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/export.h>
15 #include <linux/kmemleak.h>
16 #include <linux/range.h>
17 #include <linux/bug.h>
19 #include <linux/bootmem.h>
23 #ifndef CONFIG_NEED_MULTIPLE_NODES
24 struct pglist_data __refdata contig_page_data
= {
25 .bdata
= &bootmem_node_data
[0]
27 EXPORT_SYMBOL(contig_page_data
);
30 unsigned long max_low_pfn
;
31 unsigned long min_low_pfn
;
32 unsigned long max_pfn
;
33 unsigned long long max_possible_pfn
;
35 bootmem_data_t bootmem_node_data
[MAX_NUMNODES
] __initdata
;
37 static struct list_head bdata_list __initdata
= LIST_HEAD_INIT(bdata_list
);
39 static int bootmem_debug
;
41 static int __init
bootmem_debug_setup(char *buf
)
46 early_param("bootmem_debug", bootmem_debug_setup
);
48 #define bdebug(fmt, args...) ({ \
49 if (unlikely(bootmem_debug)) \
50 pr_info("bootmem::%s " fmt, \
54 static unsigned long __init
bootmap_bytes(unsigned long pages
)
56 unsigned long bytes
= DIV_ROUND_UP(pages
, BITS_PER_BYTE
);
58 return ALIGN(bytes
, sizeof(long));
62 * bootmem_bootmap_pages - calculate bitmap size in pages
63 * @pages: number of pages the bitmap has to represent
65 unsigned long __init
bootmem_bootmap_pages(unsigned long pages
)
67 unsigned long bytes
= bootmap_bytes(pages
);
69 return PAGE_ALIGN(bytes
) >> PAGE_SHIFT
;
75 static void __init
link_bootmem(bootmem_data_t
*bdata
)
79 list_for_each_entry(ent
, &bdata_list
, list
) {
80 if (bdata
->node_min_pfn
< ent
->node_min_pfn
) {
81 list_add_tail(&bdata
->list
, &ent
->list
);
86 list_add_tail(&bdata
->list
, &bdata_list
);
90 * Called once to set up the allocator itself.
92 static unsigned long __init
init_bootmem_core(bootmem_data_t
*bdata
,
93 unsigned long mapstart
, unsigned long start
, unsigned long end
)
95 unsigned long mapsize
;
97 mminit_validate_memmodel_limits(&start
, &end
);
98 bdata
->node_bootmem_map
= phys_to_virt(PFN_PHYS(mapstart
));
99 bdata
->node_min_pfn
= start
;
100 bdata
->node_low_pfn
= end
;
104 * Initially all pages are reserved - setup_arch() has to
105 * register free RAM areas explicitly.
107 mapsize
= bootmap_bytes(end
- start
);
108 memset(bdata
->node_bootmem_map
, 0xff, mapsize
);
110 bdebug("nid=%td start=%lx map=%lx end=%lx mapsize=%lx\n",
111 bdata
- bootmem_node_data
, start
, mapstart
, end
, mapsize
);
117 * init_bootmem_node - register a node as boot memory
118 * @pgdat: node to register
119 * @freepfn: pfn where the bitmap for this node is to be placed
120 * @startpfn: first pfn on the node
121 * @endpfn: first pfn after the node
123 * Returns the number of bytes needed to hold the bitmap for this node.
125 unsigned long __init
init_bootmem_node(pg_data_t
*pgdat
, unsigned long freepfn
,
126 unsigned long startpfn
, unsigned long endpfn
)
128 return init_bootmem_core(pgdat
->bdata
, freepfn
, startpfn
, endpfn
);
132 * init_bootmem - register boot memory
133 * @start: pfn where the bitmap is to be placed
134 * @pages: number of available physical pages
136 * Returns the number of bytes needed to hold the bitmap.
138 unsigned long __init
init_bootmem(unsigned long start
, unsigned long pages
)
142 return init_bootmem_core(NODE_DATA(0)->bdata
, start
, 0, pages
);
146 * free_bootmem_late - free bootmem pages directly to page allocator
147 * @addr: starting physical address of the range
148 * @size: size of the range in bytes
150 * This is only useful when the bootmem allocator has already been torn
151 * down, but we are still initializing the system. Pages are given directly
152 * to the page allocator, no bootmem metadata is updated because it is gone.
154 void __init
free_bootmem_late(unsigned long physaddr
, unsigned long size
)
156 unsigned long cursor
, end
;
158 kmemleak_free_part_phys(physaddr
, size
);
160 cursor
= PFN_UP(physaddr
);
161 end
= PFN_DOWN(physaddr
+ size
);
163 for (; cursor
< end
; cursor
++) {
164 __free_pages_bootmem(pfn_to_page(cursor
), cursor
, 0);
169 static unsigned long __init
free_all_bootmem_core(bootmem_data_t
*bdata
)
172 unsigned long *map
, start
, end
, pages
, cur
, count
= 0;
174 if (!bdata
->node_bootmem_map
)
177 map
= bdata
->node_bootmem_map
;
178 start
= bdata
->node_min_pfn
;
179 end
= bdata
->node_low_pfn
;
181 bdebug("nid=%td start=%lx end=%lx\n",
182 bdata
- bootmem_node_data
, start
, end
);
184 while (start
< end
) {
185 unsigned long idx
, vec
;
188 idx
= start
- bdata
->node_min_pfn
;
189 shift
= idx
& (BITS_PER_LONG
- 1);
191 * vec holds at most BITS_PER_LONG map bits,
192 * bit 0 corresponds to start.
194 vec
= ~map
[idx
/ BITS_PER_LONG
];
198 if (end
- start
>= BITS_PER_LONG
)
199 vec
|= ~map
[idx
/ BITS_PER_LONG
+ 1] <<
200 (BITS_PER_LONG
- shift
);
203 * If we have a properly aligned and fully unreserved
204 * BITS_PER_LONG block of pages in front of us, free
207 if (IS_ALIGNED(start
, BITS_PER_LONG
) && vec
== ~0UL) {
208 int order
= ilog2(BITS_PER_LONG
);
210 __free_pages_bootmem(pfn_to_page(start
), start
, order
);
211 count
+= BITS_PER_LONG
;
212 start
+= BITS_PER_LONG
;
216 start
= ALIGN(start
+ 1, BITS_PER_LONG
);
217 while (vec
&& cur
!= start
) {
219 page
= pfn_to_page(cur
);
220 __free_pages_bootmem(page
, cur
, 0);
229 cur
= bdata
->node_min_pfn
;
230 page
= virt_to_page(bdata
->node_bootmem_map
);
231 pages
= bdata
->node_low_pfn
- bdata
->node_min_pfn
;
232 pages
= bootmem_bootmap_pages(pages
);
235 __free_pages_bootmem(page
++, cur
++, 0);
236 bdata
->node_bootmem_map
= NULL
;
238 bdebug("nid=%td released=%lx\n", bdata
- bootmem_node_data
, count
);
243 static int reset_managed_pages_done __initdata
;
245 void reset_node_managed_pages(pg_data_t
*pgdat
)
249 for (z
= pgdat
->node_zones
; z
< pgdat
->node_zones
+ MAX_NR_ZONES
; z
++)
250 z
->managed_pages
= 0;
253 void __init
reset_all_zones_managed_pages(void)
255 struct pglist_data
*pgdat
;
257 if (reset_managed_pages_done
)
260 for_each_online_pgdat(pgdat
)
261 reset_node_managed_pages(pgdat
);
263 reset_managed_pages_done
= 1;
267 * free_all_bootmem - release free pages to the buddy allocator
269 * Returns the number of pages actually released.
271 unsigned long __init
free_all_bootmem(void)
273 unsigned long total_pages
= 0;
274 bootmem_data_t
*bdata
;
276 reset_all_zones_managed_pages();
278 list_for_each_entry(bdata
, &bdata_list
, list
)
279 total_pages
+= free_all_bootmem_core(bdata
);
281 totalram_pages
+= total_pages
;
286 static void __init
__free(bootmem_data_t
*bdata
,
287 unsigned long sidx
, unsigned long eidx
)
291 bdebug("nid=%td start=%lx end=%lx\n", bdata
- bootmem_node_data
,
292 sidx
+ bdata
->node_min_pfn
,
293 eidx
+ bdata
->node_min_pfn
);
295 if (WARN_ON(bdata
->node_bootmem_map
== NULL
))
298 if (bdata
->hint_idx
> sidx
)
299 bdata
->hint_idx
= sidx
;
301 for (idx
= sidx
; idx
< eidx
; idx
++)
302 if (!test_and_clear_bit(idx
, bdata
->node_bootmem_map
))
306 static int __init
__reserve(bootmem_data_t
*bdata
, unsigned long sidx
,
307 unsigned long eidx
, int flags
)
310 int exclusive
= flags
& BOOTMEM_EXCLUSIVE
;
312 bdebug("nid=%td start=%lx end=%lx flags=%x\n",
313 bdata
- bootmem_node_data
,
314 sidx
+ bdata
->node_min_pfn
,
315 eidx
+ bdata
->node_min_pfn
,
318 if (WARN_ON(bdata
->node_bootmem_map
== NULL
))
321 for (idx
= sidx
; idx
< eidx
; idx
++)
322 if (test_and_set_bit(idx
, bdata
->node_bootmem_map
)) {
324 __free(bdata
, sidx
, idx
);
327 bdebug("silent double reserve of PFN %lx\n",
328 idx
+ bdata
->node_min_pfn
);
333 static int __init
mark_bootmem_node(bootmem_data_t
*bdata
,
334 unsigned long start
, unsigned long end
,
335 int reserve
, int flags
)
337 unsigned long sidx
, eidx
;
339 bdebug("nid=%td start=%lx end=%lx reserve=%d flags=%x\n",
340 bdata
- bootmem_node_data
, start
, end
, reserve
, flags
);
342 BUG_ON(start
< bdata
->node_min_pfn
);
343 BUG_ON(end
> bdata
->node_low_pfn
);
345 sidx
= start
- bdata
->node_min_pfn
;
346 eidx
= end
- bdata
->node_min_pfn
;
349 return __reserve(bdata
, sidx
, eidx
, flags
);
351 __free(bdata
, sidx
, eidx
);
355 static int __init
mark_bootmem(unsigned long start
, unsigned long end
,
356 int reserve
, int flags
)
359 bootmem_data_t
*bdata
;
362 list_for_each_entry(bdata
, &bdata_list
, list
) {
366 if (pos
< bdata
->node_min_pfn
||
367 pos
>= bdata
->node_low_pfn
) {
368 BUG_ON(pos
!= start
);
372 max
= min(bdata
->node_low_pfn
, end
);
374 err
= mark_bootmem_node(bdata
, pos
, max
, reserve
, flags
);
375 if (reserve
&& err
) {
376 mark_bootmem(start
, pos
, 0, 0);
382 pos
= bdata
->node_low_pfn
;
388 * free_bootmem_node - mark a page range as usable
389 * @pgdat: node the range resides on
390 * @physaddr: starting address of the range
391 * @size: size of the range in bytes
393 * Partial pages will be considered reserved and left as they are.
395 * The range must reside completely on the specified node.
397 void __init
free_bootmem_node(pg_data_t
*pgdat
, unsigned long physaddr
,
400 unsigned long start
, end
;
402 kmemleak_free_part_phys(physaddr
, size
);
404 start
= PFN_UP(physaddr
);
405 end
= PFN_DOWN(physaddr
+ size
);
407 mark_bootmem_node(pgdat
->bdata
, start
, end
, 0, 0);
411 * free_bootmem - mark a page range as usable
412 * @addr: starting physical address of the range
413 * @size: size of the range in bytes
415 * Partial pages will be considered reserved and left as they are.
417 * The range must be contiguous but may span node boundaries.
419 void __init
free_bootmem(unsigned long physaddr
, unsigned long size
)
421 unsigned long start
, end
;
423 kmemleak_free_part_phys(physaddr
, size
);
425 start
= PFN_UP(physaddr
);
426 end
= PFN_DOWN(physaddr
+ size
);
428 mark_bootmem(start
, end
, 0, 0);
432 * reserve_bootmem_node - mark a page range as reserved
433 * @pgdat: node the range resides on
434 * @physaddr: starting address of the range
435 * @size: size of the range in bytes
436 * @flags: reservation flags (see linux/bootmem.h)
438 * Partial pages will be reserved.
440 * The range must reside completely on the specified node.
442 int __init
reserve_bootmem_node(pg_data_t
*pgdat
, unsigned long physaddr
,
443 unsigned long size
, int flags
)
445 unsigned long start
, end
;
447 start
= PFN_DOWN(physaddr
);
448 end
= PFN_UP(physaddr
+ size
);
450 return mark_bootmem_node(pgdat
->bdata
, start
, end
, 1, flags
);
454 * reserve_bootmem - mark a page range as reserved
455 * @addr: starting address of the range
456 * @size: size of the range in bytes
457 * @flags: reservation flags (see linux/bootmem.h)
459 * Partial pages will be reserved.
461 * The range must be contiguous but may span node boundaries.
463 int __init
reserve_bootmem(unsigned long addr
, unsigned long size
,
466 unsigned long start
, end
;
468 start
= PFN_DOWN(addr
);
469 end
= PFN_UP(addr
+ size
);
471 return mark_bootmem(start
, end
, 1, flags
);
474 static unsigned long __init
align_idx(struct bootmem_data
*bdata
,
475 unsigned long idx
, unsigned long step
)
477 unsigned long base
= bdata
->node_min_pfn
;
480 * Align the index with respect to the node start so that the
481 * combination of both satisfies the requested alignment.
484 return ALIGN(base
+ idx
, step
) - base
;
487 static unsigned long __init
align_off(struct bootmem_data
*bdata
,
488 unsigned long off
, unsigned long align
)
490 unsigned long base
= PFN_PHYS(bdata
->node_min_pfn
);
492 /* Same as align_idx for byte offsets */
494 return ALIGN(base
+ off
, align
) - base
;
497 static void * __init
alloc_bootmem_bdata(struct bootmem_data
*bdata
,
498 unsigned long size
, unsigned long align
,
499 unsigned long goal
, unsigned long limit
)
501 unsigned long fallback
= 0;
502 unsigned long min
, max
, start
, sidx
, midx
, step
;
504 bdebug("nid=%td size=%lx [%lu pages] align=%lx goal=%lx limit=%lx\n",
505 bdata
- bootmem_node_data
, size
, PAGE_ALIGN(size
) >> PAGE_SHIFT
,
509 BUG_ON(align
& (align
- 1));
510 BUG_ON(limit
&& goal
+ size
> limit
);
512 if (!bdata
->node_bootmem_map
)
515 min
= bdata
->node_min_pfn
;
516 max
= bdata
->node_low_pfn
;
519 limit
>>= PAGE_SHIFT
;
521 if (limit
&& max
> limit
)
526 step
= max(align
>> PAGE_SHIFT
, 1UL);
528 if (goal
&& min
< goal
&& goal
< max
)
529 start
= ALIGN(goal
, step
);
531 start
= ALIGN(min
, step
);
533 sidx
= start
- bdata
->node_min_pfn
;
534 midx
= max
- bdata
->node_min_pfn
;
536 if (bdata
->hint_idx
> sidx
) {
538 * Handle the valid case of sidx being zero and still
539 * catch the fallback below.
542 sidx
= align_idx(bdata
, bdata
->hint_idx
, step
);
548 unsigned long eidx
, i
, start_off
, end_off
;
550 sidx
= find_next_zero_bit(bdata
->node_bootmem_map
, midx
, sidx
);
551 sidx
= align_idx(bdata
, sidx
, step
);
552 eidx
= sidx
+ PFN_UP(size
);
554 if (sidx
>= midx
|| eidx
> midx
)
557 for (i
= sidx
; i
< eidx
; i
++)
558 if (test_bit(i
, bdata
->node_bootmem_map
)) {
559 sidx
= align_idx(bdata
, i
, step
);
565 if (bdata
->last_end_off
& (PAGE_SIZE
- 1) &&
566 PFN_DOWN(bdata
->last_end_off
) + 1 == sidx
)
567 start_off
= align_off(bdata
, bdata
->last_end_off
, align
);
569 start_off
= PFN_PHYS(sidx
);
571 merge
= PFN_DOWN(start_off
) < sidx
;
572 end_off
= start_off
+ size
;
574 bdata
->last_end_off
= end_off
;
575 bdata
->hint_idx
= PFN_UP(end_off
);
578 * Reserve the area now:
580 if (__reserve(bdata
, PFN_DOWN(start_off
) + merge
,
581 PFN_UP(end_off
), BOOTMEM_EXCLUSIVE
))
584 region
= phys_to_virt(PFN_PHYS(bdata
->node_min_pfn
) +
586 memset(region
, 0, size
);
588 * The min_count is set to 0 so that bootmem allocated blocks
589 * are never reported as leaks.
591 kmemleak_alloc(region
, size
, 0, 0);
596 sidx
= align_idx(bdata
, fallback
- 1, step
);
604 static void * __init
alloc_bootmem_core(unsigned long size
,
609 bootmem_data_t
*bdata
;
612 if (WARN_ON_ONCE(slab_is_available()))
613 return kzalloc(size
, GFP_NOWAIT
);
615 list_for_each_entry(bdata
, &bdata_list
, list
) {
616 if (goal
&& bdata
->node_low_pfn
<= PFN_DOWN(goal
))
618 if (limit
&& bdata
->node_min_pfn
>= PFN_DOWN(limit
))
621 region
= alloc_bootmem_bdata(bdata
, size
, align
, goal
, limit
);
629 static void * __init
___alloc_bootmem_nopanic(unsigned long size
,
637 ptr
= alloc_bootmem_core(size
, align
, goal
, limit
);
649 * __alloc_bootmem_nopanic - allocate boot memory without panicking
650 * @size: size of the request in bytes
651 * @align: alignment of the region
652 * @goal: preferred starting address of the region
654 * The goal is dropped if it can not be satisfied and the allocation will
655 * fall back to memory below @goal.
657 * Allocation may happen on any node in the system.
659 * Returns NULL on failure.
661 void * __init
__alloc_bootmem_nopanic(unsigned long size
, unsigned long align
,
664 unsigned long limit
= 0;
666 return ___alloc_bootmem_nopanic(size
, align
, goal
, limit
);
669 static void * __init
___alloc_bootmem(unsigned long size
, unsigned long align
,
670 unsigned long goal
, unsigned long limit
)
672 void *mem
= ___alloc_bootmem_nopanic(size
, align
, goal
, limit
);
677 * Whoops, we cannot satisfy the allocation request.
679 pr_alert("bootmem alloc of %lu bytes failed!\n", size
);
680 panic("Out of memory");
685 * __alloc_bootmem - allocate boot memory
686 * @size: size of the request in bytes
687 * @align: alignment of the region
688 * @goal: preferred starting address of the region
690 * The goal is dropped if it can not be satisfied and the allocation will
691 * fall back to memory below @goal.
693 * Allocation may happen on any node in the system.
695 * The function panics if the request can not be satisfied.
697 void * __init
__alloc_bootmem(unsigned long size
, unsigned long align
,
700 unsigned long limit
= 0;
702 return ___alloc_bootmem(size
, align
, goal
, limit
);
705 void * __init
___alloc_bootmem_node_nopanic(pg_data_t
*pgdat
,
706 unsigned long size
, unsigned long align
,
707 unsigned long goal
, unsigned long limit
)
711 if (WARN_ON_ONCE(slab_is_available()))
712 return kzalloc_node(size
, GFP_NOWAIT
, pgdat
->node_id
);
715 /* do not panic in alloc_bootmem_bdata() */
716 if (limit
&& goal
+ size
> limit
)
719 ptr
= alloc_bootmem_bdata(pgdat
->bdata
, size
, align
, goal
, limit
);
723 ptr
= alloc_bootmem_core(size
, align
, goal
, limit
);
735 void * __init
__alloc_bootmem_node_nopanic(pg_data_t
*pgdat
, unsigned long size
,
736 unsigned long align
, unsigned long goal
)
738 return ___alloc_bootmem_node_nopanic(pgdat
, size
, align
, goal
, 0);
741 void * __init
___alloc_bootmem_node(pg_data_t
*pgdat
, unsigned long size
,
742 unsigned long align
, unsigned long goal
,
747 ptr
= ___alloc_bootmem_node_nopanic(pgdat
, size
, align
, goal
, 0);
751 pr_alert("bootmem alloc of %lu bytes failed!\n", size
);
752 panic("Out of memory");
757 * __alloc_bootmem_node - allocate boot memory from a specific node
758 * @pgdat: node to allocate from
759 * @size: size of the request in bytes
760 * @align: alignment of the region
761 * @goal: preferred starting address of the region
763 * The goal is dropped if it can not be satisfied and the allocation will
764 * fall back to memory below @goal.
766 * Allocation may fall back to any node in the system if the specified node
767 * can not hold the requested memory.
769 * The function panics if the request can not be satisfied.
771 void * __init
__alloc_bootmem_node(pg_data_t
*pgdat
, unsigned long size
,
772 unsigned long align
, unsigned long goal
)
774 if (WARN_ON_ONCE(slab_is_available()))
775 return kzalloc_node(size
, GFP_NOWAIT
, pgdat
->node_id
);
777 return ___alloc_bootmem_node(pgdat
, size
, align
, goal
, 0);
780 void * __init
__alloc_bootmem_node_high(pg_data_t
*pgdat
, unsigned long size
,
781 unsigned long align
, unsigned long goal
)
784 unsigned long end_pfn
;
786 if (WARN_ON_ONCE(slab_is_available()))
787 return kzalloc_node(size
, GFP_NOWAIT
, pgdat
->node_id
);
789 /* update goal according ...MAX_DMA32_PFN */
790 end_pfn
= pgdat_end_pfn(pgdat
);
792 if (end_pfn
> MAX_DMA32_PFN
+ (128 >> (20 - PAGE_SHIFT
)) &&
793 (goal
>> PAGE_SHIFT
) < MAX_DMA32_PFN
) {
795 unsigned long new_goal
;
797 new_goal
= MAX_DMA32_PFN
<< PAGE_SHIFT
;
798 ptr
= alloc_bootmem_bdata(pgdat
->bdata
, size
, align
,
805 return __alloc_bootmem_node(pgdat
, size
, align
, goal
);
810 * __alloc_bootmem_low - allocate low boot memory
811 * @size: size of the request in bytes
812 * @align: alignment of the region
813 * @goal: preferred starting address of the region
815 * The goal is dropped if it can not be satisfied and the allocation will
816 * fall back to memory below @goal.
818 * Allocation may happen on any node in the system.
820 * The function panics if the request can not be satisfied.
822 void * __init
__alloc_bootmem_low(unsigned long size
, unsigned long align
,
825 return ___alloc_bootmem(size
, align
, goal
, ARCH_LOW_ADDRESS_LIMIT
);
828 void * __init
__alloc_bootmem_low_nopanic(unsigned long size
,
832 return ___alloc_bootmem_nopanic(size
, align
, goal
,
833 ARCH_LOW_ADDRESS_LIMIT
);
837 * __alloc_bootmem_low_node - allocate low boot memory from a specific node
838 * @pgdat: node to allocate from
839 * @size: size of the request in bytes
840 * @align: alignment of the region
841 * @goal: preferred starting address of the region
843 * The goal is dropped if it can not be satisfied and the allocation will
844 * fall back to memory below @goal.
846 * Allocation may fall back to any node in the system if the specified node
847 * can not hold the requested memory.
849 * The function panics if the request can not be satisfied.
851 void * __init
__alloc_bootmem_low_node(pg_data_t
*pgdat
, unsigned long size
,
852 unsigned long align
, unsigned long goal
)
854 if (WARN_ON_ONCE(slab_is_available()))
855 return kzalloc_node(size
, GFP_NOWAIT
, pgdat
->node_id
);
857 return ___alloc_bootmem_node(pgdat
, size
, align
,
858 goal
, ARCH_LOW_ADDRESS_LIMIT
);