blob: ec4fcb7a56c8975492d656940906af6153136e51 [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{
Uwe Kleine-König9571a982012-01-10 15:08:00 -080059 unsigned long bytes = DIV_ROUND_UP(pages, 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{
Gavin Shan5c2b8a12012-05-29 15:06:46 -070080 bootmem_data_t *ent;
Franck Bui-Huuf71bf0c2006-09-25 23:31:08 -070081
Gavin Shan5c2b8a12012-05-29 15:06:46 -070082 list_for_each_entry(ent, &bdata_list, list) {
83 if (bdata->node_min_pfn < ent->node_min_pfn) {
84 list_add_tail(&bdata->list, &ent->list);
85 return;
86 }
KAMEZAWA Hiroyuki679bc9f2006-03-27 01:15:58 -080087 }
Gavin Shan5c2b8a12012-05-29 15:06:46 -070088
89 list_add_tail(&bdata->list, &bdata_list);
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{
174 struct page *page;
Johannes Weiner41546c12008-07-23 21:28:03 -0700175 unsigned long start, end, pages, count = 0;
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700176
Johannes Weiner41546c12008-07-23 21:28:03 -0700177 if (!bdata->node_bootmem_map)
178 return 0;
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700179
Johannes Weiner3560e242008-07-23 21:28:09 -0700180 start = bdata->node_min_pfn;
Johannes Weiner41546c12008-07-23 21:28:03 -0700181 end = bdata->node_low_pfn;
182
Johannes Weiner799f9332012-01-10 15:08:15 -0800183 bdebug("nid=%td start=%lx end=%lx\n",
184 bdata - bootmem_node_data, start, end);
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700185
Johannes Weiner41546c12008-07-23 21:28:03 -0700186 while (start < end) {
187 unsigned long *map, idx, vec;
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700188
Johannes Weiner41546c12008-07-23 21:28:03 -0700189 map = bdata->node_bootmem_map;
Johannes Weiner3560e242008-07-23 21:28:09 -0700190 idx = start - bdata->node_min_pfn;
Johannes Weiner41546c12008-07-23 21:28:03 -0700191 vec = ~map[idx / BITS_PER_LONG];
Johannes Weiner799f9332012-01-10 15:08:15 -0800192 /*
193 * If we have a properly aligned and fully unreserved
194 * BITS_PER_LONG block of pages in front of us, free
195 * it in one go.
196 */
197 if (IS_ALIGNED(start, BITS_PER_LONG) && vec == ~0UL) {
Johannes Weiner41546c12008-07-23 21:28:03 -0700198 int order = ilog2(BITS_PER_LONG);
199
200 __free_pages_bootmem(pfn_to_page(start), order);
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700201 count += BITS_PER_LONG;
Johannes Weiner799f9332012-01-10 15:08:15 -0800202 start += BITS_PER_LONG;
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700203 } else {
Johannes Weiner41546c12008-07-23 21:28:03 -0700204 unsigned long off = 0;
205
Gavin Shan6dccdcb2012-05-29 15:06:32 -0700206 vec >>= start & (BITS_PER_LONG - 1);
Johannes Weiner549381e2012-05-29 15:06:32 -0700207 while (vec) {
Johannes Weiner41546c12008-07-23 21:28:03 -0700208 if (vec & 1) {
209 page = pfn_to_page(start + off);
210 __free_pages_bootmem(page, 0);
211 count++;
212 }
213 vec >>= 1;
214 off++;
215 }
Johannes Weiner799f9332012-01-10 15:08:15 -0800216 start = ALIGN(start + 1, BITS_PER_LONG);
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700217 }
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700218 }
219
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700220 page = virt_to_page(bdata->node_bootmem_map);
Johannes Weiner3560e242008-07-23 21:28:09 -0700221 pages = bdata->node_low_pfn - bdata->node_min_pfn;
Johannes Weiner41546c12008-07-23 21:28:03 -0700222 pages = bootmem_bootmap_pages(pages);
223 count += pages;
224 while (pages--)
225 __free_pages_bootmem(page++, 0);
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700226
Johannes Weiner2e5237d2008-07-23 21:28:02 -0700227 bdebug("nid=%td released=%lx\n", bdata - bootmem_node_data, count);
228
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700229 return count;
230}
231
Johannes Weinera66fd7d2008-07-23 21:28:01 -0700232/**
233 * free_all_bootmem_node - release a node's free pages to the buddy allocator
234 * @pgdat: node to be released
235 *
236 * Returns the number of pages actually released.
237 */
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700238unsigned long __init free_all_bootmem_node(pg_data_t *pgdat)
239{
240 register_page_bootmem_info_node(pgdat);
241 return free_all_bootmem_core(pgdat->bdata);
242}
243
Johannes Weinera66fd7d2008-07-23 21:28:01 -0700244/**
245 * free_all_bootmem - release free pages to the buddy allocator
246 *
247 * Returns the number of pages actually released.
248 */
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700249unsigned long __init free_all_bootmem(void)
250{
Yinghai Luaa235fc2010-03-31 20:45:27 -0700251 unsigned long total_pages = 0;
252 bootmem_data_t *bdata;
253
254 list_for_each_entry(bdata, &bdata_list, list)
255 total_pages += free_all_bootmem_core(bdata);
256
257 return total_pages;
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700258}
259
Johannes Weinerd747fa42008-07-23 21:28:05 -0700260static void __init __free(bootmem_data_t *bdata,
261 unsigned long sidx, unsigned long eidx)
262{
263 unsigned long idx;
264
265 bdebug("nid=%td start=%lx end=%lx\n", bdata - bootmem_node_data,
Johannes Weiner3560e242008-07-23 21:28:09 -0700266 sidx + bdata->node_min_pfn,
267 eidx + bdata->node_min_pfn);
Johannes Weinerd747fa42008-07-23 21:28:05 -0700268
Johannes Weinere2bf3ca2008-07-23 21:28:06 -0700269 if (bdata->hint_idx > sidx)
270 bdata->hint_idx = sidx;
271
Johannes Weinerd747fa42008-07-23 21:28:05 -0700272 for (idx = sidx; idx < eidx; idx++)
273 if (!test_and_clear_bit(idx, bdata->node_bootmem_map))
274 BUG();
275}
276
277static int __init __reserve(bootmem_data_t *bdata, unsigned long sidx,
278 unsigned long eidx, int flags)
279{
280 unsigned long idx;
281 int exclusive = flags & BOOTMEM_EXCLUSIVE;
282
283 bdebug("nid=%td start=%lx end=%lx flags=%x\n",
284 bdata - bootmem_node_data,
Johannes Weiner3560e242008-07-23 21:28:09 -0700285 sidx + bdata->node_min_pfn,
286 eidx + bdata->node_min_pfn,
Johannes Weinerd747fa42008-07-23 21:28:05 -0700287 flags);
288
289 for (idx = sidx; idx < eidx; idx++)
290 if (test_and_set_bit(idx, bdata->node_bootmem_map)) {
291 if (exclusive) {
292 __free(bdata, sidx, idx);
293 return -EBUSY;
294 }
295 bdebug("silent double reserve of PFN %lx\n",
Johannes Weiner3560e242008-07-23 21:28:09 -0700296 idx + bdata->node_min_pfn);
Johannes Weinerd747fa42008-07-23 21:28:05 -0700297 }
298 return 0;
299}
300
Johannes Weinere2bf3ca2008-07-23 21:28:06 -0700301static int __init mark_bootmem_node(bootmem_data_t *bdata,
302 unsigned long start, unsigned long end,
303 int reserve, int flags)
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700304{
305 unsigned long sidx, eidx;
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700306
Johannes Weinere2bf3ca2008-07-23 21:28:06 -0700307 bdebug("nid=%td start=%lx end=%lx reserve=%d flags=%x\n",
308 bdata - bootmem_node_data, start, end, reserve, flags);
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700309
Johannes Weiner3560e242008-07-23 21:28:09 -0700310 BUG_ON(start < bdata->node_min_pfn);
Johannes Weinere2bf3ca2008-07-23 21:28:06 -0700311 BUG_ON(end > bdata->node_low_pfn);
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700312
Johannes Weiner3560e242008-07-23 21:28:09 -0700313 sidx = start - bdata->node_min_pfn;
314 eidx = end - bdata->node_min_pfn;
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700315
Johannes Weinere2bf3ca2008-07-23 21:28:06 -0700316 if (reserve)
317 return __reserve(bdata, sidx, eidx, flags);
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700318 else
Johannes Weinere2bf3ca2008-07-23 21:28:06 -0700319 __free(bdata, sidx, eidx);
320 return 0;
321}
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700322
Johannes Weinere2bf3ca2008-07-23 21:28:06 -0700323static int __init mark_bootmem(unsigned long start, unsigned long end,
324 int reserve, int flags)
325{
326 unsigned long pos;
327 bootmem_data_t *bdata;
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700328
Johannes Weinere2bf3ca2008-07-23 21:28:06 -0700329 pos = start;
330 list_for_each_entry(bdata, &bdata_list, list) {
331 int err;
332 unsigned long max;
333
Johannes Weiner3560e242008-07-23 21:28:09 -0700334 if (pos < bdata->node_min_pfn ||
335 pos >= bdata->node_low_pfn) {
Johannes Weinere2bf3ca2008-07-23 21:28:06 -0700336 BUG_ON(pos != start);
337 continue;
338 }
339
340 max = min(bdata->node_low_pfn, end);
341
342 err = mark_bootmem_node(bdata, pos, max, reserve, flags);
343 if (reserve && err) {
344 mark_bootmem(start, pos, 0, 0);
345 return err;
346 }
347
348 if (max == end)
349 return 0;
350 pos = bdata->node_low_pfn;
351 }
352 BUG();
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700353}
354
Johannes Weinera66fd7d2008-07-23 21:28:01 -0700355/**
356 * free_bootmem_node - mark a page range as usable
357 * @pgdat: node the range resides on
358 * @physaddr: starting address of the range
359 * @size: size of the range in bytes
360 *
361 * Partial pages will be considered reserved and left as they are.
362 *
Johannes Weinere2bf3ca2008-07-23 21:28:06 -0700363 * The range must reside completely on the specified node.
Johannes Weinera66fd7d2008-07-23 21:28:01 -0700364 */
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700365void __init free_bootmem_node(pg_data_t *pgdat, unsigned long physaddr,
366 unsigned long size)
367{
Johannes Weinere2bf3ca2008-07-23 21:28:06 -0700368 unsigned long start, end;
369
Catalin Marinasec3a3542009-07-07 10:33:01 +0100370 kmemleak_free_part(__va(physaddr), size);
371
Johannes Weinere2bf3ca2008-07-23 21:28:06 -0700372 start = PFN_UP(physaddr);
373 end = PFN_DOWN(physaddr + size);
374
375 mark_bootmem_node(pgdat->bdata, start, end, 0, 0);
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700376}
377
Johannes Weinera66fd7d2008-07-23 21:28:01 -0700378/**
379 * free_bootmem - mark a page range as usable
380 * @addr: starting address of the range
381 * @size: size of the range in bytes
382 *
383 * Partial pages will be considered reserved and left as they are.
384 *
Johannes Weinere2bf3ca2008-07-23 21:28:06 -0700385 * The range must be contiguous but may span node boundaries.
Johannes Weinera66fd7d2008-07-23 21:28:01 -0700386 */
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700387void __init free_bootmem(unsigned long addr, unsigned long size)
388{
Johannes Weinere2bf3ca2008-07-23 21:28:06 -0700389 unsigned long start, end;
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700390
Catalin Marinasec3a3542009-07-07 10:33:01 +0100391 kmemleak_free_part(__va(addr), size);
392
Johannes Weinere2bf3ca2008-07-23 21:28:06 -0700393 start = PFN_UP(addr);
394 end = PFN_DOWN(addr + size);
Yinghai Lua5645a62008-03-18 12:49:12 -0700395
Johannes Weinere2bf3ca2008-07-23 21:28:06 -0700396 mark_bootmem(start, end, 0, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700397}
398
Johannes Weinera66fd7d2008-07-23 21:28:01 -0700399/**
400 * reserve_bootmem_node - mark a page range as reserved
401 * @pgdat: node the range resides on
402 * @physaddr: starting address of the range
403 * @size: size of the range in bytes
404 * @flags: reservation flags (see linux/bootmem.h)
405 *
406 * Partial pages will be reserved.
407 *
Johannes Weinere2bf3ca2008-07-23 21:28:06 -0700408 * The range must reside completely on the specified node.
Johannes Weinera66fd7d2008-07-23 21:28:01 -0700409 */
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700410int __init reserve_bootmem_node(pg_data_t *pgdat, unsigned long physaddr,
411 unsigned long size, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700412{
Johannes Weinere2bf3ca2008-07-23 21:28:06 -0700413 unsigned long start, end;
Franck Bui-Huubbc7b922006-09-25 23:31:07 -0700414
Johannes Weinere2bf3ca2008-07-23 21:28:06 -0700415 start = PFN_DOWN(physaddr);
416 end = PFN_UP(physaddr + size);
417
418 return mark_bootmem_node(pgdat->bdata, start, end, 1, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700419}
420
Johannes Weinera66fd7d2008-07-23 21:28:01 -0700421/**
422 * reserve_bootmem - mark a page range as usable
423 * @addr: starting address of the range
424 * @size: size of the range in bytes
425 * @flags: reservation flags (see linux/bootmem.h)
426 *
427 * Partial pages will be reserved.
428 *
Johannes Weinere2bf3ca2008-07-23 21:28:06 -0700429 * The range must be contiguous but may span node boundaries.
Johannes Weinera66fd7d2008-07-23 21:28:01 -0700430 */
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700431int __init reserve_bootmem(unsigned long addr, unsigned long size,
432 int flags)
433{
Johannes Weinere2bf3ca2008-07-23 21:28:06 -0700434 unsigned long start, end;
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700435
Johannes Weinere2bf3ca2008-07-23 21:28:06 -0700436 start = PFN_DOWN(addr);
437 end = PFN_UP(addr + size);
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700438
Johannes Weinere2bf3ca2008-07-23 21:28:06 -0700439 return mark_bootmem(start, end, 1, flags);
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700440}
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700441
Yinghai Luf88eff72010-08-25 13:39:15 -0700442int __weak __init reserve_bootmem_generic(unsigned long phys, unsigned long len,
443 int flags)
444{
445 return reserve_bootmem(phys, len, flags);
446}
447
Jan Beulich8aa043d2009-12-14 17:59:46 -0800448static unsigned long __init align_idx(struct bootmem_data *bdata,
449 unsigned long idx, unsigned long step)
Johannes Weiner481ebd02008-08-20 14:09:15 -0700450{
451 unsigned long base = bdata->node_min_pfn;
452
453 /*
454 * Align the index with respect to the node start so that the
455 * combination of both satisfies the requested alignment.
456 */
457
458 return ALIGN(base + idx, step) - base;
459}
460
Jan Beulich8aa043d2009-12-14 17:59:46 -0800461static unsigned long __init align_off(struct bootmem_data *bdata,
462 unsigned long off, unsigned long align)
Johannes Weiner481ebd02008-08-20 14:09:15 -0700463{
464 unsigned long base = PFN_PHYS(bdata->node_min_pfn);
465
466 /* Same as align_idx for byte offsets */
467
468 return ALIGN(base + off, align) - base;
469}
470
Johannes Weinerc6785b62012-05-29 15:06:33 -0700471static void * __init alloc_bootmem_bdata(struct bootmem_data *bdata,
Tejun Heod0c4f572009-03-01 16:06:56 +0900472 unsigned long size, unsigned long align,
473 unsigned long goal, unsigned long limit)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700474{
Johannes Weiner0f3caba2008-07-23 21:28:07 -0700475 unsigned long fallback = 0;
Johannes Weiner5f2809e2008-07-23 21:28:05 -0700476 unsigned long min, max, start, sidx, midx, step;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700477
Johannes Weiner594fe1a2009-01-06 14:40:32 -0800478 bdebug("nid=%td size=%lx [%lu pages] align=%lx goal=%lx limit=%lx\n",
479 bdata - bootmem_node_data, size, PAGE_ALIGN(size) >> PAGE_SHIFT,
480 align, goal, limit);
481
Johannes Weiner5f2809e2008-07-23 21:28:05 -0700482 BUG_ON(!size);
483 BUG_ON(align & (align - 1));
484 BUG_ON(limit && goal + size > limit);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700485
Christian Krafft7c309a62006-12-06 20:32:41 -0800486 if (!bdata->node_bootmem_map)
487 return NULL;
488
Johannes Weiner3560e242008-07-23 21:28:09 -0700489 min = bdata->node_min_pfn;
Johannes Weiner5f2809e2008-07-23 21:28:05 -0700490 max = bdata->node_low_pfn;
Yinghai Lu9a2dc042008-03-18 12:44:48 -0700491
Johannes Weiner5f2809e2008-07-23 21:28:05 -0700492 goal >>= PAGE_SHIFT;
493 limit >>= PAGE_SHIFT;
494
495 if (limit && max > limit)
496 max = limit;
497 if (max <= min)
Yinghai Lu9a2dc042008-03-18 12:44:48 -0700498 return NULL;
499
Johannes Weiner5f2809e2008-07-23 21:28:05 -0700500 step = max(align >> PAGE_SHIFT, 1UL);
Yasunori Goto281dd252005-10-19 15:52:18 -0700501
Johannes Weiner5f2809e2008-07-23 21:28:05 -0700502 if (goal && min < goal && goal < max)
503 start = ALIGN(goal, step);
504 else
505 start = ALIGN(min, step);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700506
Johannes Weiner481ebd02008-08-20 14:09:15 -0700507 sidx = start - bdata->node_min_pfn;
Johannes Weiner3560e242008-07-23 21:28:09 -0700508 midx = max - bdata->node_min_pfn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700509
Johannes Weiner5f2809e2008-07-23 21:28:05 -0700510 if (bdata->hint_idx > sidx) {
Johannes Weiner0f3caba2008-07-23 21:28:07 -0700511 /*
512 * Handle the valid case of sidx being zero and still
513 * catch the fallback below.
514 */
515 fallback = sidx + 1;
Johannes Weiner481ebd02008-08-20 14:09:15 -0700516 sidx = align_idx(bdata, bdata->hint_idx, step);
Yinghai Luad093152008-03-10 23:23:42 -0700517 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700518
Johannes Weiner5f2809e2008-07-23 21:28:05 -0700519 while (1) {
520 int merge;
521 void *region;
522 unsigned long eidx, i, start_off, end_off;
523find_block:
524 sidx = find_next_zero_bit(bdata->node_bootmem_map, midx, sidx);
Johannes Weiner481ebd02008-08-20 14:09:15 -0700525 sidx = align_idx(bdata, sidx, step);
Johannes Weiner5f2809e2008-07-23 21:28:05 -0700526 eidx = sidx + PFN_UP(size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700527
Johannes Weiner5f2809e2008-07-23 21:28:05 -0700528 if (sidx >= midx || eidx > midx)
Haren Myneni66d43e92005-12-12 00:37:39 -0800529 break;
Johannes Weiner5f2809e2008-07-23 21:28:05 -0700530
531 for (i = sidx; i < eidx; i++)
532 if (test_bit(i, bdata->node_bootmem_map)) {
Johannes Weiner481ebd02008-08-20 14:09:15 -0700533 sidx = align_idx(bdata, i, step);
Johannes Weiner5f2809e2008-07-23 21:28:05 -0700534 if (sidx == i)
535 sidx += step;
536 goto find_block;
537 }
538
Mikulas Patocka627240a2008-08-15 00:40:17 -0700539 if (bdata->last_end_off & (PAGE_SIZE - 1) &&
Johannes Weiner5f2809e2008-07-23 21:28:05 -0700540 PFN_DOWN(bdata->last_end_off) + 1 == sidx)
Johannes Weiner481ebd02008-08-20 14:09:15 -0700541 start_off = align_off(bdata, bdata->last_end_off, align);
Johannes Weiner5f2809e2008-07-23 21:28:05 -0700542 else
543 start_off = PFN_PHYS(sidx);
544
545 merge = PFN_DOWN(start_off) < sidx;
546 end_off = start_off + size;
547
548 bdata->last_end_off = end_off;
549 bdata->hint_idx = PFN_UP(end_off);
550
551 /*
552 * Reserve the area now:
553 */
Johannes Weinerd747fa42008-07-23 21:28:05 -0700554 if (__reserve(bdata, PFN_DOWN(start_off) + merge,
555 PFN_UP(end_off), BOOTMEM_EXCLUSIVE))
556 BUG();
Johannes Weiner5f2809e2008-07-23 21:28:05 -0700557
Johannes Weiner3560e242008-07-23 21:28:09 -0700558 region = phys_to_virt(PFN_PHYS(bdata->node_min_pfn) +
559 start_off);
Johannes Weiner5f2809e2008-07-23 21:28:05 -0700560 memset(region, 0, size);
Catalin Marinas008139d2009-08-27 14:29:17 +0100561 /*
562 * The min_count is set to 0 so that bootmem allocated blocks
563 * are never reported as leaks.
564 */
565 kmemleak_alloc(region, size, 0, 0);
Johannes Weiner5f2809e2008-07-23 21:28:05 -0700566 return region;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567 }
568
Johannes Weiner0f3caba2008-07-23 21:28:07 -0700569 if (fallback) {
Johannes Weiner481ebd02008-08-20 14:09:15 -0700570 sidx = align_idx(bdata, fallback - 1, step);
Johannes Weiner0f3caba2008-07-23 21:28:07 -0700571 fallback = 0;
572 goto find_block;
573 }
574
575 return NULL;
576}
577
Tejun Heod0c4f572009-03-01 16:06:56 +0900578static void * __init alloc_arch_preferred_bootmem(bootmem_data_t *bdata,
579 unsigned long size, unsigned long align,
580 unsigned long goal, unsigned long limit)
581{
Pekka Enberg441c7e02009-06-10 20:05:53 +0300582 if (WARN_ON_ONCE(slab_is_available()))
583 return kzalloc(size, GFP_NOWAIT);
584
Tejun Heod0c4f572009-03-01 16:06:56 +0900585#ifdef CONFIG_HAVE_ARCH_BOOTMEM
Joe Perches433f13a2009-06-18 16:48:58 -0700586 {
587 bootmem_data_t *p_bdata;
Tejun Heod0c4f572009-03-01 16:06:56 +0900588
Joe Perches433f13a2009-06-18 16:48:58 -0700589 p_bdata = bootmem_arch_preferred_node(bdata, size, align,
590 goal, limit);
591 if (p_bdata)
Johannes Weinerc6785b62012-05-29 15:06:33 -0700592 return alloc_bootmem_bdata(p_bdata, size, align,
Joe Perches433f13a2009-06-18 16:48:58 -0700593 goal, limit);
594 }
Tejun Heod0c4f572009-03-01 16:06:56 +0900595#endif
596 return NULL;
597}
598
Johannes Weinerc12ab502012-05-29 15:06:33 -0700599static void * __init alloc_bootmem_core(unsigned long size,
Johannes Weiner0f3caba2008-07-23 21:28:07 -0700600 unsigned long align,
601 unsigned long goal,
602 unsigned long limit)
603{
604 bootmem_data_t *bdata;
Tejun Heod0c4f572009-03-01 16:06:56 +0900605 void *region;
Johannes Weiner0f3caba2008-07-23 21:28:07 -0700606
Tejun Heod0c4f572009-03-01 16:06:56 +0900607 region = alloc_arch_preferred_bootmem(NULL, size, align, goal, limit);
608 if (region)
609 return region;
Johannes Weiner0f3caba2008-07-23 21:28:07 -0700610
Tejun Heod0c4f572009-03-01 16:06:56 +0900611 list_for_each_entry(bdata, &bdata_list, list) {
Johannes Weiner0f3caba2008-07-23 21:28:07 -0700612 if (goal && bdata->node_low_pfn <= PFN_DOWN(goal))
613 continue;
Johannes Weiner3560e242008-07-23 21:28:09 -0700614 if (limit && bdata->node_min_pfn >= PFN_DOWN(limit))
Johannes Weiner0f3caba2008-07-23 21:28:07 -0700615 break;
616
Johannes Weinerc6785b62012-05-29 15:06:33 -0700617 region = alloc_bootmem_bdata(bdata, size, align, goal, limit);
Johannes Weiner0f3caba2008-07-23 21:28:07 -0700618 if (region)
619 return region;
620 }
621
Johannes Weinerc12ab502012-05-29 15:06:33 -0700622 return NULL;
623}
624
625static void * __init ___alloc_bootmem_nopanic(unsigned long size,
626 unsigned long align,
627 unsigned long goal,
628 unsigned long limit)
629{
630 void *ptr;
631
632restart:
633 ptr = alloc_bootmem_core(size, align, goal, limit);
634 if (ptr)
635 return ptr;
Johannes Weiner5f2809e2008-07-23 21:28:05 -0700636 if (goal) {
637 goal = 0;
Johannes Weiner0f3caba2008-07-23 21:28:07 -0700638 goto restart;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700639 }
Johannes Weiner5f2809e2008-07-23 21:28:05 -0700640
Linus Torvalds1da177e2005-04-16 15:20:36 -0700641 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700642}
643
Johannes Weinera66fd7d2008-07-23 21:28:01 -0700644/**
645 * __alloc_bootmem_nopanic - allocate boot memory without panicking
646 * @size: size of the request in bytes
647 * @align: alignment of the region
648 * @goal: preferred starting address of the region
649 *
650 * The goal is dropped if it can not be satisfied and the allocation will
651 * fall back to memory below @goal.
652 *
653 * Allocation may happen on any node in the system.
654 *
655 * Returns NULL on failure.
656 */
Franck Bui-Huubb0923a2006-09-25 23:31:05 -0700657void * __init __alloc_bootmem_nopanic(unsigned long size, unsigned long align,
Johannes Weiner0f3caba2008-07-23 21:28:07 -0700658 unsigned long goal)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700659{
Yinghai Lu08677212010-02-10 01:20:20 -0800660 unsigned long limit = 0;
661
Yinghai Lu08677212010-02-10 01:20:20 -0800662 return ___alloc_bootmem_nopanic(size, align, goal, limit);
Johannes Weiner0f3caba2008-07-23 21:28:07 -0700663}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700664
Johannes Weiner0f3caba2008-07-23 21:28:07 -0700665static void * __init ___alloc_bootmem(unsigned long size, unsigned long align,
666 unsigned long goal, unsigned long limit)
667{
668 void *mem = ___alloc_bootmem_nopanic(size, align, goal, limit);
669
670 if (mem)
671 return mem;
672 /*
673 * Whoops, we cannot satisfy the allocation request.
674 */
675 printk(KERN_ALERT "bootmem alloc of %lu bytes failed!\n", size);
676 panic("Out of memory");
Andi Kleena8062232006-04-07 19:49:21 +0200677 return NULL;
678}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700679
Johannes Weinera66fd7d2008-07-23 21:28:01 -0700680/**
681 * __alloc_bootmem - allocate boot memory
682 * @size: size of the request in bytes
683 * @align: alignment of the region
684 * @goal: preferred starting address of the region
685 *
686 * The goal is dropped if it can not be satisfied and the allocation will
687 * fall back to memory below @goal.
688 *
689 * Allocation may happen on any node in the system.
690 *
691 * The function panics if the request can not be satisfied.
692 */
Franck Bui-Huubb0923a2006-09-25 23:31:05 -0700693void * __init __alloc_bootmem(unsigned long size, unsigned long align,
694 unsigned long goal)
Andi Kleena8062232006-04-07 19:49:21 +0200695{
Yinghai Lu08677212010-02-10 01:20:20 -0800696 unsigned long limit = 0;
697
Yinghai Lu08677212010-02-10 01:20:20 -0800698 return ___alloc_bootmem(size, align, goal, limit);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700699}
700
Johannes Weinere9079912012-05-29 15:06:36 -0700701static void * __init ___alloc_bootmem_node_nopanic(pg_data_t *pgdat,
Johannes Weiner4cc278b2008-07-23 21:28:08 -0700702 unsigned long size, unsigned long align,
703 unsigned long goal, unsigned long limit)
704{
705 void *ptr;
706
Johannes Weinerab381842012-05-29 15:06:34 -0700707again:
Johannes Weinere9079912012-05-29 15:06:36 -0700708 ptr = alloc_arch_preferred_bootmem(pgdat->bdata, size,
709 align, goal, limit);
Tejun Heod0c4f572009-03-01 16:06:56 +0900710 if (ptr)
711 return ptr;
712
Johannes Weinere9079912012-05-29 15:06:36 -0700713 ptr = alloc_bootmem_bdata(pgdat->bdata, size, align, goal, limit);
Johannes Weiner4cc278b2008-07-23 21:28:08 -0700714 if (ptr)
715 return ptr;
716
Johannes Weinerab381842012-05-29 15:06:34 -0700717 ptr = alloc_bootmem_core(size, align, goal, limit);
718 if (ptr)
719 return ptr;
720
721 if (goal) {
722 goal = 0;
723 goto again;
724 }
725
Johannes Weiner421456e2012-05-29 15:06:34 -0700726 return NULL;
727}
728
729void * __init __alloc_bootmem_node_nopanic(pg_data_t *pgdat, unsigned long size,
730 unsigned long align, unsigned long goal)
731{
732 if (WARN_ON_ONCE(slab_is_available()))
733 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
734
Johannes Weinere9079912012-05-29 15:06:36 -0700735 return ___alloc_bootmem_node_nopanic(pgdat, size, align, goal, 0);
Johannes Weiner421456e2012-05-29 15:06:34 -0700736}
737
Johannes Weinere9079912012-05-29 15:06:36 -0700738void * __init ___alloc_bootmem_node(pg_data_t *pgdat, unsigned long size,
Johannes Weiner421456e2012-05-29 15:06:34 -0700739 unsigned long align, unsigned long goal,
740 unsigned long limit)
741{
742 void *ptr;
743
Johannes Weinere9079912012-05-29 15:06:36 -0700744 ptr = ___alloc_bootmem_node_nopanic(pgdat, size, align, goal, 0);
Johannes Weiner421456e2012-05-29 15:06:34 -0700745 if (ptr)
746 return ptr;
747
Johannes Weinerab381842012-05-29 15:06:34 -0700748 printk(KERN_ALERT "bootmem alloc of %lu bytes failed!\n", size);
749 panic("Out of memory");
750 return NULL;
Johannes Weiner4cc278b2008-07-23 21:28:08 -0700751}
752
Johannes Weinera66fd7d2008-07-23 21:28:01 -0700753/**
754 * __alloc_bootmem_node - allocate boot memory from a specific node
755 * @pgdat: node to allocate from
756 * @size: size of the request in bytes
757 * @align: alignment of the region
758 * @goal: preferred starting address of the region
759 *
760 * The goal is dropped if it can not be satisfied and the allocation will
761 * fall back to memory below @goal.
762 *
763 * Allocation may fall back to any node in the system if the specified node
764 * can not hold the requested memory.
765 *
766 * The function panics if the request can not be satisfied.
767 */
Franck Bui-Huubb0923a2006-09-25 23:31:05 -0700768void * __init __alloc_bootmem_node(pg_data_t *pgdat, unsigned long size,
769 unsigned long align, unsigned long goal)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700770{
Pekka Enbergc91c4772009-06-11 08:10:28 +0300771 if (WARN_ON_ONCE(slab_is_available()))
772 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
773
Johannes Weinere9079912012-05-29 15:06:36 -0700774 return ___alloc_bootmem_node(pgdat, size, align, goal, 0);
Yinghai Lu08677212010-02-10 01:20:20 -0800775}
776
777void * __init __alloc_bootmem_node_high(pg_data_t *pgdat, unsigned long size,
778 unsigned long align, unsigned long goal)
779{
780#ifdef MAX_DMA32_PFN
781 unsigned long end_pfn;
782
783 if (WARN_ON_ONCE(slab_is_available()))
784 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
785
786 /* update goal according ...MAX_DMA32_PFN */
787 end_pfn = pgdat->node_start_pfn + pgdat->node_spanned_pages;
788
789 if (end_pfn > MAX_DMA32_PFN + (128 >> (20 - PAGE_SHIFT)) &&
790 (goal >> PAGE_SHIFT) < MAX_DMA32_PFN) {
791 void *ptr;
792 unsigned long new_goal;
793
794 new_goal = MAX_DMA32_PFN << PAGE_SHIFT;
Johannes Weinerc6785b62012-05-29 15:06:33 -0700795 ptr = alloc_bootmem_bdata(pgdat->bdata, size, align,
Yinghai Lu08677212010-02-10 01:20:20 -0800796 new_goal, 0);
Yinghai Lu08677212010-02-10 01:20:20 -0800797 if (ptr)
798 return ptr;
799 }
800#endif
801
802 return __alloc_bootmem_node(pgdat, size, align, goal);
803
Linus Torvalds1da177e2005-04-16 15:20:36 -0700804}
805
Heiko Carstensdfd54cb2006-09-25 23:31:33 -0700806#ifndef ARCH_LOW_ADDRESS_LIMIT
807#define ARCH_LOW_ADDRESS_LIMIT 0xffffffffUL
808#endif
Ravikiran G Thirumalai008857c2006-01-06 00:11:01 -0800809
Johannes Weinera66fd7d2008-07-23 21:28:01 -0700810/**
811 * __alloc_bootmem_low - allocate low boot memory
812 * @size: size of the request in bytes
813 * @align: alignment of the region
814 * @goal: preferred starting address of the region
815 *
816 * The goal is dropped if it can not be satisfied and the allocation will
817 * fall back to memory below @goal.
818 *
819 * Allocation may happen on any node in the system.
820 *
821 * The function panics if the request can not be satisfied.
822 */
Franck Bui-Huubb0923a2006-09-25 23:31:05 -0700823void * __init __alloc_bootmem_low(unsigned long size, unsigned long align,
824 unsigned long goal)
Ravikiran G Thirumalai008857c2006-01-06 00:11:01 -0800825{
Johannes Weiner0f3caba2008-07-23 21:28:07 -0700826 return ___alloc_bootmem(size, align, goal, ARCH_LOW_ADDRESS_LIMIT);
Ravikiran G Thirumalai008857c2006-01-06 00:11:01 -0800827}
828
Johannes Weinera66fd7d2008-07-23 21:28:01 -0700829/**
830 * __alloc_bootmem_low_node - allocate low boot memory from a specific node
831 * @pgdat: node to allocate from
832 * @size: size of the request in bytes
833 * @align: alignment of the region
834 * @goal: preferred starting address of the region
835 *
836 * The goal is dropped if it can not be satisfied and the allocation will
837 * fall back to memory below @goal.
838 *
839 * Allocation may fall back to any node in the system if the specified node
840 * can not hold the requested memory.
841 *
842 * The function panics if the request can not be satisfied.
843 */
Ravikiran G Thirumalai008857c2006-01-06 00:11:01 -0800844void * __init __alloc_bootmem_low_node(pg_data_t *pgdat, unsigned long size,
845 unsigned long align, unsigned long goal)
846{
Pekka Enbergc91c4772009-06-11 08:10:28 +0300847 if (WARN_ON_ONCE(slab_is_available()))
848 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
849
Johannes Weinere9079912012-05-29 15:06:36 -0700850 return ___alloc_bootmem_node(pgdat, size, align,
851 goal, ARCH_LOW_ADDRESS_LIMIT);
Ravikiran G Thirumalai008857c2006-01-06 00:11:01 -0800852}