Commit | Line | Data |
---|---|---|
09325873 YL |
1 | /* |
2 | * bootmem - A boot-time physical memory allocator and configurator | |
3 | * | |
4 | * Copyright (C) 1999 Ingo Molnar | |
5 | * 1999 Kanoj Sarcar, SGI | |
6 | * 2008 Johannes Weiner | |
7 | * | |
8 | * Access to this subsystem has to be serialized externally (which is true | |
9 | * for the boot process anyway). | |
10 | */ | |
11 | #include <linux/init.h> | |
12 | #include <linux/pfn.h> | |
13 | #include <linux/slab.h> | |
14 | #include <linux/bootmem.h> | |
b95f1b31 | 15 | #include <linux/export.h> |
09325873 YL |
16 | #include <linux/kmemleak.h> |
17 | #include <linux/range.h> | |
18 | #include <linux/memblock.h> | |
19 | ||
20 | #include <asm/bug.h> | |
21 | #include <asm/io.h> | |
22 | #include <asm/processor.h> | |
23 | ||
24 | #include "internal.h" | |
25 | ||
e782ab42 YL |
26 | #ifndef CONFIG_NEED_MULTIPLE_NODES |
27 | struct pglist_data __refdata contig_page_data; | |
28 | EXPORT_SYMBOL(contig_page_data); | |
29 | #endif | |
30 | ||
09325873 YL |
31 | unsigned long max_low_pfn; |
32 | unsigned long min_low_pfn; | |
33 | unsigned long max_pfn; | |
34 | ||
8bc1f91e YL |
35 | static void * __init __alloc_memory_core_early(int nid, u64 size, u64 align, |
36 | u64 goal, u64 limit) | |
37 | { | |
38 | void *ptr; | |
39 | u64 addr; | |
a3f5bafc | 40 | ulong flags = choose_memblock_flags(); |
8bc1f91e YL |
41 | |
42 | if (limit > memblock.current_limit) | |
43 | limit = memblock.current_limit; | |
44 | ||
a3f5bafc | 45 | again: |
fc6daaf9 | 46 | addr = memblock_find_in_range_node(size, align, goal, limit, nid, |
a3f5bafc TL |
47 | flags); |
48 | if (!addr && (flags & MEMBLOCK_MIRROR)) { | |
49 | flags &= ~MEMBLOCK_MIRROR; | |
50 | pr_warn("Could not allocate %pap bytes of mirrored memory\n", | |
51 | &size); | |
52 | goto again; | |
53 | } | |
1f5026a7 | 54 | if (!addr) |
8bc1f91e YL |
55 | return NULL; |
56 | ||
87379ec8 PH |
57 | if (memblock_reserve(addr, size)) |
58 | return NULL; | |
59 | ||
8bc1f91e YL |
60 | ptr = phys_to_virt(addr); |
61 | memset(ptr, 0, size); | |
8bc1f91e YL |
62 | /* |
63 | * The min_count is set to 0 so that bootmem allocated blocks | |
64 | * are never reported as leaks. | |
65 | */ | |
66 | kmemleak_alloc(ptr, size, 0, 0); | |
67 | return ptr; | |
68 | } | |
69 | ||
09325873 YL |
70 | /* |
71 | * free_bootmem_late - free bootmem pages directly to page allocator | |
72 | * @addr: starting address of the range | |
73 | * @size: size of the range in bytes | |
74 | * | |
75 | * This is only useful when the bootmem allocator has already been torn | |
76 | * down, but we are still initializing the system. Pages are given directly | |
77 | * to the page allocator, no bootmem metadata is updated because it is gone. | |
78 | */ | |
79 | void __init free_bootmem_late(unsigned long addr, unsigned long size) | |
80 | { | |
81 | unsigned long cursor, end; | |
82 | ||
83 | kmemleak_free_part(__va(addr), size); | |
84 | ||
85 | cursor = PFN_UP(addr); | |
86 | end = PFN_DOWN(addr + size); | |
87 | ||
88 | for (; cursor < end; cursor++) { | |
d70ddd7a | 89 | __free_pages_bootmem(pfn_to_page(cursor), cursor, 0); |
09325873 YL |
90 | totalram_pages++; |
91 | } | |
92 | } | |
93 | ||
94 | static void __init __free_pages_memory(unsigned long start, unsigned long end) | |
95 | { | |
309d0b39 | 96 | int order; |
09325873 | 97 | |
309d0b39 RH |
98 | while (start < end) { |
99 | order = min(MAX_ORDER - 1UL, __ffs(start)); | |
09325873 | 100 | |
309d0b39 RH |
101 | while (start + (1UL << order) > end) |
102 | order--; | |
09325873 | 103 | |
d70ddd7a | 104 | __free_pages_bootmem(pfn_to_page(start), start, order); |
09325873 | 105 | |
309d0b39 RH |
106 | start += (1UL << order); |
107 | } | |
09325873 YL |
108 | } |
109 | ||
29f67386 YL |
110 | static unsigned long __init __free_memory_core(phys_addr_t start, |
111 | phys_addr_t end) | |
112 | { | |
113 | unsigned long start_pfn = PFN_UP(start); | |
114 | unsigned long end_pfn = min_t(unsigned long, | |
115 | PFN_DOWN(end), max_low_pfn); | |
116 | ||
117 | if (start_pfn > end_pfn) | |
118 | return 0; | |
119 | ||
120 | __free_pages_memory(start_pfn, end_pfn); | |
121 | ||
122 | return end_pfn - start_pfn; | |
123 | } | |
124 | ||
b4def350 | 125 | static unsigned long __init free_low_memory_core_early(void) |
09325873 | 126 | { |
09325873 | 127 | unsigned long count = 0; |
354f17e1 | 128 | phys_addr_t start, end; |
8a9ca34c TH |
129 | u64 i; |
130 | ||
0a313a99 XQ |
131 | memblock_clear_hotplug(0, -1); |
132 | ||
92923ca3 NZ |
133 | for_each_reserved_mem_region(i, &start, &end) |
134 | reserve_bootmem_region(start, end); | |
135 | ||
fc6daaf9 TL |
136 | for_each_free_mem_range(i, NUMA_NO_NODE, MEMBLOCK_NONE, &start, &end, |
137 | NULL) | |
29f67386 YL |
138 | count += __free_memory_core(start, end); |
139 | ||
5e270e25 | 140 | #ifdef CONFIG_ARCH_DISCARD_MEMBLOCK |
354f17e1 PH |
141 | { |
142 | phys_addr_t size; | |
143 | ||
144 | /* Free memblock.reserved array if it was allocated */ | |
145 | size = get_allocated_memblock_reserved_regions_info(&start); | |
146 | if (size) | |
147 | count += __free_memory_core(start, start + size); | |
148 | ||
149 | /* Free memblock.memory array if it was allocated */ | |
150 | size = get_allocated_memblock_memory_regions_info(&start); | |
151 | if (size) | |
152 | count += __free_memory_core(start, start + size); | |
153 | } | |
5e270e25 PH |
154 | #endif |
155 | ||
09325873 YL |
156 | return count; |
157 | } | |
158 | ||
7b4b2a0d JL |
159 | static int reset_managed_pages_done __initdata; |
160 | ||
f784a3f1 | 161 | void reset_node_managed_pages(pg_data_t *pgdat) |
9feedc9d JL |
162 | { |
163 | struct zone *z; | |
164 | ||
9feedc9d | 165 | for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++) |
7b4b2a0d JL |
166 | z->managed_pages = 0; |
167 | } | |
168 | ||
169 | void __init reset_all_zones_managed_pages(void) | |
170 | { | |
171 | struct pglist_data *pgdat; | |
172 | ||
f784a3f1 TC |
173 | if (reset_managed_pages_done) |
174 | return; | |
175 | ||
7b4b2a0d JL |
176 | for_each_online_pgdat(pgdat) |
177 | reset_node_managed_pages(pgdat); | |
f784a3f1 | 178 | |
7b4b2a0d | 179 | reset_managed_pages_done = 1; |
9feedc9d JL |
180 | } |
181 | ||
09325873 YL |
182 | /** |
183 | * free_all_bootmem - release free pages to the buddy allocator | |
184 | * | |
185 | * Returns the number of pages actually released. | |
186 | */ | |
187 | unsigned long __init free_all_bootmem(void) | |
188 | { | |
0c988534 JL |
189 | unsigned long pages; |
190 | ||
7b4b2a0d | 191 | reset_all_zones_managed_pages(); |
9feedc9d | 192 | |
09325873 | 193 | /* |
b1154233 | 194 | * We need to use NUMA_NO_NODE instead of NODE_DATA(0)->node_id |
25985edc | 195 | * because in some case like Node0 doesn't have RAM installed |
09325873 | 196 | * low ram will be on Node1 |
09325873 | 197 | */ |
0c988534 JL |
198 | pages = free_low_memory_core_early(); |
199 | totalram_pages += pages; | |
200 | ||
201 | return pages; | |
09325873 YL |
202 | } |
203 | ||
204 | /** | |
205 | * free_bootmem_node - mark a page range as usable | |
206 | * @pgdat: node the range resides on | |
207 | * @physaddr: starting address of the range | |
208 | * @size: size of the range in bytes | |
209 | * | |
210 | * Partial pages will be considered reserved and left as they are. | |
211 | * | |
212 | * The range must reside completely on the specified node. | |
213 | */ | |
214 | void __init free_bootmem_node(pg_data_t *pgdat, unsigned long physaddr, | |
215 | unsigned long size) | |
216 | { | |
24aa0788 | 217 | memblock_free(physaddr, size); |
09325873 YL |
218 | } |
219 | ||
220 | /** | |
221 | * free_bootmem - mark a page range as usable | |
222 | * @addr: starting address of the range | |
223 | * @size: size of the range in bytes | |
224 | * | |
225 | * Partial pages will be considered reserved and left as they are. | |
226 | * | |
227 | * The range must be contiguous but may span node boundaries. | |
228 | */ | |
229 | void __init free_bootmem(unsigned long addr, unsigned long size) | |
230 | { | |
24aa0788 | 231 | memblock_free(addr, size); |
09325873 YL |
232 | } |
233 | ||
234 | static void * __init ___alloc_bootmem_nopanic(unsigned long size, | |
235 | unsigned long align, | |
236 | unsigned long goal, | |
237 | unsigned long limit) | |
238 | { | |
239 | void *ptr; | |
240 | ||
241 | if (WARN_ON_ONCE(slab_is_available())) | |
242 | return kzalloc(size, GFP_NOWAIT); | |
243 | ||
244 | restart: | |
245 | ||
b1154233 | 246 | ptr = __alloc_memory_core_early(NUMA_NO_NODE, size, align, goal, limit); |
09325873 YL |
247 | |
248 | if (ptr) | |
249 | return ptr; | |
250 | ||
251 | if (goal != 0) { | |
252 | goal = 0; | |
253 | goto restart; | |
254 | } | |
255 | ||
256 | return NULL; | |
257 | } | |
258 | ||
259 | /** | |
260 | * __alloc_bootmem_nopanic - allocate boot memory without panicking | |
261 | * @size: size of the request in bytes | |
262 | * @align: alignment of the region | |
263 | * @goal: preferred starting address of the region | |
264 | * | |
265 | * The goal is dropped if it can not be satisfied and the allocation will | |
266 | * fall back to memory below @goal. | |
267 | * | |
268 | * Allocation may happen on any node in the system. | |
269 | * | |
270 | * Returns NULL on failure. | |
271 | */ | |
272 | void * __init __alloc_bootmem_nopanic(unsigned long size, unsigned long align, | |
273 | unsigned long goal) | |
274 | { | |
275 | unsigned long limit = -1UL; | |
276 | ||
277 | return ___alloc_bootmem_nopanic(size, align, goal, limit); | |
278 | } | |
279 | ||
280 | static void * __init ___alloc_bootmem(unsigned long size, unsigned long align, | |
281 | unsigned long goal, unsigned long limit) | |
282 | { | |
283 | void *mem = ___alloc_bootmem_nopanic(size, align, goal, limit); | |
284 | ||
285 | if (mem) | |
286 | return mem; | |
287 | /* | |
288 | * Whoops, we cannot satisfy the allocation request. | |
289 | */ | |
290 | printk(KERN_ALERT "bootmem alloc of %lu bytes failed!\n", size); | |
291 | panic("Out of memory"); | |
292 | return NULL; | |
293 | } | |
294 | ||
295 | /** | |
296 | * __alloc_bootmem - allocate boot memory | |
297 | * @size: size of the request in bytes | |
298 | * @align: alignment of the region | |
299 | * @goal: preferred starting address of the region | |
300 | * | |
301 | * The goal is dropped if it can not be satisfied and the allocation will | |
302 | * fall back to memory below @goal. | |
303 | * | |
304 | * Allocation may happen on any node in the system. | |
305 | * | |
306 | * The function panics if the request can not be satisfied. | |
307 | */ | |
308 | void * __init __alloc_bootmem(unsigned long size, unsigned long align, | |
309 | unsigned long goal) | |
310 | { | |
311 | unsigned long limit = -1UL; | |
312 | ||
313 | return ___alloc_bootmem(size, align, goal, limit); | |
314 | } | |
315 | ||
99ab7b19 | 316 | void * __init ___alloc_bootmem_node_nopanic(pg_data_t *pgdat, |
ba539868 JW |
317 | unsigned long size, |
318 | unsigned long align, | |
319 | unsigned long goal, | |
320 | unsigned long limit) | |
321 | { | |
322 | void *ptr; | |
323 | ||
324 | again: | |
325 | ptr = __alloc_memory_core_early(pgdat->node_id, size, align, | |
326 | goal, limit); | |
327 | if (ptr) | |
328 | return ptr; | |
329 | ||
b1154233 | 330 | ptr = __alloc_memory_core_early(NUMA_NO_NODE, size, align, |
ba539868 JW |
331 | goal, limit); |
332 | if (ptr) | |
333 | return ptr; | |
334 | ||
335 | if (goal) { | |
336 | goal = 0; | |
337 | goto again; | |
338 | } | |
339 | ||
340 | return NULL; | |
341 | } | |
342 | ||
343 | void * __init __alloc_bootmem_node_nopanic(pg_data_t *pgdat, unsigned long size, | |
344 | unsigned long align, unsigned long goal) | |
345 | { | |
346 | if (WARN_ON_ONCE(slab_is_available())) | |
347 | return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id); | |
348 | ||
349 | return ___alloc_bootmem_node_nopanic(pgdat, size, align, goal, 0); | |
350 | } | |
351 | ||
de498507 | 352 | static void * __init ___alloc_bootmem_node(pg_data_t *pgdat, unsigned long size, |
ba539868 JW |
353 | unsigned long align, unsigned long goal, |
354 | unsigned long limit) | |
355 | { | |
356 | void *ptr; | |
357 | ||
358 | ptr = ___alloc_bootmem_node_nopanic(pgdat, size, align, goal, limit); | |
359 | if (ptr) | |
360 | return ptr; | |
361 | ||
362 | printk(KERN_ALERT "bootmem alloc of %lu bytes failed!\n", size); | |
363 | panic("Out of memory"); | |
364 | return NULL; | |
365 | } | |
366 | ||
09325873 YL |
367 | /** |
368 | * __alloc_bootmem_node - allocate boot memory from a specific node | |
369 | * @pgdat: node to allocate from | |
370 | * @size: size of the request in bytes | |
371 | * @align: alignment of the region | |
372 | * @goal: preferred starting address of the region | |
373 | * | |
374 | * The goal is dropped if it can not be satisfied and the allocation will | |
375 | * fall back to memory below @goal. | |
376 | * | |
377 | * Allocation may fall back to any node in the system if the specified node | |
378 | * can not hold the requested memory. | |
379 | * | |
380 | * The function panics if the request can not be satisfied. | |
381 | */ | |
382 | void * __init __alloc_bootmem_node(pg_data_t *pgdat, unsigned long size, | |
383 | unsigned long align, unsigned long goal) | |
384 | { | |
09325873 YL |
385 | if (WARN_ON_ONCE(slab_is_available())) |
386 | return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id); | |
387 | ||
ba539868 | 388 | return ___alloc_bootmem_node(pgdat, size, align, goal, 0); |
09325873 YL |
389 | } |
390 | ||
391 | void * __init __alloc_bootmem_node_high(pg_data_t *pgdat, unsigned long size, | |
392 | unsigned long align, unsigned long goal) | |
393 | { | |
09325873 | 394 | return __alloc_bootmem_node(pgdat, size, align, goal); |
09325873 YL |
395 | } |
396 | ||
09325873 YL |
397 | #ifndef ARCH_LOW_ADDRESS_LIMIT |
398 | #define ARCH_LOW_ADDRESS_LIMIT 0xffffffffUL | |
399 | #endif | |
400 | ||
401 | /** | |
402 | * __alloc_bootmem_low - allocate low boot memory | |
403 | * @size: size of the request in bytes | |
404 | * @align: alignment of the region | |
405 | * @goal: preferred starting address of the region | |
406 | * | |
407 | * The goal is dropped if it can not be satisfied and the allocation will | |
408 | * fall back to memory below @goal. | |
409 | * | |
410 | * Allocation may happen on any node in the system. | |
411 | * | |
412 | * The function panics if the request can not be satisfied. | |
413 | */ | |
414 | void * __init __alloc_bootmem_low(unsigned long size, unsigned long align, | |
415 | unsigned long goal) | |
416 | { | |
417 | return ___alloc_bootmem(size, align, goal, ARCH_LOW_ADDRESS_LIMIT); | |
418 | } | |
419 | ||
38fa4175 YL |
420 | void * __init __alloc_bootmem_low_nopanic(unsigned long size, |
421 | unsigned long align, | |
422 | unsigned long goal) | |
423 | { | |
424 | return ___alloc_bootmem_nopanic(size, align, goal, | |
425 | ARCH_LOW_ADDRESS_LIMIT); | |
426 | } | |
427 | ||
09325873 YL |
428 | /** |
429 | * __alloc_bootmem_low_node - allocate low boot memory from a specific node | |
430 | * @pgdat: node to allocate from | |
431 | * @size: size of the request in bytes | |
432 | * @align: alignment of the region | |
433 | * @goal: preferred starting address of the region | |
434 | * | |
435 | * The goal is dropped if it can not be satisfied and the allocation will | |
436 | * fall back to memory below @goal. | |
437 | * | |
438 | * Allocation may fall back to any node in the system if the specified node | |
439 | * can not hold the requested memory. | |
440 | * | |
441 | * The function panics if the request can not be satisfied. | |
442 | */ | |
443 | void * __init __alloc_bootmem_low_node(pg_data_t *pgdat, unsigned long size, | |
444 | unsigned long align, unsigned long goal) | |
445 | { | |
09325873 YL |
446 | if (WARN_ON_ONCE(slab_is_available())) |
447 | return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id); | |
448 | ||
ba539868 JW |
449 | return ___alloc_bootmem_node(pgdat, size, align, goal, |
450 | ARCH_LOW_ADDRESS_LIMIT); | |
09325873 | 451 | } |