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.
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 memblock_cap_size(base, &size);
187 for (i = 0; i < type->cnt; i++)
188 if (memblock_addrs_overlap(base, size, type->regions[i].base,
189 type->regions[i].size))
191 return i < type->cnt;
195 * __memblock_find_range_bottom_up - find free area utility in bottom-up
196 * @start: start of candidate range
197 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_ANYWHERE or
198 * %MEMBLOCK_ALLOC_ACCESSIBLE
199 * @size: size of free area to find
200 * @align: alignment of free area to find
201 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
202 * @flags: pick from blocks based on memory attributes
204 * Utility called from memblock_find_in_range_node(), find free area bottom-up.
207 * Found address on success, 0 on failure.
209 static phys_addr_t __init_memblock
210 __memblock_find_range_bottom_up(phys_addr_t start, phys_addr_t end,
211 phys_addr_t size, phys_addr_t align, int nid,
212 enum memblock_flags flags)
214 phys_addr_t this_start, this_end, cand;
217 for_each_free_mem_range(i, nid, flags, &this_start, &this_end, NULL) {
218 this_start = clamp(this_start, start, end);
219 this_end = clamp(this_end, start, end);
221 cand = round_up(this_start, align);
222 if (cand < this_end && this_end - cand >= size)
230 * __memblock_find_range_top_down - find free area utility, in top-down
231 * @start: start of candidate range
232 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_ANYWHERE or
233 * %MEMBLOCK_ALLOC_ACCESSIBLE
234 * @size: size of free area to find
235 * @align: alignment of free area to find
236 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
237 * @flags: pick from blocks based on memory attributes
239 * Utility called from memblock_find_in_range_node(), find free area top-down.
242 * Found address on success, 0 on failure.
244 static phys_addr_t __init_memblock
245 __memblock_find_range_top_down(phys_addr_t start, phys_addr_t end,
246 phys_addr_t size, phys_addr_t align, int nid,
247 enum memblock_flags flags)
249 phys_addr_t this_start, this_end, cand;
252 for_each_free_mem_range_reverse(i, nid, flags, &this_start, &this_end,
254 this_start = clamp(this_start, start, end);
255 this_end = clamp(this_end, start, end);
260 cand = round_down(this_end - size, align);
261 if (cand >= this_start)
269 * memblock_find_in_range_node - find free area in given range and node
270 * @size: size of free area to find
271 * @align: alignment of free area to find
272 * @start: start of candidate range
273 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_ANYWHERE or
274 * %MEMBLOCK_ALLOC_ACCESSIBLE
275 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
276 * @flags: pick from blocks based on memory attributes
278 * Find @size free area aligned to @align in the specified range and node.
281 * Found address on success, 0 on failure.
283 static phys_addr_t __init_memblock memblock_find_in_range_node(phys_addr_t size,
284 phys_addr_t align, phys_addr_t start,
285 phys_addr_t end, int nid,
286 enum memblock_flags flags)
289 if (end == MEMBLOCK_ALLOC_ACCESSIBLE ||
290 end == MEMBLOCK_ALLOC_KASAN)
291 end = memblock.current_limit;
293 /* avoid allocating the first page */
294 start = max_t(phys_addr_t, start, PAGE_SIZE);
295 end = max(start, end);
297 if (memblock_bottom_up())
298 return __memblock_find_range_bottom_up(start, end, size, align,
301 return __memblock_find_range_top_down(start, end, size, align,
306 * memblock_find_in_range - find free area in given range
307 * @start: start of candidate range
308 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_ANYWHERE or
309 * %MEMBLOCK_ALLOC_ACCESSIBLE
310 * @size: size of free area to find
311 * @align: alignment of free area to find
313 * Find @size free area aligned to @align in the specified range.
316 * Found address on success, 0 on failure.
318 static phys_addr_t __init_memblock memblock_find_in_range(phys_addr_t start,
319 phys_addr_t end, phys_addr_t size,
323 enum memblock_flags flags = choose_memblock_flags();
326 ret = memblock_find_in_range_node(size, align, start, end,
327 NUMA_NO_NODE, flags);
329 if (!ret && (flags & MEMBLOCK_MIRROR)) {
330 pr_warn("Could not allocate %pap bytes of mirrored memory\n",
332 flags &= ~MEMBLOCK_MIRROR;
339 static void __init_memblock memblock_remove_region(struct memblock_type *type, unsigned long r)
341 type->total_size -= type->regions[r].size;
342 memmove(&type->regions[r], &type->regions[r + 1],
343 (type->cnt - (r + 1)) * sizeof(type->regions[r]));
346 /* Special case for empty arrays */
347 if (type->cnt == 0) {
348 WARN_ON(type->total_size != 0);
350 type->regions[0].base = 0;
351 type->regions[0].size = 0;
352 type->regions[0].flags = 0;
353 memblock_set_region_node(&type->regions[0], MAX_NUMNODES);
357 #ifndef CONFIG_ARCH_KEEP_MEMBLOCK
359 * memblock_discard - discard memory and reserved arrays if they were allocated
361 void __init memblock_discard(void)
363 phys_addr_t addr, size;
365 if (memblock.reserved.regions != memblock_reserved_init_regions) {
366 addr = __pa(memblock.reserved.regions);
367 size = PAGE_ALIGN(sizeof(struct memblock_region) *
368 memblock.reserved.max);
369 if (memblock_reserved_in_slab)
370 kfree(memblock.reserved.regions);
372 __memblock_free_late(addr, size);
375 if (memblock.memory.regions != memblock_memory_init_regions) {
376 addr = __pa(memblock.memory.regions);
377 size = PAGE_ALIGN(sizeof(struct memblock_region) *
378 memblock.memory.max);
379 if (memblock_memory_in_slab)
380 kfree(memblock.memory.regions);
382 __memblock_free_late(addr, size);
385 memblock_memory = NULL;
390 * memblock_double_array - double the size of the memblock regions array
391 * @type: memblock type of the regions array being doubled
392 * @new_area_start: starting address of memory range to avoid overlap with
393 * @new_area_size: size of memory range to avoid overlap with
395 * Double the size of the @type regions array. If memblock is being used to
396 * allocate memory for a new reserved regions array and there is a previously
397 * allocated memory range [@new_area_start, @new_area_start + @new_area_size]
398 * waiting to be reserved, ensure the memory used by the new array does
402 * 0 on success, -1 on failure.
404 static int __init_memblock memblock_double_array(struct memblock_type *type,
405 phys_addr_t new_area_start,
406 phys_addr_t new_area_size)
408 struct memblock_region *new_array, *old_array;
409 phys_addr_t old_alloc_size, new_alloc_size;
410 phys_addr_t old_size, new_size, addr, new_end;
411 int use_slab = slab_is_available();
414 /* We don't allow resizing until we know about the reserved regions
415 * of memory that aren't suitable for allocation
417 if (!memblock_can_resize)
420 /* Calculate new doubled size */
421 old_size = type->max * sizeof(struct memblock_region);
422 new_size = old_size << 1;
424 * We need to allocated new one align to PAGE_SIZE,
425 * so we can free them completely later.
427 old_alloc_size = PAGE_ALIGN(old_size);
428 new_alloc_size = PAGE_ALIGN(new_size);
430 /* Retrieve the slab flag */
431 if (type == &memblock.memory)
432 in_slab = &memblock_memory_in_slab;
434 in_slab = &memblock_reserved_in_slab;
436 /* Try to find some space for it */
438 new_array = kmalloc(new_size, GFP_KERNEL);
439 addr = new_array ? __pa(new_array) : 0;
441 /* only exclude range when trying to double reserved.regions */
442 if (type != &memblock.reserved)
443 new_area_start = new_area_size = 0;
445 addr = memblock_find_in_range(new_area_start + new_area_size,
446 memblock.current_limit,
447 new_alloc_size, PAGE_SIZE);
448 if (!addr && new_area_size)
449 addr = memblock_find_in_range(0,
450 min(new_area_start, memblock.current_limit),
451 new_alloc_size, PAGE_SIZE);
453 new_array = addr ? __va(addr) : NULL;
456 pr_err("memblock: Failed to double %s array from %ld to %ld entries !\n",
457 type->name, type->max, type->max * 2);
461 new_end = addr + new_size - 1;
462 memblock_dbg("memblock: %s is doubled to %ld at [%pa-%pa]",
463 type->name, type->max * 2, &addr, &new_end);
466 * Found space, we now need to move the array over before we add the
467 * reserved region since it may be our reserved array itself that is
470 memcpy(new_array, type->regions, old_size);
471 memset(new_array + type->max, 0, old_size);
472 old_array = type->regions;
473 type->regions = new_array;
476 /* Free old array. We needn't free it if the array is the static one */
479 else if (old_array != memblock_memory_init_regions &&
480 old_array != memblock_reserved_init_regions)
481 memblock_free_ptr(old_array, old_alloc_size);
484 * Reserve the new array if that comes from the memblock. Otherwise, we
488 BUG_ON(memblock_reserve(addr, new_alloc_size));
490 /* Update slab flag */
497 * memblock_merge_regions - merge neighboring compatible regions
498 * @type: memblock type to scan
500 * Scan @type and merge neighboring compatible regions.
502 static void __init_memblock memblock_merge_regions(struct memblock_type *type)
506 /* cnt never goes below 1 */
507 while (i < type->cnt - 1) {
508 struct memblock_region *this = &type->regions[i];
509 struct memblock_region *next = &type->regions[i + 1];
511 if (this->base + this->size != next->base ||
512 memblock_get_region_node(this) !=
513 memblock_get_region_node(next) ||
514 this->flags != next->flags) {
515 BUG_ON(this->base + this->size > next->base);
520 this->size += next->size;
521 /* move forward from next + 1, index of which is i + 2 */
522 memmove(next, next + 1, (type->cnt - (i + 2)) * sizeof(*next));
528 * memblock_insert_region - insert new memblock region
529 * @type: memblock type to insert into
530 * @idx: index for the insertion point
531 * @base: base address of the new region
532 * @size: size of the new region
533 * @nid: node id of the new region
534 * @flags: flags of the new region
536 * Insert new memblock region [@base, @base + @size) into @type at @idx.
537 * @type must already have extra room to accommodate the new region.
539 static void __init_memblock memblock_insert_region(struct memblock_type *type,
540 int idx, phys_addr_t base,
543 enum memblock_flags flags)
545 struct memblock_region *rgn = &type->regions[idx];
547 BUG_ON(type->cnt >= type->max);
548 memmove(rgn + 1, rgn, (type->cnt - idx) * sizeof(*rgn));
552 memblock_set_region_node(rgn, nid);
554 type->total_size += size;
558 * memblock_add_range - add new memblock region
559 * @type: memblock type to add new region into
560 * @base: base address of the new region
561 * @size: size of the new region
562 * @nid: nid of the new region
563 * @flags: flags of the new region
565 * Add new memblock region [@base, @base + @size) into @type. The new region
566 * is allowed to overlap with existing ones - overlaps don't affect already
567 * existing regions. @type is guaranteed to be minimal (all neighbouring
568 * compatible regions are merged) after the addition.
571 * 0 on success, -errno on failure.
573 static int __init_memblock memblock_add_range(struct memblock_type *type,
574 phys_addr_t base, phys_addr_t size,
575 int nid, enum memblock_flags flags)
578 phys_addr_t obase = base;
579 phys_addr_t end = base + memblock_cap_size(base, &size);
581 struct memblock_region *rgn;
586 /* special case for empty array */
587 if (type->regions[0].size == 0) {
588 WARN_ON(type->cnt != 1 || type->total_size);
589 type->regions[0].base = base;
590 type->regions[0].size = size;
591 type->regions[0].flags = flags;
592 memblock_set_region_node(&type->regions[0], nid);
593 type->total_size = size;
598 * The following is executed twice. Once with %false @insert and
599 * then with %true. The first counts the number of regions needed
600 * to accommodate the new area. The second actually inserts them.
605 for_each_memblock_type(idx, type, rgn) {
606 phys_addr_t rbase = rgn->base;
607 phys_addr_t rend = rbase + rgn->size;
614 * @rgn overlaps. If it separates the lower part of new
615 * area, insert that portion.
619 WARN_ON(nid != memblock_get_region_node(rgn));
621 WARN_ON(flags != rgn->flags);
624 memblock_insert_region(type, idx++, base,
628 /* area below @rend is dealt with, forget about it */
629 base = min(rend, end);
632 /* insert the remaining portion */
636 memblock_insert_region(type, idx, base, end - base,
644 * If this was the first round, resize array and repeat for actual
645 * insertions; otherwise, merge and return.
648 while (type->cnt + nr_new > type->max)
649 if (memblock_double_array(type, obase, size) < 0)
654 memblock_merge_regions(type);
660 * memblock_add_node - add new memblock region within a NUMA node
661 * @base: base address of the new region
662 * @size: size of the new region
663 * @nid: nid of the new region
665 * Add new memblock region [@base, @base + @size) to the "memory"
666 * type. See memblock_add_range() description for mode details
669 * 0 on success, -errno on failure.
671 int __init_memblock memblock_add_node(phys_addr_t base, phys_addr_t size,
674 phys_addr_t end = base + size - 1;
676 memblock_dbg("%s: [%pa-%pa] nid=%d %pS\n", __func__,
677 &base, &end, nid, (void *)_RET_IP_);
679 return memblock_add_range(&memblock.memory, base, size, nid, 0);
683 * memblock_add - add new memblock region
684 * @base: base address of the new region
685 * @size: size of the new region
687 * Add new memblock region [@base, @base + @size) to the "memory"
688 * type. See memblock_add_range() description for mode details
691 * 0 on success, -errno on failure.
693 int __init_memblock memblock_add(phys_addr_t base, phys_addr_t size)
695 phys_addr_t end = base + size - 1;
697 memblock_dbg("%s: [%pa-%pa] %pS\n", __func__,
698 &base, &end, (void *)_RET_IP_);
700 return memblock_add_range(&memblock.memory, base, size, MAX_NUMNODES, 0);
704 * memblock_isolate_range - isolate given range into disjoint memblocks
705 * @type: memblock type to isolate range for
706 * @base: base of range to isolate
707 * @size: size of range to isolate
708 * @start_rgn: out parameter for the start of isolated region
709 * @end_rgn: out parameter for the end of isolated region
711 * Walk @type and ensure that regions don't cross the boundaries defined by
712 * [@base, @base + @size). Crossing regions are split at the boundaries,
713 * which may create at most two more regions. The index of the first
714 * region inside the range is returned in *@start_rgn and end in *@end_rgn.
717 * 0 on success, -errno on failure.
719 static int __init_memblock memblock_isolate_range(struct memblock_type *type,
720 phys_addr_t base, phys_addr_t size,
721 int *start_rgn, int *end_rgn)
723 phys_addr_t end = base + memblock_cap_size(base, &size);
725 struct memblock_region *rgn;
727 *start_rgn = *end_rgn = 0;
732 /* we'll create at most two more regions */
733 while (type->cnt + 2 > type->max)
734 if (memblock_double_array(type, base, size) < 0)
737 for_each_memblock_type(idx, type, rgn) {
738 phys_addr_t rbase = rgn->base;
739 phys_addr_t rend = rbase + rgn->size;
748 * @rgn intersects from below. Split and continue
749 * to process the next region - the new top half.
752 rgn->size -= base - rbase;
753 type->total_size -= base - rbase;
754 memblock_insert_region(type, idx, rbase, base - rbase,
755 memblock_get_region_node(rgn),
757 } else if (rend > end) {
759 * @rgn intersects from above. Split and redo the
760 * current region - the new bottom half.
763 rgn->size -= end - rbase;
764 type->total_size -= end - rbase;
765 memblock_insert_region(type, idx--, rbase, end - rbase,
766 memblock_get_region_node(rgn),
769 /* @rgn is fully contained, record it */
779 static int __init_memblock memblock_remove_range(struct memblock_type *type,
780 phys_addr_t base, phys_addr_t size)
782 int start_rgn, end_rgn;
785 ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
789 for (i = end_rgn - 1; i >= start_rgn; i--)
790 memblock_remove_region(type, i);
794 int __init_memblock memblock_remove(phys_addr_t base, phys_addr_t size)
796 phys_addr_t end = base + size - 1;
798 memblock_dbg("%s: [%pa-%pa] %pS\n", __func__,
799 &base, &end, (void *)_RET_IP_);
801 return memblock_remove_range(&memblock.memory, base, size);
805 * memblock_free_ptr - free boot memory allocation
806 * @ptr: starting address of the boot memory allocation
807 * @size: size of the boot memory block in bytes
809 * Free boot memory block previously allocated by memblock_alloc_xx() API.
810 * The freeing memory will not be released to the buddy allocator.
812 void __init_memblock memblock_free_ptr(void *ptr, size_t size)
815 memblock_free(__pa(ptr), size);
819 * memblock_free - free boot memory block
820 * @base: phys starting address of the boot memory block
821 * @size: size of the boot memory block in bytes
823 * Free boot memory block previously allocated by memblock_alloc_xx() API.
824 * The freeing memory will not be released to the buddy allocator.
826 int __init_memblock memblock_free(phys_addr_t base, phys_addr_t size)
828 phys_addr_t end = base + size - 1;
830 memblock_dbg("%s: [%pa-%pa] %pS\n", __func__,
831 &base, &end, (void *)_RET_IP_);
833 kmemleak_free_part_phys(base, size);
834 return memblock_remove_range(&memblock.reserved, base, size);
837 int __init_memblock memblock_reserve(phys_addr_t base, phys_addr_t size)
839 phys_addr_t end = base + size - 1;
841 memblock_dbg("%s: [%pa-%pa] %pS\n", __func__,
842 &base, &end, (void *)_RET_IP_);
844 return memblock_add_range(&memblock.reserved, base, size, MAX_NUMNODES, 0);
847 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
848 int __init_memblock memblock_physmem_add(phys_addr_t base, phys_addr_t size)
850 phys_addr_t end = base + size - 1;
852 memblock_dbg("%s: [%pa-%pa] %pS\n", __func__,
853 &base, &end, (void *)_RET_IP_);
855 return memblock_add_range(&physmem, base, size, MAX_NUMNODES, 0);
860 * memblock_setclr_flag - set or clear flag for a memory region
861 * @base: base address of the region
862 * @size: size of the region
863 * @set: set or clear the flag
864 * @flag: the flag to update
866 * This function isolates region [@base, @base + @size), and sets/clears flag
868 * Return: 0 on success, -errno on failure.
870 static int __init_memblock memblock_setclr_flag(phys_addr_t base,
871 phys_addr_t size, int set, int flag)
873 struct memblock_type *type = &memblock.memory;
874 int i, ret, start_rgn, end_rgn;
876 ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
880 for (i = start_rgn; i < end_rgn; i++) {
881 struct memblock_region *r = &type->regions[i];
889 memblock_merge_regions(type);
894 * memblock_mark_hotplug - Mark hotpluggable memory with flag MEMBLOCK_HOTPLUG.
895 * @base: the base phys addr of the region
896 * @size: the size of the region
898 * Return: 0 on success, -errno on failure.
900 int __init_memblock memblock_mark_hotplug(phys_addr_t base, phys_addr_t size)
902 return memblock_setclr_flag(base, size, 1, MEMBLOCK_HOTPLUG);
906 * memblock_clear_hotplug - Clear flag MEMBLOCK_HOTPLUG for a specified region.
907 * @base: the base phys addr of the region
908 * @size: the size of the region
910 * Return: 0 on success, -errno on failure.
912 int __init_memblock memblock_clear_hotplug(phys_addr_t base, phys_addr_t size)
914 return memblock_setclr_flag(base, size, 0, MEMBLOCK_HOTPLUG);
918 * memblock_mark_mirror - Mark mirrored memory with flag MEMBLOCK_MIRROR.
919 * @base: the base phys addr of the region
920 * @size: the size of the region
922 * Return: 0 on success, -errno on failure.
924 int __init_memblock memblock_mark_mirror(phys_addr_t base, phys_addr_t size)
926 system_has_some_mirror = true;
928 return memblock_setclr_flag(base, size, 1, MEMBLOCK_MIRROR);
932 * memblock_mark_nomap - Mark a memory region with flag MEMBLOCK_NOMAP.
933 * @base: the base phys addr of the region
934 * @size: the size of the region
936 * The memory regions marked with %MEMBLOCK_NOMAP will not be added to the
937 * direct mapping of the physical memory. These regions will still be
938 * covered by the memory map. The struct page representing NOMAP memory
939 * frames in the memory map will be PageReserved()
941 * Note: if the memory being marked %MEMBLOCK_NOMAP was allocated from
942 * memblock, the caller must inform kmemleak to ignore that memory
944 * Return: 0 on success, -errno on failure.
946 int __init_memblock memblock_mark_nomap(phys_addr_t base, phys_addr_t size)
948 return memblock_setclr_flag(base, size, 1, MEMBLOCK_NOMAP);
952 * memblock_clear_nomap - Clear flag MEMBLOCK_NOMAP for a specified region.
953 * @base: the base phys addr of the region
954 * @size: the size of the region
956 * Return: 0 on success, -errno on failure.
958 int __init_memblock memblock_clear_nomap(phys_addr_t base, phys_addr_t size)
960 return memblock_setclr_flag(base, size, 0, MEMBLOCK_NOMAP);
963 static bool should_skip_region(struct memblock_type *type,
964 struct memblock_region *m,
967 int m_nid = memblock_get_region_node(m);
969 /* we never skip regions when iterating memblock.reserved or physmem */
970 if (type != memblock_memory)
973 /* only memory regions are associated with nodes, check it */
974 if (nid != NUMA_NO_NODE && nid != m_nid)
977 /* skip hotpluggable memory regions if needed */
978 if (movable_node_is_enabled() && memblock_is_hotpluggable(m) &&
979 !(flags & MEMBLOCK_HOTPLUG))
982 /* if we want mirror memory skip non-mirror memory regions */
983 if ((flags & MEMBLOCK_MIRROR) && !memblock_is_mirror(m))
986 /* skip nomap memory unless we were asked for it explicitly */
987 if (!(flags & MEMBLOCK_NOMAP) && memblock_is_nomap(m))
994 * __next_mem_range - next function for for_each_free_mem_range() etc.
995 * @idx: pointer to u64 loop variable
996 * @nid: node selector, %NUMA_NO_NODE for all nodes
997 * @flags: pick from blocks based on memory attributes
998 * @type_a: pointer to memblock_type from where the range is taken
999 * @type_b: pointer to memblock_type which excludes memory from being taken
1000 * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL
1001 * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL
1002 * @out_nid: ptr to int for nid of the range, can be %NULL
1004 * Find the first area from *@idx which matches @nid, fill the out
1005 * parameters, and update *@idx for the next iteration. The lower 32bit of
1006 * *@idx contains index into type_a and the upper 32bit indexes the
1007 * areas before each region in type_b. For example, if type_b regions
1008 * look like the following,
1010 * 0:[0-16), 1:[32-48), 2:[128-130)
1012 * The upper 32bit indexes the following regions.
1014 * 0:[0-0), 1:[16-32), 2:[48-128), 3:[130-MAX)
1016 * As both region arrays are sorted, the function advances the two indices
1017 * in lockstep and returns each intersection.
1019 void __next_mem_range(u64 *idx, int nid, enum memblock_flags flags,
1020 struct memblock_type *type_a,
1021 struct memblock_type *type_b, phys_addr_t *out_start,
1022 phys_addr_t *out_end, int *out_nid)
1024 int idx_a = *idx & 0xffffffff;
1025 int idx_b = *idx >> 32;
1027 if (WARN_ONCE(nid == MAX_NUMNODES,
1028 "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
1031 for (; idx_a < type_a->cnt; idx_a++) {
1032 struct memblock_region *m = &type_a->regions[idx_a];
1034 phys_addr_t m_start = m->base;
1035 phys_addr_t m_end = m->base + m->size;
1036 int m_nid = memblock_get_region_node(m);
1038 if (should_skip_region(type_a, m, nid, flags))
1043 *out_start = m_start;
1049 *idx = (u32)idx_a | (u64)idx_b << 32;
1053 /* scan areas before each reservation */
1054 for (; idx_b < type_b->cnt + 1; idx_b++) {
1055 struct memblock_region *r;
1056 phys_addr_t r_start;
1059 r = &type_b->regions[idx_b];
1060 r_start = idx_b ? r[-1].base + r[-1].size : 0;
1061 r_end = idx_b < type_b->cnt ?
1062 r->base : PHYS_ADDR_MAX;
1065 * if idx_b advanced past idx_a,
1066 * break out to advance idx_a
1068 if (r_start >= m_end)
1070 /* if the two regions intersect, we're done */
1071 if (m_start < r_end) {
1074 max(m_start, r_start);
1076 *out_end = min(m_end, r_end);
1080 * The region which ends first is
1081 * advanced for the next iteration.
1087 *idx = (u32)idx_a | (u64)idx_b << 32;
1093 /* signal end of iteration */
1098 * __next_mem_range_rev - generic next function for for_each_*_range_rev()
1100 * @idx: pointer to u64 loop variable
1101 * @nid: node selector, %NUMA_NO_NODE for all nodes
1102 * @flags: pick from blocks based on memory attributes
1103 * @type_a: pointer to memblock_type from where the range is taken
1104 * @type_b: pointer to memblock_type which excludes memory from being taken
1105 * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL
1106 * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL
1107 * @out_nid: ptr to int for nid of the range, can be %NULL
1109 * Finds the next range from type_a which is not marked as unsuitable
1112 * Reverse of __next_mem_range().
1114 void __init_memblock __next_mem_range_rev(u64 *idx, int nid,
1115 enum memblock_flags flags,
1116 struct memblock_type *type_a,
1117 struct memblock_type *type_b,
1118 phys_addr_t *out_start,
1119 phys_addr_t *out_end, int *out_nid)
1121 int idx_a = *idx & 0xffffffff;
1122 int idx_b = *idx >> 32;
1124 if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
1127 if (*idx == (u64)ULLONG_MAX) {
1128 idx_a = type_a->cnt - 1;
1130 idx_b = type_b->cnt;
1135 for (; idx_a >= 0; idx_a--) {
1136 struct memblock_region *m = &type_a->regions[idx_a];
1138 phys_addr_t m_start = m->base;
1139 phys_addr_t m_end = m->base + m->size;
1140 int m_nid = memblock_get_region_node(m);
1142 if (should_skip_region(type_a, m, nid, flags))
1147 *out_start = m_start;
1153 *idx = (u32)idx_a | (u64)idx_b << 32;
1157 /* scan areas before each reservation */
1158 for (; idx_b >= 0; idx_b--) {
1159 struct memblock_region *r;
1160 phys_addr_t r_start;
1163 r = &type_b->regions[idx_b];
1164 r_start = idx_b ? r[-1].base + r[-1].size : 0;
1165 r_end = idx_b < type_b->cnt ?
1166 r->base : PHYS_ADDR_MAX;
1168 * if idx_b advanced past idx_a,
1169 * break out to advance idx_a
1172 if (r_end <= m_start)
1174 /* if the two regions intersect, we're done */
1175 if (m_end > r_start) {
1177 *out_start = max(m_start, r_start);
1179 *out_end = min(m_end, r_end);
1182 if (m_start >= r_start)
1186 *idx = (u32)idx_a | (u64)idx_b << 32;
1191 /* signal end of iteration */
1196 * Common iterator interface used to define for_each_mem_pfn_range().
1198 void __init_memblock __next_mem_pfn_range(int *idx, int nid,
1199 unsigned long *out_start_pfn,
1200 unsigned long *out_end_pfn, int *out_nid)
1202 struct memblock_type *type = &memblock.memory;
1203 struct memblock_region *r;
1206 while (++*idx < type->cnt) {
1207 r = &type->regions[*idx];
1208 r_nid = memblock_get_region_node(r);
1210 if (PFN_UP(r->base) >= PFN_DOWN(r->base + r->size))
1212 if (nid == MAX_NUMNODES || nid == r_nid)
1215 if (*idx >= type->cnt) {
1221 *out_start_pfn = PFN_UP(r->base);
1223 *out_end_pfn = PFN_DOWN(r->base + r->size);
1229 * memblock_set_node - set node ID on memblock regions
1230 * @base: base of area to set node ID for
1231 * @size: size of area to set node ID for
1232 * @type: memblock type to set node ID for
1233 * @nid: node ID to set
1235 * Set the nid of memblock @type regions in [@base, @base + @size) to @nid.
1236 * Regions which cross the area boundaries are split as necessary.
1239 * 0 on success, -errno on failure.
1241 int __init_memblock memblock_set_node(phys_addr_t base, phys_addr_t size,
1242 struct memblock_type *type, int nid)
1245 int start_rgn, end_rgn;
1248 ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
1252 for (i = start_rgn; i < end_rgn; i++)
1253 memblock_set_region_node(&type->regions[i], nid);
1255 memblock_merge_regions(type);
1260 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1262 * __next_mem_pfn_range_in_zone - iterator for for_each_*_range_in_zone()
1264 * @idx: pointer to u64 loop variable
1265 * @zone: zone in which all of the memory blocks reside
1266 * @out_spfn: ptr to ulong for start pfn of the range, can be %NULL
1267 * @out_epfn: ptr to ulong for end pfn of the range, can be %NULL
1269 * This function is meant to be a zone/pfn specific wrapper for the
1270 * for_each_mem_range type iterators. Specifically they are used in the
1271 * deferred memory init routines and as such we were duplicating much of
1272 * this logic throughout the code. So instead of having it in multiple
1273 * locations it seemed like it would make more sense to centralize this to
1274 * one new iterator that does everything they need.
1276 void __init_memblock
1277 __next_mem_pfn_range_in_zone(u64 *idx, struct zone *zone,
1278 unsigned long *out_spfn, unsigned long *out_epfn)
1280 int zone_nid = zone_to_nid(zone);
1281 phys_addr_t spa, epa;
1284 __next_mem_range(idx, zone_nid, MEMBLOCK_NONE,
1285 &memblock.memory, &memblock.reserved,
1288 while (*idx != U64_MAX) {
1289 unsigned long epfn = PFN_DOWN(epa);
1290 unsigned long spfn = PFN_UP(spa);
1293 * Verify the end is at least past the start of the zone and
1294 * that we have at least one PFN to initialize.
1296 if (zone->zone_start_pfn < epfn && spfn < epfn) {
1297 /* if we went too far just stop searching */
1298 if (zone_end_pfn(zone) <= spfn) {
1304 *out_spfn = max(zone->zone_start_pfn, spfn);
1306 *out_epfn = min(zone_end_pfn(zone), epfn);
1311 __next_mem_range(idx, zone_nid, MEMBLOCK_NONE,
1312 &memblock.memory, &memblock.reserved,
1316 /* signal end of iteration */
1318 *out_spfn = ULONG_MAX;
1323 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
1326 * memblock_alloc_range_nid - allocate boot memory block
1327 * @size: size of memory block to be allocated in bytes
1328 * @align: alignment of the region and block's size
1329 * @start: the lower bound of the memory region to allocate (phys address)
1330 * @end: the upper bound of the memory region to allocate (phys address)
1331 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1332 * @exact_nid: control the allocation fall back to other nodes
1334 * The allocation is performed from memory region limited by
1335 * memblock.current_limit if @end == %MEMBLOCK_ALLOC_ACCESSIBLE.
1337 * If the specified node can not hold the requested memory and @exact_nid
1338 * is false, the allocation falls back to any node in the system.
1340 * For systems with memory mirroring, the allocation is attempted first
1341 * from the regions with mirroring enabled and then retried from any
1344 * In addition, function sets the min_count to 0 using kmemleak_alloc_phys for
1345 * allocated boot memory block, so that it is never reported as leaks.
1348 * Physical address of allocated memory block on success, %0 on failure.
1350 phys_addr_t __init memblock_alloc_range_nid(phys_addr_t size,
1351 phys_addr_t align, phys_addr_t start,
1352 phys_addr_t end, int nid,
1355 enum memblock_flags flags = choose_memblock_flags();
1358 if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
1362 /* Can't use WARNs this early in boot on powerpc */
1364 align = SMP_CACHE_BYTES;
1368 found = memblock_find_in_range_node(size, align, start, end, nid,
1370 if (found && !memblock_reserve(found, size))
1373 if (nid != NUMA_NO_NODE && !exact_nid) {
1374 found = memblock_find_in_range_node(size, align, start,
1377 if (found && !memblock_reserve(found, size))
1381 if (flags & MEMBLOCK_MIRROR) {
1382 flags &= ~MEMBLOCK_MIRROR;
1383 pr_warn("Could not allocate %pap bytes of mirrored memory\n",
1391 /* Skip kmemleak for kasan_init() due to high volume. */
1392 if (end != MEMBLOCK_ALLOC_KASAN)
1394 * The min_count is set to 0 so that memblock allocated
1395 * blocks are never reported as leaks. This is because many
1396 * of these blocks are only referred via the physical
1397 * address which is not looked up by kmemleak.
1399 kmemleak_alloc_phys(found, size, 0, 0);
1405 * memblock_phys_alloc_range - allocate a memory block inside specified range
1406 * @size: size of memory block to be allocated in bytes
1407 * @align: alignment of the region and block's size
1408 * @start: the lower bound of the memory region to allocate (physical address)
1409 * @end: the upper bound of the memory region to allocate (physical address)
1411 * Allocate @size bytes in the between @start and @end.
1413 * Return: physical address of the allocated memory block on success,
1416 phys_addr_t __init memblock_phys_alloc_range(phys_addr_t size,
1421 memblock_dbg("%s: %llu bytes align=0x%llx from=%pa max_addr=%pa %pS\n",
1422 __func__, (u64)size, (u64)align, &start, &end,
1424 return memblock_alloc_range_nid(size, align, start, end, NUMA_NO_NODE,
1429 * memblock_phys_alloc_try_nid - allocate a memory block from specified NUMA node
1430 * @size: size of memory block to be allocated in bytes
1431 * @align: alignment of the region and block's size
1432 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1434 * Allocates memory block from the specified NUMA node. If the node
1435 * has no available memory, attempts to allocated from any node in the
1438 * Return: physical address of the allocated memory block on success,
1441 phys_addr_t __init memblock_phys_alloc_try_nid(phys_addr_t size, phys_addr_t align, int nid)
1443 return memblock_alloc_range_nid(size, align, 0,
1444 MEMBLOCK_ALLOC_ACCESSIBLE, nid, false);
1448 * memblock_alloc_internal - allocate boot memory block
1449 * @size: size of memory block to be allocated in bytes
1450 * @align: alignment of the region and block's size
1451 * @min_addr: the lower bound of the memory region to allocate (phys address)
1452 * @max_addr: the upper bound of the memory region to allocate (phys address)
1453 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1454 * @exact_nid: control the allocation fall back to other nodes
1456 * Allocates memory block using memblock_alloc_range_nid() and
1457 * converts the returned physical address to virtual.
1459 * The @min_addr limit is dropped if it can not be satisfied and the allocation
1460 * will fall back to memory below @min_addr. Other constraints, such
1461 * as node and mirrored memory will be handled again in
1462 * memblock_alloc_range_nid().
1465 * Virtual address of allocated memory block on success, NULL on failure.
1467 static void * __init memblock_alloc_internal(
1468 phys_addr_t size, phys_addr_t align,
1469 phys_addr_t min_addr, phys_addr_t max_addr,
1470 int nid, bool exact_nid)
1475 * Detect any accidental use of these APIs after slab is ready, as at
1476 * this moment memblock may be deinitialized already and its
1477 * internal data may be destroyed (after execution of memblock_free_all)
1479 if (WARN_ON_ONCE(slab_is_available()))
1480 return kzalloc_node(size, GFP_NOWAIT, nid);
1482 if (max_addr > memblock.current_limit)
1483 max_addr = memblock.current_limit;
1485 alloc = memblock_alloc_range_nid(size, align, min_addr, max_addr, nid,
1488 /* retry allocation without lower limit */
1489 if (!alloc && min_addr)
1490 alloc = memblock_alloc_range_nid(size, align, 0, max_addr, nid,
1496 return phys_to_virt(alloc);
1500 * memblock_alloc_exact_nid_raw - allocate boot memory block on the exact node
1501 * without zeroing memory
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 from where the allocation
1505 * is preferred (phys address)
1506 * @max_addr: the upper bound of the memory region from where the allocation
1507 * is preferred (phys address), or %MEMBLOCK_ALLOC_ACCESSIBLE to
1508 * allocate only from memory limited by memblock.current_limit value
1509 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1511 * Public function, provides additional debug information (including caller
1512 * info), if enabled. Does not zero allocated memory.
1515 * Virtual address of allocated memory block on success, NULL on failure.
1517 void * __init memblock_alloc_exact_nid_raw(
1518 phys_addr_t size, phys_addr_t align,
1519 phys_addr_t min_addr, phys_addr_t max_addr,
1522 memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=%pa max_addr=%pa %pS\n",
1523 __func__, (u64)size, (u64)align, nid, &min_addr,
1524 &max_addr, (void *)_RET_IP_);
1526 return memblock_alloc_internal(size, align, min_addr, max_addr, nid,
1531 * memblock_alloc_try_nid_raw - allocate boot memory block without zeroing
1532 * memory and without panicking
1533 * @size: size of memory block to be allocated in bytes
1534 * @align: alignment of the region and block's size
1535 * @min_addr: the lower bound of the memory region from where the allocation
1536 * is preferred (phys address)
1537 * @max_addr: the upper bound of the memory region from where the allocation
1538 * is preferred (phys address), or %MEMBLOCK_ALLOC_ACCESSIBLE to
1539 * allocate only from memory limited by memblock.current_limit value
1540 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1542 * Public function, provides additional debug information (including caller
1543 * info), if enabled. Does not zero allocated memory, does not panic if request
1544 * cannot be satisfied.
1547 * Virtual address of allocated memory block on success, NULL on failure.
1549 void * __init memblock_alloc_try_nid_raw(
1550 phys_addr_t size, phys_addr_t align,
1551 phys_addr_t min_addr, phys_addr_t max_addr,
1554 memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=%pa max_addr=%pa %pS\n",
1555 __func__, (u64)size, (u64)align, nid, &min_addr,
1556 &max_addr, (void *)_RET_IP_);
1558 return memblock_alloc_internal(size, align, min_addr, max_addr, nid,
1563 * memblock_alloc_try_nid - allocate boot memory block
1564 * @size: size of memory block to be allocated in bytes
1565 * @align: alignment of the region and block's size
1566 * @min_addr: the lower bound of the memory region from where the allocation
1567 * is preferred (phys address)
1568 * @max_addr: the upper bound of the memory region from where the allocation
1569 * is preferred (phys address), or %MEMBLOCK_ALLOC_ACCESSIBLE to
1570 * allocate only from memory limited by memblock.current_limit value
1571 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1573 * Public function, provides additional debug information (including caller
1574 * info), if enabled. This function zeroes the allocated memory.
1577 * Virtual address of allocated memory block on success, NULL on failure.
1579 void * __init memblock_alloc_try_nid(
1580 phys_addr_t size, phys_addr_t align,
1581 phys_addr_t min_addr, phys_addr_t max_addr,
1586 memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=%pa max_addr=%pa %pS\n",
1587 __func__, (u64)size, (u64)align, nid, &min_addr,
1588 &max_addr, (void *)_RET_IP_);
1589 ptr = memblock_alloc_internal(size, align,
1590 min_addr, max_addr, nid, false);
1592 memset(ptr, 0, size);
1598 * __memblock_free_late - free pages directly to buddy allocator
1599 * @base: phys starting address of the boot memory block
1600 * @size: size of the boot memory block in bytes
1602 * This is only useful when the memblock allocator has already been torn
1603 * down, but we are still initializing the system. Pages are released directly
1604 * to the buddy allocator.
1606 void __init __memblock_free_late(phys_addr_t base, phys_addr_t size)
1608 phys_addr_t cursor, end;
1610 end = base + size - 1;
1611 memblock_dbg("%s: [%pa-%pa] %pS\n",
1612 __func__, &base, &end, (void *)_RET_IP_);
1613 kmemleak_free_part_phys(base, size);
1614 cursor = PFN_UP(base);
1615 end = PFN_DOWN(base + size);
1617 for (; cursor < end; cursor++) {
1619 * Reserved pages are always initialized by the end of
1620 * memblock_free_all() (by memmap_init() and, if deferred
1621 * initialization is enabled, memmap_init_reserved_pages()), so
1622 * these pages can be released directly to the buddy allocator.
1624 __free_pages_core(pfn_to_page(cursor), 0);
1625 totalram_pages_inc();
1630 * Remaining API functions
1633 phys_addr_t __init_memblock memblock_phys_mem_size(void)
1635 return memblock.memory.total_size;
1638 phys_addr_t __init_memblock memblock_reserved_size(void)
1640 return memblock.reserved.total_size;
1643 /* lowest address */
1644 phys_addr_t __init_memblock memblock_start_of_DRAM(void)
1646 return memblock.memory.regions[0].base;
1649 phys_addr_t __init_memblock memblock_end_of_DRAM(void)
1651 int idx = memblock.memory.cnt - 1;
1653 return (memblock.memory.regions[idx].base + memblock.memory.regions[idx].size);
1656 static phys_addr_t __init_memblock __find_max_addr(phys_addr_t limit)
1658 phys_addr_t max_addr = PHYS_ADDR_MAX;
1659 struct memblock_region *r;
1662 * translate the memory @limit size into the max address within one of
1663 * the memory memblock regions, if the @limit exceeds the total size
1664 * of those regions, max_addr will keep original value PHYS_ADDR_MAX
1666 for_each_mem_region(r) {
1667 if (limit <= r->size) {
1668 max_addr = r->base + limit;
1677 void __init memblock_enforce_memory_limit(phys_addr_t limit)
1679 phys_addr_t max_addr;
1684 max_addr = __find_max_addr(limit);
1686 /* @limit exceeds the total size of the memory, do nothing */
1687 if (max_addr == PHYS_ADDR_MAX)
1690 /* truncate both memory and reserved regions */
1691 memblock_remove_range(&memblock.memory, max_addr,
1693 memblock_remove_range(&memblock.reserved, max_addr,
1697 void __init memblock_cap_memory_range(phys_addr_t base, phys_addr_t size)
1699 int start_rgn, end_rgn;
1705 if (!memblock_memory->total_size) {
1706 pr_warn("%s: No memory registered yet\n", __func__);
1710 ret = memblock_isolate_range(&memblock.memory, base, size,
1711 &start_rgn, &end_rgn);
1715 /* remove all the MAP regions */
1716 for (i = memblock.memory.cnt - 1; i >= end_rgn; i--)
1717 if (!memblock_is_nomap(&memblock.memory.regions[i]))
1718 memblock_remove_region(&memblock.memory, i);
1720 for (i = start_rgn - 1; i >= 0; i--)
1721 if (!memblock_is_nomap(&memblock.memory.regions[i]))
1722 memblock_remove_region(&memblock.memory, i);
1724 /* truncate the reserved regions */
1725 memblock_remove_range(&memblock.reserved, 0, base);
1726 memblock_remove_range(&memblock.reserved,
1727 base + size, PHYS_ADDR_MAX);
1730 void __init memblock_mem_limit_remove_map(phys_addr_t limit)
1732 phys_addr_t max_addr;
1737 max_addr = __find_max_addr(limit);
1739 /* @limit exceeds the total size of the memory, do nothing */
1740 if (max_addr == PHYS_ADDR_MAX)
1743 memblock_cap_memory_range(0, max_addr);
1746 static int __init_memblock memblock_search(struct memblock_type *type, phys_addr_t addr)
1748 unsigned int left = 0, right = type->cnt;
1751 unsigned int mid = (right + left) / 2;
1753 if (addr < type->regions[mid].base)
1755 else if (addr >= (type->regions[mid].base +
1756 type->regions[mid].size))
1760 } while (left < right);
1764 bool __init_memblock memblock_is_reserved(phys_addr_t addr)
1766 return memblock_search(&memblock.reserved, addr) != -1;
1769 bool __init_memblock memblock_is_memory(phys_addr_t addr)
1771 return memblock_search(&memblock.memory, addr) != -1;
1774 bool __init_memblock memblock_is_map_memory(phys_addr_t addr)
1776 int i = memblock_search(&memblock.memory, addr);
1780 return !memblock_is_nomap(&memblock.memory.regions[i]);
1783 int __init_memblock memblock_search_pfn_nid(unsigned long pfn,
1784 unsigned long *start_pfn, unsigned long *end_pfn)
1786 struct memblock_type *type = &memblock.memory;
1787 int mid = memblock_search(type, PFN_PHYS(pfn));
1792 *start_pfn = PFN_DOWN(type->regions[mid].base);
1793 *end_pfn = PFN_DOWN(type->regions[mid].base + type->regions[mid].size);
1795 return memblock_get_region_node(&type->regions[mid]);
1799 * memblock_is_region_memory - check if a region is a subset of memory
1800 * @base: base of region to check
1801 * @size: size of region to check
1803 * Check if the region [@base, @base + @size) is a subset of a memory block.
1806 * 0 if false, non-zero if true
1808 bool __init_memblock memblock_is_region_memory(phys_addr_t base, phys_addr_t size)
1810 int idx = memblock_search(&memblock.memory, base);
1811 phys_addr_t end = base + memblock_cap_size(base, &size);
1815 return (memblock.memory.regions[idx].base +
1816 memblock.memory.regions[idx].size) >= end;
1820 * memblock_is_region_reserved - check if a region intersects reserved memory
1821 * @base: base of region to check
1822 * @size: size of region to check
1824 * Check if the region [@base, @base + @size) intersects a reserved
1828 * True if they intersect, false if not.
1830 bool __init_memblock memblock_is_region_reserved(phys_addr_t base, phys_addr_t size)
1832 return memblock_overlaps_region(&memblock.reserved, base, size);
1835 void __init_memblock memblock_trim_memory(phys_addr_t align)
1837 phys_addr_t start, end, orig_start, orig_end;
1838 struct memblock_region *r;
1840 for_each_mem_region(r) {
1841 orig_start = r->base;
1842 orig_end = r->base + r->size;
1843 start = round_up(orig_start, align);
1844 end = round_down(orig_end, align);
1846 if (start == orig_start && end == orig_end)
1851 r->size = end - start;
1853 memblock_remove_region(&memblock.memory,
1854 r - memblock.memory.regions);
1860 void __init_memblock memblock_set_current_limit(phys_addr_t limit)
1862 memblock.current_limit = limit;
1865 phys_addr_t __init_memblock memblock_get_current_limit(void)
1867 return memblock.current_limit;
1870 static void __init_memblock memblock_dump(struct memblock_type *type)
1872 phys_addr_t base, end, size;
1873 enum memblock_flags flags;
1875 struct memblock_region *rgn;
1877 pr_info(" %s.cnt = 0x%lx\n", type->name, type->cnt);
1879 for_each_memblock_type(idx, type, rgn) {
1880 char nid_buf[32] = "";
1884 end = base + size - 1;
1887 if (memblock_get_region_node(rgn) != MAX_NUMNODES)
1888 snprintf(nid_buf, sizeof(nid_buf), " on node %d",
1889 memblock_get_region_node(rgn));
1891 pr_info(" %s[%#x]\t[%pa-%pa], %pa bytes%s flags: %#x\n",
1892 type->name, idx, &base, &end, &size, nid_buf, flags);
1896 static void __init_memblock __memblock_dump_all(void)
1898 pr_info("MEMBLOCK configuration:\n");
1899 pr_info(" memory size = %pa reserved size = %pa\n",
1900 &memblock.memory.total_size,
1901 &memblock.reserved.total_size);
1903 memblock_dump(&memblock.memory);
1904 memblock_dump(&memblock.reserved);
1905 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
1906 memblock_dump(&physmem);
1910 void __init_memblock memblock_dump_all(void)
1913 __memblock_dump_all();
1916 void __init memblock_allow_resize(void)
1918 memblock_can_resize = 1;
1921 static int __init early_memblock(char *p)
1923 if (p && strstr(p, "debug"))
1927 early_param("memblock", early_memblock);
1929 static void __init free_memmap(unsigned long start_pfn, unsigned long end_pfn)
1931 struct page *start_pg, *end_pg;
1932 phys_addr_t pg, pgend;
1935 * Convert start_pfn/end_pfn to a struct page pointer.
1937 start_pg = pfn_to_page(start_pfn - 1) + 1;
1938 end_pg = pfn_to_page(end_pfn - 1) + 1;
1941 * Convert to physical addresses, and round start upwards and end
1944 pg = PAGE_ALIGN(__pa(start_pg));
1945 pgend = __pa(end_pg) & PAGE_MASK;
1948 * If there are free pages between these, free the section of the
1952 memblock_free(pg, pgend - pg);
1956 * The mem_map array can get very big. Free the unused area of the memory map.
1958 static void __init free_unused_memmap(void)
1960 unsigned long start, end, prev_end = 0;
1963 if (!IS_ENABLED(CONFIG_HAVE_ARCH_PFN_VALID) ||
1964 IS_ENABLED(CONFIG_SPARSEMEM_VMEMMAP))
1968 * This relies on each bank being in address order.
1969 * The banks are sorted previously in bootmem_init().
1971 for_each_mem_pfn_range(i, MAX_NUMNODES, &start, &end, NULL) {
1972 #ifdef CONFIG_SPARSEMEM
1974 * Take care not to free memmap entries that don't exist
1975 * due to SPARSEMEM sections which aren't present.
1977 start = min(start, ALIGN(prev_end, PAGES_PER_SECTION));
1980 * Align down here since many operations in VM subsystem
1981 * presume that there are no holes in the memory map inside
1984 start = round_down(start, pageblock_nr_pages);
1987 * If we had a previous bank, and there is a space
1988 * between the current bank and the previous, free it.
1990 if (prev_end && prev_end < start)
1991 free_memmap(prev_end, start);
1994 * Align up here since many operations in VM subsystem
1995 * presume that there are no holes in the memory map inside
1998 prev_end = ALIGN(end, pageblock_nr_pages);
2001 #ifdef CONFIG_SPARSEMEM
2002 if (!IS_ALIGNED(prev_end, PAGES_PER_SECTION)) {
2003 prev_end = ALIGN(end, pageblock_nr_pages);
2004 free_memmap(prev_end, ALIGN(prev_end, PAGES_PER_SECTION));
2009 static void __init __free_pages_memory(unsigned long start, unsigned long end)
2013 while (start < end) {
2014 order = min(MAX_ORDER - 1UL, __ffs(start));
2016 while (start + (1UL << order) > end)
2019 memblock_free_pages(pfn_to_page(start), start, order);
2021 start += (1UL << order);
2025 static unsigned long __init __free_memory_core(phys_addr_t start,
2028 unsigned long start_pfn = PFN_UP(start);
2029 unsigned long end_pfn = min_t(unsigned long,
2030 PFN_DOWN(end), max_low_pfn);
2032 if (start_pfn >= end_pfn)
2035 __free_pages_memory(start_pfn, end_pfn);
2037 return end_pfn - start_pfn;
2040 static void __init memmap_init_reserved_pages(void)
2042 struct memblock_region *region;
2043 phys_addr_t start, end;
2046 /* initialize struct pages for the reserved regions */
2047 for_each_reserved_mem_range(i, &start, &end)
2048 reserve_bootmem_region(start, end);
2050 /* and also treat struct pages for the NOMAP regions as PageReserved */
2051 for_each_mem_region(region) {
2052 if (memblock_is_nomap(region)) {
2053 start = region->base;
2054 end = start + region->size;
2055 reserve_bootmem_region(start, end);
2060 static unsigned long __init free_low_memory_core_early(void)
2062 unsigned long count = 0;
2063 phys_addr_t start, end;
2066 memblock_clear_hotplug(0, -1);
2068 memmap_init_reserved_pages();
2071 * We need to use NUMA_NO_NODE instead of NODE_DATA(0)->node_id
2072 * because in some case like Node0 doesn't have RAM installed
2073 * low ram will be on Node1
2075 for_each_free_mem_range(i, NUMA_NO_NODE, MEMBLOCK_NONE, &start, &end,
2077 count += __free_memory_core(start, end);
2082 static int reset_managed_pages_done __initdata;
2084 void reset_node_managed_pages(pg_data_t *pgdat)
2088 for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++)
2089 atomic_long_set(&z->managed_pages, 0);
2092 void __init reset_all_zones_managed_pages(void)
2094 struct pglist_data *pgdat;
2096 if (reset_managed_pages_done)
2099 for_each_online_pgdat(pgdat)
2100 reset_node_managed_pages(pgdat);
2102 reset_managed_pages_done = 1;
2106 * memblock_free_all - release free pages to the buddy allocator
2108 void __init memblock_free_all(void)
2110 unsigned long pages;
2112 free_unused_memmap();
2113 reset_all_zones_managed_pages();
2115 pages = free_low_memory_core_early();
2116 totalram_pages_add(pages);
2119 #if defined(CONFIG_DEBUG_FS) && defined(CONFIG_ARCH_KEEP_MEMBLOCK)
2121 static int memblock_debug_show(struct seq_file *m, void *private)
2123 struct memblock_type *type = m->private;
2124 struct memblock_region *reg;
2128 for (i = 0; i < type->cnt; i++) {
2129 reg = &type->regions[i];
2130 end = reg->base + reg->size - 1;
2132 seq_printf(m, "%4d: ", i);
2133 seq_printf(m, "%pa..%pa\n", ®->base, &end);
2137 DEFINE_SHOW_ATTRIBUTE(memblock_debug);
2139 static int __init memblock_init_debugfs(void)
2141 struct dentry *root = debugfs_create_dir("memblock", NULL);
2143 debugfs_create_file("memory", 0444, root,
2144 &memblock.memory, &memblock_debug_fops);
2145 debugfs_create_file("reserved", 0444, root,
2146 &memblock.reserved, &memblock_debug_fops);
2147 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
2148 debugfs_create_file("physmem", 0444, root, &physmem,
2149 &memblock_debug_fops);
2154 __initcall(memblock_init_debugfs);
2156 #endif /* CONFIG_DEBUG_FS */