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 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 vec
>>= start
& (BITS_PER_LONG
- 1);
209 page
= pfn_to_page(start
+ off
);
210 __free_pages_bootmem(page
, 0);
216 start
= ALIGN(start
+ 1, BITS_PER_LONG
);
220 page
= virt_to_page(bdata
->node_bootmem_map
);
221 pages
= bdata
->node_low_pfn
- bdata
->node_min_pfn
;
222 pages
= bootmem_bootmap_pages(pages
);
225 __free_pages_bootmem(page
++, 0);
227 bdebug("nid=%td released=%lx\n", bdata
- bootmem_node_data
, count
);
233 * free_all_bootmem_node - release a node's free pages to the buddy allocator
234 * @pgdat: node to be released
236 * Returns the number of pages actually released.
238 unsigned long __init
free_all_bootmem_node(pg_data_t
*pgdat
)
240 register_page_bootmem_info_node(pgdat
);
241 return free_all_bootmem_core(pgdat
->bdata
);
245 * free_all_bootmem - release free pages to the buddy allocator
247 * Returns the number of pages actually released.
249 unsigned long __init
free_all_bootmem(void)
251 unsigned long total_pages
= 0;
252 bootmem_data_t
*bdata
;
254 list_for_each_entry(bdata
, &bdata_list
, list
)
255 total_pages
+= free_all_bootmem_core(bdata
);
260 static void __init
__free(bootmem_data_t
*bdata
,
261 unsigned long sidx
, unsigned long eidx
)
265 bdebug("nid=%td start=%lx end=%lx\n", bdata
- bootmem_node_data
,
266 sidx
+ bdata
->node_min_pfn
,
267 eidx
+ bdata
->node_min_pfn
);
269 if (bdata
->hint_idx
> sidx
)
270 bdata
->hint_idx
= sidx
;
272 for (idx
= sidx
; idx
< eidx
; idx
++)
273 if (!test_and_clear_bit(idx
, bdata
->node_bootmem_map
))
277 static int __init
__reserve(bootmem_data_t
*bdata
, unsigned long sidx
,
278 unsigned long eidx
, int flags
)
281 int exclusive
= flags
& BOOTMEM_EXCLUSIVE
;
283 bdebug("nid=%td start=%lx end=%lx flags=%x\n",
284 bdata
- bootmem_node_data
,
285 sidx
+ bdata
->node_min_pfn
,
286 eidx
+ bdata
->node_min_pfn
,
289 for (idx
= sidx
; idx
< eidx
; idx
++)
290 if (test_and_set_bit(idx
, bdata
->node_bootmem_map
)) {
292 __free(bdata
, sidx
, idx
);
295 bdebug("silent double reserve of PFN %lx\n",
296 idx
+ bdata
->node_min_pfn
);
301 static int __init
mark_bootmem_node(bootmem_data_t
*bdata
,
302 unsigned long start
, unsigned long end
,
303 int reserve
, int flags
)
305 unsigned long sidx
, eidx
;
307 bdebug("nid=%td start=%lx end=%lx reserve=%d flags=%x\n",
308 bdata
- bootmem_node_data
, start
, end
, reserve
, flags
);
310 BUG_ON(start
< bdata
->node_min_pfn
);
311 BUG_ON(end
> bdata
->node_low_pfn
);
313 sidx
= start
- bdata
->node_min_pfn
;
314 eidx
= end
- bdata
->node_min_pfn
;
317 return __reserve(bdata
, sidx
, eidx
, flags
);
319 __free(bdata
, sidx
, eidx
);
323 static int __init
mark_bootmem(unsigned long start
, unsigned long end
,
324 int reserve
, int flags
)
327 bootmem_data_t
*bdata
;
330 list_for_each_entry(bdata
, &bdata_list
, list
) {
334 if (pos
< bdata
->node_min_pfn
||
335 pos
>= bdata
->node_low_pfn
) {
336 BUG_ON(pos
!= start
);
340 max
= min(bdata
->node_low_pfn
, end
);
342 err
= mark_bootmem_node(bdata
, pos
, max
, reserve
, flags
);
343 if (reserve
&& err
) {
344 mark_bootmem(start
, pos
, 0, 0);
350 pos
= bdata
->node_low_pfn
;
356 * free_bootmem_node - mark a page range as usable
357 * @pgdat: node the range resides on
358 * @physaddr: starting address of the range
359 * @size: size of the range in bytes
361 * Partial pages will be considered reserved and left as they are.
363 * The range must reside completely on the specified node.
365 void __init
free_bootmem_node(pg_data_t
*pgdat
, unsigned long physaddr
,
368 unsigned long start
, end
;
370 kmemleak_free_part(__va(physaddr
), size
);
372 start
= PFN_UP(physaddr
);
373 end
= PFN_DOWN(physaddr
+ size
);
375 mark_bootmem_node(pgdat
->bdata
, start
, end
, 0, 0);
379 * free_bootmem - mark a page range as usable
380 * @addr: starting address of the range
381 * @size: size of the range in bytes
383 * Partial pages will be considered reserved and left as they are.
385 * The range must be contiguous but may span node boundaries.
387 void __init
free_bootmem(unsigned long addr
, unsigned long size
)
389 unsigned long start
, end
;
391 kmemleak_free_part(__va(addr
), size
);
393 start
= PFN_UP(addr
);
394 end
= PFN_DOWN(addr
+ size
);
396 mark_bootmem(start
, end
, 0, 0);
400 * reserve_bootmem_node - mark a page range as reserved
401 * @pgdat: node the range resides on
402 * @physaddr: starting address of the range
403 * @size: size of the range in bytes
404 * @flags: reservation flags (see linux/bootmem.h)
406 * Partial pages will be reserved.
408 * The range must reside completely on the specified node.
410 int __init
reserve_bootmem_node(pg_data_t
*pgdat
, unsigned long physaddr
,
411 unsigned long size
, int flags
)
413 unsigned long start
, end
;
415 start
= PFN_DOWN(physaddr
);
416 end
= PFN_UP(physaddr
+ size
);
418 return mark_bootmem_node(pgdat
->bdata
, start
, end
, 1, flags
);
422 * reserve_bootmem - mark a page range as usable
423 * @addr: starting address of the range
424 * @size: size of the range in bytes
425 * @flags: reservation flags (see linux/bootmem.h)
427 * Partial pages will be reserved.
429 * The range must be contiguous but may span node boundaries.
431 int __init
reserve_bootmem(unsigned long addr
, unsigned long size
,
434 unsigned long start
, end
;
436 start
= PFN_DOWN(addr
);
437 end
= PFN_UP(addr
+ size
);
439 return mark_bootmem(start
, end
, 1, flags
);
442 int __weak __init
reserve_bootmem_generic(unsigned long phys
, unsigned long len
,
445 return reserve_bootmem(phys
, len
, flags
);
448 static unsigned long __init
align_idx(struct bootmem_data
*bdata
,
449 unsigned long idx
, unsigned long step
)
451 unsigned long base
= bdata
->node_min_pfn
;
454 * Align the index with respect to the node start so that the
455 * combination of both satisfies the requested alignment.
458 return ALIGN(base
+ idx
, step
) - base
;
461 static unsigned long __init
align_off(struct bootmem_data
*bdata
,
462 unsigned long off
, unsigned long align
)
464 unsigned long base
= PFN_PHYS(bdata
->node_min_pfn
);
466 /* Same as align_idx for byte offsets */
468 return ALIGN(base
+ off
, align
) - base
;
471 static void * __init
alloc_bootmem_bdata(struct bootmem_data
*bdata
,
472 unsigned long size
, unsigned long align
,
473 unsigned long goal
, unsigned long limit
)
475 unsigned long fallback
= 0;
476 unsigned long min
, max
, start
, sidx
, midx
, step
;
478 bdebug("nid=%td size=%lx [%lu pages] align=%lx goal=%lx limit=%lx\n",
479 bdata
- bootmem_node_data
, size
, PAGE_ALIGN(size
) >> PAGE_SHIFT
,
483 BUG_ON(align
& (align
- 1));
484 BUG_ON(limit
&& goal
+ size
> limit
);
486 if (!bdata
->node_bootmem_map
)
489 min
= bdata
->node_min_pfn
;
490 max
= bdata
->node_low_pfn
;
493 limit
>>= PAGE_SHIFT
;
495 if (limit
&& max
> limit
)
500 step
= max(align
>> PAGE_SHIFT
, 1UL);
502 if (goal
&& min
< goal
&& goal
< max
)
503 start
= ALIGN(goal
, step
);
505 start
= ALIGN(min
, step
);
507 sidx
= start
- bdata
->node_min_pfn
;
508 midx
= max
- bdata
->node_min_pfn
;
510 if (bdata
->hint_idx
> sidx
) {
512 * Handle the valid case of sidx being zero and still
513 * catch the fallback below.
516 sidx
= align_idx(bdata
, bdata
->hint_idx
, step
);
522 unsigned long eidx
, i
, start_off
, end_off
;
524 sidx
= find_next_zero_bit(bdata
->node_bootmem_map
, midx
, sidx
);
525 sidx
= align_idx(bdata
, sidx
, step
);
526 eidx
= sidx
+ PFN_UP(size
);
528 if (sidx
>= midx
|| eidx
> midx
)
531 for (i
= sidx
; i
< eidx
; i
++)
532 if (test_bit(i
, bdata
->node_bootmem_map
)) {
533 sidx
= align_idx(bdata
, i
, step
);
539 if (bdata
->last_end_off
& (PAGE_SIZE
- 1) &&
540 PFN_DOWN(bdata
->last_end_off
) + 1 == sidx
)
541 start_off
= align_off(bdata
, bdata
->last_end_off
, align
);
543 start_off
= PFN_PHYS(sidx
);
545 merge
= PFN_DOWN(start_off
) < sidx
;
546 end_off
= start_off
+ size
;
548 bdata
->last_end_off
= end_off
;
549 bdata
->hint_idx
= PFN_UP(end_off
);
552 * Reserve the area now:
554 if (__reserve(bdata
, PFN_DOWN(start_off
) + merge
,
555 PFN_UP(end_off
), BOOTMEM_EXCLUSIVE
))
558 region
= phys_to_virt(PFN_PHYS(bdata
->node_min_pfn
) +
560 memset(region
, 0, size
);
562 * The min_count is set to 0 so that bootmem allocated blocks
563 * are never reported as leaks.
565 kmemleak_alloc(region
, size
, 0, 0);
570 sidx
= align_idx(bdata
, fallback
- 1, step
);
578 static void * __init
alloc_arch_preferred_bootmem(bootmem_data_t
*bdata
,
579 unsigned long size
, unsigned long align
,
580 unsigned long goal
, unsigned long limit
)
582 if (WARN_ON_ONCE(slab_is_available()))
583 return kzalloc(size
, GFP_NOWAIT
);
585 #ifdef CONFIG_HAVE_ARCH_BOOTMEM
587 bootmem_data_t
*p_bdata
;
589 p_bdata
= bootmem_arch_preferred_node(bdata
, size
, align
,
592 return alloc_bootmem_bdata(p_bdata
, size
, align
,
599 static void * __init
alloc_bootmem_core(unsigned long size
,
604 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_bdata(bdata
, size
, align
, goal
, limit
);
625 static void * __init
___alloc_bootmem_nopanic(unsigned long size
,
633 ptr
= alloc_bootmem_core(size
, align
, goal
, limit
);
645 * __alloc_bootmem_nopanic - allocate boot memory without panicking
646 * @size: size of the request in bytes
647 * @align: alignment of the region
648 * @goal: preferred starting address of the region
650 * The goal is dropped if it can not be satisfied and the allocation will
651 * fall back to memory below @goal.
653 * Allocation may happen on any node in the system.
655 * Returns NULL on failure.
657 void * __init
__alloc_bootmem_nopanic(unsigned long size
, unsigned long align
,
660 unsigned long limit
= 0;
662 return ___alloc_bootmem_nopanic(size
, align
, goal
, limit
);
665 static void * __init
___alloc_bootmem(unsigned long size
, unsigned long align
,
666 unsigned long goal
, unsigned long limit
)
668 void *mem
= ___alloc_bootmem_nopanic(size
, align
, goal
, limit
);
673 * Whoops, we cannot satisfy the allocation request.
675 printk(KERN_ALERT
"bootmem alloc of %lu bytes failed!\n", size
);
676 panic("Out of memory");
681 * __alloc_bootmem - allocate boot memory
682 * @size: size of the request in bytes
683 * @align: alignment of the region
684 * @goal: preferred starting address of the region
686 * The goal is dropped if it can not be satisfied and the allocation will
687 * fall back to memory below @goal.
689 * Allocation may happen on any node in the system.
691 * The function panics if the request can not be satisfied.
693 void * __init
__alloc_bootmem(unsigned long size
, unsigned long align
,
696 unsigned long limit
= 0;
698 return ___alloc_bootmem(size
, align
, goal
, limit
);
701 void * __init
___alloc_bootmem_node_nopanic(pg_data_t
*pgdat
,
702 unsigned long size
, unsigned long align
,
703 unsigned long goal
, unsigned long limit
)
708 ptr
= alloc_arch_preferred_bootmem(pgdat
->bdata
, size
,
713 /* do not panic in alloc_bootmem_bdata() */
714 if (limit
&& goal
+ size
> limit
)
717 ptr
= alloc_bootmem_bdata(pgdat
->bdata
, size
, align
, goal
, limit
);
721 ptr
= alloc_bootmem_core(size
, align
, goal
, limit
);
733 void * __init
__alloc_bootmem_node_nopanic(pg_data_t
*pgdat
, unsigned long size
,
734 unsigned long align
, unsigned long goal
)
736 if (WARN_ON_ONCE(slab_is_available()))
737 return kzalloc_node(size
, GFP_NOWAIT
, pgdat
->node_id
);
739 return ___alloc_bootmem_node_nopanic(pgdat
, size
, align
, goal
, 0);
742 void * __init
___alloc_bootmem_node(pg_data_t
*pgdat
, unsigned long size
,
743 unsigned long align
, unsigned long goal
,
748 ptr
= ___alloc_bootmem_node_nopanic(pgdat
, size
, align
, goal
, 0);
752 printk(KERN_ALERT
"bootmem alloc of %lu bytes failed!\n", size
);
753 panic("Out of memory");
758 * __alloc_bootmem_node - allocate boot memory from a specific node
759 * @pgdat: node to allocate from
760 * @size: size of the request in bytes
761 * @align: alignment of the region
762 * @goal: preferred starting address of the region
764 * The goal is dropped if it can not be satisfied and the allocation will
765 * fall back to memory below @goal.
767 * Allocation may fall back to any node in the system if the specified node
768 * can not hold the requested memory.
770 * The function panics if the request can not be satisfied.
772 void * __init
__alloc_bootmem_node(pg_data_t
*pgdat
, unsigned long size
,
773 unsigned long align
, unsigned long goal
)
775 if (WARN_ON_ONCE(slab_is_available()))
776 return kzalloc_node(size
, GFP_NOWAIT
, pgdat
->node_id
);
778 return ___alloc_bootmem_node(pgdat
, size
, align
, goal
, 0);
781 void * __init
__alloc_bootmem_node_high(pg_data_t
*pgdat
, unsigned long size
,
782 unsigned long align
, unsigned long goal
)
785 unsigned long end_pfn
;
787 if (WARN_ON_ONCE(slab_is_available()))
788 return kzalloc_node(size
, GFP_NOWAIT
, pgdat
->node_id
);
790 /* update goal according ...MAX_DMA32_PFN */
791 end_pfn
= pgdat
->node_start_pfn
+ pgdat
->node_spanned_pages
;
793 if (end_pfn
> MAX_DMA32_PFN
+ (128 >> (20 - PAGE_SHIFT
)) &&
794 (goal
>> PAGE_SHIFT
) < MAX_DMA32_PFN
) {
796 unsigned long new_goal
;
798 new_goal
= MAX_DMA32_PFN
<< PAGE_SHIFT
;
799 ptr
= alloc_bootmem_bdata(pgdat
->bdata
, size
, align
,
806 return __alloc_bootmem_node(pgdat
, size
, align
, goal
);
810 #ifndef ARCH_LOW_ADDRESS_LIMIT
811 #define ARCH_LOW_ADDRESS_LIMIT 0xffffffffUL
815 * __alloc_bootmem_low - allocate low boot memory
816 * @size: size of the request in bytes
817 * @align: alignment of the region
818 * @goal: preferred starting address of the region
820 * The goal is dropped if it can not be satisfied and the allocation will
821 * fall back to memory below @goal.
823 * Allocation may happen on any node in the system.
825 * The function panics if the request can not be satisfied.
827 void * __init
__alloc_bootmem_low(unsigned long size
, unsigned long align
,
830 return ___alloc_bootmem(size
, align
, goal
, ARCH_LOW_ADDRESS_LIMIT
);
834 * __alloc_bootmem_low_node - allocate low boot memory from a specific node
835 * @pgdat: node to allocate from
836 * @size: size of the request in bytes
837 * @align: alignment of the region
838 * @goal: preferred starting address of the region
840 * The goal is dropped if it can not be satisfied and the allocation will
841 * fall back to memory below @goal.
843 * Allocation may fall back to any node in the system if the specified node
844 * can not hold the requested memory.
846 * The function panics if the request can not be satisfied.
848 void * __init
__alloc_bootmem_low_node(pg_data_t
*pgdat
, unsigned long size
,
849 unsigned long align
, unsigned long goal
)
851 if (WARN_ON_ONCE(slab_is_available()))
852 return kzalloc_node(size
, GFP_NOWAIT
, pgdat
->node_id
);
854 return ___alloc_bootmem_node(pgdat
, size
, align
,
855 goal
, ARCH_LOW_ADDRESS_LIMIT
);