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
32 #ifndef INIT_MEMBLOCK_MEMORY_REGIONS
33 #define INIT_MEMBLOCK_MEMORY_REGIONS INIT_MEMBLOCK_REGIONS
37 * DOC: memblock overview
39 * Memblock is a method of managing memory regions during the early
40 * boot period when the usual kernel memory allocators are not up and
43 * Memblock views the system memory as collections of contiguous
44 * regions. There are several types of these collections:
46 * * ``memory`` - describes the physical memory available to the
47 * kernel; this may differ from the actual physical memory installed
48 * in the system, for instance when the memory is restricted with
49 * ``mem=`` command line parameter
50 * * ``reserved`` - describes the regions that were allocated
51 * * ``physmem`` - describes the actual physical memory available during
52 * boot regardless of the possible restrictions and memory hot(un)plug;
53 * the ``physmem`` type is only available on some architectures.
55 * Each region is represented by struct memblock_region that
56 * defines the region extents, its attributes and NUMA node id on NUMA
57 * systems. Every memory type is described by the struct memblock_type
58 * which contains an array of memory regions along with
59 * the allocator metadata. The "memory" and "reserved" types are nicely
60 * wrapped with struct memblock. This structure is statically
61 * initialized at build time. The region arrays are initially sized to
62 * %INIT_MEMBLOCK_MEMORY_REGIONS for "memory" and
63 * %INIT_MEMBLOCK_RESERVED_REGIONS for "reserved". The region array
64 * for "physmem" is initially sized to %INIT_PHYSMEM_REGIONS.
65 * The memblock_allow_resize() enables automatic resizing of the region
66 * arrays during addition of new regions. This feature should be used
67 * with care so that memory allocated for the region array will not
68 * overlap with areas that should be reserved, for example initrd.
70 * The early architecture setup should tell memblock what the physical
71 * memory layout is by using memblock_add() or memblock_add_node()
72 * functions. The first function does not assign the region to a NUMA
73 * node and it is appropriate for UMA systems. Yet, it is possible to
74 * use it on NUMA systems as well and assign the region to a NUMA node
75 * later in the setup process using memblock_set_node(). The
76 * memblock_add_node() performs such an assignment directly.
78 * Once memblock is setup the memory can be allocated using one of the
81 * * memblock_phys_alloc*() - these functions return the **physical**
82 * address of the allocated memory
83 * * memblock_alloc*() - these functions return the **virtual** address
84 * of the allocated memory.
86 * Note, that both API variants use implicit assumptions about allowed
87 * memory ranges and the fallback methods. Consult the documentation
88 * of memblock_alloc_internal() and memblock_alloc_range_nid()
89 * functions for more elaborate description.
91 * As the system boot progresses, the architecture specific mem_init()
92 * function frees all the memory to the buddy page allocator.
94 * Unless an architecture enables %CONFIG_ARCH_KEEP_MEMBLOCK, the
95 * memblock data structures (except "physmem") will be discarded after the
96 * system initialization completes.
100 struct pglist_data __refdata contig_page_data;
101 EXPORT_SYMBOL(contig_page_data);
104 unsigned long max_low_pfn;
105 unsigned long min_low_pfn;
106 unsigned long max_pfn;
107 unsigned long long max_possible_pfn;
109 static struct memblock_region memblock_memory_init_regions[INIT_MEMBLOCK_MEMORY_REGIONS] __initdata_memblock;
110 static struct memblock_region memblock_reserved_init_regions[INIT_MEMBLOCK_RESERVED_REGIONS] __initdata_memblock;
111 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
112 static struct memblock_region memblock_physmem_init_regions[INIT_PHYSMEM_REGIONS];
115 struct memblock memblock __initdata_memblock = {
116 .memory.regions = memblock_memory_init_regions,
117 .memory.cnt = 1, /* empty dummy entry */
118 .memory.max = INIT_MEMBLOCK_MEMORY_REGIONS,
119 .memory.name = "memory",
121 .reserved.regions = memblock_reserved_init_regions,
122 .reserved.cnt = 1, /* empty dummy entry */
123 .reserved.max = INIT_MEMBLOCK_RESERVED_REGIONS,
124 .reserved.name = "reserved",
127 .current_limit = MEMBLOCK_ALLOC_ANYWHERE,
130 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
131 struct memblock_type physmem = {
132 .regions = memblock_physmem_init_regions,
133 .cnt = 1, /* empty dummy entry */
134 .max = INIT_PHYSMEM_REGIONS,
140 * keep a pointer to &memblock.memory in the text section to use it in
141 * __next_mem_range() and its helpers.
142 * For architectures that do not keep memblock data after init, this
143 * pointer will be reset to NULL at memblock_discard()
145 static __refdata struct memblock_type *memblock_memory = &memblock.memory;
147 #define for_each_memblock_type(i, memblock_type, rgn) \
148 for (i = 0, rgn = &memblock_type->regions[0]; \
149 i < memblock_type->cnt; \
150 i++, rgn = &memblock_type->regions[i])
152 #define memblock_dbg(fmt, ...) \
154 if (memblock_debug) \
155 pr_info(fmt, ##__VA_ARGS__); \
158 static int memblock_debug __initdata_memblock;
159 static bool system_has_some_mirror __initdata_memblock = false;
160 static int memblock_can_resize __initdata_memblock;
161 static int memblock_memory_in_slab __initdata_memblock = 0;
162 static int memblock_reserved_in_slab __initdata_memblock = 0;
164 static enum memblock_flags __init_memblock choose_memblock_flags(void)
166 return system_has_some_mirror ? MEMBLOCK_MIRROR : MEMBLOCK_NONE;
169 /* adjust *@size so that (@base + *@size) doesn't overflow, return new size */
170 static inline phys_addr_t memblock_cap_size(phys_addr_t base, phys_addr_t *size)
172 return *size = min(*size, PHYS_ADDR_MAX - base);
176 * Address comparison utilities
178 static unsigned long __init_memblock memblock_addrs_overlap(phys_addr_t base1, phys_addr_t size1,
179 phys_addr_t base2, phys_addr_t size2)
181 return ((base1 < (base2 + size2)) && (base2 < (base1 + size1)));
184 bool __init_memblock memblock_overlaps_region(struct memblock_type *type,
185 phys_addr_t base, phys_addr_t size)
189 memblock_cap_size(base, &size);
191 for (i = 0; i < type->cnt; i++)
192 if (memblock_addrs_overlap(base, size, type->regions[i].base,
193 type->regions[i].size))
195 return i < type->cnt;
199 * __memblock_find_range_bottom_up - find free area utility in bottom-up
200 * @start: start of candidate range
201 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_ANYWHERE or
202 * %MEMBLOCK_ALLOC_ACCESSIBLE
203 * @size: size of free area to find
204 * @align: alignment of free area to find
205 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
206 * @flags: pick from blocks based on memory attributes
208 * Utility called from memblock_find_in_range_node(), find free area bottom-up.
211 * Found address on success, 0 on failure.
213 static phys_addr_t __init_memblock
214 __memblock_find_range_bottom_up(phys_addr_t start, phys_addr_t end,
215 phys_addr_t size, phys_addr_t align, int nid,
216 enum memblock_flags flags)
218 phys_addr_t this_start, this_end, cand;
221 for_each_free_mem_range(i, nid, flags, &this_start, &this_end, NULL) {
222 this_start = clamp(this_start, start, end);
223 this_end = clamp(this_end, start, end);
225 cand = round_up(this_start, align);
226 if (cand < this_end && this_end - cand >= size)
234 * __memblock_find_range_top_down - find free area utility, in top-down
235 * @start: start of candidate range
236 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_ANYWHERE or
237 * %MEMBLOCK_ALLOC_ACCESSIBLE
238 * @size: size of free area to find
239 * @align: alignment of free area to find
240 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
241 * @flags: pick from blocks based on memory attributes
243 * Utility called from memblock_find_in_range_node(), find free area top-down.
246 * Found address on success, 0 on failure.
248 static phys_addr_t __init_memblock
249 __memblock_find_range_top_down(phys_addr_t start, phys_addr_t end,
250 phys_addr_t size, phys_addr_t align, int nid,
251 enum memblock_flags flags)
253 phys_addr_t this_start, this_end, cand;
256 for_each_free_mem_range_reverse(i, nid, flags, &this_start, &this_end,
258 this_start = clamp(this_start, start, end);
259 this_end = clamp(this_end, start, end);
264 cand = round_down(this_end - size, align);
265 if (cand >= this_start)
273 * memblock_find_in_range_node - find free area in given range and node
274 * @size: size of free area to find
275 * @align: alignment of free area to find
276 * @start: start of candidate range
277 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_ANYWHERE or
278 * %MEMBLOCK_ALLOC_ACCESSIBLE
279 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
280 * @flags: pick from blocks based on memory attributes
282 * Find @size free area aligned to @align in the specified range and node.
285 * Found address on success, 0 on failure.
287 static phys_addr_t __init_memblock memblock_find_in_range_node(phys_addr_t size,
288 phys_addr_t align, phys_addr_t start,
289 phys_addr_t end, int nid,
290 enum memblock_flags flags)
293 if (end == MEMBLOCK_ALLOC_ACCESSIBLE ||
294 end == MEMBLOCK_ALLOC_NOLEAKTRACE)
295 end = memblock.current_limit;
297 /* avoid allocating the first page */
298 start = max_t(phys_addr_t, start, PAGE_SIZE);
299 end = max(start, end);
301 if (memblock_bottom_up())
302 return __memblock_find_range_bottom_up(start, end, size, align,
305 return __memblock_find_range_top_down(start, end, size, align,
310 * memblock_find_in_range - find free area in given range
311 * @start: start of candidate range
312 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_ANYWHERE or
313 * %MEMBLOCK_ALLOC_ACCESSIBLE
314 * @size: size of free area to find
315 * @align: alignment of free area to find
317 * Find @size free area aligned to @align in the specified range.
320 * Found address on success, 0 on failure.
322 static phys_addr_t __init_memblock memblock_find_in_range(phys_addr_t start,
323 phys_addr_t end, phys_addr_t size,
327 enum memblock_flags flags = choose_memblock_flags();
330 ret = memblock_find_in_range_node(size, align, start, end,
331 NUMA_NO_NODE, flags);
333 if (!ret && (flags & MEMBLOCK_MIRROR)) {
334 pr_warn_ratelimited("Could not allocate %pap bytes of mirrored memory\n",
336 flags &= ~MEMBLOCK_MIRROR;
343 static void __init_memblock memblock_remove_region(struct memblock_type *type, unsigned long r)
345 type->total_size -= type->regions[r].size;
346 memmove(&type->regions[r], &type->regions[r + 1],
347 (type->cnt - (r + 1)) * sizeof(type->regions[r]));
350 /* Special case for empty arrays */
351 if (type->cnt == 0) {
352 WARN_ON(type->total_size != 0);
354 type->regions[0].base = 0;
355 type->regions[0].size = 0;
356 type->regions[0].flags = 0;
357 memblock_set_region_node(&type->regions[0], MAX_NUMNODES);
361 #ifndef CONFIG_ARCH_KEEP_MEMBLOCK
363 * memblock_discard - discard memory and reserved arrays if they were allocated
365 void __init memblock_discard(void)
367 phys_addr_t addr, size;
369 if (memblock.reserved.regions != memblock_reserved_init_regions) {
370 addr = __pa(memblock.reserved.regions);
371 size = PAGE_ALIGN(sizeof(struct memblock_region) *
372 memblock.reserved.max);
373 if (memblock_reserved_in_slab)
374 kfree(memblock.reserved.regions);
376 memblock_free_late(addr, size);
379 if (memblock.memory.regions != memblock_memory_init_regions) {
380 addr = __pa(memblock.memory.regions);
381 size = PAGE_ALIGN(sizeof(struct memblock_region) *
382 memblock.memory.max);
383 if (memblock_memory_in_slab)
384 kfree(memblock.memory.regions);
386 memblock_free_late(addr, size);
389 memblock_memory = NULL;
394 * memblock_double_array - double the size of the memblock regions array
395 * @type: memblock type of the regions array being doubled
396 * @new_area_start: starting address of memory range to avoid overlap with
397 * @new_area_size: size of memory range to avoid overlap with
399 * Double the size of the @type regions array. If memblock is being used to
400 * allocate memory for a new reserved regions array and there is a previously
401 * allocated memory range [@new_area_start, @new_area_start + @new_area_size]
402 * waiting to be reserved, ensure the memory used by the new array does
406 * 0 on success, -1 on failure.
408 static int __init_memblock memblock_double_array(struct memblock_type *type,
409 phys_addr_t new_area_start,
410 phys_addr_t new_area_size)
412 struct memblock_region *new_array, *old_array;
413 phys_addr_t old_alloc_size, new_alloc_size;
414 phys_addr_t old_size, new_size, addr, new_end;
415 int use_slab = slab_is_available();
418 /* We don't allow resizing until we know about the reserved regions
419 * of memory that aren't suitable for allocation
421 if (!memblock_can_resize)
424 /* Calculate new doubled size */
425 old_size = type->max * sizeof(struct memblock_region);
426 new_size = old_size << 1;
428 * We need to allocated new one align to PAGE_SIZE,
429 * so we can free them completely later.
431 old_alloc_size = PAGE_ALIGN(old_size);
432 new_alloc_size = PAGE_ALIGN(new_size);
434 /* Retrieve the slab flag */
435 if (type == &memblock.memory)
436 in_slab = &memblock_memory_in_slab;
438 in_slab = &memblock_reserved_in_slab;
440 /* Try to find some space for it */
442 new_array = kmalloc(new_size, GFP_KERNEL);
443 addr = new_array ? __pa(new_array) : 0;
445 /* only exclude range when trying to double reserved.regions */
446 if (type != &memblock.reserved)
447 new_area_start = new_area_size = 0;
449 addr = memblock_find_in_range(new_area_start + new_area_size,
450 memblock.current_limit,
451 new_alloc_size, PAGE_SIZE);
452 if (!addr && new_area_size)
453 addr = memblock_find_in_range(0,
454 min(new_area_start, memblock.current_limit),
455 new_alloc_size, PAGE_SIZE);
457 new_array = addr ? __va(addr) : NULL;
460 pr_err("memblock: Failed to double %s array from %ld to %ld entries !\n",
461 type->name, type->max, type->max * 2);
465 new_end = addr + new_size - 1;
466 memblock_dbg("memblock: %s is doubled to %ld at [%pa-%pa]",
467 type->name, type->max * 2, &addr, &new_end);
470 * Found space, we now need to move the array over before we add the
471 * reserved region since it may be our reserved array itself that is
474 memcpy(new_array, type->regions, old_size);
475 memset(new_array + type->max, 0, old_size);
476 old_array = type->regions;
477 type->regions = new_array;
480 /* Free old array. We needn't free it if the array is the static one */
483 else if (old_array != memblock_memory_init_regions &&
484 old_array != memblock_reserved_init_regions)
485 memblock_free(old_array, old_alloc_size);
488 * Reserve the new array if that comes from the memblock. Otherwise, we
492 BUG_ON(memblock_reserve(addr, new_alloc_size));
494 /* Update slab flag */
501 * memblock_merge_regions - merge neighboring compatible regions
502 * @type: memblock type to scan
504 * Scan @type and merge neighboring compatible regions.
506 static void __init_memblock memblock_merge_regions(struct memblock_type *type)
510 /* cnt never goes below 1 */
511 while (i < type->cnt - 1) {
512 struct memblock_region *this = &type->regions[i];
513 struct memblock_region *next = &type->regions[i + 1];
515 if (this->base + this->size != next->base ||
516 memblock_get_region_node(this) !=
517 memblock_get_region_node(next) ||
518 this->flags != next->flags) {
519 BUG_ON(this->base + this->size > next->base);
524 this->size += next->size;
525 /* move forward from next + 1, index of which is i + 2 */
526 memmove(next, next + 1, (type->cnt - (i + 2)) * sizeof(*next));
532 * memblock_insert_region - insert new memblock region
533 * @type: memblock type to insert into
534 * @idx: index for the insertion point
535 * @base: base address of the new region
536 * @size: size of the new region
537 * @nid: node id of the new region
538 * @flags: flags of the new region
540 * Insert new memblock region [@base, @base + @size) into @type at @idx.
541 * @type must already have extra room to accommodate the new region.
543 static void __init_memblock memblock_insert_region(struct memblock_type *type,
544 int idx, phys_addr_t base,
547 enum memblock_flags flags)
549 struct memblock_region *rgn = &type->regions[idx];
551 BUG_ON(type->cnt >= type->max);
552 memmove(rgn + 1, rgn, (type->cnt - idx) * sizeof(*rgn));
556 memblock_set_region_node(rgn, nid);
558 type->total_size += size;
562 * memblock_add_range - add new memblock region
563 * @type: memblock type to add new region into
564 * @base: base address of the new region
565 * @size: size of the new region
566 * @nid: nid of the new region
567 * @flags: flags of the new region
569 * Add new memblock region [@base, @base + @size) into @type. The new region
570 * is allowed to overlap with existing ones - overlaps don't affect already
571 * existing regions. @type is guaranteed to be minimal (all neighbouring
572 * compatible regions are merged) after the addition.
575 * 0 on success, -errno on failure.
577 static int __init_memblock memblock_add_range(struct memblock_type *type,
578 phys_addr_t base, phys_addr_t size,
579 int nid, enum memblock_flags flags)
582 phys_addr_t obase = base;
583 phys_addr_t end = base + memblock_cap_size(base, &size);
585 struct memblock_region *rgn;
590 /* special case for empty array */
591 if (type->regions[0].size == 0) {
592 WARN_ON(type->cnt != 1 || type->total_size);
593 type->regions[0].base = base;
594 type->regions[0].size = size;
595 type->regions[0].flags = flags;
596 memblock_set_region_node(&type->regions[0], nid);
597 type->total_size = size;
602 * The following is executed twice. Once with %false @insert and
603 * then with %true. The first counts the number of regions needed
604 * to accommodate the new area. The second actually inserts them.
609 for_each_memblock_type(idx, type, rgn) {
610 phys_addr_t rbase = rgn->base;
611 phys_addr_t rend = rbase + rgn->size;
618 * @rgn overlaps. If it separates the lower part of new
619 * area, insert that portion.
623 WARN_ON(nid != memblock_get_region_node(rgn));
625 WARN_ON(flags != rgn->flags);
628 memblock_insert_region(type, idx++, base,
632 /* area below @rend is dealt with, forget about it */
633 base = min(rend, end);
636 /* insert the remaining portion */
640 memblock_insert_region(type, idx, base, end - base,
648 * If this was the first round, resize array and repeat for actual
649 * insertions; otherwise, merge and return.
652 while (type->cnt + nr_new > type->max)
653 if (memblock_double_array(type, obase, size) < 0)
658 memblock_merge_regions(type);
664 * memblock_add_node - add new memblock region within a NUMA node
665 * @base: base address of the new region
666 * @size: size of the new region
667 * @nid: nid of the new region
668 * @flags: flags of the new region
670 * Add new memblock region [@base, @base + @size) to the "memory"
671 * type. See memblock_add_range() description for mode details
674 * 0 on success, -errno on failure.
676 int __init_memblock memblock_add_node(phys_addr_t base, phys_addr_t size,
677 int nid, enum memblock_flags flags)
679 phys_addr_t end = base + size - 1;
681 memblock_dbg("%s: [%pa-%pa] nid=%d flags=%x %pS\n", __func__,
682 &base, &end, nid, flags, (void *)_RET_IP_);
684 return memblock_add_range(&memblock.memory, base, size, nid, flags);
688 * memblock_add - add new memblock region
689 * @base: base address of the new region
690 * @size: size of the new region
692 * Add new memblock region [@base, @base + @size) to the "memory"
693 * type. See memblock_add_range() description for mode details
696 * 0 on success, -errno on failure.
698 int __init_memblock memblock_add(phys_addr_t base, phys_addr_t size)
700 phys_addr_t end = base + size - 1;
702 memblock_dbg("%s: [%pa-%pa] %pS\n", __func__,
703 &base, &end, (void *)_RET_IP_);
705 return memblock_add_range(&memblock.memory, base, size, MAX_NUMNODES, 0);
709 * memblock_isolate_range - isolate given range into disjoint memblocks
710 * @type: memblock type to isolate range for
711 * @base: base of range to isolate
712 * @size: size of range to isolate
713 * @start_rgn: out parameter for the start of isolated region
714 * @end_rgn: out parameter for the end of isolated region
716 * Walk @type and ensure that regions don't cross the boundaries defined by
717 * [@base, @base + @size). Crossing regions are split at the boundaries,
718 * which may create at most two more regions. The index of the first
719 * region inside the range is returned in *@start_rgn and end in *@end_rgn.
722 * 0 on success, -errno on failure.
724 static int __init_memblock memblock_isolate_range(struct memblock_type *type,
725 phys_addr_t base, phys_addr_t size,
726 int *start_rgn, int *end_rgn)
728 phys_addr_t end = base + memblock_cap_size(base, &size);
730 struct memblock_region *rgn;
732 *start_rgn = *end_rgn = 0;
737 /* we'll create at most two more regions */
738 while (type->cnt + 2 > type->max)
739 if (memblock_double_array(type, base, size) < 0)
742 for_each_memblock_type(idx, type, rgn) {
743 phys_addr_t rbase = rgn->base;
744 phys_addr_t rend = rbase + rgn->size;
753 * @rgn intersects from below. Split and continue
754 * to process the next region - the new top half.
757 rgn->size -= base - rbase;
758 type->total_size -= base - rbase;
759 memblock_insert_region(type, idx, rbase, base - rbase,
760 memblock_get_region_node(rgn),
762 } else if (rend > end) {
764 * @rgn intersects from above. Split and redo the
765 * current region - the new bottom half.
768 rgn->size -= end - rbase;
769 type->total_size -= end - rbase;
770 memblock_insert_region(type, idx--, rbase, end - rbase,
771 memblock_get_region_node(rgn),
774 /* @rgn is fully contained, record it */
784 static int __init_memblock memblock_remove_range(struct memblock_type *type,
785 phys_addr_t base, phys_addr_t size)
787 int start_rgn, end_rgn;
790 ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
794 for (i = end_rgn - 1; i >= start_rgn; i--)
795 memblock_remove_region(type, i);
799 int __init_memblock memblock_remove(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 return memblock_remove_range(&memblock.memory, base, size);
810 * memblock_free - free boot memory allocation
811 * @ptr: starting address of the boot memory allocation
812 * @size: size of the boot memory block in bytes
814 * Free boot memory block previously allocated by memblock_alloc_xx() API.
815 * The freeing memory will not be released to the buddy allocator.
817 void __init_memblock memblock_free(void *ptr, size_t size)
820 memblock_phys_free(__pa(ptr), size);
824 * memblock_phys_free - free boot memory block
825 * @base: phys starting address of the boot memory block
826 * @size: size of the boot memory block in bytes
828 * Free boot memory block previously allocated by memblock_alloc_xx() API.
829 * The freeing memory will not be released to the buddy allocator.
831 int __init_memblock memblock_phys_free(phys_addr_t base, phys_addr_t size)
833 phys_addr_t end = base + size - 1;
835 memblock_dbg("%s: [%pa-%pa] %pS\n", __func__,
836 &base, &end, (void *)_RET_IP_);
838 kmemleak_free_part_phys(base, size);
839 return memblock_remove_range(&memblock.reserved, base, size);
842 int __init_memblock memblock_reserve(phys_addr_t base, phys_addr_t size)
844 phys_addr_t end = base + size - 1;
846 memblock_dbg("%s: [%pa-%pa] %pS\n", __func__,
847 &base, &end, (void *)_RET_IP_);
849 return memblock_add_range(&memblock.reserved, base, size, MAX_NUMNODES, 0);
852 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
853 int __init_memblock memblock_physmem_add(phys_addr_t base, phys_addr_t size)
855 phys_addr_t end = base + size - 1;
857 memblock_dbg("%s: [%pa-%pa] %pS\n", __func__,
858 &base, &end, (void *)_RET_IP_);
860 return memblock_add_range(&physmem, base, size, MAX_NUMNODES, 0);
865 * memblock_setclr_flag - set or clear flag for a memory region
866 * @base: base address of the region
867 * @size: size of the region
868 * @set: set or clear the flag
869 * @flag: the flag to update
871 * This function isolates region [@base, @base + @size), and sets/clears flag
873 * Return: 0 on success, -errno on failure.
875 static int __init_memblock memblock_setclr_flag(phys_addr_t base,
876 phys_addr_t size, int set, int flag)
878 struct memblock_type *type = &memblock.memory;
879 int i, ret, start_rgn, end_rgn;
881 ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
885 for (i = start_rgn; i < end_rgn; i++) {
886 struct memblock_region *r = &type->regions[i];
894 memblock_merge_regions(type);
899 * memblock_mark_hotplug - Mark hotpluggable memory with flag MEMBLOCK_HOTPLUG.
900 * @base: the base phys addr of the region
901 * @size: the size of the region
903 * Return: 0 on success, -errno on failure.
905 int __init_memblock memblock_mark_hotplug(phys_addr_t base, phys_addr_t size)
907 return memblock_setclr_flag(base, size, 1, MEMBLOCK_HOTPLUG);
911 * memblock_clear_hotplug - Clear flag MEMBLOCK_HOTPLUG for a specified region.
912 * @base: the base phys addr of the region
913 * @size: the size of the region
915 * Return: 0 on success, -errno on failure.
917 int __init_memblock memblock_clear_hotplug(phys_addr_t base, phys_addr_t size)
919 return memblock_setclr_flag(base, size, 0, MEMBLOCK_HOTPLUG);
923 * memblock_mark_mirror - Mark mirrored memory with flag MEMBLOCK_MIRROR.
924 * @base: the base phys addr of the region
925 * @size: the size of the region
927 * Return: 0 on success, -errno on failure.
929 int __init_memblock memblock_mark_mirror(phys_addr_t base, phys_addr_t size)
931 if (!mirrored_kernelcore)
934 system_has_some_mirror = true;
936 return memblock_setclr_flag(base, size, 1, MEMBLOCK_MIRROR);
940 * memblock_mark_nomap - Mark a memory region with flag MEMBLOCK_NOMAP.
941 * @base: the base phys addr of the region
942 * @size: the size of the region
944 * The memory regions marked with %MEMBLOCK_NOMAP will not be added to the
945 * direct mapping of the physical memory. These regions will still be
946 * covered by the memory map. The struct page representing NOMAP memory
947 * frames in the memory map will be PageReserved()
949 * Note: if the memory being marked %MEMBLOCK_NOMAP was allocated from
950 * memblock, the caller must inform kmemleak to ignore that memory
952 * Return: 0 on success, -errno on failure.
954 int __init_memblock memblock_mark_nomap(phys_addr_t base, phys_addr_t size)
956 return memblock_setclr_flag(base, size, 1, MEMBLOCK_NOMAP);
960 * memblock_clear_nomap - Clear flag MEMBLOCK_NOMAP for a specified region.
961 * @base: the base phys addr of the region
962 * @size: the size of the region
964 * Return: 0 on success, -errno on failure.
966 int __init_memblock memblock_clear_nomap(phys_addr_t base, phys_addr_t size)
968 return memblock_setclr_flag(base, size, 0, MEMBLOCK_NOMAP);
971 static bool should_skip_region(struct memblock_type *type,
972 struct memblock_region *m,
975 int m_nid = memblock_get_region_node(m);
977 /* we never skip regions when iterating memblock.reserved or physmem */
978 if (type != memblock_memory)
981 /* only memory regions are associated with nodes, check it */
982 if (nid != NUMA_NO_NODE && nid != m_nid)
985 /* skip hotpluggable memory regions if needed */
986 if (movable_node_is_enabled() && memblock_is_hotpluggable(m) &&
987 !(flags & MEMBLOCK_HOTPLUG))
990 /* if we want mirror memory skip non-mirror memory regions */
991 if ((flags & MEMBLOCK_MIRROR) && !memblock_is_mirror(m))
994 /* skip nomap memory unless we were asked for it explicitly */
995 if (!(flags & MEMBLOCK_NOMAP) && memblock_is_nomap(m))
998 /* skip driver-managed memory unless we were asked for it explicitly */
999 if (!(flags & MEMBLOCK_DRIVER_MANAGED) && memblock_is_driver_managed(m))
1006 * __next_mem_range - next function for for_each_free_mem_range() etc.
1007 * @idx: pointer to u64 loop variable
1008 * @nid: node selector, %NUMA_NO_NODE for all nodes
1009 * @flags: pick from blocks based on memory attributes
1010 * @type_a: pointer to memblock_type from where the range is taken
1011 * @type_b: pointer to memblock_type which excludes memory from being taken
1012 * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL
1013 * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL
1014 * @out_nid: ptr to int for nid of the range, can be %NULL
1016 * Find the first area from *@idx which matches @nid, fill the out
1017 * parameters, and update *@idx for the next iteration. The lower 32bit of
1018 * *@idx contains index into type_a and the upper 32bit indexes the
1019 * areas before each region in type_b. For example, if type_b regions
1020 * look like the following,
1022 * 0:[0-16), 1:[32-48), 2:[128-130)
1024 * The upper 32bit indexes the following regions.
1026 * 0:[0-0), 1:[16-32), 2:[48-128), 3:[130-MAX)
1028 * As both region arrays are sorted, the function advances the two indices
1029 * in lockstep and returns each intersection.
1031 void __next_mem_range(u64 *idx, int nid, enum memblock_flags flags,
1032 struct memblock_type *type_a,
1033 struct memblock_type *type_b, phys_addr_t *out_start,
1034 phys_addr_t *out_end, int *out_nid)
1036 int idx_a = *idx & 0xffffffff;
1037 int idx_b = *idx >> 32;
1039 if (WARN_ONCE(nid == MAX_NUMNODES,
1040 "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
1043 for (; idx_a < type_a->cnt; idx_a++) {
1044 struct memblock_region *m = &type_a->regions[idx_a];
1046 phys_addr_t m_start = m->base;
1047 phys_addr_t m_end = m->base + m->size;
1048 int m_nid = memblock_get_region_node(m);
1050 if (should_skip_region(type_a, m, nid, flags))
1055 *out_start = m_start;
1061 *idx = (u32)idx_a | (u64)idx_b << 32;
1065 /* scan areas before each reservation */
1066 for (; idx_b < type_b->cnt + 1; idx_b++) {
1067 struct memblock_region *r;
1068 phys_addr_t r_start;
1071 r = &type_b->regions[idx_b];
1072 r_start = idx_b ? r[-1].base + r[-1].size : 0;
1073 r_end = idx_b < type_b->cnt ?
1074 r->base : PHYS_ADDR_MAX;
1077 * if idx_b advanced past idx_a,
1078 * break out to advance idx_a
1080 if (r_start >= m_end)
1082 /* if the two regions intersect, we're done */
1083 if (m_start < r_end) {
1086 max(m_start, r_start);
1088 *out_end = min(m_end, r_end);
1092 * The region which ends first is
1093 * advanced for the next iteration.
1099 *idx = (u32)idx_a | (u64)idx_b << 32;
1105 /* signal end of iteration */
1110 * __next_mem_range_rev - generic next function for for_each_*_range_rev()
1112 * @idx: pointer to u64 loop variable
1113 * @nid: node selector, %NUMA_NO_NODE for all nodes
1114 * @flags: pick from blocks based on memory attributes
1115 * @type_a: pointer to memblock_type from where the range is taken
1116 * @type_b: pointer to memblock_type which excludes memory from being taken
1117 * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL
1118 * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL
1119 * @out_nid: ptr to int for nid of the range, can be %NULL
1121 * Finds the next range from type_a which is not marked as unsuitable
1124 * Reverse of __next_mem_range().
1126 void __init_memblock __next_mem_range_rev(u64 *idx, int nid,
1127 enum memblock_flags flags,
1128 struct memblock_type *type_a,
1129 struct memblock_type *type_b,
1130 phys_addr_t *out_start,
1131 phys_addr_t *out_end, int *out_nid)
1133 int idx_a = *idx & 0xffffffff;
1134 int idx_b = *idx >> 32;
1136 if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
1139 if (*idx == (u64)ULLONG_MAX) {
1140 idx_a = type_a->cnt - 1;
1142 idx_b = type_b->cnt;
1147 for (; idx_a >= 0; idx_a--) {
1148 struct memblock_region *m = &type_a->regions[idx_a];
1150 phys_addr_t m_start = m->base;
1151 phys_addr_t m_end = m->base + m->size;
1152 int m_nid = memblock_get_region_node(m);
1154 if (should_skip_region(type_a, m, nid, flags))
1159 *out_start = m_start;
1165 *idx = (u32)idx_a | (u64)idx_b << 32;
1169 /* scan areas before each reservation */
1170 for (; idx_b >= 0; idx_b--) {
1171 struct memblock_region *r;
1172 phys_addr_t r_start;
1175 r = &type_b->regions[idx_b];
1176 r_start = idx_b ? r[-1].base + r[-1].size : 0;
1177 r_end = idx_b < type_b->cnt ?
1178 r->base : PHYS_ADDR_MAX;
1180 * if idx_b advanced past idx_a,
1181 * break out to advance idx_a
1184 if (r_end <= m_start)
1186 /* if the two regions intersect, we're done */
1187 if (m_end > r_start) {
1189 *out_start = max(m_start, r_start);
1191 *out_end = min(m_end, r_end);
1194 if (m_start >= r_start)
1198 *idx = (u32)idx_a | (u64)idx_b << 32;
1203 /* signal end of iteration */
1208 * Common iterator interface used to define for_each_mem_pfn_range().
1210 void __init_memblock __next_mem_pfn_range(int *idx, int nid,
1211 unsigned long *out_start_pfn,
1212 unsigned long *out_end_pfn, int *out_nid)
1214 struct memblock_type *type = &memblock.memory;
1215 struct memblock_region *r;
1218 while (++*idx < type->cnt) {
1219 r = &type->regions[*idx];
1220 r_nid = memblock_get_region_node(r);
1222 if (PFN_UP(r->base) >= PFN_DOWN(r->base + r->size))
1224 if (nid == MAX_NUMNODES || nid == r_nid)
1227 if (*idx >= type->cnt) {
1233 *out_start_pfn = PFN_UP(r->base);
1235 *out_end_pfn = PFN_DOWN(r->base + r->size);
1241 * memblock_set_node - set node ID on memblock regions
1242 * @base: base of area to set node ID for
1243 * @size: size of area to set node ID for
1244 * @type: memblock type to set node ID for
1245 * @nid: node ID to set
1247 * Set the nid of memblock @type regions in [@base, @base + @size) to @nid.
1248 * Regions which cross the area boundaries are split as necessary.
1251 * 0 on success, -errno on failure.
1253 int __init_memblock memblock_set_node(phys_addr_t base, phys_addr_t size,
1254 struct memblock_type *type, int nid)
1257 int start_rgn, end_rgn;
1260 ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
1264 for (i = start_rgn; i < end_rgn; i++)
1265 memblock_set_region_node(&type->regions[i], nid);
1267 memblock_merge_regions(type);
1272 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1274 * __next_mem_pfn_range_in_zone - iterator for for_each_*_range_in_zone()
1276 * @idx: pointer to u64 loop variable
1277 * @zone: zone in which all of the memory blocks reside
1278 * @out_spfn: ptr to ulong for start pfn of the range, can be %NULL
1279 * @out_epfn: ptr to ulong for end pfn of the range, can be %NULL
1281 * This function is meant to be a zone/pfn specific wrapper for the
1282 * for_each_mem_range type iterators. Specifically they are used in the
1283 * deferred memory init routines and as such we were duplicating much of
1284 * this logic throughout the code. So instead of having it in multiple
1285 * locations it seemed like it would make more sense to centralize this to
1286 * one new iterator that does everything they need.
1288 void __init_memblock
1289 __next_mem_pfn_range_in_zone(u64 *idx, struct zone *zone,
1290 unsigned long *out_spfn, unsigned long *out_epfn)
1292 int zone_nid = zone_to_nid(zone);
1293 phys_addr_t spa, epa;
1295 __next_mem_range(idx, zone_nid, MEMBLOCK_NONE,
1296 &memblock.memory, &memblock.reserved,
1299 while (*idx != U64_MAX) {
1300 unsigned long epfn = PFN_DOWN(epa);
1301 unsigned long spfn = PFN_UP(spa);
1304 * Verify the end is at least past the start of the zone and
1305 * that we have at least one PFN to initialize.
1307 if (zone->zone_start_pfn < epfn && spfn < epfn) {
1308 /* if we went too far just stop searching */
1309 if (zone_end_pfn(zone) <= spfn) {
1315 *out_spfn = max(zone->zone_start_pfn, spfn);
1317 *out_epfn = min(zone_end_pfn(zone), epfn);
1322 __next_mem_range(idx, zone_nid, MEMBLOCK_NONE,
1323 &memblock.memory, &memblock.reserved,
1327 /* signal end of iteration */
1329 *out_spfn = ULONG_MAX;
1334 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
1337 * memblock_alloc_range_nid - allocate boot memory block
1338 * @size: size of memory block to be allocated in bytes
1339 * @align: alignment of the region and block's size
1340 * @start: the lower bound of the memory region to allocate (phys address)
1341 * @end: the upper bound of the memory region to allocate (phys address)
1342 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1343 * @exact_nid: control the allocation fall back to other nodes
1345 * The allocation is performed from memory region limited by
1346 * memblock.current_limit if @end == %MEMBLOCK_ALLOC_ACCESSIBLE.
1348 * If the specified node can not hold the requested memory and @exact_nid
1349 * is false, the allocation falls back to any node in the system.
1351 * For systems with memory mirroring, the allocation is attempted first
1352 * from the regions with mirroring enabled and then retried from any
1355 * In addition, function using kmemleak_alloc_phys for allocated boot
1356 * memory block, it is never reported as leaks.
1359 * Physical address of allocated memory block on success, %0 on failure.
1361 phys_addr_t __init memblock_alloc_range_nid(phys_addr_t size,
1362 phys_addr_t align, phys_addr_t start,
1363 phys_addr_t end, int nid,
1366 enum memblock_flags flags = choose_memblock_flags();
1369 if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
1373 /* Can't use WARNs this early in boot on powerpc */
1375 align = SMP_CACHE_BYTES;
1379 found = memblock_find_in_range_node(size, align, start, end, nid,
1381 if (found && !memblock_reserve(found, size))
1384 if (nid != NUMA_NO_NODE && !exact_nid) {
1385 found = memblock_find_in_range_node(size, align, start,
1388 if (found && !memblock_reserve(found, size))
1392 if (flags & MEMBLOCK_MIRROR) {
1393 flags &= ~MEMBLOCK_MIRROR;
1394 pr_warn_ratelimited("Could not allocate %pap bytes of mirrored memory\n",
1403 * Skip kmemleak for those places like kasan_init() and
1404 * early_pgtable_alloc() due to high volume.
1406 if (end != MEMBLOCK_ALLOC_NOLEAKTRACE)
1408 * Memblock allocated blocks are never reported as
1409 * leaks. This is because many of these blocks are
1410 * only referred via the physical address which is
1411 * not looked up by kmemleak.
1413 kmemleak_alloc_phys(found, size, 0);
1419 * memblock_phys_alloc_range - allocate a memory block inside specified range
1420 * @size: size of memory block to be allocated in bytes
1421 * @align: alignment of the region and block's size
1422 * @start: the lower bound of the memory region to allocate (physical address)
1423 * @end: the upper bound of the memory region to allocate (physical address)
1425 * Allocate @size bytes in the between @start and @end.
1427 * Return: physical address of the allocated memory block on success,
1430 phys_addr_t __init memblock_phys_alloc_range(phys_addr_t size,
1435 memblock_dbg("%s: %llu bytes align=0x%llx from=%pa max_addr=%pa %pS\n",
1436 __func__, (u64)size, (u64)align, &start, &end,
1438 return memblock_alloc_range_nid(size, align, start, end, NUMA_NO_NODE,
1443 * memblock_phys_alloc_try_nid - allocate a memory block from specified NUMA node
1444 * @size: size of memory block to be allocated in bytes
1445 * @align: alignment of the region and block's size
1446 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1448 * Allocates memory block from the specified NUMA node. If the node
1449 * has no available memory, attempts to allocated from any node in the
1452 * Return: physical address of the allocated memory block on success,
1455 phys_addr_t __init memblock_phys_alloc_try_nid(phys_addr_t size, phys_addr_t align, int nid)
1457 return memblock_alloc_range_nid(size, align, 0,
1458 MEMBLOCK_ALLOC_ACCESSIBLE, nid, false);
1462 * memblock_alloc_internal - allocate boot memory block
1463 * @size: size of memory block to be allocated in bytes
1464 * @align: alignment of the region and block's size
1465 * @min_addr: the lower bound of the memory region to allocate (phys address)
1466 * @max_addr: the upper bound of the memory region to allocate (phys address)
1467 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1468 * @exact_nid: control the allocation fall back to other nodes
1470 * Allocates memory block using memblock_alloc_range_nid() and
1471 * converts the returned physical address to virtual.
1473 * The @min_addr limit is dropped if it can not be satisfied and the allocation
1474 * will fall back to memory below @min_addr. Other constraints, such
1475 * as node and mirrored memory will be handled again in
1476 * memblock_alloc_range_nid().
1479 * Virtual address of allocated memory block on success, NULL on failure.
1481 static void * __init memblock_alloc_internal(
1482 phys_addr_t size, phys_addr_t align,
1483 phys_addr_t min_addr, phys_addr_t max_addr,
1484 int nid, bool exact_nid)
1489 * Detect any accidental use of these APIs after slab is ready, as at
1490 * this moment memblock may be deinitialized already and its
1491 * internal data may be destroyed (after execution of memblock_free_all)
1493 if (WARN_ON_ONCE(slab_is_available()))
1494 return kzalloc_node(size, GFP_NOWAIT, nid);
1496 if (max_addr > memblock.current_limit)
1497 max_addr = memblock.current_limit;
1499 alloc = memblock_alloc_range_nid(size, align, min_addr, max_addr, nid,
1502 /* retry allocation without lower limit */
1503 if (!alloc && min_addr)
1504 alloc = memblock_alloc_range_nid(size, align, 0, max_addr, nid,
1510 return phys_to_virt(alloc);
1514 * memblock_alloc_exact_nid_raw - allocate boot memory block on the exact node
1515 * without zeroing memory
1516 * @size: size of memory block to be allocated in bytes
1517 * @align: alignment of the region and block's size
1518 * @min_addr: the lower bound of the memory region from where the allocation
1519 * is preferred (phys address)
1520 * @max_addr: the upper bound of the memory region from where the allocation
1521 * is preferred (phys address), or %MEMBLOCK_ALLOC_ACCESSIBLE to
1522 * allocate only from memory limited by memblock.current_limit value
1523 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1525 * Public function, provides additional debug information (including caller
1526 * info), if enabled. Does not zero allocated memory.
1529 * Virtual address of allocated memory block on success, NULL on failure.
1531 void * __init memblock_alloc_exact_nid_raw(
1532 phys_addr_t size, phys_addr_t align,
1533 phys_addr_t min_addr, phys_addr_t max_addr,
1536 memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=%pa max_addr=%pa %pS\n",
1537 __func__, (u64)size, (u64)align, nid, &min_addr,
1538 &max_addr, (void *)_RET_IP_);
1540 return memblock_alloc_internal(size, align, min_addr, max_addr, nid,
1545 * memblock_alloc_try_nid_raw - allocate boot memory block without zeroing
1546 * memory and without panicking
1547 * @size: size of memory block to be allocated in bytes
1548 * @align: alignment of the region and block's size
1549 * @min_addr: the lower bound of the memory region from where the allocation
1550 * is preferred (phys address)
1551 * @max_addr: the upper bound of the memory region from where the allocation
1552 * is preferred (phys address), or %MEMBLOCK_ALLOC_ACCESSIBLE to
1553 * allocate only from memory limited by memblock.current_limit value
1554 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1556 * Public function, provides additional debug information (including caller
1557 * info), if enabled. Does not zero allocated memory, does not panic if request
1558 * cannot be satisfied.
1561 * Virtual address of allocated memory block on success, NULL on failure.
1563 void * __init memblock_alloc_try_nid_raw(
1564 phys_addr_t size, phys_addr_t align,
1565 phys_addr_t min_addr, phys_addr_t max_addr,
1568 memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=%pa max_addr=%pa %pS\n",
1569 __func__, (u64)size, (u64)align, nid, &min_addr,
1570 &max_addr, (void *)_RET_IP_);
1572 return memblock_alloc_internal(size, align, min_addr, max_addr, nid,
1577 * memblock_alloc_try_nid - allocate boot memory block
1578 * @size: size of memory block to be allocated in bytes
1579 * @align: alignment of the region and block's size
1580 * @min_addr: the lower bound of the memory region from where the allocation
1581 * is preferred (phys address)
1582 * @max_addr: the upper bound of the memory region from where the allocation
1583 * is preferred (phys address), or %MEMBLOCK_ALLOC_ACCESSIBLE to
1584 * allocate only from memory limited by memblock.current_limit value
1585 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1587 * Public function, provides additional debug information (including caller
1588 * info), if enabled. This function zeroes the allocated memory.
1591 * Virtual address of allocated memory block on success, NULL on failure.
1593 void * __init memblock_alloc_try_nid(
1594 phys_addr_t size, phys_addr_t align,
1595 phys_addr_t min_addr, phys_addr_t max_addr,
1600 memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=%pa max_addr=%pa %pS\n",
1601 __func__, (u64)size, (u64)align, nid, &min_addr,
1602 &max_addr, (void *)_RET_IP_);
1603 ptr = memblock_alloc_internal(size, align,
1604 min_addr, max_addr, nid, false);
1606 memset(ptr, 0, size);
1612 * memblock_free_late - free pages directly to buddy allocator
1613 * @base: phys starting address of the boot memory block
1614 * @size: size of the boot memory block in bytes
1616 * This is only useful when the memblock allocator has already been torn
1617 * down, but we are still initializing the system. Pages are released directly
1618 * to the buddy allocator.
1620 void __init memblock_free_late(phys_addr_t base, phys_addr_t size)
1622 phys_addr_t cursor, end;
1624 end = base + size - 1;
1625 memblock_dbg("%s: [%pa-%pa] %pS\n",
1626 __func__, &base, &end, (void *)_RET_IP_);
1627 kmemleak_free_part_phys(base, size);
1628 cursor = PFN_UP(base);
1629 end = PFN_DOWN(base + size);
1631 for (; cursor < end; cursor++) {
1632 memblock_free_pages(pfn_to_page(cursor), cursor, 0);
1633 totalram_pages_inc();
1638 * Remaining API functions
1641 phys_addr_t __init_memblock memblock_phys_mem_size(void)
1643 return memblock.memory.total_size;
1646 phys_addr_t __init_memblock memblock_reserved_size(void)
1648 return memblock.reserved.total_size;
1651 /* lowest address */
1652 phys_addr_t __init_memblock memblock_start_of_DRAM(void)
1654 return memblock.memory.regions[0].base;
1657 phys_addr_t __init_memblock memblock_end_of_DRAM(void)
1659 int idx = memblock.memory.cnt - 1;
1661 return (memblock.memory.regions[idx].base + memblock.memory.regions[idx].size);
1664 static phys_addr_t __init_memblock __find_max_addr(phys_addr_t limit)
1666 phys_addr_t max_addr = PHYS_ADDR_MAX;
1667 struct memblock_region *r;
1670 * translate the memory @limit size into the max address within one of
1671 * the memory memblock regions, if the @limit exceeds the total size
1672 * of those regions, max_addr will keep original value PHYS_ADDR_MAX
1674 for_each_mem_region(r) {
1675 if (limit <= r->size) {
1676 max_addr = r->base + limit;
1685 void __init memblock_enforce_memory_limit(phys_addr_t limit)
1687 phys_addr_t max_addr;
1692 max_addr = __find_max_addr(limit);
1694 /* @limit exceeds the total size of the memory, do nothing */
1695 if (max_addr == PHYS_ADDR_MAX)
1698 /* truncate both memory and reserved regions */
1699 memblock_remove_range(&memblock.memory, max_addr,
1701 memblock_remove_range(&memblock.reserved, max_addr,
1705 void __init memblock_cap_memory_range(phys_addr_t base, phys_addr_t size)
1707 int start_rgn, end_rgn;
1713 if (!memblock_memory->total_size) {
1714 pr_warn("%s: No memory registered yet\n", __func__);
1718 ret = memblock_isolate_range(&memblock.memory, base, size,
1719 &start_rgn, &end_rgn);
1723 /* remove all the MAP regions */
1724 for (i = memblock.memory.cnt - 1; i >= end_rgn; i--)
1725 if (!memblock_is_nomap(&memblock.memory.regions[i]))
1726 memblock_remove_region(&memblock.memory, i);
1728 for (i = start_rgn - 1; i >= 0; i--)
1729 if (!memblock_is_nomap(&memblock.memory.regions[i]))
1730 memblock_remove_region(&memblock.memory, i);
1732 /* truncate the reserved regions */
1733 memblock_remove_range(&memblock.reserved, 0, base);
1734 memblock_remove_range(&memblock.reserved,
1735 base + size, PHYS_ADDR_MAX);
1738 void __init memblock_mem_limit_remove_map(phys_addr_t limit)
1740 phys_addr_t max_addr;
1745 max_addr = __find_max_addr(limit);
1747 /* @limit exceeds the total size of the memory, do nothing */
1748 if (max_addr == PHYS_ADDR_MAX)
1751 memblock_cap_memory_range(0, max_addr);
1754 static int __init_memblock memblock_search(struct memblock_type *type, phys_addr_t addr)
1756 unsigned int left = 0, right = type->cnt;
1759 unsigned int mid = (right + left) / 2;
1761 if (addr < type->regions[mid].base)
1763 else if (addr >= (type->regions[mid].base +
1764 type->regions[mid].size))
1768 } while (left < right);
1772 bool __init_memblock memblock_is_reserved(phys_addr_t addr)
1774 return memblock_search(&memblock.reserved, addr) != -1;
1777 bool __init_memblock memblock_is_memory(phys_addr_t addr)
1779 return memblock_search(&memblock.memory, addr) != -1;
1782 bool __init_memblock memblock_is_map_memory(phys_addr_t addr)
1784 int i = memblock_search(&memblock.memory, addr);
1788 return !memblock_is_nomap(&memblock.memory.regions[i]);
1791 int __init_memblock memblock_search_pfn_nid(unsigned long pfn,
1792 unsigned long *start_pfn, unsigned long *end_pfn)
1794 struct memblock_type *type = &memblock.memory;
1795 int mid = memblock_search(type, PFN_PHYS(pfn));
1800 *start_pfn = PFN_DOWN(type->regions[mid].base);
1801 *end_pfn = PFN_DOWN(type->regions[mid].base + type->regions[mid].size);
1803 return memblock_get_region_node(&type->regions[mid]);
1807 * memblock_is_region_memory - check if a region is a subset of memory
1808 * @base: base of region to check
1809 * @size: size of region to check
1811 * Check if the region [@base, @base + @size) is a subset of a memory block.
1814 * 0 if false, non-zero if true
1816 bool __init_memblock memblock_is_region_memory(phys_addr_t base, phys_addr_t size)
1818 int idx = memblock_search(&memblock.memory, base);
1819 phys_addr_t end = base + memblock_cap_size(base, &size);
1823 return (memblock.memory.regions[idx].base +
1824 memblock.memory.regions[idx].size) >= end;
1828 * memblock_is_region_reserved - check if a region intersects reserved memory
1829 * @base: base of region to check
1830 * @size: size of region to check
1832 * Check if the region [@base, @base + @size) intersects a reserved
1836 * True if they intersect, false if not.
1838 bool __init_memblock memblock_is_region_reserved(phys_addr_t base, phys_addr_t size)
1840 return memblock_overlaps_region(&memblock.reserved, base, size);
1843 void __init_memblock memblock_trim_memory(phys_addr_t align)
1845 phys_addr_t start, end, orig_start, orig_end;
1846 struct memblock_region *r;
1848 for_each_mem_region(r) {
1849 orig_start = r->base;
1850 orig_end = r->base + r->size;
1851 start = round_up(orig_start, align);
1852 end = round_down(orig_end, align);
1854 if (start == orig_start && end == orig_end)
1859 r->size = end - start;
1861 memblock_remove_region(&memblock.memory,
1862 r - memblock.memory.regions);
1868 void __init_memblock memblock_set_current_limit(phys_addr_t limit)
1870 memblock.current_limit = limit;
1873 phys_addr_t __init_memblock memblock_get_current_limit(void)
1875 return memblock.current_limit;
1878 static void __init_memblock memblock_dump(struct memblock_type *type)
1880 phys_addr_t base, end, size;
1881 enum memblock_flags flags;
1883 struct memblock_region *rgn;
1885 pr_info(" %s.cnt = 0x%lx\n", type->name, type->cnt);
1887 for_each_memblock_type(idx, type, rgn) {
1888 char nid_buf[32] = "";
1892 end = base + size - 1;
1895 if (memblock_get_region_node(rgn) != MAX_NUMNODES)
1896 snprintf(nid_buf, sizeof(nid_buf), " on node %d",
1897 memblock_get_region_node(rgn));
1899 pr_info(" %s[%#x]\t[%pa-%pa], %pa bytes%s flags: %#x\n",
1900 type->name, idx, &base, &end, &size, nid_buf, flags);
1904 static void __init_memblock __memblock_dump_all(void)
1906 pr_info("MEMBLOCK configuration:\n");
1907 pr_info(" memory size = %pa reserved size = %pa\n",
1908 &memblock.memory.total_size,
1909 &memblock.reserved.total_size);
1911 memblock_dump(&memblock.memory);
1912 memblock_dump(&memblock.reserved);
1913 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
1914 memblock_dump(&physmem);
1918 void __init_memblock memblock_dump_all(void)
1921 __memblock_dump_all();
1924 void __init memblock_allow_resize(void)
1926 memblock_can_resize = 1;
1929 static int __init early_memblock(char *p)
1931 if (p && strstr(p, "debug"))
1935 early_param("memblock", early_memblock);
1937 static void __init free_memmap(unsigned long start_pfn, unsigned long end_pfn)
1939 struct page *start_pg, *end_pg;
1940 phys_addr_t pg, pgend;
1943 * Convert start_pfn/end_pfn to a struct page pointer.
1945 start_pg = pfn_to_page(start_pfn - 1) + 1;
1946 end_pg = pfn_to_page(end_pfn - 1) + 1;
1949 * Convert to physical addresses, and round start upwards and end
1952 pg = PAGE_ALIGN(__pa(start_pg));
1953 pgend = __pa(end_pg) & PAGE_MASK;
1956 * If there are free pages between these, free the section of the
1960 memblock_phys_free(pg, pgend - pg);
1964 * The mem_map array can get very big. Free the unused area of the memory map.
1966 static void __init free_unused_memmap(void)
1968 unsigned long start, end, prev_end = 0;
1971 if (!IS_ENABLED(CONFIG_HAVE_ARCH_PFN_VALID) ||
1972 IS_ENABLED(CONFIG_SPARSEMEM_VMEMMAP))
1976 * This relies on each bank being in address order.
1977 * The banks are sorted previously in bootmem_init().
1979 for_each_mem_pfn_range(i, MAX_NUMNODES, &start, &end, NULL) {
1980 #ifdef CONFIG_SPARSEMEM
1982 * Take care not to free memmap entries that don't exist
1983 * due to SPARSEMEM sections which aren't present.
1985 start = min(start, ALIGN(prev_end, PAGES_PER_SECTION));
1988 * Align down here since many operations in VM subsystem
1989 * presume that there are no holes in the memory map inside
1992 start = round_down(start, pageblock_nr_pages);
1995 * If we had a previous bank, and there is a space
1996 * between the current bank and the previous, free it.
1998 if (prev_end && prev_end < start)
1999 free_memmap(prev_end, start);
2002 * Align up here since many operations in VM subsystem
2003 * presume that there are no holes in the memory map inside
2006 prev_end = ALIGN(end, pageblock_nr_pages);
2009 #ifdef CONFIG_SPARSEMEM
2010 if (!IS_ALIGNED(prev_end, PAGES_PER_SECTION)) {
2011 prev_end = ALIGN(end, pageblock_nr_pages);
2012 free_memmap(prev_end, ALIGN(prev_end, PAGES_PER_SECTION));
2017 static void __init __free_pages_memory(unsigned long start, unsigned long end)
2021 while (start < end) {
2022 order = min(MAX_ORDER - 1UL, __ffs(start));
2024 while (start + (1UL << order) > end)
2027 memblock_free_pages(pfn_to_page(start), start, order);
2029 start += (1UL << order);
2033 static unsigned long __init __free_memory_core(phys_addr_t start,
2036 unsigned long start_pfn = PFN_UP(start);
2037 unsigned long end_pfn = min_t(unsigned long,
2038 PFN_DOWN(end), max_low_pfn);
2040 if (start_pfn >= end_pfn)
2043 __free_pages_memory(start_pfn, end_pfn);
2045 return end_pfn - start_pfn;
2048 static void __init memmap_init_reserved_pages(void)
2050 struct memblock_region *region;
2051 phys_addr_t start, end;
2054 /* initialize struct pages for the reserved regions */
2055 for_each_reserved_mem_range(i, &start, &end)
2056 reserve_bootmem_region(start, end);
2058 /* and also treat struct pages for the NOMAP regions as PageReserved */
2059 for_each_mem_region(region) {
2060 if (memblock_is_nomap(region)) {
2061 start = region->base;
2062 end = start + region->size;
2063 reserve_bootmem_region(start, end);
2068 static unsigned long __init free_low_memory_core_early(void)
2070 unsigned long count = 0;
2071 phys_addr_t start, end;
2074 memblock_clear_hotplug(0, -1);
2076 memmap_init_reserved_pages();
2079 * We need to use NUMA_NO_NODE instead of NODE_DATA(0)->node_id
2080 * because in some case like Node0 doesn't have RAM installed
2081 * low ram will be on Node1
2083 for_each_free_mem_range(i, NUMA_NO_NODE, MEMBLOCK_NONE, &start, &end,
2085 count += __free_memory_core(start, end);
2090 static int reset_managed_pages_done __initdata;
2092 void reset_node_managed_pages(pg_data_t *pgdat)
2096 for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++)
2097 atomic_long_set(&z->managed_pages, 0);
2100 void __init reset_all_zones_managed_pages(void)
2102 struct pglist_data *pgdat;
2104 if (reset_managed_pages_done)
2107 for_each_online_pgdat(pgdat)
2108 reset_node_managed_pages(pgdat);
2110 reset_managed_pages_done = 1;
2114 * memblock_free_all - release free pages to the buddy allocator
2116 void __init memblock_free_all(void)
2118 unsigned long pages;
2120 free_unused_memmap();
2121 reset_all_zones_managed_pages();
2123 pages = free_low_memory_core_early();
2124 totalram_pages_add(pages);
2127 #if defined(CONFIG_DEBUG_FS) && defined(CONFIG_ARCH_KEEP_MEMBLOCK)
2129 static int memblock_debug_show(struct seq_file *m, void *private)
2131 struct memblock_type *type = m->private;
2132 struct memblock_region *reg;
2136 for (i = 0; i < type->cnt; i++) {
2137 reg = &type->regions[i];
2138 end = reg->base + reg->size - 1;
2140 seq_printf(m, "%4d: ", i);
2141 seq_printf(m, "%pa..%pa\n", ®->base, &end);
2145 DEFINE_SHOW_ATTRIBUTE(memblock_debug);
2147 static int __init memblock_init_debugfs(void)
2149 struct dentry *root = debugfs_create_dir("memblock", NULL);
2151 debugfs_create_file("memory", 0444, root,
2152 &memblock.memory, &memblock_debug_fops);
2153 debugfs_create_file("reserved", 0444, root,
2154 &memblock.reserved, &memblock_debug_fops);
2155 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
2156 debugfs_create_file("physmem", 0444, root, &physmem,
2157 &memblock_debug_fops);
2162 __initcall(memblock_init_debugfs);
2164 #endif /* CONFIG_DEBUG_FS */