blob: 1a77012ecdb3c56a92d53ef5b779c9648d54c6a3 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
Johannes Weiner57cfc292008-07-23 21:28:00 -07002 * bootmem - A boot-time physical memory allocator and configurator
Linus Torvalds1da177e2005-04-16 15:20:36 -07003 *
4 * Copyright (C) 1999 Ingo Molnar
Johannes Weiner57cfc292008-07-23 21:28:00 -07005 * 1999 Kanoj Sarcar, SGI
6 * 2008 Johannes Weiner
Linus Torvalds1da177e2005-04-16 15:20:36 -07007 *
Johannes Weiner57cfc292008-07-23 21:28:00 -07008 * Access to this subsystem has to be serialized externally (which is true
9 * for the boot process anyway).
Linus Torvalds1da177e2005-04-16 15:20:36 -070010 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070011#include <linux/init.h>
Franck Bui-Huubbc7b922006-09-25 23:31:07 -070012#include <linux/pfn.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090013#include <linux/slab.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070014#include <linux/bootmem.h>
Paul Gortmakerb95f1b312011-10-16 02:01:52 -040015#include <linux/export.h>
Catalin Marinasec3a3542009-07-07 10:33:01 +010016#include <linux/kmemleak.h>
Yinghai Lu08677212010-02-10 01:20:20 -080017#include <linux/range.h>
Yinghai Lu72d7c3b2010-08-25 13:39:17 -070018#include <linux/memblock.h>
Franck Bui-Huue786e862006-09-25 23:31:06 -070019
20#include <asm/bug.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070021#include <asm/io.h>
Heiko Carstensdfd54cb2006-09-25 23:31:33 -070022#include <asm/processor.h>
Franck Bui-Huue786e862006-09-25 23:31:06 -070023
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#include "internal.h"
25
Yinghai Lue782ab42011-02-24 14:43:06 +010026#ifndef CONFIG_NEED_MULTIPLE_NODES
27struct pglist_data __refdata contig_page_data = {
28 .bdata = &bootmem_node_data[0]
29};
30EXPORT_SYMBOL(contig_page_data);
31#endif
32
Linus Torvalds1da177e2005-04-16 15:20:36 -070033unsigned long max_low_pfn;
34unsigned long min_low_pfn;
35unsigned long max_pfn;
36
Johannes Weinerb61bfa32008-07-23 21:26:55 -070037bootmem_data_t bootmem_node_data[MAX_NUMNODES] __initdata;
38
Johannes Weiner636cc402008-07-23 21:28:03 -070039static struct list_head bdata_list __initdata = LIST_HEAD_INIT(bdata_list);
40
Johannes Weiner2e5237d2008-07-23 21:28:02 -070041static int bootmem_debug;
42
43static int __init bootmem_debug_setup(char *buf)
44{
45 bootmem_debug = 1;
46 return 0;
47}
48early_param("bootmem_debug", bootmem_debug_setup);
49
50#define bdebug(fmt, args...) ({ \
51 if (unlikely(bootmem_debug)) \
52 printk(KERN_INFO \
53 "bootmem::%s " fmt, \
Harvey Harrison80a914d2008-10-15 22:01:25 -070054 __func__, ## args); \
Johannes Weiner2e5237d2008-07-23 21:28:02 -070055})
56
Johannes Weinerdf049a52008-07-23 21:28:02 -070057static unsigned long __init bootmap_bytes(unsigned long pages)
Johannes Weiner223e8dc2008-07-23 21:28:00 -070058{
Johannes Weinerdf049a52008-07-23 21:28:02 -070059 unsigned long bytes = (pages + 7) / 8;
Johannes Weiner223e8dc2008-07-23 21:28:00 -070060
Johannes Weinerdf049a52008-07-23 21:28:02 -070061 return ALIGN(bytes, sizeof(long));
Johannes Weiner223e8dc2008-07-23 21:28:00 -070062}
63
Johannes Weinera66fd7d2008-07-23 21:28:01 -070064/**
65 * bootmem_bootmap_pages - calculate bitmap size in pages
66 * @pages: number of pages the bitmap has to represent
67 */
Franck Bui-Huuf71bf0c2006-09-25 23:31:08 -070068unsigned long __init bootmem_bootmap_pages(unsigned long pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -070069{
Johannes Weinerdf049a52008-07-23 21:28:02 -070070 unsigned long bytes = bootmap_bytes(pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -070071
Johannes Weinerdf049a52008-07-23 21:28:02 -070072 return PAGE_ALIGN(bytes) >> PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -070073}
Franck Bui-Huuf71bf0c2006-09-25 23:31:08 -070074
KAMEZAWA Hiroyuki679bc9f2006-03-27 01:15:58 -080075/*
76 * link bdata in order
77 */
Franck Bui-Huu69d49e62006-09-25 23:31:04 -070078static void __init link_bootmem(bootmem_data_t *bdata)
KAMEZAWA Hiroyuki679bc9f2006-03-27 01:15:58 -080079{
Johannes Weiner636cc402008-07-23 21:28:03 -070080 struct list_head *iter;
Franck Bui-Huuf71bf0c2006-09-25 23:31:08 -070081
Johannes Weiner636cc402008-07-23 21:28:03 -070082 list_for_each(iter, &bdata_list) {
83 bootmem_data_t *ent;
84
85 ent = list_entry(iter, bootmem_data_t, list);
Johannes Weiner3560e242008-07-23 21:28:09 -070086 if (bdata->node_min_pfn < ent->node_min_pfn)
Johannes Weiner636cc402008-07-23 21:28:03 -070087 break;
KAMEZAWA Hiroyuki679bc9f2006-03-27 01:15:58 -080088 }
Johannes Weiner636cc402008-07-23 21:28:03 -070089 list_add_tail(&bdata->list, iter);
KAMEZAWA Hiroyuki679bc9f2006-03-27 01:15:58 -080090}
91
Franck Bui-Huubbc7b922006-09-25 23:31:07 -070092/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070093 * Called once to set up the allocator itself.
94 */
Johannes Weiner8ae04462008-07-23 21:26:59 -070095static unsigned long __init init_bootmem_core(bootmem_data_t *bdata,
Linus Torvalds1da177e2005-04-16 15:20:36 -070096 unsigned long mapstart, unsigned long start, unsigned long end)
97{
Franck Bui-Huubbc7b922006-09-25 23:31:07 -070098 unsigned long mapsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -070099
Mel Gorman2dbb51c2008-07-23 21:26:52 -0700100 mminit_validate_memmodel_limits(&start, &end);
Franck Bui-Huubbc7b922006-09-25 23:31:07 -0700101 bdata->node_bootmem_map = phys_to_virt(PFN_PHYS(mapstart));
Johannes Weiner3560e242008-07-23 21:28:09 -0700102 bdata->node_min_pfn = start;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700103 bdata->node_low_pfn = end;
KAMEZAWA Hiroyuki679bc9f2006-03-27 01:15:58 -0800104 link_bootmem(bdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700105
106 /*
107 * Initially all pages are reserved - setup_arch() has to
108 * register free RAM areas explicitly.
109 */
Johannes Weinerdf049a52008-07-23 21:28:02 -0700110 mapsize = bootmap_bytes(end - start);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700111 memset(bdata->node_bootmem_map, 0xff, mapsize);
112
Johannes Weiner2e5237d2008-07-23 21:28:02 -0700113 bdebug("nid=%td start=%lx map=%lx end=%lx mapsize=%lx\n",
114 bdata - bootmem_node_data, start, mapstart, end, mapsize);
115
Linus Torvalds1da177e2005-04-16 15:20:36 -0700116 return mapsize;
117}
118
Johannes Weinera66fd7d2008-07-23 21:28:01 -0700119/**
120 * init_bootmem_node - register a node as boot memory
121 * @pgdat: node to register
122 * @freepfn: pfn where the bitmap for this node is to be placed
123 * @startpfn: first pfn on the node
124 * @endpfn: first pfn after the node
125 *
126 * Returns the number of bytes needed to hold the bitmap for this node.
127 */
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700128unsigned long __init init_bootmem_node(pg_data_t *pgdat, unsigned long freepfn,
129 unsigned long startpfn, unsigned long endpfn)
130{
131 return init_bootmem_core(pgdat->bdata, freepfn, startpfn, endpfn);
132}
133
Johannes Weinera66fd7d2008-07-23 21:28:01 -0700134/**
135 * init_bootmem - register boot memory
136 * @start: pfn where the bitmap is to be placed
137 * @pages: number of available physical pages
138 *
139 * Returns the number of bytes needed to hold the bitmap.
140 */
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700141unsigned long __init init_bootmem(unsigned long start, unsigned long pages)
142{
143 max_low_pfn = pages;
144 min_low_pfn = start;
145 return init_bootmem_core(NODE_DATA(0)->bdata, start, 0, pages);
146}
Yinghai Lu09325872011-02-24 14:43:05 +0100147
FUJITA Tomonori9f993ac2009-11-10 19:46:17 +0900148/*
149 * free_bootmem_late - free bootmem pages directly to page allocator
150 * @addr: starting address of the range
151 * @size: size of the range in bytes
152 *
153 * This is only useful when the bootmem allocator has already been torn
154 * down, but we are still initializing the system. Pages are given directly
155 * to the page allocator, no bootmem metadata is updated because it is gone.
156 */
157void __init free_bootmem_late(unsigned long addr, unsigned long size)
158{
159 unsigned long cursor, end;
160
161 kmemleak_free_part(__va(addr), size);
162
163 cursor = PFN_UP(addr);
164 end = PFN_DOWN(addr + size);
165
166 for (; cursor < end; cursor++) {
167 __free_pages_bootmem(pfn_to_page(cursor), 0);
168 totalram_pages++;
169 }
170}
171
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700172static unsigned long __init free_all_bootmem_core(bootmem_data_t *bdata)
173{
Johannes Weiner41546c12008-07-23 21:28:03 -0700174 int aligned;
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700175 struct page *page;
Johannes Weiner41546c12008-07-23 21:28:03 -0700176 unsigned long start, end, pages, count = 0;
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700177
Johannes Weiner41546c12008-07-23 21:28:03 -0700178 if (!bdata->node_bootmem_map)
179 return 0;
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700180
Johannes Weiner3560e242008-07-23 21:28:09 -0700181 start = bdata->node_min_pfn;
Johannes Weiner41546c12008-07-23 21:28:03 -0700182 end = bdata->node_low_pfn;
183
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700184 /*
Johannes Weiner41546c12008-07-23 21:28:03 -0700185 * If the start is aligned to the machines wordsize, we might
186 * be able to free pages in bulks of that order.
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700187 */
Johannes Weiner41546c12008-07-23 21:28:03 -0700188 aligned = !(start & (BITS_PER_LONG - 1));
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700189
Johannes Weiner41546c12008-07-23 21:28:03 -0700190 bdebug("nid=%td start=%lx end=%lx aligned=%d\n",
191 bdata - bootmem_node_data, start, end, aligned);
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700192
Johannes Weiner41546c12008-07-23 21:28:03 -0700193 while (start < end) {
194 unsigned long *map, idx, vec;
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700195
Johannes Weiner41546c12008-07-23 21:28:03 -0700196 map = bdata->node_bootmem_map;
Johannes Weiner3560e242008-07-23 21:28:09 -0700197 idx = start - bdata->node_min_pfn;
Johannes Weiner41546c12008-07-23 21:28:03 -0700198 vec = ~map[idx / BITS_PER_LONG];
199
200 if (aligned && vec == ~0UL && start + BITS_PER_LONG < end) {
201 int order = ilog2(BITS_PER_LONG);
202
203 __free_pages_bootmem(pfn_to_page(start), order);
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700204 count += BITS_PER_LONG;
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700205 } else {
Johannes Weiner41546c12008-07-23 21:28:03 -0700206 unsigned long off = 0;
207
208 while (vec && off < BITS_PER_LONG) {
209 if (vec & 1) {
210 page = pfn_to_page(start + off);
211 __free_pages_bootmem(page, 0);
212 count++;
213 }
214 vec >>= 1;
215 off++;
216 }
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700217 }
Johannes Weiner41546c12008-07-23 21:28:03 -0700218 start += BITS_PER_LONG;
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700219 }
220
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700221 page = virt_to_page(bdata->node_bootmem_map);
Johannes Weiner3560e242008-07-23 21:28:09 -0700222 pages = bdata->node_low_pfn - bdata->node_min_pfn;
Johannes Weiner41546c12008-07-23 21:28:03 -0700223 pages = bootmem_bootmap_pages(pages);
224 count += pages;
225 while (pages--)
226 __free_pages_bootmem(page++, 0);
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700227
Johannes Weiner2e5237d2008-07-23 21:28:02 -0700228 bdebug("nid=%td released=%lx\n", bdata - bootmem_node_data, count);
229
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700230 return count;
231}
232
Johannes Weinera66fd7d2008-07-23 21:28:01 -0700233/**
234 * free_all_bootmem_node - release a node's free pages to the buddy allocator
235 * @pgdat: node to be released
236 *
237 * Returns the number of pages actually released.
238 */
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700239unsigned long __init free_all_bootmem_node(pg_data_t *pgdat)
240{
241 register_page_bootmem_info_node(pgdat);
242 return free_all_bootmem_core(pgdat->bdata);
243}
244
Johannes Weinera66fd7d2008-07-23 21:28:01 -0700245/**
246 * free_all_bootmem - release free pages to the buddy allocator
247 *
248 * Returns the number of pages actually released.
249 */
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700250unsigned long __init free_all_bootmem(void)
251{
Yinghai Luaa235fc2010-03-31 20:45:27 -0700252 unsigned long total_pages = 0;
253 bootmem_data_t *bdata;
254
255 list_for_each_entry(bdata, &bdata_list, list)
256 total_pages += free_all_bootmem_core(bdata);
257
258 return total_pages;
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700259}
260
Johannes Weinerd747fa42008-07-23 21:28:05 -0700261static void __init __free(bootmem_data_t *bdata,
262 unsigned long sidx, unsigned long eidx)
263{
264 unsigned long idx;
265
266 bdebug("nid=%td start=%lx end=%lx\n", bdata - bootmem_node_data,
Johannes Weiner3560e242008-07-23 21:28:09 -0700267 sidx + bdata->node_min_pfn,
268 eidx + bdata->node_min_pfn);
Johannes Weinerd747fa42008-07-23 21:28:05 -0700269
Johannes Weinere2bf3ca2008-07-23 21:28:06 -0700270 if (bdata->hint_idx > sidx)
271 bdata->hint_idx = sidx;
272
Johannes Weinerd747fa42008-07-23 21:28:05 -0700273 for (idx = sidx; idx < eidx; idx++)
274 if (!test_and_clear_bit(idx, bdata->node_bootmem_map))
275 BUG();
276}
277
278static int __init __reserve(bootmem_data_t *bdata, unsigned long sidx,
279 unsigned long eidx, int flags)
280{
281 unsigned long idx;
282 int exclusive = flags & BOOTMEM_EXCLUSIVE;
283
284 bdebug("nid=%td start=%lx end=%lx flags=%x\n",
285 bdata - bootmem_node_data,
Johannes Weiner3560e242008-07-23 21:28:09 -0700286 sidx + bdata->node_min_pfn,
287 eidx + bdata->node_min_pfn,
Johannes Weinerd747fa42008-07-23 21:28:05 -0700288 flags);
289
290 for (idx = sidx; idx < eidx; idx++)
291 if (test_and_set_bit(idx, bdata->node_bootmem_map)) {
292 if (exclusive) {
293 __free(bdata, sidx, idx);
294 return -EBUSY;
295 }
296 bdebug("silent double reserve of PFN %lx\n",
Johannes Weiner3560e242008-07-23 21:28:09 -0700297 idx + bdata->node_min_pfn);
Johannes Weinerd747fa42008-07-23 21:28:05 -0700298 }
299 return 0;
300}
301
Johannes Weinere2bf3ca2008-07-23 21:28:06 -0700302static int __init mark_bootmem_node(bootmem_data_t *bdata,
303 unsigned long start, unsigned long end,
304 int reserve, int flags)
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700305{
306 unsigned long sidx, eidx;
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700307
Johannes Weinere2bf3ca2008-07-23 21:28:06 -0700308 bdebug("nid=%td start=%lx end=%lx reserve=%d flags=%x\n",
309 bdata - bootmem_node_data, start, end, reserve, flags);
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700310
Johannes Weiner3560e242008-07-23 21:28:09 -0700311 BUG_ON(start < bdata->node_min_pfn);
Johannes Weinere2bf3ca2008-07-23 21:28:06 -0700312 BUG_ON(end > bdata->node_low_pfn);
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700313
Johannes Weiner3560e242008-07-23 21:28:09 -0700314 sidx = start - bdata->node_min_pfn;
315 eidx = end - bdata->node_min_pfn;
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700316
Johannes Weinere2bf3ca2008-07-23 21:28:06 -0700317 if (reserve)
318 return __reserve(bdata, sidx, eidx, flags);
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700319 else
Johannes Weinere2bf3ca2008-07-23 21:28:06 -0700320 __free(bdata, sidx, eidx);
321 return 0;
322}
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700323
Johannes Weinere2bf3ca2008-07-23 21:28:06 -0700324static int __init mark_bootmem(unsigned long start, unsigned long end,
325 int reserve, int flags)
326{
327 unsigned long pos;
328 bootmem_data_t *bdata;
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700329
Johannes Weinere2bf3ca2008-07-23 21:28:06 -0700330 pos = start;
331 list_for_each_entry(bdata, &bdata_list, list) {
332 int err;
333 unsigned long max;
334
Johannes Weiner3560e242008-07-23 21:28:09 -0700335 if (pos < bdata->node_min_pfn ||
336 pos >= bdata->node_low_pfn) {
Johannes Weinere2bf3ca2008-07-23 21:28:06 -0700337 BUG_ON(pos != start);
338 continue;
339 }
340
341 max = min(bdata->node_low_pfn, end);
342
343 err = mark_bootmem_node(bdata, pos, max, reserve, flags);
344 if (reserve && err) {
345 mark_bootmem(start, pos, 0, 0);
346 return err;
347 }
348
349 if (max == end)
350 return 0;
351 pos = bdata->node_low_pfn;
352 }
353 BUG();
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700354}
355
Johannes Weinera66fd7d2008-07-23 21:28:01 -0700356/**
357 * free_bootmem_node - mark a page range as usable
358 * @pgdat: node the range resides on
359 * @physaddr: starting address of the range
360 * @size: size of the range in bytes
361 *
362 * Partial pages will be considered reserved and left as they are.
363 *
Johannes Weinere2bf3ca2008-07-23 21:28:06 -0700364 * The range must reside completely on the specified node.
Johannes Weinera66fd7d2008-07-23 21:28:01 -0700365 */
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700366void __init free_bootmem_node(pg_data_t *pgdat, unsigned long physaddr,
367 unsigned long size)
368{
Johannes Weinere2bf3ca2008-07-23 21:28:06 -0700369 unsigned long start, end;
370
Catalin Marinasec3a3542009-07-07 10:33:01 +0100371 kmemleak_free_part(__va(physaddr), size);
372
Johannes Weinere2bf3ca2008-07-23 21:28:06 -0700373 start = PFN_UP(physaddr);
374 end = PFN_DOWN(physaddr + size);
375
376 mark_bootmem_node(pgdat->bdata, start, end, 0, 0);
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700377}
378
Johannes Weinera66fd7d2008-07-23 21:28:01 -0700379/**
380 * free_bootmem - mark a page range as usable
381 * @addr: starting address of the range
382 * @size: size of the range in bytes
383 *
384 * Partial pages will be considered reserved and left as they are.
385 *
Johannes Weinere2bf3ca2008-07-23 21:28:06 -0700386 * The range must be contiguous but may span node boundaries.
Johannes Weinera66fd7d2008-07-23 21:28:01 -0700387 */
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700388void __init free_bootmem(unsigned long addr, unsigned long size)
389{
Johannes Weinere2bf3ca2008-07-23 21:28:06 -0700390 unsigned long start, end;
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700391
Catalin Marinasec3a3542009-07-07 10:33:01 +0100392 kmemleak_free_part(__va(addr), size);
393
Johannes Weinere2bf3ca2008-07-23 21:28:06 -0700394 start = PFN_UP(addr);
395 end = PFN_DOWN(addr + size);
Yinghai Lua5645a62008-03-18 12:49:12 -0700396
Johannes Weinere2bf3ca2008-07-23 21:28:06 -0700397 mark_bootmem(start, end, 0, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700398}
399
Johannes Weinera66fd7d2008-07-23 21:28:01 -0700400/**
401 * reserve_bootmem_node - mark a page range as reserved
402 * @pgdat: node the range resides on
403 * @physaddr: starting address of the range
404 * @size: size of the range in bytes
405 * @flags: reservation flags (see linux/bootmem.h)
406 *
407 * Partial pages will be reserved.
408 *
Johannes Weinere2bf3ca2008-07-23 21:28:06 -0700409 * The range must reside completely on the specified node.
Johannes Weinera66fd7d2008-07-23 21:28:01 -0700410 */
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700411int __init reserve_bootmem_node(pg_data_t *pgdat, unsigned long physaddr,
412 unsigned long size, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700413{
Johannes Weinere2bf3ca2008-07-23 21:28:06 -0700414 unsigned long start, end;
Franck Bui-Huubbc7b922006-09-25 23:31:07 -0700415
Johannes Weinere2bf3ca2008-07-23 21:28:06 -0700416 start = PFN_DOWN(physaddr);
417 end = PFN_UP(physaddr + size);
418
419 return mark_bootmem_node(pgdat->bdata, start, end, 1, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700420}
421
Johannes Weinera66fd7d2008-07-23 21:28:01 -0700422/**
423 * reserve_bootmem - mark a page range as usable
424 * @addr: starting address of the range
425 * @size: size of the range in bytes
426 * @flags: reservation flags (see linux/bootmem.h)
427 *
428 * Partial pages will be reserved.
429 *
Johannes Weinere2bf3ca2008-07-23 21:28:06 -0700430 * The range must be contiguous but may span node boundaries.
Johannes Weinera66fd7d2008-07-23 21:28:01 -0700431 */
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700432int __init reserve_bootmem(unsigned long addr, unsigned long size,
433 int flags)
434{
Johannes Weinere2bf3ca2008-07-23 21:28:06 -0700435 unsigned long start, end;
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700436
Johannes Weinere2bf3ca2008-07-23 21:28:06 -0700437 start = PFN_DOWN(addr);
438 end = PFN_UP(addr + size);
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700439
Johannes Weinere2bf3ca2008-07-23 21:28:06 -0700440 return mark_bootmem(start, end, 1, flags);
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700441}
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700442
Yinghai Luf88eff72010-08-25 13:39:15 -0700443int __weak __init reserve_bootmem_generic(unsigned long phys, unsigned long len,
444 int flags)
445{
446 return reserve_bootmem(phys, len, flags);
447}
448
Jan Beulich8aa043d2009-12-14 17:59:46 -0800449static unsigned long __init align_idx(struct bootmem_data *bdata,
450 unsigned long idx, unsigned long step)
Johannes Weiner481ebd02008-08-20 14:09:15 -0700451{
452 unsigned long base = bdata->node_min_pfn;
453
454 /*
455 * Align the index with respect to the node start so that the
456 * combination of both satisfies the requested alignment.
457 */
458
459 return ALIGN(base + idx, step) - base;
460}
461
Jan Beulich8aa043d2009-12-14 17:59:46 -0800462static unsigned long __init align_off(struct bootmem_data *bdata,
463 unsigned long off, unsigned long align)
Johannes Weiner481ebd02008-08-20 14:09:15 -0700464{
465 unsigned long base = PFN_PHYS(bdata->node_min_pfn);
466
467 /* Same as align_idx for byte offsets */
468
469 return ALIGN(base + off, align) - base;
470}
471
Tejun Heod0c4f572009-03-01 16:06:56 +0900472static void * __init alloc_bootmem_core(struct bootmem_data *bdata,
473 unsigned long size, unsigned long align,
474 unsigned long goal, unsigned long limit)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700475{
Johannes Weiner0f3caba2008-07-23 21:28:07 -0700476 unsigned long fallback = 0;
Johannes Weiner5f2809e2008-07-23 21:28:05 -0700477 unsigned long min, max, start, sidx, midx, step;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700478
Johannes Weiner594fe1a2009-01-06 14:40:32 -0800479 bdebug("nid=%td size=%lx [%lu pages] align=%lx goal=%lx limit=%lx\n",
480 bdata - bootmem_node_data, size, PAGE_ALIGN(size) >> PAGE_SHIFT,
481 align, goal, limit);
482
Johannes Weiner5f2809e2008-07-23 21:28:05 -0700483 BUG_ON(!size);
484 BUG_ON(align & (align - 1));
485 BUG_ON(limit && goal + size > limit);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700486
Christian Krafft7c309a62006-12-06 20:32:41 -0800487 if (!bdata->node_bootmem_map)
488 return NULL;
489
Johannes Weiner3560e242008-07-23 21:28:09 -0700490 min = bdata->node_min_pfn;
Johannes Weiner5f2809e2008-07-23 21:28:05 -0700491 max = bdata->node_low_pfn;
Yinghai Lu9a2dc042008-03-18 12:44:48 -0700492
Johannes Weiner5f2809e2008-07-23 21:28:05 -0700493 goal >>= PAGE_SHIFT;
494 limit >>= PAGE_SHIFT;
495
496 if (limit && max > limit)
497 max = limit;
498 if (max <= min)
Yinghai Lu9a2dc042008-03-18 12:44:48 -0700499 return NULL;
500
Johannes Weiner5f2809e2008-07-23 21:28:05 -0700501 step = max(align >> PAGE_SHIFT, 1UL);
Yasunori Goto281dd252005-10-19 15:52:18 -0700502
Johannes Weiner5f2809e2008-07-23 21:28:05 -0700503 if (goal && min < goal && goal < max)
504 start = ALIGN(goal, step);
505 else
506 start = ALIGN(min, step);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700507
Johannes Weiner481ebd02008-08-20 14:09:15 -0700508 sidx = start - bdata->node_min_pfn;
Johannes Weiner3560e242008-07-23 21:28:09 -0700509 midx = max - bdata->node_min_pfn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700510
Johannes Weiner5f2809e2008-07-23 21:28:05 -0700511 if (bdata->hint_idx > sidx) {
Johannes Weiner0f3caba2008-07-23 21:28:07 -0700512 /*
513 * Handle the valid case of sidx being zero and still
514 * catch the fallback below.
515 */
516 fallback = sidx + 1;
Johannes Weiner481ebd02008-08-20 14:09:15 -0700517 sidx = align_idx(bdata, bdata->hint_idx, step);
Yinghai Luad093152008-03-10 23:23:42 -0700518 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700519
Johannes Weiner5f2809e2008-07-23 21:28:05 -0700520 while (1) {
521 int merge;
522 void *region;
523 unsigned long eidx, i, start_off, end_off;
524find_block:
525 sidx = find_next_zero_bit(bdata->node_bootmem_map, midx, sidx);
Johannes Weiner481ebd02008-08-20 14:09:15 -0700526 sidx = align_idx(bdata, sidx, step);
Johannes Weiner5f2809e2008-07-23 21:28:05 -0700527 eidx = sidx + PFN_UP(size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700528
Johannes Weiner5f2809e2008-07-23 21:28:05 -0700529 if (sidx >= midx || eidx > midx)
Haren Myneni66d43e92005-12-12 00:37:39 -0800530 break;
Johannes Weiner5f2809e2008-07-23 21:28:05 -0700531
532 for (i = sidx; i < eidx; i++)
533 if (test_bit(i, bdata->node_bootmem_map)) {
Johannes Weiner481ebd02008-08-20 14:09:15 -0700534 sidx = align_idx(bdata, i, step);
Johannes Weiner5f2809e2008-07-23 21:28:05 -0700535 if (sidx == i)
536 sidx += step;
537 goto find_block;
538 }
539
Mikulas Patocka627240a2008-08-15 00:40:17 -0700540 if (bdata->last_end_off & (PAGE_SIZE - 1) &&
Johannes Weiner5f2809e2008-07-23 21:28:05 -0700541 PFN_DOWN(bdata->last_end_off) + 1 == sidx)
Johannes Weiner481ebd02008-08-20 14:09:15 -0700542 start_off = align_off(bdata, bdata->last_end_off, align);
Johannes Weiner5f2809e2008-07-23 21:28:05 -0700543 else
544 start_off = PFN_PHYS(sidx);
545
546 merge = PFN_DOWN(start_off) < sidx;
547 end_off = start_off + size;
548
549 bdata->last_end_off = end_off;
550 bdata->hint_idx = PFN_UP(end_off);
551
552 /*
553 * Reserve the area now:
554 */
Johannes Weinerd747fa42008-07-23 21:28:05 -0700555 if (__reserve(bdata, PFN_DOWN(start_off) + merge,
556 PFN_UP(end_off), BOOTMEM_EXCLUSIVE))
557 BUG();
Johannes Weiner5f2809e2008-07-23 21:28:05 -0700558
Johannes Weiner3560e242008-07-23 21:28:09 -0700559 region = phys_to_virt(PFN_PHYS(bdata->node_min_pfn) +
560 start_off);
Johannes Weiner5f2809e2008-07-23 21:28:05 -0700561 memset(region, 0, size);
Catalin Marinas008139d2009-08-27 14:29:17 +0100562 /*
563 * The min_count is set to 0 so that bootmem allocated blocks
564 * are never reported as leaks.
565 */
566 kmemleak_alloc(region, size, 0, 0);
Johannes Weiner5f2809e2008-07-23 21:28:05 -0700567 return region;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700568 }
569
Johannes Weiner0f3caba2008-07-23 21:28:07 -0700570 if (fallback) {
Johannes Weiner481ebd02008-08-20 14:09:15 -0700571 sidx = align_idx(bdata, fallback - 1, step);
Johannes Weiner0f3caba2008-07-23 21:28:07 -0700572 fallback = 0;
573 goto find_block;
574 }
575
576 return NULL;
577}
578
Tejun Heod0c4f572009-03-01 16:06:56 +0900579static void * __init alloc_arch_preferred_bootmem(bootmem_data_t *bdata,
580 unsigned long size, unsigned long align,
581 unsigned long goal, unsigned long limit)
582{
Pekka Enberg441c7e02009-06-10 20:05:53 +0300583 if (WARN_ON_ONCE(slab_is_available()))
584 return kzalloc(size, GFP_NOWAIT);
585
Tejun Heod0c4f572009-03-01 16:06:56 +0900586#ifdef CONFIG_HAVE_ARCH_BOOTMEM
Joe Perches433f13a2009-06-18 16:48:58 -0700587 {
588 bootmem_data_t *p_bdata;
Tejun Heod0c4f572009-03-01 16:06:56 +0900589
Joe Perches433f13a2009-06-18 16:48:58 -0700590 p_bdata = bootmem_arch_preferred_node(bdata, size, align,
591 goal, limit);
592 if (p_bdata)
593 return alloc_bootmem_core(p_bdata, size, align,
594 goal, limit);
595 }
Tejun Heod0c4f572009-03-01 16:06:56 +0900596#endif
597 return NULL;
598}
599
Johannes Weiner0f3caba2008-07-23 21:28:07 -0700600static void * __init ___alloc_bootmem_nopanic(unsigned long size,
601 unsigned long align,
602 unsigned long goal,
603 unsigned long limit)
604{
605 bootmem_data_t *bdata;
Tejun Heod0c4f572009-03-01 16:06:56 +0900606 void *region;
Johannes Weiner0f3caba2008-07-23 21:28:07 -0700607
608restart:
Tejun Heod0c4f572009-03-01 16:06:56 +0900609 region = alloc_arch_preferred_bootmem(NULL, size, align, goal, limit);
610 if (region)
611 return region;
Johannes Weiner0f3caba2008-07-23 21:28:07 -0700612
Tejun Heod0c4f572009-03-01 16:06:56 +0900613 list_for_each_entry(bdata, &bdata_list, list) {
Johannes Weiner0f3caba2008-07-23 21:28:07 -0700614 if (goal && bdata->node_low_pfn <= PFN_DOWN(goal))
615 continue;
Johannes Weiner3560e242008-07-23 21:28:09 -0700616 if (limit && bdata->node_min_pfn >= PFN_DOWN(limit))
Johannes Weiner0f3caba2008-07-23 21:28:07 -0700617 break;
618
619 region = alloc_bootmem_core(bdata, size, align, goal, limit);
620 if (region)
621 return region;
622 }
623
Johannes Weiner5f2809e2008-07-23 21:28:05 -0700624 if (goal) {
625 goal = 0;
Johannes Weiner0f3caba2008-07-23 21:28:07 -0700626 goto restart;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700627 }
Johannes Weiner5f2809e2008-07-23 21:28:05 -0700628
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700630}
631
Johannes Weinera66fd7d2008-07-23 21:28:01 -0700632/**
633 * __alloc_bootmem_nopanic - allocate boot memory without panicking
634 * @size: size of the request in bytes
635 * @align: alignment of the region
636 * @goal: preferred starting address of the region
637 *
638 * The goal is dropped if it can not be satisfied and the allocation will
639 * fall back to memory below @goal.
640 *
641 * Allocation may happen on any node in the system.
642 *
643 * Returns NULL on failure.
644 */
Franck Bui-Huubb0923a2006-09-25 23:31:05 -0700645void * __init __alloc_bootmem_nopanic(unsigned long size, unsigned long align,
Johannes Weiner0f3caba2008-07-23 21:28:07 -0700646 unsigned long goal)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700647{
Yinghai Lu08677212010-02-10 01:20:20 -0800648 unsigned long limit = 0;
649
Yinghai Lu08677212010-02-10 01:20:20 -0800650 return ___alloc_bootmem_nopanic(size, align, goal, limit);
Johannes Weiner0f3caba2008-07-23 21:28:07 -0700651}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652
Johannes Weiner0f3caba2008-07-23 21:28:07 -0700653static void * __init ___alloc_bootmem(unsigned long size, unsigned long align,
654 unsigned long goal, unsigned long limit)
655{
656 void *mem = ___alloc_bootmem_nopanic(size, align, goal, limit);
657
658 if (mem)
659 return mem;
660 /*
661 * Whoops, we cannot satisfy the allocation request.
662 */
663 printk(KERN_ALERT "bootmem alloc of %lu bytes failed!\n", size);
664 panic("Out of memory");
Andi Kleena8062232006-04-07 19:49:21 +0200665 return NULL;
666}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700667
Johannes Weinera66fd7d2008-07-23 21:28:01 -0700668/**
669 * __alloc_bootmem - allocate boot memory
670 * @size: size of the request in bytes
671 * @align: alignment of the region
672 * @goal: preferred starting address of the region
673 *
674 * The goal is dropped if it can not be satisfied and the allocation will
675 * fall back to memory below @goal.
676 *
677 * Allocation may happen on any node in the system.
678 *
679 * The function panics if the request can not be satisfied.
680 */
Franck Bui-Huubb0923a2006-09-25 23:31:05 -0700681void * __init __alloc_bootmem(unsigned long size, unsigned long align,
682 unsigned long goal)
Andi Kleena8062232006-04-07 19:49:21 +0200683{
Yinghai Lu08677212010-02-10 01:20:20 -0800684 unsigned long limit = 0;
685
Yinghai Lu08677212010-02-10 01:20:20 -0800686 return ___alloc_bootmem(size, align, goal, limit);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700687}
688
Johannes Weiner4cc278b2008-07-23 21:28:08 -0700689static void * __init ___alloc_bootmem_node(bootmem_data_t *bdata,
690 unsigned long size, unsigned long align,
691 unsigned long goal, unsigned long limit)
692{
693 void *ptr;
694
Tejun Heod0c4f572009-03-01 16:06:56 +0900695 ptr = alloc_arch_preferred_bootmem(bdata, size, align, goal, limit);
696 if (ptr)
697 return ptr;
698
Johannes Weiner4cc278b2008-07-23 21:28:08 -0700699 ptr = alloc_bootmem_core(bdata, size, align, goal, limit);
700 if (ptr)
701 return ptr;
702
703 return ___alloc_bootmem(size, align, goal, limit);
704}
705
Johannes Weinera66fd7d2008-07-23 21:28:01 -0700706/**
707 * __alloc_bootmem_node - allocate boot memory from a specific node
708 * @pgdat: node to allocate from
709 * @size: size of the request in bytes
710 * @align: alignment of the region
711 * @goal: preferred starting address of the region
712 *
713 * The goal is dropped if it can not be satisfied and the allocation will
714 * fall back to memory below @goal.
715 *
716 * Allocation may fall back to any node in the system if the specified node
717 * can not hold the requested memory.
718 *
719 * The function panics if the request can not be satisfied.
720 */
Franck Bui-Huubb0923a2006-09-25 23:31:05 -0700721void * __init __alloc_bootmem_node(pg_data_t *pgdat, unsigned long size,
722 unsigned long align, unsigned long goal)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700723{
Pekka Enbergc91c4772009-06-11 08:10:28 +0300724 if (WARN_ON_ONCE(slab_is_available()))
725 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
726
Yinghai Lu09325872011-02-24 14:43:05 +0100727 return ___alloc_bootmem_node(pgdat->bdata, size, align, goal, 0);
Yinghai Lu08677212010-02-10 01:20:20 -0800728}
729
730void * __init __alloc_bootmem_node_high(pg_data_t *pgdat, unsigned long size,
731 unsigned long align, unsigned long goal)
732{
733#ifdef MAX_DMA32_PFN
734 unsigned long end_pfn;
735
736 if (WARN_ON_ONCE(slab_is_available()))
737 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
738
739 /* update goal according ...MAX_DMA32_PFN */
740 end_pfn = pgdat->node_start_pfn + pgdat->node_spanned_pages;
741
742 if (end_pfn > MAX_DMA32_PFN + (128 >> (20 - PAGE_SHIFT)) &&
743 (goal >> PAGE_SHIFT) < MAX_DMA32_PFN) {
744 void *ptr;
745 unsigned long new_goal;
746
747 new_goal = MAX_DMA32_PFN << PAGE_SHIFT;
Yinghai Lu08677212010-02-10 01:20:20 -0800748 ptr = alloc_bootmem_core(pgdat->bdata, size, align,
749 new_goal, 0);
Yinghai Lu08677212010-02-10 01:20:20 -0800750 if (ptr)
751 return ptr;
752 }
753#endif
754
755 return __alloc_bootmem_node(pgdat, size, align, goal);
756
Linus Torvalds1da177e2005-04-16 15:20:36 -0700757}
758
Yasunori Gotoe70260a2008-04-28 02:13:32 -0700759#ifdef CONFIG_SPARSEMEM
Johannes Weinera66fd7d2008-07-23 21:28:01 -0700760/**
761 * alloc_bootmem_section - allocate boot memory from a specific section
762 * @size: size of the request in bytes
763 * @section_nr: sparse map section to allocate from
764 *
765 * Return NULL on failure.
766 */
Yasunori Gotoe70260a2008-04-28 02:13:32 -0700767void * __init alloc_bootmem_section(unsigned long size,
768 unsigned long section_nr)
769{
Johannes Weiner75a56cf2008-07-23 21:28:09 -0700770 bootmem_data_t *bdata;
771 unsigned long pfn, goal, limit;
Yasunori Gotoe70260a2008-04-28 02:13:32 -0700772
773 pfn = section_nr_to_pfn(section_nr);
Johannes Weiner75a56cf2008-07-23 21:28:09 -0700774 goal = pfn << PAGE_SHIFT;
775 limit = section_nr_to_pfn(section_nr + 1) << PAGE_SHIFT;
776 bdata = &bootmem_node_data[early_pfn_to_nid(pfn)];
Yasunori Gotoe70260a2008-04-28 02:13:32 -0700777
Johannes Weiner75a56cf2008-07-23 21:28:09 -0700778 return alloc_bootmem_core(bdata, size, SMP_CACHE_BYTES, goal, limit);
Yasunori Gotoe70260a2008-04-28 02:13:32 -0700779}
780#endif
781
Andi Kleenb54bbf72008-07-23 21:27:45 -0700782void * __init __alloc_bootmem_node_nopanic(pg_data_t *pgdat, unsigned long size,
783 unsigned long align, unsigned long goal)
784{
785 void *ptr;
786
Pekka Enbergc91c4772009-06-11 08:10:28 +0300787 if (WARN_ON_ONCE(slab_is_available()))
788 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
789
Tejun Heod0c4f572009-03-01 16:06:56 +0900790 ptr = alloc_arch_preferred_bootmem(pgdat->bdata, size, align, goal, 0);
791 if (ptr)
792 return ptr;
793
Andi Kleenb54bbf72008-07-23 21:27:45 -0700794 ptr = alloc_bootmem_core(pgdat->bdata, size, align, goal, 0);
795 if (ptr)
796 return ptr;
797
798 return __alloc_bootmem_nopanic(size, align, goal);
799}
800
Heiko Carstensdfd54cb2006-09-25 23:31:33 -0700801#ifndef ARCH_LOW_ADDRESS_LIMIT
802#define ARCH_LOW_ADDRESS_LIMIT 0xffffffffUL
803#endif
Ravikiran G Thirumalai008857c2006-01-06 00:11:01 -0800804
Johannes Weinera66fd7d2008-07-23 21:28:01 -0700805/**
806 * __alloc_bootmem_low - allocate low boot memory
807 * @size: size of the request in bytes
808 * @align: alignment of the region
809 * @goal: preferred starting address of the region
810 *
811 * The goal is dropped if it can not be satisfied and the allocation will
812 * fall back to memory below @goal.
813 *
814 * Allocation may happen on any node in the system.
815 *
816 * The function panics if the request can not be satisfied.
817 */
Franck Bui-Huubb0923a2006-09-25 23:31:05 -0700818void * __init __alloc_bootmem_low(unsigned long size, unsigned long align,
819 unsigned long goal)
Ravikiran G Thirumalai008857c2006-01-06 00:11:01 -0800820{
Johannes Weiner0f3caba2008-07-23 21:28:07 -0700821 return ___alloc_bootmem(size, align, goal, ARCH_LOW_ADDRESS_LIMIT);
Ravikiran G Thirumalai008857c2006-01-06 00:11:01 -0800822}
823
Johannes Weinera66fd7d2008-07-23 21:28:01 -0700824/**
825 * __alloc_bootmem_low_node - allocate low boot memory from a specific node
826 * @pgdat: node to allocate from
827 * @size: size of the request in bytes
828 * @align: alignment of the region
829 * @goal: preferred starting address of the region
830 *
831 * The goal is dropped if it can not be satisfied and the allocation will
832 * fall back to memory below @goal.
833 *
834 * Allocation may fall back to any node in the system if the specified node
835 * can not hold the requested memory.
836 *
837 * The function panics if the request can not be satisfied.
838 */
Ravikiran G Thirumalai008857c2006-01-06 00:11:01 -0800839void * __init __alloc_bootmem_low_node(pg_data_t *pgdat, unsigned long size,
840 unsigned long align, unsigned long goal)
841{
Pekka Enbergc91c4772009-06-11 08:10:28 +0300842 if (WARN_ON_ONCE(slab_is_available()))
843 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
844
Yinghai Lu09325872011-02-24 14:43:05 +0100845 return ___alloc_bootmem_node(pgdat->bdata, size, align,
Yinghai Lub8ab9f82010-07-20 13:24:31 -0700846 goal, ARCH_LOW_ADDRESS_LIMIT);
Ravikiran G Thirumalai008857c2006-01-06 00:11:01 -0800847}