blob: 0131170c9d540a572c7b2ba3108ca5c2d9db30b7 [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{
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{
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
206 while (vec && off < BITS_PER_LONG) {
207 if (vec & 1) {
208 page = pfn_to_page(start + off);
209 __free_pages_bootmem(page, 0);
210 count++;
211 }
212 vec >>= 1;
213 off++;
214 }
Johannes Weiner799f9332012-01-10 15:08:15 -0800215 start = ALIGN(start + 1, BITS_PER_LONG);
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700216 }
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700217 }
218
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700219 page = virt_to_page(bdata->node_bootmem_map);
Johannes Weiner3560e242008-07-23 21:28:09 -0700220 pages = bdata->node_low_pfn - bdata->node_min_pfn;
Johannes Weiner41546c12008-07-23 21:28:03 -0700221 pages = bootmem_bootmap_pages(pages);
222 count += pages;
223 while (pages--)
224 __free_pages_bootmem(page++, 0);
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700225
Johannes Weiner2e5237d2008-07-23 21:28:02 -0700226 bdebug("nid=%td released=%lx\n", bdata - bootmem_node_data, count);
227
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700228 return count;
229}
230
Johannes Weinera66fd7d2008-07-23 21:28:01 -0700231/**
232 * free_all_bootmem_node - release a node's free pages to the buddy allocator
233 * @pgdat: node to be released
234 *
235 * Returns the number of pages actually released.
236 */
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700237unsigned long __init free_all_bootmem_node(pg_data_t *pgdat)
238{
239 register_page_bootmem_info_node(pgdat);
240 return free_all_bootmem_core(pgdat->bdata);
241}
242
Johannes Weinera66fd7d2008-07-23 21:28:01 -0700243/**
244 * free_all_bootmem - release free pages to the buddy allocator
245 *
246 * Returns the number of pages actually released.
247 */
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700248unsigned long __init free_all_bootmem(void)
249{
Yinghai Luaa235fc2010-03-31 20:45:27 -0700250 unsigned long total_pages = 0;
251 bootmem_data_t *bdata;
252
253 list_for_each_entry(bdata, &bdata_list, list)
254 total_pages += free_all_bootmem_core(bdata);
255
256 return total_pages;
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700257}
258
Johannes Weinerd747fa42008-07-23 21:28:05 -0700259static void __init __free(bootmem_data_t *bdata,
260 unsigned long sidx, unsigned long eidx)
261{
262 unsigned long idx;
263
264 bdebug("nid=%td start=%lx end=%lx\n", bdata - bootmem_node_data,
Johannes Weiner3560e242008-07-23 21:28:09 -0700265 sidx + bdata->node_min_pfn,
266 eidx + bdata->node_min_pfn);
Johannes Weinerd747fa42008-07-23 21:28:05 -0700267
Johannes Weinere2bf3ca2008-07-23 21:28:06 -0700268 if (bdata->hint_idx > sidx)
269 bdata->hint_idx = sidx;
270
Johannes Weinerd747fa42008-07-23 21:28:05 -0700271 for (idx = sidx; idx < eidx; idx++)
272 if (!test_and_clear_bit(idx, bdata->node_bootmem_map))
273 BUG();
274}
275
276static int __init __reserve(bootmem_data_t *bdata, unsigned long sidx,
277 unsigned long eidx, int flags)
278{
279 unsigned long idx;
280 int exclusive = flags & BOOTMEM_EXCLUSIVE;
281
282 bdebug("nid=%td start=%lx end=%lx flags=%x\n",
283 bdata - bootmem_node_data,
Johannes Weiner3560e242008-07-23 21:28:09 -0700284 sidx + bdata->node_min_pfn,
285 eidx + bdata->node_min_pfn,
Johannes Weinerd747fa42008-07-23 21:28:05 -0700286 flags);
287
288 for (idx = sidx; idx < eidx; idx++)
289 if (test_and_set_bit(idx, bdata->node_bootmem_map)) {
290 if (exclusive) {
291 __free(bdata, sidx, idx);
292 return -EBUSY;
293 }
294 bdebug("silent double reserve of PFN %lx\n",
Johannes Weiner3560e242008-07-23 21:28:09 -0700295 idx + bdata->node_min_pfn);
Johannes Weinerd747fa42008-07-23 21:28:05 -0700296 }
297 return 0;
298}
299
Johannes Weinere2bf3ca2008-07-23 21:28:06 -0700300static int __init mark_bootmem_node(bootmem_data_t *bdata,
301 unsigned long start, unsigned long end,
302 int reserve, int flags)
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700303{
304 unsigned long sidx, eidx;
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700305
Johannes Weinere2bf3ca2008-07-23 21:28:06 -0700306 bdebug("nid=%td start=%lx end=%lx reserve=%d flags=%x\n",
307 bdata - bootmem_node_data, start, end, reserve, flags);
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700308
Johannes Weiner3560e242008-07-23 21:28:09 -0700309 BUG_ON(start < bdata->node_min_pfn);
Johannes Weinere2bf3ca2008-07-23 21:28:06 -0700310 BUG_ON(end > bdata->node_low_pfn);
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700311
Johannes Weiner3560e242008-07-23 21:28:09 -0700312 sidx = start - bdata->node_min_pfn;
313 eidx = end - bdata->node_min_pfn;
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700314
Johannes Weinere2bf3ca2008-07-23 21:28:06 -0700315 if (reserve)
316 return __reserve(bdata, sidx, eidx, flags);
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700317 else
Johannes Weinere2bf3ca2008-07-23 21:28:06 -0700318 __free(bdata, sidx, eidx);
319 return 0;
320}
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700321
Johannes Weinere2bf3ca2008-07-23 21:28:06 -0700322static int __init mark_bootmem(unsigned long start, unsigned long end,
323 int reserve, int flags)
324{
325 unsigned long pos;
326 bootmem_data_t *bdata;
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700327
Johannes Weinere2bf3ca2008-07-23 21:28:06 -0700328 pos = start;
329 list_for_each_entry(bdata, &bdata_list, list) {
330 int err;
331 unsigned long max;
332
Johannes Weiner3560e242008-07-23 21:28:09 -0700333 if (pos < bdata->node_min_pfn ||
334 pos >= bdata->node_low_pfn) {
Johannes Weinere2bf3ca2008-07-23 21:28:06 -0700335 BUG_ON(pos != start);
336 continue;
337 }
338
339 max = min(bdata->node_low_pfn, end);
340
341 err = mark_bootmem_node(bdata, pos, max, reserve, flags);
342 if (reserve && err) {
343 mark_bootmem(start, pos, 0, 0);
344 return err;
345 }
346
347 if (max == end)
348 return 0;
349 pos = bdata->node_low_pfn;
350 }
351 BUG();
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700352}
353
Johannes Weinera66fd7d2008-07-23 21:28:01 -0700354/**
355 * free_bootmem_node - mark a page range as usable
356 * @pgdat: node the range resides on
357 * @physaddr: starting address of the range
358 * @size: size of the range in bytes
359 *
360 * Partial pages will be considered reserved and left as they are.
361 *
Johannes Weinere2bf3ca2008-07-23 21:28:06 -0700362 * The range must reside completely on the specified node.
Johannes Weinera66fd7d2008-07-23 21:28:01 -0700363 */
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700364void __init free_bootmem_node(pg_data_t *pgdat, unsigned long physaddr,
365 unsigned long size)
366{
Johannes Weinere2bf3ca2008-07-23 21:28:06 -0700367 unsigned long start, end;
368
Catalin Marinasec3a3542009-07-07 10:33:01 +0100369 kmemleak_free_part(__va(physaddr), size);
370
Johannes Weinere2bf3ca2008-07-23 21:28:06 -0700371 start = PFN_UP(physaddr);
372 end = PFN_DOWN(physaddr + size);
373
374 mark_bootmem_node(pgdat->bdata, start, end, 0, 0);
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700375}
376
Johannes Weinera66fd7d2008-07-23 21:28:01 -0700377/**
378 * free_bootmem - mark a page range as usable
379 * @addr: starting address of the range
380 * @size: size of the range in bytes
381 *
382 * Partial pages will be considered reserved and left as they are.
383 *
Johannes Weinere2bf3ca2008-07-23 21:28:06 -0700384 * The range must be contiguous but may span node boundaries.
Johannes Weinera66fd7d2008-07-23 21:28:01 -0700385 */
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700386void __init free_bootmem(unsigned long addr, unsigned long size)
387{
Johannes Weinere2bf3ca2008-07-23 21:28:06 -0700388 unsigned long start, end;
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700389
Catalin Marinasec3a3542009-07-07 10:33:01 +0100390 kmemleak_free_part(__va(addr), size);
391
Johannes Weinere2bf3ca2008-07-23 21:28:06 -0700392 start = PFN_UP(addr);
393 end = PFN_DOWN(addr + size);
Yinghai Lua5645a62008-03-18 12:49:12 -0700394
Johannes Weinere2bf3ca2008-07-23 21:28:06 -0700395 mark_bootmem(start, end, 0, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700396}
397
Johannes Weinera66fd7d2008-07-23 21:28:01 -0700398/**
399 * reserve_bootmem_node - mark a page range as reserved
400 * @pgdat: node the range resides on
401 * @physaddr: starting address of the range
402 * @size: size of the range in bytes
403 * @flags: reservation flags (see linux/bootmem.h)
404 *
405 * Partial pages will be reserved.
406 *
Johannes Weinere2bf3ca2008-07-23 21:28:06 -0700407 * The range must reside completely on the specified node.
Johannes Weinera66fd7d2008-07-23 21:28:01 -0700408 */
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700409int __init reserve_bootmem_node(pg_data_t *pgdat, unsigned long physaddr,
410 unsigned long size, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411{
Johannes Weinere2bf3ca2008-07-23 21:28:06 -0700412 unsigned long start, end;
Franck Bui-Huubbc7b922006-09-25 23:31:07 -0700413
Johannes Weinere2bf3ca2008-07-23 21:28:06 -0700414 start = PFN_DOWN(physaddr);
415 end = PFN_UP(physaddr + size);
416
417 return mark_bootmem_node(pgdat->bdata, start, end, 1, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700418}
419
Johannes Weinera66fd7d2008-07-23 21:28:01 -0700420/**
421 * reserve_bootmem - mark a page range as usable
422 * @addr: starting address of the range
423 * @size: size of the range in bytes
424 * @flags: reservation flags (see linux/bootmem.h)
425 *
426 * Partial pages will be reserved.
427 *
Johannes Weinere2bf3ca2008-07-23 21:28:06 -0700428 * The range must be contiguous but may span node boundaries.
Johannes Weinera66fd7d2008-07-23 21:28:01 -0700429 */
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700430int __init reserve_bootmem(unsigned long addr, unsigned long size,
431 int flags)
432{
Johannes Weinere2bf3ca2008-07-23 21:28:06 -0700433 unsigned long start, end;
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700434
Johannes Weinere2bf3ca2008-07-23 21:28:06 -0700435 start = PFN_DOWN(addr);
436 end = PFN_UP(addr + size);
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700437
Johannes Weinere2bf3ca2008-07-23 21:28:06 -0700438 return mark_bootmem(start, end, 1, flags);
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700439}
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700440
Yinghai Luf88eff72010-08-25 13:39:15 -0700441int __weak __init reserve_bootmem_generic(unsigned long phys, unsigned long len,
442 int flags)
443{
444 return reserve_bootmem(phys, len, flags);
445}
446
Jan Beulich8aa043d2009-12-14 17:59:46 -0800447static unsigned long __init align_idx(struct bootmem_data *bdata,
448 unsigned long idx, unsigned long step)
Johannes Weiner481ebd02008-08-20 14:09:15 -0700449{
450 unsigned long base = bdata->node_min_pfn;
451
452 /*
453 * Align the index with respect to the node start so that the
454 * combination of both satisfies the requested alignment.
455 */
456
457 return ALIGN(base + idx, step) - base;
458}
459
Jan Beulich8aa043d2009-12-14 17:59:46 -0800460static unsigned long __init align_off(struct bootmem_data *bdata,
461 unsigned long off, unsigned long align)
Johannes Weiner481ebd02008-08-20 14:09:15 -0700462{
463 unsigned long base = PFN_PHYS(bdata->node_min_pfn);
464
465 /* Same as align_idx for byte offsets */
466
467 return ALIGN(base + off, align) - base;
468}
469
Tejun Heod0c4f572009-03-01 16:06:56 +0900470static void * __init alloc_bootmem_core(struct bootmem_data *bdata,
471 unsigned long size, unsigned long align,
472 unsigned long goal, unsigned long limit)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473{
Johannes Weiner0f3caba2008-07-23 21:28:07 -0700474 unsigned long fallback = 0;
Johannes Weiner5f2809e2008-07-23 21:28:05 -0700475 unsigned long min, max, start, sidx, midx, step;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700476
Johannes Weiner594fe1a2009-01-06 14:40:32 -0800477 bdebug("nid=%td size=%lx [%lu pages] align=%lx goal=%lx limit=%lx\n",
478 bdata - bootmem_node_data, size, PAGE_ALIGN(size) >> PAGE_SHIFT,
479 align, goal, limit);
480
Johannes Weiner5f2809e2008-07-23 21:28:05 -0700481 BUG_ON(!size);
482 BUG_ON(align & (align - 1));
483 BUG_ON(limit && goal + size > limit);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700484
Christian Krafft7c309a62006-12-06 20:32:41 -0800485 if (!bdata->node_bootmem_map)
486 return NULL;
487
Johannes Weiner3560e242008-07-23 21:28:09 -0700488 min = bdata->node_min_pfn;
Johannes Weiner5f2809e2008-07-23 21:28:05 -0700489 max = bdata->node_low_pfn;
Yinghai Lu9a2dc042008-03-18 12:44:48 -0700490
Johannes Weiner5f2809e2008-07-23 21:28:05 -0700491 goal >>= PAGE_SHIFT;
492 limit >>= PAGE_SHIFT;
493
494 if (limit && max > limit)
495 max = limit;
496 if (max <= min)
Yinghai Lu9a2dc042008-03-18 12:44:48 -0700497 return NULL;
498
Johannes Weiner5f2809e2008-07-23 21:28:05 -0700499 step = max(align >> PAGE_SHIFT, 1UL);
Yasunori Goto281dd252005-10-19 15:52:18 -0700500
Johannes Weiner5f2809e2008-07-23 21:28:05 -0700501 if (goal && min < goal && goal < max)
502 start = ALIGN(goal, step);
503 else
504 start = ALIGN(min, step);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700505
Johannes Weiner481ebd02008-08-20 14:09:15 -0700506 sidx = start - bdata->node_min_pfn;
Johannes Weiner3560e242008-07-23 21:28:09 -0700507 midx = max - bdata->node_min_pfn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700508
Johannes Weiner5f2809e2008-07-23 21:28:05 -0700509 if (bdata->hint_idx > sidx) {
Johannes Weiner0f3caba2008-07-23 21:28:07 -0700510 /*
511 * Handle the valid case of sidx being zero and still
512 * catch the fallback below.
513 */
514 fallback = sidx + 1;
Johannes Weiner481ebd02008-08-20 14:09:15 -0700515 sidx = align_idx(bdata, bdata->hint_idx, step);
Yinghai Luad093152008-03-10 23:23:42 -0700516 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700517
Johannes Weiner5f2809e2008-07-23 21:28:05 -0700518 while (1) {
519 int merge;
520 void *region;
521 unsigned long eidx, i, start_off, end_off;
522find_block:
523 sidx = find_next_zero_bit(bdata->node_bootmem_map, midx, sidx);
Johannes Weiner481ebd02008-08-20 14:09:15 -0700524 sidx = align_idx(bdata, sidx, step);
Johannes Weiner5f2809e2008-07-23 21:28:05 -0700525 eidx = sidx + PFN_UP(size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700526
Johannes Weiner5f2809e2008-07-23 21:28:05 -0700527 if (sidx >= midx || eidx > midx)
Haren Myneni66d43e92005-12-12 00:37:39 -0800528 break;
Johannes Weiner5f2809e2008-07-23 21:28:05 -0700529
530 for (i = sidx; i < eidx; i++)
531 if (test_bit(i, bdata->node_bootmem_map)) {
Johannes Weiner481ebd02008-08-20 14:09:15 -0700532 sidx = align_idx(bdata, i, step);
Johannes Weiner5f2809e2008-07-23 21:28:05 -0700533 if (sidx == i)
534 sidx += step;
535 goto find_block;
536 }
537
Mikulas Patocka627240a2008-08-15 00:40:17 -0700538 if (bdata->last_end_off & (PAGE_SIZE - 1) &&
Johannes Weiner5f2809e2008-07-23 21:28:05 -0700539 PFN_DOWN(bdata->last_end_off) + 1 == sidx)
Johannes Weiner481ebd02008-08-20 14:09:15 -0700540 start_off = align_off(bdata, bdata->last_end_off, align);
Johannes Weiner5f2809e2008-07-23 21:28:05 -0700541 else
542 start_off = PFN_PHYS(sidx);
543
544 merge = PFN_DOWN(start_off) < sidx;
545 end_off = start_off + size;
546
547 bdata->last_end_off = end_off;
548 bdata->hint_idx = PFN_UP(end_off);
549
550 /*
551 * Reserve the area now:
552 */
Johannes Weinerd747fa42008-07-23 21:28:05 -0700553 if (__reserve(bdata, PFN_DOWN(start_off) + merge,
554 PFN_UP(end_off), BOOTMEM_EXCLUSIVE))
555 BUG();
Johannes Weiner5f2809e2008-07-23 21:28:05 -0700556
Johannes Weiner3560e242008-07-23 21:28:09 -0700557 region = phys_to_virt(PFN_PHYS(bdata->node_min_pfn) +
558 start_off);
Johannes Weiner5f2809e2008-07-23 21:28:05 -0700559 memset(region, 0, size);
Catalin Marinas008139d2009-08-27 14:29:17 +0100560 /*
561 * The min_count is set to 0 so that bootmem allocated blocks
562 * are never reported as leaks.
563 */
564 kmemleak_alloc(region, size, 0, 0);
Johannes Weiner5f2809e2008-07-23 21:28:05 -0700565 return region;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566 }
567
Johannes Weiner0f3caba2008-07-23 21:28:07 -0700568 if (fallback) {
Johannes Weiner481ebd02008-08-20 14:09:15 -0700569 sidx = align_idx(bdata, fallback - 1, step);
Johannes Weiner0f3caba2008-07-23 21:28:07 -0700570 fallback = 0;
571 goto find_block;
572 }
573
574 return NULL;
575}
576
Tejun Heod0c4f572009-03-01 16:06:56 +0900577static void * __init alloc_arch_preferred_bootmem(bootmem_data_t *bdata,
578 unsigned long size, unsigned long align,
579 unsigned long goal, unsigned long limit)
580{
Pekka Enberg441c7e02009-06-10 20:05:53 +0300581 if (WARN_ON_ONCE(slab_is_available()))
582 return kzalloc(size, GFP_NOWAIT);
583
Tejun Heod0c4f572009-03-01 16:06:56 +0900584#ifdef CONFIG_HAVE_ARCH_BOOTMEM
Joe Perches433f13a2009-06-18 16:48:58 -0700585 {
586 bootmem_data_t *p_bdata;
Tejun Heod0c4f572009-03-01 16:06:56 +0900587
Joe Perches433f13a2009-06-18 16:48:58 -0700588 p_bdata = bootmem_arch_preferred_node(bdata, size, align,
589 goal, limit);
590 if (p_bdata)
591 return alloc_bootmem_core(p_bdata, size, align,
592 goal, limit);
593 }
Tejun Heod0c4f572009-03-01 16:06:56 +0900594#endif
595 return NULL;
596}
597
Johannes Weiner0f3caba2008-07-23 21:28:07 -0700598static void * __init ___alloc_bootmem_nopanic(unsigned long size,
599 unsigned long align,
600 unsigned long goal,
601 unsigned long limit)
602{
603 bootmem_data_t *bdata;
Tejun Heod0c4f572009-03-01 16:06:56 +0900604 void *region;
Johannes Weiner0f3caba2008-07-23 21:28:07 -0700605
606restart:
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
617 region = alloc_bootmem_core(bdata, size, align, goal, limit);
618 if (region)
619 return region;
620 }
621
Johannes Weiner5f2809e2008-07-23 21:28:05 -0700622 if (goal) {
623 goal = 0;
Johannes Weiner0f3caba2008-07-23 21:28:07 -0700624 goto restart;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700625 }
Johannes Weiner5f2809e2008-07-23 21:28:05 -0700626
Linus Torvalds1da177e2005-04-16 15:20:36 -0700627 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628}
629
Johannes Weinera66fd7d2008-07-23 21:28:01 -0700630/**
631 * __alloc_bootmem_nopanic - allocate boot memory without panicking
632 * @size: size of the request in bytes
633 * @align: alignment of the region
634 * @goal: preferred starting address of the region
635 *
636 * The goal is dropped if it can not be satisfied and the allocation will
637 * fall back to memory below @goal.
638 *
639 * Allocation may happen on any node in the system.
640 *
641 * Returns NULL on failure.
642 */
Franck Bui-Huubb0923a2006-09-25 23:31:05 -0700643void * __init __alloc_bootmem_nopanic(unsigned long size, unsigned long align,
Johannes Weiner0f3caba2008-07-23 21:28:07 -0700644 unsigned long goal)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645{
Yinghai Lu08677212010-02-10 01:20:20 -0800646 unsigned long limit = 0;
647
Yinghai Lu08677212010-02-10 01:20:20 -0800648 return ___alloc_bootmem_nopanic(size, align, goal, limit);
Johannes Weiner0f3caba2008-07-23 21:28:07 -0700649}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700650
Johannes Weiner0f3caba2008-07-23 21:28:07 -0700651static void * __init ___alloc_bootmem(unsigned long size, unsigned long align,
652 unsigned long goal, unsigned long limit)
653{
654 void *mem = ___alloc_bootmem_nopanic(size, align, goal, limit);
655
656 if (mem)
657 return mem;
658 /*
659 * Whoops, we cannot satisfy the allocation request.
660 */
661 printk(KERN_ALERT "bootmem alloc of %lu bytes failed!\n", size);
662 panic("Out of memory");
Andi Kleena8062232006-04-07 19:49:21 +0200663 return NULL;
664}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665
Johannes Weinera66fd7d2008-07-23 21:28:01 -0700666/**
667 * __alloc_bootmem - allocate boot memory
668 * @size: size of the request in bytes
669 * @align: alignment of the region
670 * @goal: preferred starting address of the region
671 *
672 * The goal is dropped if it can not be satisfied and the allocation will
673 * fall back to memory below @goal.
674 *
675 * Allocation may happen on any node in the system.
676 *
677 * The function panics if the request can not be satisfied.
678 */
Franck Bui-Huubb0923a2006-09-25 23:31:05 -0700679void * __init __alloc_bootmem(unsigned long size, unsigned long align,
680 unsigned long goal)
Andi Kleena8062232006-04-07 19:49:21 +0200681{
Yinghai Lu08677212010-02-10 01:20:20 -0800682 unsigned long limit = 0;
683
Yinghai Lu08677212010-02-10 01:20:20 -0800684 return ___alloc_bootmem(size, align, goal, limit);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700685}
686
Johannes Weiner4cc278b2008-07-23 21:28:08 -0700687static void * __init ___alloc_bootmem_node(bootmem_data_t *bdata,
688 unsigned long size, unsigned long align,
689 unsigned long goal, unsigned long limit)
690{
691 void *ptr;
692
Tejun Heod0c4f572009-03-01 16:06:56 +0900693 ptr = alloc_arch_preferred_bootmem(bdata, size, align, goal, limit);
694 if (ptr)
695 return ptr;
696
Johannes Weiner4cc278b2008-07-23 21:28:08 -0700697 ptr = alloc_bootmem_core(bdata, size, align, goal, limit);
698 if (ptr)
699 return ptr;
700
701 return ___alloc_bootmem(size, align, goal, limit);
702}
703
Johannes Weinera66fd7d2008-07-23 21:28:01 -0700704/**
705 * __alloc_bootmem_node - allocate boot memory from a specific node
706 * @pgdat: node to allocate from
707 * @size: size of the request in bytes
708 * @align: alignment of the region
709 * @goal: preferred starting address of the region
710 *
711 * The goal is dropped if it can not be satisfied and the allocation will
712 * fall back to memory below @goal.
713 *
714 * Allocation may fall back to any node in the system if the specified node
715 * can not hold the requested memory.
716 *
717 * The function panics if the request can not be satisfied.
718 */
Franck Bui-Huubb0923a2006-09-25 23:31:05 -0700719void * __init __alloc_bootmem_node(pg_data_t *pgdat, unsigned long size,
720 unsigned long align, unsigned long goal)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700721{
Pekka Enbergc91c4772009-06-11 08:10:28 +0300722 if (WARN_ON_ONCE(slab_is_available()))
723 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
724
Yinghai Lu09325872011-02-24 14:43:05 +0100725 return ___alloc_bootmem_node(pgdat->bdata, size, align, goal, 0);
Yinghai Lu08677212010-02-10 01:20:20 -0800726}
727
728void * __init __alloc_bootmem_node_high(pg_data_t *pgdat, unsigned long size,
729 unsigned long align, unsigned long goal)
730{
731#ifdef MAX_DMA32_PFN
732 unsigned long end_pfn;
733
734 if (WARN_ON_ONCE(slab_is_available()))
735 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
736
737 /* update goal according ...MAX_DMA32_PFN */
738 end_pfn = pgdat->node_start_pfn + pgdat->node_spanned_pages;
739
740 if (end_pfn > MAX_DMA32_PFN + (128 >> (20 - PAGE_SHIFT)) &&
741 (goal >> PAGE_SHIFT) < MAX_DMA32_PFN) {
742 void *ptr;
743 unsigned long new_goal;
744
745 new_goal = MAX_DMA32_PFN << PAGE_SHIFT;
Yinghai Lu08677212010-02-10 01:20:20 -0800746 ptr = alloc_bootmem_core(pgdat->bdata, size, align,
747 new_goal, 0);
Yinghai Lu08677212010-02-10 01:20:20 -0800748 if (ptr)
749 return ptr;
750 }
751#endif
752
753 return __alloc_bootmem_node(pgdat, size, align, goal);
754
Linus Torvalds1da177e2005-04-16 15:20:36 -0700755}
756
Yasunori Gotoe70260a2008-04-28 02:13:32 -0700757#ifdef CONFIG_SPARSEMEM
Johannes Weinera66fd7d2008-07-23 21:28:01 -0700758/**
759 * alloc_bootmem_section - allocate boot memory from a specific section
760 * @size: size of the request in bytes
761 * @section_nr: sparse map section to allocate from
762 *
763 * Return NULL on failure.
764 */
Yasunori Gotoe70260a2008-04-28 02:13:32 -0700765void * __init alloc_bootmem_section(unsigned long size,
766 unsigned long section_nr)
767{
Johannes Weiner75a56cf2008-07-23 21:28:09 -0700768 bootmem_data_t *bdata;
Nishanth Aravamudanf5bf18f2012-03-21 16:34:07 -0700769 unsigned long pfn, goal;
Yasunori Gotoe70260a2008-04-28 02:13:32 -0700770
771 pfn = section_nr_to_pfn(section_nr);
Johannes Weiner75a56cf2008-07-23 21:28:09 -0700772 goal = pfn << PAGE_SHIFT;
Johannes Weiner75a56cf2008-07-23 21:28:09 -0700773 bdata = &bootmem_node_data[early_pfn_to_nid(pfn)];
Yasunori Gotoe70260a2008-04-28 02:13:32 -0700774
Nishanth Aravamudanf5bf18f2012-03-21 16:34:07 -0700775 return alloc_bootmem_core(bdata, size, SMP_CACHE_BYTES, goal, 0);
Yasunori Gotoe70260a2008-04-28 02:13:32 -0700776}
777#endif
778
Andi Kleenb54bbf72008-07-23 21:27:45 -0700779void * __init __alloc_bootmem_node_nopanic(pg_data_t *pgdat, unsigned long size,
780 unsigned long align, unsigned long goal)
781{
782 void *ptr;
783
Pekka Enbergc91c4772009-06-11 08:10:28 +0300784 if (WARN_ON_ONCE(slab_is_available()))
785 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
786
Tejun Heod0c4f572009-03-01 16:06:56 +0900787 ptr = alloc_arch_preferred_bootmem(pgdat->bdata, size, align, goal, 0);
788 if (ptr)
789 return ptr;
790
Andi Kleenb54bbf72008-07-23 21:27:45 -0700791 ptr = alloc_bootmem_core(pgdat->bdata, size, align, goal, 0);
792 if (ptr)
793 return ptr;
794
795 return __alloc_bootmem_nopanic(size, align, goal);
796}
797
Heiko Carstensdfd54cb2006-09-25 23:31:33 -0700798#ifndef ARCH_LOW_ADDRESS_LIMIT
799#define ARCH_LOW_ADDRESS_LIMIT 0xffffffffUL
800#endif
Ravikiran G Thirumalai008857c2006-01-06 00:11:01 -0800801
Johannes Weinera66fd7d2008-07-23 21:28:01 -0700802/**
803 * __alloc_bootmem_low - allocate low boot memory
804 * @size: size of the request in bytes
805 * @align: alignment of the region
806 * @goal: preferred starting address of the region
807 *
808 * The goal is dropped if it can not be satisfied and the allocation will
809 * fall back to memory below @goal.
810 *
811 * Allocation may happen on any node in the system.
812 *
813 * The function panics if the request can not be satisfied.
814 */
Franck Bui-Huubb0923a2006-09-25 23:31:05 -0700815void * __init __alloc_bootmem_low(unsigned long size, unsigned long align,
816 unsigned long goal)
Ravikiran G Thirumalai008857c2006-01-06 00:11:01 -0800817{
Johannes Weiner0f3caba2008-07-23 21:28:07 -0700818 return ___alloc_bootmem(size, align, goal, ARCH_LOW_ADDRESS_LIMIT);
Ravikiran G Thirumalai008857c2006-01-06 00:11:01 -0800819}
820
Johannes Weinera66fd7d2008-07-23 21:28:01 -0700821/**
822 * __alloc_bootmem_low_node - allocate low boot memory from a specific node
823 * @pgdat: node to allocate from
824 * @size: size of the request in bytes
825 * @align: alignment of the region
826 * @goal: preferred starting address of the region
827 *
828 * The goal is dropped if it can not be satisfied and the allocation will
829 * fall back to memory below @goal.
830 *
831 * Allocation may fall back to any node in the system if the specified node
832 * can not hold the requested memory.
833 *
834 * The function panics if the request can not be satisfied.
835 */
Ravikiran G Thirumalai008857c2006-01-06 00:11:01 -0800836void * __init __alloc_bootmem_low_node(pg_data_t *pgdat, unsigned long size,
837 unsigned long align, unsigned long goal)
838{
Pekka Enbergc91c4772009-06-11 08:10:28 +0300839 if (WARN_ON_ONCE(slab_is_available()))
840 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
841
Yinghai Lu09325872011-02-24 14:43:05 +0100842 return ___alloc_bootmem_node(pgdat->bdata, size, align,
Yinghai Lub8ab9f82010-07-20 13:24:31 -0700843 goal, ARCH_LOW_ADDRESS_LIMIT);
Ravikiran G Thirumalai008857c2006-01-06 00:11:01 -0800844}