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 int memblock_can_resize __initdata_memblock;
58 static int memblock_memory_in_slab __initdata_memblock = 0;
59 static int memblock_reserved_in_slab __initdata_memblock = 0;
61 /* inline so we don't get a warning when pr_debug is compiled out */
62 static __init_memblock const char *
63 memblock_type_name(struct memblock_type *type)
65 if (type == &memblock.memory)
67 else if (type == &memblock.reserved)
73 /* adjust *@size so that (@base + *@size) doesn't overflow, return new size */
74 static inline phys_addr_t memblock_cap_size(phys_addr_t base, phys_addr_t *size)
76 return *size = min(*size, (phys_addr_t)ULLONG_MAX - base);
80 * Address comparison utilities
82 static unsigned long __init_memblock memblock_addrs_overlap(phys_addr_t base1, phys_addr_t size1,
83 phys_addr_t base2, phys_addr_t size2)
85 return ((base1 < (base2 + size2)) && (base2 < (base1 + size1)));
88 static long __init_memblock memblock_overlaps_region(struct memblock_type *type,
89 phys_addr_t base, phys_addr_t size)
93 for (i = 0; i < type->cnt; i++) {
94 phys_addr_t rgnbase = type->regions[i].base;
95 phys_addr_t rgnsize = type->regions[i].size;
96 if (memblock_addrs_overlap(base, size, rgnbase, rgnsize))
100 return (i < type->cnt) ? i : -1;
104 * __memblock_find_range_bottom_up - find free area utility in bottom-up
105 * @start: start of candidate range
106 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_{ANYWHERE|ACCESSIBLE}
107 * @size: size of free area to find
108 * @align: alignment of free area to find
109 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
111 * Utility called from memblock_find_in_range_node(), find free area bottom-up.
114 * Found address on success, 0 on failure.
116 static phys_addr_t __init_memblock
117 __memblock_find_range_bottom_up(phys_addr_t start, phys_addr_t end,
118 phys_addr_t size, phys_addr_t align, int nid)
120 phys_addr_t this_start, this_end, cand;
123 for_each_free_mem_range(i, nid, &this_start, &this_end, NULL) {
124 this_start = clamp(this_start, start, end);
125 this_end = clamp(this_end, start, end);
127 cand = round_up(this_start, align);
128 if (cand < this_end && this_end - cand >= size)
136 * __memblock_find_range_top_down - find free area utility, in top-down
137 * @start: start of candidate range
138 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_{ANYWHERE|ACCESSIBLE}
139 * @size: size of free area to find
140 * @align: alignment of free area to find
141 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
143 * Utility called from memblock_find_in_range_node(), find free area top-down.
146 * Found address on success, 0 on failure.
148 static phys_addr_t __init_memblock
149 __memblock_find_range_top_down(phys_addr_t start, phys_addr_t end,
150 phys_addr_t size, phys_addr_t align, int nid)
152 phys_addr_t this_start, this_end, cand;
155 for_each_free_mem_range_reverse(i, nid, &this_start, &this_end, NULL) {
156 this_start = clamp(this_start, start, end);
157 this_end = clamp(this_end, start, end);
162 cand = round_down(this_end - size, align);
163 if (cand >= this_start)
171 * memblock_find_in_range_node - find free area in given range and node
172 * @size: size of free area to find
173 * @align: alignment of free area to find
174 * @start: start of candidate range
175 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_{ANYWHERE|ACCESSIBLE}
176 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
178 * Find @size free area aligned to @align in the specified range and node.
180 * When allocation direction is bottom-up, the @start should be greater
181 * than the end of the kernel image. Otherwise, it will be trimmed. The
182 * reason is that we want the bottom-up allocation just near the kernel
183 * image so it is highly likely that the allocated memory and the kernel
184 * will reside in the same node.
186 * If bottom-up allocation failed, will try to allocate memory top-down.
189 * Found address on success, 0 on failure.
191 phys_addr_t __init_memblock memblock_find_in_range_node(phys_addr_t size,
192 phys_addr_t align, phys_addr_t start,
193 phys_addr_t end, int nid)
195 phys_addr_t kernel_end, ret;
198 if (end == MEMBLOCK_ALLOC_ACCESSIBLE)
199 end = memblock.current_limit;
201 /* avoid allocating the first page */
202 start = max_t(phys_addr_t, start, PAGE_SIZE);
203 end = max(start, end);
204 kernel_end = __pa_symbol(_end);
207 * try bottom-up allocation only when bottom-up mode
208 * is set and @end is above the kernel image.
210 if (memblock_bottom_up() && end > kernel_end) {
211 phys_addr_t bottom_up_start;
213 /* make sure we will allocate above the kernel */
214 bottom_up_start = max(start, kernel_end);
216 /* ok, try bottom-up allocation first */
217 ret = __memblock_find_range_bottom_up(bottom_up_start, end,
223 * we always limit bottom-up allocation above the kernel,
224 * but top-down allocation doesn't have the limit, so
225 * retrying top-down allocation may succeed when bottom-up
228 * bottom-up allocation is expected to be fail very rarely,
229 * so we use WARN_ONCE() here to see the stack trace if
232 WARN_ONCE(1, "memblock: bottom-up allocation failed, "
233 "memory hotunplug may be affected\n");
236 return __memblock_find_range_top_down(start, end, size, align, nid);
240 * memblock_find_in_range - find free area in given range
241 * @start: start of candidate range
242 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_{ANYWHERE|ACCESSIBLE}
243 * @size: size of free area to find
244 * @align: alignment of free area to find
246 * Find @size free area aligned to @align in the specified range.
249 * Found address on success, 0 on failure.
251 phys_addr_t __init_memblock memblock_find_in_range(phys_addr_t start,
252 phys_addr_t end, phys_addr_t size,
255 return memblock_find_in_range_node(size, align, start, end,
259 static void __init_memblock memblock_remove_region(struct memblock_type *type, unsigned long r)
261 type->total_size -= type->regions[r].size;
262 memmove(&type->regions[r], &type->regions[r + 1],
263 (type->cnt - (r + 1)) * sizeof(type->regions[r]));
266 /* Special case for empty arrays */
267 if (type->cnt == 0) {
268 WARN_ON(type->total_size != 0);
270 type->regions[0].base = 0;
271 type->regions[0].size = 0;
272 type->regions[0].flags = 0;
273 memblock_set_region_node(&type->regions[0], MAX_NUMNODES);
277 #ifdef CONFIG_ARCH_DISCARD_MEMBLOCK
279 phys_addr_t __init_memblock get_allocated_memblock_reserved_regions_info(
282 if (memblock.reserved.regions == memblock_reserved_init_regions)
285 *addr = __pa(memblock.reserved.regions);
287 return PAGE_ALIGN(sizeof(struct memblock_region) *
288 memblock.reserved.max);
291 phys_addr_t __init_memblock get_allocated_memblock_memory_regions_info(
294 if (memblock.memory.regions == memblock_memory_init_regions)
297 *addr = __pa(memblock.memory.regions);
299 return PAGE_ALIGN(sizeof(struct memblock_region) *
300 memblock.memory.max);
306 * memblock_double_array - double the size of the memblock regions array
307 * @type: memblock type of the regions array being doubled
308 * @new_area_start: starting address of memory range to avoid overlap with
309 * @new_area_size: size of memory range to avoid overlap with
311 * Double the size of the @type regions array. If memblock is being used to
312 * allocate memory for a new reserved regions array and there is a previously
313 * allocated memory range [@new_area_start,@new_area_start+@new_area_size]
314 * waiting to be reserved, ensure the memory used by the new array does
318 * 0 on success, -1 on failure.
320 static int __init_memblock memblock_double_array(struct memblock_type *type,
321 phys_addr_t new_area_start,
322 phys_addr_t new_area_size)
324 struct memblock_region *new_array, *old_array;
325 phys_addr_t old_alloc_size, new_alloc_size;
326 phys_addr_t old_size, new_size, addr;
327 int use_slab = slab_is_available();
330 /* We don't allow resizing until we know about the reserved regions
331 * of memory that aren't suitable for allocation
333 if (!memblock_can_resize)
336 /* Calculate new doubled size */
337 old_size = type->max * sizeof(struct memblock_region);
338 new_size = old_size << 1;
340 * We need to allocated new one align to PAGE_SIZE,
341 * so we can free them completely later.
343 old_alloc_size = PAGE_ALIGN(old_size);
344 new_alloc_size = PAGE_ALIGN(new_size);
346 /* Retrieve the slab flag */
347 if (type == &memblock.memory)
348 in_slab = &memblock_memory_in_slab;
350 in_slab = &memblock_reserved_in_slab;
352 /* Try to find some space for it.
354 * WARNING: We assume that either slab_is_available() and we use it or
355 * we use MEMBLOCK for allocations. That means that this is unsafe to
356 * use when bootmem is currently active (unless bootmem itself is
357 * implemented on top of MEMBLOCK which isn't the case yet)
359 * This should however not be an issue for now, as we currently only
360 * call into MEMBLOCK while it's still active, or much later when slab
361 * is active for memory hotplug operations
364 new_array = kmalloc(new_size, GFP_KERNEL);
365 addr = new_array ? __pa(new_array) : 0;
367 /* only exclude range when trying to double reserved.regions */
368 if (type != &memblock.reserved)
369 new_area_start = new_area_size = 0;
371 addr = memblock_find_in_range(new_area_start + new_area_size,
372 memblock.current_limit,
373 new_alloc_size, PAGE_SIZE);
374 if (!addr && new_area_size)
375 addr = memblock_find_in_range(0,
376 min(new_area_start, memblock.current_limit),
377 new_alloc_size, PAGE_SIZE);
379 new_array = addr ? __va(addr) : NULL;
382 pr_err("memblock: Failed to double %s array from %ld to %ld entries !\n",
383 memblock_type_name(type), type->max, type->max * 2);
387 memblock_dbg("memblock: %s is doubled to %ld at [%#010llx-%#010llx]",
388 memblock_type_name(type), type->max * 2, (u64)addr,
389 (u64)addr + new_size - 1);
392 * Found space, we now need to move the array over before we add the
393 * reserved region since it may be our reserved array itself that is
396 memcpy(new_array, type->regions, old_size);
397 memset(new_array + type->max, 0, old_size);
398 old_array = type->regions;
399 type->regions = new_array;
402 /* Free old array. We needn't free it if the array is the static one */
405 else if (old_array != memblock_memory_init_regions &&
406 old_array != memblock_reserved_init_regions)
407 memblock_free(__pa(old_array), old_alloc_size);
410 * Reserve the new array if that comes from the memblock. Otherwise, we
414 BUG_ON(memblock_reserve(addr, new_alloc_size));
416 /* Update slab flag */
423 * memblock_merge_regions - merge neighboring compatible regions
424 * @type: memblock type to scan
426 * Scan @type and merge neighboring compatible regions.
428 static void __init_memblock memblock_merge_regions(struct memblock_type *type)
432 /* cnt never goes below 1 */
433 while (i < type->cnt - 1) {
434 struct memblock_region *this = &type->regions[i];
435 struct memblock_region *next = &type->regions[i + 1];
437 if (this->base + this->size != next->base ||
438 memblock_get_region_node(this) !=
439 memblock_get_region_node(next) ||
440 this->flags != next->flags) {
441 BUG_ON(this->base + this->size > next->base);
446 this->size += next->size;
447 /* move forward from next + 1, index of which is i + 2 */
448 memmove(next, next + 1, (type->cnt - (i + 2)) * sizeof(*next));
454 * memblock_insert_region - insert new memblock region
455 * @type: memblock type to insert into
456 * @idx: index for the insertion point
457 * @base: base address of the new region
458 * @size: size of the new region
459 * @nid: node id of the new region
460 * @flags: flags of the new region
462 * Insert new memblock region [@base,@base+@size) into @type at @idx.
463 * @type must already have extra room to accomodate the new region.
465 static void __init_memblock memblock_insert_region(struct memblock_type *type,
466 int idx, phys_addr_t base,
468 int nid, unsigned long flags)
470 struct memblock_region *rgn = &type->regions[idx];
472 BUG_ON(type->cnt >= type->max);
473 memmove(rgn + 1, rgn, (type->cnt - idx) * sizeof(*rgn));
477 memblock_set_region_node(rgn, nid);
479 type->total_size += size;
483 * memblock_add_range - add new memblock region
484 * @type: memblock type to add new region into
485 * @base: base address of the new region
486 * @size: size of the new region
487 * @nid: nid of the new region
488 * @flags: flags of the new region
490 * Add new memblock region [@base,@base+@size) into @type. The new region
491 * is allowed to overlap with existing ones - overlaps don't affect already
492 * existing regions. @type is guaranteed to be minimal (all neighbouring
493 * compatible regions are merged) after the addition.
496 * 0 on success, -errno on failure.
498 int __init_memblock memblock_add_range(struct memblock_type *type,
499 phys_addr_t base, phys_addr_t size,
500 int nid, unsigned long flags)
503 phys_addr_t obase = base;
504 phys_addr_t end = base + memblock_cap_size(base, &size);
510 /* special case for empty array */
511 if (type->regions[0].size == 0) {
512 WARN_ON(type->cnt != 1 || type->total_size);
513 type->regions[0].base = base;
514 type->regions[0].size = size;
515 type->regions[0].flags = flags;
516 memblock_set_region_node(&type->regions[0], nid);
517 type->total_size = size;
522 * The following is executed twice. Once with %false @insert and
523 * then with %true. The first counts the number of regions needed
524 * to accomodate the new area. The second actually inserts them.
529 for (i = 0; i < type->cnt; i++) {
530 struct memblock_region *rgn = &type->regions[i];
531 phys_addr_t rbase = rgn->base;
532 phys_addr_t rend = rbase + rgn->size;
539 * @rgn overlaps. If it separates the lower part of new
540 * area, insert that portion.
545 memblock_insert_region(type, i++, base,
549 /* area below @rend is dealt with, forget about it */
550 base = min(rend, end);
553 /* insert the remaining portion */
557 memblock_insert_region(type, i, base, end - base,
562 * If this was the first round, resize array and repeat for actual
563 * insertions; otherwise, merge and return.
566 while (type->cnt + nr_new > type->max)
567 if (memblock_double_array(type, obase, size) < 0)
572 memblock_merge_regions(type);
577 int __init_memblock memblock_add_node(phys_addr_t base, phys_addr_t size,
580 return memblock_add_range(&memblock.memory, base, size, nid, 0);
583 int __init_memblock memblock_add(phys_addr_t base, phys_addr_t size)
585 return memblock_add_range(&memblock.memory, base, size,
590 * memblock_isolate_range - isolate given range into disjoint memblocks
591 * @type: memblock type to isolate range for
592 * @base: base of range to isolate
593 * @size: size of range to isolate
594 * @start_rgn: out parameter for the start of isolated region
595 * @end_rgn: out parameter for the end of isolated region
597 * Walk @type and ensure that regions don't cross the boundaries defined by
598 * [@base,@base+@size). Crossing regions are split at the boundaries,
599 * which may create at most two more regions. The index of the first
600 * region inside the range is returned in *@start_rgn and end in *@end_rgn.
603 * 0 on success, -errno on failure.
605 static int __init_memblock memblock_isolate_range(struct memblock_type *type,
606 phys_addr_t base, phys_addr_t size,
607 int *start_rgn, int *end_rgn)
609 phys_addr_t end = base + memblock_cap_size(base, &size);
612 *start_rgn = *end_rgn = 0;
617 /* we'll create at most two more regions */
618 while (type->cnt + 2 > type->max)
619 if (memblock_double_array(type, base, size) < 0)
622 for (i = 0; i < type->cnt; i++) {
623 struct memblock_region *rgn = &type->regions[i];
624 phys_addr_t rbase = rgn->base;
625 phys_addr_t rend = rbase + rgn->size;
634 * @rgn intersects from below. Split and continue
635 * to process the next region - the new top half.
638 rgn->size -= base - rbase;
639 type->total_size -= base - rbase;
640 memblock_insert_region(type, i, rbase, base - rbase,
641 memblock_get_region_node(rgn),
643 } else if (rend > end) {
645 * @rgn intersects from above. Split and redo the
646 * current region - the new bottom half.
649 rgn->size -= end - rbase;
650 type->total_size -= end - rbase;
651 memblock_insert_region(type, i--, rbase, end - rbase,
652 memblock_get_region_node(rgn),
655 /* @rgn is fully contained, record it */
665 int __init_memblock memblock_remove_range(struct memblock_type *type,
666 phys_addr_t base, phys_addr_t size)
668 int start_rgn, end_rgn;
671 ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
675 for (i = end_rgn - 1; i >= start_rgn; i--)
676 memblock_remove_region(type, i);
680 int __init_memblock memblock_remove(phys_addr_t base, phys_addr_t size)
682 return memblock_remove_range(&memblock.memory, base, size);
686 int __init_memblock memblock_free(phys_addr_t base, phys_addr_t size)
688 memblock_dbg(" memblock_free: [%#016llx-%#016llx] %pF\n",
689 (unsigned long long)base,
690 (unsigned long long)base + size - 1,
693 kmemleak_free_part(__va(base), size);
694 return memblock_remove_range(&memblock.reserved, base, size);
697 static int __init_memblock memblock_reserve_region(phys_addr_t base,
702 struct memblock_type *_rgn = &memblock.reserved;
704 memblock_dbg("memblock_reserve: [%#016llx-%#016llx] flags %#02lx %pF\n",
705 (unsigned long long)base,
706 (unsigned long long)base + size - 1,
707 flags, (void *)_RET_IP_);
709 return memblock_add_range(_rgn, base, size, nid, flags);
712 int __init_memblock memblock_reserve(phys_addr_t base, phys_addr_t size)
714 return memblock_reserve_region(base, size, MAX_NUMNODES, 0);
718 * memblock_mark_hotplug - Mark hotpluggable memory with flag MEMBLOCK_HOTPLUG.
719 * @base: the base phys addr of the region
720 * @size: the size of the region
722 * This function isolates region [@base, @base + @size), and mark it with flag
725 * Return 0 on succees, -errno on failure.
727 int __init_memblock memblock_mark_hotplug(phys_addr_t base, phys_addr_t size)
729 struct memblock_type *type = &memblock.memory;
730 int i, ret, start_rgn, end_rgn;
732 ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
736 for (i = start_rgn; i < end_rgn; i++)
737 memblock_set_region_flags(&type->regions[i], MEMBLOCK_HOTPLUG);
739 memblock_merge_regions(type);
744 * memblock_clear_hotplug - Clear flag MEMBLOCK_HOTPLUG for a specified region.
745 * @base: the base phys addr of the region
746 * @size: the size of the region
748 * This function isolates region [@base, @base + @size), and clear flag
749 * MEMBLOCK_HOTPLUG for the isolated regions.
751 * Return 0 on succees, -errno on failure.
753 int __init_memblock memblock_clear_hotplug(phys_addr_t base, phys_addr_t size)
755 struct memblock_type *type = &memblock.memory;
756 int i, ret, start_rgn, end_rgn;
758 ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
762 for (i = start_rgn; i < end_rgn; i++)
763 memblock_clear_region_flags(&type->regions[i],
766 memblock_merge_regions(type);
771 * __next__mem_range - next function for for_each_free_mem_range() etc.
772 * @idx: pointer to u64 loop variable
773 * @nid: node selector, %NUMA_NO_NODE for all nodes
774 * @type_a: pointer to memblock_type from where the range is taken
775 * @type_b: pointer to memblock_type which excludes memory from being taken
776 * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL
777 * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL
778 * @out_nid: ptr to int for nid of the range, can be %NULL
780 * Find the first area from *@idx which matches @nid, fill the out
781 * parameters, and update *@idx for the next iteration. The lower 32bit of
782 * *@idx contains index into type_a and the upper 32bit indexes the
783 * areas before each region in type_b. For example, if type_b regions
784 * look like the following,
786 * 0:[0-16), 1:[32-48), 2:[128-130)
788 * The upper 32bit indexes the following regions.
790 * 0:[0-0), 1:[16-32), 2:[48-128), 3:[130-MAX)
792 * As both region arrays are sorted, the function advances the two indices
793 * in lockstep and returns each intersection.
795 void __init_memblock __next_mem_range(u64 *idx, int nid,
796 struct memblock_type *type_a,
797 struct memblock_type *type_b,
798 phys_addr_t *out_start,
799 phys_addr_t *out_end, int *out_nid)
801 int idx_a = *idx & 0xffffffff;
802 int idx_b = *idx >> 32;
804 if (WARN_ONCE(nid == MAX_NUMNODES,
805 "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
808 for (; idx_a < type_a->cnt; idx_a++) {
809 struct memblock_region *m = &type_a->regions[idx_a];
811 phys_addr_t m_start = m->base;
812 phys_addr_t m_end = m->base + m->size;
813 int m_nid = memblock_get_region_node(m);
815 /* only memory regions are associated with nodes, check it */
816 if (nid != NUMA_NO_NODE && nid != m_nid)
819 /* skip hotpluggable memory regions if needed */
820 if (movable_node_is_enabled() && memblock_is_hotpluggable(m))
825 *out_start = m_start;
831 *idx = (u32)idx_a | (u64)idx_b << 32;
835 /* scan areas before each reservation */
836 for (; idx_b < type_b->cnt + 1; idx_b++) {
837 struct memblock_region *r;
841 r = &type_b->regions[idx_b];
842 r_start = idx_b ? r[-1].base + r[-1].size : 0;
843 r_end = idx_b < type_b->cnt ?
844 r->base : ULLONG_MAX;
847 * if idx_b advanced past idx_a,
848 * break out to advance idx_a
850 if (r_start >= m_end)
852 /* if the two regions intersect, we're done */
853 if (m_start < r_end) {
856 max(m_start, r_start);
858 *out_end = min(m_end, r_end);
862 * The region which ends first is
863 * advanced for the next iteration.
869 *idx = (u32)idx_a | (u64)idx_b << 32;
875 /* signal end of iteration */
880 * __next_mem_range_rev - generic next function for for_each_*_range_rev()
882 * Finds the next range from type_a which is not marked as unsuitable
885 * @idx: pointer to u64 loop variable
886 * @nid: nid: node selector, %NUMA_NO_NODE for all nodes
887 * @type_a: pointer to memblock_type from where the range is taken
888 * @type_b: pointer to memblock_type which excludes memory from being taken
889 * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL
890 * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL
891 * @out_nid: ptr to int for nid of the range, can be %NULL
893 * Reverse of __next_mem_range().
895 void __init_memblock __next_mem_range_rev(u64 *idx, int nid,
896 struct memblock_type *type_a,
897 struct memblock_type *type_b,
898 phys_addr_t *out_start,
899 phys_addr_t *out_end, int *out_nid)
901 int idx_a = *idx & 0xffffffff;
902 int idx_b = *idx >> 32;
904 if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
907 if (*idx == (u64)ULLONG_MAX) {
908 idx_a = type_a->cnt - 1;
912 for (; idx_a >= 0; idx_a--) {
913 struct memblock_region *m = &type_a->regions[idx_a];
915 phys_addr_t m_start = m->base;
916 phys_addr_t m_end = m->base + m->size;
917 int m_nid = memblock_get_region_node(m);
919 /* only memory regions are associated with nodes, check it */
920 if (nid != NUMA_NO_NODE && nid != m_nid)
923 /* skip hotpluggable memory regions if needed */
924 if (movable_node_is_enabled() && memblock_is_hotpluggable(m))
929 *out_start = m_start;
935 *idx = (u32)idx_a | (u64)idx_b << 32;
939 /* scan areas before each reservation */
940 for (; idx_b >= 0; idx_b--) {
941 struct memblock_region *r;
945 r = &type_b->regions[idx_b];
946 r_start = idx_b ? r[-1].base + r[-1].size : 0;
947 r_end = idx_b < type_b->cnt ?
948 r->base : ULLONG_MAX;
950 * if idx_b advanced past idx_a,
951 * break out to advance idx_a
954 if (r_end <= m_start)
956 /* if the two regions intersect, we're done */
957 if (m_end > r_start) {
959 *out_start = max(m_start, r_start);
961 *out_end = min(m_end, r_end);
964 if (m_start >= r_start)
968 *idx = (u32)idx_a | (u64)idx_b << 32;
973 /* signal end of iteration */
977 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
979 * Common iterator interface used to define for_each_mem_range().
981 void __init_memblock __next_mem_pfn_range(int *idx, int nid,
982 unsigned long *out_start_pfn,
983 unsigned long *out_end_pfn, int *out_nid)
985 struct memblock_type *type = &memblock.memory;
986 struct memblock_region *r;
988 while (++*idx < type->cnt) {
989 r = &type->regions[*idx];
991 if (PFN_UP(r->base) >= PFN_DOWN(r->base + r->size))
993 if (nid == MAX_NUMNODES || nid == r->nid)
996 if (*idx >= type->cnt) {
1002 *out_start_pfn = PFN_UP(r->base);
1004 *out_end_pfn = PFN_DOWN(r->base + r->size);
1010 * memblock_set_node - set node ID on memblock regions
1011 * @base: base of area to set node ID for
1012 * @size: size of area to set node ID for
1013 * @type: memblock type to set node ID for
1014 * @nid: node ID to set
1016 * Set the nid of memblock @type regions in [@base,@base+@size) to @nid.
1017 * Regions which cross the area boundaries are split as necessary.
1020 * 0 on success, -errno on failure.
1022 int __init_memblock memblock_set_node(phys_addr_t base, phys_addr_t size,
1023 struct memblock_type *type, int nid)
1025 int start_rgn, end_rgn;
1028 ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
1032 for (i = start_rgn; i < end_rgn; i++)
1033 memblock_set_region_node(&type->regions[i], nid);
1035 memblock_merge_regions(type);
1038 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
1040 static phys_addr_t __init memblock_alloc_range_nid(phys_addr_t size,
1041 phys_addr_t align, phys_addr_t start,
1042 phys_addr_t end, int nid)
1047 align = SMP_CACHE_BYTES;
1049 found = memblock_find_in_range_node(size, align, start, end, nid);
1050 if (found && !memblock_reserve(found, size)) {
1052 * The min_count is set to 0 so that memblock allocations are
1053 * never reported as leaks.
1055 kmemleak_alloc(__va(found), size, 0, 0);
1061 phys_addr_t __init memblock_alloc_range(phys_addr_t size, phys_addr_t align,
1062 phys_addr_t start, phys_addr_t end)
1064 return memblock_alloc_range_nid(size, align, start, end, NUMA_NO_NODE);
1067 static phys_addr_t __init memblock_alloc_base_nid(phys_addr_t size,
1068 phys_addr_t align, phys_addr_t max_addr,
1071 return memblock_alloc_range_nid(size, align, 0, max_addr, nid);
1074 phys_addr_t __init memblock_alloc_nid(phys_addr_t size, phys_addr_t align, int nid)
1076 return memblock_alloc_base_nid(size, align, MEMBLOCK_ALLOC_ACCESSIBLE, nid);
1079 phys_addr_t __init __memblock_alloc_base(phys_addr_t size, phys_addr_t align, phys_addr_t max_addr)
1081 return memblock_alloc_base_nid(size, align, max_addr, NUMA_NO_NODE);
1084 phys_addr_t __init memblock_alloc_base(phys_addr_t size, phys_addr_t align, phys_addr_t max_addr)
1088 alloc = __memblock_alloc_base(size, align, max_addr);
1091 panic("ERROR: Failed to allocate 0x%llx bytes below 0x%llx.\n",
1092 (unsigned long long) size, (unsigned long long) max_addr);
1097 phys_addr_t __init memblock_alloc(phys_addr_t size, phys_addr_t align)
1099 return memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ACCESSIBLE);
1102 phys_addr_t __init memblock_alloc_try_nid(phys_addr_t size, phys_addr_t align, int nid)
1104 phys_addr_t res = memblock_alloc_nid(size, align, nid);
1108 return memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ACCESSIBLE);
1112 * memblock_virt_alloc_internal - allocate boot memory block
1113 * @size: size of memory block to be allocated in bytes
1114 * @align: alignment of the region and block's size
1115 * @min_addr: the lower bound of the memory region to allocate (phys address)
1116 * @max_addr: the upper bound of the memory region to allocate (phys address)
1117 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1119 * The @min_addr limit is dropped if it can not be satisfied and the allocation
1120 * will fall back to memory below @min_addr. Also, allocation may fall back
1121 * to any node in the system if the specified node can not
1122 * hold the requested memory.
1124 * The allocation is performed from memory region limited by
1125 * memblock.current_limit if @max_addr == %BOOTMEM_ALLOC_ACCESSIBLE.
1127 * The memory block is aligned on SMP_CACHE_BYTES if @align == 0.
1129 * The phys address of allocated boot memory block is converted to virtual and
1130 * allocated memory is reset to 0.
1132 * In addition, function sets the min_count to 0 using kmemleak_alloc for
1133 * allocated boot memory block, so that it is never reported as leaks.
1136 * Virtual address of allocated memory block on success, NULL on failure.
1138 static void * __init memblock_virt_alloc_internal(
1139 phys_addr_t size, phys_addr_t align,
1140 phys_addr_t min_addr, phys_addr_t max_addr,
1146 if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
1150 * Detect any accidental use of these APIs after slab is ready, as at
1151 * this moment memblock may be deinitialized already and its
1152 * internal data may be destroyed (after execution of free_all_bootmem)
1154 if (WARN_ON_ONCE(slab_is_available()))
1155 return kzalloc_node(size, GFP_NOWAIT, nid);
1158 align = SMP_CACHE_BYTES;
1160 if (max_addr > memblock.current_limit)
1161 max_addr = memblock.current_limit;
1164 alloc = memblock_find_in_range_node(size, align, min_addr, max_addr,
1169 if (nid != NUMA_NO_NODE) {
1170 alloc = memblock_find_in_range_node(size, align, min_addr,
1171 max_addr, NUMA_NO_NODE);
1184 memblock_reserve(alloc, size);
1185 ptr = phys_to_virt(alloc);
1186 memset(ptr, 0, size);
1189 * The min_count is set to 0 so that bootmem allocated blocks
1190 * are never reported as leaks. This is because many of these blocks
1191 * are only referred via the physical address which is not
1192 * looked up by kmemleak.
1194 kmemleak_alloc(ptr, size, 0, 0);
1203 * memblock_virt_alloc_try_nid_nopanic - allocate boot memory block
1204 * @size: size of memory block to be allocated in bytes
1205 * @align: alignment of the region and block's size
1206 * @min_addr: the lower bound of the memory region from where the allocation
1207 * is preferred (phys address)
1208 * @max_addr: the upper bound of the memory region from where the allocation
1209 * is preferred (phys address), or %BOOTMEM_ALLOC_ACCESSIBLE to
1210 * allocate only from memory limited by memblock.current_limit value
1211 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1213 * Public version of _memblock_virt_alloc_try_nid_nopanic() which provides
1214 * additional debug information (including caller info), if enabled.
1217 * Virtual address of allocated memory block on success, NULL on failure.
1219 void * __init memblock_virt_alloc_try_nid_nopanic(
1220 phys_addr_t size, phys_addr_t align,
1221 phys_addr_t min_addr, phys_addr_t max_addr,
1224 memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=0x%llx max_addr=0x%llx %pF\n",
1225 __func__, (u64)size, (u64)align, nid, (u64)min_addr,
1226 (u64)max_addr, (void *)_RET_IP_);
1227 return memblock_virt_alloc_internal(size, align, min_addr,
1232 * memblock_virt_alloc_try_nid - allocate boot memory block with panicking
1233 * @size: size of memory block to be allocated in bytes
1234 * @align: alignment of the region and block's size
1235 * @min_addr: the lower bound of the memory region from where the allocation
1236 * is preferred (phys address)
1237 * @max_addr: the upper bound of the memory region from where the allocation
1238 * is preferred (phys address), or %BOOTMEM_ALLOC_ACCESSIBLE to
1239 * allocate only from memory limited by memblock.current_limit value
1240 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1242 * Public panicking version of _memblock_virt_alloc_try_nid_nopanic()
1243 * which provides debug information (including caller info), if enabled,
1244 * and panics if the request can not be satisfied.
1247 * Virtual address of allocated memory block on success, NULL on failure.
1249 void * __init memblock_virt_alloc_try_nid(
1250 phys_addr_t size, phys_addr_t align,
1251 phys_addr_t min_addr, phys_addr_t max_addr,
1256 memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=0x%llx max_addr=0x%llx %pF\n",
1257 __func__, (u64)size, (u64)align, nid, (u64)min_addr,
1258 (u64)max_addr, (void *)_RET_IP_);
1259 ptr = memblock_virt_alloc_internal(size, align,
1260 min_addr, max_addr, nid);
1264 panic("%s: Failed to allocate %llu bytes align=0x%llx nid=%d from=0x%llx max_addr=0x%llx\n",
1265 __func__, (u64)size, (u64)align, nid, (u64)min_addr,
1271 * __memblock_free_early - free boot memory block
1272 * @base: phys starting address of the boot memory block
1273 * @size: size of the boot memory block in bytes
1275 * Free boot memory block previously allocated by memblock_virt_alloc_xx() API.
1276 * The freeing memory will not be released to the buddy allocator.
1278 void __init __memblock_free_early(phys_addr_t base, phys_addr_t size)
1280 memblock_dbg("%s: [%#016llx-%#016llx] %pF\n",
1281 __func__, (u64)base, (u64)base + size - 1,
1283 kmemleak_free_part(__va(base), size);
1284 memblock_remove_range(&memblock.reserved, base, size);
1288 * __memblock_free_late - free bootmem block pages directly to buddy allocator
1289 * @addr: phys starting address of the boot memory block
1290 * @size: size of the boot memory block in bytes
1292 * This is only useful when the bootmem allocator has already been torn
1293 * down, but we are still initializing the system. Pages are released directly
1294 * to the buddy allocator, no bootmem metadata is updated because it is gone.
1296 void __init __memblock_free_late(phys_addr_t base, phys_addr_t size)
1300 memblock_dbg("%s: [%#016llx-%#016llx] %pF\n",
1301 __func__, (u64)base, (u64)base + size - 1,
1303 kmemleak_free_part(__va(base), size);
1304 cursor = PFN_UP(base);
1305 end = PFN_DOWN(base + size);
1307 for (; cursor < end; cursor++) {
1308 __free_pages_bootmem(pfn_to_page(cursor), 0);
1314 * Remaining API functions
1317 phys_addr_t __init memblock_phys_mem_size(void)
1319 return memblock.memory.total_size;
1322 phys_addr_t __init memblock_mem_size(unsigned long limit_pfn)
1324 unsigned long pages = 0;
1325 struct memblock_region *r;
1326 unsigned long start_pfn, end_pfn;
1328 for_each_memblock(memory, r) {
1329 start_pfn = memblock_region_memory_base_pfn(r);
1330 end_pfn = memblock_region_memory_end_pfn(r);
1331 start_pfn = min_t(unsigned long, start_pfn, limit_pfn);
1332 end_pfn = min_t(unsigned long, end_pfn, limit_pfn);
1333 pages += end_pfn - start_pfn;
1336 return PFN_PHYS(pages);
1339 /* lowest address */
1340 phys_addr_t __init_memblock memblock_start_of_DRAM(void)
1342 return memblock.memory.regions[0].base;
1345 phys_addr_t __init_memblock memblock_end_of_DRAM(void)
1347 int idx = memblock.memory.cnt - 1;
1349 return (memblock.memory.regions[idx].base + memblock.memory.regions[idx].size);
1352 void __init memblock_enforce_memory_limit(phys_addr_t limit)
1354 phys_addr_t max_addr = (phys_addr_t)ULLONG_MAX;
1355 struct memblock_region *r;
1360 /* find out max address */
1361 for_each_memblock(memory, r) {
1362 if (limit <= r->size) {
1363 max_addr = r->base + limit;
1369 /* truncate both memory and reserved regions */
1370 memblock_remove_range(&memblock.memory, max_addr,
1371 (phys_addr_t)ULLONG_MAX);
1372 memblock_remove_range(&memblock.reserved, max_addr,
1373 (phys_addr_t)ULLONG_MAX);
1376 static int __init_memblock memblock_search(struct memblock_type *type, phys_addr_t addr)
1378 unsigned int left = 0, right = type->cnt;
1381 unsigned int mid = (right + left) / 2;
1383 if (addr < type->regions[mid].base)
1385 else if (addr >= (type->regions[mid].base +
1386 type->regions[mid].size))
1390 } while (left < right);
1394 int __init memblock_is_reserved(phys_addr_t addr)
1396 return memblock_search(&memblock.reserved, addr) != -1;
1399 int __init_memblock memblock_is_memory(phys_addr_t addr)
1401 return memblock_search(&memblock.memory, addr) != -1;
1404 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
1405 int __init_memblock memblock_search_pfn_nid(unsigned long pfn,
1406 unsigned long *start_pfn, unsigned long *end_pfn)
1408 struct memblock_type *type = &memblock.memory;
1409 int mid = memblock_search(type, PFN_PHYS(pfn));
1414 *start_pfn = PFN_DOWN(type->regions[mid].base);
1415 *end_pfn = PFN_DOWN(type->regions[mid].base + type->regions[mid].size);
1417 return type->regions[mid].nid;
1422 * memblock_is_region_memory - check if a region is a subset of memory
1423 * @base: base of region to check
1424 * @size: size of region to check
1426 * Check if the region [@base, @base+@size) is a subset of a memory block.
1429 * 0 if false, non-zero if true
1431 int __init_memblock memblock_is_region_memory(phys_addr_t base, phys_addr_t size)
1433 int idx = memblock_search(&memblock.memory, base);
1434 phys_addr_t end = base + memblock_cap_size(base, &size);
1438 return memblock.memory.regions[idx].base <= base &&
1439 (memblock.memory.regions[idx].base +
1440 memblock.memory.regions[idx].size) >= end;
1444 * memblock_is_region_reserved - check if a region intersects reserved memory
1445 * @base: base of region to check
1446 * @size: size of region to check
1448 * Check if the region [@base, @base+@size) intersects a reserved memory block.
1451 * 0 if false, non-zero if true
1453 int __init_memblock memblock_is_region_reserved(phys_addr_t base, phys_addr_t size)
1455 memblock_cap_size(base, &size);
1456 return memblock_overlaps_region(&memblock.reserved, base, size) >= 0;
1459 void __init_memblock memblock_trim_memory(phys_addr_t align)
1461 phys_addr_t start, end, orig_start, orig_end;
1462 struct memblock_region *r;
1464 for_each_memblock(memory, r) {
1465 orig_start = r->base;
1466 orig_end = r->base + r->size;
1467 start = round_up(orig_start, align);
1468 end = round_down(orig_end, align);
1470 if (start == orig_start && end == orig_end)
1475 r->size = end - start;
1477 memblock_remove_region(&memblock.memory,
1478 r - memblock.memory.regions);
1484 void __init_memblock memblock_set_current_limit(phys_addr_t limit)
1486 memblock.current_limit = limit;
1489 phys_addr_t __init_memblock memblock_get_current_limit(void)
1491 return memblock.current_limit;
1494 static void __init_memblock memblock_dump(struct memblock_type *type, char *name)
1496 unsigned long long base, size;
1497 unsigned long flags;
1500 pr_info(" %s.cnt = 0x%lx\n", name, type->cnt);
1502 for (i = 0; i < type->cnt; i++) {
1503 struct memblock_region *rgn = &type->regions[i];
1504 char nid_buf[32] = "";
1509 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
1510 if (memblock_get_region_node(rgn) != MAX_NUMNODES)
1511 snprintf(nid_buf, sizeof(nid_buf), " on node %d",
1512 memblock_get_region_node(rgn));
1514 pr_info(" %s[%#x]\t[%#016llx-%#016llx], %#llx bytes%s flags: %#lx\n",
1515 name, i, base, base + size - 1, size, nid_buf, flags);
1519 void __init_memblock __memblock_dump_all(void)
1521 pr_info("MEMBLOCK configuration:\n");
1522 pr_info(" memory size = %#llx reserved size = %#llx\n",
1523 (unsigned long long)memblock.memory.total_size,
1524 (unsigned long long)memblock.reserved.total_size);
1526 memblock_dump(&memblock.memory, "memory");
1527 memblock_dump(&memblock.reserved, "reserved");
1530 void __init memblock_allow_resize(void)
1532 memblock_can_resize = 1;
1535 static int __init early_memblock(char *p)
1537 if (p && strstr(p, "debug"))
1541 early_param("memblock", early_memblock);
1543 #if defined(CONFIG_DEBUG_FS) && !defined(CONFIG_ARCH_DISCARD_MEMBLOCK)
1545 static int memblock_debug_show(struct seq_file *m, void *private)
1547 struct memblock_type *type = m->private;
1548 struct memblock_region *reg;
1551 for (i = 0; i < type->cnt; i++) {
1552 reg = &type->regions[i];
1553 seq_printf(m, "%4d: ", i);
1554 if (sizeof(phys_addr_t) == 4)
1555 seq_printf(m, "0x%08lx..0x%08lx\n",
1556 (unsigned long)reg->base,
1557 (unsigned long)(reg->base + reg->size - 1));
1559 seq_printf(m, "0x%016llx..0x%016llx\n",
1560 (unsigned long long)reg->base,
1561 (unsigned long long)(reg->base + reg->size - 1));
1567 static int memblock_debug_open(struct inode *inode, struct file *file)
1569 return single_open(file, memblock_debug_show, inode->i_private);
1572 static const struct file_operations memblock_debug_fops = {
1573 .open = memblock_debug_open,
1575 .llseek = seq_lseek,
1576 .release = single_release,
1579 static int __init memblock_init_debugfs(void)
1581 struct dentry *root = debugfs_create_dir("memblock", NULL);
1584 debugfs_create_file("memory", S_IRUGO, root, &memblock.memory, &memblock_debug_fops);
1585 debugfs_create_file("reserved", S_IRUGO, root, &memblock.reserved, &memblock_debug_fops);
1586 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
1587 debugfs_create_file("physmem", S_IRUGO, root, &memblock.physmem, &memblock_debug_fops);
1592 __initcall(memblock_init_debugfs);
1594 #endif /* CONFIG_DEBUG_FS */