blob: d1dc23cc7f10e78e9598c87c70337bd3d2577d6a [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>
Linus Torvalds1da177e2005-04-16 15:20:36 -070013#include <linux/bootmem.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070014#include <linux/module.h>
Catalin Marinasec3a3542009-07-07 10:33:01 +010015#include <linux/kmemleak.h>
Franck Bui-Huue786e862006-09-25 23:31:06 -070016
17#include <asm/bug.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070018#include <asm/io.h>
Heiko Carstensdfd54cb2006-09-25 23:31:33 -070019#include <asm/processor.h>
Franck Bui-Huue786e862006-09-25 23:31:06 -070020
Linus Torvalds1da177e2005-04-16 15:20:36 -070021#include "internal.h"
22
Linus Torvalds1da177e2005-04-16 15:20:36 -070023unsigned long max_low_pfn;
24unsigned long min_low_pfn;
25unsigned long max_pfn;
26
Vivek Goyal92aa63a2005-06-25 14:58:18 -070027#ifdef CONFIG_CRASH_DUMP
28/*
29 * If we have booted due to a crash, max_pfn will be a very low value. We need
30 * to know the amount of memory that the previous kernel used.
31 */
32unsigned long saved_max_pfn;
33#endif
34
Johannes Weinerb61bfa32008-07-23 21:26:55 -070035bootmem_data_t bootmem_node_data[MAX_NUMNODES] __initdata;
36
Johannes Weiner636cc402008-07-23 21:28:03 -070037static struct list_head bdata_list __initdata = LIST_HEAD_INIT(bdata_list);
38
Johannes Weiner2e5237d2008-07-23 21:28:02 -070039static int bootmem_debug;
40
41static int __init bootmem_debug_setup(char *buf)
42{
43 bootmem_debug = 1;
44 return 0;
45}
46early_param("bootmem_debug", bootmem_debug_setup);
47
48#define bdebug(fmt, args...) ({ \
49 if (unlikely(bootmem_debug)) \
50 printk(KERN_INFO \
51 "bootmem::%s " fmt, \
Harvey Harrison80a914d2008-10-15 22:01:25 -070052 __func__, ## args); \
Johannes Weiner2e5237d2008-07-23 21:28:02 -070053})
54
Johannes Weinerdf049a52008-07-23 21:28:02 -070055static unsigned long __init bootmap_bytes(unsigned long pages)
Johannes Weiner223e8dc2008-07-23 21:28:00 -070056{
Johannes Weinerdf049a52008-07-23 21:28:02 -070057 unsigned long bytes = (pages + 7) / 8;
Johannes Weiner223e8dc2008-07-23 21:28:00 -070058
Johannes Weinerdf049a52008-07-23 21:28:02 -070059 return ALIGN(bytes, sizeof(long));
Johannes Weiner223e8dc2008-07-23 21:28:00 -070060}
61
Johannes Weinera66fd7d2008-07-23 21:28:01 -070062/**
63 * bootmem_bootmap_pages - calculate bitmap size in pages
64 * @pages: number of pages the bitmap has to represent
65 */
Franck Bui-Huuf71bf0c2006-09-25 23:31:08 -070066unsigned long __init bootmem_bootmap_pages(unsigned long pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -070067{
Johannes Weinerdf049a52008-07-23 21:28:02 -070068 unsigned long bytes = bootmap_bytes(pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -070069
Johannes Weinerdf049a52008-07-23 21:28:02 -070070 return PAGE_ALIGN(bytes) >> PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -070071}
Franck Bui-Huuf71bf0c2006-09-25 23:31:08 -070072
KAMEZAWA Hiroyuki679bc9f2006-03-27 01:15:58 -080073/*
74 * link bdata in order
75 */
Franck Bui-Huu69d49e62006-09-25 23:31:04 -070076static void __init link_bootmem(bootmem_data_t *bdata)
KAMEZAWA Hiroyuki679bc9f2006-03-27 01:15:58 -080077{
Johannes Weiner636cc402008-07-23 21:28:03 -070078 struct list_head *iter;
Franck Bui-Huuf71bf0c2006-09-25 23:31:08 -070079
Johannes Weiner636cc402008-07-23 21:28:03 -070080 list_for_each(iter, &bdata_list) {
81 bootmem_data_t *ent;
82
83 ent = list_entry(iter, bootmem_data_t, list);
Johannes Weiner3560e242008-07-23 21:28:09 -070084 if (bdata->node_min_pfn < ent->node_min_pfn)
Johannes Weiner636cc402008-07-23 21:28:03 -070085 break;
KAMEZAWA Hiroyuki679bc9f2006-03-27 01:15:58 -080086 }
Johannes Weiner636cc402008-07-23 21:28:03 -070087 list_add_tail(&bdata->list, iter);
KAMEZAWA Hiroyuki679bc9f2006-03-27 01:15:58 -080088}
89
Franck Bui-Huubbc7b922006-09-25 23:31:07 -070090/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070091 * Called once to set up the allocator itself.
92 */
Johannes Weiner8ae04462008-07-23 21:26:59 -070093static unsigned long __init init_bootmem_core(bootmem_data_t *bdata,
Linus Torvalds1da177e2005-04-16 15:20:36 -070094 unsigned long mapstart, unsigned long start, unsigned long end)
95{
Franck Bui-Huubbc7b922006-09-25 23:31:07 -070096 unsigned long mapsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -070097
Mel Gorman2dbb51c2008-07-23 21:26:52 -070098 mminit_validate_memmodel_limits(&start, &end);
Franck Bui-Huubbc7b922006-09-25 23:31:07 -070099 bdata->node_bootmem_map = phys_to_virt(PFN_PHYS(mapstart));
Johannes Weiner3560e242008-07-23 21:28:09 -0700100 bdata->node_min_pfn = start;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700101 bdata->node_low_pfn = end;
KAMEZAWA Hiroyuki679bc9f2006-03-27 01:15:58 -0800102 link_bootmem(bdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700103
104 /*
105 * Initially all pages are reserved - setup_arch() has to
106 * register free RAM areas explicitly.
107 */
Johannes Weinerdf049a52008-07-23 21:28:02 -0700108 mapsize = bootmap_bytes(end - start);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700109 memset(bdata->node_bootmem_map, 0xff, mapsize);
110
Johannes Weiner2e5237d2008-07-23 21:28:02 -0700111 bdebug("nid=%td start=%lx map=%lx end=%lx mapsize=%lx\n",
112 bdata - bootmem_node_data, start, mapstart, end, mapsize);
113
Linus Torvalds1da177e2005-04-16 15:20:36 -0700114 return mapsize;
115}
116
Johannes Weinera66fd7d2008-07-23 21:28:01 -0700117/**
118 * init_bootmem_node - register a node as boot memory
119 * @pgdat: node to register
120 * @freepfn: pfn where the bitmap for this node is to be placed
121 * @startpfn: first pfn on the node
122 * @endpfn: first pfn after the node
123 *
124 * Returns the number of bytes needed to hold the bitmap for this node.
125 */
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700126unsigned long __init init_bootmem_node(pg_data_t *pgdat, unsigned long freepfn,
127 unsigned long startpfn, unsigned long endpfn)
128{
129 return init_bootmem_core(pgdat->bdata, freepfn, startpfn, endpfn);
130}
131
Johannes Weinera66fd7d2008-07-23 21:28:01 -0700132/**
133 * init_bootmem - register boot memory
134 * @start: pfn where the bitmap is to be placed
135 * @pages: number of available physical pages
136 *
137 * Returns the number of bytes needed to hold the bitmap.
138 */
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700139unsigned long __init init_bootmem(unsigned long start, unsigned long pages)
140{
141 max_low_pfn = pages;
142 min_low_pfn = start;
143 return init_bootmem_core(NODE_DATA(0)->bdata, start, 0, pages);
144}
145
FUJITA Tomonori9f993ac2009-11-10 19:46:17 +0900146/*
147 * free_bootmem_late - free bootmem pages directly to page allocator
148 * @addr: starting address of the range
149 * @size: size of the range in bytes
150 *
151 * This is only useful when the bootmem allocator has already been torn
152 * down, but we are still initializing the system. Pages are given directly
153 * to the page allocator, no bootmem metadata is updated because it is gone.
154 */
155void __init free_bootmem_late(unsigned long addr, unsigned long size)
156{
157 unsigned long cursor, end;
158
159 kmemleak_free_part(__va(addr), size);
160
161 cursor = PFN_UP(addr);
162 end = PFN_DOWN(addr + size);
163
164 for (; cursor < end; cursor++) {
165 __free_pages_bootmem(pfn_to_page(cursor), 0);
166 totalram_pages++;
167 }
168}
169
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700170static unsigned long __init free_all_bootmem_core(bootmem_data_t *bdata)
171{
Johannes Weiner41546c12008-07-23 21:28:03 -0700172 int aligned;
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700173 struct page *page;
Johannes Weiner41546c12008-07-23 21:28:03 -0700174 unsigned long start, end, pages, count = 0;
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700175
Johannes Weiner41546c12008-07-23 21:28:03 -0700176 if (!bdata->node_bootmem_map)
177 return 0;
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700178
Johannes Weiner3560e242008-07-23 21:28:09 -0700179 start = bdata->node_min_pfn;
Johannes Weiner41546c12008-07-23 21:28:03 -0700180 end = bdata->node_low_pfn;
181
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700182 /*
Johannes Weiner41546c12008-07-23 21:28:03 -0700183 * If the start is aligned to the machines wordsize, we might
184 * be able to free pages in bulks of that order.
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700185 */
Johannes Weiner41546c12008-07-23 21:28:03 -0700186 aligned = !(start & (BITS_PER_LONG - 1));
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700187
Johannes Weiner41546c12008-07-23 21:28:03 -0700188 bdebug("nid=%td start=%lx end=%lx aligned=%d\n",
189 bdata - bootmem_node_data, start, end, aligned);
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700190
Johannes Weiner41546c12008-07-23 21:28:03 -0700191 while (start < end) {
192 unsigned long *map, idx, vec;
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700193
Johannes Weiner41546c12008-07-23 21:28:03 -0700194 map = bdata->node_bootmem_map;
Johannes Weiner3560e242008-07-23 21:28:09 -0700195 idx = start - bdata->node_min_pfn;
Johannes Weiner41546c12008-07-23 21:28:03 -0700196 vec = ~map[idx / BITS_PER_LONG];
197
198 if (aligned && vec == ~0UL && start + BITS_PER_LONG < end) {
199 int order = ilog2(BITS_PER_LONG);
200
201 __free_pages_bootmem(pfn_to_page(start), order);
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700202 count += 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 Weiner223e8dc2008-07-23 21:28:00 -0700215 }
Johannes Weiner41546c12008-07-23 21:28:03 -0700216 start += BITS_PER_LONG;
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{
250 return free_all_bootmem_core(NODE_DATA(0)->bdata);
251}
252
Johannes Weinerd747fa42008-07-23 21:28:05 -0700253static void __init __free(bootmem_data_t *bdata,
254 unsigned long sidx, unsigned long eidx)
255{
256 unsigned long idx;
257
258 bdebug("nid=%td start=%lx end=%lx\n", bdata - bootmem_node_data,
Johannes Weiner3560e242008-07-23 21:28:09 -0700259 sidx + bdata->node_min_pfn,
260 eidx + bdata->node_min_pfn);
Johannes Weinerd747fa42008-07-23 21:28:05 -0700261
Johannes Weinere2bf3ca2008-07-23 21:28:06 -0700262 if (bdata->hint_idx > sidx)
263 bdata->hint_idx = sidx;
264
Johannes Weinerd747fa42008-07-23 21:28:05 -0700265 for (idx = sidx; idx < eidx; idx++)
266 if (!test_and_clear_bit(idx, bdata->node_bootmem_map))
267 BUG();
268}
269
270static int __init __reserve(bootmem_data_t *bdata, unsigned long sidx,
271 unsigned long eidx, int flags)
272{
273 unsigned long idx;
274 int exclusive = flags & BOOTMEM_EXCLUSIVE;
275
276 bdebug("nid=%td start=%lx end=%lx flags=%x\n",
277 bdata - bootmem_node_data,
Johannes Weiner3560e242008-07-23 21:28:09 -0700278 sidx + bdata->node_min_pfn,
279 eidx + bdata->node_min_pfn,
Johannes Weinerd747fa42008-07-23 21:28:05 -0700280 flags);
281
282 for (idx = sidx; idx < eidx; idx++)
283 if (test_and_set_bit(idx, bdata->node_bootmem_map)) {
284 if (exclusive) {
285 __free(bdata, sidx, idx);
286 return -EBUSY;
287 }
288 bdebug("silent double reserve of PFN %lx\n",
Johannes Weiner3560e242008-07-23 21:28:09 -0700289 idx + bdata->node_min_pfn);
Johannes Weinerd747fa42008-07-23 21:28:05 -0700290 }
291 return 0;
292}
293
Johannes Weinere2bf3ca2008-07-23 21:28:06 -0700294static int __init mark_bootmem_node(bootmem_data_t *bdata,
295 unsigned long start, unsigned long end,
296 int reserve, int flags)
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700297{
298 unsigned long sidx, eidx;
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700299
Johannes Weinere2bf3ca2008-07-23 21:28:06 -0700300 bdebug("nid=%td start=%lx end=%lx reserve=%d flags=%x\n",
301 bdata - bootmem_node_data, start, end, reserve, flags);
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700302
Johannes Weiner3560e242008-07-23 21:28:09 -0700303 BUG_ON(start < bdata->node_min_pfn);
Johannes Weinere2bf3ca2008-07-23 21:28:06 -0700304 BUG_ON(end > bdata->node_low_pfn);
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700305
Johannes Weiner3560e242008-07-23 21:28:09 -0700306 sidx = start - bdata->node_min_pfn;
307 eidx = end - bdata->node_min_pfn;
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700308
Johannes Weinere2bf3ca2008-07-23 21:28:06 -0700309 if (reserve)
310 return __reserve(bdata, sidx, eidx, flags);
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700311 else
Johannes Weinere2bf3ca2008-07-23 21:28:06 -0700312 __free(bdata, sidx, eidx);
313 return 0;
314}
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700315
Johannes Weinere2bf3ca2008-07-23 21:28:06 -0700316static int __init mark_bootmem(unsigned long start, unsigned long end,
317 int reserve, int flags)
318{
319 unsigned long pos;
320 bootmem_data_t *bdata;
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700321
Johannes Weinere2bf3ca2008-07-23 21:28:06 -0700322 pos = start;
323 list_for_each_entry(bdata, &bdata_list, list) {
324 int err;
325 unsigned long max;
326
Johannes Weiner3560e242008-07-23 21:28:09 -0700327 if (pos < bdata->node_min_pfn ||
328 pos >= bdata->node_low_pfn) {
Johannes Weinere2bf3ca2008-07-23 21:28:06 -0700329 BUG_ON(pos != start);
330 continue;
331 }
332
333 max = min(bdata->node_low_pfn, end);
334
335 err = mark_bootmem_node(bdata, pos, max, reserve, flags);
336 if (reserve && err) {
337 mark_bootmem(start, pos, 0, 0);
338 return err;
339 }
340
341 if (max == end)
342 return 0;
343 pos = bdata->node_low_pfn;
344 }
345 BUG();
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700346}
347
Johannes Weinera66fd7d2008-07-23 21:28:01 -0700348/**
349 * free_bootmem_node - mark a page range as usable
350 * @pgdat: node the range resides on
351 * @physaddr: starting address of the range
352 * @size: size of the range in bytes
353 *
354 * Partial pages will be considered reserved and left as they are.
355 *
Johannes Weinere2bf3ca2008-07-23 21:28:06 -0700356 * The range must reside completely on the specified node.
Johannes Weinera66fd7d2008-07-23 21:28:01 -0700357 */
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700358void __init free_bootmem_node(pg_data_t *pgdat, unsigned long physaddr,
359 unsigned long size)
360{
Johannes Weinere2bf3ca2008-07-23 21:28:06 -0700361 unsigned long start, end;
362
Catalin Marinasec3a3542009-07-07 10:33:01 +0100363 kmemleak_free_part(__va(physaddr), size);
364
Johannes Weinere2bf3ca2008-07-23 21:28:06 -0700365 start = PFN_UP(physaddr);
366 end = PFN_DOWN(physaddr + size);
367
368 mark_bootmem_node(pgdat->bdata, start, end, 0, 0);
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700369}
370
Johannes Weinera66fd7d2008-07-23 21:28:01 -0700371/**
372 * free_bootmem - mark a page range as usable
373 * @addr: starting address of the range
374 * @size: size of the range in bytes
375 *
376 * Partial pages will be considered reserved and left as they are.
377 *
Johannes Weinere2bf3ca2008-07-23 21:28:06 -0700378 * The range must be contiguous but may span node boundaries.
Johannes Weinera66fd7d2008-07-23 21:28:01 -0700379 */
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700380void __init free_bootmem(unsigned long addr, unsigned long size)
381{
Johannes Weinere2bf3ca2008-07-23 21:28:06 -0700382 unsigned long start, end;
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700383
Catalin Marinasec3a3542009-07-07 10:33:01 +0100384 kmemleak_free_part(__va(addr), size);
385
Johannes Weinere2bf3ca2008-07-23 21:28:06 -0700386 start = PFN_UP(addr);
387 end = PFN_DOWN(addr + size);
Yinghai Lua5645a62008-03-18 12:49:12 -0700388
Johannes Weinere2bf3ca2008-07-23 21:28:06 -0700389 mark_bootmem(start, end, 0, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700390}
391
Johannes Weinera66fd7d2008-07-23 21:28:01 -0700392/**
393 * reserve_bootmem_node - mark a page range as reserved
394 * @pgdat: node the range resides on
395 * @physaddr: starting address of the range
396 * @size: size of the range in bytes
397 * @flags: reservation flags (see linux/bootmem.h)
398 *
399 * Partial pages will be reserved.
400 *
Johannes Weinere2bf3ca2008-07-23 21:28:06 -0700401 * The range must reside completely on the specified node.
Johannes Weinera66fd7d2008-07-23 21:28:01 -0700402 */
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700403int __init reserve_bootmem_node(pg_data_t *pgdat, unsigned long physaddr,
404 unsigned long size, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700405{
Johannes Weinere2bf3ca2008-07-23 21:28:06 -0700406 unsigned long start, end;
Franck Bui-Huubbc7b922006-09-25 23:31:07 -0700407
Johannes Weinere2bf3ca2008-07-23 21:28:06 -0700408 start = PFN_DOWN(physaddr);
409 end = PFN_UP(physaddr + size);
410
411 return mark_bootmem_node(pgdat->bdata, start, end, 1, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700412}
413
Johannes Weinera66fd7d2008-07-23 21:28:01 -0700414/**
415 * reserve_bootmem - mark a page range as usable
416 * @addr: starting address of the range
417 * @size: size of the range in bytes
418 * @flags: reservation flags (see linux/bootmem.h)
419 *
420 * Partial pages will be reserved.
421 *
Johannes Weinere2bf3ca2008-07-23 21:28:06 -0700422 * The range must be contiguous but may span node boundaries.
Johannes Weinera66fd7d2008-07-23 21:28:01 -0700423 */
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700424int __init reserve_bootmem(unsigned long addr, unsigned long size,
425 int flags)
426{
Johannes Weinere2bf3ca2008-07-23 21:28:06 -0700427 unsigned long start, end;
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700428
Johannes Weinere2bf3ca2008-07-23 21:28:06 -0700429 start = PFN_DOWN(addr);
430 end = PFN_UP(addr + size);
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700431
Johannes Weinere2bf3ca2008-07-23 21:28:06 -0700432 return mark_bootmem(start, end, 1, flags);
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700433}
Johannes Weiner223e8dc2008-07-23 21:28:00 -0700434
Johannes Weiner481ebd02008-08-20 14:09:15 -0700435static unsigned long align_idx(struct bootmem_data *bdata, unsigned long idx,
436 unsigned long step)
437{
438 unsigned long base = bdata->node_min_pfn;
439
440 /*
441 * Align the index with respect to the node start so that the
442 * combination of both satisfies the requested alignment.
443 */
444
445 return ALIGN(base + idx, step) - base;
446}
447
448static unsigned long align_off(struct bootmem_data *bdata, unsigned long off,
449 unsigned long align)
450{
451 unsigned long base = PFN_PHYS(bdata->node_min_pfn);
452
453 /* Same as align_idx for byte offsets */
454
455 return ALIGN(base + off, align) - base;
456}
457
Tejun Heod0c4f572009-03-01 16:06:56 +0900458static void * __init alloc_bootmem_core(struct bootmem_data *bdata,
459 unsigned long size, unsigned long align,
460 unsigned long goal, unsigned long limit)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461{
Johannes Weiner0f3caba2008-07-23 21:28:07 -0700462 unsigned long fallback = 0;
Johannes Weiner5f2809e2008-07-23 21:28:05 -0700463 unsigned long min, max, start, sidx, midx, step;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464
Johannes Weiner594fe1a2009-01-06 14:40:32 -0800465 bdebug("nid=%td size=%lx [%lu pages] align=%lx goal=%lx limit=%lx\n",
466 bdata - bootmem_node_data, size, PAGE_ALIGN(size) >> PAGE_SHIFT,
467 align, goal, limit);
468
Johannes Weiner5f2809e2008-07-23 21:28:05 -0700469 BUG_ON(!size);
470 BUG_ON(align & (align - 1));
471 BUG_ON(limit && goal + size > limit);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700472
Christian Krafft7c309a62006-12-06 20:32:41 -0800473 if (!bdata->node_bootmem_map)
474 return NULL;
475
Johannes Weiner3560e242008-07-23 21:28:09 -0700476 min = bdata->node_min_pfn;
Johannes Weiner5f2809e2008-07-23 21:28:05 -0700477 max = bdata->node_low_pfn;
Yinghai Lu9a2dc042008-03-18 12:44:48 -0700478
Johannes Weiner5f2809e2008-07-23 21:28:05 -0700479 goal >>= PAGE_SHIFT;
480 limit >>= PAGE_SHIFT;
481
482 if (limit && max > limit)
483 max = limit;
484 if (max <= min)
Yinghai Lu9a2dc042008-03-18 12:44:48 -0700485 return NULL;
486
Johannes Weiner5f2809e2008-07-23 21:28:05 -0700487 step = max(align >> PAGE_SHIFT, 1UL);
Yasunori Goto281dd252005-10-19 15:52:18 -0700488
Johannes Weiner5f2809e2008-07-23 21:28:05 -0700489 if (goal && min < goal && goal < max)
490 start = ALIGN(goal, step);
491 else
492 start = ALIGN(min, step);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700493
Johannes Weiner481ebd02008-08-20 14:09:15 -0700494 sidx = start - bdata->node_min_pfn;
Johannes Weiner3560e242008-07-23 21:28:09 -0700495 midx = max - bdata->node_min_pfn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700496
Johannes Weiner5f2809e2008-07-23 21:28:05 -0700497 if (bdata->hint_idx > sidx) {
Johannes Weiner0f3caba2008-07-23 21:28:07 -0700498 /*
499 * Handle the valid case of sidx being zero and still
500 * catch the fallback below.
501 */
502 fallback = sidx + 1;
Johannes Weiner481ebd02008-08-20 14:09:15 -0700503 sidx = align_idx(bdata, bdata->hint_idx, step);
Yinghai Luad093152008-03-10 23:23:42 -0700504 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700505
Johannes Weiner5f2809e2008-07-23 21:28:05 -0700506 while (1) {
507 int merge;
508 void *region;
509 unsigned long eidx, i, start_off, end_off;
510find_block:
511 sidx = find_next_zero_bit(bdata->node_bootmem_map, midx, sidx);
Johannes Weiner481ebd02008-08-20 14:09:15 -0700512 sidx = align_idx(bdata, sidx, step);
Johannes Weiner5f2809e2008-07-23 21:28:05 -0700513 eidx = sidx + PFN_UP(size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514
Johannes Weiner5f2809e2008-07-23 21:28:05 -0700515 if (sidx >= midx || eidx > midx)
Haren Myneni66d43e92005-12-12 00:37:39 -0800516 break;
Johannes Weiner5f2809e2008-07-23 21:28:05 -0700517
518 for (i = sidx; i < eidx; i++)
519 if (test_bit(i, bdata->node_bootmem_map)) {
Johannes Weiner481ebd02008-08-20 14:09:15 -0700520 sidx = align_idx(bdata, i, step);
Johannes Weiner5f2809e2008-07-23 21:28:05 -0700521 if (sidx == i)
522 sidx += step;
523 goto find_block;
524 }
525
Mikulas Patocka627240a2008-08-15 00:40:17 -0700526 if (bdata->last_end_off & (PAGE_SIZE - 1) &&
Johannes Weiner5f2809e2008-07-23 21:28:05 -0700527 PFN_DOWN(bdata->last_end_off) + 1 == sidx)
Johannes Weiner481ebd02008-08-20 14:09:15 -0700528 start_off = align_off(bdata, bdata->last_end_off, align);
Johannes Weiner5f2809e2008-07-23 21:28:05 -0700529 else
530 start_off = PFN_PHYS(sidx);
531
532 merge = PFN_DOWN(start_off) < sidx;
533 end_off = start_off + size;
534
535 bdata->last_end_off = end_off;
536 bdata->hint_idx = PFN_UP(end_off);
537
538 /*
539 * Reserve the area now:
540 */
Johannes Weinerd747fa42008-07-23 21:28:05 -0700541 if (__reserve(bdata, PFN_DOWN(start_off) + merge,
542 PFN_UP(end_off), BOOTMEM_EXCLUSIVE))
543 BUG();
Johannes Weiner5f2809e2008-07-23 21:28:05 -0700544
Johannes Weiner3560e242008-07-23 21:28:09 -0700545 region = phys_to_virt(PFN_PHYS(bdata->node_min_pfn) +
546 start_off);
Johannes Weiner5f2809e2008-07-23 21:28:05 -0700547 memset(region, 0, size);
Catalin Marinas008139d2009-08-27 14:29:17 +0100548 /*
549 * The min_count is set to 0 so that bootmem allocated blocks
550 * are never reported as leaks.
551 */
552 kmemleak_alloc(region, size, 0, 0);
Johannes Weiner5f2809e2008-07-23 21:28:05 -0700553 return region;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700554 }
555
Johannes Weiner0f3caba2008-07-23 21:28:07 -0700556 if (fallback) {
Johannes Weiner481ebd02008-08-20 14:09:15 -0700557 sidx = align_idx(bdata, fallback - 1, step);
Johannes Weiner0f3caba2008-07-23 21:28:07 -0700558 fallback = 0;
559 goto find_block;
560 }
561
562 return NULL;
563}
564
Tejun Heod0c4f572009-03-01 16:06:56 +0900565static void * __init alloc_arch_preferred_bootmem(bootmem_data_t *bdata,
566 unsigned long size, unsigned long align,
567 unsigned long goal, unsigned long limit)
568{
Pekka Enberg441c7e02009-06-10 20:05:53 +0300569 if (WARN_ON_ONCE(slab_is_available()))
570 return kzalloc(size, GFP_NOWAIT);
571
Tejun Heod0c4f572009-03-01 16:06:56 +0900572#ifdef CONFIG_HAVE_ARCH_BOOTMEM
Joe Perches433f13a2009-06-18 16:48:58 -0700573 {
574 bootmem_data_t *p_bdata;
Tejun Heod0c4f572009-03-01 16:06:56 +0900575
Joe Perches433f13a2009-06-18 16:48:58 -0700576 p_bdata = bootmem_arch_preferred_node(bdata, size, align,
577 goal, limit);
578 if (p_bdata)
579 return alloc_bootmem_core(p_bdata, size, align,
580 goal, limit);
581 }
Tejun Heod0c4f572009-03-01 16:06:56 +0900582#endif
583 return NULL;
584}
585
Johannes Weiner0f3caba2008-07-23 21:28:07 -0700586static void * __init ___alloc_bootmem_nopanic(unsigned long size,
587 unsigned long align,
588 unsigned long goal,
589 unsigned long limit)
590{
591 bootmem_data_t *bdata;
Tejun Heod0c4f572009-03-01 16:06:56 +0900592 void *region;
Johannes Weiner0f3caba2008-07-23 21:28:07 -0700593
594restart:
Tejun Heod0c4f572009-03-01 16:06:56 +0900595 region = alloc_arch_preferred_bootmem(NULL, size, align, goal, limit);
596 if (region)
597 return region;
Johannes Weiner0f3caba2008-07-23 21:28:07 -0700598
Tejun Heod0c4f572009-03-01 16:06:56 +0900599 list_for_each_entry(bdata, &bdata_list, list) {
Johannes Weiner0f3caba2008-07-23 21:28:07 -0700600 if (goal && bdata->node_low_pfn <= PFN_DOWN(goal))
601 continue;
Johannes Weiner3560e242008-07-23 21:28:09 -0700602 if (limit && bdata->node_min_pfn >= PFN_DOWN(limit))
Johannes Weiner0f3caba2008-07-23 21:28:07 -0700603 break;
604
605 region = alloc_bootmem_core(bdata, size, align, goal, limit);
606 if (region)
607 return region;
608 }
609
Johannes Weiner5f2809e2008-07-23 21:28:05 -0700610 if (goal) {
611 goal = 0;
Johannes Weiner0f3caba2008-07-23 21:28:07 -0700612 goto restart;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700613 }
Johannes Weiner5f2809e2008-07-23 21:28:05 -0700614
Linus Torvalds1da177e2005-04-16 15:20:36 -0700615 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700616}
617
Johannes Weinera66fd7d2008-07-23 21:28:01 -0700618/**
619 * __alloc_bootmem_nopanic - allocate boot memory without panicking
620 * @size: size of the request in bytes
621 * @align: alignment of the region
622 * @goal: preferred starting address of the region
623 *
624 * The goal is dropped if it can not be satisfied and the allocation will
625 * fall back to memory below @goal.
626 *
627 * Allocation may happen on any node in the system.
628 *
629 * Returns NULL on failure.
630 */
Franck Bui-Huubb0923a2006-09-25 23:31:05 -0700631void * __init __alloc_bootmem_nopanic(unsigned long size, unsigned long align,
Johannes Weiner0f3caba2008-07-23 21:28:07 -0700632 unsigned long goal)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700633{
Johannes Weiner0f3caba2008-07-23 21:28:07 -0700634 return ___alloc_bootmem_nopanic(size, align, goal, 0);
635}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700636
Johannes Weiner0f3caba2008-07-23 21:28:07 -0700637static void * __init ___alloc_bootmem(unsigned long size, unsigned long align,
638 unsigned long goal, unsigned long limit)
639{
640 void *mem = ___alloc_bootmem_nopanic(size, align, goal, limit);
641
642 if (mem)
643 return mem;
644 /*
645 * Whoops, we cannot satisfy the allocation request.
646 */
647 printk(KERN_ALERT "bootmem alloc of %lu bytes failed!\n", size);
648 panic("Out of memory");
Andi Kleena8062232006-04-07 19:49:21 +0200649 return NULL;
650}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700651
Johannes Weinera66fd7d2008-07-23 21:28:01 -0700652/**
653 * __alloc_bootmem - allocate boot memory
654 * @size: size of the request in bytes
655 * @align: alignment of the region
656 * @goal: preferred starting address of the region
657 *
658 * The goal is dropped if it can not be satisfied and the allocation will
659 * fall back to memory below @goal.
660 *
661 * Allocation may happen on any node in the system.
662 *
663 * The function panics if the request can not be satisfied.
664 */
Franck Bui-Huubb0923a2006-09-25 23:31:05 -0700665void * __init __alloc_bootmem(unsigned long size, unsigned long align,
666 unsigned long goal)
Andi Kleena8062232006-04-07 19:49:21 +0200667{
Johannes Weiner0f3caba2008-07-23 21:28:07 -0700668 return ___alloc_bootmem(size, align, goal, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700669}
670
Johannes Weiner4cc278b2008-07-23 21:28:08 -0700671static void * __init ___alloc_bootmem_node(bootmem_data_t *bdata,
672 unsigned long size, unsigned long align,
673 unsigned long goal, unsigned long limit)
674{
675 void *ptr;
676
Tejun Heod0c4f572009-03-01 16:06:56 +0900677 ptr = alloc_arch_preferred_bootmem(bdata, size, align, goal, limit);
678 if (ptr)
679 return ptr;
680
Johannes Weiner4cc278b2008-07-23 21:28:08 -0700681 ptr = alloc_bootmem_core(bdata, size, align, goal, limit);
682 if (ptr)
683 return ptr;
684
685 return ___alloc_bootmem(size, align, goal, limit);
686}
687
Johannes Weinera66fd7d2008-07-23 21:28:01 -0700688/**
689 * __alloc_bootmem_node - allocate boot memory from a specific node
690 * @pgdat: node to allocate from
691 * @size: size of the request in bytes
692 * @align: alignment of the region
693 * @goal: preferred starting address of the region
694 *
695 * The goal is dropped if it can not be satisfied and the allocation will
696 * fall back to memory below @goal.
697 *
698 * Allocation may fall back to any node in the system if the specified node
699 * can not hold the requested memory.
700 *
701 * The function panics if the request can not be satisfied.
702 */
Franck Bui-Huubb0923a2006-09-25 23:31:05 -0700703void * __init __alloc_bootmem_node(pg_data_t *pgdat, unsigned long size,
704 unsigned long align, unsigned long goal)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700705{
Pekka Enbergc91c4772009-06-11 08:10:28 +0300706 if (WARN_ON_ONCE(slab_is_available()))
707 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
708
Johannes Weiner4cc278b2008-07-23 21:28:08 -0700709 return ___alloc_bootmem_node(pgdat->bdata, size, align, goal, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700710}
711
Yasunori Gotoe70260a2008-04-28 02:13:32 -0700712#ifdef CONFIG_SPARSEMEM
Johannes Weinera66fd7d2008-07-23 21:28:01 -0700713/**
714 * alloc_bootmem_section - allocate boot memory from a specific section
715 * @size: size of the request in bytes
716 * @section_nr: sparse map section to allocate from
717 *
718 * Return NULL on failure.
719 */
Yasunori Gotoe70260a2008-04-28 02:13:32 -0700720void * __init alloc_bootmem_section(unsigned long size,
721 unsigned long section_nr)
722{
Johannes Weiner75a56cf2008-07-23 21:28:09 -0700723 bootmem_data_t *bdata;
724 unsigned long pfn, goal, limit;
Yasunori Gotoe70260a2008-04-28 02:13:32 -0700725
726 pfn = section_nr_to_pfn(section_nr);
Johannes Weiner75a56cf2008-07-23 21:28:09 -0700727 goal = pfn << PAGE_SHIFT;
728 limit = section_nr_to_pfn(section_nr + 1) << PAGE_SHIFT;
729 bdata = &bootmem_node_data[early_pfn_to_nid(pfn)];
Yasunori Gotoe70260a2008-04-28 02:13:32 -0700730
Johannes Weiner75a56cf2008-07-23 21:28:09 -0700731 return alloc_bootmem_core(bdata, size, SMP_CACHE_BYTES, goal, limit);
Yasunori Gotoe70260a2008-04-28 02:13:32 -0700732}
733#endif
734
Andi Kleenb54bbf72008-07-23 21:27:45 -0700735void * __init __alloc_bootmem_node_nopanic(pg_data_t *pgdat, unsigned long size,
736 unsigned long align, unsigned long goal)
737{
738 void *ptr;
739
Pekka Enbergc91c4772009-06-11 08:10:28 +0300740 if (WARN_ON_ONCE(slab_is_available()))
741 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
742
Tejun Heod0c4f572009-03-01 16:06:56 +0900743 ptr = alloc_arch_preferred_bootmem(pgdat->bdata, size, align, goal, 0);
744 if (ptr)
745 return ptr;
746
Andi Kleenb54bbf72008-07-23 21:27:45 -0700747 ptr = alloc_bootmem_core(pgdat->bdata, size, align, goal, 0);
748 if (ptr)
749 return ptr;
750
751 return __alloc_bootmem_nopanic(size, align, goal);
752}
753
Heiko Carstensdfd54cb2006-09-25 23:31:33 -0700754#ifndef ARCH_LOW_ADDRESS_LIMIT
755#define ARCH_LOW_ADDRESS_LIMIT 0xffffffffUL
756#endif
Ravikiran G Thirumalai008857c2006-01-06 00:11:01 -0800757
Johannes Weinera66fd7d2008-07-23 21:28:01 -0700758/**
759 * __alloc_bootmem_low - allocate low boot memory
760 * @size: size of the request in bytes
761 * @align: alignment of the region
762 * @goal: preferred starting address of the region
763 *
764 * The goal is dropped if it can not be satisfied and the allocation will
765 * fall back to memory below @goal.
766 *
767 * Allocation may happen on any node in the system.
768 *
769 * The function panics if the request can not be satisfied.
770 */
Franck Bui-Huubb0923a2006-09-25 23:31:05 -0700771void * __init __alloc_bootmem_low(unsigned long size, unsigned long align,
772 unsigned long goal)
Ravikiran G Thirumalai008857c2006-01-06 00:11:01 -0800773{
Johannes Weiner0f3caba2008-07-23 21:28:07 -0700774 return ___alloc_bootmem(size, align, goal, ARCH_LOW_ADDRESS_LIMIT);
Ravikiran G Thirumalai008857c2006-01-06 00:11:01 -0800775}
776
Johannes Weinera66fd7d2008-07-23 21:28:01 -0700777/**
778 * __alloc_bootmem_low_node - allocate low boot memory from a specific node
779 * @pgdat: node to allocate from
780 * @size: size of the request in bytes
781 * @align: alignment of the region
782 * @goal: preferred starting address of the region
783 *
784 * The goal is dropped if it can not be satisfied and the allocation will
785 * fall back to memory below @goal.
786 *
787 * Allocation may fall back to any node in the system if the specified node
788 * can not hold the requested memory.
789 *
790 * The function panics if the request can not be satisfied.
791 */
Ravikiran G Thirumalai008857c2006-01-06 00:11:01 -0800792void * __init __alloc_bootmem_low_node(pg_data_t *pgdat, unsigned long size,
793 unsigned long align, unsigned long goal)
794{
Pekka Enbergc91c4772009-06-11 08:10:28 +0300795 if (WARN_ON_ONCE(slab_is_available()))
796 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
797
Johannes Weiner4cc278b2008-07-23 21:28:08 -0700798 return ___alloc_bootmem_node(pgdat->bdata, size, align,
799 goal, ARCH_LOW_ADDRESS_LIMIT);
Ravikiran G Thirumalai008857c2006-01-06 00:11:01 -0800800}