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)
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), 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;
190 map = bdata->node_bootmem_map;
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), order);
214 count += BITS_PER_LONG;
215 start += BITS_PER_LONG;
217 unsigned long cur = start;
219 start = ALIGN(start + 1, BITS_PER_LONG);
220 while (vec && cur != start) {
222 page = pfn_to_page(cur);
223 __free_pages_bootmem(page, 0);
232 page = virt_to_page(bdata->node_bootmem_map);
233 pages = bdata->node_low_pfn - bdata->node_min_pfn;
234 pages = bootmem_bootmap_pages(pages);
237 __free_pages_bootmem(page++, 0);
239 bdebug("nid=%td released=%lx\n", bdata - bootmem_node_data, count);
244 static int reset_managed_pages_done __initdata;
246 static inline void __init reset_node_managed_pages(pg_data_t *pgdat)
250 if (reset_managed_pages_done)
253 for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++)
254 z->managed_pages = 0;
257 void __init reset_all_zones_managed_pages(void)
259 struct pglist_data *pgdat;
261 for_each_online_pgdat(pgdat)
262 reset_node_managed_pages(pgdat);
263 reset_managed_pages_done = 1;
267 * free_all_bootmem_node - release a node's free pages to the buddy allocator
268 * @pgdat: node to be released
270 * Returns the number of pages actually released.
272 unsigned long __init free_all_bootmem_node(pg_data_t *pgdat)
274 register_page_bootmem_info_node(pgdat);
275 reset_node_managed_pages(pgdat);
276 return free_all_bootmem_core(pgdat->bdata);
280 * free_all_bootmem - release free pages to the buddy allocator
282 * Returns the number of pages actually released.
284 unsigned long __init free_all_bootmem(void)
286 unsigned long total_pages = 0;
287 bootmem_data_t *bdata;
289 reset_all_zones_managed_pages();
291 list_for_each_entry(bdata, &bdata_list, list)
292 total_pages += free_all_bootmem_core(bdata);
297 static void __init __free(bootmem_data_t *bdata,
298 unsigned long sidx, unsigned long eidx)
302 bdebug("nid=%td start=%lx end=%lx\n", bdata - bootmem_node_data,
303 sidx + bdata->node_min_pfn,
304 eidx + bdata->node_min_pfn);
306 if (bdata->hint_idx > sidx)
307 bdata->hint_idx = sidx;
309 for (idx = sidx; idx < eidx; idx++)
310 if (!test_and_clear_bit(idx, bdata->node_bootmem_map))
314 static int __init __reserve(bootmem_data_t *bdata, unsigned long sidx,
315 unsigned long eidx, int flags)
318 int exclusive = flags & BOOTMEM_EXCLUSIVE;
320 bdebug("nid=%td start=%lx end=%lx flags=%x\n",
321 bdata - bootmem_node_data,
322 sidx + bdata->node_min_pfn,
323 eidx + bdata->node_min_pfn,
326 for (idx = sidx; idx < eidx; idx++)
327 if (test_and_set_bit(idx, bdata->node_bootmem_map)) {
329 __free(bdata, sidx, idx);
332 bdebug("silent double reserve of PFN %lx\n",
333 idx + bdata->node_min_pfn);
338 static int __init mark_bootmem_node(bootmem_data_t *bdata,
339 unsigned long start, unsigned long end,
340 int reserve, int flags)
342 unsigned long sidx, eidx;
344 bdebug("nid=%td start=%lx end=%lx reserve=%d flags=%x\n",
345 bdata - bootmem_node_data, start, end, reserve, flags);
347 BUG_ON(start < bdata->node_min_pfn);
348 BUG_ON(end > bdata->node_low_pfn);
350 sidx = start - bdata->node_min_pfn;
351 eidx = end - bdata->node_min_pfn;
354 return __reserve(bdata, sidx, eidx, flags);
356 __free(bdata, sidx, eidx);
360 static int __init mark_bootmem(unsigned long start, unsigned long end,
361 int reserve, int flags)
364 bootmem_data_t *bdata;
367 list_for_each_entry(bdata, &bdata_list, list) {
371 if (pos < bdata->node_min_pfn ||
372 pos >= bdata->node_low_pfn) {
373 BUG_ON(pos != start);
377 max = min(bdata->node_low_pfn, end);
379 err = mark_bootmem_node(bdata, pos, max, reserve, flags);
380 if (reserve && err) {
381 mark_bootmem(start, pos, 0, 0);
387 pos = bdata->node_low_pfn;
393 * free_bootmem_node - mark a page range as usable
394 * @pgdat: node the range resides on
395 * @physaddr: starting address of the range
396 * @size: size of the range in bytes
398 * Partial pages will be considered reserved and left as they are.
400 * The range must reside completely on the specified node.
402 void __init free_bootmem_node(pg_data_t *pgdat, unsigned long physaddr,
405 unsigned long start, end;
407 kmemleak_free_part(__va(physaddr), size);
409 start = PFN_UP(physaddr);
410 end = PFN_DOWN(physaddr + size);
412 mark_bootmem_node(pgdat->bdata, start, end, 0, 0);
416 * free_bootmem - mark a page range as usable
417 * @addr: starting physical address of the range
418 * @size: size of the range in bytes
420 * Partial pages will be considered reserved and left as they are.
422 * The range must be contiguous but may span node boundaries.
424 void __init free_bootmem(unsigned long physaddr, unsigned long size)
426 unsigned long start, end;
428 kmemleak_free_part(__va(physaddr), size);
430 start = PFN_UP(physaddr);
431 end = PFN_DOWN(physaddr + size);
433 mark_bootmem(start, end, 0, 0);
437 * reserve_bootmem_node - mark a page range as reserved
438 * @pgdat: node the range resides on
439 * @physaddr: starting address of the range
440 * @size: size of the range in bytes
441 * @flags: reservation flags (see linux/bootmem.h)
443 * Partial pages will be reserved.
445 * The range must reside completely on the specified node.
447 int __init reserve_bootmem_node(pg_data_t *pgdat, unsigned long physaddr,
448 unsigned long size, int flags)
450 unsigned long start, end;
452 start = PFN_DOWN(physaddr);
453 end = PFN_UP(physaddr + size);
455 return mark_bootmem_node(pgdat->bdata, start, end, 1, flags);
459 * reserve_bootmem - mark a page range as reserved
460 * @addr: starting address of the range
461 * @size: size of the range in bytes
462 * @flags: reservation flags (see linux/bootmem.h)
464 * Partial pages will be reserved.
466 * The range must be contiguous but may span node boundaries.
468 int __init reserve_bootmem(unsigned long addr, unsigned long size,
471 unsigned long start, end;
473 start = PFN_DOWN(addr);
474 end = PFN_UP(addr + size);
476 return mark_bootmem(start, end, 1, flags);
479 static unsigned long __init align_idx(struct bootmem_data *bdata,
480 unsigned long idx, unsigned long step)
482 unsigned long base = bdata->node_min_pfn;
485 * Align the index with respect to the node start so that the
486 * combination of both satisfies the requested alignment.
489 return ALIGN(base + idx, step) - base;
492 static unsigned long __init align_off(struct bootmem_data *bdata,
493 unsigned long off, unsigned long align)
495 unsigned long base = PFN_PHYS(bdata->node_min_pfn);
497 /* Same as align_idx for byte offsets */
499 return ALIGN(base + off, align) - base;
502 static void * __init alloc_bootmem_bdata(struct bootmem_data *bdata,
503 unsigned long size, unsigned long align,
504 unsigned long goal, unsigned long limit)
506 unsigned long fallback = 0;
507 unsigned long min, max, start, sidx, midx, step;
509 bdebug("nid=%td size=%lx [%lu pages] align=%lx goal=%lx limit=%lx\n",
510 bdata - bootmem_node_data, size, PAGE_ALIGN(size) >> PAGE_SHIFT,
514 BUG_ON(align & (align - 1));
515 BUG_ON(limit && goal + size > limit);
517 if (!bdata->node_bootmem_map)
520 min = bdata->node_min_pfn;
521 max = bdata->node_low_pfn;
524 limit >>= PAGE_SHIFT;
526 if (limit && max > limit)
531 step = max(align >> PAGE_SHIFT, 1UL);
533 if (goal && min < goal && goal < max)
534 start = ALIGN(goal, step);
536 start = ALIGN(min, step);
538 sidx = start - bdata->node_min_pfn;
539 midx = max - bdata->node_min_pfn;
541 if (bdata->hint_idx > sidx) {
543 * Handle the valid case of sidx being zero and still
544 * catch the fallback below.
547 sidx = align_idx(bdata, bdata->hint_idx, step);
553 unsigned long eidx, i, start_off, end_off;
555 sidx = find_next_zero_bit(bdata->node_bootmem_map, midx, sidx);
556 sidx = align_idx(bdata, sidx, step);
557 eidx = sidx + PFN_UP(size);
559 if (sidx >= midx || eidx > midx)
562 for (i = sidx; i < eidx; i++)
563 if (test_bit(i, bdata->node_bootmem_map)) {
564 sidx = align_idx(bdata, i, step);
570 if (bdata->last_end_off & (PAGE_SIZE - 1) &&
571 PFN_DOWN(bdata->last_end_off) + 1 == sidx)
572 start_off = align_off(bdata, bdata->last_end_off, align);
574 start_off = PFN_PHYS(sidx);
576 merge = PFN_DOWN(start_off) < sidx;
577 end_off = start_off + size;
579 bdata->last_end_off = end_off;
580 bdata->hint_idx = PFN_UP(end_off);
583 * Reserve the area now:
585 if (__reserve(bdata, PFN_DOWN(start_off) + merge,
586 PFN_UP(end_off), BOOTMEM_EXCLUSIVE))
589 region = phys_to_virt(PFN_PHYS(bdata->node_min_pfn) +
591 memset(region, 0, size);
593 * The min_count is set to 0 so that bootmem allocated blocks
594 * are never reported as leaks.
596 kmemleak_alloc(region, size, 0, 0);
601 sidx = align_idx(bdata, fallback - 1, step);
609 static void * __init alloc_bootmem_core(unsigned long size,
614 bootmem_data_t *bdata;
617 if (WARN_ON_ONCE(slab_is_available()))
618 return kzalloc(size, GFP_NOWAIT);
620 list_for_each_entry(bdata, &bdata_list, list) {
621 if (goal && bdata->node_low_pfn <= PFN_DOWN(goal))
623 if (limit && bdata->node_min_pfn >= PFN_DOWN(limit))
626 region = alloc_bootmem_bdata(bdata, size, align, goal, limit);
634 static void * __init ___alloc_bootmem_nopanic(unsigned long size,
642 ptr = alloc_bootmem_core(size, align, goal, limit);
654 * __alloc_bootmem_nopanic - allocate boot memory without panicking
655 * @size: size of the request in bytes
656 * @align: alignment of the region
657 * @goal: preferred starting address of the region
659 * The goal is dropped if it can not be satisfied and the allocation will
660 * fall back to memory below @goal.
662 * Allocation may happen on any node in the system.
664 * Returns NULL on failure.
666 void * __init __alloc_bootmem_nopanic(unsigned long size, unsigned long align,
669 unsigned long limit = 0;
671 return ___alloc_bootmem_nopanic(size, align, goal, limit);
674 static void * __init ___alloc_bootmem(unsigned long size, unsigned long align,
675 unsigned long goal, unsigned long limit)
677 void *mem = ___alloc_bootmem_nopanic(size, align, goal, limit);
682 * Whoops, we cannot satisfy the allocation request.
684 printk(KERN_ALERT "bootmem alloc of %lu bytes failed!\n", size);
685 panic("Out of memory");
690 * __alloc_bootmem - allocate boot memory
691 * @size: size of the request in bytes
692 * @align: alignment of the region
693 * @goal: preferred starting address of the region
695 * The goal is dropped if it can not be satisfied and the allocation will
696 * fall back to memory below @goal.
698 * Allocation may happen on any node in the system.
700 * The function panics if the request can not be satisfied.
702 void * __init __alloc_bootmem(unsigned long size, unsigned long align,
705 unsigned long limit = 0;
707 return ___alloc_bootmem(size, align, goal, limit);
710 void * __init ___alloc_bootmem_node_nopanic(pg_data_t *pgdat,
711 unsigned long size, unsigned long align,
712 unsigned long goal, unsigned long limit)
716 if (WARN_ON_ONCE(slab_is_available()))
717 return kzalloc(size, GFP_NOWAIT);
720 /* do not panic in alloc_bootmem_bdata() */
721 if (limit && goal + size > limit)
724 ptr = alloc_bootmem_bdata(pgdat->bdata, size, align, goal, limit);
728 ptr = alloc_bootmem_core(size, align, goal, limit);
740 void * __init __alloc_bootmem_node_nopanic(pg_data_t *pgdat, unsigned long size,
741 unsigned long align, unsigned long goal)
743 if (WARN_ON_ONCE(slab_is_available()))
744 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
746 return ___alloc_bootmem_node_nopanic(pgdat, size, align, goal, 0);
749 void * __init ___alloc_bootmem_node(pg_data_t *pgdat, unsigned long size,
750 unsigned long align, unsigned long goal,
755 ptr = ___alloc_bootmem_node_nopanic(pgdat, size, align, goal, 0);
759 printk(KERN_ALERT "bootmem alloc of %lu bytes failed!\n", size);
760 panic("Out of memory");
765 * __alloc_bootmem_node - allocate boot memory from a specific node
766 * @pgdat: node to allocate from
767 * @size: size of the request in bytes
768 * @align: alignment of the region
769 * @goal: preferred starting address of the region
771 * The goal is dropped if it can not be satisfied and the allocation will
772 * fall back to memory below @goal.
774 * Allocation may fall back to any node in the system if the specified node
775 * can not hold the requested memory.
777 * The function panics if the request can not be satisfied.
779 void * __init __alloc_bootmem_node(pg_data_t *pgdat, unsigned long size,
780 unsigned long align, unsigned long goal)
782 if (WARN_ON_ONCE(slab_is_available()))
783 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
785 return ___alloc_bootmem_node(pgdat, size, align, goal, 0);
788 void * __init __alloc_bootmem_node_high(pg_data_t *pgdat, unsigned long size,
789 unsigned long align, unsigned long goal)
792 unsigned long end_pfn;
794 if (WARN_ON_ONCE(slab_is_available()))
795 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
797 /* update goal according ...MAX_DMA32_PFN */
798 end_pfn = pgdat->node_start_pfn + pgdat->node_spanned_pages;
800 if (end_pfn > MAX_DMA32_PFN + (128 >> (20 - PAGE_SHIFT)) &&
801 (goal >> PAGE_SHIFT) < MAX_DMA32_PFN) {
803 unsigned long new_goal;
805 new_goal = MAX_DMA32_PFN << PAGE_SHIFT;
806 ptr = alloc_bootmem_bdata(pgdat->bdata, size, align,
813 return __alloc_bootmem_node(pgdat, size, align, goal);
817 #ifndef ARCH_LOW_ADDRESS_LIMIT
818 #define ARCH_LOW_ADDRESS_LIMIT 0xffffffffUL
822 * __alloc_bootmem_low - allocate low boot memory
823 * @size: size of the request in bytes
824 * @align: alignment of the region
825 * @goal: preferred starting address of the region
827 * The goal is dropped if it can not be satisfied and the allocation will
828 * fall back to memory below @goal.
830 * Allocation may happen on any node in the system.
832 * The function panics if the request can not be satisfied.
834 void * __init __alloc_bootmem_low(unsigned long size, unsigned long align,
837 return ___alloc_bootmem(size, align, goal, ARCH_LOW_ADDRESS_LIMIT);
840 void * __init __alloc_bootmem_low_nopanic(unsigned long size,
844 return ___alloc_bootmem_nopanic(size, align, goal,
845 ARCH_LOW_ADDRESS_LIMIT);
849 * __alloc_bootmem_low_node - allocate low boot memory from a specific node
850 * @pgdat: node to allocate from
851 * @size: size of the request in bytes
852 * @align: alignment of the region
853 * @goal: preferred starting address of the region
855 * The goal is dropped if it can not be satisfied and the allocation will
856 * fall back to memory below @goal.
858 * Allocation may fall back to any node in the system if the specified node
859 * can not hold the requested memory.
861 * The function panics if the request can not be satisfied.
863 void * __init __alloc_bootmem_low_node(pg_data_t *pgdat, unsigned long size,
864 unsigned long align, unsigned long goal)
866 if (WARN_ON_ONCE(slab_is_available()))
867 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
869 return ___alloc_bootmem_node(pgdat, size, align,
870 goal, ARCH_LOW_ADDRESS_LIMIT);