4 * Copyright (C) 1999 Ingo Molnar
5 * Discontiguous memory support, Kanoj Sarcar, SGI, Nov 1999
7 * simple boot-time physical memory area allocator and
8 * free memory collector. It's used to deal with reserved
9 * system memory and memory holes as well.
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>
23 * Access to this subsystem has to be serialized externally. (this is
24 * true for the boot process anyway)
26 unsigned long max_low_pfn
;
27 unsigned long min_low_pfn
;
28 unsigned long max_pfn
;
30 static LIST_HEAD(bdata_list
);
31 #ifdef CONFIG_CRASH_DUMP
33 * If we have booted due to a crash, max_pfn will be a very low value. We need
34 * to know the amount of memory that the previous kernel used.
36 unsigned long saved_max_pfn
;
39 bootmem_data_t bootmem_node_data
[MAX_NUMNODES
] __initdata
;
41 /* return the number of _pages_ that will be allocated for the boot bitmap */
42 unsigned long __init
bootmem_bootmap_pages(unsigned long pages
)
44 unsigned long mapsize
;
46 mapsize
= (pages
+7)/8;
47 mapsize
= (mapsize
+ ~PAGE_MASK
) & PAGE_MASK
;
48 mapsize
>>= PAGE_SHIFT
;
56 static void __init
link_bootmem(bootmem_data_t
*bdata
)
60 if (list_empty(&bdata_list
)) {
61 list_add(&bdata
->list
, &bdata_list
);
65 list_for_each_entry(ent
, &bdata_list
, list
) {
66 if (bdata
->node_boot_start
< ent
->node_boot_start
) {
67 list_add_tail(&bdata
->list
, &ent
->list
);
71 list_add_tail(&bdata
->list
, &bdata_list
);
75 * Given an initialised bdata, it returns the size of the boot bitmap
77 static unsigned long __init
get_mapsize(bootmem_data_t
*bdata
)
79 unsigned long mapsize
;
80 unsigned long start
= PFN_DOWN(bdata
->node_boot_start
);
81 unsigned long end
= bdata
->node_low_pfn
;
83 mapsize
= ((end
- start
) + 7) / 8;
84 return ALIGN(mapsize
, sizeof(long));
88 * Called once to set up the allocator itself.
90 static unsigned long __init
init_bootmem_core(pg_data_t
*pgdat
,
91 unsigned long mapstart
, unsigned long start
, unsigned long end
)
93 bootmem_data_t
*bdata
= pgdat
->bdata
;
94 unsigned long mapsize
;
96 mminit_validate_memmodel_limits(&start
, &end
);
97 bdata
->node_bootmem_map
= phys_to_virt(PFN_PHYS(mapstart
));
98 bdata
->node_boot_start
= PFN_PHYS(start
);
99 bdata
->node_low_pfn
= end
;
103 * Initially all pages are reserved - setup_arch() has to
104 * register free RAM areas explicitly.
106 mapsize
= get_mapsize(bdata
);
107 memset(bdata
->node_bootmem_map
, 0xff, mapsize
);
113 * Marks a particular physical memory range as unallocatable. Usable RAM
114 * might be used for boot-time allocations - or it might get added
115 * to the free page pool later on.
117 static int __init
can_reserve_bootmem_core(bootmem_data_t
*bdata
,
118 unsigned long addr
, unsigned long size
, int flags
)
120 unsigned long sidx
, eidx
;
125 /* out of range, don't hold other */
126 if (addr
+ size
< bdata
->node_boot_start
||
127 PFN_DOWN(addr
) > bdata
->node_low_pfn
)
131 * Round up to index to the range.
133 if (addr
> bdata
->node_boot_start
)
134 sidx
= PFN_DOWN(addr
- bdata
->node_boot_start
);
138 eidx
= PFN_UP(addr
+ size
- bdata
->node_boot_start
);
139 if (eidx
> bdata
->node_low_pfn
- PFN_DOWN(bdata
->node_boot_start
))
140 eidx
= bdata
->node_low_pfn
- PFN_DOWN(bdata
->node_boot_start
);
142 for (i
= sidx
; i
< eidx
; i
++) {
143 if (test_bit(i
, bdata
->node_bootmem_map
)) {
144 if (flags
& BOOTMEM_EXCLUSIVE
)
153 static void __init
reserve_bootmem_core(bootmem_data_t
*bdata
,
154 unsigned long addr
, unsigned long size
, int flags
)
156 unsigned long sidx
, eidx
;
162 if (addr
+ size
< bdata
->node_boot_start
||
163 PFN_DOWN(addr
) > bdata
->node_low_pfn
)
167 * Round up to index to the range.
169 if (addr
> bdata
->node_boot_start
)
170 sidx
= PFN_DOWN(addr
- bdata
->node_boot_start
);
174 eidx
= PFN_UP(addr
+ size
- bdata
->node_boot_start
);
175 if (eidx
> bdata
->node_low_pfn
- PFN_DOWN(bdata
->node_boot_start
))
176 eidx
= bdata
->node_low_pfn
- PFN_DOWN(bdata
->node_boot_start
);
178 for (i
= sidx
; i
< eidx
; i
++) {
179 if (test_and_set_bit(i
, bdata
->node_bootmem_map
)) {
180 #ifdef CONFIG_DEBUG_BOOTMEM
181 printk("hm, page %08lx reserved twice.\n", i
*PAGE_SIZE
);
187 static void __init
free_bootmem_core(bootmem_data_t
*bdata
, unsigned long addr
,
190 unsigned long sidx
, eidx
;
196 if (addr
+ size
< bdata
->node_boot_start
||
197 PFN_DOWN(addr
) > bdata
->node_low_pfn
)
200 * round down end of usable mem, partially free pages are
201 * considered reserved.
204 if (addr
>= bdata
->node_boot_start
&& addr
< bdata
->last_success
)
205 bdata
->last_success
= addr
;
208 * Round up to index to the range.
210 if (PFN_UP(addr
) > PFN_DOWN(bdata
->node_boot_start
))
211 sidx
= PFN_UP(addr
) - PFN_DOWN(bdata
->node_boot_start
);
215 eidx
= PFN_DOWN(addr
+ size
- bdata
->node_boot_start
);
216 if (eidx
> bdata
->node_low_pfn
- PFN_DOWN(bdata
->node_boot_start
))
217 eidx
= bdata
->node_low_pfn
- PFN_DOWN(bdata
->node_boot_start
);
219 for (i
= sidx
; i
< eidx
; i
++) {
220 if (unlikely(!test_and_clear_bit(i
, bdata
->node_bootmem_map
)))
226 * We 'merge' subsequent allocations to save space. We might 'lose'
227 * some fraction of a page if allocations cannot be satisfied due to
228 * size constraints on boxes where there is physical RAM space
229 * fragmentation - in these cases (mostly large memory boxes) this
232 * On low memory boxes we get it right in 100% of the cases.
234 * alignment has to be a power of 2 value.
236 * NOTE: This function is _not_ reentrant.
239 __alloc_bootmem_core(struct bootmem_data
*bdata
, unsigned long size
,
240 unsigned long align
, unsigned long goal
, unsigned long limit
)
242 unsigned long areasize
, preferred
;
243 unsigned long i
, start
= 0, incr
, eidx
, end_pfn
;
245 unsigned long node_boot_start
;
246 void *node_bootmem_map
;
249 printk("__alloc_bootmem_core(): zero-sized request\n");
252 BUG_ON(align
& (align
-1));
254 /* on nodes without memory - bootmem_map is NULL */
255 if (!bdata
->node_bootmem_map
)
258 /* bdata->node_boot_start is supposed to be (12+6)bits alignment on x86_64 ? */
259 node_boot_start
= bdata
->node_boot_start
;
260 node_bootmem_map
= bdata
->node_bootmem_map
;
262 node_boot_start
= ALIGN(bdata
->node_boot_start
, align
);
263 if (node_boot_start
> bdata
->node_boot_start
)
264 node_bootmem_map
= (unsigned long *)bdata
->node_bootmem_map
+
265 PFN_DOWN(node_boot_start
- bdata
->node_boot_start
)/BITS_PER_LONG
;
268 if (limit
&& node_boot_start
>= limit
)
271 end_pfn
= bdata
->node_low_pfn
;
272 limit
= PFN_DOWN(limit
);
273 if (limit
&& end_pfn
> limit
)
276 eidx
= end_pfn
- PFN_DOWN(node_boot_start
);
279 * We try to allocate bootmem pages above 'goal'
280 * first, then we try to allocate lower pages.
283 if (goal
&& PFN_DOWN(goal
) < end_pfn
) {
284 if (goal
> node_boot_start
)
285 preferred
= goal
- node_boot_start
;
287 if (bdata
->last_success
> node_boot_start
&&
288 bdata
->last_success
- node_boot_start
>= preferred
)
289 if (!limit
|| (limit
&& limit
> bdata
->last_success
))
290 preferred
= bdata
->last_success
- node_boot_start
;
293 preferred
= PFN_DOWN(ALIGN(preferred
, align
));
294 areasize
= (size
+ PAGE_SIZE
-1) / PAGE_SIZE
;
295 incr
= align
>> PAGE_SHIFT
? : 1;
298 for (i
= preferred
; i
< eidx
;) {
301 i
= find_next_zero_bit(node_bootmem_map
, eidx
, i
);
305 if (test_bit(i
, node_bootmem_map
)) {
309 for (j
= i
+ 1; j
< i
+ areasize
; ++j
) {
312 if (test_bit(j
, node_bootmem_map
))
330 bdata
->last_success
= PFN_PHYS(start
) + node_boot_start
;
331 BUG_ON(start
>= eidx
);
334 * Is the next page of the previous allocation-end the start
335 * of this allocation's buffer? If yes then we can 'merge'
336 * the previous partial page with this allocation.
338 if (align
< PAGE_SIZE
&&
339 bdata
->last_offset
&& bdata
->last_pos
+1 == start
) {
340 unsigned long offset
, remaining_size
;
341 offset
= ALIGN(bdata
->last_offset
, align
);
342 BUG_ON(offset
> PAGE_SIZE
);
343 remaining_size
= PAGE_SIZE
- offset
;
344 if (size
< remaining_size
) {
346 /* last_pos unchanged */
347 bdata
->last_offset
= offset
+ size
;
348 ret
= phys_to_virt(bdata
->last_pos
* PAGE_SIZE
+
349 offset
+ node_boot_start
);
351 remaining_size
= size
- remaining_size
;
352 areasize
= (remaining_size
+ PAGE_SIZE
-1) / PAGE_SIZE
;
353 ret
= phys_to_virt(bdata
->last_pos
* PAGE_SIZE
+
354 offset
+ node_boot_start
);
355 bdata
->last_pos
= start
+ areasize
- 1;
356 bdata
->last_offset
= remaining_size
;
358 bdata
->last_offset
&= ~PAGE_MASK
;
360 bdata
->last_pos
= start
+ areasize
- 1;
361 bdata
->last_offset
= size
& ~PAGE_MASK
;
362 ret
= phys_to_virt(start
* PAGE_SIZE
+ node_boot_start
);
366 * Reserve the area now:
368 for (i
= start
; i
< start
+ areasize
; i
++)
369 if (unlikely(test_and_set_bit(i
, node_bootmem_map
)))
371 memset(ret
, 0, size
);
375 static unsigned long __init
free_all_bootmem_core(pg_data_t
*pgdat
)
379 bootmem_data_t
*bdata
= pgdat
->bdata
;
380 unsigned long i
, count
, total
= 0;
385 BUG_ON(!bdata
->node_bootmem_map
);
388 /* first extant page of the node */
389 pfn
= PFN_DOWN(bdata
->node_boot_start
);
390 idx
= bdata
->node_low_pfn
- pfn
;
391 map
= bdata
->node_bootmem_map
;
392 /* Check physaddr is O(LOG2(BITS_PER_LONG)) page aligned */
393 if (bdata
->node_boot_start
== 0 ||
394 ffs(bdata
->node_boot_start
) - PAGE_SHIFT
> ffs(BITS_PER_LONG
))
396 for (i
= 0; i
< idx
; ) {
397 unsigned long v
= ~map
[i
/ BITS_PER_LONG
];
399 if (gofast
&& v
== ~0UL) {
402 page
= pfn_to_page(pfn
);
403 count
+= BITS_PER_LONG
;
404 order
= ffs(BITS_PER_LONG
) - 1;
405 __free_pages_bootmem(page
, order
);
407 page
+= BITS_PER_LONG
;
411 page
= pfn_to_page(pfn
);
412 for (m
= 1; m
&& i
< idx
; m
<<=1, page
++, i
++) {
415 __free_pages_bootmem(page
, 0);
421 pfn
+= BITS_PER_LONG
;
426 * Now free the allocator bitmap itself, it's not
429 page
= virt_to_page(bdata
->node_bootmem_map
);
431 idx
= (get_mapsize(bdata
) + PAGE_SIZE
-1) >> PAGE_SHIFT
;
432 for (i
= 0; i
< idx
; i
++, page
++) {
433 __free_pages_bootmem(page
, 0);
437 bdata
->node_bootmem_map
= NULL
;
442 unsigned long __init
init_bootmem_node(pg_data_t
*pgdat
, unsigned long freepfn
,
443 unsigned long startpfn
, unsigned long endpfn
)
445 return init_bootmem_core(pgdat
, freepfn
, startpfn
, endpfn
);
448 int __init
reserve_bootmem_node(pg_data_t
*pgdat
, unsigned long physaddr
,
449 unsigned long size
, int flags
)
453 ret
= can_reserve_bootmem_core(pgdat
->bdata
, physaddr
, size
, flags
);
456 reserve_bootmem_core(pgdat
->bdata
, physaddr
, size
, flags
);
461 void __init
free_bootmem_node(pg_data_t
*pgdat
, unsigned long physaddr
,
464 free_bootmem_core(pgdat
->bdata
, physaddr
, size
);
467 unsigned long __init
free_all_bootmem_node(pg_data_t
*pgdat
)
469 register_page_bootmem_info_node(pgdat
);
470 return free_all_bootmem_core(pgdat
);
473 unsigned long __init
init_bootmem(unsigned long start
, unsigned long pages
)
477 return init_bootmem_core(NODE_DATA(0), start
, 0, pages
);
480 #ifndef CONFIG_HAVE_ARCH_BOOTMEM_NODE
481 int __init
reserve_bootmem(unsigned long addr
, unsigned long size
,
484 bootmem_data_t
*bdata
;
487 list_for_each_entry(bdata
, &bdata_list
, list
) {
488 ret
= can_reserve_bootmem_core(bdata
, addr
, size
, flags
);
492 list_for_each_entry(bdata
, &bdata_list
, list
)
493 reserve_bootmem_core(bdata
, addr
, size
, flags
);
497 #endif /* !CONFIG_HAVE_ARCH_BOOTMEM_NODE */
499 void __init
free_bootmem(unsigned long addr
, unsigned long size
)
501 bootmem_data_t
*bdata
;
502 list_for_each_entry(bdata
, &bdata_list
, list
)
503 free_bootmem_core(bdata
, addr
, size
);
506 unsigned long __init
free_all_bootmem(void)
508 return free_all_bootmem_core(NODE_DATA(0));
511 void * __init
__alloc_bootmem_nopanic(unsigned long size
, unsigned long align
,
514 bootmem_data_t
*bdata
;
517 list_for_each_entry(bdata
, &bdata_list
, list
) {
518 ptr
= __alloc_bootmem_core(bdata
, size
, align
, goal
, 0);
525 void * __init
__alloc_bootmem(unsigned long size
, unsigned long align
,
528 void *mem
= __alloc_bootmem_nopanic(size
,align
,goal
);
533 * Whoops, we cannot satisfy the allocation request.
535 printk(KERN_ALERT
"bootmem alloc of %lu bytes failed!\n", size
);
536 panic("Out of memory");
541 void * __init
__alloc_bootmem_node(pg_data_t
*pgdat
, unsigned long size
,
542 unsigned long align
, unsigned long goal
)
546 ptr
= __alloc_bootmem_core(pgdat
->bdata
, size
, align
, goal
, 0);
550 return __alloc_bootmem(size
, align
, goal
);
553 #ifdef CONFIG_SPARSEMEM
554 void * __init
alloc_bootmem_section(unsigned long size
,
555 unsigned long section_nr
)
558 unsigned long limit
, goal
, start_nr
, end_nr
, pfn
;
559 struct pglist_data
*pgdat
;
561 pfn
= section_nr_to_pfn(section_nr
);
562 goal
= PFN_PHYS(pfn
);
563 limit
= PFN_PHYS(section_nr_to_pfn(section_nr
+ 1)) - 1;
564 pgdat
= NODE_DATA(early_pfn_to_nid(pfn
));
565 ptr
= __alloc_bootmem_core(pgdat
->bdata
, size
, SMP_CACHE_BYTES
, goal
,
571 start_nr
= pfn_to_section_nr(PFN_DOWN(__pa(ptr
)));
572 end_nr
= pfn_to_section_nr(PFN_DOWN(__pa(ptr
) + size
));
573 if (start_nr
!= section_nr
|| end_nr
!= section_nr
) {
574 printk(KERN_WARNING
"alloc_bootmem failed on section %ld.\n",
576 free_bootmem_core(pgdat
->bdata
, __pa(ptr
), size
);
584 #ifndef ARCH_LOW_ADDRESS_LIMIT
585 #define ARCH_LOW_ADDRESS_LIMIT 0xffffffffUL
588 void * __init
__alloc_bootmem_low(unsigned long size
, unsigned long align
,
591 bootmem_data_t
*bdata
;
594 list_for_each_entry(bdata
, &bdata_list
, list
) {
595 ptr
= __alloc_bootmem_core(bdata
, size
, align
, goal
,
596 ARCH_LOW_ADDRESS_LIMIT
);
602 * Whoops, we cannot satisfy the allocation request.
604 printk(KERN_ALERT
"low bootmem alloc of %lu bytes failed!\n", size
);
605 panic("Out of low memory");
609 void * __init
__alloc_bootmem_low_node(pg_data_t
*pgdat
, unsigned long size
,
610 unsigned long align
, unsigned long goal
)
612 return __alloc_bootmem_core(pgdat
->bdata
, size
, align
, goal
,
613 ARCH_LOW_ADDRESS_LIMIT
);