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