blob: ab998125f04d320b19626b92eba11c938e473406 [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>
Paul Gortmakerb95f1b312011-10-16 02:01:52 -040014#include <linux/export.h>
Yinghai Lu09325872011-02-24 14:43:05 +010015#include <linux/kmemleak.h>
16#include <linux/range.h>
17#include <linux/memblock.h>
zijun_hu23827052016-10-07 16:59:24 -070018#include <linux/bootmem.h>
Yinghai Lu09325872011-02-24 14:43:05 +010019
20#include <asm/bug.h>
21#include <asm/io.h>
Yinghai Lu09325872011-02-24 14:43:05 +010022
23#include "internal.h"
24
zijun_hu23827052016-10-07 16:59:24 -070025#ifndef CONFIG_HAVE_MEMBLOCK
26#error CONFIG_HAVE_MEMBLOCK not defined
27#endif
28
Yinghai Lue782ab42011-02-24 14:43:06 +010029#ifndef CONFIG_NEED_MULTIPLE_NODES
30struct pglist_data __refdata contig_page_data;
31EXPORT_SYMBOL(contig_page_data);
32#endif
33
Yinghai Lu09325872011-02-24 14:43:05 +010034unsigned long max_low_pfn;
35unsigned long min_low_pfn;
36unsigned long max_pfn;
Igor Mammedov8dd33032015-12-04 14:07:05 +010037unsigned long long max_possible_pfn;
Yinghai Lu09325872011-02-24 14:43:05 +010038
Yinghai Lu8bc1f912011-02-24 14:43:06 +010039static void * __init __alloc_memory_core_early(int nid, u64 size, u64 align,
40 u64 goal, u64 limit)
41{
42 void *ptr;
43 u64 addr;
Tony Lucka3f5baf2015-06-24 16:58:12 -070044 ulong flags = choose_memblock_flags();
Yinghai Lu8bc1f912011-02-24 14:43:06 +010045
46 if (limit > memblock.current_limit)
47 limit = memblock.current_limit;
48
Tony Lucka3f5baf2015-06-24 16:58:12 -070049again:
Tony Luckfc6daaf2015-06-24 16:58:09 -070050 addr = memblock_find_in_range_node(size, align, goal, limit, nid,
Tony Lucka3f5baf2015-06-24 16:58:12 -070051 flags);
52 if (!addr && (flags & MEMBLOCK_MIRROR)) {
53 flags &= ~MEMBLOCK_MIRROR;
54 pr_warn("Could not allocate %pap bytes of mirrored memory\n",
55 &size);
56 goto again;
57 }
Tejun Heo1f5026a2011-07-12 09:58:09 +020058 if (!addr)
Yinghai Lu8bc1f912011-02-24 14:43:06 +010059 return NULL;
60
Philipp Hachtmann87379ec2014-01-23 15:53:10 -080061 if (memblock_reserve(addr, size))
62 return NULL;
63
Yinghai Lu8bc1f912011-02-24 14:43:06 +010064 ptr = phys_to_virt(addr);
65 memset(ptr, 0, size);
Yinghai Lu8bc1f912011-02-24 14:43:06 +010066 /*
67 * The min_count is set to 0 so that bootmem allocated blocks
68 * are never reported as leaks.
69 */
70 kmemleak_alloc(ptr, size, 0, 0);
71 return ptr;
72}
73
Yinghai Lu09325872011-02-24 14:43:05 +010074/*
75 * free_bootmem_late - free bootmem pages directly to page allocator
76 * @addr: starting address of the range
77 * @size: size of the range in bytes
78 *
79 * This is only useful when the bootmem allocator has already been torn
80 * down, but we are still initializing the system. Pages are given directly
81 * to the page allocator, no bootmem metadata is updated because it is gone.
82 */
83void __init free_bootmem_late(unsigned long addr, unsigned long size)
84{
85 unsigned long cursor, end;
86
Catalin Marinas9099dae2016-10-11 13:55:11 -070087 kmemleak_free_part_phys(addr, size);
Yinghai Lu09325872011-02-24 14:43:05 +010088
89 cursor = PFN_UP(addr);
90 end = PFN_DOWN(addr + size);
91
92 for (; cursor < end; cursor++) {
Mel Gormand70ddd72015-06-30 14:56:52 -070093 __free_pages_bootmem(pfn_to_page(cursor), cursor, 0);
Yinghai Lu09325872011-02-24 14:43:05 +010094 totalram_pages++;
95 }
96}
97
98static void __init __free_pages_memory(unsigned long start, unsigned long end)
99{
Robin Holt309d0b32013-11-12 15:07:23 -0800100 int order;
Yinghai Lu09325872011-02-24 14:43:05 +0100101
Robin Holt309d0b32013-11-12 15:07:23 -0800102 while (start < end) {
103 order = min(MAX_ORDER - 1UL, __ffs(start));
Yinghai Lu09325872011-02-24 14:43:05 +0100104
Robin Holt309d0b32013-11-12 15:07:23 -0800105 while (start + (1UL << order) > end)
106 order--;
Yinghai Lu09325872011-02-24 14:43:05 +0100107
Mel Gormand70ddd72015-06-30 14:56:52 -0700108 __free_pages_bootmem(pfn_to_page(start), start, order);
Robin Holt309d0b32013-11-12 15:07:23 -0800109
110 start += (1UL << order);
Yinghai Lu09325872011-02-24 14:43:05 +0100111 }
Yinghai Lu09325872011-02-24 14:43:05 +0100112}
113
Yinghai Lu29f67382012-07-11 14:02:56 -0700114static unsigned long __init __free_memory_core(phys_addr_t start,
115 phys_addr_t end)
116{
117 unsigned long start_pfn = PFN_UP(start);
118 unsigned long end_pfn = min_t(unsigned long,
119 PFN_DOWN(end), max_low_pfn);
120
121 if (start_pfn > end_pfn)
122 return 0;
123
124 __free_pages_memory(start_pfn, end_pfn);
125
126 return end_pfn - start_pfn;
127}
128
Joonsoo Kimb4def352013-04-29 15:08:52 -0700129static unsigned long __init free_low_memory_core_early(void)
Yinghai Lu09325872011-02-24 14:43:05 +0100130{
Yinghai Lu09325872011-02-24 14:43:05 +0100131 unsigned long count = 0;
Philipp Hachtmann354f17e2014-01-23 15:53:24 -0800132 phys_addr_t start, end;
Tejun Heo8a9ca342011-07-12 11:16:02 +0200133 u64 i;
Yinghai Lu09325872011-02-24 14:43:05 +0100134
Xishi Qiu0a313a92014-09-09 14:50:46 -0700135 memblock_clear_hotplug(0, -1);
136
Nathan Zimmer92923ca2015-06-30 14:56:48 -0700137 for_each_reserved_mem_region(i, &start, &end)
138 reserve_bootmem_region(start, end);
139
Wanlong Gao914a0512016-10-07 17:01:04 -0700140 /*
141 * We need to use NUMA_NO_NODE instead of NODE_DATA(0)->node_id
142 * because in some case like Node0 doesn't have RAM installed
143 * low ram will be on Node1
144 */
Tony Luckfc6daaf2015-06-24 16:58:09 -0700145 for_each_free_mem_range(i, NUMA_NO_NODE, MEMBLOCK_NONE, &start, &end,
146 NULL)
Yinghai Lu29f67382012-07-11 14:02:56 -0700147 count += __free_memory_core(start, end);
Yinghai Lu09325872011-02-24 14:43:05 +0100148
Yinghai Lu09325872011-02-24 14:43:05 +0100149 return count;
150}
151
Jiang Liu7b4b2a02013-07-03 15:03:11 -0700152static int reset_managed_pages_done __initdata;
153
Tang Chenf784a3f2014-11-13 15:19:39 -0800154void reset_node_managed_pages(pg_data_t *pgdat)
Jiang Liu9feedc92012-12-12 13:52:12 -0800155{
156 struct zone *z;
157
Jiang Liu9feedc92012-12-12 13:52:12 -0800158 for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++)
Jiang Liu7b4b2a02013-07-03 15:03:11 -0700159 z->managed_pages = 0;
160}
161
162void __init reset_all_zones_managed_pages(void)
163{
164 struct pglist_data *pgdat;
165
Tang Chenf784a3f2014-11-13 15:19:39 -0800166 if (reset_managed_pages_done)
167 return;
168
Jiang Liu7b4b2a02013-07-03 15:03:11 -0700169 for_each_online_pgdat(pgdat)
170 reset_node_managed_pages(pgdat);
Tang Chenf784a3f2014-11-13 15:19:39 -0800171
Jiang Liu7b4b2a02013-07-03 15:03:11 -0700172 reset_managed_pages_done = 1;
Jiang Liu9feedc92012-12-12 13:52:12 -0800173}
174
Yinghai Lu09325872011-02-24 14:43:05 +0100175/**
Yinghai Lu09325872011-02-24 14:43:05 +0100176 * free_all_bootmem - release free pages to the buddy allocator
177 *
178 * Returns the number of pages actually released.
179 */
180unsigned long __init free_all_bootmem(void)
181{
Jiang Liu0c988532013-07-03 15:03:24 -0700182 unsigned long pages;
183
Jiang Liu7b4b2a02013-07-03 15:03:11 -0700184 reset_all_zones_managed_pages();
Jiang Liu9feedc92012-12-12 13:52:12 -0800185
Jiang Liu0c988532013-07-03 15:03:24 -0700186 pages = free_low_memory_core_early();
187 totalram_pages += pages;
188
189 return pages;
Yinghai Lu09325872011-02-24 14:43:05 +0100190}
191
192/**
193 * free_bootmem_node - mark a page range as usable
194 * @pgdat: node the range resides on
195 * @physaddr: starting address of the range
196 * @size: size of the range in bytes
197 *
198 * Partial pages will be considered reserved and left as they are.
199 *
200 * The range must reside completely on the specified node.
201 */
202void __init free_bootmem_node(pg_data_t *pgdat, unsigned long physaddr,
203 unsigned long size)
204{
Tejun Heo24aa0782011-07-12 11:16:06 +0200205 memblock_free(physaddr, size);
Yinghai Lu09325872011-02-24 14:43:05 +0100206}
207
208/**
209 * free_bootmem - mark a page range as usable
210 * @addr: starting address of the range
211 * @size: size of the range in bytes
212 *
213 * Partial pages will be considered reserved and left as they are.
214 *
215 * The range must be contiguous but may span node boundaries.
216 */
217void __init free_bootmem(unsigned long addr, unsigned long size)
218{
Tejun Heo24aa0782011-07-12 11:16:06 +0200219 memblock_free(addr, size);
Yinghai Lu09325872011-02-24 14:43:05 +0100220}
221
222static void * __init ___alloc_bootmem_nopanic(unsigned long size,
223 unsigned long align,
224 unsigned long goal,
225 unsigned long limit)
226{
227 void *ptr;
228
229 if (WARN_ON_ONCE(slab_is_available()))
230 return kzalloc(size, GFP_NOWAIT);
231
232restart:
233
Grygorii Strashkob1154232014-01-21 15:50:16 -0800234 ptr = __alloc_memory_core_early(NUMA_NO_NODE, size, align, goal, limit);
Yinghai Lu09325872011-02-24 14:43:05 +0100235
236 if (ptr)
237 return ptr;
238
239 if (goal != 0) {
240 goal = 0;
241 goto restart;
242 }
243
244 return NULL;
245}
246
247/**
248 * __alloc_bootmem_nopanic - allocate boot memory without panicking
249 * @size: size of the request in bytes
250 * @align: alignment of the region
251 * @goal: preferred starting address of the region
252 *
253 * The goal is dropped if it can not be satisfied and the allocation will
254 * fall back to memory below @goal.
255 *
256 * Allocation may happen on any node in the system.
257 *
258 * Returns NULL on failure.
259 */
260void * __init __alloc_bootmem_nopanic(unsigned long size, unsigned long align,
261 unsigned long goal)
262{
263 unsigned long limit = -1UL;
264
265 return ___alloc_bootmem_nopanic(size, align, goal, limit);
266}
267
268static void * __init ___alloc_bootmem(unsigned long size, unsigned long align,
269 unsigned long goal, unsigned long limit)
270{
271 void *mem = ___alloc_bootmem_nopanic(size, align, goal, limit);
272
273 if (mem)
274 return mem;
275 /*
276 * Whoops, we cannot satisfy the allocation request.
277 */
Joe Perches11705322016-03-17 14:19:50 -0700278 pr_alert("bootmem alloc of %lu bytes failed!\n", size);
Yinghai Lu09325872011-02-24 14:43:05 +0100279 panic("Out of memory");
280 return NULL;
281}
282
283/**
284 * __alloc_bootmem - allocate boot memory
285 * @size: size of the request in bytes
286 * @align: alignment of the region
287 * @goal: preferred starting address of the region
288 *
289 * The goal is dropped if it can not be satisfied and the allocation will
290 * fall back to memory below @goal.
291 *
292 * Allocation may happen on any node in the system.
293 *
294 * The function panics if the request can not be satisfied.
295 */
296void * __init __alloc_bootmem(unsigned long size, unsigned long align,
297 unsigned long goal)
298{
299 unsigned long limit = -1UL;
300
301 return ___alloc_bootmem(size, align, goal, limit);
302}
303
Yinghai Lu99ab7b12012-07-11 14:02:53 -0700304void * __init ___alloc_bootmem_node_nopanic(pg_data_t *pgdat,
Johannes Weinerba539862012-05-29 15:06:35 -0700305 unsigned long size,
306 unsigned long align,
307 unsigned long goal,
308 unsigned long limit)
309{
310 void *ptr;
311
312again:
313 ptr = __alloc_memory_core_early(pgdat->node_id, size, align,
314 goal, limit);
315 if (ptr)
316 return ptr;
317
Grygorii Strashkob1154232014-01-21 15:50:16 -0800318 ptr = __alloc_memory_core_early(NUMA_NO_NODE, size, align,
Johannes Weinerba539862012-05-29 15:06:35 -0700319 goal, limit);
320 if (ptr)
321 return ptr;
322
323 if (goal) {
324 goal = 0;
325 goto again;
326 }
327
328 return NULL;
329}
330
331void * __init __alloc_bootmem_node_nopanic(pg_data_t *pgdat, unsigned long size,
332 unsigned long align, unsigned long goal)
333{
334 if (WARN_ON_ONCE(slab_is_available()))
335 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
336
337 return ___alloc_bootmem_node_nopanic(pgdat, size, align, goal, 0);
338}
339
Rashika Kheriade498502014-04-03 14:48:06 -0700340static void * __init ___alloc_bootmem_node(pg_data_t *pgdat, unsigned long size,
Johannes Weinerba539862012-05-29 15:06:35 -0700341 unsigned long align, unsigned long goal,
342 unsigned long limit)
343{
344 void *ptr;
345
346 ptr = ___alloc_bootmem_node_nopanic(pgdat, size, align, goal, limit);
347 if (ptr)
348 return ptr;
349
Joe Perches11705322016-03-17 14:19:50 -0700350 pr_alert("bootmem alloc of %lu bytes failed!\n", size);
Johannes Weinerba539862012-05-29 15:06:35 -0700351 panic("Out of memory");
352 return NULL;
353}
354
Yinghai Lu09325872011-02-24 14:43:05 +0100355/**
356 * __alloc_bootmem_node - allocate boot memory from a specific node
357 * @pgdat: node to allocate from
358 * @size: size of the request in bytes
359 * @align: alignment of the region
360 * @goal: preferred starting address of the region
361 *
362 * The goal is dropped if it can not be satisfied and the allocation will
363 * fall back to memory below @goal.
364 *
365 * Allocation may fall back to any node in the system if the specified node
366 * can not hold the requested memory.
367 *
368 * The function panics if the request can not be satisfied.
369 */
370void * __init __alloc_bootmem_node(pg_data_t *pgdat, unsigned long size,
371 unsigned long align, unsigned long goal)
372{
Yinghai Lu09325872011-02-24 14:43:05 +0100373 if (WARN_ON_ONCE(slab_is_available()))
374 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
375
Johannes Weinerba539862012-05-29 15:06:35 -0700376 return ___alloc_bootmem_node(pgdat, size, align, goal, 0);
Yinghai Lu09325872011-02-24 14:43:05 +0100377}
378
379void * __init __alloc_bootmem_node_high(pg_data_t *pgdat, unsigned long size,
380 unsigned long align, unsigned long goal)
381{
Yinghai Lu09325872011-02-24 14:43:05 +0100382 return __alloc_bootmem_node(pgdat, size, align, goal);
Yinghai Lu09325872011-02-24 14:43:05 +0100383}
384
Yinghai Lu09325872011-02-24 14:43:05 +0100385
386/**
387 * __alloc_bootmem_low - allocate low boot memory
388 * @size: size of the request in bytes
389 * @align: alignment of the region
390 * @goal: preferred starting address of the region
391 *
392 * The goal is dropped if it can not be satisfied and the allocation will
393 * fall back to memory below @goal.
394 *
395 * Allocation may happen on any node in the system.
396 *
397 * The function panics if the request can not be satisfied.
398 */
399void * __init __alloc_bootmem_low(unsigned long size, unsigned long align,
400 unsigned long goal)
401{
402 return ___alloc_bootmem(size, align, goal, ARCH_LOW_ADDRESS_LIMIT);
403}
404
Yinghai Lu38fa4172013-01-24 12:20:15 -0800405void * __init __alloc_bootmem_low_nopanic(unsigned long size,
406 unsigned long align,
407 unsigned long goal)
408{
409 return ___alloc_bootmem_nopanic(size, align, goal,
410 ARCH_LOW_ADDRESS_LIMIT);
411}
412
Yinghai Lu09325872011-02-24 14:43:05 +0100413/**
414 * __alloc_bootmem_low_node - allocate low boot memory from a specific node
415 * @pgdat: node to allocate from
416 * @size: size of the request in bytes
417 * @align: alignment of the region
418 * @goal: preferred starting address of the region
419 *
420 * The goal is dropped if it can not be satisfied and the allocation will
421 * fall back to memory below @goal.
422 *
423 * Allocation may fall back to any node in the system if the specified node
424 * can not hold the requested memory.
425 *
426 * The function panics if the request can not be satisfied.
427 */
428void * __init __alloc_bootmem_low_node(pg_data_t *pgdat, unsigned long size,
429 unsigned long align, unsigned long goal)
430{
Yinghai Lu09325872011-02-24 14:43:05 +0100431 if (WARN_ON_ONCE(slab_is_available()))
432 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
433
Johannes Weinerba539862012-05-29 15:06:35 -0700434 return ___alloc_bootmem_node(pgdat, size, align, goal,
435 ARCH_LOW_ADDRESS_LIMIT);
Yinghai Lu09325872011-02-24 14:43:05 +0100436}