Loading...
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * bootmem - A boot-time physical memory allocator and configurator
4 *
5 * Copyright (C) 1999 Ingo Molnar
6 * 1999 Kanoj Sarcar, SGI
7 * 2008 Johannes Weiner
8 *
9 * Access to this subsystem has to be serialized externally (which is true
10 * for the boot process anyway).
11 */
12#include <linux/init.h>
13#include <linux/pfn.h>
14#include <linux/slab.h>
15#include <linux/export.h>
16#include <linux/kmemleak.h>
17#include <linux/range.h>
18#include <linux/bug.h>
19#include <linux/io.h>
20#include <linux/bootmem.h>
21
22#include "internal.h"
23
24#ifndef CONFIG_NEED_MULTIPLE_NODES
25struct pglist_data __refdata contig_page_data = {
26 .bdata = &bootmem_node_data[0]
27};
28EXPORT_SYMBOL(contig_page_data);
29#endif
30
31unsigned long max_low_pfn;
32unsigned long min_low_pfn;
33unsigned long max_pfn;
34unsigned long long max_possible_pfn;
35
36bootmem_data_t bootmem_node_data[MAX_NUMNODES] __initdata;
37
38static struct list_head bdata_list __initdata = LIST_HEAD_INIT(bdata_list);
39
40static int bootmem_debug;
41
42static int __init bootmem_debug_setup(char *buf)
43{
44 bootmem_debug = 1;
45 return 0;
46}
47early_param("bootmem_debug", bootmem_debug_setup);
48
49#define bdebug(fmt, args...) ({ \
50 if (unlikely(bootmem_debug)) \
51 pr_info("bootmem::%s " fmt, \
52 __func__, ## args); \
53})
54
55static unsigned long __init bootmap_bytes(unsigned long pages)
56{
57 unsigned long bytes = DIV_ROUND_UP(pages, BITS_PER_BYTE);
58
59 return ALIGN(bytes, sizeof(long));
60}
61
62/**
63 * bootmem_bootmap_pages - calculate bitmap size in pages
64 * @pages: number of pages the bitmap has to represent
65 */
66unsigned long __init bootmem_bootmap_pages(unsigned long pages)
67{
68 unsigned long bytes = bootmap_bytes(pages);
69
70 return PAGE_ALIGN(bytes) >> PAGE_SHIFT;
71}
72
73/*
74 * link bdata in order
75 */
76static void __init link_bootmem(bootmem_data_t *bdata)
77{
78 bootmem_data_t *ent;
79
80 list_for_each_entry(ent, &bdata_list, list) {
81 if (bdata->node_min_pfn < ent->node_min_pfn) {
82 list_add_tail(&bdata->list, &ent->list);
83 return;
84 }
85 }
86
87 list_add_tail(&bdata->list, &bdata_list);
88}
89
90/*
91 * Called once to set up the allocator itself.
92 */
93static unsigned long __init init_bootmem_core(bootmem_data_t *bdata,
94 unsigned long mapstart, unsigned long start, unsigned long end)
95{
96 unsigned long mapsize;
97
98 mminit_validate_memmodel_limits(&start, &end);
99 bdata->node_bootmem_map = phys_to_virt(PFN_PHYS(mapstart));
100 bdata->node_min_pfn = start;
101 bdata->node_low_pfn = end;
102 link_bootmem(bdata);
103
104 /*
105 * Initially all pages are reserved - setup_arch() has to
106 * register free RAM areas explicitly.
107 */
108 mapsize = bootmap_bytes(end - start);
109 memset(bdata->node_bootmem_map, 0xff, mapsize);
110
111 bdebug("nid=%td start=%lx map=%lx end=%lx mapsize=%lx\n",
112 bdata - bootmem_node_data, start, mapstart, end, mapsize);
113
114 return mapsize;
115}
116
117/**
118 * init_bootmem_node - register a node as boot memory
119 * @pgdat: node to register
120 * @freepfn: pfn where the bitmap for this node is to be placed
121 * @startpfn: first pfn on the node
122 * @endpfn: first pfn after the node
123 *
124 * Returns the number of bytes needed to hold the bitmap for this node.
125 */
126unsigned long __init init_bootmem_node(pg_data_t *pgdat, unsigned long freepfn,
127 unsigned long startpfn, unsigned long endpfn)
128{
129 return init_bootmem_core(pgdat->bdata, freepfn, startpfn, endpfn);
130}
131
132/**
133 * init_bootmem - register boot memory
134 * @start: pfn where the bitmap is to be placed
135 * @pages: number of available physical pages
136 *
137 * Returns the number of bytes needed to hold the bitmap.
138 */
139unsigned long __init init_bootmem(unsigned long start, unsigned long pages)
140{
141 max_low_pfn = pages;
142 min_low_pfn = start;
143 return init_bootmem_core(NODE_DATA(0)->bdata, start, 0, pages);
144}
145
146/*
147 * free_bootmem_late - free bootmem pages directly to page allocator
148 * @addr: starting physical address of the range
149 * @size: size of the range in bytes
150 *
151 * This is only useful when the bootmem allocator has already been torn
152 * down, but we are still initializing the system. Pages are given directly
153 * to the page allocator, no bootmem metadata is updated because it is gone.
154 */
155void __init free_bootmem_late(unsigned long physaddr, unsigned long size)
156{
157 unsigned long cursor, end;
158
159 kmemleak_free_part_phys(physaddr, size);
160
161 cursor = PFN_UP(physaddr);
162 end = PFN_DOWN(physaddr + size);
163
164 for (; cursor < end; cursor++) {
165 __free_pages_bootmem(pfn_to_page(cursor), cursor, 0);
166 totalram_pages++;
167 }
168}
169
170static unsigned long __init free_all_bootmem_core(bootmem_data_t *bdata)
171{
172 struct page *page;
173 unsigned long *map, start, end, pages, cur, count = 0;
174
175 if (!bdata->node_bootmem_map)
176 return 0;
177
178 map = bdata->node_bootmem_map;
179 start = bdata->node_min_pfn;
180 end = bdata->node_low_pfn;
181
182 bdebug("nid=%td start=%lx end=%lx\n",
183 bdata - bootmem_node_data, start, end);
184
185 while (start < end) {
186 unsigned long idx, vec;
187 unsigned shift;
188
189 idx = start - bdata->node_min_pfn;
190 shift = idx & (BITS_PER_LONG - 1);
191 /*
192 * vec holds at most BITS_PER_LONG map bits,
193 * bit 0 corresponds to start.
194 */
195 vec = ~map[idx / BITS_PER_LONG];
196
197 if (shift) {
198 vec >>= shift;
199 if (end - start >= BITS_PER_LONG)
200 vec |= ~map[idx / BITS_PER_LONG + 1] <<
201 (BITS_PER_LONG - shift);
202 }
203 /*
204 * If we have a properly aligned and fully unreserved
205 * BITS_PER_LONG block of pages in front of us, free
206 * it in one go.
207 */
208 if (IS_ALIGNED(start, BITS_PER_LONG) && vec == ~0UL) {
209 int order = ilog2(BITS_PER_LONG);
210
211 __free_pages_bootmem(pfn_to_page(start), start, order);
212 count += BITS_PER_LONG;
213 start += BITS_PER_LONG;
214 } else {
215 cur = start;
216
217 start = ALIGN(start + 1, BITS_PER_LONG);
218 while (vec && cur != start) {
219 if (vec & 1) {
220 page = pfn_to_page(cur);
221 __free_pages_bootmem(page, cur, 0);
222 count++;
223 }
224 vec >>= 1;
225 ++cur;
226 }
227 }
228 }
229
230 cur = bdata->node_min_pfn;
231 page = virt_to_page(bdata->node_bootmem_map);
232 pages = bdata->node_low_pfn - bdata->node_min_pfn;
233 pages = bootmem_bootmap_pages(pages);
234 count += pages;
235 while (pages--)
236 __free_pages_bootmem(page++, cur++, 0);
237 bdata->node_bootmem_map = NULL;
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
246void reset_node_managed_pages(pg_data_t *pgdat)
247{
248 struct zone *z;
249
250 for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++)
251 z->managed_pages = 0;
252}
253
254void __init reset_all_zones_managed_pages(void)
255{
256 struct pglist_data *pgdat;
257
258 if (reset_managed_pages_done)
259 return;
260
261 for_each_online_pgdat(pgdat)
262 reset_node_managed_pages(pgdat);
263
264 reset_managed_pages_done = 1;
265}
266
267/**
268 * free_all_bootmem - release free pages to the buddy allocator
269 *
270 * Returns the number of pages actually released.
271 */
272unsigned long __init free_all_bootmem(void)
273{
274 unsigned long total_pages = 0;
275 bootmem_data_t *bdata;
276
277 reset_all_zones_managed_pages();
278
279 list_for_each_entry(bdata, &bdata_list, list)
280 total_pages += free_all_bootmem_core(bdata);
281
282 totalram_pages += total_pages;
283
284 return total_pages;
285}
286
287static void __init __free(bootmem_data_t *bdata,
288 unsigned long sidx, unsigned long eidx)
289{
290 unsigned long idx;
291
292 bdebug("nid=%td start=%lx end=%lx\n", bdata - bootmem_node_data,
293 sidx + bdata->node_min_pfn,
294 eidx + bdata->node_min_pfn);
295
296 if (WARN_ON(bdata->node_bootmem_map == NULL))
297 return;
298
299 if (bdata->hint_idx > sidx)
300 bdata->hint_idx = sidx;
301
302 for (idx = sidx; idx < eidx; idx++)
303 if (!test_and_clear_bit(idx, bdata->node_bootmem_map))
304 BUG();
305}
306
307static int __init __reserve(bootmem_data_t *bdata, unsigned long sidx,
308 unsigned long eidx, int flags)
309{
310 unsigned long idx;
311 int exclusive = flags & BOOTMEM_EXCLUSIVE;
312
313 bdebug("nid=%td start=%lx end=%lx flags=%x\n",
314 bdata - bootmem_node_data,
315 sidx + bdata->node_min_pfn,
316 eidx + bdata->node_min_pfn,
317 flags);
318
319 if (WARN_ON(bdata->node_bootmem_map == NULL))
320 return 0;
321
322 for (idx = sidx; idx < eidx; idx++)
323 if (test_and_set_bit(idx, bdata->node_bootmem_map)) {
324 if (exclusive) {
325 __free(bdata, sidx, idx);
326 return -EBUSY;
327 }
328 bdebug("silent double reserve of PFN %lx\n",
329 idx + bdata->node_min_pfn);
330 }
331 return 0;
332}
333
334static int __init mark_bootmem_node(bootmem_data_t *bdata,
335 unsigned long start, unsigned long end,
336 int reserve, int flags)
337{
338 unsigned long sidx, eidx;
339
340 bdebug("nid=%td start=%lx end=%lx reserve=%d flags=%x\n",
341 bdata - bootmem_node_data, start, end, reserve, flags);
342
343 BUG_ON(start < bdata->node_min_pfn);
344 BUG_ON(end > bdata->node_low_pfn);
345
346 sidx = start - bdata->node_min_pfn;
347 eidx = end - bdata->node_min_pfn;
348
349 if (reserve)
350 return __reserve(bdata, sidx, eidx, flags);
351 else
352 __free(bdata, sidx, eidx);
353 return 0;
354}
355
356static int __init mark_bootmem(unsigned long start, unsigned long end,
357 int reserve, int flags)
358{
359 unsigned long pos;
360 bootmem_data_t *bdata;
361
362 pos = start;
363 list_for_each_entry(bdata, &bdata_list, list) {
364 int err;
365 unsigned long max;
366
367 if (pos < bdata->node_min_pfn ||
368 pos >= bdata->node_low_pfn) {
369 BUG_ON(pos != start);
370 continue;
371 }
372
373 max = min(bdata->node_low_pfn, end);
374
375 err = mark_bootmem_node(bdata, pos, max, reserve, flags);
376 if (reserve && err) {
377 mark_bootmem(start, pos, 0, 0);
378 return err;
379 }
380
381 if (max == end)
382 return 0;
383 pos = bdata->node_low_pfn;
384 }
385 BUG();
386}
387
388/**
389 * free_bootmem_node - mark a page range as usable
390 * @pgdat: node the range resides on
391 * @physaddr: starting address of the range
392 * @size: size of the range in bytes
393 *
394 * Partial pages will be considered reserved and left as they are.
395 *
396 * The range must reside completely on the specified node.
397 */
398void __init free_bootmem_node(pg_data_t *pgdat, unsigned long physaddr,
399 unsigned long size)
400{
401 unsigned long start, end;
402
403 kmemleak_free_part_phys(physaddr, size);
404
405 start = PFN_UP(physaddr);
406 end = PFN_DOWN(physaddr + size);
407
408 mark_bootmem_node(pgdat->bdata, start, end, 0, 0);
409}
410
411/**
412 * free_bootmem - mark a page range as usable
413 * @physaddr: starting physical address of the range
414 * @size: size of the range in bytes
415 *
416 * Partial pages will be considered reserved and left as they are.
417 *
418 * The range must be contiguous but may span node boundaries.
419 */
420void __init free_bootmem(unsigned long physaddr, unsigned long size)
421{
422 unsigned long start, end;
423
424 kmemleak_free_part_phys(physaddr, size);
425
426 start = PFN_UP(physaddr);
427 end = PFN_DOWN(physaddr + size);
428
429 mark_bootmem(start, end, 0, 0);
430}
431
432/**
433 * reserve_bootmem_node - mark a page range as reserved
434 * @pgdat: node the range resides on
435 * @physaddr: starting address of the range
436 * @size: size of the range in bytes
437 * @flags: reservation flags (see linux/bootmem.h)
438 *
439 * Partial pages will be reserved.
440 *
441 * The range must reside completely on the specified node.
442 */
443int __init reserve_bootmem_node(pg_data_t *pgdat, unsigned long physaddr,
444 unsigned long size, int flags)
445{
446 unsigned long start, end;
447
448 start = PFN_DOWN(physaddr);
449 end = PFN_UP(physaddr + size);
450
451 return mark_bootmem_node(pgdat->bdata, start, end, 1, flags);
452}
453
454/**
455 * reserve_bootmem - mark a page range as reserved
456 * @addr: starting address of the range
457 * @size: size of the range in bytes
458 * @flags: reservation flags (see linux/bootmem.h)
459 *
460 * Partial pages will be reserved.
461 *
462 * The range must be contiguous but may span node boundaries.
463 */
464int __init reserve_bootmem(unsigned long addr, unsigned long size,
465 int flags)
466{
467 unsigned long start, end;
468
469 start = PFN_DOWN(addr);
470 end = PFN_UP(addr + size);
471
472 return mark_bootmem(start, end, 1, flags);
473}
474
475static unsigned long __init align_idx(struct bootmem_data *bdata,
476 unsigned long idx, unsigned long step)
477{
478 unsigned long base = bdata->node_min_pfn;
479
480 /*
481 * Align the index with respect to the node start so that the
482 * combination of both satisfies the requested alignment.
483 */
484
485 return ALIGN(base + idx, step) - base;
486}
487
488static unsigned long __init align_off(struct bootmem_data *bdata,
489 unsigned long off, unsigned long align)
490{
491 unsigned long base = PFN_PHYS(bdata->node_min_pfn);
492
493 /* Same as align_idx for byte offsets */
494
495 return ALIGN(base + off, align) - base;
496}
497
498static void * __init alloc_bootmem_bdata(struct bootmem_data *bdata,
499 unsigned long size, unsigned long align,
500 unsigned long goal, unsigned long limit)
501{
502 unsigned long fallback = 0;
503 unsigned long min, max, start, sidx, midx, step;
504
505 bdebug("nid=%td size=%lx [%lu pages] align=%lx goal=%lx limit=%lx\n",
506 bdata - bootmem_node_data, size, PAGE_ALIGN(size) >> PAGE_SHIFT,
507 align, goal, limit);
508
509 BUG_ON(!size);
510 BUG_ON(align & (align - 1));
511 BUG_ON(limit && goal + size > limit);
512
513 if (!bdata->node_bootmem_map)
514 return NULL;
515
516 min = bdata->node_min_pfn;
517 max = bdata->node_low_pfn;
518
519 goal >>= PAGE_SHIFT;
520 limit >>= PAGE_SHIFT;
521
522 if (limit && max > limit)
523 max = limit;
524 if (max <= min)
525 return NULL;
526
527 step = max(align >> PAGE_SHIFT, 1UL);
528
529 if (goal && min < goal && goal < max)
530 start = ALIGN(goal, step);
531 else
532 start = ALIGN(min, step);
533
534 sidx = start - bdata->node_min_pfn;
535 midx = max - bdata->node_min_pfn;
536
537 if (bdata->hint_idx > sidx) {
538 /*
539 * Handle the valid case of sidx being zero and still
540 * catch the fallback below.
541 */
542 fallback = sidx + 1;
543 sidx = align_idx(bdata, bdata->hint_idx, step);
544 }
545
546 while (1) {
547 int merge;
548 void *region;
549 unsigned long eidx, i, start_off, end_off;
550find_block:
551 sidx = find_next_zero_bit(bdata->node_bootmem_map, midx, sidx);
552 sidx = align_idx(bdata, sidx, step);
553 eidx = sidx + PFN_UP(size);
554
555 if (sidx >= midx || eidx > midx)
556 break;
557
558 for (i = sidx; i < eidx; i++)
559 if (test_bit(i, bdata->node_bootmem_map)) {
560 sidx = align_idx(bdata, i, step);
561 if (sidx == i)
562 sidx += step;
563 goto find_block;
564 }
565
566 if (bdata->last_end_off & (PAGE_SIZE - 1) &&
567 PFN_DOWN(bdata->last_end_off) + 1 == sidx)
568 start_off = align_off(bdata, bdata->last_end_off, align);
569 else
570 start_off = PFN_PHYS(sidx);
571
572 merge = PFN_DOWN(start_off) < sidx;
573 end_off = start_off + size;
574
575 bdata->last_end_off = end_off;
576 bdata->hint_idx = PFN_UP(end_off);
577
578 /*
579 * Reserve the area now:
580 */
581 if (__reserve(bdata, PFN_DOWN(start_off) + merge,
582 PFN_UP(end_off), BOOTMEM_EXCLUSIVE))
583 BUG();
584
585 region = phys_to_virt(PFN_PHYS(bdata->node_min_pfn) +
586 start_off);
587 memset(region, 0, size);
588 /*
589 * The min_count is set to 0 so that bootmem allocated blocks
590 * are never reported as leaks.
591 */
592 kmemleak_alloc(region, size, 0, 0);
593 return region;
594 }
595
596 if (fallback) {
597 sidx = align_idx(bdata, fallback - 1, step);
598 fallback = 0;
599 goto find_block;
600 }
601
602 return NULL;
603}
604
605static void * __init alloc_bootmem_core(unsigned long size,
606 unsigned long align,
607 unsigned long goal,
608 unsigned long limit)
609{
610 bootmem_data_t *bdata;
611 void *region;
612
613 if (WARN_ON_ONCE(slab_is_available()))
614 return kzalloc(size, GFP_NOWAIT);
615
616 list_for_each_entry(bdata, &bdata_list, list) {
617 if (goal && bdata->node_low_pfn <= PFN_DOWN(goal))
618 continue;
619 if (limit && bdata->node_min_pfn >= PFN_DOWN(limit))
620 break;
621
622 region = alloc_bootmem_bdata(bdata, size, align, goal, limit);
623 if (region)
624 return region;
625 }
626
627 return NULL;
628}
629
630static void * __init ___alloc_bootmem_nopanic(unsigned long size,
631 unsigned long align,
632 unsigned long goal,
633 unsigned long limit)
634{
635 void *ptr;
636
637restart:
638 ptr = alloc_bootmem_core(size, align, goal, limit);
639 if (ptr)
640 return ptr;
641 if (goal) {
642 goal = 0;
643 goto restart;
644 }
645
646 return NULL;
647}
648
649/**
650 * __alloc_bootmem_nopanic - allocate boot memory without panicking
651 * @size: size of the request in bytes
652 * @align: alignment of the region
653 * @goal: preferred starting address of the region
654 *
655 * The goal is dropped if it can not be satisfied and the allocation will
656 * fall back to memory below @goal.
657 *
658 * Allocation may happen on any node in the system.
659 *
660 * Returns NULL on failure.
661 */
662void * __init __alloc_bootmem_nopanic(unsigned long size, unsigned long align,
663 unsigned long goal)
664{
665 unsigned long limit = 0;
666
667 return ___alloc_bootmem_nopanic(size, align, goal, limit);
668}
669
670static void * __init ___alloc_bootmem(unsigned long size, unsigned long align,
671 unsigned long goal, unsigned long limit)
672{
673 void *mem = ___alloc_bootmem_nopanic(size, align, goal, limit);
674
675 if (mem)
676 return mem;
677 /*
678 * Whoops, we cannot satisfy the allocation request.
679 */
680 pr_alert("bootmem alloc of %lu bytes failed!\n", size);
681 panic("Out of memory");
682 return NULL;
683}
684
685/**
686 * __alloc_bootmem - allocate boot memory
687 * @size: size of the request in bytes
688 * @align: alignment of the region
689 * @goal: preferred starting address of the region
690 *
691 * The goal is dropped if it can not be satisfied and the allocation will
692 * fall back to memory below @goal.
693 *
694 * Allocation may happen on any node in the system.
695 *
696 * The function panics if the request can not be satisfied.
697 */
698void * __init __alloc_bootmem(unsigned long size, unsigned long align,
699 unsigned long goal)
700{
701 unsigned long limit = 0;
702
703 return ___alloc_bootmem(size, align, goal, limit);
704}
705
706void * __init ___alloc_bootmem_node_nopanic(pg_data_t *pgdat,
707 unsigned long size, unsigned long align,
708 unsigned long goal, unsigned long limit)
709{
710 void *ptr;
711
712 if (WARN_ON_ONCE(slab_is_available()))
713 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
714again:
715
716 /* do not panic in alloc_bootmem_bdata() */
717 if (limit && goal + size > limit)
718 limit = 0;
719
720 ptr = alloc_bootmem_bdata(pgdat->bdata, size, align, goal, limit);
721 if (ptr)
722 return ptr;
723
724 ptr = alloc_bootmem_core(size, align, goal, limit);
725 if (ptr)
726 return ptr;
727
728 if (goal) {
729 goal = 0;
730 goto again;
731 }
732
733 return NULL;
734}
735
736void * __init __alloc_bootmem_node_nopanic(pg_data_t *pgdat, unsigned long size,
737 unsigned long align, unsigned long goal)
738{
739 return ___alloc_bootmem_node_nopanic(pgdat, size, align, goal, 0);
740}
741
742void * __init ___alloc_bootmem_node(pg_data_t *pgdat, unsigned long size,
743 unsigned long align, unsigned long goal,
744 unsigned long limit)
745{
746 void *ptr;
747
748 ptr = ___alloc_bootmem_node_nopanic(pgdat, size, align, goal, 0);
749 if (ptr)
750 return ptr;
751
752 pr_alert("bootmem alloc of %lu bytes failed!\n", size);
753 panic("Out of memory");
754 return NULL;
755}
756
757/**
758 * __alloc_bootmem_node - allocate boot memory from a specific node
759 * @pgdat: node to allocate from
760 * @size: size of the request in bytes
761 * @align: alignment of the region
762 * @goal: preferred starting address of the region
763 *
764 * The goal is dropped if it can not be satisfied and the allocation will
765 * fall back to memory below @goal.
766 *
767 * Allocation may fall back to any node in the system if the specified node
768 * can not hold the requested memory.
769 *
770 * The function panics if the request can not be satisfied.
771 */
772void * __init __alloc_bootmem_node(pg_data_t *pgdat, unsigned long size,
773 unsigned long align, unsigned long goal)
774{
775 if (WARN_ON_ONCE(slab_is_available()))
776 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
777
778 return ___alloc_bootmem_node(pgdat, size, align, goal, 0);
779}
780
781void * __init __alloc_bootmem_node_high(pg_data_t *pgdat, unsigned long size,
782 unsigned long align, unsigned long goal)
783{
784#ifdef MAX_DMA32_PFN
785 unsigned long end_pfn;
786
787 if (WARN_ON_ONCE(slab_is_available()))
788 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
789
790 /* update goal according ...MAX_DMA32_PFN */
791 end_pfn = pgdat_end_pfn(pgdat);
792
793 if (end_pfn > MAX_DMA32_PFN + (128 >> (20 - PAGE_SHIFT)) &&
794 (goal >> PAGE_SHIFT) < MAX_DMA32_PFN) {
795 void *ptr;
796 unsigned long new_goal;
797
798 new_goal = MAX_DMA32_PFN << PAGE_SHIFT;
799 ptr = alloc_bootmem_bdata(pgdat->bdata, size, align,
800 new_goal, 0);
801 if (ptr)
802 return ptr;
803 }
804#endif
805
806 return __alloc_bootmem_node(pgdat, size, align, goal);
807
808}
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}