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)
196 phys_addr_t kernel_end;
199 if (end == MEMBLOCK_ALLOC_ACCESSIBLE)
200 end = memblock.current_limit;
202 /* avoid allocating the first page */
203 start = max_t(phys_addr_t, start, PAGE_SIZE);
204 end = max(start, end);
205 kernel_end = __pa_symbol(_end);
208 * try bottom-up allocation only when bottom-up mode
209 * is set and @end is above the kernel image.
211 if (memblock_bottom_up() && end > kernel_end) {
212 phys_addr_t bottom_up_start;
214 /* make sure we will allocate above the kernel */
215 bottom_up_start = max(start, kernel_end);
217 /* ok, try bottom-up allocation first */
218 ret = __memblock_find_range_bottom_up(bottom_up_start, end,
224 * we always limit bottom-up allocation above the kernel,
225 * but top-down allocation doesn't have the limit, so
226 * retrying top-down allocation may succeed when bottom-up
229 * bottom-up allocation is expected to be fail very rarely,
230 * so we use WARN_ONCE() here to see the stack trace if
233 WARN_ONCE(1, "memblock: bottom-up allocation failed, "
234 "memory hotunplug may be affected\n");
237 return __memblock_find_range_top_down(start, end, size, align, nid);
241 * memblock_find_in_range - find free area in given range
242 * @start: start of candidate range
243 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_{ANYWHERE|ACCESSIBLE}
244 * @size: size of free area to find
245 * @align: alignment of free area to find
247 * Find @size free area aligned to @align in the specified range.
250 * Found address on success, 0 on failure.
252 phys_addr_t __init_memblock memblock_find_in_range(phys_addr_t start,
253 phys_addr_t end, phys_addr_t size,
256 return memblock_find_in_range_node(size, align, start, end,
260 static void __init_memblock memblock_remove_region(struct memblock_type *type, unsigned long r)
262 type->total_size -= type->regions[r].size;
263 memmove(&type->regions[r], &type->regions[r + 1],
264 (type->cnt - (r + 1)) * sizeof(type->regions[r]));
267 /* Special case for empty arrays */
268 if (type->cnt == 0) {
269 WARN_ON(type->total_size != 0);
271 type->regions[0].base = 0;
272 type->regions[0].size = 0;
273 type->regions[0].flags = 0;
274 memblock_set_region_node(&type->regions[0], MAX_NUMNODES);
278 #ifdef CONFIG_ARCH_DISCARD_MEMBLOCK
280 phys_addr_t __init_memblock get_allocated_memblock_reserved_regions_info(
283 if (memblock.reserved.regions == memblock_reserved_init_regions)
286 *addr = __pa(memblock.reserved.regions);
288 return PAGE_ALIGN(sizeof(struct memblock_region) *
289 memblock.reserved.max);
292 phys_addr_t __init_memblock get_allocated_memblock_memory_regions_info(
295 if (memblock.memory.regions == memblock_memory_init_regions)
298 *addr = __pa(memblock.memory.regions);
300 return PAGE_ALIGN(sizeof(struct memblock_region) *
301 memblock.memory.max);
307 * memblock_double_array - double the size of the memblock regions array
308 * @type: memblock type of the regions array being doubled
309 * @new_area_start: starting address of memory range to avoid overlap with
310 * @new_area_size: size of memory range to avoid overlap with
312 * Double the size of the @type regions array. If memblock is being used to
313 * allocate memory for a new reserved regions array and there is a previously
314 * allocated memory range [@new_area_start,@new_area_start+@new_area_size]
315 * waiting to be reserved, ensure the memory used by the new array does
319 * 0 on success, -1 on failure.
321 static int __init_memblock memblock_double_array(struct memblock_type *type,
322 phys_addr_t new_area_start,
323 phys_addr_t new_area_size)
325 struct memblock_region *new_array, *old_array;
326 phys_addr_t old_alloc_size, new_alloc_size;
327 phys_addr_t old_size, new_size, addr;
328 int use_slab = slab_is_available();
331 /* We don't allow resizing until we know about the reserved regions
332 * of memory that aren't suitable for allocation
334 if (!memblock_can_resize)
337 /* Calculate new doubled size */
338 old_size = type->max * sizeof(struct memblock_region);
339 new_size = old_size << 1;
341 * We need to allocated new one align to PAGE_SIZE,
342 * so we can free them completely later.
344 old_alloc_size = PAGE_ALIGN(old_size);
345 new_alloc_size = PAGE_ALIGN(new_size);
347 /* Retrieve the slab flag */
348 if (type == &memblock.memory)
349 in_slab = &memblock_memory_in_slab;
351 in_slab = &memblock_reserved_in_slab;
353 /* Try to find some space for it.
355 * WARNING: We assume that either slab_is_available() and we use it or
356 * we use MEMBLOCK for allocations. That means that this is unsafe to
357 * use when bootmem is currently active (unless bootmem itself is
358 * implemented on top of MEMBLOCK which isn't the case yet)
360 * This should however not be an issue for now, as we currently only
361 * call into MEMBLOCK while it's still active, or much later when slab
362 * is active for memory hotplug operations
365 new_array = kmalloc(new_size, GFP_KERNEL);
366 addr = new_array ? __pa(new_array) : 0;
368 /* only exclude range when trying to double reserved.regions */
369 if (type != &memblock.reserved)
370 new_area_start = new_area_size = 0;
372 addr = memblock_find_in_range(new_area_start + new_area_size,
373 memblock.current_limit,
374 new_alloc_size, PAGE_SIZE);
375 if (!addr && new_area_size)
376 addr = memblock_find_in_range(0,
377 min(new_area_start, memblock.current_limit),
378 new_alloc_size, PAGE_SIZE);
380 new_array = addr ? __va(addr) : NULL;
383 pr_err("memblock: Failed to double %s array from %ld to %ld entries !\n",
384 memblock_type_name(type), type->max, type->max * 2);
388 memblock_dbg("memblock: %s is doubled to %ld at [%#010llx-%#010llx]",
389 memblock_type_name(type), type->max * 2, (u64)addr,
390 (u64)addr + new_size - 1);
393 * Found space, we now need to move the array over before we add the
394 * reserved region since it may be our reserved array itself that is
397 memcpy(new_array, type->regions, old_size);
398 memset(new_array + type->max, 0, old_size);
399 old_array = type->regions;
400 type->regions = new_array;
403 /* Free old array. We needn't free it if the array is the static one */
406 else if (old_array != memblock_memory_init_regions &&
407 old_array != memblock_reserved_init_regions)
408 memblock_free(__pa(old_array), old_alloc_size);
411 * Reserve the new array if that comes from the memblock. Otherwise, we
415 BUG_ON(memblock_reserve(addr, new_alloc_size));
417 /* Update slab flag */
424 * memblock_merge_regions - merge neighboring compatible regions
425 * @type: memblock type to scan
427 * Scan @type and merge neighboring compatible regions.
429 static void __init_memblock memblock_merge_regions(struct memblock_type *type)
433 /* cnt never goes below 1 */
434 while (i < type->cnt - 1) {
435 struct memblock_region *this = &type->regions[i];
436 struct memblock_region *next = &type->regions[i + 1];
438 if (this->base + this->size != next->base ||
439 memblock_get_region_node(this) !=
440 memblock_get_region_node(next) ||
441 this->flags != next->flags) {
442 BUG_ON(this->base + this->size > next->base);
447 this->size += next->size;
448 /* move forward from next + 1, index of which is i + 2 */
449 memmove(next, next + 1, (type->cnt - (i + 2)) * sizeof(*next));
455 * memblock_insert_region - insert new memblock region
456 * @type: memblock type to insert into
457 * @idx: index for the insertion point
458 * @base: base address of the new region
459 * @size: size of the new region
460 * @nid: node id of the new region
461 * @flags: flags of the new region
463 * Insert new memblock region [@base,@base+@size) into @type at @idx.
464 * @type must already have extra room to accomodate the new region.
466 static void __init_memblock memblock_insert_region(struct memblock_type *type,
467 int idx, phys_addr_t base,
469 int nid, unsigned long flags)
471 struct memblock_region *rgn = &type->regions[idx];
473 BUG_ON(type->cnt >= type->max);
474 memmove(rgn + 1, rgn, (type->cnt - idx) * sizeof(*rgn));
478 memblock_set_region_node(rgn, nid);
480 type->total_size += size;
484 * memblock_add_range - add new memblock region
485 * @type: memblock type to add new region into
486 * @base: base address of the new region
487 * @size: size of the new region
488 * @nid: nid of the new region
489 * @flags: flags of the new region
491 * Add new memblock region [@base,@base+@size) into @type. The new region
492 * is allowed to overlap with existing ones - overlaps don't affect already
493 * existing regions. @type is guaranteed to be minimal (all neighbouring
494 * compatible regions are merged) after the addition.
497 * 0 on success, -errno on failure.
499 int __init_memblock memblock_add_range(struct memblock_type *type,
500 phys_addr_t base, phys_addr_t size,
501 int nid, unsigned long flags)
504 phys_addr_t obase = base;
505 phys_addr_t end = base + memblock_cap_size(base, &size);
511 /* special case for empty array */
512 if (type->regions[0].size == 0) {
513 WARN_ON(type->cnt != 1 || type->total_size);
514 type->regions[0].base = base;
515 type->regions[0].size = size;
516 type->regions[0].flags = flags;
517 memblock_set_region_node(&type->regions[0], nid);
518 type->total_size = size;
523 * The following is executed twice. Once with %false @insert and
524 * then with %true. The first counts the number of regions needed
525 * to accomodate the new area. The second actually inserts them.
530 for (i = 0; i < type->cnt; i++) {
531 struct memblock_region *rgn = &type->regions[i];
532 phys_addr_t rbase = rgn->base;
533 phys_addr_t rend = rbase + rgn->size;
540 * @rgn overlaps. If it separates the lower part of new
541 * area, insert that portion.
546 memblock_insert_region(type, i++, base,
550 /* area below @rend is dealt with, forget about it */
551 base = min(rend, end);
554 /* insert the remaining portion */
558 memblock_insert_region(type, i, base, end - base,
563 * If this was the first round, resize array and repeat for actual
564 * insertions; otherwise, merge and return.
567 while (type->cnt + nr_new > type->max)
568 if (memblock_double_array(type, obase, size) < 0)
573 memblock_merge_regions(type);
578 int __init_memblock memblock_add_node(phys_addr_t base, phys_addr_t size,
581 return memblock_add_range(&memblock.memory, base, size, nid, 0);
584 int __init_memblock memblock_add(phys_addr_t base, phys_addr_t size)
586 return memblock_add_range(&memblock.memory, base, size,
591 * memblock_isolate_range - isolate given range into disjoint memblocks
592 * @type: memblock type to isolate range for
593 * @base: base of range to isolate
594 * @size: size of range to isolate
595 * @start_rgn: out parameter for the start of isolated region
596 * @end_rgn: out parameter for the end of isolated region
598 * Walk @type and ensure that regions don't cross the boundaries defined by
599 * [@base,@base+@size). Crossing regions are split at the boundaries,
600 * which may create at most two more regions. The index of the first
601 * region inside the range is returned in *@start_rgn and end in *@end_rgn.
604 * 0 on success, -errno on failure.
606 static int __init_memblock memblock_isolate_range(struct memblock_type *type,
607 phys_addr_t base, phys_addr_t size,
608 int *start_rgn, int *end_rgn)
610 phys_addr_t end = base + memblock_cap_size(base, &size);
613 *start_rgn = *end_rgn = 0;
618 /* we'll create at most two more regions */
619 while (type->cnt + 2 > type->max)
620 if (memblock_double_array(type, base, size) < 0)
623 for (i = 0; i < type->cnt; i++) {
624 struct memblock_region *rgn = &type->regions[i];
625 phys_addr_t rbase = rgn->base;
626 phys_addr_t rend = rbase + rgn->size;
635 * @rgn intersects from below. Split and continue
636 * to process the next region - the new top half.
639 rgn->size -= base - rbase;
640 type->total_size -= base - rbase;
641 memblock_insert_region(type, i, rbase, base - rbase,
642 memblock_get_region_node(rgn),
644 } else if (rend > end) {
646 * @rgn intersects from above. Split and redo the
647 * current region - the new bottom half.
650 rgn->size -= end - rbase;
651 type->total_size -= end - rbase;
652 memblock_insert_region(type, i--, rbase, end - rbase,
653 memblock_get_region_node(rgn),
656 /* @rgn is fully contained, record it */
666 int __init_memblock memblock_remove_range(struct memblock_type *type,
667 phys_addr_t base, phys_addr_t size)
669 int start_rgn, end_rgn;
672 ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
676 for (i = end_rgn - 1; i >= start_rgn; i--)
677 memblock_remove_region(type, i);
681 int __init_memblock memblock_remove(phys_addr_t base, phys_addr_t size)
683 return memblock_remove_range(&memblock.memory, base, size);
687 int __init_memblock memblock_free(phys_addr_t base, phys_addr_t size)
689 memblock_dbg(" memblock_free: [%#016llx-%#016llx] %pF\n",
690 (unsigned long long)base,
691 (unsigned long long)base + size - 1,
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)
821 *out_start = m_start;
827 *idx = (u32)idx_a | (u64)idx_b << 32;
831 /* scan areas before each reservation */
832 for (; idx_b < type_b->cnt + 1; idx_b++) {
833 struct memblock_region *r;
837 r = &type_b->regions[idx_b];
838 r_start = idx_b ? r[-1].base + r[-1].size : 0;
839 r_end = idx_b < type_b->cnt ?
840 r->base : ULLONG_MAX;
843 * if idx_b advanced past idx_a,
844 * break out to advance idx_a
846 if (r_start >= m_end)
848 /* if the two regions intersect, we're done */
849 if (m_start < r_end) {
852 max(m_start, r_start);
854 *out_end = min(m_end, r_end);
858 * The region which ends first is
859 * advanced for the next iteration.
865 *idx = (u32)idx_a | (u64)idx_b << 32;
871 /* signal end of iteration */
876 * __next_mem_range_rev - generic next function for for_each_*_range_rev()
878 * Finds the next range from type_a which is not marked as unsuitable
881 * @idx: pointer to u64 loop variable
882 * @nid: nid: node selector, %NUMA_NO_NODE for all nodes
883 * @type_a: pointer to memblock_type from where the range is taken
884 * @type_b: pointer to memblock_type which excludes memory from being taken
885 * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL
886 * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL
887 * @out_nid: ptr to int for nid of the range, can be %NULL
889 * Reverse of __next_mem_range().
891 void __init_memblock __next_mem_range_rev(u64 *idx, int nid,
892 struct memblock_type *type_a,
893 struct memblock_type *type_b,
894 phys_addr_t *out_start,
895 phys_addr_t *out_end, int *out_nid)
897 int idx_a = *idx & 0xffffffff;
898 int idx_b = *idx >> 32;
900 if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
903 if (*idx == (u64)ULLONG_MAX) {
904 idx_a = type_a->cnt - 1;
908 for (; idx_a >= 0; idx_a--) {
909 struct memblock_region *m = &type_a->regions[idx_a];
911 phys_addr_t m_start = m->base;
912 phys_addr_t m_end = m->base + m->size;
913 int m_nid = memblock_get_region_node(m);
915 /* only memory regions are associated with nodes, check it */
916 if (nid != NUMA_NO_NODE && nid != m_nid)
919 /* skip hotpluggable memory regions if needed */
920 if (movable_node_is_enabled() && memblock_is_hotpluggable(m))
925 *out_start = m_start;
931 *idx = (u32)idx_a | (u64)idx_b << 32;
935 /* scan areas before each reservation */
936 for (; idx_b >= 0; idx_b--) {
937 struct memblock_region *r;
941 r = &type_b->regions[idx_b];
942 r_start = idx_b ? r[-1].base + r[-1].size : 0;
943 r_end = idx_b < type_b->cnt ?
944 r->base : ULLONG_MAX;
946 * if idx_b advanced past idx_a,
947 * break out to advance idx_a
950 if (r_end <= m_start)
952 /* if the two regions intersect, we're done */
953 if (m_end > r_start) {
955 *out_start = max(m_start, r_start);
957 *out_end = min(m_end, r_end);
960 if (m_start >= r_start)
964 *idx = (u32)idx_a | (u64)idx_b << 32;
969 /* signal end of iteration */
973 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
975 * Common iterator interface used to define for_each_mem_range().
977 void __init_memblock __next_mem_pfn_range(int *idx, int nid,
978 unsigned long *out_start_pfn,
979 unsigned long *out_end_pfn, int *out_nid)
981 struct memblock_type *type = &memblock.memory;
982 struct memblock_region *r;
984 while (++*idx < type->cnt) {
985 r = &type->regions[*idx];
987 if (PFN_UP(r->base) >= PFN_DOWN(r->base + r->size))
989 if (nid == MAX_NUMNODES || nid == r->nid)
992 if (*idx >= type->cnt) {
998 *out_start_pfn = PFN_UP(r->base);
1000 *out_end_pfn = PFN_DOWN(r->base + r->size);
1006 * memblock_set_node - set node ID on memblock regions
1007 * @base: base of area to set node ID for
1008 * @size: size of area to set node ID for
1009 * @type: memblock type to set node ID for
1010 * @nid: node ID to set
1012 * Set the nid of memblock @type regions in [@base,@base+@size) to @nid.
1013 * Regions which cross the area boundaries are split as necessary.
1016 * 0 on success, -errno on failure.
1018 int __init_memblock memblock_set_node(phys_addr_t base, phys_addr_t size,
1019 struct memblock_type *type, int nid)
1021 int start_rgn, end_rgn;
1024 ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
1028 for (i = start_rgn; i < end_rgn; i++)
1029 memblock_set_region_node(&type->regions[i], nid);
1031 memblock_merge_regions(type);
1034 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
1036 static phys_addr_t __init memblock_alloc_range_nid(phys_addr_t size,
1037 phys_addr_t align, phys_addr_t start,
1038 phys_addr_t end, int nid)
1043 align = SMP_CACHE_BYTES;
1045 found = memblock_find_in_range_node(size, align, start, end, nid);
1046 if (found && !memblock_reserve(found, size))
1052 phys_addr_t __init memblock_alloc_range(phys_addr_t size, phys_addr_t align,
1053 phys_addr_t start, phys_addr_t end)
1055 return memblock_alloc_range_nid(size, align, start, end, NUMA_NO_NODE);
1058 static phys_addr_t __init memblock_alloc_base_nid(phys_addr_t size,
1059 phys_addr_t align, phys_addr_t max_addr,
1062 return memblock_alloc_range_nid(size, align, 0, max_addr, nid);
1065 phys_addr_t __init memblock_alloc_nid(phys_addr_t size, phys_addr_t align, int nid)
1067 return memblock_alloc_base_nid(size, align, MEMBLOCK_ALLOC_ACCESSIBLE, nid);
1070 phys_addr_t __init __memblock_alloc_base(phys_addr_t size, phys_addr_t align, phys_addr_t max_addr)
1072 return memblock_alloc_base_nid(size, align, max_addr, NUMA_NO_NODE);
1075 phys_addr_t __init memblock_alloc_base(phys_addr_t size, phys_addr_t align, phys_addr_t max_addr)
1079 alloc = __memblock_alloc_base(size, align, max_addr);
1082 panic("ERROR: Failed to allocate 0x%llx bytes below 0x%llx.\n",
1083 (unsigned long long) size, (unsigned long long) max_addr);
1088 phys_addr_t __init memblock_alloc(phys_addr_t size, phys_addr_t align)
1090 return memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ACCESSIBLE);
1093 phys_addr_t __init memblock_alloc_try_nid(phys_addr_t size, phys_addr_t align, int nid)
1095 phys_addr_t res = memblock_alloc_nid(size, align, nid);
1099 return memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ACCESSIBLE);
1103 * memblock_virt_alloc_internal - allocate boot memory block
1104 * @size: size of memory block to be allocated in bytes
1105 * @align: alignment of the region and block's size
1106 * @min_addr: the lower bound of the memory region to allocate (phys address)
1107 * @max_addr: the upper bound of the memory region to allocate (phys address)
1108 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1110 * The @min_addr limit is dropped if it can not be satisfied and the allocation
1111 * will fall back to memory below @min_addr. Also, allocation may fall back
1112 * to any node in the system if the specified node can not
1113 * hold the requested memory.
1115 * The allocation is performed from memory region limited by
1116 * memblock.current_limit if @max_addr == %BOOTMEM_ALLOC_ACCESSIBLE.
1118 * The memory block is aligned on SMP_CACHE_BYTES if @align == 0.
1120 * The phys address of allocated boot memory block is converted to virtual and
1121 * allocated memory is reset to 0.
1123 * In addition, function sets the min_count to 0 using kmemleak_alloc for
1124 * allocated boot memory block, so that it is never reported as leaks.
1127 * Virtual address of allocated memory block on success, NULL on failure.
1129 static void * __init memblock_virt_alloc_internal(
1130 phys_addr_t size, phys_addr_t align,
1131 phys_addr_t min_addr, phys_addr_t max_addr,
1137 if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
1141 * Detect any accidental use of these APIs after slab is ready, as at
1142 * this moment memblock may be deinitialized already and its
1143 * internal data may be destroyed (after execution of free_all_bootmem)
1145 if (WARN_ON_ONCE(slab_is_available()))
1146 return kzalloc_node(size, GFP_NOWAIT, nid);
1149 align = SMP_CACHE_BYTES;
1151 if (max_addr > memblock.current_limit)
1152 max_addr = memblock.current_limit;
1155 alloc = memblock_find_in_range_node(size, align, min_addr, max_addr,
1160 if (nid != NUMA_NO_NODE) {
1161 alloc = memblock_find_in_range_node(size, align, min_addr,
1162 max_addr, NUMA_NO_NODE);
1175 memblock_reserve(alloc, size);
1176 ptr = phys_to_virt(alloc);
1177 memset(ptr, 0, size);
1180 * The min_count is set to 0 so that bootmem allocated blocks
1181 * are never reported as leaks. This is because many of these blocks
1182 * are only referred via the physical address which is not
1183 * looked up by kmemleak.
1185 kmemleak_alloc(ptr, size, 0, 0);
1194 * memblock_virt_alloc_try_nid_nopanic - allocate boot memory block
1195 * @size: size of memory block to be allocated in bytes
1196 * @align: alignment of the region and block's size
1197 * @min_addr: the lower bound of the memory region from where the allocation
1198 * is preferred (phys address)
1199 * @max_addr: the upper bound of the memory region from where the allocation
1200 * is preferred (phys address), or %BOOTMEM_ALLOC_ACCESSIBLE to
1201 * allocate only from memory limited by memblock.current_limit value
1202 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1204 * Public version of _memblock_virt_alloc_try_nid_nopanic() which provides
1205 * additional debug information (including caller info), if enabled.
1208 * Virtual address of allocated memory block on success, NULL on failure.
1210 void * __init memblock_virt_alloc_try_nid_nopanic(
1211 phys_addr_t size, phys_addr_t align,
1212 phys_addr_t min_addr, phys_addr_t max_addr,
1215 memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=0x%llx max_addr=0x%llx %pF\n",
1216 __func__, (u64)size, (u64)align, nid, (u64)min_addr,
1217 (u64)max_addr, (void *)_RET_IP_);
1218 return memblock_virt_alloc_internal(size, align, min_addr,
1223 * memblock_virt_alloc_try_nid - allocate boot memory block with panicking
1224 * @size: size of memory block to be allocated in bytes
1225 * @align: alignment of the region and block's size
1226 * @min_addr: the lower bound of the memory region from where the allocation
1227 * is preferred (phys address)
1228 * @max_addr: the upper bound of the memory region from where the allocation
1229 * is preferred (phys address), or %BOOTMEM_ALLOC_ACCESSIBLE to
1230 * allocate only from memory limited by memblock.current_limit value
1231 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1233 * Public panicking version of _memblock_virt_alloc_try_nid_nopanic()
1234 * which provides debug information (including caller info), if enabled,
1235 * and panics if the request can not be satisfied.
1238 * Virtual address of allocated memory block on success, NULL on failure.
1240 void * __init memblock_virt_alloc_try_nid(
1241 phys_addr_t size, phys_addr_t align,
1242 phys_addr_t min_addr, phys_addr_t max_addr,
1247 memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=0x%llx max_addr=0x%llx %pF\n",
1248 __func__, (u64)size, (u64)align, nid, (u64)min_addr,
1249 (u64)max_addr, (void *)_RET_IP_);
1250 ptr = memblock_virt_alloc_internal(size, align,
1251 min_addr, max_addr, nid);
1255 panic("%s: Failed to allocate %llu bytes align=0x%llx nid=%d from=0x%llx max_addr=0x%llx\n",
1256 __func__, (u64)size, (u64)align, nid, (u64)min_addr,
1262 * __memblock_free_early - free boot memory block
1263 * @base: phys starting address of the boot memory block
1264 * @size: size of the boot memory block in bytes
1266 * Free boot memory block previously allocated by memblock_virt_alloc_xx() API.
1267 * The freeing memory will not be released to the buddy allocator.
1269 void __init __memblock_free_early(phys_addr_t base, phys_addr_t size)
1271 memblock_dbg("%s: [%#016llx-%#016llx] %pF\n",
1272 __func__, (u64)base, (u64)base + size - 1,
1274 kmemleak_free_part(__va(base), size);
1275 memblock_remove_range(&memblock.reserved, base, size);
1279 * __memblock_free_late - free bootmem block pages directly to buddy allocator
1280 * @addr: phys starting address of the boot memory block
1281 * @size: size of the boot memory block in bytes
1283 * This is only useful when the bootmem allocator has already been torn
1284 * down, but we are still initializing the system. Pages are released directly
1285 * to the buddy allocator, no bootmem metadata is updated because it is gone.
1287 void __init __memblock_free_late(phys_addr_t base, phys_addr_t size)
1291 memblock_dbg("%s: [%#016llx-%#016llx] %pF\n",
1292 __func__, (u64)base, (u64)base + size - 1,
1294 kmemleak_free_part(__va(base), size);
1295 cursor = PFN_UP(base);
1296 end = PFN_DOWN(base + size);
1298 for (; cursor < end; cursor++) {
1299 __free_pages_bootmem(pfn_to_page(cursor), 0);
1305 * Remaining API functions
1308 phys_addr_t __init memblock_phys_mem_size(void)
1310 return memblock.memory.total_size;
1313 phys_addr_t __init memblock_mem_size(unsigned long limit_pfn)
1315 unsigned long pages = 0;
1316 struct memblock_region *r;
1317 unsigned long start_pfn, end_pfn;
1319 for_each_memblock(memory, r) {
1320 start_pfn = memblock_region_memory_base_pfn(r);
1321 end_pfn = memblock_region_memory_end_pfn(r);
1322 start_pfn = min_t(unsigned long, start_pfn, limit_pfn);
1323 end_pfn = min_t(unsigned long, end_pfn, limit_pfn);
1324 pages += end_pfn - start_pfn;
1327 return PFN_PHYS(pages);
1330 /* lowest address */
1331 phys_addr_t __init_memblock memblock_start_of_DRAM(void)
1333 return memblock.memory.regions[0].base;
1336 phys_addr_t __init_memblock memblock_end_of_DRAM(void)
1338 int idx = memblock.memory.cnt - 1;
1340 return (memblock.memory.regions[idx].base + memblock.memory.regions[idx].size);
1343 void __init memblock_enforce_memory_limit(phys_addr_t limit)
1345 phys_addr_t max_addr = (phys_addr_t)ULLONG_MAX;
1346 struct memblock_region *r;
1351 /* find out max address */
1352 for_each_memblock(memory, r) {
1353 if (limit <= r->size) {
1354 max_addr = r->base + limit;
1360 /* truncate both memory and reserved regions */
1361 memblock_remove_range(&memblock.memory, max_addr,
1362 (phys_addr_t)ULLONG_MAX);
1363 memblock_remove_range(&memblock.reserved, max_addr,
1364 (phys_addr_t)ULLONG_MAX);
1367 static int __init_memblock memblock_search(struct memblock_type *type, phys_addr_t addr)
1369 unsigned int left = 0, right = type->cnt;
1372 unsigned int mid = (right + left) / 2;
1374 if (addr < type->regions[mid].base)
1376 else if (addr >= (type->regions[mid].base +
1377 type->regions[mid].size))
1381 } while (left < right);
1385 int __init memblock_is_reserved(phys_addr_t addr)
1387 return memblock_search(&memblock.reserved, addr) != -1;
1390 int __init_memblock memblock_is_memory(phys_addr_t addr)
1392 return memblock_search(&memblock.memory, addr) != -1;
1395 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
1396 int __init_memblock memblock_search_pfn_nid(unsigned long pfn,
1397 unsigned long *start_pfn, unsigned long *end_pfn)
1399 struct memblock_type *type = &memblock.memory;
1400 int mid = memblock_search(type, PFN_PHYS(pfn));
1405 *start_pfn = PFN_DOWN(type->regions[mid].base);
1406 *end_pfn = PFN_DOWN(type->regions[mid].base + type->regions[mid].size);
1408 return type->regions[mid].nid;
1413 * memblock_is_region_memory - check if a region is a subset of memory
1414 * @base: base of region to check
1415 * @size: size of region to check
1417 * Check if the region [@base, @base+@size) is a subset of a memory block.
1420 * 0 if false, non-zero if true
1422 int __init_memblock memblock_is_region_memory(phys_addr_t base, phys_addr_t size)
1424 int idx = memblock_search(&memblock.memory, base);
1425 phys_addr_t end = base + memblock_cap_size(base, &size);
1429 return memblock.memory.regions[idx].base <= base &&
1430 (memblock.memory.regions[idx].base +
1431 memblock.memory.regions[idx].size) >= end;
1435 * memblock_is_region_reserved - check if a region intersects reserved memory
1436 * @base: base of region to check
1437 * @size: size of region to check
1439 * Check if the region [@base, @base+@size) intersects a reserved memory block.
1442 * 0 if false, non-zero if true
1444 int __init_memblock memblock_is_region_reserved(phys_addr_t base, phys_addr_t size)
1446 memblock_cap_size(base, &size);
1447 return memblock_overlaps_region(&memblock.reserved, base, size) >= 0;
1450 void __init_memblock memblock_trim_memory(phys_addr_t align)
1452 phys_addr_t start, end, orig_start, orig_end;
1453 struct memblock_region *r;
1455 for_each_memblock(memory, r) {
1456 orig_start = r->base;
1457 orig_end = r->base + r->size;
1458 start = round_up(orig_start, align);
1459 end = round_down(orig_end, align);
1461 if (start == orig_start && end == orig_end)
1466 r->size = end - start;
1468 memblock_remove_region(&memblock.memory,
1469 r - memblock.memory.regions);
1475 void __init_memblock memblock_set_current_limit(phys_addr_t limit)
1477 memblock.current_limit = limit;
1480 phys_addr_t __init_memblock memblock_get_current_limit(void)
1482 return memblock.current_limit;
1485 static void __init_memblock memblock_dump(struct memblock_type *type, char *name)
1487 unsigned long long base, size;
1488 unsigned long flags;
1491 pr_info(" %s.cnt = 0x%lx\n", name, type->cnt);
1493 for (i = 0; i < type->cnt; i++) {
1494 struct memblock_region *rgn = &type->regions[i];
1495 char nid_buf[32] = "";
1500 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
1501 if (memblock_get_region_node(rgn) != MAX_NUMNODES)
1502 snprintf(nid_buf, sizeof(nid_buf), " on node %d",
1503 memblock_get_region_node(rgn));
1505 pr_info(" %s[%#x]\t[%#016llx-%#016llx], %#llx bytes%s flags: %#lx\n",
1506 name, i, base, base + size - 1, size, nid_buf, flags);
1510 void __init_memblock __memblock_dump_all(void)
1512 pr_info("MEMBLOCK configuration:\n");
1513 pr_info(" memory size = %#llx reserved size = %#llx\n",
1514 (unsigned long long)memblock.memory.total_size,
1515 (unsigned long long)memblock.reserved.total_size);
1517 memblock_dump(&memblock.memory, "memory");
1518 memblock_dump(&memblock.reserved, "reserved");
1521 void __init memblock_allow_resize(void)
1523 memblock_can_resize = 1;
1526 static int __init early_memblock(char *p)
1528 if (p && strstr(p, "debug"))
1532 early_param("memblock", early_memblock);
1534 #if defined(CONFIG_DEBUG_FS) && !defined(CONFIG_ARCH_DISCARD_MEMBLOCK)
1536 static int memblock_debug_show(struct seq_file *m, void *private)
1538 struct memblock_type *type = m->private;
1539 struct memblock_region *reg;
1542 for (i = 0; i < type->cnt; i++) {
1543 reg = &type->regions[i];
1544 seq_printf(m, "%4d: ", i);
1545 if (sizeof(phys_addr_t) == 4)
1546 seq_printf(m, "0x%08lx..0x%08lx\n",
1547 (unsigned long)reg->base,
1548 (unsigned long)(reg->base + reg->size - 1));
1550 seq_printf(m, "0x%016llx..0x%016llx\n",
1551 (unsigned long long)reg->base,
1552 (unsigned long long)(reg->base + reg->size - 1));
1558 static int memblock_debug_open(struct inode *inode, struct file *file)
1560 return single_open(file, memblock_debug_show, inode->i_private);
1563 static const struct file_operations memblock_debug_fops = {
1564 .open = memblock_debug_open,
1566 .llseek = seq_lseek,
1567 .release = single_release,
1570 static int __init memblock_init_debugfs(void)
1572 struct dentry *root = debugfs_create_dir("memblock", NULL);
1575 debugfs_create_file("memory", S_IRUGO, root, &memblock.memory, &memblock_debug_fops);
1576 debugfs_create_file("reserved", S_IRUGO, root, &memblock.reserved, &memblock_debug_fops);
1577 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
1578 debugfs_create_file("physmem", S_IRUGO, root, &memblock.physmem, &memblock_debug_fops);
1583 __initcall(memblock_init_debugfs);
1585 #endif /* CONFIG_DEBUG_FS */