Loading...
1/*
2 * bootmem - A boot-time physical memory allocator and configurator
3 *
4 * Copyright (C) 1999 Ingo Molnar
5 * 1999 Kanoj Sarcar, SGI
6 * 2008 Johannes Weiner
7 *
8 * Access to this subsystem has to be serialized externally (which is true
9 * for the boot process anyway).
10 */
11#include <linux/init.h>
12#include <linux/pfn.h>
13#include <linux/slab.h>
14#include <linux/bootmem.h>
15#include <linux/export.h>
16#include <linux/kmemleak.h>
17#include <linux/range.h>
18#include <linux/memblock.h>
19
20#include <asm/bug.h>
21#include <asm/io.h>
22#include <asm/processor.h>
23
24#include "internal.h"
25
26#ifndef CONFIG_NEED_MULTIPLE_NODES
27struct pglist_data __refdata contig_page_data = {
28 .bdata = &bootmem_node_data[0]
29};
30EXPORT_SYMBOL(contig_page_data);
31#endif
32
33unsigned long max_low_pfn;
34unsigned long min_low_pfn;
35unsigned long max_pfn;
36
37bootmem_data_t bootmem_node_data[MAX_NUMNODES] __initdata;
38
39static struct list_head bdata_list __initdata = LIST_HEAD_INIT(bdata_list);
40
41static int bootmem_debug;
42
43static int __init bootmem_debug_setup(char *buf)
44{
45 bootmem_debug = 1;
46 return 0;
47}
48early_param("bootmem_debug", bootmem_debug_setup);
49
50#define bdebug(fmt, args...) ({ \
51 if (unlikely(bootmem_debug)) \
52 printk(KERN_INFO \
53 "bootmem::%s " fmt, \
54 __func__, ## args); \
55})
56
57static unsigned long __init bootmap_bytes(unsigned long pages)
58{
59 unsigned long bytes = DIV_ROUND_UP(pages, 8);
60
61 return ALIGN(bytes, sizeof(long));
62}
63
64/**
65 * bootmem_bootmap_pages - calculate bitmap size in pages
66 * @pages: number of pages the bitmap has to represent
67 */
68unsigned long __init bootmem_bootmap_pages(unsigned long pages)
69{
70 unsigned long bytes = bootmap_bytes(pages);
71
72 return PAGE_ALIGN(bytes) >> PAGE_SHIFT;
73}
74
75/*
76 * link bdata in order
77 */
78static void __init link_bootmem(bootmem_data_t *bdata)
79{
80 bootmem_data_t *ent;
81
82 list_for_each_entry(ent, &bdata_list, list) {
83 if (bdata->node_min_pfn < ent->node_min_pfn) {
84 list_add_tail(&bdata->list, &ent->list);
85 return;
86 }
87 }
88
89 list_add_tail(&bdata->list, &bdata_list);
90}
91
92/*
93 * Called once to set up the allocator itself.
94 */
95static unsigned long __init init_bootmem_core(bootmem_data_t *bdata,
96 unsigned long mapstart, unsigned long start, unsigned long end)
97{
98 unsigned long mapsize;
99
100 mminit_validate_memmodel_limits(&start, &end);
101 bdata->node_bootmem_map = phys_to_virt(PFN_PHYS(mapstart));
102 bdata->node_min_pfn = start;
103 bdata->node_low_pfn = end;
104 link_bootmem(bdata);
105
106 /*
107 * Initially all pages are reserved - setup_arch() has to
108 * register free RAM areas explicitly.
109 */
110 mapsize = bootmap_bytes(end - start);
111 memset(bdata->node_bootmem_map, 0xff, mapsize);
112
113 bdebug("nid=%td start=%lx map=%lx end=%lx mapsize=%lx\n",
114 bdata - bootmem_node_data, start, mapstart, end, mapsize);
115
116 return mapsize;
117}
118
119/**
120 * init_bootmem_node - register a node as boot memory
121 * @pgdat: node to register
122 * @freepfn: pfn where the bitmap for this node is to be placed
123 * @startpfn: first pfn on the node
124 * @endpfn: first pfn after the node
125 *
126 * Returns the number of bytes needed to hold the bitmap for this node.
127 */
128unsigned long __init init_bootmem_node(pg_data_t *pgdat, unsigned long freepfn,
129 unsigned long startpfn, unsigned long endpfn)
130{
131 return init_bootmem_core(pgdat->bdata, freepfn, startpfn, endpfn);
132}
133
134/**
135 * init_bootmem - register boot memory
136 * @start: pfn where the bitmap is to be placed
137 * @pages: number of available physical pages
138 *
139 * Returns the number of bytes needed to hold the bitmap.
140 */
141unsigned long __init init_bootmem(unsigned long start, unsigned long pages)
142{
143 max_low_pfn = pages;
144 min_low_pfn = start;
145 return init_bootmem_core(NODE_DATA(0)->bdata, start, 0, pages);
146}
147
148/*
149 * free_bootmem_late - free bootmem pages directly to page allocator
150 * @addr: starting physical address of the range
151 * @size: size of the range in bytes
152 *
153 * This is only useful when the bootmem allocator has already been torn
154 * down, but we are still initializing the system. Pages are given directly
155 * to the page allocator, no bootmem metadata is updated because it is gone.
156 */
157void __init free_bootmem_late(unsigned long physaddr, unsigned long size)
158{
159 unsigned long cursor, end;
160
161 kmemleak_free_part(__va(physaddr), size);
162
163 cursor = PFN_UP(physaddr);
164 end = PFN_DOWN(physaddr + size);
165
166 for (; cursor < end; cursor++) {
167 __free_pages_bootmem(pfn_to_page(cursor), 0);
168 totalram_pages++;
169 }
170}
171
172static unsigned long __init free_all_bootmem_core(bootmem_data_t *bdata)
173{
174 struct page *page;
175 unsigned long *map, start, end, pages, count = 0;
176
177 if (!bdata->node_bootmem_map)
178 return 0;
179
180 map = bdata->node_bootmem_map;
181 start = bdata->node_min_pfn;
182 end = bdata->node_low_pfn;
183
184 bdebug("nid=%td start=%lx end=%lx\n",
185 bdata - bootmem_node_data, start, end);
186
187 while (start < end) {
188 unsigned long idx, vec;
189 unsigned shift;
190
191 idx = start - bdata->node_min_pfn;
192 shift = idx & (BITS_PER_LONG - 1);
193 /*
194 * vec holds at most BITS_PER_LONG map bits,
195 * bit 0 corresponds to start.
196 */
197 vec = ~map[idx / BITS_PER_LONG];
198
199 if (shift) {
200 vec >>= shift;
201 if (end - start >= BITS_PER_LONG)
202 vec |= ~map[idx / BITS_PER_LONG + 1] <<
203 (BITS_PER_LONG - shift);
204 }
205 /*
206 * If we have a properly aligned and fully unreserved
207 * BITS_PER_LONG block of pages in front of us, free
208 * it in one go.
209 */
210 if (IS_ALIGNED(start, BITS_PER_LONG) && vec == ~0UL) {
211 int order = ilog2(BITS_PER_LONG);
212
213 __free_pages_bootmem(pfn_to_page(start), order);
214 count += BITS_PER_LONG;
215 start += BITS_PER_LONG;
216 } else {
217 unsigned long cur = start;
218
219 start = ALIGN(start + 1, BITS_PER_LONG);
220 while (vec && cur != start) {
221 if (vec & 1) {
222 page = pfn_to_page(cur);
223 __free_pages_bootmem(page, 0);
224 count++;
225 }
226 vec >>= 1;
227 ++cur;
228 }
229 }
230 }
231
232 page = virt_to_page(bdata->node_bootmem_map);
233 pages = bdata->node_low_pfn - bdata->node_min_pfn;
234 pages = bootmem_bootmap_pages(pages);
235 count += pages;
236 while (pages--)
237 __free_pages_bootmem(page++, 0);
238
239 bdebug("nid=%td released=%lx\n", bdata - bootmem_node_data, count);
240
241 return count;
242}
243
244static int reset_managed_pages_done __initdata;
245
246static inline void __init reset_node_managed_pages(pg_data_t *pgdat)
247{
248 struct zone *z;
249
250 if (reset_managed_pages_done)
251 return;
252
253 for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++)
254 z->managed_pages = 0;
255}
256
257void __init reset_all_zones_managed_pages(void)
258{
259 struct pglist_data *pgdat;
260
261 for_each_online_pgdat(pgdat)
262 reset_node_managed_pages(pgdat);
263 reset_managed_pages_done = 1;
264}
265
266/**
267 * free_all_bootmem - release free pages to the buddy allocator
268 *
269 * Returns the number of pages actually released.
270 */
271unsigned long __init free_all_bootmem(void)
272{
273 unsigned long total_pages = 0;
274 bootmem_data_t *bdata;
275
276 reset_all_zones_managed_pages();
277
278 list_for_each_entry(bdata, &bdata_list, list)
279 total_pages += free_all_bootmem_core(bdata);
280
281 totalram_pages += total_pages;
282
283 return total_pages;
284}
285
286static void __init __free(bootmem_data_t *bdata,
287 unsigned long sidx, unsigned long eidx)
288{
289 unsigned long idx;
290
291 bdebug("nid=%td start=%lx end=%lx\n", bdata - bootmem_node_data,
292 sidx + bdata->node_min_pfn,
293 eidx + bdata->node_min_pfn);
294
295 if (bdata->hint_idx > sidx)
296 bdata->hint_idx = sidx;
297
298 for (idx = sidx; idx < eidx; idx++)
299 if (!test_and_clear_bit(idx, bdata->node_bootmem_map))
300 BUG();
301}
302
303static int __init __reserve(bootmem_data_t *bdata, unsigned long sidx,
304 unsigned long eidx, int flags)
305{
306 unsigned long idx;
307 int exclusive = flags & BOOTMEM_EXCLUSIVE;
308
309 bdebug("nid=%td start=%lx end=%lx flags=%x\n",
310 bdata - bootmem_node_data,
311 sidx + bdata->node_min_pfn,
312 eidx + bdata->node_min_pfn,
313 flags);
314
315 for (idx = sidx; idx < eidx; idx++)
316 if (test_and_set_bit(idx, bdata->node_bootmem_map)) {
317 if (exclusive) {
318 __free(bdata, sidx, idx);
319 return -EBUSY;
320 }
321 bdebug("silent double reserve of PFN %lx\n",
322 idx + bdata->node_min_pfn);
323 }
324 return 0;
325}
326
327static int __init mark_bootmem_node(bootmem_data_t *bdata,
328 unsigned long start, unsigned long end,
329 int reserve, int flags)
330{
331 unsigned long sidx, eidx;
332
333 bdebug("nid=%td start=%lx end=%lx reserve=%d flags=%x\n",
334 bdata - bootmem_node_data, start, end, reserve, flags);
335
336 BUG_ON(start < bdata->node_min_pfn);
337 BUG_ON(end > bdata->node_low_pfn);
338
339 sidx = start - bdata->node_min_pfn;
340 eidx = end - bdata->node_min_pfn;
341
342 if (reserve)
343 return __reserve(bdata, sidx, eidx, flags);
344 else
345 __free(bdata, sidx, eidx);
346 return 0;
347}
348
349static int __init mark_bootmem(unsigned long start, unsigned long end,
350 int reserve, int flags)
351{
352 unsigned long pos;
353 bootmem_data_t *bdata;
354
355 pos = start;
356 list_for_each_entry(bdata, &bdata_list, list) {
357 int err;
358 unsigned long max;
359
360 if (pos < bdata->node_min_pfn ||
361 pos >= bdata->node_low_pfn) {
362 BUG_ON(pos != start);
363 continue;
364 }
365
366 max = min(bdata->node_low_pfn, end);
367
368 err = mark_bootmem_node(bdata, pos, max, reserve, flags);
369 if (reserve && err) {
370 mark_bootmem(start, pos, 0, 0);
371 return err;
372 }
373
374 if (max == end)
375 return 0;
376 pos = bdata->node_low_pfn;
377 }
378 BUG();
379}
380
381/**
382 * free_bootmem_node - mark a page range as usable
383 * @pgdat: node the range resides on
384 * @physaddr: starting address of the range
385 * @size: size of the range in bytes
386 *
387 * Partial pages will be considered reserved and left as they are.
388 *
389 * The range must reside completely on the specified node.
390 */
391void __init free_bootmem_node(pg_data_t *pgdat, unsigned long physaddr,
392 unsigned long size)
393{
394 unsigned long start, end;
395
396 kmemleak_free_part(__va(physaddr), size);
397
398 start = PFN_UP(physaddr);
399 end = PFN_DOWN(physaddr + size);
400
401 mark_bootmem_node(pgdat->bdata, start, end, 0, 0);
402}
403
404/**
405 * free_bootmem - mark a page range as usable
406 * @addr: starting physical address of the range
407 * @size: size of the range in bytes
408 *
409 * Partial pages will be considered reserved and left as they are.
410 *
411 * The range must be contiguous but may span node boundaries.
412 */
413void __init free_bootmem(unsigned long physaddr, unsigned long size)
414{
415 unsigned long start, end;
416
417 kmemleak_free_part(__va(physaddr), size);
418
419 start = PFN_UP(physaddr);
420 end = PFN_DOWN(physaddr + size);
421
422 mark_bootmem(start, end, 0, 0);
423}
424
425/**
426 * reserve_bootmem_node - mark a page range as reserved
427 * @pgdat: node the range resides on
428 * @physaddr: starting address of the range
429 * @size: size of the range in bytes
430 * @flags: reservation flags (see linux/bootmem.h)
431 *
432 * Partial pages will be reserved.
433 *
434 * The range must reside completely on the specified node.
435 */
436int __init reserve_bootmem_node(pg_data_t *pgdat, unsigned long physaddr,
437 unsigned long size, int flags)
438{
439 unsigned long start, end;
440
441 start = PFN_DOWN(physaddr);
442 end = PFN_UP(physaddr + size);
443
444 return mark_bootmem_node(pgdat->bdata, start, end, 1, flags);
445}
446
447/**
448 * reserve_bootmem - mark a page range as reserved
449 * @addr: starting address of the range
450 * @size: size of the range in bytes
451 * @flags: reservation flags (see linux/bootmem.h)
452 *
453 * Partial pages will be reserved.
454 *
455 * The range must be contiguous but may span node boundaries.
456 */
457int __init reserve_bootmem(unsigned long addr, unsigned long size,
458 int flags)
459{
460 unsigned long start, end;
461
462 start = PFN_DOWN(addr);
463 end = PFN_UP(addr + size);
464
465 return mark_bootmem(start, end, 1, flags);
466}
467
468static unsigned long __init align_idx(struct bootmem_data *bdata,
469 unsigned long idx, unsigned long step)
470{
471 unsigned long base = bdata->node_min_pfn;
472
473 /*
474 * Align the index with respect to the node start so that the
475 * combination of both satisfies the requested alignment.
476 */
477
478 return ALIGN(base + idx, step) - base;
479}
480
481static unsigned long __init align_off(struct bootmem_data *bdata,
482 unsigned long off, unsigned long align)
483{
484 unsigned long base = PFN_PHYS(bdata->node_min_pfn);
485
486 /* Same as align_idx for byte offsets */
487
488 return ALIGN(base + off, align) - base;
489}
490
491static void * __init alloc_bootmem_bdata(struct bootmem_data *bdata,
492 unsigned long size, unsigned long align,
493 unsigned long goal, unsigned long limit)
494{
495 unsigned long fallback = 0;
496 unsigned long min, max, start, sidx, midx, step;
497
498 bdebug("nid=%td size=%lx [%lu pages] align=%lx goal=%lx limit=%lx\n",
499 bdata - bootmem_node_data, size, PAGE_ALIGN(size) >> PAGE_SHIFT,
500 align, goal, limit);
501
502 BUG_ON(!size);
503 BUG_ON(align & (align - 1));
504 BUG_ON(limit && goal + size > limit);
505
506 if (!bdata->node_bootmem_map)
507 return NULL;
508
509 min = bdata->node_min_pfn;
510 max = bdata->node_low_pfn;
511
512 goal >>= PAGE_SHIFT;
513 limit >>= PAGE_SHIFT;
514
515 if (limit && max > limit)
516 max = limit;
517 if (max <= min)
518 return NULL;
519
520 step = max(align >> PAGE_SHIFT, 1UL);
521
522 if (goal && min < goal && goal < max)
523 start = ALIGN(goal, step);
524 else
525 start = ALIGN(min, step);
526
527 sidx = start - bdata->node_min_pfn;
528 midx = max - bdata->node_min_pfn;
529
530 if (bdata->hint_idx > sidx) {
531 /*
532 * Handle the valid case of sidx being zero and still
533 * catch the fallback below.
534 */
535 fallback = sidx + 1;
536 sidx = align_idx(bdata, bdata->hint_idx, step);
537 }
538
539 while (1) {
540 int merge;
541 void *region;
542 unsigned long eidx, i, start_off, end_off;
543find_block:
544 sidx = find_next_zero_bit(bdata->node_bootmem_map, midx, sidx);
545 sidx = align_idx(bdata, sidx, step);
546 eidx = sidx + PFN_UP(size);
547
548 if (sidx >= midx || eidx > midx)
549 break;
550
551 for (i = sidx; i < eidx; i++)
552 if (test_bit(i, bdata->node_bootmem_map)) {
553 sidx = align_idx(bdata, i, step);
554 if (sidx == i)
555 sidx += step;
556 goto find_block;
557 }
558
559 if (bdata->last_end_off & (PAGE_SIZE - 1) &&
560 PFN_DOWN(bdata->last_end_off) + 1 == sidx)
561 start_off = align_off(bdata, bdata->last_end_off, align);
562 else
563 start_off = PFN_PHYS(sidx);
564
565 merge = PFN_DOWN(start_off) < sidx;
566 end_off = start_off + size;
567
568 bdata->last_end_off = end_off;
569 bdata->hint_idx = PFN_UP(end_off);
570
571 /*
572 * Reserve the area now:
573 */
574 if (__reserve(bdata, PFN_DOWN(start_off) + merge,
575 PFN_UP(end_off), BOOTMEM_EXCLUSIVE))
576 BUG();
577
578 region = phys_to_virt(PFN_PHYS(bdata->node_min_pfn) +
579 start_off);
580 memset(region, 0, size);
581 /*
582 * The min_count is set to 0 so that bootmem allocated blocks
583 * are never reported as leaks.
584 */
585 kmemleak_alloc(region, size, 0, 0);
586 return region;
587 }
588
589 if (fallback) {
590 sidx = align_idx(bdata, fallback - 1, step);
591 fallback = 0;
592 goto find_block;
593 }
594
595 return NULL;
596}
597
598static void * __init alloc_bootmem_core(unsigned long size,
599 unsigned long align,
600 unsigned long goal,
601 unsigned long limit)
602{
603 bootmem_data_t *bdata;
604 void *region;
605
606 if (WARN_ON_ONCE(slab_is_available()))
607 return kzalloc(size, GFP_NOWAIT);
608
609 list_for_each_entry(bdata, &bdata_list, list) {
610 if (goal && bdata->node_low_pfn <= PFN_DOWN(goal))
611 continue;
612 if (limit && bdata->node_min_pfn >= PFN_DOWN(limit))
613 break;
614
615 region = alloc_bootmem_bdata(bdata, size, align, goal, limit);
616 if (region)
617 return region;
618 }
619
620 return NULL;
621}
622
623static void * __init ___alloc_bootmem_nopanic(unsigned long size,
624 unsigned long align,
625 unsigned long goal,
626 unsigned long limit)
627{
628 void *ptr;
629
630restart:
631 ptr = alloc_bootmem_core(size, align, goal, limit);
632 if (ptr)
633 return ptr;
634 if (goal) {
635 goal = 0;
636 goto restart;
637 }
638
639 return NULL;
640}
641
642/**
643 * __alloc_bootmem_nopanic - allocate boot memory without panicking
644 * @size: size of the request in bytes
645 * @align: alignment of the region
646 * @goal: preferred starting address of the region
647 *
648 * The goal is dropped if it can not be satisfied and the allocation will
649 * fall back to memory below @goal.
650 *
651 * Allocation may happen on any node in the system.
652 *
653 * Returns NULL on failure.
654 */
655void * __init __alloc_bootmem_nopanic(unsigned long size, unsigned long align,
656 unsigned long goal)
657{
658 unsigned long limit = 0;
659
660 return ___alloc_bootmem_nopanic(size, align, goal, limit);
661}
662
663static void * __init ___alloc_bootmem(unsigned long size, unsigned long align,
664 unsigned long goal, unsigned long limit)
665{
666 void *mem = ___alloc_bootmem_nopanic(size, align, goal, limit);
667
668 if (mem)
669 return mem;
670 /*
671 * Whoops, we cannot satisfy the allocation request.
672 */
673 printk(KERN_ALERT "bootmem alloc of %lu bytes failed!\n", size);
674 panic("Out of memory");
675 return NULL;
676}
677
678/**
679 * __alloc_bootmem - allocate boot memory
680 * @size: size of the request in bytes
681 * @align: alignment of the region
682 * @goal: preferred starting address of the region
683 *
684 * The goal is dropped if it can not be satisfied and the allocation will
685 * fall back to memory below @goal.
686 *
687 * Allocation may happen on any node in the system.
688 *
689 * The function panics if the request can not be satisfied.
690 */
691void * __init __alloc_bootmem(unsigned long size, unsigned long align,
692 unsigned long goal)
693{
694 unsigned long limit = 0;
695
696 return ___alloc_bootmem(size, align, goal, limit);
697}
698
699void * __init ___alloc_bootmem_node_nopanic(pg_data_t *pgdat,
700 unsigned long size, unsigned long align,
701 unsigned long goal, unsigned long limit)
702{
703 void *ptr;
704
705 if (WARN_ON_ONCE(slab_is_available()))
706 return kzalloc(size, GFP_NOWAIT);
707again:
708
709 /* do not panic in alloc_bootmem_bdata() */
710 if (limit && goal + size > limit)
711 limit = 0;
712
713 ptr = alloc_bootmem_bdata(pgdat->bdata, size, align, goal, limit);
714 if (ptr)
715 return ptr;
716
717 ptr = alloc_bootmem_core(size, align, goal, limit);
718 if (ptr)
719 return ptr;
720
721 if (goal) {
722 goal = 0;
723 goto again;
724 }
725
726 return NULL;
727}
728
729void * __init __alloc_bootmem_node_nopanic(pg_data_t *pgdat, unsigned long size,
730 unsigned long align, unsigned long goal)
731{
732 if (WARN_ON_ONCE(slab_is_available()))
733 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
734
735 return ___alloc_bootmem_node_nopanic(pgdat, size, align, goal, 0);
736}
737
738void * __init ___alloc_bootmem_node(pg_data_t *pgdat, unsigned long size,
739 unsigned long align, unsigned long goal,
740 unsigned long limit)
741{
742 void *ptr;
743
744 ptr = ___alloc_bootmem_node_nopanic(pgdat, size, align, goal, 0);
745 if (ptr)
746 return ptr;
747
748 printk(KERN_ALERT "bootmem alloc of %lu bytes failed!\n", size);
749 panic("Out of memory");
750 return NULL;
751}
752
753/**
754 * __alloc_bootmem_node - allocate boot memory from a specific node
755 * @pgdat: node to allocate from
756 * @size: size of the request in bytes
757 * @align: alignment of the region
758 * @goal: preferred starting address of the region
759 *
760 * The goal is dropped if it can not be satisfied and the allocation will
761 * fall back to memory below @goal.
762 *
763 * Allocation may fall back to any node in the system if the specified node
764 * can not hold the requested memory.
765 *
766 * The function panics if the request can not be satisfied.
767 */
768void * __init __alloc_bootmem_node(pg_data_t *pgdat, unsigned long size,
769 unsigned long align, unsigned long goal)
770{
771 if (WARN_ON_ONCE(slab_is_available()))
772 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
773
774 return ___alloc_bootmem_node(pgdat, size, align, goal, 0);
775}
776
777void * __init __alloc_bootmem_node_high(pg_data_t *pgdat, unsigned long size,
778 unsigned long align, unsigned long goal)
779{
780#ifdef MAX_DMA32_PFN
781 unsigned long end_pfn;
782
783 if (WARN_ON_ONCE(slab_is_available()))
784 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
785
786 /* update goal according ...MAX_DMA32_PFN */
787 end_pfn = pgdat_end_pfn(pgdat);
788
789 if (end_pfn > MAX_DMA32_PFN + (128 >> (20 - PAGE_SHIFT)) &&
790 (goal >> PAGE_SHIFT) < MAX_DMA32_PFN) {
791 void *ptr;
792 unsigned long new_goal;
793
794 new_goal = MAX_DMA32_PFN << PAGE_SHIFT;
795 ptr = alloc_bootmem_bdata(pgdat->bdata, size, align,
796 new_goal, 0);
797 if (ptr)
798 return ptr;
799 }
800#endif
801
802 return __alloc_bootmem_node(pgdat, size, align, goal);
803
804}
805
806#ifndef ARCH_LOW_ADDRESS_LIMIT
807#define ARCH_LOW_ADDRESS_LIMIT 0xffffffffUL
808#endif
809
810/**
811 * __alloc_bootmem_low - allocate low boot memory
812 * @size: size of the request in bytes
813 * @align: alignment of the region
814 * @goal: preferred starting address of the region
815 *
816 * The goal is dropped if it can not be satisfied and the allocation will
817 * fall back to memory below @goal.
818 *
819 * Allocation may happen on any node in the system.
820 *
821 * The function panics if the request can not be satisfied.
822 */
823void * __init __alloc_bootmem_low(unsigned long size, unsigned long align,
824 unsigned long goal)
825{
826 return ___alloc_bootmem(size, align, goal, ARCH_LOW_ADDRESS_LIMIT);
827}
828
829void * __init __alloc_bootmem_low_nopanic(unsigned long size,
830 unsigned long align,
831 unsigned long goal)
832{
833 return ___alloc_bootmem_nopanic(size, align, goal,
834 ARCH_LOW_ADDRESS_LIMIT);
835}
836
837/**
838 * __alloc_bootmem_low_node - allocate low boot memory from a specific node
839 * @pgdat: node to allocate from
840 * @size: size of the request in bytes
841 * @align: alignment of the region
842 * @goal: preferred starting address of the region
843 *
844 * The goal is dropped if it can not be satisfied and the allocation will
845 * fall back to memory below @goal.
846 *
847 * Allocation may fall back to any node in the system if the specified node
848 * can not hold the requested memory.
849 *
850 * The function panics if the request can not be satisfied.
851 */
852void * __init __alloc_bootmem_low_node(pg_data_t *pgdat, unsigned long size,
853 unsigned long align, unsigned long goal)
854{
855 if (WARN_ON_ONCE(slab_is_available()))
856 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
857
858 return ___alloc_bootmem_node(pgdat, size, align,
859 goal, ARCH_LOW_ADDRESS_LIMIT);
860}
1/*
2 * bootmem - A boot-time physical memory allocator and configurator
3 *
4 * Copyright (C) 1999 Ingo Molnar
5 * 1999 Kanoj Sarcar, SGI
6 * 2008 Johannes Weiner
7 *
8 * Access to this subsystem has to be serialized externally (which is true
9 * for the boot process anyway).
10 */
11#include <linux/init.h>
12#include <linux/pfn.h>
13#include <linux/slab.h>
14#include <linux/bootmem.h>
15#include <linux/module.h>
16#include <linux/kmemleak.h>
17#include <linux/range.h>
18#include <linux/memblock.h>
19
20#include <asm/bug.h>
21#include <asm/io.h>
22#include <asm/processor.h>
23
24#include "internal.h"
25
26#ifndef CONFIG_NEED_MULTIPLE_NODES
27struct pglist_data __refdata contig_page_data = {
28 .bdata = &bootmem_node_data[0]
29};
30EXPORT_SYMBOL(contig_page_data);
31#endif
32
33unsigned long max_low_pfn;
34unsigned long min_low_pfn;
35unsigned long max_pfn;
36
37bootmem_data_t bootmem_node_data[MAX_NUMNODES] __initdata;
38
39static struct list_head bdata_list __initdata = LIST_HEAD_INIT(bdata_list);
40
41static int bootmem_debug;
42
43static int __init bootmem_debug_setup(char *buf)
44{
45 bootmem_debug = 1;
46 return 0;
47}
48early_param("bootmem_debug", bootmem_debug_setup);
49
50#define bdebug(fmt, args...) ({ \
51 if (unlikely(bootmem_debug)) \
52 printk(KERN_INFO \
53 "bootmem::%s " fmt, \
54 __func__, ## args); \
55})
56
57static unsigned long __init bootmap_bytes(unsigned long pages)
58{
59 unsigned long bytes = (pages + 7) / 8;
60
61 return ALIGN(bytes, sizeof(long));
62}
63
64/**
65 * bootmem_bootmap_pages - calculate bitmap size in pages
66 * @pages: number of pages the bitmap has to represent
67 */
68unsigned long __init bootmem_bootmap_pages(unsigned long pages)
69{
70 unsigned long bytes = bootmap_bytes(pages);
71
72 return PAGE_ALIGN(bytes) >> PAGE_SHIFT;
73}
74
75/*
76 * link bdata in order
77 */
78static void __init link_bootmem(bootmem_data_t *bdata)
79{
80 struct list_head *iter;
81
82 list_for_each(iter, &bdata_list) {
83 bootmem_data_t *ent;
84
85 ent = list_entry(iter, bootmem_data_t, list);
86 if (bdata->node_min_pfn < ent->node_min_pfn)
87 break;
88 }
89 list_add_tail(&bdata->list, iter);
90}
91
92/*
93 * Called once to set up the allocator itself.
94 */
95static unsigned long __init init_bootmem_core(bootmem_data_t *bdata,
96 unsigned long mapstart, unsigned long start, unsigned long end)
97{
98 unsigned long mapsize;
99
100 mminit_validate_memmodel_limits(&start, &end);
101 bdata->node_bootmem_map = phys_to_virt(PFN_PHYS(mapstart));
102 bdata->node_min_pfn = start;
103 bdata->node_low_pfn = end;
104 link_bootmem(bdata);
105
106 /*
107 * Initially all pages are reserved - setup_arch() has to
108 * register free RAM areas explicitly.
109 */
110 mapsize = bootmap_bytes(end - start);
111 memset(bdata->node_bootmem_map, 0xff, mapsize);
112
113 bdebug("nid=%td start=%lx map=%lx end=%lx mapsize=%lx\n",
114 bdata - bootmem_node_data, start, mapstart, end, mapsize);
115
116 return mapsize;
117}
118
119/**
120 * init_bootmem_node - register a node as boot memory
121 * @pgdat: node to register
122 * @freepfn: pfn where the bitmap for this node is to be placed
123 * @startpfn: first pfn on the node
124 * @endpfn: first pfn after the node
125 *
126 * Returns the number of bytes needed to hold the bitmap for this node.
127 */
128unsigned long __init init_bootmem_node(pg_data_t *pgdat, unsigned long freepfn,
129 unsigned long startpfn, unsigned long endpfn)
130{
131 return init_bootmem_core(pgdat->bdata, freepfn, startpfn, endpfn);
132}
133
134/**
135 * init_bootmem - register boot memory
136 * @start: pfn where the bitmap is to be placed
137 * @pages: number of available physical pages
138 *
139 * Returns the number of bytes needed to hold the bitmap.
140 */
141unsigned long __init init_bootmem(unsigned long start, unsigned long pages)
142{
143 max_low_pfn = pages;
144 min_low_pfn = start;
145 return init_bootmem_core(NODE_DATA(0)->bdata, start, 0, pages);
146}
147
148/*
149 * free_bootmem_late - free bootmem pages directly to page allocator
150 * @addr: starting address of the range
151 * @size: size of the range in bytes
152 *
153 * This is only useful when the bootmem allocator has already been torn
154 * down, but we are still initializing the system. Pages are given directly
155 * to the page allocator, no bootmem metadata is updated because it is gone.
156 */
157void __init free_bootmem_late(unsigned long addr, unsigned long size)
158{
159 unsigned long cursor, end;
160
161 kmemleak_free_part(__va(addr), size);
162
163 cursor = PFN_UP(addr);
164 end = PFN_DOWN(addr + size);
165
166 for (; cursor < end; cursor++) {
167 __free_pages_bootmem(pfn_to_page(cursor), 0);
168 totalram_pages++;
169 }
170}
171
172static unsigned long __init free_all_bootmem_core(bootmem_data_t *bdata)
173{
174 int aligned;
175 struct page *page;
176 unsigned long start, end, pages, count = 0;
177
178 if (!bdata->node_bootmem_map)
179 return 0;
180
181 start = bdata->node_min_pfn;
182 end = bdata->node_low_pfn;
183
184 /*
185 * If the start is aligned to the machines wordsize, we might
186 * be able to free pages in bulks of that order.
187 */
188 aligned = !(start & (BITS_PER_LONG - 1));
189
190 bdebug("nid=%td start=%lx end=%lx aligned=%d\n",
191 bdata - bootmem_node_data, start, end, aligned);
192
193 while (start < end) {
194 unsigned long *map, idx, vec;
195
196 map = bdata->node_bootmem_map;
197 idx = start - bdata->node_min_pfn;
198 vec = ~map[idx / BITS_PER_LONG];
199
200 if (aligned && vec == ~0UL && start + BITS_PER_LONG < end) {
201 int order = ilog2(BITS_PER_LONG);
202
203 __free_pages_bootmem(pfn_to_page(start), order);
204 count += BITS_PER_LONG;
205 } else {
206 unsigned long off = 0;
207
208 while (vec && off < BITS_PER_LONG) {
209 if (vec & 1) {
210 page = pfn_to_page(start + off);
211 __free_pages_bootmem(page, 0);
212 count++;
213 }
214 vec >>= 1;
215 off++;
216 }
217 }
218 start += BITS_PER_LONG;
219 }
220
221 page = virt_to_page(bdata->node_bootmem_map);
222 pages = bdata->node_low_pfn - bdata->node_min_pfn;
223 pages = bootmem_bootmap_pages(pages);
224 count += pages;
225 while (pages--)
226 __free_pages_bootmem(page++, 0);
227
228 bdebug("nid=%td released=%lx\n", bdata - bootmem_node_data, count);
229
230 return count;
231}
232
233/**
234 * free_all_bootmem_node - release a node's free pages to the buddy allocator
235 * @pgdat: node to be released
236 *
237 * Returns the number of pages actually released.
238 */
239unsigned long __init free_all_bootmem_node(pg_data_t *pgdat)
240{
241 register_page_bootmem_info_node(pgdat);
242 return free_all_bootmem_core(pgdat->bdata);
243}
244
245/**
246 * free_all_bootmem - release free pages to the buddy allocator
247 *
248 * Returns the number of pages actually released.
249 */
250unsigned long __init free_all_bootmem(void)
251{
252 unsigned long total_pages = 0;
253 bootmem_data_t *bdata;
254
255 list_for_each_entry(bdata, &bdata_list, list)
256 total_pages += free_all_bootmem_core(bdata);
257
258 return total_pages;
259}
260
261static void __init __free(bootmem_data_t *bdata,
262 unsigned long sidx, unsigned long eidx)
263{
264 unsigned long idx;
265
266 bdebug("nid=%td start=%lx end=%lx\n", bdata - bootmem_node_data,
267 sidx + bdata->node_min_pfn,
268 eidx + bdata->node_min_pfn);
269
270 if (bdata->hint_idx > sidx)
271 bdata->hint_idx = sidx;
272
273 for (idx = sidx; idx < eidx; idx++)
274 if (!test_and_clear_bit(idx, bdata->node_bootmem_map))
275 BUG();
276}
277
278static int __init __reserve(bootmem_data_t *bdata, unsigned long sidx,
279 unsigned long eidx, int flags)
280{
281 unsigned long idx;
282 int exclusive = flags & BOOTMEM_EXCLUSIVE;
283
284 bdebug("nid=%td start=%lx end=%lx flags=%x\n",
285 bdata - bootmem_node_data,
286 sidx + bdata->node_min_pfn,
287 eidx + bdata->node_min_pfn,
288 flags);
289
290 for (idx = sidx; idx < eidx; idx++)
291 if (test_and_set_bit(idx, bdata->node_bootmem_map)) {
292 if (exclusive) {
293 __free(bdata, sidx, idx);
294 return -EBUSY;
295 }
296 bdebug("silent double reserve of PFN %lx\n",
297 idx + bdata->node_min_pfn);
298 }
299 return 0;
300}
301
302static int __init mark_bootmem_node(bootmem_data_t *bdata,
303 unsigned long start, unsigned long end,
304 int reserve, int flags)
305{
306 unsigned long sidx, eidx;
307
308 bdebug("nid=%td start=%lx end=%lx reserve=%d flags=%x\n",
309 bdata - bootmem_node_data, start, end, reserve, flags);
310
311 BUG_ON(start < bdata->node_min_pfn);
312 BUG_ON(end > bdata->node_low_pfn);
313
314 sidx = start - bdata->node_min_pfn;
315 eidx = end - bdata->node_min_pfn;
316
317 if (reserve)
318 return __reserve(bdata, sidx, eidx, flags);
319 else
320 __free(bdata, sidx, eidx);
321 return 0;
322}
323
324static int __init mark_bootmem(unsigned long start, unsigned long end,
325 int reserve, int flags)
326{
327 unsigned long pos;
328 bootmem_data_t *bdata;
329
330 pos = start;
331 list_for_each_entry(bdata, &bdata_list, list) {
332 int err;
333 unsigned long max;
334
335 if (pos < bdata->node_min_pfn ||
336 pos >= bdata->node_low_pfn) {
337 BUG_ON(pos != start);
338 continue;
339 }
340
341 max = min(bdata->node_low_pfn, end);
342
343 err = mark_bootmem_node(bdata, pos, max, reserve, flags);
344 if (reserve && err) {
345 mark_bootmem(start, pos, 0, 0);
346 return err;
347 }
348
349 if (max == end)
350 return 0;
351 pos = bdata->node_low_pfn;
352 }
353 BUG();
354}
355
356/**
357 * free_bootmem_node - mark a page range as usable
358 * @pgdat: node the range resides on
359 * @physaddr: starting address of the range
360 * @size: size of the range in bytes
361 *
362 * Partial pages will be considered reserved and left as they are.
363 *
364 * The range must reside completely on the specified node.
365 */
366void __init free_bootmem_node(pg_data_t *pgdat, unsigned long physaddr,
367 unsigned long size)
368{
369 unsigned long start, end;
370
371 kmemleak_free_part(__va(physaddr), size);
372
373 start = PFN_UP(physaddr);
374 end = PFN_DOWN(physaddr + size);
375
376 mark_bootmem_node(pgdat->bdata, start, end, 0, 0);
377}
378
379/**
380 * free_bootmem - mark a page range as usable
381 * @addr: starting address of the range
382 * @size: size of the range in bytes
383 *
384 * Partial pages will be considered reserved and left as they are.
385 *
386 * The range must be contiguous but may span node boundaries.
387 */
388void __init free_bootmem(unsigned long addr, unsigned long size)
389{
390 unsigned long start, end;
391
392 kmemleak_free_part(__va(addr), size);
393
394 start = PFN_UP(addr);
395 end = PFN_DOWN(addr + size);
396
397 mark_bootmem(start, end, 0, 0);
398}
399
400/**
401 * reserve_bootmem_node - mark a page range as reserved
402 * @pgdat: node the range resides on
403 * @physaddr: starting address of the range
404 * @size: size of the range in bytes
405 * @flags: reservation flags (see linux/bootmem.h)
406 *
407 * Partial pages will be reserved.
408 *
409 * The range must reside completely on the specified node.
410 */
411int __init reserve_bootmem_node(pg_data_t *pgdat, unsigned long physaddr,
412 unsigned long size, int flags)
413{
414 unsigned long start, end;
415
416 start = PFN_DOWN(physaddr);
417 end = PFN_UP(physaddr + size);
418
419 return mark_bootmem_node(pgdat->bdata, start, end, 1, flags);
420}
421
422/**
423 * reserve_bootmem - mark a page range as usable
424 * @addr: starting address of the range
425 * @size: size of the range in bytes
426 * @flags: reservation flags (see linux/bootmem.h)
427 *
428 * Partial pages will be reserved.
429 *
430 * The range must be contiguous but may span node boundaries.
431 */
432int __init reserve_bootmem(unsigned long addr, unsigned long size,
433 int flags)
434{
435 unsigned long start, end;
436
437 start = PFN_DOWN(addr);
438 end = PFN_UP(addr + size);
439
440 return mark_bootmem(start, end, 1, flags);
441}
442
443int __weak __init reserve_bootmem_generic(unsigned long phys, unsigned long len,
444 int flags)
445{
446 return reserve_bootmem(phys, len, flags);
447}
448
449static unsigned long __init align_idx(struct bootmem_data *bdata,
450 unsigned long idx, unsigned long step)
451{
452 unsigned long base = bdata->node_min_pfn;
453
454 /*
455 * Align the index with respect to the node start so that the
456 * combination of both satisfies the requested alignment.
457 */
458
459 return ALIGN(base + idx, step) - base;
460}
461
462static unsigned long __init align_off(struct bootmem_data *bdata,
463 unsigned long off, unsigned long align)
464{
465 unsigned long base = PFN_PHYS(bdata->node_min_pfn);
466
467 /* Same as align_idx for byte offsets */
468
469 return ALIGN(base + off, align) - base;
470}
471
472static void * __init alloc_bootmem_core(struct bootmem_data *bdata,
473 unsigned long size, unsigned long align,
474 unsigned long goal, unsigned long limit)
475{
476 unsigned long fallback = 0;
477 unsigned long min, max, start, sidx, midx, step;
478
479 bdebug("nid=%td size=%lx [%lu pages] align=%lx goal=%lx limit=%lx\n",
480 bdata - bootmem_node_data, size, PAGE_ALIGN(size) >> PAGE_SHIFT,
481 align, goal, limit);
482
483 BUG_ON(!size);
484 BUG_ON(align & (align - 1));
485 BUG_ON(limit && goal + size > limit);
486
487 if (!bdata->node_bootmem_map)
488 return NULL;
489
490 min = bdata->node_min_pfn;
491 max = bdata->node_low_pfn;
492
493 goal >>= PAGE_SHIFT;
494 limit >>= PAGE_SHIFT;
495
496 if (limit && max > limit)
497 max = limit;
498 if (max <= min)
499 return NULL;
500
501 step = max(align >> PAGE_SHIFT, 1UL);
502
503 if (goal && min < goal && goal < max)
504 start = ALIGN(goal, step);
505 else
506 start = ALIGN(min, step);
507
508 sidx = start - bdata->node_min_pfn;
509 midx = max - bdata->node_min_pfn;
510
511 if (bdata->hint_idx > sidx) {
512 /*
513 * Handle the valid case of sidx being zero and still
514 * catch the fallback below.
515 */
516 fallback = sidx + 1;
517 sidx = align_idx(bdata, bdata->hint_idx, step);
518 }
519
520 while (1) {
521 int merge;
522 void *region;
523 unsigned long eidx, i, start_off, end_off;
524find_block:
525 sidx = find_next_zero_bit(bdata->node_bootmem_map, midx, sidx);
526 sidx = align_idx(bdata, sidx, step);
527 eidx = sidx + PFN_UP(size);
528
529 if (sidx >= midx || eidx > midx)
530 break;
531
532 for (i = sidx; i < eidx; i++)
533 if (test_bit(i, bdata->node_bootmem_map)) {
534 sidx = align_idx(bdata, i, step);
535 if (sidx == i)
536 sidx += step;
537 goto find_block;
538 }
539
540 if (bdata->last_end_off & (PAGE_SIZE - 1) &&
541 PFN_DOWN(bdata->last_end_off) + 1 == sidx)
542 start_off = align_off(bdata, bdata->last_end_off, align);
543 else
544 start_off = PFN_PHYS(sidx);
545
546 merge = PFN_DOWN(start_off) < sidx;
547 end_off = start_off + size;
548
549 bdata->last_end_off = end_off;
550 bdata->hint_idx = PFN_UP(end_off);
551
552 /*
553 * Reserve the area now:
554 */
555 if (__reserve(bdata, PFN_DOWN(start_off) + merge,
556 PFN_UP(end_off), BOOTMEM_EXCLUSIVE))
557 BUG();
558
559 region = phys_to_virt(PFN_PHYS(bdata->node_min_pfn) +
560 start_off);
561 memset(region, 0, size);
562 /*
563 * The min_count is set to 0 so that bootmem allocated blocks
564 * are never reported as leaks.
565 */
566 kmemleak_alloc(region, size, 0, 0);
567 return region;
568 }
569
570 if (fallback) {
571 sidx = align_idx(bdata, fallback - 1, step);
572 fallback = 0;
573 goto find_block;
574 }
575
576 return NULL;
577}
578
579static void * __init alloc_arch_preferred_bootmem(bootmem_data_t *bdata,
580 unsigned long size, unsigned long align,
581 unsigned long goal, unsigned long limit)
582{
583 if (WARN_ON_ONCE(slab_is_available()))
584 return kzalloc(size, GFP_NOWAIT);
585
586#ifdef CONFIG_HAVE_ARCH_BOOTMEM
587 {
588 bootmem_data_t *p_bdata;
589
590 p_bdata = bootmem_arch_preferred_node(bdata, size, align,
591 goal, limit);
592 if (p_bdata)
593 return alloc_bootmem_core(p_bdata, size, align,
594 goal, limit);
595 }
596#endif
597 return NULL;
598}
599
600static void * __init ___alloc_bootmem_nopanic(unsigned long size,
601 unsigned long align,
602 unsigned long goal,
603 unsigned long limit)
604{
605 bootmem_data_t *bdata;
606 void *region;
607
608restart:
609 region = alloc_arch_preferred_bootmem(NULL, size, align, goal, limit);
610 if (region)
611 return region;
612
613 list_for_each_entry(bdata, &bdata_list, list) {
614 if (goal && bdata->node_low_pfn <= PFN_DOWN(goal))
615 continue;
616 if (limit && bdata->node_min_pfn >= PFN_DOWN(limit))
617 break;
618
619 region = alloc_bootmem_core(bdata, size, align, goal, limit);
620 if (region)
621 return region;
622 }
623
624 if (goal) {
625 goal = 0;
626 goto restart;
627 }
628
629 return NULL;
630}
631
632/**
633 * __alloc_bootmem_nopanic - allocate boot memory without panicking
634 * @size: size of the request in bytes
635 * @align: alignment of the region
636 * @goal: preferred starting address of the region
637 *
638 * The goal is dropped if it can not be satisfied and the allocation will
639 * fall back to memory below @goal.
640 *
641 * Allocation may happen on any node in the system.
642 *
643 * Returns NULL on failure.
644 */
645void * __init __alloc_bootmem_nopanic(unsigned long size, unsigned long align,
646 unsigned long goal)
647{
648 unsigned long limit = 0;
649
650 return ___alloc_bootmem_nopanic(size, align, goal, limit);
651}
652
653static void * __init ___alloc_bootmem(unsigned long size, unsigned long align,
654 unsigned long goal, unsigned long limit)
655{
656 void *mem = ___alloc_bootmem_nopanic(size, align, goal, limit);
657
658 if (mem)
659 return mem;
660 /*
661 * Whoops, we cannot satisfy the allocation request.
662 */
663 printk(KERN_ALERT "bootmem alloc of %lu bytes failed!\n", size);
664 panic("Out of memory");
665 return NULL;
666}
667
668/**
669 * __alloc_bootmem - allocate boot memory
670 * @size: size of the request in bytes
671 * @align: alignment of the region
672 * @goal: preferred starting address of the region
673 *
674 * The goal is dropped if it can not be satisfied and the allocation will
675 * fall back to memory below @goal.
676 *
677 * Allocation may happen on any node in the system.
678 *
679 * The function panics if the request can not be satisfied.
680 */
681void * __init __alloc_bootmem(unsigned long size, unsigned long align,
682 unsigned long goal)
683{
684 unsigned long limit = 0;
685
686 return ___alloc_bootmem(size, align, goal, limit);
687}
688
689static void * __init ___alloc_bootmem_node(bootmem_data_t *bdata,
690 unsigned long size, unsigned long align,
691 unsigned long goal, unsigned long limit)
692{
693 void *ptr;
694
695 ptr = alloc_arch_preferred_bootmem(bdata, size, align, goal, limit);
696 if (ptr)
697 return ptr;
698
699 ptr = alloc_bootmem_core(bdata, size, align, goal, limit);
700 if (ptr)
701 return ptr;
702
703 return ___alloc_bootmem(size, align, goal, limit);
704}
705
706/**
707 * __alloc_bootmem_node - allocate boot memory from a specific node
708 * @pgdat: node to allocate from
709 * @size: size of the request in bytes
710 * @align: alignment of the region
711 * @goal: preferred starting address of the region
712 *
713 * The goal is dropped if it can not be satisfied and the allocation will
714 * fall back to memory below @goal.
715 *
716 * Allocation may fall back to any node in the system if the specified node
717 * can not hold the requested memory.
718 *
719 * The function panics if the request can not be satisfied.
720 */
721void * __init __alloc_bootmem_node(pg_data_t *pgdat, unsigned long size,
722 unsigned long align, unsigned long goal)
723{
724 if (WARN_ON_ONCE(slab_is_available()))
725 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
726
727 return ___alloc_bootmem_node(pgdat->bdata, size, align, goal, 0);
728}
729
730void * __init __alloc_bootmem_node_high(pg_data_t *pgdat, unsigned long size,
731 unsigned long align, unsigned long goal)
732{
733#ifdef MAX_DMA32_PFN
734 unsigned long end_pfn;
735
736 if (WARN_ON_ONCE(slab_is_available()))
737 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
738
739 /* update goal according ...MAX_DMA32_PFN */
740 end_pfn = pgdat->node_start_pfn + pgdat->node_spanned_pages;
741
742 if (end_pfn > MAX_DMA32_PFN + (128 >> (20 - PAGE_SHIFT)) &&
743 (goal >> PAGE_SHIFT) < MAX_DMA32_PFN) {
744 void *ptr;
745 unsigned long new_goal;
746
747 new_goal = MAX_DMA32_PFN << PAGE_SHIFT;
748 ptr = alloc_bootmem_core(pgdat->bdata, size, align,
749 new_goal, 0);
750 if (ptr)
751 return ptr;
752 }
753#endif
754
755 return __alloc_bootmem_node(pgdat, size, align, goal);
756
757}
758
759#ifdef CONFIG_SPARSEMEM
760/**
761 * alloc_bootmem_section - allocate boot memory from a specific section
762 * @size: size of the request in bytes
763 * @section_nr: sparse map section to allocate from
764 *
765 * Return NULL on failure.
766 */
767void * __init alloc_bootmem_section(unsigned long size,
768 unsigned long section_nr)
769{
770 bootmem_data_t *bdata;
771 unsigned long pfn, goal, limit;
772
773 pfn = section_nr_to_pfn(section_nr);
774 goal = pfn << PAGE_SHIFT;
775 limit = section_nr_to_pfn(section_nr + 1) << PAGE_SHIFT;
776 bdata = &bootmem_node_data[early_pfn_to_nid(pfn)];
777
778 return alloc_bootmem_core(bdata, size, SMP_CACHE_BYTES, goal, limit);
779}
780#endif
781
782void * __init __alloc_bootmem_node_nopanic(pg_data_t *pgdat, unsigned long size,
783 unsigned long align, unsigned long goal)
784{
785 void *ptr;
786
787 if (WARN_ON_ONCE(slab_is_available()))
788 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
789
790 ptr = alloc_arch_preferred_bootmem(pgdat->bdata, size, align, goal, 0);
791 if (ptr)
792 return ptr;
793
794 ptr = alloc_bootmem_core(pgdat->bdata, size, align, goal, 0);
795 if (ptr)
796 return ptr;
797
798 return __alloc_bootmem_nopanic(size, align, goal);
799}
800
801#ifndef ARCH_LOW_ADDRESS_LIMIT
802#define ARCH_LOW_ADDRESS_LIMIT 0xffffffffUL
803#endif
804
805/**
806 * __alloc_bootmem_low - allocate low boot memory
807 * @size: size of the request in bytes
808 * @align: alignment of the region
809 * @goal: preferred starting address of the region
810 *
811 * The goal is dropped if it can not be satisfied and the allocation will
812 * fall back to memory below @goal.
813 *
814 * Allocation may happen on any node in the system.
815 *
816 * The function panics if the request can not be satisfied.
817 */
818void * __init __alloc_bootmem_low(unsigned long size, unsigned long align,
819 unsigned long goal)
820{
821 return ___alloc_bootmem(size, align, goal, ARCH_LOW_ADDRESS_LIMIT);
822}
823
824/**
825 * __alloc_bootmem_low_node - allocate low boot memory from a specific node
826 * @pgdat: node to allocate from
827 * @size: size of the request in bytes
828 * @align: alignment of the region
829 * @goal: preferred starting address of the region
830 *
831 * The goal is dropped if it can not be satisfied and the allocation will
832 * fall back to memory below @goal.
833 *
834 * Allocation may fall back to any node in the system if the specified node
835 * can not hold the requested memory.
836 *
837 * The function panics if the request can not be satisfied.
838 */
839void * __init __alloc_bootmem_low_node(pg_data_t *pgdat, unsigned long size,
840 unsigned long align, unsigned long goal)
841{
842 if (WARN_ON_ONCE(slab_is_available()))
843 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
844
845 return ___alloc_bootmem_node(pgdat->bdata, size, align,
846 goal, ARCH_LOW_ADDRESS_LIMIT);
847}