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>
24 #include <asm/sections.h>
30 * DOC: memblock overview
32 * Memblock is a method of managing memory regions during the early
33 * boot period when the usual kernel memory allocators are not up and
36 * Memblock views the system memory as collections of contiguous
37 * regions. There are several types of these collections:
39 * * ``memory`` - describes the physical memory available to the
40 * kernel; this may differ from the actual physical memory installed
41 * in the system, for instance when the memory is restricted with
42 * ``mem=`` command line parameter
43 * * ``reserved`` - describes the regions that were allocated
44 * * ``physmap`` - describes the actual physical memory regardless of
45 * the possible restrictions; the ``physmap`` type is only available
46 * on some architectures.
48 * Each region is represented by :c:type:`struct memblock_region` that
49 * defines the region extents, its attributes and NUMA node id on NUMA
50 * systems. Every memory type is described by the :c:type:`struct
51 * memblock_type` which contains an array of memory regions along with
52 * the allocator metadata. The memory types are nicely wrapped with
53 * :c:type:`struct memblock`. This structure is statically initialzed
54 * at build time. The region arrays for the "memory" and "reserved"
55 * types are initially sized to %INIT_MEMBLOCK_REGIONS and for the
56 * "physmap" type to %INIT_PHYSMEM_REGIONS.
57 * The :c:func:`memblock_allow_resize` enables automatic resizing of
58 * the region arrays during addition of new regions. This feature
59 * should be used with care so that memory allocated for the region
60 * array will not overlap with areas that should be reserved, for
63 * The early architecture setup should tell memblock what the physical
64 * memory layout is by using :c:func:`memblock_add` or
65 * :c:func:`memblock_add_node` functions. The first function does not
66 * assign the region to a NUMA node and it is appropriate for UMA
67 * systems. Yet, it is possible to use it on NUMA systems as well and
68 * assign the region to a NUMA node later in the setup process using
69 * :c:func:`memblock_set_node`. The :c:func:`memblock_add_node`
70 * performs such an assignment directly.
72 * Once memblock is setup the memory can be allocated using either
73 * memblock or bootmem APIs.
75 * As the system boot progresses, the architecture specific
76 * :c:func:`mem_init` function frees all the memory to the buddy page
79 * If an architecure enables %CONFIG_ARCH_DISCARD_MEMBLOCK, the
80 * memblock data structures will be discarded after the system
81 * initialization compltes.
84 #ifndef CONFIG_NEED_MULTIPLE_NODES
85 struct pglist_data __refdata contig_page_data;
86 EXPORT_SYMBOL(contig_page_data);
89 unsigned long max_low_pfn;
90 unsigned long min_low_pfn;
91 unsigned long max_pfn;
92 unsigned long long max_possible_pfn;
94 static struct memblock_region memblock_memory_init_regions[INIT_MEMBLOCK_REGIONS] __initdata_memblock;
95 static struct memblock_region memblock_reserved_init_regions[INIT_MEMBLOCK_REGIONS] __initdata_memblock;
96 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
97 static struct memblock_region memblock_physmem_init_regions[INIT_PHYSMEM_REGIONS] __initdata_memblock;
100 struct memblock memblock __initdata_memblock = {
101 .memory.regions = memblock_memory_init_regions,
102 .memory.cnt = 1, /* empty dummy entry */
103 .memory.max = INIT_MEMBLOCK_REGIONS,
104 .memory.name = "memory",
106 .reserved.regions = memblock_reserved_init_regions,
107 .reserved.cnt = 1, /* empty dummy entry */
108 .reserved.max = INIT_MEMBLOCK_REGIONS,
109 .reserved.name = "reserved",
111 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
112 .physmem.regions = memblock_physmem_init_regions,
113 .physmem.cnt = 1, /* empty dummy entry */
114 .physmem.max = INIT_PHYSMEM_REGIONS,
115 .physmem.name = "physmem",
119 .current_limit = MEMBLOCK_ALLOC_ANYWHERE,
122 int memblock_debug __initdata_memblock;
123 static bool system_has_some_mirror __initdata_memblock = false;
124 static int memblock_can_resize __initdata_memblock;
125 static int memblock_memory_in_slab __initdata_memblock = 0;
126 static int memblock_reserved_in_slab __initdata_memblock = 0;
128 enum memblock_flags __init_memblock choose_memblock_flags(void)
130 return system_has_some_mirror ? MEMBLOCK_MIRROR : MEMBLOCK_NONE;
133 /* adjust *@size so that (@base + *@size) doesn't overflow, return new size */
134 static inline phys_addr_t memblock_cap_size(phys_addr_t base, phys_addr_t *size)
136 return *size = min(*size, PHYS_ADDR_MAX - base);
140 * Address comparison utilities
142 static unsigned long __init_memblock memblock_addrs_overlap(phys_addr_t base1, phys_addr_t size1,
143 phys_addr_t base2, phys_addr_t size2)
145 return ((base1 < (base2 + size2)) && (base2 < (base1 + size1)));
148 bool __init_memblock memblock_overlaps_region(struct memblock_type *type,
149 phys_addr_t base, phys_addr_t size)
153 for (i = 0; i < type->cnt; i++)
154 if (memblock_addrs_overlap(base, size, type->regions[i].base,
155 type->regions[i].size))
157 return i < type->cnt;
161 * __memblock_find_range_bottom_up - find free area utility in bottom-up
162 * @start: start of candidate range
163 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_ANYWHERE or
164 * %MEMBLOCK_ALLOC_ACCESSIBLE
165 * @size: size of free area to find
166 * @align: alignment of free area to find
167 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
168 * @flags: pick from blocks based on memory attributes
170 * Utility called from memblock_find_in_range_node(), find free area bottom-up.
173 * Found address on success, 0 on failure.
175 static phys_addr_t __init_memblock
176 __memblock_find_range_bottom_up(phys_addr_t start, phys_addr_t end,
177 phys_addr_t size, phys_addr_t align, int nid,
178 enum memblock_flags flags)
180 phys_addr_t this_start, this_end, cand;
183 for_each_free_mem_range(i, nid, flags, &this_start, &this_end, NULL) {
184 this_start = clamp(this_start, start, end);
185 this_end = clamp(this_end, start, end);
187 cand = round_up(this_start, align);
188 if (cand < this_end && this_end - cand >= size)
196 * __memblock_find_range_top_down - find free area utility, in top-down
197 * @start: start of candidate range
198 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_ANYWHERE or
199 * %MEMBLOCK_ALLOC_ACCESSIBLE
200 * @size: size of free area to find
201 * @align: alignment of free area to find
202 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
203 * @flags: pick from blocks based on memory attributes
205 * Utility called from memblock_find_in_range_node(), find free area top-down.
208 * Found address on success, 0 on failure.
210 static phys_addr_t __init_memblock
211 __memblock_find_range_top_down(phys_addr_t start, phys_addr_t end,
212 phys_addr_t size, phys_addr_t align, int nid,
213 enum memblock_flags flags)
215 phys_addr_t this_start, this_end, cand;
218 for_each_free_mem_range_reverse(i, nid, flags, &this_start, &this_end,
220 this_start = clamp(this_start, start, end);
221 this_end = clamp(this_end, start, end);
226 cand = round_down(this_end - size, align);
227 if (cand >= this_start)
235 * memblock_find_in_range_node - find free area in given range and node
236 * @size: size of free area to find
237 * @align: alignment of free area to find
238 * @start: start of candidate range
239 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_ANYWHERE or
240 * %MEMBLOCK_ALLOC_ACCESSIBLE
241 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
242 * @flags: pick from blocks based on memory attributes
244 * Find @size free area aligned to @align in the specified range and node.
246 * When allocation direction is bottom-up, the @start should be greater
247 * than the end of the kernel image. Otherwise, it will be trimmed. The
248 * reason is that we want the bottom-up allocation just near the kernel
249 * image so it is highly likely that the allocated memory and the kernel
250 * will reside in the same node.
252 * If bottom-up allocation failed, will try to allocate memory top-down.
255 * Found address on success, 0 on failure.
257 phys_addr_t __init_memblock memblock_find_in_range_node(phys_addr_t size,
258 phys_addr_t align, phys_addr_t start,
259 phys_addr_t end, int nid,
260 enum memblock_flags flags)
262 phys_addr_t kernel_end, ret;
265 if (end == MEMBLOCK_ALLOC_ACCESSIBLE ||
266 end == MEMBLOCK_ALLOC_KASAN)
267 end = memblock.current_limit;
269 /* avoid allocating the first page */
270 start = max_t(phys_addr_t, start, PAGE_SIZE);
271 end = max(start, end);
272 kernel_end = __pa_symbol(_end);
275 * try bottom-up allocation only when bottom-up mode
276 * is set and @end is above the kernel image.
278 if (memblock_bottom_up() && end > kernel_end) {
279 phys_addr_t bottom_up_start;
281 /* make sure we will allocate above the kernel */
282 bottom_up_start = max(start, kernel_end);
284 /* ok, try bottom-up allocation first */
285 ret = __memblock_find_range_bottom_up(bottom_up_start, end,
286 size, align, nid, flags);
291 * we always limit bottom-up allocation above the kernel,
292 * but top-down allocation doesn't have the limit, so
293 * retrying top-down allocation may succeed when bottom-up
296 * bottom-up allocation is expected to be fail very rarely,
297 * so we use WARN_ONCE() here to see the stack trace if
300 WARN_ONCE(IS_ENABLED(CONFIG_MEMORY_HOTREMOVE),
301 "memblock: bottom-up allocation failed, memory hotremove may be affected\n");
304 return __memblock_find_range_top_down(start, end, size, align, nid,
309 * memblock_find_in_range - find free area in given range
310 * @start: start of candidate range
311 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_ANYWHERE or
312 * %MEMBLOCK_ALLOC_ACCESSIBLE
313 * @size: size of free area to find
314 * @align: alignment of free area to find
316 * Find @size free area aligned to @align in the specified range.
319 * Found address on success, 0 on failure.
321 phys_addr_t __init_memblock memblock_find_in_range(phys_addr_t start,
322 phys_addr_t end, phys_addr_t size,
326 enum memblock_flags flags = choose_memblock_flags();
329 ret = memblock_find_in_range_node(size, align, start, end,
330 NUMA_NO_NODE, flags);
332 if (!ret && (flags & MEMBLOCK_MIRROR)) {
333 pr_warn("Could not allocate %pap bytes of mirrored memory\n",
335 flags &= ~MEMBLOCK_MIRROR;
342 static void __init_memblock memblock_remove_region(struct memblock_type *type, unsigned long r)
344 type->total_size -= type->regions[r].size;
345 memmove(&type->regions[r], &type->regions[r + 1],
346 (type->cnt - (r + 1)) * sizeof(type->regions[r]));
349 /* Special case for empty arrays */
350 if (type->cnt == 0) {
351 WARN_ON(type->total_size != 0);
353 type->regions[0].base = 0;
354 type->regions[0].size = 0;
355 type->regions[0].flags = 0;
356 memblock_set_region_node(&type->regions[0], MAX_NUMNODES);
360 #ifdef CONFIG_ARCH_DISCARD_MEMBLOCK
362 * memblock_discard - discard memory and reserved arrays if they were allocated
364 void __init memblock_discard(void)
366 phys_addr_t addr, size;
368 if (memblock.reserved.regions != memblock_reserved_init_regions) {
369 addr = __pa(memblock.reserved.regions);
370 size = PAGE_ALIGN(sizeof(struct memblock_region) *
371 memblock.reserved.max);
372 __memblock_free_late(addr, size);
375 if (memblock.memory.regions != memblock_memory_init_regions) {
376 addr = __pa(memblock.memory.regions);
377 size = PAGE_ALIGN(sizeof(struct memblock_region) *
378 memblock.memory.max);
379 __memblock_free_late(addr, size);
385 * memblock_double_array - double the size of the memblock regions array
386 * @type: memblock type of the regions array being doubled
387 * @new_area_start: starting address of memory range to avoid overlap with
388 * @new_area_size: size of memory range to avoid overlap with
390 * Double the size of the @type regions array. If memblock is being used to
391 * allocate memory for a new reserved regions array and there is a previously
392 * allocated memory range [@new_area_start, @new_area_start + @new_area_size]
393 * waiting to be reserved, ensure the memory used by the new array does
397 * 0 on success, -1 on failure.
399 static int __init_memblock memblock_double_array(struct memblock_type *type,
400 phys_addr_t new_area_start,
401 phys_addr_t new_area_size)
403 struct memblock_region *new_array, *old_array;
404 phys_addr_t old_alloc_size, new_alloc_size;
405 phys_addr_t old_size, new_size, addr, new_end;
406 int use_slab = slab_is_available();
409 /* We don't allow resizing until we know about the reserved regions
410 * of memory that aren't suitable for allocation
412 if (!memblock_can_resize)
415 /* Calculate new doubled size */
416 old_size = type->max * sizeof(struct memblock_region);
417 new_size = old_size << 1;
419 * We need to allocated new one align to PAGE_SIZE,
420 * so we can free them completely later.
422 old_alloc_size = PAGE_ALIGN(old_size);
423 new_alloc_size = PAGE_ALIGN(new_size);
425 /* Retrieve the slab flag */
426 if (type == &memblock.memory)
427 in_slab = &memblock_memory_in_slab;
429 in_slab = &memblock_reserved_in_slab;
431 /* Try to find some space for it.
433 * WARNING: We assume that either slab_is_available() and we use it or
434 * we use MEMBLOCK for allocations. That means that this is unsafe to
435 * use when bootmem is currently active (unless bootmem itself is
436 * implemented on top of MEMBLOCK which isn't the case yet)
438 * This should however not be an issue for now, as we currently only
439 * call into MEMBLOCK while it's still active, or much later when slab
440 * is active for memory hotplug operations
443 new_array = kmalloc(new_size, GFP_KERNEL);
444 addr = new_array ? __pa(new_array) : 0;
446 /* only exclude range when trying to double reserved.regions */
447 if (type != &memblock.reserved)
448 new_area_start = new_area_size = 0;
450 addr = memblock_find_in_range(new_area_start + new_area_size,
451 memblock.current_limit,
452 new_alloc_size, PAGE_SIZE);
453 if (!addr && new_area_size)
454 addr = memblock_find_in_range(0,
455 min(new_area_start, memblock.current_limit),
456 new_alloc_size, PAGE_SIZE);
458 new_array = addr ? __va(addr) : NULL;
461 pr_err("memblock: Failed to double %s array from %ld to %ld entries !\n",
462 type->name, type->max, type->max * 2);
466 new_end = addr + new_size - 1;
467 memblock_dbg("memblock: %s is doubled to %ld at [%pa-%pa]",
468 type->name, type->max * 2, &addr, &new_end);
471 * Found space, we now need to move the array over before we add the
472 * reserved region since it may be our reserved array itself that is
475 memcpy(new_array, type->regions, old_size);
476 memset(new_array + type->max, 0, old_size);
477 old_array = type->regions;
478 type->regions = new_array;
481 /* Free old array. We needn't free it if the array is the static one */
484 else if (old_array != memblock_memory_init_regions &&
485 old_array != memblock_reserved_init_regions)
486 memblock_free(__pa(old_array), old_alloc_size);
489 * Reserve the new array if that comes from the memblock. Otherwise, we
493 BUG_ON(memblock_reserve(addr, new_alloc_size));
495 /* Update slab flag */
502 * memblock_merge_regions - merge neighboring compatible regions
503 * @type: memblock type to scan
505 * Scan @type and merge neighboring compatible regions.
507 static void __init_memblock memblock_merge_regions(struct memblock_type *type)
511 /* cnt never goes below 1 */
512 while (i < type->cnt - 1) {
513 struct memblock_region *this = &type->regions[i];
514 struct memblock_region *next = &type->regions[i + 1];
516 if (this->base + this->size != next->base ||
517 memblock_get_region_node(this) !=
518 memblock_get_region_node(next) ||
519 this->flags != next->flags) {
520 BUG_ON(this->base + this->size > next->base);
525 this->size += next->size;
526 /* move forward from next + 1, index of which is i + 2 */
527 memmove(next, next + 1, (type->cnt - (i + 2)) * sizeof(*next));
533 * memblock_insert_region - insert new memblock region
534 * @type: memblock type to insert into
535 * @idx: index for the insertion point
536 * @base: base address of the new region
537 * @size: size of the new region
538 * @nid: node id of the new region
539 * @flags: flags of the new region
541 * Insert new memblock region [@base, @base + @size) into @type at @idx.
542 * @type must already have extra room to accommodate the new region.
544 static void __init_memblock memblock_insert_region(struct memblock_type *type,
545 int idx, phys_addr_t base,
548 enum memblock_flags flags)
550 struct memblock_region *rgn = &type->regions[idx];
552 BUG_ON(type->cnt >= type->max);
553 memmove(rgn + 1, rgn, (type->cnt - idx) * sizeof(*rgn));
557 memblock_set_region_node(rgn, nid);
559 type->total_size += size;
563 * memblock_add_range - add new memblock region
564 * @type: memblock type to add new region into
565 * @base: base address of the new region
566 * @size: size of the new region
567 * @nid: nid of the new region
568 * @flags: flags of the new region
570 * Add new memblock region [@base, @base + @size) into @type. The new region
571 * is allowed to overlap with existing ones - overlaps don't affect already
572 * existing regions. @type is guaranteed to be minimal (all neighbouring
573 * compatible regions are merged) after the addition.
576 * 0 on success, -errno on failure.
578 int __init_memblock memblock_add_range(struct memblock_type *type,
579 phys_addr_t base, phys_addr_t size,
580 int nid, enum memblock_flags flags)
583 phys_addr_t obase = base;
584 phys_addr_t end = base + memblock_cap_size(base, &size);
586 struct memblock_region *rgn;
591 /* special case for empty array */
592 if (type->regions[0].size == 0) {
593 WARN_ON(type->cnt != 1 || type->total_size);
594 type->regions[0].base = base;
595 type->regions[0].size = size;
596 type->regions[0].flags = flags;
597 memblock_set_region_node(&type->regions[0], nid);
598 type->total_size = size;
603 * The following is executed twice. Once with %false @insert and
604 * then with %true. The first counts the number of regions needed
605 * to accommodate the new area. The second actually inserts them.
610 for_each_memblock_type(idx, type, rgn) {
611 phys_addr_t rbase = rgn->base;
612 phys_addr_t rend = rbase + rgn->size;
619 * @rgn overlaps. If it separates the lower part of new
620 * area, insert that portion.
623 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
624 WARN_ON(nid != memblock_get_region_node(rgn));
626 WARN_ON(flags != rgn->flags);
629 memblock_insert_region(type, idx++, base,
633 /* area below @rend is dealt with, forget about it */
634 base = min(rend, end);
637 /* insert the remaining portion */
641 memblock_insert_region(type, idx, base, end - base,
649 * If this was the first round, resize array and repeat for actual
650 * insertions; otherwise, merge and return.
653 while (type->cnt + nr_new > type->max)
654 if (memblock_double_array(type, obase, size) < 0)
659 memblock_merge_regions(type);
665 * memblock_add_node - add new memblock region within a NUMA node
666 * @base: base address of the new region
667 * @size: size of the new region
668 * @nid: nid of the new region
670 * Add new memblock region [@base, @base + @size) to the "memory"
671 * type. See memblock_add_range() description for mode details
674 * 0 on success, -errno on failure.
676 int __init_memblock memblock_add_node(phys_addr_t base, phys_addr_t size,
679 return memblock_add_range(&memblock.memory, base, size, nid, 0);
683 * memblock_add - add new memblock region
684 * @base: base address of the new region
685 * @size: size of the new region
687 * Add new memblock region [@base, @base + @size) to the "memory"
688 * type. See memblock_add_range() description for mode details
691 * 0 on success, -errno on failure.
693 int __init_memblock memblock_add(phys_addr_t base, phys_addr_t size)
695 phys_addr_t end = base + size - 1;
697 memblock_dbg("memblock_add: [%pa-%pa] %pF\n",
698 &base, &end, (void *)_RET_IP_);
700 return memblock_add_range(&memblock.memory, base, size, MAX_NUMNODES, 0);
704 * memblock_isolate_range - isolate given range into disjoint memblocks
705 * @type: memblock type to isolate range for
706 * @base: base of range to isolate
707 * @size: size of range to isolate
708 * @start_rgn: out parameter for the start of isolated region
709 * @end_rgn: out parameter for the end of isolated region
711 * Walk @type and ensure that regions don't cross the boundaries defined by
712 * [@base, @base + @size). Crossing regions are split at the boundaries,
713 * which may create at most two more regions. The index of the first
714 * region inside the range is returned in *@start_rgn and end in *@end_rgn.
717 * 0 on success, -errno on failure.
719 static int __init_memblock memblock_isolate_range(struct memblock_type *type,
720 phys_addr_t base, phys_addr_t size,
721 int *start_rgn, int *end_rgn)
723 phys_addr_t end = base + memblock_cap_size(base, &size);
725 struct memblock_region *rgn;
727 *start_rgn = *end_rgn = 0;
732 /* we'll create at most two more regions */
733 while (type->cnt + 2 > type->max)
734 if (memblock_double_array(type, base, size) < 0)
737 for_each_memblock_type(idx, type, rgn) {
738 phys_addr_t rbase = rgn->base;
739 phys_addr_t rend = rbase + rgn->size;
748 * @rgn intersects from below. Split and continue
749 * to process the next region - the new top half.
752 rgn->size -= base - rbase;
753 type->total_size -= base - rbase;
754 memblock_insert_region(type, idx, rbase, base - rbase,
755 memblock_get_region_node(rgn),
757 } else if (rend > end) {
759 * @rgn intersects from above. Split and redo the
760 * current region - the new bottom half.
763 rgn->size -= end - rbase;
764 type->total_size -= end - rbase;
765 memblock_insert_region(type, idx--, rbase, end - rbase,
766 memblock_get_region_node(rgn),
769 /* @rgn is fully contained, record it */
779 static int __init_memblock memblock_remove_range(struct memblock_type *type,
780 phys_addr_t base, phys_addr_t size)
782 int start_rgn, end_rgn;
785 ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
789 for (i = end_rgn - 1; i >= start_rgn; i--)
790 memblock_remove_region(type, i);
794 int __init_memblock memblock_remove(phys_addr_t base, phys_addr_t size)
796 phys_addr_t end = base + size - 1;
798 memblock_dbg("memblock_remove: [%pa-%pa] %pS\n",
799 &base, &end, (void *)_RET_IP_);
801 return memblock_remove_range(&memblock.memory, base, size);
805 * memblock_free - free boot memory block
806 * @base: phys starting address of the boot memory block
807 * @size: size of the boot memory block in bytes
809 * Free boot memory block previously allocated by memblock_alloc_xx() API.
810 * The freeing memory will not be released to the buddy allocator.
812 int __init_memblock memblock_free(phys_addr_t base, phys_addr_t size)
814 phys_addr_t end = base + size - 1;
816 memblock_dbg(" memblock_free: [%pa-%pa] %pF\n",
817 &base, &end, (void *)_RET_IP_);
819 kmemleak_free_part_phys(base, size);
820 return memblock_remove_range(&memblock.reserved, base, size);
823 int __init_memblock memblock_reserve(phys_addr_t base, phys_addr_t size)
825 phys_addr_t end = base + size - 1;
827 memblock_dbg("memblock_reserve: [%pa-%pa] %pF\n",
828 &base, &end, (void *)_RET_IP_);
830 return memblock_add_range(&memblock.reserved, base, size, MAX_NUMNODES, 0);
834 * memblock_setclr_flag - set or clear flag for a memory region
835 * @base: base address of the region
836 * @size: size of the region
837 * @set: set or clear the flag
838 * @flag: the flag to udpate
840 * This function isolates region [@base, @base + @size), and sets/clears flag
842 * Return: 0 on success, -errno on failure.
844 static int __init_memblock memblock_setclr_flag(phys_addr_t base,
845 phys_addr_t size, int set, int flag)
847 struct memblock_type *type = &memblock.memory;
848 int i, ret, start_rgn, end_rgn;
850 ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
854 for (i = start_rgn; i < end_rgn; i++)
856 memblock_set_region_flags(&type->regions[i], flag);
858 memblock_clear_region_flags(&type->regions[i], flag);
860 memblock_merge_regions(type);
865 * memblock_mark_hotplug - Mark hotpluggable memory with flag MEMBLOCK_HOTPLUG.
866 * @base: the base phys addr of the region
867 * @size: the size of the region
869 * Return: 0 on success, -errno on failure.
871 int __init_memblock memblock_mark_hotplug(phys_addr_t base, phys_addr_t size)
873 return memblock_setclr_flag(base, size, 1, MEMBLOCK_HOTPLUG);
877 * memblock_clear_hotplug - Clear flag MEMBLOCK_HOTPLUG for a specified region.
878 * @base: the base phys addr of the region
879 * @size: the size of the region
881 * Return: 0 on success, -errno on failure.
883 int __init_memblock memblock_clear_hotplug(phys_addr_t base, phys_addr_t size)
885 return memblock_setclr_flag(base, size, 0, MEMBLOCK_HOTPLUG);
889 * memblock_mark_mirror - Mark mirrored memory with flag MEMBLOCK_MIRROR.
890 * @base: the base phys addr of the region
891 * @size: the size of the region
893 * Return: 0 on success, -errno on failure.
895 int __init_memblock memblock_mark_mirror(phys_addr_t base, phys_addr_t size)
897 system_has_some_mirror = true;
899 return memblock_setclr_flag(base, size, 1, MEMBLOCK_MIRROR);
903 * memblock_mark_nomap - Mark a memory region with flag MEMBLOCK_NOMAP.
904 * @base: the base phys addr of the region
905 * @size: the size of the region
907 * Return: 0 on success, -errno on failure.
909 int __init_memblock memblock_mark_nomap(phys_addr_t base, phys_addr_t size)
911 return memblock_setclr_flag(base, size, 1, MEMBLOCK_NOMAP);
915 * memblock_clear_nomap - Clear flag MEMBLOCK_NOMAP for a specified region.
916 * @base: the base phys addr of the region
917 * @size: the size of the region
919 * Return: 0 on success, -errno on failure.
921 int __init_memblock memblock_clear_nomap(phys_addr_t base, phys_addr_t size)
923 return memblock_setclr_flag(base, size, 0, MEMBLOCK_NOMAP);
927 * __next_reserved_mem_region - next function for for_each_reserved_region()
928 * @idx: pointer to u64 loop variable
929 * @out_start: ptr to phys_addr_t for start address of the region, can be %NULL
930 * @out_end: ptr to phys_addr_t for end address of the region, can be %NULL
932 * Iterate over all reserved memory regions.
934 void __init_memblock __next_reserved_mem_region(u64 *idx,
935 phys_addr_t *out_start,
936 phys_addr_t *out_end)
938 struct memblock_type *type = &memblock.reserved;
940 if (*idx < type->cnt) {
941 struct memblock_region *r = &type->regions[*idx];
942 phys_addr_t base = r->base;
943 phys_addr_t size = r->size;
948 *out_end = base + size - 1;
954 /* signal end of iteration */
959 * __next__mem_range - next function for for_each_free_mem_range() etc.
960 * @idx: pointer to u64 loop variable
961 * @nid: node selector, %NUMA_NO_NODE for all nodes
962 * @flags: pick from blocks based on memory attributes
963 * @type_a: pointer to memblock_type from where the range is taken
964 * @type_b: pointer to memblock_type which excludes memory from being taken
965 * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL
966 * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL
967 * @out_nid: ptr to int for nid of the range, can be %NULL
969 * Find the first area from *@idx which matches @nid, fill the out
970 * parameters, and update *@idx for the next iteration. The lower 32bit of
971 * *@idx contains index into type_a and the upper 32bit indexes the
972 * areas before each region in type_b. For example, if type_b regions
973 * look like the following,
975 * 0:[0-16), 1:[32-48), 2:[128-130)
977 * The upper 32bit indexes the following regions.
979 * 0:[0-0), 1:[16-32), 2:[48-128), 3:[130-MAX)
981 * As both region arrays are sorted, the function advances the two indices
982 * in lockstep and returns each intersection.
984 void __init_memblock __next_mem_range(u64 *idx, int nid,
985 enum memblock_flags flags,
986 struct memblock_type *type_a,
987 struct memblock_type *type_b,
988 phys_addr_t *out_start,
989 phys_addr_t *out_end, int *out_nid)
991 int idx_a = *idx & 0xffffffff;
992 int idx_b = *idx >> 32;
994 if (WARN_ONCE(nid == MAX_NUMNODES,
995 "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
998 for (; idx_a < type_a->cnt; idx_a++) {
999 struct memblock_region *m = &type_a->regions[idx_a];
1001 phys_addr_t m_start = m->base;
1002 phys_addr_t m_end = m->base + m->size;
1003 int m_nid = memblock_get_region_node(m);
1005 /* only memory regions are associated with nodes, check it */
1006 if (nid != NUMA_NO_NODE && nid != m_nid)
1009 /* skip hotpluggable memory regions if needed */
1010 if (movable_node_is_enabled() && memblock_is_hotpluggable(m))
1013 /* if we want mirror memory skip non-mirror memory regions */
1014 if ((flags & MEMBLOCK_MIRROR) && !memblock_is_mirror(m))
1017 /* skip nomap memory unless we were asked for it explicitly */
1018 if (!(flags & MEMBLOCK_NOMAP) && memblock_is_nomap(m))
1023 *out_start = m_start;
1029 *idx = (u32)idx_a | (u64)idx_b << 32;
1033 /* scan areas before each reservation */
1034 for (; idx_b < type_b->cnt + 1; idx_b++) {
1035 struct memblock_region *r;
1036 phys_addr_t r_start;
1039 r = &type_b->regions[idx_b];
1040 r_start = idx_b ? r[-1].base + r[-1].size : 0;
1041 r_end = idx_b < type_b->cnt ?
1042 r->base : PHYS_ADDR_MAX;
1045 * if idx_b advanced past idx_a,
1046 * break out to advance idx_a
1048 if (r_start >= m_end)
1050 /* if the two regions intersect, we're done */
1051 if (m_start < r_end) {
1054 max(m_start, r_start);
1056 *out_end = min(m_end, r_end);
1060 * The region which ends first is
1061 * advanced for the next iteration.
1067 *idx = (u32)idx_a | (u64)idx_b << 32;
1073 /* signal end of iteration */
1078 * __next_mem_range_rev - generic next function for for_each_*_range_rev()
1080 * @idx: pointer to u64 loop variable
1081 * @nid: node selector, %NUMA_NO_NODE for all nodes
1082 * @flags: pick from blocks based on memory attributes
1083 * @type_a: pointer to memblock_type from where the range is taken
1084 * @type_b: pointer to memblock_type which excludes memory from being taken
1085 * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL
1086 * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL
1087 * @out_nid: ptr to int for nid of the range, can be %NULL
1089 * Finds the next range from type_a which is not marked as unsuitable
1092 * Reverse of __next_mem_range().
1094 void __init_memblock __next_mem_range_rev(u64 *idx, int nid,
1095 enum memblock_flags flags,
1096 struct memblock_type *type_a,
1097 struct memblock_type *type_b,
1098 phys_addr_t *out_start,
1099 phys_addr_t *out_end, int *out_nid)
1101 int idx_a = *idx & 0xffffffff;
1102 int idx_b = *idx >> 32;
1104 if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
1107 if (*idx == (u64)ULLONG_MAX) {
1108 idx_a = type_a->cnt - 1;
1110 idx_b = type_b->cnt;
1115 for (; idx_a >= 0; idx_a--) {
1116 struct memblock_region *m = &type_a->regions[idx_a];
1118 phys_addr_t m_start = m->base;
1119 phys_addr_t m_end = m->base + m->size;
1120 int m_nid = memblock_get_region_node(m);
1122 /* only memory regions are associated with nodes, check it */
1123 if (nid != NUMA_NO_NODE && nid != m_nid)
1126 /* skip hotpluggable memory regions if needed */
1127 if (movable_node_is_enabled() && memblock_is_hotpluggable(m))
1130 /* if we want mirror memory skip non-mirror memory regions */
1131 if ((flags & MEMBLOCK_MIRROR) && !memblock_is_mirror(m))
1134 /* skip nomap memory unless we were asked for it explicitly */
1135 if (!(flags & MEMBLOCK_NOMAP) && memblock_is_nomap(m))
1140 *out_start = m_start;
1146 *idx = (u32)idx_a | (u64)idx_b << 32;
1150 /* scan areas before each reservation */
1151 for (; idx_b >= 0; idx_b--) {
1152 struct memblock_region *r;
1153 phys_addr_t r_start;
1156 r = &type_b->regions[idx_b];
1157 r_start = idx_b ? r[-1].base + r[-1].size : 0;
1158 r_end = idx_b < type_b->cnt ?
1159 r->base : PHYS_ADDR_MAX;
1161 * if idx_b advanced past idx_a,
1162 * break out to advance idx_a
1165 if (r_end <= m_start)
1167 /* if the two regions intersect, we're done */
1168 if (m_end > r_start) {
1170 *out_start = max(m_start, r_start);
1172 *out_end = min(m_end, r_end);
1175 if (m_start >= r_start)
1179 *idx = (u32)idx_a | (u64)idx_b << 32;
1184 /* signal end of iteration */
1188 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
1190 * Common iterator interface used to define for_each_mem_pfn_range().
1192 void __init_memblock __next_mem_pfn_range(int *idx, int nid,
1193 unsigned long *out_start_pfn,
1194 unsigned long *out_end_pfn, int *out_nid)
1196 struct memblock_type *type = &memblock.memory;
1197 struct memblock_region *r;
1199 while (++*idx < type->cnt) {
1200 r = &type->regions[*idx];
1202 if (PFN_UP(r->base) >= PFN_DOWN(r->base + r->size))
1204 if (nid == MAX_NUMNODES || nid == r->nid)
1207 if (*idx >= type->cnt) {
1213 *out_start_pfn = PFN_UP(r->base);
1215 *out_end_pfn = PFN_DOWN(r->base + r->size);
1221 * memblock_set_node - set node ID on memblock regions
1222 * @base: base of area to set node ID for
1223 * @size: size of area to set node ID for
1224 * @type: memblock type to set node ID for
1225 * @nid: node ID to set
1227 * Set the nid of memblock @type regions in [@base, @base + @size) to @nid.
1228 * Regions which cross the area boundaries are split as necessary.
1231 * 0 on success, -errno on failure.
1233 int __init_memblock memblock_set_node(phys_addr_t base, phys_addr_t size,
1234 struct memblock_type *type, int nid)
1236 int start_rgn, end_rgn;
1239 ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
1243 for (i = start_rgn; i < end_rgn; i++)
1244 memblock_set_region_node(&type->regions[i], nid);
1246 memblock_merge_regions(type);
1249 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
1250 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1252 * __next_mem_pfn_range_in_zone - iterator for for_each_*_range_in_zone()
1254 * @idx: pointer to u64 loop variable
1255 * @zone: zone in which all of the memory blocks reside
1256 * @out_spfn: ptr to ulong for start pfn of the range, can be %NULL
1257 * @out_epfn: ptr to ulong for end pfn of the range, can be %NULL
1259 * This function is meant to be a zone/pfn specific wrapper for the
1260 * for_each_mem_range type iterators. Specifically they are used in the
1261 * deferred memory init routines and as such we were duplicating much of
1262 * this logic throughout the code. So instead of having it in multiple
1263 * locations it seemed like it would make more sense to centralize this to
1264 * one new iterator that does everything they need.
1266 void __init_memblock
1267 __next_mem_pfn_range_in_zone(u64 *idx, struct zone *zone,
1268 unsigned long *out_spfn, unsigned long *out_epfn)
1270 int zone_nid = zone_to_nid(zone);
1271 phys_addr_t spa, epa;
1274 __next_mem_range(idx, zone_nid, MEMBLOCK_NONE,
1275 &memblock.memory, &memblock.reserved,
1278 while (*idx != U64_MAX) {
1279 unsigned long epfn = PFN_DOWN(epa);
1280 unsigned long spfn = PFN_UP(spa);
1283 * Verify the end is at least past the start of the zone and
1284 * that we have at least one PFN to initialize.
1286 if (zone->zone_start_pfn < epfn && spfn < epfn) {
1287 /* if we went too far just stop searching */
1288 if (zone_end_pfn(zone) <= spfn) {
1294 *out_spfn = max(zone->zone_start_pfn, spfn);
1296 *out_epfn = min(zone_end_pfn(zone), epfn);
1301 __next_mem_range(idx, zone_nid, MEMBLOCK_NONE,
1302 &memblock.memory, &memblock.reserved,
1306 /* signal end of iteration */
1308 *out_spfn = ULONG_MAX;
1313 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
1315 static phys_addr_t __init memblock_alloc_range_nid(phys_addr_t size,
1316 phys_addr_t align, phys_addr_t start,
1317 phys_addr_t end, int nid,
1318 enum memblock_flags flags)
1323 /* Can't use WARNs this early in boot on powerpc */
1325 align = SMP_CACHE_BYTES;
1328 found = memblock_find_in_range_node(size, align, start, end, nid,
1330 if (found && !memblock_reserve(found, size)) {
1332 * The min_count is set to 0 so that memblock allocations are
1333 * never reported as leaks.
1335 kmemleak_alloc_phys(found, size, 0, 0);
1341 phys_addr_t __init memblock_phys_alloc_range(phys_addr_t size,
1346 return memblock_alloc_range_nid(size, align, start, end, NUMA_NO_NODE,
1350 phys_addr_t __init memblock_phys_alloc_nid(phys_addr_t size, phys_addr_t align, int nid)
1352 enum memblock_flags flags = choose_memblock_flags();
1356 ret = memblock_alloc_range_nid(size, align, 0,
1357 MEMBLOCK_ALLOC_ACCESSIBLE, nid, flags);
1359 if (!ret && (flags & MEMBLOCK_MIRROR)) {
1360 flags &= ~MEMBLOCK_MIRROR;
1366 phys_addr_t __init memblock_alloc_base(phys_addr_t size, phys_addr_t align, phys_addr_t max_addr)
1370 alloc = memblock_alloc_range_nid(size, align, 0, max_addr, NUMA_NO_NODE,
1374 panic("ERROR: Failed to allocate %pa bytes below %pa.\n",
1380 phys_addr_t __init memblock_phys_alloc_try_nid(phys_addr_t size, phys_addr_t align, int nid)
1382 phys_addr_t res = memblock_phys_alloc_nid(size, align, nid);
1386 return memblock_alloc_range_nid(size, align, 0,
1387 MEMBLOCK_ALLOC_ACCESSIBLE,
1388 NUMA_NO_NODE, MEMBLOCK_NONE);
1392 * memblock_alloc_internal - allocate boot memory block
1393 * @size: size of memory block to be allocated in bytes
1394 * @align: alignment of the region and block's size
1395 * @min_addr: the lower bound of the memory region to allocate (phys address)
1396 * @max_addr: the upper bound of the memory region to allocate (phys address)
1397 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1399 * The @min_addr limit is dropped if it can not be satisfied and the allocation
1400 * will fall back to memory below @min_addr. Also, allocation may fall back
1401 * to any node in the system if the specified node can not
1402 * hold the requested memory.
1404 * The allocation is performed from memory region limited by
1405 * memblock.current_limit if @max_addr == %MEMBLOCK_ALLOC_ACCESSIBLE.
1407 * The phys address of allocated boot memory block is converted to virtual and
1408 * allocated memory is reset to 0.
1410 * In addition, function sets the min_count to 0 using kmemleak_alloc for
1411 * allocated boot memory block, so that it is never reported as leaks.
1414 * Virtual address of allocated memory block on success, NULL on failure.
1416 static void * __init memblock_alloc_internal(
1417 phys_addr_t size, phys_addr_t align,
1418 phys_addr_t min_addr, phys_addr_t max_addr,
1423 enum memblock_flags flags = choose_memblock_flags();
1425 if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
1429 * Detect any accidental use of these APIs after slab is ready, as at
1430 * this moment memblock may be deinitialized already and its
1431 * internal data may be destroyed (after execution of memblock_free_all)
1433 if (WARN_ON_ONCE(slab_is_available()))
1434 return kzalloc_node(size, GFP_NOWAIT, nid);
1438 align = SMP_CACHE_BYTES;
1441 if (max_addr > memblock.current_limit)
1442 max_addr = memblock.current_limit;
1444 alloc = memblock_find_in_range_node(size, align, min_addr, max_addr,
1446 if (alloc && !memblock_reserve(alloc, size))
1449 if (nid != NUMA_NO_NODE) {
1450 alloc = memblock_find_in_range_node(size, align, min_addr,
1451 max_addr, NUMA_NO_NODE,
1453 if (alloc && !memblock_reserve(alloc, size))
1462 if (flags & MEMBLOCK_MIRROR) {
1463 flags &= ~MEMBLOCK_MIRROR;
1464 pr_warn("Could not allocate %pap bytes of mirrored memory\n",
1471 ptr = phys_to_virt(alloc);
1473 /* Skip kmemleak for kasan_init() due to high volume. */
1474 if (max_addr != MEMBLOCK_ALLOC_KASAN)
1476 * The min_count is set to 0 so that bootmem allocated
1477 * blocks are never reported as leaks. This is because many
1478 * of these blocks are only referred via the physical
1479 * address which is not looked up by kmemleak.
1481 kmemleak_alloc(ptr, size, 0, 0);
1487 * memblock_alloc_try_nid_raw - allocate boot memory block without zeroing
1488 * memory and without panicking
1489 * @size: size of memory block to be allocated in bytes
1490 * @align: alignment of the region and block's size
1491 * @min_addr: the lower bound of the memory region from where the allocation
1492 * is preferred (phys address)
1493 * @max_addr: the upper bound of the memory region from where the allocation
1494 * is preferred (phys address), or %MEMBLOCK_ALLOC_ACCESSIBLE to
1495 * allocate only from memory limited by memblock.current_limit value
1496 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1498 * Public function, provides additional debug information (including caller
1499 * info), if enabled. Does not zero allocated memory, does not panic if request
1500 * cannot be satisfied.
1503 * Virtual address of allocated memory block on success, NULL on failure.
1505 void * __init memblock_alloc_try_nid_raw(
1506 phys_addr_t size, phys_addr_t align,
1507 phys_addr_t min_addr, phys_addr_t max_addr,
1512 memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=%pa max_addr=%pa %pF\n",
1513 __func__, (u64)size, (u64)align, nid, &min_addr,
1514 &max_addr, (void *)_RET_IP_);
1516 ptr = memblock_alloc_internal(size, align,
1517 min_addr, max_addr, nid);
1518 if (ptr && size > 0)
1519 page_init_poison(ptr, size);
1525 * memblock_alloc_try_nid_nopanic - allocate boot memory block
1526 * @size: size of memory block to be allocated in bytes
1527 * @align: alignment of the region and block's size
1528 * @min_addr: the lower bound of the memory region from where the allocation
1529 * is preferred (phys address)
1530 * @max_addr: the upper bound of the memory region from where the allocation
1531 * is preferred (phys address), or %MEMBLOCK_ALLOC_ACCESSIBLE to
1532 * allocate only from memory limited by memblock.current_limit value
1533 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1535 * Public function, provides additional debug information (including caller
1536 * info), if enabled. This function zeroes the allocated memory.
1539 * Virtual address of allocated memory block on success, NULL on failure.
1541 void * __init memblock_alloc_try_nid_nopanic(
1542 phys_addr_t size, phys_addr_t align,
1543 phys_addr_t min_addr, phys_addr_t max_addr,
1548 memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=%pa max_addr=%pa %pF\n",
1549 __func__, (u64)size, (u64)align, nid, &min_addr,
1550 &max_addr, (void *)_RET_IP_);
1552 ptr = memblock_alloc_internal(size, align,
1553 min_addr, max_addr, nid);
1555 memset(ptr, 0, size);
1560 * memblock_alloc_try_nid - allocate boot memory block with panicking
1561 * @size: size of memory block to be allocated in bytes
1562 * @align: alignment of the region and block's size
1563 * @min_addr: the lower bound of the memory region from where the allocation
1564 * is preferred (phys address)
1565 * @max_addr: the upper bound of the memory region from where the allocation
1566 * is preferred (phys address), or %MEMBLOCK_ALLOC_ACCESSIBLE to
1567 * allocate only from memory limited by memblock.current_limit value
1568 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1570 * Public panicking version of memblock_alloc_try_nid_nopanic()
1571 * which provides debug information (including caller info), if enabled,
1572 * and panics if the request can not be satisfied.
1575 * Virtual address of allocated memory block on success, NULL on failure.
1577 void * __init memblock_alloc_try_nid(
1578 phys_addr_t size, phys_addr_t align,
1579 phys_addr_t min_addr, phys_addr_t max_addr,
1584 memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=%pa max_addr=%pa %pF\n",
1585 __func__, (u64)size, (u64)align, nid, &min_addr,
1586 &max_addr, (void *)_RET_IP_);
1587 ptr = memblock_alloc_internal(size, align,
1588 min_addr, max_addr, nid);
1590 memset(ptr, 0, size);
1594 panic("%s: Failed to allocate %llu bytes align=0x%llx nid=%d from=%pa max_addr=%pa\n",
1595 __func__, (u64)size, (u64)align, nid, &min_addr, &max_addr);
1600 * __memblock_free_late - free bootmem block pages directly to buddy allocator
1601 * @base: phys starting address of the boot memory block
1602 * @size: size of the boot memory block in bytes
1604 * This is only useful when the bootmem allocator has already been torn
1605 * down, but we are still initializing the system. Pages are released directly
1606 * to the buddy allocator, no bootmem metadata is updated because it is gone.
1608 void __init __memblock_free_late(phys_addr_t base, phys_addr_t size)
1610 phys_addr_t cursor, end;
1612 end = base + size - 1;
1613 memblock_dbg("%s: [%pa-%pa] %pF\n",
1614 __func__, &base, &end, (void *)_RET_IP_);
1615 kmemleak_free_part_phys(base, size);
1616 cursor = PFN_UP(base);
1617 end = PFN_DOWN(base + size);
1619 for (; cursor < end; cursor++) {
1620 memblock_free_pages(pfn_to_page(cursor), cursor, 0);
1621 totalram_pages_inc();
1626 * Remaining API functions
1629 phys_addr_t __init_memblock memblock_phys_mem_size(void)
1631 return memblock.memory.total_size;
1634 phys_addr_t __init_memblock memblock_reserved_size(void)
1636 return memblock.reserved.total_size;
1639 phys_addr_t __init memblock_mem_size(unsigned long limit_pfn)
1641 unsigned long pages = 0;
1642 struct memblock_region *r;
1643 unsigned long start_pfn, end_pfn;
1645 for_each_memblock(memory, r) {
1646 start_pfn = memblock_region_memory_base_pfn(r);
1647 end_pfn = memblock_region_memory_end_pfn(r);
1648 start_pfn = min_t(unsigned long, start_pfn, limit_pfn);
1649 end_pfn = min_t(unsigned long, end_pfn, limit_pfn);
1650 pages += end_pfn - start_pfn;
1653 return PFN_PHYS(pages);
1656 /* lowest address */
1657 phys_addr_t __init_memblock memblock_start_of_DRAM(void)
1659 return memblock.memory.regions[0].base;
1662 phys_addr_t __init_memblock memblock_end_of_DRAM(void)
1664 int idx = memblock.memory.cnt - 1;
1666 return (memblock.memory.regions[idx].base + memblock.memory.regions[idx].size);
1669 static phys_addr_t __init_memblock __find_max_addr(phys_addr_t limit)
1671 phys_addr_t max_addr = PHYS_ADDR_MAX;
1672 struct memblock_region *r;
1675 * translate the memory @limit size into the max address within one of
1676 * the memory memblock regions, if the @limit exceeds the total size
1677 * of those regions, max_addr will keep original value PHYS_ADDR_MAX
1679 for_each_memblock(memory, r) {
1680 if (limit <= r->size) {
1681 max_addr = r->base + limit;
1690 void __init memblock_enforce_memory_limit(phys_addr_t limit)
1692 phys_addr_t max_addr = PHYS_ADDR_MAX;
1697 max_addr = __find_max_addr(limit);
1699 /* @limit exceeds the total size of the memory, do nothing */
1700 if (max_addr == PHYS_ADDR_MAX)
1703 /* truncate both memory and reserved regions */
1704 memblock_remove_range(&memblock.memory, max_addr,
1706 memblock_remove_range(&memblock.reserved, max_addr,
1710 void __init memblock_cap_memory_range(phys_addr_t base, phys_addr_t size)
1712 int start_rgn, end_rgn;
1718 ret = memblock_isolate_range(&memblock.memory, base, size,
1719 &start_rgn, &end_rgn);
1723 /* remove all the MAP regions */
1724 for (i = memblock.memory.cnt - 1; i >= end_rgn; i--)
1725 if (!memblock_is_nomap(&memblock.memory.regions[i]))
1726 memblock_remove_region(&memblock.memory, i);
1728 for (i = start_rgn - 1; i >= 0; i--)
1729 if (!memblock_is_nomap(&memblock.memory.regions[i]))
1730 memblock_remove_region(&memblock.memory, i);
1732 /* truncate the reserved regions */
1733 memblock_remove_range(&memblock.reserved, 0, base);
1734 memblock_remove_range(&memblock.reserved,
1735 base + size, PHYS_ADDR_MAX);
1738 void __init memblock_mem_limit_remove_map(phys_addr_t limit)
1740 phys_addr_t max_addr;
1745 max_addr = __find_max_addr(limit);
1747 /* @limit exceeds the total size of the memory, do nothing */
1748 if (max_addr == PHYS_ADDR_MAX)
1751 memblock_cap_memory_range(0, max_addr);
1754 static int __init_memblock memblock_search(struct memblock_type *type, phys_addr_t addr)
1756 unsigned int left = 0, right = type->cnt;
1759 unsigned int mid = (right + left) / 2;
1761 if (addr < type->regions[mid].base)
1763 else if (addr >= (type->regions[mid].base +
1764 type->regions[mid].size))
1768 } while (left < right);
1772 bool __init_memblock memblock_is_reserved(phys_addr_t addr)
1774 return memblock_search(&memblock.reserved, addr) != -1;
1777 bool __init_memblock memblock_is_memory(phys_addr_t addr)
1779 return memblock_search(&memblock.memory, addr) != -1;
1782 bool __init_memblock memblock_is_map_memory(phys_addr_t addr)
1784 int i = memblock_search(&memblock.memory, addr);
1788 return !memblock_is_nomap(&memblock.memory.regions[i]);
1791 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
1792 int __init_memblock memblock_search_pfn_nid(unsigned long pfn,
1793 unsigned long *start_pfn, unsigned long *end_pfn)
1795 struct memblock_type *type = &memblock.memory;
1796 int mid = memblock_search(type, PFN_PHYS(pfn));
1801 *start_pfn = PFN_DOWN(type->regions[mid].base);
1802 *end_pfn = PFN_DOWN(type->regions[mid].base + type->regions[mid].size);
1804 return type->regions[mid].nid;
1809 * memblock_is_region_memory - check if a region is a subset of memory
1810 * @base: base of region to check
1811 * @size: size of region to check
1813 * Check if the region [@base, @base + @size) is a subset of a memory block.
1816 * 0 if false, non-zero if true
1818 bool __init_memblock memblock_is_region_memory(phys_addr_t base, phys_addr_t size)
1820 int idx = memblock_search(&memblock.memory, base);
1821 phys_addr_t end = base + memblock_cap_size(base, &size);
1825 return (memblock.memory.regions[idx].base +
1826 memblock.memory.regions[idx].size) >= end;
1830 * memblock_is_region_reserved - check if a region intersects reserved memory
1831 * @base: base of region to check
1832 * @size: size of region to check
1834 * Check if the region [@base, @base + @size) intersects a reserved
1838 * True if they intersect, false if not.
1840 bool __init_memblock memblock_is_region_reserved(phys_addr_t base, phys_addr_t size)
1842 memblock_cap_size(base, &size);
1843 return memblock_overlaps_region(&memblock.reserved, base, size);
1846 void __init_memblock memblock_trim_memory(phys_addr_t align)
1848 phys_addr_t start, end, orig_start, orig_end;
1849 struct memblock_region *r;
1851 for_each_memblock(memory, r) {
1852 orig_start = r->base;
1853 orig_end = r->base + r->size;
1854 start = round_up(orig_start, align);
1855 end = round_down(orig_end, align);
1857 if (start == orig_start && end == orig_end)
1862 r->size = end - start;
1864 memblock_remove_region(&memblock.memory,
1865 r - memblock.memory.regions);
1871 void __init_memblock memblock_set_current_limit(phys_addr_t limit)
1873 memblock.current_limit = limit;
1876 phys_addr_t __init_memblock memblock_get_current_limit(void)
1878 return memblock.current_limit;
1881 static void __init_memblock memblock_dump(struct memblock_type *type)
1883 phys_addr_t base, end, size;
1884 enum memblock_flags flags;
1886 struct memblock_region *rgn;
1888 pr_info(" %s.cnt = 0x%lx\n", type->name, type->cnt);
1890 for_each_memblock_type(idx, type, rgn) {
1891 char nid_buf[32] = "";
1895 end = base + size - 1;
1897 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
1898 if (memblock_get_region_node(rgn) != MAX_NUMNODES)
1899 snprintf(nid_buf, sizeof(nid_buf), " on node %d",
1900 memblock_get_region_node(rgn));
1902 pr_info(" %s[%#x]\t[%pa-%pa], %pa bytes%s flags: %#x\n",
1903 type->name, idx, &base, &end, &size, nid_buf, flags);
1907 void __init_memblock __memblock_dump_all(void)
1909 pr_info("MEMBLOCK configuration:\n");
1910 pr_info(" memory size = %pa reserved size = %pa\n",
1911 &memblock.memory.total_size,
1912 &memblock.reserved.total_size);
1914 memblock_dump(&memblock.memory);
1915 memblock_dump(&memblock.reserved);
1916 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
1917 memblock_dump(&memblock.physmem);
1921 void __init memblock_allow_resize(void)
1923 memblock_can_resize = 1;
1926 static int __init early_memblock(char *p)
1928 if (p && strstr(p, "debug"))
1932 early_param("memblock", early_memblock);
1934 static void __init __free_pages_memory(unsigned long start, unsigned long end)
1938 while (start < end) {
1939 order = min(MAX_ORDER - 1UL, __ffs(start));
1941 while (start + (1UL << order) > end)
1944 memblock_free_pages(pfn_to_page(start), start, order);
1946 start += (1UL << order);
1950 static unsigned long __init __free_memory_core(phys_addr_t start,
1953 unsigned long start_pfn = PFN_UP(start);
1954 unsigned long end_pfn = min_t(unsigned long,
1955 PFN_DOWN(end), max_low_pfn);
1957 if (start_pfn >= end_pfn)
1960 __free_pages_memory(start_pfn, end_pfn);
1962 return end_pfn - start_pfn;
1965 static unsigned long __init free_low_memory_core_early(void)
1967 unsigned long count = 0;
1968 phys_addr_t start, end;
1971 memblock_clear_hotplug(0, -1);
1973 for_each_reserved_mem_region(i, &start, &end)
1974 reserve_bootmem_region(start, end);
1977 * We need to use NUMA_NO_NODE instead of NODE_DATA(0)->node_id
1978 * because in some case like Node0 doesn't have RAM installed
1979 * low ram will be on Node1
1981 for_each_free_mem_range(i, NUMA_NO_NODE, MEMBLOCK_NONE, &start, &end,
1983 count += __free_memory_core(start, end);
1988 static int reset_managed_pages_done __initdata;
1990 void reset_node_managed_pages(pg_data_t *pgdat)
1994 for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++)
1995 atomic_long_set(&z->managed_pages, 0);
1998 void __init reset_all_zones_managed_pages(void)
2000 struct pglist_data *pgdat;
2002 if (reset_managed_pages_done)
2005 for_each_online_pgdat(pgdat)
2006 reset_node_managed_pages(pgdat);
2008 reset_managed_pages_done = 1;
2012 * memblock_free_all - release free pages to the buddy allocator
2014 * Return: the number of pages actually released.
2016 unsigned long __init memblock_free_all(void)
2018 unsigned long pages;
2020 reset_all_zones_managed_pages();
2022 pages = free_low_memory_core_early();
2023 totalram_pages_add(pages);
2028 #if defined(CONFIG_DEBUG_FS) && !defined(CONFIG_ARCH_DISCARD_MEMBLOCK)
2030 static int memblock_debug_show(struct seq_file *m, void *private)
2032 struct memblock_type *type = m->private;
2033 struct memblock_region *reg;
2037 for (i = 0; i < type->cnt; i++) {
2038 reg = &type->regions[i];
2039 end = reg->base + reg->size - 1;
2041 seq_printf(m, "%4d: ", i);
2042 seq_printf(m, "%pa..%pa\n", ®->base, &end);
2046 DEFINE_SHOW_ATTRIBUTE(memblock_debug);
2048 static int __init memblock_init_debugfs(void)
2050 struct dentry *root = debugfs_create_dir("memblock", NULL);
2052 debugfs_create_file("memory", 0444, root,
2053 &memblock.memory, &memblock_debug_fops);
2054 debugfs_create_file("reserved", 0444, root,
2055 &memblock.reserved, &memblock_debug_fops);
2056 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
2057 debugfs_create_file("physmem", 0444, root,
2058 &memblock.physmem, &memblock_debug_fops);
2063 __initcall(memblock_init_debugfs);
2065 #endif /* CONFIG_DEBUG_FS */