xref: /openbmc/linux/mm/memblock.c (revision b021d05e)
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 #ifndef INIT_MEMBLOCK_MEMORY_REGIONS
33 #define INIT_MEMBLOCK_MEMORY_REGIONS		INIT_MEMBLOCK_REGIONS
34 #endif
35 
36 /**
37  * DOC: memblock overview
38  *
39  * Memblock is a method of managing memory regions during the early
40  * boot period when the usual kernel memory allocators are not up and
41  * running.
42  *
43  * Memblock views the system memory as collections of contiguous
44  * regions. There are several types of these collections:
45  *
46  * * ``memory`` - describes the physical memory available to the
47  *   kernel; this may differ from the actual physical memory installed
48  *   in the system, for instance when the memory is restricted with
49  *   ``mem=`` command line parameter
50  * * ``reserved`` - describes the regions that were allocated
51  * * ``physmem`` - describes the actual physical memory available during
52  *   boot regardless of the possible restrictions and memory hot(un)plug;
53  *   the ``physmem`` type is only available on some architectures.
54  *
55  * Each region is represented by struct memblock_region that
56  * defines the region extents, its attributes and NUMA node id on NUMA
57  * systems. Every memory type is described by the struct memblock_type
58  * which contains an array of memory regions along with
59  * the allocator metadata. The "memory" and "reserved" types are nicely
60  * wrapped with struct memblock. This structure is statically
61  * initialized at build time. The region arrays are initially sized to
62  * %INIT_MEMBLOCK_MEMORY_REGIONS for "memory" and
63  * %INIT_MEMBLOCK_RESERVED_REGIONS for "reserved". The region array
64  * for "physmem" is initially sized to %INIT_PHYSMEM_REGIONS.
65  * The memblock_allow_resize() enables automatic resizing of the region
66  * arrays during addition of new regions. This feature should be used
67  * with care so that memory allocated for the region array will not
68  * overlap with areas that should be reserved, for example initrd.
69  *
70  * The early architecture setup should tell memblock what the physical
71  * memory layout is by using memblock_add() or memblock_add_node()
72  * functions. The first function does not assign the region to a NUMA
73  * node and it is appropriate for UMA systems. Yet, it is possible to
74  * use it on NUMA systems as well and assign the region to a NUMA node
75  * later in the setup process using memblock_set_node(). The
76  * memblock_add_node() performs such an assignment directly.
77  *
78  * Once memblock is setup the memory can be allocated using one of the
79  * API variants:
80  *
81  * * memblock_phys_alloc*() - these functions return the **physical**
82  *   address of the allocated memory
83  * * memblock_alloc*() - these functions return the **virtual** address
84  *   of the allocated memory.
85  *
86  * Note, that both API variants use implicit assumptions about allowed
87  * memory ranges and the fallback methods. Consult the documentation
88  * of memblock_alloc_internal() and memblock_alloc_range_nid()
89  * functions for more elaborate description.
90  *
91  * As the system boot progresses, the architecture specific mem_init()
92  * function frees all the memory to the buddy page allocator.
93  *
94  * Unless an architecture enables %CONFIG_ARCH_KEEP_MEMBLOCK, the
95  * memblock data structures (except "physmem") will be discarded after the
96  * system initialization completes.
97  */
98 
99 #ifndef CONFIG_NUMA
100 struct pglist_data __refdata contig_page_data;
101 EXPORT_SYMBOL(contig_page_data);
102 #endif
103 
104 unsigned long max_low_pfn;
105 unsigned long min_low_pfn;
106 unsigned long max_pfn;
107 unsigned long long max_possible_pfn;
108 
109 static struct memblock_region memblock_memory_init_regions[INIT_MEMBLOCK_MEMORY_REGIONS] __initdata_memblock;
110 static struct memblock_region memblock_reserved_init_regions[INIT_MEMBLOCK_RESERVED_REGIONS] __initdata_memblock;
111 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
112 static struct memblock_region memblock_physmem_init_regions[INIT_PHYSMEM_REGIONS];
113 #endif
114 
115 struct memblock memblock __initdata_memblock = {
116 	.memory.regions		= memblock_memory_init_regions,
117 	.memory.cnt		= 1,	/* empty dummy entry */
118 	.memory.max		= INIT_MEMBLOCK_MEMORY_REGIONS,
119 	.memory.name		= "memory",
120 
121 	.reserved.regions	= memblock_reserved_init_regions,
122 	.reserved.cnt		= 1,	/* empty dummy entry */
123 	.reserved.max		= INIT_MEMBLOCK_RESERVED_REGIONS,
124 	.reserved.name		= "reserved",
125 
126 	.bottom_up		= false,
127 	.current_limit		= MEMBLOCK_ALLOC_ANYWHERE,
128 };
129 
130 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
131 struct memblock_type physmem = {
132 	.regions		= memblock_physmem_init_regions,
133 	.cnt			= 1,	/* empty dummy entry */
134 	.max			= INIT_PHYSMEM_REGIONS,
135 	.name			= "physmem",
136 };
137 #endif
138 
139 /*
140  * keep a pointer to &memblock.memory in the text section to use it in
141  * __next_mem_range() and its helpers.
142  *  For architectures that do not keep memblock data after init, this
143  * pointer will be reset to NULL at memblock_discard()
144  */
145 static __refdata struct memblock_type *memblock_memory = &memblock.memory;
146 
147 #define for_each_memblock_type(i, memblock_type, rgn)			\
148 	for (i = 0, rgn = &memblock_type->regions[0];			\
149 	     i < memblock_type->cnt;					\
150 	     i++, rgn = &memblock_type->regions[i])
151 
152 #define memblock_dbg(fmt, ...)						\
153 	do {								\
154 		if (memblock_debug)					\
155 			pr_info(fmt, ##__VA_ARGS__);			\
156 	} while (0)
157 
158 static int memblock_debug __initdata_memblock;
159 static bool system_has_some_mirror __initdata_memblock;
160 static int memblock_can_resize __initdata_memblock;
161 static int memblock_memory_in_slab __initdata_memblock;
162 static int memblock_reserved_in_slab __initdata_memblock;
163 
164 static enum memblock_flags __init_memblock choose_memblock_flags(void)
165 {
166 	return system_has_some_mirror ? MEMBLOCK_MIRROR : MEMBLOCK_NONE;
167 }
168 
169 /* adjust *@size so that (@base + *@size) doesn't overflow, return new size */
170 static inline phys_addr_t memblock_cap_size(phys_addr_t base, phys_addr_t *size)
171 {
172 	return *size = min(*size, PHYS_ADDR_MAX - base);
173 }
174 
175 /*
176  * Address comparison utilities
177  */
178 static unsigned long __init_memblock memblock_addrs_overlap(phys_addr_t base1, phys_addr_t size1,
179 				       phys_addr_t base2, phys_addr_t size2)
180 {
181 	return ((base1 < (base2 + size2)) && (base2 < (base1 + size1)));
182 }
183 
184 bool __init_memblock memblock_overlaps_region(struct memblock_type *type,
185 					phys_addr_t base, phys_addr_t size)
186 {
187 	unsigned long i;
188 
189 	memblock_cap_size(base, &size);
190 
191 	for (i = 0; i < type->cnt; i++)
192 		if (memblock_addrs_overlap(base, size, type->regions[i].base,
193 					   type->regions[i].size))
194 			break;
195 	return i < type->cnt;
196 }
197 
198 /**
199  * __memblock_find_range_bottom_up - find free area utility in bottom-up
200  * @start: start of candidate range
201  * @end: end of candidate range, can be %MEMBLOCK_ALLOC_ANYWHERE or
202  *       %MEMBLOCK_ALLOC_ACCESSIBLE
203  * @size: size of free area to find
204  * @align: alignment of free area to find
205  * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
206  * @flags: pick from blocks based on memory attributes
207  *
208  * Utility called from memblock_find_in_range_node(), find free area bottom-up.
209  *
210  * Return:
211  * Found address on success, 0 on failure.
212  */
213 static phys_addr_t __init_memblock
214 __memblock_find_range_bottom_up(phys_addr_t start, phys_addr_t end,
215 				phys_addr_t size, phys_addr_t align, int nid,
216 				enum memblock_flags flags)
217 {
218 	phys_addr_t this_start, this_end, cand;
219 	u64 i;
220 
221 	for_each_free_mem_range(i, nid, flags, &this_start, &this_end, NULL) {
222 		this_start = clamp(this_start, start, end);
223 		this_end = clamp(this_end, start, end);
224 
225 		cand = round_up(this_start, align);
226 		if (cand < this_end && this_end - cand >= size)
227 			return cand;
228 	}
229 
230 	return 0;
231 }
232 
233 /**
234  * __memblock_find_range_top_down - find free area utility, in top-down
235  * @start: start of candidate range
236  * @end: end of candidate range, can be %MEMBLOCK_ALLOC_ANYWHERE or
237  *       %MEMBLOCK_ALLOC_ACCESSIBLE
238  * @size: size of free area to find
239  * @align: alignment of free area to find
240  * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
241  * @flags: pick from blocks based on memory attributes
242  *
243  * Utility called from memblock_find_in_range_node(), find free area top-down.
244  *
245  * Return:
246  * Found address on success, 0 on failure.
247  */
248 static phys_addr_t __init_memblock
249 __memblock_find_range_top_down(phys_addr_t start, phys_addr_t end,
250 			       phys_addr_t size, phys_addr_t align, int nid,
251 			       enum memblock_flags flags)
252 {
253 	phys_addr_t this_start, this_end, cand;
254 	u64 i;
255 
256 	for_each_free_mem_range_reverse(i, nid, flags, &this_start, &this_end,
257 					NULL) {
258 		this_start = clamp(this_start, start, end);
259 		this_end = clamp(this_end, start, end);
260 
261 		if (this_end < size)
262 			continue;
263 
264 		cand = round_down(this_end - size, align);
265 		if (cand >= this_start)
266 			return cand;
267 	}
268 
269 	return 0;
270 }
271 
272 /**
273  * memblock_find_in_range_node - find free area in given range and node
274  * @size: size of free area to find
275  * @align: alignment of free area to find
276  * @start: start of candidate range
277  * @end: end of candidate range, can be %MEMBLOCK_ALLOC_ANYWHERE or
278  *       %MEMBLOCK_ALLOC_ACCESSIBLE
279  * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
280  * @flags: pick from blocks based on memory attributes
281  *
282  * Find @size free area aligned to @align in the specified range and node.
283  *
284  * Return:
285  * Found address on success, 0 on failure.
286  */
287 static phys_addr_t __init_memblock memblock_find_in_range_node(phys_addr_t size,
288 					phys_addr_t align, phys_addr_t start,
289 					phys_addr_t end, int nid,
290 					enum memblock_flags flags)
291 {
292 	/* pump up @end */
293 	if (end == MEMBLOCK_ALLOC_ACCESSIBLE ||
294 	    end == MEMBLOCK_ALLOC_NOLEAKTRACE)
295 		end = memblock.current_limit;
296 
297 	/* avoid allocating the first page */
298 	start = max_t(phys_addr_t, start, PAGE_SIZE);
299 	end = max(start, end);
300 
301 	if (memblock_bottom_up())
302 		return __memblock_find_range_bottom_up(start, end, size, align,
303 						       nid, flags);
304 	else
305 		return __memblock_find_range_top_down(start, end, size, align,
306 						      nid, flags);
307 }
308 
309 /**
310  * memblock_find_in_range - find free area in given range
311  * @start: start of candidate range
312  * @end: end of candidate range, can be %MEMBLOCK_ALLOC_ANYWHERE or
313  *       %MEMBLOCK_ALLOC_ACCESSIBLE
314  * @size: size of free area to find
315  * @align: alignment of free area to find
316  *
317  * Find @size free area aligned to @align in the specified range.
318  *
319  * Return:
320  * Found address on success, 0 on failure.
321  */
322 static phys_addr_t __init_memblock memblock_find_in_range(phys_addr_t start,
323 					phys_addr_t end, phys_addr_t size,
324 					phys_addr_t align)
325 {
326 	phys_addr_t ret;
327 	enum memblock_flags flags = choose_memblock_flags();
328 
329 again:
330 	ret = memblock_find_in_range_node(size, align, start, end,
331 					    NUMA_NO_NODE, flags);
332 
333 	if (!ret && (flags & MEMBLOCK_MIRROR)) {
334 		pr_warn_ratelimited("Could not allocate %pap bytes of mirrored memory\n",
335 			&size);
336 		flags &= ~MEMBLOCK_MIRROR;
337 		goto again;
338 	}
339 
340 	return ret;
341 }
342 
343 static void __init_memblock memblock_remove_region(struct memblock_type *type, unsigned long r)
344 {
345 	type->total_size -= type->regions[r].size;
346 	memmove(&type->regions[r], &type->regions[r + 1],
347 		(type->cnt - (r + 1)) * sizeof(type->regions[r]));
348 	type->cnt--;
349 
350 	/* Special case for empty arrays */
351 	if (type->cnt == 0) {
352 		WARN_ON(type->total_size != 0);
353 		type->cnt = 1;
354 		type->regions[0].base = 0;
355 		type->regions[0].size = 0;
356 		type->regions[0].flags = 0;
357 		memblock_set_region_node(&type->regions[0], MAX_NUMNODES);
358 	}
359 }
360 
361 #ifndef CONFIG_ARCH_KEEP_MEMBLOCK
362 /**
363  * memblock_discard - discard memory and reserved arrays if they were allocated
364  */
365 void __init memblock_discard(void)
366 {
367 	phys_addr_t addr, size;
368 
369 	if (memblock.reserved.regions != memblock_reserved_init_regions) {
370 		addr = __pa(memblock.reserved.regions);
371 		size = PAGE_ALIGN(sizeof(struct memblock_region) *
372 				  memblock.reserved.max);
373 		if (memblock_reserved_in_slab)
374 			kfree(memblock.reserved.regions);
375 		else
376 			memblock_free_late(addr, size);
377 	}
378 
379 	if (memblock.memory.regions != memblock_memory_init_regions) {
380 		addr = __pa(memblock.memory.regions);
381 		size = PAGE_ALIGN(sizeof(struct memblock_region) *
382 				  memblock.memory.max);
383 		if (memblock_memory_in_slab)
384 			kfree(memblock.memory.regions);
385 		else
386 			memblock_free_late(addr, size);
387 	}
388 
389 	memblock_memory = NULL;
390 }
391 #endif
392 
393 /**
394  * memblock_double_array - double the size of the memblock regions array
395  * @type: memblock type of the regions array being doubled
396  * @new_area_start: starting address of memory range to avoid overlap with
397  * @new_area_size: size of memory range to avoid overlap with
398  *
399  * Double the size of the @type regions array. If memblock is being used to
400  * allocate memory for a new reserved regions array and there is a previously
401  * allocated memory range [@new_area_start, @new_area_start + @new_area_size]
402  * waiting to be reserved, ensure the memory used by the new array does
403  * not overlap.
404  *
405  * Return:
406  * 0 on success, -1 on failure.
407  */
408 static int __init_memblock memblock_double_array(struct memblock_type *type,
409 						phys_addr_t new_area_start,
410 						phys_addr_t new_area_size)
411 {
412 	struct memblock_region *new_array, *old_array;
413 	phys_addr_t old_alloc_size, new_alloc_size;
414 	phys_addr_t old_size, new_size, addr, new_end;
415 	int use_slab = slab_is_available();
416 	int *in_slab;
417 
418 	/* We don't allow resizing until we know about the reserved regions
419 	 * of memory that aren't suitable for allocation
420 	 */
421 	if (!memblock_can_resize)
422 		return -1;
423 
424 	/* Calculate new doubled size */
425 	old_size = type->max * sizeof(struct memblock_region);
426 	new_size = old_size << 1;
427 	/*
428 	 * We need to allocated new one align to PAGE_SIZE,
429 	 *   so we can free them completely later.
430 	 */
431 	old_alloc_size = PAGE_ALIGN(old_size);
432 	new_alloc_size = PAGE_ALIGN(new_size);
433 
434 	/* Retrieve the slab flag */
435 	if (type == &memblock.memory)
436 		in_slab = &memblock_memory_in_slab;
437 	else
438 		in_slab = &memblock_reserved_in_slab;
439 
440 	/* Try to find some space for it */
441 	if (use_slab) {
442 		new_array = kmalloc(new_size, GFP_KERNEL);
443 		addr = new_array ? __pa(new_array) : 0;
444 	} else {
445 		/* only exclude range when trying to double reserved.regions */
446 		if (type != &memblock.reserved)
447 			new_area_start = new_area_size = 0;
448 
449 		addr = memblock_find_in_range(new_area_start + new_area_size,
450 						memblock.current_limit,
451 						new_alloc_size, PAGE_SIZE);
452 		if (!addr && new_area_size)
453 			addr = memblock_find_in_range(0,
454 				min(new_area_start, memblock.current_limit),
455 				new_alloc_size, PAGE_SIZE);
456 
457 		new_array = addr ? __va(addr) : NULL;
458 	}
459 	if (!addr) {
460 		pr_err("memblock: Failed to double %s array from %ld to %ld entries !\n",
461 		       type->name, type->max, type->max * 2);
462 		return -1;
463 	}
464 
465 	new_end = addr + new_size - 1;
466 	memblock_dbg("memblock: %s is doubled to %ld at [%pa-%pa]",
467 			type->name, type->max * 2, &addr, &new_end);
468 
469 	/*
470 	 * Found space, we now need to move the array over before we add the
471 	 * reserved region since it may be our reserved array itself that is
472 	 * full.
473 	 */
474 	memcpy(new_array, type->regions, old_size);
475 	memset(new_array + type->max, 0, old_size);
476 	old_array = type->regions;
477 	type->regions = new_array;
478 	type->max <<= 1;
479 
480 	/* Free old array. We needn't free it if the array is the static one */
481 	if (*in_slab)
482 		kfree(old_array);
483 	else if (old_array != memblock_memory_init_regions &&
484 		 old_array != memblock_reserved_init_regions)
485 		memblock_free(old_array, old_alloc_size);
486 
487 	/*
488 	 * Reserve the new array if that comes from the memblock.  Otherwise, we
489 	 * needn't do it
490 	 */
491 	if (!use_slab)
492 		BUG_ON(memblock_reserve(addr, new_alloc_size));
493 
494 	/* Update slab flag */
495 	*in_slab = use_slab;
496 
497 	return 0;
498 }
499 
500 /**
501  * memblock_merge_regions - merge neighboring compatible regions
502  * @type: memblock type to scan
503  * @start_rgn: start scanning from (@start_rgn - 1)
504  * @end_rgn: end scanning at (@end_rgn - 1)
505  * Scan @type and merge neighboring compatible regions in [@start_rgn - 1, @end_rgn)
506  */
507 static void __init_memblock memblock_merge_regions(struct memblock_type *type,
508 						   unsigned long start_rgn,
509 						   unsigned long end_rgn)
510 {
511 	int i = 0;
512 	if (start_rgn)
513 		i = start_rgn - 1;
514 	end_rgn = min(end_rgn, type->cnt - 1);
515 	while (i < end_rgn) {
516 		struct memblock_region *this = &type->regions[i];
517 		struct memblock_region *next = &type->regions[i + 1];
518 
519 		if (this->base + this->size != next->base ||
520 		    memblock_get_region_node(this) !=
521 		    memblock_get_region_node(next) ||
522 		    this->flags != next->flags) {
523 			BUG_ON(this->base + this->size > next->base);
524 			i++;
525 			continue;
526 		}
527 
528 		this->size += next->size;
529 		/* move forward from next + 1, index of which is i + 2 */
530 		memmove(next, next + 1, (type->cnt - (i + 2)) * sizeof(*next));
531 		type->cnt--;
532 		end_rgn--;
533 	}
534 }
535 
536 /**
537  * memblock_insert_region - insert new memblock region
538  * @type:	memblock type to insert into
539  * @idx:	index for the insertion point
540  * @base:	base address of the new region
541  * @size:	size of the new region
542  * @nid:	node id of the new region
543  * @flags:	flags of the new region
544  *
545  * Insert new memblock region [@base, @base + @size) into @type at @idx.
546  * @type must already have extra room to accommodate the new region.
547  */
548 static void __init_memblock memblock_insert_region(struct memblock_type *type,
549 						   int idx, phys_addr_t base,
550 						   phys_addr_t size,
551 						   int nid,
552 						   enum memblock_flags flags)
553 {
554 	struct memblock_region *rgn = &type->regions[idx];
555 
556 	BUG_ON(type->cnt >= type->max);
557 	memmove(rgn + 1, rgn, (type->cnt - idx) * sizeof(*rgn));
558 	rgn->base = base;
559 	rgn->size = size;
560 	rgn->flags = flags;
561 	memblock_set_region_node(rgn, nid);
562 	type->cnt++;
563 	type->total_size += size;
564 }
565 
566 /**
567  * memblock_add_range - add new memblock region
568  * @type: memblock type to add new region into
569  * @base: base address of the new region
570  * @size: size of the new region
571  * @nid: nid of the new region
572  * @flags: flags of the new region
573  *
574  * Add new memblock region [@base, @base + @size) into @type.  The new region
575  * is allowed to overlap with existing ones - overlaps don't affect already
576  * existing regions.  @type is guaranteed to be minimal (all neighbouring
577  * compatible regions are merged) after the addition.
578  *
579  * Return:
580  * 0 on success, -errno on failure.
581  */
582 static int __init_memblock memblock_add_range(struct memblock_type *type,
583 				phys_addr_t base, phys_addr_t size,
584 				int nid, enum memblock_flags flags)
585 {
586 	bool insert = false;
587 	phys_addr_t obase = base;
588 	phys_addr_t end = base + memblock_cap_size(base, &size);
589 	int idx, nr_new, start_rgn = -1, end_rgn;
590 	struct memblock_region *rgn;
591 
592 	if (!size)
593 		return 0;
594 
595 	/* special case for empty array */
596 	if (type->regions[0].size == 0) {
597 		WARN_ON(type->cnt != 1 || type->total_size);
598 		type->regions[0].base = base;
599 		type->regions[0].size = size;
600 		type->regions[0].flags = flags;
601 		memblock_set_region_node(&type->regions[0], nid);
602 		type->total_size = size;
603 		return 0;
604 	}
605 
606 	/*
607 	 * The worst case is when new range overlaps all existing regions,
608 	 * then we'll need type->cnt + 1 empty regions in @type. So if
609 	 * type->cnt * 2 + 1 is less than or equal to type->max, we know
610 	 * that there is enough empty regions in @type, and we can insert
611 	 * regions directly.
612 	 */
613 	if (type->cnt * 2 + 1 <= type->max)
614 		insert = true;
615 
616 repeat:
617 	/*
618 	 * The following is executed twice.  Once with %false @insert and
619 	 * then with %true.  The first counts the number of regions needed
620 	 * to accommodate the new area.  The second actually inserts them.
621 	 */
622 	base = obase;
623 	nr_new = 0;
624 
625 	for_each_memblock_type(idx, type, rgn) {
626 		phys_addr_t rbase = rgn->base;
627 		phys_addr_t rend = rbase + rgn->size;
628 
629 		if (rbase >= end)
630 			break;
631 		if (rend <= base)
632 			continue;
633 		/*
634 		 * @rgn overlaps.  If it separates the lower part of new
635 		 * area, insert that portion.
636 		 */
637 		if (rbase > base) {
638 #ifdef CONFIG_NUMA
639 			WARN_ON(nid != memblock_get_region_node(rgn));
640 #endif
641 			WARN_ON(flags != rgn->flags);
642 			nr_new++;
643 			if (insert) {
644 				if (start_rgn == -1)
645 					start_rgn = idx;
646 				end_rgn = idx + 1;
647 				memblock_insert_region(type, idx++, base,
648 						       rbase - base, nid,
649 						       flags);
650 			}
651 		}
652 		/* area below @rend is dealt with, forget about it */
653 		base = min(rend, end);
654 	}
655 
656 	/* insert the remaining portion */
657 	if (base < end) {
658 		nr_new++;
659 		if (insert) {
660 			if (start_rgn == -1)
661 				start_rgn = idx;
662 			end_rgn = idx + 1;
663 			memblock_insert_region(type, idx, base, end - base,
664 					       nid, flags);
665 		}
666 	}
667 
668 	if (!nr_new)
669 		return 0;
670 
671 	/*
672 	 * If this was the first round, resize array and repeat for actual
673 	 * insertions; otherwise, merge and return.
674 	 */
675 	if (!insert) {
676 		while (type->cnt + nr_new > type->max)
677 			if (memblock_double_array(type, obase, size) < 0)
678 				return -ENOMEM;
679 		insert = true;
680 		goto repeat;
681 	} else {
682 		memblock_merge_regions(type, start_rgn, end_rgn);
683 		return 0;
684 	}
685 }
686 
687 /**
688  * memblock_add_node - add new memblock region within a NUMA node
689  * @base: base address of the new region
690  * @size: size of the new region
691  * @nid: nid of the new region
692  * @flags: flags 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, enum memblock_flags flags)
702 {
703 	phys_addr_t end = base + size - 1;
704 
705 	memblock_dbg("%s: [%pa-%pa] nid=%d flags=%x %pS\n", __func__,
706 		     &base, &end, nid, flags, (void *)_RET_IP_);
707 
708 	return memblock_add_range(&memblock.memory, base, size, nid, flags);
709 }
710 
711 /**
712  * memblock_add - add new memblock region
713  * @base: base address of the new region
714  * @size: size of the new region
715  *
716  * Add new memblock region [@base, @base + @size) to the "memory"
717  * type. See memblock_add_range() description for mode details
718  *
719  * Return:
720  * 0 on success, -errno on failure.
721  */
722 int __init_memblock memblock_add(phys_addr_t base, phys_addr_t size)
723 {
724 	phys_addr_t end = base + size - 1;
725 
726 	memblock_dbg("%s: [%pa-%pa] %pS\n", __func__,
727 		     &base, &end, (void *)_RET_IP_);
728 
729 	return memblock_add_range(&memblock.memory, base, size, MAX_NUMNODES, 0);
730 }
731 
732 /**
733  * memblock_isolate_range - isolate given range into disjoint memblocks
734  * @type: memblock type to isolate range for
735  * @base: base of range to isolate
736  * @size: size of range to isolate
737  * @start_rgn: out parameter for the start of isolated region
738  * @end_rgn: out parameter for the end of isolated region
739  *
740  * Walk @type and ensure that regions don't cross the boundaries defined by
741  * [@base, @base + @size).  Crossing regions are split at the boundaries,
742  * which may create at most two more regions.  The index of the first
743  * region inside the range is returned in *@start_rgn and end in *@end_rgn.
744  *
745  * Return:
746  * 0 on success, -errno on failure.
747  */
748 static int __init_memblock memblock_isolate_range(struct memblock_type *type,
749 					phys_addr_t base, phys_addr_t size,
750 					int *start_rgn, int *end_rgn)
751 {
752 	phys_addr_t end = base + memblock_cap_size(base, &size);
753 	int idx;
754 	struct memblock_region *rgn;
755 
756 	*start_rgn = *end_rgn = 0;
757 
758 	if (!size)
759 		return 0;
760 
761 	/* we'll create at most two more regions */
762 	while (type->cnt + 2 > type->max)
763 		if (memblock_double_array(type, base, size) < 0)
764 			return -ENOMEM;
765 
766 	for_each_memblock_type(idx, type, rgn) {
767 		phys_addr_t rbase = rgn->base;
768 		phys_addr_t rend = rbase + rgn->size;
769 
770 		if (rbase >= end)
771 			break;
772 		if (rend <= base)
773 			continue;
774 
775 		if (rbase < base) {
776 			/*
777 			 * @rgn intersects from below.  Split and continue
778 			 * to process the next region - the new top half.
779 			 */
780 			rgn->base = base;
781 			rgn->size -= base - rbase;
782 			type->total_size -= base - rbase;
783 			memblock_insert_region(type, idx, rbase, base - rbase,
784 					       memblock_get_region_node(rgn),
785 					       rgn->flags);
786 		} else if (rend > end) {
787 			/*
788 			 * @rgn intersects from above.  Split and redo the
789 			 * current region - the new bottom half.
790 			 */
791 			rgn->base = end;
792 			rgn->size -= end - rbase;
793 			type->total_size -= end - rbase;
794 			memblock_insert_region(type, idx--, rbase, end - rbase,
795 					       memblock_get_region_node(rgn),
796 					       rgn->flags);
797 		} else {
798 			/* @rgn is fully contained, record it */
799 			if (!*end_rgn)
800 				*start_rgn = idx;
801 			*end_rgn = idx + 1;
802 		}
803 	}
804 
805 	return 0;
806 }
807 
808 static int __init_memblock memblock_remove_range(struct memblock_type *type,
809 					  phys_addr_t base, phys_addr_t size)
810 {
811 	int start_rgn, end_rgn;
812 	int i, ret;
813 
814 	ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
815 	if (ret)
816 		return ret;
817 
818 	for (i = end_rgn - 1; i >= start_rgn; i--)
819 		memblock_remove_region(type, i);
820 	return 0;
821 }
822 
823 int __init_memblock memblock_remove(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_remove_range(&memblock.memory, base, size);
831 }
832 
833 /**
834  * memblock_free - free boot memory allocation
835  * @ptr: starting address of the  boot memory allocation
836  * @size: size of the boot memory block in bytes
837  *
838  * Free boot memory block previously allocated by memblock_alloc_xx() API.
839  * The freeing memory will not be released to the buddy allocator.
840  */
841 void __init_memblock memblock_free(void *ptr, size_t size)
842 {
843 	if (ptr)
844 		memblock_phys_free(__pa(ptr), size);
845 }
846 
847 /**
848  * memblock_phys_free - free boot memory block
849  * @base: phys starting address of the  boot memory block
850  * @size: size of the boot memory block in bytes
851  *
852  * Free boot memory block previously allocated by memblock_phys_alloc_xx() API.
853  * The freeing memory will not be released to the buddy allocator.
854  */
855 int __init_memblock memblock_phys_free(phys_addr_t base, phys_addr_t size)
856 {
857 	phys_addr_t end = base + size - 1;
858 
859 	memblock_dbg("%s: [%pa-%pa] %pS\n", __func__,
860 		     &base, &end, (void *)_RET_IP_);
861 
862 	kmemleak_free_part_phys(base, size);
863 	return memblock_remove_range(&memblock.reserved, base, size);
864 }
865 
866 int __init_memblock memblock_reserve(phys_addr_t base, phys_addr_t size)
867 {
868 	phys_addr_t end = base + size - 1;
869 
870 	memblock_dbg("%s: [%pa-%pa] %pS\n", __func__,
871 		     &base, &end, (void *)_RET_IP_);
872 
873 	return memblock_add_range(&memblock.reserved, base, size, MAX_NUMNODES, 0);
874 }
875 
876 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
877 int __init_memblock memblock_physmem_add(phys_addr_t base, phys_addr_t size)
878 {
879 	phys_addr_t end = base + size - 1;
880 
881 	memblock_dbg("%s: [%pa-%pa] %pS\n", __func__,
882 		     &base, &end, (void *)_RET_IP_);
883 
884 	return memblock_add_range(&physmem, base, size, MAX_NUMNODES, 0);
885 }
886 #endif
887 
888 /**
889  * memblock_setclr_flag - set or clear flag for a memory region
890  * @base: base address of the region
891  * @size: size of the region
892  * @set: set or clear the flag
893  * @flag: the flag to update
894  *
895  * This function isolates region [@base, @base + @size), and sets/clears flag
896  *
897  * Return: 0 on success, -errno on failure.
898  */
899 static int __init_memblock memblock_setclr_flag(phys_addr_t base,
900 				phys_addr_t size, int set, int flag)
901 {
902 	struct memblock_type *type = &memblock.memory;
903 	int i, ret, start_rgn, end_rgn;
904 
905 	ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
906 	if (ret)
907 		return ret;
908 
909 	for (i = start_rgn; i < end_rgn; i++) {
910 		struct memblock_region *r = &type->regions[i];
911 
912 		if (set)
913 			r->flags |= flag;
914 		else
915 			r->flags &= ~flag;
916 	}
917 
918 	memblock_merge_regions(type, start_rgn, end_rgn);
919 	return 0;
920 }
921 
922 /**
923  * memblock_mark_hotplug - Mark hotpluggable memory with flag MEMBLOCK_HOTPLUG.
924  * @base: the base phys addr of the region
925  * @size: the size of the region
926  *
927  * Return: 0 on success, -errno on failure.
928  */
929 int __init_memblock memblock_mark_hotplug(phys_addr_t base, phys_addr_t size)
930 {
931 	return memblock_setclr_flag(base, size, 1, MEMBLOCK_HOTPLUG);
932 }
933 
934 /**
935  * memblock_clear_hotplug - Clear flag MEMBLOCK_HOTPLUG for a specified region.
936  * @base: the base phys addr of the region
937  * @size: the size of the region
938  *
939  * Return: 0 on success, -errno on failure.
940  */
941 int __init_memblock memblock_clear_hotplug(phys_addr_t base, phys_addr_t size)
942 {
943 	return memblock_setclr_flag(base, size, 0, MEMBLOCK_HOTPLUG);
944 }
945 
946 /**
947  * memblock_mark_mirror - Mark mirrored memory with flag MEMBLOCK_MIRROR.
948  * @base: the base phys addr of the region
949  * @size: the size of the region
950  *
951  * Return: 0 on success, -errno on failure.
952  */
953 int __init_memblock memblock_mark_mirror(phys_addr_t base, phys_addr_t size)
954 {
955 	if (!mirrored_kernelcore)
956 		return 0;
957 
958 	system_has_some_mirror = true;
959 
960 	return memblock_setclr_flag(base, size, 1, MEMBLOCK_MIRROR);
961 }
962 
963 /**
964  * memblock_mark_nomap - Mark a memory region with flag MEMBLOCK_NOMAP.
965  * @base: the base phys addr of the region
966  * @size: the size of the region
967  *
968  * The memory regions marked with %MEMBLOCK_NOMAP will not be added to the
969  * direct mapping of the physical memory. These regions will still be
970  * covered by the memory map. The struct page representing NOMAP memory
971  * frames in the memory map will be PageReserved()
972  *
973  * Note: if the memory being marked %MEMBLOCK_NOMAP was allocated from
974  * memblock, the caller must inform kmemleak to ignore that memory
975  *
976  * Return: 0 on success, -errno on failure.
977  */
978 int __init_memblock memblock_mark_nomap(phys_addr_t base, phys_addr_t size)
979 {
980 	return memblock_setclr_flag(base, size, 1, MEMBLOCK_NOMAP);
981 }
982 
983 /**
984  * memblock_clear_nomap - Clear flag MEMBLOCK_NOMAP for a specified region.
985  * @base: the base phys addr of the region
986  * @size: the size of the region
987  *
988  * Return: 0 on success, -errno on failure.
989  */
990 int __init_memblock memblock_clear_nomap(phys_addr_t base, phys_addr_t size)
991 {
992 	return memblock_setclr_flag(base, size, 0, MEMBLOCK_NOMAP);
993 }
994 
995 static bool should_skip_region(struct memblock_type *type,
996 			       struct memblock_region *m,
997 			       int nid, int flags)
998 {
999 	int m_nid = memblock_get_region_node(m);
1000 
1001 	/* we never skip regions when iterating memblock.reserved or physmem */
1002 	if (type != memblock_memory)
1003 		return false;
1004 
1005 	/* only memory regions are associated with nodes, check it */
1006 	if (nid != NUMA_NO_NODE && nid != m_nid)
1007 		return true;
1008 
1009 	/* skip hotpluggable memory regions if needed */
1010 	if (movable_node_is_enabled() && memblock_is_hotpluggable(m) &&
1011 	    !(flags & MEMBLOCK_HOTPLUG))
1012 		return true;
1013 
1014 	/* if we want mirror memory skip non-mirror memory regions */
1015 	if ((flags & MEMBLOCK_MIRROR) && !memblock_is_mirror(m))
1016 		return true;
1017 
1018 	/* skip nomap memory unless we were asked for it explicitly */
1019 	if (!(flags & MEMBLOCK_NOMAP) && memblock_is_nomap(m))
1020 		return true;
1021 
1022 	/* skip driver-managed memory unless we were asked for it explicitly */
1023 	if (!(flags & MEMBLOCK_DRIVER_MANAGED) && memblock_is_driver_managed(m))
1024 		return true;
1025 
1026 	return false;
1027 }
1028 
1029 /**
1030  * __next_mem_range - next function for for_each_free_mem_range() etc.
1031  * @idx: pointer to u64 loop variable
1032  * @nid: node selector, %NUMA_NO_NODE for all nodes
1033  * @flags: pick from blocks based on memory attributes
1034  * @type_a: pointer to memblock_type from where the range is taken
1035  * @type_b: pointer to memblock_type which excludes memory from being taken
1036  * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL
1037  * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL
1038  * @out_nid: ptr to int for nid of the range, can be %NULL
1039  *
1040  * Find the first area from *@idx which matches @nid, fill the out
1041  * parameters, and update *@idx for the next iteration.  The lower 32bit of
1042  * *@idx contains index into type_a and the upper 32bit indexes the
1043  * areas before each region in type_b.	For example, if type_b regions
1044  * look like the following,
1045  *
1046  *	0:[0-16), 1:[32-48), 2:[128-130)
1047  *
1048  * The upper 32bit indexes the following regions.
1049  *
1050  *	0:[0-0), 1:[16-32), 2:[48-128), 3:[130-MAX)
1051  *
1052  * As both region arrays are sorted, the function advances the two indices
1053  * in lockstep and returns each intersection.
1054  */
1055 void __next_mem_range(u64 *idx, int nid, enum memblock_flags flags,
1056 		      struct memblock_type *type_a,
1057 		      struct memblock_type *type_b, phys_addr_t *out_start,
1058 		      phys_addr_t *out_end, int *out_nid)
1059 {
1060 	int idx_a = *idx & 0xffffffff;
1061 	int idx_b = *idx >> 32;
1062 
1063 	if (WARN_ONCE(nid == MAX_NUMNODES,
1064 	"Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
1065 		nid = NUMA_NO_NODE;
1066 
1067 	for (; idx_a < type_a->cnt; idx_a++) {
1068 		struct memblock_region *m = &type_a->regions[idx_a];
1069 
1070 		phys_addr_t m_start = m->base;
1071 		phys_addr_t m_end = m->base + m->size;
1072 		int	    m_nid = memblock_get_region_node(m);
1073 
1074 		if (should_skip_region(type_a, m, nid, flags))
1075 			continue;
1076 
1077 		if (!type_b) {
1078 			if (out_start)
1079 				*out_start = m_start;
1080 			if (out_end)
1081 				*out_end = m_end;
1082 			if (out_nid)
1083 				*out_nid = m_nid;
1084 			idx_a++;
1085 			*idx = (u32)idx_a | (u64)idx_b << 32;
1086 			return;
1087 		}
1088 
1089 		/* scan areas before each reservation */
1090 		for (; idx_b < type_b->cnt + 1; idx_b++) {
1091 			struct memblock_region *r;
1092 			phys_addr_t r_start;
1093 			phys_addr_t r_end;
1094 
1095 			r = &type_b->regions[idx_b];
1096 			r_start = idx_b ? r[-1].base + r[-1].size : 0;
1097 			r_end = idx_b < type_b->cnt ?
1098 				r->base : PHYS_ADDR_MAX;
1099 
1100 			/*
1101 			 * if idx_b advanced past idx_a,
1102 			 * break out to advance idx_a
1103 			 */
1104 			if (r_start >= m_end)
1105 				break;
1106 			/* if the two regions intersect, we're done */
1107 			if (m_start < r_end) {
1108 				if (out_start)
1109 					*out_start =
1110 						max(m_start, r_start);
1111 				if (out_end)
1112 					*out_end = min(m_end, r_end);
1113 				if (out_nid)
1114 					*out_nid = m_nid;
1115 				/*
1116 				 * The region which ends first is
1117 				 * advanced for the next iteration.
1118 				 */
1119 				if (m_end <= r_end)
1120 					idx_a++;
1121 				else
1122 					idx_b++;
1123 				*idx = (u32)idx_a | (u64)idx_b << 32;
1124 				return;
1125 			}
1126 		}
1127 	}
1128 
1129 	/* signal end of iteration */
1130 	*idx = ULLONG_MAX;
1131 }
1132 
1133 /**
1134  * __next_mem_range_rev - generic next function for for_each_*_range_rev()
1135  *
1136  * @idx: pointer to u64 loop variable
1137  * @nid: node selector, %NUMA_NO_NODE for all nodes
1138  * @flags: pick from blocks based on memory attributes
1139  * @type_a: pointer to memblock_type from where the range is taken
1140  * @type_b: pointer to memblock_type which excludes memory from being taken
1141  * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL
1142  * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL
1143  * @out_nid: ptr to int for nid of the range, can be %NULL
1144  *
1145  * Finds the next range from type_a which is not marked as unsuitable
1146  * in type_b.
1147  *
1148  * Reverse of __next_mem_range().
1149  */
1150 void __init_memblock __next_mem_range_rev(u64 *idx, int nid,
1151 					  enum memblock_flags flags,
1152 					  struct memblock_type *type_a,
1153 					  struct memblock_type *type_b,
1154 					  phys_addr_t *out_start,
1155 					  phys_addr_t *out_end, int *out_nid)
1156 {
1157 	int idx_a = *idx & 0xffffffff;
1158 	int idx_b = *idx >> 32;
1159 
1160 	if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
1161 		nid = NUMA_NO_NODE;
1162 
1163 	if (*idx == (u64)ULLONG_MAX) {
1164 		idx_a = type_a->cnt - 1;
1165 		if (type_b != NULL)
1166 			idx_b = type_b->cnt;
1167 		else
1168 			idx_b = 0;
1169 	}
1170 
1171 	for (; idx_a >= 0; idx_a--) {
1172 		struct memblock_region *m = &type_a->regions[idx_a];
1173 
1174 		phys_addr_t m_start = m->base;
1175 		phys_addr_t m_end = m->base + m->size;
1176 		int m_nid = memblock_get_region_node(m);
1177 
1178 		if (should_skip_region(type_a, m, nid, flags))
1179 			continue;
1180 
1181 		if (!type_b) {
1182 			if (out_start)
1183 				*out_start = m_start;
1184 			if (out_end)
1185 				*out_end = m_end;
1186 			if (out_nid)
1187 				*out_nid = m_nid;
1188 			idx_a--;
1189 			*idx = (u32)idx_a | (u64)idx_b << 32;
1190 			return;
1191 		}
1192 
1193 		/* scan areas before each reservation */
1194 		for (; idx_b >= 0; idx_b--) {
1195 			struct memblock_region *r;
1196 			phys_addr_t r_start;
1197 			phys_addr_t r_end;
1198 
1199 			r = &type_b->regions[idx_b];
1200 			r_start = idx_b ? r[-1].base + r[-1].size : 0;
1201 			r_end = idx_b < type_b->cnt ?
1202 				r->base : PHYS_ADDR_MAX;
1203 			/*
1204 			 * if idx_b advanced past idx_a,
1205 			 * break out to advance idx_a
1206 			 */
1207 
1208 			if (r_end <= m_start)
1209 				break;
1210 			/* if the two regions intersect, we're done */
1211 			if (m_end > r_start) {
1212 				if (out_start)
1213 					*out_start = max(m_start, r_start);
1214 				if (out_end)
1215 					*out_end = min(m_end, r_end);
1216 				if (out_nid)
1217 					*out_nid = m_nid;
1218 				if (m_start >= r_start)
1219 					idx_a--;
1220 				else
1221 					idx_b--;
1222 				*idx = (u32)idx_a | (u64)idx_b << 32;
1223 				return;
1224 			}
1225 		}
1226 	}
1227 	/* signal end of iteration */
1228 	*idx = ULLONG_MAX;
1229 }
1230 
1231 /*
1232  * Common iterator interface used to define for_each_mem_pfn_range().
1233  */
1234 void __init_memblock __next_mem_pfn_range(int *idx, int nid,
1235 				unsigned long *out_start_pfn,
1236 				unsigned long *out_end_pfn, int *out_nid)
1237 {
1238 	struct memblock_type *type = &memblock.memory;
1239 	struct memblock_region *r;
1240 	int r_nid;
1241 
1242 	while (++*idx < type->cnt) {
1243 		r = &type->regions[*idx];
1244 		r_nid = memblock_get_region_node(r);
1245 
1246 		if (PFN_UP(r->base) >= PFN_DOWN(r->base + r->size))
1247 			continue;
1248 		if (nid == MAX_NUMNODES || nid == r_nid)
1249 			break;
1250 	}
1251 	if (*idx >= type->cnt) {
1252 		*idx = -1;
1253 		return;
1254 	}
1255 
1256 	if (out_start_pfn)
1257 		*out_start_pfn = PFN_UP(r->base);
1258 	if (out_end_pfn)
1259 		*out_end_pfn = PFN_DOWN(r->base + r->size);
1260 	if (out_nid)
1261 		*out_nid = r_nid;
1262 }
1263 
1264 /**
1265  * memblock_set_node - set node ID on memblock regions
1266  * @base: base of area to set node ID for
1267  * @size: size of area to set node ID for
1268  * @type: memblock type to set node ID for
1269  * @nid: node ID to set
1270  *
1271  * Set the nid of memblock @type regions in [@base, @base + @size) to @nid.
1272  * Regions which cross the area boundaries are split as necessary.
1273  *
1274  * Return:
1275  * 0 on success, -errno on failure.
1276  */
1277 int __init_memblock memblock_set_node(phys_addr_t base, phys_addr_t size,
1278 				      struct memblock_type *type, int nid)
1279 {
1280 #ifdef CONFIG_NUMA
1281 	int start_rgn, end_rgn;
1282 	int i, ret;
1283 
1284 	ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
1285 	if (ret)
1286 		return ret;
1287 
1288 	for (i = start_rgn; i < end_rgn; i++)
1289 		memblock_set_region_node(&type->regions[i], nid);
1290 
1291 	memblock_merge_regions(type, start_rgn, end_rgn);
1292 #endif
1293 	return 0;
1294 }
1295 
1296 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1297 /**
1298  * __next_mem_pfn_range_in_zone - iterator for for_each_*_range_in_zone()
1299  *
1300  * @idx: pointer to u64 loop variable
1301  * @zone: zone in which all of the memory blocks reside
1302  * @out_spfn: ptr to ulong for start pfn of the range, can be %NULL
1303  * @out_epfn: ptr to ulong for end pfn of the range, can be %NULL
1304  *
1305  * This function is meant to be a zone/pfn specific wrapper for the
1306  * for_each_mem_range type iterators. Specifically they are used in the
1307  * deferred memory init routines and as such we were duplicating much of
1308  * this logic throughout the code. So instead of having it in multiple
1309  * locations it seemed like it would make more sense to centralize this to
1310  * one new iterator that does everything they need.
1311  */
1312 void __init_memblock
1313 __next_mem_pfn_range_in_zone(u64 *idx, struct zone *zone,
1314 			     unsigned long *out_spfn, unsigned long *out_epfn)
1315 {
1316 	int zone_nid = zone_to_nid(zone);
1317 	phys_addr_t spa, epa;
1318 
1319 	__next_mem_range(idx, zone_nid, MEMBLOCK_NONE,
1320 			 &memblock.memory, &memblock.reserved,
1321 			 &spa, &epa, NULL);
1322 
1323 	while (*idx != U64_MAX) {
1324 		unsigned long epfn = PFN_DOWN(epa);
1325 		unsigned long spfn = PFN_UP(spa);
1326 
1327 		/*
1328 		 * Verify the end is at least past the start of the zone and
1329 		 * that we have at least one PFN to initialize.
1330 		 */
1331 		if (zone->zone_start_pfn < epfn && spfn < epfn) {
1332 			/* if we went too far just stop searching */
1333 			if (zone_end_pfn(zone) <= spfn) {
1334 				*idx = U64_MAX;
1335 				break;
1336 			}
1337 
1338 			if (out_spfn)
1339 				*out_spfn = max(zone->zone_start_pfn, spfn);
1340 			if (out_epfn)
1341 				*out_epfn = min(zone_end_pfn(zone), epfn);
1342 
1343 			return;
1344 		}
1345 
1346 		__next_mem_range(idx, zone_nid, MEMBLOCK_NONE,
1347 				 &memblock.memory, &memblock.reserved,
1348 				 &spa, &epa, NULL);
1349 	}
1350 
1351 	/* signal end of iteration */
1352 	if (out_spfn)
1353 		*out_spfn = ULONG_MAX;
1354 	if (out_epfn)
1355 		*out_epfn = 0;
1356 }
1357 
1358 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
1359 
1360 /**
1361  * memblock_alloc_range_nid - allocate boot memory block
1362  * @size: size of memory block to be allocated in bytes
1363  * @align: alignment of the region and block's size
1364  * @start: the lower bound of the memory region to allocate (phys address)
1365  * @end: the upper bound of the memory region to allocate (phys address)
1366  * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1367  * @exact_nid: control the allocation fall back to other nodes
1368  *
1369  * The allocation is performed from memory region limited by
1370  * memblock.current_limit if @end == %MEMBLOCK_ALLOC_ACCESSIBLE.
1371  *
1372  * If the specified node can not hold the requested memory and @exact_nid
1373  * is false, the allocation falls back to any node in the system.
1374  *
1375  * For systems with memory mirroring, the allocation is attempted first
1376  * from the regions with mirroring enabled and then retried from any
1377  * memory region.
1378  *
1379  * In addition, function using kmemleak_alloc_phys for allocated boot
1380  * memory block, it is never reported as leaks.
1381  *
1382  * Return:
1383  * Physical address of allocated memory block on success, %0 on failure.
1384  */
1385 phys_addr_t __init memblock_alloc_range_nid(phys_addr_t size,
1386 					phys_addr_t align, phys_addr_t start,
1387 					phys_addr_t end, int nid,
1388 					bool exact_nid)
1389 {
1390 	enum memblock_flags flags = choose_memblock_flags();
1391 	phys_addr_t found;
1392 
1393 	if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
1394 		nid = NUMA_NO_NODE;
1395 
1396 	if (!align) {
1397 		/* Can't use WARNs this early in boot on powerpc */
1398 		dump_stack();
1399 		align = SMP_CACHE_BYTES;
1400 	}
1401 
1402 again:
1403 	found = memblock_find_in_range_node(size, align, start, end, nid,
1404 					    flags);
1405 	if (found && !memblock_reserve(found, size))
1406 		goto done;
1407 
1408 	if (nid != NUMA_NO_NODE && !exact_nid) {
1409 		found = memblock_find_in_range_node(size, align, start,
1410 						    end, NUMA_NO_NODE,
1411 						    flags);
1412 		if (found && !memblock_reserve(found, size))
1413 			goto done;
1414 	}
1415 
1416 	if (flags & MEMBLOCK_MIRROR) {
1417 		flags &= ~MEMBLOCK_MIRROR;
1418 		pr_warn_ratelimited("Could not allocate %pap bytes of mirrored memory\n",
1419 			&size);
1420 		goto again;
1421 	}
1422 
1423 	return 0;
1424 
1425 done:
1426 	/*
1427 	 * Skip kmemleak for those places like kasan_init() and
1428 	 * early_pgtable_alloc() due to high volume.
1429 	 */
1430 	if (end != MEMBLOCK_ALLOC_NOLEAKTRACE)
1431 		/*
1432 		 * Memblock allocated blocks are never reported as
1433 		 * leaks. This is because many of these blocks are
1434 		 * only referred via the physical address which is
1435 		 * not looked up by kmemleak.
1436 		 */
1437 		kmemleak_alloc_phys(found, size, 0);
1438 
1439 	/*
1440 	 * Some Virtual Machine platforms, such as Intel TDX or AMD SEV-SNP,
1441 	 * require memory to be accepted before it can be used by the
1442 	 * guest.
1443 	 *
1444 	 * Accept the memory of the allocated buffer.
1445 	 */
1446 	accept_memory(found, found + size);
1447 
1448 	return found;
1449 }
1450 
1451 /**
1452  * memblock_phys_alloc_range - allocate a memory block inside specified range
1453  * @size: size of memory block to be allocated in bytes
1454  * @align: alignment of the region and block's size
1455  * @start: the lower bound of the memory region to allocate (physical address)
1456  * @end: the upper bound of the memory region to allocate (physical address)
1457  *
1458  * Allocate @size bytes in the between @start and @end.
1459  *
1460  * Return: physical address of the allocated memory block on success,
1461  * %0 on failure.
1462  */
1463 phys_addr_t __init memblock_phys_alloc_range(phys_addr_t size,
1464 					     phys_addr_t align,
1465 					     phys_addr_t start,
1466 					     phys_addr_t end)
1467 {
1468 	memblock_dbg("%s: %llu bytes align=0x%llx from=%pa max_addr=%pa %pS\n",
1469 		     __func__, (u64)size, (u64)align, &start, &end,
1470 		     (void *)_RET_IP_);
1471 	return memblock_alloc_range_nid(size, align, start, end, NUMA_NO_NODE,
1472 					false);
1473 }
1474 
1475 /**
1476  * memblock_phys_alloc_try_nid - allocate a memory block from specified NUMA node
1477  * @size: size of memory block to be allocated in bytes
1478  * @align: alignment of the region and block's size
1479  * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1480  *
1481  * Allocates memory block from the specified NUMA node. If the node
1482  * has no available memory, attempts to allocated from any node in the
1483  * system.
1484  *
1485  * Return: physical address of the allocated memory block on success,
1486  * %0 on failure.
1487  */
1488 phys_addr_t __init memblock_phys_alloc_try_nid(phys_addr_t size, phys_addr_t align, int nid)
1489 {
1490 	return memblock_alloc_range_nid(size, align, 0,
1491 					MEMBLOCK_ALLOC_ACCESSIBLE, nid, false);
1492 }
1493 
1494 /**
1495  * memblock_alloc_internal - allocate boot memory block
1496  * @size: size of memory block to be allocated in bytes
1497  * @align: alignment of the region and block's size
1498  * @min_addr: the lower bound of the memory region to allocate (phys address)
1499  * @max_addr: the upper bound of the memory region to allocate (phys address)
1500  * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1501  * @exact_nid: control the allocation fall back to other nodes
1502  *
1503  * Allocates memory block using memblock_alloc_range_nid() and
1504  * converts the returned physical address to virtual.
1505  *
1506  * The @min_addr limit is dropped if it can not be satisfied and the allocation
1507  * will fall back to memory below @min_addr. Other constraints, such
1508  * as node and mirrored memory will be handled again in
1509  * memblock_alloc_range_nid().
1510  *
1511  * Return:
1512  * Virtual address of allocated memory block on success, NULL on failure.
1513  */
1514 static void * __init memblock_alloc_internal(
1515 				phys_addr_t size, phys_addr_t align,
1516 				phys_addr_t min_addr, phys_addr_t max_addr,
1517 				int nid, bool exact_nid)
1518 {
1519 	phys_addr_t alloc;
1520 
1521 	/*
1522 	 * Detect any accidental use of these APIs after slab is ready, as at
1523 	 * this moment memblock may be deinitialized already and its
1524 	 * internal data may be destroyed (after execution of memblock_free_all)
1525 	 */
1526 	if (WARN_ON_ONCE(slab_is_available()))
1527 		return kzalloc_node(size, GFP_NOWAIT, nid);
1528 
1529 	if (max_addr > memblock.current_limit)
1530 		max_addr = memblock.current_limit;
1531 
1532 	alloc = memblock_alloc_range_nid(size, align, min_addr, max_addr, nid,
1533 					exact_nid);
1534 
1535 	/* retry allocation without lower limit */
1536 	if (!alloc && min_addr)
1537 		alloc = memblock_alloc_range_nid(size, align, 0, max_addr, nid,
1538 						exact_nid);
1539 
1540 	if (!alloc)
1541 		return NULL;
1542 
1543 	return phys_to_virt(alloc);
1544 }
1545 
1546 /**
1547  * memblock_alloc_exact_nid_raw - allocate boot memory block on the exact node
1548  * without zeroing memory
1549  * @size: size of memory block to be allocated in bytes
1550  * @align: alignment of the region and block's size
1551  * @min_addr: the lower bound of the memory region from where the allocation
1552  *	  is preferred (phys address)
1553  * @max_addr: the upper bound of the memory region from where the allocation
1554  *	      is preferred (phys address), or %MEMBLOCK_ALLOC_ACCESSIBLE to
1555  *	      allocate only from memory limited by memblock.current_limit value
1556  * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1557  *
1558  * Public function, provides additional debug information (including caller
1559  * info), if enabled. Does not zero allocated memory.
1560  *
1561  * Return:
1562  * Virtual address of allocated memory block on success, NULL on failure.
1563  */
1564 void * __init memblock_alloc_exact_nid_raw(
1565 			phys_addr_t size, phys_addr_t align,
1566 			phys_addr_t min_addr, phys_addr_t max_addr,
1567 			int nid)
1568 {
1569 	memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=%pa max_addr=%pa %pS\n",
1570 		     __func__, (u64)size, (u64)align, nid, &min_addr,
1571 		     &max_addr, (void *)_RET_IP_);
1572 
1573 	return memblock_alloc_internal(size, align, min_addr, max_addr, nid,
1574 				       true);
1575 }
1576 
1577 /**
1578  * memblock_alloc_try_nid_raw - allocate boot memory block without zeroing
1579  * memory and without panicking
1580  * @size: size of memory block to be allocated in bytes
1581  * @align: alignment of the region and block's size
1582  * @min_addr: the lower bound of the memory region from where the allocation
1583  *	  is preferred (phys address)
1584  * @max_addr: the upper bound of the memory region from where the allocation
1585  *	      is preferred (phys address), or %MEMBLOCK_ALLOC_ACCESSIBLE to
1586  *	      allocate only from memory limited by memblock.current_limit value
1587  * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1588  *
1589  * Public function, provides additional debug information (including caller
1590  * info), if enabled. Does not zero allocated memory, does not panic if request
1591  * cannot be satisfied.
1592  *
1593  * Return:
1594  * Virtual address of allocated memory block on success, NULL on failure.
1595  */
1596 void * __init memblock_alloc_try_nid_raw(
1597 			phys_addr_t size, phys_addr_t align,
1598 			phys_addr_t min_addr, phys_addr_t max_addr,
1599 			int nid)
1600 {
1601 	memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=%pa max_addr=%pa %pS\n",
1602 		     __func__, (u64)size, (u64)align, nid, &min_addr,
1603 		     &max_addr, (void *)_RET_IP_);
1604 
1605 	return memblock_alloc_internal(size, align, min_addr, max_addr, nid,
1606 				       false);
1607 }
1608 
1609 /**
1610  * memblock_alloc_try_nid - allocate boot memory block
1611  * @size: size of memory block to be allocated in bytes
1612  * @align: alignment of the region and block's size
1613  * @min_addr: the lower bound of the memory region from where the allocation
1614  *	  is preferred (phys address)
1615  * @max_addr: the upper bound of the memory region from where the allocation
1616  *	      is preferred (phys address), or %MEMBLOCK_ALLOC_ACCESSIBLE to
1617  *	      allocate only from memory limited by memblock.current_limit value
1618  * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1619  *
1620  * Public function, provides additional debug information (including caller
1621  * info), if enabled. This function zeroes the allocated memory.
1622  *
1623  * Return:
1624  * Virtual address of allocated memory block on success, NULL on failure.
1625  */
1626 void * __init memblock_alloc_try_nid(
1627 			phys_addr_t size, phys_addr_t align,
1628 			phys_addr_t min_addr, phys_addr_t max_addr,
1629 			int nid)
1630 {
1631 	void *ptr;
1632 
1633 	memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=%pa max_addr=%pa %pS\n",
1634 		     __func__, (u64)size, (u64)align, nid, &min_addr,
1635 		     &max_addr, (void *)_RET_IP_);
1636 	ptr = memblock_alloc_internal(size, align,
1637 					   min_addr, max_addr, nid, false);
1638 	if (ptr)
1639 		memset(ptr, 0, size);
1640 
1641 	return ptr;
1642 }
1643 
1644 /**
1645  * memblock_free_late - free pages directly to buddy allocator
1646  * @base: phys starting address of the  boot memory block
1647  * @size: size of the boot memory block in bytes
1648  *
1649  * This is only useful when the memblock allocator has already been torn
1650  * down, but we are still initializing the system.  Pages are released directly
1651  * to the buddy allocator.
1652  */
1653 void __init memblock_free_late(phys_addr_t base, phys_addr_t size)
1654 {
1655 	phys_addr_t cursor, end;
1656 
1657 	end = base + size - 1;
1658 	memblock_dbg("%s: [%pa-%pa] %pS\n",
1659 		     __func__, &base, &end, (void *)_RET_IP_);
1660 	kmemleak_free_part_phys(base, size);
1661 	cursor = PFN_UP(base);
1662 	end = PFN_DOWN(base + size);
1663 
1664 	for (; cursor < end; cursor++) {
1665 		memblock_free_pages(pfn_to_page(cursor), cursor, 0);
1666 		totalram_pages_inc();
1667 	}
1668 }
1669 
1670 /*
1671  * Remaining API functions
1672  */
1673 
1674 phys_addr_t __init_memblock memblock_phys_mem_size(void)
1675 {
1676 	return memblock.memory.total_size;
1677 }
1678 
1679 phys_addr_t __init_memblock memblock_reserved_size(void)
1680 {
1681 	return memblock.reserved.total_size;
1682 }
1683 
1684 /* lowest address */
1685 phys_addr_t __init_memblock memblock_start_of_DRAM(void)
1686 {
1687 	return memblock.memory.regions[0].base;
1688 }
1689 
1690 phys_addr_t __init_memblock memblock_end_of_DRAM(void)
1691 {
1692 	int idx = memblock.memory.cnt - 1;
1693 
1694 	return (memblock.memory.regions[idx].base + memblock.memory.regions[idx].size);
1695 }
1696 
1697 static phys_addr_t __init_memblock __find_max_addr(phys_addr_t limit)
1698 {
1699 	phys_addr_t max_addr = PHYS_ADDR_MAX;
1700 	struct memblock_region *r;
1701 
1702 	/*
1703 	 * translate the memory @limit size into the max address within one of
1704 	 * the memory memblock regions, if the @limit exceeds the total size
1705 	 * of those regions, max_addr will keep original value PHYS_ADDR_MAX
1706 	 */
1707 	for_each_mem_region(r) {
1708 		if (limit <= r->size) {
1709 			max_addr = r->base + limit;
1710 			break;
1711 		}
1712 		limit -= r->size;
1713 	}
1714 
1715 	return max_addr;
1716 }
1717 
1718 void __init memblock_enforce_memory_limit(phys_addr_t limit)
1719 {
1720 	phys_addr_t max_addr;
1721 
1722 	if (!limit)
1723 		return;
1724 
1725 	max_addr = __find_max_addr(limit);
1726 
1727 	/* @limit exceeds the total size of the memory, do nothing */
1728 	if (max_addr == PHYS_ADDR_MAX)
1729 		return;
1730 
1731 	/* truncate both memory and reserved regions */
1732 	memblock_remove_range(&memblock.memory, max_addr,
1733 			      PHYS_ADDR_MAX);
1734 	memblock_remove_range(&memblock.reserved, max_addr,
1735 			      PHYS_ADDR_MAX);
1736 }
1737 
1738 void __init memblock_cap_memory_range(phys_addr_t base, phys_addr_t size)
1739 {
1740 	int start_rgn, end_rgn;
1741 	int i, ret;
1742 
1743 	if (!size)
1744 		return;
1745 
1746 	if (!memblock_memory->total_size) {
1747 		pr_warn("%s: No memory registered yet\n", __func__);
1748 		return;
1749 	}
1750 
1751 	ret = memblock_isolate_range(&memblock.memory, base, size,
1752 						&start_rgn, &end_rgn);
1753 	if (ret)
1754 		return;
1755 
1756 	/* remove all the MAP regions */
1757 	for (i = memblock.memory.cnt - 1; i >= end_rgn; i--)
1758 		if (!memblock_is_nomap(&memblock.memory.regions[i]))
1759 			memblock_remove_region(&memblock.memory, i);
1760 
1761 	for (i = start_rgn - 1; i >= 0; i--)
1762 		if (!memblock_is_nomap(&memblock.memory.regions[i]))
1763 			memblock_remove_region(&memblock.memory, i);
1764 
1765 	/* truncate the reserved regions */
1766 	memblock_remove_range(&memblock.reserved, 0, base);
1767 	memblock_remove_range(&memblock.reserved,
1768 			base + size, PHYS_ADDR_MAX);
1769 }
1770 
1771 void __init memblock_mem_limit_remove_map(phys_addr_t limit)
1772 {
1773 	phys_addr_t max_addr;
1774 
1775 	if (!limit)
1776 		return;
1777 
1778 	max_addr = __find_max_addr(limit);
1779 
1780 	/* @limit exceeds the total size of the memory, do nothing */
1781 	if (max_addr == PHYS_ADDR_MAX)
1782 		return;
1783 
1784 	memblock_cap_memory_range(0, max_addr);
1785 }
1786 
1787 static int __init_memblock memblock_search(struct memblock_type *type, phys_addr_t addr)
1788 {
1789 	unsigned int left = 0, right = type->cnt;
1790 
1791 	do {
1792 		unsigned int mid = (right + left) / 2;
1793 
1794 		if (addr < type->regions[mid].base)
1795 			right = mid;
1796 		else if (addr >= (type->regions[mid].base +
1797 				  type->regions[mid].size))
1798 			left = mid + 1;
1799 		else
1800 			return mid;
1801 	} while (left < right);
1802 	return -1;
1803 }
1804 
1805 bool __init_memblock memblock_is_reserved(phys_addr_t addr)
1806 {
1807 	return memblock_search(&memblock.reserved, addr) != -1;
1808 }
1809 
1810 bool __init_memblock memblock_is_memory(phys_addr_t addr)
1811 {
1812 	return memblock_search(&memblock.memory, addr) != -1;
1813 }
1814 
1815 bool __init_memblock memblock_is_map_memory(phys_addr_t addr)
1816 {
1817 	int i = memblock_search(&memblock.memory, addr);
1818 
1819 	if (i == -1)
1820 		return false;
1821 	return !memblock_is_nomap(&memblock.memory.regions[i]);
1822 }
1823 
1824 int __init_memblock memblock_search_pfn_nid(unsigned long pfn,
1825 			 unsigned long *start_pfn, unsigned long *end_pfn)
1826 {
1827 	struct memblock_type *type = &memblock.memory;
1828 	int mid = memblock_search(type, PFN_PHYS(pfn));
1829 
1830 	if (mid == -1)
1831 		return -1;
1832 
1833 	*start_pfn = PFN_DOWN(type->regions[mid].base);
1834 	*end_pfn = PFN_DOWN(type->regions[mid].base + type->regions[mid].size);
1835 
1836 	return memblock_get_region_node(&type->regions[mid]);
1837 }
1838 
1839 /**
1840  * memblock_is_region_memory - check if a region is a subset of memory
1841  * @base: base of region to check
1842  * @size: size of region to check
1843  *
1844  * Check if the region [@base, @base + @size) is a subset of a memory block.
1845  *
1846  * Return:
1847  * 0 if false, non-zero if true
1848  */
1849 bool __init_memblock memblock_is_region_memory(phys_addr_t base, phys_addr_t size)
1850 {
1851 	int idx = memblock_search(&memblock.memory, base);
1852 	phys_addr_t end = base + memblock_cap_size(base, &size);
1853 
1854 	if (idx == -1)
1855 		return false;
1856 	return (memblock.memory.regions[idx].base +
1857 		 memblock.memory.regions[idx].size) >= end;
1858 }
1859 
1860 /**
1861  * memblock_is_region_reserved - check if a region intersects reserved memory
1862  * @base: base of region to check
1863  * @size: size of region to check
1864  *
1865  * Check if the region [@base, @base + @size) intersects a reserved
1866  * memory block.
1867  *
1868  * Return:
1869  * True if they intersect, false if not.
1870  */
1871 bool __init_memblock memblock_is_region_reserved(phys_addr_t base, phys_addr_t size)
1872 {
1873 	return memblock_overlaps_region(&memblock.reserved, base, size);
1874 }
1875 
1876 void __init_memblock memblock_trim_memory(phys_addr_t align)
1877 {
1878 	phys_addr_t start, end, orig_start, orig_end;
1879 	struct memblock_region *r;
1880 
1881 	for_each_mem_region(r) {
1882 		orig_start = r->base;
1883 		orig_end = r->base + r->size;
1884 		start = round_up(orig_start, align);
1885 		end = round_down(orig_end, align);
1886 
1887 		if (start == orig_start && end == orig_end)
1888 			continue;
1889 
1890 		if (start < end) {
1891 			r->base = start;
1892 			r->size = end - start;
1893 		} else {
1894 			memblock_remove_region(&memblock.memory,
1895 					       r - memblock.memory.regions);
1896 			r--;
1897 		}
1898 	}
1899 }
1900 
1901 void __init_memblock memblock_set_current_limit(phys_addr_t limit)
1902 {
1903 	memblock.current_limit = limit;
1904 }
1905 
1906 phys_addr_t __init_memblock memblock_get_current_limit(void)
1907 {
1908 	return memblock.current_limit;
1909 }
1910 
1911 static void __init_memblock memblock_dump(struct memblock_type *type)
1912 {
1913 	phys_addr_t base, end, size;
1914 	enum memblock_flags flags;
1915 	int idx;
1916 	struct memblock_region *rgn;
1917 
1918 	pr_info(" %s.cnt  = 0x%lx\n", type->name, type->cnt);
1919 
1920 	for_each_memblock_type(idx, type, rgn) {
1921 		char nid_buf[32] = "";
1922 
1923 		base = rgn->base;
1924 		size = rgn->size;
1925 		end = base + size - 1;
1926 		flags = rgn->flags;
1927 #ifdef CONFIG_NUMA
1928 		if (memblock_get_region_node(rgn) != MAX_NUMNODES)
1929 			snprintf(nid_buf, sizeof(nid_buf), " on node %d",
1930 				 memblock_get_region_node(rgn));
1931 #endif
1932 		pr_info(" %s[%#x]\t[%pa-%pa], %pa bytes%s flags: %#x\n",
1933 			type->name, idx, &base, &end, &size, nid_buf, flags);
1934 	}
1935 }
1936 
1937 static void __init_memblock __memblock_dump_all(void)
1938 {
1939 	pr_info("MEMBLOCK configuration:\n");
1940 	pr_info(" memory size = %pa reserved size = %pa\n",
1941 		&memblock.memory.total_size,
1942 		&memblock.reserved.total_size);
1943 
1944 	memblock_dump(&memblock.memory);
1945 	memblock_dump(&memblock.reserved);
1946 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
1947 	memblock_dump(&physmem);
1948 #endif
1949 }
1950 
1951 void __init_memblock memblock_dump_all(void)
1952 {
1953 	if (memblock_debug)
1954 		__memblock_dump_all();
1955 }
1956 
1957 void __init memblock_allow_resize(void)
1958 {
1959 	memblock_can_resize = 1;
1960 }
1961 
1962 static int __init early_memblock(char *p)
1963 {
1964 	if (p && strstr(p, "debug"))
1965 		memblock_debug = 1;
1966 	return 0;
1967 }
1968 early_param("memblock", early_memblock);
1969 
1970 static void __init free_memmap(unsigned long start_pfn, unsigned long end_pfn)
1971 {
1972 	struct page *start_pg, *end_pg;
1973 	phys_addr_t pg, pgend;
1974 
1975 	/*
1976 	 * Convert start_pfn/end_pfn to a struct page pointer.
1977 	 */
1978 	start_pg = pfn_to_page(start_pfn - 1) + 1;
1979 	end_pg = pfn_to_page(end_pfn - 1) + 1;
1980 
1981 	/*
1982 	 * Convert to physical addresses, and round start upwards and end
1983 	 * downwards.
1984 	 */
1985 	pg = PAGE_ALIGN(__pa(start_pg));
1986 	pgend = __pa(end_pg) & PAGE_MASK;
1987 
1988 	/*
1989 	 * If there are free pages between these, free the section of the
1990 	 * memmap array.
1991 	 */
1992 	if (pg < pgend)
1993 		memblock_phys_free(pg, pgend - pg);
1994 }
1995 
1996 /*
1997  * The mem_map array can get very big.  Free the unused area of the memory map.
1998  */
1999 static void __init free_unused_memmap(void)
2000 {
2001 	unsigned long start, end, prev_end = 0;
2002 	int i;
2003 
2004 	if (!IS_ENABLED(CONFIG_HAVE_ARCH_PFN_VALID) ||
2005 	    IS_ENABLED(CONFIG_SPARSEMEM_VMEMMAP))
2006 		return;
2007 
2008 	/*
2009 	 * This relies on each bank being in address order.
2010 	 * The banks are sorted previously in bootmem_init().
2011 	 */
2012 	for_each_mem_pfn_range(i, MAX_NUMNODES, &start, &end, NULL) {
2013 #ifdef CONFIG_SPARSEMEM
2014 		/*
2015 		 * Take care not to free memmap entries that don't exist
2016 		 * due to SPARSEMEM sections which aren't present.
2017 		 */
2018 		start = min(start, ALIGN(prev_end, PAGES_PER_SECTION));
2019 #endif
2020 		/*
2021 		 * Align down here since many operations in VM subsystem
2022 		 * presume that there are no holes in the memory map inside
2023 		 * a pageblock
2024 		 */
2025 		start = pageblock_start_pfn(start);
2026 
2027 		/*
2028 		 * If we had a previous bank, and there is a space
2029 		 * between the current bank and the previous, free it.
2030 		 */
2031 		if (prev_end && prev_end < start)
2032 			free_memmap(prev_end, start);
2033 
2034 		/*
2035 		 * Align up here since many operations in VM subsystem
2036 		 * presume that there are no holes in the memory map inside
2037 		 * a pageblock
2038 		 */
2039 		prev_end = pageblock_align(end);
2040 	}
2041 
2042 #ifdef CONFIG_SPARSEMEM
2043 	if (!IS_ALIGNED(prev_end, PAGES_PER_SECTION)) {
2044 		prev_end = pageblock_align(end);
2045 		free_memmap(prev_end, ALIGN(prev_end, PAGES_PER_SECTION));
2046 	}
2047 #endif
2048 }
2049 
2050 static void __init __free_pages_memory(unsigned long start, unsigned long end)
2051 {
2052 	int order;
2053 
2054 	while (start < end) {
2055 		/*
2056 		 * Free the pages in the largest chunks alignment allows.
2057 		 *
2058 		 * __ffs() behaviour is undefined for 0. start == 0 is
2059 		 * MAX_ORDER-aligned, set order to MAX_ORDER for the case.
2060 		 */
2061 		if (start)
2062 			order = min_t(int, MAX_ORDER, __ffs(start));
2063 		else
2064 			order = MAX_ORDER;
2065 
2066 		while (start + (1UL << order) > end)
2067 			order--;
2068 
2069 		memblock_free_pages(pfn_to_page(start), start, order);
2070 
2071 		start += (1UL << order);
2072 	}
2073 }
2074 
2075 static unsigned long __init __free_memory_core(phys_addr_t start,
2076 				 phys_addr_t end)
2077 {
2078 	unsigned long start_pfn = PFN_UP(start);
2079 	unsigned long end_pfn = min_t(unsigned long,
2080 				      PFN_DOWN(end), max_low_pfn);
2081 
2082 	if (start_pfn >= end_pfn)
2083 		return 0;
2084 
2085 	__free_pages_memory(start_pfn, end_pfn);
2086 
2087 	return end_pfn - start_pfn;
2088 }
2089 
2090 static void __init memmap_init_reserved_pages(void)
2091 {
2092 	struct memblock_region *region;
2093 	phys_addr_t start, end;
2094 	int nid;
2095 
2096 	/*
2097 	 * set nid on all reserved pages and also treat struct
2098 	 * pages for the NOMAP regions as PageReserved
2099 	 */
2100 	for_each_mem_region(region) {
2101 		nid = memblock_get_region_node(region);
2102 		start = region->base;
2103 		end = start + region->size;
2104 
2105 		if (memblock_is_nomap(region))
2106 			reserve_bootmem_region(start, end, nid);
2107 
2108 		memblock_set_node(start, end, &memblock.reserved, nid);
2109 	}
2110 
2111 	/* initialize struct pages for the reserved regions */
2112 	for_each_reserved_mem_region(region) {
2113 		nid = memblock_get_region_node(region);
2114 		start = region->base;
2115 		end = start + region->size;
2116 
2117 		reserve_bootmem_region(start, end, nid);
2118 	}
2119 }
2120 
2121 static unsigned long __init free_low_memory_core_early(void)
2122 {
2123 	unsigned long count = 0;
2124 	phys_addr_t start, end;
2125 	u64 i;
2126 
2127 	memblock_clear_hotplug(0, -1);
2128 
2129 	memmap_init_reserved_pages();
2130 
2131 	/*
2132 	 * We need to use NUMA_NO_NODE instead of NODE_DATA(0)->node_id
2133 	 *  because in some case like Node0 doesn't have RAM installed
2134 	 *  low ram will be on Node1
2135 	 */
2136 	for_each_free_mem_range(i, NUMA_NO_NODE, MEMBLOCK_NONE, &start, &end,
2137 				NULL)
2138 		count += __free_memory_core(start, end);
2139 
2140 	return count;
2141 }
2142 
2143 static int reset_managed_pages_done __initdata;
2144 
2145 static void __init reset_node_managed_pages(pg_data_t *pgdat)
2146 {
2147 	struct zone *z;
2148 
2149 	for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++)
2150 		atomic_long_set(&z->managed_pages, 0);
2151 }
2152 
2153 void __init reset_all_zones_managed_pages(void)
2154 {
2155 	struct pglist_data *pgdat;
2156 
2157 	if (reset_managed_pages_done)
2158 		return;
2159 
2160 	for_each_online_pgdat(pgdat)
2161 		reset_node_managed_pages(pgdat);
2162 
2163 	reset_managed_pages_done = 1;
2164 }
2165 
2166 /**
2167  * memblock_free_all - release free pages to the buddy allocator
2168  */
2169 void __init memblock_free_all(void)
2170 {
2171 	unsigned long pages;
2172 
2173 	free_unused_memmap();
2174 	reset_all_zones_managed_pages();
2175 
2176 	pages = free_low_memory_core_early();
2177 	totalram_pages_add(pages);
2178 }
2179 
2180 #if defined(CONFIG_DEBUG_FS) && defined(CONFIG_ARCH_KEEP_MEMBLOCK)
2181 static const char * const flagname[] = {
2182 	[ilog2(MEMBLOCK_HOTPLUG)] = "HOTPLUG",
2183 	[ilog2(MEMBLOCK_MIRROR)] = "MIRROR",
2184 	[ilog2(MEMBLOCK_NOMAP)] = "NOMAP",
2185 	[ilog2(MEMBLOCK_DRIVER_MANAGED)] = "DRV_MNG",
2186 };
2187 
2188 static int memblock_debug_show(struct seq_file *m, void *private)
2189 {
2190 	struct memblock_type *type = m->private;
2191 	struct memblock_region *reg;
2192 	int i, j, nid;
2193 	unsigned int count = ARRAY_SIZE(flagname);
2194 	phys_addr_t end;
2195 
2196 	for (i = 0; i < type->cnt; i++) {
2197 		reg = &type->regions[i];
2198 		end = reg->base + reg->size - 1;
2199 		nid = memblock_get_region_node(reg);
2200 
2201 		seq_printf(m, "%4d: ", i);
2202 		seq_printf(m, "%pa..%pa ", &reg->base, &end);
2203 		if (nid != MAX_NUMNODES)
2204 			seq_printf(m, "%4d ", nid);
2205 		else
2206 			seq_printf(m, "%4c ", 'x');
2207 		if (reg->flags) {
2208 			for (j = 0; j < count; j++) {
2209 				if (reg->flags & (1U << j)) {
2210 					seq_printf(m, "%s\n", flagname[j]);
2211 					break;
2212 				}
2213 			}
2214 			if (j == count)
2215 				seq_printf(m, "%s\n", "UNKNOWN");
2216 		} else {
2217 			seq_printf(m, "%s\n", "NONE");
2218 		}
2219 	}
2220 	return 0;
2221 }
2222 DEFINE_SHOW_ATTRIBUTE(memblock_debug);
2223 
2224 static int __init memblock_init_debugfs(void)
2225 {
2226 	struct dentry *root = debugfs_create_dir("memblock", NULL);
2227 
2228 	debugfs_create_file("memory", 0444, root,
2229 			    &memblock.memory, &memblock_debug_fops);
2230 	debugfs_create_file("reserved", 0444, root,
2231 			    &memblock.reserved, &memblock_debug_fops);
2232 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
2233 	debugfs_create_file("physmem", 0444, root, &physmem,
2234 			    &memblock_debug_fops);
2235 #endif
2236 
2237 	return 0;
2238 }
2239 __initcall(memblock_init_debugfs);
2240 
2241 #endif /* CONFIG_DEBUG_FS */
2242