2 * Procedures for maintaining information about logical memory blocks.
4 * Peter Bergner, IBM Corp. June 2001.
5 * Copyright (C) 2001 Peter Bergner.
7 * This program is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU General Public License
9 * as published by the Free Software Foundation; either version
10 * 2 of the License, or (at your option) any later version.
13 #include <linux/kernel.h>
14 #include <linux/slab.h>
15 #include <linux/init.h>
16 #include <linux/bitops.h>
17 #include <linux/poison.h>
18 #include <linux/pfn.h>
19 #include <linux/debugfs.h>
20 #include <linux/kmemleak.h>
21 #include <linux/seq_file.h>
22 #include <linux/memblock.h>
23 #include <linux/bootmem.h>
25 #include <asm/sections.h>
31 * DOC: memblock overview
33 * Memblock is a method of managing memory regions during the early
34 * boot period when the usual kernel memory allocators are not up and
37 * Memblock views the system memory as collections of contiguous
38 * regions. There are several types of these collections:
40 * * ``memory`` - describes the physical memory available to the
41 * kernel; this may differ from the actual physical memory installed
42 * in the system, for instance when the memory is restricted with
43 * ``mem=`` command line parameter
44 * * ``reserved`` - describes the regions that were allocated
45 * * ``physmap`` - describes the actual physical memory regardless of
46 * the possible restrictions; the ``physmap`` type is only available
47 * on some architectures.
49 * Each region is represented by :c:type:`struct memblock_region` that
50 * defines the region extents, its attributes and NUMA node id on NUMA
51 * systems. Every memory type is described by the :c:type:`struct
52 * memblock_type` which contains an array of memory regions along with
53 * the allocator metadata. The memory types are nicely wrapped with
54 * :c:type:`struct memblock`. This structure is statically initialzed
55 * at build time. The region arrays for the "memory" and "reserved"
56 * types are initially sized to %INIT_MEMBLOCK_REGIONS and for the
57 * "physmap" type to %INIT_PHYSMEM_REGIONS.
58 * The :c:func:`memblock_allow_resize` enables automatic resizing of
59 * the region arrays during addition of new regions. This feature
60 * should be used with care so that memory allocated for the region
61 * array will not overlap with areas that should be reserved, for
64 * The early architecture setup should tell memblock what the physical
65 * memory layout is by using :c:func:`memblock_add` or
66 * :c:func:`memblock_add_node` functions. The first function does not
67 * assign the region to a NUMA node and it is appropriate for UMA
68 * systems. Yet, it is possible to use it on NUMA systems as well and
69 * assign the region to a NUMA node later in the setup process using
70 * :c:func:`memblock_set_node`. The :c:func:`memblock_add_node`
71 * performs such an assignment directly.
73 * Once memblock is setup the memory can be allocated using either
74 * memblock or bootmem APIs.
76 * As the system boot progresses, the architecture specific
77 * :c:func:`mem_init` function frees all the memory to the buddy page
80 * If an architecure enables %CONFIG_ARCH_DISCARD_MEMBLOCK, the
81 * memblock data structures will be discarded after the system
82 * initialization compltes.
85 static struct memblock_region memblock_memory_init_regions[INIT_MEMBLOCK_REGIONS] __initdata_memblock;
86 static struct memblock_region memblock_reserved_init_regions[INIT_MEMBLOCK_REGIONS] __initdata_memblock;
87 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
88 static struct memblock_region memblock_physmem_init_regions[INIT_PHYSMEM_REGIONS] __initdata_memblock;
91 struct memblock memblock __initdata_memblock = {
92 .memory.regions = memblock_memory_init_regions,
93 .memory.cnt = 1, /* empty dummy entry */
94 .memory.max = INIT_MEMBLOCK_REGIONS,
95 .memory.name = "memory",
97 .reserved.regions = memblock_reserved_init_regions,
98 .reserved.cnt = 1, /* empty dummy entry */
99 .reserved.max = INIT_MEMBLOCK_REGIONS,
100 .reserved.name = "reserved",
102 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
103 .physmem.regions = memblock_physmem_init_regions,
104 .physmem.cnt = 1, /* empty dummy entry */
105 .physmem.max = INIT_PHYSMEM_REGIONS,
106 .physmem.name = "physmem",
110 .current_limit = MEMBLOCK_ALLOC_ANYWHERE,
113 int memblock_debug __initdata_memblock;
114 static bool system_has_some_mirror __initdata_memblock = false;
115 static int memblock_can_resize __initdata_memblock;
116 static int memblock_memory_in_slab __initdata_memblock = 0;
117 static int memblock_reserved_in_slab __initdata_memblock = 0;
119 enum memblock_flags __init_memblock choose_memblock_flags(void)
121 return system_has_some_mirror ? MEMBLOCK_MIRROR : MEMBLOCK_NONE;
124 /* adjust *@size so that (@base + *@size) doesn't overflow, return new size */
125 static inline phys_addr_t memblock_cap_size(phys_addr_t base, phys_addr_t *size)
127 return *size = min(*size, PHYS_ADDR_MAX - base);
131 * Address comparison utilities
133 static unsigned long __init_memblock memblock_addrs_overlap(phys_addr_t base1, phys_addr_t size1,
134 phys_addr_t base2, phys_addr_t size2)
136 return ((base1 < (base2 + size2)) && (base2 < (base1 + size1)));
139 bool __init_memblock memblock_overlaps_region(struct memblock_type *type,
140 phys_addr_t base, phys_addr_t size)
144 for (i = 0; i < type->cnt; i++)
145 if (memblock_addrs_overlap(base, size, type->regions[i].base,
146 type->regions[i].size))
148 return i < type->cnt;
152 * __memblock_find_range_bottom_up - find free area utility in bottom-up
153 * @start: start of candidate range
154 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_ANYWHERE or
155 * %MEMBLOCK_ALLOC_ACCESSIBLE
156 * @size: size of free area to find
157 * @align: alignment of free area to find
158 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
159 * @flags: pick from blocks based on memory attributes
161 * Utility called from memblock_find_in_range_node(), find free area bottom-up.
164 * Found address on success, 0 on failure.
166 static phys_addr_t __init_memblock
167 __memblock_find_range_bottom_up(phys_addr_t start, phys_addr_t end,
168 phys_addr_t size, phys_addr_t align, int nid,
169 enum memblock_flags flags)
171 phys_addr_t this_start, this_end, cand;
174 for_each_free_mem_range(i, nid, flags, &this_start, &this_end, NULL) {
175 this_start = clamp(this_start, start, end);
176 this_end = clamp(this_end, start, end);
178 cand = round_up(this_start, align);
179 if (cand < this_end && this_end - cand >= size)
187 * __memblock_find_range_top_down - find free area utility, in top-down
188 * @start: start of candidate range
189 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_ANYWHERE or
190 * %MEMBLOCK_ALLOC_ACCESSIBLE
191 * @size: size of free area to find
192 * @align: alignment of free area to find
193 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
194 * @flags: pick from blocks based on memory attributes
196 * Utility called from memblock_find_in_range_node(), find free area top-down.
199 * Found address on success, 0 on failure.
201 static phys_addr_t __init_memblock
202 __memblock_find_range_top_down(phys_addr_t start, phys_addr_t end,
203 phys_addr_t size, phys_addr_t align, int nid,
204 enum memblock_flags flags)
206 phys_addr_t this_start, this_end, cand;
209 for_each_free_mem_range_reverse(i, nid, flags, &this_start, &this_end,
211 this_start = clamp(this_start, start, end);
212 this_end = clamp(this_end, start, end);
217 cand = round_down(this_end - size, align);
218 if (cand >= this_start)
226 * memblock_find_in_range_node - find free area in given range and node
227 * @size: size of free area to find
228 * @align: alignment of free area to find
229 * @start: start of candidate range
230 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_ANYWHERE or
231 * %MEMBLOCK_ALLOC_ACCESSIBLE
232 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
233 * @flags: pick from blocks based on memory attributes
235 * Find @size free area aligned to @align in the specified range and node.
237 * When allocation direction is bottom-up, the @start should be greater
238 * than the end of the kernel image. Otherwise, it will be trimmed. The
239 * reason is that we want the bottom-up allocation just near the kernel
240 * image so it is highly likely that the allocated memory and the kernel
241 * will reside in the same node.
243 * If bottom-up allocation failed, will try to allocate memory top-down.
246 * Found address on success, 0 on failure.
248 phys_addr_t __init_memblock memblock_find_in_range_node(phys_addr_t size,
249 phys_addr_t align, phys_addr_t start,
250 phys_addr_t end, int nid,
251 enum memblock_flags flags)
253 phys_addr_t kernel_end, ret;
256 if (end == MEMBLOCK_ALLOC_ACCESSIBLE)
257 end = memblock.current_limit;
259 /* avoid allocating the first page */
260 start = max_t(phys_addr_t, start, PAGE_SIZE);
261 end = max(start, end);
262 kernel_end = __pa_symbol(_end);
265 * try bottom-up allocation only when bottom-up mode
266 * is set and @end is above the kernel image.
268 if (memblock_bottom_up() && end > kernel_end) {
269 phys_addr_t bottom_up_start;
271 /* make sure we will allocate above the kernel */
272 bottom_up_start = max(start, kernel_end);
274 /* ok, try bottom-up allocation first */
275 ret = __memblock_find_range_bottom_up(bottom_up_start, end,
276 size, align, nid, flags);
281 * we always limit bottom-up allocation above the kernel,
282 * but top-down allocation doesn't have the limit, so
283 * retrying top-down allocation may succeed when bottom-up
286 * bottom-up allocation is expected to be fail very rarely,
287 * so we use WARN_ONCE() here to see the stack trace if
290 WARN_ONCE(IS_ENABLED(CONFIG_MEMORY_HOTREMOVE),
291 "memblock: bottom-up allocation failed, memory hotremove may be affected\n");
294 return __memblock_find_range_top_down(start, end, size, align, nid,
299 * memblock_find_in_range - find free area in given range
300 * @start: start of candidate range
301 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_ANYWHERE or
302 * %MEMBLOCK_ALLOC_ACCESSIBLE
303 * @size: size of free area to find
304 * @align: alignment of free area to find
306 * Find @size free area aligned to @align in the specified range.
309 * Found address on success, 0 on failure.
311 phys_addr_t __init_memblock memblock_find_in_range(phys_addr_t start,
312 phys_addr_t end, phys_addr_t size,
316 enum memblock_flags flags = choose_memblock_flags();
319 ret = memblock_find_in_range_node(size, align, start, end,
320 NUMA_NO_NODE, flags);
322 if (!ret && (flags & MEMBLOCK_MIRROR)) {
323 pr_warn("Could not allocate %pap bytes of mirrored memory\n",
325 flags &= ~MEMBLOCK_MIRROR;
332 static void __init_memblock memblock_remove_region(struct memblock_type *type, unsigned long r)
334 type->total_size -= type->regions[r].size;
335 memmove(&type->regions[r], &type->regions[r + 1],
336 (type->cnt - (r + 1)) * sizeof(type->regions[r]));
339 /* Special case for empty arrays */
340 if (type->cnt == 0) {
341 WARN_ON(type->total_size != 0);
343 type->regions[0].base = 0;
344 type->regions[0].size = 0;
345 type->regions[0].flags = 0;
346 memblock_set_region_node(&type->regions[0], MAX_NUMNODES);
350 #ifdef CONFIG_ARCH_DISCARD_MEMBLOCK
352 * memblock_discard - discard memory and reserved arrays if they were allocated
354 void __init memblock_discard(void)
356 phys_addr_t addr, size;
358 if (memblock.reserved.regions != memblock_reserved_init_regions) {
359 addr = __pa(memblock.reserved.regions);
360 size = PAGE_ALIGN(sizeof(struct memblock_region) *
361 memblock.reserved.max);
362 __memblock_free_late(addr, size);
365 if (memblock.memory.regions != memblock_memory_init_regions) {
366 addr = __pa(memblock.memory.regions);
367 size = PAGE_ALIGN(sizeof(struct memblock_region) *
368 memblock.memory.max);
369 __memblock_free_late(addr, size);
375 * memblock_double_array - double the size of the memblock regions array
376 * @type: memblock type of the regions array being doubled
377 * @new_area_start: starting address of memory range to avoid overlap with
378 * @new_area_size: size of memory range to avoid overlap with
380 * Double the size of the @type regions array. If memblock is being used to
381 * allocate memory for a new reserved regions array and there is a previously
382 * allocated memory range [@new_area_start, @new_area_start + @new_area_size]
383 * waiting to be reserved, ensure the memory used by the new array does
387 * 0 on success, -1 on failure.
389 static int __init_memblock memblock_double_array(struct memblock_type *type,
390 phys_addr_t new_area_start,
391 phys_addr_t new_area_size)
393 struct memblock_region *new_array, *old_array;
394 phys_addr_t old_alloc_size, new_alloc_size;
395 phys_addr_t old_size, new_size, addr, new_end;
396 int use_slab = slab_is_available();
399 /* We don't allow resizing until we know about the reserved regions
400 * of memory that aren't suitable for allocation
402 if (!memblock_can_resize)
405 /* Calculate new doubled size */
406 old_size = type->max * sizeof(struct memblock_region);
407 new_size = old_size << 1;
409 * We need to allocated new one align to PAGE_SIZE,
410 * so we can free them completely later.
412 old_alloc_size = PAGE_ALIGN(old_size);
413 new_alloc_size = PAGE_ALIGN(new_size);
415 /* Retrieve the slab flag */
416 if (type == &memblock.memory)
417 in_slab = &memblock_memory_in_slab;
419 in_slab = &memblock_reserved_in_slab;
421 /* Try to find some space for it.
423 * WARNING: We assume that either slab_is_available() and we use it or
424 * we use MEMBLOCK for allocations. That means that this is unsafe to
425 * use when bootmem is currently active (unless bootmem itself is
426 * implemented on top of MEMBLOCK which isn't the case yet)
428 * This should however not be an issue for now, as we currently only
429 * call into MEMBLOCK while it's still active, or much later when slab
430 * is active for memory hotplug operations
433 new_array = kmalloc(new_size, GFP_KERNEL);
434 addr = new_array ? __pa(new_array) : 0;
436 /* only exclude range when trying to double reserved.regions */
437 if (type != &memblock.reserved)
438 new_area_start = new_area_size = 0;
440 addr = memblock_find_in_range(new_area_start + new_area_size,
441 memblock.current_limit,
442 new_alloc_size, PAGE_SIZE);
443 if (!addr && new_area_size)
444 addr = memblock_find_in_range(0,
445 min(new_area_start, memblock.current_limit),
446 new_alloc_size, PAGE_SIZE);
448 new_array = addr ? __va(addr) : NULL;
451 pr_err("memblock: Failed to double %s array from %ld to %ld entries !\n",
452 type->name, type->max, type->max * 2);
456 new_end = addr + new_size - 1;
457 memblock_dbg("memblock: %s is doubled to %ld at [%pa-%pa]",
458 type->name, type->max * 2, &addr, &new_end);
461 * Found space, we now need to move the array over before we add the
462 * reserved region since it may be our reserved array itself that is
465 memcpy(new_array, type->regions, old_size);
466 memset(new_array + type->max, 0, old_size);
467 old_array = type->regions;
468 type->regions = new_array;
471 /* Free old array. We needn't free it if the array is the static one */
474 else if (old_array != memblock_memory_init_regions &&
475 old_array != memblock_reserved_init_regions)
476 memblock_free(__pa(old_array), old_alloc_size);
479 * Reserve the new array if that comes from the memblock. Otherwise, we
483 BUG_ON(memblock_reserve(addr, new_alloc_size));
485 /* Update slab flag */
492 * memblock_merge_regions - merge neighboring compatible regions
493 * @type: memblock type to scan
495 * Scan @type and merge neighboring compatible regions.
497 static void __init_memblock memblock_merge_regions(struct memblock_type *type)
501 /* cnt never goes below 1 */
502 while (i < type->cnt - 1) {
503 struct memblock_region *this = &type->regions[i];
504 struct memblock_region *next = &type->regions[i + 1];
506 if (this->base + this->size != next->base ||
507 memblock_get_region_node(this) !=
508 memblock_get_region_node(next) ||
509 this->flags != next->flags) {
510 BUG_ON(this->base + this->size > next->base);
515 this->size += next->size;
516 /* move forward from next + 1, index of which is i + 2 */
517 memmove(next, next + 1, (type->cnt - (i + 2)) * sizeof(*next));
523 * memblock_insert_region - insert new memblock region
524 * @type: memblock type to insert into
525 * @idx: index for the insertion point
526 * @base: base address of the new region
527 * @size: size of the new region
528 * @nid: node id of the new region
529 * @flags: flags of the new region
531 * Insert new memblock region [@base, @base + @size) into @type at @idx.
532 * @type must already have extra room to accommodate the new region.
534 static void __init_memblock memblock_insert_region(struct memblock_type *type,
535 int idx, phys_addr_t base,
538 enum memblock_flags flags)
540 struct memblock_region *rgn = &type->regions[idx];
542 BUG_ON(type->cnt >= type->max);
543 memmove(rgn + 1, rgn, (type->cnt - idx) * sizeof(*rgn));
547 memblock_set_region_node(rgn, nid);
549 type->total_size += size;
553 * memblock_add_range - add new memblock region
554 * @type: memblock type to add new region into
555 * @base: base address of the new region
556 * @size: size of the new region
557 * @nid: nid of the new region
558 * @flags: flags of the new region
560 * Add new memblock region [@base, @base + @size) into @type. The new region
561 * is allowed to overlap with existing ones - overlaps don't affect already
562 * existing regions. @type is guaranteed to be minimal (all neighbouring
563 * compatible regions are merged) after the addition.
566 * 0 on success, -errno on failure.
568 int __init_memblock memblock_add_range(struct memblock_type *type,
569 phys_addr_t base, phys_addr_t size,
570 int nid, enum memblock_flags flags)
573 phys_addr_t obase = base;
574 phys_addr_t end = base + memblock_cap_size(base, &size);
576 struct memblock_region *rgn;
581 /* special case for empty array */
582 if (type->regions[0].size == 0) {
583 WARN_ON(type->cnt != 1 || type->total_size);
584 type->regions[0].base = base;
585 type->regions[0].size = size;
586 type->regions[0].flags = flags;
587 memblock_set_region_node(&type->regions[0], nid);
588 type->total_size = size;
593 * The following is executed twice. Once with %false @insert and
594 * then with %true. The first counts the number of regions needed
595 * to accommodate the new area. The second actually inserts them.
600 for_each_memblock_type(idx, type, rgn) {
601 phys_addr_t rbase = rgn->base;
602 phys_addr_t rend = rbase + rgn->size;
609 * @rgn overlaps. If it separates the lower part of new
610 * area, insert that portion.
613 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
614 WARN_ON(nid != memblock_get_region_node(rgn));
616 WARN_ON(flags != rgn->flags);
619 memblock_insert_region(type, idx++, base,
623 /* area below @rend is dealt with, forget about it */
624 base = min(rend, end);
627 /* insert the remaining portion */
631 memblock_insert_region(type, idx, base, end - base,
639 * If this was the first round, resize array and repeat for actual
640 * insertions; otherwise, merge and return.
643 while (type->cnt + nr_new > type->max)
644 if (memblock_double_array(type, obase, size) < 0)
649 memblock_merge_regions(type);
655 * memblock_add_node - add new memblock region within a NUMA node
656 * @base: base address of the new region
657 * @size: size of the new region
658 * @nid: nid of the new region
660 * Add new memblock region [@base, @base + @size) to the "memory"
661 * type. See memblock_add_range() description for mode details
664 * 0 on success, -errno on failure.
666 int __init_memblock memblock_add_node(phys_addr_t base, phys_addr_t size,
669 return memblock_add_range(&memblock.memory, base, size, nid, 0);
673 * memblock_add - add new memblock region
674 * @base: base address of the new region
675 * @size: size of the new region
677 * Add new memblock region [@base, @base + @size) to the "memory"
678 * type. See memblock_add_range() description for mode details
681 * 0 on success, -errno on failure.
683 int __init_memblock memblock_add(phys_addr_t base, phys_addr_t size)
685 phys_addr_t end = base + size - 1;
687 memblock_dbg("memblock_add: [%pa-%pa] %pF\n",
688 &base, &end, (void *)_RET_IP_);
690 return memblock_add_range(&memblock.memory, base, size, MAX_NUMNODES, 0);
694 * memblock_isolate_range - isolate given range into disjoint memblocks
695 * @type: memblock type to isolate range for
696 * @base: base of range to isolate
697 * @size: size of range to isolate
698 * @start_rgn: out parameter for the start of isolated region
699 * @end_rgn: out parameter for the end of isolated region
701 * Walk @type and ensure that regions don't cross the boundaries defined by
702 * [@base, @base + @size). Crossing regions are split at the boundaries,
703 * which may create at most two more regions. The index of the first
704 * region inside the range is returned in *@start_rgn and end in *@end_rgn.
707 * 0 on success, -errno on failure.
709 static int __init_memblock memblock_isolate_range(struct memblock_type *type,
710 phys_addr_t base, phys_addr_t size,
711 int *start_rgn, int *end_rgn)
713 phys_addr_t end = base + memblock_cap_size(base, &size);
715 struct memblock_region *rgn;
717 *start_rgn = *end_rgn = 0;
722 /* we'll create at most two more regions */
723 while (type->cnt + 2 > type->max)
724 if (memblock_double_array(type, base, size) < 0)
727 for_each_memblock_type(idx, type, rgn) {
728 phys_addr_t rbase = rgn->base;
729 phys_addr_t rend = rbase + rgn->size;
738 * @rgn intersects from below. Split and continue
739 * to process the next region - the new top half.
742 rgn->size -= base - rbase;
743 type->total_size -= base - rbase;
744 memblock_insert_region(type, idx, rbase, base - rbase,
745 memblock_get_region_node(rgn),
747 } else if (rend > end) {
749 * @rgn intersects from above. Split and redo the
750 * current region - the new bottom half.
753 rgn->size -= end - rbase;
754 type->total_size -= end - rbase;
755 memblock_insert_region(type, idx--, rbase, end - rbase,
756 memblock_get_region_node(rgn),
759 /* @rgn is fully contained, record it */
769 static int __init_memblock memblock_remove_range(struct memblock_type *type,
770 phys_addr_t base, phys_addr_t size)
772 int start_rgn, end_rgn;
775 ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
779 for (i = end_rgn - 1; i >= start_rgn; i--)
780 memblock_remove_region(type, i);
784 int __init_memblock memblock_remove(phys_addr_t base, phys_addr_t size)
786 phys_addr_t end = base + size - 1;
788 memblock_dbg("memblock_remove: [%pa-%pa] %pS\n",
789 &base, &end, (void *)_RET_IP_);
791 return memblock_remove_range(&memblock.memory, base, size);
795 int __init_memblock memblock_free(phys_addr_t base, phys_addr_t size)
797 phys_addr_t end = base + size - 1;
799 memblock_dbg(" memblock_free: [%pa-%pa] %pF\n",
800 &base, &end, (void *)_RET_IP_);
802 kmemleak_free_part_phys(base, size);
803 return memblock_remove_range(&memblock.reserved, base, size);
806 int __init_memblock memblock_reserve(phys_addr_t base, phys_addr_t size)
808 phys_addr_t end = base + size - 1;
810 memblock_dbg("memblock_reserve: [%pa-%pa] %pF\n",
811 &base, &end, (void *)_RET_IP_);
813 return memblock_add_range(&memblock.reserved, base, size, MAX_NUMNODES, 0);
817 * memblock_setclr_flag - set or clear flag for a memory region
818 * @base: base address of the region
819 * @size: size of the region
820 * @set: set or clear the flag
821 * @flag: the flag to udpate
823 * This function isolates region [@base, @base + @size), and sets/clears flag
825 * Return: 0 on success, -errno on failure.
827 static int __init_memblock memblock_setclr_flag(phys_addr_t base,
828 phys_addr_t size, int set, int flag)
830 struct memblock_type *type = &memblock.memory;
831 int i, ret, start_rgn, end_rgn;
833 ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
837 for (i = start_rgn; i < end_rgn; i++)
839 memblock_set_region_flags(&type->regions[i], flag);
841 memblock_clear_region_flags(&type->regions[i], flag);
843 memblock_merge_regions(type);
848 * memblock_mark_hotplug - Mark hotpluggable memory with flag MEMBLOCK_HOTPLUG.
849 * @base: the base phys addr of the region
850 * @size: the size of the region
852 * Return: 0 on success, -errno on failure.
854 int __init_memblock memblock_mark_hotplug(phys_addr_t base, phys_addr_t size)
856 return memblock_setclr_flag(base, size, 1, MEMBLOCK_HOTPLUG);
860 * memblock_clear_hotplug - Clear flag MEMBLOCK_HOTPLUG for a specified region.
861 * @base: the base phys addr of the region
862 * @size: the size of the region
864 * Return: 0 on success, -errno on failure.
866 int __init_memblock memblock_clear_hotplug(phys_addr_t base, phys_addr_t size)
868 return memblock_setclr_flag(base, size, 0, MEMBLOCK_HOTPLUG);
872 * memblock_mark_mirror - Mark mirrored memory with flag MEMBLOCK_MIRROR.
873 * @base: the base phys addr of the region
874 * @size: the size of the region
876 * Return: 0 on success, -errno on failure.
878 int __init_memblock memblock_mark_mirror(phys_addr_t base, phys_addr_t size)
880 system_has_some_mirror = true;
882 return memblock_setclr_flag(base, size, 1, MEMBLOCK_MIRROR);
886 * memblock_mark_nomap - Mark a memory region with flag MEMBLOCK_NOMAP.
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_mark_nomap(phys_addr_t base, phys_addr_t size)
894 return memblock_setclr_flag(base, size, 1, MEMBLOCK_NOMAP);
898 * memblock_clear_nomap - Clear flag MEMBLOCK_NOMAP for a specified region.
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_clear_nomap(phys_addr_t base, phys_addr_t size)
906 return memblock_setclr_flag(base, size, 0, MEMBLOCK_NOMAP);
910 * __next_reserved_mem_region - next function for for_each_reserved_region()
911 * @idx: pointer to u64 loop variable
912 * @out_start: ptr to phys_addr_t for start address of the region, can be %NULL
913 * @out_end: ptr to phys_addr_t for end address of the region, can be %NULL
915 * Iterate over all reserved memory regions.
917 void __init_memblock __next_reserved_mem_region(u64 *idx,
918 phys_addr_t *out_start,
919 phys_addr_t *out_end)
921 struct memblock_type *type = &memblock.reserved;
923 if (*idx < type->cnt) {
924 struct memblock_region *r = &type->regions[*idx];
925 phys_addr_t base = r->base;
926 phys_addr_t size = r->size;
931 *out_end = base + size - 1;
937 /* signal end of iteration */
942 * __next__mem_range - next function for for_each_free_mem_range() etc.
943 * @idx: pointer to u64 loop variable
944 * @nid: node selector, %NUMA_NO_NODE for all nodes
945 * @flags: pick from blocks based on memory attributes
946 * @type_a: pointer to memblock_type from where the range is taken
947 * @type_b: pointer to memblock_type which excludes memory from being taken
948 * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL
949 * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL
950 * @out_nid: ptr to int for nid of the range, can be %NULL
952 * Find the first area from *@idx which matches @nid, fill the out
953 * parameters, and update *@idx for the next iteration. The lower 32bit of
954 * *@idx contains index into type_a and the upper 32bit indexes the
955 * areas before each region in type_b. For example, if type_b regions
956 * look like the following,
958 * 0:[0-16), 1:[32-48), 2:[128-130)
960 * The upper 32bit indexes the following regions.
962 * 0:[0-0), 1:[16-32), 2:[48-128), 3:[130-MAX)
964 * As both region arrays are sorted, the function advances the two indices
965 * in lockstep and returns each intersection.
967 void __init_memblock __next_mem_range(u64 *idx, int nid,
968 enum memblock_flags flags,
969 struct memblock_type *type_a,
970 struct memblock_type *type_b,
971 phys_addr_t *out_start,
972 phys_addr_t *out_end, int *out_nid)
974 int idx_a = *idx & 0xffffffff;
975 int idx_b = *idx >> 32;
977 if (WARN_ONCE(nid == MAX_NUMNODES,
978 "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
981 for (; idx_a < type_a->cnt; idx_a++) {
982 struct memblock_region *m = &type_a->regions[idx_a];
984 phys_addr_t m_start = m->base;
985 phys_addr_t m_end = m->base + m->size;
986 int m_nid = memblock_get_region_node(m);
988 /* only memory regions are associated with nodes, check it */
989 if (nid != NUMA_NO_NODE && nid != m_nid)
992 /* skip hotpluggable memory regions if needed */
993 if (movable_node_is_enabled() && memblock_is_hotpluggable(m))
996 /* if we want mirror memory skip non-mirror memory regions */
997 if ((flags & MEMBLOCK_MIRROR) && !memblock_is_mirror(m))
1000 /* skip nomap memory unless we were asked for it explicitly */
1001 if (!(flags & MEMBLOCK_NOMAP) && memblock_is_nomap(m))
1006 *out_start = m_start;
1012 *idx = (u32)idx_a | (u64)idx_b << 32;
1016 /* scan areas before each reservation */
1017 for (; idx_b < type_b->cnt + 1; idx_b++) {
1018 struct memblock_region *r;
1019 phys_addr_t r_start;
1022 r = &type_b->regions[idx_b];
1023 r_start = idx_b ? r[-1].base + r[-1].size : 0;
1024 r_end = idx_b < type_b->cnt ?
1025 r->base : PHYS_ADDR_MAX;
1028 * if idx_b advanced past idx_a,
1029 * break out to advance idx_a
1031 if (r_start >= m_end)
1033 /* if the two regions intersect, we're done */
1034 if (m_start < r_end) {
1037 max(m_start, r_start);
1039 *out_end = min(m_end, r_end);
1043 * The region which ends first is
1044 * advanced for the next iteration.
1050 *idx = (u32)idx_a | (u64)idx_b << 32;
1056 /* signal end of iteration */
1061 * __next_mem_range_rev - generic next function for for_each_*_range_rev()
1063 * @idx: pointer to u64 loop variable
1064 * @nid: node selector, %NUMA_NO_NODE for all nodes
1065 * @flags: pick from blocks based on memory attributes
1066 * @type_a: pointer to memblock_type from where the range is taken
1067 * @type_b: pointer to memblock_type which excludes memory from being taken
1068 * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL
1069 * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL
1070 * @out_nid: ptr to int for nid of the range, can be %NULL
1072 * Finds the next range from type_a which is not marked as unsuitable
1075 * Reverse of __next_mem_range().
1077 void __init_memblock __next_mem_range_rev(u64 *idx, int nid,
1078 enum memblock_flags flags,
1079 struct memblock_type *type_a,
1080 struct memblock_type *type_b,
1081 phys_addr_t *out_start,
1082 phys_addr_t *out_end, int *out_nid)
1084 int idx_a = *idx & 0xffffffff;
1085 int idx_b = *idx >> 32;
1087 if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
1090 if (*idx == (u64)ULLONG_MAX) {
1091 idx_a = type_a->cnt - 1;
1093 idx_b = type_b->cnt;
1098 for (; idx_a >= 0; idx_a--) {
1099 struct memblock_region *m = &type_a->regions[idx_a];
1101 phys_addr_t m_start = m->base;
1102 phys_addr_t m_end = m->base + m->size;
1103 int m_nid = memblock_get_region_node(m);
1105 /* only memory regions are associated with nodes, check it */
1106 if (nid != NUMA_NO_NODE && nid != m_nid)
1109 /* skip hotpluggable memory regions if needed */
1110 if (movable_node_is_enabled() && memblock_is_hotpluggable(m))
1113 /* if we want mirror memory skip non-mirror memory regions */
1114 if ((flags & MEMBLOCK_MIRROR) && !memblock_is_mirror(m))
1117 /* skip nomap memory unless we were asked for it explicitly */
1118 if (!(flags & MEMBLOCK_NOMAP) && memblock_is_nomap(m))
1123 *out_start = m_start;
1129 *idx = (u32)idx_a | (u64)idx_b << 32;
1133 /* scan areas before each reservation */
1134 for (; idx_b >= 0; idx_b--) {
1135 struct memblock_region *r;
1136 phys_addr_t r_start;
1139 r = &type_b->regions[idx_b];
1140 r_start = idx_b ? r[-1].base + r[-1].size : 0;
1141 r_end = idx_b < type_b->cnt ?
1142 r->base : PHYS_ADDR_MAX;
1144 * if idx_b advanced past idx_a,
1145 * break out to advance idx_a
1148 if (r_end <= m_start)
1150 /* if the two regions intersect, we're done */
1151 if (m_end > r_start) {
1153 *out_start = max(m_start, r_start);
1155 *out_end = min(m_end, r_end);
1158 if (m_start >= r_start)
1162 *idx = (u32)idx_a | (u64)idx_b << 32;
1167 /* signal end of iteration */
1171 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
1173 * Common iterator interface used to define for_each_mem_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;
1182 while (++*idx < type->cnt) {
1183 r = &type->regions[*idx];
1185 if (PFN_UP(r->base) >= PFN_DOWN(r->base + r->size))
1187 if (nid == MAX_NUMNODES || nid == r->nid)
1190 if (*idx >= type->cnt) {
1196 *out_start_pfn = PFN_UP(r->base);
1198 *out_end_pfn = PFN_DOWN(r->base + r->size);
1204 * memblock_set_node - set node ID on memblock regions
1205 * @base: base of area to set node ID for
1206 * @size: size of area to set node ID for
1207 * @type: memblock type to set node ID for
1208 * @nid: node ID to set
1210 * Set the nid of memblock @type regions in [@base, @base + @size) to @nid.
1211 * Regions which cross the area boundaries are split as necessary.
1214 * 0 on success, -errno on failure.
1216 int __init_memblock memblock_set_node(phys_addr_t base, phys_addr_t size,
1217 struct memblock_type *type, int nid)
1219 int start_rgn, end_rgn;
1222 ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
1226 for (i = start_rgn; i < end_rgn; i++)
1227 memblock_set_region_node(&type->regions[i], nid);
1229 memblock_merge_regions(type);
1232 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
1234 static phys_addr_t __init memblock_alloc_range_nid(phys_addr_t size,
1235 phys_addr_t align, phys_addr_t start,
1236 phys_addr_t end, int nid,
1237 enum memblock_flags flags)
1242 align = SMP_CACHE_BYTES;
1244 found = memblock_find_in_range_node(size, align, start, end, nid,
1246 if (found && !memblock_reserve(found, size)) {
1248 * The min_count is set to 0 so that memblock allocations are
1249 * never reported as leaks.
1251 kmemleak_alloc_phys(found, size, 0, 0);
1257 phys_addr_t __init memblock_alloc_range(phys_addr_t size, phys_addr_t align,
1258 phys_addr_t start, phys_addr_t end,
1259 enum memblock_flags flags)
1261 return memblock_alloc_range_nid(size, align, start, end, NUMA_NO_NODE,
1265 phys_addr_t __init memblock_alloc_base_nid(phys_addr_t size,
1266 phys_addr_t align, phys_addr_t max_addr,
1267 int nid, enum memblock_flags flags)
1269 return memblock_alloc_range_nid(size, align, 0, max_addr, nid, flags);
1272 phys_addr_t __init memblock_alloc_nid(phys_addr_t size, phys_addr_t align, int nid)
1274 enum memblock_flags flags = choose_memblock_flags();
1278 ret = memblock_alloc_base_nid(size, align, MEMBLOCK_ALLOC_ACCESSIBLE,
1281 if (!ret && (flags & MEMBLOCK_MIRROR)) {
1282 flags &= ~MEMBLOCK_MIRROR;
1288 phys_addr_t __init __memblock_alloc_base(phys_addr_t size, phys_addr_t align, phys_addr_t max_addr)
1290 return memblock_alloc_base_nid(size, align, max_addr, NUMA_NO_NODE,
1294 phys_addr_t __init memblock_alloc_base(phys_addr_t size, phys_addr_t align, phys_addr_t max_addr)
1298 alloc = __memblock_alloc_base(size, align, max_addr);
1301 panic("ERROR: Failed to allocate %pa bytes below %pa.\n",
1307 phys_addr_t __init memblock_alloc(phys_addr_t size, phys_addr_t align)
1309 return memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ACCESSIBLE);
1312 phys_addr_t __init memblock_alloc_try_nid(phys_addr_t size, phys_addr_t align, int nid)
1314 phys_addr_t res = memblock_alloc_nid(size, align, nid);
1318 return memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ACCESSIBLE);
1321 #if defined(CONFIG_NO_BOOTMEM)
1323 * memblock_virt_alloc_internal - allocate boot memory block
1324 * @size: size of memory block to be allocated in bytes
1325 * @align: alignment of the region and block's size
1326 * @min_addr: the lower bound of the memory region to allocate (phys address)
1327 * @max_addr: the upper bound of the memory region to allocate (phys address)
1328 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1330 * The @min_addr limit is dropped if it can not be satisfied and the allocation
1331 * will fall back to memory below @min_addr. Also, allocation may fall back
1332 * to any node in the system if the specified node can not
1333 * hold the requested memory.
1335 * The allocation is performed from memory region limited by
1336 * memblock.current_limit if @max_addr == %BOOTMEM_ALLOC_ACCESSIBLE.
1338 * The memory block is aligned on %SMP_CACHE_BYTES if @align == 0.
1340 * The phys address of allocated boot memory block is converted to virtual and
1341 * allocated memory is reset to 0.
1343 * In addition, function sets the min_count to 0 using kmemleak_alloc for
1344 * allocated boot memory block, so that it is never reported as leaks.
1347 * Virtual address of allocated memory block on success, NULL on failure.
1349 static void * __init memblock_virt_alloc_internal(
1350 phys_addr_t size, phys_addr_t align,
1351 phys_addr_t min_addr, phys_addr_t max_addr,
1356 enum memblock_flags flags = choose_memblock_flags();
1358 if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
1362 * Detect any accidental use of these APIs after slab is ready, as at
1363 * this moment memblock may be deinitialized already and its
1364 * internal data may be destroyed (after execution of free_all_bootmem)
1366 if (WARN_ON_ONCE(slab_is_available()))
1367 return kzalloc_node(size, GFP_NOWAIT, nid);
1370 align = SMP_CACHE_BYTES;
1372 if (max_addr > memblock.current_limit)
1373 max_addr = memblock.current_limit;
1375 alloc = memblock_find_in_range_node(size, align, min_addr, max_addr,
1377 if (alloc && !memblock_reserve(alloc, size))
1380 if (nid != NUMA_NO_NODE) {
1381 alloc = memblock_find_in_range_node(size, align, min_addr,
1382 max_addr, NUMA_NO_NODE,
1384 if (alloc && !memblock_reserve(alloc, size))
1393 if (flags & MEMBLOCK_MIRROR) {
1394 flags &= ~MEMBLOCK_MIRROR;
1395 pr_warn("Could not allocate %pap bytes of mirrored memory\n",
1402 ptr = phys_to_virt(alloc);
1405 * The min_count is set to 0 so that bootmem allocated blocks
1406 * are never reported as leaks. This is because many of these blocks
1407 * are only referred via the physical address which is not
1408 * looked up by kmemleak.
1410 kmemleak_alloc(ptr, size, 0, 0);
1416 * memblock_virt_alloc_try_nid_raw - allocate boot memory block without zeroing
1417 * memory and without panicking
1418 * @size: size of memory block to be allocated in bytes
1419 * @align: alignment of the region and block's size
1420 * @min_addr: the lower bound of the memory region from where the allocation
1421 * is preferred (phys address)
1422 * @max_addr: the upper bound of the memory region from where the allocation
1423 * is preferred (phys address), or %BOOTMEM_ALLOC_ACCESSIBLE to
1424 * allocate only from memory limited by memblock.current_limit value
1425 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1427 * Public function, provides additional debug information (including caller
1428 * info), if enabled. Does not zero allocated memory, does not panic if request
1429 * cannot be satisfied.
1432 * Virtual address of allocated memory block on success, NULL on failure.
1434 void * __init memblock_virt_alloc_try_nid_raw(
1435 phys_addr_t size, phys_addr_t align,
1436 phys_addr_t min_addr, phys_addr_t max_addr,
1441 memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=%pa max_addr=%pa %pF\n",
1442 __func__, (u64)size, (u64)align, nid, &min_addr,
1443 &max_addr, (void *)_RET_IP_);
1445 ptr = memblock_virt_alloc_internal(size, align,
1446 min_addr, max_addr, nid);
1447 #ifdef CONFIG_DEBUG_VM
1448 if (ptr && size > 0)
1449 memset(ptr, PAGE_POISON_PATTERN, size);
1455 * memblock_virt_alloc_try_nid_nopanic - allocate boot memory block
1456 * @size: size of memory block to be allocated in bytes
1457 * @align: alignment of the region and block's size
1458 * @min_addr: the lower bound of the memory region from where the allocation
1459 * is preferred (phys address)
1460 * @max_addr: the upper bound of the memory region from where the allocation
1461 * is preferred (phys address), or %BOOTMEM_ALLOC_ACCESSIBLE to
1462 * allocate only from memory limited by memblock.current_limit value
1463 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1465 * Public function, provides additional debug information (including caller
1466 * info), if enabled. This function zeroes the allocated memory.
1469 * Virtual address of allocated memory block on success, NULL on failure.
1471 void * __init memblock_virt_alloc_try_nid_nopanic(
1472 phys_addr_t size, phys_addr_t align,
1473 phys_addr_t min_addr, phys_addr_t max_addr,
1478 memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=%pa max_addr=%pa %pF\n",
1479 __func__, (u64)size, (u64)align, nid, &min_addr,
1480 &max_addr, (void *)_RET_IP_);
1482 ptr = memblock_virt_alloc_internal(size, align,
1483 min_addr, max_addr, nid);
1485 memset(ptr, 0, size);
1490 * memblock_virt_alloc_try_nid - allocate boot memory block with panicking
1491 * @size: size of memory block to be allocated in bytes
1492 * @align: alignment of the region and block's size
1493 * @min_addr: the lower bound of the memory region from where the allocation
1494 * is preferred (phys address)
1495 * @max_addr: the upper bound of the memory region from where the allocation
1496 * is preferred (phys address), or %BOOTMEM_ALLOC_ACCESSIBLE to
1497 * allocate only from memory limited by memblock.current_limit value
1498 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1500 * Public panicking version of memblock_virt_alloc_try_nid_nopanic()
1501 * which provides debug information (including caller info), if enabled,
1502 * and panics if the request can not be satisfied.
1505 * Virtual address of allocated memory block on success, NULL on failure.
1507 void * __init memblock_virt_alloc_try_nid(
1508 phys_addr_t size, phys_addr_t align,
1509 phys_addr_t min_addr, phys_addr_t max_addr,
1514 memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=%pa max_addr=%pa %pF\n",
1515 __func__, (u64)size, (u64)align, nid, &min_addr,
1516 &max_addr, (void *)_RET_IP_);
1517 ptr = memblock_virt_alloc_internal(size, align,
1518 min_addr, max_addr, nid);
1520 memset(ptr, 0, size);
1524 panic("%s: Failed to allocate %llu bytes align=0x%llx nid=%d from=%pa max_addr=%pa\n",
1525 __func__, (u64)size, (u64)align, nid, &min_addr, &max_addr);
1531 * __memblock_free_early - free boot memory block
1532 * @base: phys starting address of the boot memory block
1533 * @size: size of the boot memory block in bytes
1535 * Free boot memory block previously allocated by memblock_virt_alloc_xx() API.
1536 * The freeing memory will not be released to the buddy allocator.
1538 void __init __memblock_free_early(phys_addr_t base, phys_addr_t size)
1540 phys_addr_t end = base + size - 1;
1542 memblock_dbg("%s: [%pa-%pa] %pF\n",
1543 __func__, &base, &end, (void *)_RET_IP_);
1544 kmemleak_free_part_phys(base, size);
1545 memblock_remove_range(&memblock.reserved, base, size);
1549 * __memblock_free_late - free bootmem block pages directly to buddy allocator
1550 * @base: phys starting address of the boot memory block
1551 * @size: size of the boot memory block in bytes
1553 * This is only useful when the bootmem allocator has already been torn
1554 * down, but we are still initializing the system. Pages are released directly
1555 * to the buddy allocator, no bootmem metadata is updated because it is gone.
1557 void __init __memblock_free_late(phys_addr_t base, phys_addr_t size)
1559 phys_addr_t cursor, end;
1561 end = base + size - 1;
1562 memblock_dbg("%s: [%pa-%pa] %pF\n",
1563 __func__, &base, &end, (void *)_RET_IP_);
1564 kmemleak_free_part_phys(base, size);
1565 cursor = PFN_UP(base);
1566 end = PFN_DOWN(base + size);
1568 for (; cursor < end; cursor++) {
1569 __free_pages_bootmem(pfn_to_page(cursor), cursor, 0);
1575 * Remaining API functions
1578 phys_addr_t __init_memblock memblock_phys_mem_size(void)
1580 return memblock.memory.total_size;
1583 phys_addr_t __init_memblock memblock_reserved_size(void)
1585 return memblock.reserved.total_size;
1588 phys_addr_t __init memblock_mem_size(unsigned long limit_pfn)
1590 unsigned long pages = 0;
1591 struct memblock_region *r;
1592 unsigned long start_pfn, end_pfn;
1594 for_each_memblock(memory, r) {
1595 start_pfn = memblock_region_memory_base_pfn(r);
1596 end_pfn = memblock_region_memory_end_pfn(r);
1597 start_pfn = min_t(unsigned long, start_pfn, limit_pfn);
1598 end_pfn = min_t(unsigned long, end_pfn, limit_pfn);
1599 pages += end_pfn - start_pfn;
1602 return PFN_PHYS(pages);
1605 /* lowest address */
1606 phys_addr_t __init_memblock memblock_start_of_DRAM(void)
1608 return memblock.memory.regions[0].base;
1611 phys_addr_t __init_memblock memblock_end_of_DRAM(void)
1613 int idx = memblock.memory.cnt - 1;
1615 return (memblock.memory.regions[idx].base + memblock.memory.regions[idx].size);
1618 static phys_addr_t __init_memblock __find_max_addr(phys_addr_t limit)
1620 phys_addr_t max_addr = PHYS_ADDR_MAX;
1621 struct memblock_region *r;
1624 * translate the memory @limit size into the max address within one of
1625 * the memory memblock regions, if the @limit exceeds the total size
1626 * of those regions, max_addr will keep original value PHYS_ADDR_MAX
1628 for_each_memblock(memory, r) {
1629 if (limit <= r->size) {
1630 max_addr = r->base + limit;
1639 void __init memblock_enforce_memory_limit(phys_addr_t limit)
1641 phys_addr_t max_addr = PHYS_ADDR_MAX;
1646 max_addr = __find_max_addr(limit);
1648 /* @limit exceeds the total size of the memory, do nothing */
1649 if (max_addr == PHYS_ADDR_MAX)
1652 /* truncate both memory and reserved regions */
1653 memblock_remove_range(&memblock.memory, max_addr,
1655 memblock_remove_range(&memblock.reserved, max_addr,
1659 void __init memblock_cap_memory_range(phys_addr_t base, phys_addr_t size)
1661 int start_rgn, end_rgn;
1667 ret = memblock_isolate_range(&memblock.memory, base, size,
1668 &start_rgn, &end_rgn);
1672 /* remove all the MAP regions */
1673 for (i = memblock.memory.cnt - 1; i >= end_rgn; i--)
1674 if (!memblock_is_nomap(&memblock.memory.regions[i]))
1675 memblock_remove_region(&memblock.memory, i);
1677 for (i = start_rgn - 1; i >= 0; i--)
1678 if (!memblock_is_nomap(&memblock.memory.regions[i]))
1679 memblock_remove_region(&memblock.memory, i);
1681 /* truncate the reserved regions */
1682 memblock_remove_range(&memblock.reserved, 0, base);
1683 memblock_remove_range(&memblock.reserved,
1684 base + size, PHYS_ADDR_MAX);
1687 void __init memblock_mem_limit_remove_map(phys_addr_t limit)
1689 phys_addr_t max_addr;
1694 max_addr = __find_max_addr(limit);
1696 /* @limit exceeds the total size of the memory, do nothing */
1697 if (max_addr == PHYS_ADDR_MAX)
1700 memblock_cap_memory_range(0, max_addr);
1703 static int __init_memblock memblock_search(struct memblock_type *type, phys_addr_t addr)
1705 unsigned int left = 0, right = type->cnt;
1708 unsigned int mid = (right + left) / 2;
1710 if (addr < type->regions[mid].base)
1712 else if (addr >= (type->regions[mid].base +
1713 type->regions[mid].size))
1717 } while (left < right);
1721 bool __init memblock_is_reserved(phys_addr_t addr)
1723 return memblock_search(&memblock.reserved, addr) != -1;
1726 bool __init_memblock memblock_is_memory(phys_addr_t addr)
1728 return memblock_search(&memblock.memory, addr) != -1;
1731 bool __init_memblock memblock_is_map_memory(phys_addr_t addr)
1733 int i = memblock_search(&memblock.memory, addr);
1737 return !memblock_is_nomap(&memblock.memory.regions[i]);
1740 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
1741 int __init_memblock memblock_search_pfn_nid(unsigned long pfn,
1742 unsigned long *start_pfn, unsigned long *end_pfn)
1744 struct memblock_type *type = &memblock.memory;
1745 int mid = memblock_search(type, PFN_PHYS(pfn));
1750 *start_pfn = PFN_DOWN(type->regions[mid].base);
1751 *end_pfn = PFN_DOWN(type->regions[mid].base + type->regions[mid].size);
1753 return type->regions[mid].nid;
1758 * memblock_is_region_memory - check if a region is a subset of memory
1759 * @base: base of region to check
1760 * @size: size of region to check
1762 * Check if the region [@base, @base + @size) is a subset of a memory block.
1765 * 0 if false, non-zero if true
1767 bool __init_memblock memblock_is_region_memory(phys_addr_t base, phys_addr_t size)
1769 int idx = memblock_search(&memblock.memory, base);
1770 phys_addr_t end = base + memblock_cap_size(base, &size);
1774 return (memblock.memory.regions[idx].base +
1775 memblock.memory.regions[idx].size) >= end;
1779 * memblock_is_region_reserved - check if a region intersects reserved memory
1780 * @base: base of region to check
1781 * @size: size of region to check
1783 * Check if the region [@base, @base + @size) intersects a reserved
1787 * True if they intersect, false if not.
1789 bool __init_memblock memblock_is_region_reserved(phys_addr_t base, phys_addr_t size)
1791 memblock_cap_size(base, &size);
1792 return memblock_overlaps_region(&memblock.reserved, base, size);
1795 void __init_memblock memblock_trim_memory(phys_addr_t align)
1797 phys_addr_t start, end, orig_start, orig_end;
1798 struct memblock_region *r;
1800 for_each_memblock(memory, r) {
1801 orig_start = r->base;
1802 orig_end = r->base + r->size;
1803 start = round_up(orig_start, align);
1804 end = round_down(orig_end, align);
1806 if (start == orig_start && end == orig_end)
1811 r->size = end - start;
1813 memblock_remove_region(&memblock.memory,
1814 r - memblock.memory.regions);
1820 void __init_memblock memblock_set_current_limit(phys_addr_t limit)
1822 memblock.current_limit = limit;
1825 phys_addr_t __init_memblock memblock_get_current_limit(void)
1827 return memblock.current_limit;
1830 static void __init_memblock memblock_dump(struct memblock_type *type)
1832 phys_addr_t base, end, size;
1833 enum memblock_flags flags;
1835 struct memblock_region *rgn;
1837 pr_info(" %s.cnt = 0x%lx\n", type->name, type->cnt);
1839 for_each_memblock_type(idx, type, rgn) {
1840 char nid_buf[32] = "";
1844 end = base + size - 1;
1846 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
1847 if (memblock_get_region_node(rgn) != MAX_NUMNODES)
1848 snprintf(nid_buf, sizeof(nid_buf), " on node %d",
1849 memblock_get_region_node(rgn));
1851 pr_info(" %s[%#x]\t[%pa-%pa], %pa bytes%s flags: %#x\n",
1852 type->name, idx, &base, &end, &size, nid_buf, flags);
1856 void __init_memblock __memblock_dump_all(void)
1858 pr_info("MEMBLOCK configuration:\n");
1859 pr_info(" memory size = %pa reserved size = %pa\n",
1860 &memblock.memory.total_size,
1861 &memblock.reserved.total_size);
1863 memblock_dump(&memblock.memory);
1864 memblock_dump(&memblock.reserved);
1865 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
1866 memblock_dump(&memblock.physmem);
1870 void __init memblock_allow_resize(void)
1872 memblock_can_resize = 1;
1875 static int __init early_memblock(char *p)
1877 if (p && strstr(p, "debug"))
1881 early_param("memblock", early_memblock);
1883 #if defined(CONFIG_DEBUG_FS) && !defined(CONFIG_ARCH_DISCARD_MEMBLOCK)
1885 static int memblock_debug_show(struct seq_file *m, void *private)
1887 struct memblock_type *type = m->private;
1888 struct memblock_region *reg;
1892 for (i = 0; i < type->cnt; i++) {
1893 reg = &type->regions[i];
1894 end = reg->base + reg->size - 1;
1896 seq_printf(m, "%4d: ", i);
1897 seq_printf(m, "%pa..%pa\n", ®->base, &end);
1901 DEFINE_SHOW_ATTRIBUTE(memblock_debug);
1903 static int __init memblock_init_debugfs(void)
1905 struct dentry *root = debugfs_create_dir("memblock", NULL);
1908 debugfs_create_file("memory", 0444, root,
1909 &memblock.memory, &memblock_debug_fops);
1910 debugfs_create_file("reserved", 0444, root,
1911 &memblock.reserved, &memblock_debug_fops);
1912 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
1913 debugfs_create_file("physmem", 0444, root,
1914 &memblock.physmem, &memblock_debug_fops);
1919 __initcall(memblock_init_debugfs);
1921 #endif /* CONFIG_DEBUG_FS */