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
;
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
);
232 static void reset_node_lowmem_managed_pages(pg_data_t
*pgdat
)
237 * In free_area_init_core(), highmem zone's managed_pages is set to
238 * present_pages, and bootmem allocator doesn't allocate from highmem
239 * zones. So there's no need to recalculate managed_pages because all
240 * highmem pages will be managed by the buddy system. Here highmem
241 * zone also includes highmem movable zone.
243 for (z
= pgdat
->node_zones
; z
< pgdat
->node_zones
+ MAX_NR_ZONES
; z
++)
245 z
->managed_pages
= 0;
249 * free_all_bootmem_node - release a node's free pages to the buddy allocator
250 * @pgdat: node to be released
252 * Returns the number of pages actually released.
254 unsigned long __init
free_all_bootmem_node(pg_data_t
*pgdat
)
256 register_page_bootmem_info_node(pgdat
);
257 reset_node_lowmem_managed_pages(pgdat
);
258 return free_all_bootmem_core(pgdat
->bdata
);
262 * free_all_bootmem - release free pages to the buddy allocator
264 * Returns the number of pages actually released.
266 unsigned long __init
free_all_bootmem(void)
268 unsigned long total_pages
= 0;
269 bootmem_data_t
*bdata
;
270 struct pglist_data
*pgdat
;
272 for_each_online_pgdat(pgdat
)
273 reset_node_lowmem_managed_pages(pgdat
);
275 list_for_each_entry(bdata
, &bdata_list
, list
)
276 total_pages
+= free_all_bootmem_core(bdata
);
281 static void __init
__free(bootmem_data_t
*bdata
,
282 unsigned long sidx
, unsigned long eidx
)
286 bdebug("nid=%td start=%lx end=%lx\n", bdata
- bootmem_node_data
,
287 sidx
+ bdata
->node_min_pfn
,
288 eidx
+ bdata
->node_min_pfn
);
290 if (bdata
->hint_idx
> sidx
)
291 bdata
->hint_idx
= sidx
;
293 for (idx
= sidx
; idx
< eidx
; idx
++)
294 if (!test_and_clear_bit(idx
, bdata
->node_bootmem_map
))
298 static int __init
__reserve(bootmem_data_t
*bdata
, unsigned long sidx
,
299 unsigned long eidx
, int flags
)
302 int exclusive
= flags
& BOOTMEM_EXCLUSIVE
;
304 bdebug("nid=%td start=%lx end=%lx flags=%x\n",
305 bdata
- bootmem_node_data
,
306 sidx
+ bdata
->node_min_pfn
,
307 eidx
+ bdata
->node_min_pfn
,
310 for (idx
= sidx
; idx
< eidx
; idx
++)
311 if (test_and_set_bit(idx
, bdata
->node_bootmem_map
)) {
313 __free(bdata
, sidx
, idx
);
316 bdebug("silent double reserve of PFN %lx\n",
317 idx
+ bdata
->node_min_pfn
);
322 static int __init
mark_bootmem_node(bootmem_data_t
*bdata
,
323 unsigned long start
, unsigned long end
,
324 int reserve
, int flags
)
326 unsigned long sidx
, eidx
;
328 bdebug("nid=%td start=%lx end=%lx reserve=%d flags=%x\n",
329 bdata
- bootmem_node_data
, start
, end
, reserve
, flags
);
331 BUG_ON(start
< bdata
->node_min_pfn
);
332 BUG_ON(end
> bdata
->node_low_pfn
);
334 sidx
= start
- bdata
->node_min_pfn
;
335 eidx
= end
- bdata
->node_min_pfn
;
338 return __reserve(bdata
, sidx
, eidx
, flags
);
340 __free(bdata
, sidx
, eidx
);
344 static int __init
mark_bootmem(unsigned long start
, unsigned long end
,
345 int reserve
, int flags
)
348 bootmem_data_t
*bdata
;
351 list_for_each_entry(bdata
, &bdata_list
, list
) {
355 if (pos
< bdata
->node_min_pfn
||
356 pos
>= bdata
->node_low_pfn
) {
357 BUG_ON(pos
!= start
);
361 max
= min(bdata
->node_low_pfn
, end
);
363 err
= mark_bootmem_node(bdata
, pos
, max
, reserve
, flags
);
364 if (reserve
&& err
) {
365 mark_bootmem(start
, pos
, 0, 0);
371 pos
= bdata
->node_low_pfn
;
377 * free_bootmem_node - mark a page range as usable
378 * @pgdat: node the range resides on
379 * @physaddr: 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 reside completely on the specified node.
386 void __init
free_bootmem_node(pg_data_t
*pgdat
, unsigned long physaddr
,
389 unsigned long start
, end
;
391 kmemleak_free_part(__va(physaddr
), size
);
393 start
= PFN_UP(physaddr
);
394 end
= PFN_DOWN(physaddr
+ size
);
396 mark_bootmem_node(pgdat
->bdata
, start
, end
, 0, 0);
400 * free_bootmem - mark a page range as usable
401 * @addr: starting physical address of the range
402 * @size: size of the range in bytes
404 * Partial pages will be considered reserved and left as they are.
406 * The range must be contiguous but may span node boundaries.
408 void __init
free_bootmem(unsigned long physaddr
, unsigned long size
)
410 unsigned long start
, end
;
412 kmemleak_free_part(__va(physaddr
), size
);
414 start
= PFN_UP(physaddr
);
415 end
= PFN_DOWN(physaddr
+ size
);
417 mark_bootmem(start
, end
, 0, 0);
421 * reserve_bootmem_node - mark a page range as reserved
422 * @pgdat: node the range resides on
423 * @physaddr: 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 reside completely on the specified node.
431 int __init
reserve_bootmem_node(pg_data_t
*pgdat
, unsigned long physaddr
,
432 unsigned long size
, int flags
)
434 unsigned long start
, end
;
436 start
= PFN_DOWN(physaddr
);
437 end
= PFN_UP(physaddr
+ size
);
439 return mark_bootmem_node(pgdat
->bdata
, start
, end
, 1, flags
);
443 * reserve_bootmem - mark a page range as reserved
444 * @addr: starting address of the range
445 * @size: size of the range in bytes
446 * @flags: reservation flags (see linux/bootmem.h)
448 * Partial pages will be reserved.
450 * The range must be contiguous but may span node boundaries.
452 int __init
reserve_bootmem(unsigned long addr
, unsigned long size
,
455 unsigned long start
, end
;
457 start
= PFN_DOWN(addr
);
458 end
= PFN_UP(addr
+ size
);
460 return mark_bootmem(start
, end
, 1, flags
);
463 static unsigned long __init
align_idx(struct bootmem_data
*bdata
,
464 unsigned long idx
, unsigned long step
)
466 unsigned long base
= bdata
->node_min_pfn
;
469 * Align the index with respect to the node start so that the
470 * combination of both satisfies the requested alignment.
473 return ALIGN(base
+ idx
, step
) - base
;
476 static unsigned long __init
align_off(struct bootmem_data
*bdata
,
477 unsigned long off
, unsigned long align
)
479 unsigned long base
= PFN_PHYS(bdata
->node_min_pfn
);
481 /* Same as align_idx for byte offsets */
483 return ALIGN(base
+ off
, align
) - base
;
486 static void * __init
alloc_bootmem_bdata(struct bootmem_data
*bdata
,
487 unsigned long size
, unsigned long align
,
488 unsigned long goal
, unsigned long limit
)
490 unsigned long fallback
= 0;
491 unsigned long min
, max
, start
, sidx
, midx
, step
;
493 bdebug("nid=%td size=%lx [%lu pages] align=%lx goal=%lx limit=%lx\n",
494 bdata
- bootmem_node_data
, size
, PAGE_ALIGN(size
) >> PAGE_SHIFT
,
498 BUG_ON(align
& (align
- 1));
499 BUG_ON(limit
&& goal
+ size
> limit
);
501 if (!bdata
->node_bootmem_map
)
504 min
= bdata
->node_min_pfn
;
505 max
= bdata
->node_low_pfn
;
508 limit
>>= PAGE_SHIFT
;
510 if (limit
&& max
> limit
)
515 step
= max(align
>> PAGE_SHIFT
, 1UL);
517 if (goal
&& min
< goal
&& goal
< max
)
518 start
= ALIGN(goal
, step
);
520 start
= ALIGN(min
, step
);
522 sidx
= start
- bdata
->node_min_pfn
;
523 midx
= max
- bdata
->node_min_pfn
;
525 if (bdata
->hint_idx
> sidx
) {
527 * Handle the valid case of sidx being zero and still
528 * catch the fallback below.
531 sidx
= align_idx(bdata
, bdata
->hint_idx
, step
);
537 unsigned long eidx
, i
, start_off
, end_off
;
539 sidx
= find_next_zero_bit(bdata
->node_bootmem_map
, midx
, sidx
);
540 sidx
= align_idx(bdata
, sidx
, step
);
541 eidx
= sidx
+ PFN_UP(size
);
543 if (sidx
>= midx
|| eidx
> midx
)
546 for (i
= sidx
; i
< eidx
; i
++)
547 if (test_bit(i
, bdata
->node_bootmem_map
)) {
548 sidx
= align_idx(bdata
, i
, step
);
554 if (bdata
->last_end_off
& (PAGE_SIZE
- 1) &&
555 PFN_DOWN(bdata
->last_end_off
) + 1 == sidx
)
556 start_off
= align_off(bdata
, bdata
->last_end_off
, align
);
558 start_off
= PFN_PHYS(sidx
);
560 merge
= PFN_DOWN(start_off
) < sidx
;
561 end_off
= start_off
+ size
;
563 bdata
->last_end_off
= end_off
;
564 bdata
->hint_idx
= PFN_UP(end_off
);
567 * Reserve the area now:
569 if (__reserve(bdata
, PFN_DOWN(start_off
) + merge
,
570 PFN_UP(end_off
), BOOTMEM_EXCLUSIVE
))
573 region
= phys_to_virt(PFN_PHYS(bdata
->node_min_pfn
) +
575 memset(region
, 0, size
);
577 * The min_count is set to 0 so that bootmem allocated blocks
578 * are never reported as leaks.
580 kmemleak_alloc(region
, size
, 0, 0);
585 sidx
= align_idx(bdata
, fallback
- 1, step
);
593 static void * __init
alloc_bootmem_core(unsigned long size
,
598 bootmem_data_t
*bdata
;
601 if (WARN_ON_ONCE(slab_is_available()))
602 return kzalloc(size
, GFP_NOWAIT
);
604 list_for_each_entry(bdata
, &bdata_list
, list
) {
605 if (goal
&& bdata
->node_low_pfn
<= PFN_DOWN(goal
))
607 if (limit
&& bdata
->node_min_pfn
>= PFN_DOWN(limit
))
610 region
= alloc_bootmem_bdata(bdata
, size
, align
, goal
, limit
);
618 static void * __init
___alloc_bootmem_nopanic(unsigned long size
,
626 ptr
= alloc_bootmem_core(size
, align
, goal
, limit
);
638 * __alloc_bootmem_nopanic - allocate boot memory without panicking
639 * @size: size of the request in bytes
640 * @align: alignment of the region
641 * @goal: preferred starting address of the region
643 * The goal is dropped if it can not be satisfied and the allocation will
644 * fall back to memory below @goal.
646 * Allocation may happen on any node in the system.
648 * Returns NULL on failure.
650 void * __init
__alloc_bootmem_nopanic(unsigned long size
, unsigned long align
,
653 unsigned long limit
= 0;
655 return ___alloc_bootmem_nopanic(size
, align
, goal
, limit
);
658 static void * __init
___alloc_bootmem(unsigned long size
, unsigned long align
,
659 unsigned long goal
, unsigned long limit
)
661 void *mem
= ___alloc_bootmem_nopanic(size
, align
, goal
, limit
);
666 * Whoops, we cannot satisfy the allocation request.
668 printk(KERN_ALERT
"bootmem alloc of %lu bytes failed!\n", size
);
669 panic("Out of memory");
674 * __alloc_bootmem - allocate boot memory
675 * @size: size of the request in bytes
676 * @align: alignment of the region
677 * @goal: preferred starting address of the region
679 * The goal is dropped if it can not be satisfied and the allocation will
680 * fall back to memory below @goal.
682 * Allocation may happen on any node in the system.
684 * The function panics if the request can not be satisfied.
686 void * __init
__alloc_bootmem(unsigned long size
, unsigned long align
,
689 unsigned long limit
= 0;
691 return ___alloc_bootmem(size
, align
, goal
, limit
);
694 void * __init
___alloc_bootmem_node_nopanic(pg_data_t
*pgdat
,
695 unsigned long size
, unsigned long align
,
696 unsigned long goal
, unsigned long limit
)
700 if (WARN_ON_ONCE(slab_is_available()))
701 return kzalloc(size
, GFP_NOWAIT
);
704 /* do not panic in alloc_bootmem_bdata() */
705 if (limit
&& goal
+ size
> limit
)
708 ptr
= alloc_bootmem_bdata(pgdat
->bdata
, size
, align
, goal
, limit
);
712 ptr
= alloc_bootmem_core(size
, align
, goal
, limit
);
724 void * __init
__alloc_bootmem_node_nopanic(pg_data_t
*pgdat
, unsigned long size
,
725 unsigned long align
, unsigned long goal
)
727 if (WARN_ON_ONCE(slab_is_available()))
728 return kzalloc_node(size
, GFP_NOWAIT
, pgdat
->node_id
);
730 return ___alloc_bootmem_node_nopanic(pgdat
, size
, align
, goal
, 0);
733 void * __init
___alloc_bootmem_node(pg_data_t
*pgdat
, unsigned long size
,
734 unsigned long align
, unsigned long goal
,
739 ptr
= ___alloc_bootmem_node_nopanic(pgdat
, size
, align
, goal
, 0);
743 printk(KERN_ALERT
"bootmem alloc of %lu bytes failed!\n", size
);
744 panic("Out of memory");
749 * __alloc_bootmem_node - allocate boot memory from a specific node
750 * @pgdat: node to allocate from
751 * @size: size of the request in bytes
752 * @align: alignment of the region
753 * @goal: preferred starting address of the region
755 * The goal is dropped if it can not be satisfied and the allocation will
756 * fall back to memory below @goal.
758 * Allocation may fall back to any node in the system if the specified node
759 * can not hold the requested memory.
761 * The function panics if the request can not be satisfied.
763 void * __init
__alloc_bootmem_node(pg_data_t
*pgdat
, unsigned long size
,
764 unsigned long align
, unsigned long goal
)
766 if (WARN_ON_ONCE(slab_is_available()))
767 return kzalloc_node(size
, GFP_NOWAIT
, pgdat
->node_id
);
769 return ___alloc_bootmem_node(pgdat
, size
, align
, goal
, 0);
772 void * __init
__alloc_bootmem_node_high(pg_data_t
*pgdat
, unsigned long size
,
773 unsigned long align
, unsigned long goal
)
776 unsigned long end_pfn
;
778 if (WARN_ON_ONCE(slab_is_available()))
779 return kzalloc_node(size
, GFP_NOWAIT
, pgdat
->node_id
);
781 /* update goal according ...MAX_DMA32_PFN */
782 end_pfn
= pgdat
->node_start_pfn
+ pgdat
->node_spanned_pages
;
784 if (end_pfn
> MAX_DMA32_PFN
+ (128 >> (20 - PAGE_SHIFT
)) &&
785 (goal
>> PAGE_SHIFT
) < MAX_DMA32_PFN
) {
787 unsigned long new_goal
;
789 new_goal
= MAX_DMA32_PFN
<< PAGE_SHIFT
;
790 ptr
= alloc_bootmem_bdata(pgdat
->bdata
, size
, align
,
797 return __alloc_bootmem_node(pgdat
, size
, align
, goal
);
801 #ifndef ARCH_LOW_ADDRESS_LIMIT
802 #define ARCH_LOW_ADDRESS_LIMIT 0xffffffffUL
806 * __alloc_bootmem_low - allocate low boot memory
807 * @size: size of the request in bytes
808 * @align: alignment of the region
809 * @goal: preferred starting address of the region
811 * The goal is dropped if it can not be satisfied and the allocation will
812 * fall back to memory below @goal.
814 * Allocation may happen on any node in the system.
816 * The function panics if the request can not be satisfied.
818 void * __init
__alloc_bootmem_low(unsigned long size
, unsigned long align
,
821 return ___alloc_bootmem(size
, align
, goal
, ARCH_LOW_ADDRESS_LIMIT
);
825 * __alloc_bootmem_low_node - allocate low boot memory from a specific node
826 * @pgdat: node to allocate from
827 * @size: size of the request in bytes
828 * @align: alignment of the region
829 * @goal: preferred starting address of the region
831 * The goal is dropped if it can not be satisfied and the allocation will
832 * fall back to memory below @goal.
834 * Allocation may fall back to any node in the system if the specified node
835 * can not hold the requested memory.
837 * The function panics if the request can not be satisfied.
839 void * __init
__alloc_bootmem_low_node(pg_data_t
*pgdat
, unsigned long size
,
840 unsigned long align
, unsigned long goal
)
842 if (WARN_ON_ONCE(slab_is_available()))
843 return kzalloc_node(size
, GFP_NOWAIT
, pgdat
->node_id
);
845 return ___alloc_bootmem_node(pgdat
, size
, align
,
846 goal
, ARCH_LOW_ADDRESS_LIMIT
);