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
), 0);
172 static unsigned long __init
free_all_bootmem_core(bootmem_data_t
*bdata
)
175 unsigned long *map
, start
, end
, pages
, 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
), 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 void reset_node_managed_pages(pg_data_t
*pgdat
)
250 for (z
= pgdat
->node_zones
; z
< pgdat
->node_zones
+ MAX_NR_ZONES
; z
++)
251 z
->managed_pages
= 0;
254 void __init
reset_all_zones_managed_pages(void)
256 struct pglist_data
*pgdat
;
258 if (reset_managed_pages_done
)
261 for_each_online_pgdat(pgdat
)
262 reset_node_managed_pages(pgdat
);
264 reset_managed_pages_done
= 1;
268 * free_all_bootmem - release free pages to the buddy allocator
270 * Returns the number of pages actually released.
272 unsigned long __init
free_all_bootmem(void)
274 unsigned long total_pages
= 0;
275 bootmem_data_t
*bdata
;
277 reset_all_zones_managed_pages();
279 list_for_each_entry(bdata
, &bdata_list
, list
)
280 total_pages
+= free_all_bootmem_core(bdata
);
282 totalram_pages
+= total_pages
;
287 static void __init
__free(bootmem_data_t
*bdata
,
288 unsigned long sidx
, unsigned long eidx
)
292 bdebug("nid=%td start=%lx end=%lx\n", bdata
- bootmem_node_data
,
293 sidx
+ bdata
->node_min_pfn
,
294 eidx
+ bdata
->node_min_pfn
);
296 if (bdata
->hint_idx
> sidx
)
297 bdata
->hint_idx
= sidx
;
299 for (idx
= sidx
; idx
< eidx
; idx
++)
300 if (!test_and_clear_bit(idx
, bdata
->node_bootmem_map
))
304 static int __init
__reserve(bootmem_data_t
*bdata
, unsigned long sidx
,
305 unsigned long eidx
, int flags
)
308 int exclusive
= flags
& BOOTMEM_EXCLUSIVE
;
310 bdebug("nid=%td start=%lx end=%lx flags=%x\n",
311 bdata
- bootmem_node_data
,
312 sidx
+ bdata
->node_min_pfn
,
313 eidx
+ bdata
->node_min_pfn
,
316 for (idx
= sidx
; idx
< eidx
; idx
++)
317 if (test_and_set_bit(idx
, bdata
->node_bootmem_map
)) {
319 __free(bdata
, sidx
, idx
);
322 bdebug("silent double reserve of PFN %lx\n",
323 idx
+ bdata
->node_min_pfn
);
328 static int __init
mark_bootmem_node(bootmem_data_t
*bdata
,
329 unsigned long start
, unsigned long end
,
330 int reserve
, int flags
)
332 unsigned long sidx
, eidx
;
334 bdebug("nid=%td start=%lx end=%lx reserve=%d flags=%x\n",
335 bdata
- bootmem_node_data
, start
, end
, reserve
, flags
);
337 BUG_ON(start
< bdata
->node_min_pfn
);
338 BUG_ON(end
> bdata
->node_low_pfn
);
340 sidx
= start
- bdata
->node_min_pfn
;
341 eidx
= end
- bdata
->node_min_pfn
;
344 return __reserve(bdata
, sidx
, eidx
, flags
);
346 __free(bdata
, sidx
, eidx
);
350 static int __init
mark_bootmem(unsigned long start
, unsigned long end
,
351 int reserve
, int flags
)
354 bootmem_data_t
*bdata
;
357 list_for_each_entry(bdata
, &bdata_list
, list
) {
361 if (pos
< bdata
->node_min_pfn
||
362 pos
>= bdata
->node_low_pfn
) {
363 BUG_ON(pos
!= start
);
367 max
= min(bdata
->node_low_pfn
, end
);
369 err
= mark_bootmem_node(bdata
, pos
, max
, reserve
, flags
);
370 if (reserve
&& err
) {
371 mark_bootmem(start
, pos
, 0, 0);
377 pos
= bdata
->node_low_pfn
;
383 * free_bootmem_node - mark a page range as usable
384 * @pgdat: node the range resides on
385 * @physaddr: starting address of the range
386 * @size: size of the range in bytes
388 * Partial pages will be considered reserved and left as they are.
390 * The range must reside completely on the specified node.
392 void __init
free_bootmem_node(pg_data_t
*pgdat
, unsigned long physaddr
,
395 unsigned long start
, end
;
397 kmemleak_free_part(__va(physaddr
), size
);
399 start
= PFN_UP(physaddr
);
400 end
= PFN_DOWN(physaddr
+ size
);
402 mark_bootmem_node(pgdat
->bdata
, start
, end
, 0, 0);
406 * free_bootmem - mark a page range as usable
407 * @addr: starting physical address of the range
408 * @size: size of the range in bytes
410 * Partial pages will be considered reserved and left as they are.
412 * The range must be contiguous but may span node boundaries.
414 void __init
free_bootmem(unsigned long physaddr
, unsigned long size
)
416 unsigned long start
, end
;
418 kmemleak_free_part(__va(physaddr
), size
);
420 start
= PFN_UP(physaddr
);
421 end
= PFN_DOWN(physaddr
+ size
);
423 mark_bootmem(start
, end
, 0, 0);
427 * reserve_bootmem_node - mark a page range as reserved
428 * @pgdat: node the range resides on
429 * @physaddr: starting address of the range
430 * @size: size of the range in bytes
431 * @flags: reservation flags (see linux/bootmem.h)
433 * Partial pages will be reserved.
435 * The range must reside completely on the specified node.
437 int __init
reserve_bootmem_node(pg_data_t
*pgdat
, unsigned long physaddr
,
438 unsigned long size
, int flags
)
440 unsigned long start
, end
;
442 start
= PFN_DOWN(physaddr
);
443 end
= PFN_UP(physaddr
+ size
);
445 return mark_bootmem_node(pgdat
->bdata
, start
, end
, 1, flags
);
449 * reserve_bootmem - mark a page range as reserved
450 * @addr: starting address of the range
451 * @size: size of the range in bytes
452 * @flags: reservation flags (see linux/bootmem.h)
454 * Partial pages will be reserved.
456 * The range must be contiguous but may span node boundaries.
458 int __init
reserve_bootmem(unsigned long addr
, unsigned long size
,
461 unsigned long start
, end
;
463 start
= PFN_DOWN(addr
);
464 end
= PFN_UP(addr
+ size
);
466 return mark_bootmem(start
, end
, 1, flags
);
469 static unsigned long __init
align_idx(struct bootmem_data
*bdata
,
470 unsigned long idx
, unsigned long step
)
472 unsigned long base
= bdata
->node_min_pfn
;
475 * Align the index with respect to the node start so that the
476 * combination of both satisfies the requested alignment.
479 return ALIGN(base
+ idx
, step
) - base
;
482 static unsigned long __init
align_off(struct bootmem_data
*bdata
,
483 unsigned long off
, unsigned long align
)
485 unsigned long base
= PFN_PHYS(bdata
->node_min_pfn
);
487 /* Same as align_idx for byte offsets */
489 return ALIGN(base
+ off
, align
) - base
;
492 static void * __init
alloc_bootmem_bdata(struct bootmem_data
*bdata
,
493 unsigned long size
, unsigned long align
,
494 unsigned long goal
, unsigned long limit
)
496 unsigned long fallback
= 0;
497 unsigned long min
, max
, start
, sidx
, midx
, step
;
499 bdebug("nid=%td size=%lx [%lu pages] align=%lx goal=%lx limit=%lx\n",
500 bdata
- bootmem_node_data
, size
, PAGE_ALIGN(size
) >> PAGE_SHIFT
,
504 BUG_ON(align
& (align
- 1));
505 BUG_ON(limit
&& goal
+ size
> limit
);
507 if (!bdata
->node_bootmem_map
)
510 min
= bdata
->node_min_pfn
;
511 max
= bdata
->node_low_pfn
;
514 limit
>>= PAGE_SHIFT
;
516 if (limit
&& max
> limit
)
521 step
= max(align
>> PAGE_SHIFT
, 1UL);
523 if (goal
&& min
< goal
&& goal
< max
)
524 start
= ALIGN(goal
, step
);
526 start
= ALIGN(min
, step
);
528 sidx
= start
- bdata
->node_min_pfn
;
529 midx
= max
- bdata
->node_min_pfn
;
531 if (bdata
->hint_idx
> sidx
) {
533 * Handle the valid case of sidx being zero and still
534 * catch the fallback below.
537 sidx
= align_idx(bdata
, bdata
->hint_idx
, step
);
543 unsigned long eidx
, i
, start_off
, end_off
;
545 sidx
= find_next_zero_bit(bdata
->node_bootmem_map
, midx
, sidx
);
546 sidx
= align_idx(bdata
, sidx
, step
);
547 eidx
= sidx
+ PFN_UP(size
);
549 if (sidx
>= midx
|| eidx
> midx
)
552 for (i
= sidx
; i
< eidx
; i
++)
553 if (test_bit(i
, bdata
->node_bootmem_map
)) {
554 sidx
= align_idx(bdata
, i
, step
);
560 if (bdata
->last_end_off
& (PAGE_SIZE
- 1) &&
561 PFN_DOWN(bdata
->last_end_off
) + 1 == sidx
)
562 start_off
= align_off(bdata
, bdata
->last_end_off
, align
);
564 start_off
= PFN_PHYS(sidx
);
566 merge
= PFN_DOWN(start_off
) < sidx
;
567 end_off
= start_off
+ size
;
569 bdata
->last_end_off
= end_off
;
570 bdata
->hint_idx
= PFN_UP(end_off
);
573 * Reserve the area now:
575 if (__reserve(bdata
, PFN_DOWN(start_off
) + merge
,
576 PFN_UP(end_off
), BOOTMEM_EXCLUSIVE
))
579 region
= phys_to_virt(PFN_PHYS(bdata
->node_min_pfn
) +
581 memset(region
, 0, size
);
583 * The min_count is set to 0 so that bootmem allocated blocks
584 * are never reported as leaks.
586 kmemleak_alloc(region
, size
, 0, 0);
591 sidx
= align_idx(bdata
, fallback
- 1, step
);
599 static void * __init
alloc_bootmem_core(unsigned long size
,
604 bootmem_data_t
*bdata
;
607 if (WARN_ON_ONCE(slab_is_available()))
608 return kzalloc(size
, GFP_NOWAIT
);
610 list_for_each_entry(bdata
, &bdata_list
, list
) {
611 if (goal
&& bdata
->node_low_pfn
<= PFN_DOWN(goal
))
613 if (limit
&& bdata
->node_min_pfn
>= PFN_DOWN(limit
))
616 region
= alloc_bootmem_bdata(bdata
, size
, align
, goal
, limit
);
624 static void * __init
___alloc_bootmem_nopanic(unsigned long size
,
632 ptr
= alloc_bootmem_core(size
, align
, goal
, limit
);
644 * __alloc_bootmem_nopanic - allocate boot memory without panicking
645 * @size: size of the request in bytes
646 * @align: alignment of the region
647 * @goal: preferred starting address of the region
649 * The goal is dropped if it can not be satisfied and the allocation will
650 * fall back to memory below @goal.
652 * Allocation may happen on any node in the system.
654 * Returns NULL on failure.
656 void * __init
__alloc_bootmem_nopanic(unsigned long size
, unsigned long align
,
659 unsigned long limit
= 0;
661 return ___alloc_bootmem_nopanic(size
, align
, goal
, limit
);
664 static void * __init
___alloc_bootmem(unsigned long size
, unsigned long align
,
665 unsigned long goal
, unsigned long limit
)
667 void *mem
= ___alloc_bootmem_nopanic(size
, align
, goal
, limit
);
672 * Whoops, we cannot satisfy the allocation request.
674 printk(KERN_ALERT
"bootmem alloc of %lu bytes failed!\n", size
);
675 panic("Out of memory");
680 * __alloc_bootmem - allocate boot memory
681 * @size: size of the request in bytes
682 * @align: alignment of the region
683 * @goal: preferred starting address of the region
685 * The goal is dropped if it can not be satisfied and the allocation will
686 * fall back to memory below @goal.
688 * Allocation may happen on any node in the system.
690 * The function panics if the request can not be satisfied.
692 void * __init
__alloc_bootmem(unsigned long size
, unsigned long align
,
695 unsigned long limit
= 0;
697 return ___alloc_bootmem(size
, align
, goal
, limit
);
700 void * __init
___alloc_bootmem_node_nopanic(pg_data_t
*pgdat
,
701 unsigned long size
, unsigned long align
,
702 unsigned long goal
, unsigned long limit
)
706 if (WARN_ON_ONCE(slab_is_available()))
707 return kzalloc(size
, GFP_NOWAIT
);
710 /* do not panic in alloc_bootmem_bdata() */
711 if (limit
&& goal
+ size
> limit
)
714 ptr
= alloc_bootmem_bdata(pgdat
->bdata
, size
, align
, goal
, limit
);
718 ptr
= alloc_bootmem_core(size
, align
, goal
, limit
);
730 void * __init
__alloc_bootmem_node_nopanic(pg_data_t
*pgdat
, unsigned long size
,
731 unsigned long align
, unsigned long goal
)
733 if (WARN_ON_ONCE(slab_is_available()))
734 return kzalloc_node(size
, GFP_NOWAIT
, pgdat
->node_id
);
736 return ___alloc_bootmem_node_nopanic(pgdat
, size
, align
, goal
, 0);
739 void * __init
___alloc_bootmem_node(pg_data_t
*pgdat
, unsigned long size
,
740 unsigned long align
, unsigned long goal
,
745 ptr
= ___alloc_bootmem_node_nopanic(pgdat
, size
, align
, goal
, 0);
749 printk(KERN_ALERT
"bootmem alloc of %lu bytes failed!\n", size
);
750 panic("Out of memory");
755 * __alloc_bootmem_node - allocate boot memory from a specific node
756 * @pgdat: node to allocate from
757 * @size: size of the request in bytes
758 * @align: alignment of the region
759 * @goal: preferred starting address of the region
761 * The goal is dropped if it can not be satisfied and the allocation will
762 * fall back to memory below @goal.
764 * Allocation may fall back to any node in the system if the specified node
765 * can not hold the requested memory.
767 * The function panics if the request can not be satisfied.
769 void * __init
__alloc_bootmem_node(pg_data_t
*pgdat
, unsigned long size
,
770 unsigned long align
, unsigned long goal
)
772 if (WARN_ON_ONCE(slab_is_available()))
773 return kzalloc_node(size
, GFP_NOWAIT
, pgdat
->node_id
);
775 return ___alloc_bootmem_node(pgdat
, size
, align
, goal
, 0);
778 void * __init
__alloc_bootmem_node_high(pg_data_t
*pgdat
, unsigned long size
,
779 unsigned long align
, unsigned long goal
)
782 unsigned long end_pfn
;
784 if (WARN_ON_ONCE(slab_is_available()))
785 return kzalloc_node(size
, GFP_NOWAIT
, pgdat
->node_id
);
787 /* update goal according ...MAX_DMA32_PFN */
788 end_pfn
= pgdat_end_pfn(pgdat
);
790 if (end_pfn
> MAX_DMA32_PFN
+ (128 >> (20 - PAGE_SHIFT
)) &&
791 (goal
>> PAGE_SHIFT
) < MAX_DMA32_PFN
) {
793 unsigned long new_goal
;
795 new_goal
= MAX_DMA32_PFN
<< PAGE_SHIFT
;
796 ptr
= alloc_bootmem_bdata(pgdat
->bdata
, size
, align
,
803 return __alloc_bootmem_node(pgdat
, size
, align
, goal
);
807 #ifndef ARCH_LOW_ADDRESS_LIMIT
808 #define ARCH_LOW_ADDRESS_LIMIT 0xffffffffUL
812 * __alloc_bootmem_low - allocate low boot memory
813 * @size: size of the request in bytes
814 * @align: alignment of the region
815 * @goal: preferred starting address of the region
817 * The goal is dropped if it can not be satisfied and the allocation will
818 * fall back to memory below @goal.
820 * Allocation may happen on any node in the system.
822 * The function panics if the request can not be satisfied.
824 void * __init
__alloc_bootmem_low(unsigned long size
, unsigned long align
,
827 return ___alloc_bootmem(size
, align
, goal
, ARCH_LOW_ADDRESS_LIMIT
);
830 void * __init
__alloc_bootmem_low_nopanic(unsigned long size
,
834 return ___alloc_bootmem_nopanic(size
, align
, goal
,
835 ARCH_LOW_ADDRESS_LIMIT
);
839 * __alloc_bootmem_low_node - allocate low boot memory from a specific node
840 * @pgdat: node to allocate from
841 * @size: size of the request in bytes
842 * @align: alignment of the region
843 * @goal: preferred starting address of the region
845 * The goal is dropped if it can not be satisfied and the allocation will
846 * fall back to memory below @goal.
848 * Allocation may fall back to any node in the system if the specified node
849 * can not hold the requested memory.
851 * The function panics if the request can not be satisfied.
853 void * __init
__alloc_bootmem_low_node(pg_data_t
*pgdat
, unsigned long size
,
854 unsigned long align
, unsigned long goal
)
856 if (WARN_ON_ONCE(slab_is_available()))
857 return kzalloc_node(size
, GFP_NOWAIT
, pgdat
->node_id
);
859 return ___alloc_bootmem_node(pgdat
, size
, align
,
860 goal
, ARCH_LOW_ADDRESS_LIMIT
);