blob: 50353e0dac12dc41441b64c78b59c852140f76c4 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/mm/bootmem.c
3 *
4 * Copyright (C) 1999 Ingo Molnar
5 * Discontiguous memory support, Kanoj Sarcar, SGI, Nov 1999
6 *
7 * simple boot-time physical memory area allocator and
8 * free memory collector. It's used to deal with reserved
9 * system memory and memory holes as well.
10 */
11
12#include <linux/mm.h>
13#include <linux/kernel_stat.h>
14#include <linux/swap.h>
15#include <linux/interrupt.h>
16#include <linux/init.h>
17#include <linux/bootmem.h>
18#include <linux/mmzone.h>
19#include <linux/module.h>
20#include <asm/dma.h>
21#include <asm/io.h>
22#include "internal.h"
23
24/*
25 * Access to this subsystem has to be serialized externally. (this is
26 * true for the boot process anyway)
27 */
28unsigned long max_low_pfn;
29unsigned long min_low_pfn;
30unsigned long max_pfn;
31
Adrian Bunk6d46cc62006-07-10 04:44:21 -070032EXPORT_UNUSED_SYMBOL(max_pfn); /* June 2006 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070033
KAMEZAWA Hiroyuki679bc9f2006-03-27 01:15:58 -080034static LIST_HEAD(bdata_list);
Vivek Goyal92aa63a2005-06-25 14:58:18 -070035#ifdef CONFIG_CRASH_DUMP
36/*
37 * If we have booted due to a crash, max_pfn will be a very low value. We need
38 * to know the amount of memory that the previous kernel used.
39 */
40unsigned long saved_max_pfn;
41#endif
42
Linus Torvalds1da177e2005-04-16 15:20:36 -070043/* return the number of _pages_ that will be allocated for the boot bitmap */
44unsigned long __init bootmem_bootmap_pages (unsigned long pages)
45{
46 unsigned long mapsize;
47
48 mapsize = (pages+7)/8;
49 mapsize = (mapsize + ~PAGE_MASK) & PAGE_MASK;
50 mapsize >>= PAGE_SHIFT;
51
52 return mapsize;
53}
KAMEZAWA Hiroyuki679bc9f2006-03-27 01:15:58 -080054/*
55 * link bdata in order
56 */
57static void link_bootmem(bootmem_data_t *bdata)
58{
59 bootmem_data_t *ent;
60 if (list_empty(&bdata_list)) {
61 list_add(&bdata->list, &bdata_list);
62 return;
63 }
64 /* insert in order */
65 list_for_each_entry(ent, &bdata_list, list) {
66 if (bdata->node_boot_start < ent->node_boot_start) {
67 list_add_tail(&bdata->list, &ent->list);
68 return;
69 }
70 }
71 list_add_tail(&bdata->list, &bdata_list);
72 return;
73}
74
Linus Torvalds1da177e2005-04-16 15:20:36 -070075
76/*
77 * Called once to set up the allocator itself.
78 */
79static unsigned long __init init_bootmem_core (pg_data_t *pgdat,
80 unsigned long mapstart, unsigned long start, unsigned long end)
81{
82 bootmem_data_t *bdata = pgdat->bdata;
83 unsigned long mapsize = ((end - start)+7)/8;
84
Nick Wilson8c0e33c2005-06-25 14:59:00 -070085 mapsize = ALIGN(mapsize, sizeof(long));
Linus Torvalds1da177e2005-04-16 15:20:36 -070086 bdata->node_bootmem_map = phys_to_virt(mapstart << PAGE_SHIFT);
87 bdata->node_boot_start = (start << PAGE_SHIFT);
88 bdata->node_low_pfn = end;
KAMEZAWA Hiroyuki679bc9f2006-03-27 01:15:58 -080089 link_bootmem(bdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -070090
91 /*
92 * Initially all pages are reserved - setup_arch() has to
93 * register free RAM areas explicitly.
94 */
95 memset(bdata->node_bootmem_map, 0xff, mapsize);
96
97 return mapsize;
98}
99
100/*
101 * Marks a particular physical memory range as unallocatable. Usable RAM
102 * might be used for boot-time allocations - or it might get added
103 * to the free page pool later on.
104 */
105static void __init reserve_bootmem_core(bootmem_data_t *bdata, unsigned long addr, unsigned long size)
106{
107 unsigned long i;
108 /*
109 * round up, partially reserved pages are considered
110 * fully reserved.
111 */
112 unsigned long sidx = (addr - bdata->node_boot_start)/PAGE_SIZE;
113 unsigned long eidx = (addr + size - bdata->node_boot_start +
114 PAGE_SIZE-1)/PAGE_SIZE;
115 unsigned long end = (addr + size + PAGE_SIZE-1)/PAGE_SIZE;
116
117 BUG_ON(!size);
118 BUG_ON(sidx >= eidx);
119 BUG_ON((addr >> PAGE_SHIFT) >= bdata->node_low_pfn);
120 BUG_ON(end > bdata->node_low_pfn);
121
122 for (i = sidx; i < eidx; i++)
123 if (test_and_set_bit(i, bdata->node_bootmem_map)) {
124#ifdef CONFIG_DEBUG_BOOTMEM
125 printk("hm, page %08lx reserved twice.\n", i*PAGE_SIZE);
126#endif
127 }
128}
129
130static void __init free_bootmem_core(bootmem_data_t *bdata, unsigned long addr, unsigned long size)
131{
132 unsigned long i;
133 unsigned long start;
134 /*
135 * round down end of usable mem, partially free pages are
136 * considered reserved.
137 */
138 unsigned long sidx;
139 unsigned long eidx = (addr + size - bdata->node_boot_start)/PAGE_SIZE;
140 unsigned long end = (addr + size)/PAGE_SIZE;
141
142 BUG_ON(!size);
143 BUG_ON(end > bdata->node_low_pfn);
144
145 if (addr < bdata->last_success)
146 bdata->last_success = addr;
147
148 /*
149 * Round up the beginning of the address.
150 */
151 start = (addr + PAGE_SIZE-1) / PAGE_SIZE;
152 sidx = start - (bdata->node_boot_start/PAGE_SIZE);
153
154 for (i = sidx; i < eidx; i++) {
155 if (unlikely(!test_and_clear_bit(i, bdata->node_bootmem_map)))
156 BUG();
157 }
158}
159
160/*
161 * We 'merge' subsequent allocations to save space. We might 'lose'
162 * some fraction of a page if allocations cannot be satisfied due to
163 * size constraints on boxes where there is physical RAM space
164 * fragmentation - in these cases (mostly large memory boxes) this
165 * is not a problem.
166 *
167 * On low memory boxes we get it right in 100% of the cases.
168 *
169 * alignment has to be a power of 2 value.
170 *
171 * NOTE: This function is _not_ reentrant.
172 */
Andi Kleen267b4802006-03-25 16:31:10 +0100173void * __init
Linus Torvalds1da177e2005-04-16 15:20:36 -0700174__alloc_bootmem_core(struct bootmem_data *bdata, unsigned long size,
Yasunori Goto281dd252005-10-19 15:52:18 -0700175 unsigned long align, unsigned long goal, unsigned long limit)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700176{
177 unsigned long offset, remaining_size, areasize, preferred;
Yasunori Goto281dd252005-10-19 15:52:18 -0700178 unsigned long i, start = 0, incr, eidx, end_pfn = bdata->node_low_pfn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700179 void *ret;
180
181 if(!size) {
182 printk("__alloc_bootmem_core(): zero-sized request\n");
183 BUG();
184 }
185 BUG_ON(align & (align-1));
186
Yasunori Goto281dd252005-10-19 15:52:18 -0700187 if (limit && bdata->node_boot_start >= limit)
188 return NULL;
189
190 limit >>=PAGE_SHIFT;
191 if (limit && end_pfn > limit)
192 end_pfn = limit;
193
194 eidx = end_pfn - (bdata->node_boot_start >> PAGE_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700195 offset = 0;
196 if (align &&
197 (bdata->node_boot_start & (align - 1UL)) != 0)
198 offset = (align - (bdata->node_boot_start & (align - 1UL)));
199 offset >>= PAGE_SHIFT;
200
201 /*
202 * We try to allocate bootmem pages above 'goal'
203 * first, then we try to allocate lower pages.
204 */
205 if (goal && (goal >= bdata->node_boot_start) &&
Yasunori Goto281dd252005-10-19 15:52:18 -0700206 ((goal >> PAGE_SHIFT) < end_pfn)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207 preferred = goal - bdata->node_boot_start;
208
209 if (bdata->last_success >= preferred)
Yasunori Goto281dd252005-10-19 15:52:18 -0700210 if (!limit || (limit && limit > bdata->last_success))
211 preferred = bdata->last_success;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700212 } else
213 preferred = 0;
214
Nick Wilson8c0e33c2005-06-25 14:59:00 -0700215 preferred = ALIGN(preferred, align) >> PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700216 preferred += offset;
217 areasize = (size+PAGE_SIZE-1)/PAGE_SIZE;
218 incr = align >> PAGE_SHIFT ? : 1;
219
220restart_scan:
221 for (i = preferred; i < eidx; i += incr) {
222 unsigned long j;
223 i = find_next_zero_bit(bdata->node_bootmem_map, eidx, i);
224 i = ALIGN(i, incr);
Haren Myneni66d43e92005-12-12 00:37:39 -0800225 if (i >= eidx)
226 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700227 if (test_bit(i, bdata->node_bootmem_map))
228 continue;
229 for (j = i + 1; j < i + areasize; ++j) {
230 if (j >= eidx)
231 goto fail_block;
232 if (test_bit (j, bdata->node_bootmem_map))
233 goto fail_block;
234 }
235 start = i;
236 goto found;
237 fail_block:
238 i = ALIGN(j, incr);
239 }
240
241 if (preferred > offset) {
242 preferred = offset;
243 goto restart_scan;
244 }
245 return NULL;
246
247found:
248 bdata->last_success = start << PAGE_SHIFT;
249 BUG_ON(start >= eidx);
250
251 /*
252 * Is the next page of the previous allocation-end the start
253 * of this allocation's buffer? If yes then we can 'merge'
254 * the previous partial page with this allocation.
255 */
256 if (align < PAGE_SIZE &&
257 bdata->last_offset && bdata->last_pos+1 == start) {
Nick Wilson8c0e33c2005-06-25 14:59:00 -0700258 offset = ALIGN(bdata->last_offset, align);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700259 BUG_ON(offset > PAGE_SIZE);
260 remaining_size = PAGE_SIZE-offset;
261 if (size < remaining_size) {
262 areasize = 0;
263 /* last_pos unchanged */
264 bdata->last_offset = offset+size;
265 ret = phys_to_virt(bdata->last_pos*PAGE_SIZE + offset +
266 bdata->node_boot_start);
267 } else {
268 remaining_size = size - remaining_size;
269 areasize = (remaining_size+PAGE_SIZE-1)/PAGE_SIZE;
270 ret = phys_to_virt(bdata->last_pos*PAGE_SIZE + offset +
271 bdata->node_boot_start);
272 bdata->last_pos = start+areasize-1;
273 bdata->last_offset = remaining_size;
274 }
275 bdata->last_offset &= ~PAGE_MASK;
276 } else {
277 bdata->last_pos = start + areasize - 1;
278 bdata->last_offset = size & ~PAGE_MASK;
279 ret = phys_to_virt(start * PAGE_SIZE + bdata->node_boot_start);
280 }
281
282 /*
283 * Reserve the area now:
284 */
285 for (i = start; i < start+areasize; i++)
286 if (unlikely(test_and_set_bit(i, bdata->node_bootmem_map)))
287 BUG();
288 memset(ret, 0, size);
289 return ret;
290}
291
292static unsigned long __init free_all_bootmem_core(pg_data_t *pgdat)
293{
294 struct page *page;
Andy Whitcroftd41dee32005-06-23 00:07:54 -0700295 unsigned long pfn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700296 bootmem_data_t *bdata = pgdat->bdata;
297 unsigned long i, count, total = 0;
298 unsigned long idx;
299 unsigned long *map;
300 int gofast = 0;
301
302 BUG_ON(!bdata->node_bootmem_map);
303
304 count = 0;
305 /* first extant page of the node */
Andy Whitcroftd41dee32005-06-23 00:07:54 -0700306 pfn = bdata->node_boot_start >> PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700307 idx = bdata->node_low_pfn - (bdata->node_boot_start >> PAGE_SHIFT);
308 map = bdata->node_bootmem_map;
309 /* Check physaddr is O(LOG2(BITS_PER_LONG)) page aligned */
310 if (bdata->node_boot_start == 0 ||
311 ffs(bdata->node_boot_start) - PAGE_SHIFT > ffs(BITS_PER_LONG))
312 gofast = 1;
313 for (i = 0; i < idx; ) {
314 unsigned long v = ~map[i / BITS_PER_LONG];
Andy Whitcroftd41dee32005-06-23 00:07:54 -0700315
Linus Torvalds1da177e2005-04-16 15:20:36 -0700316 if (gofast && v == ~0UL) {
David Howellsa226f6c2006-01-06 00:11:08 -0800317 int order;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700318
Andy Whitcroftd41dee32005-06-23 00:07:54 -0700319 page = pfn_to_page(pfn);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700320 count += BITS_PER_LONG;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700321 order = ffs(BITS_PER_LONG) - 1;
David Howellsa226f6c2006-01-06 00:11:08 -0800322 __free_pages_bootmem(page, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700323 i += BITS_PER_LONG;
324 page += BITS_PER_LONG;
325 } else if (v) {
326 unsigned long m;
Andy Whitcroftd41dee32005-06-23 00:07:54 -0700327
328 page = pfn_to_page(pfn);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700329 for (m = 1; m && i < idx; m<<=1, page++, i++) {
330 if (v & m) {
331 count++;
David Howellsa226f6c2006-01-06 00:11:08 -0800332 __free_pages_bootmem(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700333 }
334 }
335 } else {
336 i+=BITS_PER_LONG;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337 }
Andy Whitcroftd41dee32005-06-23 00:07:54 -0700338 pfn += BITS_PER_LONG;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700339 }
340 total += count;
341
342 /*
343 * Now free the allocator bitmap itself, it's not
344 * needed anymore:
345 */
346 page = virt_to_page(bdata->node_bootmem_map);
347 count = 0;
348 for (i = 0; i < ((bdata->node_low_pfn-(bdata->node_boot_start >> PAGE_SHIFT))/8 + PAGE_SIZE-1)/PAGE_SIZE; i++,page++) {
349 count++;
David Howellsa226f6c2006-01-06 00:11:08 -0800350 __free_pages_bootmem(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700351 }
352 total += count;
353 bdata->node_bootmem_map = NULL;
354
355 return total;
356}
357
358unsigned long __init init_bootmem_node (pg_data_t *pgdat, unsigned long freepfn, unsigned long startpfn, unsigned long endpfn)
359{
360 return(init_bootmem_core(pgdat, freepfn, startpfn, endpfn));
361}
362
363void __init reserve_bootmem_node (pg_data_t *pgdat, unsigned long physaddr, unsigned long size)
364{
365 reserve_bootmem_core(pgdat->bdata, physaddr, size);
366}
367
368void __init free_bootmem_node (pg_data_t *pgdat, unsigned long physaddr, unsigned long size)
369{
370 free_bootmem_core(pgdat->bdata, physaddr, size);
371}
372
373unsigned long __init free_all_bootmem_node (pg_data_t *pgdat)
374{
375 return(free_all_bootmem_core(pgdat));
376}
377
378unsigned long __init init_bootmem (unsigned long start, unsigned long pages)
379{
380 max_low_pfn = pages;
381 min_low_pfn = start;
382 return(init_bootmem_core(NODE_DATA(0), start, 0, pages));
383}
384
385#ifndef CONFIG_HAVE_ARCH_BOOTMEM_NODE
386void __init reserve_bootmem (unsigned long addr, unsigned long size)
387{
388 reserve_bootmem_core(NODE_DATA(0)->bdata, addr, size);
389}
390#endif /* !CONFIG_HAVE_ARCH_BOOTMEM_NODE */
391
392void __init free_bootmem (unsigned long addr, unsigned long size)
393{
394 free_bootmem_core(NODE_DATA(0)->bdata, addr, size);
395}
396
397unsigned long __init free_all_bootmem (void)
398{
399 return(free_all_bootmem_core(NODE_DATA(0)));
400}
401
Andi Kleena8062232006-04-07 19:49:21 +0200402void * __init __alloc_bootmem_nopanic(unsigned long size, unsigned long align, unsigned long goal)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700403{
KAMEZAWA Hiroyuki679bc9f2006-03-27 01:15:58 -0800404 bootmem_data_t *bdata;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700405 void *ptr;
406
KAMEZAWA Hiroyuki679bc9f2006-03-27 01:15:58 -0800407 list_for_each_entry(bdata, &bdata_list, list)
408 if ((ptr = __alloc_bootmem_core(bdata, size, align, goal, 0)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700409 return(ptr);
Andi Kleena8062232006-04-07 19:49:21 +0200410 return NULL;
411}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700412
Andi Kleena8062232006-04-07 19:49:21 +0200413void * __init __alloc_bootmem(unsigned long size, unsigned long align, unsigned long goal)
414{
415 void *mem = __alloc_bootmem_nopanic(size,align,goal);
416 if (mem)
417 return mem;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700418 /*
419 * Whoops, we cannot satisfy the allocation request.
420 */
421 printk(KERN_ALERT "bootmem alloc of %lu bytes failed!\n", size);
422 panic("Out of memory");
423 return NULL;
424}
425
Yasunori Goto281dd252005-10-19 15:52:18 -0700426
Ravikiran G Thirumalai008857c2006-01-06 00:11:01 -0800427void * __init __alloc_bootmem_node(pg_data_t *pgdat, unsigned long size, unsigned long align,
428 unsigned long goal)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700429{
430 void *ptr;
431
Ravikiran G Thirumalai008857c2006-01-06 00:11:01 -0800432 ptr = __alloc_bootmem_core(pgdat->bdata, size, align, goal, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700433 if (ptr)
434 return (ptr);
435
Ravikiran G Thirumalai008857c2006-01-06 00:11:01 -0800436 return __alloc_bootmem(size, align, goal);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700437}
438
Ravikiran G Thirumalai008857c2006-01-06 00:11:01 -0800439#define LOW32LIMIT 0xffffffff
440
441void * __init __alloc_bootmem_low(unsigned long size, unsigned long align, unsigned long goal)
442{
KAMEZAWA Hiroyuki679bc9f2006-03-27 01:15:58 -0800443 bootmem_data_t *bdata;
Ravikiran G Thirumalai008857c2006-01-06 00:11:01 -0800444 void *ptr;
445
KAMEZAWA Hiroyuki679bc9f2006-03-27 01:15:58 -0800446 list_for_each_entry(bdata, &bdata_list, list)
447 if ((ptr = __alloc_bootmem_core(bdata, size,
Ravikiran G Thirumalai008857c2006-01-06 00:11:01 -0800448 align, goal, LOW32LIMIT)))
449 return(ptr);
450
451 /*
452 * Whoops, we cannot satisfy the allocation request.
453 */
454 printk(KERN_ALERT "low bootmem alloc of %lu bytes failed!\n", size);
455 panic("Out of low memory");
456 return NULL;
457}
458
459void * __init __alloc_bootmem_low_node(pg_data_t *pgdat, unsigned long size,
460 unsigned long align, unsigned long goal)
461{
462 return __alloc_bootmem_core(pgdat->bdata, size, align, goal, LOW32LIMIT);
463}