xref: /linux/mm/memblock.c (revision 24ce659dcc02c21f8d6c0a7589c3320a4dfa8152)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * Procedures for maintaining information about logical memory blocks.
4  *
5  * Peter Bergner, IBM Corp.	June 2001.
6  * Copyright (C) 2001 Peter Bergner.
7  */
8 
9 #include <linux/kernel.h>
10 #include <linux/slab.h>
11 #include <linux/init.h>
12 #include <linux/bitops.h>
13 #include <linux/poison.h>
14 #include <linux/pfn.h>
15 #include <linux/debugfs.h>
16 #include <linux/kmemleak.h>
17 #include <linux/seq_file.h>
18 #include <linux/memblock.h>
19 
20 #include <asm/sections.h>
21 #include <linux/io.h>
22 
23 #include "internal.h"
24 
25 #define INIT_MEMBLOCK_REGIONS			128
26 #define INIT_PHYSMEM_REGIONS			4
27 
28 #ifndef INIT_MEMBLOCK_RESERVED_REGIONS
29 # define INIT_MEMBLOCK_RESERVED_REGIONS		INIT_MEMBLOCK_REGIONS
30 #endif
31 
32 /**
33  * DOC: memblock overview
34  *
35  * Memblock is a method of managing memory regions during the early
36  * boot period when the usual kernel memory allocators are not up and
37  * running.
38  *
39  * Memblock views the system memory as collections of contiguous
40  * regions. There are several types of these collections:
41  *
42  * * ``memory`` - describes the physical memory available to the
43  *   kernel; this may differ from the actual physical memory installed
44  *   in the system, for instance when the memory is restricted with
45  *   ``mem=`` command line parameter
46  * * ``reserved`` - describes the regions that were allocated
47  * * ``physmap`` - describes the actual physical memory regardless of
48  *   the possible restrictions; the ``physmap`` type is only available
49  *   on some architectures.
50  *
51  * Each region is represented by :c:type:`struct memblock_region` that
52  * defines the region extents, its attributes and NUMA node id on NUMA
53  * systems. Every memory type is described by the :c:type:`struct
54  * memblock_type` which contains an array of memory regions along with
55  * the allocator metadata. The memory types are nicely wrapped with
56  * :c:type:`struct memblock`. This structure is statically initialzed
57  * at build time. The region arrays for the "memory" and "reserved"
58  * types are initially sized to %INIT_MEMBLOCK_REGIONS and for the
59  * "physmap" type to %INIT_PHYSMEM_REGIONS.
60  * The memblock_allow_resize() enables automatic resizing of the region
61  * arrays during addition of new regions. This feature should be used
62  * with care so that memory allocated for the region array will not
63  * overlap with areas that should be reserved, for example initrd.
64  *
65  * The early architecture setup should tell memblock what the physical
66  * memory layout is by using memblock_add() or memblock_add_node()
67  * functions. The first function does not assign the region to a NUMA
68  * node and it is appropriate for UMA systems. Yet, it is possible to
69  * use it on NUMA systems as well and assign the region to a NUMA node
70  * later in the setup process using memblock_set_node(). The
71  * memblock_add_node() performs such an assignment directly.
72  *
73  * Once memblock is setup the memory can be allocated using one of the
74  * API variants:
75  *
76  * * memblock_phys_alloc*() - these functions return the **physical**
77  *   address of the allocated memory
78  * * memblock_alloc*() - these functions return the **virtual** address
79  *   of the allocated memory.
80  *
81  * Note, that both API variants use implict assumptions about allowed
82  * memory ranges and the fallback methods. Consult the documentation
83  * of memblock_alloc_internal() and memblock_alloc_range_nid()
84  * functions for more elaborate description.
85  *
86  * As the system boot progresses, the architecture specific mem_init()
87  * function frees all the memory to the buddy page allocator.
88  *
89  * Unless an architecture enables %CONFIG_ARCH_KEEP_MEMBLOCK, the
90  * memblock data structures will be discarded after the system
91  * initialization completes.
92  */
93 
94 #ifndef CONFIG_NEED_MULTIPLE_NODES
95 struct pglist_data __refdata contig_page_data;
96 EXPORT_SYMBOL(contig_page_data);
97 #endif
98 
99 unsigned long max_low_pfn;
100 unsigned long min_low_pfn;
101 unsigned long max_pfn;
102 unsigned long long max_possible_pfn;
103 
104 static struct memblock_region memblock_memory_init_regions[INIT_MEMBLOCK_REGIONS] __initdata_memblock;
105 static struct memblock_region memblock_reserved_init_regions[INIT_MEMBLOCK_RESERVED_REGIONS] __initdata_memblock;
106 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
107 static struct memblock_region memblock_physmem_init_regions[INIT_PHYSMEM_REGIONS] __initdata_memblock;
108 #endif
109 
110 struct memblock memblock __initdata_memblock = {
111 	.memory.regions		= memblock_memory_init_regions,
112 	.memory.cnt		= 1,	/* empty dummy entry */
113 	.memory.max		= INIT_MEMBLOCK_REGIONS,
114 	.memory.name		= "memory",
115 
116 	.reserved.regions	= memblock_reserved_init_regions,
117 	.reserved.cnt		= 1,	/* empty dummy entry */
118 	.reserved.max		= INIT_MEMBLOCK_RESERVED_REGIONS,
119 	.reserved.name		= "reserved",
120 
121 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
122 	.physmem.regions	= memblock_physmem_init_regions,
123 	.physmem.cnt		= 1,	/* empty dummy entry */
124 	.physmem.max		= INIT_PHYSMEM_REGIONS,
125 	.physmem.name		= "physmem",
126 #endif
127 
128 	.bottom_up		= false,
129 	.current_limit		= MEMBLOCK_ALLOC_ANYWHERE,
130 };
131 
132 int memblock_debug __initdata_memblock;
133 static bool system_has_some_mirror __initdata_memblock = false;
134 static int memblock_can_resize __initdata_memblock;
135 static int memblock_memory_in_slab __initdata_memblock = 0;
136 static int memblock_reserved_in_slab __initdata_memblock = 0;
137 
138 static enum memblock_flags __init_memblock choose_memblock_flags(void)
139 {
140 	return system_has_some_mirror ? MEMBLOCK_MIRROR : MEMBLOCK_NONE;
141 }
142 
143 /* adjust *@size so that (@base + *@size) doesn't overflow, return new size */
144 static inline phys_addr_t memblock_cap_size(phys_addr_t base, phys_addr_t *size)
145 {
146 	return *size = min(*size, PHYS_ADDR_MAX - base);
147 }
148 
149 /*
150  * Address comparison utilities
151  */
152 static unsigned long __init_memblock memblock_addrs_overlap(phys_addr_t base1, phys_addr_t size1,
153 				       phys_addr_t base2, phys_addr_t size2)
154 {
155 	return ((base1 < (base2 + size2)) && (base2 < (base1 + size1)));
156 }
157 
158 bool __init_memblock memblock_overlaps_region(struct memblock_type *type,
159 					phys_addr_t base, phys_addr_t size)
160 {
161 	unsigned long i;
162 
163 	for (i = 0; i < type->cnt; i++)
164 		if (memblock_addrs_overlap(base, size, type->regions[i].base,
165 					   type->regions[i].size))
166 			break;
167 	return i < type->cnt;
168 }
169 
170 /**
171  * __memblock_find_range_bottom_up - find free area utility in bottom-up
172  * @start: start of candidate range
173  * @end: end of candidate range, can be %MEMBLOCK_ALLOC_ANYWHERE or
174  *       %MEMBLOCK_ALLOC_ACCESSIBLE
175  * @size: size of free area to find
176  * @align: alignment of free area to find
177  * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
178  * @flags: pick from blocks based on memory attributes
179  *
180  * Utility called from memblock_find_in_range_node(), find free area bottom-up.
181  *
182  * Return:
183  * Found address on success, 0 on failure.
184  */
185 static phys_addr_t __init_memblock
186 __memblock_find_range_bottom_up(phys_addr_t start, phys_addr_t end,
187 				phys_addr_t size, phys_addr_t align, int nid,
188 				enum memblock_flags flags)
189 {
190 	phys_addr_t this_start, this_end, cand;
191 	u64 i;
192 
193 	for_each_free_mem_range(i, nid, flags, &this_start, &this_end, NULL) {
194 		this_start = clamp(this_start, start, end);
195 		this_end = clamp(this_end, start, end);
196 
197 		cand = round_up(this_start, align);
198 		if (cand < this_end && this_end - cand >= size)
199 			return cand;
200 	}
201 
202 	return 0;
203 }
204 
205 /**
206  * __memblock_find_range_top_down - find free area utility, in top-down
207  * @start: start of candidate range
208  * @end: end of candidate range, can be %MEMBLOCK_ALLOC_ANYWHERE or
209  *       %MEMBLOCK_ALLOC_ACCESSIBLE
210  * @size: size of free area to find
211  * @align: alignment of free area to find
212  * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
213  * @flags: pick from blocks based on memory attributes
214  *
215  * Utility called from memblock_find_in_range_node(), find free area top-down.
216  *
217  * Return:
218  * Found address on success, 0 on failure.
219  */
220 static phys_addr_t __init_memblock
221 __memblock_find_range_top_down(phys_addr_t start, phys_addr_t end,
222 			       phys_addr_t size, phys_addr_t align, int nid,
223 			       enum memblock_flags flags)
224 {
225 	phys_addr_t this_start, this_end, cand;
226 	u64 i;
227 
228 	for_each_free_mem_range_reverse(i, nid, flags, &this_start, &this_end,
229 					NULL) {
230 		this_start = clamp(this_start, start, end);
231 		this_end = clamp(this_end, start, end);
232 
233 		if (this_end < size)
234 			continue;
235 
236 		cand = round_down(this_end - size, align);
237 		if (cand >= this_start)
238 			return cand;
239 	}
240 
241 	return 0;
242 }
243 
244 /**
245  * memblock_find_in_range_node - find free area in given range and node
246  * @size: size of free area to find
247  * @align: alignment of free area to find
248  * @start: start of candidate range
249  * @end: end of candidate range, can be %MEMBLOCK_ALLOC_ANYWHERE or
250  *       %MEMBLOCK_ALLOC_ACCESSIBLE
251  * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
252  * @flags: pick from blocks based on memory attributes
253  *
254  * Find @size free area aligned to @align in the specified range and node.
255  *
256  * When allocation direction is bottom-up, the @start should be greater
257  * than the end of the kernel image. Otherwise, it will be trimmed. The
258  * reason is that we want the bottom-up allocation just near the kernel
259  * image so it is highly likely that the allocated memory and the kernel
260  * will reside in the same node.
261  *
262  * If bottom-up allocation failed, will try to allocate memory top-down.
263  *
264  * Return:
265  * Found address on success, 0 on failure.
266  */
267 static phys_addr_t __init_memblock memblock_find_in_range_node(phys_addr_t size,
268 					phys_addr_t align, phys_addr_t start,
269 					phys_addr_t end, int nid,
270 					enum memblock_flags flags)
271 {
272 	phys_addr_t kernel_end, ret;
273 
274 	/* pump up @end */
275 	if (end == MEMBLOCK_ALLOC_ACCESSIBLE ||
276 	    end == MEMBLOCK_ALLOC_KASAN)
277 		end = memblock.current_limit;
278 
279 	/* avoid allocating the first page */
280 	start = max_t(phys_addr_t, start, PAGE_SIZE);
281 	end = max(start, end);
282 	kernel_end = __pa_symbol(_end);
283 
284 	/*
285 	 * try bottom-up allocation only when bottom-up mode
286 	 * is set and @end is above the kernel image.
287 	 */
288 	if (memblock_bottom_up() && end > kernel_end) {
289 		phys_addr_t bottom_up_start;
290 
291 		/* make sure we will allocate above the kernel */
292 		bottom_up_start = max(start, kernel_end);
293 
294 		/* ok, try bottom-up allocation first */
295 		ret = __memblock_find_range_bottom_up(bottom_up_start, end,
296 						      size, align, nid, flags);
297 		if (ret)
298 			return ret;
299 
300 		/*
301 		 * we always limit bottom-up allocation above the kernel,
302 		 * but top-down allocation doesn't have the limit, so
303 		 * retrying top-down allocation may succeed when bottom-up
304 		 * allocation failed.
305 		 *
306 		 * bottom-up allocation is expected to be fail very rarely,
307 		 * so we use WARN_ONCE() here to see the stack trace if
308 		 * fail happens.
309 		 */
310 		WARN_ONCE(IS_ENABLED(CONFIG_MEMORY_HOTREMOVE),
311 			  "memblock: bottom-up allocation failed, memory hotremove may be affected\n");
312 	}
313 
314 	return __memblock_find_range_top_down(start, end, size, align, nid,
315 					      flags);
316 }
317 
318 /**
319  * memblock_find_in_range - find free area in given range
320  * @start: start of candidate range
321  * @end: end of candidate range, can be %MEMBLOCK_ALLOC_ANYWHERE or
322  *       %MEMBLOCK_ALLOC_ACCESSIBLE
323  * @size: size of free area to find
324  * @align: alignment of free area to find
325  *
326  * Find @size free area aligned to @align in the specified range.
327  *
328  * Return:
329  * Found address on success, 0 on failure.
330  */
331 phys_addr_t __init_memblock memblock_find_in_range(phys_addr_t start,
332 					phys_addr_t end, phys_addr_t size,
333 					phys_addr_t align)
334 {
335 	phys_addr_t ret;
336 	enum memblock_flags flags = choose_memblock_flags();
337 
338 again:
339 	ret = memblock_find_in_range_node(size, align, start, end,
340 					    NUMA_NO_NODE, flags);
341 
342 	if (!ret && (flags & MEMBLOCK_MIRROR)) {
343 		pr_warn("Could not allocate %pap bytes of mirrored memory\n",
344 			&size);
345 		flags &= ~MEMBLOCK_MIRROR;
346 		goto again;
347 	}
348 
349 	return ret;
350 }
351 
352 static void __init_memblock memblock_remove_region(struct memblock_type *type, unsigned long r)
353 {
354 	type->total_size -= type->regions[r].size;
355 	memmove(&type->regions[r], &type->regions[r + 1],
356 		(type->cnt - (r + 1)) * sizeof(type->regions[r]));
357 	type->cnt--;
358 
359 	/* Special case for empty arrays */
360 	if (type->cnt == 0) {
361 		WARN_ON(type->total_size != 0);
362 		type->cnt = 1;
363 		type->regions[0].base = 0;
364 		type->regions[0].size = 0;
365 		type->regions[0].flags = 0;
366 		memblock_set_region_node(&type->regions[0], MAX_NUMNODES);
367 	}
368 }
369 
370 #ifndef CONFIG_ARCH_KEEP_MEMBLOCK
371 /**
372  * memblock_discard - discard memory and reserved arrays if they were allocated
373  */
374 void __init memblock_discard(void)
375 {
376 	phys_addr_t addr, size;
377 
378 	if (memblock.reserved.regions != memblock_reserved_init_regions) {
379 		addr = __pa(memblock.reserved.regions);
380 		size = PAGE_ALIGN(sizeof(struct memblock_region) *
381 				  memblock.reserved.max);
382 		__memblock_free_late(addr, size);
383 	}
384 
385 	if (memblock.memory.regions != memblock_memory_init_regions) {
386 		addr = __pa(memblock.memory.regions);
387 		size = PAGE_ALIGN(sizeof(struct memblock_region) *
388 				  memblock.memory.max);
389 		__memblock_free_late(addr, size);
390 	}
391 }
392 #endif
393 
394 /**
395  * memblock_double_array - double the size of the memblock regions array
396  * @type: memblock type of the regions array being doubled
397  * @new_area_start: starting address of memory range to avoid overlap with
398  * @new_area_size: size of memory range to avoid overlap with
399  *
400  * Double the size of the @type regions array. If memblock is being used to
401  * allocate memory for a new reserved regions array and there is a previously
402  * allocated memory range [@new_area_start, @new_area_start + @new_area_size]
403  * waiting to be reserved, ensure the memory used by the new array does
404  * not overlap.
405  *
406  * Return:
407  * 0 on success, -1 on failure.
408  */
409 static int __init_memblock memblock_double_array(struct memblock_type *type,
410 						phys_addr_t new_area_start,
411 						phys_addr_t new_area_size)
412 {
413 	struct memblock_region *new_array, *old_array;
414 	phys_addr_t old_alloc_size, new_alloc_size;
415 	phys_addr_t old_size, new_size, addr, new_end;
416 	int use_slab = slab_is_available();
417 	int *in_slab;
418 
419 	/* We don't allow resizing until we know about the reserved regions
420 	 * of memory that aren't suitable for allocation
421 	 */
422 	if (!memblock_can_resize)
423 		return -1;
424 
425 	/* Calculate new doubled size */
426 	old_size = type->max * sizeof(struct memblock_region);
427 	new_size = old_size << 1;
428 	/*
429 	 * We need to allocated new one align to PAGE_SIZE,
430 	 *   so we can free them completely later.
431 	 */
432 	old_alloc_size = PAGE_ALIGN(old_size);
433 	new_alloc_size = PAGE_ALIGN(new_size);
434 
435 	/* Retrieve the slab flag */
436 	if (type == &memblock.memory)
437 		in_slab = &memblock_memory_in_slab;
438 	else
439 		in_slab = &memblock_reserved_in_slab;
440 
441 	/* Try to find some space for it */
442 	if (use_slab) {
443 		new_array = kmalloc(new_size, GFP_KERNEL);
444 		addr = new_array ? __pa(new_array) : 0;
445 	} else {
446 		/* only exclude range when trying to double reserved.regions */
447 		if (type != &memblock.reserved)
448 			new_area_start = new_area_size = 0;
449 
450 		addr = memblock_find_in_range(new_area_start + new_area_size,
451 						memblock.current_limit,
452 						new_alloc_size, PAGE_SIZE);
453 		if (!addr && new_area_size)
454 			addr = memblock_find_in_range(0,
455 				min(new_area_start, memblock.current_limit),
456 				new_alloc_size, PAGE_SIZE);
457 
458 		new_array = addr ? __va(addr) : NULL;
459 	}
460 	if (!addr) {
461 		pr_err("memblock: Failed to double %s array from %ld to %ld entries !\n",
462 		       type->name, type->max, type->max * 2);
463 		return -1;
464 	}
465 
466 	new_end = addr + new_size - 1;
467 	memblock_dbg("memblock: %s is doubled to %ld at [%pa-%pa]",
468 			type->name, type->max * 2, &addr, &new_end);
469 
470 	/*
471 	 * Found space, we now need to move the array over before we add the
472 	 * reserved region since it may be our reserved array itself that is
473 	 * full.
474 	 */
475 	memcpy(new_array, type->regions, old_size);
476 	memset(new_array + type->max, 0, old_size);
477 	old_array = type->regions;
478 	type->regions = new_array;
479 	type->max <<= 1;
480 
481 	/* Free old array. We needn't free it if the array is the static one */
482 	if (*in_slab)
483 		kfree(old_array);
484 	else if (old_array != memblock_memory_init_regions &&
485 		 old_array != memblock_reserved_init_regions)
486 		memblock_free(__pa(old_array), old_alloc_size);
487 
488 	/*
489 	 * Reserve the new array if that comes from the memblock.  Otherwise, we
490 	 * needn't do it
491 	 */
492 	if (!use_slab)
493 		BUG_ON(memblock_reserve(addr, new_alloc_size));
494 
495 	/* Update slab flag */
496 	*in_slab = use_slab;
497 
498 	return 0;
499 }
500 
501 /**
502  * memblock_merge_regions - merge neighboring compatible regions
503  * @type: memblock type to scan
504  *
505  * Scan @type and merge neighboring compatible regions.
506  */
507 static void __init_memblock memblock_merge_regions(struct memblock_type *type)
508 {
509 	int i = 0;
510 
511 	/* cnt never goes below 1 */
512 	while (i < type->cnt - 1) {
513 		struct memblock_region *this = &type->regions[i];
514 		struct memblock_region *next = &type->regions[i + 1];
515 
516 		if (this->base + this->size != next->base ||
517 		    memblock_get_region_node(this) !=
518 		    memblock_get_region_node(next) ||
519 		    this->flags != next->flags) {
520 			BUG_ON(this->base + this->size > next->base);
521 			i++;
522 			continue;
523 		}
524 
525 		this->size += next->size;
526 		/* move forward from next + 1, index of which is i + 2 */
527 		memmove(next, next + 1, (type->cnt - (i + 2)) * sizeof(*next));
528 		type->cnt--;
529 	}
530 }
531 
532 /**
533  * memblock_insert_region - insert new memblock region
534  * @type:	memblock type to insert into
535  * @idx:	index for the insertion point
536  * @base:	base address of the new region
537  * @size:	size of the new region
538  * @nid:	node id of the new region
539  * @flags:	flags of the new region
540  *
541  * Insert new memblock region [@base, @base + @size) into @type at @idx.
542  * @type must already have extra room to accommodate the new region.
543  */
544 static void __init_memblock memblock_insert_region(struct memblock_type *type,
545 						   int idx, phys_addr_t base,
546 						   phys_addr_t size,
547 						   int nid,
548 						   enum memblock_flags flags)
549 {
550 	struct memblock_region *rgn = &type->regions[idx];
551 
552 	BUG_ON(type->cnt >= type->max);
553 	memmove(rgn + 1, rgn, (type->cnt - idx) * sizeof(*rgn));
554 	rgn->base = base;
555 	rgn->size = size;
556 	rgn->flags = flags;
557 	memblock_set_region_node(rgn, nid);
558 	type->cnt++;
559 	type->total_size += size;
560 }
561 
562 /**
563  * memblock_add_range - add new memblock region
564  * @type: memblock type to add new region into
565  * @base: base address of the new region
566  * @size: size of the new region
567  * @nid: nid of the new region
568  * @flags: flags of the new region
569  *
570  * Add new memblock region [@base, @base + @size) into @type.  The new region
571  * is allowed to overlap with existing ones - overlaps don't affect already
572  * existing regions.  @type is guaranteed to be minimal (all neighbouring
573  * compatible regions are merged) after the addition.
574  *
575  * Return:
576  * 0 on success, -errno on failure.
577  */
578 static int __init_memblock memblock_add_range(struct memblock_type *type,
579 				phys_addr_t base, phys_addr_t size,
580 				int nid, enum memblock_flags flags)
581 {
582 	bool insert = false;
583 	phys_addr_t obase = base;
584 	phys_addr_t end = base + memblock_cap_size(base, &size);
585 	int idx, nr_new;
586 	struct memblock_region *rgn;
587 
588 	if (!size)
589 		return 0;
590 
591 	/* special case for empty array */
592 	if (type->regions[0].size == 0) {
593 		WARN_ON(type->cnt != 1 || type->total_size);
594 		type->regions[0].base = base;
595 		type->regions[0].size = size;
596 		type->regions[0].flags = flags;
597 		memblock_set_region_node(&type->regions[0], nid);
598 		type->total_size = size;
599 		return 0;
600 	}
601 repeat:
602 	/*
603 	 * The following is executed twice.  Once with %false @insert and
604 	 * then with %true.  The first counts the number of regions needed
605 	 * to accommodate the new area.  The second actually inserts them.
606 	 */
607 	base = obase;
608 	nr_new = 0;
609 
610 	for_each_memblock_type(idx, type, rgn) {
611 		phys_addr_t rbase = rgn->base;
612 		phys_addr_t rend = rbase + rgn->size;
613 
614 		if (rbase >= end)
615 			break;
616 		if (rend <= base)
617 			continue;
618 		/*
619 		 * @rgn overlaps.  If it separates the lower part of new
620 		 * area, insert that portion.
621 		 */
622 		if (rbase > base) {
623 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
624 			WARN_ON(nid != memblock_get_region_node(rgn));
625 #endif
626 			WARN_ON(flags != rgn->flags);
627 			nr_new++;
628 			if (insert)
629 				memblock_insert_region(type, idx++, base,
630 						       rbase - base, nid,
631 						       flags);
632 		}
633 		/* area below @rend is dealt with, forget about it */
634 		base = min(rend, end);
635 	}
636 
637 	/* insert the remaining portion */
638 	if (base < end) {
639 		nr_new++;
640 		if (insert)
641 			memblock_insert_region(type, idx, base, end - base,
642 					       nid, flags);
643 	}
644 
645 	if (!nr_new)
646 		return 0;
647 
648 	/*
649 	 * If this was the first round, resize array and repeat for actual
650 	 * insertions; otherwise, merge and return.
651 	 */
652 	if (!insert) {
653 		while (type->cnt + nr_new > type->max)
654 			if (memblock_double_array(type, obase, size) < 0)
655 				return -ENOMEM;
656 		insert = true;
657 		goto repeat;
658 	} else {
659 		memblock_merge_regions(type);
660 		return 0;
661 	}
662 }
663 
664 /**
665  * memblock_add_node - add new memblock region within a NUMA node
666  * @base: base address of the new region
667  * @size: size of the new region
668  * @nid: nid of the new region
669  *
670  * Add new memblock region [@base, @base + @size) to the "memory"
671  * type. See memblock_add_range() description for mode details
672  *
673  * Return:
674  * 0 on success, -errno on failure.
675  */
676 int __init_memblock memblock_add_node(phys_addr_t base, phys_addr_t size,
677 				       int nid)
678 {
679 	return memblock_add_range(&memblock.memory, base, size, nid, 0);
680 }
681 
682 /**
683  * memblock_add - add new memblock region
684  * @base: base address of the new region
685  * @size: size of the new region
686  *
687  * Add new memblock region [@base, @base + @size) to the "memory"
688  * type. See memblock_add_range() description for mode details
689  *
690  * Return:
691  * 0 on success, -errno on failure.
692  */
693 int __init_memblock memblock_add(phys_addr_t base, phys_addr_t size)
694 {
695 	phys_addr_t end = base + size - 1;
696 
697 	memblock_dbg("%s: [%pa-%pa] %pS\n", __func__,
698 		     &base, &end, (void *)_RET_IP_);
699 
700 	return memblock_add_range(&memblock.memory, base, size, MAX_NUMNODES, 0);
701 }
702 
703 /**
704  * memblock_isolate_range - isolate given range into disjoint memblocks
705  * @type: memblock type to isolate range for
706  * @base: base of range to isolate
707  * @size: size of range to isolate
708  * @start_rgn: out parameter for the start of isolated region
709  * @end_rgn: out parameter for the end of isolated region
710  *
711  * Walk @type and ensure that regions don't cross the boundaries defined by
712  * [@base, @base + @size).  Crossing regions are split at the boundaries,
713  * which may create at most two more regions.  The index of the first
714  * region inside the range is returned in *@start_rgn and end in *@end_rgn.
715  *
716  * Return:
717  * 0 on success, -errno on failure.
718  */
719 static int __init_memblock memblock_isolate_range(struct memblock_type *type,
720 					phys_addr_t base, phys_addr_t size,
721 					int *start_rgn, int *end_rgn)
722 {
723 	phys_addr_t end = base + memblock_cap_size(base, &size);
724 	int idx;
725 	struct memblock_region *rgn;
726 
727 	*start_rgn = *end_rgn = 0;
728 
729 	if (!size)
730 		return 0;
731 
732 	/* we'll create at most two more regions */
733 	while (type->cnt + 2 > type->max)
734 		if (memblock_double_array(type, base, size) < 0)
735 			return -ENOMEM;
736 
737 	for_each_memblock_type(idx, type, rgn) {
738 		phys_addr_t rbase = rgn->base;
739 		phys_addr_t rend = rbase + rgn->size;
740 
741 		if (rbase >= end)
742 			break;
743 		if (rend <= base)
744 			continue;
745 
746 		if (rbase < base) {
747 			/*
748 			 * @rgn intersects from below.  Split and continue
749 			 * to process the next region - the new top half.
750 			 */
751 			rgn->base = base;
752 			rgn->size -= base - rbase;
753 			type->total_size -= base - rbase;
754 			memblock_insert_region(type, idx, rbase, base - rbase,
755 					       memblock_get_region_node(rgn),
756 					       rgn->flags);
757 		} else if (rend > end) {
758 			/*
759 			 * @rgn intersects from above.  Split and redo the
760 			 * current region - the new bottom half.
761 			 */
762 			rgn->base = end;
763 			rgn->size -= end - rbase;
764 			type->total_size -= end - rbase;
765 			memblock_insert_region(type, idx--, rbase, end - rbase,
766 					       memblock_get_region_node(rgn),
767 					       rgn->flags);
768 		} else {
769 			/* @rgn is fully contained, record it */
770 			if (!*end_rgn)
771 				*start_rgn = idx;
772 			*end_rgn = idx + 1;
773 		}
774 	}
775 
776 	return 0;
777 }
778 
779 static int __init_memblock memblock_remove_range(struct memblock_type *type,
780 					  phys_addr_t base, phys_addr_t size)
781 {
782 	int start_rgn, end_rgn;
783 	int i, ret;
784 
785 	ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
786 	if (ret)
787 		return ret;
788 
789 	for (i = end_rgn - 1; i >= start_rgn; i--)
790 		memblock_remove_region(type, i);
791 	return 0;
792 }
793 
794 int __init_memblock memblock_remove(phys_addr_t base, phys_addr_t size)
795 {
796 	phys_addr_t end = base + size - 1;
797 
798 	memblock_dbg("%s: [%pa-%pa] %pS\n", __func__,
799 		     &base, &end, (void *)_RET_IP_);
800 
801 	return memblock_remove_range(&memblock.memory, base, size);
802 }
803 
804 /**
805  * memblock_free - free boot memory block
806  * @base: phys starting address of the  boot memory block
807  * @size: size of the boot memory block in bytes
808  *
809  * Free boot memory block previously allocated by memblock_alloc_xx() API.
810  * The freeing memory will not be released to the buddy allocator.
811  */
812 int __init_memblock memblock_free(phys_addr_t base, phys_addr_t size)
813 {
814 	phys_addr_t end = base + size - 1;
815 
816 	memblock_dbg("%s: [%pa-%pa] %pS\n", __func__,
817 		     &base, &end, (void *)_RET_IP_);
818 
819 	kmemleak_free_part_phys(base, size);
820 	return memblock_remove_range(&memblock.reserved, base, size);
821 }
822 
823 int __init_memblock memblock_reserve(phys_addr_t base, phys_addr_t size)
824 {
825 	phys_addr_t end = base + size - 1;
826 
827 	memblock_dbg("%s: [%pa-%pa] %pS\n", __func__,
828 		     &base, &end, (void *)_RET_IP_);
829 
830 	return memblock_add_range(&memblock.reserved, base, size, MAX_NUMNODES, 0);
831 }
832 
833 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
834 int __init_memblock memblock_physmem_add(phys_addr_t base, phys_addr_t size)
835 {
836 	phys_addr_t end = base + size - 1;
837 
838 	memblock_dbg("%s: [%pa-%pa] %pS\n", __func__,
839 		     &base, &end, (void *)_RET_IP_);
840 
841 	return memblock_add_range(&memblock.physmem, base, size, MAX_NUMNODES, 0);
842 }
843 #endif
844 
845 /**
846  * memblock_setclr_flag - set or clear flag for a memory region
847  * @base: base address of the region
848  * @size: size of the region
849  * @set: set or clear the flag
850  * @flag: the flag to udpate
851  *
852  * This function isolates region [@base, @base + @size), and sets/clears flag
853  *
854  * Return: 0 on success, -errno on failure.
855  */
856 static int __init_memblock memblock_setclr_flag(phys_addr_t base,
857 				phys_addr_t size, int set, int flag)
858 {
859 	struct memblock_type *type = &memblock.memory;
860 	int i, ret, start_rgn, end_rgn;
861 
862 	ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
863 	if (ret)
864 		return ret;
865 
866 	for (i = start_rgn; i < end_rgn; i++) {
867 		struct memblock_region *r = &type->regions[i];
868 
869 		if (set)
870 			r->flags |= flag;
871 		else
872 			r->flags &= ~flag;
873 	}
874 
875 	memblock_merge_regions(type);
876 	return 0;
877 }
878 
879 /**
880  * memblock_mark_hotplug - Mark hotpluggable memory with flag MEMBLOCK_HOTPLUG.
881  * @base: the base phys addr of the region
882  * @size: the size of the region
883  *
884  * Return: 0 on success, -errno on failure.
885  */
886 int __init_memblock memblock_mark_hotplug(phys_addr_t base, phys_addr_t size)
887 {
888 	return memblock_setclr_flag(base, size, 1, MEMBLOCK_HOTPLUG);
889 }
890 
891 /**
892  * memblock_clear_hotplug - Clear flag MEMBLOCK_HOTPLUG for a specified region.
893  * @base: the base phys addr of the region
894  * @size: the size of the region
895  *
896  * Return: 0 on success, -errno on failure.
897  */
898 int __init_memblock memblock_clear_hotplug(phys_addr_t base, phys_addr_t size)
899 {
900 	return memblock_setclr_flag(base, size, 0, MEMBLOCK_HOTPLUG);
901 }
902 
903 /**
904  * memblock_mark_mirror - Mark mirrored memory with flag MEMBLOCK_MIRROR.
905  * @base: the base phys addr of the region
906  * @size: the size of the region
907  *
908  * Return: 0 on success, -errno on failure.
909  */
910 int __init_memblock memblock_mark_mirror(phys_addr_t base, phys_addr_t size)
911 {
912 	system_has_some_mirror = true;
913 
914 	return memblock_setclr_flag(base, size, 1, MEMBLOCK_MIRROR);
915 }
916 
917 /**
918  * memblock_mark_nomap - Mark a memory region with flag MEMBLOCK_NOMAP.
919  * @base: the base phys addr of the region
920  * @size: the size of the region
921  *
922  * Return: 0 on success, -errno on failure.
923  */
924 int __init_memblock memblock_mark_nomap(phys_addr_t base, phys_addr_t size)
925 {
926 	return memblock_setclr_flag(base, size, 1, MEMBLOCK_NOMAP);
927 }
928 
929 /**
930  * memblock_clear_nomap - Clear flag MEMBLOCK_NOMAP for a specified region.
931  * @base: the base phys addr of the region
932  * @size: the size of the region
933  *
934  * Return: 0 on success, -errno on failure.
935  */
936 int __init_memblock memblock_clear_nomap(phys_addr_t base, phys_addr_t size)
937 {
938 	return memblock_setclr_flag(base, size, 0, MEMBLOCK_NOMAP);
939 }
940 
941 /**
942  * __next_reserved_mem_region - next function for for_each_reserved_region()
943  * @idx: pointer to u64 loop variable
944  * @out_start: ptr to phys_addr_t for start address of the region, can be %NULL
945  * @out_end: ptr to phys_addr_t for end address of the region, can be %NULL
946  *
947  * Iterate over all reserved memory regions.
948  */
949 void __init_memblock __next_reserved_mem_region(u64 *idx,
950 					   phys_addr_t *out_start,
951 					   phys_addr_t *out_end)
952 {
953 	struct memblock_type *type = &memblock.reserved;
954 
955 	if (*idx < type->cnt) {
956 		struct memblock_region *r = &type->regions[*idx];
957 		phys_addr_t base = r->base;
958 		phys_addr_t size = r->size;
959 
960 		if (out_start)
961 			*out_start = base;
962 		if (out_end)
963 			*out_end = base + size - 1;
964 
965 		*idx += 1;
966 		return;
967 	}
968 
969 	/* signal end of iteration */
970 	*idx = ULLONG_MAX;
971 }
972 
973 static bool should_skip_region(struct memblock_region *m, int nid, int flags)
974 {
975 	int m_nid = memblock_get_region_node(m);
976 
977 	/* only memory regions are associated with nodes, check it */
978 	if (nid != NUMA_NO_NODE && nid != m_nid)
979 		return true;
980 
981 	/* skip hotpluggable memory regions if needed */
982 	if (movable_node_is_enabled() && memblock_is_hotpluggable(m))
983 		return true;
984 
985 	/* if we want mirror memory skip non-mirror memory regions */
986 	if ((flags & MEMBLOCK_MIRROR) && !memblock_is_mirror(m))
987 		return true;
988 
989 	/* skip nomap memory unless we were asked for it explicitly */
990 	if (!(flags & MEMBLOCK_NOMAP) && memblock_is_nomap(m))
991 		return true;
992 
993 	return false;
994 }
995 
996 /**
997  * __next_mem_range - next function for for_each_free_mem_range() etc.
998  * @idx: pointer to u64 loop variable
999  * @nid: node selector, %NUMA_NO_NODE for all nodes
1000  * @flags: pick from blocks based on memory attributes
1001  * @type_a: pointer to memblock_type from where the range is taken
1002  * @type_b: pointer to memblock_type which excludes memory from being taken
1003  * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL
1004  * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL
1005  * @out_nid: ptr to int for nid of the range, can be %NULL
1006  *
1007  * Find the first area from *@idx which matches @nid, fill the out
1008  * parameters, and update *@idx for the next iteration.  The lower 32bit of
1009  * *@idx contains index into type_a and the upper 32bit indexes the
1010  * areas before each region in type_b.	For example, if type_b regions
1011  * look like the following,
1012  *
1013  *	0:[0-16), 1:[32-48), 2:[128-130)
1014  *
1015  * The upper 32bit indexes the following regions.
1016  *
1017  *	0:[0-0), 1:[16-32), 2:[48-128), 3:[130-MAX)
1018  *
1019  * As both region arrays are sorted, the function advances the two indices
1020  * in lockstep and returns each intersection.
1021  */
1022 void __init_memblock __next_mem_range(u64 *idx, int nid,
1023 				      enum memblock_flags flags,
1024 				      struct memblock_type *type_a,
1025 				      struct memblock_type *type_b,
1026 				      phys_addr_t *out_start,
1027 				      phys_addr_t *out_end, int *out_nid)
1028 {
1029 	int idx_a = *idx & 0xffffffff;
1030 	int idx_b = *idx >> 32;
1031 
1032 	if (WARN_ONCE(nid == MAX_NUMNODES,
1033 	"Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
1034 		nid = NUMA_NO_NODE;
1035 
1036 	for (; idx_a < type_a->cnt; idx_a++) {
1037 		struct memblock_region *m = &type_a->regions[idx_a];
1038 
1039 		phys_addr_t m_start = m->base;
1040 		phys_addr_t m_end = m->base + m->size;
1041 		int	    m_nid = memblock_get_region_node(m);
1042 
1043 		if (should_skip_region(m, nid, flags))
1044 			continue;
1045 
1046 		if (!type_b) {
1047 			if (out_start)
1048 				*out_start = m_start;
1049 			if (out_end)
1050 				*out_end = m_end;
1051 			if (out_nid)
1052 				*out_nid = m_nid;
1053 			idx_a++;
1054 			*idx = (u32)idx_a | (u64)idx_b << 32;
1055 			return;
1056 		}
1057 
1058 		/* scan areas before each reservation */
1059 		for (; idx_b < type_b->cnt + 1; idx_b++) {
1060 			struct memblock_region *r;
1061 			phys_addr_t r_start;
1062 			phys_addr_t r_end;
1063 
1064 			r = &type_b->regions[idx_b];
1065 			r_start = idx_b ? r[-1].base + r[-1].size : 0;
1066 			r_end = idx_b < type_b->cnt ?
1067 				r->base : PHYS_ADDR_MAX;
1068 
1069 			/*
1070 			 * if idx_b advanced past idx_a,
1071 			 * break out to advance idx_a
1072 			 */
1073 			if (r_start >= m_end)
1074 				break;
1075 			/* if the two regions intersect, we're done */
1076 			if (m_start < r_end) {
1077 				if (out_start)
1078 					*out_start =
1079 						max(m_start, r_start);
1080 				if (out_end)
1081 					*out_end = min(m_end, r_end);
1082 				if (out_nid)
1083 					*out_nid = m_nid;
1084 				/*
1085 				 * The region which ends first is
1086 				 * advanced for the next iteration.
1087 				 */
1088 				if (m_end <= r_end)
1089 					idx_a++;
1090 				else
1091 					idx_b++;
1092 				*idx = (u32)idx_a | (u64)idx_b << 32;
1093 				return;
1094 			}
1095 		}
1096 	}
1097 
1098 	/* signal end of iteration */
1099 	*idx = ULLONG_MAX;
1100 }
1101 
1102 /**
1103  * __next_mem_range_rev - generic next function for for_each_*_range_rev()
1104  *
1105  * @idx: pointer to u64 loop variable
1106  * @nid: node selector, %NUMA_NO_NODE for all nodes
1107  * @flags: pick from blocks based on memory attributes
1108  * @type_a: pointer to memblock_type from where the range is taken
1109  * @type_b: pointer to memblock_type which excludes memory from being taken
1110  * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL
1111  * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL
1112  * @out_nid: ptr to int for nid of the range, can be %NULL
1113  *
1114  * Finds the next range from type_a which is not marked as unsuitable
1115  * in type_b.
1116  *
1117  * Reverse of __next_mem_range().
1118  */
1119 void __init_memblock __next_mem_range_rev(u64 *idx, int nid,
1120 					  enum memblock_flags flags,
1121 					  struct memblock_type *type_a,
1122 					  struct memblock_type *type_b,
1123 					  phys_addr_t *out_start,
1124 					  phys_addr_t *out_end, int *out_nid)
1125 {
1126 	int idx_a = *idx & 0xffffffff;
1127 	int idx_b = *idx >> 32;
1128 
1129 	if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
1130 		nid = NUMA_NO_NODE;
1131 
1132 	if (*idx == (u64)ULLONG_MAX) {
1133 		idx_a = type_a->cnt - 1;
1134 		if (type_b != NULL)
1135 			idx_b = type_b->cnt;
1136 		else
1137 			idx_b = 0;
1138 	}
1139 
1140 	for (; idx_a >= 0; idx_a--) {
1141 		struct memblock_region *m = &type_a->regions[idx_a];
1142 
1143 		phys_addr_t m_start = m->base;
1144 		phys_addr_t m_end = m->base + m->size;
1145 		int m_nid = memblock_get_region_node(m);
1146 
1147 		if (should_skip_region(m, nid, flags))
1148 			continue;
1149 
1150 		if (!type_b) {
1151 			if (out_start)
1152 				*out_start = m_start;
1153 			if (out_end)
1154 				*out_end = m_end;
1155 			if (out_nid)
1156 				*out_nid = m_nid;
1157 			idx_a--;
1158 			*idx = (u32)idx_a | (u64)idx_b << 32;
1159 			return;
1160 		}
1161 
1162 		/* scan areas before each reservation */
1163 		for (; idx_b >= 0; idx_b--) {
1164 			struct memblock_region *r;
1165 			phys_addr_t r_start;
1166 			phys_addr_t r_end;
1167 
1168 			r = &type_b->regions[idx_b];
1169 			r_start = idx_b ? r[-1].base + r[-1].size : 0;
1170 			r_end = idx_b < type_b->cnt ?
1171 				r->base : PHYS_ADDR_MAX;
1172 			/*
1173 			 * if idx_b advanced past idx_a,
1174 			 * break out to advance idx_a
1175 			 */
1176 
1177 			if (r_end <= m_start)
1178 				break;
1179 			/* if the two regions intersect, we're done */
1180 			if (m_end > r_start) {
1181 				if (out_start)
1182 					*out_start = max(m_start, r_start);
1183 				if (out_end)
1184 					*out_end = min(m_end, r_end);
1185 				if (out_nid)
1186 					*out_nid = m_nid;
1187 				if (m_start >= r_start)
1188 					idx_a--;
1189 				else
1190 					idx_b--;
1191 				*idx = (u32)idx_a | (u64)idx_b << 32;
1192 				return;
1193 			}
1194 		}
1195 	}
1196 	/* signal end of iteration */
1197 	*idx = ULLONG_MAX;
1198 }
1199 
1200 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
1201 /*
1202  * Common iterator interface used to define for_each_mem_pfn_range().
1203  */
1204 void __init_memblock __next_mem_pfn_range(int *idx, int nid,
1205 				unsigned long *out_start_pfn,
1206 				unsigned long *out_end_pfn, int *out_nid)
1207 {
1208 	struct memblock_type *type = &memblock.memory;
1209 	struct memblock_region *r;
1210 
1211 	while (++*idx < type->cnt) {
1212 		r = &type->regions[*idx];
1213 
1214 		if (PFN_UP(r->base) >= PFN_DOWN(r->base + r->size))
1215 			continue;
1216 		if (nid == MAX_NUMNODES || nid == r->nid)
1217 			break;
1218 	}
1219 	if (*idx >= type->cnt) {
1220 		*idx = -1;
1221 		return;
1222 	}
1223 
1224 	if (out_start_pfn)
1225 		*out_start_pfn = PFN_UP(r->base);
1226 	if (out_end_pfn)
1227 		*out_end_pfn = PFN_DOWN(r->base + r->size);
1228 	if (out_nid)
1229 		*out_nid = r->nid;
1230 }
1231 
1232 /**
1233  * memblock_set_node - set node ID on memblock regions
1234  * @base: base of area to set node ID for
1235  * @size: size of area to set node ID for
1236  * @type: memblock type to set node ID for
1237  * @nid: node ID to set
1238  *
1239  * Set the nid of memblock @type regions in [@base, @base + @size) to @nid.
1240  * Regions which cross the area boundaries are split as necessary.
1241  *
1242  * Return:
1243  * 0 on success, -errno on failure.
1244  */
1245 int __init_memblock memblock_set_node(phys_addr_t base, phys_addr_t size,
1246 				      struct memblock_type *type, int nid)
1247 {
1248 	int start_rgn, end_rgn;
1249 	int i, ret;
1250 
1251 	ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
1252 	if (ret)
1253 		return ret;
1254 
1255 	for (i = start_rgn; i < end_rgn; i++)
1256 		memblock_set_region_node(&type->regions[i], nid);
1257 
1258 	memblock_merge_regions(type);
1259 	return 0;
1260 }
1261 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
1262 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1263 /**
1264  * __next_mem_pfn_range_in_zone - iterator for for_each_*_range_in_zone()
1265  *
1266  * @idx: pointer to u64 loop variable
1267  * @zone: zone in which all of the memory blocks reside
1268  * @out_spfn: ptr to ulong for start pfn of the range, can be %NULL
1269  * @out_epfn: ptr to ulong for end pfn of the range, can be %NULL
1270  *
1271  * This function is meant to be a zone/pfn specific wrapper for the
1272  * for_each_mem_range type iterators. Specifically they are used in the
1273  * deferred memory init routines and as such we were duplicating much of
1274  * this logic throughout the code. So instead of having it in multiple
1275  * locations it seemed like it would make more sense to centralize this to
1276  * one new iterator that does everything they need.
1277  */
1278 void __init_memblock
1279 __next_mem_pfn_range_in_zone(u64 *idx, struct zone *zone,
1280 			     unsigned long *out_spfn, unsigned long *out_epfn)
1281 {
1282 	int zone_nid = zone_to_nid(zone);
1283 	phys_addr_t spa, epa;
1284 	int nid;
1285 
1286 	__next_mem_range(idx, zone_nid, MEMBLOCK_NONE,
1287 			 &memblock.memory, &memblock.reserved,
1288 			 &spa, &epa, &nid);
1289 
1290 	while (*idx != U64_MAX) {
1291 		unsigned long epfn = PFN_DOWN(epa);
1292 		unsigned long spfn = PFN_UP(spa);
1293 
1294 		/*
1295 		 * Verify the end is at least past the start of the zone and
1296 		 * that we have at least one PFN to initialize.
1297 		 */
1298 		if (zone->zone_start_pfn < epfn && spfn < epfn) {
1299 			/* if we went too far just stop searching */
1300 			if (zone_end_pfn(zone) <= spfn) {
1301 				*idx = U64_MAX;
1302 				break;
1303 			}
1304 
1305 			if (out_spfn)
1306 				*out_spfn = max(zone->zone_start_pfn, spfn);
1307 			if (out_epfn)
1308 				*out_epfn = min(zone_end_pfn(zone), epfn);
1309 
1310 			return;
1311 		}
1312 
1313 		__next_mem_range(idx, zone_nid, MEMBLOCK_NONE,
1314 				 &memblock.memory, &memblock.reserved,
1315 				 &spa, &epa, &nid);
1316 	}
1317 
1318 	/* signal end of iteration */
1319 	if (out_spfn)
1320 		*out_spfn = ULONG_MAX;
1321 	if (out_epfn)
1322 		*out_epfn = 0;
1323 }
1324 
1325 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
1326 
1327 /**
1328  * memblock_alloc_range_nid - allocate boot memory block
1329  * @size: size of memory block to be allocated in bytes
1330  * @align: alignment of the region and block's size
1331  * @start: the lower bound of the memory region to allocate (phys address)
1332  * @end: the upper bound of the memory region to allocate (phys address)
1333  * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1334  * @exact_nid: control the allocation fall back to other nodes
1335  *
1336  * The allocation is performed from memory region limited by
1337  * memblock.current_limit if @end == %MEMBLOCK_ALLOC_ACCESSIBLE.
1338  *
1339  * If the specified node can not hold the requested memory and @exact_nid
1340  * is false, the allocation falls back to any node in the system.
1341  *
1342  * For systems with memory mirroring, the allocation is attempted first
1343  * from the regions with mirroring enabled and then retried from any
1344  * memory region.
1345  *
1346  * In addition, function sets the min_count to 0 using kmemleak_alloc_phys for
1347  * allocated boot memory block, so that it is never reported as leaks.
1348  *
1349  * Return:
1350  * Physical address of allocated memory block on success, %0 on failure.
1351  */
1352 phys_addr_t __init memblock_alloc_range_nid(phys_addr_t size,
1353 					phys_addr_t align, phys_addr_t start,
1354 					phys_addr_t end, int nid,
1355 					bool exact_nid)
1356 {
1357 	enum memblock_flags flags = choose_memblock_flags();
1358 	phys_addr_t found;
1359 
1360 	if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
1361 		nid = NUMA_NO_NODE;
1362 
1363 	if (!align) {
1364 		/* Can't use WARNs this early in boot on powerpc */
1365 		dump_stack();
1366 		align = SMP_CACHE_BYTES;
1367 	}
1368 
1369 again:
1370 	found = memblock_find_in_range_node(size, align, start, end, nid,
1371 					    flags);
1372 	if (found && !memblock_reserve(found, size))
1373 		goto done;
1374 
1375 	if (nid != NUMA_NO_NODE && !exact_nid) {
1376 		found = memblock_find_in_range_node(size, align, start,
1377 						    end, NUMA_NO_NODE,
1378 						    flags);
1379 		if (found && !memblock_reserve(found, size))
1380 			goto done;
1381 	}
1382 
1383 	if (flags & MEMBLOCK_MIRROR) {
1384 		flags &= ~MEMBLOCK_MIRROR;
1385 		pr_warn("Could not allocate %pap bytes of mirrored memory\n",
1386 			&size);
1387 		goto again;
1388 	}
1389 
1390 	return 0;
1391 
1392 done:
1393 	/* Skip kmemleak for kasan_init() due to high volume. */
1394 	if (end != MEMBLOCK_ALLOC_KASAN)
1395 		/*
1396 		 * The min_count is set to 0 so that memblock allocated
1397 		 * blocks are never reported as leaks. This is because many
1398 		 * of these blocks are only referred via the physical
1399 		 * address which is not looked up by kmemleak.
1400 		 */
1401 		kmemleak_alloc_phys(found, size, 0, 0);
1402 
1403 	return found;
1404 }
1405 
1406 /**
1407  * memblock_phys_alloc_range - allocate a memory block inside specified range
1408  * @size: size of memory block to be allocated in bytes
1409  * @align: alignment of the region and block's size
1410  * @start: the lower bound of the memory region to allocate (physical address)
1411  * @end: the upper bound of the memory region to allocate (physical address)
1412  *
1413  * Allocate @size bytes in the between @start and @end.
1414  *
1415  * Return: physical address of the allocated memory block on success,
1416  * %0 on failure.
1417  */
1418 phys_addr_t __init memblock_phys_alloc_range(phys_addr_t size,
1419 					     phys_addr_t align,
1420 					     phys_addr_t start,
1421 					     phys_addr_t end)
1422 {
1423 	return memblock_alloc_range_nid(size, align, start, end, NUMA_NO_NODE,
1424 					false);
1425 }
1426 
1427 /**
1428  * memblock_phys_alloc_try_nid - allocate a memory block from specified MUMA node
1429  * @size: size of memory block to be allocated in bytes
1430  * @align: alignment of the region and block's size
1431  * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1432  *
1433  * Allocates memory block from the specified NUMA node. If the node
1434  * has no available memory, attempts to allocated from any node in the
1435  * system.
1436  *
1437  * Return: physical address of the allocated memory block on success,
1438  * %0 on failure.
1439  */
1440 phys_addr_t __init memblock_phys_alloc_try_nid(phys_addr_t size, phys_addr_t align, int nid)
1441 {
1442 	return memblock_alloc_range_nid(size, align, 0,
1443 					MEMBLOCK_ALLOC_ACCESSIBLE, nid, false);
1444 }
1445 
1446 /**
1447  * memblock_alloc_internal - allocate boot memory block
1448  * @size: size of memory block to be allocated in bytes
1449  * @align: alignment of the region and block's size
1450  * @min_addr: the lower bound of the memory region to allocate (phys address)
1451  * @max_addr: the upper bound of the memory region to allocate (phys address)
1452  * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1453  * @exact_nid: control the allocation fall back to other nodes
1454  *
1455  * Allocates memory block using memblock_alloc_range_nid() and
1456  * converts the returned physical address to virtual.
1457  *
1458  * The @min_addr limit is dropped if it can not be satisfied and the allocation
1459  * will fall back to memory below @min_addr. Other constraints, such
1460  * as node and mirrored memory will be handled again in
1461  * memblock_alloc_range_nid().
1462  *
1463  * Return:
1464  * Virtual address of allocated memory block on success, NULL on failure.
1465  */
1466 static void * __init memblock_alloc_internal(
1467 				phys_addr_t size, phys_addr_t align,
1468 				phys_addr_t min_addr, phys_addr_t max_addr,
1469 				int nid, bool exact_nid)
1470 {
1471 	phys_addr_t alloc;
1472 
1473 	/*
1474 	 * Detect any accidental use of these APIs after slab is ready, as at
1475 	 * this moment memblock may be deinitialized already and its
1476 	 * internal data may be destroyed (after execution of memblock_free_all)
1477 	 */
1478 	if (WARN_ON_ONCE(slab_is_available()))
1479 		return kzalloc_node(size, GFP_NOWAIT, nid);
1480 
1481 	if (max_addr > memblock.current_limit)
1482 		max_addr = memblock.current_limit;
1483 
1484 	alloc = memblock_alloc_range_nid(size, align, min_addr, max_addr, nid,
1485 					exact_nid);
1486 
1487 	/* retry allocation without lower limit */
1488 	if (!alloc && min_addr)
1489 		alloc = memblock_alloc_range_nid(size, align, 0, max_addr, nid,
1490 						exact_nid);
1491 
1492 	if (!alloc)
1493 		return NULL;
1494 
1495 	return phys_to_virt(alloc);
1496 }
1497 
1498 /**
1499  * memblock_alloc_exact_nid_raw - allocate boot memory block on the exact node
1500  * without zeroing memory
1501  * @size: size of memory block to be allocated in bytes
1502  * @align: alignment of the region and block's size
1503  * @min_addr: the lower bound of the memory region from where the allocation
1504  *	  is preferred (phys address)
1505  * @max_addr: the upper bound of the memory region from where the allocation
1506  *	      is preferred (phys address), or %MEMBLOCK_ALLOC_ACCESSIBLE to
1507  *	      allocate only from memory limited by memblock.current_limit value
1508  * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1509  *
1510  * Public function, provides additional debug information (including caller
1511  * info), if enabled. Does not zero allocated memory.
1512  *
1513  * Return:
1514  * Virtual address of allocated memory block on success, NULL on failure.
1515  */
1516 void * __init memblock_alloc_exact_nid_raw(
1517 			phys_addr_t size, phys_addr_t align,
1518 			phys_addr_t min_addr, phys_addr_t max_addr,
1519 			int nid)
1520 {
1521 	void *ptr;
1522 
1523 	memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=%pa max_addr=%pa %pS\n",
1524 		     __func__, (u64)size, (u64)align, nid, &min_addr,
1525 		     &max_addr, (void *)_RET_IP_);
1526 
1527 	ptr = memblock_alloc_internal(size, align,
1528 					   min_addr, max_addr, nid, true);
1529 	if (ptr && size > 0)
1530 		page_init_poison(ptr, size);
1531 
1532 	return ptr;
1533 }
1534 
1535 /**
1536  * memblock_alloc_try_nid_raw - allocate boot memory block without zeroing
1537  * memory and without panicking
1538  * @size: size of memory block to be allocated in bytes
1539  * @align: alignment of the region and block's size
1540  * @min_addr: the lower bound of the memory region from where the allocation
1541  *	  is preferred (phys address)
1542  * @max_addr: the upper bound of the memory region from where the allocation
1543  *	      is preferred (phys address), or %MEMBLOCK_ALLOC_ACCESSIBLE to
1544  *	      allocate only from memory limited by memblock.current_limit value
1545  * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1546  *
1547  * Public function, provides additional debug information (including caller
1548  * info), if enabled. Does not zero allocated memory, does not panic if request
1549  * cannot be satisfied.
1550  *
1551  * Return:
1552  * Virtual address of allocated memory block on success, NULL on failure.
1553  */
1554 void * __init memblock_alloc_try_nid_raw(
1555 			phys_addr_t size, phys_addr_t align,
1556 			phys_addr_t min_addr, phys_addr_t max_addr,
1557 			int nid)
1558 {
1559 	void *ptr;
1560 
1561 	memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=%pa max_addr=%pa %pS\n",
1562 		     __func__, (u64)size, (u64)align, nid, &min_addr,
1563 		     &max_addr, (void *)_RET_IP_);
1564 
1565 	ptr = memblock_alloc_internal(size, align,
1566 					   min_addr, max_addr, nid, false);
1567 	if (ptr && size > 0)
1568 		page_init_poison(ptr, size);
1569 
1570 	return ptr;
1571 }
1572 
1573 /**
1574  * memblock_alloc_try_nid - allocate boot memory block
1575  * @size: size of memory block to be allocated in bytes
1576  * @align: alignment of the region and block's size
1577  * @min_addr: the lower bound of the memory region from where the allocation
1578  *	  is preferred (phys address)
1579  * @max_addr: the upper bound of the memory region from where the allocation
1580  *	      is preferred (phys address), or %MEMBLOCK_ALLOC_ACCESSIBLE to
1581  *	      allocate only from memory limited by memblock.current_limit value
1582  * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1583  *
1584  * Public function, provides additional debug information (including caller
1585  * info), if enabled. This function zeroes the allocated memory.
1586  *
1587  * Return:
1588  * Virtual address of allocated memory block on success, NULL on failure.
1589  */
1590 void * __init memblock_alloc_try_nid(
1591 			phys_addr_t size, phys_addr_t align,
1592 			phys_addr_t min_addr, phys_addr_t max_addr,
1593 			int nid)
1594 {
1595 	void *ptr;
1596 
1597 	memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=%pa max_addr=%pa %pS\n",
1598 		     __func__, (u64)size, (u64)align, nid, &min_addr,
1599 		     &max_addr, (void *)_RET_IP_);
1600 	ptr = memblock_alloc_internal(size, align,
1601 					   min_addr, max_addr, nid, false);
1602 	if (ptr)
1603 		memset(ptr, 0, size);
1604 
1605 	return ptr;
1606 }
1607 
1608 /**
1609  * __memblock_free_late - free pages directly to buddy allocator
1610  * @base: phys starting address of the  boot memory block
1611  * @size: size of the boot memory block in bytes
1612  *
1613  * This is only useful when the memblock allocator has already been torn
1614  * down, but we are still initializing the system.  Pages are released directly
1615  * to the buddy allocator.
1616  */
1617 void __init __memblock_free_late(phys_addr_t base, phys_addr_t size)
1618 {
1619 	phys_addr_t cursor, end;
1620 
1621 	end = base + size - 1;
1622 	memblock_dbg("%s: [%pa-%pa] %pS\n",
1623 		     __func__, &base, &end, (void *)_RET_IP_);
1624 	kmemleak_free_part_phys(base, size);
1625 	cursor = PFN_UP(base);
1626 	end = PFN_DOWN(base + size);
1627 
1628 	for (; cursor < end; cursor++) {
1629 		memblock_free_pages(pfn_to_page(cursor), cursor, 0);
1630 		totalram_pages_inc();
1631 	}
1632 }
1633 
1634 /*
1635  * Remaining API functions
1636  */
1637 
1638 phys_addr_t __init_memblock memblock_phys_mem_size(void)
1639 {
1640 	return memblock.memory.total_size;
1641 }
1642 
1643 phys_addr_t __init_memblock memblock_reserved_size(void)
1644 {
1645 	return memblock.reserved.total_size;
1646 }
1647 
1648 phys_addr_t __init memblock_mem_size(unsigned long limit_pfn)
1649 {
1650 	unsigned long pages = 0;
1651 	struct memblock_region *r;
1652 	unsigned long start_pfn, end_pfn;
1653 
1654 	for_each_memblock(memory, r) {
1655 		start_pfn = memblock_region_memory_base_pfn(r);
1656 		end_pfn = memblock_region_memory_end_pfn(r);
1657 		start_pfn = min_t(unsigned long, start_pfn, limit_pfn);
1658 		end_pfn = min_t(unsigned long, end_pfn, limit_pfn);
1659 		pages += end_pfn - start_pfn;
1660 	}
1661 
1662 	return PFN_PHYS(pages);
1663 }
1664 
1665 /* lowest address */
1666 phys_addr_t __init_memblock memblock_start_of_DRAM(void)
1667 {
1668 	return memblock.memory.regions[0].base;
1669 }
1670 
1671 phys_addr_t __init_memblock memblock_end_of_DRAM(void)
1672 {
1673 	int idx = memblock.memory.cnt - 1;
1674 
1675 	return (memblock.memory.regions[idx].base + memblock.memory.regions[idx].size);
1676 }
1677 
1678 static phys_addr_t __init_memblock __find_max_addr(phys_addr_t limit)
1679 {
1680 	phys_addr_t max_addr = PHYS_ADDR_MAX;
1681 	struct memblock_region *r;
1682 
1683 	/*
1684 	 * translate the memory @limit size into the max address within one of
1685 	 * the memory memblock regions, if the @limit exceeds the total size
1686 	 * of those regions, max_addr will keep original value PHYS_ADDR_MAX
1687 	 */
1688 	for_each_memblock(memory, r) {
1689 		if (limit <= r->size) {
1690 			max_addr = r->base + limit;
1691 			break;
1692 		}
1693 		limit -= r->size;
1694 	}
1695 
1696 	return max_addr;
1697 }
1698 
1699 void __init memblock_enforce_memory_limit(phys_addr_t limit)
1700 {
1701 	phys_addr_t max_addr;
1702 
1703 	if (!limit)
1704 		return;
1705 
1706 	max_addr = __find_max_addr(limit);
1707 
1708 	/* @limit exceeds the total size of the memory, do nothing */
1709 	if (max_addr == PHYS_ADDR_MAX)
1710 		return;
1711 
1712 	/* truncate both memory and reserved regions */
1713 	memblock_remove_range(&memblock.memory, max_addr,
1714 			      PHYS_ADDR_MAX);
1715 	memblock_remove_range(&memblock.reserved, max_addr,
1716 			      PHYS_ADDR_MAX);
1717 }
1718 
1719 void __init memblock_cap_memory_range(phys_addr_t base, phys_addr_t size)
1720 {
1721 	int start_rgn, end_rgn;
1722 	int i, ret;
1723 
1724 	if (!size)
1725 		return;
1726 
1727 	ret = memblock_isolate_range(&memblock.memory, base, size,
1728 						&start_rgn, &end_rgn);
1729 	if (ret)
1730 		return;
1731 
1732 	/* remove all the MAP regions */
1733 	for (i = memblock.memory.cnt - 1; i >= end_rgn; i--)
1734 		if (!memblock_is_nomap(&memblock.memory.regions[i]))
1735 			memblock_remove_region(&memblock.memory, i);
1736 
1737 	for (i = start_rgn - 1; i >= 0; i--)
1738 		if (!memblock_is_nomap(&memblock.memory.regions[i]))
1739 			memblock_remove_region(&memblock.memory, i);
1740 
1741 	/* truncate the reserved regions */
1742 	memblock_remove_range(&memblock.reserved, 0, base);
1743 	memblock_remove_range(&memblock.reserved,
1744 			base + size, PHYS_ADDR_MAX);
1745 }
1746 
1747 void __init memblock_mem_limit_remove_map(phys_addr_t limit)
1748 {
1749 	phys_addr_t max_addr;
1750 
1751 	if (!limit)
1752 		return;
1753 
1754 	max_addr = __find_max_addr(limit);
1755 
1756 	/* @limit exceeds the total size of the memory, do nothing */
1757 	if (max_addr == PHYS_ADDR_MAX)
1758 		return;
1759 
1760 	memblock_cap_memory_range(0, max_addr);
1761 }
1762 
1763 static int __init_memblock memblock_search(struct memblock_type *type, phys_addr_t addr)
1764 {
1765 	unsigned int left = 0, right = type->cnt;
1766 
1767 	do {
1768 		unsigned int mid = (right + left) / 2;
1769 
1770 		if (addr < type->regions[mid].base)
1771 			right = mid;
1772 		else if (addr >= (type->regions[mid].base +
1773 				  type->regions[mid].size))
1774 			left = mid + 1;
1775 		else
1776 			return mid;
1777 	} while (left < right);
1778 	return -1;
1779 }
1780 
1781 bool __init_memblock memblock_is_reserved(phys_addr_t addr)
1782 {
1783 	return memblock_search(&memblock.reserved, addr) != -1;
1784 }
1785 
1786 bool __init_memblock memblock_is_memory(phys_addr_t addr)
1787 {
1788 	return memblock_search(&memblock.memory, addr) != -1;
1789 }
1790 
1791 bool __init_memblock memblock_is_map_memory(phys_addr_t addr)
1792 {
1793 	int i = memblock_search(&memblock.memory, addr);
1794 
1795 	if (i == -1)
1796 		return false;
1797 	return !memblock_is_nomap(&memblock.memory.regions[i]);
1798 }
1799 
1800 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
1801 int __init_memblock memblock_search_pfn_nid(unsigned long pfn,
1802 			 unsigned long *start_pfn, unsigned long *end_pfn)
1803 {
1804 	struct memblock_type *type = &memblock.memory;
1805 	int mid = memblock_search(type, PFN_PHYS(pfn));
1806 
1807 	if (mid == -1)
1808 		return -1;
1809 
1810 	*start_pfn = PFN_DOWN(type->regions[mid].base);
1811 	*end_pfn = PFN_DOWN(type->regions[mid].base + type->regions[mid].size);
1812 
1813 	return type->regions[mid].nid;
1814 }
1815 #endif
1816 
1817 /**
1818  * memblock_is_region_memory - check if a region is a subset of memory
1819  * @base: base of region to check
1820  * @size: size of region to check
1821  *
1822  * Check if the region [@base, @base + @size) is a subset of a memory block.
1823  *
1824  * Return:
1825  * 0 if false, non-zero if true
1826  */
1827 bool __init_memblock memblock_is_region_memory(phys_addr_t base, phys_addr_t size)
1828 {
1829 	int idx = memblock_search(&memblock.memory, base);
1830 	phys_addr_t end = base + memblock_cap_size(base, &size);
1831 
1832 	if (idx == -1)
1833 		return false;
1834 	return (memblock.memory.regions[idx].base +
1835 		 memblock.memory.regions[idx].size) >= end;
1836 }
1837 
1838 /**
1839  * memblock_is_region_reserved - check if a region intersects reserved memory
1840  * @base: base of region to check
1841  * @size: size of region to check
1842  *
1843  * Check if the region [@base, @base + @size) intersects a reserved
1844  * memory block.
1845  *
1846  * Return:
1847  * True if they intersect, false if not.
1848  */
1849 bool __init_memblock memblock_is_region_reserved(phys_addr_t base, phys_addr_t size)
1850 {
1851 	memblock_cap_size(base, &size);
1852 	return memblock_overlaps_region(&memblock.reserved, base, size);
1853 }
1854 
1855 void __init_memblock memblock_trim_memory(phys_addr_t align)
1856 {
1857 	phys_addr_t start, end, orig_start, orig_end;
1858 	struct memblock_region *r;
1859 
1860 	for_each_memblock(memory, r) {
1861 		orig_start = r->base;
1862 		orig_end = r->base + r->size;
1863 		start = round_up(orig_start, align);
1864 		end = round_down(orig_end, align);
1865 
1866 		if (start == orig_start && end == orig_end)
1867 			continue;
1868 
1869 		if (start < end) {
1870 			r->base = start;
1871 			r->size = end - start;
1872 		} else {
1873 			memblock_remove_region(&memblock.memory,
1874 					       r - memblock.memory.regions);
1875 			r--;
1876 		}
1877 	}
1878 }
1879 
1880 void __init_memblock memblock_set_current_limit(phys_addr_t limit)
1881 {
1882 	memblock.current_limit = limit;
1883 }
1884 
1885 phys_addr_t __init_memblock memblock_get_current_limit(void)
1886 {
1887 	return memblock.current_limit;
1888 }
1889 
1890 static void __init_memblock memblock_dump(struct memblock_type *type)
1891 {
1892 	phys_addr_t base, end, size;
1893 	enum memblock_flags flags;
1894 	int idx;
1895 	struct memblock_region *rgn;
1896 
1897 	pr_info(" %s.cnt  = 0x%lx\n", type->name, type->cnt);
1898 
1899 	for_each_memblock_type(idx, type, rgn) {
1900 		char nid_buf[32] = "";
1901 
1902 		base = rgn->base;
1903 		size = rgn->size;
1904 		end = base + size - 1;
1905 		flags = rgn->flags;
1906 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
1907 		if (memblock_get_region_node(rgn) != MAX_NUMNODES)
1908 			snprintf(nid_buf, sizeof(nid_buf), " on node %d",
1909 				 memblock_get_region_node(rgn));
1910 #endif
1911 		pr_info(" %s[%#x]\t[%pa-%pa], %pa bytes%s flags: %#x\n",
1912 			type->name, idx, &base, &end, &size, nid_buf, flags);
1913 	}
1914 }
1915 
1916 void __init_memblock __memblock_dump_all(void)
1917 {
1918 	pr_info("MEMBLOCK configuration:\n");
1919 	pr_info(" memory size = %pa reserved size = %pa\n",
1920 		&memblock.memory.total_size,
1921 		&memblock.reserved.total_size);
1922 
1923 	memblock_dump(&memblock.memory);
1924 	memblock_dump(&memblock.reserved);
1925 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
1926 	memblock_dump(&memblock.physmem);
1927 #endif
1928 }
1929 
1930 void __init memblock_allow_resize(void)
1931 {
1932 	memblock_can_resize = 1;
1933 }
1934 
1935 static int __init early_memblock(char *p)
1936 {
1937 	if (p && strstr(p, "debug"))
1938 		memblock_debug = 1;
1939 	return 0;
1940 }
1941 early_param("memblock", early_memblock);
1942 
1943 static void __init __free_pages_memory(unsigned long start, unsigned long end)
1944 {
1945 	int order;
1946 
1947 	while (start < end) {
1948 		order = min(MAX_ORDER - 1UL, __ffs(start));
1949 
1950 		while (start + (1UL << order) > end)
1951 			order--;
1952 
1953 		memblock_free_pages(pfn_to_page(start), start, order);
1954 
1955 		start += (1UL << order);
1956 	}
1957 }
1958 
1959 static unsigned long __init __free_memory_core(phys_addr_t start,
1960 				 phys_addr_t end)
1961 {
1962 	unsigned long start_pfn = PFN_UP(start);
1963 	unsigned long end_pfn = min_t(unsigned long,
1964 				      PFN_DOWN(end), max_low_pfn);
1965 
1966 	if (start_pfn >= end_pfn)
1967 		return 0;
1968 
1969 	__free_pages_memory(start_pfn, end_pfn);
1970 
1971 	return end_pfn - start_pfn;
1972 }
1973 
1974 static unsigned long __init free_low_memory_core_early(void)
1975 {
1976 	unsigned long count = 0;
1977 	phys_addr_t start, end;
1978 	u64 i;
1979 
1980 	memblock_clear_hotplug(0, -1);
1981 
1982 	for_each_reserved_mem_region(i, &start, &end)
1983 		reserve_bootmem_region(start, end);
1984 
1985 	/*
1986 	 * We need to use NUMA_NO_NODE instead of NODE_DATA(0)->node_id
1987 	 *  because in some case like Node0 doesn't have RAM installed
1988 	 *  low ram will be on Node1
1989 	 */
1990 	for_each_free_mem_range(i, NUMA_NO_NODE, MEMBLOCK_NONE, &start, &end,
1991 				NULL)
1992 		count += __free_memory_core(start, end);
1993 
1994 	return count;
1995 }
1996 
1997 static int reset_managed_pages_done __initdata;
1998 
1999 void reset_node_managed_pages(pg_data_t *pgdat)
2000 {
2001 	struct zone *z;
2002 
2003 	for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++)
2004 		atomic_long_set(&z->managed_pages, 0);
2005 }
2006 
2007 void __init reset_all_zones_managed_pages(void)
2008 {
2009 	struct pglist_data *pgdat;
2010 
2011 	if (reset_managed_pages_done)
2012 		return;
2013 
2014 	for_each_online_pgdat(pgdat)
2015 		reset_node_managed_pages(pgdat);
2016 
2017 	reset_managed_pages_done = 1;
2018 }
2019 
2020 /**
2021  * memblock_free_all - release free pages to the buddy allocator
2022  *
2023  * Return: the number of pages actually released.
2024  */
2025 unsigned long __init memblock_free_all(void)
2026 {
2027 	unsigned long pages;
2028 
2029 	reset_all_zones_managed_pages();
2030 
2031 	pages = free_low_memory_core_early();
2032 	totalram_pages_add(pages);
2033 
2034 	return pages;
2035 }
2036 
2037 #if defined(CONFIG_DEBUG_FS) && defined(CONFIG_ARCH_KEEP_MEMBLOCK)
2038 
2039 static int memblock_debug_show(struct seq_file *m, void *private)
2040 {
2041 	struct memblock_type *type = m->private;
2042 	struct memblock_region *reg;
2043 	int i;
2044 	phys_addr_t end;
2045 
2046 	for (i = 0; i < type->cnt; i++) {
2047 		reg = &type->regions[i];
2048 		end = reg->base + reg->size - 1;
2049 
2050 		seq_printf(m, "%4d: ", i);
2051 		seq_printf(m, "%pa..%pa\n", &reg->base, &end);
2052 	}
2053 	return 0;
2054 }
2055 DEFINE_SHOW_ATTRIBUTE(memblock_debug);
2056 
2057 static int __init memblock_init_debugfs(void)
2058 {
2059 	struct dentry *root = debugfs_create_dir("memblock", NULL);
2060 
2061 	debugfs_create_file("memory", 0444, root,
2062 			    &memblock.memory, &memblock_debug_fops);
2063 	debugfs_create_file("reserved", 0444, root,
2064 			    &memblock.reserved, &memblock_debug_fops);
2065 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
2066 	debugfs_create_file("physmem", 0444, root,
2067 			    &memblock.physmem, &memblock_debug_fops);
2068 #endif
2069 
2070 	return 0;
2071 }
2072 __initcall(memblock_init_debugfs);
2073 
2074 #endif /* CONFIG_DEBUG_FS */
2075