page_cgroup.c revision dc19f9db38295f811d9041bd89b113beccbd763a 
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
12static 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}
19static unsigned long total_usage;
20
21#if !defined(CONFIG_SPARSEMEM)
22
23
24void __meminit pgdat_page_cgroup_init(struct pglist_data *pgdat)
25{
26	pgdat->node_page_cgroup = NULL;
27}
28
29struct 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
43static 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
67void __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;
84fail:
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
92struct 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/* __alloc_bootmem...() is protected by !slab_available() */
101int __init_refok init_section_page_cgroup(unsigned long pfn)
102{
103	struct mem_section *section;
104	struct page_cgroup *base, *pc;
105	unsigned long table_size;
106	int nid, index;
107
108	section = __pfn_to_section(pfn);
109
110	if (!section->page_cgroup) {
111		nid = page_to_nid(pfn_to_page(pfn));
112		table_size = sizeof(struct page_cgroup) * PAGES_PER_SECTION;
113		if (slab_is_available()) {
114			base = kmalloc_node(table_size, GFP_KERNEL, nid);
115			if (!base)
116				base = vmalloc_node(table_size, nid);
117		} else {
118			base = __alloc_bootmem_node_nopanic(NODE_DATA(nid),
119				table_size,
120				PAGE_SIZE, __pa(MAX_DMA_ADDRESS));
121		}
122	} else {
123		/*
124 		 * We don't have to allocate page_cgroup again, but
125		 * address of memmap may be changed. So, we have to initialize
126		 * again.
127		 */
128		base = section->page_cgroup + pfn;
129		table_size = 0;
130		/* check address of memmap is changed or not. */
131		if (base->page == pfn_to_page(pfn))
132			return 0;
133	}
134
135	if (!base) {
136		printk(KERN_ERR "page cgroup allocation failure\n");
137		return -ENOMEM;
138	}
139
140	for (index = 0; index < PAGES_PER_SECTION; index++) {
141		pc = base + index;
142		__init_page_cgroup(pc, pfn + index);
143	}
144
145	section = __pfn_to_section(pfn);
146	section->page_cgroup = base - pfn;
147	total_usage += table_size;
148	return 0;
149}
150#ifdef CONFIG_MEMORY_HOTPLUG
151void __free_page_cgroup(unsigned long pfn)
152{
153	struct mem_section *ms;
154	struct page_cgroup *base;
155
156	ms = __pfn_to_section(pfn);
157	if (!ms || !ms->page_cgroup)
158		return;
159	base = ms->page_cgroup + pfn;
160	if (is_vmalloc_addr(base)) {
161		vfree(base);
162		ms->page_cgroup = NULL;
163	} else {
164		struct page *page = virt_to_page(base);
165		if (!PageReserved(page)) { /* Is bootmem ? */
166			kfree(base);
167			ms->page_cgroup = NULL;
168		}
169	}
170}
171
172int __meminit online_page_cgroup(unsigned long start_pfn,
173			unsigned long nr_pages,
174			int nid)
175{
176	unsigned long start, end, pfn;
177	int fail = 0;
178
179	start = start_pfn & ~(PAGES_PER_SECTION - 1);
180	end = ALIGN(start_pfn + nr_pages, PAGES_PER_SECTION);
181
182	for (pfn = start; !fail && pfn < end; pfn += PAGES_PER_SECTION) {
183		if (!pfn_present(pfn))
184			continue;
185		fail = init_section_page_cgroup(pfn);
186	}
187	if (!fail)
188		return 0;
189
190	/* rollback */
191	for (pfn = start; pfn < end; pfn += PAGES_PER_SECTION)
192		__free_page_cgroup(pfn);
193
194	return -ENOMEM;
195}
196
197int __meminit offline_page_cgroup(unsigned long start_pfn,
198		unsigned long nr_pages, int nid)
199{
200	unsigned long start, end, pfn;
201
202	start = start_pfn & ~(PAGES_PER_SECTION - 1);
203	end = ALIGN(start_pfn + nr_pages, PAGES_PER_SECTION);
204
205	for (pfn = start; pfn < end; pfn += PAGES_PER_SECTION)
206		__free_page_cgroup(pfn);
207	return 0;
208
209}
210
211static int __meminit page_cgroup_callback(struct notifier_block *self,
212			       unsigned long action, void *arg)
213{
214	struct memory_notify *mn = arg;
215	int ret = 0;
216	switch (action) {
217	case MEM_GOING_ONLINE:
218		ret = online_page_cgroup(mn->start_pfn,
219				   mn->nr_pages, mn->status_change_nid);
220		break;
221	case MEM_OFFLINE:
222		offline_page_cgroup(mn->start_pfn,
223				mn->nr_pages, mn->status_change_nid);
224		break;
225	case MEM_CANCEL_ONLINE:
226	case MEM_GOING_OFFLINE:
227		break;
228	case MEM_ONLINE:
229	case MEM_CANCEL_OFFLINE:
230		break;
231	}
232
233	if (ret)
234		ret = notifier_from_errno(ret);
235	else
236		ret = NOTIFY_OK;
237
238	return ret;
239}
240
241#endif
242
243void __init page_cgroup_init(void)
244{
245	unsigned long pfn;
246	int fail = 0;
247
248	if (mem_cgroup_subsys.disabled)
249		return;
250
251	for (pfn = 0; !fail && pfn < max_pfn; pfn += PAGES_PER_SECTION) {
252		if (!pfn_present(pfn))
253			continue;
254		fail = init_section_page_cgroup(pfn);
255	}
256	if (fail) {
257		printk(KERN_CRIT "try cgroup_disable=memory boot option\n");
258		panic("Out of memory");
259	} else {
260		hotplug_memory_notifier(page_cgroup_callback, 0);
261	}
262	printk(KERN_INFO "allocated %ld bytes of page_cgroup\n", total_usage);
263	printk(KERN_INFO "please try cgroup_disable=memory option if you don't"
264	" want\n");
265}
266
267void __meminit pgdat_page_cgroup_init(struct pglist_data *pgdat)
268{
269	return;
270}
271
272#endif
273