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
;
42 * Given an initialised bdata, it returns the size of the boot bitmap
44 static unsigned long __init
get_mapsize(bootmem_data_t
*bdata
)
46 unsigned long mapsize
;
47 unsigned long start
= PFN_DOWN(bdata
->node_boot_start
);
48 unsigned long end
= bdata
->node_low_pfn
;
50 mapsize
= ((end
- start
) + 7) / 8;
51 return ALIGN(mapsize
, sizeof(long));
54 /* return the number of _pages_ that will be allocated for the boot bitmap */
55 unsigned long __init
bootmem_bootmap_pages(unsigned long pages
)
57 unsigned long mapsize
;
59 mapsize
= (pages
+7)/8;
60 mapsize
= (mapsize
+ ~PAGE_MASK
) & PAGE_MASK
;
61 mapsize
>>= PAGE_SHIFT
;
69 static void __init
link_bootmem(bootmem_data_t
*bdata
)
73 if (list_empty(&bdata_list
)) {
74 list_add(&bdata
->list
, &bdata_list
);
78 list_for_each_entry(ent
, &bdata_list
, list
) {
79 if (bdata
->node_boot_start
< ent
->node_boot_start
) {
80 list_add_tail(&bdata
->list
, &ent
->list
);
84 list_add_tail(&bdata
->list
, &bdata_list
);
88 * Called once to set up the allocator itself.
90 static unsigned long __init
init_bootmem_core(bootmem_data_t
*bdata
,
91 unsigned long mapstart
, unsigned long start
, unsigned long end
)
93 unsigned long mapsize
;
95 mminit_validate_memmodel_limits(&start
, &end
);
96 bdata
->node_bootmem_map
= phys_to_virt(PFN_PHYS(mapstart
));
97 bdata
->node_boot_start
= PFN_PHYS(start
);
98 bdata
->node_low_pfn
= end
;
102 * Initially all pages are reserved - setup_arch() has to
103 * register free RAM areas explicitly.
105 mapsize
= get_mapsize(bdata
);
106 memset(bdata
->node_bootmem_map
, 0xff, mapsize
);
111 unsigned long __init
init_bootmem_node(pg_data_t
*pgdat
, unsigned long freepfn
,
112 unsigned long startpfn
, unsigned long endpfn
)
114 return init_bootmem_core(pgdat
->bdata
, freepfn
, startpfn
, endpfn
);
117 unsigned long __init
init_bootmem(unsigned long start
, unsigned long pages
)
121 return init_bootmem_core(NODE_DATA(0)->bdata
, start
, 0, pages
);
124 static unsigned long __init
free_all_bootmem_core(bootmem_data_t
*bdata
)
128 unsigned long i
, count
;
133 BUG_ON(!bdata
->node_bootmem_map
);
136 /* first extant page of the node */
137 pfn
= PFN_DOWN(bdata
->node_boot_start
);
138 idx
= bdata
->node_low_pfn
- pfn
;
139 map
= bdata
->node_bootmem_map
;
141 * Check if we are aligned to BITS_PER_LONG pages. If so, we might
142 * be able to free page orders of that size at once.
144 if (!(pfn
& (BITS_PER_LONG
-1)))
147 for (i
= 0; i
< idx
; ) {
148 unsigned long v
= ~map
[i
/ BITS_PER_LONG
];
150 if (gofast
&& v
== ~0UL) {
153 page
= pfn_to_page(pfn
);
154 count
+= BITS_PER_LONG
;
155 order
= ffs(BITS_PER_LONG
) - 1;
156 __free_pages_bootmem(page
, order
);
158 page
+= BITS_PER_LONG
;
162 page
= pfn_to_page(pfn
);
163 for (m
= 1; m
&& i
< idx
; m
<<=1, page
++, i
++) {
166 __free_pages_bootmem(page
, 0);
172 pfn
+= BITS_PER_LONG
;
176 * Now free the allocator bitmap itself, it's not
179 page
= virt_to_page(bdata
->node_bootmem_map
);
180 idx
= (get_mapsize(bdata
) + PAGE_SIZE
-1) >> PAGE_SHIFT
;
181 for (i
= 0; i
< idx
; i
++, page
++)
182 __free_pages_bootmem(page
, 0);
184 bdata
->node_bootmem_map
= NULL
;
189 unsigned long __init
free_all_bootmem_node(pg_data_t
*pgdat
)
191 register_page_bootmem_info_node(pgdat
);
192 return free_all_bootmem_core(pgdat
->bdata
);
195 unsigned long __init
free_all_bootmem(void)
197 return free_all_bootmem_core(NODE_DATA(0)->bdata
);
200 static void __init
free_bootmem_core(bootmem_data_t
*bdata
, unsigned long addr
,
203 unsigned long sidx
, eidx
;
209 if (addr
+ size
< bdata
->node_boot_start
||
210 PFN_DOWN(addr
) > bdata
->node_low_pfn
)
213 * round down end of usable mem, partially free pages are
214 * considered reserved.
217 if (addr
>= bdata
->node_boot_start
&& addr
< bdata
->last_success
)
218 bdata
->last_success
= addr
;
221 * Round up to index to the range.
223 if (PFN_UP(addr
) > PFN_DOWN(bdata
->node_boot_start
))
224 sidx
= PFN_UP(addr
) - PFN_DOWN(bdata
->node_boot_start
);
228 eidx
= PFN_DOWN(addr
+ size
- bdata
->node_boot_start
);
229 if (eidx
> bdata
->node_low_pfn
- PFN_DOWN(bdata
->node_boot_start
))
230 eidx
= bdata
->node_low_pfn
- PFN_DOWN(bdata
->node_boot_start
);
232 for (i
= sidx
; i
< eidx
; i
++) {
233 if (unlikely(!test_and_clear_bit(i
, bdata
->node_bootmem_map
)))
238 void __init
free_bootmem_node(pg_data_t
*pgdat
, unsigned long physaddr
,
241 free_bootmem_core(pgdat
->bdata
, physaddr
, size
);
244 void __init
free_bootmem(unsigned long addr
, unsigned long size
)
246 bootmem_data_t
*bdata
;
247 list_for_each_entry(bdata
, &bdata_list
, list
)
248 free_bootmem_core(bdata
, addr
, size
);
252 * Marks a particular physical memory range as unallocatable. Usable RAM
253 * might be used for boot-time allocations - or it might get added
254 * to the free page pool later on.
256 static int __init
can_reserve_bootmem_core(bootmem_data_t
*bdata
,
257 unsigned long addr
, unsigned long size
, int flags
)
259 unsigned long sidx
, eidx
;
264 /* out of range, don't hold other */
265 if (addr
+ size
< bdata
->node_boot_start
||
266 PFN_DOWN(addr
) > bdata
->node_low_pfn
)
270 * Round up to index to the range.
272 if (addr
> bdata
->node_boot_start
)
273 sidx
= PFN_DOWN(addr
- bdata
->node_boot_start
);
277 eidx
= PFN_UP(addr
+ size
- bdata
->node_boot_start
);
278 if (eidx
> bdata
->node_low_pfn
- PFN_DOWN(bdata
->node_boot_start
))
279 eidx
= bdata
->node_low_pfn
- PFN_DOWN(bdata
->node_boot_start
);
281 for (i
= sidx
; i
< eidx
; i
++) {
282 if (test_bit(i
, bdata
->node_bootmem_map
)) {
283 if (flags
& BOOTMEM_EXCLUSIVE
)
292 static void __init
reserve_bootmem_core(bootmem_data_t
*bdata
,
293 unsigned long addr
, unsigned long size
, int flags
)
295 unsigned long sidx
, eidx
;
301 if (addr
+ size
< bdata
->node_boot_start
||
302 PFN_DOWN(addr
) > bdata
->node_low_pfn
)
306 * Round up to index to the range.
308 if (addr
> bdata
->node_boot_start
)
309 sidx
= PFN_DOWN(addr
- bdata
->node_boot_start
);
313 eidx
= PFN_UP(addr
+ size
- bdata
->node_boot_start
);
314 if (eidx
> bdata
->node_low_pfn
- PFN_DOWN(bdata
->node_boot_start
))
315 eidx
= bdata
->node_low_pfn
- PFN_DOWN(bdata
->node_boot_start
);
317 for (i
= sidx
; i
< eidx
; i
++) {
318 if (test_and_set_bit(i
, bdata
->node_bootmem_map
)) {
319 #ifdef CONFIG_DEBUG_BOOTMEM
320 printk("hm, page %08lx reserved twice.\n", i
*PAGE_SIZE
);
326 int __init
reserve_bootmem_node(pg_data_t
*pgdat
, unsigned long physaddr
,
327 unsigned long size
, int flags
)
331 ret
= can_reserve_bootmem_core(pgdat
->bdata
, physaddr
, size
, flags
);
334 reserve_bootmem_core(pgdat
->bdata
, physaddr
, size
, flags
);
338 #ifndef CONFIG_HAVE_ARCH_BOOTMEM_NODE
339 int __init
reserve_bootmem(unsigned long addr
, unsigned long size
,
342 bootmem_data_t
*bdata
;
345 list_for_each_entry(bdata
, &bdata_list
, list
) {
346 ret
= can_reserve_bootmem_core(bdata
, addr
, size
, flags
);
350 list_for_each_entry(bdata
, &bdata_list
, list
)
351 reserve_bootmem_core(bdata
, addr
, size
, flags
);
355 #endif /* !CONFIG_HAVE_ARCH_BOOTMEM_NODE */
358 * We 'merge' subsequent allocations to save space. We might 'lose'
359 * some fraction of a page if allocations cannot be satisfied due to
360 * size constraints on boxes where there is physical RAM space
361 * fragmentation - in these cases (mostly large memory boxes) this
364 * On low memory boxes we get it right in 100% of the cases.
366 * alignment has to be a power of 2 value.
368 * NOTE: This function is _not_ reentrant.
371 alloc_bootmem_core(struct bootmem_data
*bdata
, unsigned long size
,
372 unsigned long align
, unsigned long goal
, unsigned long limit
)
374 unsigned long areasize
, preferred
;
375 unsigned long i
, start
= 0, incr
, eidx
, end_pfn
;
377 unsigned long node_boot_start
;
378 void *node_bootmem_map
;
381 printk("alloc_bootmem_core(): zero-sized request\n");
384 BUG_ON(align
& (align
-1));
386 /* on nodes without memory - bootmem_map is NULL */
387 if (!bdata
->node_bootmem_map
)
390 /* bdata->node_boot_start is supposed to be (12+6)bits alignment on x86_64 ? */
391 node_boot_start
= bdata
->node_boot_start
;
392 node_bootmem_map
= bdata
->node_bootmem_map
;
394 node_boot_start
= ALIGN(bdata
->node_boot_start
, align
);
395 if (node_boot_start
> bdata
->node_boot_start
)
396 node_bootmem_map
= (unsigned long *)bdata
->node_bootmem_map
+
397 PFN_DOWN(node_boot_start
- bdata
->node_boot_start
)/BITS_PER_LONG
;
400 if (limit
&& node_boot_start
>= limit
)
403 end_pfn
= bdata
->node_low_pfn
;
404 limit
= PFN_DOWN(limit
);
405 if (limit
&& end_pfn
> limit
)
408 eidx
= end_pfn
- PFN_DOWN(node_boot_start
);
411 * We try to allocate bootmem pages above 'goal'
412 * first, then we try to allocate lower pages.
415 if (goal
&& PFN_DOWN(goal
) < end_pfn
) {
416 if (goal
> node_boot_start
)
417 preferred
= goal
- node_boot_start
;
419 if (bdata
->last_success
> node_boot_start
&&
420 bdata
->last_success
- node_boot_start
>= preferred
)
421 if (!limit
|| (limit
&& limit
> bdata
->last_success
))
422 preferred
= bdata
->last_success
- node_boot_start
;
425 preferred
= PFN_DOWN(ALIGN(preferred
, align
));
426 areasize
= (size
+ PAGE_SIZE
-1) / PAGE_SIZE
;
427 incr
= align
>> PAGE_SHIFT
? : 1;
430 for (i
= preferred
; i
< eidx
;) {
433 i
= find_next_zero_bit(node_bootmem_map
, eidx
, i
);
437 if (test_bit(i
, node_bootmem_map
)) {
441 for (j
= i
+ 1; j
< i
+ areasize
; ++j
) {
444 if (test_bit(j
, node_bootmem_map
))
462 bdata
->last_success
= PFN_PHYS(start
) + node_boot_start
;
463 BUG_ON(start
>= eidx
);
466 * Is the next page of the previous allocation-end the start
467 * of this allocation's buffer? If yes then we can 'merge'
468 * the previous partial page with this allocation.
470 if (align
< PAGE_SIZE
&&
471 bdata
->last_offset
&& bdata
->last_pos
+1 == start
) {
472 unsigned long offset
, remaining_size
;
473 offset
= ALIGN(bdata
->last_offset
, align
);
474 BUG_ON(offset
> PAGE_SIZE
);
475 remaining_size
= PAGE_SIZE
- offset
;
476 if (size
< remaining_size
) {
478 /* last_pos unchanged */
479 bdata
->last_offset
= offset
+ size
;
480 ret
= phys_to_virt(bdata
->last_pos
* PAGE_SIZE
+
481 offset
+ node_boot_start
);
483 remaining_size
= size
- remaining_size
;
484 areasize
= (remaining_size
+ PAGE_SIZE
-1) / PAGE_SIZE
;
485 ret
= phys_to_virt(bdata
->last_pos
* PAGE_SIZE
+
486 offset
+ node_boot_start
);
487 bdata
->last_pos
= start
+ areasize
- 1;
488 bdata
->last_offset
= remaining_size
;
490 bdata
->last_offset
&= ~PAGE_MASK
;
492 bdata
->last_pos
= start
+ areasize
- 1;
493 bdata
->last_offset
= size
& ~PAGE_MASK
;
494 ret
= phys_to_virt(start
* PAGE_SIZE
+ node_boot_start
);
498 * Reserve the area now:
500 for (i
= start
; i
< start
+ areasize
; i
++)
501 if (unlikely(test_and_set_bit(i
, node_bootmem_map
)))
503 memset(ret
, 0, size
);
507 void * __init
__alloc_bootmem_nopanic(unsigned long size
, unsigned long align
,
510 bootmem_data_t
*bdata
;
513 list_for_each_entry(bdata
, &bdata_list
, list
) {
514 ptr
= alloc_bootmem_core(bdata
, size
, align
, goal
, 0);
521 void * __init
__alloc_bootmem(unsigned long size
, unsigned long align
,
524 void *mem
= __alloc_bootmem_nopanic(size
,align
,goal
);
529 * Whoops, we cannot satisfy the allocation request.
531 printk(KERN_ALERT
"bootmem alloc of %lu bytes failed!\n", size
);
532 panic("Out of memory");
536 void * __init
__alloc_bootmem_node(pg_data_t
*pgdat
, unsigned long size
,
537 unsigned long align
, unsigned long goal
)
541 ptr
= alloc_bootmem_core(pgdat
->bdata
, size
, align
, goal
, 0);
545 return __alloc_bootmem(size
, align
, goal
);
548 #ifdef CONFIG_SPARSEMEM
549 void * __init
alloc_bootmem_section(unsigned long size
,
550 unsigned long section_nr
)
553 unsigned long limit
, goal
, start_nr
, end_nr
, pfn
;
554 struct pglist_data
*pgdat
;
556 pfn
= section_nr_to_pfn(section_nr
);
557 goal
= PFN_PHYS(pfn
);
558 limit
= PFN_PHYS(section_nr_to_pfn(section_nr
+ 1)) - 1;
559 pgdat
= NODE_DATA(early_pfn_to_nid(pfn
));
560 ptr
= alloc_bootmem_core(pgdat
->bdata
, size
, SMP_CACHE_BYTES
, goal
,
566 start_nr
= pfn_to_section_nr(PFN_DOWN(__pa(ptr
)));
567 end_nr
= pfn_to_section_nr(PFN_DOWN(__pa(ptr
) + size
));
568 if (start_nr
!= section_nr
|| end_nr
!= section_nr
) {
569 printk(KERN_WARNING
"alloc_bootmem failed on section %ld.\n",
571 free_bootmem_core(pgdat
->bdata
, __pa(ptr
), size
);
579 void * __init
__alloc_bootmem_node_nopanic(pg_data_t
*pgdat
, unsigned long size
,
580 unsigned long align
, unsigned long goal
)
584 ptr
= alloc_bootmem_core(pgdat
->bdata
, size
, align
, goal
, 0);
588 return __alloc_bootmem_nopanic(size
, align
, goal
);
591 #ifndef ARCH_LOW_ADDRESS_LIMIT
592 #define ARCH_LOW_ADDRESS_LIMIT 0xffffffffUL
595 void * __init
__alloc_bootmem_low(unsigned long size
, unsigned long align
,
598 bootmem_data_t
*bdata
;
601 list_for_each_entry(bdata
, &bdata_list
, list
) {
602 ptr
= alloc_bootmem_core(bdata
, size
, align
, goal
,
603 ARCH_LOW_ADDRESS_LIMIT
);
609 * Whoops, we cannot satisfy the allocation request.
611 printk(KERN_ALERT
"low bootmem alloc of %lu bytes failed!\n", size
);
612 panic("Out of low memory");
616 void * __init
__alloc_bootmem_low_node(pg_data_t
*pgdat
, unsigned long size
,
617 unsigned long align
, unsigned long goal
)
619 return alloc_bootmem_core(pgdat
->bdata
, size
, align
, goal
,
620 ARCH_LOW_ADDRESS_LIMIT
);