1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * Procedures for maintaining information about logical memory blocks.
5 * Peter Bergner, IBM Corp. June 2001.
6 * Copyright (C) 2001 Peter Bergner.
9 #include <linux/kernel.h>
10 #include <linux/slab.h>
11 #include <linux/init.h>
12 #include <linux/bitops.h>
13 #include <linux/poison.h>
14 #include <linux/pfn.h>
15 #include <linux/debugfs.h>
16 #include <linux/kmemleak.h>
17 #include <linux/seq_file.h>
18 #include <linux/memblock.h>
20 #include <asm/sections.h>
25 #define INIT_MEMBLOCK_REGIONS 128
26 #define INIT_PHYSMEM_REGIONS 4
28 #ifndef INIT_MEMBLOCK_RESERVED_REGIONS
29 # define INIT_MEMBLOCK_RESERVED_REGIONS INIT_MEMBLOCK_REGIONS
33 * DOC: memblock overview
35 * Memblock is a method of managing memory regions during the early
36 * boot period when the usual kernel memory allocators are not up and
39 * Memblock views the system memory as collections of contiguous
40 * regions. There are several types of these collections:
42 * * ``memory`` - describes the physical memory available to the
43 * kernel; this may differ from the actual physical memory installed
44 * in the system, for instance when the memory is restricted with
45 * ``mem=`` command line parameter
46 * * ``reserved`` - describes the regions that were allocated
47 * * ``physmem`` - describes the actual physical memory available during
48 * boot regardless of the possible restrictions and memory hot(un)plug;
49 * the ``physmem`` type is only available on some architectures.
51 * Each region is represented by struct memblock_region that
52 * defines the region extents, its attributes and NUMA node id on NUMA
53 * systems. Every memory type is described by the struct memblock_type
54 * which contains an array of memory regions along with
55 * the allocator metadata. The "memory" and "reserved" types are nicely
56 * wrapped with struct memblock. This structure is statically
57 * initialized at build time. The region arrays are initially sized to
58 * %INIT_MEMBLOCK_REGIONS for "memory" and %INIT_MEMBLOCK_RESERVED_REGIONS
59 * for "reserved". The region array for "physmem" is initially sized to
60 * %INIT_PHYSMEM_REGIONS.
61 * The memblock_allow_resize() enables automatic resizing of the region
62 * arrays during addition of new regions. This feature should be used
63 * with care so that memory allocated for the region array will not
64 * overlap with areas that should be reserved, for example initrd.
66 * The early architecture setup should tell memblock what the physical
67 * memory layout is by using memblock_add() or memblock_add_node()
68 * functions. The first function does not assign the region to a NUMA
69 * node and it is appropriate for UMA systems. Yet, it is possible to
70 * use it on NUMA systems as well and assign the region to a NUMA node
71 * later in the setup process using memblock_set_node(). The
72 * memblock_add_node() performs such an assignment directly.
74 * Once memblock is setup the memory can be allocated using one of the
77 * * memblock_phys_alloc*() - these functions return the **physical**
78 * address of the allocated memory
79 * * memblock_alloc*() - these functions return the **virtual** address
80 * of the allocated memory.
82 * Note, that both API variants use implicit assumptions about allowed
83 * memory ranges and the fallback methods. Consult the documentation
84 * of memblock_alloc_internal() and memblock_alloc_range_nid()
85 * functions for more elaborate description.
87 * As the system boot progresses, the architecture specific mem_init()
88 * function frees all the memory to the buddy page allocator.
90 * Unless an architecture enables %CONFIG_ARCH_KEEP_MEMBLOCK, the
91 * memblock data structures (except "physmem") will be discarded after the
92 * system initialization completes.
95 #ifndef CONFIG_NEED_MULTIPLE_NODES
96 struct pglist_data __refdata contig_page_data;
97 EXPORT_SYMBOL(contig_page_data);
100 unsigned long max_low_pfn;
101 unsigned long min_low_pfn;
102 unsigned long max_pfn;
103 unsigned long long max_possible_pfn;
105 static struct memblock_region memblock_memory_init_regions[INIT_MEMBLOCK_REGIONS] __initdata_memblock;
106 static struct memblock_region memblock_reserved_init_regions[INIT_MEMBLOCK_RESERVED_REGIONS] __initdata_memblock;
107 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
108 static struct memblock_region memblock_physmem_init_regions[INIT_PHYSMEM_REGIONS];
111 struct memblock memblock __initdata_memblock = {
112 .memory.regions = memblock_memory_init_regions,
113 .memory.cnt = 1, /* empty dummy entry */
114 .memory.max = INIT_MEMBLOCK_REGIONS,
115 .memory.name = "memory",
117 .reserved.regions = memblock_reserved_init_regions,
118 .reserved.cnt = 1, /* empty dummy entry */
119 .reserved.max = INIT_MEMBLOCK_RESERVED_REGIONS,
120 .reserved.name = "reserved",
123 .current_limit = MEMBLOCK_ALLOC_ANYWHERE,
126 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
127 struct memblock_type physmem = {
128 .regions = memblock_physmem_init_regions,
129 .cnt = 1, /* empty dummy entry */
130 .max = INIT_PHYSMEM_REGIONS,
136 * keep a pointer to &memblock.memory in the text section to use it in
137 * __next_mem_range() and its helpers.
138 * For architectures that do not keep memblock data after init, this
139 * pointer will be reset to NULL at memblock_discard()
141 static __refdata struct memblock_type *memblock_memory = &memblock.memory;
143 #define for_each_memblock_type(i, memblock_type, rgn) \
144 for (i = 0, rgn = &memblock_type->regions[0]; \
145 i < memblock_type->cnt; \
146 i++, rgn = &memblock_type->regions[i])
148 #define memblock_dbg(fmt, ...) \
150 if (memblock_debug) \
151 pr_info(fmt, ##__VA_ARGS__); \
154 static int memblock_debug __initdata_memblock;
155 static bool system_has_some_mirror __initdata_memblock = false;
156 static int memblock_can_resize __initdata_memblock;
157 static int memblock_memory_in_slab __initdata_memblock = 0;
158 static int memblock_reserved_in_slab __initdata_memblock = 0;
160 static enum memblock_flags __init_memblock choose_memblock_flags(void)
162 return system_has_some_mirror ? MEMBLOCK_MIRROR : MEMBLOCK_NONE;
165 /* adjust *@size so that (@base + *@size) doesn't overflow, return new size */
166 static inline phys_addr_t memblock_cap_size(phys_addr_t base, phys_addr_t *size)
168 return *size = min(*size, PHYS_ADDR_MAX - base);
172 * Address comparison utilities
174 static unsigned long __init_memblock memblock_addrs_overlap(phys_addr_t base1, phys_addr_t size1,
175 phys_addr_t base2, phys_addr_t size2)
177 return ((base1 < (base2 + size2)) && (base2 < (base1 + size1)));
180 bool __init_memblock memblock_overlaps_region(struct memblock_type *type,
181 phys_addr_t base, phys_addr_t size)
185 for (i = 0; i < type->cnt; i++)
186 if (memblock_addrs_overlap(base, size, type->regions[i].base,
187 type->regions[i].size))
189 return i < type->cnt;
193 * __memblock_find_range_bottom_up - find free area utility in bottom-up
194 * @start: start of candidate range
195 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_ANYWHERE or
196 * %MEMBLOCK_ALLOC_ACCESSIBLE
197 * @size: size of free area to find
198 * @align: alignment of free area to find
199 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
200 * @flags: pick from blocks based on memory attributes
202 * Utility called from memblock_find_in_range_node(), find free area bottom-up.
205 * Found address on success, 0 on failure.
207 static phys_addr_t __init_memblock
208 __memblock_find_range_bottom_up(phys_addr_t start, phys_addr_t end,
209 phys_addr_t size, phys_addr_t align, int nid,
210 enum memblock_flags flags)
212 phys_addr_t this_start, this_end, cand;
215 for_each_free_mem_range(i, nid, flags, &this_start, &this_end, NULL) {
216 this_start = clamp(this_start, start, end);
217 this_end = clamp(this_end, start, end);
219 cand = round_up(this_start, align);
220 if (cand < this_end && this_end - cand >= size)
228 * __memblock_find_range_top_down - find free area utility, in top-down
229 * @start: start of candidate range
230 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_ANYWHERE or
231 * %MEMBLOCK_ALLOC_ACCESSIBLE
232 * @size: size of free area to find
233 * @align: alignment of free area to find
234 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
235 * @flags: pick from blocks based on memory attributes
237 * Utility called from memblock_find_in_range_node(), find free area top-down.
240 * Found address on success, 0 on failure.
242 static phys_addr_t __init_memblock
243 __memblock_find_range_top_down(phys_addr_t start, phys_addr_t end,
244 phys_addr_t size, phys_addr_t align, int nid,
245 enum memblock_flags flags)
247 phys_addr_t this_start, this_end, cand;
250 for_each_free_mem_range_reverse(i, nid, flags, &this_start, &this_end,
252 this_start = clamp(this_start, start, end);
253 this_end = clamp(this_end, start, end);
258 cand = round_down(this_end - size, align);
259 if (cand >= this_start)
267 * memblock_find_in_range_node - find free area in given range and node
268 * @size: size of free area to find
269 * @align: alignment of free area to find
270 * @start: start of candidate range
271 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_ANYWHERE or
272 * %MEMBLOCK_ALLOC_ACCESSIBLE
273 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
274 * @flags: pick from blocks based on memory attributes
276 * Find @size free area aligned to @align in the specified range and node.
278 * When allocation direction is bottom-up, the @start should be greater
279 * than the end of the kernel image. Otherwise, it will be trimmed. The
280 * reason is that we want the bottom-up allocation just near the kernel
281 * image so it is highly likely that the allocated memory and the kernel
282 * will reside in the same node.
284 * If bottom-up allocation failed, will try to allocate memory top-down.
287 * Found address on success, 0 on failure.
289 static phys_addr_t __init_memblock memblock_find_in_range_node(phys_addr_t size,
290 phys_addr_t align, phys_addr_t start,
291 phys_addr_t end, int nid,
292 enum memblock_flags flags)
294 phys_addr_t kernel_end, ret;
297 if (end == MEMBLOCK_ALLOC_ACCESSIBLE ||
298 end == MEMBLOCK_ALLOC_KASAN)
299 end = memblock.current_limit;
301 /* avoid allocating the first page */
302 start = max_t(phys_addr_t, start, PAGE_SIZE);
303 end = max(start, end);
304 kernel_end = __pa_symbol(_end);
307 * try bottom-up allocation only when bottom-up mode
308 * is set and @end is above the kernel image.
310 if (memblock_bottom_up() && end > kernel_end) {
311 phys_addr_t bottom_up_start;
313 /* make sure we will allocate above the kernel */
314 bottom_up_start = max(start, kernel_end);
316 /* ok, try bottom-up allocation first */
317 ret = __memblock_find_range_bottom_up(bottom_up_start, end,
318 size, align, nid, flags);
323 * we always limit bottom-up allocation above the kernel,
324 * but top-down allocation doesn't have the limit, so
325 * retrying top-down allocation may succeed when bottom-up
328 * bottom-up allocation is expected to be fail very rarely,
329 * so we use WARN_ONCE() here to see the stack trace if
332 WARN_ONCE(IS_ENABLED(CONFIG_MEMORY_HOTREMOVE),
333 "memblock: bottom-up allocation failed, memory hotremove may be affected\n");
336 return __memblock_find_range_top_down(start, end, size, align, nid,
341 * memblock_find_in_range - find free area in given range
342 * @start: start of candidate range
343 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_ANYWHERE or
344 * %MEMBLOCK_ALLOC_ACCESSIBLE
345 * @size: size of free area to find
346 * @align: alignment of free area to find
348 * Find @size free area aligned to @align in the specified range.
351 * Found address on success, 0 on failure.
353 phys_addr_t __init_memblock memblock_find_in_range(phys_addr_t start,
354 phys_addr_t end, phys_addr_t size,
358 enum memblock_flags flags = choose_memblock_flags();
361 ret = memblock_find_in_range_node(size, align, start, end,
362 NUMA_NO_NODE, flags);
364 if (!ret && (flags & MEMBLOCK_MIRROR)) {
365 pr_warn("Could not allocate %pap bytes of mirrored memory\n",
367 flags &= ~MEMBLOCK_MIRROR;
374 static void __init_memblock memblock_remove_region(struct memblock_type *type, unsigned long r)
376 type->total_size -= type->regions[r].size;
377 memmove(&type->regions[r], &type->regions[r + 1],
378 (type->cnt - (r + 1)) * sizeof(type->regions[r]));
381 /* Special case for empty arrays */
382 if (type->cnt == 0) {
383 WARN_ON(type->total_size != 0);
385 type->regions[0].base = 0;
386 type->regions[0].size = 0;
387 type->regions[0].flags = 0;
388 memblock_set_region_node(&type->regions[0], MAX_NUMNODES);
392 #ifndef CONFIG_ARCH_KEEP_MEMBLOCK
394 * memblock_discard - discard memory and reserved arrays if they were allocated
396 void __init memblock_discard(void)
398 phys_addr_t addr, size;
400 if (memblock.reserved.regions != memblock_reserved_init_regions) {
401 addr = __pa(memblock.reserved.regions);
402 size = PAGE_ALIGN(sizeof(struct memblock_region) *
403 memblock.reserved.max);
404 __memblock_free_late(addr, size);
407 if (memblock.memory.regions != memblock_memory_init_regions) {
408 addr = __pa(memblock.memory.regions);
409 size = PAGE_ALIGN(sizeof(struct memblock_region) *
410 memblock.memory.max);
411 __memblock_free_late(addr, size);
414 memblock_memory = NULL;
419 * memblock_double_array - double the size of the memblock regions array
420 * @type: memblock type of the regions array being doubled
421 * @new_area_start: starting address of memory range to avoid overlap with
422 * @new_area_size: size of memory range to avoid overlap with
424 * Double the size of the @type regions array. If memblock is being used to
425 * allocate memory for a new reserved regions array and there is a previously
426 * allocated memory range [@new_area_start, @new_area_start + @new_area_size]
427 * waiting to be reserved, ensure the memory used by the new array does
431 * 0 on success, -1 on failure.
433 static int __init_memblock memblock_double_array(struct memblock_type *type,
434 phys_addr_t new_area_start,
435 phys_addr_t new_area_size)
437 struct memblock_region *new_array, *old_array;
438 phys_addr_t old_alloc_size, new_alloc_size;
439 phys_addr_t old_size, new_size, addr, new_end;
440 int use_slab = slab_is_available();
443 /* We don't allow resizing until we know about the reserved regions
444 * of memory that aren't suitable for allocation
446 if (!memblock_can_resize)
449 /* Calculate new doubled size */
450 old_size = type->max * sizeof(struct memblock_region);
451 new_size = old_size << 1;
453 * We need to allocated new one align to PAGE_SIZE,
454 * so we can free them completely later.
456 old_alloc_size = PAGE_ALIGN(old_size);
457 new_alloc_size = PAGE_ALIGN(new_size);
459 /* Retrieve the slab flag */
460 if (type == &memblock.memory)
461 in_slab = &memblock_memory_in_slab;
463 in_slab = &memblock_reserved_in_slab;
465 /* Try to find some space for it */
467 new_array = kmalloc(new_size, GFP_KERNEL);
468 addr = new_array ? __pa(new_array) : 0;
470 /* only exclude range when trying to double reserved.regions */
471 if (type != &memblock.reserved)
472 new_area_start = new_area_size = 0;
474 addr = memblock_find_in_range(new_area_start + new_area_size,
475 memblock.current_limit,
476 new_alloc_size, PAGE_SIZE);
477 if (!addr && new_area_size)
478 addr = memblock_find_in_range(0,
479 min(new_area_start, memblock.current_limit),
480 new_alloc_size, PAGE_SIZE);
482 new_array = addr ? __va(addr) : NULL;
485 pr_err("memblock: Failed to double %s array from %ld to %ld entries !\n",
486 type->name, type->max, type->max * 2);
490 new_end = addr + new_size - 1;
491 memblock_dbg("memblock: %s is doubled to %ld at [%pa-%pa]",
492 type->name, type->max * 2, &addr, &new_end);
495 * Found space, we now need to move the array over before we add the
496 * reserved region since it may be our reserved array itself that is
499 memcpy(new_array, type->regions, old_size);
500 memset(new_array + type->max, 0, old_size);
501 old_array = type->regions;
502 type->regions = new_array;
505 /* Free old array. We needn't free it if the array is the static one */
508 else if (old_array != memblock_memory_init_regions &&
509 old_array != memblock_reserved_init_regions)
510 memblock_free(__pa(old_array), old_alloc_size);
513 * Reserve the new array if that comes from the memblock. Otherwise, we
517 BUG_ON(memblock_reserve(addr, new_alloc_size));
519 /* Update slab flag */
526 * memblock_merge_regions - merge neighboring compatible regions
527 * @type: memblock type to scan
529 * Scan @type and merge neighboring compatible regions.
531 static void __init_memblock memblock_merge_regions(struct memblock_type *type)
535 /* cnt never goes below 1 */
536 while (i < type->cnt - 1) {
537 struct memblock_region *this = &type->regions[i];
538 struct memblock_region *next = &type->regions[i + 1];
540 if (this->base + this->size != next->base ||
541 memblock_get_region_node(this) !=
542 memblock_get_region_node(next) ||
543 this->flags != next->flags) {
544 BUG_ON(this->base + this->size > next->base);
549 this->size += next->size;
550 /* move forward from next + 1, index of which is i + 2 */
551 memmove(next, next + 1, (type->cnt - (i + 2)) * sizeof(*next));
557 * memblock_insert_region - insert new memblock region
558 * @type: memblock type to insert into
559 * @idx: index for the insertion point
560 * @base: base address of the new region
561 * @size: size of the new region
562 * @nid: node id of the new region
563 * @flags: flags of the new region
565 * Insert new memblock region [@base, @base + @size) into @type at @idx.
566 * @type must already have extra room to accommodate the new region.
568 static void __init_memblock memblock_insert_region(struct memblock_type *type,
569 int idx, phys_addr_t base,
572 enum memblock_flags flags)
574 struct memblock_region *rgn = &type->regions[idx];
576 BUG_ON(type->cnt >= type->max);
577 memmove(rgn + 1, rgn, (type->cnt - idx) * sizeof(*rgn));
581 memblock_set_region_node(rgn, nid);
583 type->total_size += size;
587 * memblock_add_range - add new memblock region
588 * @type: memblock type to add new region into
589 * @base: base address of the new region
590 * @size: size of the new region
591 * @nid: nid of the new region
592 * @flags: flags of the new region
594 * Add new memblock region [@base, @base + @size) into @type. The new region
595 * is allowed to overlap with existing ones - overlaps don't affect already
596 * existing regions. @type is guaranteed to be minimal (all neighbouring
597 * compatible regions are merged) after the addition.
600 * 0 on success, -errno on failure.
602 static int __init_memblock memblock_add_range(struct memblock_type *type,
603 phys_addr_t base, phys_addr_t size,
604 int nid, enum memblock_flags flags)
607 phys_addr_t obase = base;
608 phys_addr_t end = base + memblock_cap_size(base, &size);
610 struct memblock_region *rgn;
615 /* special case for empty array */
616 if (type->regions[0].size == 0) {
617 WARN_ON(type->cnt != 1 || type->total_size);
618 type->regions[0].base = base;
619 type->regions[0].size = size;
620 type->regions[0].flags = flags;
621 memblock_set_region_node(&type->regions[0], nid);
622 type->total_size = size;
627 * The following is executed twice. Once with %false @insert and
628 * then with %true. The first counts the number of regions needed
629 * to accommodate the new area. The second actually inserts them.
634 for_each_memblock_type(idx, type, rgn) {
635 phys_addr_t rbase = rgn->base;
636 phys_addr_t rend = rbase + rgn->size;
643 * @rgn overlaps. If it separates the lower part of new
644 * area, insert that portion.
647 #ifdef CONFIG_NEED_MULTIPLE_NODES
648 WARN_ON(nid != memblock_get_region_node(rgn));
650 WARN_ON(flags != rgn->flags);
653 memblock_insert_region(type, idx++, base,
657 /* area below @rend is dealt with, forget about it */
658 base = min(rend, end);
661 /* insert the remaining portion */
665 memblock_insert_region(type, idx, base, end - base,
673 * If this was the first round, resize array and repeat for actual
674 * insertions; otherwise, merge and return.
677 while (type->cnt + nr_new > type->max)
678 if (memblock_double_array(type, obase, size) < 0)
683 memblock_merge_regions(type);
689 * memblock_add_node - add new memblock region within a NUMA node
690 * @base: base address of the new region
691 * @size: size of the new region
692 * @nid: nid of the new region
694 * Add new memblock region [@base, @base + @size) to the "memory"
695 * type. See memblock_add_range() description for mode details
698 * 0 on success, -errno on failure.
700 int __init_memblock memblock_add_node(phys_addr_t base, phys_addr_t size,
703 return memblock_add_range(&memblock.memory, base, size, nid, 0);
707 * memblock_add - add new memblock region
708 * @base: base address of the new region
709 * @size: size of the new region
711 * Add new memblock region [@base, @base + @size) to the "memory"
712 * type. See memblock_add_range() description for mode details
715 * 0 on success, -errno on failure.
717 int __init_memblock memblock_add(phys_addr_t base, phys_addr_t size)
719 phys_addr_t end = base + size - 1;
721 memblock_dbg("%s: [%pa-%pa] %pS\n", __func__,
722 &base, &end, (void *)_RET_IP_);
724 return memblock_add_range(&memblock.memory, base, size, MAX_NUMNODES, 0);
728 * memblock_isolate_range - isolate given range into disjoint memblocks
729 * @type: memblock type to isolate range for
730 * @base: base of range to isolate
731 * @size: size of range to isolate
732 * @start_rgn: out parameter for the start of isolated region
733 * @end_rgn: out parameter for the end of isolated region
735 * Walk @type and ensure that regions don't cross the boundaries defined by
736 * [@base, @base + @size). Crossing regions are split at the boundaries,
737 * which may create at most two more regions. The index of the first
738 * region inside the range is returned in *@start_rgn and end in *@end_rgn.
741 * 0 on success, -errno on failure.
743 static int __init_memblock memblock_isolate_range(struct memblock_type *type,
744 phys_addr_t base, phys_addr_t size,
745 int *start_rgn, int *end_rgn)
747 phys_addr_t end = base + memblock_cap_size(base, &size);
749 struct memblock_region *rgn;
751 *start_rgn = *end_rgn = 0;
756 /* we'll create at most two more regions */
757 while (type->cnt + 2 > type->max)
758 if (memblock_double_array(type, base, size) < 0)
761 for_each_memblock_type(idx, type, rgn) {
762 phys_addr_t rbase = rgn->base;
763 phys_addr_t rend = rbase + rgn->size;
772 * @rgn intersects from below. Split and continue
773 * to process the next region - the new top half.
776 rgn->size -= base - rbase;
777 type->total_size -= base - rbase;
778 memblock_insert_region(type, idx, rbase, base - rbase,
779 memblock_get_region_node(rgn),
781 } else if (rend > end) {
783 * @rgn intersects from above. Split and redo the
784 * current region - the new bottom half.
787 rgn->size -= end - rbase;
788 type->total_size -= end - rbase;
789 memblock_insert_region(type, idx--, rbase, end - rbase,
790 memblock_get_region_node(rgn),
793 /* @rgn is fully contained, record it */
803 static int __init_memblock memblock_remove_range(struct memblock_type *type,
804 phys_addr_t base, phys_addr_t size)
806 int start_rgn, end_rgn;
809 ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
813 for (i = end_rgn - 1; i >= start_rgn; i--)
814 memblock_remove_region(type, i);
818 int __init_memblock memblock_remove(phys_addr_t base, phys_addr_t size)
820 phys_addr_t end = base + size - 1;
822 memblock_dbg("%s: [%pa-%pa] %pS\n", __func__,
823 &base, &end, (void *)_RET_IP_);
825 return memblock_remove_range(&memblock.memory, base, size);
829 * memblock_free - free boot memory block
830 * @base: phys starting address of the boot memory block
831 * @size: size of the boot memory block in bytes
833 * Free boot memory block previously allocated by memblock_alloc_xx() API.
834 * The freeing memory will not be released to the buddy allocator.
836 int __init_memblock memblock_free(phys_addr_t base, phys_addr_t size)
838 phys_addr_t end = base + size - 1;
840 memblock_dbg("%s: [%pa-%pa] %pS\n", __func__,
841 &base, &end, (void *)_RET_IP_);
843 kmemleak_free_part_phys(base, size);
844 return memblock_remove_range(&memblock.reserved, base, size);
847 int __init_memblock memblock_reserve(phys_addr_t base, phys_addr_t size)
849 phys_addr_t end = base + size - 1;
851 memblock_dbg("%s: [%pa-%pa] %pS\n", __func__,
852 &base, &end, (void *)_RET_IP_);
854 return memblock_add_range(&memblock.reserved, base, size, MAX_NUMNODES, 0);
857 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
858 int __init_memblock memblock_physmem_add(phys_addr_t base, phys_addr_t size)
860 phys_addr_t end = base + size - 1;
862 memblock_dbg("%s: [%pa-%pa] %pS\n", __func__,
863 &base, &end, (void *)_RET_IP_);
865 return memblock_add_range(&physmem, base, size, MAX_NUMNODES, 0);
870 * memblock_setclr_flag - set or clear flag for a memory region
871 * @base: base address of the region
872 * @size: size of the region
873 * @set: set or clear the flag
874 * @flag: the flag to udpate
876 * This function isolates region [@base, @base + @size), and sets/clears flag
878 * Return: 0 on success, -errno on failure.
880 static int __init_memblock memblock_setclr_flag(phys_addr_t base,
881 phys_addr_t size, int set, int flag)
883 struct memblock_type *type = &memblock.memory;
884 int i, ret, start_rgn, end_rgn;
886 ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
890 for (i = start_rgn; i < end_rgn; i++) {
891 struct memblock_region *r = &type->regions[i];
899 memblock_merge_regions(type);
904 * memblock_mark_hotplug - Mark hotpluggable memory with flag MEMBLOCK_HOTPLUG.
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_hotplug(phys_addr_t base, phys_addr_t size)
912 return memblock_setclr_flag(base, size, 1, MEMBLOCK_HOTPLUG);
916 * memblock_clear_hotplug - Clear flag MEMBLOCK_HOTPLUG for a specified region.
917 * @base: the base phys addr of the region
918 * @size: the size of the region
920 * Return: 0 on success, -errno on failure.
922 int __init_memblock memblock_clear_hotplug(phys_addr_t base, phys_addr_t size)
924 return memblock_setclr_flag(base, size, 0, MEMBLOCK_HOTPLUG);
928 * memblock_mark_mirror - Mark mirrored memory with flag MEMBLOCK_MIRROR.
929 * @base: the base phys addr of the region
930 * @size: the size of the region
932 * Return: 0 on success, -errno on failure.
934 int __init_memblock memblock_mark_mirror(phys_addr_t base, phys_addr_t size)
936 system_has_some_mirror = true;
938 return memblock_setclr_flag(base, size, 1, MEMBLOCK_MIRROR);
942 * memblock_mark_nomap - Mark a memory region with flag MEMBLOCK_NOMAP.
943 * @base: the base phys addr of the region
944 * @size: the size of the region
946 * Return: 0 on success, -errno on failure.
948 int __init_memblock memblock_mark_nomap(phys_addr_t base, phys_addr_t size)
950 return memblock_setclr_flag(base, size, 1, MEMBLOCK_NOMAP);
954 * memblock_clear_nomap - Clear flag MEMBLOCK_NOMAP for a specified region.
955 * @base: the base phys addr of the region
956 * @size: the size of the region
958 * Return: 0 on success, -errno on failure.
960 int __init_memblock memblock_clear_nomap(phys_addr_t base, phys_addr_t size)
962 return memblock_setclr_flag(base, size, 0, MEMBLOCK_NOMAP);
965 static bool should_skip_region(struct memblock_type *type,
966 struct memblock_region *m,
969 int m_nid = memblock_get_region_node(m);
971 /* we never skip regions when iterating memblock.reserved or physmem */
972 if (type != memblock_memory)
975 /* only memory regions are associated with nodes, check it */
976 if (nid != NUMA_NO_NODE && nid != m_nid)
979 /* skip hotpluggable memory regions if needed */
980 if (movable_node_is_enabled() && memblock_is_hotpluggable(m))
983 /* if we want mirror memory skip non-mirror memory regions */
984 if ((flags & MEMBLOCK_MIRROR) && !memblock_is_mirror(m))
987 /* skip nomap memory unless we were asked for it explicitly */
988 if (!(flags & MEMBLOCK_NOMAP) && memblock_is_nomap(m))
995 * __next_mem_range - next function for for_each_free_mem_range() etc.
996 * @idx: pointer to u64 loop variable
997 * @nid: node selector, %NUMA_NO_NODE for all nodes
998 * @flags: pick from blocks based on memory attributes
999 * @type_a: pointer to memblock_type from where the range is taken
1000 * @type_b: pointer to memblock_type which excludes memory from being taken
1001 * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL
1002 * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL
1003 * @out_nid: ptr to int for nid of the range, can be %NULL
1005 * Find the first area from *@idx which matches @nid, fill the out
1006 * parameters, and update *@idx for the next iteration. The lower 32bit of
1007 * *@idx contains index into type_a and the upper 32bit indexes the
1008 * areas before each region in type_b. For example, if type_b regions
1009 * look like the following,
1011 * 0:[0-16), 1:[32-48), 2:[128-130)
1013 * The upper 32bit indexes the following regions.
1015 * 0:[0-0), 1:[16-32), 2:[48-128), 3:[130-MAX)
1017 * As both region arrays are sorted, the function advances the two indices
1018 * in lockstep and returns each intersection.
1020 void __next_mem_range(u64 *idx, int nid, enum memblock_flags flags,
1021 struct memblock_type *type_a,
1022 struct memblock_type *type_b, phys_addr_t *out_start,
1023 phys_addr_t *out_end, int *out_nid)
1025 int idx_a = *idx & 0xffffffff;
1026 int idx_b = *idx >> 32;
1028 if (WARN_ONCE(nid == MAX_NUMNODES,
1029 "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
1032 for (; idx_a < type_a->cnt; idx_a++) {
1033 struct memblock_region *m = &type_a->regions[idx_a];
1035 phys_addr_t m_start = m->base;
1036 phys_addr_t m_end = m->base + m->size;
1037 int m_nid = memblock_get_region_node(m);
1039 if (should_skip_region(type_a, m, nid, flags))
1044 *out_start = m_start;
1050 *idx = (u32)idx_a | (u64)idx_b << 32;
1054 /* scan areas before each reservation */
1055 for (; idx_b < type_b->cnt + 1; idx_b++) {
1056 struct memblock_region *r;
1057 phys_addr_t r_start;
1060 r = &type_b->regions[idx_b];
1061 r_start = idx_b ? r[-1].base + r[-1].size : 0;
1062 r_end = idx_b < type_b->cnt ?
1063 r->base : PHYS_ADDR_MAX;
1066 * if idx_b advanced past idx_a,
1067 * break out to advance idx_a
1069 if (r_start >= m_end)
1071 /* if the two regions intersect, we're done */
1072 if (m_start < r_end) {
1075 max(m_start, r_start);
1077 *out_end = min(m_end, r_end);
1081 * The region which ends first is
1082 * advanced for the next iteration.
1088 *idx = (u32)idx_a | (u64)idx_b << 32;
1094 /* signal end of iteration */
1099 * __next_mem_range_rev - generic next function for for_each_*_range_rev()
1101 * @idx: pointer to u64 loop variable
1102 * @nid: node selector, %NUMA_NO_NODE for all nodes
1103 * @flags: pick from blocks based on memory attributes
1104 * @type_a: pointer to memblock_type from where the range is taken
1105 * @type_b: pointer to memblock_type which excludes memory from being taken
1106 * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL
1107 * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL
1108 * @out_nid: ptr to int for nid of the range, can be %NULL
1110 * Finds the next range from type_a which is not marked as unsuitable
1113 * Reverse of __next_mem_range().
1115 void __init_memblock __next_mem_range_rev(u64 *idx, int nid,
1116 enum memblock_flags flags,
1117 struct memblock_type *type_a,
1118 struct memblock_type *type_b,
1119 phys_addr_t *out_start,
1120 phys_addr_t *out_end, int *out_nid)
1122 int idx_a = *idx & 0xffffffff;
1123 int idx_b = *idx >> 32;
1125 if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
1128 if (*idx == (u64)ULLONG_MAX) {
1129 idx_a = type_a->cnt - 1;
1131 idx_b = type_b->cnt;
1136 for (; idx_a >= 0; idx_a--) {
1137 struct memblock_region *m = &type_a->regions[idx_a];
1139 phys_addr_t m_start = m->base;
1140 phys_addr_t m_end = m->base + m->size;
1141 int m_nid = memblock_get_region_node(m);
1143 if (should_skip_region(type_a, m, nid, flags))
1148 *out_start = m_start;
1154 *idx = (u32)idx_a | (u64)idx_b << 32;
1158 /* scan areas before each reservation */
1159 for (; idx_b >= 0; idx_b--) {
1160 struct memblock_region *r;
1161 phys_addr_t r_start;
1164 r = &type_b->regions[idx_b];
1165 r_start = idx_b ? r[-1].base + r[-1].size : 0;
1166 r_end = idx_b < type_b->cnt ?
1167 r->base : PHYS_ADDR_MAX;
1169 * if idx_b advanced past idx_a,
1170 * break out to advance idx_a
1173 if (r_end <= m_start)
1175 /* if the two regions intersect, we're done */
1176 if (m_end > r_start) {
1178 *out_start = max(m_start, r_start);
1180 *out_end = min(m_end, r_end);
1183 if (m_start >= r_start)
1187 *idx = (u32)idx_a | (u64)idx_b << 32;
1192 /* signal end of iteration */
1197 * Common iterator interface used to define for_each_mem_pfn_range().
1199 void __init_memblock __next_mem_pfn_range(int *idx, int nid,
1200 unsigned long *out_start_pfn,
1201 unsigned long *out_end_pfn, int *out_nid)
1203 struct memblock_type *type = &memblock.memory;
1204 struct memblock_region *r;
1207 while (++*idx < type->cnt) {
1208 r = &type->regions[*idx];
1209 r_nid = memblock_get_region_node(r);
1211 if (PFN_UP(r->base) >= PFN_DOWN(r->base + r->size))
1213 if (nid == MAX_NUMNODES || nid == r_nid)
1216 if (*idx >= type->cnt) {
1222 *out_start_pfn = PFN_UP(r->base);
1224 *out_end_pfn = PFN_DOWN(r->base + r->size);
1230 * memblock_set_node - set node ID on memblock regions
1231 * @base: base of area to set node ID for
1232 * @size: size of area to set node ID for
1233 * @type: memblock type to set node ID for
1234 * @nid: node ID to set
1236 * Set the nid of memblock @type regions in [@base, @base + @size) to @nid.
1237 * Regions which cross the area boundaries are split as necessary.
1240 * 0 on success, -errno on failure.
1242 int __init_memblock memblock_set_node(phys_addr_t base, phys_addr_t size,
1243 struct memblock_type *type, int nid)
1245 #ifdef CONFIG_NEED_MULTIPLE_NODES
1246 int start_rgn, end_rgn;
1249 ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
1253 for (i = start_rgn; i < end_rgn; i++)
1254 memblock_set_region_node(&type->regions[i], nid);
1256 memblock_merge_regions(type);
1261 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1263 * __next_mem_pfn_range_in_zone - iterator for for_each_*_range_in_zone()
1265 * @idx: pointer to u64 loop variable
1266 * @zone: zone in which all of the memory blocks reside
1267 * @out_spfn: ptr to ulong for start pfn of the range, can be %NULL
1268 * @out_epfn: ptr to ulong for end pfn of the range, can be %NULL
1270 * This function is meant to be a zone/pfn specific wrapper for the
1271 * for_each_mem_range type iterators. Specifically they are used in the
1272 * deferred memory init routines and as such we were duplicating much of
1273 * this logic throughout the code. So instead of having it in multiple
1274 * locations it seemed like it would make more sense to centralize this to
1275 * one new iterator that does everything they need.
1277 void __init_memblock
1278 __next_mem_pfn_range_in_zone(u64 *idx, struct zone *zone,
1279 unsigned long *out_spfn, unsigned long *out_epfn)
1281 int zone_nid = zone_to_nid(zone);
1282 phys_addr_t spa, epa;
1285 __next_mem_range(idx, zone_nid, MEMBLOCK_NONE,
1286 &memblock.memory, &memblock.reserved,
1289 while (*idx != U64_MAX) {
1290 unsigned long epfn = PFN_DOWN(epa);
1291 unsigned long spfn = PFN_UP(spa);
1294 * Verify the end is at least past the start of the zone and
1295 * that we have at least one PFN to initialize.
1297 if (zone->zone_start_pfn < epfn && spfn < epfn) {
1298 /* if we went too far just stop searching */
1299 if (zone_end_pfn(zone) <= spfn) {
1305 *out_spfn = max(zone->zone_start_pfn, spfn);
1307 *out_epfn = min(zone_end_pfn(zone), epfn);
1312 __next_mem_range(idx, zone_nid, MEMBLOCK_NONE,
1313 &memblock.memory, &memblock.reserved,
1317 /* signal end of iteration */
1319 *out_spfn = ULONG_MAX;
1324 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
1327 * memblock_alloc_range_nid - allocate boot memory block
1328 * @size: size of memory block to be allocated in bytes
1329 * @align: alignment of the region and block's size
1330 * @start: the lower bound of the memory region to allocate (phys address)
1331 * @end: the upper bound of the memory region to allocate (phys address)
1332 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1333 * @exact_nid: control the allocation fall back to other nodes
1335 * The allocation is performed from memory region limited by
1336 * memblock.current_limit if @end == %MEMBLOCK_ALLOC_ACCESSIBLE.
1338 * If the specified node can not hold the requested memory and @exact_nid
1339 * is false, the allocation falls back to any node in the system.
1341 * For systems with memory mirroring, the allocation is attempted first
1342 * from the regions with mirroring enabled and then retried from any
1345 * In addition, function sets the min_count to 0 using kmemleak_alloc_phys for
1346 * allocated boot memory block, so that it is never reported as leaks.
1349 * Physical address of allocated memory block on success, %0 on failure.
1351 phys_addr_t __init memblock_alloc_range_nid(phys_addr_t size,
1352 phys_addr_t align, phys_addr_t start,
1353 phys_addr_t end, int nid,
1356 enum memblock_flags flags = choose_memblock_flags();
1359 if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
1363 /* Can't use WARNs this early in boot on powerpc */
1365 align = SMP_CACHE_BYTES;
1369 found = memblock_find_in_range_node(size, align, start, end, nid,
1371 if (found && !memblock_reserve(found, size))
1374 if (nid != NUMA_NO_NODE && !exact_nid) {
1375 found = memblock_find_in_range_node(size, align, start,
1378 if (found && !memblock_reserve(found, size))
1382 if (flags & MEMBLOCK_MIRROR) {
1383 flags &= ~MEMBLOCK_MIRROR;
1384 pr_warn("Could not allocate %pap bytes of mirrored memory\n",
1392 /* Skip kmemleak for kasan_init() due to high volume. */
1393 if (end != MEMBLOCK_ALLOC_KASAN)
1395 * The min_count is set to 0 so that memblock allocated
1396 * blocks are never reported as leaks. This is because many
1397 * of these blocks are only referred via the physical
1398 * address which is not looked up by kmemleak.
1400 kmemleak_alloc_phys(found, size, 0, 0);
1406 * memblock_phys_alloc_range - allocate a memory block inside specified range
1407 * @size: size of memory block to be allocated in bytes
1408 * @align: alignment of the region and block's size
1409 * @start: the lower bound of the memory region to allocate (physical address)
1410 * @end: the upper bound of the memory region to allocate (physical address)
1412 * Allocate @size bytes in the between @start and @end.
1414 * Return: physical address of the allocated memory block on success,
1417 phys_addr_t __init memblock_phys_alloc_range(phys_addr_t size,
1422 return memblock_alloc_range_nid(size, align, start, end, NUMA_NO_NODE,
1427 * memblock_phys_alloc_try_nid - allocate a memory block from specified MUMA node
1428 * @size: size of memory block to be allocated in bytes
1429 * @align: alignment of the region and block's size
1430 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1432 * Allocates memory block from the specified NUMA node. If the node
1433 * has no available memory, attempts to allocated from any node in the
1436 * Return: physical address of the allocated memory block on success,
1439 phys_addr_t __init memblock_phys_alloc_try_nid(phys_addr_t size, phys_addr_t align, int nid)
1441 return memblock_alloc_range_nid(size, align, 0,
1442 MEMBLOCK_ALLOC_ACCESSIBLE, nid, false);
1446 * memblock_alloc_internal - allocate boot memory block
1447 * @size: size of memory block to be allocated in bytes
1448 * @align: alignment of the region and block's size
1449 * @min_addr: the lower bound of the memory region to allocate (phys address)
1450 * @max_addr: the upper bound of the memory region to allocate (phys address)
1451 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1452 * @exact_nid: control the allocation fall back to other nodes
1454 * Allocates memory block using memblock_alloc_range_nid() and
1455 * converts the returned physical address to virtual.
1457 * The @min_addr limit is dropped if it can not be satisfied and the allocation
1458 * will fall back to memory below @min_addr. Other constraints, such
1459 * as node and mirrored memory will be handled again in
1460 * memblock_alloc_range_nid().
1463 * Virtual address of allocated memory block on success, NULL on failure.
1465 static void * __init memblock_alloc_internal(
1466 phys_addr_t size, phys_addr_t align,
1467 phys_addr_t min_addr, phys_addr_t max_addr,
1468 int nid, bool exact_nid)
1473 * Detect any accidental use of these APIs after slab is ready, as at
1474 * this moment memblock may be deinitialized already and its
1475 * internal data may be destroyed (after execution of memblock_free_all)
1477 if (WARN_ON_ONCE(slab_is_available()))
1478 return kzalloc_node(size, GFP_NOWAIT, nid);
1480 if (max_addr > memblock.current_limit)
1481 max_addr = memblock.current_limit;
1483 alloc = memblock_alloc_range_nid(size, align, min_addr, max_addr, nid,
1486 /* retry allocation without lower limit */
1487 if (!alloc && min_addr)
1488 alloc = memblock_alloc_range_nid(size, align, 0, max_addr, nid,
1494 return phys_to_virt(alloc);
1498 * memblock_alloc_exact_nid_raw - allocate boot memory block on the exact node
1499 * without zeroing memory
1500 * @size: size of memory block to be allocated in bytes
1501 * @align: alignment of the region and block's size
1502 * @min_addr: the lower bound of the memory region from where the allocation
1503 * is preferred (phys address)
1504 * @max_addr: the upper bound of the memory region from where the allocation
1505 * is preferred (phys address), or %MEMBLOCK_ALLOC_ACCESSIBLE to
1506 * allocate only from memory limited by memblock.current_limit value
1507 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1509 * Public function, provides additional debug information (including caller
1510 * info), if enabled. Does not zero allocated memory.
1513 * Virtual address of allocated memory block on success, NULL on failure.
1515 void * __init memblock_alloc_exact_nid_raw(
1516 phys_addr_t size, phys_addr_t align,
1517 phys_addr_t min_addr, phys_addr_t max_addr,
1522 memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=%pa max_addr=%pa %pS\n",
1523 __func__, (u64)size, (u64)align, nid, &min_addr,
1524 &max_addr, (void *)_RET_IP_);
1526 ptr = memblock_alloc_internal(size, align,
1527 min_addr, max_addr, nid, true);
1528 if (ptr && size > 0)
1529 page_init_poison(ptr, size);
1535 * memblock_alloc_try_nid_raw - allocate boot memory block without zeroing
1536 * memory and without panicking
1537 * @size: size of memory block to be allocated in bytes
1538 * @align: alignment of the region and block's size
1539 * @min_addr: the lower bound of the memory region from where the allocation
1540 * is preferred (phys address)
1541 * @max_addr: the upper bound of the memory region from where the allocation
1542 * is preferred (phys address), or %MEMBLOCK_ALLOC_ACCESSIBLE to
1543 * allocate only from memory limited by memblock.current_limit value
1544 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1546 * Public function, provides additional debug information (including caller
1547 * info), if enabled. Does not zero allocated memory, does not panic if request
1548 * cannot be satisfied.
1551 * Virtual address of allocated memory block on success, NULL on failure.
1553 void * __init memblock_alloc_try_nid_raw(
1554 phys_addr_t size, phys_addr_t align,
1555 phys_addr_t min_addr, phys_addr_t max_addr,
1560 memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=%pa max_addr=%pa %pS\n",
1561 __func__, (u64)size, (u64)align, nid, &min_addr,
1562 &max_addr, (void *)_RET_IP_);
1564 ptr = memblock_alloc_internal(size, align,
1565 min_addr, max_addr, nid, false);
1566 if (ptr && size > 0)
1567 page_init_poison(ptr, size);
1573 * memblock_alloc_try_nid - allocate boot memory block
1574 * @size: size of memory block to be allocated in bytes
1575 * @align: alignment of the region and block's size
1576 * @min_addr: the lower bound of the memory region from where the allocation
1577 * is preferred (phys address)
1578 * @max_addr: the upper bound of the memory region from where the allocation
1579 * is preferred (phys address), or %MEMBLOCK_ALLOC_ACCESSIBLE to
1580 * allocate only from memory limited by memblock.current_limit value
1581 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1583 * Public function, provides additional debug information (including caller
1584 * info), if enabled. This function zeroes the allocated memory.
1587 * Virtual address of allocated memory block on success, NULL on failure.
1589 void * __init memblock_alloc_try_nid(
1590 phys_addr_t size, phys_addr_t align,
1591 phys_addr_t min_addr, phys_addr_t max_addr,
1596 memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=%pa max_addr=%pa %pS\n",
1597 __func__, (u64)size, (u64)align, nid, &min_addr,
1598 &max_addr, (void *)_RET_IP_);
1599 ptr = memblock_alloc_internal(size, align,
1600 min_addr, max_addr, nid, false);
1602 memset(ptr, 0, size);
1608 * __memblock_free_late - free pages directly to buddy allocator
1609 * @base: phys starting address of the boot memory block
1610 * @size: size of the boot memory block in bytes
1612 * This is only useful when the memblock allocator has already been torn
1613 * down, but we are still initializing the system. Pages are released directly
1614 * to the buddy allocator.
1616 void __init __memblock_free_late(phys_addr_t base, phys_addr_t size)
1618 phys_addr_t cursor, end;
1620 end = base + size - 1;
1621 memblock_dbg("%s: [%pa-%pa] %pS\n",
1622 __func__, &base, &end, (void *)_RET_IP_);
1623 kmemleak_free_part_phys(base, size);
1624 cursor = PFN_UP(base);
1625 end = PFN_DOWN(base + size);
1627 for (; cursor < end; cursor++) {
1628 memblock_free_pages(pfn_to_page(cursor), cursor, 0);
1629 totalram_pages_inc();
1634 * Remaining API functions
1637 phys_addr_t __init_memblock memblock_phys_mem_size(void)
1639 return memblock.memory.total_size;
1642 phys_addr_t __init_memblock memblock_reserved_size(void)
1644 return memblock.reserved.total_size;
1647 /* lowest address */
1648 phys_addr_t __init_memblock memblock_start_of_DRAM(void)
1650 return memblock.memory.regions[0].base;
1653 phys_addr_t __init_memblock memblock_end_of_DRAM(void)
1655 int idx = memblock.memory.cnt - 1;
1657 return (memblock.memory.regions[idx].base + memblock.memory.regions[idx].size);
1660 static phys_addr_t __init_memblock __find_max_addr(phys_addr_t limit)
1662 phys_addr_t max_addr = PHYS_ADDR_MAX;
1663 struct memblock_region *r;
1666 * translate the memory @limit size into the max address within one of
1667 * the memory memblock regions, if the @limit exceeds the total size
1668 * of those regions, max_addr will keep original value PHYS_ADDR_MAX
1670 for_each_mem_region(r) {
1671 if (limit <= r->size) {
1672 max_addr = r->base + limit;
1681 void __init memblock_enforce_memory_limit(phys_addr_t limit)
1683 phys_addr_t max_addr;
1688 max_addr = __find_max_addr(limit);
1690 /* @limit exceeds the total size of the memory, do nothing */
1691 if (max_addr == PHYS_ADDR_MAX)
1694 /* truncate both memory and reserved regions */
1695 memblock_remove_range(&memblock.memory, max_addr,
1697 memblock_remove_range(&memblock.reserved, max_addr,
1701 void __init memblock_cap_memory_range(phys_addr_t base, phys_addr_t size)
1703 int start_rgn, end_rgn;
1709 ret = memblock_isolate_range(&memblock.memory, base, size,
1710 &start_rgn, &end_rgn);
1714 /* remove all the MAP regions */
1715 for (i = memblock.memory.cnt - 1; i >= end_rgn; i--)
1716 if (!memblock_is_nomap(&memblock.memory.regions[i]))
1717 memblock_remove_region(&memblock.memory, i);
1719 for (i = start_rgn - 1; i >= 0; i--)
1720 if (!memblock_is_nomap(&memblock.memory.regions[i]))
1721 memblock_remove_region(&memblock.memory, i);
1723 /* truncate the reserved regions */
1724 memblock_remove_range(&memblock.reserved, 0, base);
1725 memblock_remove_range(&memblock.reserved,
1726 base + size, PHYS_ADDR_MAX);
1729 void __init memblock_mem_limit_remove_map(phys_addr_t limit)
1731 phys_addr_t max_addr;
1736 max_addr = __find_max_addr(limit);
1738 /* @limit exceeds the total size of the memory, do nothing */
1739 if (max_addr == PHYS_ADDR_MAX)
1742 memblock_cap_memory_range(0, max_addr);
1745 static int __init_memblock memblock_search(struct memblock_type *type, phys_addr_t addr)
1747 unsigned int left = 0, right = type->cnt;
1750 unsigned int mid = (right + left) / 2;
1752 if (addr < type->regions[mid].base)
1754 else if (addr >= (type->regions[mid].base +
1755 type->regions[mid].size))
1759 } while (left < right);
1763 bool __init_memblock memblock_is_reserved(phys_addr_t addr)
1765 return memblock_search(&memblock.reserved, addr) != -1;
1768 bool __init_memblock memblock_is_memory(phys_addr_t addr)
1770 return memblock_search(&memblock.memory, addr) != -1;
1773 bool __init_memblock memblock_is_map_memory(phys_addr_t addr)
1775 int i = memblock_search(&memblock.memory, addr);
1779 return !memblock_is_nomap(&memblock.memory.regions[i]);
1782 int __init_memblock memblock_search_pfn_nid(unsigned long pfn,
1783 unsigned long *start_pfn, unsigned long *end_pfn)
1785 struct memblock_type *type = &memblock.memory;
1786 int mid = memblock_search(type, PFN_PHYS(pfn));
1791 *start_pfn = PFN_DOWN(type->regions[mid].base);
1792 *end_pfn = PFN_DOWN(type->regions[mid].base + type->regions[mid].size);
1794 return memblock_get_region_node(&type->regions[mid]);
1798 * memblock_is_region_memory - check if a region is a subset of memory
1799 * @base: base of region to check
1800 * @size: size of region to check
1802 * Check if the region [@base, @base + @size) is a subset of a memory block.
1805 * 0 if false, non-zero if true
1807 bool __init_memblock memblock_is_region_memory(phys_addr_t base, phys_addr_t size)
1809 int idx = memblock_search(&memblock.memory, base);
1810 phys_addr_t end = base + memblock_cap_size(base, &size);
1814 return (memblock.memory.regions[idx].base +
1815 memblock.memory.regions[idx].size) >= end;
1819 * memblock_is_region_reserved - check if a region intersects reserved memory
1820 * @base: base of region to check
1821 * @size: size of region to check
1823 * Check if the region [@base, @base + @size) intersects a reserved
1827 * True if they intersect, false if not.
1829 bool __init_memblock memblock_is_region_reserved(phys_addr_t base, phys_addr_t size)
1831 memblock_cap_size(base, &size);
1832 return memblock_overlaps_region(&memblock.reserved, base, size);
1835 void __init_memblock memblock_trim_memory(phys_addr_t align)
1837 phys_addr_t start, end, orig_start, orig_end;
1838 struct memblock_region *r;
1840 for_each_mem_region(r) {
1841 orig_start = r->base;
1842 orig_end = r->base + r->size;
1843 start = round_up(orig_start, align);
1844 end = round_down(orig_end, align);
1846 if (start == orig_start && end == orig_end)
1851 r->size = end - start;
1853 memblock_remove_region(&memblock.memory,
1854 r - memblock.memory.regions);
1860 void __init_memblock memblock_set_current_limit(phys_addr_t limit)
1862 memblock.current_limit = limit;
1865 phys_addr_t __init_memblock memblock_get_current_limit(void)
1867 return memblock.current_limit;
1870 static void __init_memblock memblock_dump(struct memblock_type *type)
1872 phys_addr_t base, end, size;
1873 enum memblock_flags flags;
1875 struct memblock_region *rgn;
1877 pr_info(" %s.cnt = 0x%lx\n", type->name, type->cnt);
1879 for_each_memblock_type(idx, type, rgn) {
1880 char nid_buf[32] = "";
1884 end = base + size - 1;
1886 #ifdef CONFIG_NEED_MULTIPLE_NODES
1887 if (memblock_get_region_node(rgn) != MAX_NUMNODES)
1888 snprintf(nid_buf, sizeof(nid_buf), " on node %d",
1889 memblock_get_region_node(rgn));
1891 pr_info(" %s[%#x]\t[%pa-%pa], %pa bytes%s flags: %#x\n",
1892 type->name, idx, &base, &end, &size, nid_buf, flags);
1896 static void __init_memblock __memblock_dump_all(void)
1898 pr_info("MEMBLOCK configuration:\n");
1899 pr_info(" memory size = %pa reserved size = %pa\n",
1900 &memblock.memory.total_size,
1901 &memblock.reserved.total_size);
1903 memblock_dump(&memblock.memory);
1904 memblock_dump(&memblock.reserved);
1905 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
1906 memblock_dump(&physmem);
1910 void __init_memblock memblock_dump_all(void)
1913 __memblock_dump_all();
1916 void __init memblock_allow_resize(void)
1918 memblock_can_resize = 1;
1921 static int __init early_memblock(char *p)
1923 if (p && strstr(p, "debug"))
1927 early_param("memblock", early_memblock);
1929 static void __init __free_pages_memory(unsigned long start, unsigned long end)
1933 while (start < end) {
1934 order = min(MAX_ORDER - 1UL, __ffs(start));
1936 while (start + (1UL << order) > end)
1939 memblock_free_pages(pfn_to_page(start), start, order);
1941 start += (1UL << order);
1945 static unsigned long __init __free_memory_core(phys_addr_t start,
1948 unsigned long start_pfn = PFN_UP(start);
1949 unsigned long end_pfn = min_t(unsigned long,
1950 PFN_DOWN(end), max_low_pfn);
1952 if (start_pfn >= end_pfn)
1955 __free_pages_memory(start_pfn, end_pfn);
1957 return end_pfn - start_pfn;
1960 static unsigned long __init free_low_memory_core_early(void)
1962 unsigned long count = 0;
1963 phys_addr_t start, end;
1966 memblock_clear_hotplug(0, -1);
1968 for_each_reserved_mem_range(i, &start, &end)
1969 reserve_bootmem_region(start, end);
1972 * We need to use NUMA_NO_NODE instead of NODE_DATA(0)->node_id
1973 * because in some case like Node0 doesn't have RAM installed
1974 * low ram will be on Node1
1976 for_each_free_mem_range(i, NUMA_NO_NODE, MEMBLOCK_NONE, &start, &end,
1978 count += __free_memory_core(start, end);
1983 static int reset_managed_pages_done __initdata;
1985 void reset_node_managed_pages(pg_data_t *pgdat)
1989 for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++)
1990 atomic_long_set(&z->managed_pages, 0);
1993 void __init reset_all_zones_managed_pages(void)
1995 struct pglist_data *pgdat;
1997 if (reset_managed_pages_done)
2000 for_each_online_pgdat(pgdat)
2001 reset_node_managed_pages(pgdat);
2003 reset_managed_pages_done = 1;
2007 * memblock_free_all - release free pages to the buddy allocator
2009 * Return: the number of pages actually released.
2011 unsigned long __init memblock_free_all(void)
2013 unsigned long pages;
2015 reset_all_zones_managed_pages();
2017 pages = free_low_memory_core_early();
2018 totalram_pages_add(pages);
2023 #if defined(CONFIG_DEBUG_FS) && defined(CONFIG_ARCH_KEEP_MEMBLOCK)
2025 static int memblock_debug_show(struct seq_file *m, void *private)
2027 struct memblock_type *type = m->private;
2028 struct memblock_region *reg;
2032 for (i = 0; i < type->cnt; i++) {
2033 reg = &type->regions[i];
2034 end = reg->base + reg->size - 1;
2036 seq_printf(m, "%4d: ", i);
2037 seq_printf(m, "%pa..%pa\n", ®->base, &end);
2041 DEFINE_SHOW_ATTRIBUTE(memblock_debug);
2043 static int __init memblock_init_debugfs(void)
2045 struct dentry *root = debugfs_create_dir("memblock", NULL);
2047 debugfs_create_file("memory", 0444, root,
2048 &memblock.memory, &memblock_debug_fops);
2049 debugfs_create_file("reserved", 0444, root,
2050 &memblock.reserved, &memblock_debug_fops);
2051 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
2052 debugfs_create_file("physmem", 0444, root, &physmem,
2053 &memblock_debug_fops);
2058 __initcall(memblock_init_debugfs);
2060 #endif /* CONFIG_DEBUG_FS */