Commit 7952f98818d561ed0e11434a7a16acd9a7bae859
1 parent
815c4163b6
Exists in
master
and in
4 other branches
kmemleak: Annotate false positive in init_section_page_cgroup()
The pointer to the page_cgroup table allocated in init_section_page_cgroup() is stored in section->page_cgroup as (base - pfn). Since this value does not point to the beginning or inside the allocated memory block, kmemleak reports a false positive. This was reported in bugzilla.kernel.org as #16297. Signed-off-by: Catalin Marinas <catalin.marinas@arm.com> Reported-by: Adrien Dessemond <adrien.dessemond@gmail.com> Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Cc: Pekka Enberg <penberg@cs.helsinki.fi> Cc: Andrew Morton <akpm@linux-foundation.org>
Showing 1 changed file with 7 additions and 0 deletions Inline Diff
mm/page_cgroup.c
1 | #include <linux/mm.h> | 1 | #include <linux/mm.h> |
2 | #include <linux/mmzone.h> | 2 | #include <linux/mmzone.h> |
3 | #include <linux/bootmem.h> | 3 | #include <linux/bootmem.h> |
4 | #include <linux/bit_spinlock.h> | 4 | #include <linux/bit_spinlock.h> |
5 | #include <linux/page_cgroup.h> | 5 | #include <linux/page_cgroup.h> |
6 | #include <linux/hash.h> | 6 | #include <linux/hash.h> |
7 | #include <linux/slab.h> | 7 | #include <linux/slab.h> |
8 | #include <linux/memory.h> | 8 | #include <linux/memory.h> |
9 | #include <linux/vmalloc.h> | 9 | #include <linux/vmalloc.h> |
10 | #include <linux/cgroup.h> | 10 | #include <linux/cgroup.h> |
11 | #include <linux/swapops.h> | 11 | #include <linux/swapops.h> |
12 | #include <linux/kmemleak.h> | ||
12 | 13 | ||
13 | static void __meminit | 14 | static void __meminit |
14 | __init_page_cgroup(struct page_cgroup *pc, unsigned long pfn) | 15 | __init_page_cgroup(struct page_cgroup *pc, unsigned long pfn) |
15 | { | 16 | { |
16 | pc->flags = 0; | 17 | pc->flags = 0; |
17 | pc->mem_cgroup = NULL; | 18 | pc->mem_cgroup = NULL; |
18 | pc->page = pfn_to_page(pfn); | 19 | pc->page = pfn_to_page(pfn); |
19 | INIT_LIST_HEAD(&pc->lru); | 20 | INIT_LIST_HEAD(&pc->lru); |
20 | } | 21 | } |
21 | static unsigned long total_usage; | 22 | static unsigned long total_usage; |
22 | 23 | ||
23 | #if !defined(CONFIG_SPARSEMEM) | 24 | #if !defined(CONFIG_SPARSEMEM) |
24 | 25 | ||
25 | 26 | ||
26 | void __meminit pgdat_page_cgroup_init(struct pglist_data *pgdat) | 27 | void __meminit pgdat_page_cgroup_init(struct pglist_data *pgdat) |
27 | { | 28 | { |
28 | pgdat->node_page_cgroup = NULL; | 29 | pgdat->node_page_cgroup = NULL; |
29 | } | 30 | } |
30 | 31 | ||
31 | struct page_cgroup *lookup_page_cgroup(struct page *page) | 32 | struct page_cgroup *lookup_page_cgroup(struct page *page) |
32 | { | 33 | { |
33 | unsigned long pfn = page_to_pfn(page); | 34 | unsigned long pfn = page_to_pfn(page); |
34 | unsigned long offset; | 35 | unsigned long offset; |
35 | struct page_cgroup *base; | 36 | struct page_cgroup *base; |
36 | 37 | ||
37 | base = NODE_DATA(page_to_nid(page))->node_page_cgroup; | 38 | base = NODE_DATA(page_to_nid(page))->node_page_cgroup; |
38 | if (unlikely(!base)) | 39 | if (unlikely(!base)) |
39 | return NULL; | 40 | return NULL; |
40 | 41 | ||
41 | offset = pfn - NODE_DATA(page_to_nid(page))->node_start_pfn; | 42 | offset = pfn - NODE_DATA(page_to_nid(page))->node_start_pfn; |
42 | return base + offset; | 43 | return base + offset; |
43 | } | 44 | } |
44 | 45 | ||
45 | static int __init alloc_node_page_cgroup(int nid) | 46 | static int __init alloc_node_page_cgroup(int nid) |
46 | { | 47 | { |
47 | struct page_cgroup *base, *pc; | 48 | struct page_cgroup *base, *pc; |
48 | unsigned long table_size; | 49 | unsigned long table_size; |
49 | unsigned long start_pfn, nr_pages, index; | 50 | unsigned long start_pfn, nr_pages, index; |
50 | 51 | ||
51 | start_pfn = NODE_DATA(nid)->node_start_pfn; | 52 | start_pfn = NODE_DATA(nid)->node_start_pfn; |
52 | nr_pages = NODE_DATA(nid)->node_spanned_pages; | 53 | nr_pages = NODE_DATA(nid)->node_spanned_pages; |
53 | 54 | ||
54 | if (!nr_pages) | 55 | if (!nr_pages) |
55 | return 0; | 56 | return 0; |
56 | 57 | ||
57 | table_size = sizeof(struct page_cgroup) * nr_pages; | 58 | table_size = sizeof(struct page_cgroup) * nr_pages; |
58 | 59 | ||
59 | base = __alloc_bootmem_node_nopanic(NODE_DATA(nid), | 60 | base = __alloc_bootmem_node_nopanic(NODE_DATA(nid), |
60 | table_size, PAGE_SIZE, __pa(MAX_DMA_ADDRESS)); | 61 | table_size, PAGE_SIZE, __pa(MAX_DMA_ADDRESS)); |
61 | if (!base) | 62 | if (!base) |
62 | return -ENOMEM; | 63 | return -ENOMEM; |
63 | for (index = 0; index < nr_pages; index++) { | 64 | for (index = 0; index < nr_pages; index++) { |
64 | pc = base + index; | 65 | pc = base + index; |
65 | __init_page_cgroup(pc, start_pfn + index); | 66 | __init_page_cgroup(pc, start_pfn + index); |
66 | } | 67 | } |
67 | NODE_DATA(nid)->node_page_cgroup = base; | 68 | NODE_DATA(nid)->node_page_cgroup = base; |
68 | total_usage += table_size; | 69 | total_usage += table_size; |
69 | return 0; | 70 | return 0; |
70 | } | 71 | } |
71 | 72 | ||
72 | void __init page_cgroup_init_flatmem(void) | 73 | void __init page_cgroup_init_flatmem(void) |
73 | { | 74 | { |
74 | 75 | ||
75 | int nid, fail; | 76 | int nid, fail; |
76 | 77 | ||
77 | if (mem_cgroup_disabled()) | 78 | if (mem_cgroup_disabled()) |
78 | return; | 79 | return; |
79 | 80 | ||
80 | for_each_online_node(nid) { | 81 | for_each_online_node(nid) { |
81 | fail = alloc_node_page_cgroup(nid); | 82 | fail = alloc_node_page_cgroup(nid); |
82 | if (fail) | 83 | if (fail) |
83 | goto fail; | 84 | goto fail; |
84 | } | 85 | } |
85 | printk(KERN_INFO "allocated %ld bytes of page_cgroup\n", total_usage); | 86 | 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 | printk(KERN_INFO "please try 'cgroup_disable=memory' option if you" |
87 | " don't want memory cgroups\n"); | 88 | " don't want memory cgroups\n"); |
88 | return; | 89 | return; |
89 | fail: | 90 | fail: |
90 | printk(KERN_CRIT "allocation of page_cgroup failed.\n"); | 91 | printk(KERN_CRIT "allocation of page_cgroup failed.\n"); |
91 | printk(KERN_CRIT "please try 'cgroup_disable=memory' boot option\n"); | 92 | printk(KERN_CRIT "please try 'cgroup_disable=memory' boot option\n"); |
92 | panic("Out of memory"); | 93 | panic("Out of memory"); |
93 | } | 94 | } |
94 | 95 | ||
95 | #else /* CONFIG_FLAT_NODE_MEM_MAP */ | 96 | #else /* CONFIG_FLAT_NODE_MEM_MAP */ |
96 | 97 | ||
97 | struct page_cgroup *lookup_page_cgroup(struct page *page) | 98 | struct page_cgroup *lookup_page_cgroup(struct page *page) |
98 | { | 99 | { |
99 | unsigned long pfn = page_to_pfn(page); | 100 | unsigned long pfn = page_to_pfn(page); |
100 | struct mem_section *section = __pfn_to_section(pfn); | 101 | struct mem_section *section = __pfn_to_section(pfn); |
101 | 102 | ||
102 | if (!section->page_cgroup) | 103 | if (!section->page_cgroup) |
103 | return NULL; | 104 | return NULL; |
104 | return section->page_cgroup + pfn; | 105 | return section->page_cgroup + pfn; |
105 | } | 106 | } |
106 | 107 | ||
107 | /* __alloc_bootmem...() is protected by !slab_available() */ | 108 | /* __alloc_bootmem...() is protected by !slab_available() */ |
108 | static int __init_refok init_section_page_cgroup(unsigned long pfn) | 109 | static int __init_refok init_section_page_cgroup(unsigned long pfn) |
109 | { | 110 | { |
110 | struct mem_section *section = __pfn_to_section(pfn); | 111 | struct mem_section *section = __pfn_to_section(pfn); |
111 | struct page_cgroup *base, *pc; | 112 | struct page_cgroup *base, *pc; |
112 | unsigned long table_size; | 113 | unsigned long table_size; |
113 | int nid, index; | 114 | int nid, index; |
114 | 115 | ||
115 | if (!section->page_cgroup) { | 116 | if (!section->page_cgroup) { |
116 | nid = page_to_nid(pfn_to_page(pfn)); | 117 | nid = page_to_nid(pfn_to_page(pfn)); |
117 | table_size = sizeof(struct page_cgroup) * PAGES_PER_SECTION; | 118 | table_size = sizeof(struct page_cgroup) * PAGES_PER_SECTION; |
118 | VM_BUG_ON(!slab_is_available()); | 119 | VM_BUG_ON(!slab_is_available()); |
119 | if (node_state(nid, N_HIGH_MEMORY)) { | 120 | if (node_state(nid, N_HIGH_MEMORY)) { |
120 | base = kmalloc_node(table_size, | 121 | base = kmalloc_node(table_size, |
121 | GFP_KERNEL | __GFP_NOWARN, nid); | 122 | GFP_KERNEL | __GFP_NOWARN, nid); |
122 | if (!base) | 123 | if (!base) |
123 | base = vmalloc_node(table_size, nid); | 124 | base = vmalloc_node(table_size, nid); |
124 | } else { | 125 | } else { |
125 | base = kmalloc(table_size, GFP_KERNEL | __GFP_NOWARN); | 126 | base = kmalloc(table_size, GFP_KERNEL | __GFP_NOWARN); |
126 | if (!base) | 127 | if (!base) |
127 | base = vmalloc(table_size); | 128 | base = vmalloc(table_size); |
128 | } | 129 | } |
130 | /* | ||
131 | * The value stored in section->page_cgroup is (base - pfn) | ||
132 | * and it does not point to the memory block allocated above, | ||
133 | * causing kmemleak false positives. | ||
134 | */ | ||
135 | kmemleak_not_leak(base); | ||
129 | } else { | 136 | } else { |
130 | /* | 137 | /* |
131 | * We don't have to allocate page_cgroup again, but | 138 | * We don't have to allocate page_cgroup again, but |
132 | * address of memmap may be changed. So, we have to initialize | 139 | * address of memmap may be changed. So, we have to initialize |
133 | * again. | 140 | * again. |
134 | */ | 141 | */ |
135 | base = section->page_cgroup + pfn; | 142 | base = section->page_cgroup + pfn; |
136 | table_size = 0; | 143 | table_size = 0; |
137 | /* check address of memmap is changed or not. */ | 144 | /* check address of memmap is changed or not. */ |
138 | if (base->page == pfn_to_page(pfn)) | 145 | if (base->page == pfn_to_page(pfn)) |
139 | return 0; | 146 | return 0; |
140 | } | 147 | } |
141 | 148 | ||
142 | if (!base) { | 149 | if (!base) { |
143 | printk(KERN_ERR "page cgroup allocation failure\n"); | 150 | printk(KERN_ERR "page cgroup allocation failure\n"); |
144 | return -ENOMEM; | 151 | return -ENOMEM; |
145 | } | 152 | } |
146 | 153 | ||
147 | for (index = 0; index < PAGES_PER_SECTION; index++) { | 154 | for (index = 0; index < PAGES_PER_SECTION; index++) { |
148 | pc = base + index; | 155 | pc = base + index; |
149 | __init_page_cgroup(pc, pfn + index); | 156 | __init_page_cgroup(pc, pfn + index); |
150 | } | 157 | } |
151 | 158 | ||
152 | section->page_cgroup = base - pfn; | 159 | section->page_cgroup = base - pfn; |
153 | total_usage += table_size; | 160 | total_usage += table_size; |
154 | return 0; | 161 | return 0; |
155 | } | 162 | } |
156 | #ifdef CONFIG_MEMORY_HOTPLUG | 163 | #ifdef CONFIG_MEMORY_HOTPLUG |
157 | void __free_page_cgroup(unsigned long pfn) | 164 | void __free_page_cgroup(unsigned long pfn) |
158 | { | 165 | { |
159 | struct mem_section *ms; | 166 | struct mem_section *ms; |
160 | struct page_cgroup *base; | 167 | struct page_cgroup *base; |
161 | 168 | ||
162 | ms = __pfn_to_section(pfn); | 169 | ms = __pfn_to_section(pfn); |
163 | if (!ms || !ms->page_cgroup) | 170 | if (!ms || !ms->page_cgroup) |
164 | return; | 171 | return; |
165 | base = ms->page_cgroup + pfn; | 172 | base = ms->page_cgroup + pfn; |
166 | if (is_vmalloc_addr(base)) { | 173 | if (is_vmalloc_addr(base)) { |
167 | vfree(base); | 174 | vfree(base); |
168 | ms->page_cgroup = NULL; | 175 | ms->page_cgroup = NULL; |
169 | } else { | 176 | } else { |
170 | struct page *page = virt_to_page(base); | 177 | struct page *page = virt_to_page(base); |
171 | if (!PageReserved(page)) { /* Is bootmem ? */ | 178 | if (!PageReserved(page)) { /* Is bootmem ? */ |
172 | kfree(base); | 179 | kfree(base); |
173 | ms->page_cgroup = NULL; | 180 | ms->page_cgroup = NULL; |
174 | } | 181 | } |
175 | } | 182 | } |
176 | } | 183 | } |
177 | 184 | ||
178 | int __meminit online_page_cgroup(unsigned long start_pfn, | 185 | int __meminit online_page_cgroup(unsigned long start_pfn, |
179 | unsigned long nr_pages, | 186 | unsigned long nr_pages, |
180 | int nid) | 187 | int nid) |
181 | { | 188 | { |
182 | unsigned long start, end, pfn; | 189 | unsigned long start, end, pfn; |
183 | int fail = 0; | 190 | int fail = 0; |
184 | 191 | ||
185 | start = start_pfn & ~(PAGES_PER_SECTION - 1); | 192 | start = start_pfn & ~(PAGES_PER_SECTION - 1); |
186 | end = ALIGN(start_pfn + nr_pages, PAGES_PER_SECTION); | 193 | end = ALIGN(start_pfn + nr_pages, PAGES_PER_SECTION); |
187 | 194 | ||
188 | for (pfn = start; !fail && pfn < end; pfn += PAGES_PER_SECTION) { | 195 | for (pfn = start; !fail && pfn < end; pfn += PAGES_PER_SECTION) { |
189 | if (!pfn_present(pfn)) | 196 | if (!pfn_present(pfn)) |
190 | continue; | 197 | continue; |
191 | fail = init_section_page_cgroup(pfn); | 198 | fail = init_section_page_cgroup(pfn); |
192 | } | 199 | } |
193 | if (!fail) | 200 | if (!fail) |
194 | return 0; | 201 | return 0; |
195 | 202 | ||
196 | /* rollback */ | 203 | /* rollback */ |
197 | for (pfn = start; pfn < end; pfn += PAGES_PER_SECTION) | 204 | for (pfn = start; pfn < end; pfn += PAGES_PER_SECTION) |
198 | __free_page_cgroup(pfn); | 205 | __free_page_cgroup(pfn); |
199 | 206 | ||
200 | return -ENOMEM; | 207 | return -ENOMEM; |
201 | } | 208 | } |
202 | 209 | ||
203 | int __meminit offline_page_cgroup(unsigned long start_pfn, | 210 | int __meminit offline_page_cgroup(unsigned long start_pfn, |
204 | unsigned long nr_pages, int nid) | 211 | unsigned long nr_pages, int nid) |
205 | { | 212 | { |
206 | unsigned long start, end, pfn; | 213 | unsigned long start, end, pfn; |
207 | 214 | ||
208 | start = start_pfn & ~(PAGES_PER_SECTION - 1); | 215 | start = start_pfn & ~(PAGES_PER_SECTION - 1); |
209 | end = ALIGN(start_pfn + nr_pages, PAGES_PER_SECTION); | 216 | end = ALIGN(start_pfn + nr_pages, PAGES_PER_SECTION); |
210 | 217 | ||
211 | for (pfn = start; pfn < end; pfn += PAGES_PER_SECTION) | 218 | for (pfn = start; pfn < end; pfn += PAGES_PER_SECTION) |
212 | __free_page_cgroup(pfn); | 219 | __free_page_cgroup(pfn); |
213 | return 0; | 220 | return 0; |
214 | 221 | ||
215 | } | 222 | } |
216 | 223 | ||
217 | static int __meminit page_cgroup_callback(struct notifier_block *self, | 224 | static int __meminit page_cgroup_callback(struct notifier_block *self, |
218 | unsigned long action, void *arg) | 225 | unsigned long action, void *arg) |
219 | { | 226 | { |
220 | struct memory_notify *mn = arg; | 227 | struct memory_notify *mn = arg; |
221 | int ret = 0; | 228 | int ret = 0; |
222 | switch (action) { | 229 | switch (action) { |
223 | case MEM_GOING_ONLINE: | 230 | case MEM_GOING_ONLINE: |
224 | ret = online_page_cgroup(mn->start_pfn, | 231 | ret = online_page_cgroup(mn->start_pfn, |
225 | mn->nr_pages, mn->status_change_nid); | 232 | mn->nr_pages, mn->status_change_nid); |
226 | break; | 233 | break; |
227 | case MEM_OFFLINE: | 234 | case MEM_OFFLINE: |
228 | offline_page_cgroup(mn->start_pfn, | 235 | offline_page_cgroup(mn->start_pfn, |
229 | mn->nr_pages, mn->status_change_nid); | 236 | mn->nr_pages, mn->status_change_nid); |
230 | break; | 237 | break; |
231 | case MEM_CANCEL_ONLINE: | 238 | case MEM_CANCEL_ONLINE: |
232 | case MEM_GOING_OFFLINE: | 239 | case MEM_GOING_OFFLINE: |
233 | break; | 240 | break; |
234 | case MEM_ONLINE: | 241 | case MEM_ONLINE: |
235 | case MEM_CANCEL_OFFLINE: | 242 | case MEM_CANCEL_OFFLINE: |
236 | break; | 243 | break; |
237 | } | 244 | } |
238 | 245 | ||
239 | if (ret) | 246 | if (ret) |
240 | ret = notifier_from_errno(ret); | 247 | ret = notifier_from_errno(ret); |
241 | else | 248 | else |
242 | ret = NOTIFY_OK; | 249 | ret = NOTIFY_OK; |
243 | 250 | ||
244 | return ret; | 251 | return ret; |
245 | } | 252 | } |
246 | 253 | ||
247 | #endif | 254 | #endif |
248 | 255 | ||
249 | void __init page_cgroup_init(void) | 256 | void __init page_cgroup_init(void) |
250 | { | 257 | { |
251 | unsigned long pfn; | 258 | unsigned long pfn; |
252 | int fail = 0; | 259 | int fail = 0; |
253 | 260 | ||
254 | if (mem_cgroup_disabled()) | 261 | if (mem_cgroup_disabled()) |
255 | return; | 262 | return; |
256 | 263 | ||
257 | for (pfn = 0; !fail && pfn < max_pfn; pfn += PAGES_PER_SECTION) { | 264 | for (pfn = 0; !fail && pfn < max_pfn; pfn += PAGES_PER_SECTION) { |
258 | if (!pfn_present(pfn)) | 265 | if (!pfn_present(pfn)) |
259 | continue; | 266 | continue; |
260 | fail = init_section_page_cgroup(pfn); | 267 | fail = init_section_page_cgroup(pfn); |
261 | } | 268 | } |
262 | if (fail) { | 269 | if (fail) { |
263 | printk(KERN_CRIT "try 'cgroup_disable=memory' boot option\n"); | 270 | printk(KERN_CRIT "try 'cgroup_disable=memory' boot option\n"); |
264 | panic("Out of memory"); | 271 | panic("Out of memory"); |
265 | } else { | 272 | } else { |
266 | hotplug_memory_notifier(page_cgroup_callback, 0); | 273 | hotplug_memory_notifier(page_cgroup_callback, 0); |
267 | } | 274 | } |
268 | printk(KERN_INFO "allocated %ld bytes of page_cgroup\n", total_usage); | 275 | printk(KERN_INFO "allocated %ld bytes of page_cgroup\n", total_usage); |
269 | printk(KERN_INFO "please try 'cgroup_disable=memory' option if you don't" | 276 | printk(KERN_INFO "please try 'cgroup_disable=memory' option if you don't" |
270 | " want memory cgroups\n"); | 277 | " want memory cgroups\n"); |
271 | } | 278 | } |
272 | 279 | ||
273 | void __meminit pgdat_page_cgroup_init(struct pglist_data *pgdat) | 280 | void __meminit pgdat_page_cgroup_init(struct pglist_data *pgdat) |
274 | { | 281 | { |
275 | return; | 282 | return; |
276 | } | 283 | } |
277 | 284 | ||
278 | #endif | 285 | #endif |
279 | 286 | ||
280 | 287 | ||
281 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP | 288 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP |
282 | 289 | ||
283 | static DEFINE_MUTEX(swap_cgroup_mutex); | 290 | static DEFINE_MUTEX(swap_cgroup_mutex); |
284 | struct swap_cgroup_ctrl { | 291 | struct swap_cgroup_ctrl { |
285 | struct page **map; | 292 | struct page **map; |
286 | unsigned long length; | 293 | unsigned long length; |
287 | spinlock_t lock; | 294 | spinlock_t lock; |
288 | }; | 295 | }; |
289 | 296 | ||
290 | struct swap_cgroup_ctrl swap_cgroup_ctrl[MAX_SWAPFILES]; | 297 | struct swap_cgroup_ctrl swap_cgroup_ctrl[MAX_SWAPFILES]; |
291 | 298 | ||
292 | struct swap_cgroup { | 299 | struct swap_cgroup { |
293 | unsigned short id; | 300 | unsigned short id; |
294 | }; | 301 | }; |
295 | #define SC_PER_PAGE (PAGE_SIZE/sizeof(struct swap_cgroup)) | 302 | #define SC_PER_PAGE (PAGE_SIZE/sizeof(struct swap_cgroup)) |
296 | #define SC_POS_MASK (SC_PER_PAGE - 1) | 303 | #define SC_POS_MASK (SC_PER_PAGE - 1) |
297 | 304 | ||
298 | /* | 305 | /* |
299 | * SwapCgroup implements "lookup" and "exchange" operations. | 306 | * SwapCgroup implements "lookup" and "exchange" operations. |
300 | * In typical usage, this swap_cgroup is accessed via memcg's charge/uncharge | 307 | * In typical usage, this swap_cgroup is accessed via memcg's charge/uncharge |
301 | * against SwapCache. At swap_free(), this is accessed directly from swap. | 308 | * against SwapCache. At swap_free(), this is accessed directly from swap. |
302 | * | 309 | * |
303 | * This means, | 310 | * This means, |
304 | * - we have no race in "exchange" when we're accessed via SwapCache because | 311 | * - we have no race in "exchange" when we're accessed via SwapCache because |
305 | * SwapCache(and its swp_entry) is under lock. | 312 | * SwapCache(and its swp_entry) is under lock. |
306 | * - When called via swap_free(), there is no user of this entry and no race. | 313 | * - When called via swap_free(), there is no user of this entry and no race. |
307 | * Then, we don't need lock around "exchange". | 314 | * Then, we don't need lock around "exchange". |
308 | * | 315 | * |
309 | * TODO: we can push these buffers out to HIGHMEM. | 316 | * TODO: we can push these buffers out to HIGHMEM. |
310 | */ | 317 | */ |
311 | 318 | ||
312 | /* | 319 | /* |
313 | * allocate buffer for swap_cgroup. | 320 | * allocate buffer for swap_cgroup. |
314 | */ | 321 | */ |
315 | static int swap_cgroup_prepare(int type) | 322 | static int swap_cgroup_prepare(int type) |
316 | { | 323 | { |
317 | struct page *page; | 324 | struct page *page; |
318 | struct swap_cgroup_ctrl *ctrl; | 325 | struct swap_cgroup_ctrl *ctrl; |
319 | unsigned long idx, max; | 326 | unsigned long idx, max; |
320 | 327 | ||
321 | ctrl = &swap_cgroup_ctrl[type]; | 328 | ctrl = &swap_cgroup_ctrl[type]; |
322 | 329 | ||
323 | for (idx = 0; idx < ctrl->length; idx++) { | 330 | for (idx = 0; idx < ctrl->length; idx++) { |
324 | page = alloc_page(GFP_KERNEL | __GFP_ZERO); | 331 | page = alloc_page(GFP_KERNEL | __GFP_ZERO); |
325 | if (!page) | 332 | if (!page) |
326 | goto not_enough_page; | 333 | goto not_enough_page; |
327 | ctrl->map[idx] = page; | 334 | ctrl->map[idx] = page; |
328 | } | 335 | } |
329 | return 0; | 336 | return 0; |
330 | not_enough_page: | 337 | not_enough_page: |
331 | max = idx; | 338 | max = idx; |
332 | for (idx = 0; idx < max; idx++) | 339 | for (idx = 0; idx < max; idx++) |
333 | __free_page(ctrl->map[idx]); | 340 | __free_page(ctrl->map[idx]); |
334 | 341 | ||
335 | return -ENOMEM; | 342 | return -ENOMEM; |
336 | } | 343 | } |
337 | 344 | ||
338 | /** | 345 | /** |
339 | * swap_cgroup_cmpxchg - cmpxchg mem_cgroup's id for this swp_entry. | 346 | * swap_cgroup_cmpxchg - cmpxchg mem_cgroup's id for this swp_entry. |
340 | * @end: swap entry to be cmpxchged | 347 | * @end: swap entry to be cmpxchged |
341 | * @old: old id | 348 | * @old: old id |
342 | * @new: new id | 349 | * @new: new id |
343 | * | 350 | * |
344 | * Returns old id at success, 0 at failure. | 351 | * Returns old id at success, 0 at failure. |
345 | * (There is no mem_cgroup useing 0 as its id) | 352 | * (There is no mem_cgroup useing 0 as its id) |
346 | */ | 353 | */ |
347 | unsigned short swap_cgroup_cmpxchg(swp_entry_t ent, | 354 | unsigned short swap_cgroup_cmpxchg(swp_entry_t ent, |
348 | unsigned short old, unsigned short new) | 355 | unsigned short old, unsigned short new) |
349 | { | 356 | { |
350 | int type = swp_type(ent); | 357 | int type = swp_type(ent); |
351 | unsigned long offset = swp_offset(ent); | 358 | unsigned long offset = swp_offset(ent); |
352 | unsigned long idx = offset / SC_PER_PAGE; | 359 | unsigned long idx = offset / SC_PER_PAGE; |
353 | unsigned long pos = offset & SC_POS_MASK; | 360 | unsigned long pos = offset & SC_POS_MASK; |
354 | struct swap_cgroup_ctrl *ctrl; | 361 | struct swap_cgroup_ctrl *ctrl; |
355 | struct page *mappage; | 362 | struct page *mappage; |
356 | struct swap_cgroup *sc; | 363 | struct swap_cgroup *sc; |
357 | unsigned long flags; | 364 | unsigned long flags; |
358 | unsigned short retval; | 365 | unsigned short retval; |
359 | 366 | ||
360 | ctrl = &swap_cgroup_ctrl[type]; | 367 | ctrl = &swap_cgroup_ctrl[type]; |
361 | 368 | ||
362 | mappage = ctrl->map[idx]; | 369 | mappage = ctrl->map[idx]; |
363 | sc = page_address(mappage); | 370 | sc = page_address(mappage); |
364 | sc += pos; | 371 | sc += pos; |
365 | spin_lock_irqsave(&ctrl->lock, flags); | 372 | spin_lock_irqsave(&ctrl->lock, flags); |
366 | retval = sc->id; | 373 | retval = sc->id; |
367 | if (retval == old) | 374 | if (retval == old) |
368 | sc->id = new; | 375 | sc->id = new; |
369 | else | 376 | else |
370 | retval = 0; | 377 | retval = 0; |
371 | spin_unlock_irqrestore(&ctrl->lock, flags); | 378 | spin_unlock_irqrestore(&ctrl->lock, flags); |
372 | return retval; | 379 | return retval; |
373 | } | 380 | } |
374 | 381 | ||
375 | /** | 382 | /** |
376 | * swap_cgroup_record - record mem_cgroup for this swp_entry. | 383 | * swap_cgroup_record - record mem_cgroup for this swp_entry. |
377 | * @ent: swap entry to be recorded into | 384 | * @ent: swap entry to be recorded into |
378 | * @mem: mem_cgroup to be recorded | 385 | * @mem: mem_cgroup to be recorded |
379 | * | 386 | * |
380 | * Returns old value at success, 0 at failure. | 387 | * Returns old value at success, 0 at failure. |
381 | * (Of course, old value can be 0.) | 388 | * (Of course, old value can be 0.) |
382 | */ | 389 | */ |
383 | unsigned short swap_cgroup_record(swp_entry_t ent, unsigned short id) | 390 | unsigned short swap_cgroup_record(swp_entry_t ent, unsigned short id) |
384 | { | 391 | { |
385 | int type = swp_type(ent); | 392 | int type = swp_type(ent); |
386 | unsigned long offset = swp_offset(ent); | 393 | unsigned long offset = swp_offset(ent); |
387 | unsigned long idx = offset / SC_PER_PAGE; | 394 | unsigned long idx = offset / SC_PER_PAGE; |
388 | unsigned long pos = offset & SC_POS_MASK; | 395 | unsigned long pos = offset & SC_POS_MASK; |
389 | struct swap_cgroup_ctrl *ctrl; | 396 | struct swap_cgroup_ctrl *ctrl; |
390 | struct page *mappage; | 397 | struct page *mappage; |
391 | struct swap_cgroup *sc; | 398 | struct swap_cgroup *sc; |
392 | unsigned short old; | 399 | unsigned short old; |
393 | unsigned long flags; | 400 | unsigned long flags; |
394 | 401 | ||
395 | ctrl = &swap_cgroup_ctrl[type]; | 402 | ctrl = &swap_cgroup_ctrl[type]; |
396 | 403 | ||
397 | mappage = ctrl->map[idx]; | 404 | mappage = ctrl->map[idx]; |
398 | sc = page_address(mappage); | 405 | sc = page_address(mappage); |
399 | sc += pos; | 406 | sc += pos; |
400 | spin_lock_irqsave(&ctrl->lock, flags); | 407 | spin_lock_irqsave(&ctrl->lock, flags); |
401 | old = sc->id; | 408 | old = sc->id; |
402 | sc->id = id; | 409 | sc->id = id; |
403 | spin_unlock_irqrestore(&ctrl->lock, flags); | 410 | spin_unlock_irqrestore(&ctrl->lock, flags); |
404 | 411 | ||
405 | return old; | 412 | return old; |
406 | } | 413 | } |
407 | 414 | ||
408 | /** | 415 | /** |
409 | * lookup_swap_cgroup - lookup mem_cgroup tied to swap entry | 416 | * lookup_swap_cgroup - lookup mem_cgroup tied to swap entry |
410 | * @ent: swap entry to be looked up. | 417 | * @ent: swap entry to be looked up. |
411 | * | 418 | * |
412 | * Returns CSS ID of mem_cgroup at success. 0 at failure. (0 is invalid ID) | 419 | * Returns CSS ID of mem_cgroup at success. 0 at failure. (0 is invalid ID) |
413 | */ | 420 | */ |
414 | unsigned short lookup_swap_cgroup(swp_entry_t ent) | 421 | unsigned short lookup_swap_cgroup(swp_entry_t ent) |
415 | { | 422 | { |
416 | int type = swp_type(ent); | 423 | int type = swp_type(ent); |
417 | unsigned long offset = swp_offset(ent); | 424 | unsigned long offset = swp_offset(ent); |
418 | unsigned long idx = offset / SC_PER_PAGE; | 425 | unsigned long idx = offset / SC_PER_PAGE; |
419 | unsigned long pos = offset & SC_POS_MASK; | 426 | unsigned long pos = offset & SC_POS_MASK; |
420 | struct swap_cgroup_ctrl *ctrl; | 427 | struct swap_cgroup_ctrl *ctrl; |
421 | struct page *mappage; | 428 | struct page *mappage; |
422 | struct swap_cgroup *sc; | 429 | struct swap_cgroup *sc; |
423 | unsigned short ret; | 430 | unsigned short ret; |
424 | 431 | ||
425 | ctrl = &swap_cgroup_ctrl[type]; | 432 | ctrl = &swap_cgroup_ctrl[type]; |
426 | mappage = ctrl->map[idx]; | 433 | mappage = ctrl->map[idx]; |
427 | sc = page_address(mappage); | 434 | sc = page_address(mappage); |
428 | sc += pos; | 435 | sc += pos; |
429 | ret = sc->id; | 436 | ret = sc->id; |
430 | return ret; | 437 | return ret; |
431 | } | 438 | } |
432 | 439 | ||
433 | int swap_cgroup_swapon(int type, unsigned long max_pages) | 440 | int swap_cgroup_swapon(int type, unsigned long max_pages) |
434 | { | 441 | { |
435 | void *array; | 442 | void *array; |
436 | unsigned long array_size; | 443 | unsigned long array_size; |
437 | unsigned long length; | 444 | unsigned long length; |
438 | struct swap_cgroup_ctrl *ctrl; | 445 | struct swap_cgroup_ctrl *ctrl; |
439 | 446 | ||
440 | if (!do_swap_account) | 447 | if (!do_swap_account) |
441 | return 0; | 448 | return 0; |
442 | 449 | ||
443 | length = ((max_pages/SC_PER_PAGE) + 1); | 450 | length = ((max_pages/SC_PER_PAGE) + 1); |
444 | array_size = length * sizeof(void *); | 451 | array_size = length * sizeof(void *); |
445 | 452 | ||
446 | array = vmalloc(array_size); | 453 | array = vmalloc(array_size); |
447 | if (!array) | 454 | if (!array) |
448 | goto nomem; | 455 | goto nomem; |
449 | 456 | ||
450 | memset(array, 0, array_size); | 457 | memset(array, 0, array_size); |
451 | ctrl = &swap_cgroup_ctrl[type]; | 458 | ctrl = &swap_cgroup_ctrl[type]; |
452 | mutex_lock(&swap_cgroup_mutex); | 459 | mutex_lock(&swap_cgroup_mutex); |
453 | ctrl->length = length; | 460 | ctrl->length = length; |
454 | ctrl->map = array; | 461 | ctrl->map = array; |
455 | spin_lock_init(&ctrl->lock); | 462 | spin_lock_init(&ctrl->lock); |
456 | if (swap_cgroup_prepare(type)) { | 463 | if (swap_cgroup_prepare(type)) { |
457 | /* memory shortage */ | 464 | /* memory shortage */ |
458 | ctrl->map = NULL; | 465 | ctrl->map = NULL; |
459 | ctrl->length = 0; | 466 | ctrl->length = 0; |
460 | vfree(array); | 467 | vfree(array); |
461 | mutex_unlock(&swap_cgroup_mutex); | 468 | mutex_unlock(&swap_cgroup_mutex); |
462 | goto nomem; | 469 | goto nomem; |
463 | } | 470 | } |
464 | mutex_unlock(&swap_cgroup_mutex); | 471 | mutex_unlock(&swap_cgroup_mutex); |
465 | 472 | ||
466 | return 0; | 473 | return 0; |
467 | nomem: | 474 | nomem: |
468 | printk(KERN_INFO "couldn't allocate enough memory for swap_cgroup.\n"); | 475 | printk(KERN_INFO "couldn't allocate enough memory for swap_cgroup.\n"); |
469 | printk(KERN_INFO | 476 | printk(KERN_INFO |
470 | "swap_cgroup can be disabled by noswapaccount boot option\n"); | 477 | "swap_cgroup can be disabled by noswapaccount boot option\n"); |
471 | return -ENOMEM; | 478 | return -ENOMEM; |
472 | } | 479 | } |
473 | 480 | ||
474 | void swap_cgroup_swapoff(int type) | 481 | void swap_cgroup_swapoff(int type) |
475 | { | 482 | { |
476 | int i; | 483 | int i; |
477 | struct swap_cgroup_ctrl *ctrl; | 484 | struct swap_cgroup_ctrl *ctrl; |
478 | 485 | ||
479 | if (!do_swap_account) | 486 | if (!do_swap_account) |
480 | return; | 487 | return; |
481 | 488 | ||
482 | mutex_lock(&swap_cgroup_mutex); | 489 | mutex_lock(&swap_cgroup_mutex); |
483 | ctrl = &swap_cgroup_ctrl[type]; | 490 | ctrl = &swap_cgroup_ctrl[type]; |
484 | if (ctrl->map) { | 491 | if (ctrl->map) { |
485 | for (i = 0; i < ctrl->length; i++) { | 492 | for (i = 0; i < ctrl->length; i++) { |
486 | struct page *page = ctrl->map[i]; | 493 | struct page *page = ctrl->map[i]; |
487 | if (page) | 494 | if (page) |
488 | __free_page(page); | 495 | __free_page(page); |
489 | } | 496 | } |
490 | vfree(ctrl->map); | 497 | vfree(ctrl->map); |
491 | ctrl->map = NULL; | 498 | ctrl->map = NULL; |
492 | ctrl->length = 0; | 499 | ctrl->length = 0; |
493 | } | 500 | } |
494 | mutex_unlock(&swap_cgroup_mutex); | 501 | mutex_unlock(&swap_cgroup_mutex); |
495 | } | 502 | } |
496 | 503 | ||
497 | #endif | 504 | #endif |
498 | 505 |