2 * bootmem - A boot-time physical memory allocator and configurator
4 * Copyright (C) 1999 Ingo Molnar
5 * 1999 Kanoj Sarcar, SGI
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/slab.h>
14 #include <linux/bootmem.h>
15 #include <linux/export.h>
16 #include <linux/kmemleak.h>
17 #include <linux/range.h>
18 #include <linux/memblock.h>
22 #include <asm/processor.h>
26 #ifndef CONFIG_NEED_MULTIPLE_NODES
27 struct pglist_data __refdata contig_page_data = {
28 .bdata = &bootmem_node_data[0]
30 EXPORT_SYMBOL(contig_page_data);
33 unsigned long max_low_pfn;
34 unsigned long min_low_pfn;
35 unsigned long max_pfn;
37 bootmem_data_t bootmem_node_data[MAX_NUMNODES] __initdata;
39 static struct list_head bdata_list __initdata = LIST_HEAD_INIT(bdata_list);
41 static int bootmem_debug;
43 static int __init bootmem_debug_setup(char *buf)
48 early_param("bootmem_debug", bootmem_debug_setup);
50 #define bdebug(fmt, args...) ({ \
51 if (unlikely(bootmem_debug)) \
57 static unsigned long __init bootmap_bytes(unsigned long pages)
59 unsigned long bytes = DIV_ROUND_UP(pages, 8);
61 return ALIGN(bytes, sizeof(long));
65 * bootmem_bootmap_pages - calculate bitmap size in pages
66 * @pages: number of pages the bitmap has to represent
68 unsigned long __init bootmem_bootmap_pages(unsigned long pages)
70 unsigned long bytes = bootmap_bytes(pages);
72 return PAGE_ALIGN(bytes) >> PAGE_SHIFT;
78 static void __init link_bootmem(bootmem_data_t *bdata)
80 struct list_head *iter;
82 list_for_each(iter, &bdata_list) {
85 ent = list_entry(iter, bootmem_data_t, list);
86 if (bdata->node_min_pfn < ent->node_min_pfn)
89 list_add_tail(&bdata->list, iter);
93 * Called once to set up the allocator itself.
95 static unsigned long __init init_bootmem_core(bootmem_data_t *bdata,
96 unsigned long mapstart, unsigned long start, unsigned long end)
98 unsigned long mapsize;
100 mminit_validate_memmodel_limits(&start, &end);
101 bdata->node_bootmem_map = phys_to_virt(PFN_PHYS(mapstart));
102 bdata->node_min_pfn = start;
103 bdata->node_low_pfn = end;
107 * Initially all pages are reserved - setup_arch() has to
108 * register free RAM areas explicitly.
110 mapsize = bootmap_bytes(end - start);
111 memset(bdata->node_bootmem_map, 0xff, mapsize);
113 bdebug("nid=%td start=%lx map=%lx end=%lx mapsize=%lx\n",
114 bdata - bootmem_node_data, start, mapstart, end, mapsize);
120 * init_bootmem_node - register a node as boot memory
121 * @pgdat: node to register
122 * @freepfn: pfn where the bitmap for this node is to be placed
123 * @startpfn: first pfn on the node
124 * @endpfn: first pfn after the node
126 * Returns the number of bytes needed to hold the bitmap for this node.
128 unsigned long __init init_bootmem_node(pg_data_t *pgdat, unsigned long freepfn,
129 unsigned long startpfn, unsigned long endpfn)
131 return init_bootmem_core(pgdat->bdata, freepfn, startpfn, endpfn);
135 * init_bootmem - register boot memory
136 * @start: pfn where the bitmap is to be placed
137 * @pages: number of available physical pages
139 * Returns the number of bytes needed to hold the bitmap.
141 unsigned long __init init_bootmem(unsigned long start, unsigned long pages)
145 return init_bootmem_core(NODE_DATA(0)->bdata, start, 0, pages);
149 * free_bootmem_late - free bootmem pages directly to page allocator
150 * @addr: starting address of the range
151 * @size: size of the range in bytes
153 * This is only useful when the bootmem allocator has already been torn
154 * down, but we are still initializing the system. Pages are given directly
155 * to the page allocator, no bootmem metadata is updated because it is gone.
157 void __init free_bootmem_late(unsigned long addr, unsigned long size)
159 unsigned long cursor, end;
161 kmemleak_free_part(__va(addr), size);
163 cursor = PFN_UP(addr);
164 end = PFN_DOWN(addr + size);
166 for (; cursor < end; cursor++) {
167 __free_pages_bootmem(pfn_to_page(cursor), 0);
172 static unsigned long __init free_all_bootmem_core(bootmem_data_t *bdata)
176 unsigned long start, end, pages, count = 0;
178 if (!bdata->node_bootmem_map)
181 start = bdata->node_min_pfn;
182 end = bdata->node_low_pfn;
185 * If the start is aligned to the machines wordsize, we might
186 * be able to free pages in bulks of that order.
188 aligned = !(start & (BITS_PER_LONG - 1));
190 bdebug("nid=%td start=%lx end=%lx aligned=%d\n",
191 bdata - bootmem_node_data, start, end, aligned);
193 while (start < end) {
194 unsigned long *map, idx, vec;
196 map = bdata->node_bootmem_map;
197 idx = start - bdata->node_min_pfn;
198 vec = ~map[idx / BITS_PER_LONG];
200 if (aligned && vec == ~0UL) {
201 int order = ilog2(BITS_PER_LONG);
203 __free_pages_bootmem(pfn_to_page(start), order);
204 count += BITS_PER_LONG;
206 unsigned long off = 0;
208 while (vec && off < BITS_PER_LONG) {
210 page = pfn_to_page(start + off);
211 __free_pages_bootmem(page, 0);
218 start += BITS_PER_LONG;
221 page = virt_to_page(bdata->node_bootmem_map);
222 pages = bdata->node_low_pfn - bdata->node_min_pfn;
223 pages = bootmem_bootmap_pages(pages);
226 __free_pages_bootmem(page++, 0);
228 bdebug("nid=%td released=%lx\n", bdata - bootmem_node_data, count);
234 * free_all_bootmem_node - release a node's free pages to the buddy allocator
235 * @pgdat: node to be released
237 * Returns the number of pages actually released.
239 unsigned long __init free_all_bootmem_node(pg_data_t *pgdat)
241 register_page_bootmem_info_node(pgdat);
242 return free_all_bootmem_core(pgdat->bdata);
246 * free_all_bootmem - release free pages to the buddy allocator
248 * Returns the number of pages actually released.
250 unsigned long __init free_all_bootmem(void)
252 unsigned long total_pages = 0;
253 bootmem_data_t *bdata;
255 list_for_each_entry(bdata, &bdata_list, list)
256 total_pages += free_all_bootmem_core(bdata);
261 static void __init __free(bootmem_data_t *bdata,
262 unsigned long sidx, unsigned long eidx)
266 bdebug("nid=%td start=%lx end=%lx\n", bdata - bootmem_node_data,
267 sidx + bdata->node_min_pfn,
268 eidx + bdata->node_min_pfn);
270 if (bdata->hint_idx > sidx)
271 bdata->hint_idx = sidx;
273 for (idx = sidx; idx < eidx; idx++)
274 if (!test_and_clear_bit(idx, bdata->node_bootmem_map))
278 static int __init __reserve(bootmem_data_t *bdata, unsigned long sidx,
279 unsigned long eidx, int flags)
282 int exclusive = flags & BOOTMEM_EXCLUSIVE;
284 bdebug("nid=%td start=%lx end=%lx flags=%x\n",
285 bdata - bootmem_node_data,
286 sidx + bdata->node_min_pfn,
287 eidx + bdata->node_min_pfn,
290 for (idx = sidx; idx < eidx; idx++)
291 if (test_and_set_bit(idx, bdata->node_bootmem_map)) {
293 __free(bdata, sidx, idx);
296 bdebug("silent double reserve of PFN %lx\n",
297 idx + bdata->node_min_pfn);
302 static int __init mark_bootmem_node(bootmem_data_t *bdata,
303 unsigned long start, unsigned long end,
304 int reserve, int flags)
306 unsigned long sidx, eidx;
308 bdebug("nid=%td start=%lx end=%lx reserve=%d flags=%x\n",
309 bdata - bootmem_node_data, start, end, reserve, flags);
311 BUG_ON(start < bdata->node_min_pfn);
312 BUG_ON(end > bdata->node_low_pfn);
314 sidx = start - bdata->node_min_pfn;
315 eidx = end - bdata->node_min_pfn;
318 return __reserve(bdata, sidx, eidx, flags);
320 __free(bdata, sidx, eidx);
324 static int __init mark_bootmem(unsigned long start, unsigned long end,
325 int reserve, int flags)
328 bootmem_data_t *bdata;
331 list_for_each_entry(bdata, &bdata_list, list) {
335 if (pos < bdata->node_min_pfn ||
336 pos >= bdata->node_low_pfn) {
337 BUG_ON(pos != start);
341 max = min(bdata->node_low_pfn, end);
343 err = mark_bootmem_node(bdata, pos, max, reserve, flags);
344 if (reserve && err) {
345 mark_bootmem(start, pos, 0, 0);
351 pos = bdata->node_low_pfn;
357 * free_bootmem_node - mark a page range as usable
358 * @pgdat: node the range resides on
359 * @physaddr: starting address of the range
360 * @size: size of the range in bytes
362 * Partial pages will be considered reserved and left as they are.
364 * The range must reside completely on the specified node.
366 void __init free_bootmem_node(pg_data_t *pgdat, unsigned long physaddr,
369 unsigned long start, end;
371 kmemleak_free_part(__va(physaddr), size);
373 start = PFN_UP(physaddr);
374 end = PFN_DOWN(physaddr + size);
376 mark_bootmem_node(pgdat->bdata, start, end, 0, 0);
380 * free_bootmem - mark a page range as usable
381 * @addr: starting address of the range
382 * @size: size of the range in bytes
384 * Partial pages will be considered reserved and left as they are.
386 * The range must be contiguous but may span node boundaries.
388 void __init free_bootmem(unsigned long addr, unsigned long size)
390 unsigned long start, end;
392 kmemleak_free_part(__va(addr), size);
394 start = PFN_UP(addr);
395 end = PFN_DOWN(addr + size);
397 mark_bootmem(start, end, 0, 0);
401 * reserve_bootmem_node - mark a page range as reserved
402 * @pgdat: node the range resides on
403 * @physaddr: starting address of the range
404 * @size: size of the range in bytes
405 * @flags: reservation flags (see linux/bootmem.h)
407 * Partial pages will be reserved.
409 * The range must reside completely on the specified node.
411 int __init reserve_bootmem_node(pg_data_t *pgdat, unsigned long physaddr,
412 unsigned long size, int flags)
414 unsigned long start, end;
416 start = PFN_DOWN(physaddr);
417 end = PFN_UP(physaddr + size);
419 return mark_bootmem_node(pgdat->bdata, start, end, 1, flags);
423 * reserve_bootmem - mark a page range as usable
424 * @addr: starting address of the range
425 * @size: size of the range in bytes
426 * @flags: reservation flags (see linux/bootmem.h)
428 * Partial pages will be reserved.
430 * The range must be contiguous but may span node boundaries.
432 int __init reserve_bootmem(unsigned long addr, unsigned long size,
435 unsigned long start, end;
437 start = PFN_DOWN(addr);
438 end = PFN_UP(addr + size);
440 return mark_bootmem(start, end, 1, flags);
443 int __weak __init reserve_bootmem_generic(unsigned long phys, unsigned long len,
446 return reserve_bootmem(phys, len, flags);
449 static unsigned long __init align_idx(struct bootmem_data *bdata,
450 unsigned long idx, unsigned long step)
452 unsigned long base = bdata->node_min_pfn;
455 * Align the index with respect to the node start so that the
456 * combination of both satisfies the requested alignment.
459 return ALIGN(base + idx, step) - base;
462 static unsigned long __init align_off(struct bootmem_data *bdata,
463 unsigned long off, unsigned long align)
465 unsigned long base = PFN_PHYS(bdata->node_min_pfn);
467 /* Same as align_idx for byte offsets */
469 return ALIGN(base + off, align) - base;
472 static void * __init alloc_bootmem_core(struct bootmem_data *bdata,
473 unsigned long size, unsigned long align,
474 unsigned long goal, unsigned long limit)
476 unsigned long fallback = 0;
477 unsigned long min, max, start, sidx, midx, step;
479 bdebug("nid=%td size=%lx [%lu pages] align=%lx goal=%lx limit=%lx\n",
480 bdata - bootmem_node_data, size, PAGE_ALIGN(size) >> PAGE_SHIFT,
484 BUG_ON(align & (align - 1));
485 BUG_ON(limit && goal + size > limit);
487 if (!bdata->node_bootmem_map)
490 min = bdata->node_min_pfn;
491 max = bdata->node_low_pfn;
494 limit >>= PAGE_SHIFT;
496 if (limit && max > limit)
501 step = max(align >> PAGE_SHIFT, 1UL);
503 if (goal && min < goal && goal < max)
504 start = ALIGN(goal, step);
506 start = ALIGN(min, step);
508 sidx = start - bdata->node_min_pfn;
509 midx = max - bdata->node_min_pfn;
511 if (bdata->hint_idx > sidx) {
513 * Handle the valid case of sidx being zero and still
514 * catch the fallback below.
517 sidx = align_idx(bdata, bdata->hint_idx, step);
523 unsigned long eidx, i, start_off, end_off;
525 sidx = find_next_zero_bit(bdata->node_bootmem_map, midx, sidx);
526 sidx = align_idx(bdata, sidx, step);
527 eidx = sidx + PFN_UP(size);
529 if (sidx >= midx || eidx > midx)
532 for (i = sidx; i < eidx; i++)
533 if (test_bit(i, bdata->node_bootmem_map)) {
534 sidx = align_idx(bdata, i, step);
540 if (bdata->last_end_off & (PAGE_SIZE - 1) &&
541 PFN_DOWN(bdata->last_end_off) + 1 == sidx)
542 start_off = align_off(bdata, bdata->last_end_off, align);
544 start_off = PFN_PHYS(sidx);
546 merge = PFN_DOWN(start_off) < sidx;
547 end_off = start_off + size;
549 bdata->last_end_off = end_off;
550 bdata->hint_idx = PFN_UP(end_off);
553 * Reserve the area now:
555 if (__reserve(bdata, PFN_DOWN(start_off) + merge,
556 PFN_UP(end_off), BOOTMEM_EXCLUSIVE))
559 region = phys_to_virt(PFN_PHYS(bdata->node_min_pfn) +
561 memset(region, 0, size);
563 * The min_count is set to 0 so that bootmem allocated blocks
564 * are never reported as leaks.
566 kmemleak_alloc(region, size, 0, 0);
571 sidx = align_idx(bdata, fallback - 1, step);
579 static void * __init alloc_arch_preferred_bootmem(bootmem_data_t *bdata,
580 unsigned long size, unsigned long align,
581 unsigned long goal, unsigned long limit)
583 if (WARN_ON_ONCE(slab_is_available()))
584 return kzalloc(size, GFP_NOWAIT);
586 #ifdef CONFIG_HAVE_ARCH_BOOTMEM
588 bootmem_data_t *p_bdata;
590 p_bdata = bootmem_arch_preferred_node(bdata, size, align,
593 return alloc_bootmem_core(p_bdata, size, align,
600 static void * __init ___alloc_bootmem_nopanic(unsigned long size,
605 bootmem_data_t *bdata;
609 region = alloc_arch_preferred_bootmem(NULL, size, align, goal, limit);
613 list_for_each_entry(bdata, &bdata_list, list) {
614 if (goal && bdata->node_low_pfn <= PFN_DOWN(goal))
616 if (limit && bdata->node_min_pfn >= PFN_DOWN(limit))
619 region = alloc_bootmem_core(bdata, size, align, goal, limit);
633 * __alloc_bootmem_nopanic - allocate boot memory without panicking
634 * @size: size of the request in bytes
635 * @align: alignment of the region
636 * @goal: preferred starting address of the region
638 * The goal is dropped if it can not be satisfied and the allocation will
639 * fall back to memory below @goal.
641 * Allocation may happen on any node in the system.
643 * Returns NULL on failure.
645 void * __init __alloc_bootmem_nopanic(unsigned long size, unsigned long align,
648 unsigned long limit = 0;
650 return ___alloc_bootmem_nopanic(size, align, goal, limit);
653 static void * __init ___alloc_bootmem(unsigned long size, unsigned long align,
654 unsigned long goal, unsigned long limit)
656 void *mem = ___alloc_bootmem_nopanic(size, align, goal, limit);
661 * Whoops, we cannot satisfy the allocation request.
663 printk(KERN_ALERT "bootmem alloc of %lu bytes failed!\n", size);
664 panic("Out of memory");
669 * __alloc_bootmem - allocate boot memory
670 * @size: size of the request in bytes
671 * @align: alignment of the region
672 * @goal: preferred starting address of the region
674 * The goal is dropped if it can not be satisfied and the allocation will
675 * fall back to memory below @goal.
677 * Allocation may happen on any node in the system.
679 * The function panics if the request can not be satisfied.
681 void * __init __alloc_bootmem(unsigned long size, unsigned long align,
684 unsigned long limit = 0;
686 return ___alloc_bootmem(size, align, goal, limit);
689 static void * __init ___alloc_bootmem_node(bootmem_data_t *bdata,
690 unsigned long size, unsigned long align,
691 unsigned long goal, unsigned long limit)
695 ptr = alloc_arch_preferred_bootmem(bdata, size, align, goal, limit);
699 ptr = alloc_bootmem_core(bdata, size, align, goal, limit);
703 return ___alloc_bootmem(size, align, goal, limit);
707 * __alloc_bootmem_node - allocate boot memory from a specific node
708 * @pgdat: node to allocate from
709 * @size: size of the request in bytes
710 * @align: alignment of the region
711 * @goal: preferred starting address of the region
713 * The goal is dropped if it can not be satisfied and the allocation will
714 * fall back to memory below @goal.
716 * Allocation may fall back to any node in the system if the specified node
717 * can not hold the requested memory.
719 * The function panics if the request can not be satisfied.
721 void * __init __alloc_bootmem_node(pg_data_t *pgdat, unsigned long size,
722 unsigned long align, unsigned long goal)
724 if (WARN_ON_ONCE(slab_is_available()))
725 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
727 return ___alloc_bootmem_node(pgdat->bdata, size, align, goal, 0);
730 void * __init __alloc_bootmem_node_high(pg_data_t *pgdat, unsigned long size,
731 unsigned long align, unsigned long goal)
734 unsigned long end_pfn;
736 if (WARN_ON_ONCE(slab_is_available()))
737 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
739 /* update goal according ...MAX_DMA32_PFN */
740 end_pfn = pgdat->node_start_pfn + pgdat->node_spanned_pages;
742 if (end_pfn > MAX_DMA32_PFN + (128 >> (20 - PAGE_SHIFT)) &&
743 (goal >> PAGE_SHIFT) < MAX_DMA32_PFN) {
745 unsigned long new_goal;
747 new_goal = MAX_DMA32_PFN << PAGE_SHIFT;
748 ptr = alloc_bootmem_core(pgdat->bdata, size, align,
755 return __alloc_bootmem_node(pgdat, size, align, goal);
759 #ifdef CONFIG_SPARSEMEM
761 * alloc_bootmem_section - allocate boot memory from a specific section
762 * @size: size of the request in bytes
763 * @section_nr: sparse map section to allocate from
765 * Return NULL on failure.
767 void * __init alloc_bootmem_section(unsigned long size,
768 unsigned long section_nr)
770 bootmem_data_t *bdata;
771 unsigned long pfn, goal, limit;
773 pfn = section_nr_to_pfn(section_nr);
774 goal = pfn << PAGE_SHIFT;
775 limit = section_nr_to_pfn(section_nr + 1) << PAGE_SHIFT;
776 bdata = &bootmem_node_data[early_pfn_to_nid(pfn)];
778 return alloc_bootmem_core(bdata, size, SMP_CACHE_BYTES, goal, limit);
782 void * __init __alloc_bootmem_node_nopanic(pg_data_t *pgdat, unsigned long size,
783 unsigned long align, unsigned long goal)
787 if (WARN_ON_ONCE(slab_is_available()))
788 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
790 ptr = alloc_arch_preferred_bootmem(pgdat->bdata, size, align, goal, 0);
794 ptr = alloc_bootmem_core(pgdat->bdata, size, align, goal, 0);
798 return __alloc_bootmem_nopanic(size, align, goal);
801 #ifndef ARCH_LOW_ADDRESS_LIMIT
802 #define ARCH_LOW_ADDRESS_LIMIT 0xffffffffUL
806 * __alloc_bootmem_low - allocate low boot memory
807 * @size: size of the request in bytes
808 * @align: alignment of the region
809 * @goal: preferred starting address of the region
811 * The goal is dropped if it can not be satisfied and the allocation will
812 * fall back to memory below @goal.
814 * Allocation may happen on any node in the system.
816 * The function panics if the request can not be satisfied.
818 void * __init __alloc_bootmem_low(unsigned long size, unsigned long align,
821 return ___alloc_bootmem(size, align, goal, ARCH_LOW_ADDRESS_LIMIT);
825 * __alloc_bootmem_low_node - allocate low boot memory from a specific node
826 * @pgdat: node to allocate from
827 * @size: size of the request in bytes
828 * @align: alignment of the region
829 * @goal: preferred starting address of the region
831 * The goal is dropped if it can not be satisfied and the allocation will
832 * fall back to memory below @goal.
834 * Allocation may fall back to any node in the system if the specified node
835 * can not hold the requested memory.
837 * The function panics if the request can not be satisfied.
839 void * __init __alloc_bootmem_low_node(pg_data_t *pgdat, unsigned long size,
840 unsigned long align, unsigned long goal)
842 if (WARN_ON_ONCE(slab_is_available()))
843 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
845 return ___alloc_bootmem_node(pgdat->bdata, size, align,
846 goal, ARCH_LOW_ADDRESS_LIMIT);