cgroups: introduce link_css_set() to remove duplicate code
[linux-2.6/verdex.git] / mm / bootmem.c
blob51a0ccf61e0e97284cf3be1d05d4f43feb293d93
1 /*
2 * bootmem - A boot-time physical memory allocator and configurator
4 * Copyright (C) 1999 Ingo Molnar
5 * 1999 Kanoj Sarcar, SGI
6 * 2008 Johannes Weiner
8 * Access to this subsystem has to be serialized externally (which is true
9 * for the boot process anyway).
11 #include <linux/init.h>
12 #include <linux/pfn.h>
13 #include <linux/bootmem.h>
14 #include <linux/module.h>
16 #include <asm/bug.h>
17 #include <asm/io.h>
18 #include <asm/processor.h>
20 #include "internal.h"
22 unsigned long max_low_pfn;
23 unsigned long min_low_pfn;
24 unsigned long max_pfn;
26 #ifdef CONFIG_CRASH_DUMP
28 * If we have booted due to a crash, max_pfn will be a very low value. We need
29 * to know the amount of memory that the previous kernel used.
31 unsigned long saved_max_pfn;
32 #endif
34 bootmem_data_t bootmem_node_data[MAX_NUMNODES] __initdata;
36 static struct list_head bdata_list __initdata = LIST_HEAD_INIT(bdata_list);
38 static int bootmem_debug;
40 static int __init bootmem_debug_setup(char *buf)
42 bootmem_debug = 1;
43 return 0;
45 early_param("bootmem_debug", bootmem_debug_setup);
47 #define bdebug(fmt, args...) ({ \
48 if (unlikely(bootmem_debug)) \
49 printk(KERN_INFO \
50 "bootmem::%s " fmt, \
51 __func__, ## args); \
54 static unsigned long __init bootmap_bytes(unsigned long pages)
56 unsigned long bytes = (pages + 7) / 8;
58 return ALIGN(bytes, sizeof(long));
61 /**
62 * bootmem_bootmap_pages - calculate bitmap size in pages
63 * @pages: number of pages the bitmap has to represent
65 unsigned long __init bootmem_bootmap_pages(unsigned long pages)
67 unsigned long bytes = bootmap_bytes(pages);
69 return PAGE_ALIGN(bytes) >> PAGE_SHIFT;
73 * link bdata in order
75 static void __init link_bootmem(bootmem_data_t *bdata)
77 struct list_head *iter;
79 list_for_each(iter, &bdata_list) {
80 bootmem_data_t *ent;
82 ent = list_entry(iter, bootmem_data_t, list);
83 if (bdata->node_min_pfn < ent->node_min_pfn)
84 break;
86 list_add_tail(&bdata->list, iter);
90 * Called once to set up the allocator itself.
92 static unsigned long __init init_bootmem_core(bootmem_data_t *bdata,
93 unsigned long mapstart, unsigned long start, unsigned long end)
95 unsigned long mapsize;
97 mminit_validate_memmodel_limits(&start, &end);
98 bdata->node_bootmem_map = phys_to_virt(PFN_PHYS(mapstart));
99 bdata->node_min_pfn = start;
100 bdata->node_low_pfn = end;
101 link_bootmem(bdata);
104 * Initially all pages are reserved - setup_arch() has to
105 * register free RAM areas explicitly.
107 mapsize = bootmap_bytes(end - start);
108 memset(bdata->node_bootmem_map, 0xff, mapsize);
110 bdebug("nid=%td start=%lx map=%lx end=%lx mapsize=%lx\n",
111 bdata - bootmem_node_data, start, mapstart, end, mapsize);
113 return mapsize;
117 * init_bootmem_node - register a node as boot memory
118 * @pgdat: node to register
119 * @freepfn: pfn where the bitmap for this node is to be placed
120 * @startpfn: first pfn on the node
121 * @endpfn: first pfn after the node
123 * Returns the number of bytes needed to hold the bitmap for this node.
125 unsigned long __init init_bootmem_node(pg_data_t *pgdat, unsigned long freepfn,
126 unsigned long startpfn, unsigned long endpfn)
128 return init_bootmem_core(pgdat->bdata, freepfn, startpfn, endpfn);
132 * init_bootmem - register boot memory
133 * @start: pfn where the bitmap is to be placed
134 * @pages: number of available physical pages
136 * Returns the number of bytes needed to hold the bitmap.
138 unsigned long __init init_bootmem(unsigned long start, unsigned long pages)
140 max_low_pfn = pages;
141 min_low_pfn = start;
142 return init_bootmem_core(NODE_DATA(0)->bdata, start, 0, pages);
145 static unsigned long __init free_all_bootmem_core(bootmem_data_t *bdata)
147 int aligned;
148 struct page *page;
149 unsigned long start, end, pages, count = 0;
151 if (!bdata->node_bootmem_map)
152 return 0;
154 start = bdata->node_min_pfn;
155 end = bdata->node_low_pfn;
158 * If the start is aligned to the machines wordsize, we might
159 * be able to free pages in bulks of that order.
161 aligned = !(start & (BITS_PER_LONG - 1));
163 bdebug("nid=%td start=%lx end=%lx aligned=%d\n",
164 bdata - bootmem_node_data, start, end, aligned);
166 while (start < end) {
167 unsigned long *map, idx, vec;
169 map = bdata->node_bootmem_map;
170 idx = start - bdata->node_min_pfn;
171 vec = ~map[idx / BITS_PER_LONG];
173 if (aligned && vec == ~0UL && start + BITS_PER_LONG < end) {
174 int order = ilog2(BITS_PER_LONG);
176 __free_pages_bootmem(pfn_to_page(start), order);
177 count += BITS_PER_LONG;
178 } else {
179 unsigned long off = 0;
181 while (vec && off < BITS_PER_LONG) {
182 if (vec & 1) {
183 page = pfn_to_page(start + off);
184 __free_pages_bootmem(page, 0);
185 count++;
187 vec >>= 1;
188 off++;
191 start += BITS_PER_LONG;
194 page = virt_to_page(bdata->node_bootmem_map);
195 pages = bdata->node_low_pfn - bdata->node_min_pfn;
196 pages = bootmem_bootmap_pages(pages);
197 count += pages;
198 while (pages--)
199 __free_pages_bootmem(page++, 0);
201 bdebug("nid=%td released=%lx\n", bdata - bootmem_node_data, count);
203 return count;
207 * free_all_bootmem_node - release a node's free pages to the buddy allocator
208 * @pgdat: node to be released
210 * Returns the number of pages actually released.
212 unsigned long __init free_all_bootmem_node(pg_data_t *pgdat)
214 register_page_bootmem_info_node(pgdat);
215 return free_all_bootmem_core(pgdat->bdata);
219 * free_all_bootmem - release free pages to the buddy allocator
221 * Returns the number of pages actually released.
223 unsigned long __init free_all_bootmem(void)
225 return free_all_bootmem_core(NODE_DATA(0)->bdata);
228 static void __init __free(bootmem_data_t *bdata,
229 unsigned long sidx, unsigned long eidx)
231 unsigned long idx;
233 bdebug("nid=%td start=%lx end=%lx\n", bdata - bootmem_node_data,
234 sidx + bdata->node_min_pfn,
235 eidx + bdata->node_min_pfn);
237 if (bdata->hint_idx > sidx)
238 bdata->hint_idx = sidx;
240 for (idx = sidx; idx < eidx; idx++)
241 if (!test_and_clear_bit(idx, bdata->node_bootmem_map))
242 BUG();
245 static int __init __reserve(bootmem_data_t *bdata, unsigned long sidx,
246 unsigned long eidx, int flags)
248 unsigned long idx;
249 int exclusive = flags & BOOTMEM_EXCLUSIVE;
251 bdebug("nid=%td start=%lx end=%lx flags=%x\n",
252 bdata - bootmem_node_data,
253 sidx + bdata->node_min_pfn,
254 eidx + bdata->node_min_pfn,
255 flags);
257 for (idx = sidx; idx < eidx; idx++)
258 if (test_and_set_bit(idx, bdata->node_bootmem_map)) {
259 if (exclusive) {
260 __free(bdata, sidx, idx);
261 return -EBUSY;
263 bdebug("silent double reserve of PFN %lx\n",
264 idx + bdata->node_min_pfn);
266 return 0;
269 static int __init mark_bootmem_node(bootmem_data_t *bdata,
270 unsigned long start, unsigned long end,
271 int reserve, int flags)
273 unsigned long sidx, eidx;
275 bdebug("nid=%td start=%lx end=%lx reserve=%d flags=%x\n",
276 bdata - bootmem_node_data, start, end, reserve, flags);
278 BUG_ON(start < bdata->node_min_pfn);
279 BUG_ON(end > bdata->node_low_pfn);
281 sidx = start - bdata->node_min_pfn;
282 eidx = end - bdata->node_min_pfn;
284 if (reserve)
285 return __reserve(bdata, sidx, eidx, flags);
286 else
287 __free(bdata, sidx, eidx);
288 return 0;
291 static int __init mark_bootmem(unsigned long start, unsigned long end,
292 int reserve, int flags)
294 unsigned long pos;
295 bootmem_data_t *bdata;
297 pos = start;
298 list_for_each_entry(bdata, &bdata_list, list) {
299 int err;
300 unsigned long max;
302 if (pos < bdata->node_min_pfn ||
303 pos >= bdata->node_low_pfn) {
304 BUG_ON(pos != start);
305 continue;
308 max = min(bdata->node_low_pfn, end);
310 err = mark_bootmem_node(bdata, pos, max, reserve, flags);
311 if (reserve && err) {
312 mark_bootmem(start, pos, 0, 0);
313 return err;
316 if (max == end)
317 return 0;
318 pos = bdata->node_low_pfn;
320 BUG();
324 * free_bootmem_node - mark a page range as usable
325 * @pgdat: node the range resides on
326 * @physaddr: starting address of the range
327 * @size: size of the range in bytes
329 * Partial pages will be considered reserved and left as they are.
331 * The range must reside completely on the specified node.
333 void __init free_bootmem_node(pg_data_t *pgdat, unsigned long physaddr,
334 unsigned long size)
336 unsigned long start, end;
338 start = PFN_UP(physaddr);
339 end = PFN_DOWN(physaddr + size);
341 mark_bootmem_node(pgdat->bdata, start, end, 0, 0);
345 * free_bootmem - mark a page range as usable
346 * @addr: starting address of the range
347 * @size: size of the range in bytes
349 * Partial pages will be considered reserved and left as they are.
351 * The range must be contiguous but may span node boundaries.
353 void __init free_bootmem(unsigned long addr, unsigned long size)
355 unsigned long start, end;
357 start = PFN_UP(addr);
358 end = PFN_DOWN(addr + size);
360 mark_bootmem(start, end, 0, 0);
364 * reserve_bootmem_node - mark a page range as reserved
365 * @pgdat: node the range resides on
366 * @physaddr: starting address of the range
367 * @size: size of the range in bytes
368 * @flags: reservation flags (see linux/bootmem.h)
370 * Partial pages will be reserved.
372 * The range must reside completely on the specified node.
374 int __init reserve_bootmem_node(pg_data_t *pgdat, unsigned long physaddr,
375 unsigned long size, int flags)
377 unsigned long start, end;
379 start = PFN_DOWN(physaddr);
380 end = PFN_UP(physaddr + size);
382 return mark_bootmem_node(pgdat->bdata, start, end, 1, flags);
385 #ifndef CONFIG_HAVE_ARCH_BOOTMEM_NODE
387 * reserve_bootmem - mark a page range as usable
388 * @addr: starting address of the range
389 * @size: size of the range in bytes
390 * @flags: reservation flags (see linux/bootmem.h)
392 * Partial pages will be reserved.
394 * The range must be contiguous but may span node boundaries.
396 int __init reserve_bootmem(unsigned long addr, unsigned long size,
397 int flags)
399 unsigned long start, end;
401 start = PFN_DOWN(addr);
402 end = PFN_UP(addr + size);
404 return mark_bootmem(start, end, 1, flags);
406 #endif /* !CONFIG_HAVE_ARCH_BOOTMEM_NODE */
408 static unsigned long align_idx(struct bootmem_data *bdata, unsigned long idx,
409 unsigned long step)
411 unsigned long base = bdata->node_min_pfn;
414 * Align the index with respect to the node start so that the
415 * combination of both satisfies the requested alignment.
418 return ALIGN(base + idx, step) - base;
421 static unsigned long align_off(struct bootmem_data *bdata, unsigned long off,
422 unsigned long align)
424 unsigned long base = PFN_PHYS(bdata->node_min_pfn);
426 /* Same as align_idx for byte offsets */
428 return ALIGN(base + off, align) - base;
431 static void * __init alloc_bootmem_core(struct bootmem_data *bdata,
432 unsigned long size, unsigned long align,
433 unsigned long goal, unsigned long limit)
435 unsigned long fallback = 0;
436 unsigned long min, max, start, sidx, midx, step;
438 bdebug("nid=%td size=%lx [%lu pages] align=%lx goal=%lx limit=%lx\n",
439 bdata - bootmem_node_data, size, PAGE_ALIGN(size) >> PAGE_SHIFT,
440 align, goal, limit);
442 BUG_ON(!size);
443 BUG_ON(align & (align - 1));
444 BUG_ON(limit && goal + size > limit);
446 if (!bdata->node_bootmem_map)
447 return NULL;
449 min = bdata->node_min_pfn;
450 max = bdata->node_low_pfn;
452 goal >>= PAGE_SHIFT;
453 limit >>= PAGE_SHIFT;
455 if (limit && max > limit)
456 max = limit;
457 if (max <= min)
458 return NULL;
460 step = max(align >> PAGE_SHIFT, 1UL);
462 if (goal && min < goal && goal < max)
463 start = ALIGN(goal, step);
464 else
465 start = ALIGN(min, step);
467 sidx = start - bdata->node_min_pfn;
468 midx = max - bdata->node_min_pfn;
470 if (bdata->hint_idx > sidx) {
472 * Handle the valid case of sidx being zero and still
473 * catch the fallback below.
475 fallback = sidx + 1;
476 sidx = align_idx(bdata, bdata->hint_idx, step);
479 while (1) {
480 int merge;
481 void *region;
482 unsigned long eidx, i, start_off, end_off;
483 find_block:
484 sidx = find_next_zero_bit(bdata->node_bootmem_map, midx, sidx);
485 sidx = align_idx(bdata, sidx, step);
486 eidx = sidx + PFN_UP(size);
488 if (sidx >= midx || eidx > midx)
489 break;
491 for (i = sidx; i < eidx; i++)
492 if (test_bit(i, bdata->node_bootmem_map)) {
493 sidx = align_idx(bdata, i, step);
494 if (sidx == i)
495 sidx += step;
496 goto find_block;
499 if (bdata->last_end_off & (PAGE_SIZE - 1) &&
500 PFN_DOWN(bdata->last_end_off) + 1 == sidx)
501 start_off = align_off(bdata, bdata->last_end_off, align);
502 else
503 start_off = PFN_PHYS(sidx);
505 merge = PFN_DOWN(start_off) < sidx;
506 end_off = start_off + size;
508 bdata->last_end_off = end_off;
509 bdata->hint_idx = PFN_UP(end_off);
512 * Reserve the area now:
514 if (__reserve(bdata, PFN_DOWN(start_off) + merge,
515 PFN_UP(end_off), BOOTMEM_EXCLUSIVE))
516 BUG();
518 region = phys_to_virt(PFN_PHYS(bdata->node_min_pfn) +
519 start_off);
520 memset(region, 0, size);
521 return region;
524 if (fallback) {
525 sidx = align_idx(bdata, fallback - 1, step);
526 fallback = 0;
527 goto find_block;
530 return NULL;
533 static void * __init ___alloc_bootmem_nopanic(unsigned long size,
534 unsigned long align,
535 unsigned long goal,
536 unsigned long limit)
538 bootmem_data_t *bdata;
540 restart:
541 list_for_each_entry(bdata, &bdata_list, list) {
542 void *region;
544 if (goal && bdata->node_low_pfn <= PFN_DOWN(goal))
545 continue;
546 if (limit && bdata->node_min_pfn >= PFN_DOWN(limit))
547 break;
549 region = alloc_bootmem_core(bdata, size, align, goal, limit);
550 if (region)
551 return region;
554 if (goal) {
555 goal = 0;
556 goto restart;
559 return NULL;
563 * __alloc_bootmem_nopanic - allocate boot memory without panicking
564 * @size: size of the request in bytes
565 * @align: alignment of the region
566 * @goal: preferred starting address of the region
568 * The goal is dropped if it can not be satisfied and the allocation will
569 * fall back to memory below @goal.
571 * Allocation may happen on any node in the system.
573 * Returns NULL on failure.
575 void * __init __alloc_bootmem_nopanic(unsigned long size, unsigned long align,
576 unsigned long goal)
578 return ___alloc_bootmem_nopanic(size, align, goal, 0);
581 static void * __init ___alloc_bootmem(unsigned long size, unsigned long align,
582 unsigned long goal, unsigned long limit)
584 void *mem = ___alloc_bootmem_nopanic(size, align, goal, limit);
586 if (mem)
587 return mem;
589 * Whoops, we cannot satisfy the allocation request.
591 printk(KERN_ALERT "bootmem alloc of %lu bytes failed!\n", size);
592 panic("Out of memory");
593 return NULL;
597 * __alloc_bootmem - allocate boot memory
598 * @size: size of the request in bytes
599 * @align: alignment of the region
600 * @goal: preferred starting address of the region
602 * The goal is dropped if it can not be satisfied and the allocation will
603 * fall back to memory below @goal.
605 * Allocation may happen on any node in the system.
607 * The function panics if the request can not be satisfied.
609 void * __init __alloc_bootmem(unsigned long size, unsigned long align,
610 unsigned long goal)
612 return ___alloc_bootmem(size, align, goal, 0);
615 static void * __init ___alloc_bootmem_node(bootmem_data_t *bdata,
616 unsigned long size, unsigned long align,
617 unsigned long goal, unsigned long limit)
619 void *ptr;
621 ptr = alloc_bootmem_core(bdata, size, align, goal, limit);
622 if (ptr)
623 return ptr;
625 return ___alloc_bootmem(size, align, goal, limit);
629 * __alloc_bootmem_node - allocate boot memory from a specific node
630 * @pgdat: node to allocate from
631 * @size: size of the request in bytes
632 * @align: alignment of the region
633 * @goal: preferred starting address of the region
635 * The goal is dropped if it can not be satisfied and the allocation will
636 * fall back to memory below @goal.
638 * Allocation may fall back to any node in the system if the specified node
639 * can not hold the requested memory.
641 * The function panics if the request can not be satisfied.
643 void * __init __alloc_bootmem_node(pg_data_t *pgdat, unsigned long size,
644 unsigned long align, unsigned long goal)
646 return ___alloc_bootmem_node(pgdat->bdata, size, align, goal, 0);
649 #ifdef CONFIG_SPARSEMEM
651 * alloc_bootmem_section - allocate boot memory from a specific section
652 * @size: size of the request in bytes
653 * @section_nr: sparse map section to allocate from
655 * Return NULL on failure.
657 void * __init alloc_bootmem_section(unsigned long size,
658 unsigned long section_nr)
660 bootmem_data_t *bdata;
661 unsigned long pfn, goal, limit;
663 pfn = section_nr_to_pfn(section_nr);
664 goal = pfn << PAGE_SHIFT;
665 limit = section_nr_to_pfn(section_nr + 1) << PAGE_SHIFT;
666 bdata = &bootmem_node_data[early_pfn_to_nid(pfn)];
668 return alloc_bootmem_core(bdata, size, SMP_CACHE_BYTES, goal, limit);
670 #endif
672 void * __init __alloc_bootmem_node_nopanic(pg_data_t *pgdat, unsigned long size,
673 unsigned long align, unsigned long goal)
675 void *ptr;
677 ptr = alloc_bootmem_core(pgdat->bdata, size, align, goal, 0);
678 if (ptr)
679 return ptr;
681 return __alloc_bootmem_nopanic(size, align, goal);
684 #ifndef ARCH_LOW_ADDRESS_LIMIT
685 #define ARCH_LOW_ADDRESS_LIMIT 0xffffffffUL
686 #endif
689 * __alloc_bootmem_low - allocate low boot memory
690 * @size: size of the request in bytes
691 * @align: alignment of the region
692 * @goal: preferred starting address of the region
694 * The goal is dropped if it can not be satisfied and the allocation will
695 * fall back to memory below @goal.
697 * Allocation may happen on any node in the system.
699 * The function panics if the request can not be satisfied.
701 void * __init __alloc_bootmem_low(unsigned long size, unsigned long align,
702 unsigned long goal)
704 return ___alloc_bootmem(size, align, goal, ARCH_LOW_ADDRESS_LIMIT);
708 * __alloc_bootmem_low_node - allocate low boot memory from a specific node
709 * @pgdat: node to allocate from
710 * @size: size of the request in bytes
711 * @align: alignment of the region
712 * @goal: preferred starting address of the region
714 * The goal is dropped if it can not be satisfied and the allocation will
715 * fall back to memory below @goal.
717 * Allocation may fall back to any node in the system if the specified node
718 * can not hold the requested memory.
720 * The function panics if the request can not be satisfied.
722 void * __init __alloc_bootmem_low_node(pg_data_t *pgdat, unsigned long size,
723 unsigned long align, unsigned long goal)
725 return ___alloc_bootmem_node(pgdat->bdata, size, align,
726 goal, ARCH_LOW_ADDRESS_LIMIT);