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/bootmem.h>
14 #include <linux/module.h>
18 #include <asm/processor.h>
22 unsigned long max_low_pfn
;
23 unsigned long min_low_pfn
;
24 unsigned long max_pfn
;
26 #ifdef CONFIG_CRASH_DUMP
28 * If we have booted due to a crash, max_pfn will be a very low value. We need
29 * to know the amount of memory that the previous kernel used.
31 unsigned long saved_max_pfn
;
34 bootmem_data_t bootmem_node_data
[MAX_NUMNODES
] __initdata
;
36 static struct list_head bdata_list __initdata
= LIST_HEAD_INIT(bdata_list
);
38 static int bootmem_debug
;
40 static int __init
bootmem_debug_setup(char *buf
)
45 early_param("bootmem_debug", bootmem_debug_setup
);
47 #define bdebug(fmt, args...) ({ \
48 if (unlikely(bootmem_debug)) \
54 static unsigned long __init
bootmap_bytes(unsigned long pages
)
56 unsigned long bytes
= (pages
+ 7) / 8;
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
)
77 struct list_head
*iter
;
79 list_for_each(iter
, &bdata_list
) {
82 ent
= list_entry(iter
, bootmem_data_t
, list
);
83 if (bdata
->node_min_pfn
< ent
->node_min_pfn
)
86 list_add_tail(&bdata
->list
, iter
);
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
);
145 static unsigned long __init
free_all_bootmem_core(bootmem_data_t
*bdata
)
149 unsigned long start
, end
, pages
, count
= 0;
151 if (!bdata
->node_bootmem_map
)
154 start
= bdata
->node_min_pfn
;
155 end
= bdata
->node_low_pfn
;
158 * If the start is aligned to the machines wordsize, we might
159 * be able to free pages in bulks of that order.
161 aligned
= !(start
& (BITS_PER_LONG
- 1));
163 bdebug("nid=%td start=%lx end=%lx aligned=%d\n",
164 bdata
- bootmem_node_data
, start
, end
, aligned
);
166 while (start
< end
) {
167 unsigned long *map
, idx
, vec
;
169 map
= bdata
->node_bootmem_map
;
170 idx
= start
- bdata
->node_min_pfn
;
171 vec
= ~map
[idx
/ BITS_PER_LONG
];
173 if (aligned
&& vec
== ~0UL && start
+ BITS_PER_LONG
< end
) {
174 int order
= ilog2(BITS_PER_LONG
);
176 __free_pages_bootmem(pfn_to_page(start
), order
);
177 count
+= BITS_PER_LONG
;
179 unsigned long off
= 0;
181 while (vec
&& off
< BITS_PER_LONG
) {
183 page
= pfn_to_page(start
+ off
);
184 __free_pages_bootmem(page
, 0);
191 start
+= BITS_PER_LONG
;
194 page
= virt_to_page(bdata
->node_bootmem_map
);
195 pages
= bdata
->node_low_pfn
- bdata
->node_min_pfn
;
196 pages
= bootmem_bootmap_pages(pages
);
199 __free_pages_bootmem(page
++, 0);
201 bdebug("nid=%td released=%lx\n", bdata
- bootmem_node_data
, count
);
207 * free_all_bootmem_node - release a node's free pages to the buddy allocator
208 * @pgdat: node to be released
210 * Returns the number of pages actually released.
212 unsigned long __init
free_all_bootmem_node(pg_data_t
*pgdat
)
214 register_page_bootmem_info_node(pgdat
);
215 return free_all_bootmem_core(pgdat
->bdata
);
219 * free_all_bootmem - release free pages to the buddy allocator
221 * Returns the number of pages actually released.
223 unsigned long __init
free_all_bootmem(void)
225 return free_all_bootmem_core(NODE_DATA(0)->bdata
);
228 static void __init
__free(bootmem_data_t
*bdata
,
229 unsigned long sidx
, unsigned long eidx
)
233 bdebug("nid=%td start=%lx end=%lx\n", bdata
- bootmem_node_data
,
234 sidx
+ bdata
->node_min_pfn
,
235 eidx
+ bdata
->node_min_pfn
);
237 if (bdata
->hint_idx
> sidx
)
238 bdata
->hint_idx
= sidx
;
240 for (idx
= sidx
; idx
< eidx
; idx
++)
241 if (!test_and_clear_bit(idx
, bdata
->node_bootmem_map
))
245 static int __init
__reserve(bootmem_data_t
*bdata
, unsigned long sidx
,
246 unsigned long eidx
, int flags
)
249 int exclusive
= flags
& BOOTMEM_EXCLUSIVE
;
251 bdebug("nid=%td start=%lx end=%lx flags=%x\n",
252 bdata
- bootmem_node_data
,
253 sidx
+ bdata
->node_min_pfn
,
254 eidx
+ bdata
->node_min_pfn
,
257 for (idx
= sidx
; idx
< eidx
; idx
++)
258 if (test_and_set_bit(idx
, bdata
->node_bootmem_map
)) {
260 __free(bdata
, sidx
, idx
);
263 bdebug("silent double reserve of PFN %lx\n",
264 idx
+ bdata
->node_min_pfn
);
269 static int __init
mark_bootmem_node(bootmem_data_t
*bdata
,
270 unsigned long start
, unsigned long end
,
271 int reserve
, int flags
)
273 unsigned long sidx
, eidx
;
275 bdebug("nid=%td start=%lx end=%lx reserve=%d flags=%x\n",
276 bdata
- bootmem_node_data
, start
, end
, reserve
, flags
);
278 BUG_ON(start
< bdata
->node_min_pfn
);
279 BUG_ON(end
> bdata
->node_low_pfn
);
281 sidx
= start
- bdata
->node_min_pfn
;
282 eidx
= end
- bdata
->node_min_pfn
;
285 return __reserve(bdata
, sidx
, eidx
, flags
);
287 __free(bdata
, sidx
, eidx
);
291 static int __init
mark_bootmem(unsigned long start
, unsigned long end
,
292 int reserve
, int flags
)
295 bootmem_data_t
*bdata
;
298 list_for_each_entry(bdata
, &bdata_list
, list
) {
302 if (pos
< bdata
->node_min_pfn
||
303 pos
>= bdata
->node_low_pfn
) {
304 BUG_ON(pos
!= start
);
308 max
= min(bdata
->node_low_pfn
, end
);
310 err
= mark_bootmem_node(bdata
, pos
, max
, reserve
, flags
);
311 if (reserve
&& err
) {
312 mark_bootmem(start
, pos
, 0, 0);
318 pos
= bdata
->node_low_pfn
;
324 * free_bootmem_node - mark a page range as usable
325 * @pgdat: node the range resides on
326 * @physaddr: starting address of the range
327 * @size: size of the range in bytes
329 * Partial pages will be considered reserved and left as they are.
331 * The range must reside completely on the specified node.
333 void __init
free_bootmem_node(pg_data_t
*pgdat
, unsigned long physaddr
,
336 unsigned long start
, end
;
338 start
= PFN_UP(physaddr
);
339 end
= PFN_DOWN(physaddr
+ size
);
341 mark_bootmem_node(pgdat
->bdata
, start
, end
, 0, 0);
345 * free_bootmem - mark a page range as usable
346 * @addr: starting address of the range
347 * @size: size of the range in bytes
349 * Partial pages will be considered reserved and left as they are.
351 * The range must be contiguous but may span node boundaries.
353 void __init
free_bootmem(unsigned long addr
, unsigned long size
)
355 unsigned long start
, end
;
357 start
= PFN_UP(addr
);
358 end
= PFN_DOWN(addr
+ size
);
360 mark_bootmem(start
, end
, 0, 0);
364 * reserve_bootmem_node - mark a page range as reserved
365 * @pgdat: node the range resides on
366 * @physaddr: starting address of the range
367 * @size: size of the range in bytes
368 * @flags: reservation flags (see linux/bootmem.h)
370 * Partial pages will be reserved.
372 * The range must reside completely on the specified node.
374 int __init
reserve_bootmem_node(pg_data_t
*pgdat
, unsigned long physaddr
,
375 unsigned long size
, int flags
)
377 unsigned long start
, end
;
379 start
= PFN_DOWN(physaddr
);
380 end
= PFN_UP(physaddr
+ size
);
382 return mark_bootmem_node(pgdat
->bdata
, start
, end
, 1, flags
);
386 * reserve_bootmem - mark a page range as usable
387 * @addr: starting address of the range
388 * @size: size of the range in bytes
389 * @flags: reservation flags (see linux/bootmem.h)
391 * Partial pages will be reserved.
393 * The range must be contiguous but may span node boundaries.
395 int __init
reserve_bootmem(unsigned long addr
, unsigned long size
,
398 unsigned long start
, end
;
400 start
= PFN_DOWN(addr
);
401 end
= PFN_UP(addr
+ size
);
403 return mark_bootmem(start
, end
, 1, flags
);
406 static unsigned long align_idx(struct bootmem_data
*bdata
, unsigned long idx
,
409 unsigned long base
= bdata
->node_min_pfn
;
412 * Align the index with respect to the node start so that the
413 * combination of both satisfies the requested alignment.
416 return ALIGN(base
+ idx
, step
) - base
;
419 static unsigned long align_off(struct bootmem_data
*bdata
, unsigned long off
,
422 unsigned long base
= PFN_PHYS(bdata
->node_min_pfn
);
424 /* Same as align_idx for byte offsets */
426 return ALIGN(base
+ off
, align
) - base
;
429 static void * __init
alloc_bootmem_core(struct bootmem_data
*bdata
,
430 unsigned long size
, unsigned long align
,
431 unsigned long goal
, unsigned long limit
)
433 unsigned long fallback
= 0;
434 unsigned long min
, max
, start
, sidx
, midx
, step
;
436 bdebug("nid=%td size=%lx [%lu pages] align=%lx goal=%lx limit=%lx\n",
437 bdata
- bootmem_node_data
, size
, PAGE_ALIGN(size
) >> PAGE_SHIFT
,
441 BUG_ON(align
& (align
- 1));
442 BUG_ON(limit
&& goal
+ size
> limit
);
444 if (!bdata
->node_bootmem_map
)
447 min
= bdata
->node_min_pfn
;
448 max
= bdata
->node_low_pfn
;
451 limit
>>= PAGE_SHIFT
;
453 if (limit
&& max
> limit
)
458 step
= max(align
>> PAGE_SHIFT
, 1UL);
460 if (goal
&& min
< goal
&& goal
< max
)
461 start
= ALIGN(goal
, step
);
463 start
= ALIGN(min
, step
);
465 sidx
= start
- bdata
->node_min_pfn
;
466 midx
= max
- bdata
->node_min_pfn
;
468 if (bdata
->hint_idx
> sidx
) {
470 * Handle the valid case of sidx being zero and still
471 * catch the fallback below.
474 sidx
= align_idx(bdata
, bdata
->hint_idx
, step
);
480 unsigned long eidx
, i
, start_off
, end_off
;
482 sidx
= find_next_zero_bit(bdata
->node_bootmem_map
, midx
, sidx
);
483 sidx
= align_idx(bdata
, sidx
, step
);
484 eidx
= sidx
+ PFN_UP(size
);
486 if (sidx
>= midx
|| eidx
> midx
)
489 for (i
= sidx
; i
< eidx
; i
++)
490 if (test_bit(i
, bdata
->node_bootmem_map
)) {
491 sidx
= align_idx(bdata
, i
, step
);
497 if (bdata
->last_end_off
& (PAGE_SIZE
- 1) &&
498 PFN_DOWN(bdata
->last_end_off
) + 1 == sidx
)
499 start_off
= align_off(bdata
, bdata
->last_end_off
, align
);
501 start_off
= PFN_PHYS(sidx
);
503 merge
= PFN_DOWN(start_off
) < sidx
;
504 end_off
= start_off
+ size
;
506 bdata
->last_end_off
= end_off
;
507 bdata
->hint_idx
= PFN_UP(end_off
);
510 * Reserve the area now:
512 if (__reserve(bdata
, PFN_DOWN(start_off
) + merge
,
513 PFN_UP(end_off
), BOOTMEM_EXCLUSIVE
))
516 region
= phys_to_virt(PFN_PHYS(bdata
->node_min_pfn
) +
518 memset(region
, 0, size
);
523 sidx
= align_idx(bdata
, fallback
- 1, step
);
531 static void * __init
alloc_arch_preferred_bootmem(bootmem_data_t
*bdata
,
532 unsigned long size
, unsigned long align
,
533 unsigned long goal
, unsigned long limit
)
535 if (WARN_ON_ONCE(slab_is_available()))
536 return kzalloc(size
, GFP_NOWAIT
);
538 #ifdef CONFIG_HAVE_ARCH_BOOTMEM
540 bootmem_data_t
*p_bdata
;
542 p_bdata
= bootmem_arch_preferred_node(bdata
, size
, align
,
545 return alloc_bootmem_core(p_bdata
, size
, align
,
552 static void * __init
___alloc_bootmem_nopanic(unsigned long size
,
557 bootmem_data_t
*bdata
;
561 region
= alloc_arch_preferred_bootmem(NULL
, size
, align
, goal
, limit
);
565 list_for_each_entry(bdata
, &bdata_list
, list
) {
566 if (goal
&& bdata
->node_low_pfn
<= PFN_DOWN(goal
))
568 if (limit
&& bdata
->node_min_pfn
>= PFN_DOWN(limit
))
571 region
= alloc_bootmem_core(bdata
, size
, align
, goal
, limit
);
585 * __alloc_bootmem_nopanic - allocate boot memory without panicking
586 * @size: size of the request in bytes
587 * @align: alignment of the region
588 * @goal: preferred starting address of the region
590 * The goal is dropped if it can not be satisfied and the allocation will
591 * fall back to memory below @goal.
593 * Allocation may happen on any node in the system.
595 * Returns NULL on failure.
597 void * __init
__alloc_bootmem_nopanic(unsigned long size
, unsigned long align
,
600 return ___alloc_bootmem_nopanic(size
, align
, goal
, 0);
603 static void * __init
___alloc_bootmem(unsigned long size
, unsigned long align
,
604 unsigned long goal
, unsigned long limit
)
606 void *mem
= ___alloc_bootmem_nopanic(size
, align
, goal
, limit
);
611 * Whoops, we cannot satisfy the allocation request.
613 printk(KERN_ALERT
"bootmem alloc of %lu bytes failed!\n", size
);
614 panic("Out of memory");
619 * __alloc_bootmem - allocate boot memory
620 * @size: size of the request in bytes
621 * @align: alignment of the region
622 * @goal: preferred starting address of the region
624 * The goal is dropped if it can not be satisfied and the allocation will
625 * fall back to memory below @goal.
627 * Allocation may happen on any node in the system.
629 * The function panics if the request can not be satisfied.
631 void * __init
__alloc_bootmem(unsigned long size
, unsigned long align
,
634 return ___alloc_bootmem(size
, align
, goal
, 0);
637 static void * __init
___alloc_bootmem_node(bootmem_data_t
*bdata
,
638 unsigned long size
, unsigned long align
,
639 unsigned long goal
, unsigned long limit
)
643 ptr
= alloc_arch_preferred_bootmem(bdata
, size
, align
, goal
, limit
);
647 ptr
= alloc_bootmem_core(bdata
, size
, align
, goal
, limit
);
651 return ___alloc_bootmem(size
, align
, goal
, limit
);
655 * __alloc_bootmem_node - allocate boot memory from a specific node
656 * @pgdat: node to allocate from
657 * @size: size of the request in bytes
658 * @align: alignment of the region
659 * @goal: preferred starting address of the region
661 * The goal is dropped if it can not be satisfied and the allocation will
662 * fall back to memory below @goal.
664 * Allocation may fall back to any node in the system if the specified node
665 * can not hold the requested memory.
667 * The function panics if the request can not be satisfied.
669 void * __init
__alloc_bootmem_node(pg_data_t
*pgdat
, unsigned long size
,
670 unsigned long align
, unsigned long goal
)
672 if (WARN_ON_ONCE(slab_is_available()))
673 return kzalloc_node(size
, GFP_NOWAIT
, pgdat
->node_id
);
675 return ___alloc_bootmem_node(pgdat
->bdata
, size
, align
, goal
, 0);
678 #ifdef CONFIG_SPARSEMEM
680 * alloc_bootmem_section - allocate boot memory from a specific section
681 * @size: size of the request in bytes
682 * @section_nr: sparse map section to allocate from
684 * Return NULL on failure.
686 void * __init
alloc_bootmem_section(unsigned long size
,
687 unsigned long section_nr
)
689 bootmem_data_t
*bdata
;
690 unsigned long pfn
, goal
, limit
;
692 pfn
= section_nr_to_pfn(section_nr
);
693 goal
= pfn
<< PAGE_SHIFT
;
694 limit
= section_nr_to_pfn(section_nr
+ 1) << PAGE_SHIFT
;
695 bdata
= &bootmem_node_data
[early_pfn_to_nid(pfn
)];
697 return alloc_bootmem_core(bdata
, size
, SMP_CACHE_BYTES
, goal
, limit
);
701 void * __init
__alloc_bootmem_node_nopanic(pg_data_t
*pgdat
, unsigned long size
,
702 unsigned long align
, unsigned long goal
)
706 if (WARN_ON_ONCE(slab_is_available()))
707 return kzalloc_node(size
, GFP_NOWAIT
, pgdat
->node_id
);
709 ptr
= alloc_arch_preferred_bootmem(pgdat
->bdata
, size
, align
, goal
, 0);
713 ptr
= alloc_bootmem_core(pgdat
->bdata
, size
, align
, goal
, 0);
717 return __alloc_bootmem_nopanic(size
, align
, goal
);
720 #ifndef ARCH_LOW_ADDRESS_LIMIT
721 #define ARCH_LOW_ADDRESS_LIMIT 0xffffffffUL
725 * __alloc_bootmem_low - allocate low boot memory
726 * @size: size of the request in bytes
727 * @align: alignment of the region
728 * @goal: preferred starting address of the region
730 * The goal is dropped if it can not be satisfied and the allocation will
731 * fall back to memory below @goal.
733 * Allocation may happen on any node in the system.
735 * The function panics if the request can not be satisfied.
737 void * __init
__alloc_bootmem_low(unsigned long size
, unsigned long align
,
740 return ___alloc_bootmem(size
, align
, goal
, ARCH_LOW_ADDRESS_LIMIT
);
744 * __alloc_bootmem_low_node - allocate low boot memory from a specific node
745 * @pgdat: node to allocate from
746 * @size: size of the request in bytes
747 * @align: alignment of the region
748 * @goal: preferred starting address of the region
750 * The goal is dropped if it can not be satisfied and the allocation will
751 * fall back to memory below @goal.
753 * Allocation may fall back to any node in the system if the specified node
754 * can not hold the requested memory.
756 * The function panics if the request can not be satisfied.
758 void * __init
__alloc_bootmem_low_node(pg_data_t
*pgdat
, unsigned long size
,
759 unsigned long align
, unsigned long goal
)
761 if (WARN_ON_ONCE(slab_is_available()))
762 return kzalloc_node(size
, GFP_NOWAIT
, pgdat
->node_id
);
764 return ___alloc_bootmem_node(pgdat
->bdata
, size
, align
,
765 goal
, ARCH_LOW_ADDRESS_LIMIT
);