfn) |
52d4b9ac | 107 | { |
0753b0ef | 108 | struct mem_section *section = __pfn_to_section(pfn); |
52d4b9ac KH |
109 | struct page_cgroup *base, *pc; |
110 | unsigned long table_size; | |
111 | int nid, index; | |
112 | ||
dc19f9db KH |
113 | if (!section->page_cgroup) { |
114 | nid = page_to_nid(pfn_to_page(pfn)); | |
115 | table_size = sizeof(struct page_cgroup) * PAGES_PER_SECTION; | |
116 | if (slab_is_available()) { | |
2e9c2372 KH |
117 | base = kmalloc_node(table_size, |
118 | GFP_KERNEL | __GFP_NOWARN, nid); | |
dc19f9db KH |
119 | if (!base) |
120 | base = vmalloc_node(table_size, nid); | |
121 | } else { | |
122 | base = __alloc_bootmem_node_nopanic(NODE_DATA(nid), | |
123 | table_size, | |
94b6da5a | 124 | PAGE_SIZE, __pa(MAX_DMA_ADDRESS)); |
dc19f9db KH |
125 | } |
126 | } else { | |
127 | /* | |
128 | * We don't have to allocate page_cgroup again, but | |
129 | * address of memmap may be changed. So, we have to initialize | |
130 | * again. | |
131 | */ | |
132 | base = section->page_cgroup + pfn; | |
133 | table_size = 0; | |
134 | /* check address of memmap is changed or not. */ | |
135 | if (base->page == pfn_to_page(pfn)) | |
136 | return 0; | |
94b6da5a | 137 | } |
52d4b9ac KH |
138 | |
139 | if (!base) { | |
140 | printk(KERN_ERR "page cgroup allocation failure\n"); | |
141 | return -ENOMEM; | |
142 | } | |
143 | ||
144 | for (index = 0; index < PAGES_PER_SECTION; index++) { | |
145 | pc = base + index; | |
146 | __init_page_cgroup(pc, pfn + index); | |
147 | } | |
148 | ||
52d4b9ac KH |
149 | section->page_cgroup = base - pfn; |
150 | total_usage += table_size; | |
151 | return 0; | |
152 | } | |
153 | #ifdef CONFIG_MEMORY_HOTPLUG | |
154 | void __free_page_cgroup(unsigned long pfn) | |
155 | { | |
156 | struct mem_section *ms; | |
157 | struct page_cgroup *base; | |
158 | ||
159 | ms = __pfn_to_section(pfn); | |
160 | if (!ms || !ms->page_cgroup) | |
161 | return; | |
162 | base = ms->page_cgroup + pfn; | |
94b6da5a | 163 | if (is_vmalloc_addr(base)) { |
52d4b9ac | 164 | vfree(base); |
94b6da5a KH |
165 | ms->page_cgroup = NULL; |
166 | } else { | |
167 | struct page *page = virt_to_page(base); | |
168 | if (!PageReserved(page)) { /* Is bootmem ? */ | |
169 | kfree(base); | |
170 | ms->page_cgroup = NULL; | |
171 | } | |
172 | } | |
52d4b9ac KH |
173 | } |
174 | ||
31168481 | 175 | int __meminit online_page_cgroup(unsigned long start_pfn, |
52d4b9ac KH |
176 | unsigned long nr_pages, |
177 | int nid) | |
178 | { | |
179 | unsigned long start, end, pfn; | |
180 | int fail = 0; | |
181 | ||
33c5d3d6 | 182 | start = start_pfn & ~(PAGES_PER_SECTION - 1); |
52d4b9ac KH |
183 | end = ALIGN(start_pfn + nr_pages, PAGES_PER_SECTION); |
184 | ||
185 | for (pfn = start; !fail && pfn < end; pfn += PAGES_PER_SECTION) { | |
186 | if (!pfn_present(pfn)) | |
187 | continue; | |
188 | fail = init_section_page_cgroup(pfn); | |
189 | } | |
190 | if (!fail) | |
191 | return 0; | |
192 | ||
193 | /* rollback */ | |
194 | for (pfn = start; pfn < end; pfn += PAGES_PER_SECTION) | |
195 | __free_page_cgroup(pfn); | |
196 | ||
197 | return -ENOMEM; | |
198 | } | |
199 | ||
31168481 | 200 | int __meminit offline_page_cgroup(unsigned long start_pfn, |
52d4b9ac KH |
201 | unsigned long nr_pages, int nid) |
202 | { | |
203 | unsigned long start, end, pfn; | |
204 | ||
33c5d3d6 | 205 | start = start_pfn & ~(PAGES_PER_SECTION - 1); |
52d4b9ac KH |
206 | end = ALIGN(start_pfn + nr_pages, PAGES_PER_SECTION); |
207 | ||
208 | for (pfn = start; pfn < end; pfn += PAGES_PER_SECTION) | |
209 | __free_page_cgroup(pfn); | |
210 | return 0; | |
211 | ||
212 | } | |
213 | ||
31168481 | 214 | static int __meminit page_cgroup_callback(struct notifier_block *self, |
52d4b9ac KH |
215 | unsigned long action, void *arg) |
216 | { | |
217 | struct memory_notify *mn = arg; | |
218 | int ret = 0; | |
219 | switch (action) { | |
220 | case MEM_GOING_ONLINE: | |
221 | ret = online_page_cgroup(mn->start_pfn, | |
222 | mn->nr_pages, mn->status_change_nid); | |
223 | break; | |
52d4b9ac KH |
224 | case MEM_OFFLINE: |
225 | offline_page_cgroup(mn->start_pfn, | |
226 | mn->nr_pages, mn->status_change_nid); | |
227 | break; | |
dc19f9db | 228 | case MEM_CANCEL_ONLINE: |
52d4b9ac KH |
229 | case MEM_GOING_OFFLINE: |
230 | break; | |
231 | case MEM_ONLINE: | |
232 | case MEM_CANCEL_OFFLINE: | |
233 | break; | |
234 | } | |
dc19f9db KH |
235 | |
236 | if (ret) | |
237 | ret = notifier_from_errno(ret); | |
238 | else | |
239 | ret = NOTIFY_OK; | |
240 | ||
52d4b9ac KH |
241 | return ret; |
242 | } | |
243 | ||
244 | #endif | |
245 | ||
246 | void __init page_cgroup_init(void) | |
247 | { | |
248 | unsigned long pfn; | |
249 | int fail = 0; | |
250 | ||
f8d66542 | 251 | if (mem_cgroup_disabled()) |
94b6da5a KH |
252 | return; |
253 | ||
52d4b9ac KH |
254 | for (pfn = 0; !fail && pfn < max_pfn; pfn += PAGES_PER_SECTION) { |
255 | if (!pfn_present(pfn)) | |
256 | continue; | |
257 | fail = init_section_page_cgroup(pfn); | |
258 | } | |
259 | if (fail) { | |
260 | printk(KERN_CRIT "try cgroup_disable=memory boot option\n"); | |
261 | panic("Out of memory"); | |
262 | } else { | |
263 | hotplug_memory_notifier(page_cgroup_callback, 0); | |
264 | } | |
265 | printk(KERN_INFO "allocated %ld bytes of page_cgroup\n", total_usage); | |
266 | printk(KERN_INFO "please try cgroup_disable=memory option if you don't" | |
267 | " want\n"); | |
268 | } | |
269 | ||
31168481 | 270 | void __meminit pgdat_page_cgroup_init(struct pglist_data *pgdat) |
52d4b9ac KH |
271 | { |
272 | return; | |
273 | } | |
274 | ||
275 | #endif | |
27a7faa0 KH |
276 | |
277 | ||
278 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP | |
279 | ||
280 | static DEFINE_MUTEX(swap_cgroup_mutex); | |
281 | struct swap_cgroup_ctrl { | |
282 | struct page **map; | |
283 | unsigned long length; | |
284 | }; | |
285 | ||
286 | struct swap_cgroup_ctrl swap_cgroup_ctrl[MAX_SWAPFILES]; | |
287 | ||
27a7faa0 | 288 | struct swap_cgroup { |
a3b2d692 | 289 | unsigned short id; |
27a7faa0 KH |
290 | }; |
291 | #define SC_PER_PAGE (PAGE_SIZE/sizeof(struct swap_cgroup)) | |
292 | #define SC_POS_MASK (SC_PER_PAGE - 1) | |
293 | ||
294 | /* | |
295 | * SwapCgroup implements "lookup" and "exchange" operations. | |
296 | * In typical usage, this swap_cgroup is accessed via memcg's charge/uncharge | |
297 | * against SwapCache. At swap_free(), this is accessed directly from swap. | |
298 | * | |
299 | * This means, | |
300 | * - we have no race in "exchange" when we're accessed via SwapCache because | |
301 | * SwapCache(and its swp_entry) is under lock. | |
302 | * - When called via swap_free(), there is no user of this entry and no race. | |
303 | * Then, we don't need lock around "exchange". | |
304 | * | |
305 | * TODO: we can push these buffers out to HIGHMEM. | |
306 | */ | |
307 | ||
308 | /* | |
309 | * allocate buffer for swap_cgroup. | |
310 | */ | |
311 | static int swap_cgroup_prepare(int type) | |
312 | { | |
313 | struct page *page; | |
314 | struct swap_cgroup_ctrl *ctrl; | |
315 | unsigned long idx, max; | |
316 | ||
317 | if (!do_swap_account) | |
318 | return 0; | |
319 | ctrl = &swap_cgroup_ctrl[type]; | |
320 | ||
321 | for (idx = 0; idx < ctrl->length; idx++) { | |
322 | page = alloc_page(GFP_KERNEL | __GFP_ZERO); | |
323 | if (!page) | |
324 | goto not_enough_page; | |
325 | ctrl->map[idx] = page; | |
326 | } | |
327 | return 0; | |
328 | not_enough_page: | |
329 | max = idx; | |
330 | for (idx = 0; idx < max; idx++) | |
331 | __free_page(ctrl->map[idx]); | |
332 | ||
333 | return -ENOMEM; | |
334 | } | |
335 | ||
336 | /** | |
337 | * swap_cgroup_record - record mem_cgroup for this swp_entry. | |
338 | * @ent: swap entry to be recorded into | |
339 | * @mem: mem_cgroup to be recorded | |
340 | * | |
a3b2d692 KH |
341 | * Returns old value at success, 0 at failure. |
342 | * (Of course, old value can be 0.) | |
27a7faa0 | 343 | */ |
a3b2d692 | 344 | unsigned short swap_cgroup_record(swp_entry_t ent, unsigned short id) |
27a7faa0 KH |
345 | { |
346 | int type = swp_type(ent); | |
347 | unsigned long offset = swp_offset(ent); | |
348 | unsigned long idx = offset / SC_PER_PAGE; | |
349 | unsigned long pos = offset & SC_POS_MASK; | |
350 | struct swap_cgroup_ctrl *ctrl; | |
351 | struct page *mappage; | |
352 | struct swap_cgroup *sc; | |
a3b2d692 | 353 | unsigned short old; |
27a7faa0 KH |
354 | |
355 | if (!do_swap_account) | |
a3b2d692 | 356 | return 0; |
27a7faa0 KH |
357 | |
358 | ctrl = &swap_cgroup_ctrl[type]; | |
359 | ||
360 | mappage = ctrl->map[idx]; | |
361 | sc = page_address(mappage); | |
362 | sc += pos; | |
a3b2d692 KH |
363 | old = sc->id; |
364 | sc->id = id; | |
27a7faa0 KH |
365 | |
366 | return old; | |
367 | } | |
368 | ||
369 | /** | |
370 | * lookup_swap_cgroup - lookup mem_cgroup tied to swap entry | |
371 | * @ent: swap entry to be looked up. | |
372 | * | |
a3b2d692 | 373 | * Returns CSS ID of mem_cgroup at success. 0 at failure. (0 is invalid ID) |
27a7faa0 | 374 | */ |
a3b2d692 | 375 | unsigned short lookup_swap_cgroup(swp_entry_t ent) |
27a7faa0 KH |
376 | { |
377 | int type = swp_type(ent); | |
378 | unsigned long offset = swp_offset(ent); | |
379 | unsigned long idx = offset / SC_PER_PAGE; | |
380 | unsigned long pos = offset & SC_POS_MASK; | |
381 | struct swap_cgroup_ctrl *ctrl; | |
382 | struct page *mappage; | |
383 | struct swap_cgroup *sc; | |
a3b2d692 | 384 | unsigned short ret; |
27a7faa0 KH |
385 | |
386 | if (!do_swap_account) | |
a3b2d692 | 387 | return 0; |
27a7faa0 KH |
388 | |
389 | ctrl = &swap_cgroup_ctrl[type]; | |
390 | mappage = ctrl->map[idx]; | |
391 | sc = page_address(mappage); | |
392 | sc += pos; | |
a3b2d692 | 393 | ret = sc->id; |
27a7faa0 KH |
394 | return ret; |
395 | } | |
396 | ||
397 | int swap_cgroup_swapon(int type, unsigned long max_pages) | |
398 | { | |
399 | void *array; | |
400 | unsigned long array_size; | |
401 | unsigned long length; | |
402 | struct swap_cgroup_ctrl *ctrl; | |
403 | ||
404 | if (!do_swap_account) | |
405 | return 0; | |
406 | ||
407 | length = ((max_pages/SC_PER_PAGE) + 1); | |
408 | array_size = length * sizeof(void *); | |
409 | ||
410 | array = vmalloc(array_size); | |
411 | if (!array) | |
412 | goto nomem; | |
413 | ||
414 | memset(array, 0, array_size); | |
415 | ctrl = &swap_cgroup_ctrl[type]; | |
416 | mutex_lock(&swap_cgroup_mutex); | |
417 | ctrl->length = length; | |
418 | ctrl->map = array; | |
419 | if (swap_cgroup_prepare(type)) { | |
420 | /* memory shortage */ | |
421 | ctrl->map = NULL; | |
422 | ctrl->length = 0; | |
423 | vfree(array); | |
424 | mutex_unlock(&swap_cgroup_mutex); | |
425 | goto nomem; | |
426 | } | |
427 | mutex_unlock(&swap_cgroup_mutex); | |
428 | ||
27a7faa0 KH |
429 | return 0; |
430 | nomem: | |
431 | printk(KERN_INFO "couldn't allocate enough memory for swap_cgroup.\n"); | |
432 | printk(KERN_INFO | |
433 | "swap_cgroup can be disabled by noswapaccount boot option\n"); | |
434 | return -ENOMEM; | |
435 | } | |
436 | ||
437 | void swap_cgroup_swapoff(int type) | |
438 | { | |
439 | int i; | |
440 | struct swap_cgroup_ctrl *ctrl; | |
441 | ||
442 | if (!do_swap_account) | |
443 | return; | |
444 | ||
445 | mutex_lock(&swap_cgroup_mutex); | |
446 | ctrl = &swap_cgroup_ctrl[type]; | |
447 | if (ctrl->map) { | |
448 | for (i = 0; i < ctrl->length; i++) { | |
449 | struct page *page = ctrl->map[i]; | |
450 | if (page) | |
451 | __free_page(page); | |
452 | } | |
453 | vfree(ctrl->map); | |
454 | ctrl->map = NULL; | |
455 | ctrl->length = 0; | |
456 | } | |
457 | mutex_unlock(&swap_cgroup_mutex); | |
458 | } | |
459 | ||
460 | #endif |