1
2#include <linux/mm.h>
3#include <linux/mmzone.h>
4#include <linux/memblock.h>
5#include <linux/page_ext.h>
6#include <linux/memory.h>
7#include <linux/vmalloc.h>
8#include <linux/kmemleak.h>
9#include <linux/page_owner.h>
10#include <linux/page_idle.h>
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61static struct page_ext_operations *page_ext_ops[] = {
62#ifdef CONFIG_PAGE_OWNER
63 &page_owner_ops,
64#endif
65#if defined(CONFIG_IDLE_PAGE_TRACKING) && !defined(CONFIG_64BIT)
66 &page_idle_ops,
67#endif
68};
69
70static unsigned long total_usage;
71static unsigned long extra_mem;
72
73static bool __init invoke_need_callbacks(void)
74{
75 int i;
76 int entries = ARRAY_SIZE(page_ext_ops);
77 bool need = false;
78
79 for (i = 0; i < entries; i++) {
80 if (page_ext_ops[i]->need && page_ext_ops[i]->need()) {
81 page_ext_ops[i]->offset = sizeof(struct page_ext) +
82 extra_mem;
83 extra_mem += page_ext_ops[i]->size;
84 need = true;
85 }
86 }
87
88 return need;
89}
90
91static void __init invoke_init_callbacks(void)
92{
93 int i;
94 int entries = ARRAY_SIZE(page_ext_ops);
95
96 for (i = 0; i < entries; i++) {
97 if (page_ext_ops[i]->init)
98 page_ext_ops[i]->init();
99 }
100}
101
102static unsigned long get_entry_size(void)
103{
104 return sizeof(struct page_ext) + extra_mem;
105}
106
107static inline struct page_ext *get_entry(void *base, unsigned long index)
108{
109 return base + get_entry_size() * index;
110}
111
112#if !defined(CONFIG_SPARSEMEM)
113
114
115void __meminit pgdat_page_ext_init(struct pglist_data *pgdat)
116{
117 pgdat->node_page_ext = NULL;
118}
119
120struct page_ext *lookup_page_ext(const struct page *page)
121{
122 unsigned long pfn = page_to_pfn(page);
123 unsigned long index;
124 struct page_ext *base;
125
126 base = NODE_DATA(page_to_nid(page))->node_page_ext;
127
128
129
130
131
132
133 if (unlikely(!base))
134 return NULL;
135 index = pfn - round_down(node_start_pfn(page_to_nid(page)),
136 MAX_ORDER_NR_PAGES);
137 return get_entry(base, index);
138}
139
140static int __init alloc_node_page_ext(int nid)
141{
142 struct page_ext *base;
143 unsigned long table_size;
144 unsigned long nr_pages;
145
146 nr_pages = NODE_DATA(nid)->node_spanned_pages;
147 if (!nr_pages)
148 return 0;
149
150
151
152
153
154
155 if (!IS_ALIGNED(node_start_pfn(nid), MAX_ORDER_NR_PAGES) ||
156 !IS_ALIGNED(node_end_pfn(nid), MAX_ORDER_NR_PAGES))
157 nr_pages += MAX_ORDER_NR_PAGES;
158
159 table_size = get_entry_size() * nr_pages;
160
161 base = memblock_alloc_try_nid(
162 table_size, PAGE_SIZE, __pa(MAX_DMA_ADDRESS),
163 MEMBLOCK_ALLOC_ACCESSIBLE, nid);
164 if (!base)
165 return -ENOMEM;
166 NODE_DATA(nid)->node_page_ext = base;
167 total_usage += table_size;
168 return 0;
169}
170
171void __init page_ext_init_flatmem(void)
172{
173
174 int nid, fail;
175
176 if (!invoke_need_callbacks())
177 return;
178
179 for_each_online_node(nid) {
180 fail = alloc_node_page_ext(nid);
181 if (fail)
182 goto fail;
183 }
184 pr_info("allocated %ld bytes of page_ext\n", total_usage);
185 invoke_init_callbacks();
186 return;
187
188fail:
189 pr_crit("allocation of page_ext failed.\n");
190 panic("Out of memory");
191}
192
193#else
194
195struct page_ext *lookup_page_ext(const struct page *page)
196{
197 unsigned long pfn = page_to_pfn(page);
198 struct mem_section *section = __pfn_to_section(pfn);
199
200
201
202
203
204
205 if (!section->page_ext)
206 return NULL;
207 return get_entry(section->page_ext, pfn);
208}
209
210static void *__meminit alloc_page_ext(size_t size, int nid)
211{
212 gfp_t flags = GFP_KERNEL | __GFP_ZERO | __GFP_NOWARN;
213 void *addr = NULL;
214
215 addr = alloc_pages_exact_nid(nid, size, flags);
216 if (addr) {
217 kmemleak_alloc(addr, size, 1, flags);
218 return addr;
219 }
220
221 addr = vzalloc_node(size, nid);
222
223 return addr;
224}
225
226static int __meminit init_section_page_ext(unsigned long pfn, int nid)
227{
228 struct mem_section *section;
229 struct page_ext *base;
230 unsigned long table_size;
231
232 section = __pfn_to_section(pfn);
233
234 if (section->page_ext)
235 return 0;
236
237 table_size = get_entry_size() * PAGES_PER_SECTION;
238 base = alloc_page_ext(table_size, nid);
239
240
241
242
243
244
245 kmemleak_not_leak(base);
246
247 if (!base) {
248 pr_err("page ext allocation failure\n");
249 return -ENOMEM;
250 }
251
252
253
254
255
256 pfn &= PAGE_SECTION_MASK;
257 section->page_ext = (void *)base - get_entry_size() * pfn;
258 total_usage += table_size;
259 return 0;
260}
261#ifdef CONFIG_MEMORY_HOTPLUG
262static void free_page_ext(void *addr)
263{
264 if (is_vmalloc_addr(addr)) {
265 vfree(addr);
266 } else {
267 struct page *page = virt_to_page(addr);
268 size_t table_size;
269
270 table_size = get_entry_size() * PAGES_PER_SECTION;
271
272 BUG_ON(PageReserved(page));
273 kmemleak_free(addr);
274 free_pages_exact(addr, table_size);
275 }
276}
277
278static void __free_page_ext(unsigned long pfn)
279{
280 struct mem_section *ms;
281 struct page_ext *base;
282
283 ms = __pfn_to_section(pfn);
284 if (!ms || !ms->page_ext)
285 return;
286 base = get_entry(ms->page_ext, pfn);
287 free_page_ext(base);
288 ms->page_ext = NULL;
289}
290
291static int __meminit online_page_ext(unsigned long start_pfn,
292 unsigned long nr_pages,
293 int nid)
294{
295 unsigned long start, end, pfn;
296 int fail = 0;
297
298 start = SECTION_ALIGN_DOWN(start_pfn);
299 end = SECTION_ALIGN_UP(start_pfn + nr_pages);
300
301 if (nid == NUMA_NO_NODE) {
302
303
304
305
306
307 nid = pfn_to_nid(start_pfn);
308 VM_BUG_ON(!node_state(nid, N_ONLINE));
309 }
310
311 for (pfn = start; !fail && pfn < end; pfn += PAGES_PER_SECTION) {
312 if (!pfn_present(pfn))
313 continue;
314 fail = init_section_page_ext(pfn, nid);
315 }
316 if (!fail)
317 return 0;
318
319
320 for (pfn = start; pfn < end; pfn += PAGES_PER_SECTION)
321 __free_page_ext(pfn);
322
323 return -ENOMEM;
324}
325
326static int __meminit offline_page_ext(unsigned long start_pfn,
327 unsigned long nr_pages, int nid)
328{
329 unsigned long start, end, pfn;
330
331 start = SECTION_ALIGN_DOWN(start_pfn);
332 end = SECTION_ALIGN_UP(start_pfn + nr_pages);
333
334 for (pfn = start; pfn < end; pfn += PAGES_PER_SECTION)
335 __free_page_ext(pfn);
336 return 0;
337
338}
339
340static int __meminit page_ext_callback(struct notifier_block *self,
341 unsigned long action, void *arg)
342{
343 struct memory_notify *mn = arg;
344 int ret = 0;
345
346 switch (action) {
347 case MEM_GOING_ONLINE:
348 ret = online_page_ext(mn->start_pfn,
349 mn->nr_pages, mn->status_change_nid);
350 break;
351 case MEM_OFFLINE:
352 offline_page_ext(mn->start_pfn,
353 mn->nr_pages, mn->status_change_nid);
354 break;
355 case MEM_CANCEL_ONLINE:
356 offline_page_ext(mn->start_pfn,
357 mn->nr_pages, mn->status_change_nid);
358 break;
359 case MEM_GOING_OFFLINE:
360 break;
361 case MEM_ONLINE:
362 case MEM_CANCEL_OFFLINE:
363 break;
364 }
365
366 return notifier_from_errno(ret);
367}
368
369#endif
370
371void __init page_ext_init(void)
372{
373 unsigned long pfn;
374 int nid;
375
376 if (!invoke_need_callbacks())
377 return;
378
379 for_each_node_state(nid, N_MEMORY) {
380 unsigned long start_pfn, end_pfn;
381
382 start_pfn = node_start_pfn(nid);
383 end_pfn = node_end_pfn(nid);
384
385
386
387
388
389 for (pfn = start_pfn; pfn < end_pfn;
390 pfn = ALIGN(pfn + 1, PAGES_PER_SECTION)) {
391
392 if (!pfn_valid(pfn))
393 continue;
394
395
396
397
398
399
400 if (pfn_to_nid(pfn) != nid)
401 continue;
402 if (init_section_page_ext(pfn, nid))
403 goto oom;
404 cond_resched();
405 }
406 }
407 hotplug_memory_notifier(page_ext_callback, 0);
408 pr_info("allocated %ld bytes of page_ext\n", total_usage);
409 invoke_init_callbacks();
410 return;
411
412oom:
413 panic("Out of memory");
414}
415
416void __meminit pgdat_page_ext_init(struct pglist_data *pgdat)
417{
418}
419
420#endif
421