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>
25 static struct memblock_region memblock_memory_init_regions[INIT_MEMBLOCK_REGIONS] __initdata_memblock;
26 static struct memblock_region memblock_reserved_init_regions[INIT_MEMBLOCK_REGIONS] __initdata_memblock;
28 struct memblock memblock __initdata_memblock = {
29 .memory.regions = memblock_memory_init_regions,
30 .memory.cnt = 1, /* empty dummy entry */
31 .memory.max = INIT_MEMBLOCK_REGIONS,
33 .reserved.regions = memblock_reserved_init_regions,
34 .reserved.cnt = 1, /* empty dummy entry */
35 .reserved.max = INIT_MEMBLOCK_REGIONS,
38 .current_limit = MEMBLOCK_ALLOC_ANYWHERE,
41 int memblock_debug __initdata_memblock;
42 #ifdef CONFIG_MOVABLE_NODE
43 bool movable_node_enabled __initdata_memblock = false;
45 static int memblock_can_resize __initdata_memblock;
46 static int memblock_memory_in_slab __initdata_memblock = 0;
47 static int memblock_reserved_in_slab __initdata_memblock = 0;
49 /* inline so we don't get a warning when pr_debug is compiled out */
50 static __init_memblock const char *
51 memblock_type_name(struct memblock_type *type)
53 if (type == &memblock.memory)
55 else if (type == &memblock.reserved)
61 /* adjust *@size so that (@base + *@size) doesn't overflow, return new size */
62 static inline phys_addr_t memblock_cap_size(phys_addr_t base, phys_addr_t *size)
64 return *size = min(*size, (phys_addr_t)ULLONG_MAX - base);
68 * Address comparison utilities
70 static unsigned long __init_memblock memblock_addrs_overlap(phys_addr_t base1, phys_addr_t size1,
71 phys_addr_t base2, phys_addr_t size2)
73 return ((base1 < (base2 + size2)) && (base2 < (base1 + size1)));
76 static long __init_memblock memblock_overlaps_region(struct memblock_type *type,
77 phys_addr_t base, phys_addr_t size)
81 for (i = 0; i < type->cnt; i++) {
82 phys_addr_t rgnbase = type->regions[i].base;
83 phys_addr_t rgnsize = type->regions[i].size;
84 if (memblock_addrs_overlap(base, size, rgnbase, rgnsize))
88 return (i < type->cnt) ? i : -1;
92 * __memblock_find_range_bottom_up - find free area utility in bottom-up
93 * @start: start of candidate range
94 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_{ANYWHERE|ACCESSIBLE}
95 * @size: size of free area to find
96 * @align: alignment of free area to find
97 * @nid: nid of the free area to find, %MAX_NUMNODES for any node
99 * Utility called from memblock_find_in_range_node(), find free area bottom-up.
102 * Found address on success, 0 on failure.
104 static phys_addr_t __init_memblock
105 __memblock_find_range_bottom_up(phys_addr_t start, phys_addr_t end,
106 phys_addr_t size, phys_addr_t align, int nid)
108 phys_addr_t this_start, this_end, cand;
111 for_each_free_mem_range(i, nid, &this_start, &this_end, NULL) {
112 this_start = clamp(this_start, start, end);
113 this_end = clamp(this_end, start, end);
115 cand = round_up(this_start, align);
116 if (cand < this_end && this_end - cand >= size)
124 * __memblock_find_range_top_down - find free area utility, in top-down
125 * @start: start of candidate range
126 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_{ANYWHERE|ACCESSIBLE}
127 * @size: size of free area to find
128 * @align: alignment of free area to find
129 * @nid: nid of the free area to find, %MAX_NUMNODES for any node
131 * Utility called from memblock_find_in_range_node(), find free area top-down.
134 * Found address on success, 0 on failure.
136 static phys_addr_t __init_memblock
137 __memblock_find_range_top_down(phys_addr_t start, phys_addr_t end,
138 phys_addr_t size, phys_addr_t align, int nid)
140 phys_addr_t this_start, this_end, cand;
143 for_each_free_mem_range_reverse(i, nid, &this_start, &this_end, NULL) {
144 this_start = clamp(this_start, start, end);
145 this_end = clamp(this_end, start, end);
150 cand = round_down(this_end - size, align);
151 if (cand >= this_start)
159 * memblock_find_in_range_node - find free area in given range and node
160 * @start: start of candidate range
161 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_{ANYWHERE|ACCESSIBLE}
162 * @size: size of free area to find
163 * @align: alignment of free area to find
164 * @nid: nid of the free area to find, %MAX_NUMNODES for any node
166 * Find @size free area aligned to @align in the specified range and node.
168 * When allocation direction is bottom-up, the @start should be greater
169 * than the end of the kernel image. Otherwise, it will be trimmed. The
170 * reason is that we want the bottom-up allocation just near the kernel
171 * image so it is highly likely that the allocated memory and the kernel
172 * will reside in the same node.
174 * If bottom-up allocation failed, will try to allocate memory top-down.
177 * Found address on success, 0 on failure.
179 phys_addr_t __init_memblock memblock_find_in_range_node(phys_addr_t start,
180 phys_addr_t end, phys_addr_t size,
181 phys_addr_t align, int nid)
184 phys_addr_t kernel_end;
187 if (end == MEMBLOCK_ALLOC_ACCESSIBLE)
188 end = memblock.current_limit;
190 /* avoid allocating the first page */
191 start = max_t(phys_addr_t, start, PAGE_SIZE);
192 end = max(start, end);
193 kernel_end = __pa_symbol(_end);
196 * try bottom-up allocation only when bottom-up mode
197 * is set and @end is above the kernel image.
199 if (memblock_bottom_up() && end > kernel_end) {
200 phys_addr_t bottom_up_start;
202 /* make sure we will allocate above the kernel */
203 bottom_up_start = max(start, kernel_end);
205 /* ok, try bottom-up allocation first */
206 ret = __memblock_find_range_bottom_up(bottom_up_start, end,
212 * we always limit bottom-up allocation above the kernel,
213 * but top-down allocation doesn't have the limit, so
214 * retrying top-down allocation may succeed when bottom-up
217 * bottom-up allocation is expected to be fail very rarely,
218 * so we use WARN_ONCE() here to see the stack trace if
221 WARN_ONCE(1, "memblock: bottom-up allocation failed, "
222 "memory hotunplug may be affected\n");
225 return __memblock_find_range_top_down(start, end, size, align, nid);
229 * memblock_find_in_range - find free area in given range
230 * @start: start of candidate range
231 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_{ANYWHERE|ACCESSIBLE}
232 * @size: size of free area to find
233 * @align: alignment of free area to find
235 * Find @size free area aligned to @align in the specified range.
238 * Found address on success, 0 on failure.
240 phys_addr_t __init_memblock memblock_find_in_range(phys_addr_t start,
241 phys_addr_t end, phys_addr_t size,
244 return memblock_find_in_range_node(start, end, size, align,
248 static void __init_memblock memblock_remove_region(struct memblock_type *type, unsigned long r)
250 type->total_size -= type->regions[r].size;
251 memmove(&type->regions[r], &type->regions[r + 1],
252 (type->cnt - (r + 1)) * sizeof(type->regions[r]));
255 /* Special case for empty arrays */
256 if (type->cnt == 0) {
257 WARN_ON(type->total_size != 0);
259 type->regions[0].base = 0;
260 type->regions[0].size = 0;
261 type->regions[0].flags = 0;
262 memblock_set_region_node(&type->regions[0], MAX_NUMNODES);
266 phys_addr_t __init_memblock get_allocated_memblock_reserved_regions_info(
269 if (memblock.reserved.regions == memblock_reserved_init_regions)
272 *addr = __pa(memblock.reserved.regions);
274 return PAGE_ALIGN(sizeof(struct memblock_region) *
275 memblock.reserved.max);
279 * memblock_double_array - double the size of the memblock regions array
280 * @type: memblock type of the regions array being doubled
281 * @new_area_start: starting address of memory range to avoid overlap with
282 * @new_area_size: size of memory range to avoid overlap with
284 * Double the size of the @type regions array. If memblock is being used to
285 * allocate memory for a new reserved regions array and there is a previously
286 * allocated memory range [@new_area_start,@new_area_start+@new_area_size]
287 * waiting to be reserved, ensure the memory used by the new array does
291 * 0 on success, -1 on failure.
293 static int __init_memblock memblock_double_array(struct memblock_type *type,
294 phys_addr_t new_area_start,
295 phys_addr_t new_area_size)
297 struct memblock_region *new_array, *old_array;
298 phys_addr_t old_alloc_size, new_alloc_size;
299 phys_addr_t old_size, new_size, addr;
300 int use_slab = slab_is_available();
303 /* We don't allow resizing until we know about the reserved regions
304 * of memory that aren't suitable for allocation
306 if (!memblock_can_resize)
309 /* Calculate new doubled size */
310 old_size = type->max * sizeof(struct memblock_region);
311 new_size = old_size << 1;
313 * We need to allocated new one align to PAGE_SIZE,
314 * so we can free them completely later.
316 old_alloc_size = PAGE_ALIGN(old_size);
317 new_alloc_size = PAGE_ALIGN(new_size);
319 /* Retrieve the slab flag */
320 if (type == &memblock.memory)
321 in_slab = &memblock_memory_in_slab;
323 in_slab = &memblock_reserved_in_slab;
325 /* Try to find some space for it.
327 * WARNING: We assume that either slab_is_available() and we use it or
328 * we use MEMBLOCK for allocations. That means that this is unsafe to
329 * use when bootmem is currently active (unless bootmem itself is
330 * implemented on top of MEMBLOCK which isn't the case yet)
332 * This should however not be an issue for now, as we currently only
333 * call into MEMBLOCK while it's still active, or much later when slab
334 * is active for memory hotplug operations
337 new_array = kmalloc(new_size, GFP_KERNEL);
338 addr = new_array ? __pa(new_array) : 0;
340 /* only exclude range when trying to double reserved.regions */
341 if (type != &memblock.reserved)
342 new_area_start = new_area_size = 0;
344 addr = memblock_find_in_range(new_area_start + new_area_size,
345 memblock.current_limit,
346 new_alloc_size, PAGE_SIZE);
347 if (!addr && new_area_size)
348 addr = memblock_find_in_range(0,
349 min(new_area_start, memblock.current_limit),
350 new_alloc_size, PAGE_SIZE);
352 new_array = addr ? __va(addr) : NULL;
355 pr_err("memblock: Failed to double %s array from %ld to %ld entries !\n",
356 memblock_type_name(type), type->max, type->max * 2);
360 memblock_dbg("memblock: %s is doubled to %ld at [%#010llx-%#010llx]",
361 memblock_type_name(type), type->max * 2, (u64)addr,
362 (u64)addr + new_size - 1);
365 * Found space, we now need to move the array over before we add the
366 * reserved region since it may be our reserved array itself that is
369 memcpy(new_array, type->regions, old_size);
370 memset(new_array + type->max, 0, old_size);
371 old_array = type->regions;
372 type->regions = new_array;
375 /* Free old array. We needn't free it if the array is the static one */
378 else if (old_array != memblock_memory_init_regions &&
379 old_array != memblock_reserved_init_regions)
380 memblock_free(__pa(old_array), old_alloc_size);
383 * Reserve the new array if that comes from the memblock. Otherwise, we
387 BUG_ON(memblock_reserve(addr, new_alloc_size));
389 /* Update slab flag */
396 * memblock_merge_regions - merge neighboring compatible regions
397 * @type: memblock type to scan
399 * Scan @type and merge neighboring compatible regions.
401 static void __init_memblock memblock_merge_regions(struct memblock_type *type)
405 /* cnt never goes below 1 */
406 while (i < type->cnt - 1) {
407 struct memblock_region *this = &type->regions[i];
408 struct memblock_region *next = &type->regions[i + 1];
410 if (this->base + this->size != next->base ||
411 memblock_get_region_node(this) !=
412 memblock_get_region_node(next) ||
413 this->flags != next->flags) {
414 BUG_ON(this->base + this->size > next->base);
419 this->size += next->size;
420 /* move forward from next + 1, index of which is i + 2 */
421 memmove(next, next + 1, (type->cnt - (i + 2)) * sizeof(*next));
427 * memblock_insert_region - insert new memblock region
428 * @type: memblock type to insert into
429 * @idx: index for the insertion point
430 * @base: base address of the new region
431 * @size: size of the new region
432 * @nid: node id of the new region
433 * @flags: flags of the new region
435 * Insert new memblock region [@base,@base+@size) into @type at @idx.
436 * @type must already have extra room to accomodate the new region.
438 static void __init_memblock memblock_insert_region(struct memblock_type *type,
439 int idx, phys_addr_t base,
441 int nid, unsigned long flags)
443 struct memblock_region *rgn = &type->regions[idx];
445 BUG_ON(type->cnt >= type->max);
446 memmove(rgn + 1, rgn, (type->cnt - idx) * sizeof(*rgn));
450 memblock_set_region_node(rgn, nid);
452 type->total_size += size;
456 * memblock_add_region - add new memblock region
457 * @type: memblock type to add new region into
458 * @base: base address of the new region
459 * @size: size of the new region
460 * @nid: nid of the new region
461 * @flags: flags of the new region
463 * Add new memblock region [@base,@base+@size) into @type. The new region
464 * is allowed to overlap with existing ones - overlaps don't affect already
465 * existing regions. @type is guaranteed to be minimal (all neighbouring
466 * compatible regions are merged) after the addition.
469 * 0 on success, -errno on failure.
471 static int __init_memblock memblock_add_region(struct memblock_type *type,
472 phys_addr_t base, phys_addr_t size,
473 int nid, unsigned long flags)
476 phys_addr_t obase = base;
477 phys_addr_t end = base + memblock_cap_size(base, &size);
483 /* special case for empty array */
484 if (type->regions[0].size == 0) {
485 WARN_ON(type->cnt != 1 || type->total_size);
486 type->regions[0].base = base;
487 type->regions[0].size = size;
488 type->regions[0].flags = flags;
489 memblock_set_region_node(&type->regions[0], nid);
490 type->total_size = size;
495 * The following is executed twice. Once with %false @insert and
496 * then with %true. The first counts the number of regions needed
497 * to accomodate the new area. The second actually inserts them.
502 for (i = 0; i < type->cnt; i++) {
503 struct memblock_region *rgn = &type->regions[i];
504 phys_addr_t rbase = rgn->base;
505 phys_addr_t rend = rbase + rgn->size;
512 * @rgn overlaps. If it separates the lower part of new
513 * area, insert that portion.
518 memblock_insert_region(type, i++, base,
522 /* area below @rend is dealt with, forget about it */
523 base = min(rend, end);
526 /* insert the remaining portion */
530 memblock_insert_region(type, i, base, end - base,
535 * If this was the first round, resize array and repeat for actual
536 * insertions; otherwise, merge and return.
539 while (type->cnt + nr_new > type->max)
540 if (memblock_double_array(type, obase, size) < 0)
545 memblock_merge_regions(type);
550 int __init_memblock memblock_add_node(phys_addr_t base, phys_addr_t size,
553 return memblock_add_region(&memblock.memory, base, size, nid, 0);
556 int __init_memblock memblock_add(phys_addr_t base, phys_addr_t size)
558 return memblock_add_region(&memblock.memory, base, size,
563 * memblock_isolate_range - isolate given range into disjoint memblocks
564 * @type: memblock type to isolate range for
565 * @base: base of range to isolate
566 * @size: size of range to isolate
567 * @start_rgn: out parameter for the start of isolated region
568 * @end_rgn: out parameter for the end of isolated region
570 * Walk @type and ensure that regions don't cross the boundaries defined by
571 * [@base,@base+@size). Crossing regions are split at the boundaries,
572 * which may create at most two more regions. The index of the first
573 * region inside the range is returned in *@start_rgn and end in *@end_rgn.
576 * 0 on success, -errno on failure.
578 static int __init_memblock memblock_isolate_range(struct memblock_type *type,
579 phys_addr_t base, phys_addr_t size,
580 int *start_rgn, int *end_rgn)
582 phys_addr_t end = base + memblock_cap_size(base, &size);
585 *start_rgn = *end_rgn = 0;
590 /* we'll create at most two more regions */
591 while (type->cnt + 2 > type->max)
592 if (memblock_double_array(type, base, size) < 0)
595 for (i = 0; i < type->cnt; i++) {
596 struct memblock_region *rgn = &type->regions[i];
597 phys_addr_t rbase = rgn->base;
598 phys_addr_t rend = rbase + rgn->size;
607 * @rgn intersects from below. Split and continue
608 * to process the next region - the new top half.
611 rgn->size -= base - rbase;
612 type->total_size -= base - rbase;
613 memblock_insert_region(type, i, rbase, base - rbase,
614 memblock_get_region_node(rgn),
616 } else if (rend > end) {
618 * @rgn intersects from above. Split and redo the
619 * current region - the new bottom half.
622 rgn->size -= end - rbase;
623 type->total_size -= end - rbase;
624 memblock_insert_region(type, i--, rbase, end - rbase,
625 memblock_get_region_node(rgn),
628 /* @rgn is fully contained, record it */
638 static int __init_memblock __memblock_remove(struct memblock_type *type,
639 phys_addr_t base, phys_addr_t size)
641 int start_rgn, end_rgn;
644 ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
648 for (i = end_rgn - 1; i >= start_rgn; i--)
649 memblock_remove_region(type, i);
653 int __init_memblock memblock_remove(phys_addr_t base, phys_addr_t size)
655 return __memblock_remove(&memblock.memory, base, size);
658 int __init_memblock memblock_free(phys_addr_t base, phys_addr_t size)
660 memblock_dbg(" memblock_free: [%#016llx-%#016llx] %pF\n",
661 (unsigned long long)base,
662 (unsigned long long)base + size - 1,
665 return __memblock_remove(&memblock.reserved, base, size);
668 static int __init_memblock memblock_reserve_region(phys_addr_t base,
673 struct memblock_type *_rgn = &memblock.reserved;
675 memblock_dbg("memblock_reserve: [%#016llx-%#016llx] flags %#02lx %pF\n",
676 (unsigned long long)base,
677 (unsigned long long)base + size - 1,
678 flags, (void *)_RET_IP_);
680 return memblock_add_region(_rgn, base, size, nid, flags);
683 int __init_memblock memblock_reserve(phys_addr_t base, phys_addr_t size)
685 return memblock_reserve_region(base, size, MAX_NUMNODES, 0);
689 * memblock_mark_hotplug - Mark hotpluggable memory with flag MEMBLOCK_HOTPLUG.
690 * @base: the base phys addr of the region
691 * @size: the size of the region
693 * This function isolates region [@base, @base + @size), and mark it with flag
696 * Return 0 on succees, -errno on failure.
698 int __init_memblock memblock_mark_hotplug(phys_addr_t base, phys_addr_t size)
700 struct memblock_type *type = &memblock.memory;
701 int i, ret, start_rgn, end_rgn;
703 ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
707 for (i = start_rgn; i < end_rgn; i++)
708 memblock_set_region_flags(&type->regions[i], MEMBLOCK_HOTPLUG);
710 memblock_merge_regions(type);
715 * memblock_clear_hotplug - Clear flag MEMBLOCK_HOTPLUG for a specified region.
716 * @base: the base phys addr of the region
717 * @size: the size of the region
719 * This function isolates region [@base, @base + @size), and clear flag
720 * MEMBLOCK_HOTPLUG for the isolated regions.
722 * Return 0 on succees, -errno on failure.
724 int __init_memblock memblock_clear_hotplug(phys_addr_t base, phys_addr_t size)
726 struct memblock_type *type = &memblock.memory;
727 int i, ret, start_rgn, end_rgn;
729 ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
733 for (i = start_rgn; i < end_rgn; i++)
734 memblock_clear_region_flags(&type->regions[i],
737 memblock_merge_regions(type);
742 * __next_free_mem_range - next function for for_each_free_mem_range()
743 * @idx: pointer to u64 loop variable
744 * @nid: node selector, %MAX_NUMNODES for all nodes
745 * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL
746 * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL
747 * @out_nid: ptr to int for nid of the range, can be %NULL
749 * Find the first free area from *@idx which matches @nid, fill the out
750 * parameters, and update *@idx for the next iteration. The lower 32bit of
751 * *@idx contains index into memory region and the upper 32bit indexes the
752 * areas before each reserved region. For example, if reserved regions
753 * look like the following,
755 * 0:[0-16), 1:[32-48), 2:[128-130)
757 * The upper 32bit indexes the following regions.
759 * 0:[0-0), 1:[16-32), 2:[48-128), 3:[130-MAX)
761 * As both region arrays are sorted, the function advances the two indices
762 * in lockstep and returns each intersection.
764 void __init_memblock __next_free_mem_range(u64 *idx, int nid,
765 phys_addr_t *out_start,
766 phys_addr_t *out_end, int *out_nid)
768 struct memblock_type *mem = &memblock.memory;
769 struct memblock_type *rsv = &memblock.reserved;
770 int mi = *idx & 0xffffffff;
773 for ( ; mi < mem->cnt; mi++) {
774 struct memblock_region *m = &mem->regions[mi];
775 phys_addr_t m_start = m->base;
776 phys_addr_t m_end = m->base + m->size;
778 /* only memory regions are associated with nodes, check it */
779 if (nid != MAX_NUMNODES && nid != memblock_get_region_node(m))
782 /* scan areas before each reservation for intersection */
783 for ( ; ri < rsv->cnt + 1; ri++) {
784 struct memblock_region *r = &rsv->regions[ri];
785 phys_addr_t r_start = ri ? r[-1].base + r[-1].size : 0;
786 phys_addr_t r_end = ri < rsv->cnt ? r->base : ULLONG_MAX;
788 /* if ri advanced past mi, break out to advance mi */
789 if (r_start >= m_end)
791 /* if the two regions intersect, we're done */
792 if (m_start < r_end) {
794 *out_start = max(m_start, r_start);
796 *out_end = min(m_end, r_end);
798 *out_nid = memblock_get_region_node(m);
800 * The region which ends first is advanced
801 * for the next iteration.
807 *idx = (u32)mi | (u64)ri << 32;
813 /* signal end of iteration */
818 * __next_free_mem_range_rev - next function for for_each_free_mem_range_reverse()
819 * @idx: pointer to u64 loop variable
820 * @nid: nid: node selector, %MAX_NUMNODES for all nodes
821 * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL
822 * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL
823 * @out_nid: ptr to int for nid of the range, can be %NULL
825 * Reverse of __next_free_mem_range().
827 * Linux kernel cannot migrate pages used by itself. Memory hotplug users won't
828 * be able to hot-remove hotpluggable memory used by the kernel. So this
829 * function skip hotpluggable regions if needed when allocating memory for the
832 void __init_memblock __next_free_mem_range_rev(u64 *idx, int nid,
833 phys_addr_t *out_start,
834 phys_addr_t *out_end, int *out_nid)
836 struct memblock_type *mem = &memblock.memory;
837 struct memblock_type *rsv = &memblock.reserved;
838 int mi = *idx & 0xffffffff;
841 if (*idx == (u64)ULLONG_MAX) {
846 for ( ; mi >= 0; mi--) {
847 struct memblock_region *m = &mem->regions[mi];
848 phys_addr_t m_start = m->base;
849 phys_addr_t m_end = m->base + m->size;
851 /* only memory regions are associated with nodes, check it */
852 if (nid != MAX_NUMNODES && nid != memblock_get_region_node(m))
855 /* skip hotpluggable memory regions if needed */
856 if (movable_node_is_enabled() && memblock_is_hotpluggable(m))
859 /* scan areas before each reservation for intersection */
860 for ( ; ri >= 0; ri--) {
861 struct memblock_region *r = &rsv->regions[ri];
862 phys_addr_t r_start = ri ? r[-1].base + r[-1].size : 0;
863 phys_addr_t r_end = ri < rsv->cnt ? r->base : ULLONG_MAX;
865 /* if ri advanced past mi, break out to advance mi */
866 if (r_end <= m_start)
868 /* if the two regions intersect, we're done */
869 if (m_end > r_start) {
871 *out_start = max(m_start, r_start);
873 *out_end = min(m_end, r_end);
875 *out_nid = memblock_get_region_node(m);
877 if (m_start >= r_start)
881 *idx = (u32)mi | (u64)ri << 32;
890 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
892 * Common iterator interface used to define for_each_mem_range().
894 void __init_memblock __next_mem_pfn_range(int *idx, int nid,
895 unsigned long *out_start_pfn,
896 unsigned long *out_end_pfn, int *out_nid)
898 struct memblock_type *type = &memblock.memory;
899 struct memblock_region *r;
901 while (++*idx < type->cnt) {
902 r = &type->regions[*idx];
904 if (PFN_UP(r->base) >= PFN_DOWN(r->base + r->size))
906 if (nid == MAX_NUMNODES || nid == r->nid)
909 if (*idx >= type->cnt) {
915 *out_start_pfn = PFN_UP(r->base);
917 *out_end_pfn = PFN_DOWN(r->base + r->size);
923 * memblock_set_node - set node ID on memblock regions
924 * @base: base of area to set node ID for
925 * @size: size of area to set node ID for
926 * @type: memblock type to set node ID for
927 * @nid: node ID to set
929 * Set the nid of memblock @type regions in [@base,@base+@size) to @nid.
930 * Regions which cross the area boundaries are split as necessary.
933 * 0 on success, -errno on failure.
935 int __init_memblock memblock_set_node(phys_addr_t base, phys_addr_t size,
936 struct memblock_type *type, int nid)
938 int start_rgn, end_rgn;
941 ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
945 for (i = start_rgn; i < end_rgn; i++)
946 memblock_set_region_node(&type->regions[i], nid);
948 memblock_merge_regions(type);
951 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
953 static phys_addr_t __init memblock_alloc_base_nid(phys_addr_t size,
954 phys_addr_t align, phys_addr_t max_addr,
960 align = __alignof__(long long);
962 /* align @size to avoid excessive fragmentation on reserved array */
963 size = round_up(size, align);
965 found = memblock_find_in_range_node(0, max_addr, size, align, nid);
966 if (found && !memblock_reserve(found, size))
972 phys_addr_t __init memblock_alloc_nid(phys_addr_t size, phys_addr_t align, int nid)
974 return memblock_alloc_base_nid(size, align, MEMBLOCK_ALLOC_ACCESSIBLE, nid);
977 phys_addr_t __init __memblock_alloc_base(phys_addr_t size, phys_addr_t align, phys_addr_t max_addr)
979 return memblock_alloc_base_nid(size, align, max_addr, MAX_NUMNODES);
982 phys_addr_t __init memblock_alloc_base(phys_addr_t size, phys_addr_t align, phys_addr_t max_addr)
986 alloc = __memblock_alloc_base(size, align, max_addr);
989 panic("ERROR: Failed to allocate 0x%llx bytes below 0x%llx.\n",
990 (unsigned long long) size, (unsigned long long) max_addr);
995 phys_addr_t __init memblock_alloc(phys_addr_t size, phys_addr_t align)
997 return memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ACCESSIBLE);
1000 phys_addr_t __init memblock_alloc_try_nid(phys_addr_t size, phys_addr_t align, int nid)
1002 phys_addr_t res = memblock_alloc_nid(size, align, nid);
1006 return memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ACCESSIBLE);
1011 * Remaining API functions
1014 phys_addr_t __init memblock_phys_mem_size(void)
1016 return memblock.memory.total_size;
1019 phys_addr_t __init memblock_mem_size(unsigned long limit_pfn)
1021 unsigned long pages = 0;
1022 struct memblock_region *r;
1023 unsigned long start_pfn, end_pfn;
1025 for_each_memblock(memory, r) {
1026 start_pfn = memblock_region_memory_base_pfn(r);
1027 end_pfn = memblock_region_memory_end_pfn(r);
1028 start_pfn = min_t(unsigned long, start_pfn, limit_pfn);
1029 end_pfn = min_t(unsigned long, end_pfn, limit_pfn);
1030 pages += end_pfn - start_pfn;
1033 return (phys_addr_t)pages << PAGE_SHIFT;
1036 /* lowest address */
1037 phys_addr_t __init_memblock memblock_start_of_DRAM(void)
1039 return memblock.memory.regions[0].base;
1042 phys_addr_t __init_memblock memblock_end_of_DRAM(void)
1044 int idx = memblock.memory.cnt - 1;
1046 return (memblock.memory.regions[idx].base + memblock.memory.regions[idx].size);
1049 void __init memblock_enforce_memory_limit(phys_addr_t limit)
1052 phys_addr_t max_addr = (phys_addr_t)ULLONG_MAX;
1057 /* find out max address */
1058 for (i = 0; i < memblock.memory.cnt; i++) {
1059 struct memblock_region *r = &memblock.memory.regions[i];
1061 if (limit <= r->size) {
1062 max_addr = r->base + limit;
1068 /* truncate both memory and reserved regions */
1069 __memblock_remove(&memblock.memory, max_addr, (phys_addr_t)ULLONG_MAX);
1070 __memblock_remove(&memblock.reserved, max_addr, (phys_addr_t)ULLONG_MAX);
1073 static int __init_memblock memblock_search(struct memblock_type *type, phys_addr_t addr)
1075 unsigned int left = 0, right = type->cnt;
1078 unsigned int mid = (right + left) / 2;
1080 if (addr < type->regions[mid].base)
1082 else if (addr >= (type->regions[mid].base +
1083 type->regions[mid].size))
1087 } while (left < right);
1091 int __init memblock_is_reserved(phys_addr_t addr)
1093 return memblock_search(&memblock.reserved, addr) != -1;
1096 int __init_memblock memblock_is_memory(phys_addr_t addr)
1098 return memblock_search(&memblock.memory, addr) != -1;
1101 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
1102 int __init_memblock memblock_search_pfn_nid(unsigned long pfn,
1103 unsigned long *start_pfn, unsigned long *end_pfn)
1105 struct memblock_type *type = &memblock.memory;
1106 int mid = memblock_search(type, (phys_addr_t)pfn << PAGE_SHIFT);
1111 *start_pfn = type->regions[mid].base >> PAGE_SHIFT;
1112 *end_pfn = (type->regions[mid].base + type->regions[mid].size)
1115 return type->regions[mid].nid;
1120 * memblock_is_region_memory - check if a region is a subset of memory
1121 * @base: base of region to check
1122 * @size: size of region to check
1124 * Check if the region [@base, @base+@size) is a subset of a memory block.
1127 * 0 if false, non-zero if true
1129 int __init_memblock memblock_is_region_memory(phys_addr_t base, phys_addr_t size)
1131 int idx = memblock_search(&memblock.memory, base);
1132 phys_addr_t end = base + memblock_cap_size(base, &size);
1136 return memblock.memory.regions[idx].base <= base &&
1137 (memblock.memory.regions[idx].base +
1138 memblock.memory.regions[idx].size) >= end;
1142 * memblock_is_region_reserved - check if a region intersects reserved memory
1143 * @base: base of region to check
1144 * @size: size of region to check
1146 * Check if the region [@base, @base+@size) intersects a reserved memory block.
1149 * 0 if false, non-zero if true
1151 int __init_memblock memblock_is_region_reserved(phys_addr_t base, phys_addr_t size)
1153 memblock_cap_size(base, &size);
1154 return memblock_overlaps_region(&memblock.reserved, base, size) >= 0;
1157 void __init_memblock memblock_trim_memory(phys_addr_t align)
1160 phys_addr_t start, end, orig_start, orig_end;
1161 struct memblock_type *mem = &memblock.memory;
1163 for (i = 0; i < mem->cnt; i++) {
1164 orig_start = mem->regions[i].base;
1165 orig_end = mem->regions[i].base + mem->regions[i].size;
1166 start = round_up(orig_start, align);
1167 end = round_down(orig_end, align);
1169 if (start == orig_start && end == orig_end)
1173 mem->regions[i].base = start;
1174 mem->regions[i].size = end - start;
1176 memblock_remove_region(mem, i);
1182 void __init_memblock memblock_set_current_limit(phys_addr_t limit)
1184 memblock.current_limit = limit;
1187 static void __init_memblock memblock_dump(struct memblock_type *type, char *name)
1189 unsigned long long base, size;
1190 unsigned long flags;
1193 pr_info(" %s.cnt = 0x%lx\n", name, type->cnt);
1195 for (i = 0; i < type->cnt; i++) {
1196 struct memblock_region *rgn = &type->regions[i];
1197 char nid_buf[32] = "";
1202 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
1203 if (memblock_get_region_node(rgn) != MAX_NUMNODES)
1204 snprintf(nid_buf, sizeof(nid_buf), " on node %d",
1205 memblock_get_region_node(rgn));
1207 pr_info(" %s[%#x]\t[%#016llx-%#016llx], %#llx bytes%s flags: %#lx\n",
1208 name, i, base, base + size - 1, size, nid_buf, flags);
1212 void __init_memblock __memblock_dump_all(void)
1214 pr_info("MEMBLOCK configuration:\n");
1215 pr_info(" memory size = %#llx reserved size = %#llx\n",
1216 (unsigned long long)memblock.memory.total_size,
1217 (unsigned long long)memblock.reserved.total_size);
1219 memblock_dump(&memblock.memory, "memory");
1220 memblock_dump(&memblock.reserved, "reserved");
1223 void __init memblock_allow_resize(void)
1225 memblock_can_resize = 1;
1228 static int __init early_memblock(char *p)
1230 if (p && strstr(p, "debug"))
1234 early_param("memblock", early_memblock);
1236 #if defined(CONFIG_DEBUG_FS) && !defined(CONFIG_ARCH_DISCARD_MEMBLOCK)
1238 static int memblock_debug_show(struct seq_file *m, void *private)
1240 struct memblock_type *type = m->private;
1241 struct memblock_region *reg;
1244 for (i = 0; i < type->cnt; i++) {
1245 reg = &type->regions[i];
1246 seq_printf(m, "%4d: ", i);
1247 if (sizeof(phys_addr_t) == 4)
1248 seq_printf(m, "0x%08lx..0x%08lx\n",
1249 (unsigned long)reg->base,
1250 (unsigned long)(reg->base + reg->size - 1));
1252 seq_printf(m, "0x%016llx..0x%016llx\n",
1253 (unsigned long long)reg->base,
1254 (unsigned long long)(reg->base + reg->size - 1));
1260 static int memblock_debug_open(struct inode *inode, struct file *file)
1262 return single_open(file, memblock_debug_show, inode->i_private);
1265 static const struct file_operations memblock_debug_fops = {
1266 .open = memblock_debug_open,
1268 .llseek = seq_lseek,
1269 .release = single_release,
1272 static int __init memblock_init_debugfs(void)
1274 struct dentry *root = debugfs_create_dir("memblock", NULL);
1277 debugfs_create_file("memory", S_IRUGO, root, &memblock.memory, &memblock_debug_fops);
1278 debugfs_create_file("reserved", S_IRUGO, root, &memblock.reserved, &memblock_debug_fops);
1282 __initcall(memblock_init_debugfs);
1284 #endif /* CONFIG_DEBUG_FS */