xref: /linux/mm/memblock.c (revision 26fbb4c8c7c3ee9a4c3b4de555a8587b5a19154e)
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  * * ``physmem`` - describes the actual physical memory available during
48  *   boot regardless of the possible restrictions and memory hot(un)plug;
49  *   the ``physmem`` type is only available on some architectures.
50  *
51  * Each region is represented by 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 struct memblock_type
54  * which contains an array of memory regions along with
55  * the allocator metadata. The "memory" and "reserved" types are nicely
56  * wrapped with struct memblock. This structure is statically
57  * initialized at build time. The region arrays are initially sized to
58  * %INIT_MEMBLOCK_REGIONS for "memory" and %INIT_MEMBLOCK_RESERVED_REGIONS
59  * for "reserved". The region array for "physmem" is initially sized to
60  * %INIT_PHYSMEM_REGIONS.
61  * The memblock_allow_resize() enables automatic resizing of the region
62  * arrays during addition of new regions. This feature should be used
63  * with care so that memory allocated for the region array will not
64  * overlap with areas that should be reserved, for example initrd.
65  *
66  * The early architecture setup should tell memblock what the physical
67  * memory layout is by using memblock_add() or memblock_add_node()
68  * functions. The first function does not assign the region to a NUMA
69  * node and it is appropriate for UMA systems. Yet, it is possible to
70  * use it on NUMA systems as well and assign the region to a NUMA node
71  * later in the setup process using memblock_set_node(). The
72  * memblock_add_node() performs such an assignment directly.
73  *
74  * Once memblock is setup the memory can be allocated using one of the
75  * API variants:
76  *
77  * * memblock_phys_alloc*() - these functions return the **physical**
78  *   address of the allocated memory
79  * * memblock_alloc*() - these functions return the **virtual** address
80  *   of the allocated memory.
81  *
82  * Note, that both API variants use implicit assumptions about allowed
83  * memory ranges and the fallback methods. Consult the documentation
84  * of memblock_alloc_internal() and memblock_alloc_range_nid()
85  * functions for more elaborate description.
86  *
87  * As the system boot progresses, the architecture specific mem_init()
88  * function frees all the memory to the buddy page allocator.
89  *
90  * Unless an architecture enables %CONFIG_ARCH_KEEP_MEMBLOCK, the
91  * memblock data structures (except "physmem") will be discarded after the
92  * system initialization completes.
93  */
94 
95 #ifndef CONFIG_NEED_MULTIPLE_NODES
96 struct pglist_data __refdata contig_page_data;
97 EXPORT_SYMBOL(contig_page_data);
98 #endif
99 
100 unsigned long max_low_pfn;
101 unsigned long min_low_pfn;
102 unsigned long max_pfn;
103 unsigned long long max_possible_pfn;
104 
105 static struct memblock_region memblock_memory_init_regions[INIT_MEMBLOCK_REGIONS] __initdata_memblock;
106 static struct memblock_region memblock_reserved_init_regions[INIT_MEMBLOCK_RESERVED_REGIONS] __initdata_memblock;
107 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
108 static struct memblock_region memblock_physmem_init_regions[INIT_PHYSMEM_REGIONS];
109 #endif
110 
111 struct memblock memblock __initdata_memblock = {
112 	.memory.regions		= memblock_memory_init_regions,
113 	.memory.cnt		= 1,	/* empty dummy entry */
114 	.memory.max		= INIT_MEMBLOCK_REGIONS,
115 	.memory.name		= "memory",
116 
117 	.reserved.regions	= memblock_reserved_init_regions,
118 	.reserved.cnt		= 1,	/* empty dummy entry */
119 	.reserved.max		= INIT_MEMBLOCK_RESERVED_REGIONS,
120 	.reserved.name		= "reserved",
121 
122 	.bottom_up		= false,
123 	.current_limit		= MEMBLOCK_ALLOC_ANYWHERE,
124 };
125 
126 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
127 struct memblock_type physmem = {
128 	.regions		= memblock_physmem_init_regions,
129 	.cnt			= 1,	/* empty dummy entry */
130 	.max			= INIT_PHYSMEM_REGIONS,
131 	.name			= "physmem",
132 };
133 #endif
134 
135 /*
136  * keep a pointer to &memblock.memory in the text section to use it in
137  * __next_mem_range() and its helpers.
138  *  For architectures that do not keep memblock data after init, this
139  * pointer will be reset to NULL at memblock_discard()
140  */
141 static __refdata struct memblock_type *memblock_memory = &memblock.memory;
142 
143 #define for_each_memblock_type(i, memblock_type, rgn)			\
144 	for (i = 0, rgn = &memblock_type->regions[0];			\
145 	     i < memblock_type->cnt;					\
146 	     i++, rgn = &memblock_type->regions[i])
147 
148 #define memblock_dbg(fmt, ...)						\
149 	do {								\
150 		if (memblock_debug)					\
151 			pr_info(fmt, ##__VA_ARGS__);			\
152 	} while (0)
153 
154 static int memblock_debug __initdata_memblock;
155 static bool system_has_some_mirror __initdata_memblock = false;
156 static int memblock_can_resize __initdata_memblock;
157 static int memblock_memory_in_slab __initdata_memblock = 0;
158 static int memblock_reserved_in_slab __initdata_memblock = 0;
159 
160 static enum memblock_flags __init_memblock choose_memblock_flags(void)
161 {
162 	return system_has_some_mirror ? MEMBLOCK_MIRROR : MEMBLOCK_NONE;
163 }
164 
165 /* adjust *@size so that (@base + *@size) doesn't overflow, return new size */
166 static inline phys_addr_t memblock_cap_size(phys_addr_t base, phys_addr_t *size)
167 {
168 	return *size = min(*size, PHYS_ADDR_MAX - base);
169 }
170 
171 /*
172  * Address comparison utilities
173  */
174 static unsigned long __init_memblock memblock_addrs_overlap(phys_addr_t base1, phys_addr_t size1,
175 				       phys_addr_t base2, phys_addr_t size2)
176 {
177 	return ((base1 < (base2 + size2)) && (base2 < (base1 + size1)));
178 }
179 
180 bool __init_memblock memblock_overlaps_region(struct memblock_type *type,
181 					phys_addr_t base, phys_addr_t size)
182 {
183 	unsigned long i;
184 
185 	for (i = 0; i < type->cnt; i++)
186 		if (memblock_addrs_overlap(base, size, type->regions[i].base,
187 					   type->regions[i].size))
188 			break;
189 	return i < type->cnt;
190 }
191 
192 /**
193  * __memblock_find_range_bottom_up - find free area utility in bottom-up
194  * @start: start of candidate range
195  * @end: end of candidate range, can be %MEMBLOCK_ALLOC_ANYWHERE or
196  *       %MEMBLOCK_ALLOC_ACCESSIBLE
197  * @size: size of free area to find
198  * @align: alignment of free area to find
199  * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
200  * @flags: pick from blocks based on memory attributes
201  *
202  * Utility called from memblock_find_in_range_node(), find free area bottom-up.
203  *
204  * Return:
205  * Found address on success, 0 on failure.
206  */
207 static phys_addr_t __init_memblock
208 __memblock_find_range_bottom_up(phys_addr_t start, phys_addr_t end,
209 				phys_addr_t size, phys_addr_t align, int nid,
210 				enum memblock_flags flags)
211 {
212 	phys_addr_t this_start, this_end, cand;
213 	u64 i;
214 
215 	for_each_free_mem_range(i, nid, flags, &this_start, &this_end, NULL) {
216 		this_start = clamp(this_start, start, end);
217 		this_end = clamp(this_end, start, end);
218 
219 		cand = round_up(this_start, align);
220 		if (cand < this_end && this_end - cand >= size)
221 			return cand;
222 	}
223 
224 	return 0;
225 }
226 
227 /**
228  * __memblock_find_range_top_down - find free area utility, in top-down
229  * @start: start of candidate range
230  * @end: end of candidate range, can be %MEMBLOCK_ALLOC_ANYWHERE or
231  *       %MEMBLOCK_ALLOC_ACCESSIBLE
232  * @size: size of free area to find
233  * @align: alignment of free area to find
234  * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
235  * @flags: pick from blocks based on memory attributes
236  *
237  * Utility called from memblock_find_in_range_node(), find free area top-down.
238  *
239  * Return:
240  * Found address on success, 0 on failure.
241  */
242 static phys_addr_t __init_memblock
243 __memblock_find_range_top_down(phys_addr_t start, phys_addr_t end,
244 			       phys_addr_t size, phys_addr_t align, int nid,
245 			       enum memblock_flags flags)
246 {
247 	phys_addr_t this_start, this_end, cand;
248 	u64 i;
249 
250 	for_each_free_mem_range_reverse(i, nid, flags, &this_start, &this_end,
251 					NULL) {
252 		this_start = clamp(this_start, start, end);
253 		this_end = clamp(this_end, start, end);
254 
255 		if (this_end < size)
256 			continue;
257 
258 		cand = round_down(this_end - size, align);
259 		if (cand >= this_start)
260 			return cand;
261 	}
262 
263 	return 0;
264 }
265 
266 /**
267  * memblock_find_in_range_node - find free area in given range and node
268  * @size: size of free area to find
269  * @align: alignment of free area to find
270  * @start: start of candidate range
271  * @end: end of candidate range, can be %MEMBLOCK_ALLOC_ANYWHERE or
272  *       %MEMBLOCK_ALLOC_ACCESSIBLE
273  * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
274  * @flags: pick from blocks based on memory attributes
275  *
276  * Find @size free area aligned to @align in the specified range and node.
277  *
278  * When allocation direction is bottom-up, the @start should be greater
279  * than the end of the kernel image. Otherwise, it will be trimmed. The
280  * reason is that we want the bottom-up allocation just near the kernel
281  * image so it is highly likely that the allocated memory and the kernel
282  * will reside in the same node.
283  *
284  * If bottom-up allocation failed, will try to allocate memory top-down.
285  *
286  * Return:
287  * Found address on success, 0 on failure.
288  */
289 static phys_addr_t __init_memblock memblock_find_in_range_node(phys_addr_t size,
290 					phys_addr_t align, phys_addr_t start,
291 					phys_addr_t end, int nid,
292 					enum memblock_flags flags)
293 {
294 	phys_addr_t kernel_end, ret;
295 
296 	/* pump up @end */
297 	if (end == MEMBLOCK_ALLOC_ACCESSIBLE ||
298 	    end == MEMBLOCK_ALLOC_KASAN)
299 		end = memblock.current_limit;
300 
301 	/* avoid allocating the first page */
302 	start = max_t(phys_addr_t, start, PAGE_SIZE);
303 	end = max(start, end);
304 	kernel_end = __pa_symbol(_end);
305 
306 	/*
307 	 * try bottom-up allocation only when bottom-up mode
308 	 * is set and @end is above the kernel image.
309 	 */
310 	if (memblock_bottom_up() && end > kernel_end) {
311 		phys_addr_t bottom_up_start;
312 
313 		/* make sure we will allocate above the kernel */
314 		bottom_up_start = max(start, kernel_end);
315 
316 		/* ok, try bottom-up allocation first */
317 		ret = __memblock_find_range_bottom_up(bottom_up_start, end,
318 						      size, align, nid, flags);
319 		if (ret)
320 			return ret;
321 
322 		/*
323 		 * we always limit bottom-up allocation above the kernel,
324 		 * but top-down allocation doesn't have the limit, so
325 		 * retrying top-down allocation may succeed when bottom-up
326 		 * allocation failed.
327 		 *
328 		 * bottom-up allocation is expected to be fail very rarely,
329 		 * so we use WARN_ONCE() here to see the stack trace if
330 		 * fail happens.
331 		 */
332 		WARN_ONCE(IS_ENABLED(CONFIG_MEMORY_HOTREMOVE),
333 			  "memblock: bottom-up allocation failed, memory hotremove may be affected\n");
334 	}
335 
336 	return __memblock_find_range_top_down(start, end, size, align, nid,
337 					      flags);
338 }
339 
340 /**
341  * memblock_find_in_range - find free area in given range
342  * @start: start of candidate range
343  * @end: end of candidate range, can be %MEMBLOCK_ALLOC_ANYWHERE or
344  *       %MEMBLOCK_ALLOC_ACCESSIBLE
345  * @size: size of free area to find
346  * @align: alignment of free area to find
347  *
348  * Find @size free area aligned to @align in the specified range.
349  *
350  * Return:
351  * Found address on success, 0 on failure.
352  */
353 phys_addr_t __init_memblock memblock_find_in_range(phys_addr_t start,
354 					phys_addr_t end, phys_addr_t size,
355 					phys_addr_t align)
356 {
357 	phys_addr_t ret;
358 	enum memblock_flags flags = choose_memblock_flags();
359 
360 again:
361 	ret = memblock_find_in_range_node(size, align, start, end,
362 					    NUMA_NO_NODE, flags);
363 
364 	if (!ret && (flags & MEMBLOCK_MIRROR)) {
365 		pr_warn("Could not allocate %pap bytes of mirrored memory\n",
366 			&size);
367 		flags &= ~MEMBLOCK_MIRROR;
368 		goto again;
369 	}
370 
371 	return ret;
372 }
373 
374 static void __init_memblock memblock_remove_region(struct memblock_type *type, unsigned long r)
375 {
376 	type->total_size -= type->regions[r].size;
377 	memmove(&type->regions[r], &type->regions[r + 1],
378 		(type->cnt - (r + 1)) * sizeof(type->regions[r]));
379 	type->cnt--;
380 
381 	/* Special case for empty arrays */
382 	if (type->cnt == 0) {
383 		WARN_ON(type->total_size != 0);
384 		type->cnt = 1;
385 		type->regions[0].base = 0;
386 		type->regions[0].size = 0;
387 		type->regions[0].flags = 0;
388 		memblock_set_region_node(&type->regions[0], MAX_NUMNODES);
389 	}
390 }
391 
392 #ifndef CONFIG_ARCH_KEEP_MEMBLOCK
393 /**
394  * memblock_discard - discard memory and reserved arrays if they were allocated
395  */
396 void __init memblock_discard(void)
397 {
398 	phys_addr_t addr, size;
399 
400 	if (memblock.reserved.regions != memblock_reserved_init_regions) {
401 		addr = __pa(memblock.reserved.regions);
402 		size = PAGE_ALIGN(sizeof(struct memblock_region) *
403 				  memblock.reserved.max);
404 		__memblock_free_late(addr, size);
405 	}
406 
407 	if (memblock.memory.regions != memblock_memory_init_regions) {
408 		addr = __pa(memblock.memory.regions);
409 		size = PAGE_ALIGN(sizeof(struct memblock_region) *
410 				  memblock.memory.max);
411 		__memblock_free_late(addr, size);
412 	}
413 
414 	memblock_memory = NULL;
415 }
416 #endif
417 
418 /**
419  * memblock_double_array - double the size of the memblock regions array
420  * @type: memblock type of the regions array being doubled
421  * @new_area_start: starting address of memory range to avoid overlap with
422  * @new_area_size: size of memory range to avoid overlap with
423  *
424  * Double the size of the @type regions array. If memblock is being used to
425  * allocate memory for a new reserved regions array and there is a previously
426  * allocated memory range [@new_area_start, @new_area_start + @new_area_size]
427  * waiting to be reserved, ensure the memory used by the new array does
428  * not overlap.
429  *
430  * Return:
431  * 0 on success, -1 on failure.
432  */
433 static int __init_memblock memblock_double_array(struct memblock_type *type,
434 						phys_addr_t new_area_start,
435 						phys_addr_t new_area_size)
436 {
437 	struct memblock_region *new_array, *old_array;
438 	phys_addr_t old_alloc_size, new_alloc_size;
439 	phys_addr_t old_size, new_size, addr, new_end;
440 	int use_slab = slab_is_available();
441 	int *in_slab;
442 
443 	/* We don't allow resizing until we know about the reserved regions
444 	 * of memory that aren't suitable for allocation
445 	 */
446 	if (!memblock_can_resize)
447 		return -1;
448 
449 	/* Calculate new doubled size */
450 	old_size = type->max * sizeof(struct memblock_region);
451 	new_size = old_size << 1;
452 	/*
453 	 * We need to allocated new one align to PAGE_SIZE,
454 	 *   so we can free them completely later.
455 	 */
456 	old_alloc_size = PAGE_ALIGN(old_size);
457 	new_alloc_size = PAGE_ALIGN(new_size);
458 
459 	/* Retrieve the slab flag */
460 	if (type == &memblock.memory)
461 		in_slab = &memblock_memory_in_slab;
462 	else
463 		in_slab = &memblock_reserved_in_slab;
464 
465 	/* Try to find some space for it */
466 	if (use_slab) {
467 		new_array = kmalloc(new_size, GFP_KERNEL);
468 		addr = new_array ? __pa(new_array) : 0;
469 	} else {
470 		/* only exclude range when trying to double reserved.regions */
471 		if (type != &memblock.reserved)
472 			new_area_start = new_area_size = 0;
473 
474 		addr = memblock_find_in_range(new_area_start + new_area_size,
475 						memblock.current_limit,
476 						new_alloc_size, PAGE_SIZE);
477 		if (!addr && new_area_size)
478 			addr = memblock_find_in_range(0,
479 				min(new_area_start, memblock.current_limit),
480 				new_alloc_size, PAGE_SIZE);
481 
482 		new_array = addr ? __va(addr) : NULL;
483 	}
484 	if (!addr) {
485 		pr_err("memblock: Failed to double %s array from %ld to %ld entries !\n",
486 		       type->name, type->max, type->max * 2);
487 		return -1;
488 	}
489 
490 	new_end = addr + new_size - 1;
491 	memblock_dbg("memblock: %s is doubled to %ld at [%pa-%pa]",
492 			type->name, type->max * 2, &addr, &new_end);
493 
494 	/*
495 	 * Found space, we now need to move the array over before we add the
496 	 * reserved region since it may be our reserved array itself that is
497 	 * full.
498 	 */
499 	memcpy(new_array, type->regions, old_size);
500 	memset(new_array + type->max, 0, old_size);
501 	old_array = type->regions;
502 	type->regions = new_array;
503 	type->max <<= 1;
504 
505 	/* Free old array. We needn't free it if the array is the static one */
506 	if (*in_slab)
507 		kfree(old_array);
508 	else if (old_array != memblock_memory_init_regions &&
509 		 old_array != memblock_reserved_init_regions)
510 		memblock_free(__pa(old_array), old_alloc_size);
511 
512 	/*
513 	 * Reserve the new array if that comes from the memblock.  Otherwise, we
514 	 * needn't do it
515 	 */
516 	if (!use_slab)
517 		BUG_ON(memblock_reserve(addr, new_alloc_size));
518 
519 	/* Update slab flag */
520 	*in_slab = use_slab;
521 
522 	return 0;
523 }
524 
525 /**
526  * memblock_merge_regions - merge neighboring compatible regions
527  * @type: memblock type to scan
528  *
529  * Scan @type and merge neighboring compatible regions.
530  */
531 static void __init_memblock memblock_merge_regions(struct memblock_type *type)
532 {
533 	int i = 0;
534 
535 	/* cnt never goes below 1 */
536 	while (i < type->cnt - 1) {
537 		struct memblock_region *this = &type->regions[i];
538 		struct memblock_region *next = &type->regions[i + 1];
539 
540 		if (this->base + this->size != next->base ||
541 		    memblock_get_region_node(this) !=
542 		    memblock_get_region_node(next) ||
543 		    this->flags != next->flags) {
544 			BUG_ON(this->base + this->size > next->base);
545 			i++;
546 			continue;
547 		}
548 
549 		this->size += next->size;
550 		/* move forward from next + 1, index of which is i + 2 */
551 		memmove(next, next + 1, (type->cnt - (i + 2)) * sizeof(*next));
552 		type->cnt--;
553 	}
554 }
555 
556 /**
557  * memblock_insert_region - insert new memblock region
558  * @type:	memblock type to insert into
559  * @idx:	index for the insertion point
560  * @base:	base address of the new region
561  * @size:	size of the new region
562  * @nid:	node id of the new region
563  * @flags:	flags of the new region
564  *
565  * Insert new memblock region [@base, @base + @size) into @type at @idx.
566  * @type must already have extra room to accommodate the new region.
567  */
568 static void __init_memblock memblock_insert_region(struct memblock_type *type,
569 						   int idx, phys_addr_t base,
570 						   phys_addr_t size,
571 						   int nid,
572 						   enum memblock_flags flags)
573 {
574 	struct memblock_region *rgn = &type->regions[idx];
575 
576 	BUG_ON(type->cnt >= type->max);
577 	memmove(rgn + 1, rgn, (type->cnt - idx) * sizeof(*rgn));
578 	rgn->base = base;
579 	rgn->size = size;
580 	rgn->flags = flags;
581 	memblock_set_region_node(rgn, nid);
582 	type->cnt++;
583 	type->total_size += size;
584 }
585 
586 /**
587  * memblock_add_range - add new memblock region
588  * @type: memblock type to add new region into
589  * @base: base address of the new region
590  * @size: size of the new region
591  * @nid: nid of the new region
592  * @flags: flags of the new region
593  *
594  * Add new memblock region [@base, @base + @size) into @type.  The new region
595  * is allowed to overlap with existing ones - overlaps don't affect already
596  * existing regions.  @type is guaranteed to be minimal (all neighbouring
597  * compatible regions are merged) after the addition.
598  *
599  * Return:
600  * 0 on success, -errno on failure.
601  */
602 static int __init_memblock memblock_add_range(struct memblock_type *type,
603 				phys_addr_t base, phys_addr_t size,
604 				int nid, enum memblock_flags flags)
605 {
606 	bool insert = false;
607 	phys_addr_t obase = base;
608 	phys_addr_t end = base + memblock_cap_size(base, &size);
609 	int idx, nr_new;
610 	struct memblock_region *rgn;
611 
612 	if (!size)
613 		return 0;
614 
615 	/* special case for empty array */
616 	if (type->regions[0].size == 0) {
617 		WARN_ON(type->cnt != 1 || type->total_size);
618 		type->regions[0].base = base;
619 		type->regions[0].size = size;
620 		type->regions[0].flags = flags;
621 		memblock_set_region_node(&type->regions[0], nid);
622 		type->total_size = size;
623 		return 0;
624 	}
625 repeat:
626 	/*
627 	 * The following is executed twice.  Once with %false @insert and
628 	 * then with %true.  The first counts the number of regions needed
629 	 * to accommodate the new area.  The second actually inserts them.
630 	 */
631 	base = obase;
632 	nr_new = 0;
633 
634 	for_each_memblock_type(idx, type, rgn) {
635 		phys_addr_t rbase = rgn->base;
636 		phys_addr_t rend = rbase + rgn->size;
637 
638 		if (rbase >= end)
639 			break;
640 		if (rend <= base)
641 			continue;
642 		/*
643 		 * @rgn overlaps.  If it separates the lower part of new
644 		 * area, insert that portion.
645 		 */
646 		if (rbase > base) {
647 #ifdef CONFIG_NEED_MULTIPLE_NODES
648 			WARN_ON(nid != memblock_get_region_node(rgn));
649 #endif
650 			WARN_ON(flags != rgn->flags);
651 			nr_new++;
652 			if (insert)
653 				memblock_insert_region(type, idx++, base,
654 						       rbase - base, nid,
655 						       flags);
656 		}
657 		/* area below @rend is dealt with, forget about it */
658 		base = min(rend, end);
659 	}
660 
661 	/* insert the remaining portion */
662 	if (base < end) {
663 		nr_new++;
664 		if (insert)
665 			memblock_insert_region(type, idx, base, end - base,
666 					       nid, flags);
667 	}
668 
669 	if (!nr_new)
670 		return 0;
671 
672 	/*
673 	 * If this was the first round, resize array and repeat for actual
674 	 * insertions; otherwise, merge and return.
675 	 */
676 	if (!insert) {
677 		while (type->cnt + nr_new > type->max)
678 			if (memblock_double_array(type, obase, size) < 0)
679 				return -ENOMEM;
680 		insert = true;
681 		goto repeat;
682 	} else {
683 		memblock_merge_regions(type);
684 		return 0;
685 	}
686 }
687 
688 /**
689  * memblock_add_node - add new memblock region within a NUMA node
690  * @base: base address of the new region
691  * @size: size of the new region
692  * @nid: nid of the new region
693  *
694  * Add new memblock region [@base, @base + @size) to the "memory"
695  * type. See memblock_add_range() description for mode details
696  *
697  * Return:
698  * 0 on success, -errno on failure.
699  */
700 int __init_memblock memblock_add_node(phys_addr_t base, phys_addr_t size,
701 				       int nid)
702 {
703 	return memblock_add_range(&memblock.memory, base, size, nid, 0);
704 }
705 
706 /**
707  * memblock_add - add new memblock region
708  * @base: base address of the new region
709  * @size: size of the new region
710  *
711  * Add new memblock region [@base, @base + @size) to the "memory"
712  * type. See memblock_add_range() description for mode details
713  *
714  * Return:
715  * 0 on success, -errno on failure.
716  */
717 int __init_memblock memblock_add(phys_addr_t base, phys_addr_t size)
718 {
719 	phys_addr_t end = base + size - 1;
720 
721 	memblock_dbg("%s: [%pa-%pa] %pS\n", __func__,
722 		     &base, &end, (void *)_RET_IP_);
723 
724 	return memblock_add_range(&memblock.memory, base, size, MAX_NUMNODES, 0);
725 }
726 
727 /**
728  * memblock_isolate_range - isolate given range into disjoint memblocks
729  * @type: memblock type to isolate range for
730  * @base: base of range to isolate
731  * @size: size of range to isolate
732  * @start_rgn: out parameter for the start of isolated region
733  * @end_rgn: out parameter for the end of isolated region
734  *
735  * Walk @type and ensure that regions don't cross the boundaries defined by
736  * [@base, @base + @size).  Crossing regions are split at the boundaries,
737  * which may create at most two more regions.  The index of the first
738  * region inside the range is returned in *@start_rgn and end in *@end_rgn.
739  *
740  * Return:
741  * 0 on success, -errno on failure.
742  */
743 static int __init_memblock memblock_isolate_range(struct memblock_type *type,
744 					phys_addr_t base, phys_addr_t size,
745 					int *start_rgn, int *end_rgn)
746 {
747 	phys_addr_t end = base + memblock_cap_size(base, &size);
748 	int idx;
749 	struct memblock_region *rgn;
750 
751 	*start_rgn = *end_rgn = 0;
752 
753 	if (!size)
754 		return 0;
755 
756 	/* we'll create at most two more regions */
757 	while (type->cnt + 2 > type->max)
758 		if (memblock_double_array(type, base, size) < 0)
759 			return -ENOMEM;
760 
761 	for_each_memblock_type(idx, type, rgn) {
762 		phys_addr_t rbase = rgn->base;
763 		phys_addr_t rend = rbase + rgn->size;
764 
765 		if (rbase >= end)
766 			break;
767 		if (rend <= base)
768 			continue;
769 
770 		if (rbase < base) {
771 			/*
772 			 * @rgn intersects from below.  Split and continue
773 			 * to process the next region - the new top half.
774 			 */
775 			rgn->base = base;
776 			rgn->size -= base - rbase;
777 			type->total_size -= base - rbase;
778 			memblock_insert_region(type, idx, rbase, base - rbase,
779 					       memblock_get_region_node(rgn),
780 					       rgn->flags);
781 		} else if (rend > end) {
782 			/*
783 			 * @rgn intersects from above.  Split and redo the
784 			 * current region - the new bottom half.
785 			 */
786 			rgn->base = end;
787 			rgn->size -= end - rbase;
788 			type->total_size -= end - rbase;
789 			memblock_insert_region(type, idx--, rbase, end - rbase,
790 					       memblock_get_region_node(rgn),
791 					       rgn->flags);
792 		} else {
793 			/* @rgn is fully contained, record it */
794 			if (!*end_rgn)
795 				*start_rgn = idx;
796 			*end_rgn = idx + 1;
797 		}
798 	}
799 
800 	return 0;
801 }
802 
803 static int __init_memblock memblock_remove_range(struct memblock_type *type,
804 					  phys_addr_t base, phys_addr_t size)
805 {
806 	int start_rgn, end_rgn;
807 	int i, ret;
808 
809 	ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
810 	if (ret)
811 		return ret;
812 
813 	for (i = end_rgn - 1; i >= start_rgn; i--)
814 		memblock_remove_region(type, i);
815 	return 0;
816 }
817 
818 int __init_memblock memblock_remove(phys_addr_t base, phys_addr_t size)
819 {
820 	phys_addr_t end = base + size - 1;
821 
822 	memblock_dbg("%s: [%pa-%pa] %pS\n", __func__,
823 		     &base, &end, (void *)_RET_IP_);
824 
825 	return memblock_remove_range(&memblock.memory, base, size);
826 }
827 
828 /**
829  * memblock_free - free boot memory block
830  * @base: phys starting address of the  boot memory block
831  * @size: size of the boot memory block in bytes
832  *
833  * Free boot memory block previously allocated by memblock_alloc_xx() API.
834  * The freeing memory will not be released to the buddy allocator.
835  */
836 int __init_memblock memblock_free(phys_addr_t base, phys_addr_t size)
837 {
838 	phys_addr_t end = base + size - 1;
839 
840 	memblock_dbg("%s: [%pa-%pa] %pS\n", __func__,
841 		     &base, &end, (void *)_RET_IP_);
842 
843 	kmemleak_free_part_phys(base, size);
844 	return memblock_remove_range(&memblock.reserved, base, size);
845 }
846 
847 int __init_memblock memblock_reserve(phys_addr_t base, phys_addr_t size)
848 {
849 	phys_addr_t end = base + size - 1;
850 
851 	memblock_dbg("%s: [%pa-%pa] %pS\n", __func__,
852 		     &base, &end, (void *)_RET_IP_);
853 
854 	return memblock_add_range(&memblock.reserved, base, size, MAX_NUMNODES, 0);
855 }
856 
857 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
858 int __init_memblock memblock_physmem_add(phys_addr_t base, phys_addr_t size)
859 {
860 	phys_addr_t end = base + size - 1;
861 
862 	memblock_dbg("%s: [%pa-%pa] %pS\n", __func__,
863 		     &base, &end, (void *)_RET_IP_);
864 
865 	return memblock_add_range(&physmem, base, size, MAX_NUMNODES, 0);
866 }
867 #endif
868 
869 /**
870  * memblock_setclr_flag - set or clear flag for a memory region
871  * @base: base address of the region
872  * @size: size of the region
873  * @set: set or clear the flag
874  * @flag: the flag to update
875  *
876  * This function isolates region [@base, @base + @size), and sets/clears flag
877  *
878  * Return: 0 on success, -errno on failure.
879  */
880 static int __init_memblock memblock_setclr_flag(phys_addr_t base,
881 				phys_addr_t size, int set, int flag)
882 {
883 	struct memblock_type *type = &memblock.memory;
884 	int i, ret, start_rgn, end_rgn;
885 
886 	ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
887 	if (ret)
888 		return ret;
889 
890 	for (i = start_rgn; i < end_rgn; i++) {
891 		struct memblock_region *r = &type->regions[i];
892 
893 		if (set)
894 			r->flags |= flag;
895 		else
896 			r->flags &= ~flag;
897 	}
898 
899 	memblock_merge_regions(type);
900 	return 0;
901 }
902 
903 /**
904  * memblock_mark_hotplug - Mark hotpluggable memory with flag MEMBLOCK_HOTPLUG.
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_hotplug(phys_addr_t base, phys_addr_t size)
911 {
912 	return memblock_setclr_flag(base, size, 1, MEMBLOCK_HOTPLUG);
913 }
914 
915 /**
916  * memblock_clear_hotplug - Clear flag MEMBLOCK_HOTPLUG for a specified region.
917  * @base: the base phys addr of the region
918  * @size: the size of the region
919  *
920  * Return: 0 on success, -errno on failure.
921  */
922 int __init_memblock memblock_clear_hotplug(phys_addr_t base, phys_addr_t size)
923 {
924 	return memblock_setclr_flag(base, size, 0, MEMBLOCK_HOTPLUG);
925 }
926 
927 /**
928  * memblock_mark_mirror - Mark mirrored memory with flag MEMBLOCK_MIRROR.
929  * @base: the base phys addr of the region
930  * @size: the size of the region
931  *
932  * Return: 0 on success, -errno on failure.
933  */
934 int __init_memblock memblock_mark_mirror(phys_addr_t base, phys_addr_t size)
935 {
936 	system_has_some_mirror = true;
937 
938 	return memblock_setclr_flag(base, size, 1, MEMBLOCK_MIRROR);
939 }
940 
941 /**
942  * memblock_mark_nomap - Mark a memory region with flag MEMBLOCK_NOMAP.
943  * @base: the base phys addr of the region
944  * @size: the size of the region
945  *
946  * Return: 0 on success, -errno on failure.
947  */
948 int __init_memblock memblock_mark_nomap(phys_addr_t base, phys_addr_t size)
949 {
950 	return memblock_setclr_flag(base, size, 1, MEMBLOCK_NOMAP);
951 }
952 
953 /**
954  * memblock_clear_nomap - Clear flag MEMBLOCK_NOMAP for a specified region.
955  * @base: the base phys addr of the region
956  * @size: the size of the region
957  *
958  * Return: 0 on success, -errno on failure.
959  */
960 int __init_memblock memblock_clear_nomap(phys_addr_t base, phys_addr_t size)
961 {
962 	return memblock_setclr_flag(base, size, 0, MEMBLOCK_NOMAP);
963 }
964 
965 static bool should_skip_region(struct memblock_type *type,
966 			       struct memblock_region *m,
967 			       int nid, int flags)
968 {
969 	int m_nid = memblock_get_region_node(m);
970 
971 	/* we never skip regions when iterating memblock.reserved or physmem */
972 	if (type != memblock_memory)
973 		return false;
974 
975 	/* only memory regions are associated with nodes, check it */
976 	if (nid != NUMA_NO_NODE && nid != m_nid)
977 		return true;
978 
979 	/* skip hotpluggable memory regions if needed */
980 	if (movable_node_is_enabled() && memblock_is_hotpluggable(m))
981 		return true;
982 
983 	/* if we want mirror memory skip non-mirror memory regions */
984 	if ((flags & MEMBLOCK_MIRROR) && !memblock_is_mirror(m))
985 		return true;
986 
987 	/* skip nomap memory unless we were asked for it explicitly */
988 	if (!(flags & MEMBLOCK_NOMAP) && memblock_is_nomap(m))
989 		return true;
990 
991 	return false;
992 }
993 
994 /**
995  * __next_mem_range - next function for for_each_free_mem_range() etc.
996  * @idx: pointer to u64 loop variable
997  * @nid: node selector, %NUMA_NO_NODE for all nodes
998  * @flags: pick from blocks based on memory attributes
999  * @type_a: pointer to memblock_type from where the range is taken
1000  * @type_b: pointer to memblock_type which excludes memory from being taken
1001  * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL
1002  * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL
1003  * @out_nid: ptr to int for nid of the range, can be %NULL
1004  *
1005  * Find the first area from *@idx which matches @nid, fill the out
1006  * parameters, and update *@idx for the next iteration.  The lower 32bit of
1007  * *@idx contains index into type_a and the upper 32bit indexes the
1008  * areas before each region in type_b.	For example, if type_b regions
1009  * look like the following,
1010  *
1011  *	0:[0-16), 1:[32-48), 2:[128-130)
1012  *
1013  * The upper 32bit indexes the following regions.
1014  *
1015  *	0:[0-0), 1:[16-32), 2:[48-128), 3:[130-MAX)
1016  *
1017  * As both region arrays are sorted, the function advances the two indices
1018  * in lockstep and returns each intersection.
1019  */
1020 void __next_mem_range(u64 *idx, int nid, enum memblock_flags flags,
1021 		      struct memblock_type *type_a,
1022 		      struct memblock_type *type_b, phys_addr_t *out_start,
1023 		      phys_addr_t *out_end, int *out_nid)
1024 {
1025 	int idx_a = *idx & 0xffffffff;
1026 	int idx_b = *idx >> 32;
1027 
1028 	if (WARN_ONCE(nid == MAX_NUMNODES,
1029 	"Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
1030 		nid = NUMA_NO_NODE;
1031 
1032 	for (; idx_a < type_a->cnt; idx_a++) {
1033 		struct memblock_region *m = &type_a->regions[idx_a];
1034 
1035 		phys_addr_t m_start = m->base;
1036 		phys_addr_t m_end = m->base + m->size;
1037 		int	    m_nid = memblock_get_region_node(m);
1038 
1039 		if (should_skip_region(type_a, m, nid, flags))
1040 			continue;
1041 
1042 		if (!type_b) {
1043 			if (out_start)
1044 				*out_start = m_start;
1045 			if (out_end)
1046 				*out_end = m_end;
1047 			if (out_nid)
1048 				*out_nid = m_nid;
1049 			idx_a++;
1050 			*idx = (u32)idx_a | (u64)idx_b << 32;
1051 			return;
1052 		}
1053 
1054 		/* scan areas before each reservation */
1055 		for (; idx_b < type_b->cnt + 1; idx_b++) {
1056 			struct memblock_region *r;
1057 			phys_addr_t r_start;
1058 			phys_addr_t r_end;
1059 
1060 			r = &type_b->regions[idx_b];
1061 			r_start = idx_b ? r[-1].base + r[-1].size : 0;
1062 			r_end = idx_b < type_b->cnt ?
1063 				r->base : PHYS_ADDR_MAX;
1064 
1065 			/*
1066 			 * if idx_b advanced past idx_a,
1067 			 * break out to advance idx_a
1068 			 */
1069 			if (r_start >= m_end)
1070 				break;
1071 			/* if the two regions intersect, we're done */
1072 			if (m_start < r_end) {
1073 				if (out_start)
1074 					*out_start =
1075 						max(m_start, r_start);
1076 				if (out_end)
1077 					*out_end = min(m_end, r_end);
1078 				if (out_nid)
1079 					*out_nid = m_nid;
1080 				/*
1081 				 * The region which ends first is
1082 				 * advanced for the next iteration.
1083 				 */
1084 				if (m_end <= r_end)
1085 					idx_a++;
1086 				else
1087 					idx_b++;
1088 				*idx = (u32)idx_a | (u64)idx_b << 32;
1089 				return;
1090 			}
1091 		}
1092 	}
1093 
1094 	/* signal end of iteration */
1095 	*idx = ULLONG_MAX;
1096 }
1097 
1098 /**
1099  * __next_mem_range_rev - generic next function for for_each_*_range_rev()
1100  *
1101  * @idx: pointer to u64 loop variable
1102  * @nid: node selector, %NUMA_NO_NODE for all nodes
1103  * @flags: pick from blocks based on memory attributes
1104  * @type_a: pointer to memblock_type from where the range is taken
1105  * @type_b: pointer to memblock_type which excludes memory from being taken
1106  * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL
1107  * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL
1108  * @out_nid: ptr to int for nid of the range, can be %NULL
1109  *
1110  * Finds the next range from type_a which is not marked as unsuitable
1111  * in type_b.
1112  *
1113  * Reverse of __next_mem_range().
1114  */
1115 void __init_memblock __next_mem_range_rev(u64 *idx, int nid,
1116 					  enum memblock_flags flags,
1117 					  struct memblock_type *type_a,
1118 					  struct memblock_type *type_b,
1119 					  phys_addr_t *out_start,
1120 					  phys_addr_t *out_end, int *out_nid)
1121 {
1122 	int idx_a = *idx & 0xffffffff;
1123 	int idx_b = *idx >> 32;
1124 
1125 	if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
1126 		nid = NUMA_NO_NODE;
1127 
1128 	if (*idx == (u64)ULLONG_MAX) {
1129 		idx_a = type_a->cnt - 1;
1130 		if (type_b != NULL)
1131 			idx_b = type_b->cnt;
1132 		else
1133 			idx_b = 0;
1134 	}
1135 
1136 	for (; idx_a >= 0; idx_a--) {
1137 		struct memblock_region *m = &type_a->regions[idx_a];
1138 
1139 		phys_addr_t m_start = m->base;
1140 		phys_addr_t m_end = m->base + m->size;
1141 		int m_nid = memblock_get_region_node(m);
1142 
1143 		if (should_skip_region(type_a, m, nid, flags))
1144 			continue;
1145 
1146 		if (!type_b) {
1147 			if (out_start)
1148 				*out_start = m_start;
1149 			if (out_end)
1150 				*out_end = m_end;
1151 			if (out_nid)
1152 				*out_nid = m_nid;
1153 			idx_a--;
1154 			*idx = (u32)idx_a | (u64)idx_b << 32;
1155 			return;
1156 		}
1157 
1158 		/* scan areas before each reservation */
1159 		for (; idx_b >= 0; idx_b--) {
1160 			struct memblock_region *r;
1161 			phys_addr_t r_start;
1162 			phys_addr_t r_end;
1163 
1164 			r = &type_b->regions[idx_b];
1165 			r_start = idx_b ? r[-1].base + r[-1].size : 0;
1166 			r_end = idx_b < type_b->cnt ?
1167 				r->base : PHYS_ADDR_MAX;
1168 			/*
1169 			 * if idx_b advanced past idx_a,
1170 			 * break out to advance idx_a
1171 			 */
1172 
1173 			if (r_end <= m_start)
1174 				break;
1175 			/* if the two regions intersect, we're done */
1176 			if (m_end > r_start) {
1177 				if (out_start)
1178 					*out_start = max(m_start, r_start);
1179 				if (out_end)
1180 					*out_end = min(m_end, r_end);
1181 				if (out_nid)
1182 					*out_nid = m_nid;
1183 				if (m_start >= r_start)
1184 					idx_a--;
1185 				else
1186 					idx_b--;
1187 				*idx = (u32)idx_a | (u64)idx_b << 32;
1188 				return;
1189 			}
1190 		}
1191 	}
1192 	/* signal end of iteration */
1193 	*idx = ULLONG_MAX;
1194 }
1195 
1196 /*
1197  * Common iterator interface used to define for_each_mem_pfn_range().
1198  */
1199 void __init_memblock __next_mem_pfn_range(int *idx, int nid,
1200 				unsigned long *out_start_pfn,
1201 				unsigned long *out_end_pfn, int *out_nid)
1202 {
1203 	struct memblock_type *type = &memblock.memory;
1204 	struct memblock_region *r;
1205 	int r_nid;
1206 
1207 	while (++*idx < type->cnt) {
1208 		r = &type->regions[*idx];
1209 		r_nid = memblock_get_region_node(r);
1210 
1211 		if (PFN_UP(r->base) >= PFN_DOWN(r->base + r->size))
1212 			continue;
1213 		if (nid == MAX_NUMNODES || nid == r_nid)
1214 			break;
1215 	}
1216 	if (*idx >= type->cnt) {
1217 		*idx = -1;
1218 		return;
1219 	}
1220 
1221 	if (out_start_pfn)
1222 		*out_start_pfn = PFN_UP(r->base);
1223 	if (out_end_pfn)
1224 		*out_end_pfn = PFN_DOWN(r->base + r->size);
1225 	if (out_nid)
1226 		*out_nid = r_nid;
1227 }
1228 
1229 /**
1230  * memblock_set_node - set node ID on memblock regions
1231  * @base: base of area to set node ID for
1232  * @size: size of area to set node ID for
1233  * @type: memblock type to set node ID for
1234  * @nid: node ID to set
1235  *
1236  * Set the nid of memblock @type regions in [@base, @base + @size) to @nid.
1237  * Regions which cross the area boundaries are split as necessary.
1238  *
1239  * Return:
1240  * 0 on success, -errno on failure.
1241  */
1242 int __init_memblock memblock_set_node(phys_addr_t base, phys_addr_t size,
1243 				      struct memblock_type *type, int nid)
1244 {
1245 #ifdef CONFIG_NEED_MULTIPLE_NODES
1246 	int start_rgn, end_rgn;
1247 	int i, ret;
1248 
1249 	ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
1250 	if (ret)
1251 		return ret;
1252 
1253 	for (i = start_rgn; i < end_rgn; i++)
1254 		memblock_set_region_node(&type->regions[i], nid);
1255 
1256 	memblock_merge_regions(type);
1257 #endif
1258 	return 0;
1259 }
1260 
1261 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1262 /**
1263  * __next_mem_pfn_range_in_zone - iterator for for_each_*_range_in_zone()
1264  *
1265  * @idx: pointer to u64 loop variable
1266  * @zone: zone in which all of the memory blocks reside
1267  * @out_spfn: ptr to ulong for start pfn of the range, can be %NULL
1268  * @out_epfn: ptr to ulong for end pfn of the range, can be %NULL
1269  *
1270  * This function is meant to be a zone/pfn specific wrapper for the
1271  * for_each_mem_range type iterators. Specifically they are used in the
1272  * deferred memory init routines and as such we were duplicating much of
1273  * this logic throughout the code. So instead of having it in multiple
1274  * locations it seemed like it would make more sense to centralize this to
1275  * one new iterator that does everything they need.
1276  */
1277 void __init_memblock
1278 __next_mem_pfn_range_in_zone(u64 *idx, struct zone *zone,
1279 			     unsigned long *out_spfn, unsigned long *out_epfn)
1280 {
1281 	int zone_nid = zone_to_nid(zone);
1282 	phys_addr_t spa, epa;
1283 	int nid;
1284 
1285 	__next_mem_range(idx, zone_nid, MEMBLOCK_NONE,
1286 			 &memblock.memory, &memblock.reserved,
1287 			 &spa, &epa, &nid);
1288 
1289 	while (*idx != U64_MAX) {
1290 		unsigned long epfn = PFN_DOWN(epa);
1291 		unsigned long spfn = PFN_UP(spa);
1292 
1293 		/*
1294 		 * Verify the end is at least past the start of the zone and
1295 		 * that we have at least one PFN to initialize.
1296 		 */
1297 		if (zone->zone_start_pfn < epfn && spfn < epfn) {
1298 			/* if we went too far just stop searching */
1299 			if (zone_end_pfn(zone) <= spfn) {
1300 				*idx = U64_MAX;
1301 				break;
1302 			}
1303 
1304 			if (out_spfn)
1305 				*out_spfn = max(zone->zone_start_pfn, spfn);
1306 			if (out_epfn)
1307 				*out_epfn = min(zone_end_pfn(zone), epfn);
1308 
1309 			return;
1310 		}
1311 
1312 		__next_mem_range(idx, zone_nid, MEMBLOCK_NONE,
1313 				 &memblock.memory, &memblock.reserved,
1314 				 &spa, &epa, &nid);
1315 	}
1316 
1317 	/* signal end of iteration */
1318 	if (out_spfn)
1319 		*out_spfn = ULONG_MAX;
1320 	if (out_epfn)
1321 		*out_epfn = 0;
1322 }
1323 
1324 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
1325 
1326 /**
1327  * memblock_alloc_range_nid - allocate boot memory block
1328  * @size: size of memory block to be allocated in bytes
1329  * @align: alignment of the region and block's size
1330  * @start: the lower bound of the memory region to allocate (phys address)
1331  * @end: the upper bound of the memory region to allocate (phys address)
1332  * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1333  * @exact_nid: control the allocation fall back to other nodes
1334  *
1335  * The allocation is performed from memory region limited by
1336  * memblock.current_limit if @end == %MEMBLOCK_ALLOC_ACCESSIBLE.
1337  *
1338  * If the specified node can not hold the requested memory and @exact_nid
1339  * is false, the allocation falls back to any node in the system.
1340  *
1341  * For systems with memory mirroring, the allocation is attempted first
1342  * from the regions with mirroring enabled and then retried from any
1343  * memory region.
1344  *
1345  * In addition, function sets the min_count to 0 using kmemleak_alloc_phys for
1346  * allocated boot memory block, so that it is never reported as leaks.
1347  *
1348  * Return:
1349  * Physical address of allocated memory block on success, %0 on failure.
1350  */
1351 phys_addr_t __init memblock_alloc_range_nid(phys_addr_t size,
1352 					phys_addr_t align, phys_addr_t start,
1353 					phys_addr_t end, int nid,
1354 					bool exact_nid)
1355 {
1356 	enum memblock_flags flags = choose_memblock_flags();
1357 	phys_addr_t found;
1358 
1359 	if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
1360 		nid = NUMA_NO_NODE;
1361 
1362 	if (!align) {
1363 		/* Can't use WARNs this early in boot on powerpc */
1364 		dump_stack();
1365 		align = SMP_CACHE_BYTES;
1366 	}
1367 
1368 again:
1369 	found = memblock_find_in_range_node(size, align, start, end, nid,
1370 					    flags);
1371 	if (found && !memblock_reserve(found, size))
1372 		goto done;
1373 
1374 	if (nid != NUMA_NO_NODE && !exact_nid) {
1375 		found = memblock_find_in_range_node(size, align, start,
1376 						    end, NUMA_NO_NODE,
1377 						    flags);
1378 		if (found && !memblock_reserve(found, size))
1379 			goto done;
1380 	}
1381 
1382 	if (flags & MEMBLOCK_MIRROR) {
1383 		flags &= ~MEMBLOCK_MIRROR;
1384 		pr_warn("Could not allocate %pap bytes of mirrored memory\n",
1385 			&size);
1386 		goto again;
1387 	}
1388 
1389 	return 0;
1390 
1391 done:
1392 	/* Skip kmemleak for kasan_init() due to high volume. */
1393 	if (end != MEMBLOCK_ALLOC_KASAN)
1394 		/*
1395 		 * The min_count is set to 0 so that memblock allocated
1396 		 * blocks are never reported as leaks. This is because many
1397 		 * of these blocks are only referred via the physical
1398 		 * address which is not looked up by kmemleak.
1399 		 */
1400 		kmemleak_alloc_phys(found, size, 0, 0);
1401 
1402 	return found;
1403 }
1404 
1405 /**
1406  * memblock_phys_alloc_range - allocate a memory block inside specified range
1407  * @size: size of memory block to be allocated in bytes
1408  * @align: alignment of the region and block's size
1409  * @start: the lower bound of the memory region to allocate (physical address)
1410  * @end: the upper bound of the memory region to allocate (physical address)
1411  *
1412  * Allocate @size bytes in the between @start and @end.
1413  *
1414  * Return: physical address of the allocated memory block on success,
1415  * %0 on failure.
1416  */
1417 phys_addr_t __init memblock_phys_alloc_range(phys_addr_t size,
1418 					     phys_addr_t align,
1419 					     phys_addr_t start,
1420 					     phys_addr_t end)
1421 {
1422 	memblock_dbg("%s: %llu bytes align=0x%llx from=%pa max_addr=%pa %pS\n",
1423 		     __func__, (u64)size, (u64)align, &start, &end,
1424 		     (void *)_RET_IP_);
1425 	return memblock_alloc_range_nid(size, align, start, end, NUMA_NO_NODE,
1426 					false);
1427 }
1428 
1429 /**
1430  * memblock_phys_alloc_try_nid - allocate a memory block from specified NUMA node
1431  * @size: size of memory block to be allocated in bytes
1432  * @align: alignment of the region and block's size
1433  * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1434  *
1435  * Allocates memory block from the specified NUMA node. If the node
1436  * has no available memory, attempts to allocated from any node in the
1437  * system.
1438  *
1439  * Return: physical address of the allocated memory block on success,
1440  * %0 on failure.
1441  */
1442 phys_addr_t __init memblock_phys_alloc_try_nid(phys_addr_t size, phys_addr_t align, int nid)
1443 {
1444 	return memblock_alloc_range_nid(size, align, 0,
1445 					MEMBLOCK_ALLOC_ACCESSIBLE, nid, false);
1446 }
1447 
1448 /**
1449  * memblock_alloc_internal - allocate boot memory block
1450  * @size: size of memory block to be allocated in bytes
1451  * @align: alignment of the region and block's size
1452  * @min_addr: the lower bound of the memory region to allocate (phys address)
1453  * @max_addr: the upper bound of the memory region to allocate (phys address)
1454  * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1455  * @exact_nid: control the allocation fall back to other nodes
1456  *
1457  * Allocates memory block using memblock_alloc_range_nid() and
1458  * converts the returned physical address to virtual.
1459  *
1460  * The @min_addr limit is dropped if it can not be satisfied and the allocation
1461  * will fall back to memory below @min_addr. Other constraints, such
1462  * as node and mirrored memory will be handled again in
1463  * memblock_alloc_range_nid().
1464  *
1465  * Return:
1466  * Virtual address of allocated memory block on success, NULL on failure.
1467  */
1468 static void * __init memblock_alloc_internal(
1469 				phys_addr_t size, phys_addr_t align,
1470 				phys_addr_t min_addr, phys_addr_t max_addr,
1471 				int nid, bool exact_nid)
1472 {
1473 	phys_addr_t alloc;
1474 
1475 	/*
1476 	 * Detect any accidental use of these APIs after slab is ready, as at
1477 	 * this moment memblock may be deinitialized already and its
1478 	 * internal data may be destroyed (after execution of memblock_free_all)
1479 	 */
1480 	if (WARN_ON_ONCE(slab_is_available()))
1481 		return kzalloc_node(size, GFP_NOWAIT, nid);
1482 
1483 	if (max_addr > memblock.current_limit)
1484 		max_addr = memblock.current_limit;
1485 
1486 	alloc = memblock_alloc_range_nid(size, align, min_addr, max_addr, nid,
1487 					exact_nid);
1488 
1489 	/* retry allocation without lower limit */
1490 	if (!alloc && min_addr)
1491 		alloc = memblock_alloc_range_nid(size, align, 0, max_addr, nid,
1492 						exact_nid);
1493 
1494 	if (!alloc)
1495 		return NULL;
1496 
1497 	return phys_to_virt(alloc);
1498 }
1499 
1500 /**
1501  * memblock_alloc_exact_nid_raw - allocate boot memory block on the exact node
1502  * without zeroing memory
1503  * @size: size of memory block to be allocated in bytes
1504  * @align: alignment of the region and block's size
1505  * @min_addr: the lower bound of the memory region from where the allocation
1506  *	  is preferred (phys address)
1507  * @max_addr: the upper bound of the memory region from where the allocation
1508  *	      is preferred (phys address), or %MEMBLOCK_ALLOC_ACCESSIBLE to
1509  *	      allocate only from memory limited by memblock.current_limit value
1510  * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1511  *
1512  * Public function, provides additional debug information (including caller
1513  * info), if enabled. Does not zero allocated memory.
1514  *
1515  * Return:
1516  * Virtual address of allocated memory block on success, NULL on failure.
1517  */
1518 void * __init memblock_alloc_exact_nid_raw(
1519 			phys_addr_t size, phys_addr_t align,
1520 			phys_addr_t min_addr, phys_addr_t max_addr,
1521 			int nid)
1522 {
1523 	void *ptr;
1524 
1525 	memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=%pa max_addr=%pa %pS\n",
1526 		     __func__, (u64)size, (u64)align, nid, &min_addr,
1527 		     &max_addr, (void *)_RET_IP_);
1528 
1529 	ptr = memblock_alloc_internal(size, align,
1530 					   min_addr, max_addr, nid, true);
1531 	if (ptr && size > 0)
1532 		page_init_poison(ptr, size);
1533 
1534 	return ptr;
1535 }
1536 
1537 /**
1538  * memblock_alloc_try_nid_raw - allocate boot memory block without zeroing
1539  * memory and without panicking
1540  * @size: size of memory block to be allocated in bytes
1541  * @align: alignment of the region and block's size
1542  * @min_addr: the lower bound of the memory region from where the allocation
1543  *	  is preferred (phys address)
1544  * @max_addr: the upper bound of the memory region from where the allocation
1545  *	      is preferred (phys address), or %MEMBLOCK_ALLOC_ACCESSIBLE to
1546  *	      allocate only from memory limited by memblock.current_limit value
1547  * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1548  *
1549  * Public function, provides additional debug information (including caller
1550  * info), if enabled. Does not zero allocated memory, does not panic if request
1551  * cannot be satisfied.
1552  *
1553  * Return:
1554  * Virtual address of allocated memory block on success, NULL on failure.
1555  */
1556 void * __init memblock_alloc_try_nid_raw(
1557 			phys_addr_t size, phys_addr_t align,
1558 			phys_addr_t min_addr, phys_addr_t max_addr,
1559 			int nid)
1560 {
1561 	void *ptr;
1562 
1563 	memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=%pa max_addr=%pa %pS\n",
1564 		     __func__, (u64)size, (u64)align, nid, &min_addr,
1565 		     &max_addr, (void *)_RET_IP_);
1566 
1567 	ptr = memblock_alloc_internal(size, align,
1568 					   min_addr, max_addr, nid, false);
1569 	if (ptr && size > 0)
1570 		page_init_poison(ptr, size);
1571 
1572 	return ptr;
1573 }
1574 
1575 /**
1576  * memblock_alloc_try_nid - allocate boot memory block
1577  * @size: size of memory block to be allocated in bytes
1578  * @align: alignment of the region and block's size
1579  * @min_addr: the lower bound of the memory region from where the allocation
1580  *	  is preferred (phys address)
1581  * @max_addr: the upper bound of the memory region from where the allocation
1582  *	      is preferred (phys address), or %MEMBLOCK_ALLOC_ACCESSIBLE to
1583  *	      allocate only from memory limited by memblock.current_limit value
1584  * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1585  *
1586  * Public function, provides additional debug information (including caller
1587  * info), if enabled. This function zeroes the allocated memory.
1588  *
1589  * Return:
1590  * Virtual address of allocated memory block on success, NULL on failure.
1591  */
1592 void * __init memblock_alloc_try_nid(
1593 			phys_addr_t size, phys_addr_t align,
1594 			phys_addr_t min_addr, phys_addr_t max_addr,
1595 			int nid)
1596 {
1597 	void *ptr;
1598 
1599 	memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=%pa max_addr=%pa %pS\n",
1600 		     __func__, (u64)size, (u64)align, nid, &min_addr,
1601 		     &max_addr, (void *)_RET_IP_);
1602 	ptr = memblock_alloc_internal(size, align,
1603 					   min_addr, max_addr, nid, false);
1604 	if (ptr)
1605 		memset(ptr, 0, size);
1606 
1607 	return ptr;
1608 }
1609 
1610 /**
1611  * __memblock_free_late - free pages directly to buddy allocator
1612  * @base: phys starting address of the  boot memory block
1613  * @size: size of the boot memory block in bytes
1614  *
1615  * This is only useful when the memblock allocator has already been torn
1616  * down, but we are still initializing the system.  Pages are released directly
1617  * to the buddy allocator.
1618  */
1619 void __init __memblock_free_late(phys_addr_t base, phys_addr_t size)
1620 {
1621 	phys_addr_t cursor, end;
1622 
1623 	end = base + size - 1;
1624 	memblock_dbg("%s: [%pa-%pa] %pS\n",
1625 		     __func__, &base, &end, (void *)_RET_IP_);
1626 	kmemleak_free_part_phys(base, size);
1627 	cursor = PFN_UP(base);
1628 	end = PFN_DOWN(base + size);
1629 
1630 	for (; cursor < end; cursor++) {
1631 		memblock_free_pages(pfn_to_page(cursor), cursor, 0);
1632 		totalram_pages_inc();
1633 	}
1634 }
1635 
1636 /*
1637  * Remaining API functions
1638  */
1639 
1640 phys_addr_t __init_memblock memblock_phys_mem_size(void)
1641 {
1642 	return memblock.memory.total_size;
1643 }
1644 
1645 phys_addr_t __init_memblock memblock_reserved_size(void)
1646 {
1647 	return memblock.reserved.total_size;
1648 }
1649 
1650 /* lowest address */
1651 phys_addr_t __init_memblock memblock_start_of_DRAM(void)
1652 {
1653 	return memblock.memory.regions[0].base;
1654 }
1655 
1656 phys_addr_t __init_memblock memblock_end_of_DRAM(void)
1657 {
1658 	int idx = memblock.memory.cnt - 1;
1659 
1660 	return (memblock.memory.regions[idx].base + memblock.memory.regions[idx].size);
1661 }
1662 
1663 static phys_addr_t __init_memblock __find_max_addr(phys_addr_t limit)
1664 {
1665 	phys_addr_t max_addr = PHYS_ADDR_MAX;
1666 	struct memblock_region *r;
1667 
1668 	/*
1669 	 * translate the memory @limit size into the max address within one of
1670 	 * the memory memblock regions, if the @limit exceeds the total size
1671 	 * of those regions, max_addr will keep original value PHYS_ADDR_MAX
1672 	 */
1673 	for_each_mem_region(r) {
1674 		if (limit <= r->size) {
1675 			max_addr = r->base + limit;
1676 			break;
1677 		}
1678 		limit -= r->size;
1679 	}
1680 
1681 	return max_addr;
1682 }
1683 
1684 void __init memblock_enforce_memory_limit(phys_addr_t limit)
1685 {
1686 	phys_addr_t max_addr;
1687 
1688 	if (!limit)
1689 		return;
1690 
1691 	max_addr = __find_max_addr(limit);
1692 
1693 	/* @limit exceeds the total size of the memory, do nothing */
1694 	if (max_addr == PHYS_ADDR_MAX)
1695 		return;
1696 
1697 	/* truncate both memory and reserved regions */
1698 	memblock_remove_range(&memblock.memory, max_addr,
1699 			      PHYS_ADDR_MAX);
1700 	memblock_remove_range(&memblock.reserved, max_addr,
1701 			      PHYS_ADDR_MAX);
1702 }
1703 
1704 void __init memblock_cap_memory_range(phys_addr_t base, phys_addr_t size)
1705 {
1706 	int start_rgn, end_rgn;
1707 	int i, ret;
1708 
1709 	if (!size)
1710 		return;
1711 
1712 	ret = memblock_isolate_range(&memblock.memory, base, size,
1713 						&start_rgn, &end_rgn);
1714 	if (ret)
1715 		return;
1716 
1717 	/* remove all the MAP regions */
1718 	for (i = memblock.memory.cnt - 1; i >= end_rgn; i--)
1719 		if (!memblock_is_nomap(&memblock.memory.regions[i]))
1720 			memblock_remove_region(&memblock.memory, i);
1721 
1722 	for (i = start_rgn - 1; i >= 0; i--)
1723 		if (!memblock_is_nomap(&memblock.memory.regions[i]))
1724 			memblock_remove_region(&memblock.memory, i);
1725 
1726 	/* truncate the reserved regions */
1727 	memblock_remove_range(&memblock.reserved, 0, base);
1728 	memblock_remove_range(&memblock.reserved,
1729 			base + size, PHYS_ADDR_MAX);
1730 }
1731 
1732 void __init memblock_mem_limit_remove_map(phys_addr_t limit)
1733 {
1734 	phys_addr_t max_addr;
1735 
1736 	if (!limit)
1737 		return;
1738 
1739 	max_addr = __find_max_addr(limit);
1740 
1741 	/* @limit exceeds the total size of the memory, do nothing */
1742 	if (max_addr == PHYS_ADDR_MAX)
1743 		return;
1744 
1745 	memblock_cap_memory_range(0, max_addr);
1746 }
1747 
1748 static int __init_memblock memblock_search(struct memblock_type *type, phys_addr_t addr)
1749 {
1750 	unsigned int left = 0, right = type->cnt;
1751 
1752 	do {
1753 		unsigned int mid = (right + left) / 2;
1754 
1755 		if (addr < type->regions[mid].base)
1756 			right = mid;
1757 		else if (addr >= (type->regions[mid].base +
1758 				  type->regions[mid].size))
1759 			left = mid + 1;
1760 		else
1761 			return mid;
1762 	} while (left < right);
1763 	return -1;
1764 }
1765 
1766 bool __init_memblock memblock_is_reserved(phys_addr_t addr)
1767 {
1768 	return memblock_search(&memblock.reserved, addr) != -1;
1769 }
1770 
1771 bool __init_memblock memblock_is_memory(phys_addr_t addr)
1772 {
1773 	return memblock_search(&memblock.memory, addr) != -1;
1774 }
1775 
1776 bool __init_memblock memblock_is_map_memory(phys_addr_t addr)
1777 {
1778 	int i = memblock_search(&memblock.memory, addr);
1779 
1780 	if (i == -1)
1781 		return false;
1782 	return !memblock_is_nomap(&memblock.memory.regions[i]);
1783 }
1784 
1785 int __init_memblock memblock_search_pfn_nid(unsigned long pfn,
1786 			 unsigned long *start_pfn, unsigned long *end_pfn)
1787 {
1788 	struct memblock_type *type = &memblock.memory;
1789 	int mid = memblock_search(type, PFN_PHYS(pfn));
1790 
1791 	if (mid == -1)
1792 		return -1;
1793 
1794 	*start_pfn = PFN_DOWN(type->regions[mid].base);
1795 	*end_pfn = PFN_DOWN(type->regions[mid].base + type->regions[mid].size);
1796 
1797 	return memblock_get_region_node(&type->regions[mid]);
1798 }
1799 
1800 /**
1801  * memblock_is_region_memory - check if a region is a subset of memory
1802  * @base: base of region to check
1803  * @size: size of region to check
1804  *
1805  * Check if the region [@base, @base + @size) is a subset of a memory block.
1806  *
1807  * Return:
1808  * 0 if false, non-zero if true
1809  */
1810 bool __init_memblock memblock_is_region_memory(phys_addr_t base, phys_addr_t size)
1811 {
1812 	int idx = memblock_search(&memblock.memory, base);
1813 	phys_addr_t end = base + memblock_cap_size(base, &size);
1814 
1815 	if (idx == -1)
1816 		return false;
1817 	return (memblock.memory.regions[idx].base +
1818 		 memblock.memory.regions[idx].size) >= end;
1819 }
1820 
1821 /**
1822  * memblock_is_region_reserved - check if a region intersects reserved memory
1823  * @base: base of region to check
1824  * @size: size of region to check
1825  *
1826  * Check if the region [@base, @base + @size) intersects a reserved
1827  * memory block.
1828  *
1829  * Return:
1830  * True if they intersect, false if not.
1831  */
1832 bool __init_memblock memblock_is_region_reserved(phys_addr_t base, phys_addr_t size)
1833 {
1834 	memblock_cap_size(base, &size);
1835 	return memblock_overlaps_region(&memblock.reserved, base, size);
1836 }
1837 
1838 void __init_memblock memblock_trim_memory(phys_addr_t align)
1839 {
1840 	phys_addr_t start, end, orig_start, orig_end;
1841 	struct memblock_region *r;
1842 
1843 	for_each_mem_region(r) {
1844 		orig_start = r->base;
1845 		orig_end = r->base + r->size;
1846 		start = round_up(orig_start, align);
1847 		end = round_down(orig_end, align);
1848 
1849 		if (start == orig_start && end == orig_end)
1850 			continue;
1851 
1852 		if (start < end) {
1853 			r->base = start;
1854 			r->size = end - start;
1855 		} else {
1856 			memblock_remove_region(&memblock.memory,
1857 					       r - memblock.memory.regions);
1858 			r--;
1859 		}
1860 	}
1861 }
1862 
1863 void __init_memblock memblock_set_current_limit(phys_addr_t limit)
1864 {
1865 	memblock.current_limit = limit;
1866 }
1867 
1868 phys_addr_t __init_memblock memblock_get_current_limit(void)
1869 {
1870 	return memblock.current_limit;
1871 }
1872 
1873 static void __init_memblock memblock_dump(struct memblock_type *type)
1874 {
1875 	phys_addr_t base, end, size;
1876 	enum memblock_flags flags;
1877 	int idx;
1878 	struct memblock_region *rgn;
1879 
1880 	pr_info(" %s.cnt  = 0x%lx\n", type->name, type->cnt);
1881 
1882 	for_each_memblock_type(idx, type, rgn) {
1883 		char nid_buf[32] = "";
1884 
1885 		base = rgn->base;
1886 		size = rgn->size;
1887 		end = base + size - 1;
1888 		flags = rgn->flags;
1889 #ifdef CONFIG_NEED_MULTIPLE_NODES
1890 		if (memblock_get_region_node(rgn) != MAX_NUMNODES)
1891 			snprintf(nid_buf, sizeof(nid_buf), " on node %d",
1892 				 memblock_get_region_node(rgn));
1893 #endif
1894 		pr_info(" %s[%#x]\t[%pa-%pa], %pa bytes%s flags: %#x\n",
1895 			type->name, idx, &base, &end, &size, nid_buf, flags);
1896 	}
1897 }
1898 
1899 static void __init_memblock __memblock_dump_all(void)
1900 {
1901 	pr_info("MEMBLOCK configuration:\n");
1902 	pr_info(" memory size = %pa reserved size = %pa\n",
1903 		&memblock.memory.total_size,
1904 		&memblock.reserved.total_size);
1905 
1906 	memblock_dump(&memblock.memory);
1907 	memblock_dump(&memblock.reserved);
1908 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
1909 	memblock_dump(&physmem);
1910 #endif
1911 }
1912 
1913 void __init_memblock memblock_dump_all(void)
1914 {
1915 	if (memblock_debug)
1916 		__memblock_dump_all();
1917 }
1918 
1919 void __init memblock_allow_resize(void)
1920 {
1921 	memblock_can_resize = 1;
1922 }
1923 
1924 static int __init early_memblock(char *p)
1925 {
1926 	if (p && strstr(p, "debug"))
1927 		memblock_debug = 1;
1928 	return 0;
1929 }
1930 early_param("memblock", early_memblock);
1931 
1932 static void __init free_memmap(unsigned long start_pfn, unsigned long end_pfn)
1933 {
1934 	struct page *start_pg, *end_pg;
1935 	phys_addr_t pg, pgend;
1936 
1937 	/*
1938 	 * Convert start_pfn/end_pfn to a struct page pointer.
1939 	 */
1940 	start_pg = pfn_to_page(start_pfn - 1) + 1;
1941 	end_pg = pfn_to_page(end_pfn - 1) + 1;
1942 
1943 	/*
1944 	 * Convert to physical addresses, and round start upwards and end
1945 	 * downwards.
1946 	 */
1947 	pg = PAGE_ALIGN(__pa(start_pg));
1948 	pgend = __pa(end_pg) & PAGE_MASK;
1949 
1950 	/*
1951 	 * If there are free pages between these, free the section of the
1952 	 * memmap array.
1953 	 */
1954 	if (pg < pgend)
1955 		memblock_free(pg, pgend - pg);
1956 }
1957 
1958 /*
1959  * The mem_map array can get very big.  Free the unused area of the memory map.
1960  */
1961 static void __init free_unused_memmap(void)
1962 {
1963 	unsigned long start, end, prev_end = 0;
1964 	int i;
1965 
1966 	if (!IS_ENABLED(CONFIG_HAVE_ARCH_PFN_VALID) ||
1967 	    IS_ENABLED(CONFIG_SPARSEMEM_VMEMMAP))
1968 		return;
1969 
1970 	/*
1971 	 * This relies on each bank being in address order.
1972 	 * The banks are sorted previously in bootmem_init().
1973 	 */
1974 	for_each_mem_pfn_range(i, MAX_NUMNODES, &start, &end, NULL) {
1975 #ifdef CONFIG_SPARSEMEM
1976 		/*
1977 		 * Take care not to free memmap entries that don't exist
1978 		 * due to SPARSEMEM sections which aren't present.
1979 		 */
1980 		start = min(start, ALIGN(prev_end, PAGES_PER_SECTION));
1981 #else
1982 		/*
1983 		 * Align down here since the VM subsystem insists that the
1984 		 * memmap entries are valid from the bank start aligned to
1985 		 * MAX_ORDER_NR_PAGES.
1986 		 */
1987 		start = round_down(start, MAX_ORDER_NR_PAGES);
1988 #endif
1989 
1990 		/*
1991 		 * If we had a previous bank, and there is a space
1992 		 * between the current bank and the previous, free it.
1993 		 */
1994 		if (prev_end && prev_end < start)
1995 			free_memmap(prev_end, start);
1996 
1997 		/*
1998 		 * Align up here since the VM subsystem insists that the
1999 		 * memmap entries are valid from the bank end aligned to
2000 		 * MAX_ORDER_NR_PAGES.
2001 		 */
2002 		prev_end = ALIGN(end, MAX_ORDER_NR_PAGES);
2003 	}
2004 
2005 #ifdef CONFIG_SPARSEMEM
2006 	if (!IS_ALIGNED(prev_end, PAGES_PER_SECTION))
2007 		free_memmap(prev_end, ALIGN(prev_end, PAGES_PER_SECTION));
2008 #endif
2009 }
2010 
2011 static void __init __free_pages_memory(unsigned long start, unsigned long end)
2012 {
2013 	int order;
2014 
2015 	while (start < end) {
2016 		order = min(MAX_ORDER - 1UL, __ffs(start));
2017 
2018 		while (start + (1UL << order) > end)
2019 			order--;
2020 
2021 		memblock_free_pages(pfn_to_page(start), start, order);
2022 
2023 		start += (1UL << order);
2024 	}
2025 }
2026 
2027 static unsigned long __init __free_memory_core(phys_addr_t start,
2028 				 phys_addr_t end)
2029 {
2030 	unsigned long start_pfn = PFN_UP(start);
2031 	unsigned long end_pfn = min_t(unsigned long,
2032 				      PFN_DOWN(end), max_low_pfn);
2033 
2034 	if (start_pfn >= end_pfn)
2035 		return 0;
2036 
2037 	__free_pages_memory(start_pfn, end_pfn);
2038 
2039 	return end_pfn - start_pfn;
2040 }
2041 
2042 static unsigned long __init free_low_memory_core_early(void)
2043 {
2044 	unsigned long count = 0;
2045 	phys_addr_t start, end;
2046 	u64 i;
2047 
2048 	memblock_clear_hotplug(0, -1);
2049 
2050 	for_each_reserved_mem_range(i, &start, &end)
2051 		reserve_bootmem_region(start, end);
2052 
2053 	/*
2054 	 * We need to use NUMA_NO_NODE instead of NODE_DATA(0)->node_id
2055 	 *  because in some case like Node0 doesn't have RAM installed
2056 	 *  low ram will be on Node1
2057 	 */
2058 	for_each_free_mem_range(i, NUMA_NO_NODE, MEMBLOCK_NONE, &start, &end,
2059 				NULL)
2060 		count += __free_memory_core(start, end);
2061 
2062 	return count;
2063 }
2064 
2065 static int reset_managed_pages_done __initdata;
2066 
2067 void reset_node_managed_pages(pg_data_t *pgdat)
2068 {
2069 	struct zone *z;
2070 
2071 	for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++)
2072 		atomic_long_set(&z->managed_pages, 0);
2073 }
2074 
2075 void __init reset_all_zones_managed_pages(void)
2076 {
2077 	struct pglist_data *pgdat;
2078 
2079 	if (reset_managed_pages_done)
2080 		return;
2081 
2082 	for_each_online_pgdat(pgdat)
2083 		reset_node_managed_pages(pgdat);
2084 
2085 	reset_managed_pages_done = 1;
2086 }
2087 
2088 /**
2089  * memblock_free_all - release free pages to the buddy allocator
2090  *
2091  * Return: the number of pages actually released.
2092  */
2093 unsigned long __init memblock_free_all(void)
2094 {
2095 	unsigned long pages;
2096 
2097 	free_unused_memmap();
2098 	reset_all_zones_managed_pages();
2099 
2100 	pages = free_low_memory_core_early();
2101 	totalram_pages_add(pages);
2102 
2103 	return pages;
2104 }
2105 
2106 #if defined(CONFIG_DEBUG_FS) && defined(CONFIG_ARCH_KEEP_MEMBLOCK)
2107 
2108 static int memblock_debug_show(struct seq_file *m, void *private)
2109 {
2110 	struct memblock_type *type = m->private;
2111 	struct memblock_region *reg;
2112 	int i;
2113 	phys_addr_t end;
2114 
2115 	for (i = 0; i < type->cnt; i++) {
2116 		reg = &type->regions[i];
2117 		end = reg->base + reg->size - 1;
2118 
2119 		seq_printf(m, "%4d: ", i);
2120 		seq_printf(m, "%pa..%pa\n", &reg->base, &end);
2121 	}
2122 	return 0;
2123 }
2124 DEFINE_SHOW_ATTRIBUTE(memblock_debug);
2125 
2126 static int __init memblock_init_debugfs(void)
2127 {
2128 	struct dentry *root = debugfs_create_dir("memblock", NULL);
2129 
2130 	debugfs_create_file("memory", 0444, root,
2131 			    &memblock.memory, &memblock_debug_fops);
2132 	debugfs_create_file("reserved", 0444, root,
2133 			    &memblock.reserved, &memblock_debug_fops);
2134 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
2135 	debugfs_create_file("physmem", 0444, root, &physmem,
2136 			    &memblock_debug_fops);
2137 #endif
2138 
2139 	return 0;
2140 }
2141 __initcall(memblock_init_debugfs);
2142 
2143 #endif /* CONFIG_DEBUG_FS */
2144