bootmem.c revision 337998587f802535896e9ed16d19f97915ccd368
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/bootmem.h>
14#include <linux/module.h>
15#include <linux/kmemleak.h>
16#include <linux/range.h>
17
18#include <asm/bug.h>
19#include <asm/io.h>
20#include <asm/processor.h>
21
22#include "internal.h"
23
24unsigned long max_low_pfn;
25unsigned long min_low_pfn;
26unsigned long max_pfn;
27
28#ifdef CONFIG_CRASH_DUMP
29/*
30 * If we have booted due to a crash, max_pfn will be a very low value. We need
31 * to know the amount of memory that the previous kernel used.
32 */
33unsigned long saved_max_pfn;
34#endif
35
36#ifndef CONFIG_NO_BOOTMEM
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#endif
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
172#ifdef CONFIG_NO_BOOTMEM
173static void __init __free_pages_memory(unsigned long start, unsigned long end)
174{
175	int i;
176	unsigned long start_aligned, end_aligned;
177	int order = ilog2(BITS_PER_LONG);
178
179	start_aligned = (start + (BITS_PER_LONG - 1)) & ~(BITS_PER_LONG - 1);
180	end_aligned = end & ~(BITS_PER_LONG - 1);
181
182	if (end_aligned <= start_aligned) {
183		for (i = start; i < end; i++)
184			__free_pages_bootmem(pfn_to_page(i), 0);
185
186		return;
187	}
188
189	for (i = start; i < start_aligned; i++)
190		__free_pages_bootmem(pfn_to_page(i), 0);
191
192	for (i = start_aligned; i < end_aligned; i += BITS_PER_LONG)
193		__free_pages_bootmem(pfn_to_page(i), order);
194
195	for (i = end_aligned; i < end; i++)
196		__free_pages_bootmem(pfn_to_page(i), 0);
197}
198
199unsigned long __init free_all_memory_core_early(int nodeid)
200{
201	int i;
202	u64 start, end;
203	unsigned long count = 0;
204	struct range *range = NULL;
205	int nr_range;
206
207	nr_range = get_free_all_memory_range(&range, nodeid);
208
209	for (i = 0; i < nr_range; i++) {
210		start = range[i].start;
211		end = range[i].end;
212		count += end - start;
213		__free_pages_memory(start, end);
214	}
215
216	return count;
217}
218#else
219static unsigned long __init free_all_bootmem_core(bootmem_data_t *bdata)
220{
221	int aligned;
222	struct page *page;
223	unsigned long start, end, pages, count = 0;
224
225	if (!bdata->node_bootmem_map)
226		return 0;
227
228	start = bdata->node_min_pfn;
229	end = bdata->node_low_pfn;
230
231	/*
232	 * If the start is aligned to the machines wordsize, we might
233	 * be able to free pages in bulks of that order.
234	 */
235	aligned = !(start & (BITS_PER_LONG - 1));
236
237	bdebug("nid=%td start=%lx end=%lx aligned=%d\n",
238		bdata - bootmem_node_data, start, end, aligned);
239
240	while (start < end) {
241		unsigned long *map, idx, vec;
242
243		map = bdata->node_bootmem_map;
244		idx = start - bdata->node_min_pfn;
245		vec = ~map[idx / BITS_PER_LONG];
246
247		if (aligned && vec == ~0UL && start + BITS_PER_LONG < end) {
248			int order = ilog2(BITS_PER_LONG);
249
250			__free_pages_bootmem(pfn_to_page(start), order);
251			count += BITS_PER_LONG;
252		} else {
253			unsigned long off = 0;
254
255			while (vec && off < BITS_PER_LONG) {
256				if (vec & 1) {
257					page = pfn_to_page(start + off);
258					__free_pages_bootmem(page, 0);
259					count++;
260				}
261				vec >>= 1;
262				off++;
263			}
264		}
265		start += BITS_PER_LONG;
266	}
267
268	page = virt_to_page(bdata->node_bootmem_map);
269	pages = bdata->node_low_pfn - bdata->node_min_pfn;
270	pages = bootmem_bootmap_pages(pages);
271	count += pages;
272	while (pages--)
273		__free_pages_bootmem(page++, 0);
274
275	bdebug("nid=%td released=%lx\n", bdata - bootmem_node_data, count);
276
277	return count;
278}
279#endif
280
281/**
282 * free_all_bootmem_node - release a node's free pages to the buddy allocator
283 * @pgdat: node to be released
284 *
285 * Returns the number of pages actually released.
286 */
287unsigned long __init free_all_bootmem_node(pg_data_t *pgdat)
288{
289	register_page_bootmem_info_node(pgdat);
290#ifdef CONFIG_NO_BOOTMEM
291	/* free_all_memory_core_early(MAX_NUMNODES) will be called later */
292	return 0;
293#else
294	return free_all_bootmem_core(pgdat->bdata);
295#endif
296}
297
298/**
299 * free_all_bootmem - release free pages to the buddy allocator
300 *
301 * Returns the number of pages actually released.
302 */
303unsigned long __init free_all_bootmem(void)
304{
305#ifdef CONFIG_NO_BOOTMEM
306	/*
307	 * We need to use MAX_NUMNODES instead of NODE_DATA(0)->node_id
308	 *  because in some case like Node0 doesnt have RAM installed
309	 *  low ram will be on Node1
310	 * Use MAX_NUMNODES will make sure all ranges in early_node_map[]
311	 *  will be used instead of only Node0 related
312	 */
313	return free_all_memory_core_early(MAX_NUMNODES);
314#else
315	return free_all_bootmem_core(NODE_DATA(0)->bdata);
316#endif
317}
318
319#ifndef CONFIG_NO_BOOTMEM
320static void __init __free(bootmem_data_t *bdata,
321			unsigned long sidx, unsigned long eidx)
322{
323	unsigned long idx;
324
325	bdebug("nid=%td start=%lx end=%lx\n", bdata - bootmem_node_data,
326		sidx + bdata->node_min_pfn,
327		eidx + bdata->node_min_pfn);
328
329	if (bdata->hint_idx > sidx)
330		bdata->hint_idx = sidx;
331
332	for (idx = sidx; idx < eidx; idx++)
333		if (!test_and_clear_bit(idx, bdata->node_bootmem_map))
334			BUG();
335}
336
337static int __init __reserve(bootmem_data_t *bdata, unsigned long sidx,
338			unsigned long eidx, int flags)
339{
340	unsigned long idx;
341	int exclusive = flags & BOOTMEM_EXCLUSIVE;
342
343	bdebug("nid=%td start=%lx end=%lx flags=%x\n",
344		bdata - bootmem_node_data,
345		sidx + bdata->node_min_pfn,
346		eidx + bdata->node_min_pfn,
347		flags);
348
349	for (idx = sidx; idx < eidx; idx++)
350		if (test_and_set_bit(idx, bdata->node_bootmem_map)) {
351			if (exclusive) {
352				__free(bdata, sidx, idx);
353				return -EBUSY;
354			}
355			bdebug("silent double reserve of PFN %lx\n",
356				idx + bdata->node_min_pfn);
357		}
358	return 0;
359}
360
361static int __init mark_bootmem_node(bootmem_data_t *bdata,
362				unsigned long start, unsigned long end,
363				int reserve, int flags)
364{
365	unsigned long sidx, eidx;
366
367	bdebug("nid=%td start=%lx end=%lx reserve=%d flags=%x\n",
368		bdata - bootmem_node_data, start, end, reserve, flags);
369
370	BUG_ON(start < bdata->node_min_pfn);
371	BUG_ON(end > bdata->node_low_pfn);
372
373	sidx = start - bdata->node_min_pfn;
374	eidx = end - bdata->node_min_pfn;
375
376	if (reserve)
377		return __reserve(bdata, sidx, eidx, flags);
378	else
379		__free(bdata, sidx, eidx);
380	return 0;
381}
382
383static int __init mark_bootmem(unsigned long start, unsigned long end,
384				int reserve, int flags)
385{
386	unsigned long pos;
387	bootmem_data_t *bdata;
388
389	pos = start;
390	list_for_each_entry(bdata, &bdata_list, list) {
391		int err;
392		unsigned long max;
393
394		if (pos < bdata->node_min_pfn ||
395		    pos >= bdata->node_low_pfn) {
396			BUG_ON(pos != start);
397			continue;
398		}
399
400		max = min(bdata->node_low_pfn, end);
401
402		err = mark_bootmem_node(bdata, pos, max, reserve, flags);
403		if (reserve && err) {
404			mark_bootmem(start, pos, 0, 0);
405			return err;
406		}
407
408		if (max == end)
409			return 0;
410		pos = bdata->node_low_pfn;
411	}
412	BUG();
413}
414#endif
415
416/**
417 * free_bootmem_node - mark a page range as usable
418 * @pgdat: node the range resides on
419 * @physaddr: starting address of the range
420 * @size: size of the range in bytes
421 *
422 * Partial pages will be considered reserved and left as they are.
423 *
424 * The range must reside completely on the specified node.
425 */
426void __init free_bootmem_node(pg_data_t *pgdat, unsigned long physaddr,
427			      unsigned long size)
428{
429#ifdef CONFIG_NO_BOOTMEM
430	free_early(physaddr, physaddr + size);
431#else
432	unsigned long start, end;
433
434	kmemleak_free_part(__va(physaddr), size);
435
436	start = PFN_UP(physaddr);
437	end = PFN_DOWN(physaddr + size);
438
439	mark_bootmem_node(pgdat->bdata, start, end, 0, 0);
440#endif
441}
442
443/**
444 * free_bootmem - mark a page range as usable
445 * @addr: starting address of the range
446 * @size: size of the range in bytes
447 *
448 * Partial pages will be considered reserved and left as they are.
449 *
450 * The range must be contiguous but may span node boundaries.
451 */
452void __init free_bootmem(unsigned long addr, unsigned long size)
453{
454#ifdef CONFIG_NO_BOOTMEM
455	free_early(addr, addr + size);
456#else
457	unsigned long start, end;
458
459	kmemleak_free_part(__va(addr), size);
460
461	start = PFN_UP(addr);
462	end = PFN_DOWN(addr + size);
463
464	mark_bootmem(start, end, 0, 0);
465#endif
466}
467
468/**
469 * reserve_bootmem_node - mark a page range as reserved
470 * @pgdat: node the range resides on
471 * @physaddr: starting address of the range
472 * @size: size of the range in bytes
473 * @flags: reservation flags (see linux/bootmem.h)
474 *
475 * Partial pages will be reserved.
476 *
477 * The range must reside completely on the specified node.
478 */
479int __init reserve_bootmem_node(pg_data_t *pgdat, unsigned long physaddr,
480				 unsigned long size, int flags)
481{
482#ifdef CONFIG_NO_BOOTMEM
483	panic("no bootmem");
484	return 0;
485#else
486	unsigned long start, end;
487
488	start = PFN_DOWN(physaddr);
489	end = PFN_UP(physaddr + size);
490
491	return mark_bootmem_node(pgdat->bdata, start, end, 1, flags);
492#endif
493}
494
495/**
496 * reserve_bootmem - mark a page range as usable
497 * @addr: starting address of the range
498 * @size: size of the range in bytes
499 * @flags: reservation flags (see linux/bootmem.h)
500 *
501 * Partial pages will be reserved.
502 *
503 * The range must be contiguous but may span node boundaries.
504 */
505int __init reserve_bootmem(unsigned long addr, unsigned long size,
506			    int flags)
507{
508#ifdef CONFIG_NO_BOOTMEM
509	panic("no bootmem");
510	return 0;
511#else
512	unsigned long start, end;
513
514	start = PFN_DOWN(addr);
515	end = PFN_UP(addr + size);
516
517	return mark_bootmem(start, end, 1, flags);
518#endif
519}
520
521#ifndef CONFIG_NO_BOOTMEM
522static unsigned long __init align_idx(struct bootmem_data *bdata,
523				      unsigned long idx, unsigned long step)
524{
525	unsigned long base = bdata->node_min_pfn;
526
527	/*
528	 * Align the index with respect to the node start so that the
529	 * combination of both satisfies the requested alignment.
530	 */
531
532	return ALIGN(base + idx, step) - base;
533}
534
535static unsigned long __init align_off(struct bootmem_data *bdata,
536				      unsigned long off, unsigned long align)
537{
538	unsigned long base = PFN_PHYS(bdata->node_min_pfn);
539
540	/* Same as align_idx for byte offsets */
541
542	return ALIGN(base + off, align) - base;
543}
544
545static void * __init alloc_bootmem_core(struct bootmem_data *bdata,
546					unsigned long size, unsigned long align,
547					unsigned long goal, unsigned long limit)
548{
549	unsigned long fallback = 0;
550	unsigned long min, max, start, sidx, midx, step;
551
552	bdebug("nid=%td size=%lx [%lu pages] align=%lx goal=%lx limit=%lx\n",
553		bdata - bootmem_node_data, size, PAGE_ALIGN(size) >> PAGE_SHIFT,
554		align, goal, limit);
555
556	BUG_ON(!size);
557	BUG_ON(align & (align - 1));
558	BUG_ON(limit && goal + size > limit);
559
560	if (!bdata->node_bootmem_map)
561		return NULL;
562
563	min = bdata->node_min_pfn;
564	max = bdata->node_low_pfn;
565
566	goal >>= PAGE_SHIFT;
567	limit >>= PAGE_SHIFT;
568
569	if (limit && max > limit)
570		max = limit;
571	if (max <= min)
572		return NULL;
573
574	step = max(align >> PAGE_SHIFT, 1UL);
575
576	if (goal && min < goal && goal < max)
577		start = ALIGN(goal, step);
578	else
579		start = ALIGN(min, step);
580
581	sidx = start - bdata->node_min_pfn;
582	midx = max - bdata->node_min_pfn;
583
584	if (bdata->hint_idx > sidx) {
585		/*
586		 * Handle the valid case of sidx being zero and still
587		 * catch the fallback below.
588		 */
589		fallback = sidx + 1;
590		sidx = align_idx(bdata, bdata->hint_idx, step);
591	}
592
593	while (1) {
594		int merge;
595		void *region;
596		unsigned long eidx, i, start_off, end_off;
597find_block:
598		sidx = find_next_zero_bit(bdata->node_bootmem_map, midx, sidx);
599		sidx = align_idx(bdata, sidx, step);
600		eidx = sidx + PFN_UP(size);
601
602		if (sidx >= midx || eidx > midx)
603			break;
604
605		for (i = sidx; i < eidx; i++)
606			if (test_bit(i, bdata->node_bootmem_map)) {
607				sidx = align_idx(bdata, i, step);
608				if (sidx == i)
609					sidx += step;
610				goto find_block;
611			}
612
613		if (bdata->last_end_off & (PAGE_SIZE - 1) &&
614				PFN_DOWN(bdata->last_end_off) + 1 == sidx)
615			start_off = align_off(bdata, bdata->last_end_off, align);
616		else
617			start_off = PFN_PHYS(sidx);
618
619		merge = PFN_DOWN(start_off) < sidx;
620		end_off = start_off + size;
621
622		bdata->last_end_off = end_off;
623		bdata->hint_idx = PFN_UP(end_off);
624
625		/*
626		 * Reserve the area now:
627		 */
628		if (__reserve(bdata, PFN_DOWN(start_off) + merge,
629				PFN_UP(end_off), BOOTMEM_EXCLUSIVE))
630			BUG();
631
632		region = phys_to_virt(PFN_PHYS(bdata->node_min_pfn) +
633				start_off);
634		memset(region, 0, size);
635		/*
636		 * The min_count is set to 0 so that bootmem allocated blocks
637		 * are never reported as leaks.
638		 */
639		kmemleak_alloc(region, size, 0, 0);
640		return region;
641	}
642
643	if (fallback) {
644		sidx = align_idx(bdata, fallback - 1, step);
645		fallback = 0;
646		goto find_block;
647	}
648
649	return NULL;
650}
651
652static void * __init alloc_arch_preferred_bootmem(bootmem_data_t *bdata,
653					unsigned long size, unsigned long align,
654					unsigned long goal, unsigned long limit)
655{
656	if (WARN_ON_ONCE(slab_is_available()))
657		return kzalloc(size, GFP_NOWAIT);
658
659#ifdef CONFIG_HAVE_ARCH_BOOTMEM
660	{
661		bootmem_data_t *p_bdata;
662
663		p_bdata = bootmem_arch_preferred_node(bdata, size, align,
664							goal, limit);
665		if (p_bdata)
666			return alloc_bootmem_core(p_bdata, size, align,
667							goal, limit);
668	}
669#endif
670	return NULL;
671}
672#endif
673
674static void * __init ___alloc_bootmem_nopanic(unsigned long size,
675					unsigned long align,
676					unsigned long goal,
677					unsigned long limit)
678{
679#ifdef CONFIG_NO_BOOTMEM
680	void *ptr;
681
682	if (WARN_ON_ONCE(slab_is_available()))
683		return kzalloc(size, GFP_NOWAIT);
684
685restart:
686
687	ptr = __alloc_memory_core_early(MAX_NUMNODES, size, align, goal, limit);
688
689	if (ptr)
690		return ptr;
691
692	if (goal != 0) {
693		goal = 0;
694		goto restart;
695	}
696
697	return NULL;
698#else
699	bootmem_data_t *bdata;
700	void *region;
701
702restart:
703	region = alloc_arch_preferred_bootmem(NULL, size, align, goal, limit);
704	if (region)
705		return region;
706
707	list_for_each_entry(bdata, &bdata_list, list) {
708		if (goal && bdata->node_low_pfn <= PFN_DOWN(goal))
709			continue;
710		if (limit && bdata->node_min_pfn >= PFN_DOWN(limit))
711			break;
712
713		region = alloc_bootmem_core(bdata, size, align, goal, limit);
714		if (region)
715			return region;
716	}
717
718	if (goal) {
719		goal = 0;
720		goto restart;
721	}
722
723	return NULL;
724#endif
725}
726
727/**
728 * __alloc_bootmem_nopanic - allocate boot memory without panicking
729 * @size: size of the request in bytes
730 * @align: alignment of the region
731 * @goal: preferred starting address of the region
732 *
733 * The goal is dropped if it can not be satisfied and the allocation will
734 * fall back to memory below @goal.
735 *
736 * Allocation may happen on any node in the system.
737 *
738 * Returns NULL on failure.
739 */
740void * __init __alloc_bootmem_nopanic(unsigned long size, unsigned long align,
741					unsigned long goal)
742{
743	unsigned long limit = 0;
744
745#ifdef CONFIG_NO_BOOTMEM
746	limit = -1UL;
747#endif
748
749	return ___alloc_bootmem_nopanic(size, align, goal, limit);
750}
751
752static void * __init ___alloc_bootmem(unsigned long size, unsigned long align,
753					unsigned long goal, unsigned long limit)
754{
755	void *mem = ___alloc_bootmem_nopanic(size, align, goal, limit);
756
757	if (mem)
758		return mem;
759	/*
760	 * Whoops, we cannot satisfy the allocation request.
761	 */
762	printk(KERN_ALERT "bootmem alloc of %lu bytes failed!\n", size);
763	panic("Out of memory");
764	return NULL;
765}
766
767/**
768 * __alloc_bootmem - allocate boot memory
769 * @size: size of the request in bytes
770 * @align: alignment of the region
771 * @goal: preferred starting address of the region
772 *
773 * The goal is dropped if it can not be satisfied and the allocation will
774 * fall back to memory below @goal.
775 *
776 * Allocation may happen on any node in the system.
777 *
778 * The function panics if the request can not be satisfied.
779 */
780void * __init __alloc_bootmem(unsigned long size, unsigned long align,
781			      unsigned long goal)
782{
783	unsigned long limit = 0;
784
785#ifdef CONFIG_NO_BOOTMEM
786	limit = -1UL;
787#endif
788
789	return ___alloc_bootmem(size, align, goal, limit);
790}
791
792#ifndef CONFIG_NO_BOOTMEM
793static void * __init ___alloc_bootmem_node(bootmem_data_t *bdata,
794				unsigned long size, unsigned long align,
795				unsigned long goal, unsigned long limit)
796{
797	void *ptr;
798
799	ptr = alloc_arch_preferred_bootmem(bdata, size, align, goal, limit);
800	if (ptr)
801		return ptr;
802
803	ptr = alloc_bootmem_core(bdata, size, align, goal, limit);
804	if (ptr)
805		return ptr;
806
807	return ___alloc_bootmem(size, align, goal, limit);
808}
809#endif
810
811/**
812 * __alloc_bootmem_node - allocate boot memory from a specific node
813 * @pgdat: node to allocate from
814 * @size: size of the request in bytes
815 * @align: alignment of the region
816 * @goal: preferred starting address of the region
817 *
818 * The goal is dropped if it can not be satisfied and the allocation will
819 * fall back to memory below @goal.
820 *
821 * Allocation may fall back to any node in the system if the specified node
822 * can not hold the requested memory.
823 *
824 * The function panics if the request can not be satisfied.
825 */
826void * __init __alloc_bootmem_node(pg_data_t *pgdat, unsigned long size,
827				   unsigned long align, unsigned long goal)
828{
829	if (WARN_ON_ONCE(slab_is_available()))
830		return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
831
832#ifdef CONFIG_NO_BOOTMEM
833	return __alloc_memory_core_early(pgdat->node_id, size, align,
834					 goal, -1ULL);
835#else
836	return ___alloc_bootmem_node(pgdat->bdata, size, align, goal, 0);
837#endif
838}
839
840void * __init __alloc_bootmem_node_high(pg_data_t *pgdat, unsigned long size,
841				   unsigned long align, unsigned long goal)
842{
843#ifdef MAX_DMA32_PFN
844	unsigned long end_pfn;
845
846	if (WARN_ON_ONCE(slab_is_available()))
847		return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
848
849	/* update goal according ...MAX_DMA32_PFN */
850	end_pfn = pgdat->node_start_pfn + pgdat->node_spanned_pages;
851
852	if (end_pfn > MAX_DMA32_PFN + (128 >> (20 - PAGE_SHIFT)) &&
853	    (goal >> PAGE_SHIFT) < MAX_DMA32_PFN) {
854		void *ptr;
855		unsigned long new_goal;
856
857		new_goal = MAX_DMA32_PFN << PAGE_SHIFT;
858#ifdef CONFIG_NO_BOOTMEM
859		ptr =  __alloc_memory_core_early(pgdat->node_id, size, align,
860						 new_goal, -1ULL);
861#else
862		ptr = alloc_bootmem_core(pgdat->bdata, size, align,
863						 new_goal, 0);
864#endif
865		if (ptr)
866			return ptr;
867	}
868#endif
869
870	return __alloc_bootmem_node(pgdat, size, align, goal);
871
872}
873
874#ifdef CONFIG_SPARSEMEM
875/**
876 * alloc_bootmem_section - allocate boot memory from a specific section
877 * @size: size of the request in bytes
878 * @section_nr: sparse map section to allocate from
879 *
880 * Return NULL on failure.
881 */
882void * __init alloc_bootmem_section(unsigned long size,
883				    unsigned long section_nr)
884{
885#ifdef CONFIG_NO_BOOTMEM
886	unsigned long pfn, goal, limit;
887
888	pfn = section_nr_to_pfn(section_nr);
889	goal = pfn << PAGE_SHIFT;
890	limit = section_nr_to_pfn(section_nr + 1) << PAGE_SHIFT;
891
892	return __alloc_memory_core_early(early_pfn_to_nid(pfn), size,
893					 SMP_CACHE_BYTES, goal, limit);
894#else
895	bootmem_data_t *bdata;
896	unsigned long pfn, goal, limit;
897
898	pfn = section_nr_to_pfn(section_nr);
899	goal = pfn << PAGE_SHIFT;
900	limit = section_nr_to_pfn(section_nr + 1) << PAGE_SHIFT;
901	bdata = &bootmem_node_data[early_pfn_to_nid(pfn)];
902
903	return alloc_bootmem_core(bdata, size, SMP_CACHE_BYTES, goal, limit);
904#endif
905}
906#endif
907
908void * __init __alloc_bootmem_node_nopanic(pg_data_t *pgdat, unsigned long size,
909				   unsigned long align, unsigned long goal)
910{
911	void *ptr;
912
913	if (WARN_ON_ONCE(slab_is_available()))
914		return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
915
916#ifdef CONFIG_NO_BOOTMEM
917	ptr =  __alloc_memory_core_early(pgdat->node_id, size, align,
918						 goal, -1ULL);
919#else
920	ptr = alloc_arch_preferred_bootmem(pgdat->bdata, size, align, goal, 0);
921	if (ptr)
922		return ptr;
923
924	ptr = alloc_bootmem_core(pgdat->bdata, size, align, goal, 0);
925#endif
926	if (ptr)
927		return ptr;
928
929	return __alloc_bootmem_nopanic(size, align, goal);
930}
931
932#ifndef ARCH_LOW_ADDRESS_LIMIT
933#define ARCH_LOW_ADDRESS_LIMIT	0xffffffffUL
934#endif
935
936/**
937 * __alloc_bootmem_low - allocate low boot memory
938 * @size: size of the request in bytes
939 * @align: alignment of the region
940 * @goal: preferred starting address of the region
941 *
942 * The goal is dropped if it can not be satisfied and the allocation will
943 * fall back to memory below @goal.
944 *
945 * Allocation may happen on any node in the system.
946 *
947 * The function panics if the request can not be satisfied.
948 */
949void * __init __alloc_bootmem_low(unsigned long size, unsigned long align,
950				  unsigned long goal)
951{
952	return ___alloc_bootmem(size, align, goal, ARCH_LOW_ADDRESS_LIMIT);
953}
954
955/**
956 * __alloc_bootmem_low_node - allocate low boot memory from a specific node
957 * @pgdat: node to allocate from
958 * @size: size of the request in bytes
959 * @align: alignment of the region
960 * @goal: preferred starting address of the region
961 *
962 * The goal is dropped if it can not be satisfied and the allocation will
963 * fall back to memory below @goal.
964 *
965 * Allocation may fall back to any node in the system if the specified node
966 * can not hold the requested memory.
967 *
968 * The function panics if the request can not be satisfied.
969 */
970void * __init __alloc_bootmem_low_node(pg_data_t *pgdat, unsigned long size,
971				       unsigned long align, unsigned long goal)
972{
973	if (WARN_ON_ONCE(slab_is_available()))
974		return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
975
976#ifdef CONFIG_NO_BOOTMEM
977	return __alloc_memory_core_early(pgdat->node_id, size, align,
978				goal, ARCH_LOW_ADDRESS_LIMIT);
979#else
980	return ___alloc_bootmem_node(pgdat->bdata, size, align,
981				goal, ARCH_LOW_ADDRESS_LIMIT);
982#endif
983}
984