1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * Procedures for maintaining information about logical memory blocks.
5 * Peter Bergner, IBM Corp. June 2001.
6 * Copyright (C) 2001 Peter Bergner.
9 #include <linux/kernel.h>
10 #include <linux/slab.h>
11 #include <linux/init.h>
12 #include <linux/bitops.h>
13 #include <linux/poison.h>
14 #include <linux/pfn.h>
15 #include <linux/debugfs.h>
16 #include <linux/kmemleak.h>
17 #include <linux/seq_file.h>
18 #include <linux/memblock.h>
20 #include <asm/sections.h>
25 #define INIT_MEMBLOCK_REGIONS 128
26 #define INIT_PHYSMEM_REGIONS 4
28 #ifndef INIT_MEMBLOCK_RESERVED_REGIONS
29 # define INIT_MEMBLOCK_RESERVED_REGIONS INIT_MEMBLOCK_REGIONS
33 * DOC: memblock overview
35 * Memblock is a method of managing memory regions during the early
36 * boot period when the usual kernel memory allocators are not up and
39 * Memblock views the system memory as collections of contiguous
40 * regions. There are several types of these collections:
42 * * ``memory`` - describes the physical memory available to the
43 * kernel; this may differ from the actual physical memory installed
44 * in the system, for instance when the memory is restricted with
45 * ``mem=`` command line parameter
46 * * ``reserved`` - describes the regions that were allocated
47 * * ``physmap`` - describes the actual physical memory regardless of
48 * the possible restrictions; the ``physmap`` type is only available
49 * on some architectures.
51 * Each region is represented by :c:type:`struct memblock_region` that
52 * defines the region extents, its attributes and NUMA node id on NUMA
53 * systems. Every memory type is described by the :c:type:`struct
54 * memblock_type` which contains an array of memory regions along with
55 * the allocator metadata. The memory types are nicely wrapped with
56 * :c:type:`struct memblock`. This structure is statically initialzed
57 * at build time. The region arrays for the "memory" and "reserved"
58 * types are initially sized to %INIT_MEMBLOCK_REGIONS and for the
59 * "physmap" type to %INIT_PHYSMEM_REGIONS.
60 * The memblock_allow_resize() enables automatic resizing of the region
61 * arrays during addition of new regions. This feature should be used
62 * with care so that memory allocated for the region array will not
63 * overlap with areas that should be reserved, for example initrd.
65 * The early architecture setup should tell memblock what the physical
66 * memory layout is by using memblock_add() or memblock_add_node()
67 * functions. The first function does not assign the region to a NUMA
68 * node and it is appropriate for UMA systems. Yet, it is possible to
69 * use it on NUMA systems as well and assign the region to a NUMA node
70 * later in the setup process using memblock_set_node(). The
71 * memblock_add_node() performs such an assignment directly.
73 * Once memblock is setup the memory can be allocated using one of the
76 * * memblock_phys_alloc*() - these functions return the **physical**
77 * address of the allocated memory
78 * * memblock_alloc*() - these functions return the **virtual** address
79 * of the allocated memory.
81 * Note, that both API variants use implicit assumptions about allowed
82 * memory ranges and the fallback methods. Consult the documentation
83 * of memblock_alloc_internal() and memblock_alloc_range_nid()
84 * functions for more elaborate description.
86 * As the system boot progresses, the architecture specific mem_init()
87 * function frees all the memory to the buddy page allocator.
89 * Unless an architecture enables %CONFIG_ARCH_KEEP_MEMBLOCK, the
90 * memblock data structures will be discarded after the system
91 * initialization completes.
94 #ifndef CONFIG_NEED_MULTIPLE_NODES
95 struct pglist_data __refdata contig_page_data;
96 EXPORT_SYMBOL(contig_page_data);
99 unsigned long max_low_pfn;
100 unsigned long min_low_pfn;
101 unsigned long max_pfn;
102 unsigned long long max_possible_pfn;
104 static struct memblock_region memblock_memory_init_regions[INIT_MEMBLOCK_REGIONS] __initdata_memblock;
105 static struct memblock_region memblock_reserved_init_regions[INIT_MEMBLOCK_RESERVED_REGIONS] __initdata_memblock;
106 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
107 static struct memblock_region memblock_physmem_init_regions[INIT_PHYSMEM_REGIONS] __initdata_memblock;
110 struct memblock memblock __initdata_memblock = {
111 .memory.regions = memblock_memory_init_regions,
112 .memory.cnt = 1, /* empty dummy entry */
113 .memory.max = INIT_MEMBLOCK_REGIONS,
114 .memory.name = "memory",
116 .reserved.regions = memblock_reserved_init_regions,
117 .reserved.cnt = 1, /* empty dummy entry */
118 .reserved.max = INIT_MEMBLOCK_RESERVED_REGIONS,
119 .reserved.name = "reserved",
121 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
122 .physmem.regions = memblock_physmem_init_regions,
123 .physmem.cnt = 1, /* empty dummy entry */
124 .physmem.max = INIT_PHYSMEM_REGIONS,
125 .physmem.name = "physmem",
129 .current_limit = MEMBLOCK_ALLOC_ANYWHERE,
132 int memblock_debug __initdata_memblock;
133 static bool system_has_some_mirror __initdata_memblock = false;
134 static int memblock_can_resize __initdata_memblock;
135 static int memblock_memory_in_slab __initdata_memblock = 0;
136 static int memblock_reserved_in_slab __initdata_memblock = 0;
138 static enum memblock_flags __init_memblock choose_memblock_flags(void)
140 return system_has_some_mirror ? MEMBLOCK_MIRROR : MEMBLOCK_NONE;
143 /* adjust *@size so that (@base + *@size) doesn't overflow, return new size */
144 static inline phys_addr_t memblock_cap_size(phys_addr_t base, phys_addr_t *size)
146 return *size = min(*size, PHYS_ADDR_MAX - base);
150 * Address comparison utilities
152 static unsigned long __init_memblock memblock_addrs_overlap(phys_addr_t base1, phys_addr_t size1,
153 phys_addr_t base2, phys_addr_t size2)
155 return ((base1 < (base2 + size2)) && (base2 < (base1 + size1)));
158 bool __init_memblock memblock_overlaps_region(struct memblock_type *type,
159 phys_addr_t base, phys_addr_t size)
163 for (i = 0; i < type->cnt; i++)
164 if (memblock_addrs_overlap(base, size, type->regions[i].base,
165 type->regions[i].size))
167 return i < type->cnt;
171 * __memblock_find_range_bottom_up - find free area utility in bottom-up
172 * @start: start of candidate range
173 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_ANYWHERE or
174 * %MEMBLOCK_ALLOC_ACCESSIBLE
175 * @size: size of free area to find
176 * @align: alignment of free area to find
177 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
178 * @flags: pick from blocks based on memory attributes
180 * Utility called from memblock_find_in_range_node(), find free area bottom-up.
183 * Found address on success, 0 on failure.
185 static phys_addr_t __init_memblock
186 __memblock_find_range_bottom_up(phys_addr_t start, phys_addr_t end,
187 phys_addr_t size, phys_addr_t align, int nid,
188 enum memblock_flags flags)
190 phys_addr_t this_start, this_end, cand;
193 for_each_free_mem_range(i, nid, flags, &this_start, &this_end, NULL) {
194 this_start = clamp(this_start, start, end);
195 this_end = clamp(this_end, start, end);
197 cand = round_up(this_start, align);
198 if (cand < this_end && this_end - cand >= size)
206 * __memblock_find_range_top_down - find free area utility, in top-down
207 * @start: start of candidate range
208 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_ANYWHERE or
209 * %MEMBLOCK_ALLOC_ACCESSIBLE
210 * @size: size of free area to find
211 * @align: alignment of free area to find
212 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
213 * @flags: pick from blocks based on memory attributes
215 * Utility called from memblock_find_in_range_node(), find free area top-down.
218 * Found address on success, 0 on failure.
220 static phys_addr_t __init_memblock
221 __memblock_find_range_top_down(phys_addr_t start, phys_addr_t end,
222 phys_addr_t size, phys_addr_t align, int nid,
223 enum memblock_flags flags)
225 phys_addr_t this_start, this_end, cand;
228 for_each_free_mem_range_reverse(i, nid, flags, &this_start, &this_end,
230 this_start = clamp(this_start, start, end);
231 this_end = clamp(this_end, start, end);
236 cand = round_down(this_end - size, align);
237 if (cand >= this_start)
245 * memblock_find_in_range_node - find free area in given range and node
246 * @size: size of free area to find
247 * @align: alignment of free area to find
248 * @start: start of candidate range
249 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_ANYWHERE or
250 * %MEMBLOCK_ALLOC_ACCESSIBLE
251 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
252 * @flags: pick from blocks based on memory attributes
254 * Find @size free area aligned to @align in the specified range and node.
256 * When allocation direction is bottom-up, the @start should be greater
257 * than the end of the kernel image. Otherwise, it will be trimmed. The
258 * reason is that we want the bottom-up allocation just near the kernel
259 * image so it is highly likely that the allocated memory and the kernel
260 * will reside in the same node.
262 * If bottom-up allocation failed, will try to allocate memory top-down.
265 * Found address on success, 0 on failure.
267 static phys_addr_t __init_memblock memblock_find_in_range_node(phys_addr_t size,
268 phys_addr_t align, phys_addr_t start,
269 phys_addr_t end, int nid,
270 enum memblock_flags flags)
272 phys_addr_t kernel_end, ret;
275 if (end == MEMBLOCK_ALLOC_ACCESSIBLE ||
276 end == MEMBLOCK_ALLOC_KASAN)
277 end = memblock.current_limit;
279 /* avoid allocating the first page */
280 start = max_t(phys_addr_t, start, PAGE_SIZE);
281 end = max(start, end);
282 kernel_end = __pa_symbol(_end);
285 * try bottom-up allocation only when bottom-up mode
286 * is set and @end is above the kernel image.
288 if (memblock_bottom_up() && end > kernel_end) {
289 phys_addr_t bottom_up_start;
291 /* make sure we will allocate above the kernel */
292 bottom_up_start = max(start, kernel_end);
294 /* ok, try bottom-up allocation first */
295 ret = __memblock_find_range_bottom_up(bottom_up_start, end,
296 size, align, nid, flags);
301 * we always limit bottom-up allocation above the kernel,
302 * but top-down allocation doesn't have the limit, so
303 * retrying top-down allocation may succeed when bottom-up
306 * bottom-up allocation is expected to be fail very rarely,
307 * so we use WARN_ONCE() here to see the stack trace if
310 WARN_ONCE(IS_ENABLED(CONFIG_MEMORY_HOTREMOVE),
311 "memblock: bottom-up allocation failed, memory hotremove may be affected\n");
314 return __memblock_find_range_top_down(start, end, size, align, nid,
319 * memblock_find_in_range - find free area in given range
320 * @start: start of candidate range
321 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_ANYWHERE or
322 * %MEMBLOCK_ALLOC_ACCESSIBLE
323 * @size: size of free area to find
324 * @align: alignment of free area to find
326 * Find @size free area aligned to @align in the specified range.
329 * Found address on success, 0 on failure.
331 phys_addr_t __init_memblock memblock_find_in_range(phys_addr_t start,
332 phys_addr_t end, phys_addr_t size,
336 enum memblock_flags flags = choose_memblock_flags();
339 ret = memblock_find_in_range_node(size, align, start, end,
340 NUMA_NO_NODE, flags);
342 if (!ret && (flags & MEMBLOCK_MIRROR)) {
343 pr_warn("Could not allocate %pap bytes of mirrored memory\n",
345 flags &= ~MEMBLOCK_MIRROR;
352 static void __init_memblock memblock_remove_region(struct memblock_type *type, unsigned long r)
354 type->total_size -= type->regions[r].size;
355 memmove(&type->regions[r], &type->regions[r + 1],
356 (type->cnt - (r + 1)) * sizeof(type->regions[r]));
359 /* Special case for empty arrays */
360 if (type->cnt == 0) {
361 WARN_ON(type->total_size != 0);
363 type->regions[0].base = 0;
364 type->regions[0].size = 0;
365 type->regions[0].flags = 0;
366 memblock_set_region_node(&type->regions[0], MAX_NUMNODES);
370 #ifndef CONFIG_ARCH_KEEP_MEMBLOCK
372 * memblock_discard - discard memory and reserved arrays if they were allocated
374 void __init memblock_discard(void)
376 phys_addr_t addr, size;
378 if (memblock.reserved.regions != memblock_reserved_init_regions) {
379 addr = __pa(memblock.reserved.regions);
380 size = PAGE_ALIGN(sizeof(struct memblock_region) *
381 memblock.reserved.max);
382 __memblock_free_late(addr, size);
385 if (memblock.memory.regions != memblock_memory_init_regions) {
386 addr = __pa(memblock.memory.regions);
387 size = PAGE_ALIGN(sizeof(struct memblock_region) *
388 memblock.memory.max);
389 __memblock_free_late(addr, size);
395 * memblock_double_array - double the size of the memblock regions array
396 * @type: memblock type of the regions array being doubled
397 * @new_area_start: starting address of memory range to avoid overlap with
398 * @new_area_size: size of memory range to avoid overlap with
400 * Double the size of the @type regions array. If memblock is being used to
401 * allocate memory for a new reserved regions array and there is a previously
402 * allocated memory range [@new_area_start, @new_area_start + @new_area_size]
403 * waiting to be reserved, ensure the memory used by the new array does
407 * 0 on success, -1 on failure.
409 static int __init_memblock memblock_double_array(struct memblock_type *type,
410 phys_addr_t new_area_start,
411 phys_addr_t new_area_size)
413 struct memblock_region *new_array, *old_array;
414 phys_addr_t old_alloc_size, new_alloc_size;
415 phys_addr_t old_size, new_size, addr, new_end;
416 int use_slab = slab_is_available();
419 /* We don't allow resizing until we know about the reserved regions
420 * of memory that aren't suitable for allocation
422 if (!memblock_can_resize)
425 /* Calculate new doubled size */
426 old_size = type->max * sizeof(struct memblock_region);
427 new_size = old_size << 1;
429 * We need to allocated new one align to PAGE_SIZE,
430 * so we can free them completely later.
432 old_alloc_size = PAGE_ALIGN(old_size);
433 new_alloc_size = PAGE_ALIGN(new_size);
435 /* Retrieve the slab flag */
436 if (type == &memblock.memory)
437 in_slab = &memblock_memory_in_slab;
439 in_slab = &memblock_reserved_in_slab;
441 /* Try to find some space for it */
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 static 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_NEED_MULTIPLE_NODES
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("%s: [%pa-%pa] %pS\n", __func__,
698 &base, &end, (void *)_RET_IP_);
700 return memblock_add_range(&memblock.memory, base, size, MAX_NUMNODES, 0);
704 * memblock_isolate_range - isolate given range into disjoint memblocks
705 * @type: memblock type to isolate range for
706 * @base: base of range to isolate
707 * @size: size of range to isolate
708 * @start_rgn: out parameter for the start of isolated region
709 * @end_rgn: out parameter for the end of isolated region
711 * Walk @type and ensure that regions don't cross the boundaries defined by
712 * [@base, @base + @size). Crossing regions are split at the boundaries,
713 * which may create at most two more regions. The index of the first
714 * region inside the range is returned in *@start_rgn and end in *@end_rgn.
717 * 0 on success, -errno on failure.
719 static int __init_memblock memblock_isolate_range(struct memblock_type *type,
720 phys_addr_t base, phys_addr_t size,
721 int *start_rgn, int *end_rgn)
723 phys_addr_t end = base + memblock_cap_size(base, &size);
725 struct memblock_region *rgn;
727 *start_rgn = *end_rgn = 0;
732 /* we'll create at most two more regions */
733 while (type->cnt + 2 > type->max)
734 if (memblock_double_array(type, base, size) < 0)
737 for_each_memblock_type(idx, type, rgn) {
738 phys_addr_t rbase = rgn->base;
739 phys_addr_t rend = rbase + rgn->size;
748 * @rgn intersects from below. Split and continue
749 * to process the next region - the new top half.
752 rgn->size -= base - rbase;
753 type->total_size -= base - rbase;
754 memblock_insert_region(type, idx, rbase, base - rbase,
755 memblock_get_region_node(rgn),
757 } else if (rend > end) {
759 * @rgn intersects from above. Split and redo the
760 * current region - the new bottom half.
763 rgn->size -= end - rbase;
764 type->total_size -= end - rbase;
765 memblock_insert_region(type, idx--, rbase, end - rbase,
766 memblock_get_region_node(rgn),
769 /* @rgn is fully contained, record it */
779 static int __init_memblock memblock_remove_range(struct memblock_type *type,
780 phys_addr_t base, phys_addr_t size)
782 int start_rgn, end_rgn;
785 ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
789 for (i = end_rgn - 1; i >= start_rgn; i--)
790 memblock_remove_region(type, i);
794 int __init_memblock memblock_remove(phys_addr_t base, phys_addr_t size)
796 phys_addr_t end = base + size - 1;
798 memblock_dbg("%s: [%pa-%pa] %pS\n", __func__,
799 &base, &end, (void *)_RET_IP_);
801 return memblock_remove_range(&memblock.memory, base, size);
805 * memblock_free - 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("%s: [%pa-%pa] %pS\n", __func__,
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("%s: [%pa-%pa] %pS\n", __func__,
828 &base, &end, (void *)_RET_IP_);
830 return memblock_add_range(&memblock.reserved, base, size, MAX_NUMNODES, 0);
833 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
834 int __init_memblock memblock_physmem_add(phys_addr_t base, phys_addr_t size)
836 phys_addr_t end = base + size - 1;
838 memblock_dbg("%s: [%pa-%pa] %pS\n", __func__,
839 &base, &end, (void *)_RET_IP_);
841 return memblock_add_range(&memblock.physmem, base, size, MAX_NUMNODES, 0);
846 * memblock_setclr_flag - set or clear flag for a memory region
847 * @base: base address of the region
848 * @size: size of the region
849 * @set: set or clear the flag
850 * @flag: the flag to udpate
852 * This function isolates region [@base, @base + @size), and sets/clears flag
854 * Return: 0 on success, -errno on failure.
856 static int __init_memblock memblock_setclr_flag(phys_addr_t base,
857 phys_addr_t size, int set, int flag)
859 struct memblock_type *type = &memblock.memory;
860 int i, ret, start_rgn, end_rgn;
862 ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
866 for (i = start_rgn; i < end_rgn; i++) {
867 struct memblock_region *r = &type->regions[i];
875 memblock_merge_regions(type);
880 * memblock_mark_hotplug - Mark hotpluggable memory with flag MEMBLOCK_HOTPLUG.
881 * @base: the base phys addr of the region
882 * @size: the size of the region
884 * Return: 0 on success, -errno on failure.
886 int __init_memblock memblock_mark_hotplug(phys_addr_t base, phys_addr_t size)
888 return memblock_setclr_flag(base, size, 1, MEMBLOCK_HOTPLUG);
892 * memblock_clear_hotplug - Clear flag MEMBLOCK_HOTPLUG for a specified region.
893 * @base: the base phys addr of the region
894 * @size: the size of the region
896 * Return: 0 on success, -errno on failure.
898 int __init_memblock memblock_clear_hotplug(phys_addr_t base, phys_addr_t size)
900 return memblock_setclr_flag(base, size, 0, MEMBLOCK_HOTPLUG);
904 * memblock_mark_mirror - Mark mirrored memory with flag MEMBLOCK_MIRROR.
905 * @base: the base phys addr of the region
906 * @size: the size of the region
908 * Return: 0 on success, -errno on failure.
910 int __init_memblock memblock_mark_mirror(phys_addr_t base, phys_addr_t size)
912 system_has_some_mirror = true;
914 return memblock_setclr_flag(base, size, 1, MEMBLOCK_MIRROR);
918 * memblock_mark_nomap - Mark a memory region with flag MEMBLOCK_NOMAP.
919 * @base: the base phys addr of the region
920 * @size: the size of the region
922 * Return: 0 on success, -errno on failure.
924 int __init_memblock memblock_mark_nomap(phys_addr_t base, phys_addr_t size)
926 return memblock_setclr_flag(base, size, 1, MEMBLOCK_NOMAP);
930 * memblock_clear_nomap - Clear flag MEMBLOCK_NOMAP for a specified region.
931 * @base: the base phys addr of the region
932 * @size: the size of the region
934 * Return: 0 on success, -errno on failure.
936 int __init_memblock memblock_clear_nomap(phys_addr_t base, phys_addr_t size)
938 return memblock_setclr_flag(base, size, 0, MEMBLOCK_NOMAP);
942 * __next_reserved_mem_region - next function for for_each_reserved_region()
943 * @idx: pointer to u64 loop variable
944 * @out_start: ptr to phys_addr_t for start address of the region, can be %NULL
945 * @out_end: ptr to phys_addr_t for end address of the region, can be %NULL
947 * Iterate over all reserved memory regions.
949 void __init_memblock __next_reserved_mem_region(u64 *idx,
950 phys_addr_t *out_start,
951 phys_addr_t *out_end)
953 struct memblock_type *type = &memblock.reserved;
955 if (*idx < type->cnt) {
956 struct memblock_region *r = &type->regions[*idx];
957 phys_addr_t base = r->base;
958 phys_addr_t size = r->size;
963 *out_end = base + size - 1;
969 /* signal end of iteration */
973 static bool should_skip_region(struct memblock_region *m, int nid, int flags)
975 int m_nid = memblock_get_region_node(m);
977 /* only memory regions are associated with nodes, check it */
978 if (nid != NUMA_NO_NODE && nid != m_nid)
981 /* skip hotpluggable memory regions if needed */
982 if (movable_node_is_enabled() && memblock_is_hotpluggable(m))
985 /* if we want mirror memory skip non-mirror memory regions */
986 if ((flags & MEMBLOCK_MIRROR) && !memblock_is_mirror(m))
989 /* skip nomap memory unless we were asked for it explicitly */
990 if (!(flags & MEMBLOCK_NOMAP) && memblock_is_nomap(m))
997 * __next_mem_range - next function for for_each_free_mem_range() etc.
998 * @idx: pointer to u64 loop variable
999 * @nid: node selector, %NUMA_NO_NODE for all nodes
1000 * @flags: pick from blocks based on memory attributes
1001 * @type_a: pointer to memblock_type from where the range is taken
1002 * @type_b: pointer to memblock_type which excludes memory from being taken
1003 * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL
1004 * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL
1005 * @out_nid: ptr to int for nid of the range, can be %NULL
1007 * Find the first area from *@idx which matches @nid, fill the out
1008 * parameters, and update *@idx for the next iteration. The lower 32bit of
1009 * *@idx contains index into type_a and the upper 32bit indexes the
1010 * areas before each region in type_b. For example, if type_b regions
1011 * look like the following,
1013 * 0:[0-16), 1:[32-48), 2:[128-130)
1015 * The upper 32bit indexes the following regions.
1017 * 0:[0-0), 1:[16-32), 2:[48-128), 3:[130-MAX)
1019 * As both region arrays are sorted, the function advances the two indices
1020 * in lockstep and returns each intersection.
1022 void __init_memblock __next_mem_range(u64 *idx, int nid,
1023 enum memblock_flags flags,
1024 struct memblock_type *type_a,
1025 struct memblock_type *type_b,
1026 phys_addr_t *out_start,
1027 phys_addr_t *out_end, int *out_nid)
1029 int idx_a = *idx & 0xffffffff;
1030 int idx_b = *idx >> 32;
1032 if (WARN_ONCE(nid == MAX_NUMNODES,
1033 "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
1036 for (; idx_a < type_a->cnt; idx_a++) {
1037 struct memblock_region *m = &type_a->regions[idx_a];
1039 phys_addr_t m_start = m->base;
1040 phys_addr_t m_end = m->base + m->size;
1041 int m_nid = memblock_get_region_node(m);
1043 if (should_skip_region(m, nid, flags))
1048 *out_start = m_start;
1054 *idx = (u32)idx_a | (u64)idx_b << 32;
1058 /* scan areas before each reservation */
1059 for (; idx_b < type_b->cnt + 1; idx_b++) {
1060 struct memblock_region *r;
1061 phys_addr_t r_start;
1064 r = &type_b->regions[idx_b];
1065 r_start = idx_b ? r[-1].base + r[-1].size : 0;
1066 r_end = idx_b < type_b->cnt ?
1067 r->base : PHYS_ADDR_MAX;
1070 * if idx_b advanced past idx_a,
1071 * break out to advance idx_a
1073 if (r_start >= m_end)
1075 /* if the two regions intersect, we're done */
1076 if (m_start < r_end) {
1079 max(m_start, r_start);
1081 *out_end = min(m_end, r_end);
1085 * The region which ends first is
1086 * advanced for the next iteration.
1092 *idx = (u32)idx_a | (u64)idx_b << 32;
1098 /* signal end of iteration */
1103 * __next_mem_range_rev - generic next function for for_each_*_range_rev()
1105 * @idx: pointer to u64 loop variable
1106 * @nid: node selector, %NUMA_NO_NODE for all nodes
1107 * @flags: pick from blocks based on memory attributes
1108 * @type_a: pointer to memblock_type from where the range is taken
1109 * @type_b: pointer to memblock_type which excludes memory from being taken
1110 * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL
1111 * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL
1112 * @out_nid: ptr to int for nid of the range, can be %NULL
1114 * Finds the next range from type_a which is not marked as unsuitable
1117 * Reverse of __next_mem_range().
1119 void __init_memblock __next_mem_range_rev(u64 *idx, int nid,
1120 enum memblock_flags flags,
1121 struct memblock_type *type_a,
1122 struct memblock_type *type_b,
1123 phys_addr_t *out_start,
1124 phys_addr_t *out_end, int *out_nid)
1126 int idx_a = *idx & 0xffffffff;
1127 int idx_b = *idx >> 32;
1129 if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
1132 if (*idx == (u64)ULLONG_MAX) {
1133 idx_a = type_a->cnt - 1;
1135 idx_b = type_b->cnt;
1140 for (; idx_a >= 0; idx_a--) {
1141 struct memblock_region *m = &type_a->regions[idx_a];
1143 phys_addr_t m_start = m->base;
1144 phys_addr_t m_end = m->base + m->size;
1145 int m_nid = memblock_get_region_node(m);
1147 if (should_skip_region(m, nid, flags))
1152 *out_start = m_start;
1158 *idx = (u32)idx_a | (u64)idx_b << 32;
1162 /* scan areas before each reservation */
1163 for (; idx_b >= 0; idx_b--) {
1164 struct memblock_region *r;
1165 phys_addr_t r_start;
1168 r = &type_b->regions[idx_b];
1169 r_start = idx_b ? r[-1].base + r[-1].size : 0;
1170 r_end = idx_b < type_b->cnt ?
1171 r->base : PHYS_ADDR_MAX;
1173 * if idx_b advanced past idx_a,
1174 * break out to advance idx_a
1177 if (r_end <= m_start)
1179 /* if the two regions intersect, we're done */
1180 if (m_end > r_start) {
1182 *out_start = max(m_start, r_start);
1184 *out_end = min(m_end, r_end);
1187 if (m_start >= r_start)
1191 *idx = (u32)idx_a | (u64)idx_b << 32;
1196 /* signal end of iteration */
1201 * Common iterator interface used to define for_each_mem_pfn_range().
1203 void __init_memblock __next_mem_pfn_range(int *idx, int nid,
1204 unsigned long *out_start_pfn,
1205 unsigned long *out_end_pfn, int *out_nid)
1207 struct memblock_type *type = &memblock.memory;
1208 struct memblock_region *r;
1211 while (++*idx < type->cnt) {
1212 r = &type->regions[*idx];
1213 r_nid = memblock_get_region_node(r);
1215 if (PFN_UP(r->base) >= PFN_DOWN(r->base + r->size))
1217 if (nid == MAX_NUMNODES || nid == r_nid)
1220 if (*idx >= type->cnt) {
1226 *out_start_pfn = PFN_UP(r->base);
1228 *out_end_pfn = PFN_DOWN(r->base + r->size);
1234 * memblock_set_node - set node ID on memblock regions
1235 * @base: base of area to set node ID for
1236 * @size: size of area to set node ID for
1237 * @type: memblock type to set node ID for
1238 * @nid: node ID to set
1240 * Set the nid of memblock @type regions in [@base, @base + @size) to @nid.
1241 * Regions which cross the area boundaries are split as necessary.
1244 * 0 on success, -errno on failure.
1246 int __init_memblock memblock_set_node(phys_addr_t base, phys_addr_t size,
1247 struct memblock_type *type, int nid)
1249 #ifdef CONFIG_NEED_MULTIPLE_NODES
1250 int start_rgn, end_rgn;
1253 ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
1257 for (i = start_rgn; i < end_rgn; i++)
1258 memblock_set_region_node(&type->regions[i], nid);
1260 memblock_merge_regions(type);
1265 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1267 * __next_mem_pfn_range_in_zone - iterator for for_each_*_range_in_zone()
1269 * @idx: pointer to u64 loop variable
1270 * @zone: zone in which all of the memory blocks reside
1271 * @out_spfn: ptr to ulong for start pfn of the range, can be %NULL
1272 * @out_epfn: ptr to ulong for end pfn of the range, can be %NULL
1274 * This function is meant to be a zone/pfn specific wrapper for the
1275 * for_each_mem_range type iterators. Specifically they are used in the
1276 * deferred memory init routines and as such we were duplicating much of
1277 * this logic throughout the code. So instead of having it in multiple
1278 * locations it seemed like it would make more sense to centralize this to
1279 * one new iterator that does everything they need.
1281 void __init_memblock
1282 __next_mem_pfn_range_in_zone(u64 *idx, struct zone *zone,
1283 unsigned long *out_spfn, unsigned long *out_epfn)
1285 int zone_nid = zone_to_nid(zone);
1286 phys_addr_t spa, epa;
1289 __next_mem_range(idx, zone_nid, MEMBLOCK_NONE,
1290 &memblock.memory, &memblock.reserved,
1293 while (*idx != U64_MAX) {
1294 unsigned long epfn = PFN_DOWN(epa);
1295 unsigned long spfn = PFN_UP(spa);
1298 * Verify the end is at least past the start of the zone and
1299 * that we have at least one PFN to initialize.
1301 if (zone->zone_start_pfn < epfn && spfn < epfn) {
1302 /* if we went too far just stop searching */
1303 if (zone_end_pfn(zone) <= spfn) {
1309 *out_spfn = max(zone->zone_start_pfn, spfn);
1311 *out_epfn = min(zone_end_pfn(zone), epfn);
1316 __next_mem_range(idx, zone_nid, MEMBLOCK_NONE,
1317 &memblock.memory, &memblock.reserved,
1321 /* signal end of iteration */
1323 *out_spfn = ULONG_MAX;
1328 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
1331 * memblock_alloc_range_nid - allocate boot memory block
1332 * @size: size of memory block to be allocated in bytes
1333 * @align: alignment of the region and block's size
1334 * @start: the lower bound of the memory region to allocate (phys address)
1335 * @end: the upper bound of the memory region to allocate (phys address)
1336 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1337 * @exact_nid: control the allocation fall back to other nodes
1339 * The allocation is performed from memory region limited by
1340 * memblock.current_limit if @end == %MEMBLOCK_ALLOC_ACCESSIBLE.
1342 * If the specified node can not hold the requested memory and @exact_nid
1343 * is false, the allocation falls back to any node in the system.
1345 * For systems with memory mirroring, the allocation is attempted first
1346 * from the regions with mirroring enabled and then retried from any
1349 * In addition, function sets the min_count to 0 using kmemleak_alloc_phys for
1350 * allocated boot memory block, so that it is never reported as leaks.
1353 * Physical address of allocated memory block on success, %0 on failure.
1355 phys_addr_t __init memblock_alloc_range_nid(phys_addr_t size,
1356 phys_addr_t align, phys_addr_t start,
1357 phys_addr_t end, int nid,
1360 enum memblock_flags flags = choose_memblock_flags();
1363 if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
1367 /* Can't use WARNs this early in boot on powerpc */
1369 align = SMP_CACHE_BYTES;
1373 found = memblock_find_in_range_node(size, align, start, end, nid,
1375 if (found && !memblock_reserve(found, size))
1378 if (nid != NUMA_NO_NODE && !exact_nid) {
1379 found = memblock_find_in_range_node(size, align, start,
1382 if (found && !memblock_reserve(found, size))
1386 if (flags & MEMBLOCK_MIRROR) {
1387 flags &= ~MEMBLOCK_MIRROR;
1388 pr_warn("Could not allocate %pap bytes of mirrored memory\n",
1396 /* Skip kmemleak for kasan_init() due to high volume. */
1397 if (end != MEMBLOCK_ALLOC_KASAN)
1399 * The min_count is set to 0 so that memblock allocated
1400 * blocks are never reported as leaks. This is because many
1401 * of these blocks are only referred via the physical
1402 * address which is not looked up by kmemleak.
1404 kmemleak_alloc_phys(found, size, 0, 0);
1410 * memblock_phys_alloc_range - allocate a memory block inside specified range
1411 * @size: size of memory block to be allocated in bytes
1412 * @align: alignment of the region and block's size
1413 * @start: the lower bound of the memory region to allocate (physical address)
1414 * @end: the upper bound of the memory region to allocate (physical address)
1416 * Allocate @size bytes in the between @start and @end.
1418 * Return: physical address of the allocated memory block on success,
1421 phys_addr_t __init memblock_phys_alloc_range(phys_addr_t size,
1426 return memblock_alloc_range_nid(size, align, start, end, NUMA_NO_NODE,
1431 * memblock_phys_alloc_try_nid - allocate a memory block from specified MUMA node
1432 * @size: size of memory block to be allocated in bytes
1433 * @align: alignment of the region and block's size
1434 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1436 * Allocates memory block from the specified NUMA node. If the node
1437 * has no available memory, attempts to allocated from any node in the
1440 * Return: physical address of the allocated memory block on success,
1443 phys_addr_t __init memblock_phys_alloc_try_nid(phys_addr_t size, phys_addr_t align, int nid)
1445 return memblock_alloc_range_nid(size, align, 0,
1446 MEMBLOCK_ALLOC_ACCESSIBLE, nid, false);
1450 * memblock_alloc_internal - allocate boot memory block
1451 * @size: size of memory block to be allocated in bytes
1452 * @align: alignment of the region and block's size
1453 * @min_addr: the lower bound of the memory region to allocate (phys address)
1454 * @max_addr: the upper bound of the memory region to allocate (phys address)
1455 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1456 * @exact_nid: control the allocation fall back to other nodes
1458 * Allocates memory block using memblock_alloc_range_nid() and
1459 * converts the returned physical address to virtual.
1461 * The @min_addr limit is dropped if it can not be satisfied and the allocation
1462 * will fall back to memory below @min_addr. Other constraints, such
1463 * as node and mirrored memory will be handled again in
1464 * memblock_alloc_range_nid().
1467 * Virtual address of allocated memory block on success, NULL on failure.
1469 static void * __init memblock_alloc_internal(
1470 phys_addr_t size, phys_addr_t align,
1471 phys_addr_t min_addr, phys_addr_t max_addr,
1472 int nid, bool exact_nid)
1477 * Detect any accidental use of these APIs after slab is ready, as at
1478 * this moment memblock may be deinitialized already and its
1479 * internal data may be destroyed (after execution of memblock_free_all)
1481 if (WARN_ON_ONCE(slab_is_available()))
1482 return kzalloc_node(size, GFP_NOWAIT, nid);
1484 if (max_addr > memblock.current_limit)
1485 max_addr = memblock.current_limit;
1487 alloc = memblock_alloc_range_nid(size, align, min_addr, max_addr, nid,
1490 /* retry allocation without lower limit */
1491 if (!alloc && min_addr)
1492 alloc = memblock_alloc_range_nid(size, align, 0, max_addr, nid,
1498 return phys_to_virt(alloc);
1502 * memblock_alloc_exact_nid_raw - allocate boot memory block on the exact node
1503 * without zeroing memory
1504 * @size: size of memory block to be allocated in bytes
1505 * @align: alignment of the region and block's size
1506 * @min_addr: the lower bound of the memory region from where the allocation
1507 * is preferred (phys address)
1508 * @max_addr: the upper bound of the memory region from where the allocation
1509 * is preferred (phys address), or %MEMBLOCK_ALLOC_ACCESSIBLE to
1510 * allocate only from memory limited by memblock.current_limit value
1511 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1513 * Public function, provides additional debug information (including caller
1514 * info), if enabled. Does not zero allocated memory.
1517 * Virtual address of allocated memory block on success, NULL on failure.
1519 void * __init memblock_alloc_exact_nid_raw(
1520 phys_addr_t size, phys_addr_t align,
1521 phys_addr_t min_addr, phys_addr_t max_addr,
1526 memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=%pa max_addr=%pa %pS\n",
1527 __func__, (u64)size, (u64)align, nid, &min_addr,
1528 &max_addr, (void *)_RET_IP_);
1530 ptr = memblock_alloc_internal(size, align,
1531 min_addr, max_addr, nid, true);
1532 if (ptr && size > 0)
1533 page_init_poison(ptr, size);
1539 * memblock_alloc_try_nid_raw - allocate boot memory block without zeroing
1540 * memory and without panicking
1541 * @size: size of memory block to be allocated in bytes
1542 * @align: alignment of the region and block's size
1543 * @min_addr: the lower bound of the memory region from where the allocation
1544 * is preferred (phys address)
1545 * @max_addr: the upper bound of the memory region from where the allocation
1546 * is preferred (phys address), or %MEMBLOCK_ALLOC_ACCESSIBLE to
1547 * allocate only from memory limited by memblock.current_limit value
1548 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1550 * Public function, provides additional debug information (including caller
1551 * info), if enabled. Does not zero allocated memory, does not panic if request
1552 * cannot be satisfied.
1555 * Virtual address of allocated memory block on success, NULL on failure.
1557 void * __init memblock_alloc_try_nid_raw(
1558 phys_addr_t size, phys_addr_t align,
1559 phys_addr_t min_addr, phys_addr_t max_addr,
1564 memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=%pa max_addr=%pa %pS\n",
1565 __func__, (u64)size, (u64)align, nid, &min_addr,
1566 &max_addr, (void *)_RET_IP_);
1568 ptr = memblock_alloc_internal(size, align,
1569 min_addr, max_addr, nid, false);
1570 if (ptr && size > 0)
1571 page_init_poison(ptr, size);
1577 * memblock_alloc_try_nid - allocate boot memory block
1578 * @size: size of memory block to be allocated in bytes
1579 * @align: alignment of the region and block's size
1580 * @min_addr: the lower bound of the memory region from where the allocation
1581 * is preferred (phys address)
1582 * @max_addr: the upper bound of the memory region from where the allocation
1583 * is preferred (phys address), or %MEMBLOCK_ALLOC_ACCESSIBLE to
1584 * allocate only from memory limited by memblock.current_limit value
1585 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1587 * Public function, provides additional debug information (including caller
1588 * info), if enabled. This function zeroes the allocated memory.
1591 * Virtual address of allocated memory block on success, NULL on failure.
1593 void * __init memblock_alloc_try_nid(
1594 phys_addr_t size, phys_addr_t align,
1595 phys_addr_t min_addr, phys_addr_t max_addr,
1600 memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=%pa max_addr=%pa %pS\n",
1601 __func__, (u64)size, (u64)align, nid, &min_addr,
1602 &max_addr, (void *)_RET_IP_);
1603 ptr = memblock_alloc_internal(size, align,
1604 min_addr, max_addr, nid, false);
1606 memset(ptr, 0, size);
1612 * __memblock_free_late - free pages directly to buddy allocator
1613 * @base: phys starting address of the boot memory block
1614 * @size: size of the boot memory block in bytes
1616 * This is only useful when the memblock allocator has already been torn
1617 * down, but we are still initializing the system. Pages are released directly
1618 * to the buddy allocator.
1620 void __init __memblock_free_late(phys_addr_t base, phys_addr_t size)
1622 phys_addr_t cursor, end;
1624 end = base + size - 1;
1625 memblock_dbg("%s: [%pa-%pa] %pS\n",
1626 __func__, &base, &end, (void *)_RET_IP_);
1627 kmemleak_free_part_phys(base, size);
1628 cursor = PFN_UP(base);
1629 end = PFN_DOWN(base + size);
1631 for (; cursor < end; cursor++) {
1632 memblock_free_pages(pfn_to_page(cursor), cursor, 0);
1633 totalram_pages_inc();
1638 * Remaining API functions
1641 phys_addr_t __init_memblock memblock_phys_mem_size(void)
1643 return memblock.memory.total_size;
1646 phys_addr_t __init_memblock memblock_reserved_size(void)
1648 return memblock.reserved.total_size;
1651 phys_addr_t __init memblock_mem_size(unsigned long limit_pfn)
1653 unsigned long pages = 0;
1654 struct memblock_region *r;
1655 unsigned long start_pfn, end_pfn;
1657 for_each_memblock(memory, r) {
1658 start_pfn = memblock_region_memory_base_pfn(r);
1659 end_pfn = memblock_region_memory_end_pfn(r);
1660 start_pfn = min_t(unsigned long, start_pfn, limit_pfn);
1661 end_pfn = min_t(unsigned long, end_pfn, limit_pfn);
1662 pages += end_pfn - start_pfn;
1665 return PFN_PHYS(pages);
1668 /* lowest address */
1669 phys_addr_t __init_memblock memblock_start_of_DRAM(void)
1671 return memblock.memory.regions[0].base;
1674 phys_addr_t __init_memblock memblock_end_of_DRAM(void)
1676 int idx = memblock.memory.cnt - 1;
1678 return (memblock.memory.regions[idx].base + memblock.memory.regions[idx].size);
1681 static phys_addr_t __init_memblock __find_max_addr(phys_addr_t limit)
1683 phys_addr_t max_addr = PHYS_ADDR_MAX;
1684 struct memblock_region *r;
1687 * translate the memory @limit size into the max address within one of
1688 * the memory memblock regions, if the @limit exceeds the total size
1689 * of those regions, max_addr will keep original value PHYS_ADDR_MAX
1691 for_each_memblock(memory, r) {
1692 if (limit <= r->size) {
1693 max_addr = r->base + limit;
1702 void __init memblock_enforce_memory_limit(phys_addr_t limit)
1704 phys_addr_t max_addr;
1709 max_addr = __find_max_addr(limit);
1711 /* @limit exceeds the total size of the memory, do nothing */
1712 if (max_addr == PHYS_ADDR_MAX)
1715 /* truncate both memory and reserved regions */
1716 memblock_remove_range(&memblock.memory, max_addr,
1718 memblock_remove_range(&memblock.reserved, max_addr,
1722 void __init memblock_cap_memory_range(phys_addr_t base, phys_addr_t size)
1724 int start_rgn, end_rgn;
1730 ret = memblock_isolate_range(&memblock.memory, base, size,
1731 &start_rgn, &end_rgn);
1735 /* remove all the MAP regions */
1736 for (i = memblock.memory.cnt - 1; i >= end_rgn; i--)
1737 if (!memblock_is_nomap(&memblock.memory.regions[i]))
1738 memblock_remove_region(&memblock.memory, i);
1740 for (i = start_rgn - 1; i >= 0; i--)
1741 if (!memblock_is_nomap(&memblock.memory.regions[i]))
1742 memblock_remove_region(&memblock.memory, i);
1744 /* truncate the reserved regions */
1745 memblock_remove_range(&memblock.reserved, 0, base);
1746 memblock_remove_range(&memblock.reserved,
1747 base + size, PHYS_ADDR_MAX);
1750 void __init memblock_mem_limit_remove_map(phys_addr_t limit)
1752 phys_addr_t max_addr;
1757 max_addr = __find_max_addr(limit);
1759 /* @limit exceeds the total size of the memory, do nothing */
1760 if (max_addr == PHYS_ADDR_MAX)
1763 memblock_cap_memory_range(0, max_addr);
1766 static int __init_memblock memblock_search(struct memblock_type *type, phys_addr_t addr)
1768 unsigned int left = 0, right = type->cnt;
1771 unsigned int mid = (right + left) / 2;
1773 if (addr < type->regions[mid].base)
1775 else if (addr >= (type->regions[mid].base +
1776 type->regions[mid].size))
1780 } while (left < right);
1784 bool __init_memblock memblock_is_reserved(phys_addr_t addr)
1786 return memblock_search(&memblock.reserved, addr) != -1;
1789 bool __init_memblock memblock_is_memory(phys_addr_t addr)
1791 return memblock_search(&memblock.memory, addr) != -1;
1794 bool __init_memblock memblock_is_map_memory(phys_addr_t addr)
1796 int i = memblock_search(&memblock.memory, addr);
1800 return !memblock_is_nomap(&memblock.memory.regions[i]);
1803 int __init_memblock memblock_search_pfn_nid(unsigned long pfn,
1804 unsigned long *start_pfn, unsigned long *end_pfn)
1806 struct memblock_type *type = &memblock.memory;
1807 int mid = memblock_search(type, PFN_PHYS(pfn));
1812 *start_pfn = PFN_DOWN(type->regions[mid].base);
1813 *end_pfn = PFN_DOWN(type->regions[mid].base + type->regions[mid].size);
1815 return memblock_get_region_node(&type->regions[mid]);
1819 * memblock_is_region_memory - check if a region is a subset of memory
1820 * @base: base of region to check
1821 * @size: size of region to check
1823 * Check if the region [@base, @base + @size) is a subset of a memory block.
1826 * 0 if false, non-zero if true
1828 bool __init_memblock memblock_is_region_memory(phys_addr_t base, phys_addr_t size)
1830 int idx = memblock_search(&memblock.memory, base);
1831 phys_addr_t end = base + memblock_cap_size(base, &size);
1835 return (memblock.memory.regions[idx].base +
1836 memblock.memory.regions[idx].size) >= end;
1840 * memblock_is_region_reserved - check if a region intersects reserved memory
1841 * @base: base of region to check
1842 * @size: size of region to check
1844 * Check if the region [@base, @base + @size) intersects a reserved
1848 * True if they intersect, false if not.
1850 bool __init_memblock memblock_is_region_reserved(phys_addr_t base, phys_addr_t size)
1852 memblock_cap_size(base, &size);
1853 return memblock_overlaps_region(&memblock.reserved, base, size);
1856 void __init_memblock memblock_trim_memory(phys_addr_t align)
1858 phys_addr_t start, end, orig_start, orig_end;
1859 struct memblock_region *r;
1861 for_each_memblock(memory, r) {
1862 orig_start = r->base;
1863 orig_end = r->base + r->size;
1864 start = round_up(orig_start, align);
1865 end = round_down(orig_end, align);
1867 if (start == orig_start && end == orig_end)
1872 r->size = end - start;
1874 memblock_remove_region(&memblock.memory,
1875 r - memblock.memory.regions);
1881 void __init_memblock memblock_set_current_limit(phys_addr_t limit)
1883 memblock.current_limit = limit;
1886 phys_addr_t __init_memblock memblock_get_current_limit(void)
1888 return memblock.current_limit;
1891 static void __init_memblock memblock_dump(struct memblock_type *type)
1893 phys_addr_t base, end, size;
1894 enum memblock_flags flags;
1896 struct memblock_region *rgn;
1898 pr_info(" %s.cnt = 0x%lx\n", type->name, type->cnt);
1900 for_each_memblock_type(idx, type, rgn) {
1901 char nid_buf[32] = "";
1905 end = base + size - 1;
1907 #ifdef CONFIG_NEED_MULTIPLE_NODES
1908 if (memblock_get_region_node(rgn) != MAX_NUMNODES)
1909 snprintf(nid_buf, sizeof(nid_buf), " on node %d",
1910 memblock_get_region_node(rgn));
1912 pr_info(" %s[%#x]\t[%pa-%pa], %pa bytes%s flags: %#x\n",
1913 type->name, idx, &base, &end, &size, nid_buf, flags);
1917 void __init_memblock __memblock_dump_all(void)
1919 pr_info("MEMBLOCK configuration:\n");
1920 pr_info(" memory size = %pa reserved size = %pa\n",
1921 &memblock.memory.total_size,
1922 &memblock.reserved.total_size);
1924 memblock_dump(&memblock.memory);
1925 memblock_dump(&memblock.reserved);
1926 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
1927 memblock_dump(&memblock.physmem);
1931 void __init memblock_allow_resize(void)
1933 memblock_can_resize = 1;
1936 static int __init early_memblock(char *p)
1938 if (p && strstr(p, "debug"))
1942 early_param("memblock", early_memblock);
1944 static void __init __free_pages_memory(unsigned long start, unsigned long end)
1948 while (start < end) {
1949 order = min(MAX_ORDER - 1UL, __ffs(start));
1951 while (start + (1UL << order) > end)
1954 memblock_free_pages(pfn_to_page(start), start, order);
1956 start += (1UL << order);
1960 static unsigned long __init __free_memory_core(phys_addr_t start,
1963 unsigned long start_pfn = PFN_UP(start);
1964 unsigned long end_pfn = min_t(unsigned long,
1965 PFN_DOWN(end), max_low_pfn);
1967 if (start_pfn >= end_pfn)
1970 __free_pages_memory(start_pfn, end_pfn);
1972 return end_pfn - start_pfn;
1975 static unsigned long __init free_low_memory_core_early(void)
1977 unsigned long count = 0;
1978 phys_addr_t start, end;
1981 memblock_clear_hotplug(0, -1);
1983 for_each_reserved_mem_region(i, &start, &end)
1984 reserve_bootmem_region(start, end);
1987 * We need to use NUMA_NO_NODE instead of NODE_DATA(0)->node_id
1988 * because in some case like Node0 doesn't have RAM installed
1989 * low ram will be on Node1
1991 for_each_free_mem_range(i, NUMA_NO_NODE, MEMBLOCK_NONE, &start, &end,
1993 count += __free_memory_core(start, end);
1998 static int reset_managed_pages_done __initdata;
2000 void reset_node_managed_pages(pg_data_t *pgdat)
2004 for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++)
2005 atomic_long_set(&z->managed_pages, 0);
2008 void __init reset_all_zones_managed_pages(void)
2010 struct pglist_data *pgdat;
2012 if (reset_managed_pages_done)
2015 for_each_online_pgdat(pgdat)
2016 reset_node_managed_pages(pgdat);
2018 reset_managed_pages_done = 1;
2022 * memblock_free_all - release free pages to the buddy allocator
2024 * Return: the number of pages actually released.
2026 unsigned long __init memblock_free_all(void)
2028 unsigned long pages;
2030 reset_all_zones_managed_pages();
2032 pages = free_low_memory_core_early();
2033 totalram_pages_add(pages);
2038 #if defined(CONFIG_DEBUG_FS) && defined(CONFIG_ARCH_KEEP_MEMBLOCK)
2040 static int memblock_debug_show(struct seq_file *m, void *private)
2042 struct memblock_type *type = m->private;
2043 struct memblock_region *reg;
2047 for (i = 0; i < type->cnt; i++) {
2048 reg = &type->regions[i];
2049 end = reg->base + reg->size - 1;
2051 seq_printf(m, "%4d: ", i);
2052 seq_printf(m, "%pa..%pa\n", ®->base, &end);
2056 DEFINE_SHOW_ATTRIBUTE(memblock_debug);
2058 static int __init memblock_init_debugfs(void)
2060 struct dentry *root = debugfs_create_dir("memblock", NULL);
2062 debugfs_create_file("memory", 0444, root,
2063 &memblock.memory, &memblock_debug_fops);
2064 debugfs_create_file("reserved", 0444, root,
2065 &memblock.reserved, &memblock_debug_fops);
2066 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
2067 debugfs_create_file("physmem", 0444, root,
2068 &memblock.physmem, &memblock_debug_fops);
2073 __initcall(memblock_init_debugfs);
2075 #endif /* CONFIG_DEBUG_FS */