bootmem.c 12.7 KB
Newer Older
L
Linus Torvalds 已提交
1 2 3 4 5 6 7 8 9 10 11
/*
 *  linux/mm/bootmem.c
 *
 *  Copyright (C) 1999 Ingo Molnar
 *  Discontiguous memory support, Kanoj Sarcar, SGI, Nov 1999
 *
 *  simple boot-time physical memory area allocator and
 *  free memory collector. It's used to deal with reserved
 *  system memory and memory holes as well.
 */
#include <linux/init.h>
12
#include <linux/pfn.h>
L
Linus Torvalds 已提交
13 14
#include <linux/bootmem.h>
#include <linux/module.h>
15 16

#include <asm/bug.h>
L
Linus Torvalds 已提交
17
#include <asm/io.h>
18
#include <asm/processor.h>
19

L
Linus Torvalds 已提交
20 21 22 23 24 25 26 27 28 29
#include "internal.h"

/*
 * Access to this subsystem has to be serialized externally. (this is
 * true for the boot process anyway)
 */
unsigned long max_low_pfn;
unsigned long min_low_pfn;
unsigned long max_pfn;

30
static LIST_HEAD(bdata_list);
31 32 33 34 35 36 37 38
#ifdef CONFIG_CRASH_DUMP
/*
 * If we have booted due to a crash, max_pfn will be a very low value. We need
 * to know the amount of memory that the previous kernel used.
 */
unsigned long saved_max_pfn;
#endif

L
Linus Torvalds 已提交
39
/* return the number of _pages_ that will be allocated for the boot bitmap */
40
unsigned long __init bootmem_bootmap_pages(unsigned long pages)
L
Linus Torvalds 已提交
41 42 43 44 45 46 47 48 49
{
	unsigned long mapsize;

	mapsize = (pages+7)/8;
	mapsize = (mapsize + ~PAGE_MASK) & PAGE_MASK;
	mapsize >>= PAGE_SHIFT;

	return mapsize;
}
50

51 52 53
/*
 * link bdata in order
 */
54
static void __init link_bootmem(bootmem_data_t *bdata)
55 56
{
	bootmem_data_t *ent;
57

58 59 60 61 62 63 64 65 66 67 68 69 70 71
	if (list_empty(&bdata_list)) {
		list_add(&bdata->list, &bdata_list);
		return;
	}
	/* insert in order */
	list_for_each_entry(ent, &bdata_list, list) {
		if (bdata->node_boot_start < ent->node_boot_start) {
			list_add_tail(&bdata->list, &ent->list);
			return;
		}
	}
	list_add_tail(&bdata->list, &bdata_list);
}

72 73 74 75 76 77 78 79 80 81 82 83
/*
 * Given an initialised bdata, it returns the size of the boot bitmap
 */
static unsigned long __init get_mapsize(bootmem_data_t *bdata)
{
	unsigned long mapsize;
	unsigned long start = PFN_DOWN(bdata->node_boot_start);
	unsigned long end = bdata->node_low_pfn;

	mapsize = ((end - start) + 7) / 8;
	return ALIGN(mapsize, sizeof(long));
}
L
Linus Torvalds 已提交
84 85 86 87

/*
 * Called once to set up the allocator itself.
 */
88
static unsigned long __init init_bootmem_core(pg_data_t *pgdat,
L
Linus Torvalds 已提交
89 90 91
	unsigned long mapstart, unsigned long start, unsigned long end)
{
	bootmem_data_t *bdata = pgdat->bdata;
92
	unsigned long mapsize;
L
Linus Torvalds 已提交
93

94 95
	bdata->node_bootmem_map = phys_to_virt(PFN_PHYS(mapstart));
	bdata->node_boot_start = PFN_PHYS(start);
L
Linus Torvalds 已提交
96
	bdata->node_low_pfn = end;
97
	link_bootmem(bdata);
L
Linus Torvalds 已提交
98 99 100 101 102

	/*
	 * Initially all pages are reserved - setup_arch() has to
	 * register free RAM areas explicitly.
	 */
103
	mapsize = get_mapsize(bdata);
L
Linus Torvalds 已提交
104 105 106 107 108 109 110 111 112 113
	memset(bdata->node_bootmem_map, 0xff, mapsize);

	return mapsize;
}

/*
 * Marks a particular physical memory range as unallocatable. Usable RAM
 * might be used for boot-time allocations - or it might get added
 * to the free page pool later on.
 */
114 115
static int __init reserve_bootmem_core(bootmem_data_t *bdata,
			unsigned long addr, unsigned long size, int flags)
L
Linus Torvalds 已提交
116
{
117
	unsigned long sidx, eidx;
L
Linus Torvalds 已提交
118
	unsigned long i;
119
	int ret;
120

L
Linus Torvalds 已提交
121 122 123 124 125
	/*
	 * round up, partially reserved pages are considered
	 * fully reserved.
	 */
	BUG_ON(!size);
126 127
	BUG_ON(PFN_DOWN(addr) >= bdata->node_low_pfn);
	BUG_ON(PFN_UP(addr + size) > bdata->node_low_pfn);
128
	BUG_ON(addr < bdata->node_boot_start);
129 130 131

	sidx = PFN_DOWN(addr - bdata->node_boot_start);
	eidx = PFN_UP(addr + size - bdata->node_boot_start);
L
Linus Torvalds 已提交
132 133 134 135 136 137

	for (i = sidx; i < eidx; i++)
		if (test_and_set_bit(i, bdata->node_bootmem_map)) {
#ifdef CONFIG_DEBUG_BOOTMEM
			printk("hm, page %08lx reserved twice.\n", i*PAGE_SIZE);
#endif
138 139 140 141
			if (flags & BOOTMEM_EXCLUSIVE) {
				ret = -EBUSY;
				goto err;
			}
L
Linus Torvalds 已提交
142
		}
143 144 145 146 147 148 149 150 151

	return 0;

err:
	/* unreserve memory we accidentally reserved */
	for (i--; i >= sidx; i--)
		clear_bit(i, bdata->node_bootmem_map);

	return ret;
L
Linus Torvalds 已提交
152 153
}

154 155
static void __init free_bootmem_core(bootmem_data_t *bdata, unsigned long addr,
				     unsigned long size)
L
Linus Torvalds 已提交
156
{
157
	unsigned long sidx, eidx;
L
Linus Torvalds 已提交
158
	unsigned long i;
159

160 161 162 163 164 165
	BUG_ON(!size);

	/* out range */
	if (addr + size < bdata->node_boot_start ||
		PFN_DOWN(addr) > bdata->node_low_pfn)
		return;
L
Linus Torvalds 已提交
166 167 168 169 170
	/*
	 * round down end of usable mem, partially free pages are
	 * considered reserved.
	 */

171
	if (addr >= bdata->node_boot_start && addr < bdata->last_success)
L
Linus Torvalds 已提交
172 173 174
		bdata->last_success = addr;

	/*
175
	 * Round up to index to the range.
L
Linus Torvalds 已提交
176
	 */
177 178 179 180 181
	if (PFN_UP(addr) > PFN_DOWN(bdata->node_boot_start))
		sidx = PFN_UP(addr) - PFN_DOWN(bdata->node_boot_start);
	else
		sidx = 0;

182
	eidx = PFN_DOWN(addr + size - bdata->node_boot_start);
183 184
	if (eidx > bdata->node_low_pfn - PFN_DOWN(bdata->node_boot_start))
		eidx = bdata->node_low_pfn - PFN_DOWN(bdata->node_boot_start);
L
Linus Torvalds 已提交
185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204

	for (i = sidx; i < eidx; i++) {
		if (unlikely(!test_and_clear_bit(i, bdata->node_bootmem_map)))
			BUG();
	}
}

/*
 * We 'merge' subsequent allocations to save space. We might 'lose'
 * some fraction of a page if allocations cannot be satisfied due to
 * size constraints on boxes where there is physical RAM space
 * fragmentation - in these cases (mostly large memory boxes) this
 * is not a problem.
 *
 * On low memory boxes we get it right in 100% of the cases.
 *
 * alignment has to be a power of 2 value.
 *
 * NOTE:  This function is _not_ reentrant.
 */
205
void * __init
L
Linus Torvalds 已提交
206
__alloc_bootmem_core(struct bootmem_data *bdata, unsigned long size,
207
	      unsigned long align, unsigned long goal, unsigned long limit)
L
Linus Torvalds 已提交
208 209
{
	unsigned long offset, remaining_size, areasize, preferred;
210
	unsigned long i, start = 0, incr, eidx, end_pfn;
L
Linus Torvalds 已提交
211 212
	void *ret;

213
	if (!size) {
L
Linus Torvalds 已提交
214 215 216 217 218
		printk("__alloc_bootmem_core(): zero-sized request\n");
		BUG();
	}
	BUG_ON(align & (align-1));

219 220 221
	if (limit && bdata->node_boot_start >= limit)
		return NULL;

222 223 224 225
	/* on nodes without memory - bootmem_map is NULL */
	if (!bdata->node_bootmem_map)
		return NULL;

226 227
	end_pfn = bdata->node_low_pfn;
	limit = PFN_DOWN(limit);
228 229 230
	if (limit && end_pfn > limit)
		end_pfn = limit;

231
	eidx = end_pfn - PFN_DOWN(bdata->node_boot_start);
L
Linus Torvalds 已提交
232
	offset = 0;
233 234 235
	if (align && (bdata->node_boot_start & (align - 1UL)) != 0)
		offset = align - (bdata->node_boot_start & (align - 1UL));
	offset = PFN_DOWN(offset);
L
Linus Torvalds 已提交
236 237 238 239 240

	/*
	 * We try to allocate bootmem pages above 'goal'
	 * first, then we try to allocate lower pages.
	 */
241
	if (goal && goal >= bdata->node_boot_start && PFN_DOWN(goal) < end_pfn) {
L
Linus Torvalds 已提交
242 243 244
		preferred = goal - bdata->node_boot_start;

		if (bdata->last_success >= preferred)
245 246
			if (!limit || (limit && limit > bdata->last_success))
				preferred = bdata->last_success;
L
Linus Torvalds 已提交
247 248 249
	} else
		preferred = 0;

250 251
	preferred = PFN_DOWN(ALIGN(preferred, align)) + offset;
	areasize = (size + PAGE_SIZE-1) / PAGE_SIZE;
L
Linus Torvalds 已提交
252 253 254 255 256 257 258
	incr = align >> PAGE_SHIFT ? : 1;

restart_scan:
	for (i = preferred; i < eidx; i += incr) {
		unsigned long j;
		i = find_next_zero_bit(bdata->node_bootmem_map, eidx, i);
		i = ALIGN(i, incr);
259 260
		if (i >= eidx)
			break;
L
Linus Torvalds 已提交
261 262 263 264 265
		if (test_bit(i, bdata->node_bootmem_map))
			continue;
		for (j = i + 1; j < i + areasize; ++j) {
			if (j >= eidx)
				goto fail_block;
266
			if (test_bit(j, bdata->node_bootmem_map))
L
Linus Torvalds 已提交
267 268 269 270 271 272 273 274 275 276 277 278 279 280 281
				goto fail_block;
		}
		start = i;
		goto found;
	fail_block:
		i = ALIGN(j, incr);
	}

	if (preferred > offset) {
		preferred = offset;
		goto restart_scan;
	}
	return NULL;

found:
282
	bdata->last_success = PFN_PHYS(start);
L
Linus Torvalds 已提交
283 284 285 286 287 288 289 290 291
	BUG_ON(start >= eidx);

	/*
	 * Is the next page of the previous allocation-end the start
	 * of this allocation's buffer? If yes then we can 'merge'
	 * the previous partial page with this allocation.
	 */
	if (align < PAGE_SIZE &&
	    bdata->last_offset && bdata->last_pos+1 == start) {
292
		offset = ALIGN(bdata->last_offset, align);
L
Linus Torvalds 已提交
293
		BUG_ON(offset > PAGE_SIZE);
294
		remaining_size = PAGE_SIZE - offset;
L
Linus Torvalds 已提交
295 296 297
		if (size < remaining_size) {
			areasize = 0;
			/* last_pos unchanged */
298 299 300 301
			bdata->last_offset = offset + size;
			ret = phys_to_virt(bdata->last_pos * PAGE_SIZE +
					   offset +
					   bdata->node_boot_start);
L
Linus Torvalds 已提交
302 303
		} else {
			remaining_size = size - remaining_size;
304 305 306 307 308
			areasize = (remaining_size + PAGE_SIZE-1) / PAGE_SIZE;
			ret = phys_to_virt(bdata->last_pos * PAGE_SIZE +
					   offset +
					   bdata->node_boot_start);
			bdata->last_pos = start + areasize - 1;
L
Linus Torvalds 已提交
309 310 311 312 313 314 315 316 317 318 319 320
			bdata->last_offset = remaining_size;
		}
		bdata->last_offset &= ~PAGE_MASK;
	} else {
		bdata->last_pos = start + areasize - 1;
		bdata->last_offset = size & ~PAGE_MASK;
		ret = phys_to_virt(start * PAGE_SIZE + bdata->node_boot_start);
	}

	/*
	 * Reserve the area now:
	 */
321
	for (i = start; i < start + areasize; i++)
L
Linus Torvalds 已提交
322 323 324 325 326 327 328 329 330
		if (unlikely(test_and_set_bit(i, bdata->node_bootmem_map)))
			BUG();
	memset(ret, 0, size);
	return ret;
}

static unsigned long __init free_all_bootmem_core(pg_data_t *pgdat)
{
	struct page *page;
A
Andy Whitcroft 已提交
331
	unsigned long pfn;
L
Linus Torvalds 已提交
332 333 334 335 336 337 338 339 340 341
	bootmem_data_t *bdata = pgdat->bdata;
	unsigned long i, count, total = 0;
	unsigned long idx;
	unsigned long *map; 
	int gofast = 0;

	BUG_ON(!bdata->node_bootmem_map);

	count = 0;
	/* first extant page of the node */
342 343
	pfn = PFN_DOWN(bdata->node_boot_start);
	idx = bdata->node_low_pfn - pfn;
L
Linus Torvalds 已提交
344 345 346 347 348 349 350
	map = bdata->node_bootmem_map;
	/* Check physaddr is O(LOG2(BITS_PER_LONG)) page aligned */
	if (bdata->node_boot_start == 0 ||
	    ffs(bdata->node_boot_start) - PAGE_SHIFT > ffs(BITS_PER_LONG))
		gofast = 1;
	for (i = 0; i < idx; ) {
		unsigned long v = ~map[i / BITS_PER_LONG];
A
Andy Whitcroft 已提交
351

L
Linus Torvalds 已提交
352
		if (gofast && v == ~0UL) {
353
			int order;
L
Linus Torvalds 已提交
354

A
Andy Whitcroft 已提交
355
			page = pfn_to_page(pfn);
L
Linus Torvalds 已提交
356 357
			count += BITS_PER_LONG;
			order = ffs(BITS_PER_LONG) - 1;
358
			__free_pages_bootmem(page, order);
L
Linus Torvalds 已提交
359 360 361 362
			i += BITS_PER_LONG;
			page += BITS_PER_LONG;
		} else if (v) {
			unsigned long m;
A
Andy Whitcroft 已提交
363 364

			page = pfn_to_page(pfn);
L
Linus Torvalds 已提交
365 366 367
			for (m = 1; m && i < idx; m<<=1, page++, i++) {
				if (v & m) {
					count++;
368
					__free_pages_bootmem(page, 0);
L
Linus Torvalds 已提交
369 370 371
				}
			}
		} else {
372
			i += BITS_PER_LONG;
L
Linus Torvalds 已提交
373
		}
A
Andy Whitcroft 已提交
374
		pfn += BITS_PER_LONG;
L
Linus Torvalds 已提交
375 376 377 378 379 380 381 382 383
	}
	total += count;

	/*
	 * Now free the allocator bitmap itself, it's not
	 * needed anymore:
	 */
	page = virt_to_page(bdata->node_bootmem_map);
	count = 0;
384 385
	idx = (get_mapsize(bdata) + PAGE_SIZE-1) >> PAGE_SHIFT;
	for (i = 0; i < idx; i++, page++) {
386
		__free_pages_bootmem(page, 0);
387
		count++;
L
Linus Torvalds 已提交
388 389 390 391 392 393 394
	}
	total += count;
	bdata->node_bootmem_map = NULL;

	return total;
}

395
unsigned long __init init_bootmem_node(pg_data_t *pgdat, unsigned long freepfn,
396
				unsigned long startpfn, unsigned long endpfn)
L
Linus Torvalds 已提交
397
{
398
	return init_bootmem_core(pgdat, freepfn, startpfn, endpfn);
L
Linus Torvalds 已提交
399 400
}

401
void __init reserve_bootmem_node(pg_data_t *pgdat, unsigned long physaddr,
402
				 unsigned long size, int flags)
L
Linus Torvalds 已提交
403
{
404
	reserve_bootmem_core(pgdat->bdata, physaddr, size, flags);
L
Linus Torvalds 已提交
405 406
}

407 408
void __init free_bootmem_node(pg_data_t *pgdat, unsigned long physaddr,
			      unsigned long size)
L
Linus Torvalds 已提交
409 410 411 412
{
	free_bootmem_core(pgdat->bdata, physaddr, size);
}

413
unsigned long __init free_all_bootmem_node(pg_data_t *pgdat)
L
Linus Torvalds 已提交
414
{
415
	return free_all_bootmem_core(pgdat);
L
Linus Torvalds 已提交
416 417
}

418
unsigned long __init init_bootmem(unsigned long start, unsigned long pages)
L
Linus Torvalds 已提交
419 420 421
{
	max_low_pfn = pages;
	min_low_pfn = start;
422
	return init_bootmem_core(NODE_DATA(0), start, 0, pages);
L
Linus Torvalds 已提交
423 424 425
}

#ifndef CONFIG_HAVE_ARCH_BOOTMEM_NODE
426 427
int __init reserve_bootmem(unsigned long addr, unsigned long size,
			    int flags)
L
Linus Torvalds 已提交
428
{
429
	return reserve_bootmem_core(NODE_DATA(0)->bdata, addr, size, flags);
L
Linus Torvalds 已提交
430 431 432
}
#endif /* !CONFIG_HAVE_ARCH_BOOTMEM_NODE */

433
void __init free_bootmem(unsigned long addr, unsigned long size)
L
Linus Torvalds 已提交
434
{
435 436 437
	bootmem_data_t *bdata;
	list_for_each_entry(bdata, &bdata_list, list)
		free_bootmem_core(bdata, addr, size);
L
Linus Torvalds 已提交
438 439
}

440
unsigned long __init free_all_bootmem(void)
L
Linus Torvalds 已提交
441
{
442
	return free_all_bootmem_core(NODE_DATA(0));
L
Linus Torvalds 已提交
443 444
}

445 446
void * __init __alloc_bootmem_nopanic(unsigned long size, unsigned long align,
				      unsigned long goal)
L
Linus Torvalds 已提交
447
{
448
	bootmem_data_t *bdata;
L
Linus Torvalds 已提交
449 450
	void *ptr;

451 452 453 454 455
	list_for_each_entry(bdata, &bdata_list, list) {
		ptr = __alloc_bootmem_core(bdata, size, align, goal, 0);
		if (ptr)
			return ptr;
	}
456 457
	return NULL;
}
L
Linus Torvalds 已提交
458

459 460
void * __init __alloc_bootmem(unsigned long size, unsigned long align,
			      unsigned long goal)
461 462
{
	void *mem = __alloc_bootmem_nopanic(size,align,goal);
463

464 465
	if (mem)
		return mem;
L
Linus Torvalds 已提交
466 467 468 469 470 471 472 473
	/*
	 * Whoops, we cannot satisfy the allocation request.
	 */
	printk(KERN_ALERT "bootmem alloc of %lu bytes failed!\n", size);
	panic("Out of memory");
	return NULL;
}

474

475 476
void * __init __alloc_bootmem_node(pg_data_t *pgdat, unsigned long size,
				   unsigned long align, unsigned long goal)
L
Linus Torvalds 已提交
477 478 479
{
	void *ptr;

480
	ptr = __alloc_bootmem_core(pgdat->bdata, size, align, goal, 0);
L
Linus Torvalds 已提交
481
	if (ptr)
482
		return ptr;
L
Linus Torvalds 已提交
483

484
	return __alloc_bootmem(size, align, goal);
L
Linus Torvalds 已提交
485 486
}

487 488 489
#ifndef ARCH_LOW_ADDRESS_LIMIT
#define ARCH_LOW_ADDRESS_LIMIT	0xffffffffUL
#endif
490

491 492
void * __init __alloc_bootmem_low(unsigned long size, unsigned long align,
				  unsigned long goal)
493
{
494
	bootmem_data_t *bdata;
495 496
	void *ptr;

497
	list_for_each_entry(bdata, &bdata_list, list) {
498 499
		ptr = __alloc_bootmem_core(bdata, size, align, goal,
						ARCH_LOW_ADDRESS_LIMIT);
500 501 502
		if (ptr)
			return ptr;
	}
503 504 505 506 507 508 509 510 511 512 513 514

	/*
	 * Whoops, we cannot satisfy the allocation request.
	 */
	printk(KERN_ALERT "low bootmem alloc of %lu bytes failed!\n", size);
	panic("Out of low memory");
	return NULL;
}

void * __init __alloc_bootmem_low_node(pg_data_t *pgdat, unsigned long size,
				       unsigned long align, unsigned long goal)
{
515 516
	return __alloc_bootmem_core(pgdat->bdata, size, align, goal,
				    ARCH_LOW_ADDRESS_LIMIT);
517
}