blob: e8c567177dcf83e0b8d7352c581c0565c400844c [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
32EXPORT_SYMBOL(max_pfn); /* This is exported so
33 * dma_get_required_mask(), which uses
34 * it, can be an inline function */
35
Vivek Goyal92aa63a2005-06-25 14:58:18 -070036#ifdef CONFIG_CRASH_DUMP
37/*
38 * If we have booted due to a crash, max_pfn will be a very low value. We need
39 * to know the amount of memory that the previous kernel used.
40 */
41unsigned long saved_max_pfn;
42#endif
43
Linus Torvalds1da177e2005-04-16 15:20:36 -070044/* return the number of _pages_ that will be allocated for the boot bitmap */
45unsigned long __init bootmem_bootmap_pages (unsigned long pages)
46{
47 unsigned long mapsize;
48
49 mapsize = (pages+7)/8;
50 mapsize = (mapsize + ~PAGE_MASK) & PAGE_MASK;
51 mapsize >>= PAGE_SHIFT;
52
53 return mapsize;
54}
55
56/*
57 * Called once to set up the allocator itself.
58 */
59static unsigned long __init init_bootmem_core (pg_data_t *pgdat,
60 unsigned long mapstart, unsigned long start, unsigned long end)
61{
62 bootmem_data_t *bdata = pgdat->bdata;
63 unsigned long mapsize = ((end - start)+7)/8;
64
Linus Torvalds6e3254c2005-09-30 12:38:27 -070065 pgdat->pgdat_next = pgdat_list;
66 pgdat_list = pgdat;
Linus Torvalds1da177e2005-04-16 15:20:36 -070067
Nick Wilson8c0e33c2005-06-25 14:59:00 -070068 mapsize = ALIGN(mapsize, sizeof(long));
Linus Torvalds1da177e2005-04-16 15:20:36 -070069 bdata->node_bootmem_map = phys_to_virt(mapstart << PAGE_SHIFT);
70 bdata->node_boot_start = (start << PAGE_SHIFT);
71 bdata->node_low_pfn = end;
72
73 /*
74 * Initially all pages are reserved - setup_arch() has to
75 * register free RAM areas explicitly.
76 */
77 memset(bdata->node_bootmem_map, 0xff, mapsize);
78
79 return mapsize;
80}
81
82/*
83 * Marks a particular physical memory range as unallocatable. Usable RAM
84 * might be used for boot-time allocations - or it might get added
85 * to the free page pool later on.
86 */
87static void __init reserve_bootmem_core(bootmem_data_t *bdata, unsigned long addr, unsigned long size)
88{
89 unsigned long i;
90 /*
91 * round up, partially reserved pages are considered
92 * fully reserved.
93 */
94 unsigned long sidx = (addr - bdata->node_boot_start)/PAGE_SIZE;
95 unsigned long eidx = (addr + size - bdata->node_boot_start +
96 PAGE_SIZE-1)/PAGE_SIZE;
97 unsigned long end = (addr + size + PAGE_SIZE-1)/PAGE_SIZE;
98
99 BUG_ON(!size);
100 BUG_ON(sidx >= eidx);
101 BUG_ON((addr >> PAGE_SHIFT) >= bdata->node_low_pfn);
102 BUG_ON(end > bdata->node_low_pfn);
103
104 for (i = sidx; i < eidx; i++)
105 if (test_and_set_bit(i, bdata->node_bootmem_map)) {
106#ifdef CONFIG_DEBUG_BOOTMEM
107 printk("hm, page %08lx reserved twice.\n", i*PAGE_SIZE);
108#endif
109 }
110}
111
112static void __init free_bootmem_core(bootmem_data_t *bdata, unsigned long addr, unsigned long size)
113{
114 unsigned long i;
115 unsigned long start;
116 /*
117 * round down end of usable mem, partially free pages are
118 * considered reserved.
119 */
120 unsigned long sidx;
121 unsigned long eidx = (addr + size - bdata->node_boot_start)/PAGE_SIZE;
122 unsigned long end = (addr + size)/PAGE_SIZE;
123
124 BUG_ON(!size);
125 BUG_ON(end > bdata->node_low_pfn);
126
127 if (addr < bdata->last_success)
128 bdata->last_success = addr;
129
130 /*
131 * Round up the beginning of the address.
132 */
133 start = (addr + PAGE_SIZE-1) / PAGE_SIZE;
134 sidx = start - (bdata->node_boot_start/PAGE_SIZE);
135
136 for (i = sidx; i < eidx; i++) {
137 if (unlikely(!test_and_clear_bit(i, bdata->node_bootmem_map)))
138 BUG();
139 }
140}
141
142/*
143 * We 'merge' subsequent allocations to save space. We might 'lose'
144 * some fraction of a page if allocations cannot be satisfied due to
145 * size constraints on boxes where there is physical RAM space
146 * fragmentation - in these cases (mostly large memory boxes) this
147 * is not a problem.
148 *
149 * On low memory boxes we get it right in 100% of the cases.
150 *
151 * alignment has to be a power of 2 value.
152 *
153 * NOTE: This function is _not_ reentrant.
154 */
155static void * __init
156__alloc_bootmem_core(struct bootmem_data *bdata, unsigned long size,
Yasunori Goto281dd252005-10-19 15:52:18 -0700157 unsigned long align, unsigned long goal, unsigned long limit)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700158{
159 unsigned long offset, remaining_size, areasize, preferred;
Yasunori Goto281dd252005-10-19 15:52:18 -0700160 unsigned long i, start = 0, incr, eidx, end_pfn = bdata->node_low_pfn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700161 void *ret;
162
163 if(!size) {
164 printk("__alloc_bootmem_core(): zero-sized request\n");
165 BUG();
166 }
167 BUG_ON(align & (align-1));
168
Yasunori Goto281dd252005-10-19 15:52:18 -0700169 if (limit && bdata->node_boot_start >= limit)
170 return NULL;
171
172 limit >>=PAGE_SHIFT;
173 if (limit && end_pfn > limit)
174 end_pfn = limit;
175
176 eidx = end_pfn - (bdata->node_boot_start >> PAGE_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700177 offset = 0;
178 if (align &&
179 (bdata->node_boot_start & (align - 1UL)) != 0)
180 offset = (align - (bdata->node_boot_start & (align - 1UL)));
181 offset >>= PAGE_SHIFT;
182
183 /*
184 * We try to allocate bootmem pages above 'goal'
185 * first, then we try to allocate lower pages.
186 */
187 if (goal && (goal >= bdata->node_boot_start) &&
Yasunori Goto281dd252005-10-19 15:52:18 -0700188 ((goal >> PAGE_SHIFT) < end_pfn)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700189 preferred = goal - bdata->node_boot_start;
190
191 if (bdata->last_success >= preferred)
Yasunori Goto281dd252005-10-19 15:52:18 -0700192 if (!limit || (limit && limit > bdata->last_success))
193 preferred = bdata->last_success;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700194 } else
195 preferred = 0;
196
Nick Wilson8c0e33c2005-06-25 14:59:00 -0700197 preferred = ALIGN(preferred, align) >> PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700198 preferred += offset;
199 areasize = (size+PAGE_SIZE-1)/PAGE_SIZE;
200 incr = align >> PAGE_SHIFT ? : 1;
201
202restart_scan:
203 for (i = preferred; i < eidx; i += incr) {
204 unsigned long j;
205 i = find_next_zero_bit(bdata->node_bootmem_map, eidx, i);
206 i = ALIGN(i, incr);
207 if (test_bit(i, bdata->node_bootmem_map))
208 continue;
209 for (j = i + 1; j < i + areasize; ++j) {
210 if (j >= eidx)
211 goto fail_block;
212 if (test_bit (j, bdata->node_bootmem_map))
213 goto fail_block;
214 }
215 start = i;
216 goto found;
217 fail_block:
218 i = ALIGN(j, incr);
219 }
220
221 if (preferred > offset) {
222 preferred = offset;
223 goto restart_scan;
224 }
225 return NULL;
226
227found:
228 bdata->last_success = start << PAGE_SHIFT;
229 BUG_ON(start >= eidx);
230
231 /*
232 * Is the next page of the previous allocation-end the start
233 * of this allocation's buffer? If yes then we can 'merge'
234 * the previous partial page with this allocation.
235 */
236 if (align < PAGE_SIZE &&
237 bdata->last_offset && bdata->last_pos+1 == start) {
Nick Wilson8c0e33c2005-06-25 14:59:00 -0700238 offset = ALIGN(bdata->last_offset, align);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239 BUG_ON(offset > PAGE_SIZE);
240 remaining_size = PAGE_SIZE-offset;
241 if (size < remaining_size) {
242 areasize = 0;
243 /* last_pos unchanged */
244 bdata->last_offset = offset+size;
245 ret = phys_to_virt(bdata->last_pos*PAGE_SIZE + offset +
246 bdata->node_boot_start);
247 } else {
248 remaining_size = size - remaining_size;
249 areasize = (remaining_size+PAGE_SIZE-1)/PAGE_SIZE;
250 ret = phys_to_virt(bdata->last_pos*PAGE_SIZE + offset +
251 bdata->node_boot_start);
252 bdata->last_pos = start+areasize-1;
253 bdata->last_offset = remaining_size;
254 }
255 bdata->last_offset &= ~PAGE_MASK;
256 } else {
257 bdata->last_pos = start + areasize - 1;
258 bdata->last_offset = size & ~PAGE_MASK;
259 ret = phys_to_virt(start * PAGE_SIZE + bdata->node_boot_start);
260 }
261
262 /*
263 * Reserve the area now:
264 */
265 for (i = start; i < start+areasize; i++)
266 if (unlikely(test_and_set_bit(i, bdata->node_bootmem_map)))
267 BUG();
268 memset(ret, 0, size);
269 return ret;
270}
271
272static unsigned long __init free_all_bootmem_core(pg_data_t *pgdat)
273{
274 struct page *page;
Andy Whitcroftd41dee32005-06-23 00:07:54 -0700275 unsigned long pfn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700276 bootmem_data_t *bdata = pgdat->bdata;
277 unsigned long i, count, total = 0;
278 unsigned long idx;
279 unsigned long *map;
280 int gofast = 0;
281
282 BUG_ON(!bdata->node_bootmem_map);
283
284 count = 0;
285 /* first extant page of the node */
Andy Whitcroftd41dee32005-06-23 00:07:54 -0700286 pfn = bdata->node_boot_start >> PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700287 idx = bdata->node_low_pfn - (bdata->node_boot_start >> PAGE_SHIFT);
288 map = bdata->node_bootmem_map;
289 /* Check physaddr is O(LOG2(BITS_PER_LONG)) page aligned */
290 if (bdata->node_boot_start == 0 ||
291 ffs(bdata->node_boot_start) - PAGE_SHIFT > ffs(BITS_PER_LONG))
292 gofast = 1;
293 for (i = 0; i < idx; ) {
294 unsigned long v = ~map[i / BITS_PER_LONG];
Andy Whitcroftd41dee32005-06-23 00:07:54 -0700295
Linus Torvalds1da177e2005-04-16 15:20:36 -0700296 if (gofast && v == ~0UL) {
297 int j, order;
298
Andy Whitcroftd41dee32005-06-23 00:07:54 -0700299 page = pfn_to_page(pfn);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700300 count += BITS_PER_LONG;
301 __ClearPageReserved(page);
302 order = ffs(BITS_PER_LONG) - 1;
303 set_page_refs(page, order);
304 for (j = 1; j < BITS_PER_LONG; j++) {
305 if (j + 16 < BITS_PER_LONG)
306 prefetchw(page + j + 16);
307 __ClearPageReserved(page + j);
Nick Pigginb5810032005-10-29 18:16:12 -0700308 set_page_count(page + j, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700309 }
310 __free_pages(page, order);
311 i += BITS_PER_LONG;
312 page += BITS_PER_LONG;
313 } else if (v) {
314 unsigned long m;
Andy Whitcroftd41dee32005-06-23 00:07:54 -0700315
316 page = pfn_to_page(pfn);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700317 for (m = 1; m && i < idx; m<<=1, page++, i++) {
318 if (v & m) {
319 count++;
320 __ClearPageReserved(page);
321 set_page_refs(page, 0);
322 __free_page(page);
323 }
324 }
325 } else {
326 i+=BITS_PER_LONG;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700327 }
Andy Whitcroftd41dee32005-06-23 00:07:54 -0700328 pfn += BITS_PER_LONG;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700329 }
330 total += count;
331
332 /*
333 * Now free the allocator bitmap itself, it's not
334 * needed anymore:
335 */
336 page = virt_to_page(bdata->node_bootmem_map);
337 count = 0;
338 for (i = 0; i < ((bdata->node_low_pfn-(bdata->node_boot_start >> PAGE_SHIFT))/8 + PAGE_SIZE-1)/PAGE_SIZE; i++,page++) {
339 count++;
340 __ClearPageReserved(page);
341 set_page_count(page, 1);
342 __free_page(page);
343 }
344 total += count;
345 bdata->node_bootmem_map = NULL;
346
347 return total;
348}
349
350unsigned long __init init_bootmem_node (pg_data_t *pgdat, unsigned long freepfn, unsigned long startpfn, unsigned long endpfn)
351{
352 return(init_bootmem_core(pgdat, freepfn, startpfn, endpfn));
353}
354
355void __init reserve_bootmem_node (pg_data_t *pgdat, unsigned long physaddr, unsigned long size)
356{
357 reserve_bootmem_core(pgdat->bdata, physaddr, size);
358}
359
360void __init free_bootmem_node (pg_data_t *pgdat, unsigned long physaddr, unsigned long size)
361{
362 free_bootmem_core(pgdat->bdata, physaddr, size);
363}
364
365unsigned long __init free_all_bootmem_node (pg_data_t *pgdat)
366{
367 return(free_all_bootmem_core(pgdat));
368}
369
370unsigned long __init init_bootmem (unsigned long start, unsigned long pages)
371{
372 max_low_pfn = pages;
373 min_low_pfn = start;
374 return(init_bootmem_core(NODE_DATA(0), start, 0, pages));
375}
376
377#ifndef CONFIG_HAVE_ARCH_BOOTMEM_NODE
378void __init reserve_bootmem (unsigned long addr, unsigned long size)
379{
380 reserve_bootmem_core(NODE_DATA(0)->bdata, addr, size);
381}
382#endif /* !CONFIG_HAVE_ARCH_BOOTMEM_NODE */
383
384void __init free_bootmem (unsigned long addr, unsigned long size)
385{
386 free_bootmem_core(NODE_DATA(0)->bdata, addr, size);
387}
388
389unsigned long __init free_all_bootmem (void)
390{
391 return(free_all_bootmem_core(NODE_DATA(0)));
392}
393
Yasunori Goto281dd252005-10-19 15:52:18 -0700394void * __init __alloc_bootmem_limit (unsigned long size, unsigned long align, unsigned long goal,
395 unsigned long limit)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700396{
397 pg_data_t *pgdat = pgdat_list;
398 void *ptr;
399
400 for_each_pgdat(pgdat)
401 if ((ptr = __alloc_bootmem_core(pgdat->bdata, size,
Yasunori Goto281dd252005-10-19 15:52:18 -0700402 align, goal, limit)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700403 return(ptr);
404
405 /*
406 * Whoops, we cannot satisfy the allocation request.
407 */
408 printk(KERN_ALERT "bootmem alloc of %lu bytes failed!\n", size);
409 panic("Out of memory");
410 return NULL;
411}
412
Yasunori Goto281dd252005-10-19 15:52:18 -0700413
414void * __init __alloc_bootmem_node_limit (pg_data_t *pgdat, unsigned long size, unsigned long align,
415 unsigned long goal, unsigned long limit)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700416{
417 void *ptr;
418
Yasunori Goto281dd252005-10-19 15:52:18 -0700419 ptr = __alloc_bootmem_core(pgdat->bdata, size, align, goal, limit);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700420 if (ptr)
421 return (ptr);
422
Yasunori Goto281dd252005-10-19 15:52:18 -0700423 return __alloc_bootmem_limit(size, align, goal, limit);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700424}
425