• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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 #include <linux/bug.h>
20 #include <linux/io.h>
21 
22 #include <asm/processor.h>
23 
24 #include "internal.h"
25 
26 #ifndef CONFIG_NEED_MULTIPLE_NODES
27 struct pglist_data __refdata contig_page_data = {
28 	.bdata = &bootmem_node_data[0]
29 };
30 EXPORT_SYMBOL(contig_page_data);
31 #endif
32 
33 unsigned long max_low_pfn;
34 unsigned long min_low_pfn;
35 unsigned long max_pfn;
36 
37 bootmem_data_t bootmem_node_data[MAX_NUMNODES] __initdata;
38 
39 static struct list_head bdata_list __initdata = LIST_HEAD_INIT(bdata_list);
40 
41 static int bootmem_debug;
42 
bootmem_debug_setup(char * buf)43 static int __init bootmem_debug_setup(char *buf)
44 {
45 	bootmem_debug = 1;
46 	return 0;
47 }
48 early_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 
bootmap_bytes(unsigned long pages)57 static 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  */
bootmem_bootmap_pages(unsigned long pages)68 unsigned 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  */
link_bootmem(bootmem_data_t * bdata)78 static 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  */
init_bootmem_core(bootmem_data_t * bdata,unsigned long mapstart,unsigned long start,unsigned long end)95 static 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  */
init_bootmem_node(pg_data_t * pgdat,unsigned long freepfn,unsigned long startpfn,unsigned long endpfn)128 unsigned 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  */
init_bootmem(unsigned long start,unsigned long pages)141 unsigned 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  */
free_bootmem_late(unsigned long physaddr,unsigned long size)157 void __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), cursor, 0);
168 		totalram_pages++;
169 	}
170 }
171 
free_all_bootmem_core(bootmem_data_t * bdata)172 static unsigned long __init free_all_bootmem_core(bootmem_data_t *bdata)
173 {
174 	struct page *page;
175 	unsigned long *map, start, end, pages, cur, 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), start, order);
214 			count += BITS_PER_LONG;
215 			start += BITS_PER_LONG;
216 		} else {
217 			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, cur, 0);
224 					count++;
225 				}
226 				vec >>= 1;
227 				++cur;
228 			}
229 		}
230 	}
231 
232 	cur = bdata->node_min_pfn;
233 	page = virt_to_page(bdata->node_bootmem_map);
234 	pages = bdata->node_low_pfn - bdata->node_min_pfn;
235 	pages = bootmem_bootmap_pages(pages);
236 	count += pages;
237 	while (pages--)
238 		__free_pages_bootmem(page++, cur++, 0);
239 
240 	bdebug("nid=%td released=%lx\n", bdata - bootmem_node_data, count);
241 
242 	return count;
243 }
244 
245 static int reset_managed_pages_done __initdata;
246 
reset_node_managed_pages(pg_data_t * pgdat)247 void reset_node_managed_pages(pg_data_t *pgdat)
248 {
249 	struct zone *z;
250 
251 	for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++)
252 		z->managed_pages = 0;
253 }
254 
reset_all_zones_managed_pages(void)255 void __init reset_all_zones_managed_pages(void)
256 {
257 	struct pglist_data *pgdat;
258 
259 	if (reset_managed_pages_done)
260 		return;
261 
262 	for_each_online_pgdat(pgdat)
263 		reset_node_managed_pages(pgdat);
264 
265 	reset_managed_pages_done = 1;
266 }
267 
268 /**
269  * free_all_bootmem - release free pages to the buddy allocator
270  *
271  * Returns the number of pages actually released.
272  */
free_all_bootmem(void)273 unsigned long __init free_all_bootmem(void)
274 {
275 	unsigned long total_pages = 0;
276 	bootmem_data_t *bdata;
277 
278 	reset_all_zones_managed_pages();
279 
280 	list_for_each_entry(bdata, &bdata_list, list)
281 		total_pages += free_all_bootmem_core(bdata);
282 
283 	totalram_pages += total_pages;
284 
285 	return total_pages;
286 }
287 
__free(bootmem_data_t * bdata,unsigned long sidx,unsigned long eidx)288 static void __init __free(bootmem_data_t *bdata,
289 			unsigned long sidx, unsigned long eidx)
290 {
291 	unsigned long idx;
292 
293 	bdebug("nid=%td start=%lx end=%lx\n", bdata - bootmem_node_data,
294 		sidx + bdata->node_min_pfn,
295 		eidx + bdata->node_min_pfn);
296 
297 	if (bdata->hint_idx > sidx)
298 		bdata->hint_idx = sidx;
299 
300 	for (idx = sidx; idx < eidx; idx++)
301 		if (!test_and_clear_bit(idx, bdata->node_bootmem_map))
302 			BUG();
303 }
304 
__reserve(bootmem_data_t * bdata,unsigned long sidx,unsigned long eidx,int flags)305 static int __init __reserve(bootmem_data_t *bdata, unsigned long sidx,
306 			unsigned long eidx, int flags)
307 {
308 	unsigned long idx;
309 	int exclusive = flags & BOOTMEM_EXCLUSIVE;
310 
311 	bdebug("nid=%td start=%lx end=%lx flags=%x\n",
312 		bdata - bootmem_node_data,
313 		sidx + bdata->node_min_pfn,
314 		eidx + bdata->node_min_pfn,
315 		flags);
316 
317 	for (idx = sidx; idx < eidx; idx++)
318 		if (test_and_set_bit(idx, bdata->node_bootmem_map)) {
319 			if (exclusive) {
320 				__free(bdata, sidx, idx);
321 				return -EBUSY;
322 			}
323 			bdebug("silent double reserve of PFN %lx\n",
324 				idx + bdata->node_min_pfn);
325 		}
326 	return 0;
327 }
328 
mark_bootmem_node(bootmem_data_t * bdata,unsigned long start,unsigned long end,int reserve,int flags)329 static int __init mark_bootmem_node(bootmem_data_t *bdata,
330 				unsigned long start, unsigned long end,
331 				int reserve, int flags)
332 {
333 	unsigned long sidx, eidx;
334 
335 	bdebug("nid=%td start=%lx end=%lx reserve=%d flags=%x\n",
336 		bdata - bootmem_node_data, start, end, reserve, flags);
337 
338 	BUG_ON(start < bdata->node_min_pfn);
339 	BUG_ON(end > bdata->node_low_pfn);
340 
341 	sidx = start - bdata->node_min_pfn;
342 	eidx = end - bdata->node_min_pfn;
343 
344 	if (reserve)
345 		return __reserve(bdata, sidx, eidx, flags);
346 	else
347 		__free(bdata, sidx, eidx);
348 	return 0;
349 }
350 
mark_bootmem(unsigned long start,unsigned long end,int reserve,int flags)351 static int __init mark_bootmem(unsigned long start, unsigned long end,
352 				int reserve, int flags)
353 {
354 	unsigned long pos;
355 	bootmem_data_t *bdata;
356 
357 	pos = start;
358 	list_for_each_entry(bdata, &bdata_list, list) {
359 		int err;
360 		unsigned long max;
361 
362 		if (pos < bdata->node_min_pfn ||
363 		    pos >= bdata->node_low_pfn) {
364 			BUG_ON(pos != start);
365 			continue;
366 		}
367 
368 		max = min(bdata->node_low_pfn, end);
369 
370 		err = mark_bootmem_node(bdata, pos, max, reserve, flags);
371 		if (reserve && err) {
372 			mark_bootmem(start, pos, 0, 0);
373 			return err;
374 		}
375 
376 		if (max == end)
377 			return 0;
378 		pos = bdata->node_low_pfn;
379 	}
380 	BUG();
381 }
382 
383 /**
384  * free_bootmem_node - mark a page range as usable
385  * @pgdat: node the range resides on
386  * @physaddr: starting address of the range
387  * @size: size of the range in bytes
388  *
389  * Partial pages will be considered reserved and left as they are.
390  *
391  * The range must reside completely on the specified node.
392  */
free_bootmem_node(pg_data_t * pgdat,unsigned long physaddr,unsigned long size)393 void __init free_bootmem_node(pg_data_t *pgdat, unsigned long physaddr,
394 			      unsigned long size)
395 {
396 	unsigned long start, end;
397 
398 	kmemleak_free_part(__va(physaddr), size);
399 
400 	start = PFN_UP(physaddr);
401 	end = PFN_DOWN(physaddr + size);
402 
403 	mark_bootmem_node(pgdat->bdata, start, end, 0, 0);
404 }
405 
406 /**
407  * free_bootmem - mark a page range as usable
408  * @addr: starting physical address of the range
409  * @size: size of the range in bytes
410  *
411  * Partial pages will be considered reserved and left as they are.
412  *
413  * The range must be contiguous but may span node boundaries.
414  */
free_bootmem(unsigned long physaddr,unsigned long size)415 void __init free_bootmem(unsigned long physaddr, unsigned long size)
416 {
417 	unsigned long start, end;
418 
419 	kmemleak_free_part(__va(physaddr), size);
420 
421 	start = PFN_UP(physaddr);
422 	end = PFN_DOWN(physaddr + size);
423 
424 	mark_bootmem(start, end, 0, 0);
425 }
426 
427 /**
428  * reserve_bootmem_node - mark a page range as reserved
429  * @pgdat: node the range resides on
430  * @physaddr: starting address of the range
431  * @size: size of the range in bytes
432  * @flags: reservation flags (see linux/bootmem.h)
433  *
434  * Partial pages will be reserved.
435  *
436  * The range must reside completely on the specified node.
437  */
reserve_bootmem_node(pg_data_t * pgdat,unsigned long physaddr,unsigned long size,int flags)438 int __init reserve_bootmem_node(pg_data_t *pgdat, unsigned long physaddr,
439 				 unsigned long size, int flags)
440 {
441 	unsigned long start, end;
442 
443 	start = PFN_DOWN(physaddr);
444 	end = PFN_UP(physaddr + size);
445 
446 	return mark_bootmem_node(pgdat->bdata, start, end, 1, flags);
447 }
448 
449 /**
450  * reserve_bootmem - mark a page range as reserved
451  * @addr: starting address of the range
452  * @size: size of the range in bytes
453  * @flags: reservation flags (see linux/bootmem.h)
454  *
455  * Partial pages will be reserved.
456  *
457  * The range must be contiguous but may span node boundaries.
458  */
reserve_bootmem(unsigned long addr,unsigned long size,int flags)459 int __init reserve_bootmem(unsigned long addr, unsigned long size,
460 			    int flags)
461 {
462 	unsigned long start, end;
463 
464 	start = PFN_DOWN(addr);
465 	end = PFN_UP(addr + size);
466 
467 	return mark_bootmem(start, end, 1, flags);
468 }
469 
align_idx(struct bootmem_data * bdata,unsigned long idx,unsigned long step)470 static unsigned long __init align_idx(struct bootmem_data *bdata,
471 				      unsigned long idx, unsigned long step)
472 {
473 	unsigned long base = bdata->node_min_pfn;
474 
475 	/*
476 	 * Align the index with respect to the node start so that the
477 	 * combination of both satisfies the requested alignment.
478 	 */
479 
480 	return ALIGN(base + idx, step) - base;
481 }
482 
align_off(struct bootmem_data * bdata,unsigned long off,unsigned long align)483 static unsigned long __init align_off(struct bootmem_data *bdata,
484 				      unsigned long off, unsigned long align)
485 {
486 	unsigned long base = PFN_PHYS(bdata->node_min_pfn);
487 
488 	/* Same as align_idx for byte offsets */
489 
490 	return ALIGN(base + off, align) - base;
491 }
492 
alloc_bootmem_bdata(struct bootmem_data * bdata,unsigned long size,unsigned long align,unsigned long goal,unsigned long limit)493 static void * __init alloc_bootmem_bdata(struct bootmem_data *bdata,
494 					unsigned long size, unsigned long align,
495 					unsigned long goal, unsigned long limit)
496 {
497 	unsigned long fallback = 0;
498 	unsigned long min, max, start, sidx, midx, step;
499 
500 	bdebug("nid=%td size=%lx [%lu pages] align=%lx goal=%lx limit=%lx\n",
501 		bdata - bootmem_node_data, size, PAGE_ALIGN(size) >> PAGE_SHIFT,
502 		align, goal, limit);
503 
504 	BUG_ON(!size);
505 	BUG_ON(align & (align - 1));
506 	BUG_ON(limit && goal + size > limit);
507 
508 	if (!bdata->node_bootmem_map)
509 		return NULL;
510 
511 	min = bdata->node_min_pfn;
512 	max = bdata->node_low_pfn;
513 
514 	goal >>= PAGE_SHIFT;
515 	limit >>= PAGE_SHIFT;
516 
517 	if (limit && max > limit)
518 		max = limit;
519 	if (max <= min)
520 		return NULL;
521 
522 	step = max(align >> PAGE_SHIFT, 1UL);
523 
524 	if (goal && min < goal && goal < max)
525 		start = ALIGN(goal, step);
526 	else
527 		start = ALIGN(min, step);
528 
529 	sidx = start - bdata->node_min_pfn;
530 	midx = max - bdata->node_min_pfn;
531 
532 	if (bdata->hint_idx > sidx) {
533 		/*
534 		 * Handle the valid case of sidx being zero and still
535 		 * catch the fallback below.
536 		 */
537 		fallback = sidx + 1;
538 		sidx = align_idx(bdata, bdata->hint_idx, step);
539 	}
540 
541 	while (1) {
542 		int merge;
543 		void *region;
544 		unsigned long eidx, i, start_off, end_off;
545 find_block:
546 		sidx = find_next_zero_bit(bdata->node_bootmem_map, midx, sidx);
547 		sidx = align_idx(bdata, sidx, step);
548 		eidx = sidx + PFN_UP(size);
549 
550 		if (sidx >= midx || eidx > midx)
551 			break;
552 
553 		for (i = sidx; i < eidx; i++)
554 			if (test_bit(i, bdata->node_bootmem_map)) {
555 				sidx = align_idx(bdata, i, step);
556 				if (sidx == i)
557 					sidx += step;
558 				goto find_block;
559 			}
560 
561 		if (bdata->last_end_off & (PAGE_SIZE - 1) &&
562 				PFN_DOWN(bdata->last_end_off) + 1 == sidx)
563 			start_off = align_off(bdata, bdata->last_end_off, align);
564 		else
565 			start_off = PFN_PHYS(sidx);
566 
567 		merge = PFN_DOWN(start_off) < sidx;
568 		end_off = start_off + size;
569 
570 		bdata->last_end_off = end_off;
571 		bdata->hint_idx = PFN_UP(end_off);
572 
573 		/*
574 		 * Reserve the area now:
575 		 */
576 		if (__reserve(bdata, PFN_DOWN(start_off) + merge,
577 				PFN_UP(end_off), BOOTMEM_EXCLUSIVE))
578 			BUG();
579 
580 		region = phys_to_virt(PFN_PHYS(bdata->node_min_pfn) +
581 				start_off);
582 		memset(region, 0, size);
583 		/*
584 		 * The min_count is set to 0 so that bootmem allocated blocks
585 		 * are never reported as leaks.
586 		 */
587 		kmemleak_alloc(region, size, 0, 0);
588 		return region;
589 	}
590 
591 	if (fallback) {
592 		sidx = align_idx(bdata, fallback - 1, step);
593 		fallback = 0;
594 		goto find_block;
595 	}
596 
597 	return NULL;
598 }
599 
alloc_bootmem_core(unsigned long size,unsigned long align,unsigned long goal,unsigned long limit)600 static void * __init alloc_bootmem_core(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 
608 	if (WARN_ON_ONCE(slab_is_available()))
609 		return kzalloc(size, GFP_NOWAIT);
610 
611 	list_for_each_entry(bdata, &bdata_list, list) {
612 		if (goal && bdata->node_low_pfn <= PFN_DOWN(goal))
613 			continue;
614 		if (limit && bdata->node_min_pfn >= PFN_DOWN(limit))
615 			break;
616 
617 		region = alloc_bootmem_bdata(bdata, size, align, goal, limit);
618 		if (region)
619 			return region;
620 	}
621 
622 	return NULL;
623 }
624 
___alloc_bootmem_nopanic(unsigned long size,unsigned long align,unsigned long goal,unsigned long limit)625 static void * __init ___alloc_bootmem_nopanic(unsigned long size,
626 					      unsigned long align,
627 					      unsigned long goal,
628 					      unsigned long limit)
629 {
630 	void *ptr;
631 
632 restart:
633 	ptr = alloc_bootmem_core(size, align, goal, limit);
634 	if (ptr)
635 		return ptr;
636 	if (goal) {
637 		goal = 0;
638 		goto restart;
639 	}
640 
641 	return NULL;
642 }
643 
644 /**
645  * __alloc_bootmem_nopanic - allocate boot memory without panicking
646  * @size: size of the request in bytes
647  * @align: alignment of the region
648  * @goal: preferred starting address of the region
649  *
650  * The goal is dropped if it can not be satisfied and the allocation will
651  * fall back to memory below @goal.
652  *
653  * Allocation may happen on any node in the system.
654  *
655  * Returns NULL on failure.
656  */
__alloc_bootmem_nopanic(unsigned long size,unsigned long align,unsigned long goal)657 void * __init __alloc_bootmem_nopanic(unsigned long size, unsigned long align,
658 					unsigned long goal)
659 {
660 	unsigned long limit = 0;
661 
662 	return ___alloc_bootmem_nopanic(size, align, goal, limit);
663 }
664 
___alloc_bootmem(unsigned long size,unsigned long align,unsigned long goal,unsigned long limit)665 static void * __init ___alloc_bootmem(unsigned long size, unsigned long align,
666 					unsigned long goal, unsigned long limit)
667 {
668 	void *mem = ___alloc_bootmem_nopanic(size, align, goal, limit);
669 
670 	if (mem)
671 		return mem;
672 	/*
673 	 * Whoops, we cannot satisfy the allocation request.
674 	 */
675 	printk(KERN_ALERT "bootmem alloc of %lu bytes failed!\n", size);
676 	panic("Out of memory");
677 	return NULL;
678 }
679 
680 /**
681  * __alloc_bootmem - allocate boot memory
682  * @size: size of the request in bytes
683  * @align: alignment of the region
684  * @goal: preferred starting address of the region
685  *
686  * The goal is dropped if it can not be satisfied and the allocation will
687  * fall back to memory below @goal.
688  *
689  * Allocation may happen on any node in the system.
690  *
691  * The function panics if the request can not be satisfied.
692  */
__alloc_bootmem(unsigned long size,unsigned long align,unsigned long goal)693 void * __init __alloc_bootmem(unsigned long size, unsigned long align,
694 			      unsigned long goal)
695 {
696 	unsigned long limit = 0;
697 
698 	return ___alloc_bootmem(size, align, goal, limit);
699 }
700 
___alloc_bootmem_node_nopanic(pg_data_t * pgdat,unsigned long size,unsigned long align,unsigned long goal,unsigned long limit)701 void * __init ___alloc_bootmem_node_nopanic(pg_data_t *pgdat,
702 				unsigned long size, unsigned long align,
703 				unsigned long goal, unsigned long limit)
704 {
705 	void *ptr;
706 
707 	if (WARN_ON_ONCE(slab_is_available()))
708 		return kzalloc(size, GFP_NOWAIT);
709 again:
710 
711 	/* do not panic in alloc_bootmem_bdata() */
712 	if (limit && goal + size > limit)
713 		limit = 0;
714 
715 	ptr = alloc_bootmem_bdata(pgdat->bdata, size, align, goal, limit);
716 	if (ptr)
717 		return ptr;
718 
719 	ptr = alloc_bootmem_core(size, align, goal, limit);
720 	if (ptr)
721 		return ptr;
722 
723 	if (goal) {
724 		goal = 0;
725 		goto again;
726 	}
727 
728 	return NULL;
729 }
730 
__alloc_bootmem_node_nopanic(pg_data_t * pgdat,unsigned long size,unsigned long align,unsigned long goal)731 void * __init __alloc_bootmem_node_nopanic(pg_data_t *pgdat, unsigned long size,
732 				   unsigned long align, unsigned long goal)
733 {
734 	if (WARN_ON_ONCE(slab_is_available()))
735 		return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
736 
737 	return ___alloc_bootmem_node_nopanic(pgdat, size, align, goal, 0);
738 }
739 
___alloc_bootmem_node(pg_data_t * pgdat,unsigned long size,unsigned long align,unsigned long goal,unsigned long limit)740 void * __init ___alloc_bootmem_node(pg_data_t *pgdat, unsigned long size,
741 				    unsigned long align, unsigned long goal,
742 				    unsigned long limit)
743 {
744 	void *ptr;
745 
746 	ptr = ___alloc_bootmem_node_nopanic(pgdat, size, align, goal, 0);
747 	if (ptr)
748 		return ptr;
749 
750 	printk(KERN_ALERT "bootmem alloc of %lu bytes failed!\n", size);
751 	panic("Out of memory");
752 	return NULL;
753 }
754 
755 /**
756  * __alloc_bootmem_node - allocate boot memory from a specific node
757  * @pgdat: node to allocate from
758  * @size: size of the request in bytes
759  * @align: alignment of the region
760  * @goal: preferred starting address of the region
761  *
762  * The goal is dropped if it can not be satisfied and the allocation will
763  * fall back to memory below @goal.
764  *
765  * Allocation may fall back to any node in the system if the specified node
766  * can not hold the requested memory.
767  *
768  * The function panics if the request can not be satisfied.
769  */
__alloc_bootmem_node(pg_data_t * pgdat,unsigned long size,unsigned long align,unsigned long goal)770 void * __init __alloc_bootmem_node(pg_data_t *pgdat, unsigned long size,
771 				   unsigned long align, unsigned long goal)
772 {
773 	if (WARN_ON_ONCE(slab_is_available()))
774 		return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
775 
776 	return  ___alloc_bootmem_node(pgdat, size, align, goal, 0);
777 }
778 
__alloc_bootmem_node_high(pg_data_t * pgdat,unsigned long size,unsigned long align,unsigned long goal)779 void * __init __alloc_bootmem_node_high(pg_data_t *pgdat, unsigned long size,
780 				   unsigned long align, unsigned long goal)
781 {
782 #ifdef MAX_DMA32_PFN
783 	unsigned long end_pfn;
784 
785 	if (WARN_ON_ONCE(slab_is_available()))
786 		return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
787 
788 	/* update goal according ...MAX_DMA32_PFN */
789 	end_pfn = pgdat_end_pfn(pgdat);
790 
791 	if (end_pfn > MAX_DMA32_PFN + (128 >> (20 - PAGE_SHIFT)) &&
792 	    (goal >> PAGE_SHIFT) < MAX_DMA32_PFN) {
793 		void *ptr;
794 		unsigned long new_goal;
795 
796 		new_goal = MAX_DMA32_PFN << PAGE_SHIFT;
797 		ptr = alloc_bootmem_bdata(pgdat->bdata, size, align,
798 						 new_goal, 0);
799 		if (ptr)
800 			return ptr;
801 	}
802 #endif
803 
804 	return __alloc_bootmem_node(pgdat, size, align, goal);
805 
806 }
807 
808 #ifndef ARCH_LOW_ADDRESS_LIMIT
809 #define ARCH_LOW_ADDRESS_LIMIT	0xffffffffUL
810 #endif
811 
812 /**
813  * __alloc_bootmem_low - allocate low boot memory
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 happen on any node in the system.
822  *
823  * The function panics if the request can not be satisfied.
824  */
__alloc_bootmem_low(unsigned long size,unsigned long align,unsigned long goal)825 void * __init __alloc_bootmem_low(unsigned long size, unsigned long align,
826 				  unsigned long goal)
827 {
828 	return ___alloc_bootmem(size, align, goal, ARCH_LOW_ADDRESS_LIMIT);
829 }
830 
__alloc_bootmem_low_nopanic(unsigned long size,unsigned long align,unsigned long goal)831 void * __init __alloc_bootmem_low_nopanic(unsigned long size,
832 					  unsigned long align,
833 					  unsigned long goal)
834 {
835 	return ___alloc_bootmem_nopanic(size, align, goal,
836 					ARCH_LOW_ADDRESS_LIMIT);
837 }
838 
839 /**
840  * __alloc_bootmem_low_node - allocate low boot memory from a specific node
841  * @pgdat: node to allocate from
842  * @size: size of the request in bytes
843  * @align: alignment of the region
844  * @goal: preferred starting address of the region
845  *
846  * The goal is dropped if it can not be satisfied and the allocation will
847  * fall back to memory below @goal.
848  *
849  * Allocation may fall back to any node in the system if the specified node
850  * can not hold the requested memory.
851  *
852  * The function panics if the request can not be satisfied.
853  */
__alloc_bootmem_low_node(pg_data_t * pgdat,unsigned long size,unsigned long align,unsigned long goal)854 void * __init __alloc_bootmem_low_node(pg_data_t *pgdat, unsigned long size,
855 				       unsigned long align, unsigned long goal)
856 {
857 	if (WARN_ON_ONCE(slab_is_available()))
858 		return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
859 
860 	return ___alloc_bootmem_node(pgdat, size, align,
861 				     goal, ARCH_LOW_ADDRESS_LIMIT);
862 }
863