1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * Procedures for maintaining information about logical memory blocks.
4  *
5  * Peter Bergner, IBM Corp.	June 2001.
6  * Copyright (C) 2001 Peter Bergner.
7  */
8 
9 #include <linux/kernel.h>
10 #include <linux/slab.h>
11 #include <linux/init.h>
12 #include <linux/bitops.h>
13 #include <linux/poison.h>
14 #include <linux/pfn.h>
15 #include <linux/debugfs.h>
16 #include <linux/kmemleak.h>
17 #include <linux/seq_file.h>
18 #include <linux/memblock.h>
19 
20 #include <asm/sections.h>
21 #include <linux/io.h>
22 
23 #include "internal.h"
24 
25 #define INIT_MEMBLOCK_REGIONS			128
26 #define INIT_PHYSMEM_REGIONS			4
27 
28 #ifndef INIT_MEMBLOCK_RESERVED_REGIONS
29 # define INIT_MEMBLOCK_RESERVED_REGIONS		INIT_MEMBLOCK_REGIONS
30 #endif
31 
32 #ifndef INIT_MEMBLOCK_MEMORY_REGIONS
33 #define INIT_MEMBLOCK_MEMORY_REGIONS		INIT_MEMBLOCK_REGIONS
34 #endif
35 
36 /**
37  * DOC: memblock overview
38  *
39  * Memblock is a method of managing memory regions during the early
40  * boot period when the usual kernel memory allocators are not up and
41  * running.
42  *
43  * Memblock views the system memory as collections of contiguous
44  * regions. There are several types of these collections:
45  *
46  * * ``memory`` - describes the physical memory available to the
47  *   kernel; this may differ from the actual physical memory installed
48  *   in the system, for instance when the memory is restricted with
49  *   ``mem=`` command line parameter
50  * * ``reserved`` - describes the regions that were allocated
51  * * ``physmem`` - describes the actual physical memory available during
52  *   boot regardless of the possible restrictions and memory hot(un)plug;
53  *   the ``physmem`` type is only available on some architectures.
54  *
55  * Each region is represented by struct memblock_region that
56  * defines the region extents, its attributes and NUMA node id on NUMA
57  * systems. Every memory type is described by the struct memblock_type
58  * which contains an array of memory regions along with
59  * the allocator metadata. The "memory" and "reserved" types are nicely
60  * wrapped with struct memblock. This structure is statically
61  * initialized at build time. The region arrays are initially sized to
62  * %INIT_MEMBLOCK_MEMORY_REGIONS for "memory" and
63  * %INIT_MEMBLOCK_RESERVED_REGIONS for "reserved". The region array
64  * for "physmem" is initially sized to %INIT_PHYSMEM_REGIONS.
65  * The memblock_allow_resize() enables automatic resizing of the region
66  * arrays during addition of new regions. This feature should be used
67  * with care so that memory allocated for the region array will not
68  * overlap with areas that should be reserved, for example initrd.
69  *
70  * The early architecture setup should tell memblock what the physical
71  * memory layout is by using memblock_add() or memblock_add_node()
72  * functions. The first function does not assign the region to a NUMA
73  * node and it is appropriate for UMA systems. Yet, it is possible to
74  * use it on NUMA systems as well and assign the region to a NUMA node
75  * later in the setup process using memblock_set_node(). The
76  * memblock_add_node() performs such an assignment directly.
77  *
78  * Once memblock is setup the memory can be allocated using one of the
79  * API variants:
80  *
81  * * memblock_phys_alloc*() - these functions return the **physical**
82  *   address of the allocated memory
83  * * memblock_alloc*() - these functions return the **virtual** address
84  *   of the allocated memory.
85  *
86  * Note, that both API variants use implicit assumptions about allowed
87  * memory ranges and the fallback methods. Consult the documentation
88  * of memblock_alloc_internal() and memblock_alloc_range_nid()
89  * functions for more elaborate description.
90  *
91  * As the system boot progresses, the architecture specific mem_init()
92  * function frees all the memory to the buddy page allocator.
93  *
94  * Unless an architecture enables %CONFIG_ARCH_KEEP_MEMBLOCK, the
95  * memblock data structures (except "physmem") will be discarded after the
96  * system initialization completes.
97  */
98 
99 #ifndef CONFIG_NUMA
100 struct pglist_data __refdata contig_page_data;
101 EXPORT_SYMBOL(contig_page_data);
102 #endif
103 
104 unsigned long max_low_pfn;
105 unsigned long min_low_pfn;
106 unsigned long max_pfn;
107 unsigned long long max_possible_pfn;
108 
109 static struct memblock_region memblock_memory_init_regions[INIT_MEMBLOCK_MEMORY_REGIONS] __initdata_memblock;
110 static struct memblock_region memblock_reserved_init_regions[INIT_MEMBLOCK_RESERVED_REGIONS] __initdata_memblock;
111 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
112 static struct memblock_region memblock_physmem_init_regions[INIT_PHYSMEM_REGIONS];
113 #endif
114 
115 struct memblock memblock __initdata_memblock = {
116 	.memory.regions		= memblock_memory_init_regions,
117 	.memory.cnt		= 1,	/* empty dummy entry */
118 	.memory.max		= INIT_MEMBLOCK_MEMORY_REGIONS,
119 	.memory.name		= "memory",
120 
121 	.reserved.regions	= memblock_reserved_init_regions,
122 	.reserved.cnt		= 1,	/* empty dummy entry */
123 	.reserved.max		= INIT_MEMBLOCK_RESERVED_REGIONS,
124 	.reserved.name		= "reserved",
125 
126 	.bottom_up		= false,
127 	.current_limit		= MEMBLOCK_ALLOC_ANYWHERE,
128 };
129 
130 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
131 struct memblock_type physmem = {
132 	.regions		= memblock_physmem_init_regions,
133 	.cnt			= 1,	/* empty dummy entry */
134 	.max			= INIT_PHYSMEM_REGIONS,
135 	.name			= "physmem",
136 };
137 #endif
138 
139 /*
140  * keep a pointer to &memblock.memory in the text section to use it in
141  * __next_mem_range() and its helpers.
142  *  For architectures that do not keep memblock data after init, this
143  * pointer will be reset to NULL at memblock_discard()
144  */
145 static __refdata struct memblock_type *memblock_memory = &memblock.memory;
146 
147 #define for_each_memblock_type(i, memblock_type, rgn)			\
148 	for (i = 0, rgn = &memblock_type->regions[0];			\
149 	     i < memblock_type->cnt;					\
150 	     i++, rgn = &memblock_type->regions[i])
151 
152 #define memblock_dbg(fmt, ...)						\
153 	do {								\
154 		if (memblock_debug)					\
155 			pr_info(fmt, ##__VA_ARGS__);			\
156 	} while (0)
157 
158 static int memblock_debug __initdata_memblock;
159 static bool system_has_some_mirror __initdata_memblock;
160 static int memblock_can_resize __initdata_memblock;
161 static int memblock_memory_in_slab __initdata_memblock;
162 static int memblock_reserved_in_slab __initdata_memblock;
163 
memblock_has_mirror(void)164 bool __init_memblock memblock_has_mirror(void)
165 {
166 	return system_has_some_mirror;
167 }
168 
choose_memblock_flags(void)169 static enum memblock_flags __init_memblock choose_memblock_flags(void)
170 {
171 	return system_has_some_mirror ? MEMBLOCK_MIRROR : MEMBLOCK_NONE;
172 }
173 
174 /* adjust *@size so that (@base + *@size) doesn't overflow, return new size */
memblock_cap_size(phys_addr_t base,phys_addr_t * size)175 static inline phys_addr_t memblock_cap_size(phys_addr_t base, phys_addr_t *size)
176 {
177 	return *size = min(*size, PHYS_ADDR_MAX - base);
178 }
179 
180 /*
181  * Address comparison utilities
182  */
183 unsigned long __init_memblock
memblock_addrs_overlap(phys_addr_t base1,phys_addr_t size1,phys_addr_t base2,phys_addr_t size2)184 memblock_addrs_overlap(phys_addr_t base1, phys_addr_t size1, phys_addr_t base2,
185 		       phys_addr_t size2)
186 {
187 	return ((base1 < (base2 + size2)) && (base2 < (base1 + size1)));
188 }
189 
memblock_overlaps_region(struct memblock_type * type,phys_addr_t base,phys_addr_t size)190 bool __init_memblock memblock_overlaps_region(struct memblock_type *type,
191 					phys_addr_t base, phys_addr_t size)
192 {
193 	unsigned long i;
194 
195 	memblock_cap_size(base, &size);
196 
197 	for (i = 0; i < type->cnt; i++)
198 		if (memblock_addrs_overlap(base, size, type->regions[i].base,
199 					   type->regions[i].size))
200 			break;
201 	return i < type->cnt;
202 }
203 
204 /**
205  * __memblock_find_range_bottom_up - find free area utility in bottom-up
206  * @start: start of candidate range
207  * @end: end of candidate range, can be %MEMBLOCK_ALLOC_ANYWHERE or
208  *       %MEMBLOCK_ALLOC_ACCESSIBLE
209  * @size: size of free area to find
210  * @align: alignment of free area to find
211  * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
212  * @flags: pick from blocks based on memory attributes
213  *
214  * Utility called from memblock_find_in_range_node(), find free area bottom-up.
215  *
216  * Return:
217  * Found address on success, 0 on failure.
218  */
219 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)220 __memblock_find_range_bottom_up(phys_addr_t start, phys_addr_t end,
221 				phys_addr_t size, phys_addr_t align, int nid,
222 				enum memblock_flags flags)
223 {
224 	phys_addr_t this_start, this_end, cand;
225 	u64 i;
226 
227 	for_each_free_mem_range(i, nid, flags, &this_start, &this_end, NULL) {
228 		this_start = clamp(this_start, start, end);
229 		this_end = clamp(this_end, start, end);
230 
231 		cand = round_up(this_start, align);
232 		if (cand < this_end && this_end - cand >= size)
233 			return cand;
234 	}
235 
236 	return 0;
237 }
238 
239 /**
240  * __memblock_find_range_top_down - find free area utility, in top-down
241  * @start: start of candidate range
242  * @end: end of candidate range, can be %MEMBLOCK_ALLOC_ANYWHERE or
243  *       %MEMBLOCK_ALLOC_ACCESSIBLE
244  * @size: size of free area to find
245  * @align: alignment of free area to find
246  * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
247  * @flags: pick from blocks based on memory attributes
248  *
249  * Utility called from memblock_find_in_range_node(), find free area top-down.
250  *
251  * Return:
252  * Found address on success, 0 on failure.
253  */
254 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)255 __memblock_find_range_top_down(phys_addr_t start, phys_addr_t end,
256 			       phys_addr_t size, phys_addr_t align, int nid,
257 			       enum memblock_flags flags)
258 {
259 	phys_addr_t this_start, this_end, cand;
260 	u64 i;
261 
262 	for_each_free_mem_range_reverse(i, nid, flags, &this_start, &this_end,
263 					NULL) {
264 		this_start = clamp(this_start, start, end);
265 		this_end = clamp(this_end, start, end);
266 
267 		if (this_end < size)
268 			continue;
269 
270 		cand = round_down(this_end - size, align);
271 		if (cand >= this_start)
272 			return cand;
273 	}
274 
275 	return 0;
276 }
277 
278 /**
279  * memblock_find_in_range_node - find free area in given range and node
280  * @size: size of free area to find
281  * @align: alignment of free area to find
282  * @start: start of candidate range
283  * @end: end of candidate range, can be %MEMBLOCK_ALLOC_ANYWHERE or
284  *       %MEMBLOCK_ALLOC_ACCESSIBLE
285  * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
286  * @flags: pick from blocks based on memory attributes
287  *
288  * Find @size free area aligned to @align in the specified range and node.
289  *
290  * Return:
291  * Found address on success, 0 on failure.
292  */
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)293 static phys_addr_t __init_memblock memblock_find_in_range_node(phys_addr_t size,
294 					phys_addr_t align, phys_addr_t start,
295 					phys_addr_t end, int nid,
296 					enum memblock_flags flags)
297 {
298 	/* pump up @end */
299 	if (end == MEMBLOCK_ALLOC_ACCESSIBLE ||
300 	    end == MEMBLOCK_ALLOC_NOLEAKTRACE)
301 		end = memblock.current_limit;
302 
303 	/* avoid allocating the first page */
304 	start = max_t(phys_addr_t, start, PAGE_SIZE);
305 	end = max(start, end);
306 
307 	if (memblock_bottom_up())
308 		return __memblock_find_range_bottom_up(start, end, size, align,
309 						       nid, flags);
310 	else
311 		return __memblock_find_range_top_down(start, end, size, align,
312 						      nid, flags);
313 }
314 
315 /**
316  * memblock_find_in_range - find free area in given range
317  * @start: start of candidate range
318  * @end: end of candidate range, can be %MEMBLOCK_ALLOC_ANYWHERE or
319  *       %MEMBLOCK_ALLOC_ACCESSIBLE
320  * @size: size of free area to find
321  * @align: alignment of free area to find
322  *
323  * Find @size free area aligned to @align in the specified range.
324  *
325  * Return:
326  * Found address on success, 0 on failure.
327  */
memblock_find_in_range(phys_addr_t start,phys_addr_t end,phys_addr_t size,phys_addr_t align)328 static phys_addr_t __init_memblock memblock_find_in_range(phys_addr_t start,
329 					phys_addr_t end, phys_addr_t size,
330 					phys_addr_t align)
331 {
332 	phys_addr_t ret;
333 	enum memblock_flags flags = choose_memblock_flags();
334 
335 again:
336 	ret = memblock_find_in_range_node(size, align, start, end,
337 					    NUMA_NO_NODE, flags);
338 
339 	if (!ret && (flags & MEMBLOCK_MIRROR)) {
340 		pr_warn_ratelimited("Could not allocate %pap bytes of mirrored memory\n",
341 			&size);
342 		flags &= ~MEMBLOCK_MIRROR;
343 		goto again;
344 	}
345 
346 	return ret;
347 }
348 
memblock_remove_region(struct memblock_type * type,unsigned long r)349 static void __init_memblock memblock_remove_region(struct memblock_type *type, unsigned long r)
350 {
351 	type->total_size -= type->regions[r].size;
352 	memmove(&type->regions[r], &type->regions[r + 1],
353 		(type->cnt - (r + 1)) * sizeof(type->regions[r]));
354 	type->cnt--;
355 
356 	/* Special case for empty arrays */
357 	if (type->cnt == 0) {
358 		WARN_ON(type->total_size != 0);
359 		type->cnt = 1;
360 		type->regions[0].base = 0;
361 		type->regions[0].size = 0;
362 		type->regions[0].flags = 0;
363 		memblock_set_region_node(&type->regions[0], MAX_NUMNODES);
364 	}
365 }
366 
367 #ifndef CONFIG_ARCH_KEEP_MEMBLOCK
368 /**
369  * memblock_discard - discard memory and reserved arrays if they were allocated
370  */
memblock_discard(void)371 void __init memblock_discard(void)
372 {
373 	phys_addr_t addr, size;
374 
375 	if (memblock.reserved.regions != memblock_reserved_init_regions) {
376 		addr = __pa(memblock.reserved.regions);
377 		size = PAGE_ALIGN(sizeof(struct memblock_region) *
378 				  memblock.reserved.max);
379 		if (memblock_reserved_in_slab)
380 			kfree(memblock.reserved.regions);
381 		else
382 			memblock_free_late(addr, size);
383 	}
384 
385 	if (memblock.memory.regions != memblock_memory_init_regions) {
386 		addr = __pa(memblock.memory.regions);
387 		size = PAGE_ALIGN(sizeof(struct memblock_region) *
388 				  memblock.memory.max);
389 		if (memblock_memory_in_slab)
390 			kfree(memblock.memory.regions);
391 		else
392 			memblock_free_late(addr, size);
393 	}
394 
395 	memblock_memory = NULL;
396 }
397 #endif
398 
399 /**
400  * memblock_double_array - double the size of the memblock regions array
401  * @type: memblock type of the regions array being doubled
402  * @new_area_start: starting address of memory range to avoid overlap with
403  * @new_area_size: size of memory range to avoid overlap with
404  *
405  * Double the size of the @type regions array. If memblock is being used to
406  * allocate memory for a new reserved regions array and there is a previously
407  * allocated memory range [@new_area_start, @new_area_start + @new_area_size]
408  * waiting to be reserved, ensure the memory used by the new array does
409  * not overlap.
410  *
411  * Return:
412  * 0 on success, -1 on failure.
413  */
memblock_double_array(struct memblock_type * type,phys_addr_t new_area_start,phys_addr_t new_area_size)414 static int __init_memblock memblock_double_array(struct memblock_type *type,
415 						phys_addr_t new_area_start,
416 						phys_addr_t new_area_size)
417 {
418 	struct memblock_region *new_array, *old_array;
419 	phys_addr_t old_alloc_size, new_alloc_size;
420 	phys_addr_t old_size, new_size, addr, new_end;
421 	int use_slab = slab_is_available();
422 	int *in_slab;
423 
424 	/* We don't allow resizing until we know about the reserved regions
425 	 * of memory that aren't suitable for allocation
426 	 */
427 	if (!memblock_can_resize)
428 		return -1;
429 
430 	/* Calculate new doubled size */
431 	old_size = type->max * sizeof(struct memblock_region);
432 	new_size = old_size << 1;
433 	/*
434 	 * We need to allocated new one align to PAGE_SIZE,
435 	 *   so we can free them completely later.
436 	 */
437 	old_alloc_size = PAGE_ALIGN(old_size);
438 	new_alloc_size = PAGE_ALIGN(new_size);
439 
440 	/* Retrieve the slab flag */
441 	if (type == &memblock.memory)
442 		in_slab = &memblock_memory_in_slab;
443 	else
444 		in_slab = &memblock_reserved_in_slab;
445 
446 	/* Try to find some space for it */
447 	if (use_slab) {
448 		new_array = kmalloc(new_size, GFP_KERNEL);
449 		addr = new_array ? __pa(new_array) : 0;
450 	} else {
451 		/* only exclude range when trying to double reserved.regions */
452 		if (type != &memblock.reserved)
453 			new_area_start = new_area_size = 0;
454 
455 		addr = memblock_find_in_range(new_area_start + new_area_size,
456 						memblock.current_limit,
457 						new_alloc_size, PAGE_SIZE);
458 		if (!addr && new_area_size)
459 			addr = memblock_find_in_range(0,
460 				min(new_area_start, memblock.current_limit),
461 				new_alloc_size, PAGE_SIZE);
462 
463 		new_array = addr ? __va(addr) : NULL;
464 	}
465 	if (!addr) {
466 		pr_err("memblock: Failed to double %s array from %ld to %ld entries !\n",
467 		       type->name, type->max, type->max * 2);
468 		return -1;
469 	}
470 
471 	new_end = addr + new_size - 1;
472 	memblock_dbg("memblock: %s is doubled to %ld at [%pa-%pa]",
473 			type->name, type->max * 2, &addr, &new_end);
474 
475 	/*
476 	 * Found space, we now need to move the array over before we add the
477 	 * reserved region since it may be our reserved array itself that is
478 	 * full.
479 	 */
480 	memcpy(new_array, type->regions, old_size);
481 	memset(new_array + type->max, 0, old_size);
482 	old_array = type->regions;
483 	type->regions = new_array;
484 	type->max <<= 1;
485 
486 	/* Free old array. We needn't free it if the array is the static one */
487 	if (*in_slab)
488 		kfree(old_array);
489 	else if (old_array != memblock_memory_init_regions &&
490 		 old_array != memblock_reserved_init_regions)
491 		memblock_free(old_array, old_alloc_size);
492 
493 	/*
494 	 * Reserve the new array if that comes from the memblock.  Otherwise, we
495 	 * needn't do it
496 	 */
497 	if (!use_slab)
498 		BUG_ON(memblock_reserve(addr, new_alloc_size));
499 
500 	/* Update slab flag */
501 	*in_slab = use_slab;
502 
503 	return 0;
504 }
505 
506 /**
507  * memblock_merge_regions - merge neighboring compatible regions
508  * @type: memblock type to scan
509  * @start_rgn: start scanning from (@start_rgn - 1)
510  * @end_rgn: end scanning at (@end_rgn - 1)
511  * Scan @type and merge neighboring compatible regions in [@start_rgn - 1, @end_rgn)
512  */
memblock_merge_regions(struct memblock_type * type,unsigned long start_rgn,unsigned long end_rgn)513 static void __init_memblock memblock_merge_regions(struct memblock_type *type,
514 						   unsigned long start_rgn,
515 						   unsigned long end_rgn)
516 {
517 	int i = 0;
518 	if (start_rgn)
519 		i = start_rgn - 1;
520 	end_rgn = min(end_rgn, type->cnt - 1);
521 	while (i < end_rgn) {
522 		struct memblock_region *this = &type->regions[i];
523 		struct memblock_region *next = &type->regions[i + 1];
524 
525 		if (this->base + this->size != next->base ||
526 		    memblock_get_region_node(this) !=
527 		    memblock_get_region_node(next) ||
528 		    this->flags != next->flags) {
529 			BUG_ON(this->base + this->size > next->base);
530 			i++;
531 			continue;
532 		}
533 
534 		this->size += next->size;
535 		/* move forward from next + 1, index of which is i + 2 */
536 		memmove(next, next + 1, (type->cnt - (i + 2)) * sizeof(*next));
537 		type->cnt--;
538 		end_rgn--;
539 	}
540 }
541 
542 /**
543  * memblock_insert_region - insert new memblock region
544  * @type:	memblock type to insert into
545  * @idx:	index for the insertion point
546  * @base:	base address of the new region
547  * @size:	size of the new region
548  * @nid:	node id of the new region
549  * @flags:	flags of the new region
550  *
551  * Insert new memblock region [@base, @base + @size) into @type at @idx.
552  * @type must already have extra room to accommodate the new region.
553  */
memblock_insert_region(struct memblock_type * type,int idx,phys_addr_t base,phys_addr_t size,int nid,enum memblock_flags flags)554 static void __init_memblock memblock_insert_region(struct memblock_type *type,
555 						   int idx, phys_addr_t base,
556 						   phys_addr_t size,
557 						   int nid,
558 						   enum memblock_flags flags)
559 {
560 	struct memblock_region *rgn = &type->regions[idx];
561 
562 	BUG_ON(type->cnt >= type->max);
563 	memmove(rgn + 1, rgn, (type->cnt - idx) * sizeof(*rgn));
564 	rgn->base = base;
565 	rgn->size = size;
566 	rgn->flags = flags;
567 	memblock_set_region_node(rgn, nid);
568 	type->cnt++;
569 	type->total_size += size;
570 }
571 
572 /**
573  * memblock_add_range - add new memblock region
574  * @type: memblock type to add new region into
575  * @base: base address of the new region
576  * @size: size of the new region
577  * @nid: nid of the new region
578  * @flags: flags of the new region
579  *
580  * Add new memblock region [@base, @base + @size) into @type.  The new region
581  * is allowed to overlap with existing ones - overlaps don't affect already
582  * existing regions.  @type is guaranteed to be minimal (all neighbouring
583  * compatible regions are merged) after the addition.
584  *
585  * Return:
586  * 0 on success, -errno on failure.
587  */
memblock_add_range(struct memblock_type * type,phys_addr_t base,phys_addr_t size,int nid,enum memblock_flags flags)588 static int __init_memblock memblock_add_range(struct memblock_type *type,
589 				phys_addr_t base, phys_addr_t size,
590 				int nid, enum memblock_flags flags)
591 {
592 	bool insert = false;
593 	phys_addr_t obase = base;
594 	phys_addr_t end = base + memblock_cap_size(base, &size);
595 	int idx, nr_new, start_rgn = -1, end_rgn;
596 	struct memblock_region *rgn;
597 
598 	if (!size)
599 		return 0;
600 
601 	/* special case for empty array */
602 	if (type->regions[0].size == 0) {
603 		WARN_ON(type->cnt != 1 || type->total_size);
604 		type->regions[0].base = base;
605 		type->regions[0].size = size;
606 		type->regions[0].flags = flags;
607 		memblock_set_region_node(&type->regions[0], nid);
608 		type->total_size = size;
609 		return 0;
610 	}
611 
612 	/*
613 	 * The worst case is when new range overlaps all existing regions,
614 	 * then we'll need type->cnt + 1 empty regions in @type. So if
615 	 * type->cnt * 2 + 1 is less than or equal to type->max, we know
616 	 * that there is enough empty regions in @type, and we can insert
617 	 * regions directly.
618 	 */
619 	if (type->cnt * 2 + 1 <= type->max)
620 		insert = true;
621 
622 repeat:
623 	/*
624 	 * The following is executed twice.  Once with %false @insert and
625 	 * then with %true.  The first counts the number of regions needed
626 	 * to accommodate the new area.  The second actually inserts them.
627 	 */
628 	base = obase;
629 	nr_new = 0;
630 
631 	for_each_memblock_type(idx, type, rgn) {
632 		phys_addr_t rbase = rgn->base;
633 		phys_addr_t rend = rbase + rgn->size;
634 
635 		if (rbase >= end)
636 			break;
637 		if (rend <= base)
638 			continue;
639 		/*
640 		 * @rgn overlaps.  If it separates the lower part of new
641 		 * area, insert that portion.
642 		 */
643 		if (rbase > base) {
644 #ifdef CONFIG_NUMA
645 			WARN_ON(nid != memblock_get_region_node(rgn));
646 #endif
647 			WARN_ON(flags != rgn->flags);
648 			nr_new++;
649 			if (insert) {
650 				if (start_rgn == -1)
651 					start_rgn = idx;
652 				end_rgn = idx + 1;
653 				memblock_insert_region(type, idx++, base,
654 						       rbase - base, nid,
655 						       flags);
656 			}
657 		}
658 		/* area below @rend is dealt with, forget about it */
659 		base = min(rend, end);
660 	}
661 
662 	/* insert the remaining portion */
663 	if (base < end) {
664 		nr_new++;
665 		if (insert) {
666 			if (start_rgn == -1)
667 				start_rgn = idx;
668 			end_rgn = idx + 1;
669 			memblock_insert_region(type, idx, base, end - base,
670 					       nid, flags);
671 		}
672 	}
673 
674 	if (!nr_new)
675 		return 0;
676 
677 	/*
678 	 * If this was the first round, resize array and repeat for actual
679 	 * insertions; otherwise, merge and return.
680 	 */
681 	if (!insert) {
682 		while (type->cnt + nr_new > type->max)
683 			if (memblock_double_array(type, obase, size) < 0)
684 				return -ENOMEM;
685 		insert = true;
686 		goto repeat;
687 	} else {
688 		memblock_merge_regions(type, start_rgn, end_rgn);
689 		return 0;
690 	}
691 }
692 
693 /**
694  * memblock_add_node - add new memblock region within a NUMA node
695  * @base: base address of the new region
696  * @size: size of the new region
697  * @nid: nid of the new region
698  * @flags: flags of the new region
699  *
700  * Add new memblock region [@base, @base + @size) to the "memory"
701  * type. See memblock_add_range() description for mode details
702  *
703  * Return:
704  * 0 on success, -errno on failure.
705  */
memblock_add_node(phys_addr_t base,phys_addr_t size,int nid,enum memblock_flags flags)706 int __init_memblock memblock_add_node(phys_addr_t base, phys_addr_t size,
707 				      int nid, enum memblock_flags flags)
708 {
709 	phys_addr_t end = base + size - 1;
710 
711 	memblock_dbg("%s: [%pa-%pa] nid=%d flags=%x %pS\n", __func__,
712 		     &base, &end, nid, flags, (void *)_RET_IP_);
713 
714 	return memblock_add_range(&memblock.memory, base, size, nid, flags);
715 }
716 
717 /**
718  * memblock_add - add new memblock region
719  * @base: base address of the new region
720  * @size: size of the new region
721  *
722  * Add new memblock region [@base, @base + @size) to the "memory"
723  * type. See memblock_add_range() description for mode details
724  *
725  * Return:
726  * 0 on success, -errno on failure.
727  */
memblock_add(phys_addr_t base,phys_addr_t size)728 int __init_memblock memblock_add(phys_addr_t base, phys_addr_t size)
729 {
730 	phys_addr_t end = base + size - 1;
731 
732 	memblock_dbg("%s: [%pa-%pa] %pS\n", __func__,
733 		     &base, &end, (void *)_RET_IP_);
734 
735 	return memblock_add_range(&memblock.memory, base, size, MAX_NUMNODES, 0);
736 }
737 
738 /**
739  * memblock_isolate_range - isolate given range into disjoint memblocks
740  * @type: memblock type to isolate range for
741  * @base: base of range to isolate
742  * @size: size of range to isolate
743  * @start_rgn: out parameter for the start of isolated region
744  * @end_rgn: out parameter for the end of isolated region
745  *
746  * Walk @type and ensure that regions don't cross the boundaries defined by
747  * [@base, @base + @size).  Crossing regions are split at the boundaries,
748  * which may create at most two more regions.  The index of the first
749  * region inside the range is returned in *@start_rgn and end in *@end_rgn.
750  *
751  * Return:
752  * 0 on success, -errno on failure.
753  */
memblock_isolate_range(struct memblock_type * type,phys_addr_t base,phys_addr_t size,int * start_rgn,int * end_rgn)754 static int __init_memblock memblock_isolate_range(struct memblock_type *type,
755 					phys_addr_t base, phys_addr_t size,
756 					int *start_rgn, int *end_rgn)
757 {
758 	phys_addr_t end = base + memblock_cap_size(base, &size);
759 	int idx;
760 	struct memblock_region *rgn;
761 
762 	*start_rgn = *end_rgn = 0;
763 
764 	if (!size)
765 		return 0;
766 
767 	/* we'll create at most two more regions */
768 	while (type->cnt + 2 > type->max)
769 		if (memblock_double_array(type, base, size) < 0)
770 			return -ENOMEM;
771 
772 	for_each_memblock_type(idx, type, rgn) {
773 		phys_addr_t rbase = rgn->base;
774 		phys_addr_t rend = rbase + rgn->size;
775 
776 		if (rbase >= end)
777 			break;
778 		if (rend <= base)
779 			continue;
780 
781 		if (rbase < base) {
782 			/*
783 			 * @rgn intersects from below.  Split and continue
784 			 * to process the next region - the new top half.
785 			 */
786 			rgn->base = base;
787 			rgn->size -= base - rbase;
788 			type->total_size -= base - rbase;
789 			memblock_insert_region(type, idx, rbase, base - rbase,
790 					       memblock_get_region_node(rgn),
791 					       rgn->flags);
792 		} else if (rend > end) {
793 			/*
794 			 * @rgn intersects from above.  Split and redo the
795 			 * current region - the new bottom half.
796 			 */
797 			rgn->base = end;
798 			rgn->size -= end - rbase;
799 			type->total_size -= end - rbase;
800 			memblock_insert_region(type, idx--, rbase, end - rbase,
801 					       memblock_get_region_node(rgn),
802 					       rgn->flags);
803 		} else {
804 			/* @rgn is fully contained, record it */
805 			if (!*end_rgn)
806 				*start_rgn = idx;
807 			*end_rgn = idx + 1;
808 		}
809 	}
810 
811 	return 0;
812 }
813 
memblock_remove_range(struct memblock_type * type,phys_addr_t base,phys_addr_t size)814 static int __init_memblock memblock_remove_range(struct memblock_type *type,
815 					  phys_addr_t base, phys_addr_t size)
816 {
817 	int start_rgn, end_rgn;
818 	int i, ret;
819 
820 	ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
821 	if (ret)
822 		return ret;
823 
824 	for (i = end_rgn - 1; i >= start_rgn; i--)
825 		memblock_remove_region(type, i);
826 	return 0;
827 }
828 
memblock_remove(phys_addr_t base,phys_addr_t size)829 int __init_memblock memblock_remove(phys_addr_t base, phys_addr_t size)
830 {
831 	phys_addr_t end = base + size - 1;
832 
833 	memblock_dbg("%s: [%pa-%pa] %pS\n", __func__,
834 		     &base, &end, (void *)_RET_IP_);
835 
836 	return memblock_remove_range(&memblock.memory, base, size);
837 }
838 
839 /**
840  * memblock_free - free boot memory allocation
841  * @ptr: starting address of the  boot memory allocation
842  * @size: size of the boot memory block in bytes
843  *
844  * Free boot memory block previously allocated by memblock_alloc_xx() API.
845  * The freeing memory will not be released to the buddy allocator.
846  */
memblock_free(void * ptr,size_t size)847 void __init_memblock memblock_free(void *ptr, size_t size)
848 {
849 	if (ptr)
850 		memblock_phys_free(__pa(ptr), size);
851 }
852 
853 /**
854  * memblock_phys_free - free boot memory block
855  * @base: phys starting address of the  boot memory block
856  * @size: size of the boot memory block in bytes
857  *
858  * Free boot memory block previously allocated by memblock_phys_alloc_xx() API.
859  * The freeing memory will not be released to the buddy allocator.
860  */
memblock_phys_free(phys_addr_t base,phys_addr_t size)861 int __init_memblock memblock_phys_free(phys_addr_t base, phys_addr_t size)
862 {
863 	phys_addr_t end = base + size - 1;
864 
865 	memblock_dbg("%s: [%pa-%pa] %pS\n", __func__,
866 		     &base, &end, (void *)_RET_IP_);
867 
868 	kmemleak_free_part_phys(base, size);
869 	return memblock_remove_range(&memblock.reserved, base, size);
870 }
871 
memblock_reserve(phys_addr_t base,phys_addr_t size)872 int __init_memblock memblock_reserve(phys_addr_t base, phys_addr_t size)
873 {
874 	phys_addr_t end = base + size - 1;
875 
876 	memblock_dbg("%s: [%pa-%pa] %pS\n", __func__,
877 		     &base, &end, (void *)_RET_IP_);
878 
879 	return memblock_add_range(&memblock.reserved, base, size, MAX_NUMNODES, 0);
880 }
881 
882 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
memblock_physmem_add(phys_addr_t base,phys_addr_t size)883 int __init_memblock memblock_physmem_add(phys_addr_t base, phys_addr_t size)
884 {
885 	phys_addr_t end = base + size - 1;
886 
887 	memblock_dbg("%s: [%pa-%pa] %pS\n", __func__,
888 		     &base, &end, (void *)_RET_IP_);
889 
890 	return memblock_add_range(&physmem, base, size, MAX_NUMNODES, 0);
891 }
892 #endif
893 
894 /**
895  * memblock_setclr_flag - set or clear flag for a memory region
896  * @base: base address of the region
897  * @size: size of the region
898  * @set: set or clear the flag
899  * @flag: the flag to update
900  *
901  * This function isolates region [@base, @base + @size), and sets/clears flag
902  *
903  * Return: 0 on success, -errno on failure.
904  */
memblock_setclr_flag(phys_addr_t base,phys_addr_t size,int set,int flag)905 static int __init_memblock memblock_setclr_flag(phys_addr_t base,
906 				phys_addr_t size, int set, int flag)
907 {
908 	struct memblock_type *type = &memblock.memory;
909 	int i, ret, start_rgn, end_rgn;
910 
911 	ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
912 	if (ret)
913 		return ret;
914 
915 	for (i = start_rgn; i < end_rgn; i++) {
916 		struct memblock_region *r = &type->regions[i];
917 
918 		if (set)
919 			r->flags |= flag;
920 		else
921 			r->flags &= ~flag;
922 	}
923 
924 	memblock_merge_regions(type, start_rgn, end_rgn);
925 	return 0;
926 }
927 
928 /**
929  * memblock_mark_hotplug - Mark hotpluggable memory with flag MEMBLOCK_HOTPLUG.
930  * @base: the base phys addr of the region
931  * @size: the size of the region
932  *
933  * Return: 0 on success, -errno on failure.
934  */
memblock_mark_hotplug(phys_addr_t base,phys_addr_t size)935 int __init_memblock memblock_mark_hotplug(phys_addr_t base, phys_addr_t size)
936 {
937 	return memblock_setclr_flag(base, size, 1, MEMBLOCK_HOTPLUG);
938 }
939 
940 /**
941  * memblock_clear_hotplug - Clear flag MEMBLOCK_HOTPLUG for a specified region.
942  * @base: the base phys addr of the region
943  * @size: the size of the region
944  *
945  * Return: 0 on success, -errno on failure.
946  */
memblock_clear_hotplug(phys_addr_t base,phys_addr_t size)947 int __init_memblock memblock_clear_hotplug(phys_addr_t base, phys_addr_t size)
948 {
949 	return memblock_setclr_flag(base, size, 0, MEMBLOCK_HOTPLUG);
950 }
951 
952 /**
953  * memblock_mark_mirror - Mark mirrored memory with flag MEMBLOCK_MIRROR.
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_mirror(phys_addr_t base,phys_addr_t size)959 int __init_memblock memblock_mark_mirror(phys_addr_t base, phys_addr_t size)
960 {
961 	if (!mirrored_kernelcore)
962 		return 0;
963 
964 	system_has_some_mirror = true;
965 
966 	return memblock_setclr_flag(base, size, 1, MEMBLOCK_MIRROR);
967 }
968 
969 /**
970  * memblock_mark_nomap - Mark a memory region with flag MEMBLOCK_NOMAP.
971  * @base: the base phys addr of the region
972  * @size: the size of the region
973  *
974  * The memory regions marked with %MEMBLOCK_NOMAP will not be added to the
975  * direct mapping of the physical memory. These regions will still be
976  * covered by the memory map. The struct page representing NOMAP memory
977  * frames in the memory map will be PageReserved()
978  *
979  * Note: if the memory being marked %MEMBLOCK_NOMAP was allocated from
980  * memblock, the caller must inform kmemleak to ignore that memory
981  *
982  * Return: 0 on success, -errno on failure.
983  */
memblock_mark_nomap(phys_addr_t base,phys_addr_t size)984 int __init_memblock memblock_mark_nomap(phys_addr_t base, phys_addr_t size)
985 {
986 	return memblock_setclr_flag(base, size, 1, MEMBLOCK_NOMAP);
987 }
988 
989 /**
990  * memblock_clear_nomap - Clear flag MEMBLOCK_NOMAP for a specified region.
991  * @base: the base phys addr of the region
992  * @size: the size of the region
993  *
994  * Return: 0 on success, -errno on failure.
995  */
memblock_clear_nomap(phys_addr_t base,phys_addr_t size)996 int __init_memblock memblock_clear_nomap(phys_addr_t base, phys_addr_t size)
997 {
998 	return memblock_setclr_flag(base, size, 0, MEMBLOCK_NOMAP);
999 }
1000 
should_skip_region(struct memblock_type * type,struct memblock_region * m,int nid,int flags)1001 static bool should_skip_region(struct memblock_type *type,
1002 			       struct memblock_region *m,
1003 			       int nid, int flags)
1004 {
1005 	int m_nid = memblock_get_region_node(m);
1006 
1007 	/* we never skip regions when iterating memblock.reserved or physmem */
1008 	if (type != memblock_memory)
1009 		return false;
1010 
1011 	/* only memory regions are associated with nodes, check it */
1012 	if (nid != NUMA_NO_NODE && nid != m_nid)
1013 		return true;
1014 
1015 	/* skip hotpluggable memory regions if needed */
1016 	if (movable_node_is_enabled() && memblock_is_hotpluggable(m) &&
1017 	    !(flags & MEMBLOCK_HOTPLUG))
1018 		return true;
1019 
1020 	/* if we want mirror memory skip non-mirror memory regions */
1021 	if ((flags & MEMBLOCK_MIRROR) && !memblock_is_mirror(m))
1022 		return true;
1023 
1024 	/* skip nomap memory unless we were asked for it explicitly */
1025 	if (!(flags & MEMBLOCK_NOMAP) && memblock_is_nomap(m))
1026 		return true;
1027 
1028 	/* skip driver-managed memory unless we were asked for it explicitly */
1029 	if (!(flags & MEMBLOCK_DRIVER_MANAGED) && memblock_is_driver_managed(m))
1030 		return true;
1031 
1032 	return false;
1033 }
1034 
1035 /**
1036  * __next_mem_range - next function for for_each_free_mem_range() etc.
1037  * @idx: pointer to u64 loop variable
1038  * @nid: node selector, %NUMA_NO_NODE for all nodes
1039  * @flags: pick from blocks based on memory attributes
1040  * @type_a: pointer to memblock_type from where the range is taken
1041  * @type_b: pointer to memblock_type which excludes memory from being taken
1042  * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL
1043  * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL
1044  * @out_nid: ptr to int for nid of the range, can be %NULL
1045  *
1046  * Find the first area from *@idx which matches @nid, fill the out
1047  * parameters, and update *@idx for the next iteration.  The lower 32bit of
1048  * *@idx contains index into type_a and the upper 32bit indexes the
1049  * areas before each region in type_b.	For example, if type_b regions
1050  * look like the following,
1051  *
1052  *	0:[0-16), 1:[32-48), 2:[128-130)
1053  *
1054  * The upper 32bit indexes the following regions.
1055  *
1056  *	0:[0-0), 1:[16-32), 2:[48-128), 3:[130-MAX)
1057  *
1058  * As both region arrays are sorted, the function advances the two indices
1059  * in lockstep and returns each intersection.
1060  */
__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)1061 void __next_mem_range(u64 *idx, int nid, enum memblock_flags flags,
1062 		      struct memblock_type *type_a,
1063 		      struct memblock_type *type_b, phys_addr_t *out_start,
1064 		      phys_addr_t *out_end, int *out_nid)
1065 {
1066 	int idx_a = *idx & 0xffffffff;
1067 	int idx_b = *idx >> 32;
1068 
1069 	if (WARN_ONCE(nid == MAX_NUMNODES,
1070 	"Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
1071 		nid = NUMA_NO_NODE;
1072 
1073 	for (; idx_a < type_a->cnt; idx_a++) {
1074 		struct memblock_region *m = &type_a->regions[idx_a];
1075 
1076 		phys_addr_t m_start = m->base;
1077 		phys_addr_t m_end = m->base + m->size;
1078 		int	    m_nid = memblock_get_region_node(m);
1079 
1080 		if (should_skip_region(type_a, m, nid, flags))
1081 			continue;
1082 
1083 		if (!type_b) {
1084 			if (out_start)
1085 				*out_start = m_start;
1086 			if (out_end)
1087 				*out_end = m_end;
1088 			if (out_nid)
1089 				*out_nid = m_nid;
1090 			idx_a++;
1091 			*idx = (u32)idx_a | (u64)idx_b << 32;
1092 			return;
1093 		}
1094 
1095 		/* scan areas before each reservation */
1096 		for (; idx_b < type_b->cnt + 1; idx_b++) {
1097 			struct memblock_region *r;
1098 			phys_addr_t r_start;
1099 			phys_addr_t r_end;
1100 
1101 			r = &type_b->regions[idx_b];
1102 			r_start = idx_b ? r[-1].base + r[-1].size : 0;
1103 			r_end = idx_b < type_b->cnt ?
1104 				r->base : PHYS_ADDR_MAX;
1105 
1106 			/*
1107 			 * if idx_b advanced past idx_a,
1108 			 * break out to advance idx_a
1109 			 */
1110 			if (r_start >= m_end)
1111 				break;
1112 			/* if the two regions intersect, we're done */
1113 			if (m_start < r_end) {
1114 				if (out_start)
1115 					*out_start =
1116 						max(m_start, r_start);
1117 				if (out_end)
1118 					*out_end = min(m_end, r_end);
1119 				if (out_nid)
1120 					*out_nid = m_nid;
1121 				/*
1122 				 * The region which ends first is
1123 				 * advanced for the next iteration.
1124 				 */
1125 				if (m_end <= r_end)
1126 					idx_a++;
1127 				else
1128 					idx_b++;
1129 				*idx = (u32)idx_a | (u64)idx_b << 32;
1130 				return;
1131 			}
1132 		}
1133 	}
1134 
1135 	/* signal end of iteration */
1136 	*idx = ULLONG_MAX;
1137 }
1138 
1139 /**
1140  * __next_mem_range_rev - generic next function for for_each_*_range_rev()
1141  *
1142  * @idx: pointer to u64 loop variable
1143  * @nid: node selector, %NUMA_NO_NODE for all nodes
1144  * @flags: pick from blocks based on memory attributes
1145  * @type_a: pointer to memblock_type from where the range is taken
1146  * @type_b: pointer to memblock_type which excludes memory from being taken
1147  * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL
1148  * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL
1149  * @out_nid: ptr to int for nid of the range, can be %NULL
1150  *
1151  * Finds the next range from type_a which is not marked as unsuitable
1152  * in type_b.
1153  *
1154  * Reverse of __next_mem_range().
1155  */
__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)1156 void __init_memblock __next_mem_range_rev(u64 *idx, int nid,
1157 					  enum memblock_flags flags,
1158 					  struct memblock_type *type_a,
1159 					  struct memblock_type *type_b,
1160 					  phys_addr_t *out_start,
1161 					  phys_addr_t *out_end, int *out_nid)
1162 {
1163 	int idx_a = *idx & 0xffffffff;
1164 	int idx_b = *idx >> 32;
1165 
1166 	if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
1167 		nid = NUMA_NO_NODE;
1168 
1169 	if (*idx == (u64)ULLONG_MAX) {
1170 		idx_a = type_a->cnt - 1;
1171 		if (type_b != NULL)
1172 			idx_b = type_b->cnt;
1173 		else
1174 			idx_b = 0;
1175 	}
1176 
1177 	for (; idx_a >= 0; idx_a--) {
1178 		struct memblock_region *m = &type_a->regions[idx_a];
1179 
1180 		phys_addr_t m_start = m->base;
1181 		phys_addr_t m_end = m->base + m->size;
1182 		int m_nid = memblock_get_region_node(m);
1183 
1184 		if (should_skip_region(type_a, m, nid, flags))
1185 			continue;
1186 
1187 		if (!type_b) {
1188 			if (out_start)
1189 				*out_start = m_start;
1190 			if (out_end)
1191 				*out_end = m_end;
1192 			if (out_nid)
1193 				*out_nid = m_nid;
1194 			idx_a--;
1195 			*idx = (u32)idx_a | (u64)idx_b << 32;
1196 			return;
1197 		}
1198 
1199 		/* scan areas before each reservation */
1200 		for (; idx_b >= 0; idx_b--) {
1201 			struct memblock_region *r;
1202 			phys_addr_t r_start;
1203 			phys_addr_t r_end;
1204 
1205 			r = &type_b->regions[idx_b];
1206 			r_start = idx_b ? r[-1].base + r[-1].size : 0;
1207 			r_end = idx_b < type_b->cnt ?
1208 				r->base : PHYS_ADDR_MAX;
1209 			/*
1210 			 * if idx_b advanced past idx_a,
1211 			 * break out to advance idx_a
1212 			 */
1213 
1214 			if (r_end <= m_start)
1215 				break;
1216 			/* if the two regions intersect, we're done */
1217 			if (m_end > r_start) {
1218 				if (out_start)
1219 					*out_start = max(m_start, r_start);
1220 				if (out_end)
1221 					*out_end = min(m_end, r_end);
1222 				if (out_nid)
1223 					*out_nid = m_nid;
1224 				if (m_start >= r_start)
1225 					idx_a--;
1226 				else
1227 					idx_b--;
1228 				*idx = (u32)idx_a | (u64)idx_b << 32;
1229 				return;
1230 			}
1231 		}
1232 	}
1233 	/* signal end of iteration */
1234 	*idx = ULLONG_MAX;
1235 }
1236 
1237 /*
1238  * Common iterator interface used to define for_each_mem_pfn_range().
1239  */
__next_mem_pfn_range(int * idx,int nid,unsigned long * out_start_pfn,unsigned long * out_end_pfn,int * out_nid)1240 void __init_memblock __next_mem_pfn_range(int *idx, int nid,
1241 				unsigned long *out_start_pfn,
1242 				unsigned long *out_end_pfn, int *out_nid)
1243 {
1244 	struct memblock_type *type = &memblock.memory;
1245 	struct memblock_region *r;
1246 	int r_nid;
1247 
1248 	while (++*idx < type->cnt) {
1249 		r = &type->regions[*idx];
1250 		r_nid = memblock_get_region_node(r);
1251 
1252 		if (PFN_UP(r->base) >= PFN_DOWN(r->base + r->size))
1253 			continue;
1254 		if (nid == MAX_NUMNODES || nid == r_nid)
1255 			break;
1256 	}
1257 	if (*idx >= type->cnt) {
1258 		*idx = -1;
1259 		return;
1260 	}
1261 
1262 	if (out_start_pfn)
1263 		*out_start_pfn = PFN_UP(r->base);
1264 	if (out_end_pfn)
1265 		*out_end_pfn = PFN_DOWN(r->base + r->size);
1266 	if (out_nid)
1267 		*out_nid = r_nid;
1268 }
1269 
1270 /**
1271  * memblock_set_node - set node ID on memblock regions
1272  * @base: base of area to set node ID for
1273  * @size: size of area to set node ID for
1274  * @type: memblock type to set node ID for
1275  * @nid: node ID to set
1276  *
1277  * Set the nid of memblock @type regions in [@base, @base + @size) to @nid.
1278  * Regions which cross the area boundaries are split as necessary.
1279  *
1280  * Return:
1281  * 0 on success, -errno on failure.
1282  */
memblock_set_node(phys_addr_t base,phys_addr_t size,struct memblock_type * type,int nid)1283 int __init_memblock memblock_set_node(phys_addr_t base, phys_addr_t size,
1284 				      struct memblock_type *type, int nid)
1285 {
1286 #ifdef CONFIG_NUMA
1287 	int start_rgn, end_rgn;
1288 	int i, ret;
1289 
1290 	ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
1291 	if (ret)
1292 		return ret;
1293 
1294 	for (i = start_rgn; i < end_rgn; i++)
1295 		memblock_set_region_node(&type->regions[i], nid);
1296 
1297 	memblock_merge_regions(type, start_rgn, end_rgn);
1298 #endif
1299 	return 0;
1300 }
1301 
1302 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1303 /**
1304  * __next_mem_pfn_range_in_zone - iterator for for_each_*_range_in_zone()
1305  *
1306  * @idx: pointer to u64 loop variable
1307  * @zone: zone in which all of the memory blocks reside
1308  * @out_spfn: ptr to ulong for start pfn of the range, can be %NULL
1309  * @out_epfn: ptr to ulong for end pfn of the range, can be %NULL
1310  *
1311  * This function is meant to be a zone/pfn specific wrapper for the
1312  * for_each_mem_range type iterators. Specifically they are used in the
1313  * deferred memory init routines and as such we were duplicating much of
1314  * this logic throughout the code. So instead of having it in multiple
1315  * locations it seemed like it would make more sense to centralize this to
1316  * one new iterator that does everything they need.
1317  */
1318 void __init_memblock
__next_mem_pfn_range_in_zone(u64 * idx,struct zone * zone,unsigned long * out_spfn,unsigned long * out_epfn)1319 __next_mem_pfn_range_in_zone(u64 *idx, struct zone *zone,
1320 			     unsigned long *out_spfn, unsigned long *out_epfn)
1321 {
1322 	int zone_nid = zone_to_nid(zone);
1323 	phys_addr_t spa, epa;
1324 
1325 	__next_mem_range(idx, zone_nid, MEMBLOCK_NONE,
1326 			 &memblock.memory, &memblock.reserved,
1327 			 &spa, &epa, NULL);
1328 
1329 	while (*idx != U64_MAX) {
1330 		unsigned long epfn = PFN_DOWN(epa);
1331 		unsigned long spfn = PFN_UP(spa);
1332 
1333 		/*
1334 		 * Verify the end is at least past the start of the zone and
1335 		 * that we have at least one PFN to initialize.
1336 		 */
1337 		if (zone->zone_start_pfn < epfn && spfn < epfn) {
1338 			/* if we went too far just stop searching */
1339 			if (zone_end_pfn(zone) <= spfn) {
1340 				*idx = U64_MAX;
1341 				break;
1342 			}
1343 
1344 			if (out_spfn)
1345 				*out_spfn = max(zone->zone_start_pfn, spfn);
1346 			if (out_epfn)
1347 				*out_epfn = min(zone_end_pfn(zone), epfn);
1348 
1349 			return;
1350 		}
1351 
1352 		__next_mem_range(idx, zone_nid, MEMBLOCK_NONE,
1353 				 &memblock.memory, &memblock.reserved,
1354 				 &spa, &epa, NULL);
1355 	}
1356 
1357 	/* signal end of iteration */
1358 	if (out_spfn)
1359 		*out_spfn = ULONG_MAX;
1360 	if (out_epfn)
1361 		*out_epfn = 0;
1362 }
1363 
1364 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
1365 
1366 /**
1367  * memblock_alloc_range_nid - allocate boot memory block
1368  * @size: size of memory block to be allocated in bytes
1369  * @align: alignment of the region and block's size
1370  * @start: the lower bound of the memory region to allocate (phys address)
1371  * @end: the upper bound of the memory region to allocate (phys address)
1372  * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1373  * @exact_nid: control the allocation fall back to other nodes
1374  *
1375  * The allocation is performed from memory region limited by
1376  * memblock.current_limit if @end == %MEMBLOCK_ALLOC_ACCESSIBLE.
1377  *
1378  * If the specified node can not hold the requested memory and @exact_nid
1379  * is false, the allocation falls back to any node in the system.
1380  *
1381  * For systems with memory mirroring, the allocation is attempted first
1382  * from the regions with mirroring enabled and then retried from any
1383  * memory region.
1384  *
1385  * In addition, function using kmemleak_alloc_phys for allocated boot
1386  * memory block, it is never reported as leaks.
1387  *
1388  * Return:
1389  * Physical address of allocated memory block on success, %0 on failure.
1390  */
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)1391 phys_addr_t __init memblock_alloc_range_nid(phys_addr_t size,
1392 					phys_addr_t align, phys_addr_t start,
1393 					phys_addr_t end, int nid,
1394 					bool exact_nid)
1395 {
1396 	enum memblock_flags flags = choose_memblock_flags();
1397 	phys_addr_t found;
1398 
1399 	if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
1400 		nid = NUMA_NO_NODE;
1401 
1402 	if (!align) {
1403 		/* Can't use WARNs this early in boot on powerpc */
1404 		dump_stack();
1405 		align = SMP_CACHE_BYTES;
1406 	}
1407 
1408 again:
1409 	found = memblock_find_in_range_node(size, align, start, end, nid,
1410 					    flags);
1411 	if (found && !memblock_reserve(found, size))
1412 		goto done;
1413 
1414 	if (nid != NUMA_NO_NODE && !exact_nid) {
1415 		found = memblock_find_in_range_node(size, align, start,
1416 						    end, NUMA_NO_NODE,
1417 						    flags);
1418 		if (found && !memblock_reserve(found, size))
1419 			goto done;
1420 	}
1421 
1422 	if (flags & MEMBLOCK_MIRROR) {
1423 		flags &= ~MEMBLOCK_MIRROR;
1424 		pr_warn_ratelimited("Could not allocate %pap bytes of mirrored memory\n",
1425 			&size);
1426 		goto again;
1427 	}
1428 
1429 	return 0;
1430 
1431 done:
1432 	/*
1433 	 * Skip kmemleak for those places like kasan_init() and
1434 	 * early_pgtable_alloc() due to high volume.
1435 	 */
1436 	if (end != MEMBLOCK_ALLOC_NOLEAKTRACE)
1437 		/*
1438 		 * Memblock allocated blocks are never reported as
1439 		 * leaks. This is because many of these blocks are
1440 		 * only referred via the physical address which is
1441 		 * not looked up by kmemleak.
1442 		 */
1443 		kmemleak_alloc_phys(found, size, 0);
1444 
1445 	/*
1446 	 * Some Virtual Machine platforms, such as Intel TDX or AMD SEV-SNP,
1447 	 * require memory to be accepted before it can be used by the
1448 	 * guest.
1449 	 *
1450 	 * Accept the memory of the allocated buffer.
1451 	 */
1452 	accept_memory(found, found + size);
1453 
1454 	return found;
1455 }
1456 
1457 /**
1458  * memblock_phys_alloc_range - allocate a memory block inside specified range
1459  * @size: size of memory block to be allocated in bytes
1460  * @align: alignment of the region and block's size
1461  * @start: the lower bound of the memory region to allocate (physical address)
1462  * @end: the upper bound of the memory region to allocate (physical address)
1463  *
1464  * Allocate @size bytes in the between @start and @end.
1465  *
1466  * Return: physical address of the allocated memory block on success,
1467  * %0 on failure.
1468  */
memblock_phys_alloc_range(phys_addr_t size,phys_addr_t align,phys_addr_t start,phys_addr_t end)1469 phys_addr_t __init memblock_phys_alloc_range(phys_addr_t size,
1470 					     phys_addr_t align,
1471 					     phys_addr_t start,
1472 					     phys_addr_t end)
1473 {
1474 	memblock_dbg("%s: %llu bytes align=0x%llx from=%pa max_addr=%pa %pS\n",
1475 		     __func__, (u64)size, (u64)align, &start, &end,
1476 		     (void *)_RET_IP_);
1477 	return memblock_alloc_range_nid(size, align, start, end, NUMA_NO_NODE,
1478 					false);
1479 }
1480 
1481 /**
1482  * memblock_phys_alloc_try_nid - allocate a memory block from specified NUMA node
1483  * @size: size of memory block to be allocated in bytes
1484  * @align: alignment of the region and block's size
1485  * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1486  *
1487  * Allocates memory block from the specified NUMA node. If the node
1488  * has no available memory, attempts to allocated from any node in the
1489  * system.
1490  *
1491  * Return: physical address of the allocated memory block on success,
1492  * %0 on failure.
1493  */
memblock_phys_alloc_try_nid(phys_addr_t size,phys_addr_t align,int nid)1494 phys_addr_t __init memblock_phys_alloc_try_nid(phys_addr_t size, phys_addr_t align, int nid)
1495 {
1496 	return memblock_alloc_range_nid(size, align, 0,
1497 					MEMBLOCK_ALLOC_ACCESSIBLE, nid, false);
1498 }
1499 
1500 /**
1501  * memblock_alloc_internal - allocate boot memory block
1502  * @size: size of memory block to be allocated in bytes
1503  * @align: alignment of the region and block's size
1504  * @min_addr: the lower bound of the memory region to allocate (phys address)
1505  * @max_addr: the upper bound of the memory region to allocate (phys address)
1506  * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1507  * @exact_nid: control the allocation fall back to other nodes
1508  *
1509  * Allocates memory block using memblock_alloc_range_nid() and
1510  * converts the returned physical address to virtual.
1511  *
1512  * The @min_addr limit is dropped if it can not be satisfied and the allocation
1513  * will fall back to memory below @min_addr. Other constraints, such
1514  * as node and mirrored memory will be handled again in
1515  * memblock_alloc_range_nid().
1516  *
1517  * Return:
1518  * Virtual address of allocated memory block on success, NULL on failure.
1519  */
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)1520 static void * __init memblock_alloc_internal(
1521 				phys_addr_t size, phys_addr_t align,
1522 				phys_addr_t min_addr, phys_addr_t max_addr,
1523 				int nid, bool exact_nid)
1524 {
1525 	phys_addr_t alloc;
1526 
1527 	/*
1528 	 * Detect any accidental use of these APIs after slab is ready, as at
1529 	 * this moment memblock may be deinitialized already and its
1530 	 * internal data may be destroyed (after execution of memblock_free_all)
1531 	 */
1532 	if (WARN_ON_ONCE(slab_is_available()))
1533 		return kzalloc_node(size, GFP_NOWAIT, nid);
1534 
1535 	if (max_addr > memblock.current_limit)
1536 		max_addr = memblock.current_limit;
1537 
1538 	alloc = memblock_alloc_range_nid(size, align, min_addr, max_addr, nid,
1539 					exact_nid);
1540 
1541 	/* retry allocation without lower limit */
1542 	if (!alloc && min_addr)
1543 		alloc = memblock_alloc_range_nid(size, align, 0, max_addr, nid,
1544 						exact_nid);
1545 
1546 	if (!alloc)
1547 		return NULL;
1548 
1549 	return phys_to_virt(alloc);
1550 }
1551 
1552 /**
1553  * memblock_alloc_exact_nid_raw - allocate boot memory block on the exact node
1554  * without zeroing memory
1555  * @size: size of memory block to be allocated in bytes
1556  * @align: alignment of the region and block's size
1557  * @min_addr: the lower bound of the memory region from where the allocation
1558  *	  is preferred (phys address)
1559  * @max_addr: the upper bound of the memory region from where the allocation
1560  *	      is preferred (phys address), or %MEMBLOCK_ALLOC_ACCESSIBLE to
1561  *	      allocate only from memory limited by memblock.current_limit value
1562  * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1563  *
1564  * Public function, provides additional debug information (including caller
1565  * info), if enabled. Does not zero allocated memory.
1566  *
1567  * Return:
1568  * Virtual address of allocated memory block on success, NULL on failure.
1569  */
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)1570 void * __init memblock_alloc_exact_nid_raw(
1571 			phys_addr_t size, phys_addr_t align,
1572 			phys_addr_t min_addr, phys_addr_t max_addr,
1573 			int nid)
1574 {
1575 	memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=%pa max_addr=%pa %pS\n",
1576 		     __func__, (u64)size, (u64)align, nid, &min_addr,
1577 		     &max_addr, (void *)_RET_IP_);
1578 
1579 	return memblock_alloc_internal(size, align, min_addr, max_addr, nid,
1580 				       true);
1581 }
1582 
1583 /**
1584  * memblock_alloc_try_nid_raw - allocate boot memory block without zeroing
1585  * memory and without panicking
1586  * @size: size of memory block to be allocated in bytes
1587  * @align: alignment of the region and block's size
1588  * @min_addr: the lower bound of the memory region from where the allocation
1589  *	  is preferred (phys address)
1590  * @max_addr: the upper bound of the memory region from where the allocation
1591  *	      is preferred (phys address), or %MEMBLOCK_ALLOC_ACCESSIBLE to
1592  *	      allocate only from memory limited by memblock.current_limit value
1593  * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1594  *
1595  * Public function, provides additional debug information (including caller
1596  * info), if enabled. Does not zero allocated memory, does not panic if request
1597  * cannot be satisfied.
1598  *
1599  * Return:
1600  * Virtual address of allocated memory block on success, NULL on failure.
1601  */
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)1602 void * __init memblock_alloc_try_nid_raw(
1603 			phys_addr_t size, phys_addr_t align,
1604 			phys_addr_t min_addr, phys_addr_t max_addr,
1605 			int nid)
1606 {
1607 	memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=%pa max_addr=%pa %pS\n",
1608 		     __func__, (u64)size, (u64)align, nid, &min_addr,
1609 		     &max_addr, (void *)_RET_IP_);
1610 
1611 	return memblock_alloc_internal(size, align, min_addr, max_addr, nid,
1612 				       false);
1613 }
1614 
1615 /**
1616  * memblock_alloc_try_nid - allocate boot memory block
1617  * @size: size of memory block to be allocated in bytes
1618  * @align: alignment of the region and block's size
1619  * @min_addr: the lower bound of the memory region from where the allocation
1620  *	  is preferred (phys address)
1621  * @max_addr: the upper bound of the memory region from where the allocation
1622  *	      is preferred (phys address), or %MEMBLOCK_ALLOC_ACCESSIBLE to
1623  *	      allocate only from memory limited by memblock.current_limit value
1624  * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1625  *
1626  * Public function, provides additional debug information (including caller
1627  * info), if enabled. This function zeroes the allocated memory.
1628  *
1629  * Return:
1630  * Virtual address of allocated memory block on success, NULL on failure.
1631  */
memblock_alloc_try_nid(phys_addr_t size,phys_addr_t align,phys_addr_t min_addr,phys_addr_t max_addr,int nid)1632 void * __init memblock_alloc_try_nid(
1633 			phys_addr_t size, phys_addr_t align,
1634 			phys_addr_t min_addr, phys_addr_t max_addr,
1635 			int nid)
1636 {
1637 	void *ptr;
1638 
1639 	memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=%pa max_addr=%pa %pS\n",
1640 		     __func__, (u64)size, (u64)align, nid, &min_addr,
1641 		     &max_addr, (void *)_RET_IP_);
1642 	ptr = memblock_alloc_internal(size, align,
1643 					   min_addr, max_addr, nid, false);
1644 	if (ptr)
1645 		memset(ptr, 0, size);
1646 
1647 	return ptr;
1648 }
1649 
1650 /**
1651  * memblock_free_late - free pages directly to buddy allocator
1652  * @base: phys starting address of the  boot memory block
1653  * @size: size of the boot memory block in bytes
1654  *
1655  * This is only useful when the memblock allocator has already been torn
1656  * down, but we are still initializing the system.  Pages are released directly
1657  * to the buddy allocator.
1658  */
memblock_free_late(phys_addr_t base,phys_addr_t size)1659 void __init memblock_free_late(phys_addr_t base, phys_addr_t size)
1660 {
1661 	phys_addr_t cursor, end;
1662 
1663 	end = base + size - 1;
1664 	memblock_dbg("%s: [%pa-%pa] %pS\n",
1665 		     __func__, &base, &end, (void *)_RET_IP_);
1666 	kmemleak_free_part_phys(base, size);
1667 	cursor = PFN_UP(base);
1668 	end = PFN_DOWN(base + size);
1669 
1670 	for (; cursor < end; cursor++) {
1671 		memblock_free_pages(pfn_to_page(cursor), cursor, 0);
1672 		totalram_pages_inc();
1673 	}
1674 }
1675 
1676 /*
1677  * Remaining API functions
1678  */
1679 
memblock_phys_mem_size(void)1680 phys_addr_t __init_memblock memblock_phys_mem_size(void)
1681 {
1682 	return memblock.memory.total_size;
1683 }
1684 
memblock_reserved_size(void)1685 phys_addr_t __init_memblock memblock_reserved_size(void)
1686 {
1687 	return memblock.reserved.total_size;
1688 }
1689 
1690 /* lowest address */
memblock_start_of_DRAM(void)1691 phys_addr_t __init_memblock memblock_start_of_DRAM(void)
1692 {
1693 	return memblock.memory.regions[0].base;
1694 }
1695 
memblock_end_of_DRAM(void)1696 phys_addr_t __init_memblock memblock_end_of_DRAM(void)
1697 {
1698 	int idx = memblock.memory.cnt - 1;
1699 
1700 	return (memblock.memory.regions[idx].base + memblock.memory.regions[idx].size);
1701 }
1702 
__find_max_addr(phys_addr_t limit)1703 static phys_addr_t __init_memblock __find_max_addr(phys_addr_t limit)
1704 {
1705 	phys_addr_t max_addr = PHYS_ADDR_MAX;
1706 	struct memblock_region *r;
1707 
1708 	/*
1709 	 * translate the memory @limit size into the max address within one of
1710 	 * the memory memblock regions, if the @limit exceeds the total size
1711 	 * of those regions, max_addr will keep original value PHYS_ADDR_MAX
1712 	 */
1713 	for_each_mem_region(r) {
1714 		if (limit <= r->size) {
1715 			max_addr = r->base + limit;
1716 			break;
1717 		}
1718 		limit -= r->size;
1719 	}
1720 
1721 	return max_addr;
1722 }
1723 
memblock_enforce_memory_limit(phys_addr_t limit)1724 void __init memblock_enforce_memory_limit(phys_addr_t limit)
1725 {
1726 	phys_addr_t max_addr;
1727 
1728 	if (!limit)
1729 		return;
1730 
1731 	max_addr = __find_max_addr(limit);
1732 
1733 	/* @limit exceeds the total size of the memory, do nothing */
1734 	if (max_addr == PHYS_ADDR_MAX)
1735 		return;
1736 
1737 	/* truncate both memory and reserved regions */
1738 	memblock_remove_range(&memblock.memory, max_addr,
1739 			      PHYS_ADDR_MAX);
1740 	memblock_remove_range(&memblock.reserved, max_addr,
1741 			      PHYS_ADDR_MAX);
1742 }
1743 
memblock_cap_memory_range(phys_addr_t base,phys_addr_t size)1744 void __init memblock_cap_memory_range(phys_addr_t base, phys_addr_t size)
1745 {
1746 	int start_rgn, end_rgn;
1747 	int i, ret;
1748 
1749 	if (!size)
1750 		return;
1751 
1752 	if (!memblock_memory->total_size) {
1753 		pr_warn("%s: No memory registered yet\n", __func__);
1754 		return;
1755 	}
1756 
1757 	ret = memblock_isolate_range(&memblock.memory, base, size,
1758 						&start_rgn, &end_rgn);
1759 	if (ret)
1760 		return;
1761 
1762 	/* remove all the MAP regions */
1763 	for (i = memblock.memory.cnt - 1; i >= end_rgn; i--)
1764 		if (!memblock_is_nomap(&memblock.memory.regions[i]))
1765 			memblock_remove_region(&memblock.memory, i);
1766 
1767 	for (i = start_rgn - 1; i >= 0; i--)
1768 		if (!memblock_is_nomap(&memblock.memory.regions[i]))
1769 			memblock_remove_region(&memblock.memory, i);
1770 
1771 	/* truncate the reserved regions */
1772 	memblock_remove_range(&memblock.reserved, 0, base);
1773 	memblock_remove_range(&memblock.reserved,
1774 			base + size, PHYS_ADDR_MAX);
1775 }
1776 
memblock_mem_limit_remove_map(phys_addr_t limit)1777 void __init memblock_mem_limit_remove_map(phys_addr_t limit)
1778 {
1779 	phys_addr_t max_addr;
1780 
1781 	if (!limit)
1782 		return;
1783 
1784 	max_addr = __find_max_addr(limit);
1785 
1786 	/* @limit exceeds the total size of the memory, do nothing */
1787 	if (max_addr == PHYS_ADDR_MAX)
1788 		return;
1789 
1790 	memblock_cap_memory_range(0, max_addr);
1791 }
1792 
memblock_search(struct memblock_type * type,phys_addr_t addr)1793 static int __init_memblock memblock_search(struct memblock_type *type, phys_addr_t addr)
1794 {
1795 	unsigned int left = 0, right = type->cnt;
1796 
1797 	do {
1798 		unsigned int mid = (right + left) / 2;
1799 
1800 		if (addr < type->regions[mid].base)
1801 			right = mid;
1802 		else if (addr >= (type->regions[mid].base +
1803 				  type->regions[mid].size))
1804 			left = mid + 1;
1805 		else
1806 			return mid;
1807 	} while (left < right);
1808 	return -1;
1809 }
1810 
memblock_is_reserved(phys_addr_t addr)1811 bool __init_memblock memblock_is_reserved(phys_addr_t addr)
1812 {
1813 	return memblock_search(&memblock.reserved, addr) != -1;
1814 }
1815 
memblock_is_memory(phys_addr_t addr)1816 bool __init_memblock memblock_is_memory(phys_addr_t addr)
1817 {
1818 	return memblock_search(&memblock.memory, addr) != -1;
1819 }
1820 
memblock_is_map_memory(phys_addr_t addr)1821 bool __init_memblock memblock_is_map_memory(phys_addr_t addr)
1822 {
1823 	int i = memblock_search(&memblock.memory, addr);
1824 
1825 	if (i == -1)
1826 		return false;
1827 	return !memblock_is_nomap(&memblock.memory.regions[i]);
1828 }
1829 
memblock_search_pfn_nid(unsigned long pfn,unsigned long * start_pfn,unsigned long * end_pfn)1830 int __init_memblock memblock_search_pfn_nid(unsigned long pfn,
1831 			 unsigned long *start_pfn, unsigned long *end_pfn)
1832 {
1833 	struct memblock_type *type = &memblock.memory;
1834 	int mid = memblock_search(type, PFN_PHYS(pfn));
1835 
1836 	if (mid == -1)
1837 		return -1;
1838 
1839 	*start_pfn = PFN_DOWN(type->regions[mid].base);
1840 	*end_pfn = PFN_DOWN(type->regions[mid].base + type->regions[mid].size);
1841 
1842 	return memblock_get_region_node(&type->regions[mid]);
1843 }
1844 
1845 /**
1846  * memblock_is_region_memory - check if a region is a subset of memory
1847  * @base: base of region to check
1848  * @size: size of region to check
1849  *
1850  * Check if the region [@base, @base + @size) is a subset of a memory block.
1851  *
1852  * Return:
1853  * 0 if false, non-zero if true
1854  */
memblock_is_region_memory(phys_addr_t base,phys_addr_t size)1855 bool __init_memblock memblock_is_region_memory(phys_addr_t base, phys_addr_t size)
1856 {
1857 	int idx = memblock_search(&memblock.memory, base);
1858 	phys_addr_t end = base + memblock_cap_size(base, &size);
1859 
1860 	if (idx == -1)
1861 		return false;
1862 	return (memblock.memory.regions[idx].base +
1863 		 memblock.memory.regions[idx].size) >= end;
1864 }
1865 
1866 /**
1867  * memblock_is_region_reserved - check if a region intersects reserved memory
1868  * @base: base of region to check
1869  * @size: size of region to check
1870  *
1871  * Check if the region [@base, @base + @size) intersects a reserved
1872  * memory block.
1873  *
1874  * Return:
1875  * True if they intersect, false if not.
1876  */
memblock_is_region_reserved(phys_addr_t base,phys_addr_t size)1877 bool __init_memblock memblock_is_region_reserved(phys_addr_t base, phys_addr_t size)
1878 {
1879 	return memblock_overlaps_region(&memblock.reserved, base, size);
1880 }
1881 
memblock_trim_memory(phys_addr_t align)1882 void __init_memblock memblock_trim_memory(phys_addr_t align)
1883 {
1884 	phys_addr_t start, end, orig_start, orig_end;
1885 	struct memblock_region *r;
1886 
1887 	for_each_mem_region(r) {
1888 		orig_start = r->base;
1889 		orig_end = r->base + r->size;
1890 		start = round_up(orig_start, align);
1891 		end = round_down(orig_end, align);
1892 
1893 		if (start == orig_start && end == orig_end)
1894 			continue;
1895 
1896 		if (start < end) {
1897 			r->base = start;
1898 			r->size = end - start;
1899 		} else {
1900 			memblock_remove_region(&memblock.memory,
1901 					       r - memblock.memory.regions);
1902 			r--;
1903 		}
1904 	}
1905 }
1906 
memblock_set_current_limit(phys_addr_t limit)1907 void __init_memblock memblock_set_current_limit(phys_addr_t limit)
1908 {
1909 	memblock.current_limit = limit;
1910 }
1911 
memblock_get_current_limit(void)1912 phys_addr_t __init_memblock memblock_get_current_limit(void)
1913 {
1914 	return memblock.current_limit;
1915 }
1916 
memblock_dump(struct memblock_type * type)1917 static void __init_memblock memblock_dump(struct memblock_type *type)
1918 {
1919 	phys_addr_t base, end, size;
1920 	enum memblock_flags flags;
1921 	int idx;
1922 	struct memblock_region *rgn;
1923 
1924 	pr_info(" %s.cnt  = 0x%lx\n", type->name, type->cnt);
1925 
1926 	for_each_memblock_type(idx, type, rgn) {
1927 		char nid_buf[32] = "";
1928 
1929 		base = rgn->base;
1930 		size = rgn->size;
1931 		end = base + size - 1;
1932 		flags = rgn->flags;
1933 #ifdef CONFIG_NUMA
1934 		if (memblock_get_region_node(rgn) != MAX_NUMNODES)
1935 			snprintf(nid_buf, sizeof(nid_buf), " on node %d",
1936 				 memblock_get_region_node(rgn));
1937 #endif
1938 		pr_info(" %s[%#x]\t[%pa-%pa], %pa bytes%s flags: %#x\n",
1939 			type->name, idx, &base, &end, &size, nid_buf, flags);
1940 	}
1941 }
1942 
__memblock_dump_all(void)1943 static void __init_memblock __memblock_dump_all(void)
1944 {
1945 	pr_info("MEMBLOCK configuration:\n");
1946 	pr_info(" memory size = %pa reserved size = %pa\n",
1947 		&memblock.memory.total_size,
1948 		&memblock.reserved.total_size);
1949 
1950 	memblock_dump(&memblock.memory);
1951 	memblock_dump(&memblock.reserved);
1952 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
1953 	memblock_dump(&physmem);
1954 #endif
1955 }
1956 
memblock_dump_all(void)1957 void __init_memblock memblock_dump_all(void)
1958 {
1959 	if (memblock_debug)
1960 		__memblock_dump_all();
1961 }
1962 
memblock_allow_resize(void)1963 void __init memblock_allow_resize(void)
1964 {
1965 	memblock_can_resize = 1;
1966 }
1967 
early_memblock(char * p)1968 static int __init early_memblock(char *p)
1969 {
1970 	if (p && strstr(p, "debug"))
1971 		memblock_debug = 1;
1972 	return 0;
1973 }
1974 early_param("memblock", early_memblock);
1975 
free_memmap(unsigned long start_pfn,unsigned long end_pfn)1976 static void __init free_memmap(unsigned long start_pfn, unsigned long end_pfn)
1977 {
1978 	struct page *start_pg, *end_pg;
1979 	phys_addr_t pg, pgend;
1980 
1981 	/*
1982 	 * Convert start_pfn/end_pfn to a struct page pointer.
1983 	 */
1984 	start_pg = pfn_to_page(start_pfn - 1) + 1;
1985 	end_pg = pfn_to_page(end_pfn - 1) + 1;
1986 
1987 	/*
1988 	 * Convert to physical addresses, and round start upwards and end
1989 	 * downwards.
1990 	 */
1991 	pg = PAGE_ALIGN(__pa(start_pg));
1992 	pgend = __pa(end_pg) & PAGE_MASK;
1993 
1994 	/*
1995 	 * If there are free pages between these, free the section of the
1996 	 * memmap array.
1997 	 */
1998 	if (pg < pgend)
1999 		memblock_phys_free(pg, pgend - pg);
2000 }
2001 
2002 /*
2003  * The mem_map array can get very big.  Free the unused area of the memory map.
2004  */
free_unused_memmap(void)2005 static void __init free_unused_memmap(void)
2006 {
2007 	unsigned long start, end, prev_end = 0;
2008 	int i;
2009 
2010 	if (!IS_ENABLED(CONFIG_HAVE_ARCH_PFN_VALID) ||
2011 	    IS_ENABLED(CONFIG_SPARSEMEM_VMEMMAP))
2012 		return;
2013 
2014 	/*
2015 	 * This relies on each bank being in address order.
2016 	 * The banks are sorted previously in bootmem_init().
2017 	 */
2018 	for_each_mem_pfn_range(i, MAX_NUMNODES, &start, &end, NULL) {
2019 #ifdef CONFIG_SPARSEMEM
2020 		/*
2021 		 * Take care not to free memmap entries that don't exist
2022 		 * due to SPARSEMEM sections which aren't present.
2023 		 */
2024 		start = min(start, ALIGN(prev_end, PAGES_PER_SECTION));
2025 #endif
2026 		/*
2027 		 * Align down here since many operations in VM subsystem
2028 		 * presume that there are no holes in the memory map inside
2029 		 * a pageblock
2030 		 */
2031 		start = pageblock_start_pfn(start);
2032 
2033 		/*
2034 		 * If we had a previous bank, and there is a space
2035 		 * between the current bank and the previous, free it.
2036 		 */
2037 		if (prev_end && prev_end < start)
2038 			free_memmap(prev_end, start);
2039 
2040 		/*
2041 		 * Align up here since many operations in VM subsystem
2042 		 * presume that there are no holes in the memory map inside
2043 		 * a pageblock
2044 		 */
2045 		prev_end = pageblock_align(end);
2046 	}
2047 
2048 #ifdef CONFIG_SPARSEMEM
2049 	if (!IS_ALIGNED(prev_end, PAGES_PER_SECTION)) {
2050 		prev_end = pageblock_align(end);
2051 		free_memmap(prev_end, ALIGN(prev_end, PAGES_PER_SECTION));
2052 	}
2053 #endif
2054 }
2055 
__free_pages_memory(unsigned long start,unsigned long end)2056 static void __init __free_pages_memory(unsigned long start, unsigned long end)
2057 {
2058 	int order;
2059 
2060 	while (start < end) {
2061 		/*
2062 		 * Free the pages in the largest chunks alignment allows.
2063 		 *
2064 		 * __ffs() behaviour is undefined for 0. start == 0 is
2065 		 * MAX_ORDER-aligned, set order to MAX_ORDER for the case.
2066 		 */
2067 		if (start)
2068 			order = min_t(int, MAX_ORDER, __ffs(start));
2069 		else
2070 			order = MAX_ORDER;
2071 
2072 		while (start + (1UL << order) > end)
2073 			order--;
2074 
2075 		memblock_free_pages(pfn_to_page(start), start, order);
2076 
2077 		start += (1UL << order);
2078 	}
2079 }
2080 
__free_memory_core(phys_addr_t start,phys_addr_t end)2081 static unsigned long __init __free_memory_core(phys_addr_t start,
2082 				 phys_addr_t end)
2083 {
2084 	unsigned long start_pfn = PFN_UP(start);
2085 	unsigned long end_pfn = min_t(unsigned long,
2086 				      PFN_DOWN(end), max_low_pfn);
2087 
2088 	if (start_pfn >= end_pfn)
2089 		return 0;
2090 
2091 	__free_pages_memory(start_pfn, end_pfn);
2092 
2093 	return end_pfn - start_pfn;
2094 }
2095 
memmap_init_reserved_pages(void)2096 static void __init memmap_init_reserved_pages(void)
2097 {
2098 	struct memblock_region *region;
2099 	phys_addr_t start, end;
2100 	int nid;
2101 
2102 	/*
2103 	 * set nid on all reserved pages and also treat struct
2104 	 * pages for the NOMAP regions as PageReserved
2105 	 */
2106 	for_each_mem_region(region) {
2107 		nid = memblock_get_region_node(region);
2108 		start = region->base;
2109 		end = start + region->size;
2110 
2111 		if (memblock_is_nomap(region))
2112 			reserve_bootmem_region(start, end, nid);
2113 
2114 		memblock_set_node(start, end, &memblock.reserved, nid);
2115 	}
2116 
2117 	/* initialize struct pages for the reserved regions */
2118 	for_each_reserved_mem_region(region) {
2119 		nid = memblock_get_region_node(region);
2120 		start = region->base;
2121 		end = start + region->size;
2122 
2123 		if (nid == NUMA_NO_NODE || nid >= MAX_NUMNODES)
2124 			nid = early_pfn_to_nid(PFN_DOWN(start));
2125 
2126 		reserve_bootmem_region(start, end, nid);
2127 	}
2128 }
2129 
free_low_memory_core_early(void)2130 static unsigned long __init free_low_memory_core_early(void)
2131 {
2132 	unsigned long count = 0;
2133 	phys_addr_t start, end;
2134 	u64 i;
2135 
2136 	memblock_clear_hotplug(0, -1);
2137 
2138 	memmap_init_reserved_pages();
2139 
2140 	/*
2141 	 * We need to use NUMA_NO_NODE instead of NODE_DATA(0)->node_id
2142 	 *  because in some case like Node0 doesn't have RAM installed
2143 	 *  low ram will be on Node1
2144 	 */
2145 	for_each_free_mem_range(i, NUMA_NO_NODE, MEMBLOCK_NONE, &start, &end,
2146 				NULL)
2147 		count += __free_memory_core(start, end);
2148 
2149 	return count;
2150 }
2151 
2152 static int reset_managed_pages_done __initdata;
2153 
reset_node_managed_pages(pg_data_t * pgdat)2154 static void __init reset_node_managed_pages(pg_data_t *pgdat)
2155 {
2156 	struct zone *z;
2157 
2158 	for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++)
2159 		atomic_long_set(&z->managed_pages, 0);
2160 }
2161 
reset_all_zones_managed_pages(void)2162 void __init reset_all_zones_managed_pages(void)
2163 {
2164 	struct pglist_data *pgdat;
2165 
2166 	if (reset_managed_pages_done)
2167 		return;
2168 
2169 	for_each_online_pgdat(pgdat)
2170 		reset_node_managed_pages(pgdat);
2171 
2172 	reset_managed_pages_done = 1;
2173 }
2174 
2175 /**
2176  * memblock_free_all - release free pages to the buddy allocator
2177  */
memblock_free_all(void)2178 void __init memblock_free_all(void)
2179 {
2180 	unsigned long pages;
2181 
2182 	free_unused_memmap();
2183 	reset_all_zones_managed_pages();
2184 
2185 	pages = free_low_memory_core_early();
2186 	totalram_pages_add(pages);
2187 }
2188 
2189 #if defined(CONFIG_DEBUG_FS) && defined(CONFIG_ARCH_KEEP_MEMBLOCK)
2190 static const char * const flagname[] = {
2191 	[ilog2(MEMBLOCK_HOTPLUG)] = "HOTPLUG",
2192 	[ilog2(MEMBLOCK_MIRROR)] = "MIRROR",
2193 	[ilog2(MEMBLOCK_NOMAP)] = "NOMAP",
2194 	[ilog2(MEMBLOCK_DRIVER_MANAGED)] = "DRV_MNG",
2195 };
2196 
memblock_debug_show(struct seq_file * m,void * private)2197 static int memblock_debug_show(struct seq_file *m, void *private)
2198 {
2199 	struct memblock_type *type = m->private;
2200 	struct memblock_region *reg;
2201 	int i, j, nid;
2202 	unsigned int count = ARRAY_SIZE(flagname);
2203 	phys_addr_t end;
2204 
2205 	for (i = 0; i < type->cnt; i++) {
2206 		reg = &type->regions[i];
2207 		end = reg->base + reg->size - 1;
2208 		nid = memblock_get_region_node(reg);
2209 
2210 		seq_printf(m, "%4d: ", i);
2211 		seq_printf(m, "%pa..%pa ", &reg->base, &end);
2212 		if (nid != MAX_NUMNODES)
2213 			seq_printf(m, "%4d ", nid);
2214 		else
2215 			seq_printf(m, "%4c ", 'x');
2216 		if (reg->flags) {
2217 			for (j = 0; j < count; j++) {
2218 				if (reg->flags & (1U << j)) {
2219 					seq_printf(m, "%s\n", flagname[j]);
2220 					break;
2221 				}
2222 			}
2223 			if (j == count)
2224 				seq_printf(m, "%s\n", "UNKNOWN");
2225 		} else {
2226 			seq_printf(m, "%s\n", "NONE");
2227 		}
2228 	}
2229 	return 0;
2230 }
2231 DEFINE_SHOW_ATTRIBUTE(memblock_debug);
2232 
memblock_init_debugfs(void)2233 static int __init memblock_init_debugfs(void)
2234 {
2235 	struct dentry *root = debugfs_create_dir("memblock", NULL);
2236 
2237 	debugfs_create_file("memory", 0444, root,
2238 			    &memblock.memory, &memblock_debug_fops);
2239 	debugfs_create_file("reserved", 0444, root,
2240 			    &memblock.reserved, &memblock_debug_fops);
2241 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
2242 	debugfs_create_file("physmem", 0444, root, &physmem,
2243 			    &memblock_debug_fops);
2244 #endif
2245 
2246 	return 0;
2247 }
2248 __initcall(memblock_init_debugfs);
2249 
2250 #endif /* CONFIG_DEBUG_FS */
2251