2 * Procedures for maintaining information about logical memory blocks.
4 * Peter Bergner, IBM Corp. June 2001.
5 * Copyright (C) 2001 Peter Bergner.
7 * This program is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU General Public License
9 * as published by the Free Software Foundation; either version
10 * 2 of the License, or (at your option) any later version.
13 #include <linux/kernel.h>
14 #include <linux/slab.h>
15 #include <linux/init.h>
16 #include <linux/bitops.h>
17 #include <linux/poison.h>
18 #include <linux/pfn.h>
19 #include <linux/debugfs.h>
20 #include <linux/seq_file.h>
21 #include <linux/memblock.h>
23 #include <asm-generic/sections.h>
28 static struct memblock_region memblock_memory_init_regions[INIT_MEMBLOCK_REGIONS] __initdata_memblock;
29 static struct memblock_region memblock_reserved_init_regions[INIT_MEMBLOCK_REGIONS] __initdata_memblock;
30 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
31 static struct memblock_region memblock_physmem_init_regions[INIT_PHYSMEM_REGIONS] __initdata_memblock;
34 struct memblock memblock __initdata_memblock = {
35 .memory.regions = memblock_memory_init_regions,
36 .memory.cnt = 1, /* empty dummy entry */
37 .memory.max = INIT_MEMBLOCK_REGIONS,
39 .reserved.regions = memblock_reserved_init_regions,
40 .reserved.cnt = 1, /* empty dummy entry */
41 .reserved.max = INIT_MEMBLOCK_REGIONS,
43 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
44 .physmem.regions = memblock_physmem_init_regions,
45 .physmem.cnt = 1, /* empty dummy entry */
46 .physmem.max = INIT_PHYSMEM_REGIONS,
50 .current_limit = MEMBLOCK_ALLOC_ANYWHERE,
53 int memblock_debug __initdata_memblock;
54 #ifdef CONFIG_MOVABLE_NODE
55 bool movable_node_enabled __initdata_memblock = false;
57 static bool system_has_some_mirror __initdata_memblock = false;
58 static int memblock_can_resize __initdata_memblock;
59 static int memblock_memory_in_slab __initdata_memblock = 0;
60 static int memblock_reserved_in_slab __initdata_memblock = 0;
62 ulong __init_memblock choose_memblock_flags(void)
64 return system_has_some_mirror ? MEMBLOCK_MIRROR : MEMBLOCK_NONE;
67 /* inline so we don't get a warning when pr_debug is compiled out */
68 static __init_memblock const char *
69 memblock_type_name(struct memblock_type *type)
71 if (type == &memblock.memory)
73 else if (type == &memblock.reserved)
79 /* adjust *@size so that (@base + *@size) doesn't overflow, return new size */
80 static inline phys_addr_t memblock_cap_size(phys_addr_t base, phys_addr_t *size)
82 return *size = min(*size, (phys_addr_t)ULLONG_MAX - base);
86 * Address comparison utilities
88 static unsigned long __init_memblock memblock_addrs_overlap(phys_addr_t base1, phys_addr_t size1,
89 phys_addr_t base2, phys_addr_t size2)
91 return ((base1 < (base2 + size2)) && (base2 < (base1 + size1)));
94 bool __init_memblock memblock_overlaps_region(struct memblock_type *type,
95 phys_addr_t base, phys_addr_t size)
99 for (i = 0; i < type->cnt; i++) {
100 phys_addr_t rgnbase = type->regions[i].base;
101 phys_addr_t rgnsize = type->regions[i].size;
102 if (memblock_addrs_overlap(base, size, rgnbase, rgnsize))
106 return i < type->cnt;
110 * __memblock_find_range_bottom_up - find free area utility in bottom-up
111 * @start: start of candidate range
112 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_{ANYWHERE|ACCESSIBLE}
113 * @size: size of free area to find
114 * @align: alignment of free area to find
115 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
116 * @flags: pick from blocks based on memory attributes
118 * Utility called from memblock_find_in_range_node(), find free area bottom-up.
121 * Found address on success, 0 on failure.
123 static phys_addr_t __init_memblock
124 __memblock_find_range_bottom_up(phys_addr_t start, phys_addr_t end,
125 phys_addr_t size, phys_addr_t align, int nid,
128 phys_addr_t this_start, this_end, cand;
131 for_each_free_mem_range(i, nid, flags, &this_start, &this_end, NULL) {
132 this_start = clamp(this_start, start, end);
133 this_end = clamp(this_end, start, end);
135 cand = round_up(this_start, align);
136 if (cand < this_end && this_end - cand >= size)
144 * __memblock_find_range_top_down - find free area utility, in top-down
145 * @start: start of candidate range
146 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_{ANYWHERE|ACCESSIBLE}
147 * @size: size of free area to find
148 * @align: alignment of free area to find
149 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
150 * @flags: pick from blocks based on memory attributes
152 * Utility called from memblock_find_in_range_node(), find free area top-down.
155 * Found address on success, 0 on failure.
157 static phys_addr_t __init_memblock
158 __memblock_find_range_top_down(phys_addr_t start, phys_addr_t end,
159 phys_addr_t size, phys_addr_t align, int nid,
162 phys_addr_t this_start, this_end, cand;
165 for_each_free_mem_range_reverse(i, nid, flags, &this_start, &this_end,
167 this_start = clamp(this_start, start, end);
168 this_end = clamp(this_end, start, end);
173 cand = round_down(this_end - size, align);
174 if (cand >= this_start)
182 * memblock_find_in_range_node - find free area in given range and node
183 * @size: size of free area to find
184 * @align: alignment of free area to find
185 * @start: start of candidate range
186 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_{ANYWHERE|ACCESSIBLE}
187 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
188 * @flags: pick from blocks based on memory attributes
190 * Find @size free area aligned to @align in the specified range and node.
193 * Found address on success, 0 on failure.
195 phys_addr_t __init_memblock memblock_find_in_range_node(phys_addr_t size,
196 phys_addr_t align, phys_addr_t start,
197 phys_addr_t end, int nid, ulong flags)
200 if (end == MEMBLOCK_ALLOC_ACCESSIBLE)
201 end = memblock.current_limit;
203 /* avoid allocating the first page */
204 start = max_t(phys_addr_t, start, PAGE_SIZE);
205 end = max(start, end);
207 if (memblock_bottom_up())
208 return __memblock_find_range_bottom_up(start, end, size, align,
211 return __memblock_find_range_top_down(start, end, size, align,
216 * memblock_find_in_range - find free area in given range
217 * @start: start of candidate range
218 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_{ANYWHERE|ACCESSIBLE}
219 * @size: size of free area to find
220 * @align: alignment of free area to find
222 * Find @size free area aligned to @align in the specified range.
225 * Found address on success, 0 on failure.
227 phys_addr_t __init_memblock memblock_find_in_range(phys_addr_t start,
228 phys_addr_t end, phys_addr_t size,
232 ulong flags = choose_memblock_flags();
235 ret = memblock_find_in_range_node(size, align, start, end,
236 NUMA_NO_NODE, flags);
238 if (!ret && (flags & MEMBLOCK_MIRROR)) {
239 pr_warn("Could not allocate %pap bytes of mirrored memory\n",
241 flags &= ~MEMBLOCK_MIRROR;
248 static void __init_memblock memblock_remove_region(struct memblock_type *type, unsigned long r)
250 type->total_size -= type->regions[r].size;
251 memmove(&type->regions[r], &type->regions[r + 1],
252 (type->cnt - (r + 1)) * sizeof(type->regions[r]));
255 /* Special case for empty arrays */
256 if (type->cnt == 0) {
257 WARN_ON(type->total_size != 0);
259 type->regions[0].base = 0;
260 type->regions[0].size = 0;
261 type->regions[0].flags = 0;
262 memblock_set_region_node(&type->regions[0], MAX_NUMNODES);
266 #ifdef CONFIG_ARCH_DISCARD_MEMBLOCK
268 phys_addr_t __init_memblock get_allocated_memblock_reserved_regions_info(
271 if (memblock.reserved.regions == memblock_reserved_init_regions)
274 *addr = __pa(memblock.reserved.regions);
276 return PAGE_ALIGN(sizeof(struct memblock_region) *
277 memblock.reserved.max);
280 phys_addr_t __init_memblock get_allocated_memblock_memory_regions_info(
283 if (memblock.memory.regions == memblock_memory_init_regions)
286 *addr = __pa(memblock.memory.regions);
288 return PAGE_ALIGN(sizeof(struct memblock_region) *
289 memblock.memory.max);
295 * memblock_double_array - double the size of the memblock regions array
296 * @type: memblock type of the regions array being doubled
297 * @new_area_start: starting address of memory range to avoid overlap with
298 * @new_area_size: size of memory range to avoid overlap with
300 * Double the size of the @type regions array. If memblock is being used to
301 * allocate memory for a new reserved regions array and there is a previously
302 * allocated memory range [@new_area_start,@new_area_start+@new_area_size]
303 * waiting to be reserved, ensure the memory used by the new array does
307 * 0 on success, -1 on failure.
309 static int __init_memblock memblock_double_array(struct memblock_type *type,
310 phys_addr_t new_area_start,
311 phys_addr_t new_area_size)
313 struct memblock_region *new_array, *old_array;
314 phys_addr_t old_alloc_size, new_alloc_size;
315 phys_addr_t old_size, new_size, addr;
316 int use_slab = slab_is_available();
319 /* We don't allow resizing until we know about the reserved regions
320 * of memory that aren't suitable for allocation
322 if (!memblock_can_resize)
325 /* Calculate new doubled size */
326 old_size = type->max * sizeof(struct memblock_region);
327 new_size = old_size << 1;
329 * We need to allocated new one align to PAGE_SIZE,
330 * so we can free them completely later.
332 old_alloc_size = PAGE_ALIGN(old_size);
333 new_alloc_size = PAGE_ALIGN(new_size);
335 /* Retrieve the slab flag */
336 if (type == &memblock.memory)
337 in_slab = &memblock_memory_in_slab;
339 in_slab = &memblock_reserved_in_slab;
341 /* Try to find some space for it.
343 * WARNING: We assume that either slab_is_available() and we use it or
344 * we use MEMBLOCK for allocations. That means that this is unsafe to
345 * use when bootmem is currently active (unless bootmem itself is
346 * implemented on top of MEMBLOCK which isn't the case yet)
348 * This should however not be an issue for now, as we currently only
349 * call into MEMBLOCK while it's still active, or much later when slab
350 * is active for memory hotplug operations
353 new_array = kmalloc(new_size, GFP_KERNEL);
354 addr = new_array ? __pa(new_array) : 0;
356 /* only exclude range when trying to double reserved.regions */
357 if (type != &memblock.reserved)
358 new_area_start = new_area_size = 0;
360 addr = memblock_find_in_range(new_area_start + new_area_size,
361 memblock.current_limit,
362 new_alloc_size, PAGE_SIZE);
363 if (!addr && new_area_size)
364 addr = memblock_find_in_range(0,
365 min(new_area_start, memblock.current_limit),
366 new_alloc_size, PAGE_SIZE);
368 new_array = addr ? __va(addr) : NULL;
371 pr_err("memblock: Failed to double %s array from %ld to %ld entries !\n",
372 memblock_type_name(type), type->max, type->max * 2);
376 memblock_dbg("memblock: %s is doubled to %ld at [%#010llx-%#010llx]",
377 memblock_type_name(type), type->max * 2, (u64)addr,
378 (u64)addr + new_size - 1);
381 * Found space, we now need to move the array over before we add the
382 * reserved region since it may be our reserved array itself that is
385 memcpy(new_array, type->regions, old_size);
386 memset(new_array + type->max, 0, old_size);
387 old_array = type->regions;
388 type->regions = new_array;
391 /* Free old array. We needn't free it if the array is the static one */
394 else if (old_array != memblock_memory_init_regions &&
395 old_array != memblock_reserved_init_regions)
396 memblock_free(__pa(old_array), old_alloc_size);
399 * Reserve the new array if that comes from the memblock. Otherwise, we
403 BUG_ON(memblock_reserve(addr, new_alloc_size));
405 /* Update slab flag */
412 * memblock_merge_regions - merge neighboring compatible regions
413 * @type: memblock type to scan
415 * Scan @type and merge neighboring compatible regions.
417 static void __init_memblock memblock_merge_regions(struct memblock_type *type)
421 /* cnt never goes below 1 */
422 while (i < type->cnt - 1) {
423 struct memblock_region *this = &type->regions[i];
424 struct memblock_region *next = &type->regions[i + 1];
426 if (this->base + this->size != next->base ||
427 memblock_get_region_node(this) !=
428 memblock_get_region_node(next) ||
429 this->flags != next->flags) {
430 BUG_ON(this->base + this->size > next->base);
435 this->size += next->size;
436 /* move forward from next + 1, index of which is i + 2 */
437 memmove(next, next + 1, (type->cnt - (i + 2)) * sizeof(*next));
443 * memblock_insert_region - insert new memblock region
444 * @type: memblock type to insert into
445 * @idx: index for the insertion point
446 * @base: base address of the new region
447 * @size: size of the new region
448 * @nid: node id of the new region
449 * @flags: flags of the new region
451 * Insert new memblock region [@base,@base+@size) into @type at @idx.
452 * @type must already have extra room to accomodate the new region.
454 static void __init_memblock memblock_insert_region(struct memblock_type *type,
455 int idx, phys_addr_t base,
457 int nid, unsigned long flags)
459 struct memblock_region *rgn = &type->regions[idx];
461 BUG_ON(type->cnt >= type->max);
462 memmove(rgn + 1, rgn, (type->cnt - idx) * sizeof(*rgn));
466 memblock_set_region_node(rgn, nid);
468 type->total_size += size;
472 * memblock_add_range - add new memblock region
473 * @type: memblock type to add new region into
474 * @base: base address of the new region
475 * @size: size of the new region
476 * @nid: nid of the new region
477 * @flags: flags of the new region
479 * Add new memblock region [@base,@base+@size) into @type. The new region
480 * is allowed to overlap with existing ones - overlaps don't affect already
481 * existing regions. @type is guaranteed to be minimal (all neighbouring
482 * compatible regions are merged) after the addition.
485 * 0 on success, -errno on failure.
487 int __init_memblock memblock_add_range(struct memblock_type *type,
488 phys_addr_t base, phys_addr_t size,
489 int nid, unsigned long flags)
492 phys_addr_t obase = base;
493 phys_addr_t end = base + memblock_cap_size(base, &size);
499 /* special case for empty array */
500 if (type->regions[0].size == 0) {
501 WARN_ON(type->cnt != 1 || type->total_size);
502 type->regions[0].base = base;
503 type->regions[0].size = size;
504 type->regions[0].flags = flags;
505 memblock_set_region_node(&type->regions[0], nid);
506 type->total_size = size;
511 * The following is executed twice. Once with %false @insert and
512 * then with %true. The first counts the number of regions needed
513 * to accomodate the new area. The second actually inserts them.
518 for (i = 0; i < type->cnt; i++) {
519 struct memblock_region *rgn = &type->regions[i];
520 phys_addr_t rbase = rgn->base;
521 phys_addr_t rend = rbase + rgn->size;
528 * @rgn overlaps. If it separates the lower part of new
529 * area, insert that portion.
532 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
533 WARN_ON(nid != memblock_get_region_node(rgn));
535 WARN_ON(flags != rgn->flags);
538 memblock_insert_region(type, i++, base,
542 /* area below @rend is dealt with, forget about it */
543 base = min(rend, end);
546 /* insert the remaining portion */
550 memblock_insert_region(type, i, base, end - base,
555 * If this was the first round, resize array and repeat for actual
556 * insertions; otherwise, merge and return.
559 while (type->cnt + nr_new > type->max)
560 if (memblock_double_array(type, obase, size) < 0)
565 memblock_merge_regions(type);
570 int __init_memblock memblock_add_node(phys_addr_t base, phys_addr_t size,
573 return memblock_add_range(&memblock.memory, base, size, nid, 0);
576 static int __init_memblock memblock_add_region(phys_addr_t base,
581 struct memblock_type *type = &memblock.memory;
583 memblock_dbg("memblock_add: [%#016llx-%#016llx] flags %#02lx %pF\n",
584 (unsigned long long)base,
585 (unsigned long long)base + size - 1,
586 flags, (void *)_RET_IP_);
588 return memblock_add_range(type, base, size, nid, flags);
591 int __init_memblock memblock_add(phys_addr_t base, phys_addr_t size)
593 return memblock_add_region(base, size, MAX_NUMNODES, 0);
597 * memblock_isolate_range - isolate given range into disjoint memblocks
598 * @type: memblock type to isolate range for
599 * @base: base of range to isolate
600 * @size: size of range to isolate
601 * @start_rgn: out parameter for the start of isolated region
602 * @end_rgn: out parameter for the end of isolated region
604 * Walk @type and ensure that regions don't cross the boundaries defined by
605 * [@base,@base+@size). Crossing regions are split at the boundaries,
606 * which may create at most two more regions. The index of the first
607 * region inside the range is returned in *@start_rgn and end in *@end_rgn.
610 * 0 on success, -errno on failure.
612 static int __init_memblock memblock_isolate_range(struct memblock_type *type,
613 phys_addr_t base, phys_addr_t size,
614 int *start_rgn, int *end_rgn)
616 phys_addr_t end = base + memblock_cap_size(base, &size);
619 *start_rgn = *end_rgn = 0;
624 /* we'll create at most two more regions */
625 while (type->cnt + 2 > type->max)
626 if (memblock_double_array(type, base, size) < 0)
629 for (i = 0; i < type->cnt; i++) {
630 struct memblock_region *rgn = &type->regions[i];
631 phys_addr_t rbase = rgn->base;
632 phys_addr_t rend = rbase + rgn->size;
641 * @rgn intersects from below. Split and continue
642 * to process the next region - the new top half.
645 rgn->size -= base - rbase;
646 type->total_size -= base - rbase;
647 memblock_insert_region(type, i, rbase, base - rbase,
648 memblock_get_region_node(rgn),
650 } else if (rend > end) {
652 * @rgn intersects from above. Split and redo the
653 * current region - the new bottom half.
656 rgn->size -= end - rbase;
657 type->total_size -= end - rbase;
658 memblock_insert_region(type, i--, rbase, end - rbase,
659 memblock_get_region_node(rgn),
662 /* @rgn is fully contained, record it */
672 static int __init_memblock memblock_remove_range(struct memblock_type *type,
673 phys_addr_t base, phys_addr_t size)
675 int start_rgn, end_rgn;
678 ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
682 for (i = end_rgn - 1; i >= start_rgn; i--)
683 memblock_remove_region(type, i);
687 int __init_memblock memblock_remove(phys_addr_t base, phys_addr_t size)
689 return memblock_remove_range(&memblock.memory, base, size);
693 int __init_memblock memblock_free(phys_addr_t base, phys_addr_t size)
695 memblock_dbg(" memblock_free: [%#016llx-%#016llx] %pF\n",
696 (unsigned long long)base,
697 (unsigned long long)base + size - 1,
700 kmemleak_free_part(__va(base), size);
701 return memblock_remove_range(&memblock.reserved, base, size);
704 static int __init_memblock memblock_reserve_region(phys_addr_t base,
709 struct memblock_type *type = &memblock.reserved;
711 memblock_dbg("memblock_reserve: [%#016llx-%#016llx] flags %#02lx %pF\n",
712 (unsigned long long)base,
713 (unsigned long long)base + size - 1,
714 flags, (void *)_RET_IP_);
716 return memblock_add_range(type, base, size, nid, flags);
719 int __init_memblock memblock_reserve(phys_addr_t base, phys_addr_t size)
721 return memblock_reserve_region(base, size, MAX_NUMNODES, 0);
726 * This function isolates region [@base, @base + @size), and sets/clears flag
728 * Return 0 on success, -errno on failure.
730 static int __init_memblock memblock_setclr_flag(phys_addr_t base,
731 phys_addr_t size, int set, int flag)
733 struct memblock_type *type = &memblock.memory;
734 int i, ret, start_rgn, end_rgn;
736 ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
740 for (i = start_rgn; i < end_rgn; i++)
742 memblock_set_region_flags(&type->regions[i], flag);
744 memblock_clear_region_flags(&type->regions[i], flag);
746 memblock_merge_regions(type);
751 * memblock_mark_hotplug - Mark hotpluggable memory with flag MEMBLOCK_HOTPLUG.
752 * @base: the base phys addr of the region
753 * @size: the size of the region
755 * Return 0 on success, -errno on failure.
757 int __init_memblock memblock_mark_hotplug(phys_addr_t base, phys_addr_t size)
759 return memblock_setclr_flag(base, size, 1, MEMBLOCK_HOTPLUG);
763 * memblock_clear_hotplug - Clear flag MEMBLOCK_HOTPLUG for a specified region.
764 * @base: the base phys addr of the region
765 * @size: the size of the region
767 * Return 0 on success, -errno on failure.
769 int __init_memblock memblock_clear_hotplug(phys_addr_t base, phys_addr_t size)
771 return memblock_setclr_flag(base, size, 0, MEMBLOCK_HOTPLUG);
775 * memblock_mark_mirror - Mark mirrored memory with flag MEMBLOCK_MIRROR.
776 * @base: the base phys addr of the region
777 * @size: the size of the region
779 * Return 0 on success, -errno on failure.
781 int __init_memblock memblock_mark_mirror(phys_addr_t base, phys_addr_t size)
783 system_has_some_mirror = true;
785 return memblock_setclr_flag(base, size, 1, MEMBLOCK_MIRROR);
790 * __next_reserved_mem_region - next function for for_each_reserved_region()
791 * @idx: pointer to u64 loop variable
792 * @out_start: ptr to phys_addr_t for start address of the region, can be %NULL
793 * @out_end: ptr to phys_addr_t for end address of the region, can be %NULL
795 * Iterate over all reserved memory regions.
797 void __init_memblock __next_reserved_mem_region(u64 *idx,
798 phys_addr_t *out_start,
799 phys_addr_t *out_end)
801 struct memblock_type *type = &memblock.reserved;
803 if (*idx >= 0 && *idx < type->cnt) {
804 struct memblock_region *r = &type->regions[*idx];
805 phys_addr_t base = r->base;
806 phys_addr_t size = r->size;
811 *out_end = base + size - 1;
817 /* signal end of iteration */
822 * __next__mem_range - next function for for_each_free_mem_range() etc.
823 * @idx: pointer to u64 loop variable
824 * @nid: node selector, %NUMA_NO_NODE for all nodes
825 * @flags: pick from blocks based on memory attributes
826 * @type_a: pointer to memblock_type from where the range is taken
827 * @type_b: pointer to memblock_type which excludes memory from being taken
828 * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL
829 * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL
830 * @out_nid: ptr to int for nid of the range, can be %NULL
832 * Find the first area from *@idx which matches @nid, fill the out
833 * parameters, and update *@idx for the next iteration. The lower 32bit of
834 * *@idx contains index into type_a and the upper 32bit indexes the
835 * areas before each region in type_b. For example, if type_b regions
836 * look like the following,
838 * 0:[0-16), 1:[32-48), 2:[128-130)
840 * The upper 32bit indexes the following regions.
842 * 0:[0-0), 1:[16-32), 2:[48-128), 3:[130-MAX)
844 * As both region arrays are sorted, the function advances the two indices
845 * in lockstep and returns each intersection.
847 void __init_memblock __next_mem_range(u64 *idx, int nid, ulong flags,
848 struct memblock_type *type_a,
849 struct memblock_type *type_b,
850 phys_addr_t *out_start,
851 phys_addr_t *out_end, int *out_nid)
853 int idx_a = *idx & 0xffffffff;
854 int idx_b = *idx >> 32;
856 if (WARN_ONCE(nid == MAX_NUMNODES,
857 "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
860 for (; idx_a < type_a->cnt; idx_a++) {
861 struct memblock_region *m = &type_a->regions[idx_a];
863 phys_addr_t m_start = m->base;
864 phys_addr_t m_end = m->base + m->size;
865 int m_nid = memblock_get_region_node(m);
867 /* only memory regions are associated with nodes, check it */
868 if (nid != NUMA_NO_NODE && nid != m_nid)
871 /* skip hotpluggable memory regions if needed */
872 if (movable_node_is_enabled() && memblock_is_hotpluggable(m))
875 /* if we want mirror memory skip non-mirror memory regions */
876 if ((flags & MEMBLOCK_MIRROR) && !memblock_is_mirror(m))
881 *out_start = m_start;
887 *idx = (u32)idx_a | (u64)idx_b << 32;
891 /* scan areas before each reservation */
892 for (; idx_b < type_b->cnt + 1; idx_b++) {
893 struct memblock_region *r;
897 r = &type_b->regions[idx_b];
898 r_start = idx_b ? r[-1].base + r[-1].size : 0;
899 r_end = idx_b < type_b->cnt ?
900 r->base : ULLONG_MAX;
903 * if idx_b advanced past idx_a,
904 * break out to advance idx_a
906 if (r_start >= m_end)
908 /* if the two regions intersect, we're done */
909 if (m_start < r_end) {
912 max(m_start, r_start);
914 *out_end = min(m_end, r_end);
918 * The region which ends first is
919 * advanced for the next iteration.
925 *idx = (u32)idx_a | (u64)idx_b << 32;
931 /* signal end of iteration */
936 * __next_mem_range_rev - generic next function for for_each_*_range_rev()
938 * Finds the next range from type_a which is not marked as unsuitable
941 * @idx: pointer to u64 loop variable
942 * @nid: node selector, %NUMA_NO_NODE for all nodes
943 * @flags: pick from blocks based on memory attributes
944 * @type_a: pointer to memblock_type from where the range is taken
945 * @type_b: pointer to memblock_type which excludes memory from being taken
946 * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL
947 * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL
948 * @out_nid: ptr to int for nid of the range, can be %NULL
950 * Reverse of __next_mem_range().
952 void __init_memblock __next_mem_range_rev(u64 *idx, int nid, ulong flags,
953 struct memblock_type *type_a,
954 struct memblock_type *type_b,
955 phys_addr_t *out_start,
956 phys_addr_t *out_end, int *out_nid)
958 int idx_a = *idx & 0xffffffff;
959 int idx_b = *idx >> 32;
961 if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
964 if (*idx == (u64)ULLONG_MAX) {
965 idx_a = type_a->cnt - 1;
969 for (; idx_a >= 0; idx_a--) {
970 struct memblock_region *m = &type_a->regions[idx_a];
972 phys_addr_t m_start = m->base;
973 phys_addr_t m_end = m->base + m->size;
974 int m_nid = memblock_get_region_node(m);
976 /* only memory regions are associated with nodes, check it */
977 if (nid != NUMA_NO_NODE && nid != m_nid)
980 /* skip hotpluggable memory regions if needed */
981 if (movable_node_is_enabled() && memblock_is_hotpluggable(m))
984 /* if we want mirror memory skip non-mirror memory regions */
985 if ((flags & MEMBLOCK_MIRROR) && !memblock_is_mirror(m))
990 *out_start = m_start;
996 *idx = (u32)idx_a | (u64)idx_b << 32;
1000 /* scan areas before each reservation */
1001 for (; idx_b >= 0; idx_b--) {
1002 struct memblock_region *r;
1003 phys_addr_t r_start;
1006 r = &type_b->regions[idx_b];
1007 r_start = idx_b ? r[-1].base + r[-1].size : 0;
1008 r_end = idx_b < type_b->cnt ?
1009 r->base : ULLONG_MAX;
1011 * if idx_b advanced past idx_a,
1012 * break out to advance idx_a
1015 if (r_end <= m_start)
1017 /* if the two regions intersect, we're done */
1018 if (m_end > r_start) {
1020 *out_start = max(m_start, r_start);
1022 *out_end = min(m_end, r_end);
1025 if (m_start >= r_start)
1029 *idx = (u32)idx_a | (u64)idx_b << 32;
1034 /* signal end of iteration */
1038 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
1040 * Common iterator interface used to define for_each_mem_range().
1042 void __init_memblock __next_mem_pfn_range(int *idx, int nid,
1043 unsigned long *out_start_pfn,
1044 unsigned long *out_end_pfn, int *out_nid)
1046 struct memblock_type *type = &memblock.memory;
1047 struct memblock_region *r;
1049 while (++*idx < type->cnt) {
1050 r = &type->regions[*idx];
1052 if (PFN_UP(r->base) >= PFN_DOWN(r->base + r->size))
1054 if (nid == MAX_NUMNODES || nid == r->nid)
1057 if (*idx >= type->cnt) {
1063 *out_start_pfn = PFN_UP(r->base);
1065 *out_end_pfn = PFN_DOWN(r->base + r->size);
1071 * memblock_set_node - set node ID on memblock regions
1072 * @base: base of area to set node ID for
1073 * @size: size of area to set node ID for
1074 * @type: memblock type to set node ID for
1075 * @nid: node ID to set
1077 * Set the nid of memblock @type regions in [@base,@base+@size) to @nid.
1078 * Regions which cross the area boundaries are split as necessary.
1081 * 0 on success, -errno on failure.
1083 int __init_memblock memblock_set_node(phys_addr_t base, phys_addr_t size,
1084 struct memblock_type *type, int nid)
1086 int start_rgn, end_rgn;
1089 ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
1093 for (i = start_rgn; i < end_rgn; i++)
1094 memblock_set_region_node(&type->regions[i], nid);
1096 memblock_merge_regions(type);
1099 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
1101 static phys_addr_t __init memblock_alloc_range_nid(phys_addr_t size,
1102 phys_addr_t align, phys_addr_t start,
1103 phys_addr_t end, int nid, ulong flags)
1108 align = SMP_CACHE_BYTES;
1110 found = memblock_find_in_range_node(size, align, start, end, nid,
1112 if (found && !memblock_reserve(found, size)) {
1114 * The min_count is set to 0 so that memblock allocations are
1115 * never reported as leaks.
1117 kmemleak_alloc(__va(found), size, 0, 0);
1123 phys_addr_t __init memblock_alloc_range(phys_addr_t size, phys_addr_t align,
1124 phys_addr_t start, phys_addr_t end,
1127 return memblock_alloc_range_nid(size, align, start, end, NUMA_NO_NODE,
1131 static phys_addr_t __init memblock_alloc_base_nid(phys_addr_t size,
1132 phys_addr_t align, phys_addr_t max_addr,
1133 int nid, ulong flags)
1135 return memblock_alloc_range_nid(size, align, 0, max_addr, nid, flags);
1138 phys_addr_t __init memblock_alloc_nid(phys_addr_t size, phys_addr_t align, int nid)
1140 ulong flags = choose_memblock_flags();
1144 ret = memblock_alloc_base_nid(size, align, MEMBLOCK_ALLOC_ACCESSIBLE,
1147 if (!ret && (flags & MEMBLOCK_MIRROR)) {
1148 flags &= ~MEMBLOCK_MIRROR;
1154 phys_addr_t __init __memblock_alloc_base(phys_addr_t size, phys_addr_t align, phys_addr_t max_addr)
1156 return memblock_alloc_base_nid(size, align, max_addr, NUMA_NO_NODE,
1160 phys_addr_t __init memblock_alloc_base(phys_addr_t size, phys_addr_t align, phys_addr_t max_addr)
1164 alloc = __memblock_alloc_base(size, align, max_addr);
1167 panic("ERROR: Failed to allocate 0x%llx bytes below 0x%llx.\n",
1168 (unsigned long long) size, (unsigned long long) max_addr);
1173 phys_addr_t __init memblock_alloc(phys_addr_t size, phys_addr_t align)
1175 return memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ACCESSIBLE);
1178 phys_addr_t __init memblock_alloc_try_nid(phys_addr_t size, phys_addr_t align, int nid)
1180 phys_addr_t res = memblock_alloc_nid(size, align, nid);
1184 return memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ACCESSIBLE);
1188 * memblock_virt_alloc_internal - allocate boot memory block
1189 * @size: size of memory block to be allocated in bytes
1190 * @align: alignment of the region and block's size
1191 * @min_addr: the lower bound of the memory region to allocate (phys address)
1192 * @max_addr: the upper bound of the memory region to allocate (phys address)
1193 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1195 * The @min_addr limit is dropped if it can not be satisfied and the allocation
1196 * will fall back to memory below @min_addr. Also, allocation may fall back
1197 * to any node in the system if the specified node can not
1198 * hold the requested memory.
1200 * The allocation is performed from memory region limited by
1201 * memblock.current_limit if @max_addr == %BOOTMEM_ALLOC_ACCESSIBLE.
1203 * The memory block is aligned on SMP_CACHE_BYTES if @align == 0.
1205 * The phys address of allocated boot memory block is converted to virtual and
1206 * allocated memory is reset to 0.
1208 * In addition, function sets the min_count to 0 using kmemleak_alloc for
1209 * allocated boot memory block, so that it is never reported as leaks.
1212 * Virtual address of allocated memory block on success, NULL on failure.
1214 static void * __init memblock_virt_alloc_internal(
1215 phys_addr_t size, phys_addr_t align,
1216 phys_addr_t min_addr, phys_addr_t max_addr,
1221 ulong flags = choose_memblock_flags();
1223 if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
1227 * Detect any accidental use of these APIs after slab is ready, as at
1228 * this moment memblock may be deinitialized already and its
1229 * internal data may be destroyed (after execution of free_all_bootmem)
1231 if (WARN_ON_ONCE(slab_is_available()))
1232 return kzalloc_node(size, GFP_NOWAIT, nid);
1235 align = SMP_CACHE_BYTES;
1237 if (max_addr > memblock.current_limit)
1238 max_addr = memblock.current_limit;
1241 alloc = memblock_find_in_range_node(size, align, min_addr, max_addr,
1246 if (nid != NUMA_NO_NODE) {
1247 alloc = memblock_find_in_range_node(size, align, min_addr,
1248 max_addr, NUMA_NO_NODE,
1259 if (flags & MEMBLOCK_MIRROR) {
1260 flags &= ~MEMBLOCK_MIRROR;
1261 pr_warn("Could not allocate %pap bytes of mirrored memory\n",
1268 memblock_reserve(alloc, size);
1269 ptr = phys_to_virt(alloc);
1270 memset(ptr, 0, size);
1273 * The min_count is set to 0 so that bootmem allocated blocks
1274 * are never reported as leaks. This is because many of these blocks
1275 * are only referred via the physical address which is not
1276 * looked up by kmemleak.
1278 kmemleak_alloc(ptr, size, 0, 0);
1284 * memblock_virt_alloc_try_nid_nopanic - allocate boot memory block
1285 * @size: size of memory block to be allocated in bytes
1286 * @align: alignment of the region and block's size
1287 * @min_addr: the lower bound of the memory region from where the allocation
1288 * is preferred (phys address)
1289 * @max_addr: the upper bound of the memory region from where the allocation
1290 * is preferred (phys address), or %BOOTMEM_ALLOC_ACCESSIBLE to
1291 * allocate only from memory limited by memblock.current_limit value
1292 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1294 * Public version of _memblock_virt_alloc_try_nid_nopanic() which provides
1295 * additional debug information (including caller info), if enabled.
1298 * Virtual address of allocated memory block on success, NULL on failure.
1300 void * __init memblock_virt_alloc_try_nid_nopanic(
1301 phys_addr_t size, phys_addr_t align,
1302 phys_addr_t min_addr, phys_addr_t max_addr,
1305 memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=0x%llx max_addr=0x%llx %pF\n",
1306 __func__, (u64)size, (u64)align, nid, (u64)min_addr,
1307 (u64)max_addr, (void *)_RET_IP_);
1308 return memblock_virt_alloc_internal(size, align, min_addr,
1313 * memblock_virt_alloc_try_nid - allocate boot memory block with panicking
1314 * @size: size of memory block to be allocated in bytes
1315 * @align: alignment of the region and block's size
1316 * @min_addr: the lower bound of the memory region from where the allocation
1317 * is preferred (phys address)
1318 * @max_addr: the upper bound of the memory region from where the allocation
1319 * is preferred (phys address), or %BOOTMEM_ALLOC_ACCESSIBLE to
1320 * allocate only from memory limited by memblock.current_limit value
1321 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1323 * Public panicking version of _memblock_virt_alloc_try_nid_nopanic()
1324 * which provides debug information (including caller info), if enabled,
1325 * and panics if the request can not be satisfied.
1328 * Virtual address of allocated memory block on success, NULL on failure.
1330 void * __init memblock_virt_alloc_try_nid(
1331 phys_addr_t size, phys_addr_t align,
1332 phys_addr_t min_addr, phys_addr_t max_addr,
1337 memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=0x%llx max_addr=0x%llx %pF\n",
1338 __func__, (u64)size, (u64)align, nid, (u64)min_addr,
1339 (u64)max_addr, (void *)_RET_IP_);
1340 ptr = memblock_virt_alloc_internal(size, align,
1341 min_addr, max_addr, nid);
1345 panic("%s: Failed to allocate %llu bytes align=0x%llx nid=%d from=0x%llx max_addr=0x%llx\n",
1346 __func__, (u64)size, (u64)align, nid, (u64)min_addr,
1352 * __memblock_free_early - free boot memory block
1353 * @base: phys starting address of the boot memory block
1354 * @size: size of the boot memory block in bytes
1356 * Free boot memory block previously allocated by memblock_virt_alloc_xx() API.
1357 * The freeing memory will not be released to the buddy allocator.
1359 void __init __memblock_free_early(phys_addr_t base, phys_addr_t size)
1361 memblock_dbg("%s: [%#016llx-%#016llx] %pF\n",
1362 __func__, (u64)base, (u64)base + size - 1,
1364 kmemleak_free_part(__va(base), size);
1365 memblock_remove_range(&memblock.reserved, base, size);
1369 * __memblock_free_late - free bootmem block pages directly to buddy allocator
1370 * @addr: phys starting address of the boot memory block
1371 * @size: size of the boot memory block in bytes
1373 * This is only useful when the bootmem allocator has already been torn
1374 * down, but we are still initializing the system. Pages are released directly
1375 * to the buddy allocator, no bootmem metadata is updated because it is gone.
1377 void __init __memblock_free_late(phys_addr_t base, phys_addr_t size)
1381 memblock_dbg("%s: [%#016llx-%#016llx] %pF\n",
1382 __func__, (u64)base, (u64)base + size - 1,
1384 kmemleak_free_part(__va(base), size);
1385 cursor = PFN_UP(base);
1386 end = PFN_DOWN(base + size);
1388 for (; cursor < end; cursor++) {
1389 __free_pages_bootmem(pfn_to_page(cursor), cursor, 0);
1395 * Remaining API functions
1398 phys_addr_t __init memblock_phys_mem_size(void)
1400 return memblock.memory.total_size;
1403 phys_addr_t __init memblock_mem_size(unsigned long limit_pfn)
1405 unsigned long pages = 0;
1406 struct memblock_region *r;
1407 unsigned long start_pfn, end_pfn;
1409 for_each_memblock(memory, r) {
1410 start_pfn = memblock_region_memory_base_pfn(r);
1411 end_pfn = memblock_region_memory_end_pfn(r);
1412 start_pfn = min_t(unsigned long, start_pfn, limit_pfn);
1413 end_pfn = min_t(unsigned long, end_pfn, limit_pfn);
1414 pages += end_pfn - start_pfn;
1417 return PFN_PHYS(pages);
1420 /* lowest address */
1421 phys_addr_t __init_memblock memblock_start_of_DRAM(void)
1423 return memblock.memory.regions[0].base;
1426 phys_addr_t __init_memblock memblock_end_of_DRAM(void)
1428 int idx = memblock.memory.cnt - 1;
1430 return (memblock.memory.regions[idx].base + memblock.memory.regions[idx].size);
1433 void __init memblock_enforce_memory_limit(phys_addr_t limit)
1435 phys_addr_t max_addr = (phys_addr_t)ULLONG_MAX;
1436 struct memblock_region *r;
1441 /* find out max address */
1442 for_each_memblock(memory, r) {
1443 if (limit <= r->size) {
1444 max_addr = r->base + limit;
1450 /* truncate both memory and reserved regions */
1451 memblock_remove_range(&memblock.memory, max_addr,
1452 (phys_addr_t)ULLONG_MAX);
1453 memblock_remove_range(&memblock.reserved, max_addr,
1454 (phys_addr_t)ULLONG_MAX);
1457 static int __init_memblock memblock_search(struct memblock_type *type, phys_addr_t addr)
1459 unsigned int left = 0, right = type->cnt;
1462 unsigned int mid = (right + left) / 2;
1464 if (addr < type->regions[mid].base)
1466 else if (addr >= (type->regions[mid].base +
1467 type->regions[mid].size))
1471 } while (left < right);
1475 int __init memblock_is_reserved(phys_addr_t addr)
1477 return memblock_search(&memblock.reserved, addr) != -1;
1480 int __init_memblock memblock_is_memory(phys_addr_t addr)
1482 return memblock_search(&memblock.memory, addr) != -1;
1485 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
1486 int __init_memblock memblock_search_pfn_nid(unsigned long pfn,
1487 unsigned long *start_pfn, unsigned long *end_pfn)
1489 struct memblock_type *type = &memblock.memory;
1490 int mid = memblock_search(type, PFN_PHYS(pfn));
1495 *start_pfn = PFN_DOWN(type->regions[mid].base);
1496 *end_pfn = PFN_DOWN(type->regions[mid].base + type->regions[mid].size);
1498 return type->regions[mid].nid;
1503 * memblock_is_region_memory - check if a region is a subset of memory
1504 * @base: base of region to check
1505 * @size: size of region to check
1507 * Check if the region [@base, @base+@size) is a subset of a memory block.
1510 * 0 if false, non-zero if true
1512 int __init_memblock memblock_is_region_memory(phys_addr_t base, phys_addr_t size)
1514 int idx = memblock_search(&memblock.memory, base);
1515 phys_addr_t end = base + memblock_cap_size(base, &size);
1519 return memblock.memory.regions[idx].base <= base &&
1520 (memblock.memory.regions[idx].base +
1521 memblock.memory.regions[idx].size) >= end;
1525 * memblock_is_region_reserved - check if a region intersects reserved memory
1526 * @base: base of region to check
1527 * @size: size of region to check
1529 * Check if the region [@base, @base+@size) intersects a reserved memory block.
1532 * True if they intersect, false if not.
1534 bool __init_memblock memblock_is_region_reserved(phys_addr_t base, phys_addr_t size)
1536 memblock_cap_size(base, &size);
1537 return memblock_overlaps_region(&memblock.reserved, base, size);
1540 void __init_memblock memblock_trim_memory(phys_addr_t align)
1542 phys_addr_t start, end, orig_start, orig_end;
1543 struct memblock_region *r;
1545 for_each_memblock(memory, r) {
1546 orig_start = r->base;
1547 orig_end = r->base + r->size;
1548 start = round_up(orig_start, align);
1549 end = round_down(orig_end, align);
1551 if (start == orig_start && end == orig_end)
1556 r->size = end - start;
1558 memblock_remove_region(&memblock.memory,
1559 r - memblock.memory.regions);
1565 void __init_memblock memblock_set_current_limit(phys_addr_t limit)
1567 memblock.current_limit = limit;
1570 phys_addr_t __init_memblock memblock_get_current_limit(void)
1572 return memblock.current_limit;
1575 static void __init_memblock memblock_dump(struct memblock_type *type, char *name)
1577 unsigned long long base, size;
1578 unsigned long flags;
1581 pr_info(" %s.cnt = 0x%lx\n", name, type->cnt);
1583 for (i = 0; i < type->cnt; i++) {
1584 struct memblock_region *rgn = &type->regions[i];
1585 char nid_buf[32] = "";
1590 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
1591 if (memblock_get_region_node(rgn) != MAX_NUMNODES)
1592 snprintf(nid_buf, sizeof(nid_buf), " on node %d",
1593 memblock_get_region_node(rgn));
1595 pr_info(" %s[%#x]\t[%#016llx-%#016llx], %#llx bytes%s flags: %#lx\n",
1596 name, i, base, base + size - 1, size, nid_buf, flags);
1600 extern unsigned long __init_memblock
1601 memblock_reserved_memory_within(phys_addr_t start_addr, phys_addr_t end_addr)
1603 struct memblock_type *type = &memblock.reserved;
1604 unsigned long size = 0;
1607 for (idx = 0; idx < type->cnt; idx++) {
1608 struct memblock_region *rgn = &type->regions[idx];
1609 phys_addr_t start, end;
1611 if (rgn->base + rgn->size < start_addr)
1613 if (rgn->base > end_addr)
1617 end = start + rgn->size;
1618 size += end - start;
1624 void __init_memblock __memblock_dump_all(void)
1626 pr_info("MEMBLOCK configuration:\n");
1627 pr_info(" memory size = %#llx reserved size = %#llx\n",
1628 (unsigned long long)memblock.memory.total_size,
1629 (unsigned long long)memblock.reserved.total_size);
1631 memblock_dump(&memblock.memory, "memory");
1632 memblock_dump(&memblock.reserved, "reserved");
1635 void __init memblock_allow_resize(void)
1637 memblock_can_resize = 1;
1640 static int __init early_memblock(char *p)
1642 if (p && strstr(p, "debug"))
1646 early_param("memblock", early_memblock);
1648 #if defined(CONFIG_DEBUG_FS) && !defined(CONFIG_ARCH_DISCARD_MEMBLOCK)
1650 static int memblock_debug_show(struct seq_file *m, void *private)
1652 struct memblock_type *type = m->private;
1653 struct memblock_region *reg;
1656 for (i = 0; i < type->cnt; i++) {
1657 reg = &type->regions[i];
1658 seq_printf(m, "%4d: ", i);
1659 if (sizeof(phys_addr_t) == 4)
1660 seq_printf(m, "0x%08lx..0x%08lx\n",
1661 (unsigned long)reg->base,
1662 (unsigned long)(reg->base + reg->size - 1));
1664 seq_printf(m, "0x%016llx..0x%016llx\n",
1665 (unsigned long long)reg->base,
1666 (unsigned long long)(reg->base + reg->size - 1));
1672 static int memblock_debug_open(struct inode *inode, struct file *file)
1674 return single_open(file, memblock_debug_show, inode->i_private);
1677 static const struct file_operations memblock_debug_fops = {
1678 .open = memblock_debug_open,
1680 .llseek = seq_lseek,
1681 .release = single_release,
1684 static int __init memblock_init_debugfs(void)
1686 struct dentry *root = debugfs_create_dir("memblock", NULL);
1689 debugfs_create_file("memory", S_IRUGO, root, &memblock.memory, &memblock_debug_fops);
1690 debugfs_create_file("reserved", S_IRUGO, root, &memblock.reserved, &memblock_debug_fops);
1691 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
1692 debugfs_create_file("physmem", S_IRUGO, root, &memblock.physmem, &memblock_debug_fops);
1697 __initcall(memblock_init_debugfs);
1699 #endif /* CONFIG_DEBUG_FS */