memcg: fix page_cgroup allocation
[linux-2.6.git] / mm / page_cgroup.c
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>
7 #include <linux/slab.h>
8 #include <linux/memory.h>
9 #include <linux/vmalloc.h>
10 #include <linux/cgroup.h>
11
12 static void __meminit
13 __init_page_cgroup(struct page_cgroup *pc, unsigned long pfn)
14 {
15         pc->flags = 0;
16         pc->mem_cgroup = NULL;
17         pc->page = pfn_to_page(pfn);
18 }
19 static unsigned long total_usage;
20
21 #if !defined(CONFIG_SPARSEMEM)
22
23
24 void __init pgdat_page_cgroup_init(struct pglist_data *pgdat)
25 {
26         pgdat->node_page_cgroup = NULL;
27 }
28
29 struct page_cgroup *lookup_page_cgroup(struct page *page)
30 {
31         unsigned long pfn = page_to_pfn(page);
32         unsigned long offset;
33         struct page_cgroup *base;
34
35         base = NODE_DATA(page_to_nid(page))->node_page_cgroup;
36         if (unlikely(!base))
37                 return NULL;
38
39         offset = pfn - NODE_DATA(page_to_nid(page))->node_start_pfn;
40         return base + offset;
41 }
42
43 static int __init alloc_node_page_cgroup(int nid)
44 {
45         struct page_cgroup *base, *pc;
46         unsigned long table_size;
47         unsigned long start_pfn, nr_pages, index;
48
49         start_pfn = NODE_DATA(nid)->node_start_pfn;
50         nr_pages = NODE_DATA(nid)->node_spanned_pages;
51
52         table_size = sizeof(struct page_cgroup) * nr_pages;
53
54         base = __alloc_bootmem_node_nopanic(NODE_DATA(nid),
55                         table_size, PAGE_SIZE, __pa(MAX_DMA_ADDRESS));
56         if (!base)
57                 return -ENOMEM;
58         for (index = 0; index < nr_pages; index++) {
59                 pc = base + index;
60                 __init_page_cgroup(pc, start_pfn + index);
61         }
62         NODE_DATA(nid)->node_page_cgroup = base;
63         total_usage += table_size;
64         return 0;
65 }
66
67 void __init page_cgroup_init(void)
68 {
69
70         int nid, fail;
71
72         if (mem_cgroup_subsys.disabled)
73                 return;
74
75         for_each_online_node(nid)  {
76                 fail = alloc_node_page_cgroup(nid);
77                 if (fail)
78                         goto fail;
79         }
80         printk(KERN_INFO "allocated %ld bytes of page_cgroup\n", total_usage);
81         printk(KERN_INFO "please try cgroup_disable=memory option if you"
82         " don't want\n");
83         return;
84 fail:
85         printk(KERN_CRIT "allocation of page_cgroup was failed.\n");
86         printk(KERN_CRIT "please try cgroup_disable=memory boot option\n");
87         panic("Out of memory");
88 }
89
90 #else /* CONFIG_FLAT_NODE_MEM_MAP */
91
92 struct page_cgroup *lookup_page_cgroup(struct page *page)
93 {
94         unsigned long pfn = page_to_pfn(page);
95         struct mem_section *section = __pfn_to_section(pfn);
96
97         return section->page_cgroup + pfn;
98 }
99
100 int __meminit init_section_page_cgroup(unsigned long pfn)
101 {
102         struct mem_section *section;
103         struct page_cgroup *base, *pc;
104         unsigned long table_size;
105         int nid, index;
106
107         section = __pfn_to_section(pfn);
108
109         if (section->page_cgroup)
110                 return 0;
111
112         nid = page_to_nid(pfn_to_page(pfn));
113
114         table_size = sizeof(struct page_cgroup) * PAGES_PER_SECTION;
115         if (slab_is_available()) {
116                 base = kmalloc_node(table_size, GFP_KERNEL, nid);
117                 if (!base)
118                         base = vmalloc_node(table_size, nid);
119         } else {
120                 base = __alloc_bootmem_node_nopanic(NODE_DATA(nid), table_size,
121                                 PAGE_SIZE, __pa(MAX_DMA_ADDRESS));
122         }
123
124         if (!base) {
125                 printk(KERN_ERR "page cgroup allocation failure\n");
126                 return -ENOMEM;
127         }
128
129         for (index = 0; index < PAGES_PER_SECTION; index++) {
130                 pc = base + index;
131                 __init_page_cgroup(pc, pfn + index);
132         }
133
134         section = __pfn_to_section(pfn);
135         section->page_cgroup = base - pfn;
136         total_usage += table_size;
137         return 0;
138 }
139 #ifdef CONFIG_MEMORY_HOTPLUG
140 void __free_page_cgroup(unsigned long pfn)
141 {
142         struct mem_section *ms;
143         struct page_cgroup *base;
144
145         ms = __pfn_to_section(pfn);
146         if (!ms || !ms->page_cgroup)
147                 return;
148         base = ms->page_cgroup + pfn;
149         if (is_vmalloc_addr(base)) {
150                 vfree(base);
151                 ms->page_cgroup = NULL;
152         } else {
153                 struct page *page = virt_to_page(base);
154                 if (!PageReserved(page)) { /* Is bootmem ? */
155                         kfree(base);
156                         ms->page_cgroup = NULL;
157                 }
158         }
159 }
160
161 int online_page_cgroup(unsigned long start_pfn,
162                         unsigned long nr_pages,
163                         int nid)
164 {
165         unsigned long start, end, pfn;
166         int fail = 0;
167
168         start = start_pfn & (PAGES_PER_SECTION - 1);
169         end = ALIGN(start_pfn + nr_pages, PAGES_PER_SECTION);
170
171         for (pfn = start; !fail && pfn < end; pfn += PAGES_PER_SECTION) {
172                 if (!pfn_present(pfn))
173                         continue;
174                 fail = init_section_page_cgroup(pfn);
175         }
176         if (!fail)
177                 return 0;
178
179         /* rollback */
180         for (pfn = start; pfn < end; pfn += PAGES_PER_SECTION)
181                 __free_page_cgroup(pfn);
182
183         return -ENOMEM;
184 }
185
186 int offline_page_cgroup(unsigned long start_pfn,
187                 unsigned long nr_pages, int nid)
188 {
189         unsigned long start, end, pfn;
190
191         start = start_pfn & (PAGES_PER_SECTION - 1);
192         end = ALIGN(start_pfn + nr_pages, PAGES_PER_SECTION);
193
194         for (pfn = start; pfn < end; pfn += PAGES_PER_SECTION)
195                 __free_page_cgroup(pfn);
196         return 0;
197
198 }
199
200 static int page_cgroup_callback(struct notifier_block *self,
201                                unsigned long action, void *arg)
202 {
203         struct memory_notify *mn = arg;
204         int ret = 0;
205         switch (action) {
206         case MEM_GOING_ONLINE:
207                 ret = online_page_cgroup(mn->start_pfn,
208                                    mn->nr_pages, mn->status_change_nid);
209                 break;
210         case MEM_CANCEL_ONLINE:
211         case MEM_OFFLINE:
212                 offline_page_cgroup(mn->start_pfn,
213                                 mn->nr_pages, mn->status_change_nid);
214                 break;
215         case MEM_GOING_OFFLINE:
216                 break;
217         case MEM_ONLINE:
218         case MEM_CANCEL_OFFLINE:
219                 break;
220         }
221         ret = notifier_from_errno(ret);
222         return ret;
223 }
224
225 #endif
226
227 void __init page_cgroup_init(void)
228 {
229         unsigned long pfn;
230         int fail = 0;
231
232         if (mem_cgroup_subsys.disabled)
233                 return;
234
235         for (pfn = 0; !fail && pfn < max_pfn; pfn += PAGES_PER_SECTION) {
236                 if (!pfn_present(pfn))
237                         continue;
238                 fail = init_section_page_cgroup(pfn);
239         }
240         if (fail) {
241                 printk(KERN_CRIT "try cgroup_disable=memory boot option\n");
242                 panic("Out of memory");
243         } else {
244                 hotplug_memory_notifier(page_cgroup_callback, 0);
245         }
246         printk(KERN_INFO "allocated %ld bytes of page_cgroup\n", total_usage);
247         printk(KERN_INFO "please try cgroup_disable=memory option if you don't"
248         " want\n");
249 }
250
251 void __init pgdat_page_cgroup_init(struct pglist_data *pgdat)
252 {
253         return;
254 }
255
256 #endif