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 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 __memblock_free_late(addr, size);
372 if (memblock.memory.regions != memblock_memory_init_regions) {
373 addr = __pa(memblock.memory.regions);
374 size = PAGE_ALIGN(sizeof(struct memblock_region) *
375 memblock.memory.max);
376 __memblock_free_late(addr, size);
379 memblock_memory = NULL;
384 * memblock_double_array - double the size of the memblock regions array
385 * @type: memblock type of the regions array being doubled
386 * @new_area_start: starting address of memory range to avoid overlap with
387 * @new_area_size: size of memory range to avoid overlap with
389 * Double the size of the @type regions array. If memblock is being used to
390 * allocate memory for a new reserved regions array and there is a previously
391 * allocated memory range [@new_area_start, @new_area_start + @new_area_size]
392 * waiting to be reserved, ensure the memory used by the new array does
396 * 0 on success, -1 on failure.
398 static int __init_memblock memblock_double_array(struct memblock_type *type,
399 phys_addr_t new_area_start,
400 phys_addr_t new_area_size)
402 struct memblock_region *new_array, *old_array;
403 phys_addr_t old_alloc_size, new_alloc_size;
404 phys_addr_t old_size, new_size, addr, new_end;
405 int use_slab = slab_is_available();
408 /* We don't allow resizing until we know about the reserved regions
409 * of memory that aren't suitable for allocation
411 if (!memblock_can_resize)
414 /* Calculate new doubled size */
415 old_size = type->max * sizeof(struct memblock_region);
416 new_size = old_size << 1;
418 * We need to allocated new one align to PAGE_SIZE,
419 * so we can free them completely later.
421 old_alloc_size = PAGE_ALIGN(old_size);
422 new_alloc_size = PAGE_ALIGN(new_size);
424 /* Retrieve the slab flag */
425 if (type == &memblock.memory)
426 in_slab = &memblock_memory_in_slab;
428 in_slab = &memblock_reserved_in_slab;
430 /* Try to find some space for it */
432 new_array = kmalloc(new_size, GFP_KERNEL);
433 addr = new_array ? __pa(new_array) : 0;
435 /* only exclude range when trying to double reserved.regions */
436 if (type != &memblock.reserved)
437 new_area_start = new_area_size = 0;
439 addr = memblock_find_in_range(new_area_start + new_area_size,
440 memblock.current_limit,
441 new_alloc_size, PAGE_SIZE);
442 if (!addr && new_area_size)
443 addr = memblock_find_in_range(0,
444 min(new_area_start, memblock.current_limit),
445 new_alloc_size, PAGE_SIZE);
447 new_array = addr ? __va(addr) : NULL;
450 pr_err("memblock: Failed to double %s array from %ld to %ld entries !\n",
451 type->name, type->max, type->max * 2);
455 new_end = addr + new_size - 1;
456 memblock_dbg("memblock: %s is doubled to %ld at [%pa-%pa]",
457 type->name, type->max * 2, &addr, &new_end);
460 * Found space, we now need to move the array over before we add the
461 * reserved region since it may be our reserved array itself that is
464 memcpy(new_array, type->regions, old_size);
465 memset(new_array + type->max, 0, old_size);
466 old_array = type->regions;
467 type->regions = new_array;
470 /* Free old array. We needn't free it if the array is the static one */
473 else if (old_array != memblock_memory_init_regions &&
474 old_array != memblock_reserved_init_regions)
475 memblock_free(__pa(old_array), old_alloc_size);
478 * Reserve the new array if that comes from the memblock. Otherwise, we
482 BUG_ON(memblock_reserve(addr, new_alloc_size));
484 /* Update slab flag */
491 * memblock_merge_regions - merge neighboring compatible regions
492 * @type: memblock type to scan
494 * Scan @type and merge neighboring compatible regions.
496 static void __init_memblock memblock_merge_regions(struct memblock_type *type)
500 /* cnt never goes below 1 */
501 while (i < type->cnt - 1) {
502 struct memblock_region *this = &type->regions[i];
503 struct memblock_region *next = &type->regions[i + 1];
505 if (this->base + this->size != next->base ||
506 memblock_get_region_node(this) !=
507 memblock_get_region_node(next) ||
508 this->flags != next->flags) {
509 BUG_ON(this->base + this->size > next->base);
514 this->size += next->size;
515 /* move forward from next + 1, index of which is i + 2 */
516 memmove(next, next + 1, (type->cnt - (i + 2)) * sizeof(*next));
522 * memblock_insert_region - insert new memblock region
523 * @type: memblock type to insert into
524 * @idx: index for the insertion point
525 * @base: base address of the new region
526 * @size: size of the new region
527 * @nid: node id of the new region
528 * @flags: flags of the new region
530 * Insert new memblock region [@base, @base + @size) into @type at @idx.
531 * @type must already have extra room to accommodate the new region.
533 static void __init_memblock memblock_insert_region(struct memblock_type *type,
534 int idx, phys_addr_t base,
537 enum memblock_flags flags)
539 struct memblock_region *rgn = &type->regions[idx];
541 BUG_ON(type->cnt >= type->max);
542 memmove(rgn + 1, rgn, (type->cnt - idx) * sizeof(*rgn));
546 memblock_set_region_node(rgn, nid);
548 type->total_size += size;
552 * memblock_add_range - add new memblock region
553 * @type: memblock type to add new region into
554 * @base: base address of the new region
555 * @size: size of the new region
556 * @nid: nid of the new region
557 * @flags: flags of the new region
559 * Add new memblock region [@base, @base + @size) into @type. The new region
560 * is allowed to overlap with existing ones - overlaps don't affect already
561 * existing regions. @type is guaranteed to be minimal (all neighbouring
562 * compatible regions are merged) after the addition.
565 * 0 on success, -errno on failure.
567 static int __init_memblock memblock_add_range(struct memblock_type *type,
568 phys_addr_t base, phys_addr_t size,
569 int nid, enum memblock_flags flags)
572 phys_addr_t obase = base;
573 phys_addr_t end = base + memblock_cap_size(base, &size);
575 struct memblock_region *rgn;
580 /* special case for empty array */
581 if (type->regions[0].size == 0) {
582 WARN_ON(type->cnt != 1 || type->total_size);
583 type->regions[0].base = base;
584 type->regions[0].size = size;
585 type->regions[0].flags = flags;
586 memblock_set_region_node(&type->regions[0], nid);
587 type->total_size = size;
592 * The following is executed twice. Once with %false @insert and
593 * then with %true. The first counts the number of regions needed
594 * to accommodate the new area. The second actually inserts them.
599 for_each_memblock_type(idx, type, rgn) {
600 phys_addr_t rbase = rgn->base;
601 phys_addr_t rend = rbase + rgn->size;
608 * @rgn overlaps. If it separates the lower part of new
609 * area, insert that portion.
613 WARN_ON(nid != memblock_get_region_node(rgn));
615 WARN_ON(flags != rgn->flags);
618 memblock_insert_region(type, idx++, base,
622 /* area below @rend is dealt with, forget about it */
623 base = min(rend, end);
626 /* insert the remaining portion */
630 memblock_insert_region(type, idx, base, end - base,
638 * If this was the first round, resize array and repeat for actual
639 * insertions; otherwise, merge and return.
642 while (type->cnt + nr_new > type->max)
643 if (memblock_double_array(type, obase, size) < 0)
648 memblock_merge_regions(type);
654 * memblock_add_node - add new memblock region within a NUMA node
655 * @base: base address of the new region
656 * @size: size of the new region
657 * @nid: nid of the new region
659 * Add new memblock region [@base, @base + @size) to the "memory"
660 * type. See memblock_add_range() description for mode details
663 * 0 on success, -errno on failure.
665 int __init_memblock memblock_add_node(phys_addr_t base, phys_addr_t size,
668 phys_addr_t end = base + size - 1;
670 memblock_dbg("%s: [%pa-%pa] nid=%d %pS\n", __func__,
671 &base, &end, nid, (void *)_RET_IP_);
673 return memblock_add_range(&memblock.memory, base, size, nid, 0);
677 * memblock_add - add new memblock region
678 * @base: base address of the new region
679 * @size: size of the new region
681 * Add new memblock region [@base, @base + @size) to the "memory"
682 * type. See memblock_add_range() description for mode details
685 * 0 on success, -errno on failure.
687 int __init_memblock memblock_add(phys_addr_t base, phys_addr_t size)
689 phys_addr_t end = base + size - 1;
691 memblock_dbg("%s: [%pa-%pa] %pS\n", __func__,
692 &base, &end, (void *)_RET_IP_);
694 return memblock_add_range(&memblock.memory, base, size, MAX_NUMNODES, 0);
698 * memblock_isolate_range - isolate given range into disjoint memblocks
699 * @type: memblock type to isolate range for
700 * @base: base of range to isolate
701 * @size: size of range to isolate
702 * @start_rgn: out parameter for the start of isolated region
703 * @end_rgn: out parameter for the end of isolated region
705 * Walk @type and ensure that regions don't cross the boundaries defined by
706 * [@base, @base + @size). Crossing regions are split at the boundaries,
707 * which may create at most two more regions. The index of the first
708 * region inside the range is returned in *@start_rgn and end in *@end_rgn.
711 * 0 on success, -errno on failure.
713 static int __init_memblock memblock_isolate_range(struct memblock_type *type,
714 phys_addr_t base, phys_addr_t size,
715 int *start_rgn, int *end_rgn)
717 phys_addr_t end = base + memblock_cap_size(base, &size);
719 struct memblock_region *rgn;
721 *start_rgn = *end_rgn = 0;
726 /* we'll create at most two more regions */
727 while (type->cnt + 2 > type->max)
728 if (memblock_double_array(type, base, size) < 0)
731 for_each_memblock_type(idx, type, rgn) {
732 phys_addr_t rbase = rgn->base;
733 phys_addr_t rend = rbase + rgn->size;
742 * @rgn intersects from below. Split and continue
743 * to process the next region - the new top half.
746 rgn->size -= base - rbase;
747 type->total_size -= base - rbase;
748 memblock_insert_region(type, idx, rbase, base - rbase,
749 memblock_get_region_node(rgn),
751 } else if (rend > end) {
753 * @rgn intersects from above. Split and redo the
754 * current region - the new bottom half.
757 rgn->size -= end - rbase;
758 type->total_size -= end - rbase;
759 memblock_insert_region(type, idx--, rbase, end - rbase,
760 memblock_get_region_node(rgn),
763 /* @rgn is fully contained, record it */
773 static int __init_memblock memblock_remove_range(struct memblock_type *type,
774 phys_addr_t base, phys_addr_t size)
776 int start_rgn, end_rgn;
779 ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
783 for (i = end_rgn - 1; i >= start_rgn; i--)
784 memblock_remove_region(type, i);
788 int __init_memblock memblock_remove(phys_addr_t base, phys_addr_t size)
790 phys_addr_t end = base + size - 1;
792 memblock_dbg("%s: [%pa-%pa] %pS\n", __func__,
793 &base, &end, (void *)_RET_IP_);
795 return memblock_remove_range(&memblock.memory, base, size);
799 * memblock_free - free boot memory block
800 * @base: phys starting address of the boot memory block
801 * @size: size of the boot memory block in bytes
803 * Free boot memory block previously allocated by memblock_alloc_xx() API.
804 * The freeing memory will not be released to the buddy allocator.
806 int __init_memblock memblock_free(phys_addr_t base, phys_addr_t size)
808 phys_addr_t end = base + size - 1;
810 memblock_dbg("%s: [%pa-%pa] %pS\n", __func__,
811 &base, &end, (void *)_RET_IP_);
813 kmemleak_free_part_phys(base, size);
814 return memblock_remove_range(&memblock.reserved, base, size);
817 int __init_memblock memblock_reserve(phys_addr_t base, phys_addr_t size)
819 phys_addr_t end = base + size - 1;
821 memblock_dbg("%s: [%pa-%pa] %pS\n", __func__,
822 &base, &end, (void *)_RET_IP_);
824 return memblock_add_range(&memblock.reserved, base, size, MAX_NUMNODES, 0);
827 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
828 int __init_memblock memblock_physmem_add(phys_addr_t base, phys_addr_t size)
830 phys_addr_t end = base + size - 1;
832 memblock_dbg("%s: [%pa-%pa] %pS\n", __func__,
833 &base, &end, (void *)_RET_IP_);
835 return memblock_add_range(&physmem, base, size, MAX_NUMNODES, 0);
840 * memblock_setclr_flag - set or clear flag for a memory region
841 * @base: base address of the region
842 * @size: size of the region
843 * @set: set or clear the flag
844 * @flag: the flag to update
846 * This function isolates region [@base, @base + @size), and sets/clears flag
848 * Return: 0 on success, -errno on failure.
850 static int __init_memblock memblock_setclr_flag(phys_addr_t base,
851 phys_addr_t size, int set, int flag)
853 struct memblock_type *type = &memblock.memory;
854 int i, ret, start_rgn, end_rgn;
856 ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
860 for (i = start_rgn; i < end_rgn; i++) {
861 struct memblock_region *r = &type->regions[i];
869 memblock_merge_regions(type);
874 * memblock_mark_hotplug - Mark hotpluggable memory with flag MEMBLOCK_HOTPLUG.
875 * @base: the base phys addr of the region
876 * @size: the size of the region
878 * Return: 0 on success, -errno on failure.
880 int __init_memblock memblock_mark_hotplug(phys_addr_t base, phys_addr_t size)
882 return memblock_setclr_flag(base, size, 1, MEMBLOCK_HOTPLUG);
886 * memblock_clear_hotplug - Clear flag MEMBLOCK_HOTPLUG for a specified region.
887 * @base: the base phys addr of the region
888 * @size: the size of the region
890 * Return: 0 on success, -errno on failure.
892 int __init_memblock memblock_clear_hotplug(phys_addr_t base, phys_addr_t size)
894 return memblock_setclr_flag(base, size, 0, MEMBLOCK_HOTPLUG);
898 * memblock_mark_mirror - Mark mirrored memory with flag MEMBLOCK_MIRROR.
899 * @base: the base phys addr of the region
900 * @size: the size of the region
902 * Return: 0 on success, -errno on failure.
904 int __init_memblock memblock_mark_mirror(phys_addr_t base, phys_addr_t size)
906 system_has_some_mirror = true;
908 return memblock_setclr_flag(base, size, 1, MEMBLOCK_MIRROR);
912 * memblock_mark_nomap - Mark a memory region with flag MEMBLOCK_NOMAP.
913 * @base: the base phys addr of the region
914 * @size: the size of the region
916 * The memory regions marked with %MEMBLOCK_NOMAP will not be added to the
917 * direct mapping of the physical memory. These regions will still be
918 * covered by the memory map. The struct page representing NOMAP memory
919 * frames in the memory map will be PageReserved()
921 * Return: 0 on success, -errno on failure.
923 int __init_memblock memblock_mark_nomap(phys_addr_t base, phys_addr_t size)
925 return memblock_setclr_flag(base, size, 1, MEMBLOCK_NOMAP);
929 * memblock_clear_nomap - Clear flag MEMBLOCK_NOMAP for a specified region.
930 * @base: the base phys addr of the region
931 * @size: the size of the region
933 * Return: 0 on success, -errno on failure.
935 int __init_memblock memblock_clear_nomap(phys_addr_t base, phys_addr_t size)
937 return memblock_setclr_flag(base, size, 0, MEMBLOCK_NOMAP);
940 static bool should_skip_region(struct memblock_type *type,
941 struct memblock_region *m,
944 int m_nid = memblock_get_region_node(m);
946 /* we never skip regions when iterating memblock.reserved or physmem */
947 if (type != memblock_memory)
950 /* only memory regions are associated with nodes, check it */
951 if (nid != NUMA_NO_NODE && nid != m_nid)
954 /* skip hotpluggable memory regions if needed */
955 if (movable_node_is_enabled() && memblock_is_hotpluggable(m) &&
956 !(flags & MEMBLOCK_HOTPLUG))
959 /* if we want mirror memory skip non-mirror memory regions */
960 if ((flags & MEMBLOCK_MIRROR) && !memblock_is_mirror(m))
963 /* skip nomap memory unless we were asked for it explicitly */
964 if (!(flags & MEMBLOCK_NOMAP) && memblock_is_nomap(m))
971 * __next_mem_range - next function for for_each_free_mem_range() etc.
972 * @idx: pointer to u64 loop variable
973 * @nid: node selector, %NUMA_NO_NODE for all nodes
974 * @flags: pick from blocks based on memory attributes
975 * @type_a: pointer to memblock_type from where the range is taken
976 * @type_b: pointer to memblock_type which excludes memory from being taken
977 * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL
978 * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL
979 * @out_nid: ptr to int for nid of the range, can be %NULL
981 * Find the first area from *@idx which matches @nid, fill the out
982 * parameters, and update *@idx for the next iteration. The lower 32bit of
983 * *@idx contains index into type_a and the upper 32bit indexes the
984 * areas before each region in type_b. For example, if type_b regions
985 * look like the following,
987 * 0:[0-16), 1:[32-48), 2:[128-130)
989 * The upper 32bit indexes the following regions.
991 * 0:[0-0), 1:[16-32), 2:[48-128), 3:[130-MAX)
993 * As both region arrays are sorted, the function advances the two indices
994 * in lockstep and returns each intersection.
996 void __next_mem_range(u64 *idx, int nid, enum memblock_flags flags,
997 struct memblock_type *type_a,
998 struct memblock_type *type_b, phys_addr_t *out_start,
999 phys_addr_t *out_end, int *out_nid)
1001 int idx_a = *idx & 0xffffffff;
1002 int idx_b = *idx >> 32;
1004 if (WARN_ONCE(nid == MAX_NUMNODES,
1005 "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
1008 for (; idx_a < type_a->cnt; idx_a++) {
1009 struct memblock_region *m = &type_a->regions[idx_a];
1011 phys_addr_t m_start = m->base;
1012 phys_addr_t m_end = m->base + m->size;
1013 int m_nid = memblock_get_region_node(m);
1015 if (should_skip_region(type_a, m, nid, flags))
1020 *out_start = m_start;
1026 *idx = (u32)idx_a | (u64)idx_b << 32;
1030 /* scan areas before each reservation */
1031 for (; idx_b < type_b->cnt + 1; idx_b++) {
1032 struct memblock_region *r;
1033 phys_addr_t r_start;
1036 r = &type_b->regions[idx_b];
1037 r_start = idx_b ? r[-1].base + r[-1].size : 0;
1038 r_end = idx_b < type_b->cnt ?
1039 r->base : PHYS_ADDR_MAX;
1042 * if idx_b advanced past idx_a,
1043 * break out to advance idx_a
1045 if (r_start >= m_end)
1047 /* if the two regions intersect, we're done */
1048 if (m_start < r_end) {
1051 max(m_start, r_start);
1053 *out_end = min(m_end, r_end);
1057 * The region which ends first is
1058 * advanced for the next iteration.
1064 *idx = (u32)idx_a | (u64)idx_b << 32;
1070 /* signal end of iteration */
1075 * __next_mem_range_rev - generic next function for for_each_*_range_rev()
1077 * @idx: pointer to u64 loop variable
1078 * @nid: node selector, %NUMA_NO_NODE for all nodes
1079 * @flags: pick from blocks based on memory attributes
1080 * @type_a: pointer to memblock_type from where the range is taken
1081 * @type_b: pointer to memblock_type which excludes memory from being taken
1082 * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL
1083 * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL
1084 * @out_nid: ptr to int for nid of the range, can be %NULL
1086 * Finds the next range from type_a which is not marked as unsuitable
1089 * Reverse of __next_mem_range().
1091 void __init_memblock __next_mem_range_rev(u64 *idx, int nid,
1092 enum memblock_flags flags,
1093 struct memblock_type *type_a,
1094 struct memblock_type *type_b,
1095 phys_addr_t *out_start,
1096 phys_addr_t *out_end, int *out_nid)
1098 int idx_a = *idx & 0xffffffff;
1099 int idx_b = *idx >> 32;
1101 if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
1104 if (*idx == (u64)ULLONG_MAX) {
1105 idx_a = type_a->cnt - 1;
1107 idx_b = type_b->cnt;
1112 for (; idx_a >= 0; idx_a--) {
1113 struct memblock_region *m = &type_a->regions[idx_a];
1115 phys_addr_t m_start = m->base;
1116 phys_addr_t m_end = m->base + m->size;
1117 int m_nid = memblock_get_region_node(m);
1119 if (should_skip_region(type_a, m, nid, flags))
1124 *out_start = m_start;
1130 *idx = (u32)idx_a | (u64)idx_b << 32;
1134 /* scan areas before each reservation */
1135 for (; idx_b >= 0; idx_b--) {
1136 struct memblock_region *r;
1137 phys_addr_t r_start;
1140 r = &type_b->regions[idx_b];
1141 r_start = idx_b ? r[-1].base + r[-1].size : 0;
1142 r_end = idx_b < type_b->cnt ?
1143 r->base : PHYS_ADDR_MAX;
1145 * if idx_b advanced past idx_a,
1146 * break out to advance idx_a
1149 if (r_end <= m_start)
1151 /* if the two regions intersect, we're done */
1152 if (m_end > r_start) {
1154 *out_start = max(m_start, r_start);
1156 *out_end = min(m_end, r_end);
1159 if (m_start >= r_start)
1163 *idx = (u32)idx_a | (u64)idx_b << 32;
1168 /* signal end of iteration */
1173 * Common iterator interface used to define for_each_mem_pfn_range().
1175 void __init_memblock __next_mem_pfn_range(int *idx, int nid,
1176 unsigned long *out_start_pfn,
1177 unsigned long *out_end_pfn, int *out_nid)
1179 struct memblock_type *type = &memblock.memory;
1180 struct memblock_region *r;
1183 while (++*idx < type->cnt) {
1184 r = &type->regions[*idx];
1185 r_nid = memblock_get_region_node(r);
1187 if (PFN_UP(r->base) >= PFN_DOWN(r->base + r->size))
1189 if (nid == MAX_NUMNODES || nid == r_nid)
1192 if (*idx >= type->cnt) {
1198 *out_start_pfn = PFN_UP(r->base);
1200 *out_end_pfn = PFN_DOWN(r->base + r->size);
1206 * memblock_set_node - set node ID on memblock regions
1207 * @base: base of area to set node ID for
1208 * @size: size of area to set node ID for
1209 * @type: memblock type to set node ID for
1210 * @nid: node ID to set
1212 * Set the nid of memblock @type regions in [@base, @base + @size) to @nid.
1213 * Regions which cross the area boundaries are split as necessary.
1216 * 0 on success, -errno on failure.
1218 int __init_memblock memblock_set_node(phys_addr_t base, phys_addr_t size,
1219 struct memblock_type *type, int nid)
1222 int start_rgn, end_rgn;
1225 ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
1229 for (i = start_rgn; i < end_rgn; i++)
1230 memblock_set_region_node(&type->regions[i], nid);
1232 memblock_merge_regions(type);
1237 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1239 * __next_mem_pfn_range_in_zone - iterator for for_each_*_range_in_zone()
1241 * @idx: pointer to u64 loop variable
1242 * @zone: zone in which all of the memory blocks reside
1243 * @out_spfn: ptr to ulong for start pfn of the range, can be %NULL
1244 * @out_epfn: ptr to ulong for end pfn of the range, can be %NULL
1246 * This function is meant to be a zone/pfn specific wrapper for the
1247 * for_each_mem_range type iterators. Specifically they are used in the
1248 * deferred memory init routines and as such we were duplicating much of
1249 * this logic throughout the code. So instead of having it in multiple
1250 * locations it seemed like it would make more sense to centralize this to
1251 * one new iterator that does everything they need.
1253 void __init_memblock
1254 __next_mem_pfn_range_in_zone(u64 *idx, struct zone *zone,
1255 unsigned long *out_spfn, unsigned long *out_epfn)
1257 int zone_nid = zone_to_nid(zone);
1258 phys_addr_t spa, epa;
1261 __next_mem_range(idx, zone_nid, MEMBLOCK_NONE,
1262 &memblock.memory, &memblock.reserved,
1265 while (*idx != U64_MAX) {
1266 unsigned long epfn = PFN_DOWN(epa);
1267 unsigned long spfn = PFN_UP(spa);
1270 * Verify the end is at least past the start of the zone and
1271 * that we have at least one PFN to initialize.
1273 if (zone->zone_start_pfn < epfn && spfn < epfn) {
1274 /* if we went too far just stop searching */
1275 if (zone_end_pfn(zone) <= spfn) {
1281 *out_spfn = max(zone->zone_start_pfn, spfn);
1283 *out_epfn = min(zone_end_pfn(zone), epfn);
1288 __next_mem_range(idx, zone_nid, MEMBLOCK_NONE,
1289 &memblock.memory, &memblock.reserved,
1293 /* signal end of iteration */
1295 *out_spfn = ULONG_MAX;
1300 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
1303 * memblock_alloc_range_nid - allocate boot memory block
1304 * @size: size of memory block to be allocated in bytes
1305 * @align: alignment of the region and block's size
1306 * @start: the lower bound of the memory region to allocate (phys address)
1307 * @end: the upper bound of the memory region to allocate (phys address)
1308 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1309 * @exact_nid: control the allocation fall back to other nodes
1311 * The allocation is performed from memory region limited by
1312 * memblock.current_limit if @end == %MEMBLOCK_ALLOC_ACCESSIBLE.
1314 * If the specified node can not hold the requested memory and @exact_nid
1315 * is false, the allocation falls back to any node in the system.
1317 * For systems with memory mirroring, the allocation is attempted first
1318 * from the regions with mirroring enabled and then retried from any
1321 * In addition, function sets the min_count to 0 using kmemleak_alloc_phys for
1322 * allocated boot memory block, so that it is never reported as leaks.
1325 * Physical address of allocated memory block on success, %0 on failure.
1327 phys_addr_t __init memblock_alloc_range_nid(phys_addr_t size,
1328 phys_addr_t align, phys_addr_t start,
1329 phys_addr_t end, int nid,
1332 enum memblock_flags flags = choose_memblock_flags();
1335 if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
1339 /* Can't use WARNs this early in boot on powerpc */
1341 align = SMP_CACHE_BYTES;
1345 found = memblock_find_in_range_node(size, align, start, end, nid,
1347 if (found && !memblock_reserve(found, size))
1350 if (nid != NUMA_NO_NODE && !exact_nid) {
1351 found = memblock_find_in_range_node(size, align, start,
1354 if (found && !memblock_reserve(found, size))
1358 if (flags & MEMBLOCK_MIRROR) {
1359 flags &= ~MEMBLOCK_MIRROR;
1360 pr_warn("Could not allocate %pap bytes of mirrored memory\n",
1368 /* Skip kmemleak for kasan_init() due to high volume. */
1369 if (end != MEMBLOCK_ALLOC_KASAN)
1371 * The min_count is set to 0 so that memblock allocated
1372 * blocks are never reported as leaks. This is because many
1373 * of these blocks are only referred via the physical
1374 * address which is not looked up by kmemleak.
1376 kmemleak_alloc_phys(found, size, 0, 0);
1382 * memblock_phys_alloc_range - allocate a memory block inside specified range
1383 * @size: size of memory block to be allocated in bytes
1384 * @align: alignment of the region and block's size
1385 * @start: the lower bound of the memory region to allocate (physical address)
1386 * @end: the upper bound of the memory region to allocate (physical address)
1388 * Allocate @size bytes in the between @start and @end.
1390 * Return: physical address of the allocated memory block on success,
1393 phys_addr_t __init memblock_phys_alloc_range(phys_addr_t size,
1398 memblock_dbg("%s: %llu bytes align=0x%llx from=%pa max_addr=%pa %pS\n",
1399 __func__, (u64)size, (u64)align, &start, &end,
1401 return memblock_alloc_range_nid(size, align, start, end, NUMA_NO_NODE,
1406 * memblock_phys_alloc_try_nid - allocate a memory block from specified NUMA node
1407 * @size: size of memory block to be allocated in bytes
1408 * @align: alignment of the region and block's size
1409 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1411 * Allocates memory block from the specified NUMA node. If the node
1412 * has no available memory, attempts to allocated from any node in the
1415 * Return: physical address of the allocated memory block on success,
1418 phys_addr_t __init memblock_phys_alloc_try_nid(phys_addr_t size, phys_addr_t align, int nid)
1420 return memblock_alloc_range_nid(size, align, 0,
1421 MEMBLOCK_ALLOC_ACCESSIBLE, nid, false);
1425 * memblock_alloc_internal - allocate boot memory block
1426 * @size: size of memory block to be allocated in bytes
1427 * @align: alignment of the region and block's size
1428 * @min_addr: the lower bound of the memory region to allocate (phys address)
1429 * @max_addr: the upper bound of the memory region to allocate (phys address)
1430 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1431 * @exact_nid: control the allocation fall back to other nodes
1433 * Allocates memory block using memblock_alloc_range_nid() and
1434 * converts the returned physical address to virtual.
1436 * The @min_addr limit is dropped if it can not be satisfied and the allocation
1437 * will fall back to memory below @min_addr. Other constraints, such
1438 * as node and mirrored memory will be handled again in
1439 * memblock_alloc_range_nid().
1442 * Virtual address of allocated memory block on success, NULL on failure.
1444 static void * __init memblock_alloc_internal(
1445 phys_addr_t size, phys_addr_t align,
1446 phys_addr_t min_addr, phys_addr_t max_addr,
1447 int nid, bool exact_nid)
1452 * Detect any accidental use of these APIs after slab is ready, as at
1453 * this moment memblock may be deinitialized already and its
1454 * internal data may be destroyed (after execution of memblock_free_all)
1456 if (WARN_ON_ONCE(slab_is_available()))
1457 return kzalloc_node(size, GFP_NOWAIT, nid);
1459 if (max_addr > memblock.current_limit)
1460 max_addr = memblock.current_limit;
1462 alloc = memblock_alloc_range_nid(size, align, min_addr, max_addr, nid,
1465 /* retry allocation without lower limit */
1466 if (!alloc && min_addr)
1467 alloc = memblock_alloc_range_nid(size, align, 0, max_addr, nid,
1473 return phys_to_virt(alloc);
1477 * memblock_alloc_exact_nid_raw - allocate boot memory block on the exact node
1478 * without zeroing memory
1479 * @size: size of memory block to be allocated in bytes
1480 * @align: alignment of the region and block's size
1481 * @min_addr: the lower bound of the memory region from where the allocation
1482 * is preferred (phys address)
1483 * @max_addr: the upper bound of the memory region from where the allocation
1484 * is preferred (phys address), or %MEMBLOCK_ALLOC_ACCESSIBLE to
1485 * allocate only from memory limited by memblock.current_limit value
1486 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1488 * Public function, provides additional debug information (including caller
1489 * info), if enabled. Does not zero allocated memory.
1492 * Virtual address of allocated memory block on success, NULL on failure.
1494 void * __init memblock_alloc_exact_nid_raw(
1495 phys_addr_t size, phys_addr_t align,
1496 phys_addr_t min_addr, phys_addr_t max_addr,
1501 memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=%pa max_addr=%pa %pS\n",
1502 __func__, (u64)size, (u64)align, nid, &min_addr,
1503 &max_addr, (void *)_RET_IP_);
1505 ptr = memblock_alloc_internal(size, align,
1506 min_addr, max_addr, nid, true);
1507 if (ptr && size > 0)
1508 page_init_poison(ptr, size);
1514 * memblock_alloc_try_nid_raw - allocate boot memory block without zeroing
1515 * memory and without panicking
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, does not panic if request
1527 * cannot be satisfied.
1530 * Virtual address of allocated memory block on success, NULL on failure.
1532 void * __init memblock_alloc_try_nid_raw(
1533 phys_addr_t size, phys_addr_t align,
1534 phys_addr_t min_addr, phys_addr_t max_addr,
1539 memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=%pa max_addr=%pa %pS\n",
1540 __func__, (u64)size, (u64)align, nid, &min_addr,
1541 &max_addr, (void *)_RET_IP_);
1543 ptr = memblock_alloc_internal(size, align,
1544 min_addr, max_addr, nid, false);
1545 if (ptr && size > 0)
1546 page_init_poison(ptr, size);
1552 * memblock_alloc_try_nid - allocate boot memory block
1553 * @size: size of memory block to be allocated in bytes
1554 * @align: alignment of the region and block's size
1555 * @min_addr: the lower bound of the memory region from where the allocation
1556 * is preferred (phys address)
1557 * @max_addr: the upper bound of the memory region from where the allocation
1558 * is preferred (phys address), or %MEMBLOCK_ALLOC_ACCESSIBLE to
1559 * allocate only from memory limited by memblock.current_limit value
1560 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1562 * Public function, provides additional debug information (including caller
1563 * info), if enabled. This function zeroes the allocated memory.
1566 * Virtual address of allocated memory block on success, NULL on failure.
1568 void * __init memblock_alloc_try_nid(
1569 phys_addr_t size, phys_addr_t align,
1570 phys_addr_t min_addr, phys_addr_t max_addr,
1575 memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=%pa max_addr=%pa %pS\n",
1576 __func__, (u64)size, (u64)align, nid, &min_addr,
1577 &max_addr, (void *)_RET_IP_);
1578 ptr = memblock_alloc_internal(size, align,
1579 min_addr, max_addr, nid, false);
1581 memset(ptr, 0, size);
1587 * __memblock_free_late - free pages directly to buddy allocator
1588 * @base: phys starting address of the boot memory block
1589 * @size: size of the boot memory block in bytes
1591 * This is only useful when the memblock allocator has already been torn
1592 * down, but we are still initializing the system. Pages are released directly
1593 * to the buddy allocator.
1595 void __init __memblock_free_late(phys_addr_t base, phys_addr_t size)
1597 phys_addr_t cursor, end;
1599 end = base + size - 1;
1600 memblock_dbg("%s: [%pa-%pa] %pS\n",
1601 __func__, &base, &end, (void *)_RET_IP_);
1602 kmemleak_free_part_phys(base, size);
1603 cursor = PFN_UP(base);
1604 end = PFN_DOWN(base + size);
1606 for (; cursor < end; cursor++) {
1607 memblock_free_pages(pfn_to_page(cursor), cursor, 0);
1608 totalram_pages_inc();
1613 * Remaining API functions
1616 phys_addr_t __init_memblock memblock_phys_mem_size(void)
1618 return memblock.memory.total_size;
1621 phys_addr_t __init_memblock memblock_reserved_size(void)
1623 return memblock.reserved.total_size;
1626 /* lowest address */
1627 phys_addr_t __init_memblock memblock_start_of_DRAM(void)
1629 return memblock.memory.regions[0].base;
1632 phys_addr_t __init_memblock memblock_end_of_DRAM(void)
1634 int idx = memblock.memory.cnt - 1;
1636 return (memblock.memory.regions[idx].base + memblock.memory.regions[idx].size);
1639 static phys_addr_t __init_memblock __find_max_addr(phys_addr_t limit)
1641 phys_addr_t max_addr = PHYS_ADDR_MAX;
1642 struct memblock_region *r;
1645 * translate the memory @limit size into the max address within one of
1646 * the memory memblock regions, if the @limit exceeds the total size
1647 * of those regions, max_addr will keep original value PHYS_ADDR_MAX
1649 for_each_mem_region(r) {
1650 if (limit <= r->size) {
1651 max_addr = r->base + limit;
1660 void __init memblock_enforce_memory_limit(phys_addr_t limit)
1662 phys_addr_t max_addr;
1667 max_addr = __find_max_addr(limit);
1669 /* @limit exceeds the total size of the memory, do nothing */
1670 if (max_addr == PHYS_ADDR_MAX)
1673 /* truncate both memory and reserved regions */
1674 memblock_remove_range(&memblock.memory, max_addr,
1676 memblock_remove_range(&memblock.reserved, max_addr,
1680 void __init memblock_cap_memory_range(phys_addr_t base, phys_addr_t size)
1682 int start_rgn, end_rgn;
1688 if (memblock.memory.cnt <= 1) {
1689 pr_warn("%s: No memory registered yet\n", __func__);
1693 ret = memblock_isolate_range(&memblock.memory, base, size,
1694 &start_rgn, &end_rgn);
1698 /* remove all the MAP regions */
1699 for (i = memblock.memory.cnt - 1; i >= end_rgn; i--)
1700 if (!memblock_is_nomap(&memblock.memory.regions[i]))
1701 memblock_remove_region(&memblock.memory, i);
1703 for (i = start_rgn - 1; i >= 0; i--)
1704 if (!memblock_is_nomap(&memblock.memory.regions[i]))
1705 memblock_remove_region(&memblock.memory, i);
1707 /* truncate the reserved regions */
1708 memblock_remove_range(&memblock.reserved, 0, base);
1709 memblock_remove_range(&memblock.reserved,
1710 base + size, PHYS_ADDR_MAX);
1713 void __init memblock_mem_limit_remove_map(phys_addr_t limit)
1715 phys_addr_t max_addr;
1720 max_addr = __find_max_addr(limit);
1722 /* @limit exceeds the total size of the memory, do nothing */
1723 if (max_addr == PHYS_ADDR_MAX)
1726 memblock_cap_memory_range(0, max_addr);
1729 static int __init_memblock memblock_search(struct memblock_type *type, phys_addr_t addr)
1731 unsigned int left = 0, right = type->cnt;
1734 unsigned int mid = (right + left) / 2;
1736 if (addr < type->regions[mid].base)
1738 else if (addr >= (type->regions[mid].base +
1739 type->regions[mid].size))
1743 } while (left < right);
1747 bool __init_memblock memblock_is_reserved(phys_addr_t addr)
1749 return memblock_search(&memblock.reserved, addr) != -1;
1752 bool __init_memblock memblock_is_memory(phys_addr_t addr)
1754 return memblock_search(&memblock.memory, addr) != -1;
1757 bool __init_memblock memblock_is_map_memory(phys_addr_t addr)
1759 int i = memblock_search(&memblock.memory, addr);
1763 return !memblock_is_nomap(&memblock.memory.regions[i]);
1766 int __init_memblock memblock_search_pfn_nid(unsigned long pfn,
1767 unsigned long *start_pfn, unsigned long *end_pfn)
1769 struct memblock_type *type = &memblock.memory;
1770 int mid = memblock_search(type, PFN_PHYS(pfn));
1775 *start_pfn = PFN_DOWN(type->regions[mid].base);
1776 *end_pfn = PFN_DOWN(type->regions[mid].base + type->regions[mid].size);
1778 return memblock_get_region_node(&type->regions[mid]);
1782 * memblock_is_region_memory - check if a region is a subset of memory
1783 * @base: base of region to check
1784 * @size: size of region to check
1786 * Check if the region [@base, @base + @size) is a subset of a memory block.
1789 * 0 if false, non-zero if true
1791 bool __init_memblock memblock_is_region_memory(phys_addr_t base, phys_addr_t size)
1793 int idx = memblock_search(&memblock.memory, base);
1794 phys_addr_t end = base + memblock_cap_size(base, &size);
1798 return (memblock.memory.regions[idx].base +
1799 memblock.memory.regions[idx].size) >= end;
1803 * memblock_is_region_reserved - check if a region intersects reserved memory
1804 * @base: base of region to check
1805 * @size: size of region to check
1807 * Check if the region [@base, @base + @size) intersects a reserved
1811 * True if they intersect, false if not.
1813 bool __init_memblock memblock_is_region_reserved(phys_addr_t base, phys_addr_t size)
1815 return memblock_overlaps_region(&memblock.reserved, base, size);
1818 void __init_memblock memblock_trim_memory(phys_addr_t align)
1820 phys_addr_t start, end, orig_start, orig_end;
1821 struct memblock_region *r;
1823 for_each_mem_region(r) {
1824 orig_start = r->base;
1825 orig_end = r->base + r->size;
1826 start = round_up(orig_start, align);
1827 end = round_down(orig_end, align);
1829 if (start == orig_start && end == orig_end)
1834 r->size = end - start;
1836 memblock_remove_region(&memblock.memory,
1837 r - memblock.memory.regions);
1843 void __init_memblock memblock_set_current_limit(phys_addr_t limit)
1845 memblock.current_limit = limit;
1848 phys_addr_t __init_memblock memblock_get_current_limit(void)
1850 return memblock.current_limit;
1853 static void __init_memblock memblock_dump(struct memblock_type *type)
1855 phys_addr_t base, end, size;
1856 enum memblock_flags flags;
1858 struct memblock_region *rgn;
1860 pr_info(" %s.cnt = 0x%lx\n", type->name, type->cnt);
1862 for_each_memblock_type(idx, type, rgn) {
1863 char nid_buf[32] = "";
1867 end = base + size - 1;
1870 if (memblock_get_region_node(rgn) != MAX_NUMNODES)
1871 snprintf(nid_buf, sizeof(nid_buf), " on node %d",
1872 memblock_get_region_node(rgn));
1874 pr_info(" %s[%#x]\t[%pa-%pa], %pa bytes%s flags: %#x\n",
1875 type->name, idx, &base, &end, &size, nid_buf, flags);
1879 static void __init_memblock __memblock_dump_all(void)
1881 pr_info("MEMBLOCK configuration:\n");
1882 pr_info(" memory size = %pa reserved size = %pa\n",
1883 &memblock.memory.total_size,
1884 &memblock.reserved.total_size);
1886 memblock_dump(&memblock.memory);
1887 memblock_dump(&memblock.reserved);
1888 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
1889 memblock_dump(&physmem);
1893 void __init_memblock memblock_dump_all(void)
1896 __memblock_dump_all();
1899 void __init memblock_allow_resize(void)
1901 memblock_can_resize = 1;
1904 static int __init early_memblock(char *p)
1906 if (p && strstr(p, "debug"))
1910 early_param("memblock", early_memblock);
1912 static void __init free_memmap(unsigned long start_pfn, unsigned long end_pfn)
1914 struct page *start_pg, *end_pg;
1915 phys_addr_t pg, pgend;
1918 * Convert start_pfn/end_pfn to a struct page pointer.
1920 start_pg = pfn_to_page(start_pfn - 1) + 1;
1921 end_pg = pfn_to_page(end_pfn - 1) + 1;
1924 * Convert to physical addresses, and round start upwards and end
1927 pg = PAGE_ALIGN(__pa(start_pg));
1928 pgend = __pa(end_pg) & PAGE_MASK;
1931 * If there are free pages between these, free the section of the
1935 memblock_free(pg, pgend - pg);
1939 * The mem_map array can get very big. Free the unused area of the memory map.
1941 static void __init free_unused_memmap(void)
1943 unsigned long start, end, prev_end = 0;
1946 if (!IS_ENABLED(CONFIG_HAVE_ARCH_PFN_VALID) ||
1947 IS_ENABLED(CONFIG_SPARSEMEM_VMEMMAP))
1951 * This relies on each bank being in address order.
1952 * The banks are sorted previously in bootmem_init().
1954 for_each_mem_pfn_range(i, MAX_NUMNODES, &start, &end, NULL) {
1955 #ifdef CONFIG_SPARSEMEM
1957 * Take care not to free memmap entries that don't exist
1958 * due to SPARSEMEM sections which aren't present.
1960 start = min(start, ALIGN(prev_end, PAGES_PER_SECTION));
1963 * Align down here since many operations in VM subsystem
1964 * presume that there are no holes in the memory map inside
1967 start = round_down(start, pageblock_nr_pages);
1970 * If we had a previous bank, and there is a space
1971 * between the current bank and the previous, free it.
1973 if (prev_end && prev_end < start)
1974 free_memmap(prev_end, start);
1977 * Align up here since many operations in VM subsystem
1978 * presume that there are no holes in the memory map inside
1981 prev_end = ALIGN(end, pageblock_nr_pages);
1984 #ifdef CONFIG_SPARSEMEM
1985 if (!IS_ALIGNED(prev_end, PAGES_PER_SECTION)) {
1986 prev_end = ALIGN(end, pageblock_nr_pages);
1987 free_memmap(prev_end, ALIGN(prev_end, PAGES_PER_SECTION));
1992 static void __init __free_pages_memory(unsigned long start, unsigned long end)
1996 while (start < end) {
1997 order = min(MAX_ORDER - 1UL, __ffs(start));
1999 while (start + (1UL << order) > end)
2002 memblock_free_pages(pfn_to_page(start), start, order);
2004 start += (1UL << order);
2008 static unsigned long __init __free_memory_core(phys_addr_t start,
2011 unsigned long start_pfn = PFN_UP(start);
2012 unsigned long end_pfn = min_t(unsigned long,
2013 PFN_DOWN(end), max_low_pfn);
2015 if (start_pfn >= end_pfn)
2018 __free_pages_memory(start_pfn, end_pfn);
2020 return end_pfn - start_pfn;
2023 static void __init memmap_init_reserved_pages(void)
2025 struct memblock_region *region;
2026 phys_addr_t start, end;
2029 /* initialize struct pages for the reserved regions */
2030 for_each_reserved_mem_range(i, &start, &end)
2031 reserve_bootmem_region(start, end);
2033 /* and also treat struct pages for the NOMAP regions as PageReserved */
2034 for_each_mem_region(region) {
2035 if (memblock_is_nomap(region)) {
2036 start = region->base;
2037 end = start + region->size;
2038 reserve_bootmem_region(start, end);
2043 static unsigned long __init free_low_memory_core_early(void)
2045 unsigned long count = 0;
2046 phys_addr_t start, end;
2049 memblock_clear_hotplug(0, -1);
2051 memmap_init_reserved_pages();
2054 * We need to use NUMA_NO_NODE instead of NODE_DATA(0)->node_id
2055 * because in some case like Node0 doesn't have RAM installed
2056 * low ram will be on Node1
2058 for_each_free_mem_range(i, NUMA_NO_NODE, MEMBLOCK_NONE, &start, &end,
2060 count += __free_memory_core(start, end);
2065 static int reset_managed_pages_done __initdata;
2067 void reset_node_managed_pages(pg_data_t *pgdat)
2071 for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++)
2072 atomic_long_set(&z->managed_pages, 0);
2075 void __init reset_all_zones_managed_pages(void)
2077 struct pglist_data *pgdat;
2079 if (reset_managed_pages_done)
2082 for_each_online_pgdat(pgdat)
2083 reset_node_managed_pages(pgdat);
2085 reset_managed_pages_done = 1;
2089 * memblock_free_all - release free pages to the buddy allocator
2091 void __init memblock_free_all(void)
2093 unsigned long pages;
2095 free_unused_memmap();
2096 reset_all_zones_managed_pages();
2098 pages = free_low_memory_core_early();
2099 totalram_pages_add(pages);
2102 #if defined(CONFIG_DEBUG_FS) && defined(CONFIG_ARCH_KEEP_MEMBLOCK)
2104 static int memblock_debug_show(struct seq_file *m, void *private)
2106 struct memblock_type *type = m->private;
2107 struct memblock_region *reg;
2111 for (i = 0; i < type->cnt; i++) {
2112 reg = &type->regions[i];
2113 end = reg->base + reg->size - 1;
2115 seq_printf(m, "%4d: ", i);
2116 seq_printf(m, "%pa..%pa\n", ®->base, &end);
2120 DEFINE_SHOW_ATTRIBUTE(memblock_debug);
2122 static int __init memblock_init_debugfs(void)
2124 struct dentry *root = debugfs_create_dir("memblock", NULL);
2126 debugfs_create_file("memory", 0444, root,
2127 &memblock.memory, &memblock_debug_fops);
2128 debugfs_create_file("reserved", 0444, root,
2129 &memblock.reserved, &memblock_debug_fops);
2130 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
2131 debugfs_create_file("physmem", 0444, root, &physmem,
2132 &memblock_debug_fops);
2137 __initcall(memblock_init_debugfs);
2139 #endif /* CONFIG_DEBUG_FS */