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 /* return the number of _pages_ that will be allocated for the boot bitmap */
40 unsigned long __init
bootmem_bootmap_pages(unsigned long pages
)
42 unsigned long mapsize
;
44 mapsize
= (pages
+7)/8;
45 mapsize
= (mapsize
+ ~PAGE_MASK
) & PAGE_MASK
;
46 mapsize
>>= PAGE_SHIFT
;
54 static void __init
link_bootmem(bootmem_data_t
*bdata
)
58 if (list_empty(&bdata_list
)) {
59 list_add(&bdata
->list
, &bdata_list
);
63 list_for_each_entry(ent
, &bdata_list
, list
) {
64 if (bdata
->node_boot_start
< ent
->node_boot_start
) {
65 list_add_tail(&bdata
->list
, &ent
->list
);
69 list_add_tail(&bdata
->list
, &bdata_list
);
73 * Given an initialised bdata, it returns the size of the boot bitmap
75 static unsigned long __init
get_mapsize(bootmem_data_t
*bdata
)
77 unsigned long mapsize
;
78 unsigned long start
= PFN_DOWN(bdata
->node_boot_start
);
79 unsigned long end
= bdata
->node_low_pfn
;
81 mapsize
= ((end
- start
) + 7) / 8;
82 return ALIGN(mapsize
, sizeof(long));
86 * Called once to set up the allocator itself.
88 static unsigned long __init
init_bootmem_core(pg_data_t
*pgdat
,
89 unsigned long mapstart
, unsigned long start
, unsigned long end
)
91 bootmem_data_t
*bdata
= pgdat
->bdata
;
92 unsigned long mapsize
;
94 mminit_validate_memmodel_limits(&start
, &end
);
95 bdata
->node_bootmem_map
= phys_to_virt(PFN_PHYS(mapstart
));
96 bdata
->node_boot_start
= PFN_PHYS(start
);
97 bdata
->node_low_pfn
= end
;
101 * Initially all pages are reserved - setup_arch() has to
102 * register free RAM areas explicitly.
104 mapsize
= get_mapsize(bdata
);
105 memset(bdata
->node_bootmem_map
, 0xff, mapsize
);
111 * Marks a particular physical memory range as unallocatable. Usable RAM
112 * might be used for boot-time allocations - or it might get added
113 * to the free page pool later on.
115 static int __init
can_reserve_bootmem_core(bootmem_data_t
*bdata
,
116 unsigned long addr
, unsigned long size
, int flags
)
118 unsigned long sidx
, eidx
;
123 /* out of range, don't hold other */
124 if (addr
+ size
< bdata
->node_boot_start
||
125 PFN_DOWN(addr
) > bdata
->node_low_pfn
)
129 * Round up to index to the range.
131 if (addr
> bdata
->node_boot_start
)
132 sidx
= PFN_DOWN(addr
- bdata
->node_boot_start
);
136 eidx
= PFN_UP(addr
+ size
- bdata
->node_boot_start
);
137 if (eidx
> bdata
->node_low_pfn
- PFN_DOWN(bdata
->node_boot_start
))
138 eidx
= bdata
->node_low_pfn
- PFN_DOWN(bdata
->node_boot_start
);
140 for (i
= sidx
; i
< eidx
; i
++) {
141 if (test_bit(i
, bdata
->node_bootmem_map
)) {
142 if (flags
& BOOTMEM_EXCLUSIVE
)
151 static void __init
reserve_bootmem_core(bootmem_data_t
*bdata
,
152 unsigned long addr
, unsigned long size
, int flags
)
154 unsigned long sidx
, eidx
;
160 if (addr
+ size
< bdata
->node_boot_start
||
161 PFN_DOWN(addr
) > bdata
->node_low_pfn
)
165 * Round up to index to the range.
167 if (addr
> bdata
->node_boot_start
)
168 sidx
= PFN_DOWN(addr
- bdata
->node_boot_start
);
172 eidx
= PFN_UP(addr
+ size
- bdata
->node_boot_start
);
173 if (eidx
> bdata
->node_low_pfn
- PFN_DOWN(bdata
->node_boot_start
))
174 eidx
= bdata
->node_low_pfn
- PFN_DOWN(bdata
->node_boot_start
);
176 for (i
= sidx
; i
< eidx
; i
++) {
177 if (test_and_set_bit(i
, bdata
->node_bootmem_map
)) {
178 #ifdef CONFIG_DEBUG_BOOTMEM
179 printk("hm, page %08lx reserved twice.\n", i
*PAGE_SIZE
);
185 static void __init
free_bootmem_core(bootmem_data_t
*bdata
, unsigned long addr
,
188 unsigned long sidx
, eidx
;
194 if (addr
+ size
< bdata
->node_boot_start
||
195 PFN_DOWN(addr
) > bdata
->node_low_pfn
)
198 * round down end of usable mem, partially free pages are
199 * considered reserved.
202 if (addr
>= bdata
->node_boot_start
&& addr
< bdata
->last_success
)
203 bdata
->last_success
= addr
;
206 * Round up to index to the range.
208 if (PFN_UP(addr
) > PFN_DOWN(bdata
->node_boot_start
))
209 sidx
= PFN_UP(addr
) - PFN_DOWN(bdata
->node_boot_start
);
213 eidx
= PFN_DOWN(addr
+ size
- bdata
->node_boot_start
);
214 if (eidx
> bdata
->node_low_pfn
- PFN_DOWN(bdata
->node_boot_start
))
215 eidx
= bdata
->node_low_pfn
- PFN_DOWN(bdata
->node_boot_start
);
217 for (i
= sidx
; i
< eidx
; i
++) {
218 if (unlikely(!test_and_clear_bit(i
, bdata
->node_bootmem_map
)))
224 * We 'merge' subsequent allocations to save space. We might 'lose'
225 * some fraction of a page if allocations cannot be satisfied due to
226 * size constraints on boxes where there is physical RAM space
227 * fragmentation - in these cases (mostly large memory boxes) this
230 * On low memory boxes we get it right in 100% of the cases.
232 * alignment has to be a power of 2 value.
234 * NOTE: This function is _not_ reentrant.
237 __alloc_bootmem_core(struct bootmem_data
*bdata
, unsigned long size
,
238 unsigned long align
, unsigned long goal
, unsigned long limit
)
240 unsigned long areasize
, preferred
;
241 unsigned long i
, start
= 0, incr
, eidx
, end_pfn
;
243 unsigned long node_boot_start
;
244 void *node_bootmem_map
;
247 printk("__alloc_bootmem_core(): zero-sized request\n");
250 BUG_ON(align
& (align
-1));
252 /* on nodes without memory - bootmem_map is NULL */
253 if (!bdata
->node_bootmem_map
)
256 /* bdata->node_boot_start is supposed to be (12+6)bits alignment on x86_64 ? */
257 node_boot_start
= bdata
->node_boot_start
;
258 node_bootmem_map
= bdata
->node_bootmem_map
;
260 node_boot_start
= ALIGN(bdata
->node_boot_start
, align
);
261 if (node_boot_start
> bdata
->node_boot_start
)
262 node_bootmem_map
= (unsigned long *)bdata
->node_bootmem_map
+
263 PFN_DOWN(node_boot_start
- bdata
->node_boot_start
)/BITS_PER_LONG
;
266 if (limit
&& node_boot_start
>= limit
)
269 end_pfn
= bdata
->node_low_pfn
;
270 limit
= PFN_DOWN(limit
);
271 if (limit
&& end_pfn
> limit
)
274 eidx
= end_pfn
- PFN_DOWN(node_boot_start
);
277 * We try to allocate bootmem pages above 'goal'
278 * first, then we try to allocate lower pages.
281 if (goal
&& PFN_DOWN(goal
) < end_pfn
) {
282 if (goal
> node_boot_start
)
283 preferred
= goal
- node_boot_start
;
285 if (bdata
->last_success
> node_boot_start
&&
286 bdata
->last_success
- node_boot_start
>= preferred
)
287 if (!limit
|| (limit
&& limit
> bdata
->last_success
))
288 preferred
= bdata
->last_success
- node_boot_start
;
291 preferred
= PFN_DOWN(ALIGN(preferred
, align
));
292 areasize
= (size
+ PAGE_SIZE
-1) / PAGE_SIZE
;
293 incr
= align
>> PAGE_SHIFT
? : 1;
296 for (i
= preferred
; i
< eidx
;) {
299 i
= find_next_zero_bit(node_bootmem_map
, eidx
, i
);
303 if (test_bit(i
, node_bootmem_map
)) {
307 for (j
= i
+ 1; j
< i
+ areasize
; ++j
) {
310 if (test_bit(j
, node_bootmem_map
))
328 bdata
->last_success
= PFN_PHYS(start
) + node_boot_start
;
329 BUG_ON(start
>= eidx
);
332 * Is the next page of the previous allocation-end the start
333 * of this allocation's buffer? If yes then we can 'merge'
334 * the previous partial page with this allocation.
336 if (align
< PAGE_SIZE
&&
337 bdata
->last_offset
&& bdata
->last_pos
+1 == start
) {
338 unsigned long offset
, remaining_size
;
339 offset
= ALIGN(bdata
->last_offset
, align
);
340 BUG_ON(offset
> PAGE_SIZE
);
341 remaining_size
= PAGE_SIZE
- offset
;
342 if (size
< remaining_size
) {
344 /* last_pos unchanged */
345 bdata
->last_offset
= offset
+ size
;
346 ret
= phys_to_virt(bdata
->last_pos
* PAGE_SIZE
+
347 offset
+ node_boot_start
);
349 remaining_size
= size
- remaining_size
;
350 areasize
= (remaining_size
+ PAGE_SIZE
-1) / PAGE_SIZE
;
351 ret
= phys_to_virt(bdata
->last_pos
* PAGE_SIZE
+
352 offset
+ node_boot_start
);
353 bdata
->last_pos
= start
+ areasize
- 1;
354 bdata
->last_offset
= remaining_size
;
356 bdata
->last_offset
&= ~PAGE_MASK
;
358 bdata
->last_pos
= start
+ areasize
- 1;
359 bdata
->last_offset
= size
& ~PAGE_MASK
;
360 ret
= phys_to_virt(start
* PAGE_SIZE
+ node_boot_start
);
364 * Reserve the area now:
366 for (i
= start
; i
< start
+ areasize
; i
++)
367 if (unlikely(test_and_set_bit(i
, node_bootmem_map
)))
369 memset(ret
, 0, size
);
373 static unsigned long __init
free_all_bootmem_core(pg_data_t
*pgdat
)
377 bootmem_data_t
*bdata
= pgdat
->bdata
;
378 unsigned long i
, count
, total
= 0;
383 BUG_ON(!bdata
->node_bootmem_map
);
386 /* first extant page of the node */
387 pfn
= PFN_DOWN(bdata
->node_boot_start
);
388 idx
= bdata
->node_low_pfn
- pfn
;
389 map
= bdata
->node_bootmem_map
;
390 /* Check physaddr is O(LOG2(BITS_PER_LONG)) page aligned */
391 if (bdata
->node_boot_start
== 0 ||
392 ffs(bdata
->node_boot_start
) - PAGE_SHIFT
> ffs(BITS_PER_LONG
))
394 for (i
= 0; i
< idx
; ) {
395 unsigned long v
= ~map
[i
/ BITS_PER_LONG
];
397 if (gofast
&& v
== ~0UL) {
400 page
= pfn_to_page(pfn
);
401 count
+= BITS_PER_LONG
;
402 order
= ffs(BITS_PER_LONG
) - 1;
403 __free_pages_bootmem(page
, order
);
405 page
+= BITS_PER_LONG
;
409 page
= pfn_to_page(pfn
);
410 for (m
= 1; m
&& i
< idx
; m
<<=1, page
++, i
++) {
413 __free_pages_bootmem(page
, 0);
419 pfn
+= BITS_PER_LONG
;
424 * Now free the allocator bitmap itself, it's not
427 page
= virt_to_page(bdata
->node_bootmem_map
);
429 idx
= (get_mapsize(bdata
) + PAGE_SIZE
-1) >> PAGE_SHIFT
;
430 for (i
= 0; i
< idx
; i
++, page
++) {
431 __free_pages_bootmem(page
, 0);
435 bdata
->node_bootmem_map
= NULL
;
440 unsigned long __init
init_bootmem_node(pg_data_t
*pgdat
, unsigned long freepfn
,
441 unsigned long startpfn
, unsigned long endpfn
)
443 return init_bootmem_core(pgdat
, freepfn
, startpfn
, endpfn
);
446 int __init
reserve_bootmem_node(pg_data_t
*pgdat
, unsigned long physaddr
,
447 unsigned long size
, int flags
)
451 ret
= can_reserve_bootmem_core(pgdat
->bdata
, physaddr
, size
, flags
);
454 reserve_bootmem_core(pgdat
->bdata
, physaddr
, size
, flags
);
459 void __init
free_bootmem_node(pg_data_t
*pgdat
, unsigned long physaddr
,
462 free_bootmem_core(pgdat
->bdata
, physaddr
, size
);
465 unsigned long __init
free_all_bootmem_node(pg_data_t
*pgdat
)
467 register_page_bootmem_info_node(pgdat
);
468 return free_all_bootmem_core(pgdat
);
471 unsigned long __init
init_bootmem(unsigned long start
, unsigned long pages
)
475 return init_bootmem_core(NODE_DATA(0), start
, 0, pages
);
478 #ifndef CONFIG_HAVE_ARCH_BOOTMEM_NODE
479 int __init
reserve_bootmem(unsigned long addr
, unsigned long size
,
482 bootmem_data_t
*bdata
;
485 list_for_each_entry(bdata
, &bdata_list
, list
) {
486 ret
= can_reserve_bootmem_core(bdata
, addr
, size
, flags
);
490 list_for_each_entry(bdata
, &bdata_list
, list
)
491 reserve_bootmem_core(bdata
, addr
, size
, flags
);
495 #endif /* !CONFIG_HAVE_ARCH_BOOTMEM_NODE */
497 void __init
free_bootmem(unsigned long addr
, unsigned long size
)
499 bootmem_data_t
*bdata
;
500 list_for_each_entry(bdata
, &bdata_list
, list
)
501 free_bootmem_core(bdata
, addr
, size
);
504 unsigned long __init
free_all_bootmem(void)
506 return free_all_bootmem_core(NODE_DATA(0));
509 void * __init
__alloc_bootmem_nopanic(unsigned long size
, unsigned long align
,
512 bootmem_data_t
*bdata
;
515 list_for_each_entry(bdata
, &bdata_list
, list
) {
516 ptr
= __alloc_bootmem_core(bdata
, size
, align
, goal
, 0);
523 void * __init
__alloc_bootmem(unsigned long size
, unsigned long align
,
526 void *mem
= __alloc_bootmem_nopanic(size
,align
,goal
);
531 * Whoops, we cannot satisfy the allocation request.
533 printk(KERN_ALERT
"bootmem alloc of %lu bytes failed!\n", size
);
534 panic("Out of memory");
539 void * __init
__alloc_bootmem_node(pg_data_t
*pgdat
, unsigned long size
,
540 unsigned long align
, unsigned long goal
)
544 ptr
= __alloc_bootmem_core(pgdat
->bdata
, size
, align
, goal
, 0);
548 return __alloc_bootmem(size
, align
, goal
);
551 #ifdef CONFIG_SPARSEMEM
552 void * __init
alloc_bootmem_section(unsigned long size
,
553 unsigned long section_nr
)
556 unsigned long limit
, goal
, start_nr
, end_nr
, pfn
;
557 struct pglist_data
*pgdat
;
559 pfn
= section_nr_to_pfn(section_nr
);
560 goal
= PFN_PHYS(pfn
);
561 limit
= PFN_PHYS(section_nr_to_pfn(section_nr
+ 1)) - 1;
562 pgdat
= NODE_DATA(early_pfn_to_nid(pfn
));
563 ptr
= __alloc_bootmem_core(pgdat
->bdata
, size
, SMP_CACHE_BYTES
, goal
,
569 start_nr
= pfn_to_section_nr(PFN_DOWN(__pa(ptr
)));
570 end_nr
= pfn_to_section_nr(PFN_DOWN(__pa(ptr
) + size
));
571 if (start_nr
!= section_nr
|| end_nr
!= section_nr
) {
572 printk(KERN_WARNING
"alloc_bootmem failed on section %ld.\n",
574 free_bootmem_core(pgdat
->bdata
, __pa(ptr
), size
);
582 #ifndef ARCH_LOW_ADDRESS_LIMIT
583 #define ARCH_LOW_ADDRESS_LIMIT 0xffffffffUL
586 void * __init
__alloc_bootmem_low(unsigned long size
, unsigned long align
,
589 bootmem_data_t
*bdata
;
592 list_for_each_entry(bdata
, &bdata_list
, list
) {
593 ptr
= __alloc_bootmem_core(bdata
, size
, align
, goal
,
594 ARCH_LOW_ADDRESS_LIMIT
);
600 * Whoops, we cannot satisfy the allocation request.
602 printk(KERN_ALERT
"low bootmem alloc of %lu bytes failed!\n", size
);
603 panic("Out of low memory");
607 void * __init
__alloc_bootmem_low_node(pg_data_t
*pgdat
, unsigned long size
,
608 unsigned long align
, unsigned long goal
)
610 return __alloc_bootmem_core(pgdat
->bdata
, size
, align
, goal
,
611 ARCH_LOW_ADDRESS_LIMIT
);