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)
175 unsigned long start, end, pages, count = 0;
177 if (!bdata->node_bootmem_map)
180 start = bdata->node_min_pfn;
181 end = bdata->node_low_pfn;
183 bdebug("nid=%td start=%lx end=%lx\n",
184 bdata - bootmem_node_data, start, end);
186 while (start < end) {
187 unsigned long *map, idx, vec;
189 map = bdata->node_bootmem_map;
190 idx = start - bdata->node_min_pfn;
191 vec = ~map[idx / BITS_PER_LONG];
193 * If we have a properly aligned and fully unreserved
194 * BITS_PER_LONG block of pages in front of us, free
197 if (IS_ALIGNED(start, BITS_PER_LONG) && vec == ~0UL) {
198 int order = ilog2(BITS_PER_LONG);
200 __free_pages_bootmem(pfn_to_page(start), order);
201 count += BITS_PER_LONG;
202 start += BITS_PER_LONG;
204 unsigned long off = 0;
206 while (vec && off < BITS_PER_LONG) {
208 page = pfn_to_page(start + off);
209 __free_pages_bootmem(page, 0);
215 start = ALIGN(start + 1, BITS_PER_LONG);
219 page = virt_to_page(bdata->node_bootmem_map);
220 pages = bdata->node_low_pfn - bdata->node_min_pfn;
221 pages = bootmem_bootmap_pages(pages);
224 __free_pages_bootmem(page++, 0);
226 bdebug("nid=%td released=%lx\n", bdata - bootmem_node_data, count);
232 * free_all_bootmem_node - release a node's free pages to the buddy allocator
233 * @pgdat: node to be released
235 * Returns the number of pages actually released.
237 unsigned long __init free_all_bootmem_node(pg_data_t *pgdat)
239 register_page_bootmem_info_node(pgdat);
240 return free_all_bootmem_core(pgdat->bdata);
244 * free_all_bootmem - release free pages to the buddy allocator
246 * Returns the number of pages actually released.
248 unsigned long __init free_all_bootmem(void)
250 unsigned long total_pages = 0;
251 bootmem_data_t *bdata;
253 list_for_each_entry(bdata, &bdata_list, list)
254 total_pages += free_all_bootmem_core(bdata);
259 static void __init __free(bootmem_data_t *bdata,
260 unsigned long sidx, unsigned long eidx)
264 bdebug("nid=%td start=%lx end=%lx\n", bdata - bootmem_node_data,
265 sidx + bdata->node_min_pfn,
266 eidx + bdata->node_min_pfn);
268 if (bdata->hint_idx > sidx)
269 bdata->hint_idx = sidx;
271 for (idx = sidx; idx < eidx; idx++)
272 if (!test_and_clear_bit(idx, bdata->node_bootmem_map))
276 static int __init __reserve(bootmem_data_t *bdata, unsigned long sidx,
277 unsigned long eidx, int flags)
280 int exclusive = flags & BOOTMEM_EXCLUSIVE;
282 bdebug("nid=%td start=%lx end=%lx flags=%x\n",
283 bdata - bootmem_node_data,
284 sidx + bdata->node_min_pfn,
285 eidx + bdata->node_min_pfn,
288 for (idx = sidx; idx < eidx; idx++)
289 if (test_and_set_bit(idx, bdata->node_bootmem_map)) {
291 __free(bdata, sidx, idx);
294 bdebug("silent double reserve of PFN %lx\n",
295 idx + bdata->node_min_pfn);
300 static int __init mark_bootmem_node(bootmem_data_t *bdata,
301 unsigned long start, unsigned long end,
302 int reserve, int flags)
304 unsigned long sidx, eidx;
306 bdebug("nid=%td start=%lx end=%lx reserve=%d flags=%x\n",
307 bdata - bootmem_node_data, start, end, reserve, flags);
309 BUG_ON(start < bdata->node_min_pfn);
310 BUG_ON(end > bdata->node_low_pfn);
312 sidx = start - bdata->node_min_pfn;
313 eidx = end - bdata->node_min_pfn;
316 return __reserve(bdata, sidx, eidx, flags);
318 __free(bdata, sidx, eidx);
322 static int __init mark_bootmem(unsigned long start, unsigned long end,
323 int reserve, int flags)
326 bootmem_data_t *bdata;
329 list_for_each_entry(bdata, &bdata_list, list) {
333 if (pos < bdata->node_min_pfn ||
334 pos >= bdata->node_low_pfn) {
335 BUG_ON(pos != start);
339 max = min(bdata->node_low_pfn, end);
341 err = mark_bootmem_node(bdata, pos, max, reserve, flags);
342 if (reserve && err) {
343 mark_bootmem(start, pos, 0, 0);
349 pos = bdata->node_low_pfn;
355 * free_bootmem_node - mark a page range as usable
356 * @pgdat: node the range resides on
357 * @physaddr: starting address of the range
358 * @size: size of the range in bytes
360 * Partial pages will be considered reserved and left as they are.
362 * The range must reside completely on the specified node.
364 void __init free_bootmem_node(pg_data_t *pgdat, unsigned long physaddr,
367 unsigned long start, end;
369 kmemleak_free_part(__va(physaddr), size);
371 start = PFN_UP(physaddr);
372 end = PFN_DOWN(physaddr + size);
374 mark_bootmem_node(pgdat->bdata, start, end, 0, 0);
378 * free_bootmem - mark a page range as usable
379 * @addr: starting address of the range
380 * @size: size of the range in bytes
382 * Partial pages will be considered reserved and left as they are.
384 * The range must be contiguous but may span node boundaries.
386 void __init free_bootmem(unsigned long addr, unsigned long size)
388 unsigned long start, end;
390 kmemleak_free_part(__va(addr), size);
392 start = PFN_UP(addr);
393 end = PFN_DOWN(addr + size);
395 mark_bootmem(start, end, 0, 0);
399 * reserve_bootmem_node - mark a page range as reserved
400 * @pgdat: node the range resides on
401 * @physaddr: starting address of the range
402 * @size: size of the range in bytes
403 * @flags: reservation flags (see linux/bootmem.h)
405 * Partial pages will be reserved.
407 * The range must reside completely on the specified node.
409 int __init reserve_bootmem_node(pg_data_t *pgdat, unsigned long physaddr,
410 unsigned long size, int flags)
412 unsigned long start, end;
414 start = PFN_DOWN(physaddr);
415 end = PFN_UP(physaddr + size);
417 return mark_bootmem_node(pgdat->bdata, start, end, 1, flags);
421 * reserve_bootmem - mark a page range as usable
422 * @addr: starting address of the range
423 * @size: size of the range in bytes
424 * @flags: reservation flags (see linux/bootmem.h)
426 * Partial pages will be reserved.
428 * The range must be contiguous but may span node boundaries.
430 int __init reserve_bootmem(unsigned long addr, unsigned long size,
433 unsigned long start, end;
435 start = PFN_DOWN(addr);
436 end = PFN_UP(addr + size);
438 return mark_bootmem(start, end, 1, flags);
441 int __weak __init reserve_bootmem_generic(unsigned long phys, unsigned long len,
444 return reserve_bootmem(phys, len, flags);
447 static unsigned long __init align_idx(struct bootmem_data *bdata,
448 unsigned long idx, unsigned long step)
450 unsigned long base = bdata->node_min_pfn;
453 * Align the index with respect to the node start so that the
454 * combination of both satisfies the requested alignment.
457 return ALIGN(base + idx, step) - base;
460 static unsigned long __init align_off(struct bootmem_data *bdata,
461 unsigned long off, unsigned long align)
463 unsigned long base = PFN_PHYS(bdata->node_min_pfn);
465 /* Same as align_idx for byte offsets */
467 return ALIGN(base + off, align) - base;
470 static void * __init alloc_bootmem_core(struct bootmem_data *bdata,
471 unsigned long size, unsigned long align,
472 unsigned long goal, unsigned long limit)
474 unsigned long fallback = 0;
475 unsigned long min, max, start, sidx, midx, step;
477 bdebug("nid=%td size=%lx [%lu pages] align=%lx goal=%lx limit=%lx\n",
478 bdata - bootmem_node_data, size, PAGE_ALIGN(size) >> PAGE_SHIFT,
482 BUG_ON(align & (align - 1));
483 BUG_ON(limit && goal + size > limit);
485 if (!bdata->node_bootmem_map)
488 min = bdata->node_min_pfn;
489 max = bdata->node_low_pfn;
492 limit >>= PAGE_SHIFT;
494 if (limit && max > limit)
499 step = max(align >> PAGE_SHIFT, 1UL);
501 if (goal && min < goal && goal < max)
502 start = ALIGN(goal, step);
504 start = ALIGN(min, step);
506 sidx = start - bdata->node_min_pfn;
507 midx = max - bdata->node_min_pfn;
509 if (bdata->hint_idx > sidx) {
511 * Handle the valid case of sidx being zero and still
512 * catch the fallback below.
515 sidx = align_idx(bdata, bdata->hint_idx, step);
521 unsigned long eidx, i, start_off, end_off;
523 sidx = find_next_zero_bit(bdata->node_bootmem_map, midx, sidx);
524 sidx = align_idx(bdata, sidx, step);
525 eidx = sidx + PFN_UP(size);
527 if (sidx >= midx || eidx > midx)
530 for (i = sidx; i < eidx; i++)
531 if (test_bit(i, bdata->node_bootmem_map)) {
532 sidx = align_idx(bdata, i, step);
538 if (bdata->last_end_off & (PAGE_SIZE - 1) &&
539 PFN_DOWN(bdata->last_end_off) + 1 == sidx)
540 start_off = align_off(bdata, bdata->last_end_off, align);
542 start_off = PFN_PHYS(sidx);
544 merge = PFN_DOWN(start_off) < sidx;
545 end_off = start_off + size;
547 bdata->last_end_off = end_off;
548 bdata->hint_idx = PFN_UP(end_off);
551 * Reserve the area now:
553 if (__reserve(bdata, PFN_DOWN(start_off) + merge,
554 PFN_UP(end_off), BOOTMEM_EXCLUSIVE))
557 region = phys_to_virt(PFN_PHYS(bdata->node_min_pfn) +
559 memset(region, 0, size);
561 * The min_count is set to 0 so that bootmem allocated blocks
562 * are never reported as leaks.
564 kmemleak_alloc(region, size, 0, 0);
569 sidx = align_idx(bdata, fallback - 1, step);
577 static void * __init alloc_arch_preferred_bootmem(bootmem_data_t *bdata,
578 unsigned long size, unsigned long align,
579 unsigned long goal, unsigned long limit)
581 if (WARN_ON_ONCE(slab_is_available()))
582 return kzalloc(size, GFP_NOWAIT);
584 #ifdef CONFIG_HAVE_ARCH_BOOTMEM
586 bootmem_data_t *p_bdata;
588 p_bdata = bootmem_arch_preferred_node(bdata, size, align,
591 return alloc_bootmem_core(p_bdata, size, align,
598 static void * __init ___alloc_bootmem_nopanic(unsigned long size,
603 bootmem_data_t *bdata;
607 region = alloc_arch_preferred_bootmem(NULL, size, align, goal, limit);
611 list_for_each_entry(bdata, &bdata_list, list) {
612 if (goal && bdata->node_low_pfn <= PFN_DOWN(goal))
614 if (limit && bdata->node_min_pfn >= PFN_DOWN(limit))
617 region = alloc_bootmem_core(bdata, size, align, goal, limit);
631 * __alloc_bootmem_nopanic - allocate boot memory without panicking
632 * @size: size of the request in bytes
633 * @align: alignment of the region
634 * @goal: preferred starting address of the region
636 * The goal is dropped if it can not be satisfied and the allocation will
637 * fall back to memory below @goal.
639 * Allocation may happen on any node in the system.
641 * Returns NULL on failure.
643 void * __init __alloc_bootmem_nopanic(unsigned long size, unsigned long align,
646 unsigned long limit = 0;
648 return ___alloc_bootmem_nopanic(size, align, goal, limit);
651 static void * __init ___alloc_bootmem(unsigned long size, unsigned long align,
652 unsigned long goal, unsigned long limit)
654 void *mem = ___alloc_bootmem_nopanic(size, align, goal, limit);
659 * Whoops, we cannot satisfy the allocation request.
661 printk(KERN_ALERT "bootmem alloc of %lu bytes failed!\n", size);
662 panic("Out of memory");
667 * __alloc_bootmem - allocate boot memory
668 * @size: size of the request in bytes
669 * @align: alignment of the region
670 * @goal: preferred starting address of the region
672 * The goal is dropped if it can not be satisfied and the allocation will
673 * fall back to memory below @goal.
675 * Allocation may happen on any node in the system.
677 * The function panics if the request can not be satisfied.
679 void * __init __alloc_bootmem(unsigned long size, unsigned long align,
682 unsigned long limit = 0;
684 return ___alloc_bootmem(size, align, goal, limit);
687 static void * __init ___alloc_bootmem_node(bootmem_data_t *bdata,
688 unsigned long size, unsigned long align,
689 unsigned long goal, unsigned long limit)
693 ptr = alloc_arch_preferred_bootmem(bdata, size, align, goal, limit);
697 ptr = alloc_bootmem_core(bdata, size, align, goal, limit);
701 return ___alloc_bootmem(size, align, goal, limit);
705 * __alloc_bootmem_node - allocate boot memory from a specific node
706 * @pgdat: node to allocate from
707 * @size: size of the request in bytes
708 * @align: alignment of the region
709 * @goal: preferred starting address of the region
711 * The goal is dropped if it can not be satisfied and the allocation will
712 * fall back to memory below @goal.
714 * Allocation may fall back to any node in the system if the specified node
715 * can not hold the requested memory.
717 * The function panics if the request can not be satisfied.
719 void * __init __alloc_bootmem_node(pg_data_t *pgdat, unsigned long size,
720 unsigned long align, unsigned long goal)
722 if (WARN_ON_ONCE(slab_is_available()))
723 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
725 return ___alloc_bootmem_node(pgdat->bdata, size, align, goal, 0);
728 void * __init __alloc_bootmem_node_high(pg_data_t *pgdat, unsigned long size,
729 unsigned long align, unsigned long goal)
732 unsigned long end_pfn;
734 if (WARN_ON_ONCE(slab_is_available()))
735 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
737 /* update goal according ...MAX_DMA32_PFN */
738 end_pfn = pgdat->node_start_pfn + pgdat->node_spanned_pages;
740 if (end_pfn > MAX_DMA32_PFN + (128 >> (20 - PAGE_SHIFT)) &&
741 (goal >> PAGE_SHIFT) < MAX_DMA32_PFN) {
743 unsigned long new_goal;
745 new_goal = MAX_DMA32_PFN << PAGE_SHIFT;
746 ptr = alloc_bootmem_core(pgdat->bdata, size, align,
753 return __alloc_bootmem_node(pgdat, size, align, goal);
757 #ifdef CONFIG_SPARSEMEM
759 * alloc_bootmem_section - allocate boot memory from a specific section
760 * @size: size of the request in bytes
761 * @section_nr: sparse map section to allocate from
763 * Return NULL on failure.
765 void * __init alloc_bootmem_section(unsigned long size,
766 unsigned long section_nr)
768 bootmem_data_t *bdata;
769 unsigned long pfn, goal, limit;
771 pfn = section_nr_to_pfn(section_nr);
772 goal = pfn << PAGE_SHIFT;
773 limit = section_nr_to_pfn(section_nr + 1) << PAGE_SHIFT;
774 bdata = &bootmem_node_data[early_pfn_to_nid(pfn)];
776 return alloc_bootmem_core(bdata, size, SMP_CACHE_BYTES, goal, limit);
780 void * __init __alloc_bootmem_node_nopanic(pg_data_t *pgdat, unsigned long size,
781 unsigned long align, unsigned long goal)
785 if (WARN_ON_ONCE(slab_is_available()))
786 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
788 ptr = alloc_arch_preferred_bootmem(pgdat->bdata, size, align, goal, 0);
792 ptr = alloc_bootmem_core(pgdat->bdata, size, align, goal, 0);
796 return __alloc_bootmem_nopanic(size, align, goal);
799 #ifndef ARCH_LOW_ADDRESS_LIMIT
800 #define ARCH_LOW_ADDRESS_LIMIT 0xffffffffUL
804 * __alloc_bootmem_low - allocate low boot memory
805 * @size: size of the request in bytes
806 * @align: alignment of the region
807 * @goal: preferred starting address of the region
809 * The goal is dropped if it can not be satisfied and the allocation will
810 * fall back to memory below @goal.
812 * Allocation may happen on any node in the system.
814 * The function panics if the request can not be satisfied.
816 void * __init __alloc_bootmem_low(unsigned long size, unsigned long align,
819 return ___alloc_bootmem(size, align, goal, ARCH_LOW_ADDRESS_LIMIT);
823 * __alloc_bootmem_low_node - allocate low boot memory from a specific node
824 * @pgdat: node to allocate from
825 * @size: size of the request in bytes
826 * @align: alignment of the region
827 * @goal: preferred starting address of the region
829 * The goal is dropped if it can not be satisfied and the allocation will
830 * fall back to memory below @goal.
832 * Allocation may fall back to any node in the system if the specified node
833 * can not hold the requested memory.
835 * The function panics if the request can not be satisfied.
837 void * __init __alloc_bootmem_low_node(pg_data_t *pgdat, unsigned long size,
838 unsigned long align, unsigned long goal)
840 if (WARN_ON_ONCE(slab_is_available()))
841 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
843 return ___alloc_bootmem_node(pgdat->bdata, size, align,
844 goal, ARCH_LOW_ADDRESS_LIMIT);