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>
19 #include <linux/bug.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)
82 list_for_each_entry(ent, &bdata_list, list) {
83 if (bdata->node_min_pfn < ent->node_min_pfn) {
84 list_add_tail(&bdata->list, &ent->list);
89 list_add_tail(&bdata->list, &bdata_list);
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 physical 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 physaddr, unsigned long size)
159 unsigned long cursor, end;
161 kmemleak_free_part(__va(physaddr), size);
163 cursor = PFN_UP(physaddr);
164 end = PFN_DOWN(physaddr + size);
166 for (; cursor < end; cursor++) {
167 __free_pages_bootmem(pfn_to_page(cursor), cursor, 0);
172 static unsigned long __init free_all_bootmem_core(bootmem_data_t *bdata)
175 unsigned long *map, start, end, pages, cur, count = 0;
177 if (!bdata->node_bootmem_map)
180 map = bdata->node_bootmem_map;
181 start = bdata->node_min_pfn;
182 end = bdata->node_low_pfn;
184 bdebug("nid=%td start=%lx end=%lx\n",
185 bdata - bootmem_node_data, start, end);
187 while (start < end) {
188 unsigned long idx, vec;
191 idx = start - bdata->node_min_pfn;
192 shift = idx & (BITS_PER_LONG - 1);
194 * vec holds at most BITS_PER_LONG map bits,
195 * bit 0 corresponds to start.
197 vec = ~map[idx / BITS_PER_LONG];
201 if (end - start >= BITS_PER_LONG)
202 vec |= ~map[idx / BITS_PER_LONG + 1] <<
203 (BITS_PER_LONG - shift);
206 * If we have a properly aligned and fully unreserved
207 * BITS_PER_LONG block of pages in front of us, free
210 if (IS_ALIGNED(start, BITS_PER_LONG) && vec == ~0UL) {
211 int order = ilog2(BITS_PER_LONG);
213 __free_pages_bootmem(pfn_to_page(start), start, order);
214 count += BITS_PER_LONG;
215 start += BITS_PER_LONG;
219 start = ALIGN(start + 1, BITS_PER_LONG);
220 while (vec && cur != start) {
222 page = pfn_to_page(cur);
223 __free_pages_bootmem(page, cur, 0);
232 cur = bdata->node_min_pfn;
233 page = virt_to_page(bdata->node_bootmem_map);
234 pages = bdata->node_low_pfn - bdata->node_min_pfn;
235 pages = bootmem_bootmap_pages(pages);
238 __free_pages_bootmem(page++, cur++, 0);
239 bdata->node_bootmem_map = NULL;
241 bdebug("nid=%td released=%lx\n", bdata - bootmem_node_data, count);
246 static int reset_managed_pages_done __initdata;
248 void reset_node_managed_pages(pg_data_t *pgdat)
252 for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++)
253 z->managed_pages = 0;
256 void __init reset_all_zones_managed_pages(void)
258 struct pglist_data *pgdat;
260 if (reset_managed_pages_done)
263 for_each_online_pgdat(pgdat)
264 reset_node_managed_pages(pgdat);
266 reset_managed_pages_done = 1;
270 * free_all_bootmem - release free pages to the buddy allocator
272 * Returns the number of pages actually released.
274 unsigned long __init free_all_bootmem(void)
276 unsigned long total_pages = 0;
277 bootmem_data_t *bdata;
279 reset_all_zones_managed_pages();
281 list_for_each_entry(bdata, &bdata_list, list)
282 total_pages += free_all_bootmem_core(bdata);
284 totalram_pages += total_pages;
289 static void __init __free(bootmem_data_t *bdata,
290 unsigned long sidx, unsigned long eidx)
294 bdebug("nid=%td start=%lx end=%lx\n", bdata - bootmem_node_data,
295 sidx + bdata->node_min_pfn,
296 eidx + bdata->node_min_pfn);
298 if (WARN_ON(bdata->node_bootmem_map == NULL))
301 if (bdata->hint_idx > sidx)
302 bdata->hint_idx = sidx;
304 for (idx = sidx; idx < eidx; idx++)
305 if (!test_and_clear_bit(idx, bdata->node_bootmem_map))
309 static int __init __reserve(bootmem_data_t *bdata, unsigned long sidx,
310 unsigned long eidx, int flags)
313 int exclusive = flags & BOOTMEM_EXCLUSIVE;
315 bdebug("nid=%td start=%lx end=%lx flags=%x\n",
316 bdata - bootmem_node_data,
317 sidx + bdata->node_min_pfn,
318 eidx + bdata->node_min_pfn,
321 if (WARN_ON(bdata->node_bootmem_map == NULL))
324 for (idx = sidx; idx < eidx; idx++)
325 if (test_and_set_bit(idx, bdata->node_bootmem_map)) {
327 __free(bdata, sidx, idx);
330 bdebug("silent double reserve of PFN %lx\n",
331 idx + bdata->node_min_pfn);
336 static int __init mark_bootmem_node(bootmem_data_t *bdata,
337 unsigned long start, unsigned long end,
338 int reserve, int flags)
340 unsigned long sidx, eidx;
342 bdebug("nid=%td start=%lx end=%lx reserve=%d flags=%x\n",
343 bdata - bootmem_node_data, start, end, reserve, flags);
345 BUG_ON(start < bdata->node_min_pfn);
346 BUG_ON(end > bdata->node_low_pfn);
348 sidx = start - bdata->node_min_pfn;
349 eidx = end - bdata->node_min_pfn;
352 return __reserve(bdata, sidx, eidx, flags);
354 __free(bdata, sidx, eidx);
358 static int __init mark_bootmem(unsigned long start, unsigned long end,
359 int reserve, int flags)
362 bootmem_data_t *bdata;
365 list_for_each_entry(bdata, &bdata_list, list) {
369 if (pos < bdata->node_min_pfn ||
370 pos >= bdata->node_low_pfn) {
371 BUG_ON(pos != start);
375 max = min(bdata->node_low_pfn, end);
377 err = mark_bootmem_node(bdata, pos, max, reserve, flags);
378 if (reserve && err) {
379 mark_bootmem(start, pos, 0, 0);
385 pos = bdata->node_low_pfn;
391 * free_bootmem_node - mark a page range as usable
392 * @pgdat: node the range resides on
393 * @physaddr: starting address of the range
394 * @size: size of the range in bytes
396 * Partial pages will be considered reserved and left as they are.
398 * The range must reside completely on the specified node.
400 void __init free_bootmem_node(pg_data_t *pgdat, unsigned long physaddr,
403 unsigned long start, end;
405 kmemleak_free_part(__va(physaddr), size);
407 start = PFN_UP(physaddr);
408 end = PFN_DOWN(physaddr + size);
410 mark_bootmem_node(pgdat->bdata, start, end, 0, 0);
414 * free_bootmem - mark a page range as usable
415 * @addr: starting physical address of the range
416 * @size: size of the range in bytes
418 * Partial pages will be considered reserved and left as they are.
420 * The range must be contiguous but may span node boundaries.
422 void __init free_bootmem(unsigned long physaddr, unsigned long size)
424 unsigned long start, end;
426 kmemleak_free_part(__va(physaddr), size);
428 start = PFN_UP(physaddr);
429 end = PFN_DOWN(physaddr + size);
431 mark_bootmem(start, end, 0, 0);
435 * reserve_bootmem_node - mark a page range as reserved
436 * @pgdat: node the range resides on
437 * @physaddr: starting address of the range
438 * @size: size of the range in bytes
439 * @flags: reservation flags (see linux/bootmem.h)
441 * Partial pages will be reserved.
443 * The range must reside completely on the specified node.
445 int __init reserve_bootmem_node(pg_data_t *pgdat, unsigned long physaddr,
446 unsigned long size, int flags)
448 unsigned long start, end;
450 start = PFN_DOWN(physaddr);
451 end = PFN_UP(physaddr + size);
453 return mark_bootmem_node(pgdat->bdata, start, end, 1, flags);
457 * reserve_bootmem - mark a page range as reserved
458 * @addr: starting address of the range
459 * @size: size of the range in bytes
460 * @flags: reservation flags (see linux/bootmem.h)
462 * Partial pages will be reserved.
464 * The range must be contiguous but may span node boundaries.
466 int __init reserve_bootmem(unsigned long addr, unsigned long size,
469 unsigned long start, end;
471 start = PFN_DOWN(addr);
472 end = PFN_UP(addr + size);
474 return mark_bootmem(start, end, 1, flags);
477 static unsigned long __init align_idx(struct bootmem_data *bdata,
478 unsigned long idx, unsigned long step)
480 unsigned long base = bdata->node_min_pfn;
483 * Align the index with respect to the node start so that the
484 * combination of both satisfies the requested alignment.
487 return ALIGN(base + idx, step) - base;
490 static unsigned long __init align_off(struct bootmem_data *bdata,
491 unsigned long off, unsigned long align)
493 unsigned long base = PFN_PHYS(bdata->node_min_pfn);
495 /* Same as align_idx for byte offsets */
497 return ALIGN(base + off, align) - base;
500 static void * __init alloc_bootmem_bdata(struct bootmem_data *bdata,
501 unsigned long size, unsigned long align,
502 unsigned long goal, unsigned long limit)
504 unsigned long fallback = 0;
505 unsigned long min, max, start, sidx, midx, step;
507 bdebug("nid=%td size=%lx [%lu pages] align=%lx goal=%lx limit=%lx\n",
508 bdata - bootmem_node_data, size, PAGE_ALIGN(size) >> PAGE_SHIFT,
512 BUG_ON(align & (align - 1));
513 BUG_ON(limit && goal + size > limit);
515 if (!bdata->node_bootmem_map)
518 min = bdata->node_min_pfn;
519 max = bdata->node_low_pfn;
522 limit >>= PAGE_SHIFT;
524 if (limit && max > limit)
529 step = max(align >> PAGE_SHIFT, 1UL);
531 if (goal && min < goal && goal < max)
532 start = ALIGN(goal, step);
534 start = ALIGN(min, step);
536 sidx = start - bdata->node_min_pfn;
537 midx = max - bdata->node_min_pfn;
539 if (bdata->hint_idx > sidx) {
541 * Handle the valid case of sidx being zero and still
542 * catch the fallback below.
545 sidx = align_idx(bdata, bdata->hint_idx, step);
551 unsigned long eidx, i, start_off, end_off;
553 sidx = find_next_zero_bit(bdata->node_bootmem_map, midx, sidx);
554 sidx = align_idx(bdata, sidx, step);
555 eidx = sidx + PFN_UP(size);
557 if (sidx >= midx || eidx > midx)
560 for (i = sidx; i < eidx; i++)
561 if (test_bit(i, bdata->node_bootmem_map)) {
562 sidx = align_idx(bdata, i, step);
568 if (bdata->last_end_off & (PAGE_SIZE - 1) &&
569 PFN_DOWN(bdata->last_end_off) + 1 == sidx)
570 start_off = align_off(bdata, bdata->last_end_off, align);
572 start_off = PFN_PHYS(sidx);
574 merge = PFN_DOWN(start_off) < sidx;
575 end_off = start_off + size;
577 bdata->last_end_off = end_off;
578 bdata->hint_idx = PFN_UP(end_off);
581 * Reserve the area now:
583 if (__reserve(bdata, PFN_DOWN(start_off) + merge,
584 PFN_UP(end_off), BOOTMEM_EXCLUSIVE))
587 region = phys_to_virt(PFN_PHYS(bdata->node_min_pfn) +
589 memset(region, 0, size);
591 * The min_count is set to 0 so that bootmem allocated blocks
592 * are never reported as leaks.
594 kmemleak_alloc(region, size, 0, 0);
599 sidx = align_idx(bdata, fallback - 1, step);
607 static void * __init alloc_bootmem_core(unsigned long size,
612 bootmem_data_t *bdata;
615 if (WARN_ON_ONCE(slab_is_available()))
616 return kzalloc(size, GFP_NOWAIT);
618 list_for_each_entry(bdata, &bdata_list, list) {
619 if (goal && bdata->node_low_pfn <= PFN_DOWN(goal))
621 if (limit && bdata->node_min_pfn >= PFN_DOWN(limit))
624 region = alloc_bootmem_bdata(bdata, size, align, goal, limit);
632 static void * __init ___alloc_bootmem_nopanic(unsigned long size,
640 ptr = alloc_bootmem_core(size, align, goal, limit);
652 * __alloc_bootmem_nopanic - allocate boot memory without panicking
653 * @size: size of the request in bytes
654 * @align: alignment of the region
655 * @goal: preferred starting address of the region
657 * The goal is dropped if it can not be satisfied and the allocation will
658 * fall back to memory below @goal.
660 * Allocation may happen on any node in the system.
662 * Returns NULL on failure.
664 void * __init __alloc_bootmem_nopanic(unsigned long size, unsigned long align,
667 unsigned long limit = 0;
669 return ___alloc_bootmem_nopanic(size, align, goal, limit);
672 static void * __init ___alloc_bootmem(unsigned long size, unsigned long align,
673 unsigned long goal, unsigned long limit)
675 void *mem = ___alloc_bootmem_nopanic(size, align, goal, limit);
680 * Whoops, we cannot satisfy the allocation request.
682 printk(KERN_ALERT "bootmem alloc of %lu bytes failed!\n", size);
683 panic("Out of memory");
688 * __alloc_bootmem - allocate boot memory
689 * @size: size of the request in bytes
690 * @align: alignment of the region
691 * @goal: preferred starting address of the region
693 * The goal is dropped if it can not be satisfied and the allocation will
694 * fall back to memory below @goal.
696 * Allocation may happen on any node in the system.
698 * The function panics if the request can not be satisfied.
700 void * __init __alloc_bootmem(unsigned long size, unsigned long align,
703 unsigned long limit = 0;
705 return ___alloc_bootmem(size, align, goal, limit);
708 void * __init ___alloc_bootmem_node_nopanic(pg_data_t *pgdat,
709 unsigned long size, unsigned long align,
710 unsigned long goal, unsigned long limit)
714 if (WARN_ON_ONCE(slab_is_available()))
715 return kzalloc(size, GFP_NOWAIT);
718 /* do not panic in alloc_bootmem_bdata() */
719 if (limit && goal + size > limit)
722 ptr = alloc_bootmem_bdata(pgdat->bdata, size, align, goal, limit);
726 ptr = alloc_bootmem_core(size, align, goal, limit);
738 void * __init __alloc_bootmem_node_nopanic(pg_data_t *pgdat, unsigned long size,
739 unsigned long align, unsigned long goal)
741 if (WARN_ON_ONCE(slab_is_available()))
742 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
744 return ___alloc_bootmem_node_nopanic(pgdat, size, align, goal, 0);
747 void * __init ___alloc_bootmem_node(pg_data_t *pgdat, unsigned long size,
748 unsigned long align, unsigned long goal,
753 ptr = ___alloc_bootmem_node_nopanic(pgdat, size, align, goal, 0);
757 printk(KERN_ALERT "bootmem alloc of %lu bytes failed!\n", size);
758 panic("Out of memory");
763 * __alloc_bootmem_node - allocate boot memory from a specific node
764 * @pgdat: node to allocate from
765 * @size: size of the request in bytes
766 * @align: alignment of the region
767 * @goal: preferred starting address of the region
769 * The goal is dropped if it can not be satisfied and the allocation will
770 * fall back to memory below @goal.
772 * Allocation may fall back to any node in the system if the specified node
773 * can not hold the requested memory.
775 * The function panics if the request can not be satisfied.
777 void * __init __alloc_bootmem_node(pg_data_t *pgdat, unsigned long size,
778 unsigned long align, unsigned long goal)
780 if (WARN_ON_ONCE(slab_is_available()))
781 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
783 return ___alloc_bootmem_node(pgdat, size, align, goal, 0);
786 void * __init __alloc_bootmem_node_high(pg_data_t *pgdat, unsigned long size,
787 unsigned long align, unsigned long goal)
790 unsigned long end_pfn;
792 if (WARN_ON_ONCE(slab_is_available()))
793 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
795 /* update goal according ...MAX_DMA32_PFN */
796 end_pfn = pgdat_end_pfn(pgdat);
798 if (end_pfn > MAX_DMA32_PFN + (128 >> (20 - PAGE_SHIFT)) &&
799 (goal >> PAGE_SHIFT) < MAX_DMA32_PFN) {
801 unsigned long new_goal;
803 new_goal = MAX_DMA32_PFN << PAGE_SHIFT;
804 ptr = alloc_bootmem_bdata(pgdat->bdata, size, align,
811 return __alloc_bootmem_node(pgdat, size, align, goal);
815 #ifndef ARCH_LOW_ADDRESS_LIMIT
816 #define ARCH_LOW_ADDRESS_LIMIT 0xffffffffUL
820 * __alloc_bootmem_low - allocate low boot memory
821 * @size: size of the request in bytes
822 * @align: alignment of the region
823 * @goal: preferred starting address of the region
825 * The goal is dropped if it can not be satisfied and the allocation will
826 * fall back to memory below @goal.
828 * Allocation may happen on any node in the system.
830 * The function panics if the request can not be satisfied.
832 void * __init __alloc_bootmem_low(unsigned long size, unsigned long align,
835 return ___alloc_bootmem(size, align, goal, ARCH_LOW_ADDRESS_LIMIT);
838 void * __init __alloc_bootmem_low_nopanic(unsigned long size,
842 return ___alloc_bootmem_nopanic(size, align, goal,
843 ARCH_LOW_ADDRESS_LIMIT);
847 * __alloc_bootmem_low_node - allocate low boot memory from a specific node
848 * @pgdat: node to allocate from
849 * @size: size of the request in bytes
850 * @align: alignment of the region
851 * @goal: preferred starting address of the region
853 * The goal is dropped if it can not be satisfied and the allocation will
854 * fall back to memory below @goal.
856 * Allocation may fall back to any node in the system if the specified node
857 * can not hold the requested memory.
859 * The function panics if the request can not be satisfied.
861 void * __init __alloc_bootmem_low_node(pg_data_t *pgdat, unsigned long size,
862 unsigned long align, unsigned long goal)
864 if (WARN_ON_ONCE(slab_is_available()))
865 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
867 return ___alloc_bootmem_node(pgdat, size, align,
868 goal, ARCH_LOW_ADDRESS_LIMIT);