1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * Procedures for maintaining information about logical memory blocks.
5 * Peter Bergner, IBM Corp. June 2001.
6 * Copyright (C) 2001 Peter Bergner.
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>
20 #include <asm/sections.h>
25 #define INIT_MEMBLOCK_REGIONS 128
26 #define INIT_PHYSMEM_REGIONS 4
28 #ifndef INIT_MEMBLOCK_RESERVED_REGIONS
29 # define INIT_MEMBLOCK_RESERVED_REGIONS INIT_MEMBLOCK_REGIONS
33 * DOC: memblock overview
35 * Memblock is a method of managing memory regions during the early
36 * boot period when the usual kernel memory allocators are not up and
39 * Memblock views the system memory as collections of contiguous
40 * regions. There are several types of these collections:
42 * * ``memory`` - describes the physical memory available to the
43 * kernel; this may differ from the actual physical memory installed
44 * in the system, for instance when the memory is restricted with
45 * ``mem=`` command line parameter
46 * * ``reserved`` - describes the regions that were allocated
47 * * ``physmem`` - describes the actual physical memory available during
48 * boot regardless of the possible restrictions and memory hot(un)plug;
49 * the ``physmem`` type is only available on some architectures.
51 * Each region is represented by struct memblock_region that
52 * defines the region extents, its attributes and NUMA node id on NUMA
53 * systems. Every memory type is described by the struct memblock_type
54 * which contains an array of memory regions along with
55 * the allocator metadata. The "memory" and "reserved" types are nicely
56 * wrapped with struct memblock. This structure is statically
57 * initialized at build time. The region arrays are initially sized to
58 * %INIT_MEMBLOCK_REGIONS for "memory" and %INIT_MEMBLOCK_RESERVED_REGIONS
59 * for "reserved". The region array for "physmem" is initially sized to
60 * %INIT_PHYSMEM_REGIONS.
61 * The memblock_allow_resize() enables automatic resizing of the region
62 * arrays during addition of new regions. This feature should be used
63 * with care so that memory allocated for the region array will not
64 * overlap with areas that should be reserved, for example initrd.
66 * The early architecture setup should tell memblock what the physical
67 * memory layout is by using memblock_add() or memblock_add_node()
68 * functions. The first function does not assign the region to a NUMA
69 * node and it is appropriate for UMA systems. Yet, it is possible to
70 * use it on NUMA systems as well and assign the region to a NUMA node
71 * later in the setup process using memblock_set_node(). The
72 * memblock_add_node() performs such an assignment directly.
74 * Once memblock is setup the memory can be allocated using one of the
77 * * memblock_phys_alloc*() - these functions return the **physical**
78 * address of the allocated memory
79 * * memblock_alloc*() - these functions return the **virtual** address
80 * of the allocated memory.
82 * Note, that both API variants use implicit assumptions about allowed
83 * memory ranges and the fallback methods. Consult the documentation
84 * of memblock_alloc_internal() and memblock_alloc_range_nid()
85 * functions for more elaborate description.
87 * As the system boot progresses, the architecture specific mem_init()
88 * function frees all the memory to the buddy page allocator.
90 * Unless an architecture enables %CONFIG_ARCH_KEEP_MEMBLOCK, the
91 * memblock data structures (except "physmem") will be discarded after the
92 * system initialization completes.
95 #ifndef CONFIG_NEED_MULTIPLE_NODES
96 struct pglist_data __refdata contig_page_data;
97 EXPORT_SYMBOL(contig_page_data);
100 unsigned long max_low_pfn;
101 unsigned long min_low_pfn;
102 unsigned long max_pfn;
103 unsigned long long max_possible_pfn;
105 static struct memblock_region memblock_memory_init_regions[INIT_MEMBLOCK_REGIONS] __initdata_memblock;
106 static struct memblock_region memblock_reserved_init_regions[INIT_MEMBLOCK_RESERVED_REGIONS] __initdata_memblock;
107 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
108 static struct memblock_region memblock_physmem_init_regions[INIT_PHYSMEM_REGIONS];
111 struct memblock memblock __initdata_memblock = {
112 .memory.regions = memblock_memory_init_regions,
113 .memory.cnt = 1, /* empty dummy entry */
114 .memory.max = INIT_MEMBLOCK_REGIONS,
115 .memory.name = "memory",
117 .reserved.regions = memblock_reserved_init_regions,
118 .reserved.cnt = 1, /* empty dummy entry */
119 .reserved.max = INIT_MEMBLOCK_RESERVED_REGIONS,
120 .reserved.name = "reserved",
123 .current_limit = MEMBLOCK_ALLOC_ANYWHERE,
126 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
127 struct memblock_type physmem = {
128 .regions = memblock_physmem_init_regions,
129 .cnt = 1, /* empty dummy entry */
130 .max = INIT_PHYSMEM_REGIONS,
136 * keep a pointer to &memblock.memory in the text section to use it in
137 * __next_mem_range() and its helpers.
138 * For architectures that do not keep memblock data after init, this
139 * pointer will be reset to NULL at memblock_discard()
141 static __refdata struct memblock_type *memblock_memory = &memblock.memory;
143 #define for_each_memblock_type(i, memblock_type, rgn) \
144 for (i = 0, rgn = &memblock_type->regions[0]; \
145 i < memblock_type->cnt; \
146 i++, rgn = &memblock_type->regions[i])
148 #define memblock_dbg(fmt, ...) \
150 if (memblock_debug) \
151 pr_info(fmt, ##__VA_ARGS__); \
154 static int memblock_debug __initdata_memblock;
155 static bool system_has_some_mirror __initdata_memblock = false;
156 static int memblock_can_resize __initdata_memblock;
157 static int memblock_memory_in_slab __initdata_memblock = 0;
158 static int memblock_reserved_in_slab __initdata_memblock = 0;
160 static enum memblock_flags __init_memblock choose_memblock_flags(void)
162 return system_has_some_mirror ? MEMBLOCK_MIRROR : MEMBLOCK_NONE;
165 /* adjust *@size so that (@base + *@size) doesn't overflow, return new size */
166 static inline phys_addr_t memblock_cap_size(phys_addr_t base, phys_addr_t *size)
168 return *size = min(*size, PHYS_ADDR_MAX - base);
172 * Address comparison utilities
174 static unsigned long __init_memblock memblock_addrs_overlap(phys_addr_t base1, phys_addr_t size1,
175 phys_addr_t base2, phys_addr_t size2)
177 return ((base1 < (base2 + size2)) && (base2 < (base1 + size1)));
180 bool __init_memblock memblock_overlaps_region(struct memblock_type *type,
181 phys_addr_t base, phys_addr_t size)
185 for (i = 0; i < type->cnt; i++)
186 if (memblock_addrs_overlap(base, size, type->regions[i].base,
187 type->regions[i].size))
189 return i < type->cnt;
193 * __memblock_find_range_bottom_up - find free area utility in bottom-up
194 * @start: start of candidate range
195 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_ANYWHERE or
196 * %MEMBLOCK_ALLOC_ACCESSIBLE
197 * @size: size of free area to find
198 * @align: alignment of free area to find
199 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
200 * @flags: pick from blocks based on memory attributes
202 * Utility called from memblock_find_in_range_node(), find free area bottom-up.
205 * Found address on success, 0 on failure.
207 static phys_addr_t __init_memblock
208 __memblock_find_range_bottom_up(phys_addr_t start, phys_addr_t end,
209 phys_addr_t size, phys_addr_t align, int nid,
210 enum memblock_flags flags)
212 phys_addr_t this_start, this_end, cand;
215 for_each_free_mem_range(i, nid, flags, &this_start, &this_end, NULL) {
216 this_start = clamp(this_start, start, end);
217 this_end = clamp(this_end, start, end);
219 cand = round_up(this_start, align);
220 if (cand < this_end && this_end - cand >= size)
228 * __memblock_find_range_top_down - find free area utility, in top-down
229 * @start: start of candidate range
230 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_ANYWHERE or
231 * %MEMBLOCK_ALLOC_ACCESSIBLE
232 * @size: size of free area to find
233 * @align: alignment of free area to find
234 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
235 * @flags: pick from blocks based on memory attributes
237 * Utility called from memblock_find_in_range_node(), find free area top-down.
240 * Found address on success, 0 on failure.
242 static phys_addr_t __init_memblock
243 __memblock_find_range_top_down(phys_addr_t start, phys_addr_t end,
244 phys_addr_t size, phys_addr_t align, int nid,
245 enum memblock_flags flags)
247 phys_addr_t this_start, this_end, cand;
250 for_each_free_mem_range_reverse(i, nid, flags, &this_start, &this_end,
252 this_start = clamp(this_start, start, end);
253 this_end = clamp(this_end, start, end);
258 cand = round_down(this_end - size, align);
259 if (cand >= this_start)
267 * memblock_find_in_range_node - find free area in given range and node
268 * @size: size of free area to find
269 * @align: alignment of free area to find
270 * @start: start of candidate range
271 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_ANYWHERE or
272 * %MEMBLOCK_ALLOC_ACCESSIBLE
273 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
274 * @flags: pick from blocks based on memory attributes
276 * Find @size free area aligned to @align in the specified range and node.
279 * Found address on success, 0 on failure.
281 static phys_addr_t __init_memblock memblock_find_in_range_node(phys_addr_t size,
282 phys_addr_t align, phys_addr_t start,
283 phys_addr_t end, int nid,
284 enum memblock_flags flags)
287 if (end == MEMBLOCK_ALLOC_ACCESSIBLE ||
288 end == MEMBLOCK_ALLOC_KASAN)
289 end = memblock.current_limit;
291 /* avoid allocating the first page */
292 start = max_t(phys_addr_t, start, PAGE_SIZE);
293 end = max(start, end);
295 if (memblock_bottom_up())
296 return __memblock_find_range_bottom_up(start, end, size, align,
299 return __memblock_find_range_top_down(start, end, size, align,
304 * memblock_find_in_range - find free area in given range
305 * @start: start of candidate range
306 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_ANYWHERE or
307 * %MEMBLOCK_ALLOC_ACCESSIBLE
308 * @size: size of free area to find
309 * @align: alignment of free area to find
311 * Find @size free area aligned to @align in the specified range.
314 * Found address on success, 0 on failure.
316 phys_addr_t __init_memblock memblock_find_in_range(phys_addr_t start,
317 phys_addr_t end, phys_addr_t size,
321 enum memblock_flags flags = choose_memblock_flags();
324 ret = memblock_find_in_range_node(size, align, start, end,
325 NUMA_NO_NODE, flags);
327 if (!ret && (flags & MEMBLOCK_MIRROR)) {
328 pr_warn("Could not allocate %pap bytes of mirrored memory\n",
330 flags &= ~MEMBLOCK_MIRROR;
337 static void __init_memblock memblock_remove_region(struct memblock_type *type, unsigned long r)
339 type->total_size -= type->regions[r].size;
340 memmove(&type->regions[r], &type->regions[r + 1],
341 (type->cnt - (r + 1)) * sizeof(type->regions[r]));
344 /* Special case for empty arrays */
345 if (type->cnt == 0) {
346 WARN_ON(type->total_size != 0);
348 type->regions[0].base = 0;
349 type->regions[0].size = 0;
350 type->regions[0].flags = 0;
351 memblock_set_region_node(&type->regions[0], MAX_NUMNODES);
355 #ifndef CONFIG_ARCH_KEEP_MEMBLOCK
357 * memblock_discard - discard memory and reserved arrays if they were allocated
359 void __init memblock_discard(void)
361 phys_addr_t addr, size;
363 if (memblock.reserved.regions != memblock_reserved_init_regions) {
364 addr = __pa(memblock.reserved.regions);
365 size = PAGE_ALIGN(sizeof(struct memblock_region) *
366 memblock.reserved.max);
367 __memblock_free_late(addr, size);
370 if (memblock.memory.regions != memblock_memory_init_regions) {
371 addr = __pa(memblock.memory.regions);
372 size = PAGE_ALIGN(sizeof(struct memblock_region) *
373 memblock.memory.max);
374 __memblock_free_late(addr, size);
377 memblock_memory = NULL;
382 * memblock_double_array - double the size of the memblock regions array
383 * @type: memblock type of the regions array being doubled
384 * @new_area_start: starting address of memory range to avoid overlap with
385 * @new_area_size: size of memory range to avoid overlap with
387 * Double the size of the @type regions array. If memblock is being used to
388 * allocate memory for a new reserved regions array and there is a previously
389 * allocated memory range [@new_area_start, @new_area_start + @new_area_size]
390 * waiting to be reserved, ensure the memory used by the new array does
394 * 0 on success, -1 on failure.
396 static int __init_memblock memblock_double_array(struct memblock_type *type,
397 phys_addr_t new_area_start,
398 phys_addr_t new_area_size)
400 struct memblock_region *new_array, *old_array;
401 phys_addr_t old_alloc_size, new_alloc_size;
402 phys_addr_t old_size, new_size, addr, new_end;
403 int use_slab = slab_is_available();
406 /* We don't allow resizing until we know about the reserved regions
407 * of memory that aren't suitable for allocation
409 if (!memblock_can_resize)
412 /* Calculate new doubled size */
413 old_size = type->max * sizeof(struct memblock_region);
414 new_size = old_size << 1;
416 * We need to allocated new one align to PAGE_SIZE,
417 * so we can free them completely later.
419 old_alloc_size = PAGE_ALIGN(old_size);
420 new_alloc_size = PAGE_ALIGN(new_size);
422 /* Retrieve the slab flag */
423 if (type == &memblock.memory)
424 in_slab = &memblock_memory_in_slab;
426 in_slab = &memblock_reserved_in_slab;
428 /* Try to find some space for it */
430 new_array = kmalloc(new_size, GFP_KERNEL);
431 addr = new_array ? __pa(new_array) : 0;
433 /* only exclude range when trying to double reserved.regions */
434 if (type != &memblock.reserved)
435 new_area_start = new_area_size = 0;
437 addr = memblock_find_in_range(new_area_start + new_area_size,
438 memblock.current_limit,
439 new_alloc_size, PAGE_SIZE);
440 if (!addr && new_area_size)
441 addr = memblock_find_in_range(0,
442 min(new_area_start, memblock.current_limit),
443 new_alloc_size, PAGE_SIZE);
445 new_array = addr ? __va(addr) : NULL;
448 pr_err("memblock: Failed to double %s array from %ld to %ld entries !\n",
449 type->name, type->max, type->max * 2);
453 new_end = addr + new_size - 1;
454 memblock_dbg("memblock: %s is doubled to %ld at [%pa-%pa]",
455 type->name, type->max * 2, &addr, &new_end);
458 * Found space, we now need to move the array over before we add the
459 * reserved region since it may be our reserved array itself that is
462 memcpy(new_array, type->regions, old_size);
463 memset(new_array + type->max, 0, old_size);
464 old_array = type->regions;
465 type->regions = new_array;
468 /* Free old array. We needn't free it if the array is the static one */
471 else if (old_array != memblock_memory_init_regions &&
472 old_array != memblock_reserved_init_regions)
473 memblock_free(__pa(old_array), old_alloc_size);
476 * Reserve the new array if that comes from the memblock. Otherwise, we
480 BUG_ON(memblock_reserve(addr, new_alloc_size));
482 /* Update slab flag */
489 * memblock_merge_regions - merge neighboring compatible regions
490 * @type: memblock type to scan
492 * Scan @type and merge neighboring compatible regions.
494 static void __init_memblock memblock_merge_regions(struct memblock_type *type)
498 /* cnt never goes below 1 */
499 while (i < type->cnt - 1) {
500 struct memblock_region *this = &type->regions[i];
501 struct memblock_region *next = &type->regions[i + 1];
503 if (this->base + this->size != next->base ||
504 memblock_get_region_node(this) !=
505 memblock_get_region_node(next) ||
506 this->flags != next->flags) {
507 BUG_ON(this->base + this->size > next->base);
512 this->size += next->size;
513 /* move forward from next + 1, index of which is i + 2 */
514 memmove(next, next + 1, (type->cnt - (i + 2)) * sizeof(*next));
520 * memblock_insert_region - insert new memblock region
521 * @type: memblock type to insert into
522 * @idx: index for the insertion point
523 * @base: base address of the new region
524 * @size: size of the new region
525 * @nid: node id of the new region
526 * @flags: flags of the new region
528 * Insert new memblock region [@base, @base + @size) into @type at @idx.
529 * @type must already have extra room to accommodate the new region.
531 static void __init_memblock memblock_insert_region(struct memblock_type *type,
532 int idx, phys_addr_t base,
535 enum memblock_flags flags)
537 struct memblock_region *rgn = &type->regions[idx];
539 BUG_ON(type->cnt >= type->max);
540 memmove(rgn + 1, rgn, (type->cnt - idx) * sizeof(*rgn));
544 memblock_set_region_node(rgn, nid);
546 type->total_size += size;
550 * memblock_add_range - add new memblock region
551 * @type: memblock type to add new region into
552 * @base: base address of the new region
553 * @size: size of the new region
554 * @nid: nid of the new region
555 * @flags: flags of the new region
557 * Add new memblock region [@base, @base + @size) into @type. The new region
558 * is allowed to overlap with existing ones - overlaps don't affect already
559 * existing regions. @type is guaranteed to be minimal (all neighbouring
560 * compatible regions are merged) after the addition.
563 * 0 on success, -errno on failure.
565 static int __init_memblock memblock_add_range(struct memblock_type *type,
566 phys_addr_t base, phys_addr_t size,
567 int nid, enum memblock_flags flags)
570 phys_addr_t obase = base;
571 phys_addr_t end = base + memblock_cap_size(base, &size);
573 struct memblock_region *rgn;
578 /* special case for empty array */
579 if (type->regions[0].size == 0) {
580 WARN_ON(type->cnt != 1 || type->total_size);
581 type->regions[0].base = base;
582 type->regions[0].size = size;
583 type->regions[0].flags = flags;
584 memblock_set_region_node(&type->regions[0], nid);
585 type->total_size = size;
590 * The following is executed twice. Once with %false @insert and
591 * then with %true. The first counts the number of regions needed
592 * to accommodate the new area. The second actually inserts them.
597 for_each_memblock_type(idx, type, rgn) {
598 phys_addr_t rbase = rgn->base;
599 phys_addr_t rend = rbase + rgn->size;
606 * @rgn overlaps. If it separates the lower part of new
607 * area, insert that portion.
610 #ifdef CONFIG_NEED_MULTIPLE_NODES
611 WARN_ON(nid != memblock_get_region_node(rgn));
613 WARN_ON(flags != rgn->flags);
616 memblock_insert_region(type, idx++, base,
620 /* area below @rend is dealt with, forget about it */
621 base = min(rend, end);
624 /* insert the remaining portion */
628 memblock_insert_region(type, idx, base, end - base,
636 * If this was the first round, resize array and repeat for actual
637 * insertions; otherwise, merge and return.
640 while (type->cnt + nr_new > type->max)
641 if (memblock_double_array(type, obase, size) < 0)
646 memblock_merge_regions(type);
652 * memblock_add_node - add new memblock region within a NUMA node
653 * @base: base address of the new region
654 * @size: size of the new region
655 * @nid: nid of the new region
657 * Add new memblock region [@base, @base + @size) to the "memory"
658 * type. See memblock_add_range() description for mode details
661 * 0 on success, -errno on failure.
663 int __init_memblock memblock_add_node(phys_addr_t base, phys_addr_t size,
666 return memblock_add_range(&memblock.memory, base, size, nid, 0);
670 * memblock_add - add new memblock region
671 * @base: base address of the new region
672 * @size: size of the new region
674 * Add new memblock region [@base, @base + @size) to the "memory"
675 * type. See memblock_add_range() description for mode details
678 * 0 on success, -errno on failure.
680 int __init_memblock memblock_add(phys_addr_t base, phys_addr_t size)
682 phys_addr_t end = base + size - 1;
684 memblock_dbg("%s: [%pa-%pa] %pS\n", __func__,
685 &base, &end, (void *)_RET_IP_);
687 return memblock_add_range(&memblock.memory, base, size, MAX_NUMNODES, 0);
691 * memblock_isolate_range - isolate given range into disjoint memblocks
692 * @type: memblock type to isolate range for
693 * @base: base of range to isolate
694 * @size: size of range to isolate
695 * @start_rgn: out parameter for the start of isolated region
696 * @end_rgn: out parameter for the end of isolated region
698 * Walk @type and ensure that regions don't cross the boundaries defined by
699 * [@base, @base + @size). Crossing regions are split at the boundaries,
700 * which may create at most two more regions. The index of the first
701 * region inside the range is returned in *@start_rgn and end in *@end_rgn.
704 * 0 on success, -errno on failure.
706 static int __init_memblock memblock_isolate_range(struct memblock_type *type,
707 phys_addr_t base, phys_addr_t size,
708 int *start_rgn, int *end_rgn)
710 phys_addr_t end = base + memblock_cap_size(base, &size);
712 struct memblock_region *rgn;
714 *start_rgn = *end_rgn = 0;
719 /* we'll create at most two more regions */
720 while (type->cnt + 2 > type->max)
721 if (memblock_double_array(type, base, size) < 0)
724 for_each_memblock_type(idx, type, rgn) {
725 phys_addr_t rbase = rgn->base;
726 phys_addr_t rend = rbase + rgn->size;
735 * @rgn intersects from below. Split and continue
736 * to process the next region - the new top half.
739 rgn->size -= base - rbase;
740 type->total_size -= base - rbase;
741 memblock_insert_region(type, idx, rbase, base - rbase,
742 memblock_get_region_node(rgn),
744 } else if (rend > end) {
746 * @rgn intersects from above. Split and redo the
747 * current region - the new bottom half.
750 rgn->size -= end - rbase;
751 type->total_size -= end - rbase;
752 memblock_insert_region(type, idx--, rbase, end - rbase,
753 memblock_get_region_node(rgn),
756 /* @rgn is fully contained, record it */
766 static int __init_memblock memblock_remove_range(struct memblock_type *type,
767 phys_addr_t base, phys_addr_t size)
769 int start_rgn, end_rgn;
772 ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
776 for (i = end_rgn - 1; i >= start_rgn; i--)
777 memblock_remove_region(type, i);
781 int __init_memblock memblock_remove(phys_addr_t base, phys_addr_t size)
783 phys_addr_t end = base + size - 1;
785 memblock_dbg("%s: [%pa-%pa] %pS\n", __func__,
786 &base, &end, (void *)_RET_IP_);
788 return memblock_remove_range(&memblock.memory, base, size);
792 * memblock_free - free boot memory block
793 * @base: phys starting address of the boot memory block
794 * @size: size of the boot memory block in bytes
796 * Free boot memory block previously allocated by memblock_alloc_xx() API.
797 * The freeing memory will not be released to the buddy allocator.
799 int __init_memblock memblock_free(phys_addr_t base, phys_addr_t size)
801 phys_addr_t end = base + size - 1;
803 memblock_dbg("%s: [%pa-%pa] %pS\n", __func__,
804 &base, &end, (void *)_RET_IP_);
806 kmemleak_free_part_phys(base, size);
807 return memblock_remove_range(&memblock.reserved, base, size);
810 int __init_memblock memblock_reserve(phys_addr_t base, phys_addr_t size)
812 phys_addr_t end = base + size - 1;
814 memblock_dbg("%s: [%pa-%pa] %pS\n", __func__,
815 &base, &end, (void *)_RET_IP_);
817 return memblock_add_range(&memblock.reserved, base, size, MAX_NUMNODES, 0);
820 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
821 int __init_memblock memblock_physmem_add(phys_addr_t base, phys_addr_t size)
823 phys_addr_t end = base + size - 1;
825 memblock_dbg("%s: [%pa-%pa] %pS\n", __func__,
826 &base, &end, (void *)_RET_IP_);
828 return memblock_add_range(&physmem, base, size, MAX_NUMNODES, 0);
833 * memblock_setclr_flag - set or clear flag for a memory region
834 * @base: base address of the region
835 * @size: size of the region
836 * @set: set or clear the flag
837 * @flag: the flag to update
839 * This function isolates region [@base, @base + @size), and sets/clears flag
841 * Return: 0 on success, -errno on failure.
843 static int __init_memblock memblock_setclr_flag(phys_addr_t base,
844 phys_addr_t size, int set, int flag)
846 struct memblock_type *type = &memblock.memory;
847 int i, ret, start_rgn, end_rgn;
849 ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
853 for (i = start_rgn; i < end_rgn; i++) {
854 struct memblock_region *r = &type->regions[i];
862 memblock_merge_regions(type);
867 * memblock_mark_hotplug - Mark hotpluggable memory with flag MEMBLOCK_HOTPLUG.
868 * @base: the base phys addr of the region
869 * @size: the size of the region
871 * Return: 0 on success, -errno on failure.
873 int __init_memblock memblock_mark_hotplug(phys_addr_t base, phys_addr_t size)
875 return memblock_setclr_flag(base, size, 1, MEMBLOCK_HOTPLUG);
879 * memblock_clear_hotplug - Clear flag MEMBLOCK_HOTPLUG for a specified region.
880 * @base: the base phys addr of the region
881 * @size: the size of the region
883 * Return: 0 on success, -errno on failure.
885 int __init_memblock memblock_clear_hotplug(phys_addr_t base, phys_addr_t size)
887 return memblock_setclr_flag(base, size, 0, MEMBLOCK_HOTPLUG);
891 * memblock_mark_mirror - Mark mirrored memory with flag MEMBLOCK_MIRROR.
892 * @base: the base phys addr of the region
893 * @size: the size of the region
895 * Return: 0 on success, -errno on failure.
897 int __init_memblock memblock_mark_mirror(phys_addr_t base, phys_addr_t size)
899 system_has_some_mirror = true;
901 return memblock_setclr_flag(base, size, 1, MEMBLOCK_MIRROR);
905 * memblock_mark_nomap - Mark a memory region with flag MEMBLOCK_NOMAP.
906 * @base: the base phys addr of the region
907 * @size: the size of the region
909 * Return: 0 on success, -errno on failure.
911 int __init_memblock memblock_mark_nomap(phys_addr_t base, phys_addr_t size)
913 return memblock_setclr_flag(base, size, 1, MEMBLOCK_NOMAP);
917 * memblock_clear_nomap - Clear flag MEMBLOCK_NOMAP for a specified region.
918 * @base: the base phys addr of the region
919 * @size: the size of the region
921 * Return: 0 on success, -errno on failure.
923 int __init_memblock memblock_clear_nomap(phys_addr_t base, phys_addr_t size)
925 return memblock_setclr_flag(base, size, 0, MEMBLOCK_NOMAP);
928 static bool should_skip_region(struct memblock_type *type,
929 struct memblock_region *m,
932 int m_nid = memblock_get_region_node(m);
934 /* we never skip regions when iterating memblock.reserved or physmem */
935 if (type != memblock_memory)
938 /* only memory regions are associated with nodes, check it */
939 if (nid != NUMA_NO_NODE && nid != m_nid)
942 /* skip hotpluggable memory regions if needed */
943 if (movable_node_is_enabled() && memblock_is_hotpluggable(m))
946 /* if we want mirror memory skip non-mirror memory regions */
947 if ((flags & MEMBLOCK_MIRROR) && !memblock_is_mirror(m))
950 /* skip nomap memory unless we were asked for it explicitly */
951 if (!(flags & MEMBLOCK_NOMAP) && memblock_is_nomap(m))
958 * __next_mem_range - next function for for_each_free_mem_range() etc.
959 * @idx: pointer to u64 loop variable
960 * @nid: node selector, %NUMA_NO_NODE for all nodes
961 * @flags: pick from blocks based on memory attributes
962 * @type_a: pointer to memblock_type from where the range is taken
963 * @type_b: pointer to memblock_type which excludes memory from being taken
964 * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL
965 * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL
966 * @out_nid: ptr to int for nid of the range, can be %NULL
968 * Find the first area from *@idx which matches @nid, fill the out
969 * parameters, and update *@idx for the next iteration. The lower 32bit of
970 * *@idx contains index into type_a and the upper 32bit indexes the
971 * areas before each region in type_b. For example, if type_b regions
972 * look like the following,
974 * 0:[0-16), 1:[32-48), 2:[128-130)
976 * The upper 32bit indexes the following regions.
978 * 0:[0-0), 1:[16-32), 2:[48-128), 3:[130-MAX)
980 * As both region arrays are sorted, the function advances the two indices
981 * in lockstep and returns each intersection.
983 void __next_mem_range(u64 *idx, int nid, enum memblock_flags flags,
984 struct memblock_type *type_a,
985 struct memblock_type *type_b, phys_addr_t *out_start,
986 phys_addr_t *out_end, int *out_nid)
988 int idx_a = *idx & 0xffffffff;
989 int idx_b = *idx >> 32;
991 if (WARN_ONCE(nid == MAX_NUMNODES,
992 "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
995 for (; idx_a < type_a->cnt; idx_a++) {
996 struct memblock_region *m = &type_a->regions[idx_a];
998 phys_addr_t m_start = m->base;
999 phys_addr_t m_end = m->base + m->size;
1000 int m_nid = memblock_get_region_node(m);
1002 if (should_skip_region(type_a, m, nid, flags))
1007 *out_start = m_start;
1013 *idx = (u32)idx_a | (u64)idx_b << 32;
1017 /* scan areas before each reservation */
1018 for (; idx_b < type_b->cnt + 1; idx_b++) {
1019 struct memblock_region *r;
1020 phys_addr_t r_start;
1023 r = &type_b->regions[idx_b];
1024 r_start = idx_b ? r[-1].base + r[-1].size : 0;
1025 r_end = idx_b < type_b->cnt ?
1026 r->base : PHYS_ADDR_MAX;
1029 * if idx_b advanced past idx_a,
1030 * break out to advance idx_a
1032 if (r_start >= m_end)
1034 /* if the two regions intersect, we're done */
1035 if (m_start < r_end) {
1038 max(m_start, r_start);
1040 *out_end = min(m_end, r_end);
1044 * The region which ends first is
1045 * advanced for the next iteration.
1051 *idx = (u32)idx_a | (u64)idx_b << 32;
1057 /* signal end of iteration */
1062 * __next_mem_range_rev - generic next function for for_each_*_range_rev()
1064 * @idx: pointer to u64 loop variable
1065 * @nid: node selector, %NUMA_NO_NODE for all nodes
1066 * @flags: pick from blocks based on memory attributes
1067 * @type_a: pointer to memblock_type from where the range is taken
1068 * @type_b: pointer to memblock_type which excludes memory from being taken
1069 * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL
1070 * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL
1071 * @out_nid: ptr to int for nid of the range, can be %NULL
1073 * Finds the next range from type_a which is not marked as unsuitable
1076 * Reverse of __next_mem_range().
1078 void __init_memblock __next_mem_range_rev(u64 *idx, int nid,
1079 enum memblock_flags flags,
1080 struct memblock_type *type_a,
1081 struct memblock_type *type_b,
1082 phys_addr_t *out_start,
1083 phys_addr_t *out_end, int *out_nid)
1085 int idx_a = *idx & 0xffffffff;
1086 int idx_b = *idx >> 32;
1088 if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
1091 if (*idx == (u64)ULLONG_MAX) {
1092 idx_a = type_a->cnt - 1;
1094 idx_b = type_b->cnt;
1099 for (; idx_a >= 0; idx_a--) {
1100 struct memblock_region *m = &type_a->regions[idx_a];
1102 phys_addr_t m_start = m->base;
1103 phys_addr_t m_end = m->base + m->size;
1104 int m_nid = memblock_get_region_node(m);
1106 if (should_skip_region(type_a, m, nid, flags))
1111 *out_start = m_start;
1117 *idx = (u32)idx_a | (u64)idx_b << 32;
1121 /* scan areas before each reservation */
1122 for (; idx_b >= 0; idx_b--) {
1123 struct memblock_region *r;
1124 phys_addr_t r_start;
1127 r = &type_b->regions[idx_b];
1128 r_start = idx_b ? r[-1].base + r[-1].size : 0;
1129 r_end = idx_b < type_b->cnt ?
1130 r->base : PHYS_ADDR_MAX;
1132 * if idx_b advanced past idx_a,
1133 * break out to advance idx_a
1136 if (r_end <= m_start)
1138 /* if the two regions intersect, we're done */
1139 if (m_end > r_start) {
1141 *out_start = max(m_start, r_start);
1143 *out_end = min(m_end, r_end);
1146 if (m_start >= r_start)
1150 *idx = (u32)idx_a | (u64)idx_b << 32;
1155 /* signal end of iteration */
1160 * Common iterator interface used to define for_each_mem_pfn_range().
1162 void __init_memblock __next_mem_pfn_range(int *idx, int nid,
1163 unsigned long *out_start_pfn,
1164 unsigned long *out_end_pfn, int *out_nid)
1166 struct memblock_type *type = &memblock.memory;
1167 struct memblock_region *r;
1170 while (++*idx < type->cnt) {
1171 r = &type->regions[*idx];
1172 r_nid = memblock_get_region_node(r);
1174 if (PFN_UP(r->base) >= PFN_DOWN(r->base + r->size))
1176 if (nid == MAX_NUMNODES || nid == r_nid)
1179 if (*idx >= type->cnt) {
1185 *out_start_pfn = PFN_UP(r->base);
1187 *out_end_pfn = PFN_DOWN(r->base + r->size);
1193 * memblock_set_node - set node ID on memblock regions
1194 * @base: base of area to set node ID for
1195 * @size: size of area to set node ID for
1196 * @type: memblock type to set node ID for
1197 * @nid: node ID to set
1199 * Set the nid of memblock @type regions in [@base, @base + @size) to @nid.
1200 * Regions which cross the area boundaries are split as necessary.
1203 * 0 on success, -errno on failure.
1205 int __init_memblock memblock_set_node(phys_addr_t base, phys_addr_t size,
1206 struct memblock_type *type, int nid)
1208 #ifdef CONFIG_NEED_MULTIPLE_NODES
1209 int start_rgn, end_rgn;
1212 ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
1216 for (i = start_rgn; i < end_rgn; i++)
1217 memblock_set_region_node(&type->regions[i], nid);
1219 memblock_merge_regions(type);
1224 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1226 * __next_mem_pfn_range_in_zone - iterator for for_each_*_range_in_zone()
1228 * @idx: pointer to u64 loop variable
1229 * @zone: zone in which all of the memory blocks reside
1230 * @out_spfn: ptr to ulong for start pfn of the range, can be %NULL
1231 * @out_epfn: ptr to ulong for end pfn of the range, can be %NULL
1233 * This function is meant to be a zone/pfn specific wrapper for the
1234 * for_each_mem_range type iterators. Specifically they are used in the
1235 * deferred memory init routines and as such we were duplicating much of
1236 * this logic throughout the code. So instead of having it in multiple
1237 * locations it seemed like it would make more sense to centralize this to
1238 * one new iterator that does everything they need.
1240 void __init_memblock
1241 __next_mem_pfn_range_in_zone(u64 *idx, struct zone *zone,
1242 unsigned long *out_spfn, unsigned long *out_epfn)
1244 int zone_nid = zone_to_nid(zone);
1245 phys_addr_t spa, epa;
1248 __next_mem_range(idx, zone_nid, MEMBLOCK_NONE,
1249 &memblock.memory, &memblock.reserved,
1252 while (*idx != U64_MAX) {
1253 unsigned long epfn = PFN_DOWN(epa);
1254 unsigned long spfn = PFN_UP(spa);
1257 * Verify the end is at least past the start of the zone and
1258 * that we have at least one PFN to initialize.
1260 if (zone->zone_start_pfn < epfn && spfn < epfn) {
1261 /* if we went too far just stop searching */
1262 if (zone_end_pfn(zone) <= spfn) {
1268 *out_spfn = max(zone->zone_start_pfn, spfn);
1270 *out_epfn = min(zone_end_pfn(zone), epfn);
1275 __next_mem_range(idx, zone_nid, MEMBLOCK_NONE,
1276 &memblock.memory, &memblock.reserved,
1280 /* signal end of iteration */
1282 *out_spfn = ULONG_MAX;
1287 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
1290 * memblock_alloc_range_nid - allocate boot memory block
1291 * @size: size of memory block to be allocated in bytes
1292 * @align: alignment of the region and block's size
1293 * @start: the lower bound of the memory region to allocate (phys address)
1294 * @end: the upper bound of the memory region to allocate (phys address)
1295 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1296 * @exact_nid: control the allocation fall back to other nodes
1298 * The allocation is performed from memory region limited by
1299 * memblock.current_limit if @end == %MEMBLOCK_ALLOC_ACCESSIBLE.
1301 * If the specified node can not hold the requested memory and @exact_nid
1302 * is false, the allocation falls back to any node in the system.
1304 * For systems with memory mirroring, the allocation is attempted first
1305 * from the regions with mirroring enabled and then retried from any
1308 * In addition, function sets the min_count to 0 using kmemleak_alloc_phys for
1309 * allocated boot memory block, so that it is never reported as leaks.
1312 * Physical address of allocated memory block on success, %0 on failure.
1314 phys_addr_t __init memblock_alloc_range_nid(phys_addr_t size,
1315 phys_addr_t align, phys_addr_t start,
1316 phys_addr_t end, int nid,
1319 enum memblock_flags flags = choose_memblock_flags();
1322 if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
1326 /* Can't use WARNs this early in boot on powerpc */
1328 align = SMP_CACHE_BYTES;
1332 found = memblock_find_in_range_node(size, align, start, end, nid,
1334 if (found && !memblock_reserve(found, size))
1337 if (nid != NUMA_NO_NODE && !exact_nid) {
1338 found = memblock_find_in_range_node(size, align, start,
1341 if (found && !memblock_reserve(found, size))
1345 if (flags & MEMBLOCK_MIRROR) {
1346 flags &= ~MEMBLOCK_MIRROR;
1347 pr_warn("Could not allocate %pap bytes of mirrored memory\n",
1355 /* Skip kmemleak for kasan_init() due to high volume. */
1356 if (end != MEMBLOCK_ALLOC_KASAN)
1358 * The min_count is set to 0 so that memblock allocated
1359 * blocks are never reported as leaks. This is because many
1360 * of these blocks are only referred via the physical
1361 * address which is not looked up by kmemleak.
1363 kmemleak_alloc_phys(found, size, 0, 0);
1369 * memblock_phys_alloc_range - allocate a memory block inside specified range
1370 * @size: size of memory block to be allocated in bytes
1371 * @align: alignment of the region and block's size
1372 * @start: the lower bound of the memory region to allocate (physical address)
1373 * @end: the upper bound of the memory region to allocate (physical address)
1375 * Allocate @size bytes in the between @start and @end.
1377 * Return: physical address of the allocated memory block on success,
1380 phys_addr_t __init memblock_phys_alloc_range(phys_addr_t size,
1385 memblock_dbg("%s: %llu bytes align=0x%llx from=%pa max_addr=%pa %pS\n",
1386 __func__, (u64)size, (u64)align, &start, &end,
1388 return memblock_alloc_range_nid(size, align, start, end, NUMA_NO_NODE,
1393 * memblock_phys_alloc_try_nid - allocate a memory block from specified NUMA node
1394 * @size: size of memory block to be allocated in bytes
1395 * @align: alignment of the region and block's size
1396 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1398 * Allocates memory block from the specified NUMA node. If the node
1399 * has no available memory, attempts to allocated from any node in the
1402 * Return: physical address of the allocated memory block on success,
1405 phys_addr_t __init memblock_phys_alloc_try_nid(phys_addr_t size, phys_addr_t align, int nid)
1407 return memblock_alloc_range_nid(size, align, 0,
1408 MEMBLOCK_ALLOC_ACCESSIBLE, nid, false);
1412 * memblock_alloc_internal - allocate boot memory block
1413 * @size: size of memory block to be allocated in bytes
1414 * @align: alignment of the region and block's size
1415 * @min_addr: the lower bound of the memory region to allocate (phys address)
1416 * @max_addr: the upper bound of the memory region to allocate (phys address)
1417 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1418 * @exact_nid: control the allocation fall back to other nodes
1420 * Allocates memory block using memblock_alloc_range_nid() and
1421 * converts the returned physical address to virtual.
1423 * The @min_addr limit is dropped if it can not be satisfied and the allocation
1424 * will fall back to memory below @min_addr. Other constraints, such
1425 * as node and mirrored memory will be handled again in
1426 * memblock_alloc_range_nid().
1429 * Virtual address of allocated memory block on success, NULL on failure.
1431 static void * __init memblock_alloc_internal(
1432 phys_addr_t size, phys_addr_t align,
1433 phys_addr_t min_addr, phys_addr_t max_addr,
1434 int nid, bool exact_nid)
1439 * Detect any accidental use of these APIs after slab is ready, as at
1440 * this moment memblock may be deinitialized already and its
1441 * internal data may be destroyed (after execution of memblock_free_all)
1443 if (WARN_ON_ONCE(slab_is_available()))
1444 return kzalloc_node(size, GFP_NOWAIT, nid);
1446 if (max_addr > memblock.current_limit)
1447 max_addr = memblock.current_limit;
1449 alloc = memblock_alloc_range_nid(size, align, min_addr, max_addr, nid,
1452 /* retry allocation without lower limit */
1453 if (!alloc && min_addr)
1454 alloc = memblock_alloc_range_nid(size, align, 0, max_addr, nid,
1460 return phys_to_virt(alloc);
1464 * memblock_alloc_exact_nid_raw - allocate boot memory block on the exact node
1465 * without zeroing memory
1466 * @size: size of memory block to be allocated in bytes
1467 * @align: alignment of the region and block's size
1468 * @min_addr: the lower bound of the memory region from where the allocation
1469 * is preferred (phys address)
1470 * @max_addr: the upper bound of the memory region from where the allocation
1471 * is preferred (phys address), or %MEMBLOCK_ALLOC_ACCESSIBLE to
1472 * allocate only from memory limited by memblock.current_limit value
1473 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1475 * Public function, provides additional debug information (including caller
1476 * info), if enabled. Does not zero allocated memory.
1479 * Virtual address of allocated memory block on success, NULL on failure.
1481 void * __init memblock_alloc_exact_nid_raw(
1482 phys_addr_t size, phys_addr_t align,
1483 phys_addr_t min_addr, phys_addr_t max_addr,
1488 memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=%pa max_addr=%pa %pS\n",
1489 __func__, (u64)size, (u64)align, nid, &min_addr,
1490 &max_addr, (void *)_RET_IP_);
1492 ptr = memblock_alloc_internal(size, align,
1493 min_addr, max_addr, nid, true);
1494 if (ptr && size > 0)
1495 page_init_poison(ptr, size);
1501 * memblock_alloc_try_nid_raw - allocate boot memory block without zeroing
1502 * memory and without panicking
1503 * @size: size of memory block to be allocated in bytes
1504 * @align: alignment of the region and block's size
1505 * @min_addr: the lower bound of the memory region from where the allocation
1506 * is preferred (phys address)
1507 * @max_addr: the upper bound of the memory region from where the allocation
1508 * is preferred (phys address), or %MEMBLOCK_ALLOC_ACCESSIBLE to
1509 * allocate only from memory limited by memblock.current_limit value
1510 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1512 * Public function, provides additional debug information (including caller
1513 * info), if enabled. Does not zero allocated memory, does not panic if request
1514 * cannot be satisfied.
1517 * Virtual address of allocated memory block on success, NULL on failure.
1519 void * __init memblock_alloc_try_nid_raw(
1520 phys_addr_t size, phys_addr_t align,
1521 phys_addr_t min_addr, phys_addr_t max_addr,
1526 memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=%pa max_addr=%pa %pS\n",
1527 __func__, (u64)size, (u64)align, nid, &min_addr,
1528 &max_addr, (void *)_RET_IP_);
1530 ptr = memblock_alloc_internal(size, align,
1531 min_addr, max_addr, nid, false);
1532 if (ptr && size > 0)
1533 page_init_poison(ptr, size);
1539 * memblock_alloc_try_nid - allocate boot memory block
1540 * @size: size of memory block to be allocated in bytes
1541 * @align: alignment of the region and block's size
1542 * @min_addr: the lower bound of the memory region from where the allocation
1543 * is preferred (phys address)
1544 * @max_addr: the upper bound of the memory region from where the allocation
1545 * is preferred (phys address), or %MEMBLOCK_ALLOC_ACCESSIBLE to
1546 * allocate only from memory limited by memblock.current_limit value
1547 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1549 * Public function, provides additional debug information (including caller
1550 * info), if enabled. This function zeroes the allocated memory.
1553 * Virtual address of allocated memory block on success, NULL on failure.
1555 void * __init memblock_alloc_try_nid(
1556 phys_addr_t size, phys_addr_t align,
1557 phys_addr_t min_addr, phys_addr_t max_addr,
1562 memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=%pa max_addr=%pa %pS\n",
1563 __func__, (u64)size, (u64)align, nid, &min_addr,
1564 &max_addr, (void *)_RET_IP_);
1565 ptr = memblock_alloc_internal(size, align,
1566 min_addr, max_addr, nid, false);
1568 memset(ptr, 0, size);
1574 * __memblock_free_late - free pages directly to buddy allocator
1575 * @base: phys starting address of the boot memory block
1576 * @size: size of the boot memory block in bytes
1578 * This is only useful when the memblock allocator has already been torn
1579 * down, but we are still initializing the system. Pages are released directly
1580 * to the buddy allocator.
1582 void __init __memblock_free_late(phys_addr_t base, phys_addr_t size)
1584 phys_addr_t cursor, end;
1586 end = base + size - 1;
1587 memblock_dbg("%s: [%pa-%pa] %pS\n",
1588 __func__, &base, &end, (void *)_RET_IP_);
1589 kmemleak_free_part_phys(base, size);
1590 cursor = PFN_UP(base);
1591 end = PFN_DOWN(base + size);
1593 for (; cursor < end; cursor++) {
1594 memblock_free_pages(pfn_to_page(cursor), cursor, 0);
1595 totalram_pages_inc();
1600 * Remaining API functions
1603 phys_addr_t __init_memblock memblock_phys_mem_size(void)
1605 return memblock.memory.total_size;
1608 phys_addr_t __init_memblock memblock_reserved_size(void)
1610 return memblock.reserved.total_size;
1613 /* lowest address */
1614 phys_addr_t __init_memblock memblock_start_of_DRAM(void)
1616 return memblock.memory.regions[0].base;
1619 phys_addr_t __init_memblock memblock_end_of_DRAM(void)
1621 int idx = memblock.memory.cnt - 1;
1623 return (memblock.memory.regions[idx].base + memblock.memory.regions[idx].size);
1626 static phys_addr_t __init_memblock __find_max_addr(phys_addr_t limit)
1628 phys_addr_t max_addr = PHYS_ADDR_MAX;
1629 struct memblock_region *r;
1632 * translate the memory @limit size into the max address within one of
1633 * the memory memblock regions, if the @limit exceeds the total size
1634 * of those regions, max_addr will keep original value PHYS_ADDR_MAX
1636 for_each_mem_region(r) {
1637 if (limit <= r->size) {
1638 max_addr = r->base + limit;
1647 void __init memblock_enforce_memory_limit(phys_addr_t limit)
1649 phys_addr_t max_addr;
1654 max_addr = __find_max_addr(limit);
1656 /* @limit exceeds the total size of the memory, do nothing */
1657 if (max_addr == PHYS_ADDR_MAX)
1660 /* truncate both memory and reserved regions */
1661 memblock_remove_range(&memblock.memory, max_addr,
1663 memblock_remove_range(&memblock.reserved, max_addr,
1667 void __init memblock_cap_memory_range(phys_addr_t base, phys_addr_t size)
1669 int start_rgn, end_rgn;
1675 ret = memblock_isolate_range(&memblock.memory, base, size,
1676 &start_rgn, &end_rgn);
1680 /* remove all the MAP regions */
1681 for (i = memblock.memory.cnt - 1; i >= end_rgn; i--)
1682 if (!memblock_is_nomap(&memblock.memory.regions[i]))
1683 memblock_remove_region(&memblock.memory, i);
1685 for (i = start_rgn - 1; i >= 0; i--)
1686 if (!memblock_is_nomap(&memblock.memory.regions[i]))
1687 memblock_remove_region(&memblock.memory, i);
1689 /* truncate the reserved regions */
1690 memblock_remove_range(&memblock.reserved, 0, base);
1691 memblock_remove_range(&memblock.reserved,
1692 base + size, PHYS_ADDR_MAX);
1695 void __init memblock_mem_limit_remove_map(phys_addr_t limit)
1697 phys_addr_t max_addr;
1702 max_addr = __find_max_addr(limit);
1704 /* @limit exceeds the total size of the memory, do nothing */
1705 if (max_addr == PHYS_ADDR_MAX)
1708 memblock_cap_memory_range(0, max_addr);
1711 static int __init_memblock memblock_search(struct memblock_type *type, phys_addr_t addr)
1713 unsigned int left = 0, right = type->cnt;
1716 unsigned int mid = (right + left) / 2;
1718 if (addr < type->regions[mid].base)
1720 else if (addr >= (type->regions[mid].base +
1721 type->regions[mid].size))
1725 } while (left < right);
1729 bool __init_memblock memblock_is_reserved(phys_addr_t addr)
1731 return memblock_search(&memblock.reserved, addr) != -1;
1734 bool __init_memblock memblock_is_memory(phys_addr_t addr)
1736 return memblock_search(&memblock.memory, addr) != -1;
1739 bool __init_memblock memblock_is_map_memory(phys_addr_t addr)
1741 int i = memblock_search(&memblock.memory, addr);
1745 return !memblock_is_nomap(&memblock.memory.regions[i]);
1748 int __init_memblock memblock_search_pfn_nid(unsigned long pfn,
1749 unsigned long *start_pfn, unsigned long *end_pfn)
1751 struct memblock_type *type = &memblock.memory;
1752 int mid = memblock_search(type, PFN_PHYS(pfn));
1757 *start_pfn = PFN_DOWN(type->regions[mid].base);
1758 *end_pfn = PFN_DOWN(type->regions[mid].base + type->regions[mid].size);
1760 return memblock_get_region_node(&type->regions[mid]);
1764 * memblock_is_region_memory - check if a region is a subset of memory
1765 * @base: base of region to check
1766 * @size: size of region to check
1768 * Check if the region [@base, @base + @size) is a subset of a memory block.
1771 * 0 if false, non-zero if true
1773 bool __init_memblock memblock_is_region_memory(phys_addr_t base, phys_addr_t size)
1775 int idx = memblock_search(&memblock.memory, base);
1776 phys_addr_t end = base + memblock_cap_size(base, &size);
1780 return (memblock.memory.regions[idx].base +
1781 memblock.memory.regions[idx].size) >= end;
1785 * memblock_is_region_reserved - check if a region intersects reserved memory
1786 * @base: base of region to check
1787 * @size: size of region to check
1789 * Check if the region [@base, @base + @size) intersects a reserved
1793 * True if they intersect, false if not.
1795 bool __init_memblock memblock_is_region_reserved(phys_addr_t base, phys_addr_t size)
1797 memblock_cap_size(base, &size);
1798 return memblock_overlaps_region(&memblock.reserved, base, size);
1801 void __init_memblock memblock_trim_memory(phys_addr_t align)
1803 phys_addr_t start, end, orig_start, orig_end;
1804 struct memblock_region *r;
1806 for_each_mem_region(r) {
1807 orig_start = r->base;
1808 orig_end = r->base + r->size;
1809 start = round_up(orig_start, align);
1810 end = round_down(orig_end, align);
1812 if (start == orig_start && end == orig_end)
1817 r->size = end - start;
1819 memblock_remove_region(&memblock.memory,
1820 r - memblock.memory.regions);
1826 void __init_memblock memblock_set_current_limit(phys_addr_t limit)
1828 memblock.current_limit = limit;
1831 phys_addr_t __init_memblock memblock_get_current_limit(void)
1833 return memblock.current_limit;
1836 static void __init_memblock memblock_dump(struct memblock_type *type)
1838 phys_addr_t base, end, size;
1839 enum memblock_flags flags;
1841 struct memblock_region *rgn;
1843 pr_info(" %s.cnt = 0x%lx\n", type->name, type->cnt);
1845 for_each_memblock_type(idx, type, rgn) {
1846 char nid_buf[32] = "";
1850 end = base + size - 1;
1852 #ifdef CONFIG_NEED_MULTIPLE_NODES
1853 if (memblock_get_region_node(rgn) != MAX_NUMNODES)
1854 snprintf(nid_buf, sizeof(nid_buf), " on node %d",
1855 memblock_get_region_node(rgn));
1857 pr_info(" %s[%#x]\t[%pa-%pa], %pa bytes%s flags: %#x\n",
1858 type->name, idx, &base, &end, &size, nid_buf, flags);
1862 static void __init_memblock __memblock_dump_all(void)
1864 pr_info("MEMBLOCK configuration:\n");
1865 pr_info(" memory size = %pa reserved size = %pa\n",
1866 &memblock.memory.total_size,
1867 &memblock.reserved.total_size);
1869 memblock_dump(&memblock.memory);
1870 memblock_dump(&memblock.reserved);
1871 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
1872 memblock_dump(&physmem);
1876 void __init_memblock memblock_dump_all(void)
1879 __memblock_dump_all();
1882 void __init memblock_allow_resize(void)
1884 memblock_can_resize = 1;
1887 static int __init early_memblock(char *p)
1889 if (p && strstr(p, "debug"))
1893 early_param("memblock", early_memblock);
1895 static void __init free_memmap(unsigned long start_pfn, unsigned long end_pfn)
1897 struct page *start_pg, *end_pg;
1898 phys_addr_t pg, pgend;
1901 * Convert start_pfn/end_pfn to a struct page pointer.
1903 start_pg = pfn_to_page(start_pfn - 1) + 1;
1904 end_pg = pfn_to_page(end_pfn - 1) + 1;
1907 * Convert to physical addresses, and round start upwards and end
1910 pg = PAGE_ALIGN(__pa(start_pg));
1911 pgend = __pa(end_pg) & PAGE_MASK;
1914 * If there are free pages between these, free the section of the
1918 memblock_free(pg, pgend - pg);
1922 * The mem_map array can get very big. Free the unused area of the memory map.
1924 static void __init free_unused_memmap(void)
1926 unsigned long start, end, prev_end = 0;
1929 if (!IS_ENABLED(CONFIG_HAVE_ARCH_PFN_VALID) ||
1930 IS_ENABLED(CONFIG_SPARSEMEM_VMEMMAP))
1934 * This relies on each bank being in address order.
1935 * The banks are sorted previously in bootmem_init().
1937 for_each_mem_pfn_range(i, MAX_NUMNODES, &start, &end, NULL) {
1938 #ifdef CONFIG_SPARSEMEM
1940 * Take care not to free memmap entries that don't exist
1941 * due to SPARSEMEM sections which aren't present.
1943 start = min(start, ALIGN(prev_end, PAGES_PER_SECTION));
1946 * Align down here since the VM subsystem insists that the
1947 * memmap entries are valid from the bank start aligned to
1948 * MAX_ORDER_NR_PAGES.
1950 start = round_down(start, MAX_ORDER_NR_PAGES);
1954 * If we had a previous bank, and there is a space
1955 * between the current bank and the previous, free it.
1957 if (prev_end && prev_end < start)
1958 free_memmap(prev_end, start);
1961 * Align up here since the VM subsystem insists that the
1962 * memmap entries are valid from the bank end aligned to
1963 * MAX_ORDER_NR_PAGES.
1965 prev_end = ALIGN(end, MAX_ORDER_NR_PAGES);
1968 #ifdef CONFIG_SPARSEMEM
1969 if (!IS_ALIGNED(prev_end, PAGES_PER_SECTION))
1970 free_memmap(prev_end, ALIGN(prev_end, PAGES_PER_SECTION));
1974 static void __init __free_pages_memory(unsigned long start, unsigned long end)
1978 while (start < end) {
1979 order = min(MAX_ORDER - 1UL, __ffs(start));
1981 while (start + (1UL << order) > end)
1984 memblock_free_pages(pfn_to_page(start), start, order);
1986 start += (1UL << order);
1990 static unsigned long __init __free_memory_core(phys_addr_t start,
1993 unsigned long start_pfn = PFN_UP(start);
1994 unsigned long end_pfn = min_t(unsigned long,
1995 PFN_DOWN(end), max_low_pfn);
1997 if (start_pfn >= end_pfn)
2000 __free_pages_memory(start_pfn, end_pfn);
2002 return end_pfn - start_pfn;
2005 static unsigned long __init free_low_memory_core_early(void)
2007 unsigned long count = 0;
2008 phys_addr_t start, end;
2011 memblock_clear_hotplug(0, -1);
2013 for_each_reserved_mem_range(i, &start, &end)
2014 reserve_bootmem_region(start, end);
2017 * We need to use NUMA_NO_NODE instead of NODE_DATA(0)->node_id
2018 * because in some case like Node0 doesn't have RAM installed
2019 * low ram will be on Node1
2021 for_each_free_mem_range(i, NUMA_NO_NODE, MEMBLOCK_NONE, &start, &end,
2023 count += __free_memory_core(start, end);
2028 static int reset_managed_pages_done __initdata;
2030 void reset_node_managed_pages(pg_data_t *pgdat)
2034 for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++)
2035 atomic_long_set(&z->managed_pages, 0);
2038 void __init reset_all_zones_managed_pages(void)
2040 struct pglist_data *pgdat;
2042 if (reset_managed_pages_done)
2045 for_each_online_pgdat(pgdat)
2046 reset_node_managed_pages(pgdat);
2048 reset_managed_pages_done = 1;
2052 * memblock_free_all - release free pages to the buddy allocator
2054 * Return: the number of pages actually released.
2056 unsigned long __init memblock_free_all(void)
2058 unsigned long pages;
2060 free_unused_memmap();
2061 reset_all_zones_managed_pages();
2063 pages = free_low_memory_core_early();
2064 totalram_pages_add(pages);
2069 #if defined(CONFIG_DEBUG_FS) && defined(CONFIG_ARCH_KEEP_MEMBLOCK)
2071 static int memblock_debug_show(struct seq_file *m, void *private)
2073 struct memblock_type *type = m->private;
2074 struct memblock_region *reg;
2078 for (i = 0; i < type->cnt; i++) {
2079 reg = &type->regions[i];
2080 end = reg->base + reg->size - 1;
2082 seq_printf(m, "%4d: ", i);
2083 seq_printf(m, "%pa..%pa\n", ®->base, &end);
2087 DEFINE_SHOW_ATTRIBUTE(memblock_debug);
2089 static int __init memblock_init_debugfs(void)
2091 struct dentry *root = debugfs_create_dir("memblock", NULL);
2093 debugfs_create_file("memory", 0444, root,
2094 &memblock.memory, &memblock_debug_fops);
2095 debugfs_create_file("reserved", 0444, root,
2096 &memblock.reserved, &memblock_debug_fops);
2097 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
2098 debugfs_create_file("physmem", 0444, root, &physmem,
2099 &memblock_debug_fops);
2104 __initcall(memblock_init_debugfs);
2106 #endif /* CONFIG_DEBUG_FS */