blob: e53bb8a256b196018c26899427897a0f471d8193 [file] [log] [blame]
Yinghai Lu09325872011-02-24 14:43:05 +01001/*
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>
Paul Gortmakerb95f1b312011-10-16 02:01:52 -040015#include <linux/export.h>
Yinghai Lu09325872011-02-24 14:43:05 +010016#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
Yinghai Lue782ab42011-02-24 14:43:06 +010026#ifndef CONFIG_NEED_MULTIPLE_NODES
27struct pglist_data __refdata contig_page_data;
28EXPORT_SYMBOL(contig_page_data);
29#endif
30
Yinghai Lu09325872011-02-24 14:43:05 +010031unsigned long max_low_pfn;
32unsigned long min_low_pfn;
33unsigned long max_pfn;
34
Yinghai Lu8bc1f912011-02-24 14:43:06 +010035static void * __init __alloc_memory_core_early(int nid, u64 size, u64 align,
36 u64 goal, u64 limit)
37{
38 void *ptr;
39 u64 addr;
40
41 if (limit > memblock.current_limit)
42 limit = memblock.current_limit;
43
Tejun Heoeb40c4c2011-07-12 10:46:35 +020044 addr = memblock_find_in_range_node(goal, limit, size, align, nid);
Tejun Heo1f5026a2011-07-12 09:58:09 +020045 if (!addr)
Yinghai Lu8bc1f912011-02-24 14:43:06 +010046 return NULL;
47
48 ptr = phys_to_virt(addr);
49 memset(ptr, 0, size);
Tejun Heo24aa0782011-07-12 11:16:06 +020050 memblock_reserve(addr, size);
Yinghai Lu8bc1f912011-02-24 14:43:06 +010051 /*
52 * The min_count is set to 0 so that bootmem allocated blocks
53 * are never reported as leaks.
54 */
55 kmemleak_alloc(ptr, size, 0, 0);
56 return ptr;
57}
58
Yinghai Lu09325872011-02-24 14:43:05 +010059/*
60 * free_bootmem_late - free bootmem pages directly to page allocator
61 * @addr: starting address of the range
62 * @size: size of the range in bytes
63 *
64 * This is only useful when the bootmem allocator has already been torn
65 * down, but we are still initializing the system. Pages are given directly
66 * to the page allocator, no bootmem metadata is updated because it is gone.
67 */
68void __init free_bootmem_late(unsigned long addr, unsigned long size)
69{
70 unsigned long cursor, end;
71
72 kmemleak_free_part(__va(addr), size);
73
74 cursor = PFN_UP(addr);
75 end = PFN_DOWN(addr + size);
76
77 for (; cursor < end; cursor++) {
78 __free_pages_bootmem(pfn_to_page(cursor), 0);
79 totalram_pages++;
80 }
81}
82
83static void __init __free_pages_memory(unsigned long start, unsigned long end)
84{
85 int i;
86 unsigned long start_aligned, end_aligned;
87 int order = ilog2(BITS_PER_LONG);
88
89 start_aligned = (start + (BITS_PER_LONG - 1)) & ~(BITS_PER_LONG - 1);
90 end_aligned = end & ~(BITS_PER_LONG - 1);
91
92 if (end_aligned <= start_aligned) {
93 for (i = start; i < end; i++)
94 __free_pages_bootmem(pfn_to_page(i), 0);
95
96 return;
97 }
98
99 for (i = start; i < start_aligned; i++)
100 __free_pages_bootmem(pfn_to_page(i), 0);
101
102 for (i = start_aligned; i < end_aligned; i += BITS_PER_LONG)
103 __free_pages_bootmem(pfn_to_page(i), order);
104
105 for (i = end_aligned; i < end; i++)
106 __free_pages_bootmem(pfn_to_page(i), 0);
107}
108
Tejun Heo64a02da2011-07-12 11:16:01 +0200109unsigned long __init free_low_memory_core_early(int nodeid)
Yinghai Lu09325872011-02-24 14:43:05 +0100110{
Yinghai Lu09325872011-02-24 14:43:05 +0100111 unsigned long count = 0;
Tejun Heo8a9ca342011-07-12 11:16:02 +0200112 phys_addr_t start, end;
113 u64 i;
Yinghai Lu09325872011-02-24 14:43:05 +0100114
Tejun Heo8a9ca342011-07-12 11:16:02 +0200115 /* free reserved array temporarily so that it's treated as free area */
116 memblock_free_reserved_regions();
Yinghai Lu09325872011-02-24 14:43:05 +0100117
Tejun Heo8a9ca342011-07-12 11:16:02 +0200118 for_each_free_mem_range(i, MAX_NUMNODES, &start, &end, NULL) {
119 unsigned long start_pfn = PFN_UP(start);
120 unsigned long end_pfn = min_t(unsigned long,
121 PFN_DOWN(end), max_low_pfn);
122 if (start_pfn < end_pfn) {
123 __free_pages_memory(start_pfn, end_pfn);
124 count += end_pfn - start_pfn;
125 }
Yinghai Lu09325872011-02-24 14:43:05 +0100126 }
127
Tejun Heo8a9ca342011-07-12 11:16:02 +0200128 /* put region array back? */
129 memblock_reserve_reserved_regions();
Yinghai Lu09325872011-02-24 14:43:05 +0100130 return count;
131}
132
133/**
134 * free_all_bootmem_node - release a node's free pages to the buddy allocator
135 * @pgdat: node to be released
136 *
137 * Returns the number of pages actually released.
138 */
139unsigned long __init free_all_bootmem_node(pg_data_t *pgdat)
140{
141 register_page_bootmem_info_node(pgdat);
142
Tejun Heo64a02da2011-07-12 11:16:01 +0200143 /* free_low_memory_core_early(MAX_NUMNODES) will be called later */
Yinghai Lu09325872011-02-24 14:43:05 +0100144 return 0;
145}
146
147/**
148 * free_all_bootmem - release free pages to the buddy allocator
149 *
150 * Returns the number of pages actually released.
151 */
152unsigned long __init free_all_bootmem(void)
153{
154 /*
155 * We need to use MAX_NUMNODES instead of NODE_DATA(0)->node_id
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300156 * because in some case like Node0 doesn't have RAM installed
Yinghai Lu09325872011-02-24 14:43:05 +0100157 * low ram will be on Node1
158 * Use MAX_NUMNODES will make sure all ranges in early_node_map[]
159 * will be used instead of only Node0 related
160 */
Tejun Heo64a02da2011-07-12 11:16:01 +0200161 return free_low_memory_core_early(MAX_NUMNODES);
Yinghai Lu09325872011-02-24 14:43:05 +0100162}
163
164/**
165 * free_bootmem_node - mark a page range as usable
166 * @pgdat: node the range resides on
167 * @physaddr: starting address of the range
168 * @size: size of the range in bytes
169 *
170 * Partial pages will be considered reserved and left as they are.
171 *
172 * The range must reside completely on the specified node.
173 */
174void __init free_bootmem_node(pg_data_t *pgdat, unsigned long physaddr,
175 unsigned long size)
176{
177 kmemleak_free_part(__va(physaddr), size);
Tejun Heo24aa0782011-07-12 11:16:06 +0200178 memblock_free(physaddr, size);
Yinghai Lu09325872011-02-24 14:43:05 +0100179}
180
181/**
182 * free_bootmem - mark a page range as usable
183 * @addr: starting address of the range
184 * @size: size of the range in bytes
185 *
186 * Partial pages will be considered reserved and left as they are.
187 *
188 * The range must be contiguous but may span node boundaries.
189 */
190void __init free_bootmem(unsigned long addr, unsigned long size)
191{
192 kmemleak_free_part(__va(addr), size);
Tejun Heo24aa0782011-07-12 11:16:06 +0200193 memblock_free(addr, size);
Yinghai Lu09325872011-02-24 14:43:05 +0100194}
195
196static void * __init ___alloc_bootmem_nopanic(unsigned long size,
197 unsigned long align,
198 unsigned long goal,
199 unsigned long limit)
200{
201 void *ptr;
202
203 if (WARN_ON_ONCE(slab_is_available()))
204 return kzalloc(size, GFP_NOWAIT);
205
206restart:
207
208 ptr = __alloc_memory_core_early(MAX_NUMNODES, size, align, goal, limit);
209
210 if (ptr)
211 return ptr;
212
213 if (goal != 0) {
214 goal = 0;
215 goto restart;
216 }
217
218 return NULL;
219}
220
221/**
222 * __alloc_bootmem_nopanic - allocate boot memory without panicking
223 * @size: size of the request in bytes
224 * @align: alignment of the region
225 * @goal: preferred starting address of the region
226 *
227 * The goal is dropped if it can not be satisfied and the allocation will
228 * fall back to memory below @goal.
229 *
230 * Allocation may happen on any node in the system.
231 *
232 * Returns NULL on failure.
233 */
234void * __init __alloc_bootmem_nopanic(unsigned long size, unsigned long align,
235 unsigned long goal)
236{
237 unsigned long limit = -1UL;
238
239 return ___alloc_bootmem_nopanic(size, align, goal, limit);
240}
241
242static void * __init ___alloc_bootmem(unsigned long size, unsigned long align,
243 unsigned long goal, unsigned long limit)
244{
245 void *mem = ___alloc_bootmem_nopanic(size, align, goal, limit);
246
247 if (mem)
248 return mem;
249 /*
250 * Whoops, we cannot satisfy the allocation request.
251 */
252 printk(KERN_ALERT "bootmem alloc of %lu bytes failed!\n", size);
253 panic("Out of memory");
254 return NULL;
255}
256
257/**
258 * __alloc_bootmem - allocate boot memory
259 * @size: size of the request in bytes
260 * @align: alignment of the region
261 * @goal: preferred starting address of the region
262 *
263 * The goal is dropped if it can not be satisfied and the allocation will
264 * fall back to memory below @goal.
265 *
266 * Allocation may happen on any node in the system.
267 *
268 * The function panics if the request can not be satisfied.
269 */
270void * __init __alloc_bootmem(unsigned long size, unsigned long align,
271 unsigned long goal)
272{
273 unsigned long limit = -1UL;
274
275 return ___alloc_bootmem(size, align, goal, limit);
276}
277
278/**
279 * __alloc_bootmem_node - allocate boot memory from a specific node
280 * @pgdat: node to allocate from
281 * @size: size of the request in bytes
282 * @align: alignment of the region
283 * @goal: preferred starting address of the region
284 *
285 * The goal is dropped if it can not be satisfied and the allocation will
286 * fall back to memory below @goal.
287 *
288 * Allocation may fall back to any node in the system if the specified node
289 * can not hold the requested memory.
290 *
291 * The function panics if the request can not be satisfied.
292 */
293void * __init __alloc_bootmem_node(pg_data_t *pgdat, unsigned long size,
294 unsigned long align, unsigned long goal)
295{
296 void *ptr;
297
298 if (WARN_ON_ONCE(slab_is_available()))
299 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
300
David Miller4e1c2b22012-04-25 16:10:50 -0400301again:
Yinghai Lu09325872011-02-24 14:43:05 +0100302 ptr = __alloc_memory_core_early(pgdat->node_id, size, align,
303 goal, -1ULL);
304 if (ptr)
305 return ptr;
306
David Miller4e1c2b22012-04-25 16:10:50 -0400307 ptr = __alloc_memory_core_early(MAX_NUMNODES, size, align,
308 goal, -1ULL);
309 if (!ptr && goal) {
310 goal = 0;
311 goto again;
312 }
313 return ptr;
Yinghai Lu09325872011-02-24 14:43:05 +0100314}
315
316void * __init __alloc_bootmem_node_high(pg_data_t *pgdat, unsigned long size,
317 unsigned long align, unsigned long goal)
318{
Yinghai Lu09325872011-02-24 14:43:05 +0100319 return __alloc_bootmem_node(pgdat, size, align, goal);
Yinghai Lu09325872011-02-24 14:43:05 +0100320}
321
322#ifdef CONFIG_SPARSEMEM
323/**
324 * alloc_bootmem_section - allocate boot memory from a specific section
325 * @size: size of the request in bytes
326 * @section_nr: sparse map section to allocate from
327 *
328 * Return NULL on failure.
329 */
330void * __init alloc_bootmem_section(unsigned long size,
331 unsigned long section_nr)
332{
333 unsigned long pfn, goal, limit;
334
335 pfn = section_nr_to_pfn(section_nr);
336 goal = pfn << PAGE_SHIFT;
337 limit = section_nr_to_pfn(section_nr + 1) << PAGE_SHIFT;
338
339 return __alloc_memory_core_early(early_pfn_to_nid(pfn), size,
340 SMP_CACHE_BYTES, goal, limit);
341}
342#endif
343
344void * __init __alloc_bootmem_node_nopanic(pg_data_t *pgdat, unsigned long size,
345 unsigned long align, unsigned long goal)
346{
347 void *ptr;
348
349 if (WARN_ON_ONCE(slab_is_available()))
350 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
351
352 ptr = __alloc_memory_core_early(pgdat->node_id, size, align,
353 goal, -1ULL);
354 if (ptr)
355 return ptr;
356
357 return __alloc_bootmem_nopanic(size, align, goal);
358}
359
360#ifndef ARCH_LOW_ADDRESS_LIMIT
361#define ARCH_LOW_ADDRESS_LIMIT 0xffffffffUL
362#endif
363
364/**
365 * __alloc_bootmem_low - allocate low boot memory
366 * @size: size of the request in bytes
367 * @align: alignment of the region
368 * @goal: preferred starting address of the region
369 *
370 * The goal is dropped if it can not be satisfied and the allocation will
371 * fall back to memory below @goal.
372 *
373 * Allocation may happen on any node in the system.
374 *
375 * The function panics if the request can not be satisfied.
376 */
377void * __init __alloc_bootmem_low(unsigned long size, unsigned long align,
378 unsigned long goal)
379{
380 return ___alloc_bootmem(size, align, goal, ARCH_LOW_ADDRESS_LIMIT);
381}
382
383/**
384 * __alloc_bootmem_low_node - allocate low boot memory from a specific node
385 * @pgdat: node to allocate from
386 * @size: size of the request in bytes
387 * @align: alignment of the region
388 * @goal: preferred starting address of the region
389 *
390 * The goal is dropped if it can not be satisfied and the allocation will
391 * fall back to memory below @goal.
392 *
393 * Allocation may fall back to any node in the system if the specified node
394 * can not hold the requested memory.
395 *
396 * The function panics if the request can not be satisfied.
397 */
398void * __init __alloc_bootmem_low_node(pg_data_t *pgdat, unsigned long size,
399 unsigned long align, unsigned long goal)
400{
401 void *ptr;
402
403 if (WARN_ON_ONCE(slab_is_available()))
404 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
405
406 ptr = __alloc_memory_core_early(pgdat->node_id, size, align,
407 goal, ARCH_LOW_ADDRESS_LIMIT);
408 if (ptr)
409 return ptr;
410
411 return __alloc_memory_core_early(MAX_NUMNODES, size, align,
412 goal, ARCH_LOW_ADDRESS_LIMIT);
413}