blob: 6d757e3a872ad52adea55aabecee1ec571c3970d [file] [log] [blame]
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07001#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>
KAMEZAWA Hiroyuki94b6da52008-10-22 14:15:05 -07007#include <linux/slab.h>
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07008#include <linux/memory.h>
Paul Mundt4c8210422008-10-22 14:14:58 -07009#include <linux/vmalloc.h>
KAMEZAWA Hiroyuki94b6da52008-10-22 14:15:05 -070010#include <linux/cgroup.h>
KAMEZAWA Hiroyuki27a7faa2009-01-07 18:07:58 -080011#include <linux/swapops.h>
Catalin Marinas7952f982010-07-19 11:54:14 +010012#include <linux/kmemleak.h>
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -070013
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -070014static unsigned long total_usage;
15
16#if !defined(CONFIG_SPARSEMEM)
17
18
Al Viro31168482008-11-22 17:33:24 +000019void __meminit pgdat_page_cgroup_init(struct pglist_data *pgdat)
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -070020{
21 pgdat->node_page_cgroup = NULL;
22}
23
24struct page_cgroup *lookup_page_cgroup(struct page *page)
25{
26 unsigned long pfn = page_to_pfn(page);
27 unsigned long offset;
28 struct page_cgroup *base;
29
30 base = NODE_DATA(page_to_nid(page))->node_page_cgroup;
Johannes Weiner00c54c02012-01-12 17:18:40 -080031#ifdef CONFIG_DEBUG_VM
32 /*
33 * The sanity checks the page allocator does upon freeing a
34 * page can reach here before the page_cgroup arrays are
35 * allocated when feeding a range of pages to the allocator
36 * for the first time during bootup or memory hotplug.
37 */
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -070038 if (unlikely(!base))
39 return NULL;
Johannes Weiner00c54c02012-01-12 17:18:40 -080040#endif
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -070041 offset = pfn - NODE_DATA(page_to_nid(page))->node_start_pfn;
42 return base + offset;
43}
44
45static int __init alloc_node_page_cgroup(int nid)
46{
Johannes Weiner6b208e32012-01-12 17:18:18 -080047 struct page_cgroup *base;
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -070048 unsigned long table_size;
Johannes Weiner6b208e32012-01-12 17:18:18 -080049 unsigned long nr_pages;
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -070050
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -070051 nr_pages = NODE_DATA(nid)->node_spanned_pages;
KAMEZAWA Hiroyuki653d22c2008-12-09 13:14:20 -080052 if (!nr_pages)
53 return 0;
54
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -070055 table_size = sizeof(struct page_cgroup) * nr_pages;
KAMEZAWA Hiroyukica371c02009-06-12 10:33:53 +030056
57 base = __alloc_bootmem_node_nopanic(NODE_DATA(nid),
58 table_size, PAGE_SIZE, __pa(MAX_DMA_ADDRESS));
59 if (!base)
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -070060 return -ENOMEM;
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -070061 NODE_DATA(nid)->node_page_cgroup = base;
62 total_usage += table_size;
63 return 0;
64}
65
KAMEZAWA Hiroyukica371c02009-06-12 10:33:53 +030066void __init page_cgroup_init_flatmem(void)
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -070067{
68
69 int nid, fail;
70
Hirokazu Takahashif8d665422009-01-07 18:08:02 -080071 if (mem_cgroup_disabled())
KAMEZAWA Hiroyuki94b6da52008-10-22 14:15:05 -070072 return;
73
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -070074 for_each_online_node(nid) {
75 fail = alloc_node_page_cgroup(nid);
76 if (fail)
77 goto fail;
78 }
79 printk(KERN_INFO "allocated %ld bytes of page_cgroup\n", total_usage);
Randy Dunlap8ca739e2009-06-17 16:26:32 -070080 printk(KERN_INFO "please try 'cgroup_disable=memory' option if you"
81 " don't want memory cgroups\n");
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -070082 return;
83fail:
Randy Dunlap8ca739e2009-06-17 16:26:32 -070084 printk(KERN_CRIT "allocation of page_cgroup failed.\n");
85 printk(KERN_CRIT "please try 'cgroup_disable=memory' boot option\n");
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -070086 panic("Out of memory");
87}
88
89#else /* CONFIG_FLAT_NODE_MEM_MAP */
90
91struct page_cgroup *lookup_page_cgroup(struct page *page)
92{
93 unsigned long pfn = page_to_pfn(page);
94 struct mem_section *section = __pfn_to_section(pfn);
Johannes Weiner00c54c02012-01-12 17:18:40 -080095#ifdef CONFIG_DEBUG_VM
96 /*
97 * The sanity checks the page allocator does upon freeing a
98 * page can reach here before the page_cgroup arrays are
99 * allocated when feeding a range of pages to the allocator
100 * for the first time during bootup or memory hotplug.
101 */
Balbir Singhd69b0422009-06-17 16:26:34 -0700102 if (!section->page_cgroup)
103 return NULL;
Johannes Weiner00c54c02012-01-12 17:18:40 -0800104#endif
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -0700105 return section->page_cgroup + pfn;
106}
107
Namhyung Kim268433b2011-05-26 16:25:29 -0700108static void *__meminit alloc_page_cgroup(size_t size, int nid)
Michal Hockodde79e02011-03-23 16:42:40 -0700109{
Johannes Weiner6b208e32012-01-12 17:18:18 -0800110 gfp_t flags = GFP_KERNEL | __GFP_ZERO | __GFP_NOWARN;
Michal Hockodde79e02011-03-23 16:42:40 -0700111 void *addr = NULL;
112
Steven Rostedtff7ee932011-11-02 13:38:11 -0700113 addr = alloc_pages_exact_nid(nid, size, flags);
114 if (addr) {
115 kmemleak_alloc(addr, size, 1, flags);
Michal Hockodde79e02011-03-23 16:42:40 -0700116 return addr;
Steven Rostedtff7ee932011-11-02 13:38:11 -0700117 }
Michal Hockodde79e02011-03-23 16:42:40 -0700118
119 if (node_state(nid, N_HIGH_MEMORY))
Johannes Weiner6b208e32012-01-12 17:18:18 -0800120 addr = vzalloc_node(size, nid);
Michal Hockodde79e02011-03-23 16:42:40 -0700121 else
Johannes Weiner6b208e32012-01-12 17:18:18 -0800122 addr = vzalloc(size);
Michal Hockodde79e02011-03-23 16:42:40 -0700123
124 return addr;
125}
126
KAMEZAWA Hiroyuki37573e82011-06-15 15:08:42 -0700127static int __meminit init_section_page_cgroup(unsigned long pfn, int nid)
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -0700128{
Johannes Weiner6b3ae582011-03-23 16:42:30 -0700129 struct mem_section *section;
Johannes Weiner6b208e32012-01-12 17:18:18 -0800130 struct page_cgroup *base;
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -0700131 unsigned long table_size;
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -0700132
Johannes Weiner6b208e32012-01-12 17:18:18 -0800133 section = __pfn_to_section(pfn);
Johannes Weiner6b3ae582011-03-23 16:42:30 -0700134
135 if (section->page_cgroup)
136 return 0;
137
Johannes Weiner6b3ae582011-03-23 16:42:30 -0700138 table_size = sizeof(struct page_cgroup) * PAGES_PER_SECTION;
Michal Hockodde79e02011-03-23 16:42:40 -0700139 base = alloc_page_cgroup(table_size, nid);
140
Johannes Weiner6b3ae582011-03-23 16:42:30 -0700141 /*
142 * The value stored in section->page_cgroup is (base - pfn)
143 * and it does not point to the memory block allocated above,
144 * causing kmemleak false positives.
145 */
146 kmemleak_not_leak(base);
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -0700147
148 if (!base) {
149 printk(KERN_ERR "page cgroup allocation failure\n");
150 return -ENOMEM;
151 }
152
KAMEZAWA Hiroyuki37573e82011-06-15 15:08:42 -0700153 /*
154 * The passed "pfn" may not be aligned to SECTION. For the calculation
155 * we need to apply a mask.
156 */
157 pfn &= PAGE_SECTION_MASK;
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -0700158 section->page_cgroup = base - pfn;
159 total_usage += table_size;
160 return 0;
161}
162#ifdef CONFIG_MEMORY_HOTPLUG
Bob Liu0efc8eb2012-01-12 17:19:08 -0800163static void free_page_cgroup(void *addr)
164{
165 if (is_vmalloc_addr(addr)) {
166 vfree(addr);
167 } else {
168 struct page *page = virt_to_page(addr);
169 size_t table_size =
170 sizeof(struct page_cgroup) * PAGES_PER_SECTION;
171
172 BUG_ON(PageReserved(page));
173 free_pages_exact(addr, table_size);
174 }
175}
176
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -0700177void __free_page_cgroup(unsigned long pfn)
178{
179 struct mem_section *ms;
180 struct page_cgroup *base;
181
182 ms = __pfn_to_section(pfn);
183 if (!ms || !ms->page_cgroup)
184 return;
185 base = ms->page_cgroup + pfn;
Michal Hockodde79e02011-03-23 16:42:40 -0700186 free_page_cgroup(base);
187 ms->page_cgroup = NULL;
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -0700188}
189
Al Viro31168482008-11-22 17:33:24 +0000190int __meminit online_page_cgroup(unsigned long start_pfn,
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -0700191 unsigned long nr_pages,
192 int nid)
193{
194 unsigned long start, end, pfn;
195 int fail = 0;
196
Daniel Kiper1bb36fb2011-07-25 17:12:13 -0700197 start = SECTION_ALIGN_DOWN(start_pfn);
198 end = SECTION_ALIGN_UP(start_pfn + nr_pages);
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -0700199
KAMEZAWA Hiroyuki37573e82011-06-15 15:08:42 -0700200 if (nid == -1) {
201 /*
202 * In this case, "nid" already exists and contains valid memory.
203 * "start_pfn" passed to us is a pfn which is an arg for
204 * online__pages(), and start_pfn should exist.
205 */
206 nid = pfn_to_nid(start_pfn);
207 VM_BUG_ON(!node_state(nid, N_ONLINE));
208 }
209
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -0700210 for (pfn = start; !fail && pfn < end; pfn += PAGES_PER_SECTION) {
211 if (!pfn_present(pfn))
212 continue;
KAMEZAWA Hiroyuki37573e82011-06-15 15:08:42 -0700213 fail = init_section_page_cgroup(pfn, nid);
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -0700214 }
215 if (!fail)
216 return 0;
217
218 /* rollback */
219 for (pfn = start; pfn < end; pfn += PAGES_PER_SECTION)
220 __free_page_cgroup(pfn);
221
222 return -ENOMEM;
223}
224
Al Viro31168482008-11-22 17:33:24 +0000225int __meminit offline_page_cgroup(unsigned long start_pfn,
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -0700226 unsigned long nr_pages, int nid)
227{
228 unsigned long start, end, pfn;
229
Daniel Kiper1bb36fb2011-07-25 17:12:13 -0700230 start = SECTION_ALIGN_DOWN(start_pfn);
231 end = SECTION_ALIGN_UP(start_pfn + nr_pages);
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -0700232
233 for (pfn = start; pfn < end; pfn += PAGES_PER_SECTION)
234 __free_page_cgroup(pfn);
235 return 0;
236
237}
238
Al Viro31168482008-11-22 17:33:24 +0000239static int __meminit page_cgroup_callback(struct notifier_block *self,
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -0700240 unsigned long action, void *arg)
241{
242 struct memory_notify *mn = arg;
243 int ret = 0;
244 switch (action) {
245 case MEM_GOING_ONLINE:
246 ret = online_page_cgroup(mn->start_pfn,
247 mn->nr_pages, mn->status_change_nid);
248 break;
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -0700249 case MEM_OFFLINE:
250 offline_page_cgroup(mn->start_pfn,
251 mn->nr_pages, mn->status_change_nid);
252 break;
KAMEZAWA Hiroyukidc19f9d2008-12-01 13:13:48 -0800253 case MEM_CANCEL_ONLINE:
Wen Congyang7c72eb32012-12-11 16:00:49 -0800254 offline_page_cgroup(mn->start_pfn,
255 mn->nr_pages, mn->status_change_nid);
256 break;
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -0700257 case MEM_GOING_OFFLINE:
258 break;
259 case MEM_ONLINE:
260 case MEM_CANCEL_OFFLINE:
261 break;
262 }
KAMEZAWA Hiroyukidc19f9d2008-12-01 13:13:48 -0800263
Prarit Bhargava5fda1bd2011-03-22 16:30:49 -0700264 return notifier_from_errno(ret);
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -0700265}
266
267#endif
268
269void __init page_cgroup_init(void)
270{
271 unsigned long pfn;
KAMEZAWA Hiroyuki37573e82011-06-15 15:08:42 -0700272 int nid;
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -0700273
Hirokazu Takahashif8d665422009-01-07 18:08:02 -0800274 if (mem_cgroup_disabled())
KAMEZAWA Hiroyuki94b6da52008-10-22 14:15:05 -0700275 return;
276
Lai Jiangshan31aaea42012-12-12 13:51:27 -0800277 for_each_node_state(nid, N_MEMORY) {
KAMEZAWA Hiroyuki37573e82011-06-15 15:08:42 -0700278 unsigned long start_pfn, end_pfn;
279
280 start_pfn = node_start_pfn(nid);
281 end_pfn = node_end_pfn(nid);
282 /*
283 * start_pfn and end_pfn may not be aligned to SECTION and the
284 * page->flags of out of node pages are not initialized. So we
285 * scan [start_pfn, the biggest section's pfn < end_pfn) here.
286 */
287 for (pfn = start_pfn;
288 pfn < end_pfn;
289 pfn = ALIGN(pfn + 1, PAGES_PER_SECTION)) {
290
291 if (!pfn_valid(pfn))
292 continue;
293 /*
294 * Nodes's pfns can be overlapping.
295 * We know some arch can have a nodes layout such as
296 * -------------pfn-------------->
297 * N0 | N1 | N2 | N0 | N1 | N2|....
298 */
299 if (pfn_to_nid(pfn) != nid)
300 continue;
301 if (init_section_page_cgroup(pfn, nid))
302 goto oom;
303 }
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -0700304 }
KAMEZAWA Hiroyuki37573e82011-06-15 15:08:42 -0700305 hotplug_memory_notifier(page_cgroup_callback, 0);
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -0700306 printk(KERN_INFO "allocated %ld bytes of page_cgroup\n", total_usage);
KAMEZAWA Hiroyuki37573e82011-06-15 15:08:42 -0700307 printk(KERN_INFO "please try 'cgroup_disable=memory' option if you "
308 "don't want memory cgroups\n");
309 return;
310oom:
311 printk(KERN_CRIT "try 'cgroup_disable=memory' boot option\n");
312 panic("Out of memory");
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -0700313}
314
Al Viro31168482008-11-22 17:33:24 +0000315void __meminit pgdat_page_cgroup_init(struct pglist_data *pgdat)
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -0700316{
317 return;
318}
319
320#endif
KAMEZAWA Hiroyuki27a7faa2009-01-07 18:07:58 -0800321
322
Andrew Mortonc255a452012-07-31 16:43:02 -0700323#ifdef CONFIG_MEMCG_SWAP
KAMEZAWA Hiroyuki27a7faa2009-01-07 18:07:58 -0800324
325static DEFINE_MUTEX(swap_cgroup_mutex);
326struct swap_cgroup_ctrl {
327 struct page **map;
328 unsigned long length;
KAMEZAWA Hiroyukie9e58a42010-03-15 00:34:57 -0400329 spinlock_t lock;
KAMEZAWA Hiroyuki27a7faa2009-01-07 18:07:58 -0800330};
331
H Hartley Sweeten61600f52011-11-02 13:38:36 -0700332static struct swap_cgroup_ctrl swap_cgroup_ctrl[MAX_SWAPFILES];
KAMEZAWA Hiroyuki27a7faa2009-01-07 18:07:58 -0800333
KAMEZAWA Hiroyuki27a7faa2009-01-07 18:07:58 -0800334struct swap_cgroup {
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -0700335 unsigned short id;
KAMEZAWA Hiroyuki27a7faa2009-01-07 18:07:58 -0800336};
337#define SC_PER_PAGE (PAGE_SIZE/sizeof(struct swap_cgroup))
KAMEZAWA Hiroyuki27a7faa2009-01-07 18:07:58 -0800338
339/*
340 * SwapCgroup implements "lookup" and "exchange" operations.
341 * In typical usage, this swap_cgroup is accessed via memcg's charge/uncharge
342 * against SwapCache. At swap_free(), this is accessed directly from swap.
343 *
344 * This means,
345 * - we have no race in "exchange" when we're accessed via SwapCache because
346 * SwapCache(and its swp_entry) is under lock.
347 * - When called via swap_free(), there is no user of this entry and no race.
348 * Then, we don't need lock around "exchange".
349 *
350 * TODO: we can push these buffers out to HIGHMEM.
351 */
352
353/*
354 * allocate buffer for swap_cgroup.
355 */
356static int swap_cgroup_prepare(int type)
357{
358 struct page *page;
359 struct swap_cgroup_ctrl *ctrl;
360 unsigned long idx, max;
361
KAMEZAWA Hiroyuki27a7faa2009-01-07 18:07:58 -0800362 ctrl = &swap_cgroup_ctrl[type];
363
364 for (idx = 0; idx < ctrl->length; idx++) {
365 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
366 if (!page)
367 goto not_enough_page;
368 ctrl->map[idx] = page;
369 }
370 return 0;
371not_enough_page:
372 max = idx;
373 for (idx = 0; idx < max; idx++)
374 __free_page(ctrl->map[idx]);
375
376 return -ENOMEM;
377}
378
Bob Liu9fb4b7c2012-01-12 17:18:48 -0800379static struct swap_cgroup *lookup_swap_cgroup(swp_entry_t ent,
380 struct swap_cgroup_ctrl **ctrlp)
381{
382 pgoff_t offset = swp_offset(ent);
383 struct swap_cgroup_ctrl *ctrl;
384 struct page *mappage;
Hugh Dickinsc09ff082012-03-05 20:52:55 -0800385 struct swap_cgroup *sc;
Bob Liu9fb4b7c2012-01-12 17:18:48 -0800386
387 ctrl = &swap_cgroup_ctrl[swp_type(ent)];
388 if (ctrlp)
389 *ctrlp = ctrl;
390
391 mappage = ctrl->map[offset / SC_PER_PAGE];
Hugh Dickinsc09ff082012-03-05 20:52:55 -0800392 sc = page_address(mappage);
393 return sc + offset % SC_PER_PAGE;
Bob Liu9fb4b7c2012-01-12 17:18:48 -0800394}
395
KAMEZAWA Hiroyuki27a7faa2009-01-07 18:07:58 -0800396/**
Daisuke Nishimura02491442010-03-10 15:22:17 -0800397 * swap_cgroup_cmpxchg - cmpxchg mem_cgroup's id for this swp_entry.
Wanpeng Lidad75572012-06-20 12:53:01 -0700398 * @ent: swap entry to be cmpxchged
Daisuke Nishimura02491442010-03-10 15:22:17 -0800399 * @old: old id
400 * @new: new id
401 *
402 * Returns old id at success, 0 at failure.
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300403 * (There is no mem_cgroup using 0 as its id)
Daisuke Nishimura02491442010-03-10 15:22:17 -0800404 */
405unsigned short swap_cgroup_cmpxchg(swp_entry_t ent,
406 unsigned short old, unsigned short new)
407{
Daisuke Nishimura02491442010-03-10 15:22:17 -0800408 struct swap_cgroup_ctrl *ctrl;
Daisuke Nishimura02491442010-03-10 15:22:17 -0800409 struct swap_cgroup *sc;
KAMEZAWA Hiroyukie9e58a42010-03-15 00:34:57 -0400410 unsigned long flags;
411 unsigned short retval;
Daisuke Nishimura02491442010-03-10 15:22:17 -0800412
Bob Liu9fb4b7c2012-01-12 17:18:48 -0800413 sc = lookup_swap_cgroup(ent, &ctrl);
Daisuke Nishimura02491442010-03-10 15:22:17 -0800414
KAMEZAWA Hiroyukie9e58a42010-03-15 00:34:57 -0400415 spin_lock_irqsave(&ctrl->lock, flags);
416 retval = sc->id;
417 if (retval == old)
418 sc->id = new;
Daisuke Nishimura02491442010-03-10 15:22:17 -0800419 else
KAMEZAWA Hiroyukie9e58a42010-03-15 00:34:57 -0400420 retval = 0;
421 spin_unlock_irqrestore(&ctrl->lock, flags);
422 return retval;
Daisuke Nishimura02491442010-03-10 15:22:17 -0800423}
424
425/**
KAMEZAWA Hiroyuki27a7faa2009-01-07 18:07:58 -0800426 * swap_cgroup_record - record mem_cgroup for this swp_entry.
427 * @ent: swap entry to be recorded into
Wanpeng Lidad75572012-06-20 12:53:01 -0700428 * @id: mem_cgroup to be recorded
KAMEZAWA Hiroyuki27a7faa2009-01-07 18:07:58 -0800429 *
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -0700430 * Returns old value at success, 0 at failure.
431 * (Of course, old value can be 0.)
KAMEZAWA Hiroyuki27a7faa2009-01-07 18:07:58 -0800432 */
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -0700433unsigned short swap_cgroup_record(swp_entry_t ent, unsigned short id)
KAMEZAWA Hiroyuki27a7faa2009-01-07 18:07:58 -0800434{
KAMEZAWA Hiroyuki27a7faa2009-01-07 18:07:58 -0800435 struct swap_cgroup_ctrl *ctrl;
KAMEZAWA Hiroyuki27a7faa2009-01-07 18:07:58 -0800436 struct swap_cgroup *sc;
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -0700437 unsigned short old;
KAMEZAWA Hiroyukie9e58a42010-03-15 00:34:57 -0400438 unsigned long flags;
KAMEZAWA Hiroyuki27a7faa2009-01-07 18:07:58 -0800439
Bob Liu9fb4b7c2012-01-12 17:18:48 -0800440 sc = lookup_swap_cgroup(ent, &ctrl);
KAMEZAWA Hiroyuki27a7faa2009-01-07 18:07:58 -0800441
KAMEZAWA Hiroyukie9e58a42010-03-15 00:34:57 -0400442 spin_lock_irqsave(&ctrl->lock, flags);
443 old = sc->id;
444 sc->id = id;
445 spin_unlock_irqrestore(&ctrl->lock, flags);
KAMEZAWA Hiroyuki27a7faa2009-01-07 18:07:58 -0800446
447 return old;
448}
449
450/**
Bob Liu9fb4b7c2012-01-12 17:18:48 -0800451 * lookup_swap_cgroup_id - lookup mem_cgroup id tied to swap entry
KAMEZAWA Hiroyuki27a7faa2009-01-07 18:07:58 -0800452 * @ent: swap entry to be looked up.
453 *
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -0700454 * Returns CSS ID of mem_cgroup at success. 0 at failure. (0 is invalid ID)
KAMEZAWA Hiroyuki27a7faa2009-01-07 18:07:58 -0800455 */
Bob Liu9fb4b7c2012-01-12 17:18:48 -0800456unsigned short lookup_swap_cgroup_id(swp_entry_t ent)
KAMEZAWA Hiroyuki27a7faa2009-01-07 18:07:58 -0800457{
Bob Liu9fb4b7c2012-01-12 17:18:48 -0800458 return lookup_swap_cgroup(ent, NULL)->id;
KAMEZAWA Hiroyuki27a7faa2009-01-07 18:07:58 -0800459}
460
461int swap_cgroup_swapon(int type, unsigned long max_pages)
462{
463 void *array;
464 unsigned long array_size;
465 unsigned long length;
466 struct swap_cgroup_ctrl *ctrl;
467
468 if (!do_swap_account)
469 return 0;
470
Namhyung Kim33278f72011-05-26 16:25:30 -0700471 length = DIV_ROUND_UP(max_pages, SC_PER_PAGE);
KAMEZAWA Hiroyuki27a7faa2009-01-07 18:07:58 -0800472 array_size = length * sizeof(void *);
473
Joe Perches8c1fec12011-05-28 10:36:34 -0700474 array = vzalloc(array_size);
KAMEZAWA Hiroyuki27a7faa2009-01-07 18:07:58 -0800475 if (!array)
476 goto nomem;
477
KAMEZAWA Hiroyuki27a7faa2009-01-07 18:07:58 -0800478 ctrl = &swap_cgroup_ctrl[type];
479 mutex_lock(&swap_cgroup_mutex);
480 ctrl->length = length;
481 ctrl->map = array;
KAMEZAWA Hiroyukie9e58a42010-03-15 00:34:57 -0400482 spin_lock_init(&ctrl->lock);
KAMEZAWA Hiroyuki27a7faa2009-01-07 18:07:58 -0800483 if (swap_cgroup_prepare(type)) {
484 /* memory shortage */
485 ctrl->map = NULL;
486 ctrl->length = 0;
KAMEZAWA Hiroyuki27a7faa2009-01-07 18:07:58 -0800487 mutex_unlock(&swap_cgroup_mutex);
Namhyung Kim6a5b18d2011-05-26 16:25:31 -0700488 vfree(array);
KAMEZAWA Hiroyuki27a7faa2009-01-07 18:07:58 -0800489 goto nomem;
490 }
491 mutex_unlock(&swap_cgroup_mutex);
492
KAMEZAWA Hiroyuki27a7faa2009-01-07 18:07:58 -0800493 return 0;
494nomem:
495 printk(KERN_INFO "couldn't allocate enough memory for swap_cgroup.\n");
496 printk(KERN_INFO
WANG Cong00a66d22011-07-25 17:12:12 -0700497 "swap_cgroup can be disabled by swapaccount=0 boot option\n");
KAMEZAWA Hiroyuki27a7faa2009-01-07 18:07:58 -0800498 return -ENOMEM;
499}
500
501void swap_cgroup_swapoff(int type)
502{
Namhyung Kim6a5b18d2011-05-26 16:25:31 -0700503 struct page **map;
504 unsigned long i, length;
KAMEZAWA Hiroyuki27a7faa2009-01-07 18:07:58 -0800505 struct swap_cgroup_ctrl *ctrl;
506
507 if (!do_swap_account)
508 return;
509
510 mutex_lock(&swap_cgroup_mutex);
511 ctrl = &swap_cgroup_ctrl[type];
Namhyung Kim6a5b18d2011-05-26 16:25:31 -0700512 map = ctrl->map;
513 length = ctrl->length;
514 ctrl->map = NULL;
515 ctrl->length = 0;
516 mutex_unlock(&swap_cgroup_mutex);
517
518 if (map) {
519 for (i = 0; i < length; i++) {
520 struct page *page = map[i];
KAMEZAWA Hiroyuki27a7faa2009-01-07 18:07:58 -0800521 if (page)
522 __free_page(page);
523 }
Namhyung Kim6a5b18d2011-05-26 16:25:31 -0700524 vfree(map);
KAMEZAWA Hiroyuki27a7faa2009-01-07 18:07:58 -0800525 }
KAMEZAWA Hiroyuki27a7faa2009-01-07 18:07:58 -0800526}
527
528#endif