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)
186 phys_addr_t kernel_end, ret;
189 if (end == MEMBLOCK_ALLOC_ACCESSIBLE)
190 end = memblock.current_limit;
192 /* avoid allocating the first page */
193 start = max_t(phys_addr_t, start, PAGE_SIZE);
194 end = max(start, end);
195 kernel_end = __pa_symbol(_end);
198 * try bottom-up allocation only when bottom-up mode
199 * is set and @end is above the kernel image.
201 if (memblock_bottom_up() && end > kernel_end) {
202 phys_addr_t bottom_up_start;
204 /* make sure we will allocate above the kernel */
205 bottom_up_start = max(start, kernel_end);
207 /* ok, try bottom-up allocation first */
208 ret = __memblock_find_range_bottom_up(bottom_up_start, end,
214 * we always limit bottom-up allocation above the kernel,
215 * but top-down allocation doesn't have the limit, so
216 * retrying top-down allocation may succeed when bottom-up
219 * bottom-up allocation is expected to be fail very rarely,
220 * so we use WARN_ONCE() here to see the stack trace if
223 WARN_ONCE(1, "memblock: bottom-up allocation failed, "
224 "memory hotunplug may be affected\n");
227 return __memblock_find_range_top_down(start, end, size, align, nid);
231 * memblock_find_in_range - find free area in given range
232 * @start: start of candidate range
233 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_{ANYWHERE|ACCESSIBLE}
234 * @size: size of free area to find
235 * @align: alignment of free area to find
237 * Find @size free area aligned to @align in the specified range.
240 * Found address on success, 0 on failure.
242 phys_addr_t __init_memblock memblock_find_in_range(phys_addr_t start,
243 phys_addr_t end, phys_addr_t size,
246 return memblock_find_in_range_node(size, align, start, end,
250 static void __init_memblock memblock_remove_region(struct memblock_type *type, unsigned long r)
252 type->total_size -= type->regions[r].size;
253 memmove(&type->regions[r], &type->regions[r + 1],
254 (type->cnt - (r + 1)) * sizeof(type->regions[r]));
257 /* Special case for empty arrays */
258 if (type->cnt == 0) {
259 WARN_ON(type->total_size != 0);
261 type->regions[0].base = 0;
262 type->regions[0].size = 0;
263 type->regions[0].flags = 0;
264 memblock_set_region_node(&type->regions[0], MAX_NUMNODES);
268 #ifdef CONFIG_ARCH_DISCARD_MEMBLOCK
270 phys_addr_t __init_memblock get_allocated_memblock_reserved_regions_info(
273 if (memblock.reserved.regions == memblock_reserved_init_regions)
276 *addr = __pa(memblock.reserved.regions);
278 return PAGE_ALIGN(sizeof(struct memblock_region) *
279 memblock.reserved.max);
282 phys_addr_t __init_memblock get_allocated_memblock_memory_regions_info(
285 if (memblock.memory.regions == memblock_memory_init_regions)
288 *addr = __pa(memblock.memory.regions);
290 return PAGE_ALIGN(sizeof(struct memblock_region) *
291 memblock.memory.max);
297 * memblock_double_array - double the size of the memblock regions array
298 * @type: memblock type of the regions array being doubled
299 * @new_area_start: starting address of memory range to avoid overlap with
300 * @new_area_size: size of memory range to avoid overlap with
302 * Double the size of the @type regions array. If memblock is being used to
303 * allocate memory for a new reserved regions array and there is a previously
304 * allocated memory range [@new_area_start,@new_area_start+@new_area_size]
305 * waiting to be reserved, ensure the memory used by the new array does
309 * 0 on success, -1 on failure.
311 static int __init_memblock memblock_double_array(struct memblock_type *type,
312 phys_addr_t new_area_start,
313 phys_addr_t new_area_size)
315 struct memblock_region *new_array, *old_array;
316 phys_addr_t old_alloc_size, new_alloc_size;
317 phys_addr_t old_size, new_size, addr;
318 int use_slab = slab_is_available();
321 /* We don't allow resizing until we know about the reserved regions
322 * of memory that aren't suitable for allocation
324 if (!memblock_can_resize)
327 /* Calculate new doubled size */
328 old_size = type->max * sizeof(struct memblock_region);
329 new_size = old_size << 1;
331 * We need to allocated new one align to PAGE_SIZE,
332 * so we can free them completely later.
334 old_alloc_size = PAGE_ALIGN(old_size);
335 new_alloc_size = PAGE_ALIGN(new_size);
337 /* Retrieve the slab flag */
338 if (type == &memblock.memory)
339 in_slab = &memblock_memory_in_slab;
341 in_slab = &memblock_reserved_in_slab;
343 /* Try to find some space for it.
345 * WARNING: We assume that either slab_is_available() and we use it or
346 * we use MEMBLOCK for allocations. That means that this is unsafe to
347 * use when bootmem is currently active (unless bootmem itself is
348 * implemented on top of MEMBLOCK which isn't the case yet)
350 * This should however not be an issue for now, as we currently only
351 * call into MEMBLOCK while it's still active, or much later when slab
352 * is active for memory hotplug operations
355 new_array = kmalloc(new_size, GFP_KERNEL);
356 addr = new_array ? __pa(new_array) : 0;
358 /* only exclude range when trying to double reserved.regions */
359 if (type != &memblock.reserved)
360 new_area_start = new_area_size = 0;
362 addr = memblock_find_in_range(new_area_start + new_area_size,
363 memblock.current_limit,
364 new_alloc_size, PAGE_SIZE);
365 if (!addr && new_area_size)
366 addr = memblock_find_in_range(0,
367 min(new_area_start, memblock.current_limit),
368 new_alloc_size, PAGE_SIZE);
370 new_array = addr ? __va(addr) : NULL;
373 pr_err("memblock: Failed to double %s array from %ld to %ld entries !\n",
374 memblock_type_name(type), type->max, type->max * 2);
378 memblock_dbg("memblock: %s is doubled to %ld at [%#010llx-%#010llx]",
379 memblock_type_name(type), type->max * 2, (u64)addr,
380 (u64)addr + new_size - 1);
383 * Found space, we now need to move the array over before we add the
384 * reserved region since it may be our reserved array itself that is
387 memcpy(new_array, type->regions, old_size);
388 memset(new_array + type->max, 0, old_size);
389 old_array = type->regions;
390 type->regions = new_array;
393 /* Free old array. We needn't free it if the array is the static one */
396 else if (old_array != memblock_memory_init_regions &&
397 old_array != memblock_reserved_init_regions)
398 memblock_free(__pa(old_array), old_alloc_size);
401 * Reserve the new array if that comes from the memblock. Otherwise, we
405 BUG_ON(memblock_reserve(addr, new_alloc_size));
407 /* Update slab flag */
414 * memblock_merge_regions - merge neighboring compatible regions
415 * @type: memblock type to scan
417 * Scan @type and merge neighboring compatible regions.
419 static void __init_memblock memblock_merge_regions(struct memblock_type *type)
423 /* cnt never goes below 1 */
424 while (i < type->cnt - 1) {
425 struct memblock_region *this = &type->regions[i];
426 struct memblock_region *next = &type->regions[i + 1];
428 if (this->base + this->size != next->base ||
429 memblock_get_region_node(this) !=
430 memblock_get_region_node(next) ||
431 this->flags != next->flags) {
432 BUG_ON(this->base + this->size > next->base);
437 this->size += next->size;
438 /* move forward from next + 1, index of which is i + 2 */
439 memmove(next, next + 1, (type->cnt - (i + 2)) * sizeof(*next));
445 * memblock_insert_region - insert new memblock region
446 * @type: memblock type to insert into
447 * @idx: index for the insertion point
448 * @base: base address of the new region
449 * @size: size of the new region
450 * @nid: node id of the new region
451 * @flags: flags of the new region
453 * Insert new memblock region [@base,@base+@size) into @type at @idx.
454 * @type must already have extra room to accomodate the new region.
456 static void __init_memblock memblock_insert_region(struct memblock_type *type,
457 int idx, phys_addr_t base,
459 int nid, unsigned long flags)
461 struct memblock_region *rgn = &type->regions[idx];
463 BUG_ON(type->cnt >= type->max);
464 memmove(rgn + 1, rgn, (type->cnt - idx) * sizeof(*rgn));
468 memblock_set_region_node(rgn, nid);
470 type->total_size += size;
474 * memblock_add_region - add new memblock region
475 * @type: memblock type to add new region into
476 * @base: base address of the new region
477 * @size: size of the new region
478 * @nid: nid of the new region
479 * @flags: flags of the new region
481 * Add new memblock region [@base,@base+@size) into @type. The new region
482 * is allowed to overlap with existing ones - overlaps don't affect already
483 * existing regions. @type is guaranteed to be minimal (all neighbouring
484 * compatible regions are merged) after the addition.
487 * 0 on success, -errno on failure.
489 static int __init_memblock memblock_add_region(struct memblock_type *type,
490 phys_addr_t base, phys_addr_t size,
491 int nid, unsigned long flags)
494 phys_addr_t obase = base;
495 phys_addr_t end = base + memblock_cap_size(base, &size);
501 /* special case for empty array */
502 if (type->regions[0].size == 0) {
503 WARN_ON(type->cnt != 1 || type->total_size);
504 type->regions[0].base = base;
505 type->regions[0].size = size;
506 type->regions[0].flags = flags;
507 memblock_set_region_node(&type->regions[0], nid);
508 type->total_size = size;
513 * The following is executed twice. Once with %false @insert and
514 * then with %true. The first counts the number of regions needed
515 * to accomodate the new area. The second actually inserts them.
520 for (i = 0; i < type->cnt; i++) {
521 struct memblock_region *rgn = &type->regions[i];
522 phys_addr_t rbase = rgn->base;
523 phys_addr_t rend = rbase + rgn->size;
530 * @rgn overlaps. If it separates the lower part of new
531 * area, insert that portion.
536 memblock_insert_region(type, i++, base,
540 /* area below @rend is dealt with, forget about it */
541 base = min(rend, end);
544 /* insert the remaining portion */
548 memblock_insert_region(type, i, base, end - base,
553 * If this was the first round, resize array and repeat for actual
554 * insertions; otherwise, merge and return.
557 while (type->cnt + nr_new > type->max)
558 if (memblock_double_array(type, obase, size) < 0)
563 memblock_merge_regions(type);
568 int __init_memblock memblock_add_node(phys_addr_t base, phys_addr_t size,
571 return memblock_add_region(&memblock.memory, base, size, nid, 0);
574 int __init_memblock memblock_add(phys_addr_t base, phys_addr_t size)
576 return memblock_add_region(&memblock.memory, base, size,
581 * memblock_isolate_range - isolate given range into disjoint memblocks
582 * @type: memblock type to isolate range for
583 * @base: base of range to isolate
584 * @size: size of range to isolate
585 * @start_rgn: out parameter for the start of isolated region
586 * @end_rgn: out parameter for the end of isolated region
588 * Walk @type and ensure that regions don't cross the boundaries defined by
589 * [@base,@base+@size). Crossing regions are split at the boundaries,
590 * which may create at most two more regions. The index of the first
591 * region inside the range is returned in *@start_rgn and end in *@end_rgn.
594 * 0 on success, -errno on failure.
596 static int __init_memblock memblock_isolate_range(struct memblock_type *type,
597 phys_addr_t base, phys_addr_t size,
598 int *start_rgn, int *end_rgn)
600 phys_addr_t end = base + memblock_cap_size(base, &size);
603 *start_rgn = *end_rgn = 0;
608 /* we'll create at most two more regions */
609 while (type->cnt + 2 > type->max)
610 if (memblock_double_array(type, base, size) < 0)
613 for (i = 0; i < type->cnt; i++) {
614 struct memblock_region *rgn = &type->regions[i];
615 phys_addr_t rbase = rgn->base;
616 phys_addr_t rend = rbase + rgn->size;
625 * @rgn intersects from below. Split and continue
626 * to process the next region - the new top half.
629 rgn->size -= base - rbase;
630 type->total_size -= base - rbase;
631 memblock_insert_region(type, i, rbase, base - rbase,
632 memblock_get_region_node(rgn),
634 } else if (rend > end) {
636 * @rgn intersects from above. Split and redo the
637 * current region - the new bottom half.
640 rgn->size -= end - rbase;
641 type->total_size -= end - rbase;
642 memblock_insert_region(type, i--, rbase, end - rbase,
643 memblock_get_region_node(rgn),
646 /* @rgn is fully contained, record it */
656 static int __init_memblock __memblock_remove(struct memblock_type *type,
657 phys_addr_t base, phys_addr_t size)
659 int start_rgn, end_rgn;
662 ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
666 for (i = end_rgn - 1; i >= start_rgn; i--)
667 memblock_remove_region(type, i);
671 int __init_memblock memblock_remove(phys_addr_t base, phys_addr_t size)
673 return __memblock_remove(&memblock.memory, base, size);
676 int __init_memblock memblock_free(phys_addr_t base, phys_addr_t size)
678 memblock_dbg(" memblock_free: [%#016llx-%#016llx] %pF\n",
679 (unsigned long long)base,
680 (unsigned long long)base + size - 1,
683 return __memblock_remove(&memblock.reserved, base, size);
686 static int __init_memblock memblock_reserve_region(phys_addr_t base,
691 struct memblock_type *_rgn = &memblock.reserved;
693 memblock_dbg("memblock_reserve: [%#016llx-%#016llx] flags %#02lx %pF\n",
694 (unsigned long long)base,
695 (unsigned long long)base + size - 1,
696 flags, (void *)_RET_IP_);
698 return memblock_add_region(_rgn, base, size, nid, flags);
701 int __init_memblock memblock_reserve(phys_addr_t base, phys_addr_t size)
703 return memblock_reserve_region(base, size, MAX_NUMNODES, 0);
707 * memblock_mark_hotplug - Mark hotpluggable memory with flag MEMBLOCK_HOTPLUG.
708 * @base: the base phys addr of the region
709 * @size: the size of the region
711 * This function isolates region [@base, @base + @size), and mark it with flag
714 * Return 0 on succees, -errno on failure.
716 int __init_memblock memblock_mark_hotplug(phys_addr_t base, phys_addr_t size)
718 struct memblock_type *type = &memblock.memory;
719 int i, ret, start_rgn, end_rgn;
721 ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
725 for (i = start_rgn; i < end_rgn; i++)
726 memblock_set_region_flags(&type->regions[i], MEMBLOCK_HOTPLUG);
728 memblock_merge_regions(type);
733 * memblock_clear_hotplug - Clear flag MEMBLOCK_HOTPLUG for a specified region.
734 * @base: the base phys addr of the region
735 * @size: the size of the region
737 * This function isolates region [@base, @base + @size), and clear flag
738 * MEMBLOCK_HOTPLUG for the isolated regions.
740 * Return 0 on succees, -errno on failure.
742 int __init_memblock memblock_clear_hotplug(phys_addr_t base, phys_addr_t size)
744 struct memblock_type *type = &memblock.memory;
745 int i, ret, start_rgn, end_rgn;
747 ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
751 for (i = start_rgn; i < end_rgn; i++)
752 memblock_clear_region_flags(&type->regions[i],
755 memblock_merge_regions(type);
760 * __next_free_mem_range - next function for for_each_free_mem_range()
761 * @idx: pointer to u64 loop variable
762 * @nid: node selector, %NUMA_NO_NODE for all nodes
763 * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL
764 * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL
765 * @out_nid: ptr to int for nid of the range, can be %NULL
767 * Find the first free area from *@idx which matches @nid, fill the out
768 * parameters, and update *@idx for the next iteration. The lower 32bit of
769 * *@idx contains index into memory region and the upper 32bit indexes the
770 * areas before each reserved region. For example, if reserved regions
771 * look like the following,
773 * 0:[0-16), 1:[32-48), 2:[128-130)
775 * The upper 32bit indexes the following regions.
777 * 0:[0-0), 1:[16-32), 2:[48-128), 3:[130-MAX)
779 * As both region arrays are sorted, the function advances the two indices
780 * in lockstep and returns each intersection.
782 void __init_memblock __next_free_mem_range(u64 *idx, int nid,
783 phys_addr_t *out_start,
784 phys_addr_t *out_end, int *out_nid)
786 struct memblock_type *mem = &memblock.memory;
787 struct memblock_type *rsv = &memblock.reserved;
788 int mi = *idx & 0xffffffff;
791 if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
794 for ( ; mi < mem->cnt; mi++) {
795 struct memblock_region *m = &mem->regions[mi];
796 phys_addr_t m_start = m->base;
797 phys_addr_t m_end = m->base + m->size;
799 /* only memory regions are associated with nodes, check it */
800 if (nid != NUMA_NO_NODE && nid != memblock_get_region_node(m))
803 /* scan areas before each reservation for intersection */
804 for ( ; ri < rsv->cnt + 1; ri++) {
805 struct memblock_region *r = &rsv->regions[ri];
806 phys_addr_t r_start = ri ? r[-1].base + r[-1].size : 0;
807 phys_addr_t r_end = ri < rsv->cnt ? r->base : ULLONG_MAX;
809 /* if ri advanced past mi, break out to advance mi */
810 if (r_start >= m_end)
812 /* if the two regions intersect, we're done */
813 if (m_start < r_end) {
815 *out_start = max(m_start, r_start);
817 *out_end = min(m_end, r_end);
819 *out_nid = memblock_get_region_node(m);
821 * The region which ends first is advanced
822 * for the next iteration.
828 *idx = (u32)mi | (u64)ri << 32;
834 /* signal end of iteration */
839 * __next_free_mem_range_rev - next function for for_each_free_mem_range_reverse()
840 * @idx: pointer to u64 loop variable
841 * @nid: nid: node selector, %NUMA_NO_NODE for all nodes
842 * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL
843 * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL
844 * @out_nid: ptr to int for nid of the range, can be %NULL
846 * Reverse of __next_free_mem_range().
848 * Linux kernel cannot migrate pages used by itself. Memory hotplug users won't
849 * be able to hot-remove hotpluggable memory used by the kernel. So this
850 * function skip hotpluggable regions if needed when allocating memory for the
853 void __init_memblock __next_free_mem_range_rev(u64 *idx, int nid,
854 phys_addr_t *out_start,
855 phys_addr_t *out_end, int *out_nid)
857 struct memblock_type *mem = &memblock.memory;
858 struct memblock_type *rsv = &memblock.reserved;
859 int mi = *idx & 0xffffffff;
862 if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
865 if (*idx == (u64)ULLONG_MAX) {
870 for ( ; mi >= 0; mi--) {
871 struct memblock_region *m = &mem->regions[mi];
872 phys_addr_t m_start = m->base;
873 phys_addr_t m_end = m->base + m->size;
875 /* only memory regions are associated with nodes, check it */
876 if (nid != NUMA_NO_NODE && nid != memblock_get_region_node(m))
879 /* skip hotpluggable memory regions if needed */
880 if (movable_node_is_enabled() && memblock_is_hotpluggable(m))
883 /* scan areas before each reservation for intersection */
884 for ( ; ri >= 0; ri--) {
885 struct memblock_region *r = &rsv->regions[ri];
886 phys_addr_t r_start = ri ? r[-1].base + r[-1].size : 0;
887 phys_addr_t r_end = ri < rsv->cnt ? r->base : ULLONG_MAX;
889 /* if ri advanced past mi, break out to advance mi */
890 if (r_end <= m_start)
892 /* if the two regions intersect, we're done */
893 if (m_end > r_start) {
895 *out_start = max(m_start, r_start);
897 *out_end = min(m_end, r_end);
899 *out_nid = memblock_get_region_node(m);
901 if (m_start >= r_start)
905 *idx = (u32)mi | (u64)ri << 32;
914 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
916 * Common iterator interface used to define for_each_mem_range().
918 void __init_memblock __next_mem_pfn_range(int *idx, int nid,
919 unsigned long *out_start_pfn,
920 unsigned long *out_end_pfn, int *out_nid)
922 struct memblock_type *type = &memblock.memory;
923 struct memblock_region *r;
925 while (++*idx < type->cnt) {
926 r = &type->regions[*idx];
928 if (PFN_UP(r->base) >= PFN_DOWN(r->base + r->size))
930 if (nid == MAX_NUMNODES || nid == r->nid)
933 if (*idx >= type->cnt) {
939 *out_start_pfn = PFN_UP(r->base);
941 *out_end_pfn = PFN_DOWN(r->base + r->size);
947 * memblock_set_node - set node ID on memblock regions
948 * @base: base of area to set node ID for
949 * @size: size of area to set node ID for
950 * @type: memblock type to set node ID for
951 * @nid: node ID to set
953 * Set the nid of memblock @type regions in [@base,@base+@size) to @nid.
954 * Regions which cross the area boundaries are split as necessary.
957 * 0 on success, -errno on failure.
959 int __init_memblock memblock_set_node(phys_addr_t base, phys_addr_t size,
960 struct memblock_type *type, int nid)
962 int start_rgn, end_rgn;
965 ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
969 for (i = start_rgn; i < end_rgn; i++)
970 memblock_set_region_node(&type->regions[i], nid);
972 memblock_merge_regions(type);
975 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
977 static phys_addr_t __init memblock_alloc_base_nid(phys_addr_t size,
978 phys_addr_t align, phys_addr_t max_addr,
984 align = SMP_CACHE_BYTES;
986 found = memblock_find_in_range_node(size, align, 0, max_addr, nid);
987 if (found && !memblock_reserve(found, size))
993 phys_addr_t __init memblock_alloc_nid(phys_addr_t size, phys_addr_t align, int nid)
995 return memblock_alloc_base_nid(size, align, MEMBLOCK_ALLOC_ACCESSIBLE, nid);
998 phys_addr_t __init __memblock_alloc_base(phys_addr_t size, phys_addr_t align, phys_addr_t max_addr)
1000 return memblock_alloc_base_nid(size, align, max_addr, NUMA_NO_NODE);
1003 phys_addr_t __init memblock_alloc_base(phys_addr_t size, phys_addr_t align, phys_addr_t max_addr)
1007 alloc = __memblock_alloc_base(size, align, max_addr);
1010 panic("ERROR: Failed to allocate 0x%llx bytes below 0x%llx.\n",
1011 (unsigned long long) size, (unsigned long long) max_addr);
1016 phys_addr_t __init memblock_alloc(phys_addr_t size, phys_addr_t align)
1018 return memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ACCESSIBLE);
1021 phys_addr_t __init memblock_alloc_try_nid(phys_addr_t size, phys_addr_t align, int nid)
1023 phys_addr_t res = memblock_alloc_nid(size, align, nid);
1027 return memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ACCESSIBLE);
1031 * memblock_virt_alloc_internal - allocate boot memory block
1032 * @size: size of memory block to be allocated in bytes
1033 * @align: alignment of the region and block's size
1034 * @min_addr: the lower bound of the memory region to allocate (phys address)
1035 * @max_addr: the upper bound of the memory region to allocate (phys address)
1036 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1038 * The @min_addr limit is dropped if it can not be satisfied and the allocation
1039 * will fall back to memory below @min_addr. Also, allocation may fall back
1040 * to any node in the system if the specified node can not
1041 * hold the requested memory.
1043 * The allocation is performed from memory region limited by
1044 * memblock.current_limit if @max_addr == %BOOTMEM_ALLOC_ACCESSIBLE.
1046 * The memory block is aligned on SMP_CACHE_BYTES if @align == 0.
1048 * The phys address of allocated boot memory block is converted to virtual and
1049 * allocated memory is reset to 0.
1051 * In addition, function sets the min_count to 0 using kmemleak_alloc for
1052 * allocated boot memory block, so that it is never reported as leaks.
1055 * Virtual address of allocated memory block on success, NULL on failure.
1057 static void * __init memblock_virt_alloc_internal(
1058 phys_addr_t size, phys_addr_t align,
1059 phys_addr_t min_addr, phys_addr_t max_addr,
1065 if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
1069 * Detect any accidental use of these APIs after slab is ready, as at
1070 * this moment memblock may be deinitialized already and its
1071 * internal data may be destroyed (after execution of free_all_bootmem)
1073 if (WARN_ON_ONCE(slab_is_available()))
1074 return kzalloc_node(size, GFP_NOWAIT, nid);
1077 align = SMP_CACHE_BYTES;
1079 if (max_addr > memblock.current_limit)
1080 max_addr = memblock.current_limit;
1083 alloc = memblock_find_in_range_node(size, align, min_addr, max_addr,
1088 if (nid != NUMA_NO_NODE) {
1089 alloc = memblock_find_in_range_node(size, align, min_addr,
1090 max_addr, NUMA_NO_NODE);
1103 memblock_reserve(alloc, size);
1104 ptr = phys_to_virt(alloc);
1105 memset(ptr, 0, size);
1108 * The min_count is set to 0 so that bootmem allocated blocks
1109 * are never reported as leaks. This is because many of these blocks
1110 * are only referred via the physical address which is not
1111 * looked up by kmemleak.
1113 kmemleak_alloc(ptr, size, 0, 0);
1122 * memblock_virt_alloc_try_nid_nopanic - allocate boot memory block
1123 * @size: size of memory block to be allocated in bytes
1124 * @align: alignment of the region and block's size
1125 * @min_addr: the lower bound of the memory region from where the allocation
1126 * is preferred (phys address)
1127 * @max_addr: the upper bound of the memory region from where the allocation
1128 * is preferred (phys address), or %BOOTMEM_ALLOC_ACCESSIBLE to
1129 * allocate only from memory limited by memblock.current_limit value
1130 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1132 * Public version of _memblock_virt_alloc_try_nid_nopanic() which provides
1133 * additional debug information (including caller info), if enabled.
1136 * Virtual address of allocated memory block on success, NULL on failure.
1138 void * __init memblock_virt_alloc_try_nid_nopanic(
1139 phys_addr_t size, phys_addr_t align,
1140 phys_addr_t min_addr, phys_addr_t max_addr,
1143 memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=0x%llx max_addr=0x%llx %pF\n",
1144 __func__, (u64)size, (u64)align, nid, (u64)min_addr,
1145 (u64)max_addr, (void *)_RET_IP_);
1146 return memblock_virt_alloc_internal(size, align, min_addr,
1151 * memblock_virt_alloc_try_nid - allocate boot memory block with panicking
1152 * @size: size of memory block to be allocated in bytes
1153 * @align: alignment of the region and block's size
1154 * @min_addr: the lower bound of the memory region from where the allocation
1155 * is preferred (phys address)
1156 * @max_addr: the upper bound of the memory region from where the allocation
1157 * is preferred (phys address), or %BOOTMEM_ALLOC_ACCESSIBLE to
1158 * allocate only from memory limited by memblock.current_limit value
1159 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1161 * Public panicking version of _memblock_virt_alloc_try_nid_nopanic()
1162 * which provides debug information (including caller info), if enabled,
1163 * and panics if the request can not be satisfied.
1166 * Virtual address of allocated memory block on success, NULL on failure.
1168 void * __init memblock_virt_alloc_try_nid(
1169 phys_addr_t size, phys_addr_t align,
1170 phys_addr_t min_addr, phys_addr_t max_addr,
1175 memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=0x%llx max_addr=0x%llx %pF\n",
1176 __func__, (u64)size, (u64)align, nid, (u64)min_addr,
1177 (u64)max_addr, (void *)_RET_IP_);
1178 ptr = memblock_virt_alloc_internal(size, align,
1179 min_addr, max_addr, nid);
1183 panic("%s: Failed to allocate %llu bytes align=0x%llx nid=%d from=0x%llx max_addr=0x%llx\n",
1184 __func__, (u64)size, (u64)align, nid, (u64)min_addr,
1190 * __memblock_free_early - free boot memory block
1191 * @base: phys starting address of the boot memory block
1192 * @size: size of the boot memory block in bytes
1194 * Free boot memory block previously allocated by memblock_virt_alloc_xx() API.
1195 * The freeing memory will not be released to the buddy allocator.
1197 void __init __memblock_free_early(phys_addr_t base, phys_addr_t size)
1199 memblock_dbg("%s: [%#016llx-%#016llx] %pF\n",
1200 __func__, (u64)base, (u64)base + size - 1,
1202 kmemleak_free_part(__va(base), size);
1203 __memblock_remove(&memblock.reserved, base, size);
1207 * __memblock_free_late - free bootmem block pages directly to buddy allocator
1208 * @addr: phys starting address of the boot memory block
1209 * @size: size of the boot memory block in bytes
1211 * This is only useful when the bootmem allocator has already been torn
1212 * down, but we are still initializing the system. Pages are released directly
1213 * to the buddy allocator, no bootmem metadata is updated because it is gone.
1215 void __init __memblock_free_late(phys_addr_t base, phys_addr_t size)
1219 memblock_dbg("%s: [%#016llx-%#016llx] %pF\n",
1220 __func__, (u64)base, (u64)base + size - 1,
1222 kmemleak_free_part(__va(base), size);
1223 cursor = PFN_UP(base);
1224 end = PFN_DOWN(base + size);
1226 for (; cursor < end; cursor++) {
1227 __free_pages_bootmem(pfn_to_page(cursor), 0);
1233 * Remaining API functions
1236 phys_addr_t __init memblock_phys_mem_size(void)
1238 return memblock.memory.total_size;
1241 phys_addr_t __init memblock_mem_size(unsigned long limit_pfn)
1243 unsigned long pages = 0;
1244 struct memblock_region *r;
1245 unsigned long start_pfn, end_pfn;
1247 for_each_memblock(memory, r) {
1248 start_pfn = memblock_region_memory_base_pfn(r);
1249 end_pfn = memblock_region_memory_end_pfn(r);
1250 start_pfn = min_t(unsigned long, start_pfn, limit_pfn);
1251 end_pfn = min_t(unsigned long, end_pfn, limit_pfn);
1252 pages += end_pfn - start_pfn;
1255 return (phys_addr_t)pages << PAGE_SHIFT;
1258 /* lowest address */
1259 phys_addr_t __init_memblock memblock_start_of_DRAM(void)
1261 return memblock.memory.regions[0].base;
1264 phys_addr_t __init_memblock memblock_end_of_DRAM(void)
1266 int idx = memblock.memory.cnt - 1;
1268 return (memblock.memory.regions[idx].base + memblock.memory.regions[idx].size);
1271 void __init memblock_enforce_memory_limit(phys_addr_t limit)
1274 phys_addr_t max_addr = (phys_addr_t)ULLONG_MAX;
1279 /* find out max address */
1280 for (i = 0; i < memblock.memory.cnt; i++) {
1281 struct memblock_region *r = &memblock.memory.regions[i];
1283 if (limit <= r->size) {
1284 max_addr = r->base + limit;
1290 /* truncate both memory and reserved regions */
1291 __memblock_remove(&memblock.memory, max_addr, (phys_addr_t)ULLONG_MAX);
1292 __memblock_remove(&memblock.reserved, max_addr, (phys_addr_t)ULLONG_MAX);
1295 static int __init_memblock memblock_search(struct memblock_type *type, phys_addr_t addr)
1297 unsigned int left = 0, right = type->cnt;
1300 unsigned int mid = (right + left) / 2;
1302 if (addr < type->regions[mid].base)
1304 else if (addr >= (type->regions[mid].base +
1305 type->regions[mid].size))
1309 } while (left < right);
1313 int __init memblock_is_reserved(phys_addr_t addr)
1315 return memblock_search(&memblock.reserved, addr) != -1;
1318 int __init_memblock memblock_is_memory(phys_addr_t addr)
1320 return memblock_search(&memblock.memory, addr) != -1;
1323 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
1324 int __init_memblock memblock_search_pfn_nid(unsigned long pfn,
1325 unsigned long *start_pfn, unsigned long *end_pfn)
1327 struct memblock_type *type = &memblock.memory;
1328 int mid = memblock_search(type, (phys_addr_t)pfn << PAGE_SHIFT);
1333 *start_pfn = type->regions[mid].base >> PAGE_SHIFT;
1334 *end_pfn = (type->regions[mid].base + type->regions[mid].size)
1337 return type->regions[mid].nid;
1342 * memblock_is_region_memory - check if a region is a subset of memory
1343 * @base: base of region to check
1344 * @size: size of region to check
1346 * Check if the region [@base, @base+@size) is a subset of a memory block.
1349 * 0 if false, non-zero if true
1351 int __init_memblock memblock_is_region_memory(phys_addr_t base, phys_addr_t size)
1353 int idx = memblock_search(&memblock.memory, base);
1354 phys_addr_t end = base + memblock_cap_size(base, &size);
1358 return memblock.memory.regions[idx].base <= base &&
1359 (memblock.memory.regions[idx].base +
1360 memblock.memory.regions[idx].size) >= end;
1364 * memblock_is_region_reserved - check if a region intersects reserved memory
1365 * @base: base of region to check
1366 * @size: size of region to check
1368 * Check if the region [@base, @base+@size) intersects a reserved memory block.
1371 * 0 if false, non-zero if true
1373 int __init_memblock memblock_is_region_reserved(phys_addr_t base, phys_addr_t size)
1375 memblock_cap_size(base, &size);
1376 return memblock_overlaps_region(&memblock.reserved, base, size) >= 0;
1379 void __init_memblock memblock_trim_memory(phys_addr_t align)
1382 phys_addr_t start, end, orig_start, orig_end;
1383 struct memblock_type *mem = &memblock.memory;
1385 for (i = 0; i < mem->cnt; i++) {
1386 orig_start = mem->regions[i].base;
1387 orig_end = mem->regions[i].base + mem->regions[i].size;
1388 start = round_up(orig_start, align);
1389 end = round_down(orig_end, align);
1391 if (start == orig_start && end == orig_end)
1395 mem->regions[i].base = start;
1396 mem->regions[i].size = end - start;
1398 memblock_remove_region(mem, i);
1404 void __init_memblock memblock_set_current_limit(phys_addr_t limit)
1406 memblock.current_limit = limit;
1409 static void __init_memblock memblock_dump(struct memblock_type *type, char *name)
1411 unsigned long long base, size;
1412 unsigned long flags;
1415 pr_info(" %s.cnt = 0x%lx\n", name, type->cnt);
1417 for (i = 0; i < type->cnt; i++) {
1418 struct memblock_region *rgn = &type->regions[i];
1419 char nid_buf[32] = "";
1424 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
1425 if (memblock_get_region_node(rgn) != MAX_NUMNODES)
1426 snprintf(nid_buf, sizeof(nid_buf), " on node %d",
1427 memblock_get_region_node(rgn));
1429 pr_info(" %s[%#x]\t[%#016llx-%#016llx], %#llx bytes%s flags: %#lx\n",
1430 name, i, base, base + size - 1, size, nid_buf, flags);
1434 void __init_memblock __memblock_dump_all(void)
1436 pr_info("MEMBLOCK configuration:\n");
1437 pr_info(" memory size = %#llx reserved size = %#llx\n",
1438 (unsigned long long)memblock.memory.total_size,
1439 (unsigned long long)memblock.reserved.total_size);
1441 memblock_dump(&memblock.memory, "memory");
1442 memblock_dump(&memblock.reserved, "reserved");
1445 void __init memblock_allow_resize(void)
1447 memblock_can_resize = 1;
1450 static int __init early_memblock(char *p)
1452 if (p && strstr(p, "debug"))
1456 early_param("memblock", early_memblock);
1458 #if defined(CONFIG_DEBUG_FS) && !defined(CONFIG_ARCH_DISCARD_MEMBLOCK)
1460 static int memblock_debug_show(struct seq_file *m, void *private)
1462 struct memblock_type *type = m->private;
1463 struct memblock_region *reg;
1466 for (i = 0; i < type->cnt; i++) {
1467 reg = &type->regions[i];
1468 seq_printf(m, "%4d: ", i);
1469 if (sizeof(phys_addr_t) == 4)
1470 seq_printf(m, "0x%08lx..0x%08lx\n",
1471 (unsigned long)reg->base,
1472 (unsigned long)(reg->base + reg->size - 1));
1474 seq_printf(m, "0x%016llx..0x%016llx\n",
1475 (unsigned long long)reg->base,
1476 (unsigned long long)(reg->base + reg->size - 1));
1482 static int memblock_debug_open(struct inode *inode, struct file *file)
1484 return single_open(file, memblock_debug_show, inode->i_private);
1487 static const struct file_operations memblock_debug_fops = {
1488 .open = memblock_debug_open,
1490 .llseek = seq_lseek,
1491 .release = single_release,
1494 static int __init memblock_init_debugfs(void)
1496 struct dentry *root = debugfs_create_dir("memblock", NULL);
1499 debugfs_create_file("memory", S_IRUGO, root, &memblock.memory, &memblock_debug_fops);
1500 debugfs_create_file("reserved", S_IRUGO, root, &memblock.reserved, &memblock_debug_fops);
1504 __initcall(memblock_init_debugfs);
1506 #endif /* CONFIG_DEBUG_FS */