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)) \
51 __FUNCTION__, ## args); \
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_boot_start
< ent
->node_boot_start
)
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_boot_start
= PFN_PHYS(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 i
, count
;
150 unsigned long idx
, pages
;
154 BUG_ON(!bdata
->node_bootmem_map
);
157 /* first extant page of the node */
158 pfn
= PFN_DOWN(bdata
->node_boot_start
);
159 idx
= bdata
->node_low_pfn
- pfn
;
160 map
= bdata
->node_bootmem_map
;
162 * Check if we are aligned to BITS_PER_LONG pages. If so, we might
163 * be able to free page orders of that size at once.
165 if (!(pfn
& (BITS_PER_LONG
-1)))
168 for (i
= 0; i
< idx
; ) {
169 unsigned long v
= ~map
[i
/ BITS_PER_LONG
];
171 if (gofast
&& v
== ~0UL) {
174 page
= pfn_to_page(pfn
);
175 count
+= BITS_PER_LONG
;
176 order
= ffs(BITS_PER_LONG
) - 1;
177 __free_pages_bootmem(page
, order
);
179 page
+= BITS_PER_LONG
;
183 page
= pfn_to_page(pfn
);
184 for (m
= 1; m
&& i
< idx
; m
<<=1, page
++, i
++) {
187 __free_pages_bootmem(page
, 0);
193 pfn
+= BITS_PER_LONG
;
197 * Now free the allocator bitmap itself, it's not
200 page
= virt_to_page(bdata
->node_bootmem_map
);
201 pages
= bdata
->node_low_pfn
- PFN_DOWN(bdata
->node_boot_start
);
202 idx
= bootmem_bootmap_pages(pages
);
203 for (i
= 0; i
< idx
; i
++, page
++)
204 __free_pages_bootmem(page
, 0);
206 bdata
->node_bootmem_map
= NULL
;
208 bdebug("nid=%td released=%lx\n", bdata
- bootmem_node_data
, count
);
214 * free_all_bootmem_node - release a node's free pages to the buddy allocator
215 * @pgdat: node to be released
217 * Returns the number of pages actually released.
219 unsigned long __init
free_all_bootmem_node(pg_data_t
*pgdat
)
221 register_page_bootmem_info_node(pgdat
);
222 return free_all_bootmem_core(pgdat
->bdata
);
226 * free_all_bootmem - release free pages to the buddy allocator
228 * Returns the number of pages actually released.
230 unsigned long __init
free_all_bootmem(void)
232 return free_all_bootmem_core(NODE_DATA(0)->bdata
);
235 static void __init
free_bootmem_core(bootmem_data_t
*bdata
, unsigned long addr
,
238 unsigned long sidx
, eidx
;
244 if (addr
+ size
< bdata
->node_boot_start
||
245 PFN_DOWN(addr
) > bdata
->node_low_pfn
)
248 * round down end of usable mem, partially free pages are
249 * considered reserved.
252 if (addr
>= bdata
->node_boot_start
&& addr
< bdata
->last_success
)
253 bdata
->last_success
= addr
;
256 * Round up to index to the range.
258 if (PFN_UP(addr
) > PFN_DOWN(bdata
->node_boot_start
))
259 sidx
= PFN_UP(addr
) - PFN_DOWN(bdata
->node_boot_start
);
263 eidx
= PFN_DOWN(addr
+ size
- bdata
->node_boot_start
);
264 if (eidx
> bdata
->node_low_pfn
- PFN_DOWN(bdata
->node_boot_start
))
265 eidx
= bdata
->node_low_pfn
- PFN_DOWN(bdata
->node_boot_start
);
267 bdebug("nid=%td start=%lx end=%lx\n", bdata
- bootmem_node_data
,
268 sidx
+ PFN_DOWN(bdata
->node_boot_start
),
269 eidx
+ PFN_DOWN(bdata
->node_boot_start
));
271 for (i
= sidx
; i
< eidx
; i
++) {
272 if (unlikely(!test_and_clear_bit(i
, bdata
->node_bootmem_map
)))
278 * free_bootmem_node - mark a page range as usable
279 * @pgdat: node the range resides on
280 * @physaddr: starting address of the range
281 * @size: size of the range in bytes
283 * Partial pages will be considered reserved and left as they are.
285 * Only physical pages that actually reside on @pgdat are marked.
287 void __init
free_bootmem_node(pg_data_t
*pgdat
, unsigned long physaddr
,
290 free_bootmem_core(pgdat
->bdata
, physaddr
, size
);
294 * free_bootmem - mark a page range as usable
295 * @addr: starting address of the range
296 * @size: size of the range in bytes
298 * Partial pages will be considered reserved and left as they are.
300 * All physical pages within the range are marked, no matter what
301 * node they reside on.
303 void __init
free_bootmem(unsigned long addr
, unsigned long size
)
305 bootmem_data_t
*bdata
;
306 list_for_each_entry(bdata
, &bdata_list
, list
)
307 free_bootmem_core(bdata
, addr
, size
);
311 * Marks a particular physical memory range as unallocatable. Usable RAM
312 * might be used for boot-time allocations - or it might get added
313 * to the free page pool later on.
315 static int __init
can_reserve_bootmem_core(bootmem_data_t
*bdata
,
316 unsigned long addr
, unsigned long size
, int flags
)
318 unsigned long sidx
, eidx
;
323 /* out of range, don't hold other */
324 if (addr
+ size
< bdata
->node_boot_start
||
325 PFN_DOWN(addr
) > bdata
->node_low_pfn
)
329 * Round up to index to the range.
331 if (addr
> bdata
->node_boot_start
)
332 sidx
= PFN_DOWN(addr
- bdata
->node_boot_start
);
336 eidx
= PFN_UP(addr
+ size
- bdata
->node_boot_start
);
337 if (eidx
> bdata
->node_low_pfn
- PFN_DOWN(bdata
->node_boot_start
))
338 eidx
= bdata
->node_low_pfn
- PFN_DOWN(bdata
->node_boot_start
);
340 for (i
= sidx
; i
< eidx
; i
++) {
341 if (test_bit(i
, bdata
->node_bootmem_map
)) {
342 if (flags
& BOOTMEM_EXCLUSIVE
)
351 static void __init
reserve_bootmem_core(bootmem_data_t
*bdata
,
352 unsigned long addr
, unsigned long size
, int flags
)
354 unsigned long sidx
, eidx
;
360 if (addr
+ size
< bdata
->node_boot_start
||
361 PFN_DOWN(addr
) > bdata
->node_low_pfn
)
365 * Round up to index to the range.
367 if (addr
> bdata
->node_boot_start
)
368 sidx
= PFN_DOWN(addr
- bdata
->node_boot_start
);
372 eidx
= PFN_UP(addr
+ size
- bdata
->node_boot_start
);
373 if (eidx
> bdata
->node_low_pfn
- PFN_DOWN(bdata
->node_boot_start
))
374 eidx
= bdata
->node_low_pfn
- PFN_DOWN(bdata
->node_boot_start
);
376 bdebug("nid=%td start=%lx end=%lx flags=%x\n",
377 bdata
- bootmem_node_data
,
378 sidx
+ PFN_DOWN(bdata
->node_boot_start
),
379 eidx
+ PFN_DOWN(bdata
->node_boot_start
),
382 for (i
= sidx
; i
< eidx
; i
++)
383 if (test_and_set_bit(i
, bdata
->node_bootmem_map
))
384 bdebug("hm, page %lx reserved twice.\n",
385 PFN_DOWN(bdata
->node_boot_start
) + i
);
389 * reserve_bootmem_node - mark a page range as reserved
390 * @pgdat: node the range resides on
391 * @physaddr: starting address of the range
392 * @size: size of the range in bytes
393 * @flags: reservation flags (see linux/bootmem.h)
395 * Partial pages will be reserved.
397 * Only physical pages that actually reside on @pgdat are marked.
399 int __init
reserve_bootmem_node(pg_data_t
*pgdat
, unsigned long physaddr
,
400 unsigned long size
, int flags
)
404 ret
= can_reserve_bootmem_core(pgdat
->bdata
, physaddr
, size
, flags
);
407 reserve_bootmem_core(pgdat
->bdata
, physaddr
, size
, flags
);
411 #ifndef CONFIG_HAVE_ARCH_BOOTMEM_NODE
413 * reserve_bootmem - mark a page range as usable
414 * @addr: starting address of the range
415 * @size: size of the range in bytes
416 * @flags: reservation flags (see linux/bootmem.h)
418 * Partial pages will be reserved.
420 * All physical pages within the range are marked, no matter what
421 * node they reside on.
423 int __init
reserve_bootmem(unsigned long addr
, unsigned long size
,
426 bootmem_data_t
*bdata
;
429 list_for_each_entry(bdata
, &bdata_list
, list
) {
430 ret
= can_reserve_bootmem_core(bdata
, addr
, size
, flags
);
434 list_for_each_entry(bdata
, &bdata_list
, list
)
435 reserve_bootmem_core(bdata
, addr
, size
, flags
);
439 #endif /* !CONFIG_HAVE_ARCH_BOOTMEM_NODE */
442 * We 'merge' subsequent allocations to save space. We might 'lose'
443 * some fraction of a page if allocations cannot be satisfied due to
444 * size constraints on boxes where there is physical RAM space
445 * fragmentation - in these cases (mostly large memory boxes) this
448 * On low memory boxes we get it right in 100% of the cases.
450 * alignment has to be a power of 2 value.
452 * NOTE: This function is _not_ reentrant.
455 alloc_bootmem_core(struct bootmem_data
*bdata
, unsigned long size
,
456 unsigned long align
, unsigned long goal
, unsigned long limit
)
458 unsigned long areasize
, preferred
;
459 unsigned long i
, start
= 0, incr
, eidx
, end_pfn
;
461 unsigned long node_boot_start
;
462 void *node_bootmem_map
;
465 printk("alloc_bootmem_core(): zero-sized request\n");
468 BUG_ON(align
& (align
-1));
470 /* on nodes without memory - bootmem_map is NULL */
471 if (!bdata
->node_bootmem_map
)
474 bdebug("nid=%td size=%lx [%lu pages] align=%lx goal=%lx limit=%lx\n",
475 bdata
- bootmem_node_data
, size
, PAGE_ALIGN(size
) >> PAGE_SHIFT
,
478 /* bdata->node_boot_start is supposed to be (12+6)bits alignment on x86_64 ? */
479 node_boot_start
= bdata
->node_boot_start
;
480 node_bootmem_map
= bdata
->node_bootmem_map
;
482 node_boot_start
= ALIGN(bdata
->node_boot_start
, align
);
483 if (node_boot_start
> bdata
->node_boot_start
)
484 node_bootmem_map
= (unsigned long *)bdata
->node_bootmem_map
+
485 PFN_DOWN(node_boot_start
- bdata
->node_boot_start
)/BITS_PER_LONG
;
488 if (limit
&& node_boot_start
>= limit
)
491 end_pfn
= bdata
->node_low_pfn
;
492 limit
= PFN_DOWN(limit
);
493 if (limit
&& end_pfn
> limit
)
496 eidx
= end_pfn
- PFN_DOWN(node_boot_start
);
499 * We try to allocate bootmem pages above 'goal'
500 * first, then we try to allocate lower pages.
503 if (goal
&& PFN_DOWN(goal
) < end_pfn
) {
504 if (goal
> node_boot_start
)
505 preferred
= goal
- node_boot_start
;
507 if (bdata
->last_success
> node_boot_start
&&
508 bdata
->last_success
- node_boot_start
>= preferred
)
509 if (!limit
|| (limit
&& limit
> bdata
->last_success
))
510 preferred
= bdata
->last_success
- node_boot_start
;
513 preferred
= PFN_DOWN(ALIGN(preferred
, align
));
514 areasize
= (size
+ PAGE_SIZE
-1) / PAGE_SIZE
;
515 incr
= align
>> PAGE_SHIFT
? : 1;
518 for (i
= preferred
; i
< eidx
;) {
521 i
= find_next_zero_bit(node_bootmem_map
, eidx
, i
);
525 if (test_bit(i
, node_bootmem_map
)) {
529 for (j
= i
+ 1; j
< i
+ areasize
; ++j
) {
532 if (test_bit(j
, node_bootmem_map
))
550 bdata
->last_success
= PFN_PHYS(start
) + node_boot_start
;
551 BUG_ON(start
>= eidx
);
554 * Is the next page of the previous allocation-end the start
555 * of this allocation's buffer? If yes then we can 'merge'
556 * the previous partial page with this allocation.
558 if (align
< PAGE_SIZE
&&
559 bdata
->last_offset
&& bdata
->last_pos
+1 == start
) {
560 unsigned long offset
, remaining_size
;
561 offset
= ALIGN(bdata
->last_offset
, align
);
562 BUG_ON(offset
> PAGE_SIZE
);
563 remaining_size
= PAGE_SIZE
- offset
;
564 if (size
< remaining_size
) {
566 /* last_pos unchanged */
567 bdata
->last_offset
= offset
+ size
;
568 ret
= phys_to_virt(bdata
->last_pos
* PAGE_SIZE
+
569 offset
+ node_boot_start
);
571 remaining_size
= size
- remaining_size
;
572 areasize
= (remaining_size
+ PAGE_SIZE
-1) / PAGE_SIZE
;
573 ret
= phys_to_virt(bdata
->last_pos
* PAGE_SIZE
+
574 offset
+ node_boot_start
);
575 bdata
->last_pos
= start
+ areasize
- 1;
576 bdata
->last_offset
= remaining_size
;
578 bdata
->last_offset
&= ~PAGE_MASK
;
580 bdata
->last_pos
= start
+ areasize
- 1;
581 bdata
->last_offset
= size
& ~PAGE_MASK
;
582 ret
= phys_to_virt(start
* PAGE_SIZE
+ node_boot_start
);
585 bdebug("nid=%td start=%lx end=%lx\n",
586 bdata
- bootmem_node_data
,
587 start
+ PFN_DOWN(bdata
->node_boot_start
),
588 start
+ areasize
+ PFN_DOWN(bdata
->node_boot_start
));
591 * Reserve the area now:
593 for (i
= start
; i
< start
+ areasize
; i
++)
594 if (unlikely(test_and_set_bit(i
, node_bootmem_map
)))
596 memset(ret
, 0, size
);
601 * __alloc_bootmem_nopanic - allocate boot memory without panicking
602 * @size: size of the request in bytes
603 * @align: alignment of the region
604 * @goal: preferred starting address of the region
606 * The goal is dropped if it can not be satisfied and the allocation will
607 * fall back to memory below @goal.
609 * Allocation may happen on any node in the system.
611 * Returns NULL on failure.
613 void * __init
__alloc_bootmem_nopanic(unsigned long size
, unsigned long align
,
616 bootmem_data_t
*bdata
;
619 list_for_each_entry(bdata
, &bdata_list
, list
) {
620 ptr
= alloc_bootmem_core(bdata
, size
, align
, goal
, 0);
628 * __alloc_bootmem - allocate boot memory
629 * @size: size of the request in bytes
630 * @align: alignment of the region
631 * @goal: preferred starting address of the region
633 * The goal is dropped if it can not be satisfied and the allocation will
634 * fall back to memory below @goal.
636 * Allocation may happen on any node in the system.
638 * The function panics if the request can not be satisfied.
640 void * __init
__alloc_bootmem(unsigned long size
, unsigned long align
,
643 void *mem
= __alloc_bootmem_nopanic(size
,align
,goal
);
648 * Whoops, we cannot satisfy the allocation request.
650 printk(KERN_ALERT
"bootmem alloc of %lu bytes failed!\n", size
);
651 panic("Out of memory");
656 * __alloc_bootmem_node - allocate boot memory from a specific node
657 * @pgdat: node to allocate from
658 * @size: size of the request in bytes
659 * @align: alignment of the region
660 * @goal: preferred starting address of the region
662 * The goal is dropped if it can not be satisfied and the allocation will
663 * fall back to memory below @goal.
665 * Allocation may fall back to any node in the system if the specified node
666 * can not hold the requested memory.
668 * The function panics if the request can not be satisfied.
670 void * __init
__alloc_bootmem_node(pg_data_t
*pgdat
, unsigned long size
,
671 unsigned long align
, unsigned long goal
)
675 ptr
= alloc_bootmem_core(pgdat
->bdata
, size
, align
, goal
, 0);
679 return __alloc_bootmem(size
, align
, goal
);
682 #ifdef CONFIG_SPARSEMEM
684 * alloc_bootmem_section - allocate boot memory from a specific section
685 * @size: size of the request in bytes
686 * @section_nr: sparse map section to allocate from
688 * Return NULL on failure.
690 void * __init
alloc_bootmem_section(unsigned long size
,
691 unsigned long section_nr
)
694 unsigned long limit
, goal
, start_nr
, end_nr
, pfn
;
695 struct pglist_data
*pgdat
;
697 pfn
= section_nr_to_pfn(section_nr
);
698 goal
= PFN_PHYS(pfn
);
699 limit
= PFN_PHYS(section_nr_to_pfn(section_nr
+ 1)) - 1;
700 pgdat
= NODE_DATA(early_pfn_to_nid(pfn
));
701 ptr
= alloc_bootmem_core(pgdat
->bdata
, size
, SMP_CACHE_BYTES
, goal
,
707 start_nr
= pfn_to_section_nr(PFN_DOWN(__pa(ptr
)));
708 end_nr
= pfn_to_section_nr(PFN_DOWN(__pa(ptr
) + size
));
709 if (start_nr
!= section_nr
|| end_nr
!= section_nr
) {
710 printk(KERN_WARNING
"alloc_bootmem failed on section %ld.\n",
712 free_bootmem_core(pgdat
->bdata
, __pa(ptr
), size
);
720 void * __init
__alloc_bootmem_node_nopanic(pg_data_t
*pgdat
, unsigned long size
,
721 unsigned long align
, unsigned long goal
)
725 ptr
= alloc_bootmem_core(pgdat
->bdata
, size
, align
, goal
, 0);
729 return __alloc_bootmem_nopanic(size
, align
, goal
);
732 #ifndef ARCH_LOW_ADDRESS_LIMIT
733 #define ARCH_LOW_ADDRESS_LIMIT 0xffffffffUL
737 * __alloc_bootmem_low - allocate low boot memory
738 * @size: size of the request in bytes
739 * @align: alignment of the region
740 * @goal: preferred starting address of the region
742 * The goal is dropped if it can not be satisfied and the allocation will
743 * fall back to memory below @goal.
745 * Allocation may happen on any node in the system.
747 * The function panics if the request can not be satisfied.
749 void * __init
__alloc_bootmem_low(unsigned long size
, unsigned long align
,
752 bootmem_data_t
*bdata
;
755 list_for_each_entry(bdata
, &bdata_list
, list
) {
756 ptr
= alloc_bootmem_core(bdata
, size
, align
, goal
,
757 ARCH_LOW_ADDRESS_LIMIT
);
763 * Whoops, we cannot satisfy the allocation request.
765 printk(KERN_ALERT
"low bootmem alloc of %lu bytes failed!\n", size
);
766 panic("Out of low memory");
771 * __alloc_bootmem_low_node - allocate low boot memory from a specific node
772 * @pgdat: node to allocate from
773 * @size: size of the request in bytes
774 * @align: alignment of the region
775 * @goal: preferred starting address of the region
777 * The goal is dropped if it can not be satisfied and the allocation will
778 * fall back to memory below @goal.
780 * Allocation may fall back to any node in the system if the specified node
781 * can not hold the requested memory.
783 * The function panics if the request can not be satisfied.
785 void * __init
__alloc_bootmem_low_node(pg_data_t
*pgdat
, unsigned long size
,
786 unsigned long align
, unsigned long goal
)
788 return alloc_bootmem_core(pgdat
->bdata
, size
, align
, goal
,
789 ARCH_LOW_ADDRESS_LIMIT
);