• 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 #include <linux/sort.h>
23 #include <linux/proc_fs.h>
24 
25 #include "internal.h"
26 
27 #define INIT_MEMBLOCK_REGIONS			128
28 #define INIT_PHYSMEM_REGIONS			4
29 
30 #ifndef INIT_MEMBLOCK_RESERVED_REGIONS
31 # define INIT_MEMBLOCK_RESERVED_REGIONS		INIT_MEMBLOCK_REGIONS
32 #endif
33 
34 #ifndef INIT_MEMBLOCK_MEMORY_REGIONS
35 #define INIT_MEMBLOCK_MEMORY_REGIONS		INIT_MEMBLOCK_REGIONS
36 #endif
37 
38 /**
39  * DOC: memblock overview
40  *
41  * Memblock is a method of managing memory regions during the early
42  * boot period when the usual kernel memory allocators are not up and
43  * running.
44  *
45  * Memblock views the system memory as collections of contiguous
46  * regions. There are several types of these collections:
47  *
48  * * ``memory`` - describes the physical memory available to the
49  *   kernel; this may differ from the actual physical memory installed
50  *   in the system, for instance when the memory is restricted with
51  *   ``mem=`` command line parameter
52  * * ``reserved`` - describes the regions that were allocated
53  * * ``physmem`` - describes the actual physical memory available during
54  *   boot regardless of the possible restrictions and memory hot(un)plug;
55  *   the ``physmem`` type is only available on some architectures.
56  *
57  * Each region is represented by struct memblock_region that
58  * defines the region extents, its attributes and NUMA node id on NUMA
59  * systems. Every memory type is described by the struct memblock_type
60  * which contains an array of memory regions along with
61  * the allocator metadata. The "memory" and "reserved" types are nicely
62  * wrapped with struct memblock. This structure is statically
63  * initialized at build time. The region arrays are initially sized to
64  * %INIT_MEMBLOCK_MEMORY_REGIONS for "memory" and
65  * %INIT_MEMBLOCK_RESERVED_REGIONS for "reserved". The region array
66  * for "physmem" is initially sized to %INIT_PHYSMEM_REGIONS.
67  * The memblock_allow_resize() enables automatic resizing of the region
68  * arrays during addition of new regions. This feature should be used
69  * with care so that memory allocated for the region array will not
70  * overlap with areas that should be reserved, for example initrd.
71  *
72  * The early architecture setup should tell memblock what the physical
73  * memory layout is by using memblock_add() or memblock_add_node()
74  * functions. The first function does not assign the region to a NUMA
75  * node and it is appropriate for UMA systems. Yet, it is possible to
76  * use it on NUMA systems as well and assign the region to a NUMA node
77  * later in the setup process using memblock_set_node(). The
78  * memblock_add_node() performs such an assignment directly.
79  *
80  * Once memblock is setup the memory can be allocated using one of the
81  * API variants:
82  *
83  * * memblock_phys_alloc*() - these functions return the **physical**
84  *   address of the allocated memory
85  * * memblock_alloc*() - these functions return the **virtual** address
86  *   of the allocated memory.
87  *
88  * Note, that both API variants use implicit assumptions about allowed
89  * memory ranges and the fallback methods. Consult the documentation
90  * of memblock_alloc_internal() and memblock_alloc_range_nid()
91  * functions for more elaborate description.
92  *
93  * As the system boot progresses, the architecture specific mem_init()
94  * function frees all the memory to the buddy page allocator.
95  *
96  * Unless an architecture enables %CONFIG_ARCH_KEEP_MEMBLOCK, the
97  * memblock data structures (except "physmem") will be discarded after the
98  * system initialization completes.
99  */
100 
101 #ifndef CONFIG_NUMA
102 struct pglist_data __refdata contig_page_data;
103 EXPORT_SYMBOL(contig_page_data);
104 #endif
105 
106 unsigned long max_low_pfn;
107 unsigned long min_low_pfn;
108 unsigned long max_pfn;
109 unsigned long long max_possible_pfn;
110 
111 static struct memblock_region memblock_memory_init_regions[INIT_MEMBLOCK_MEMORY_REGIONS] __initdata_memblock;
112 static struct memblock_region memblock_reserved_init_regions[INIT_MEMBLOCK_RESERVED_REGIONS] __initdata_memblock;
113 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
114 static struct memblock_region memblock_physmem_init_regions[INIT_PHYSMEM_REGIONS];
115 #endif
116 
117 struct memblock memblock __initdata_memblock = {
118 	.memory.regions		= memblock_memory_init_regions,
119 	.memory.cnt		= 1,	/* empty dummy entry */
120 	.memory.max		= INIT_MEMBLOCK_MEMORY_REGIONS,
121 	.memory.name		= "memory",
122 
123 	.reserved.regions	= memblock_reserved_init_regions,
124 	.reserved.cnt		= 1,	/* empty dummy entry */
125 	.reserved.max		= INIT_MEMBLOCK_RESERVED_REGIONS,
126 	.reserved.name		= "reserved",
127 
128 	.bottom_up		= false,
129 	.current_limit		= MEMBLOCK_ALLOC_ANYWHERE,
130 };
131 
132 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
133 struct memblock_type physmem = {
134 	.regions		= memblock_physmem_init_regions,
135 	.cnt			= 1,	/* empty dummy entry */
136 	.max			= INIT_PHYSMEM_REGIONS,
137 	.name			= "physmem",
138 };
139 #endif
140 
141 static long memsize_kinit;
142 static bool memblock_memsize_tracking __initdata_memblock = true;
143 
144 /*
145  * keep a pointer to &memblock.memory in the text section to use it in
146  * __next_mem_range() and its helpers.
147  *  For architectures that do not keep memblock data after init, this
148  * pointer will be reset to NULL at memblock_discard()
149  */
150 static __refdata struct memblock_type *memblock_memory = &memblock.memory;
151 
152 #define for_each_memblock_type(i, memblock_type, rgn)			\
153 	for (i = 0, rgn = &memblock_type->regions[0];			\
154 	     i < memblock_type->cnt;					\
155 	     i++, rgn = &memblock_type->regions[i])
156 
157 #define memblock_dbg(fmt, ...)						\
158 	do {								\
159 		if (memblock_debug)					\
160 			pr_info(fmt, ##__VA_ARGS__);			\
161 	} while (0)
162 
163 static int memblock_debug __initdata_memblock;
164 static bool system_has_some_mirror __initdata_memblock;
165 static int memblock_can_resize __initdata_memblock;
166 static int memblock_memory_in_slab __initdata_memblock;
167 static int memblock_reserved_in_slab __initdata_memblock;
168 
memblock_has_mirror(void)169 bool __init_memblock memblock_has_mirror(void)
170 {
171 	return system_has_some_mirror;
172 }
173 
choose_memblock_flags(void)174 static enum memblock_flags __init_memblock choose_memblock_flags(void)
175 {
176 	return system_has_some_mirror ? MEMBLOCK_MIRROR : MEMBLOCK_NONE;
177 }
178 
179 /* adjust *@size so that (@base + *@size) doesn't overflow, return new size */
memblock_cap_size(phys_addr_t base,phys_addr_t * size)180 static inline phys_addr_t memblock_cap_size(phys_addr_t base, phys_addr_t *size)
181 {
182 	return *size = min(*size, PHYS_ADDR_MAX - base);
183 }
184 
185 /*
186  * Address comparison utilities
187  */
188 unsigned long __init_memblock
memblock_addrs_overlap(phys_addr_t base1,phys_addr_t size1,phys_addr_t base2,phys_addr_t size2)189 memblock_addrs_overlap(phys_addr_t base1, phys_addr_t size1, phys_addr_t base2,
190 		       phys_addr_t size2)
191 {
192 	return ((base1 < (base2 + size2)) && (base2 < (base1 + size1)));
193 }
194 
memblock_overlaps_region(struct memblock_type * type,phys_addr_t base,phys_addr_t size)195 bool __init_memblock memblock_overlaps_region(struct memblock_type *type,
196 					phys_addr_t base, phys_addr_t size)
197 {
198 	unsigned long i;
199 
200 	memblock_cap_size(base, &size);
201 
202 	for (i = 0; i < type->cnt; i++)
203 		if (memblock_addrs_overlap(base, size, type->regions[i].base,
204 					   type->regions[i].size))
205 			break;
206 	return i < type->cnt;
207 }
208 
209 /**
210  * __memblock_find_range_bottom_up - find free area utility in bottom-up
211  * @start: start of candidate range
212  * @end: end of candidate range, can be %MEMBLOCK_ALLOC_ANYWHERE or
213  *       %MEMBLOCK_ALLOC_ACCESSIBLE
214  * @size: size of free area to find
215  * @align: alignment of free area to find
216  * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
217  * @flags: pick from blocks based on memory attributes
218  *
219  * Utility called from memblock_find_in_range_node(), find free area bottom-up.
220  *
221  * Return:
222  * Found address on success, 0 on failure.
223  */
224 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)225 __memblock_find_range_bottom_up(phys_addr_t start, phys_addr_t end,
226 				phys_addr_t size, phys_addr_t align, int nid,
227 				enum memblock_flags flags)
228 {
229 	phys_addr_t this_start, this_end, cand;
230 	u64 i;
231 
232 	for_each_free_mem_range(i, nid, flags, &this_start, &this_end, NULL) {
233 		this_start = clamp(this_start, start, end);
234 		this_end = clamp(this_end, start, end);
235 
236 		cand = round_up(this_start, align);
237 		if (cand < this_end && this_end - cand >= size)
238 			return cand;
239 	}
240 
241 	return 0;
242 }
243 
244 /**
245  * __memblock_find_range_top_down - find free area utility, in top-down
246  * @start: start of candidate range
247  * @end: end of candidate range, can be %MEMBLOCK_ALLOC_ANYWHERE or
248  *       %MEMBLOCK_ALLOC_ACCESSIBLE
249  * @size: size of free area to find
250  * @align: alignment of free area to find
251  * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
252  * @flags: pick from blocks based on memory attributes
253  *
254  * Utility called from memblock_find_in_range_node(), find free area top-down.
255  *
256  * Return:
257  * Found address on success, 0 on failure.
258  */
259 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)260 __memblock_find_range_top_down(phys_addr_t start, phys_addr_t end,
261 			       phys_addr_t size, phys_addr_t align, int nid,
262 			       enum memblock_flags flags)
263 {
264 	phys_addr_t this_start, this_end, cand;
265 	u64 i;
266 
267 	for_each_free_mem_range_reverse(i, nid, flags, &this_start, &this_end,
268 					NULL) {
269 		this_start = clamp(this_start, start, end);
270 		this_end = clamp(this_end, start, end);
271 
272 		if (this_end < size)
273 			continue;
274 
275 		cand = round_down(this_end - size, align);
276 		if (cand >= this_start)
277 			return cand;
278 	}
279 
280 	return 0;
281 }
282 
283 /**
284  * memblock_find_in_range_node - find free area in given range and node
285  * @size: size of free area to find
286  * @align: alignment of free area to find
287  * @start: start of candidate range
288  * @end: end of candidate range, can be %MEMBLOCK_ALLOC_ANYWHERE or
289  *       %MEMBLOCK_ALLOC_ACCESSIBLE
290  * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
291  * @flags: pick from blocks based on memory attributes
292  *
293  * Find @size free area aligned to @align in the specified range and node.
294  *
295  * Return:
296  * Found address on success, 0 on failure.
297  */
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)298 static phys_addr_t __init_memblock memblock_find_in_range_node(phys_addr_t size,
299 					phys_addr_t align, phys_addr_t start,
300 					phys_addr_t end, int nid,
301 					enum memblock_flags flags)
302 {
303 	/* pump up @end */
304 	if (end == MEMBLOCK_ALLOC_ACCESSIBLE ||
305 	    end == MEMBLOCK_ALLOC_NOLEAKTRACE)
306 		end = memblock.current_limit;
307 
308 	/* avoid allocating the first page */
309 	start = max_t(phys_addr_t, start, PAGE_SIZE);
310 	end = max(start, end);
311 
312 	if (memblock_bottom_up())
313 		return __memblock_find_range_bottom_up(start, end, size, align,
314 						       nid, flags);
315 	else
316 		return __memblock_find_range_top_down(start, end, size, align,
317 						      nid, flags);
318 }
319 
320 /**
321  * memblock_find_in_range - find free area in given range
322  * @start: start of candidate range
323  * @end: end of candidate range, can be %MEMBLOCK_ALLOC_ANYWHERE or
324  *       %MEMBLOCK_ALLOC_ACCESSIBLE
325  * @size: size of free area to find
326  * @align: alignment of free area to find
327  *
328  * Find @size free area aligned to @align in the specified range.
329  *
330  * Return:
331  * Found address on success, 0 on failure.
332  */
memblock_find_in_range(phys_addr_t start,phys_addr_t end,phys_addr_t size,phys_addr_t align)333 static phys_addr_t __init_memblock memblock_find_in_range(phys_addr_t start,
334 					phys_addr_t end, phys_addr_t size,
335 					phys_addr_t align)
336 {
337 	phys_addr_t ret;
338 	enum memblock_flags flags = choose_memblock_flags();
339 
340 again:
341 	ret = memblock_find_in_range_node(size, align, start, end,
342 					    NUMA_NO_NODE, flags);
343 
344 	if (!ret && (flags & MEMBLOCK_MIRROR)) {
345 		pr_warn_ratelimited("Could not allocate %pap bytes of mirrored memory\n",
346 			&size);
347 		flags &= ~MEMBLOCK_MIRROR;
348 		goto again;
349 	}
350 
351 	return ret;
352 }
353 
memblock_remove_region(struct memblock_type * type,unsigned long r)354 static void __init_memblock memblock_remove_region(struct memblock_type *type, unsigned long r)
355 {
356 	type->total_size -= type->regions[r].size;
357 	memmove(&type->regions[r], &type->regions[r + 1],
358 		(type->cnt - (r + 1)) * sizeof(type->regions[r]));
359 	type->cnt--;
360 
361 	/* Special case for empty arrays */
362 	if (type->cnt == 0) {
363 		WARN_ON(type->total_size != 0);
364 		type->cnt = 1;
365 		type->regions[0].base = 0;
366 		type->regions[0].size = 0;
367 		type->regions[0].flags = 0;
368 		memblock_set_region_node(&type->regions[0], MAX_NUMNODES);
369 	}
370 }
371 
372 #ifndef CONFIG_ARCH_KEEP_MEMBLOCK
373 /**
374  * memblock_discard - discard memory and reserved arrays if they were allocated
375  */
memblock_discard(void)376 void __init memblock_discard(void)
377 {
378 	phys_addr_t addr, size;
379 
380 	if (memblock.reserved.regions != memblock_reserved_init_regions) {
381 		addr = __pa(memblock.reserved.regions);
382 		size = PAGE_ALIGN(sizeof(struct memblock_region) *
383 				  memblock.reserved.max);
384 		if (memblock_reserved_in_slab)
385 			kfree(memblock.reserved.regions);
386 		else
387 			memblock_free_late(addr, size);
388 	}
389 
390 	if (memblock.memory.regions != memblock_memory_init_regions) {
391 		addr = __pa(memblock.memory.regions);
392 		size = PAGE_ALIGN(sizeof(struct memblock_region) *
393 				  memblock.memory.max);
394 		if (memblock_memory_in_slab)
395 			kfree(memblock.memory.regions);
396 		else
397 			memblock_free_late(addr, size);
398 	}
399 
400 	memblock_memory = NULL;
401 }
402 #endif
403 
404 /**
405  * memblock_double_array - double the size of the memblock regions array
406  * @type: memblock type of the regions array being doubled
407  * @new_area_start: starting address of memory range to avoid overlap with
408  * @new_area_size: size of memory range to avoid overlap with
409  *
410  * Double the size of the @type regions array. If memblock is being used to
411  * allocate memory for a new reserved regions array and there is a previously
412  * allocated memory range [@new_area_start, @new_area_start + @new_area_size]
413  * waiting to be reserved, ensure the memory used by the new array does
414  * not overlap.
415  *
416  * Return:
417  * 0 on success, -1 on failure.
418  */
memblock_double_array(struct memblock_type * type,phys_addr_t new_area_start,phys_addr_t new_area_size)419 static int __init_memblock memblock_double_array(struct memblock_type *type,
420 						phys_addr_t new_area_start,
421 						phys_addr_t new_area_size)
422 {
423 	struct memblock_region *new_array, *old_array;
424 	phys_addr_t old_alloc_size, new_alloc_size;
425 	phys_addr_t old_size, new_size, addr, new_end;
426 	int use_slab = slab_is_available();
427 	int *in_slab;
428 
429 	/* We don't allow resizing until we know about the reserved regions
430 	 * of memory that aren't suitable for allocation
431 	 */
432 	if (!memblock_can_resize)
433 		return -1;
434 
435 	/* Calculate new doubled size */
436 	old_size = type->max * sizeof(struct memblock_region);
437 	new_size = old_size << 1;
438 	/*
439 	 * We need to allocated new one align to PAGE_SIZE,
440 	 *   so we can free them completely later.
441 	 */
442 	old_alloc_size = PAGE_ALIGN(old_size);
443 	new_alloc_size = PAGE_ALIGN(new_size);
444 
445 	/* Retrieve the slab flag */
446 	if (type == &memblock.memory)
447 		in_slab = &memblock_memory_in_slab;
448 	else
449 		in_slab = &memblock_reserved_in_slab;
450 
451 	/* Try to find some space for it */
452 	if (use_slab) {
453 		new_array = kmalloc(new_size, GFP_KERNEL);
454 		addr = new_array ? __pa(new_array) : 0;
455 	} else {
456 		/* only exclude range when trying to double reserved.regions */
457 		if (type != &memblock.reserved)
458 			new_area_start = new_area_size = 0;
459 
460 		addr = memblock_find_in_range(new_area_start + new_area_size,
461 						memblock.current_limit,
462 						new_alloc_size, PAGE_SIZE);
463 		if (!addr && new_area_size)
464 			addr = memblock_find_in_range(0,
465 				min(new_area_start, memblock.current_limit),
466 				new_alloc_size, PAGE_SIZE);
467 
468 		new_array = addr ? __va(addr) : NULL;
469 	}
470 	if (!addr) {
471 		pr_err("memblock: Failed to double %s array from %ld to %ld entries !\n",
472 		       type->name, type->max, type->max * 2);
473 		return -1;
474 	}
475 
476 	new_end = addr + new_size - 1;
477 	memblock_dbg("memblock: %s is doubled to %ld at [%pa-%pa]",
478 			type->name, type->max * 2, &addr, &new_end);
479 
480 	/*
481 	 * Found space, we now need to move the array over before we add the
482 	 * reserved region since it may be our reserved array itself that is
483 	 * full.
484 	 */
485 	memcpy(new_array, type->regions, old_size);
486 	memset(new_array + type->max, 0, old_size);
487 	old_array = type->regions;
488 	type->regions = new_array;
489 	type->max <<= 1;
490 
491 	/* Free old array. We needn't free it if the array is the static one */
492 	if (*in_slab)
493 		kfree(old_array);
494 	else if (old_array != memblock_memory_init_regions &&
495 		 old_array != memblock_reserved_init_regions)
496 		memblock_free(old_array, old_alloc_size);
497 
498 	/*
499 	 * Reserve the new array if that comes from the memblock.  Otherwise, we
500 	 * needn't do it
501 	 */
502 	if (!use_slab)
503 		BUG_ON(memblock_reserve(addr, new_alloc_size));
504 
505 	/* Update slab flag */
506 	*in_slab = use_slab;
507 
508 	return 0;
509 }
510 
511 /**
512  * memblock_merge_regions - merge neighboring compatible regions
513  * @type: memblock type to scan
514  * @start_rgn: start scanning from (@start_rgn - 1)
515  * @end_rgn: end scanning at (@end_rgn - 1)
516  * Scan @type and merge neighboring compatible regions in [@start_rgn - 1, @end_rgn)
517  */
memblock_merge_regions(struct memblock_type * type,unsigned long start_rgn,unsigned long end_rgn)518 static void __init_memblock memblock_merge_regions(struct memblock_type *type,
519 						   unsigned long start_rgn,
520 						   unsigned long end_rgn)
521 {
522 	int i = 0;
523 	if (start_rgn)
524 		i = start_rgn - 1;
525 	end_rgn = min(end_rgn, type->cnt - 1);
526 	while (i < end_rgn) {
527 		struct memblock_region *this = &type->regions[i];
528 		struct memblock_region *next = &type->regions[i + 1];
529 
530 		if (this->base + this->size != next->base ||
531 		    memblock_get_region_node(this) !=
532 		    memblock_get_region_node(next) ||
533 		    this->flags != next->flags) {
534 			BUG_ON(this->base + this->size > next->base);
535 			i++;
536 			continue;
537 		}
538 
539 		this->size += next->size;
540 		/* move forward from next + 1, index of which is i + 2 */
541 		memmove(next, next + 1, (type->cnt - (i + 2)) * sizeof(*next));
542 		type->cnt--;
543 		end_rgn--;
544 	}
545 }
546 
547 /**
548  * memblock_insert_region - insert new memblock region
549  * @type:	memblock type to insert into
550  * @idx:	index for the insertion point
551  * @base:	base address of the new region
552  * @size:	size of the new region
553  * @nid:	node id of the new region
554  * @flags:	flags of the new region
555  *
556  * Insert new memblock region [@base, @base + @size) into @type at @idx.
557  * @type must already have extra room to accommodate the new region.
558  */
memblock_insert_region(struct memblock_type * type,int idx,phys_addr_t base,phys_addr_t size,int nid,enum memblock_flags flags)559 static void __init_memblock memblock_insert_region(struct memblock_type *type,
560 						   int idx, phys_addr_t base,
561 						   phys_addr_t size,
562 						   int nid,
563 						   enum memblock_flags flags)
564 {
565 	struct memblock_region *rgn = &type->regions[idx];
566 
567 	BUG_ON(type->cnt >= type->max);
568 	memmove(rgn + 1, rgn, (type->cnt - idx) * sizeof(*rgn));
569 	rgn->base = base;
570 	rgn->size = size;
571 	rgn->flags = flags;
572 	memblock_set_region_node(rgn, nid);
573 	type->cnt++;
574 	type->total_size += size;
575 }
576 
577 /**
578  * memblock_add_range - add new memblock region
579  * @type: memblock type to add new region into
580  * @base: base address of the new region
581  * @size: size of the new region
582  * @nid: nid of the new region
583  * @flags: flags of the new region
584  *
585  * Add new memblock region [@base, @base + @size) into @type.  The new region
586  * is allowed to overlap with existing ones - overlaps don't affect already
587  * existing regions.  @type is guaranteed to be minimal (all neighbouring
588  * compatible regions are merged) after the addition.
589  *
590  * Return:
591  * 0 on success, -errno on failure.
592  */
memblock_add_range(struct memblock_type * type,phys_addr_t base,phys_addr_t size,int nid,enum memblock_flags flags)593 static int __init_memblock memblock_add_range(struct memblock_type *type,
594 				phys_addr_t base, phys_addr_t size,
595 				int nid, enum memblock_flags flags)
596 {
597 	bool insert = false;
598 	phys_addr_t obase = base;
599 	phys_addr_t end = base + memblock_cap_size(base, &size);
600 	int idx, nr_new, start_rgn = -1, end_rgn;
601 	struct memblock_region *rgn;
602 	phys_addr_t new_size = 0;
603 
604 	if (!size)
605 		return 0;
606 
607 	/* special case for empty array */
608 	if (type->regions[0].size == 0) {
609 		WARN_ON(type->cnt != 1 || type->total_size);
610 		type->regions[0].base = base;
611 		type->regions[0].size = size;
612 		type->regions[0].flags = flags;
613 		memblock_set_region_node(&type->regions[0], nid);
614 		type->total_size = size;
615 		new_size = size;
616 		goto done;
617 	}
618 
619 	/*
620 	 * The worst case is when new range overlaps all existing regions,
621 	 * then we'll need type->cnt + 1 empty regions in @type. So if
622 	 * type->cnt * 2 + 1 is less than or equal to type->max, we know
623 	 * that there is enough empty regions in @type, and we can insert
624 	 * regions directly.
625 	 */
626 	if (type->cnt * 2 + 1 <= type->max)
627 		insert = true;
628 
629 repeat:
630 	/*
631 	 * The following is executed twice.  Once with %false @insert and
632 	 * then with %true.  The first counts the number of regions needed
633 	 * to accommodate the new area.  The second actually inserts them.
634 	 */
635 	base = obase;
636 	nr_new = 0;
637 
638 	for_each_memblock_type(idx, type, rgn) {
639 		phys_addr_t rbase = rgn->base;
640 		phys_addr_t rend = rbase + rgn->size;
641 
642 		if (rbase >= end)
643 			break;
644 		if (rend <= base)
645 			continue;
646 		/*
647 		 * @rgn overlaps.  If it separates the lower part of new
648 		 * area, insert that portion.
649 		 */
650 		if (rbase > base) {
651 #ifdef CONFIG_NUMA
652 			WARN_ON(nid != memblock_get_region_node(rgn));
653 #endif
654 			WARN_ON(flags != rgn->flags);
655 			nr_new++;
656 			if (insert) {
657 				if (start_rgn == -1)
658 					start_rgn = idx;
659 				end_rgn = idx + 1;
660 				memblock_insert_region(type, idx++, base,
661 						       rbase - base, nid,
662 						       flags);
663 				new_size += rbase - base;
664 			}
665 		}
666 		/* area below @rend is dealt with, forget about it */
667 		base = min(rend, end);
668 	}
669 
670 	/* insert the remaining portion */
671 	if (base < end) {
672 		nr_new++;
673 		if (insert) {
674 			if (start_rgn == -1)
675 				start_rgn = idx;
676 			end_rgn = idx + 1;
677 			memblock_insert_region(type, idx, base, end - base,
678 					       nid, flags);
679 			new_size += end - base;
680 		}
681 	}
682 
683 	if (!nr_new)
684 		return 0;
685 
686 	/*
687 	 * If this was the first round, resize array and repeat for actual
688 	 * insertions; otherwise, merge and return.
689 	 */
690 	if (!insert) {
691 		while (type->cnt + nr_new > type->max)
692 			if (memblock_double_array(type, obase, size) < 0)
693 				return -ENOMEM;
694 		insert = true;
695 		goto repeat;
696 	} else {
697 		memblock_merge_regions(type, start_rgn, end_rgn);
698 	}
699 done:
700 	if (memblock_memsize_tracking) {
701 		if (new_size && type == &memblock.reserved) {
702 			memblock_dbg("%s: kernel %lu %+ld\n", __func__,
703 				     memsize_kinit, (unsigned long)new_size);
704 			memsize_kinit += size;
705 		}
706 	}
707 	return 0;
708 }
709 
710 /**
711  * memblock_add_node - add new memblock region within a NUMA node
712  * @base: base address of the new region
713  * @size: size of the new region
714  * @nid: nid of the new region
715  * @flags: flags of the new region
716  *
717  * Add new memblock region [@base, @base + @size) to the "memory"
718  * type. See memblock_add_range() description for mode details
719  *
720  * Return:
721  * 0 on success, -errno on failure.
722  */
memblock_add_node(phys_addr_t base,phys_addr_t size,int nid,enum memblock_flags flags)723 int __init_memblock memblock_add_node(phys_addr_t base, phys_addr_t size,
724 				      int nid, enum memblock_flags flags)
725 {
726 	phys_addr_t end = base + size - 1;
727 
728 	memblock_dbg("%s: [%pa-%pa] nid=%d flags=%x %pS\n", __func__,
729 		     &base, &end, nid, flags, (void *)_RET_IP_);
730 
731 	return memblock_add_range(&memblock.memory, base, size, nid, flags);
732 }
733 
734 /**
735  * memblock_add - add new memblock region
736  * @base: base address of the new region
737  * @size: size of the new region
738  *
739  * Add new memblock region [@base, @base + @size) to the "memory"
740  * type. See memblock_add_range() description for mode details
741  *
742  * Return:
743  * 0 on success, -errno on failure.
744  */
memblock_add(phys_addr_t base,phys_addr_t size)745 int __init_memblock memblock_add(phys_addr_t base, phys_addr_t size)
746 {
747 	phys_addr_t end = base + size - 1;
748 
749 	memblock_dbg("%s: [%pa-%pa] %pS\n", __func__,
750 		     &base, &end, (void *)_RET_IP_);
751 
752 	return memblock_add_range(&memblock.memory, base, size, MAX_NUMNODES, 0);
753 }
754 
755 /**
756  * memblock_isolate_range - isolate given range into disjoint memblocks
757  * @type: memblock type to isolate range for
758  * @base: base of range to isolate
759  * @size: size of range to isolate
760  * @start_rgn: out parameter for the start of isolated region
761  * @end_rgn: out parameter for the end of isolated region
762  *
763  * Walk @type and ensure that regions don't cross the boundaries defined by
764  * [@base, @base + @size).  Crossing regions are split at the boundaries,
765  * which may create at most two more regions.  The index of the first
766  * region inside the range is returned in *@start_rgn and end in *@end_rgn.
767  *
768  * Return:
769  * 0 on success, -errno on failure.
770  */
memblock_isolate_range(struct memblock_type * type,phys_addr_t base,phys_addr_t size,int * start_rgn,int * end_rgn)771 static int __init_memblock memblock_isolate_range(struct memblock_type *type,
772 					phys_addr_t base, phys_addr_t size,
773 					int *start_rgn, int *end_rgn)
774 {
775 	phys_addr_t end = base + memblock_cap_size(base, &size);
776 	int idx;
777 	struct memblock_region *rgn;
778 
779 	*start_rgn = *end_rgn = 0;
780 
781 	if (!size)
782 		return 0;
783 
784 	/* we'll create at most two more regions */
785 	while (type->cnt + 2 > type->max)
786 		if (memblock_double_array(type, base, size) < 0)
787 			return -ENOMEM;
788 
789 	for_each_memblock_type(idx, type, rgn) {
790 		phys_addr_t rbase = rgn->base;
791 		phys_addr_t rend = rbase + rgn->size;
792 
793 		if (rbase >= end)
794 			break;
795 		if (rend <= base)
796 			continue;
797 
798 		if (rbase < base) {
799 			/*
800 			 * @rgn intersects from below.  Split and continue
801 			 * to process the next region - the new top half.
802 			 */
803 			rgn->base = base;
804 			rgn->size -= base - rbase;
805 			type->total_size -= base - rbase;
806 			memblock_insert_region(type, idx, rbase, base - rbase,
807 					       memblock_get_region_node(rgn),
808 					       rgn->flags);
809 		} else if (rend > end) {
810 			/*
811 			 * @rgn intersects from above.  Split and redo the
812 			 * current region - the new bottom half.
813 			 */
814 			rgn->base = end;
815 			rgn->size -= end - rbase;
816 			type->total_size -= end - rbase;
817 			memblock_insert_region(type, idx--, rbase, end - rbase,
818 					       memblock_get_region_node(rgn),
819 					       rgn->flags);
820 		} else {
821 			/* @rgn is fully contained, record it */
822 			if (!*end_rgn)
823 				*start_rgn = idx;
824 			*end_rgn = idx + 1;
825 		}
826 	}
827 
828 	return 0;
829 }
830 
memblock_remove_range(struct memblock_type * type,phys_addr_t base,phys_addr_t size)831 static int __init_memblock memblock_remove_range(struct memblock_type *type,
832 					  phys_addr_t base, phys_addr_t size)
833 {
834 	int start_rgn, end_rgn;
835 	int i, ret;
836 
837 	ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
838 	if (ret)
839 		return ret;
840 
841 	for (i = end_rgn - 1; i >= start_rgn; i--)
842 		memblock_remove_region(type, i);
843 	if (memblock_memsize_tracking) {
844 		if (type == &memblock.reserved) {
845 			memblock_dbg("%s: kernel %lu %+ld\n", __func__,
846 				     memsize_kinit, (unsigned long)size);
847 			memsize_kinit -= size;
848 		}
849 	}
850 	return 0;
851 }
852 
memblock_remove(phys_addr_t base,phys_addr_t size)853 int __init_memblock memblock_remove(phys_addr_t base, phys_addr_t size)
854 {
855 	phys_addr_t end = base + size - 1;
856 
857 	memblock_dbg("%s: [%pa-%pa] %pS\n", __func__,
858 		     &base, &end, (void *)_RET_IP_);
859 
860 	return memblock_remove_range(&memblock.memory, base, size);
861 }
862 
863 /**
864  * memblock_free - free boot memory allocation
865  * @ptr: starting address of the  boot memory allocation
866  * @size: size of the boot memory block in bytes
867  *
868  * Free boot memory block previously allocated by memblock_alloc_xx() API.
869  * The freeing memory will not be released to the buddy allocator.
870  */
memblock_free(void * ptr,size_t size)871 void __init_memblock memblock_free(void *ptr, size_t size)
872 {
873 	if (ptr)
874 		memblock_phys_free(__pa(ptr), size);
875 }
876 
877 /**
878  * memblock_phys_free - free boot memory block
879  * @base: phys starting address of the  boot memory block
880  * @size: size of the boot memory block in bytes
881  *
882  * Free boot memory block previously allocated by memblock_phys_alloc_xx() API.
883  * The freeing memory will not be released to the buddy allocator.
884  */
memblock_phys_free(phys_addr_t base,phys_addr_t size)885 int __init_memblock memblock_phys_free(phys_addr_t base, phys_addr_t size)
886 {
887 	phys_addr_t end = base + size - 1;
888 
889 	memblock_dbg("%s: [%pa-%pa] %pS\n", __func__,
890 		     &base, &end, (void *)_RET_IP_);
891 
892 	kmemleak_free_part_phys(base, size);
893 	return memblock_remove_range(&memblock.reserved, base, size);
894 }
895 
memblock_reserve(phys_addr_t base,phys_addr_t size)896 int __init_memblock memblock_reserve(phys_addr_t base, phys_addr_t size)
897 {
898 	phys_addr_t end = base + size - 1;
899 
900 	memblock_dbg("%s: [%pa-%pa] %pS\n", __func__,
901 		     &base, &end, (void *)_RET_IP_);
902 
903 	return memblock_add_range(&memblock.reserved, base, size, MAX_NUMNODES, 0);
904 }
905 
906 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
memblock_physmem_add(phys_addr_t base,phys_addr_t size)907 int __init_memblock memblock_physmem_add(phys_addr_t base, phys_addr_t size)
908 {
909 	phys_addr_t end = base + size - 1;
910 
911 	memblock_dbg("%s: [%pa-%pa] %pS\n", __func__,
912 		     &base, &end, (void *)_RET_IP_);
913 
914 	return memblock_add_range(&physmem, base, size, MAX_NUMNODES, 0);
915 }
916 #endif
917 
918 /**
919  * memblock_setclr_flag - set or clear flag for a memory region
920  * @base: base address of the region
921  * @size: size of the region
922  * @set: set or clear the flag
923  * @flag: the flag to update
924  *
925  * This function isolates region [@base, @base + @size), and sets/clears flag
926  *
927  * Return: 0 on success, -errno on failure.
928  */
memblock_setclr_flag(phys_addr_t base,phys_addr_t size,int set,int flag)929 static int __init_memblock memblock_setclr_flag(phys_addr_t base,
930 				phys_addr_t size, int set, int flag)
931 {
932 	struct memblock_type *type = &memblock.memory;
933 	int i, ret, start_rgn, end_rgn;
934 
935 	ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
936 	if (ret)
937 		return ret;
938 
939 	for (i = start_rgn; i < end_rgn; i++) {
940 		struct memblock_region *r = &type->regions[i];
941 
942 		if (set)
943 			r->flags |= flag;
944 		else
945 			r->flags &= ~flag;
946 	}
947 
948 	memblock_merge_regions(type, start_rgn, end_rgn);
949 	return 0;
950 }
951 
952 /**
953  * memblock_mark_hotplug - Mark hotpluggable memory with flag MEMBLOCK_HOTPLUG.
954  * @base: the base phys addr of the region
955  * @size: the size of the region
956  *
957  * Return: 0 on success, -errno on failure.
958  */
memblock_mark_hotplug(phys_addr_t base,phys_addr_t size)959 int __init_memblock memblock_mark_hotplug(phys_addr_t base, phys_addr_t size)
960 {
961 	return memblock_setclr_flag(base, size, 1, MEMBLOCK_HOTPLUG);
962 }
963 
964 /**
965  * memblock_clear_hotplug - Clear flag MEMBLOCK_HOTPLUG for a specified region.
966  * @base: the base phys addr of the region
967  * @size: the size of the region
968  *
969  * Return: 0 on success, -errno on failure.
970  */
memblock_clear_hotplug(phys_addr_t base,phys_addr_t size)971 int __init_memblock memblock_clear_hotplug(phys_addr_t base, phys_addr_t size)
972 {
973 	return memblock_setclr_flag(base, size, 0, MEMBLOCK_HOTPLUG);
974 }
975 
976 /**
977  * memblock_mark_mirror - Mark mirrored memory with flag MEMBLOCK_MIRROR.
978  * @base: the base phys addr of the region
979  * @size: the size of the region
980  *
981  * Return: 0 on success, -errno on failure.
982  */
memblock_mark_mirror(phys_addr_t base,phys_addr_t size)983 int __init_memblock memblock_mark_mirror(phys_addr_t base, phys_addr_t size)
984 {
985 	if (!mirrored_kernelcore)
986 		return 0;
987 
988 	system_has_some_mirror = true;
989 
990 	return memblock_setclr_flag(base, size, 1, MEMBLOCK_MIRROR);
991 }
992 
993 /**
994  * memblock_mark_nomap - Mark a memory region with flag MEMBLOCK_NOMAP.
995  * @base: the base phys addr of the region
996  * @size: the size of the region
997  *
998  * The memory regions marked with %MEMBLOCK_NOMAP will not be added to the
999  * direct mapping of the physical memory. These regions will still be
1000  * covered by the memory map. The struct page representing NOMAP memory
1001  * frames in the memory map will be PageReserved()
1002  *
1003  * Note: if the memory being marked %MEMBLOCK_NOMAP was allocated from
1004  * memblock, the caller must inform kmemleak to ignore that memory
1005  *
1006  * Return: 0 on success, -errno on failure.
1007  */
memblock_mark_nomap(phys_addr_t base,phys_addr_t size)1008 int __init_memblock memblock_mark_nomap(phys_addr_t base, phys_addr_t size)
1009 {
1010 	return memblock_setclr_flag(base, size, 1, MEMBLOCK_NOMAP);
1011 }
1012 
1013 /**
1014  * memblock_clear_nomap - Clear flag MEMBLOCK_NOMAP for a specified region.
1015  * @base: the base phys addr of the region
1016  * @size: the size of the region
1017  *
1018  * Return: 0 on success, -errno on failure.
1019  */
memblock_clear_nomap(phys_addr_t base,phys_addr_t size)1020 int __init_memblock memblock_clear_nomap(phys_addr_t base, phys_addr_t size)
1021 {
1022 	return memblock_setclr_flag(base, size, 0, MEMBLOCK_NOMAP);
1023 }
1024 
should_skip_region(struct memblock_type * type,struct memblock_region * m,int nid,int flags)1025 static bool should_skip_region(struct memblock_type *type,
1026 			       struct memblock_region *m,
1027 			       int nid, int flags)
1028 {
1029 	int m_nid = memblock_get_region_node(m);
1030 
1031 	/* we never skip regions when iterating memblock.reserved or physmem */
1032 	if (type != memblock_memory)
1033 		return false;
1034 
1035 	/* only memory regions are associated with nodes, check it */
1036 	if (nid != NUMA_NO_NODE && nid != m_nid)
1037 		return true;
1038 
1039 	/* skip hotpluggable memory regions if needed */
1040 	if (movable_node_is_enabled() && memblock_is_hotpluggable(m) &&
1041 	    !(flags & MEMBLOCK_HOTPLUG))
1042 		return true;
1043 
1044 	/* if we want mirror memory skip non-mirror memory regions */
1045 	if ((flags & MEMBLOCK_MIRROR) && !memblock_is_mirror(m))
1046 		return true;
1047 
1048 	/* skip nomap memory unless we were asked for it explicitly */
1049 	if (!(flags & MEMBLOCK_NOMAP) && memblock_is_nomap(m))
1050 		return true;
1051 
1052 	/* skip driver-managed memory unless we were asked for it explicitly */
1053 	if (!(flags & MEMBLOCK_DRIVER_MANAGED) && memblock_is_driver_managed(m))
1054 		return true;
1055 
1056 	return false;
1057 }
1058 
1059 /**
1060  * __next_mem_range - next function for for_each_free_mem_range() etc.
1061  * @idx: pointer to u64 loop variable
1062  * @nid: node selector, %NUMA_NO_NODE for all nodes
1063  * @flags: pick from blocks based on memory attributes
1064  * @type_a: pointer to memblock_type from where the range is taken
1065  * @type_b: pointer to memblock_type which excludes memory from being taken
1066  * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL
1067  * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL
1068  * @out_nid: ptr to int for nid of the range, can be %NULL
1069  *
1070  * Find the first area from *@idx which matches @nid, fill the out
1071  * parameters, and update *@idx for the next iteration.  The lower 32bit of
1072  * *@idx contains index into type_a and the upper 32bit indexes the
1073  * areas before each region in type_b.	For example, if type_b regions
1074  * look like the following,
1075  *
1076  *	0:[0-16), 1:[32-48), 2:[128-130)
1077  *
1078  * The upper 32bit indexes the following regions.
1079  *
1080  *	0:[0-0), 1:[16-32), 2:[48-128), 3:[130-MAX)
1081  *
1082  * As both region arrays are sorted, the function advances the two indices
1083  * in lockstep and returns each intersection.
1084  */
__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)1085 void __next_mem_range(u64 *idx, int nid, enum memblock_flags flags,
1086 		      struct memblock_type *type_a,
1087 		      struct memblock_type *type_b, phys_addr_t *out_start,
1088 		      phys_addr_t *out_end, int *out_nid)
1089 {
1090 	int idx_a = *idx & 0xffffffff;
1091 	int idx_b = *idx >> 32;
1092 
1093 	if (WARN_ONCE(nid == MAX_NUMNODES,
1094 	"Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
1095 		nid = NUMA_NO_NODE;
1096 
1097 	for (; idx_a < type_a->cnt; idx_a++) {
1098 		struct memblock_region *m = &type_a->regions[idx_a];
1099 
1100 		phys_addr_t m_start = m->base;
1101 		phys_addr_t m_end = m->base + m->size;
1102 		int	    m_nid = memblock_get_region_node(m);
1103 
1104 		if (should_skip_region(type_a, m, nid, flags))
1105 			continue;
1106 
1107 		if (!type_b) {
1108 			if (out_start)
1109 				*out_start = m_start;
1110 			if (out_end)
1111 				*out_end = m_end;
1112 			if (out_nid)
1113 				*out_nid = m_nid;
1114 			idx_a++;
1115 			*idx = (u32)idx_a | (u64)idx_b << 32;
1116 			return;
1117 		}
1118 
1119 		/* scan areas before each reservation */
1120 		for (; idx_b < type_b->cnt + 1; idx_b++) {
1121 			struct memblock_region *r;
1122 			phys_addr_t r_start;
1123 			phys_addr_t r_end;
1124 
1125 			r = &type_b->regions[idx_b];
1126 			r_start = idx_b ? r[-1].base + r[-1].size : 0;
1127 			r_end = idx_b < type_b->cnt ?
1128 				r->base : PHYS_ADDR_MAX;
1129 
1130 			/*
1131 			 * if idx_b advanced past idx_a,
1132 			 * break out to advance idx_a
1133 			 */
1134 			if (r_start >= m_end)
1135 				break;
1136 			/* if the two regions intersect, we're done */
1137 			if (m_start < r_end) {
1138 				if (out_start)
1139 					*out_start =
1140 						max(m_start, r_start);
1141 				if (out_end)
1142 					*out_end = min(m_end, r_end);
1143 				if (out_nid)
1144 					*out_nid = m_nid;
1145 				/*
1146 				 * The region which ends first is
1147 				 * advanced for the next iteration.
1148 				 */
1149 				if (m_end <= r_end)
1150 					idx_a++;
1151 				else
1152 					idx_b++;
1153 				*idx = (u32)idx_a | (u64)idx_b << 32;
1154 				return;
1155 			}
1156 		}
1157 	}
1158 
1159 	/* signal end of iteration */
1160 	*idx = ULLONG_MAX;
1161 }
1162 
1163 /**
1164  * __next_mem_range_rev - generic next function for for_each_*_range_rev()
1165  *
1166  * @idx: pointer to u64 loop variable
1167  * @nid: node selector, %NUMA_NO_NODE for all nodes
1168  * @flags: pick from blocks based on memory attributes
1169  * @type_a: pointer to memblock_type from where the range is taken
1170  * @type_b: pointer to memblock_type which excludes memory from being taken
1171  * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL
1172  * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL
1173  * @out_nid: ptr to int for nid of the range, can be %NULL
1174  *
1175  * Finds the next range from type_a which is not marked as unsuitable
1176  * in type_b.
1177  *
1178  * Reverse of __next_mem_range().
1179  */
__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)1180 void __init_memblock __next_mem_range_rev(u64 *idx, int nid,
1181 					  enum memblock_flags flags,
1182 					  struct memblock_type *type_a,
1183 					  struct memblock_type *type_b,
1184 					  phys_addr_t *out_start,
1185 					  phys_addr_t *out_end, int *out_nid)
1186 {
1187 	int idx_a = *idx & 0xffffffff;
1188 	int idx_b = *idx >> 32;
1189 
1190 	if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
1191 		nid = NUMA_NO_NODE;
1192 
1193 	if (*idx == (u64)ULLONG_MAX) {
1194 		idx_a = type_a->cnt - 1;
1195 		if (type_b != NULL)
1196 			idx_b = type_b->cnt;
1197 		else
1198 			idx_b = 0;
1199 	}
1200 
1201 	for (; idx_a >= 0; idx_a--) {
1202 		struct memblock_region *m = &type_a->regions[idx_a];
1203 
1204 		phys_addr_t m_start = m->base;
1205 		phys_addr_t m_end = m->base + m->size;
1206 		int m_nid = memblock_get_region_node(m);
1207 
1208 		if (should_skip_region(type_a, m, nid, flags))
1209 			continue;
1210 
1211 		if (!type_b) {
1212 			if (out_start)
1213 				*out_start = m_start;
1214 			if (out_end)
1215 				*out_end = m_end;
1216 			if (out_nid)
1217 				*out_nid = m_nid;
1218 			idx_a--;
1219 			*idx = (u32)idx_a | (u64)idx_b << 32;
1220 			return;
1221 		}
1222 
1223 		/* scan areas before each reservation */
1224 		for (; idx_b >= 0; idx_b--) {
1225 			struct memblock_region *r;
1226 			phys_addr_t r_start;
1227 			phys_addr_t r_end;
1228 
1229 			r = &type_b->regions[idx_b];
1230 			r_start = idx_b ? r[-1].base + r[-1].size : 0;
1231 			r_end = idx_b < type_b->cnt ?
1232 				r->base : PHYS_ADDR_MAX;
1233 			/*
1234 			 * if idx_b advanced past idx_a,
1235 			 * break out to advance idx_a
1236 			 */
1237 
1238 			if (r_end <= m_start)
1239 				break;
1240 			/* if the two regions intersect, we're done */
1241 			if (m_end > r_start) {
1242 				if (out_start)
1243 					*out_start = max(m_start, r_start);
1244 				if (out_end)
1245 					*out_end = min(m_end, r_end);
1246 				if (out_nid)
1247 					*out_nid = m_nid;
1248 				if (m_start >= r_start)
1249 					idx_a--;
1250 				else
1251 					idx_b--;
1252 				*idx = (u32)idx_a | (u64)idx_b << 32;
1253 				return;
1254 			}
1255 		}
1256 	}
1257 	/* signal end of iteration */
1258 	*idx = ULLONG_MAX;
1259 }
1260 
1261 /*
1262  * Common iterator interface used to define for_each_mem_pfn_range().
1263  */
__next_mem_pfn_range(int * idx,int nid,unsigned long * out_start_pfn,unsigned long * out_end_pfn,int * out_nid)1264 void __init_memblock __next_mem_pfn_range(int *idx, int nid,
1265 				unsigned long *out_start_pfn,
1266 				unsigned long *out_end_pfn, int *out_nid)
1267 {
1268 	struct memblock_type *type = &memblock.memory;
1269 	struct memblock_region *r;
1270 	int r_nid;
1271 
1272 	while (++*idx < type->cnt) {
1273 		r = &type->regions[*idx];
1274 		r_nid = memblock_get_region_node(r);
1275 
1276 		if (PFN_UP(r->base) >= PFN_DOWN(r->base + r->size))
1277 			continue;
1278 		if (nid == MAX_NUMNODES || nid == r_nid)
1279 			break;
1280 	}
1281 	if (*idx >= type->cnt) {
1282 		*idx = -1;
1283 		return;
1284 	}
1285 
1286 	if (out_start_pfn)
1287 		*out_start_pfn = PFN_UP(r->base);
1288 	if (out_end_pfn)
1289 		*out_end_pfn = PFN_DOWN(r->base + r->size);
1290 	if (out_nid)
1291 		*out_nid = r_nid;
1292 }
1293 
1294 /**
1295  * memblock_set_node - set node ID on memblock regions
1296  * @base: base of area to set node ID for
1297  * @size: size of area to set node ID for
1298  * @type: memblock type to set node ID for
1299  * @nid: node ID to set
1300  *
1301  * Set the nid of memblock @type regions in [@base, @base + @size) to @nid.
1302  * Regions which cross the area boundaries are split as necessary.
1303  *
1304  * Return:
1305  * 0 on success, -errno on failure.
1306  */
memblock_set_node(phys_addr_t base,phys_addr_t size,struct memblock_type * type,int nid)1307 int __init_memblock memblock_set_node(phys_addr_t base, phys_addr_t size,
1308 				      struct memblock_type *type, int nid)
1309 {
1310 #ifdef CONFIG_NUMA
1311 	int start_rgn, end_rgn;
1312 	int i, ret;
1313 
1314 	ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
1315 	if (ret)
1316 		return ret;
1317 
1318 	for (i = start_rgn; i < end_rgn; i++)
1319 		memblock_set_region_node(&type->regions[i], nid);
1320 
1321 	memblock_merge_regions(type, start_rgn, end_rgn);
1322 #endif
1323 	return 0;
1324 }
1325 
1326 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1327 /**
1328  * __next_mem_pfn_range_in_zone - iterator for for_each_*_range_in_zone()
1329  *
1330  * @idx: pointer to u64 loop variable
1331  * @zone: zone in which all of the memory blocks reside
1332  * @out_spfn: ptr to ulong for start pfn of the range, can be %NULL
1333  * @out_epfn: ptr to ulong for end pfn of the range, can be %NULL
1334  *
1335  * This function is meant to be a zone/pfn specific wrapper for the
1336  * for_each_mem_range type iterators. Specifically they are used in the
1337  * deferred memory init routines and as such we were duplicating much of
1338  * this logic throughout the code. So instead of having it in multiple
1339  * locations it seemed like it would make more sense to centralize this to
1340  * one new iterator that does everything they need.
1341  */
1342 void __init_memblock
__next_mem_pfn_range_in_zone(u64 * idx,struct zone * zone,unsigned long * out_spfn,unsigned long * out_epfn)1343 __next_mem_pfn_range_in_zone(u64 *idx, struct zone *zone,
1344 			     unsigned long *out_spfn, unsigned long *out_epfn)
1345 {
1346 	int zone_nid = zone_to_nid(zone);
1347 	phys_addr_t spa, epa;
1348 
1349 	__next_mem_range(idx, zone_nid, MEMBLOCK_NONE,
1350 			 &memblock.memory, &memblock.reserved,
1351 			 &spa, &epa, NULL);
1352 
1353 	while (*idx != U64_MAX) {
1354 		unsigned long epfn = PFN_DOWN(epa);
1355 		unsigned long spfn = PFN_UP(spa);
1356 
1357 		/*
1358 		 * Verify the end is at least past the start of the zone and
1359 		 * that we have at least one PFN to initialize.
1360 		 */
1361 		if (zone->zone_start_pfn < epfn && spfn < epfn) {
1362 			/* if we went too far just stop searching */
1363 			if (zone_end_pfn(zone) <= spfn) {
1364 				*idx = U64_MAX;
1365 				break;
1366 			}
1367 
1368 			if (out_spfn)
1369 				*out_spfn = max(zone->zone_start_pfn, spfn);
1370 			if (out_epfn)
1371 				*out_epfn = min(zone_end_pfn(zone), epfn);
1372 
1373 			return;
1374 		}
1375 
1376 		__next_mem_range(idx, zone_nid, MEMBLOCK_NONE,
1377 				 &memblock.memory, &memblock.reserved,
1378 				 &spa, &epa, NULL);
1379 	}
1380 
1381 	/* signal end of iteration */
1382 	if (out_spfn)
1383 		*out_spfn = ULONG_MAX;
1384 	if (out_epfn)
1385 		*out_epfn = 0;
1386 }
1387 
1388 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
1389 
1390 /**
1391  * memblock_alloc_range_nid - allocate boot memory block
1392  * @size: size of memory block to be allocated in bytes
1393  * @align: alignment of the region and block's size
1394  * @start: the lower bound of the memory region to allocate (phys address)
1395  * @end: the upper bound of the memory region to allocate (phys address)
1396  * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1397  * @exact_nid: control the allocation fall back to other nodes
1398  *
1399  * The allocation is performed from memory region limited by
1400  * memblock.current_limit if @end == %MEMBLOCK_ALLOC_ACCESSIBLE.
1401  *
1402  * If the specified node can not hold the requested memory and @exact_nid
1403  * is false, the allocation falls back to any node in the system.
1404  *
1405  * For systems with memory mirroring, the allocation is attempted first
1406  * from the regions with mirroring enabled and then retried from any
1407  * memory region.
1408  *
1409  * In addition, function using kmemleak_alloc_phys for allocated boot
1410  * memory block, it is never reported as leaks.
1411  *
1412  * Return:
1413  * Physical address of allocated memory block on success, %0 on failure.
1414  */
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)1415 phys_addr_t __init memblock_alloc_range_nid(phys_addr_t size,
1416 					phys_addr_t align, phys_addr_t start,
1417 					phys_addr_t end, int nid,
1418 					bool exact_nid)
1419 {
1420 	enum memblock_flags flags = choose_memblock_flags();
1421 	phys_addr_t found;
1422 
1423 	if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
1424 		nid = NUMA_NO_NODE;
1425 
1426 	if (!align) {
1427 		/* Can't use WARNs this early in boot on powerpc */
1428 		dump_stack();
1429 		align = SMP_CACHE_BYTES;
1430 	}
1431 
1432 again:
1433 	found = memblock_find_in_range_node(size, align, start, end, nid,
1434 					    flags);
1435 	if (found && !memblock_reserve(found, size))
1436 		goto done;
1437 
1438 	if (nid != NUMA_NO_NODE && !exact_nid) {
1439 		found = memblock_find_in_range_node(size, align, start,
1440 						    end, NUMA_NO_NODE,
1441 						    flags);
1442 		if (found && !memblock_reserve(found, size))
1443 			goto done;
1444 	}
1445 
1446 	if (flags & MEMBLOCK_MIRROR) {
1447 		flags &= ~MEMBLOCK_MIRROR;
1448 		pr_warn_ratelimited("Could not allocate %pap bytes of mirrored memory\n",
1449 			&size);
1450 		goto again;
1451 	}
1452 
1453 	return 0;
1454 
1455 done:
1456 	/*
1457 	 * Skip kmemleak for those places like kasan_init() and
1458 	 * early_pgtable_alloc() due to high volume.
1459 	 */
1460 	if (end != MEMBLOCK_ALLOC_NOLEAKTRACE)
1461 		/*
1462 		 * Memblock allocated blocks are never reported as
1463 		 * leaks. This is because many of these blocks are
1464 		 * only referred via the physical address which is
1465 		 * not looked up by kmemleak.
1466 		 */
1467 		kmemleak_alloc_phys(found, size, 0);
1468 
1469 	/*
1470 	 * Some Virtual Machine platforms, such as Intel TDX or AMD SEV-SNP,
1471 	 * require memory to be accepted before it can be used by the
1472 	 * guest.
1473 	 *
1474 	 * Accept the memory of the allocated buffer.
1475 	 */
1476 	accept_memory(found, found + size);
1477 
1478 	return found;
1479 }
1480 
1481 /**
1482  * memblock_phys_alloc_range - allocate a memory block inside specified range
1483  * @size: size of memory block to be allocated in bytes
1484  * @align: alignment of the region and block's size
1485  * @start: the lower bound of the memory region to allocate (physical address)
1486  * @end: the upper bound of the memory region to allocate (physical address)
1487  *
1488  * Allocate @size bytes in the between @start and @end.
1489  *
1490  * Return: physical address of the allocated memory block on success,
1491  * %0 on failure.
1492  */
memblock_phys_alloc_range(phys_addr_t size,phys_addr_t align,phys_addr_t start,phys_addr_t end)1493 phys_addr_t __init memblock_phys_alloc_range(phys_addr_t size,
1494 					     phys_addr_t align,
1495 					     phys_addr_t start,
1496 					     phys_addr_t end)
1497 {
1498 	memblock_dbg("%s: %llu bytes align=0x%llx from=%pa max_addr=%pa %pS\n",
1499 		     __func__, (u64)size, (u64)align, &start, &end,
1500 		     (void *)_RET_IP_);
1501 	return memblock_alloc_range_nid(size, align, start, end, NUMA_NO_NODE,
1502 					false);
1503 }
1504 
1505 /**
1506  * memblock_phys_alloc_try_nid - allocate a memory block from specified NUMA node
1507  * @size: size of memory block to be allocated in bytes
1508  * @align: alignment of the region and block's size
1509  * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1510  *
1511  * Allocates memory block from the specified NUMA node. If the node
1512  * has no available memory, attempts to allocated from any node in the
1513  * system.
1514  *
1515  * Return: physical address of the allocated memory block on success,
1516  * %0 on failure.
1517  */
memblock_phys_alloc_try_nid(phys_addr_t size,phys_addr_t align,int nid)1518 phys_addr_t __init memblock_phys_alloc_try_nid(phys_addr_t size, phys_addr_t align, int nid)
1519 {
1520 	return memblock_alloc_range_nid(size, align, 0,
1521 					MEMBLOCK_ALLOC_ACCESSIBLE, nid, false);
1522 }
1523 
1524 /**
1525  * memblock_alloc_internal - allocate boot memory block
1526  * @size: size of memory block to be allocated in bytes
1527  * @align: alignment of the region and block's size
1528  * @min_addr: the lower bound of the memory region to allocate (phys address)
1529  * @max_addr: the upper bound of the memory region to allocate (phys address)
1530  * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1531  * @exact_nid: control the allocation fall back to other nodes
1532  *
1533  * Allocates memory block using memblock_alloc_range_nid() and
1534  * converts the returned physical address to virtual.
1535  *
1536  * The @min_addr limit is dropped if it can not be satisfied and the allocation
1537  * will fall back to memory below @min_addr. Other constraints, such
1538  * as node and mirrored memory will be handled again in
1539  * memblock_alloc_range_nid().
1540  *
1541  * Return:
1542  * Virtual address of allocated memory block on success, NULL on failure.
1543  */
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)1544 static void * __init memblock_alloc_internal(
1545 				phys_addr_t size, phys_addr_t align,
1546 				phys_addr_t min_addr, phys_addr_t max_addr,
1547 				int nid, bool exact_nid)
1548 {
1549 	phys_addr_t alloc;
1550 
1551 	/*
1552 	 * Detect any accidental use of these APIs after slab is ready, as at
1553 	 * this moment memblock may be deinitialized already and its
1554 	 * internal data may be destroyed (after execution of memblock_free_all)
1555 	 */
1556 	if (WARN_ON_ONCE(slab_is_available()))
1557 		return kzalloc_node(size, GFP_NOWAIT, nid);
1558 
1559 	if (max_addr > memblock.current_limit)
1560 		max_addr = memblock.current_limit;
1561 
1562 	alloc = memblock_alloc_range_nid(size, align, min_addr, max_addr, nid,
1563 					exact_nid);
1564 
1565 	/* retry allocation without lower limit */
1566 	if (!alloc && min_addr)
1567 		alloc = memblock_alloc_range_nid(size, align, 0, max_addr, nid,
1568 						exact_nid);
1569 
1570 	if (!alloc)
1571 		return NULL;
1572 
1573 	return phys_to_virt(alloc);
1574 }
1575 
1576 /**
1577  * memblock_alloc_exact_nid_raw - allocate boot memory block on the exact node
1578  * without zeroing memory
1579  * @size: size of memory block to be allocated in bytes
1580  * @align: alignment of the region and block's size
1581  * @min_addr: the lower bound of the memory region from where the allocation
1582  *	  is preferred (phys address)
1583  * @max_addr: the upper bound of the memory region from where the allocation
1584  *	      is preferred (phys address), or %MEMBLOCK_ALLOC_ACCESSIBLE to
1585  *	      allocate only from memory limited by memblock.current_limit value
1586  * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1587  *
1588  * Public function, provides additional debug information (including caller
1589  * info), if enabled. Does not zero allocated memory.
1590  *
1591  * Return:
1592  * Virtual address of allocated memory block on success, NULL on failure.
1593  */
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)1594 void * __init memblock_alloc_exact_nid_raw(
1595 			phys_addr_t size, phys_addr_t align,
1596 			phys_addr_t min_addr, phys_addr_t max_addr,
1597 			int nid)
1598 {
1599 	memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=%pa max_addr=%pa %pS\n",
1600 		     __func__, (u64)size, (u64)align, nid, &min_addr,
1601 		     &max_addr, (void *)_RET_IP_);
1602 
1603 	return memblock_alloc_internal(size, align, min_addr, max_addr, nid,
1604 				       true);
1605 }
1606 
1607 /**
1608  * memblock_alloc_try_nid_raw - allocate boot memory block without zeroing
1609  * memory and without panicking
1610  * @size: size of memory block to be allocated in bytes
1611  * @align: alignment of the region and block's size
1612  * @min_addr: the lower bound of the memory region from where the allocation
1613  *	  is preferred (phys address)
1614  * @max_addr: the upper bound of the memory region from where the allocation
1615  *	      is preferred (phys address), or %MEMBLOCK_ALLOC_ACCESSIBLE to
1616  *	      allocate only from memory limited by memblock.current_limit value
1617  * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1618  *
1619  * Public function, provides additional debug information (including caller
1620  * info), if enabled. Does not zero allocated memory, does not panic if request
1621  * cannot be satisfied.
1622  *
1623  * Return:
1624  * Virtual address of allocated memory block on success, NULL on failure.
1625  */
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)1626 void * __init memblock_alloc_try_nid_raw(
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 	memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=%pa max_addr=%pa %pS\n",
1632 		     __func__, (u64)size, (u64)align, nid, &min_addr,
1633 		     &max_addr, (void *)_RET_IP_);
1634 
1635 	return memblock_alloc_internal(size, align, min_addr, max_addr, nid,
1636 				       false);
1637 }
1638 
1639 /**
1640  * memblock_alloc_try_nid - allocate boot memory block
1641  * @size: size of memory block to be allocated in bytes
1642  * @align: alignment of the region and block's size
1643  * @min_addr: the lower bound of the memory region from where the allocation
1644  *	  is preferred (phys address)
1645  * @max_addr: the upper bound of the memory region from where the allocation
1646  *	      is preferred (phys address), or %MEMBLOCK_ALLOC_ACCESSIBLE to
1647  *	      allocate only from memory limited by memblock.current_limit value
1648  * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1649  *
1650  * Public function, provides additional debug information (including caller
1651  * info), if enabled. This function zeroes the allocated memory.
1652  *
1653  * Return:
1654  * Virtual address of allocated memory block on success, NULL on failure.
1655  */
memblock_alloc_try_nid(phys_addr_t size,phys_addr_t align,phys_addr_t min_addr,phys_addr_t max_addr,int nid)1656 void * __init memblock_alloc_try_nid(
1657 			phys_addr_t size, phys_addr_t align,
1658 			phys_addr_t min_addr, phys_addr_t max_addr,
1659 			int nid)
1660 {
1661 	void *ptr;
1662 
1663 	memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=%pa max_addr=%pa %pS\n",
1664 		     __func__, (u64)size, (u64)align, nid, &min_addr,
1665 		     &max_addr, (void *)_RET_IP_);
1666 	ptr = memblock_alloc_internal(size, align,
1667 					   min_addr, max_addr, nid, false);
1668 	if (ptr)
1669 		memset(ptr, 0, size);
1670 
1671 	return ptr;
1672 }
1673 
1674 /**
1675  * memblock_free_late - free pages directly to buddy allocator
1676  * @base: phys starting address of the  boot memory block
1677  * @size: size of the boot memory block in bytes
1678  *
1679  * This is only useful when the memblock allocator has already been torn
1680  * down, but we are still initializing the system.  Pages are released directly
1681  * to the buddy allocator.
1682  */
memblock_free_late(phys_addr_t base,phys_addr_t size)1683 void __init memblock_free_late(phys_addr_t base, phys_addr_t size)
1684 {
1685 	phys_addr_t cursor, end;
1686 
1687 	end = base + size - 1;
1688 	memblock_dbg("%s: [%pa-%pa] %pS\n",
1689 		     __func__, &base, &end, (void *)_RET_IP_);
1690 	kmemleak_free_part_phys(base, size);
1691 	cursor = PFN_UP(base);
1692 	end = PFN_DOWN(base + size);
1693 
1694 	memblock_memsize_mod_kernel_size(-1 * ((long)(end - cursor) << PAGE_SHIFT));
1695 
1696 	for (; cursor < end; cursor++) {
1697 		memblock_free_pages(pfn_to_page(cursor), cursor, 0);
1698 		totalram_pages_inc();
1699 	}
1700 }
1701 
1702 /*
1703  * Remaining API functions
1704  */
1705 
memblock_phys_mem_size(void)1706 phys_addr_t __init_memblock memblock_phys_mem_size(void)
1707 {
1708 	return memblock.memory.total_size;
1709 }
1710 
memblock_reserved_size(void)1711 phys_addr_t __init_memblock memblock_reserved_size(void)
1712 {
1713 	return memblock.reserved.total_size;
1714 }
1715 
1716 /* lowest address */
memblock_start_of_DRAM(void)1717 phys_addr_t __init_memblock memblock_start_of_DRAM(void)
1718 {
1719 	return memblock.memory.regions[0].base;
1720 }
1721 
memblock_end_of_DRAM(void)1722 phys_addr_t __init_memblock memblock_end_of_DRAM(void)
1723 {
1724 	int idx = memblock.memory.cnt - 1;
1725 
1726 	return (memblock.memory.regions[idx].base + memblock.memory.regions[idx].size);
1727 }
1728 EXPORT_SYMBOL_GPL(memblock_end_of_DRAM);
1729 
__find_max_addr(phys_addr_t limit)1730 static phys_addr_t __init_memblock __find_max_addr(phys_addr_t limit)
1731 {
1732 	phys_addr_t max_addr = PHYS_ADDR_MAX;
1733 	struct memblock_region *r;
1734 
1735 	/*
1736 	 * translate the memory @limit size into the max address within one of
1737 	 * the memory memblock regions, if the @limit exceeds the total size
1738 	 * of those regions, max_addr will keep original value PHYS_ADDR_MAX
1739 	 */
1740 	for_each_mem_region(r) {
1741 		if (limit <= r->size) {
1742 			max_addr = r->base + limit;
1743 			break;
1744 		}
1745 		limit -= r->size;
1746 	}
1747 
1748 	return max_addr;
1749 }
1750 
memblock_enforce_memory_limit(phys_addr_t limit)1751 void __init memblock_enforce_memory_limit(phys_addr_t limit)
1752 {
1753 	phys_addr_t max_addr;
1754 
1755 	if (!limit)
1756 		return;
1757 
1758 	max_addr = __find_max_addr(limit);
1759 
1760 	/* @limit exceeds the total size of the memory, do nothing */
1761 	if (max_addr == PHYS_ADDR_MAX)
1762 		return;
1763 
1764 	/* truncate both memory and reserved regions */
1765 	memblock_remove_range(&memblock.memory, max_addr,
1766 			      PHYS_ADDR_MAX);
1767 	memblock_remove_range(&memblock.reserved, max_addr,
1768 			      PHYS_ADDR_MAX);
1769 }
1770 
memblock_cap_memory_range(phys_addr_t base,phys_addr_t size)1771 void __init memblock_cap_memory_range(phys_addr_t base, phys_addr_t size)
1772 {
1773 	int start_rgn, end_rgn;
1774 	int i, ret;
1775 
1776 	if (!size)
1777 		return;
1778 
1779 	if (!memblock_memory->total_size) {
1780 		pr_warn("%s: No memory registered yet\n", __func__);
1781 		return;
1782 	}
1783 
1784 	ret = memblock_isolate_range(&memblock.memory, base, size,
1785 						&start_rgn, &end_rgn);
1786 	if (ret)
1787 		return;
1788 
1789 	/* remove all the MAP regions */
1790 	for (i = memblock.memory.cnt - 1; i >= end_rgn; i--)
1791 		if (!memblock_is_nomap(&memblock.memory.regions[i]))
1792 			memblock_remove_region(&memblock.memory, i);
1793 
1794 	for (i = start_rgn - 1; i >= 0; i--)
1795 		if (!memblock_is_nomap(&memblock.memory.regions[i]))
1796 			memblock_remove_region(&memblock.memory, i);
1797 
1798 	/* truncate the reserved regions */
1799 	memblock_remove_range(&memblock.reserved, 0, base);
1800 	memblock_remove_range(&memblock.reserved,
1801 			base + size, PHYS_ADDR_MAX);
1802 }
1803 
memblock_mem_limit_remove_map(phys_addr_t limit)1804 void __init memblock_mem_limit_remove_map(phys_addr_t limit)
1805 {
1806 	phys_addr_t max_addr;
1807 
1808 	if (!limit)
1809 		return;
1810 
1811 	max_addr = __find_max_addr(limit);
1812 
1813 	/* @limit exceeds the total size of the memory, do nothing */
1814 	if (max_addr == PHYS_ADDR_MAX)
1815 		return;
1816 
1817 	memblock_cap_memory_range(0, max_addr);
1818 }
1819 
memblock_search(struct memblock_type * type,phys_addr_t addr)1820 static int __init_memblock memblock_search(struct memblock_type *type, phys_addr_t addr)
1821 {
1822 	unsigned int left = 0, right = type->cnt;
1823 
1824 	do {
1825 		unsigned int mid = (right + left) / 2;
1826 
1827 		if (addr < type->regions[mid].base)
1828 			right = mid;
1829 		else if (addr >= (type->regions[mid].base +
1830 				  type->regions[mid].size))
1831 			left = mid + 1;
1832 		else
1833 			return mid;
1834 	} while (left < right);
1835 	return -1;
1836 }
1837 
memblock_is_reserved(phys_addr_t addr)1838 bool __init_memblock memblock_is_reserved(phys_addr_t addr)
1839 {
1840 	return memblock_search(&memblock.reserved, addr) != -1;
1841 }
1842 
memblock_is_memory(phys_addr_t addr)1843 bool __init_memblock memblock_is_memory(phys_addr_t addr)
1844 {
1845 	return memblock_search(&memblock.memory, addr) != -1;
1846 }
1847 
memblock_is_map_memory(phys_addr_t addr)1848 bool __init_memblock memblock_is_map_memory(phys_addr_t addr)
1849 {
1850 	int i = memblock_search(&memblock.memory, addr);
1851 
1852 	if (i == -1)
1853 		return false;
1854 	return !memblock_is_nomap(&memblock.memory.regions[i]);
1855 }
1856 
memblock_search_pfn_nid(unsigned long pfn,unsigned long * start_pfn,unsigned long * end_pfn)1857 int __init_memblock memblock_search_pfn_nid(unsigned long pfn,
1858 			 unsigned long *start_pfn, unsigned long *end_pfn)
1859 {
1860 	struct memblock_type *type = &memblock.memory;
1861 	int mid = memblock_search(type, PFN_PHYS(pfn));
1862 
1863 	if (mid == -1)
1864 		return -1;
1865 
1866 	*start_pfn = PFN_DOWN(type->regions[mid].base);
1867 	*end_pfn = PFN_DOWN(type->regions[mid].base + type->regions[mid].size);
1868 
1869 	return memblock_get_region_node(&type->regions[mid]);
1870 }
1871 
1872 /**
1873  * memblock_is_region_memory - check if a region is a subset of memory
1874  * @base: base of region to check
1875  * @size: size of region to check
1876  *
1877  * Check if the region [@base, @base + @size) is a subset of a memory block.
1878  *
1879  * Return:
1880  * 0 if false, non-zero if true
1881  */
memblock_is_region_memory(phys_addr_t base,phys_addr_t size)1882 bool __init_memblock memblock_is_region_memory(phys_addr_t base, phys_addr_t size)
1883 {
1884 	int idx = memblock_search(&memblock.memory, base);
1885 	phys_addr_t end = base + memblock_cap_size(base, &size);
1886 
1887 	if (idx == -1)
1888 		return false;
1889 	return (memblock.memory.regions[idx].base +
1890 		 memblock.memory.regions[idx].size) >= end;
1891 }
1892 
1893 /**
1894  * memblock_is_region_reserved - check if a region intersects reserved memory
1895  * @base: base of region to check
1896  * @size: size of region to check
1897  *
1898  * Check if the region [@base, @base + @size) intersects a reserved
1899  * memory block.
1900  *
1901  * Return:
1902  * True if they intersect, false if not.
1903  */
memblock_is_region_reserved(phys_addr_t base,phys_addr_t size)1904 bool __init_memblock memblock_is_region_reserved(phys_addr_t base, phys_addr_t size)
1905 {
1906 	return memblock_overlaps_region(&memblock.reserved, base, size);
1907 }
1908 
memblock_trim_memory(phys_addr_t align)1909 void __init_memblock memblock_trim_memory(phys_addr_t align)
1910 {
1911 	phys_addr_t start, end, orig_start, orig_end;
1912 	struct memblock_region *r;
1913 
1914 	for_each_mem_region(r) {
1915 		orig_start = r->base;
1916 		orig_end = r->base + r->size;
1917 		start = round_up(orig_start, align);
1918 		end = round_down(orig_end, align);
1919 
1920 		if (start == orig_start && end == orig_end)
1921 			continue;
1922 
1923 		if (start < end) {
1924 			r->base = start;
1925 			r->size = end - start;
1926 		} else {
1927 			memblock_remove_region(&memblock.memory,
1928 					       r - memblock.memory.regions);
1929 			r--;
1930 		}
1931 	}
1932 }
1933 
memblock_set_current_limit(phys_addr_t limit)1934 void __init_memblock memblock_set_current_limit(phys_addr_t limit)
1935 {
1936 	memblock.current_limit = limit;
1937 }
1938 
memblock_get_current_limit(void)1939 phys_addr_t __init_memblock memblock_get_current_limit(void)
1940 {
1941 	return memblock.current_limit;
1942 }
1943 
memblock_dump(struct memblock_type * type)1944 static void __init_memblock memblock_dump(struct memblock_type *type)
1945 {
1946 	phys_addr_t base, end, size;
1947 	enum memblock_flags flags;
1948 	int idx;
1949 	struct memblock_region *rgn;
1950 
1951 	pr_info(" %s.cnt  = 0x%lx\n", type->name, type->cnt);
1952 
1953 	for_each_memblock_type(idx, type, rgn) {
1954 		char nid_buf[32] = "";
1955 
1956 		base = rgn->base;
1957 		size = rgn->size;
1958 		end = base + size - 1;
1959 		flags = rgn->flags;
1960 #ifdef CONFIG_NUMA
1961 		if (memblock_get_region_node(rgn) != MAX_NUMNODES)
1962 			snprintf(nid_buf, sizeof(nid_buf), " on node %d",
1963 				 memblock_get_region_node(rgn));
1964 #endif
1965 		pr_info(" %s[%#x]\t[%pa-%pa], %pa bytes%s flags: %#x\n",
1966 			type->name, idx, &base, &end, &size, nid_buf, flags);
1967 	}
1968 }
1969 
__memblock_dump_all(void)1970 static void __init_memblock __memblock_dump_all(void)
1971 {
1972 	pr_info("MEMBLOCK configuration:\n");
1973 	pr_info(" memory size = %pa reserved size = %pa\n",
1974 		&memblock.memory.total_size,
1975 		&memblock.reserved.total_size);
1976 
1977 	memblock_dump(&memblock.memory);
1978 	memblock_dump(&memblock.reserved);
1979 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
1980 	memblock_dump(&physmem);
1981 #endif
1982 }
1983 
memblock_dump_all(void)1984 void __init_memblock memblock_dump_all(void)
1985 {
1986 	if (memblock_debug)
1987 		__memblock_dump_all();
1988 }
1989 
memblock_allow_resize(void)1990 void __init memblock_allow_resize(void)
1991 {
1992 	memblock_can_resize = 1;
1993 }
1994 
early_memblock(char * p)1995 static int __init early_memblock(char *p)
1996 {
1997 	if (p && strstr(p, "debug"))
1998 		memblock_debug = 1;
1999 	return 0;
2000 }
2001 early_param("memblock", early_memblock);
2002 
2003 #define NAME_SIZE	100
2004 struct memsize_rgn_struct {
2005 	phys_addr_t	base;
2006 	long		size;
2007 	bool		nomap;			/*  1/32 byte */
2008 	bool		reusable;		/*  1/32 byte */
2009 	char		name[NAME_SIZE];	/* 30/32 byte */
2010 };
2011 
2012 #define MAX_MEMBLOCK_MEMSIZE	100
2013 
2014 static struct memsize_rgn_struct memsize_rgn[MAX_MEMBLOCK_MEMSIZE] __initdata_memblock;
2015 static int memsize_rgn_count __initdata_memblock;
2016 static long memsize_memmap;
2017 static unsigned long memsize_code __initdata_memblock;
2018 static unsigned long memsize_data __initdata_memblock;
2019 static unsigned long memsize_ro __initdata_memblock;
2020 static unsigned long memsize_bss __initdata_memblock;
2021 static long memsize_reusable_size;
2022 
2023 enum memblock_memsize_state {
2024 	MEMBLOCK_MEMSIZE_NONE = 0,
2025 	MEMBLOCK_MEMSIZE_DEBUGFS,
2026 	MEMBLOCK_MEMSIZE_PROCFS,
2027 };
2028 
2029 static enum memblock_memsize_state memsize_state __initdata_memblock = MEMBLOCK_MEMSIZE_NONE;
2030 
early_memblock_memsize(char * str)2031 static int __init early_memblock_memsize(char *str)
2032 {
2033 	if (!str)
2034 		return -EINVAL;
2035 	if (strcmp(str, "none") == 0)
2036 		memsize_state = MEMBLOCK_MEMSIZE_NONE;
2037 	else if (strcmp(str, "debugfs") == 0)
2038 		memsize_state = MEMBLOCK_MEMSIZE_DEBUGFS;
2039 	else if (strcmp(str, "procfs") == 0)
2040 		memsize_state = MEMBLOCK_MEMSIZE_PROCFS;
2041 	else
2042 		return -EINVAL;
2043 	return 0;
2044 }
2045 early_param("memblock_memsize", early_memblock_memsize);
2046 
memblock_memsize_enable_tracking(void)2047 void __init memblock_memsize_enable_tracking(void)
2048 {
2049 	memblock_memsize_tracking = true;
2050 }
2051 
memblock_memsize_disable_tracking(void)2052 void __init memblock_memsize_disable_tracking(void)
2053 {
2054 	memblock_memsize_tracking = false;
2055 }
2056 
memblock_memsize_mod_memmap_size(long size)2057 void __init memblock_memsize_mod_memmap_size(long size)
2058 {
2059 	memsize_memmap += size;
2060 }
2061 
memblock_memsize_mod_kernel_size(long size)2062 void memblock_memsize_mod_kernel_size(long size)
2063 {
2064 	memsize_kinit += size;
2065 }
2066 
memblock_memsize_kernel_code_data(unsigned long code,unsigned long data,unsigned long ro,unsigned long bss)2067 void __init memblock_memsize_kernel_code_data(unsigned long code, unsigned long data,
2068 		unsigned long ro, unsigned long bss)
2069 {
2070 	memsize_code = code;
2071 	memsize_data = data;
2072 	memsize_ro = ro;
2073 	memsize_bss = bss;
2074 }
2075 
memsize_get_valid_name(char * valid_name,const char * name)2076 static void __init_memblock memsize_get_valid_name(char *valid_name, const char *name)
2077 {
2078 	char *head, *tail, *found;
2079 	int valid_size;
2080 
2081 	head = (char *)name;
2082 	tail = head + strlen(name);
2083 
2084 	/* get tail position after valid char */
2085 	found = strchr(name, '@');
2086 	if (found)
2087 		tail = found;
2088 
2089 	valid_size = tail - head + 1;
2090 	if (valid_size > NAME_SIZE)
2091 		valid_size = NAME_SIZE;
2092 	strscpy(valid_name, head, valid_size);
2093 }
2094 
memblock_memsize_mod_reusable_size(long size)2095 void memblock_memsize_mod_reusable_size(long size)
2096 {
2097 	memsize_reusable_size += size;
2098 }
2099 
memsize_get_new_rgn(void)2100 static inline struct memsize_rgn_struct * __init_memblock memsize_get_new_rgn(void)
2101 {
2102 	if (memsize_rgn_count == ARRAY_SIZE(memsize_rgn)) {
2103 		pr_err("not enough space on memsize_rgn\n");
2104 		return NULL;
2105 	}
2106 	return &memsize_rgn[memsize_rgn_count++];
2107 }
2108 
memsize_update_nomap_region(const char * name,phys_addr_t base,phys_addr_t size,bool nomap)2109 static bool __init_memblock memsize_update_nomap_region(const char *name, phys_addr_t base,
2110 					phys_addr_t size, bool nomap)
2111 {
2112 	int i;
2113 	struct memsize_rgn_struct *rmem_rgn, *new_rgn;
2114 
2115 	if (!name)
2116 		return false;
2117 
2118 	for (i = 0; i < memsize_rgn_count; i++)	{
2119 		rmem_rgn = &memsize_rgn[i];
2120 
2121 		/* skip either !nomap, !unknown, !overlap */
2122 		if (!rmem_rgn->nomap)
2123 			continue;
2124 		if (strcmp(rmem_rgn->name, "unknown"))
2125 			continue;
2126 		if (base + size <= rmem_rgn->base)
2127 			continue;
2128 		if (base >= rmem_rgn->base + rmem_rgn->size)
2129 			continue;
2130 
2131 		/* exactly same */
2132 		if (base == rmem_rgn->base && size == rmem_rgn->size) {
2133 			memsize_get_valid_name(rmem_rgn->name, name);
2134 			return true;
2135 		}
2136 
2137 		/* bigger */
2138 		if (base <= rmem_rgn->base &&
2139 				base + size >= rmem_rgn->base + rmem_rgn->size) {
2140 			memsize_get_valid_name(rmem_rgn->name, name);
2141 			rmem_rgn->base = base;
2142 			rmem_rgn->size = size;
2143 			return true;
2144 		}
2145 
2146 		/* intersect */
2147 		if (base < rmem_rgn->base ||
2148 				base + size > rmem_rgn->base + rmem_rgn->size) {
2149 			new_rgn = memsize_get_new_rgn();
2150 			if (!new_rgn)
2151 				return true;
2152 			new_rgn->base = base;
2153 			new_rgn->size = size;
2154 			new_rgn->nomap = nomap;
2155 			new_rgn->reusable = false;
2156 			memsize_get_valid_name(new_rgn->name, name);
2157 
2158 			if (base < rmem_rgn->base) {
2159 				rmem_rgn->size -= base + size - rmem_rgn->base;
2160 				rmem_rgn->base = base + size;
2161 			} else {
2162 				rmem_rgn->size -= rmem_rgn->base
2163 							+ rmem_rgn->size - base;
2164 			}
2165 			return true;
2166 		}
2167 
2168 		/* smaller */
2169 		new_rgn = memsize_get_new_rgn();
2170 		if (!new_rgn)
2171 			return true;
2172 		new_rgn->base = base;
2173 		new_rgn->size = size;
2174 		new_rgn->nomap = nomap;
2175 		new_rgn->reusable = false;
2176 		memsize_get_valid_name(new_rgn->name, name);
2177 
2178 		if (base == rmem_rgn->base && size < rmem_rgn->size) {
2179 			rmem_rgn->base = base + size;
2180 			rmem_rgn->size -= size;
2181 		} else if (base + size == rmem_rgn->base + rmem_rgn->size) {
2182 			rmem_rgn->size -= size;
2183 		} else {
2184 			new_rgn = memsize_get_new_rgn();
2185 			if (!new_rgn)
2186 				return true;
2187 			new_rgn->base = base + size;
2188 			new_rgn->size = (rmem_rgn->base + rmem_rgn->size)
2189 					- (base + size);
2190 			new_rgn->nomap = nomap;
2191 			new_rgn->reusable = false;
2192 			strscpy(new_rgn->name, "unknown", sizeof(new_rgn->name));
2193 			rmem_rgn->size = base - rmem_rgn->base;
2194 		}
2195 		return true;
2196 	}
2197 
2198 	return false;
2199 }
2200 
memblock_memsize_record(const char * name,phys_addr_t base,phys_addr_t size,bool nomap,bool reusable)2201 void __init_memblock memblock_memsize_record(const char *name, phys_addr_t base,
2202 			     phys_addr_t size, bool nomap, bool reusable)
2203 {
2204 	struct memsize_rgn_struct *rgn;
2205 	phys_addr_t end;
2206 
2207 	if (name && memsize_state == MEMBLOCK_MEMSIZE_NONE)
2208 		return;
2209 
2210 	if (memsize_rgn_count == MAX_MEMBLOCK_MEMSIZE) {
2211 		pr_err("not enough space on memsize_rgn\n");
2212 		return;
2213 	}
2214 
2215 	if (memsize_update_nomap_region(name, base, size, nomap))
2216 		return;
2217 
2218 	rgn = memsize_get_new_rgn();
2219 	if (!rgn)
2220 		return;
2221 
2222 	rgn->base = base;
2223 	rgn->size = size;
2224 	rgn->nomap = nomap;
2225 	rgn->reusable = reusable;
2226 
2227 	if (!name)
2228 		strscpy(rgn->name, "unknown", sizeof(rgn->name));
2229 	else
2230 		memsize_get_valid_name(rgn->name, name);
2231 	end = base + size - 1;
2232 	memblock_dbg("%s %pa..%pa nomap:%d reusable:%d\n",
2233 		     __func__, &base, &end, nomap, reusable);
2234 }
2235 
memblock_memsize_detect_hole(void)2236 void __init memblock_memsize_detect_hole(void)
2237 {
2238 	phys_addr_t base, end;
2239 	phys_addr_t prev_end, hole_sz;
2240 	int idx;
2241 	struct memblock_region *rgn;
2242 	int memblock_cnt = (int)memblock.memory.cnt;
2243 
2244 	/* assume that the hole size is less than 1 GB */
2245 	for_each_memblock_type(idx, (&memblock.memory), rgn) {
2246 		prev_end = (idx == 0) ? round_down(rgn->base, SZ_1G) : end;
2247 		base = rgn->base;
2248 		end = rgn->base + rgn->size;
2249 
2250 		/* only for the last region, check a hole after the region */
2251 		if (idx + 1 == memblock_cnt) {
2252 			hole_sz = round_up(end, SZ_1G) - end;
2253 			if (hole_sz)
2254 				memblock_memsize_record(NULL, end, hole_sz,
2255 							true, false);
2256 		}
2257 
2258 		/* for each region, check a hole prior to the region */
2259 		hole_sz = base - prev_end;
2260 		if (!hole_sz)
2261 			continue;
2262 		if (hole_sz < SZ_1G) {
2263 			memblock_memsize_record(NULL, prev_end, hole_sz, true,
2264 						false);
2265 		} else {
2266 			phys_addr_t hole_sz1, hole_sz2;
2267 
2268 			hole_sz1 = round_up(prev_end, SZ_1G) - prev_end;
2269 			if (hole_sz1)
2270 				memblock_memsize_record(NULL, prev_end,
2271 							hole_sz1, true, false);
2272 			hole_sz2 = base % SZ_1G;
2273 			if (hole_sz2)
2274 				memblock_memsize_record(NULL, base - hole_sz2,
2275 							hole_sz2, true, false);
2276 		}
2277 	}
2278 }
2279 
free_memmap(unsigned long start_pfn,unsigned long end_pfn)2280 static void __init free_memmap(unsigned long start_pfn, unsigned long end_pfn)
2281 {
2282 	struct page *start_pg, *end_pg;
2283 	phys_addr_t pg, pgend;
2284 
2285 	/*
2286 	 * Convert start_pfn/end_pfn to a struct page pointer.
2287 	 */
2288 	start_pg = pfn_to_page(start_pfn - 1) + 1;
2289 	end_pg = pfn_to_page(end_pfn - 1) + 1;
2290 
2291 	/*
2292 	 * Convert to physical addresses, and round start upwards and end
2293 	 * downwards.
2294 	 */
2295 	pg = PAGE_ALIGN(__pa(start_pg));
2296 	pgend = __pa(end_pg) & PAGE_MASK;
2297 
2298 	/*
2299 	 * If there are free pages between these, free the section of the
2300 	 * memmap array.
2301 	 */
2302 	if (pg < pgend)
2303 		memblock_phys_free(pg, pgend - pg);
2304 }
2305 
2306 /*
2307  * The mem_map array can get very big.  Free the unused area of the memory map.
2308  */
free_unused_memmap(void)2309 static void __init free_unused_memmap(void)
2310 {
2311 	unsigned long start, end, prev_end = 0;
2312 	int i;
2313 
2314 	if (!IS_ENABLED(CONFIG_HAVE_ARCH_PFN_VALID) ||
2315 	    IS_ENABLED(CONFIG_SPARSEMEM_VMEMMAP))
2316 		return;
2317 
2318 	/*
2319 	 * This relies on each bank being in address order.
2320 	 * The banks are sorted previously in bootmem_init().
2321 	 */
2322 	for_each_mem_pfn_range(i, MAX_NUMNODES, &start, &end, NULL) {
2323 #ifdef CONFIG_SPARSEMEM
2324 		/*
2325 		 * Take care not to free memmap entries that don't exist
2326 		 * due to SPARSEMEM sections which aren't present.
2327 		 */
2328 		start = min(start, ALIGN(prev_end, PAGES_PER_SECTION));
2329 #endif
2330 		/*
2331 		 * Align down here since many operations in VM subsystem
2332 		 * presume that there are no holes in the memory map inside
2333 		 * a pageblock
2334 		 */
2335 		start = pageblock_start_pfn(start);
2336 
2337 		/*
2338 		 * If we had a previous bank, and there is a space
2339 		 * between the current bank and the previous, free it.
2340 		 */
2341 		if (prev_end && prev_end < start)
2342 			free_memmap(prev_end, start);
2343 
2344 		/*
2345 		 * Align up here since many operations in VM subsystem
2346 		 * presume that there are no holes in the memory map inside
2347 		 * a pageblock
2348 		 */
2349 		prev_end = pageblock_align(end);
2350 	}
2351 
2352 #ifdef CONFIG_SPARSEMEM
2353 	if (!IS_ALIGNED(prev_end, PAGES_PER_SECTION)) {
2354 		prev_end = pageblock_align(end);
2355 		free_memmap(prev_end, ALIGN(prev_end, PAGES_PER_SECTION));
2356 	}
2357 #endif
2358 }
2359 
__free_pages_memory(unsigned long start,unsigned long end)2360 static void __init __free_pages_memory(unsigned long start, unsigned long end)
2361 {
2362 	int order;
2363 
2364 	while (start < end) {
2365 		/*
2366 		 * Free the pages in the largest chunks alignment allows.
2367 		 *
2368 		 * __ffs() behaviour is undefined for 0. start == 0 is
2369 		 * MAX_ORDER-aligned, set order to MAX_ORDER for the case.
2370 		 */
2371 		if (start)
2372 			order = min_t(int, MAX_ORDER, __ffs(start));
2373 		else
2374 			order = MAX_ORDER;
2375 
2376 		while (start + (1UL << order) > end)
2377 			order--;
2378 
2379 		memblock_free_pages(pfn_to_page(start), start, order);
2380 
2381 		start += (1UL << order);
2382 	}
2383 }
2384 
__free_memory_core(phys_addr_t start,phys_addr_t end)2385 static unsigned long __init __free_memory_core(phys_addr_t start,
2386 				 phys_addr_t end)
2387 {
2388 	unsigned long start_pfn = PFN_UP(start);
2389 	unsigned long end_pfn = min_t(unsigned long,
2390 				      PFN_DOWN(end), max_low_pfn);
2391 
2392 	unsigned long start_align_up = PFN_ALIGN(start);
2393 	unsigned long end_align_down = PFN_PHYS(end_pfn);
2394 
2395 	if (start_pfn >= end_pfn) {
2396 		memblock_memsize_mod_kernel_size(end - start);
2397 	} else {
2398 		if (start_align_up > start)
2399 			memblock_memsize_mod_kernel_size(start_align_up - start);
2400 		if (end_pfn != max_low_pfn && end_align_down < end)
2401 			memblock_memsize_mod_kernel_size(end - end_align_down);
2402 	}
2403 	if (start_pfn >= end_pfn)
2404 		return 0;
2405 
2406 	__free_pages_memory(start_pfn, end_pfn);
2407 
2408 	return end_pfn - start_pfn;
2409 }
2410 
memmap_init_reserved_pages(void)2411 static void __init memmap_init_reserved_pages(void)
2412 {
2413 	struct memblock_region *region;
2414 	phys_addr_t start, end;
2415 	int nid;
2416 
2417 	/*
2418 	 * set nid on all reserved pages and also treat struct
2419 	 * pages for the NOMAP regions as PageReserved
2420 	 */
2421 	for_each_mem_region(region) {
2422 		nid = memblock_get_region_node(region);
2423 		start = region->base;
2424 		end = start + region->size;
2425 
2426 		if (memblock_is_nomap(region))
2427 			reserve_bootmem_region(start, end, nid);
2428 
2429 		memblock_set_node(start, end, &memblock.reserved, nid);
2430 	}
2431 
2432 	/* initialize struct pages for the reserved regions */
2433 	for_each_reserved_mem_region(region) {
2434 		nid = memblock_get_region_node(region);
2435 		start = region->base;
2436 		end = start + region->size;
2437 
2438 		if (nid == NUMA_NO_NODE || nid >= MAX_NUMNODES)
2439 			nid = early_pfn_to_nid(PFN_DOWN(start));
2440 
2441 		reserve_bootmem_region(start, end, nid);
2442 	}
2443 }
2444 
free_low_memory_core_early(void)2445 static unsigned long __init free_low_memory_core_early(void)
2446 {
2447 	unsigned long count = 0;
2448 	phys_addr_t start, end;
2449 	u64 i;
2450 
2451 	memblock_clear_hotplug(0, -1);
2452 
2453 	memmap_init_reserved_pages();
2454 
2455 	/*
2456 	 * We need to use NUMA_NO_NODE instead of NODE_DATA(0)->node_id
2457 	 *  because in some case like Node0 doesn't have RAM installed
2458 	 *  low ram will be on Node1
2459 	 */
2460 	for_each_free_mem_range(i, NUMA_NO_NODE, MEMBLOCK_NONE, &start, &end,
2461 				NULL)
2462 		count += __free_memory_core(start, end);
2463 
2464 	return count;
2465 }
2466 
2467 static int reset_managed_pages_done __initdata;
2468 
reset_node_managed_pages(pg_data_t * pgdat)2469 static void __init reset_node_managed_pages(pg_data_t *pgdat)
2470 {
2471 	struct zone *z;
2472 
2473 	for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++)
2474 		atomic_long_set(&z->managed_pages, 0);
2475 }
2476 
reset_all_zones_managed_pages(void)2477 void __init reset_all_zones_managed_pages(void)
2478 {
2479 	struct pglist_data *pgdat;
2480 
2481 	if (reset_managed_pages_done)
2482 		return;
2483 
2484 	for_each_online_pgdat(pgdat)
2485 		reset_node_managed_pages(pgdat);
2486 
2487 	reset_managed_pages_done = 1;
2488 }
2489 
2490 /**
2491  * memblock_free_all - release free pages to the buddy allocator
2492  */
memblock_free_all(void)2493 void __init memblock_free_all(void)
2494 {
2495 	unsigned long pages;
2496 
2497 	free_unused_memmap();
2498 	reset_all_zones_managed_pages();
2499 
2500 	pages = free_low_memory_core_early();
2501 	totalram_pages_add(pages);
2502 
2503 	memblock_memsize_disable_tracking();
2504 }
2505 
2506 #if defined(CONFIG_DEBUG_FS) && defined(CONFIG_ARCH_KEEP_MEMBLOCK)
2507 static const char * const flagname[] = {
2508 	[ilog2(MEMBLOCK_HOTPLUG)] = "HOTPLUG",
2509 	[ilog2(MEMBLOCK_MIRROR)] = "MIRROR",
2510 	[ilog2(MEMBLOCK_NOMAP)] = "NOMAP",
2511 	[ilog2(MEMBLOCK_DRIVER_MANAGED)] = "DRV_MNG",
2512 };
2513 
memblock_debug_show(struct seq_file * m,void * private)2514 static int memblock_debug_show(struct seq_file *m, void *private)
2515 {
2516 	struct memblock_type *type = m->private;
2517 	struct memblock_region *reg;
2518 	int i, j, nid;
2519 	unsigned int count = ARRAY_SIZE(flagname);
2520 	phys_addr_t end;
2521 
2522 	for (i = 0; i < type->cnt; i++) {
2523 		reg = &type->regions[i];
2524 		end = reg->base + reg->size - 1;
2525 		nid = memblock_get_region_node(reg);
2526 
2527 		seq_printf(m, "%4d: ", i);
2528 		seq_printf(m, "%pa..%pa ", &reg->base, &end);
2529 		if (nid != MAX_NUMNODES)
2530 			seq_printf(m, "%4d ", nid);
2531 		else
2532 			seq_printf(m, "%4c ", 'x');
2533 		if (reg->flags) {
2534 			for (j = 0; j < count; j++) {
2535 				if (reg->flags & (1U << j)) {
2536 					seq_printf(m, "%s\n", flagname[j]);
2537 					break;
2538 				}
2539 			}
2540 			if (j == count)
2541 				seq_printf(m, "%s\n", "UNKNOWN");
2542 		} else {
2543 			seq_printf(m, "%s\n", "NONE");
2544 		}
2545 	}
2546 	return 0;
2547 }
2548 DEFINE_SHOW_ATTRIBUTE(memblock_debug);
2549 
2550 /* assume that freeing region is NOT bigger than the previous region */
memblock_memsize_free(phys_addr_t free_base,phys_addr_t free_size)2551 static void memblock_memsize_free(phys_addr_t free_base,
2552 						  phys_addr_t free_size)
2553 {
2554 	int i;
2555 	struct memsize_rgn_struct *rgn;
2556 	phys_addr_t free_end, end;
2557 
2558 	free_end = free_base + free_size - 1;
2559 	memblock_dbg("%s %pa..%pa\n",
2560 		     __func__, &free_base, &free_end);
2561 
2562 	for (i = 0; i < memsize_rgn_count; i++) {
2563 		rgn = &memsize_rgn[i];
2564 
2565 		end = rgn->base + rgn->size;
2566 		if (free_base < rgn->base ||
2567 		    free_base >= end)
2568 			continue;
2569 
2570 		free_end = free_base + free_size;
2571 		if (free_base == rgn->base) {
2572 			rgn->size -= free_size;
2573 			if (rgn->size != 0)
2574 				rgn->base += free_size;
2575 		} else if (free_end == end) {
2576 			rgn->size -= free_size;
2577 		} else {
2578 			memblock_memsize_record(rgn->name, free_end,
2579 				end - free_end, rgn->nomap, rgn->reusable);
2580 			rgn->size = free_base - rgn->base;
2581 		}
2582 	}
2583 }
2584 
memsize_rgn_cmp(const void * a,const void * b)2585 static int memsize_rgn_cmp(const void *a, const void *b)
2586 {
2587 	const struct memsize_rgn_struct *ra = a, *rb = b;
2588 
2589 	if (ra->base > rb->base)
2590 		return -1;
2591 
2592 	if (ra->base < rb->base)
2593 		return 1;
2594 
2595 	return 0;
2596 }
2597 
2598 /* assume that freed size is always 64 KB aligned */
memblock_memsize_check_size(struct memsize_rgn_struct * rgn)2599 static inline void memblock_memsize_check_size(struct memsize_rgn_struct *rgn)
2600 {
2601 	phys_addr_t phy, end, freed = 0;
2602 	bool has_freed = false;
2603 	struct page *page;
2604 
2605 	if (rgn->reusable || rgn->nomap)
2606 		return;
2607 
2608 	phy = rgn->base;
2609 	end = rgn->base + rgn->size;
2610 	while (phy < end) {
2611 		unsigned long pfn = __phys_to_pfn(phy);
2612 
2613 		if (!pfn_valid(pfn))
2614 			return;
2615 		page = pfn_to_page(pfn);
2616 		if (!has_freed && !PageReserved(page)) {
2617 			has_freed = true;
2618 			freed = phy;
2619 		} else if (has_freed && PageReserved(page)) {
2620 			has_freed = false;
2621 			memblock_memsize_free(freed, phy - freed);
2622 		}
2623 
2624 		if (has_freed && (phy + SZ_64K >= end))
2625 			memblock_memsize_free(freed, end - freed);
2626 
2627 		/* check the first page only */
2628 		phy += SZ_64K;
2629 	}
2630 }
2631 
memblock_memsize_show(struct seq_file * m,void * private)2632 static int memblock_memsize_show(struct seq_file *m, void *private)
2633 {
2634 	int i;
2635 	struct memsize_rgn_struct *rgn;
2636 	unsigned long reserved = 0, reusable = 0, total;
2637 	unsigned long system = totalram_pages() << PAGE_SHIFT;
2638 	unsigned long etc;
2639 
2640 	etc = memsize_kinit;
2641 	etc -= memsize_code + memsize_data + memsize_ro + memsize_bss +
2642 		memsize_memmap;
2643 
2644 	system += memsize_reusable_size;
2645 	sort(memsize_rgn, memsize_rgn_count,
2646 	     sizeof(memsize_rgn[0]), memsize_rgn_cmp, NULL);
2647 	for (i = 0; i < memsize_rgn_count; i++) {
2648 		phys_addr_t base, end;
2649 		long size;
2650 
2651 		rgn = &memsize_rgn[i];
2652 		memblock_memsize_check_size(rgn);
2653 		base = rgn->base;
2654 		size = rgn->size;
2655 		end = base + size;
2656 
2657 		seq_printf(m, "0x%pK-0x%pK 0x%08lx ( %7lu KB ) %s %s %s\n",
2658 			   (void *)base, (void *)end,
2659 			   size, DIV_ROUND_UP(size, SZ_1K),
2660 			   rgn->nomap ? "nomap" : "  map",
2661 			   rgn->reusable ? "reusable" : "unusable",
2662 			   rgn->name);
2663 		if (rgn->reusable)
2664 			reusable += (unsigned long)rgn->size;
2665 		else
2666 			reserved += (unsigned long)rgn->size;
2667 	}
2668 
2669 	total = memsize_kinit + reserved + system;
2670 
2671 	seq_puts(m, "\n");
2672 	seq_printf(m, "Reserved    : %7lu KB\n",
2673 		   DIV_ROUND_UP(memsize_kinit + reserved, SZ_1K));
2674 	seq_printf(m, " .kernel    : %7lu KB\n",
2675 		   DIV_ROUND_UP(memsize_kinit, SZ_1K));
2676 	seq_printf(m, "  .text     : %7lu KB\n"
2677 		      "  .rwdata   : %7lu KB\n"
2678 		      "  .rodata   : %7lu KB\n"
2679 		      "  .bss      : %7lu KB\n"
2680 		      "  .memmap   : %7lu KB\n"
2681 		      "  .etc      : %7lu KB\n",
2682 			DIV_ROUND_UP(memsize_code, SZ_1K),
2683 			DIV_ROUND_UP(memsize_data, SZ_1K),
2684 			DIV_ROUND_UP(memsize_ro, SZ_1K),
2685 			DIV_ROUND_UP(memsize_bss, SZ_1K),
2686 			DIV_ROUND_UP(memsize_memmap, SZ_1K),
2687 			DIV_ROUND_UP(etc, SZ_1K));
2688 	seq_printf(m, " .unusable  : %7lu KB\n",
2689 		   DIV_ROUND_UP(reserved, SZ_1K));
2690 	seq_printf(m, "System      : %7lu KB\n",
2691 		   DIV_ROUND_UP(system, SZ_1K));
2692 	seq_printf(m, " .common    : %7lu KB\n",
2693 		   DIV_ROUND_UP(system - reusable, SZ_1K));
2694 	seq_printf(m, " .reusable  : %7lu KB\n",
2695 		   DIV_ROUND_UP(reusable, SZ_1K));
2696 	seq_printf(m, "Total       : %7lu KB ( %5lu.%02lu MB )\n",
2697 		   DIV_ROUND_UP(total, SZ_1K),
2698 		   total >> 20, ((total % SZ_1M) * 100) >> 20);
2699 	return 0;
2700 }
2701 
2702 DEFINE_SHOW_ATTRIBUTE(memblock_memsize);
2703 
memblock_init_debugfs(void)2704 static int __init memblock_init_debugfs(void)
2705 {
2706 	struct dentry *root = debugfs_create_dir("memblock", NULL);
2707 
2708 	debugfs_create_file("memory", 0444, root,
2709 			    &memblock.memory, &memblock_debug_fops);
2710 	debugfs_create_file("reserved", 0444, root,
2711 			    &memblock.reserved, &memblock_debug_fops);
2712 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
2713 	debugfs_create_file("physmem", 0444, root, &physmem,
2714 			    &memblock_debug_fops);
2715 #endif
2716 	if (memsize_state == MEMBLOCK_MEMSIZE_DEBUGFS)
2717 		debugfs_create_file("memsize", 0444, root, NULL,
2718 				    &memblock_memsize_fops);
2719 	else if (memsize_state == MEMBLOCK_MEMSIZE_PROCFS)
2720 		proc_create_single("memsize", 0, NULL,
2721 				    memblock_memsize_show);
2722 
2723 	return 0;
2724 }
2725 __initcall(memblock_init_debugfs);
2726 
2727 #endif /* CONFIG_DEBUG_FS */
2728