blob: 714d5d6504708c83c64365618749a84b1da4cd66 [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
Yinghai Lu29f67382012-07-11 14:02:56 -0700108static unsigned long __init __free_memory_core(phys_addr_t start,
109 phys_addr_t end)
110{
111 unsigned long start_pfn = PFN_UP(start);
112 unsigned long end_pfn = min_t(unsigned long,
113 PFN_DOWN(end), max_low_pfn);
114
115 if (start_pfn > end_pfn)
116 return 0;
117
118 __free_pages_memory(start_pfn, end_pfn);
Jianguo Wu7f1290f2012-10-08 16:33:06 -0700119 fixup_zone_present_pages(pfn_to_nid(start >> PAGE_SHIFT),
120 start_pfn, end_pfn);
Yinghai Lu29f67382012-07-11 14:02:56 -0700121
122 return end_pfn - start_pfn;
123}
124
Tejun Heo64a02da2011-07-12 11:16:01 +0200125unsigned long __init free_low_memory_core_early(int nodeid)
Yinghai Lu09325872011-02-24 14:43:05 +0100126{
Yinghai Lu09325872011-02-24 14:43:05 +0100127 unsigned long count = 0;
Yinghai Lu29f67382012-07-11 14:02:56 -0700128 phys_addr_t start, end, size;
Tejun Heo8a9ca342011-07-12 11:16:02 +0200129 u64 i;
Yinghai Lu09325872011-02-24 14:43:05 +0100130
Jianguo Wu7f1290f2012-10-08 16:33:06 -0700131 reset_zone_present_pages();
Yinghai Lu29f67382012-07-11 14:02:56 -0700132 for_each_free_mem_range(i, MAX_NUMNODES, &start, &end, NULL)
133 count += __free_memory_core(start, end);
Yinghai Lu09325872011-02-24 14:43:05 +0100134
Yinghai Lu29f67382012-07-11 14:02:56 -0700135 /* free range that is used for reserved array if we allocate it */
136 size = get_allocated_memblock_reserved_regions_info(&start);
137 if (size)
138 count += __free_memory_core(start, start + size);
Yinghai Lu09325872011-02-24 14:43:05 +0100139
140 return count;
141}
142
143/**
144 * free_all_bootmem_node - release a node's free pages to the buddy allocator
145 * @pgdat: node to be released
146 *
147 * Returns the number of pages actually released.
148 */
149unsigned long __init free_all_bootmem_node(pg_data_t *pgdat)
150{
151 register_page_bootmem_info_node(pgdat);
152
Tejun Heo64a02da2011-07-12 11:16:01 +0200153 /* free_low_memory_core_early(MAX_NUMNODES) will be called later */
Yinghai Lu09325872011-02-24 14:43:05 +0100154 return 0;
155}
156
157/**
158 * free_all_bootmem - release free pages to the buddy allocator
159 *
160 * Returns the number of pages actually released.
161 */
162unsigned long __init free_all_bootmem(void)
163{
164 /*
165 * We need to use MAX_NUMNODES instead of NODE_DATA(0)->node_id
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300166 * because in some case like Node0 doesn't have RAM installed
Yinghai Lu09325872011-02-24 14:43:05 +0100167 * low ram will be on Node1
Yinghai Lu09325872011-02-24 14:43:05 +0100168 */
Tejun Heo64a02da2011-07-12 11:16:01 +0200169 return free_low_memory_core_early(MAX_NUMNODES);
Yinghai Lu09325872011-02-24 14:43:05 +0100170}
171
172/**
173 * free_bootmem_node - mark a page range as usable
174 * @pgdat: node the range resides on
175 * @physaddr: starting address of the range
176 * @size: size of the range in bytes
177 *
178 * Partial pages will be considered reserved and left as they are.
179 *
180 * The range must reside completely on the specified node.
181 */
182void __init free_bootmem_node(pg_data_t *pgdat, unsigned long physaddr,
183 unsigned long size)
184{
185 kmemleak_free_part(__va(physaddr), size);
Tejun Heo24aa0782011-07-12 11:16:06 +0200186 memblock_free(physaddr, size);
Yinghai Lu09325872011-02-24 14:43:05 +0100187}
188
189/**
190 * free_bootmem - mark a page range as usable
191 * @addr: starting address of the range
192 * @size: size of the range in bytes
193 *
194 * Partial pages will be considered reserved and left as they are.
195 *
196 * The range must be contiguous but may span node boundaries.
197 */
198void __init free_bootmem(unsigned long addr, unsigned long size)
199{
200 kmemleak_free_part(__va(addr), size);
Tejun Heo24aa0782011-07-12 11:16:06 +0200201 memblock_free(addr, size);
Yinghai Lu09325872011-02-24 14:43:05 +0100202}
203
204static void * __init ___alloc_bootmem_nopanic(unsigned long size,
205 unsigned long align,
206 unsigned long goal,
207 unsigned long limit)
208{
209 void *ptr;
210
211 if (WARN_ON_ONCE(slab_is_available()))
212 return kzalloc(size, GFP_NOWAIT);
213
214restart:
215
216 ptr = __alloc_memory_core_early(MAX_NUMNODES, size, align, goal, limit);
217
218 if (ptr)
219 return ptr;
220
221 if (goal != 0) {
222 goal = 0;
223 goto restart;
224 }
225
226 return NULL;
227}
228
229/**
230 * __alloc_bootmem_nopanic - allocate boot memory without panicking
231 * @size: size of the request in bytes
232 * @align: alignment of the region
233 * @goal: preferred starting address of the region
234 *
235 * The goal is dropped if it can not be satisfied and the allocation will
236 * fall back to memory below @goal.
237 *
238 * Allocation may happen on any node in the system.
239 *
240 * Returns NULL on failure.
241 */
242void * __init __alloc_bootmem_nopanic(unsigned long size, unsigned long align,
243 unsigned long goal)
244{
245 unsigned long limit = -1UL;
246
247 return ___alloc_bootmem_nopanic(size, align, goal, limit);
248}
249
250static void * __init ___alloc_bootmem(unsigned long size, unsigned long align,
251 unsigned long goal, unsigned long limit)
252{
253 void *mem = ___alloc_bootmem_nopanic(size, align, goal, limit);
254
255 if (mem)
256 return mem;
257 /*
258 * Whoops, we cannot satisfy the allocation request.
259 */
260 printk(KERN_ALERT "bootmem alloc of %lu bytes failed!\n", size);
261 panic("Out of memory");
262 return NULL;
263}
264
265/**
266 * __alloc_bootmem - allocate boot memory
267 * @size: size of the request in bytes
268 * @align: alignment of the region
269 * @goal: preferred starting address of the region
270 *
271 * The goal is dropped if it can not be satisfied and the allocation will
272 * fall back to memory below @goal.
273 *
274 * Allocation may happen on any node in the system.
275 *
276 * The function panics if the request can not be satisfied.
277 */
278void * __init __alloc_bootmem(unsigned long size, unsigned long align,
279 unsigned long goal)
280{
281 unsigned long limit = -1UL;
282
283 return ___alloc_bootmem(size, align, goal, limit);
284}
285
Yinghai Lu99ab7b12012-07-11 14:02:53 -0700286void * __init ___alloc_bootmem_node_nopanic(pg_data_t *pgdat,
Johannes Weinerba539862012-05-29 15:06:35 -0700287 unsigned long size,
288 unsigned long align,
289 unsigned long goal,
290 unsigned long limit)
291{
292 void *ptr;
293
294again:
295 ptr = __alloc_memory_core_early(pgdat->node_id, size, align,
296 goal, limit);
297 if (ptr)
298 return ptr;
299
300 ptr = __alloc_memory_core_early(MAX_NUMNODES, size, align,
301 goal, limit);
302 if (ptr)
303 return ptr;
304
305 if (goal) {
306 goal = 0;
307 goto again;
308 }
309
310 return NULL;
311}
312
313void * __init __alloc_bootmem_node_nopanic(pg_data_t *pgdat, unsigned long size,
314 unsigned long align, unsigned long goal)
315{
316 if (WARN_ON_ONCE(slab_is_available()))
317 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
318
319 return ___alloc_bootmem_node_nopanic(pgdat, size, align, goal, 0);
320}
321
322void * __init ___alloc_bootmem_node(pg_data_t *pgdat, unsigned long size,
323 unsigned long align, unsigned long goal,
324 unsigned long limit)
325{
326 void *ptr;
327
328 ptr = ___alloc_bootmem_node_nopanic(pgdat, size, align, goal, limit);
329 if (ptr)
330 return ptr;
331
332 printk(KERN_ALERT "bootmem alloc of %lu bytes failed!\n", size);
333 panic("Out of memory");
334 return NULL;
335}
336
Yinghai Lu09325872011-02-24 14:43:05 +0100337/**
338 * __alloc_bootmem_node - allocate boot memory from a specific node
339 * @pgdat: node to allocate from
340 * @size: size of the request in bytes
341 * @align: alignment of the region
342 * @goal: preferred starting address of the region
343 *
344 * The goal is dropped if it can not be satisfied and the allocation will
345 * fall back to memory below @goal.
346 *
347 * Allocation may fall back to any node in the system if the specified node
348 * can not hold the requested memory.
349 *
350 * The function panics if the request can not be satisfied.
351 */
352void * __init __alloc_bootmem_node(pg_data_t *pgdat, unsigned long size,
353 unsigned long align, unsigned long goal)
354{
Yinghai Lu09325872011-02-24 14:43:05 +0100355 if (WARN_ON_ONCE(slab_is_available()))
356 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
357
Johannes Weinerba539862012-05-29 15:06:35 -0700358 return ___alloc_bootmem_node(pgdat, size, align, goal, 0);
Yinghai Lu09325872011-02-24 14:43:05 +0100359}
360
361void * __init __alloc_bootmem_node_high(pg_data_t *pgdat, unsigned long size,
362 unsigned long align, unsigned long goal)
363{
Yinghai Lu09325872011-02-24 14:43:05 +0100364 return __alloc_bootmem_node(pgdat, size, align, goal);
Yinghai Lu09325872011-02-24 14:43:05 +0100365}
366
Yinghai Lu09325872011-02-24 14:43:05 +0100367#ifndef ARCH_LOW_ADDRESS_LIMIT
368#define ARCH_LOW_ADDRESS_LIMIT 0xffffffffUL
369#endif
370
371/**
372 * __alloc_bootmem_low - allocate low boot memory
373 * @size: size of the request in bytes
374 * @align: alignment of the region
375 * @goal: preferred starting address of the region
376 *
377 * The goal is dropped if it can not be satisfied and the allocation will
378 * fall back to memory below @goal.
379 *
380 * Allocation may happen on any node in the system.
381 *
382 * The function panics if the request can not be satisfied.
383 */
384void * __init __alloc_bootmem_low(unsigned long size, unsigned long align,
385 unsigned long goal)
386{
387 return ___alloc_bootmem(size, align, goal, ARCH_LOW_ADDRESS_LIMIT);
388}
389
390/**
391 * __alloc_bootmem_low_node - allocate low boot memory from a specific node
392 * @pgdat: node to allocate from
393 * @size: size of the request in bytes
394 * @align: alignment of the region
395 * @goal: preferred starting address of the region
396 *
397 * The goal is dropped if it can not be satisfied and the allocation will
398 * fall back to memory below @goal.
399 *
400 * Allocation may fall back to any node in the system if the specified node
401 * can not hold the requested memory.
402 *
403 * The function panics if the request can not be satisfied.
404 */
405void * __init __alloc_bootmem_low_node(pg_data_t *pgdat, unsigned long size,
406 unsigned long align, unsigned long goal)
407{
Yinghai Lu09325872011-02-24 14:43:05 +0100408 if (WARN_ON_ONCE(slab_is_available()))
409 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
410
Johannes Weinerba539862012-05-29 15:06:35 -0700411 return ___alloc_bootmem_node(pgdat, size, align, goal,
412 ARCH_LOW_ADDRESS_LIMIT);
Yinghai Lu09325872011-02-24 14:43:05 +0100413}