blob: 1983fb1c7026c0ef4c1e705298876d0ab2449c55 [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{
Russ Anderson6bc2e852012-05-10 13:01:46 -070085 unsigned long i, start_aligned, end_aligned;
Yinghai Lu09325872011-02-24 14:43:05 +010086 int order = ilog2(BITS_PER_LONG);
87
88 start_aligned = (start + (BITS_PER_LONG - 1)) & ~(BITS_PER_LONG - 1);
89 end_aligned = end & ~(BITS_PER_LONG - 1);
90
91 if (end_aligned <= start_aligned) {
92 for (i = start; i < end; i++)
93 __free_pages_bootmem(pfn_to_page(i), 0);
94
95 return;
96 }
97
98 for (i = start; i < start_aligned; i++)
99 __free_pages_bootmem(pfn_to_page(i), 0);
100
101 for (i = start_aligned; i < end_aligned; i += BITS_PER_LONG)
102 __free_pages_bootmem(pfn_to_page(i), order);
103
104 for (i = end_aligned; i < end; i++)
105 __free_pages_bootmem(pfn_to_page(i), 0);
106}
107
Tejun Heo64a02da2011-07-12 11:16:01 +0200108unsigned long __init free_low_memory_core_early(int nodeid)
Yinghai Lu09325872011-02-24 14:43:05 +0100109{
Yinghai Lu09325872011-02-24 14:43:05 +0100110 unsigned long count = 0;
Tejun Heo8a9ca342011-07-12 11:16:02 +0200111 phys_addr_t start, end;
112 u64 i;
Yinghai Lu09325872011-02-24 14:43:05 +0100113
Tejun Heo8a9ca342011-07-12 11:16:02 +0200114 /* free reserved array temporarily so that it's treated as free area */
115 memblock_free_reserved_regions();
Yinghai Lu09325872011-02-24 14:43:05 +0100116
Tejun Heo8a9ca342011-07-12 11:16:02 +0200117 for_each_free_mem_range(i, MAX_NUMNODES, &start, &end, NULL) {
118 unsigned long start_pfn = PFN_UP(start);
119 unsigned long end_pfn = min_t(unsigned long,
120 PFN_DOWN(end), max_low_pfn);
121 if (start_pfn < end_pfn) {
122 __free_pages_memory(start_pfn, end_pfn);
123 count += end_pfn - start_pfn;
124 }
Yinghai Lu09325872011-02-24 14:43:05 +0100125 }
126
Tejun Heo8a9ca342011-07-12 11:16:02 +0200127 /* put region array back? */
128 memblock_reserve_reserved_regions();
Yinghai Lu09325872011-02-24 14:43:05 +0100129 return count;
130}
131
132/**
133 * free_all_bootmem_node - release a node's free pages to the buddy allocator
134 * @pgdat: node to be released
135 *
136 * Returns the number of pages actually released.
137 */
138unsigned long __init free_all_bootmem_node(pg_data_t *pgdat)
139{
140 register_page_bootmem_info_node(pgdat);
141
Tejun Heo64a02da2011-07-12 11:16:01 +0200142 /* free_low_memory_core_early(MAX_NUMNODES) will be called later */
Yinghai Lu09325872011-02-24 14:43:05 +0100143 return 0;
144}
145
146/**
147 * free_all_bootmem - release free pages to the buddy allocator
148 *
149 * Returns the number of pages actually released.
150 */
151unsigned long __init free_all_bootmem(void)
152{
153 /*
154 * We need to use MAX_NUMNODES instead of NODE_DATA(0)->node_id
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300155 * because in some case like Node0 doesn't have RAM installed
Yinghai Lu09325872011-02-24 14:43:05 +0100156 * low ram will be on Node1
157 * Use MAX_NUMNODES will make sure all ranges in early_node_map[]
158 * will be used instead of only Node0 related
159 */
Tejun Heo64a02da2011-07-12 11:16:01 +0200160 return free_low_memory_core_early(MAX_NUMNODES);
Yinghai Lu09325872011-02-24 14:43:05 +0100161}
162
163/**
164 * free_bootmem_node - mark a page range as usable
165 * @pgdat: node the range resides on
166 * @physaddr: starting address of the range
167 * @size: size of the range in bytes
168 *
169 * Partial pages will be considered reserved and left as they are.
170 *
171 * The range must reside completely on the specified node.
172 */
173void __init free_bootmem_node(pg_data_t *pgdat, unsigned long physaddr,
174 unsigned long size)
175{
176 kmemleak_free_part(__va(physaddr), size);
Tejun Heo24aa0782011-07-12 11:16:06 +0200177 memblock_free(physaddr, size);
Yinghai Lu09325872011-02-24 14:43:05 +0100178}
179
180/**
181 * free_bootmem - mark a page range as usable
182 * @addr: starting address of the range
183 * @size: size of the range in bytes
184 *
185 * Partial pages will be considered reserved and left as they are.
186 *
187 * The range must be contiguous but may span node boundaries.
188 */
189void __init free_bootmem(unsigned long addr, unsigned long size)
190{
191 kmemleak_free_part(__va(addr), size);
Tejun Heo24aa0782011-07-12 11:16:06 +0200192 memblock_free(addr, size);
Yinghai Lu09325872011-02-24 14:43:05 +0100193}
194
195static void * __init ___alloc_bootmem_nopanic(unsigned long size,
196 unsigned long align,
197 unsigned long goal,
198 unsigned long limit)
199{
200 void *ptr;
201
202 if (WARN_ON_ONCE(slab_is_available()))
203 return kzalloc(size, GFP_NOWAIT);
204
205restart:
206
207 ptr = __alloc_memory_core_early(MAX_NUMNODES, size, align, goal, limit);
208
209 if (ptr)
210 return ptr;
211
212 if (goal != 0) {
213 goal = 0;
214 goto restart;
215 }
216
217 return NULL;
218}
219
220/**
221 * __alloc_bootmem_nopanic - allocate boot memory without panicking
222 * @size: size of the request in bytes
223 * @align: alignment of the region
224 * @goal: preferred starting address of the region
225 *
226 * The goal is dropped if it can not be satisfied and the allocation will
227 * fall back to memory below @goal.
228 *
229 * Allocation may happen on any node in the system.
230 *
231 * Returns NULL on failure.
232 */
233void * __init __alloc_bootmem_nopanic(unsigned long size, unsigned long align,
234 unsigned long goal)
235{
236 unsigned long limit = -1UL;
237
238 return ___alloc_bootmem_nopanic(size, align, goal, limit);
239}
240
241static void * __init ___alloc_bootmem(unsigned long size, unsigned long align,
242 unsigned long goal, unsigned long limit)
243{
244 void *mem = ___alloc_bootmem_nopanic(size, align, goal, limit);
245
246 if (mem)
247 return mem;
248 /*
249 * Whoops, we cannot satisfy the allocation request.
250 */
251 printk(KERN_ALERT "bootmem alloc of %lu bytes failed!\n", size);
252 panic("Out of memory");
253 return NULL;
254}
255
256/**
257 * __alloc_bootmem - allocate boot memory
258 * @size: size of the request in bytes
259 * @align: alignment of the region
260 * @goal: preferred starting address of the region
261 *
262 * The goal is dropped if it can not be satisfied and the allocation will
263 * fall back to memory below @goal.
264 *
265 * Allocation may happen on any node in the system.
266 *
267 * The function panics if the request can not be satisfied.
268 */
269void * __init __alloc_bootmem(unsigned long size, unsigned long align,
270 unsigned long goal)
271{
272 unsigned long limit = -1UL;
273
274 return ___alloc_bootmem(size, align, goal, limit);
275}
276
277/**
278 * __alloc_bootmem_node - allocate boot memory from a specific node
279 * @pgdat: node to allocate from
280 * @size: size of the request in bytes
281 * @align: alignment of the region
282 * @goal: preferred starting address of the region
283 *
284 * The goal is dropped if it can not be satisfied and the allocation will
285 * fall back to memory below @goal.
286 *
287 * Allocation may fall back to any node in the system if the specified node
288 * can not hold the requested memory.
289 *
290 * The function panics if the request can not be satisfied.
291 */
292void * __init __alloc_bootmem_node(pg_data_t *pgdat, unsigned long size,
293 unsigned long align, unsigned long goal)
294{
295 void *ptr;
296
297 if (WARN_ON_ONCE(slab_is_available()))
298 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
299
David Miller4e1c2b22012-04-25 16:10:50 -0400300again:
Yinghai Lu09325872011-02-24 14:43:05 +0100301 ptr = __alloc_memory_core_early(pgdat->node_id, size, align,
302 goal, -1ULL);
303 if (ptr)
304 return ptr;
305
David Miller4e1c2b22012-04-25 16:10:50 -0400306 ptr = __alloc_memory_core_early(MAX_NUMNODES, size, align,
307 goal, -1ULL);
308 if (!ptr && goal) {
309 goal = 0;
310 goto again;
311 }
312 return ptr;
Yinghai Lu09325872011-02-24 14:43:05 +0100313}
314
315void * __init __alloc_bootmem_node_high(pg_data_t *pgdat, unsigned long size,
316 unsigned long align, unsigned long goal)
317{
Yinghai Lu09325872011-02-24 14:43:05 +0100318 return __alloc_bootmem_node(pgdat, size, align, goal);
Yinghai Lu09325872011-02-24 14:43:05 +0100319}
320
321#ifdef CONFIG_SPARSEMEM
322/**
323 * alloc_bootmem_section - allocate boot memory from a specific section
324 * @size: size of the request in bytes
325 * @section_nr: sparse map section to allocate from
326 *
327 * Return NULL on failure.
328 */
329void * __init alloc_bootmem_section(unsigned long size,
330 unsigned long section_nr)
331{
332 unsigned long pfn, goal, limit;
333
334 pfn = section_nr_to_pfn(section_nr);
335 goal = pfn << PAGE_SHIFT;
336 limit = section_nr_to_pfn(section_nr + 1) << PAGE_SHIFT;
337
338 return __alloc_memory_core_early(early_pfn_to_nid(pfn), size,
339 SMP_CACHE_BYTES, goal, limit);
340}
341#endif
342
343void * __init __alloc_bootmem_node_nopanic(pg_data_t *pgdat, unsigned long size,
344 unsigned long align, unsigned long goal)
345{
346 void *ptr;
347
348 if (WARN_ON_ONCE(slab_is_available()))
349 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
350
351 ptr = __alloc_memory_core_early(pgdat->node_id, size, align,
352 goal, -1ULL);
353 if (ptr)
354 return ptr;
355
356 return __alloc_bootmem_nopanic(size, align, goal);
357}
358
359#ifndef ARCH_LOW_ADDRESS_LIMIT
360#define ARCH_LOW_ADDRESS_LIMIT 0xffffffffUL
361#endif
362
363/**
364 * __alloc_bootmem_low - allocate low boot memory
365 * @size: size of the request in bytes
366 * @align: alignment of the region
367 * @goal: preferred starting address of the region
368 *
369 * The goal is dropped if it can not be satisfied and the allocation will
370 * fall back to memory below @goal.
371 *
372 * Allocation may happen on any node in the system.
373 *
374 * The function panics if the request can not be satisfied.
375 */
376void * __init __alloc_bootmem_low(unsigned long size, unsigned long align,
377 unsigned long goal)
378{
379 return ___alloc_bootmem(size, align, goal, ARCH_LOW_ADDRESS_LIMIT);
380}
381
382/**
383 * __alloc_bootmem_low_node - allocate low boot memory from a specific node
384 * @pgdat: node to allocate from
385 * @size: size of the request in bytes
386 * @align: alignment of the region
387 * @goal: preferred starting address of the region
388 *
389 * The goal is dropped if it can not be satisfied and the allocation will
390 * fall back to memory below @goal.
391 *
392 * Allocation may fall back to any node in the system if the specified node
393 * can not hold the requested memory.
394 *
395 * The function panics if the request can not be satisfied.
396 */
397void * __init __alloc_bootmem_low_node(pg_data_t *pgdat, unsigned long size,
398 unsigned long align, unsigned long goal)
399{
400 void *ptr;
401
402 if (WARN_ON_ONCE(slab_is_available()))
403 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
404
405 ptr = __alloc_memory_core_early(pgdat->node_id, size, align,
406 goal, ARCH_LOW_ADDRESS_LIMIT);
407 if (ptr)
408 return ptr;
409
410 return __alloc_memory_core_early(MAX_NUMNODES, size, align,
411 goal, ARCH_LOW_ADDRESS_LIMIT);
412}