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;
31 struct memblock memblock __initdata_memblock = {
32 .memory.regions = memblock_memory_init_regions,
33 .memory.cnt = 1, /* empty dummy entry */
34 .memory.max = INIT_MEMBLOCK_REGIONS,
36 .reserved.regions = memblock_reserved_init_regions,
37 .reserved.cnt = 1, /* empty dummy entry */
38 .reserved.max = INIT_MEMBLOCK_REGIONS,
41 .current_limit = MEMBLOCK_ALLOC_ANYWHERE,
44 int memblock_debug __initdata_memblock;
45 #ifdef CONFIG_MOVABLE_NODE
46 bool movable_node_enabled __initdata_memblock = false;
48 static int memblock_can_resize __initdata_memblock;
49 static int memblock_memory_in_slab __initdata_memblock = 0;
50 static int memblock_reserved_in_slab __initdata_memblock = 0;
52 /* inline so we don't get a warning when pr_debug is compiled out */
53 static __init_memblock const char *
54 memblock_type_name(struct memblock_type *type)
56 if (type == &memblock.memory)
58 else if (type == &memblock.reserved)
64 /* adjust *@size so that (@base + *@size) doesn't overflow, return new size */
65 static inline phys_addr_t memblock_cap_size(phys_addr_t base, phys_addr_t *size)
67 return *size = min(*size, (phys_addr_t)ULLONG_MAX - base);
71 * Address comparison utilities
73 static unsigned long __init_memblock memblock_addrs_overlap(phys_addr_t base1, phys_addr_t size1,
74 phys_addr_t base2, phys_addr_t size2)
76 return ((base1 < (base2 + size2)) && (base2 < (base1 + size1)));
79 static long __init_memblock memblock_overlaps_region(struct memblock_type *type,
80 phys_addr_t base, phys_addr_t size)
84 for (i = 0; i < type->cnt; i++) {
85 phys_addr_t rgnbase = type->regions[i].base;
86 phys_addr_t rgnsize = type->regions[i].size;
87 if (memblock_addrs_overlap(base, size, rgnbase, rgnsize))
91 return (i < type->cnt) ? i : -1;
95 * __memblock_find_range_bottom_up - find free area utility in bottom-up
96 * @start: start of candidate range
97 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_{ANYWHERE|ACCESSIBLE}
98 * @size: size of free area to find
99 * @align: alignment of free area to find
100 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
102 * Utility called from memblock_find_in_range_node(), find free area bottom-up.
105 * Found address on success, 0 on failure.
107 static phys_addr_t __init_memblock
108 __memblock_find_range_bottom_up(phys_addr_t start, phys_addr_t end,
109 phys_addr_t size, phys_addr_t align, int nid)
111 phys_addr_t this_start, this_end, cand;
114 for_each_free_mem_range(i, nid, &this_start, &this_end, NULL) {
115 this_start = clamp(this_start, start, end);
116 this_end = clamp(this_end, start, end);
118 cand = round_up(this_start, align);
119 if (cand < this_end && this_end - cand >= size)
127 * __memblock_find_range_top_down - find free area utility, in top-down
128 * @start: start of candidate range
129 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_{ANYWHERE|ACCESSIBLE}
130 * @size: size of free area to find
131 * @align: alignment of free area to find
132 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
134 * Utility called from memblock_find_in_range_node(), find free area top-down.
137 * Found address on success, 0 on failure.
139 static phys_addr_t __init_memblock
140 __memblock_find_range_top_down(phys_addr_t start, phys_addr_t end,
141 phys_addr_t size, phys_addr_t align, int nid)
143 phys_addr_t this_start, this_end, cand;
146 for_each_free_mem_range_reverse(i, nid, &this_start, &this_end, NULL) {
147 this_start = clamp(this_start, start, end);
148 this_end = clamp(this_end, start, end);
153 cand = round_down(this_end - size, align);
154 if (cand >= this_start)
162 * memblock_find_in_range_node - find free area in given range and node
163 * @size: size of free area to find
164 * @align: alignment of free area to find
165 * @start: start of candidate range
166 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_{ANYWHERE|ACCESSIBLE}
167 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
169 * Find @size free area aligned to @align in the specified range and node.
171 * When allocation direction is bottom-up, the @start should be greater
172 * than the end of the kernel image. Otherwise, it will be trimmed. The
173 * reason is that we want the bottom-up allocation just near the kernel
174 * image so it is highly likely that the allocated memory and the kernel
175 * will reside in the same node.
177 * If bottom-up allocation failed, will try to allocate memory top-down.
180 * Found address on success, 0 on failure.
182 phys_addr_t __init_memblock memblock_find_in_range_node(phys_addr_t size,
183 phys_addr_t align, phys_addr_t start,
184 phys_addr_t end, int nid)
187 phys_addr_t kernel_end;
190 if (end == MEMBLOCK_ALLOC_ACCESSIBLE)
191 end = memblock.current_limit;
193 /* avoid allocating the first page */
194 start = max_t(phys_addr_t, start, PAGE_SIZE);
195 end = max(start, end);
196 kernel_end = __pa_symbol(_end);
199 * try bottom-up allocation only when bottom-up mode
200 * is set and @end is above the kernel image.
202 if (memblock_bottom_up() && end > kernel_end) {
203 phys_addr_t bottom_up_start;
205 /* make sure we will allocate above the kernel */
206 bottom_up_start = max(start, kernel_end);
208 /* ok, try bottom-up allocation first */
209 ret = __memblock_find_range_bottom_up(bottom_up_start, end,
215 * we always limit bottom-up allocation above the kernel,
216 * but top-down allocation doesn't have the limit, so
217 * retrying top-down allocation may succeed when bottom-up
220 * bottom-up allocation is expected to be fail very rarely,
221 * so we use WARN_ONCE() here to see the stack trace if
224 WARN_ONCE(1, "memblock: bottom-up allocation failed, "
225 "memory hotunplug may be affected\n");
228 return __memblock_find_range_top_down(start, end, size, align, nid);
232 * memblock_find_in_range - find free area in given range
233 * @start: start of candidate range
234 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_{ANYWHERE|ACCESSIBLE}
235 * @size: size of free area to find
236 * @align: alignment of free area to find
238 * Find @size free area aligned to @align in the specified range.
241 * Found address on success, 0 on failure.
243 phys_addr_t __init_memblock memblock_find_in_range(phys_addr_t start,
244 phys_addr_t end, phys_addr_t size,
247 return memblock_find_in_range_node(size, align, start, end,
251 static void __init_memblock memblock_remove_region(struct memblock_type *type, unsigned long r)
253 type->total_size -= type->regions[r].size;
254 memmove(&type->regions[r], &type->regions[r + 1],
255 (type->cnt - (r + 1)) * sizeof(type->regions[r]));
258 /* Special case for empty arrays */
259 if (type->cnt == 0) {
260 WARN_ON(type->total_size != 0);
262 type->regions[0].base = 0;
263 type->regions[0].size = 0;
264 type->regions[0].flags = 0;
265 memblock_set_region_node(&type->regions[0], MAX_NUMNODES);
269 phys_addr_t __init_memblock get_allocated_memblock_reserved_regions_info(
272 if (memblock.reserved.regions == memblock_reserved_init_regions)
276 * Don't allow nobootmem allocator to free reserved memory regions
278 * - CONFIG_DEBUG_FS is enabled;
279 * - CONFIG_ARCH_DISCARD_MEMBLOCK is not enabled;
280 * - reserved memory regions array have been resized during boot.
281 * Otherwise debug_fs entry "sys/kernel/debug/memblock/reserved"
282 * will show garbage instead of state of memory reservations.
284 if (IS_ENABLED(CONFIG_DEBUG_FS) &&
285 !IS_ENABLED(CONFIG_ARCH_DISCARD_MEMBLOCK))
288 *addr = __pa(memblock.reserved.regions);
290 return PAGE_ALIGN(sizeof(struct memblock_region) *
291 memblock.reserved.max);
294 #ifdef CONFIG_ARCH_DISCARD_MEMBLOCK
296 phys_addr_t __init_memblock get_allocated_memblock_memory_regions_info(
299 if (memblock.memory.regions == memblock_memory_init_regions)
302 *addr = __pa(memblock.memory.regions);
304 return PAGE_ALIGN(sizeof(struct memblock_region) *
305 memblock.memory.max);
311 * memblock_double_array - double the size of the memblock regions array
312 * @type: memblock type of the regions array being doubled
313 * @new_area_start: starting address of memory range to avoid overlap with
314 * @new_area_size: size of memory range to avoid overlap with
316 * Double the size of the @type regions array. If memblock is being used to
317 * allocate memory for a new reserved regions array and there is a previously
318 * allocated memory range [@new_area_start,@new_area_start+@new_area_size]
319 * waiting to be reserved, ensure the memory used by the new array does
323 * 0 on success, -1 on failure.
325 static int __init_memblock memblock_double_array(struct memblock_type *type,
326 phys_addr_t new_area_start,
327 phys_addr_t new_area_size)
329 struct memblock_region *new_array, *old_array;
330 phys_addr_t old_alloc_size, new_alloc_size;
331 phys_addr_t old_size, new_size, addr;
332 int use_slab = slab_is_available();
335 /* We don't allow resizing until we know about the reserved regions
336 * of memory that aren't suitable for allocation
338 if (!memblock_can_resize)
341 /* Calculate new doubled size */
342 old_size = type->max * sizeof(struct memblock_region);
343 new_size = old_size << 1;
345 * We need to allocated new one align to PAGE_SIZE,
346 * so we can free them completely later.
348 old_alloc_size = PAGE_ALIGN(old_size);
349 new_alloc_size = PAGE_ALIGN(new_size);
351 /* Retrieve the slab flag */
352 if (type == &memblock.memory)
353 in_slab = &memblock_memory_in_slab;
355 in_slab = &memblock_reserved_in_slab;
357 /* Try to find some space for it.
359 * WARNING: We assume that either slab_is_available() and we use it or
360 * we use MEMBLOCK for allocations. That means that this is unsafe to
361 * use when bootmem is currently active (unless bootmem itself is
362 * implemented on top of MEMBLOCK which isn't the case yet)
364 * This should however not be an issue for now, as we currently only
365 * call into MEMBLOCK while it's still active, or much later when slab
366 * is active for memory hotplug operations
369 new_array = kmalloc(new_size, GFP_KERNEL);
370 addr = new_array ? __pa(new_array) : 0;
372 /* only exclude range when trying to double reserved.regions */
373 if (type != &memblock.reserved)
374 new_area_start = new_area_size = 0;
376 addr = memblock_find_in_range(new_area_start + new_area_size,
377 memblock.current_limit,
378 new_alloc_size, PAGE_SIZE);
379 if (!addr && new_area_size)
380 addr = memblock_find_in_range(0,
381 min(new_area_start, memblock.current_limit),
382 new_alloc_size, PAGE_SIZE);
384 new_array = addr ? __va(addr) : NULL;
387 pr_err("memblock: Failed to double %s array from %ld to %ld entries !\n",
388 memblock_type_name(type), type->max, type->max * 2);
392 memblock_dbg("memblock: %s is doubled to %ld at [%#010llx-%#010llx]",
393 memblock_type_name(type), type->max * 2, (u64)addr,
394 (u64)addr + new_size - 1);
397 * Found space, we now need to move the array over before we add the
398 * reserved region since it may be our reserved array itself that is
401 memcpy(new_array, type->regions, old_size);
402 memset(new_array + type->max, 0, old_size);
403 old_array = type->regions;
404 type->regions = new_array;
407 /* Free old array. We needn't free it if the array is the static one */
410 else if (old_array != memblock_memory_init_regions &&
411 old_array != memblock_reserved_init_regions)
412 memblock_free(__pa(old_array), old_alloc_size);
415 * Reserve the new array if that comes from the memblock. Otherwise, we
419 BUG_ON(memblock_reserve(addr, new_alloc_size));
421 /* Update slab flag */
428 * memblock_merge_regions - merge neighboring compatible regions
429 * @type: memblock type to scan
431 * Scan @type and merge neighboring compatible regions.
433 static void __init_memblock memblock_merge_regions(struct memblock_type *type)
437 /* cnt never goes below 1 */
438 while (i < type->cnt - 1) {
439 struct memblock_region *this = &type->regions[i];
440 struct memblock_region *next = &type->regions[i + 1];
442 if (this->base + this->size != next->base ||
443 memblock_get_region_node(this) !=
444 memblock_get_region_node(next) ||
445 this->flags != next->flags) {
446 BUG_ON(this->base + this->size > next->base);
451 this->size += next->size;
452 /* move forward from next + 1, index of which is i + 2 */
453 memmove(next, next + 1, (type->cnt - (i + 2)) * sizeof(*next));
459 * memblock_insert_region - insert new memblock region
460 * @type: memblock type to insert into
461 * @idx: index for the insertion point
462 * @base: base address of the new region
463 * @size: size of the new region
464 * @nid: node id of the new region
465 * @flags: flags of the new region
467 * Insert new memblock region [@base,@base+@size) into @type at @idx.
468 * @type must already have extra room to accomodate the new region.
470 static void __init_memblock memblock_insert_region(struct memblock_type *type,
471 int idx, phys_addr_t base,
473 int nid, unsigned long flags)
475 struct memblock_region *rgn = &type->regions[idx];
477 BUG_ON(type->cnt >= type->max);
478 memmove(rgn + 1, rgn, (type->cnt - idx) * sizeof(*rgn));
482 memblock_set_region_node(rgn, nid);
484 type->total_size += size;
488 * memblock_add_region - add new memblock region
489 * @type: memblock type to add new region into
490 * @base: base address of the new region
491 * @size: size of the new region
492 * @nid: nid of the new region
493 * @flags: flags of the new region
495 * Add new memblock region [@base,@base+@size) into @type. The new region
496 * is allowed to overlap with existing ones - overlaps don't affect already
497 * existing regions. @type is guaranteed to be minimal (all neighbouring
498 * compatible regions are merged) after the addition.
501 * 0 on success, -errno on failure.
503 static int __init_memblock memblock_add_region(struct memblock_type *type,
504 phys_addr_t base, phys_addr_t size,
505 int nid, unsigned long flags)
508 phys_addr_t obase = base;
509 phys_addr_t end = base + memblock_cap_size(base, &size);
515 /* special case for empty array */
516 if (type->regions[0].size == 0) {
517 WARN_ON(type->cnt != 1 || type->total_size);
518 type->regions[0].base = base;
519 type->regions[0].size = size;
520 type->regions[0].flags = flags;
521 memblock_set_region_node(&type->regions[0], nid);
522 type->total_size = size;
527 * The following is executed twice. Once with %false @insert and
528 * then with %true. The first counts the number of regions needed
529 * to accomodate the new area. The second actually inserts them.
534 for (i = 0; i < type->cnt; i++) {
535 struct memblock_region *rgn = &type->regions[i];
536 phys_addr_t rbase = rgn->base;
537 phys_addr_t rend = rbase + rgn->size;
544 * @rgn overlaps. If it separates the lower part of new
545 * area, insert that portion.
550 memblock_insert_region(type, i++, base,
554 /* area below @rend is dealt with, forget about it */
555 base = min(rend, end);
558 /* insert the remaining portion */
562 memblock_insert_region(type, i, base, end - base,
567 * If this was the first round, resize array and repeat for actual
568 * insertions; otherwise, merge and return.
571 while (type->cnt + nr_new > type->max)
572 if (memblock_double_array(type, obase, size) < 0)
577 memblock_merge_regions(type);
582 int __init_memblock memblock_add_node(phys_addr_t base, phys_addr_t size,
585 return memblock_add_region(&memblock.memory, base, size, nid, 0);
588 int __init_memblock memblock_add(phys_addr_t base, phys_addr_t size)
590 return memblock_add_region(&memblock.memory, base, size,
595 * memblock_isolate_range - isolate given range into disjoint memblocks
596 * @type: memblock type to isolate range for
597 * @base: base of range to isolate
598 * @size: size of range to isolate
599 * @start_rgn: out parameter for the start of isolated region
600 * @end_rgn: out parameter for the end of isolated region
602 * Walk @type and ensure that regions don't cross the boundaries defined by
603 * [@base,@base+@size). Crossing regions are split at the boundaries,
604 * which may create at most two more regions. The index of the first
605 * region inside the range is returned in *@start_rgn and end in *@end_rgn.
608 * 0 on success, -errno on failure.
610 static int __init_memblock memblock_isolate_range(struct memblock_type *type,
611 phys_addr_t base, phys_addr_t size,
612 int *start_rgn, int *end_rgn)
614 phys_addr_t end = base + memblock_cap_size(base, &size);
617 *start_rgn = *end_rgn = 0;
622 /* we'll create at most two more regions */
623 while (type->cnt + 2 > type->max)
624 if (memblock_double_array(type, base, size) < 0)
627 for (i = 0; i < type->cnt; i++) {
628 struct memblock_region *rgn = &type->regions[i];
629 phys_addr_t rbase = rgn->base;
630 phys_addr_t rend = rbase + rgn->size;
639 * @rgn intersects from below. Split and continue
640 * to process the next region - the new top half.
643 rgn->size -= base - rbase;
644 type->total_size -= base - rbase;
645 memblock_insert_region(type, i, rbase, base - rbase,
646 memblock_get_region_node(rgn),
648 } else if (rend > end) {
650 * @rgn intersects from above. Split and redo the
651 * current region - the new bottom half.
654 rgn->size -= end - rbase;
655 type->total_size -= end - rbase;
656 memblock_insert_region(type, i--, rbase, end - rbase,
657 memblock_get_region_node(rgn),
660 /* @rgn is fully contained, record it */
670 static int __init_memblock __memblock_remove(struct memblock_type *type,
671 phys_addr_t base, phys_addr_t size)
673 int start_rgn, end_rgn;
676 ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
680 for (i = end_rgn - 1; i >= start_rgn; i--)
681 memblock_remove_region(type, i);
685 int __init_memblock memblock_remove(phys_addr_t base, phys_addr_t size)
687 return __memblock_remove(&memblock.memory, base, size);
690 int __init_memblock memblock_free(phys_addr_t base, phys_addr_t size)
692 memblock_dbg(" memblock_free: [%#016llx-%#016llx] %pF\n",
693 (unsigned long long)base,
694 (unsigned long long)base + size - 1,
697 return __memblock_remove(&memblock.reserved, base, size);
700 static int __init_memblock memblock_reserve_region(phys_addr_t base,
705 struct memblock_type *_rgn = &memblock.reserved;
707 memblock_dbg("memblock_reserve: [%#016llx-%#016llx] flags %#02lx %pF\n",
708 (unsigned long long)base,
709 (unsigned long long)base + size - 1,
710 flags, (void *)_RET_IP_);
712 return memblock_add_region(_rgn, base, size, nid, flags);
715 int __init_memblock memblock_reserve(phys_addr_t base, phys_addr_t size)
717 return memblock_reserve_region(base, size, MAX_NUMNODES, 0);
721 * memblock_mark_hotplug - Mark hotpluggable memory with flag MEMBLOCK_HOTPLUG.
722 * @base: the base phys addr of the region
723 * @size: the size of the region
725 * This function isolates region [@base, @base + @size), and mark it with flag
728 * Return 0 on succees, -errno on failure.
730 int __init_memblock memblock_mark_hotplug(phys_addr_t base, phys_addr_t size)
732 struct memblock_type *type = &memblock.memory;
733 int i, ret, start_rgn, end_rgn;
735 ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
739 for (i = start_rgn; i < end_rgn; i++)
740 memblock_set_region_flags(&type->regions[i], MEMBLOCK_HOTPLUG);
742 memblock_merge_regions(type);
747 * memblock_clear_hotplug - Clear flag MEMBLOCK_HOTPLUG for a specified region.
748 * @base: the base phys addr of the region
749 * @size: the size of the region
751 * This function isolates region [@base, @base + @size), and clear flag
752 * MEMBLOCK_HOTPLUG for the isolated regions.
754 * Return 0 on succees, -errno on failure.
756 int __init_memblock memblock_clear_hotplug(phys_addr_t base, phys_addr_t size)
758 struct memblock_type *type = &memblock.memory;
759 int i, ret, start_rgn, end_rgn;
761 ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
765 for (i = start_rgn; i < end_rgn; i++)
766 memblock_clear_region_flags(&type->regions[i],
769 memblock_merge_regions(type);
774 * __next_free_mem_range - next function for for_each_free_mem_range()
775 * @idx: pointer to u64 loop variable
776 * @nid: node selector, %NUMA_NO_NODE for all nodes
777 * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL
778 * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL
779 * @out_nid: ptr to int for nid of the range, can be %NULL
781 * Find the first free area from *@idx which matches @nid, fill the out
782 * parameters, and update *@idx for the next iteration. The lower 32bit of
783 * *@idx contains index into memory region and the upper 32bit indexes the
784 * areas before each reserved region. For example, if reserved regions
785 * look like the following,
787 * 0:[0-16), 1:[32-48), 2:[128-130)
789 * The upper 32bit indexes the following regions.
791 * 0:[0-0), 1:[16-32), 2:[48-128), 3:[130-MAX)
793 * As both region arrays are sorted, the function advances the two indices
794 * in lockstep and returns each intersection.
796 void __init_memblock __next_free_mem_range(u64 *idx, int nid,
797 phys_addr_t *out_start,
798 phys_addr_t *out_end, int *out_nid)
800 struct memblock_type *mem = &memblock.memory;
801 struct memblock_type *rsv = &memblock.reserved;
802 int mi = *idx & 0xffffffff;
805 if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
808 for ( ; mi < mem->cnt; mi++) {
809 struct memblock_region *m = &mem->regions[mi];
810 phys_addr_t m_start = m->base;
811 phys_addr_t m_end = m->base + m->size;
813 /* only memory regions are associated with nodes, check it */
814 if (nid != NUMA_NO_NODE && nid != memblock_get_region_node(m))
817 /* scan areas before each reservation for intersection */
818 for ( ; ri < rsv->cnt + 1; ri++) {
819 struct memblock_region *r = &rsv->regions[ri];
820 phys_addr_t r_start = ri ? r[-1].base + r[-1].size : 0;
821 phys_addr_t r_end = ri < rsv->cnt ? r->base : ULLONG_MAX;
823 /* if ri advanced past mi, break out to advance mi */
824 if (r_start >= m_end)
826 /* if the two regions intersect, we're done */
827 if (m_start < r_end) {
829 *out_start = max(m_start, r_start);
831 *out_end = min(m_end, r_end);
833 *out_nid = memblock_get_region_node(m);
835 * The region which ends first is advanced
836 * for the next iteration.
842 *idx = (u32)mi | (u64)ri << 32;
848 /* signal end of iteration */
853 * __next_free_mem_range_rev - next function for for_each_free_mem_range_reverse()
854 * @idx: pointer to u64 loop variable
855 * @nid: nid: node selector, %NUMA_NO_NODE for all nodes
856 * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL
857 * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL
858 * @out_nid: ptr to int for nid of the range, can be %NULL
860 * Reverse of __next_free_mem_range().
862 * Linux kernel cannot migrate pages used by itself. Memory hotplug users won't
863 * be able to hot-remove hotpluggable memory used by the kernel. So this
864 * function skip hotpluggable regions if needed when allocating memory for the
867 void __init_memblock __next_free_mem_range_rev(u64 *idx, int nid,
868 phys_addr_t *out_start,
869 phys_addr_t *out_end, int *out_nid)
871 struct memblock_type *mem = &memblock.memory;
872 struct memblock_type *rsv = &memblock.reserved;
873 int mi = *idx & 0xffffffff;
876 if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
879 if (*idx == (u64)ULLONG_MAX) {
884 for ( ; mi >= 0; mi--) {
885 struct memblock_region *m = &mem->regions[mi];
886 phys_addr_t m_start = m->base;
887 phys_addr_t m_end = m->base + m->size;
889 /* only memory regions are associated with nodes, check it */
890 if (nid != NUMA_NO_NODE && nid != memblock_get_region_node(m))
893 /* skip hotpluggable memory regions if needed */
894 if (movable_node_is_enabled() && memblock_is_hotpluggable(m))
897 /* scan areas before each reservation for intersection */
898 for ( ; ri >= 0; ri--) {
899 struct memblock_region *r = &rsv->regions[ri];
900 phys_addr_t r_start = ri ? r[-1].base + r[-1].size : 0;
901 phys_addr_t r_end = ri < rsv->cnt ? r->base : ULLONG_MAX;
903 /* if ri advanced past mi, break out to advance mi */
904 if (r_end <= m_start)
906 /* if the two regions intersect, we're done */
907 if (m_end > r_start) {
909 *out_start = max(m_start, r_start);
911 *out_end = min(m_end, r_end);
913 *out_nid = memblock_get_region_node(m);
915 if (m_start >= r_start)
919 *idx = (u32)mi | (u64)ri << 32;
928 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
930 * Common iterator interface used to define for_each_mem_range().
932 void __init_memblock __next_mem_pfn_range(int *idx, int nid,
933 unsigned long *out_start_pfn,
934 unsigned long *out_end_pfn, int *out_nid)
936 struct memblock_type *type = &memblock.memory;
937 struct memblock_region *r;
939 while (++*idx < type->cnt) {
940 r = &type->regions[*idx];
942 if (PFN_UP(r->base) >= PFN_DOWN(r->base + r->size))
944 if (nid == MAX_NUMNODES || nid == r->nid)
947 if (*idx >= type->cnt) {
953 *out_start_pfn = PFN_UP(r->base);
955 *out_end_pfn = PFN_DOWN(r->base + r->size);
961 * memblock_set_node - set node ID on memblock regions
962 * @base: base of area to set node ID for
963 * @size: size of area to set node ID for
964 * @type: memblock type to set node ID for
965 * @nid: node ID to set
967 * Set the nid of memblock @type regions in [@base,@base+@size) to @nid.
968 * Regions which cross the area boundaries are split as necessary.
971 * 0 on success, -errno on failure.
973 int __init_memblock memblock_set_node(phys_addr_t base, phys_addr_t size,
974 struct memblock_type *type, int nid)
976 int start_rgn, end_rgn;
979 ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
983 for (i = start_rgn; i < end_rgn; i++)
984 memblock_set_region_node(&type->regions[i], nid);
986 memblock_merge_regions(type);
989 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
991 static phys_addr_t __init memblock_alloc_base_nid(phys_addr_t size,
992 phys_addr_t align, phys_addr_t max_addr,
998 align = SMP_CACHE_BYTES;
1000 /* align @size to avoid excessive fragmentation on reserved array */
1001 size = round_up(size, align);
1003 found = memblock_find_in_range_node(size, align, 0, max_addr, nid);
1004 if (found && !memblock_reserve(found, size))
1010 phys_addr_t __init memblock_alloc_nid(phys_addr_t size, phys_addr_t align, int nid)
1012 return memblock_alloc_base_nid(size, align, MEMBLOCK_ALLOC_ACCESSIBLE, nid);
1015 phys_addr_t __init __memblock_alloc_base(phys_addr_t size, phys_addr_t align, phys_addr_t max_addr)
1017 return memblock_alloc_base_nid(size, align, max_addr, NUMA_NO_NODE);
1020 phys_addr_t __init memblock_alloc_base(phys_addr_t size, phys_addr_t align, phys_addr_t max_addr)
1024 alloc = __memblock_alloc_base(size, align, max_addr);
1027 panic("ERROR: Failed to allocate 0x%llx bytes below 0x%llx.\n",
1028 (unsigned long long) size, (unsigned long long) max_addr);
1033 phys_addr_t __init memblock_alloc(phys_addr_t size, phys_addr_t align)
1035 return memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ACCESSIBLE);
1038 phys_addr_t __init memblock_alloc_try_nid(phys_addr_t size, phys_addr_t align, int nid)
1040 phys_addr_t res = memblock_alloc_nid(size, align, nid);
1044 return memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ACCESSIBLE);
1048 * memblock_virt_alloc_internal - allocate boot memory block
1049 * @size: size of memory block to be allocated in bytes
1050 * @align: alignment of the region and block's size
1051 * @min_addr: the lower bound of the memory region to allocate (phys address)
1052 * @max_addr: the upper bound of the memory region to allocate (phys address)
1053 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1055 * The @min_addr limit is dropped if it can not be satisfied and the allocation
1056 * will fall back to memory below @min_addr. Also, allocation may fall back
1057 * to any node in the system if the specified node can not
1058 * hold the requested memory.
1060 * The allocation is performed from memory region limited by
1061 * memblock.current_limit if @max_addr == %BOOTMEM_ALLOC_ACCESSIBLE.
1063 * The memory block is aligned on SMP_CACHE_BYTES if @align == 0.
1065 * The phys address of allocated boot memory block is converted to virtual and
1066 * allocated memory is reset to 0.
1068 * In addition, function sets the min_count to 0 using kmemleak_alloc for
1069 * allocated boot memory block, so that it is never reported as leaks.
1072 * Virtual address of allocated memory block on success, NULL on failure.
1074 static void * __init memblock_virt_alloc_internal(
1075 phys_addr_t size, phys_addr_t align,
1076 phys_addr_t min_addr, phys_addr_t max_addr,
1082 if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
1086 * Detect any accidental use of these APIs after slab is ready, as at
1087 * this moment memblock may be deinitialized already and its
1088 * internal data may be destroyed (after execution of free_all_bootmem)
1090 if (WARN_ON_ONCE(slab_is_available()))
1091 return kzalloc_node(size, GFP_NOWAIT, nid);
1094 align = SMP_CACHE_BYTES;
1096 /* align @size to avoid excessive fragmentation on reserved array */
1097 size = round_up(size, align);
1100 alloc = memblock_find_in_range_node(size, align, min_addr, max_addr,
1105 if (nid != NUMA_NO_NODE) {
1106 alloc = memblock_find_in_range_node(size, align, min_addr,
1107 max_addr, NUMA_NO_NODE);
1120 memblock_reserve(alloc, size);
1121 ptr = phys_to_virt(alloc);
1122 memset(ptr, 0, size);
1125 * The min_count is set to 0 so that bootmem allocated blocks
1126 * are never reported as leaks. This is because many of these blocks
1127 * are only referred via the physical address which is not
1128 * looked up by kmemleak.
1130 kmemleak_alloc(ptr, size, 0, 0);
1139 * memblock_virt_alloc_try_nid_nopanic - allocate boot memory block
1140 * @size: size of memory block to be allocated in bytes
1141 * @align: alignment of the region and block's size
1142 * @min_addr: the lower bound of the memory region from where the allocation
1143 * is preferred (phys address)
1144 * @max_addr: the upper bound of the memory region from where the allocation
1145 * is preferred (phys address), or %BOOTMEM_ALLOC_ACCESSIBLE to
1146 * allocate only from memory limited by memblock.current_limit value
1147 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1149 * Public version of _memblock_virt_alloc_try_nid_nopanic() which provides
1150 * additional debug information (including caller info), if enabled.
1153 * Virtual address of allocated memory block on success, NULL on failure.
1155 void * __init memblock_virt_alloc_try_nid_nopanic(
1156 phys_addr_t size, phys_addr_t align,
1157 phys_addr_t min_addr, phys_addr_t max_addr,
1160 memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=0x%llx max_addr=0x%llx %pF\n",
1161 __func__, (u64)size, (u64)align, nid, (u64)min_addr,
1162 (u64)max_addr, (void *)_RET_IP_);
1163 return memblock_virt_alloc_internal(size, align, min_addr,
1168 * memblock_virt_alloc_try_nid - allocate boot memory block with panicking
1169 * @size: size of memory block to be allocated in bytes
1170 * @align: alignment of the region and block's size
1171 * @min_addr: the lower bound of the memory region from where the allocation
1172 * is preferred (phys address)
1173 * @max_addr: the upper bound of the memory region from where the allocation
1174 * is preferred (phys address), or %BOOTMEM_ALLOC_ACCESSIBLE to
1175 * allocate only from memory limited by memblock.current_limit value
1176 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1178 * Public panicking version of _memblock_virt_alloc_try_nid_nopanic()
1179 * which provides debug information (including caller info), if enabled,
1180 * and panics if the request can not be satisfied.
1183 * Virtual address of allocated memory block on success, NULL on failure.
1185 void * __init memblock_virt_alloc_try_nid(
1186 phys_addr_t size, phys_addr_t align,
1187 phys_addr_t min_addr, phys_addr_t max_addr,
1192 memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=0x%llx max_addr=0x%llx %pF\n",
1193 __func__, (u64)size, (u64)align, nid, (u64)min_addr,
1194 (u64)max_addr, (void *)_RET_IP_);
1195 ptr = memblock_virt_alloc_internal(size, align,
1196 min_addr, max_addr, nid);
1200 panic("%s: Failed to allocate %llu bytes align=0x%llx nid=%d from=0x%llx max_addr=0x%llx\n",
1201 __func__, (u64)size, (u64)align, nid, (u64)min_addr,
1207 * __memblock_free_early - free boot memory block
1208 * @base: phys starting address of the boot memory block
1209 * @size: size of the boot memory block in bytes
1211 * Free boot memory block previously allocated by memblock_virt_alloc_xx() API.
1212 * The freeing memory will not be released to the buddy allocator.
1214 void __init __memblock_free_early(phys_addr_t base, phys_addr_t size)
1216 memblock_dbg("%s: [%#016llx-%#016llx] %pF\n",
1217 __func__, (u64)base, (u64)base + size - 1,
1219 kmemleak_free_part(__va(base), size);
1220 __memblock_remove(&memblock.reserved, base, size);
1224 * __memblock_free_late - free bootmem block pages directly to buddy allocator
1225 * @addr: phys starting address of the boot memory block
1226 * @size: size of the boot memory block in bytes
1228 * This is only useful when the bootmem allocator has already been torn
1229 * down, but we are still initializing the system. Pages are released directly
1230 * to the buddy allocator, no bootmem metadata is updated because it is gone.
1232 void __init __memblock_free_late(phys_addr_t base, phys_addr_t size)
1236 memblock_dbg("%s: [%#016llx-%#016llx] %pF\n",
1237 __func__, (u64)base, (u64)base + size - 1,
1239 kmemleak_free_part(__va(base), size);
1240 cursor = PFN_UP(base);
1241 end = PFN_DOWN(base + size);
1243 for (; cursor < end; cursor++) {
1244 __free_pages_bootmem(pfn_to_page(cursor), 0);
1250 * Remaining API functions
1253 phys_addr_t __init memblock_phys_mem_size(void)
1255 return memblock.memory.total_size;
1258 phys_addr_t __init memblock_mem_size(unsigned long limit_pfn)
1260 unsigned long pages = 0;
1261 struct memblock_region *r;
1262 unsigned long start_pfn, end_pfn;
1264 for_each_memblock(memory, r) {
1265 start_pfn = memblock_region_memory_base_pfn(r);
1266 end_pfn = memblock_region_memory_end_pfn(r);
1267 start_pfn = min_t(unsigned long, start_pfn, limit_pfn);
1268 end_pfn = min_t(unsigned long, end_pfn, limit_pfn);
1269 pages += end_pfn - start_pfn;
1272 return (phys_addr_t)pages << PAGE_SHIFT;
1275 /* lowest address */
1276 phys_addr_t __init_memblock memblock_start_of_DRAM(void)
1278 return memblock.memory.regions[0].base;
1281 phys_addr_t __init_memblock memblock_end_of_DRAM(void)
1283 int idx = memblock.memory.cnt - 1;
1285 return (memblock.memory.regions[idx].base + memblock.memory.regions[idx].size);
1288 void __init memblock_enforce_memory_limit(phys_addr_t limit)
1291 phys_addr_t max_addr = (phys_addr_t)ULLONG_MAX;
1296 /* find out max address */
1297 for (i = 0; i < memblock.memory.cnt; i++) {
1298 struct memblock_region *r = &memblock.memory.regions[i];
1300 if (limit <= r->size) {
1301 max_addr = r->base + limit;
1307 /* truncate both memory and reserved regions */
1308 __memblock_remove(&memblock.memory, max_addr, (phys_addr_t)ULLONG_MAX);
1309 __memblock_remove(&memblock.reserved, max_addr, (phys_addr_t)ULLONG_MAX);
1312 static int __init_memblock memblock_search(struct memblock_type *type, phys_addr_t addr)
1314 unsigned int left = 0, right = type->cnt;
1317 unsigned int mid = (right + left) / 2;
1319 if (addr < type->regions[mid].base)
1321 else if (addr >= (type->regions[mid].base +
1322 type->regions[mid].size))
1326 } while (left < right);
1330 int __init memblock_is_reserved(phys_addr_t addr)
1332 return memblock_search(&memblock.reserved, addr) != -1;
1335 int __init_memblock memblock_is_memory(phys_addr_t addr)
1337 return memblock_search(&memblock.memory, addr) != -1;
1340 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
1341 int __init_memblock memblock_search_pfn_nid(unsigned long pfn,
1342 unsigned long *start_pfn, unsigned long *end_pfn)
1344 struct memblock_type *type = &memblock.memory;
1345 int mid = memblock_search(type, (phys_addr_t)pfn << PAGE_SHIFT);
1350 *start_pfn = type->regions[mid].base >> PAGE_SHIFT;
1351 *end_pfn = (type->regions[mid].base + type->regions[mid].size)
1354 return type->regions[mid].nid;
1359 * memblock_is_region_memory - check if a region is a subset of memory
1360 * @base: base of region to check
1361 * @size: size of region to check
1363 * Check if the region [@base, @base+@size) is a subset of a memory block.
1366 * 0 if false, non-zero if true
1368 int __init_memblock memblock_is_region_memory(phys_addr_t base, phys_addr_t size)
1370 int idx = memblock_search(&memblock.memory, base);
1371 phys_addr_t end = base + memblock_cap_size(base, &size);
1375 return memblock.memory.regions[idx].base <= base &&
1376 (memblock.memory.regions[idx].base +
1377 memblock.memory.regions[idx].size) >= end;
1381 * memblock_is_region_reserved - check if a region intersects reserved memory
1382 * @base: base of region to check
1383 * @size: size of region to check
1385 * Check if the region [@base, @base+@size) intersects a reserved memory block.
1388 * 0 if false, non-zero if true
1390 int __init_memblock memblock_is_region_reserved(phys_addr_t base, phys_addr_t size)
1392 memblock_cap_size(base, &size);
1393 return memblock_overlaps_region(&memblock.reserved, base, size) >= 0;
1396 void __init_memblock memblock_trim_memory(phys_addr_t align)
1399 phys_addr_t start, end, orig_start, orig_end;
1400 struct memblock_type *mem = &memblock.memory;
1402 for (i = 0; i < mem->cnt; i++) {
1403 orig_start = mem->regions[i].base;
1404 orig_end = mem->regions[i].base + mem->regions[i].size;
1405 start = round_up(orig_start, align);
1406 end = round_down(orig_end, align);
1408 if (start == orig_start && end == orig_end)
1412 mem->regions[i].base = start;
1413 mem->regions[i].size = end - start;
1415 memblock_remove_region(mem, i);
1421 void __init_memblock memblock_set_current_limit(phys_addr_t limit)
1423 memblock.current_limit = limit;
1426 static void __init_memblock memblock_dump(struct memblock_type *type, char *name)
1428 unsigned long long base, size;
1429 unsigned long flags;
1432 pr_info(" %s.cnt = 0x%lx\n", name, type->cnt);
1434 for (i = 0; i < type->cnt; i++) {
1435 struct memblock_region *rgn = &type->regions[i];
1436 char nid_buf[32] = "";
1441 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
1442 if (memblock_get_region_node(rgn) != MAX_NUMNODES)
1443 snprintf(nid_buf, sizeof(nid_buf), " on node %d",
1444 memblock_get_region_node(rgn));
1446 pr_info(" %s[%#x]\t[%#016llx-%#016llx], %#llx bytes%s flags: %#lx\n",
1447 name, i, base, base + size - 1, size, nid_buf, flags);
1451 void __init_memblock __memblock_dump_all(void)
1453 pr_info("MEMBLOCK configuration:\n");
1454 pr_info(" memory size = %#llx reserved size = %#llx\n",
1455 (unsigned long long)memblock.memory.total_size,
1456 (unsigned long long)memblock.reserved.total_size);
1458 memblock_dump(&memblock.memory, "memory");
1459 memblock_dump(&memblock.reserved, "reserved");
1462 void __init memblock_allow_resize(void)
1464 memblock_can_resize = 1;
1467 static int __init early_memblock(char *p)
1469 if (p && strstr(p, "debug"))
1473 early_param("memblock", early_memblock);
1475 #if defined(CONFIG_DEBUG_FS) && !defined(CONFIG_ARCH_DISCARD_MEMBLOCK)
1477 static int memblock_debug_show(struct seq_file *m, void *private)
1479 struct memblock_type *type = m->private;
1480 struct memblock_region *reg;
1483 for (i = 0; i < type->cnt; i++) {
1484 reg = &type->regions[i];
1485 seq_printf(m, "%4d: ", i);
1486 if (sizeof(phys_addr_t) == 4)
1487 seq_printf(m, "0x%08lx..0x%08lx\n",
1488 (unsigned long)reg->base,
1489 (unsigned long)(reg->base + reg->size - 1));
1491 seq_printf(m, "0x%016llx..0x%016llx\n",
1492 (unsigned long long)reg->base,
1493 (unsigned long long)(reg->base + reg->size - 1));
1499 static int memblock_debug_open(struct inode *inode, struct file *file)
1501 return single_open(file, memblock_debug_show, inode->i_private);
1504 static const struct file_operations memblock_debug_fops = {
1505 .open = memblock_debug_open,
1507 .llseek = seq_lseek,
1508 .release = single_release,
1511 static int __init memblock_init_debugfs(void)
1513 struct dentry *root = debugfs_create_dir("memblock", NULL);
1516 debugfs_create_file("memory", S_IRUGO, root, &memblock.memory, &memblock_debug_fops);
1517 debugfs_create_file("reserved", S_IRUGO, root, &memblock.reserved, &memblock_debug_fops);
1521 __initcall(memblock_init_debugfs);
1523 #endif /* CONFIG_DEBUG_FS */