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>
7 #include <linux/slab.h>
8 #include <linux/memory.h>
9 #include <linux/vmalloc.h>
10 #include <linux/cgroup.h>
11 #include <linux/swapops.h>
14 __init_page_cgroup(struct page_cgroup *pc, unsigned long pfn)
17 pc->mem_cgroup = NULL;
18 pc->page = pfn_to_page(pfn);
19 INIT_LIST_HEAD(&pc->lru);
21 static unsigned long total_usage;
23 #if !defined(CONFIG_SPARSEMEM)
26 void __meminit pgdat_page_cgroup_init(struct pglist_data *pgdat)
28 pgdat->node_page_cgroup = NULL;
31 struct page_cgroup *lookup_page_cgroup(struct page *page)
33 unsigned long pfn = page_to_pfn(page);
35 struct page_cgroup *base;
37 base = NODE_DATA(page_to_nid(page))->node_page_cgroup;
41 offset = pfn - NODE_DATA(page_to_nid(page))->node_start_pfn;
45 static int __init alloc_node_page_cgroup(int nid)
47 struct page_cgroup *base, *pc;
48 unsigned long table_size;
49 unsigned long start_pfn, nr_pages, index;
51 start_pfn = NODE_DATA(nid)->node_start_pfn;
52 nr_pages = NODE_DATA(nid)->node_spanned_pages;
57 table_size = sizeof(struct page_cgroup) * nr_pages;
59 base = __alloc_bootmem_node_nopanic(NODE_DATA(nid),
60 table_size, PAGE_SIZE, __pa(MAX_DMA_ADDRESS));
63 for (index = 0; index < nr_pages; index++) {
65 __init_page_cgroup(pc, start_pfn + index);
67 NODE_DATA(nid)->node_page_cgroup = base;
68 total_usage += table_size;
72 void __init page_cgroup_init(void)
77 if (mem_cgroup_disabled())
80 for_each_online_node(nid) {
81 fail = alloc_node_page_cgroup(nid);
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"
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");
95 #else /* CONFIG_FLAT_NODE_MEM_MAP */
97 struct page_cgroup *lookup_page_cgroup(struct page *page)
99 unsigned long pfn = page_to_pfn(page);
100 struct mem_section *section = __pfn_to_section(pfn);
102 return section->page_cgroup + pfn;
105 /* __alloc_bootmem...() is protected by !slab_available() */
106 static int __init_refok init_section_page_cgroup(unsigned long pfn)
108 struct mem_section *section = __pfn_to_section(pfn);
109 struct page_cgroup *base, *pc;
110 unsigned long table_size;
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()) {
117 base = kmalloc_node(table_size,
118 GFP_KERNEL | __GFP_NOWARN, nid);
120 base = vmalloc_node(table_size, nid);
122 base = __alloc_bootmem_node_nopanic(NODE_DATA(nid),
124 PAGE_SIZE, __pa(MAX_DMA_ADDRESS));
128 * We don't have to allocate page_cgroup again, but
129 * address of memmap may be changed. So, we have to initialize
132 base = section->page_cgroup + pfn;
134 /* check address of memmap is changed or not. */
135 if (base->page == pfn_to_page(pfn))
140 printk(KERN_ERR "page cgroup allocation failure\n");
144 for (index = 0; index < PAGES_PER_SECTION; index++) {
146 __init_page_cgroup(pc, pfn + index);
149 section->page_cgroup = base - pfn;
150 total_usage += table_size;
153 #ifdef CONFIG_MEMORY_HOTPLUG
154 void __free_page_cgroup(unsigned long pfn)
156 struct mem_section *ms;
157 struct page_cgroup *base;
159 ms = __pfn_to_section(pfn);
160 if (!ms || !ms->page_cgroup)
162 base = ms->page_cgroup + pfn;
163 if (is_vmalloc_addr(base)) {
165 ms->page_cgroup = NULL;
167 struct page *page = virt_to_page(base);
168 if (!PageReserved(page)) { /* Is bootmem ? */
170 ms->page_cgroup = NULL;
175 int __meminit online_page_cgroup(unsigned long start_pfn,
176 unsigned long nr_pages,
179 unsigned long start, end, pfn;
182 start = start_pfn & ~(PAGES_PER_SECTION - 1);
183 end = ALIGN(start_pfn + nr_pages, PAGES_PER_SECTION);
185 for (pfn = start; !fail && pfn < end; pfn += PAGES_PER_SECTION) {
186 if (!pfn_present(pfn))
188 fail = init_section_page_cgroup(pfn);
194 for (pfn = start; pfn < end; pfn += PAGES_PER_SECTION)
195 __free_page_cgroup(pfn);
200 int __meminit offline_page_cgroup(unsigned long start_pfn,
201 unsigned long nr_pages, int nid)
203 unsigned long start, end, pfn;
205 start = start_pfn & ~(PAGES_PER_SECTION - 1);
206 end = ALIGN(start_pfn + nr_pages, PAGES_PER_SECTION);
208 for (pfn = start; pfn < end; pfn += PAGES_PER_SECTION)
209 __free_page_cgroup(pfn);
214 static int __meminit page_cgroup_callback(struct notifier_block *self,
215 unsigned long action, void *arg)
217 struct memory_notify *mn = arg;
220 case MEM_GOING_ONLINE:
221 ret = online_page_cgroup(mn->start_pfn,
222 mn->nr_pages, mn->status_change_nid);
225 offline_page_cgroup(mn->start_pfn,
226 mn->nr_pages, mn->status_change_nid);
228 case MEM_CANCEL_ONLINE:
229 case MEM_GOING_OFFLINE:
232 case MEM_CANCEL_OFFLINE:
237 ret = notifier_from_errno(ret);
246 void __init page_cgroup_init(void)
251 if (mem_cgroup_disabled())
254 for (pfn = 0; !fail && pfn < max_pfn; pfn += PAGES_PER_SECTION) {
255 if (!pfn_present(pfn))
257 fail = init_section_page_cgroup(pfn);
260 printk(KERN_CRIT "try cgroup_disable=memory boot option\n");
261 panic("Out of memory");
263 hotplug_memory_notifier(page_cgroup_callback, 0);
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"
270 void __meminit pgdat_page_cgroup_init(struct pglist_data *pgdat)
278 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
280 static DEFINE_MUTEX(swap_cgroup_mutex);
281 struct swap_cgroup_ctrl {
283 unsigned long length;
286 struct swap_cgroup_ctrl swap_cgroup_ctrl[MAX_SWAPFILES];
291 #define SC_PER_PAGE (PAGE_SIZE/sizeof(struct swap_cgroup))
292 #define SC_POS_MASK (SC_PER_PAGE - 1)
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.
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".
305 * TODO: we can push these buffers out to HIGHMEM.
309 * allocate buffer for swap_cgroup.
311 static int swap_cgroup_prepare(int type)
314 struct swap_cgroup_ctrl *ctrl;
315 unsigned long idx, max;
317 if (!do_swap_account)
319 ctrl = &swap_cgroup_ctrl[type];
321 for (idx = 0; idx < ctrl->length; idx++) {
322 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
324 goto not_enough_page;
325 ctrl->map[idx] = page;
330 for (idx = 0; idx < max; idx++)
331 __free_page(ctrl->map[idx]);
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
341 * Returns old value at success, 0 at failure.
342 * (Of course, old value can be 0.)
344 unsigned short swap_cgroup_record(swp_entry_t ent, unsigned short id)
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;
355 if (!do_swap_account)
358 ctrl = &swap_cgroup_ctrl[type];
360 mappage = ctrl->map[idx];
361 sc = page_address(mappage);
370 * lookup_swap_cgroup - lookup mem_cgroup tied to swap entry
371 * @ent: swap entry to be looked up.
373 * Returns CSS ID of mem_cgroup at success. 0 at failure. (0 is invalid ID)
375 unsigned short lookup_swap_cgroup(swp_entry_t ent)
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;
386 if (!do_swap_account)
389 ctrl = &swap_cgroup_ctrl[type];
390 mappage = ctrl->map[idx];
391 sc = page_address(mappage);
397 int swap_cgroup_swapon(int type, unsigned long max_pages)
400 unsigned long array_size;
401 unsigned long length;
402 struct swap_cgroup_ctrl *ctrl;
404 if (!do_swap_account)
407 length = ((max_pages/SC_PER_PAGE) + 1);
408 array_size = length * sizeof(void *);
410 array = vmalloc(array_size);
414 memset(array, 0, array_size);
415 ctrl = &swap_cgroup_ctrl[type];
416 mutex_lock(&swap_cgroup_mutex);
417 ctrl->length = length;
419 if (swap_cgroup_prepare(type)) {
420 /* memory shortage */
424 mutex_unlock(&swap_cgroup_mutex);
427 mutex_unlock(&swap_cgroup_mutex);
431 printk(KERN_INFO "couldn't allocate enough memory for swap_cgroup.\n");
433 "swap_cgroup can be disabled by noswapaccount boot option\n");
437 void swap_cgroup_swapoff(int type)
440 struct swap_cgroup_ctrl *ctrl;
442 if (!do_swap_account)
445 mutex_lock(&swap_cgroup_mutex);
446 ctrl = &swap_cgroup_ctrl[type];
448 for (i = 0; i < ctrl->length; i++) {
449 struct page *page = ctrl->map[i];
457 mutex_unlock(&swap_cgroup_mutex);