leds: triggers: define module_led_trigger helper
[linux-2.6/btrfs-unstable.git] / mm / memblock.c
blobcc16d70b8333890730d16c08b858631947e38d70
1 /*
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/kmemleak.h>
21 #include <linux/seq_file.h>
22 #include <linux/memblock.h>
23 #include <linux/bootmem.h>
25 #include <asm/sections.h>
26 #include <linux/io.h>
28 #include "internal.h"
30 static struct memblock_region memblock_memory_init_regions[INIT_MEMBLOCK_REGIONS] __initdata_memblock;
31 static struct memblock_region memblock_reserved_init_regions[INIT_MEMBLOCK_REGIONS] __initdata_memblock;
32 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
33 static struct memblock_region memblock_physmem_init_regions[INIT_PHYSMEM_REGIONS] __initdata_memblock;
34 #endif
36 struct memblock memblock __initdata_memblock = {
37 .memory.regions = memblock_memory_init_regions,
38 .memory.cnt = 1, /* empty dummy entry */
39 .memory.max = INIT_MEMBLOCK_REGIONS,
40 .memory.name = "memory",
42 .reserved.regions = memblock_reserved_init_regions,
43 .reserved.cnt = 1, /* empty dummy entry */
44 .reserved.max = INIT_MEMBLOCK_REGIONS,
45 .reserved.name = "reserved",
47 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
48 .physmem.regions = memblock_physmem_init_regions,
49 .physmem.cnt = 1, /* empty dummy entry */
50 .physmem.max = INIT_PHYSMEM_REGIONS,
51 .physmem.name = "physmem",
52 #endif
54 .bottom_up = false,
55 .current_limit = MEMBLOCK_ALLOC_ANYWHERE,
58 int memblock_debug __initdata_memblock;
59 static bool system_has_some_mirror __initdata_memblock = false;
60 static int memblock_can_resize __initdata_memblock;
61 static int memblock_memory_in_slab __initdata_memblock = 0;
62 static int memblock_reserved_in_slab __initdata_memblock = 0;
64 ulong __init_memblock choose_memblock_flags(void)
66 return system_has_some_mirror ? MEMBLOCK_MIRROR : MEMBLOCK_NONE;
69 /* adjust *@size so that (@base + *@size) doesn't overflow, return new size */
70 static inline phys_addr_t memblock_cap_size(phys_addr_t base, phys_addr_t *size)
72 return *size = min(*size, PHYS_ADDR_MAX - base);
76 * Address comparison utilities
78 static unsigned long __init_memblock memblock_addrs_overlap(phys_addr_t base1, phys_addr_t size1,
79 phys_addr_t base2, phys_addr_t size2)
81 return ((base1 < (base2 + size2)) && (base2 < (base1 + size1)));
84 bool __init_memblock memblock_overlaps_region(struct memblock_type *type,
85 phys_addr_t base, phys_addr_t size)
87 unsigned long i;
89 for (i = 0; i < type->cnt; i++)
90 if (memblock_addrs_overlap(base, size, type->regions[i].base,
91 type->regions[i].size))
92 break;
93 return i < type->cnt;
97 * __memblock_find_range_bottom_up - find free area utility in bottom-up
98 * @start: start of candidate range
99 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_{ANYWHERE|ACCESSIBLE}
100 * @size: size of free area to find
101 * @align: alignment of free area to find
102 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
103 * @flags: pick from blocks based on memory attributes
105 * Utility called from memblock_find_in_range_node(), find free area bottom-up.
107 * RETURNS:
108 * Found address on success, 0 on failure.
110 static phys_addr_t __init_memblock
111 __memblock_find_range_bottom_up(phys_addr_t start, phys_addr_t end,
112 phys_addr_t size, phys_addr_t align, int nid,
113 ulong flags)
115 phys_addr_t this_start, this_end, cand;
116 u64 i;
118 for_each_free_mem_range(i, nid, flags, &this_start, &this_end, NULL) {
119 this_start = clamp(this_start, start, end);
120 this_end = clamp(this_end, start, end);
122 cand = round_up(this_start, align);
123 if (cand < this_end && this_end - cand >= size)
124 return cand;
127 return 0;
131 * __memblock_find_range_top_down - find free area utility, in top-down
132 * @start: start of candidate range
133 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_{ANYWHERE|ACCESSIBLE}
134 * @size: size of free area to find
135 * @align: alignment of free area to find
136 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
137 * @flags: pick from blocks based on memory attributes
139 * Utility called from memblock_find_in_range_node(), find free area top-down.
141 * RETURNS:
142 * Found address on success, 0 on failure.
144 static phys_addr_t __init_memblock
145 __memblock_find_range_top_down(phys_addr_t start, phys_addr_t end,
146 phys_addr_t size, phys_addr_t align, int nid,
147 ulong flags)
149 phys_addr_t this_start, this_end, cand;
150 u64 i;
152 for_each_free_mem_range_reverse(i, nid, flags, &this_start, &this_end,
153 NULL) {
154 this_start = clamp(this_start, start, end);
155 this_end = clamp(this_end, start, end);
157 if (this_end < size)
158 continue;
160 cand = round_down(this_end - size, align);
161 if (cand >= this_start)
162 return cand;
165 return 0;
169 * memblock_find_in_range_node - find free area in given range and node
170 * @size: size of free area to find
171 * @align: alignment of free area to find
172 * @start: start of candidate range
173 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_{ANYWHERE|ACCESSIBLE}
174 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
175 * @flags: pick from blocks based on memory attributes
177 * Find @size free area aligned to @align in the specified range and node.
179 * When allocation direction is bottom-up, the @start should be greater
180 * than the end of the kernel image. Otherwise, it will be trimmed. The
181 * reason is that we want the bottom-up allocation just near the kernel
182 * image so it is highly likely that the allocated memory and the kernel
183 * will reside in the same node.
185 * If bottom-up allocation failed, will try to allocate memory top-down.
187 * RETURNS:
188 * Found address on success, 0 on failure.
190 phys_addr_t __init_memblock memblock_find_in_range_node(phys_addr_t size,
191 phys_addr_t align, phys_addr_t start,
192 phys_addr_t end, int nid, ulong flags)
194 phys_addr_t kernel_end, ret;
196 /* pump up @end */
197 if (end == MEMBLOCK_ALLOC_ACCESSIBLE)
198 end = memblock.current_limit;
200 /* avoid allocating the first page */
201 start = max_t(phys_addr_t, start, PAGE_SIZE);
202 end = max(start, end);
203 kernel_end = __pa_symbol(_end);
206 * try bottom-up allocation only when bottom-up mode
207 * is set and @end is above the kernel image.
209 if (memblock_bottom_up() && end > kernel_end) {
210 phys_addr_t bottom_up_start;
212 /* make sure we will allocate above the kernel */
213 bottom_up_start = max(start, kernel_end);
215 /* ok, try bottom-up allocation first */
216 ret = __memblock_find_range_bottom_up(bottom_up_start, end,
217 size, align, nid, flags);
218 if (ret)
219 return ret;
222 * we always limit bottom-up allocation above the kernel,
223 * but top-down allocation doesn't have the limit, so
224 * retrying top-down allocation may succeed when bottom-up
225 * allocation failed.
227 * bottom-up allocation is expected to be fail very rarely,
228 * so we use WARN_ONCE() here to see the stack trace if
229 * fail happens.
231 WARN_ONCE(1, "memblock: bottom-up allocation failed, memory hotunplug may be affected\n");
234 return __memblock_find_range_top_down(start, end, size, align, nid,
235 flags);
239 * memblock_find_in_range - find free area in given range
240 * @start: start of candidate range
241 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_{ANYWHERE|ACCESSIBLE}
242 * @size: size of free area to find
243 * @align: alignment of free area to find
245 * Find @size free area aligned to @align in the specified range.
247 * RETURNS:
248 * Found address on success, 0 on failure.
250 phys_addr_t __init_memblock memblock_find_in_range(phys_addr_t start,
251 phys_addr_t end, phys_addr_t size,
252 phys_addr_t align)
254 phys_addr_t ret;
255 ulong flags = choose_memblock_flags();
257 again:
258 ret = memblock_find_in_range_node(size, align, start, end,
259 NUMA_NO_NODE, flags);
261 if (!ret && (flags & MEMBLOCK_MIRROR)) {
262 pr_warn("Could not allocate %pap bytes of mirrored memory\n",
263 &size);
264 flags &= ~MEMBLOCK_MIRROR;
265 goto again;
268 return ret;
271 static void __init_memblock memblock_remove_region(struct memblock_type *type, unsigned long r)
273 type->total_size -= type->regions[r].size;
274 memmove(&type->regions[r], &type->regions[r + 1],
275 (type->cnt - (r + 1)) * sizeof(type->regions[r]));
276 type->cnt--;
278 /* Special case for empty arrays */
279 if (type->cnt == 0) {
280 WARN_ON(type->total_size != 0);
281 type->cnt = 1;
282 type->regions[0].base = 0;
283 type->regions[0].size = 0;
284 type->regions[0].flags = 0;
285 memblock_set_region_node(&type->regions[0], MAX_NUMNODES);
289 #ifdef CONFIG_ARCH_DISCARD_MEMBLOCK
291 * Discard memory and reserved arrays if they were allocated
293 void __init memblock_discard(void)
295 phys_addr_t addr, size;
297 if (memblock.reserved.regions != memblock_reserved_init_regions) {
298 addr = __pa(memblock.reserved.regions);
299 size = PAGE_ALIGN(sizeof(struct memblock_region) *
300 memblock.reserved.max);
301 __memblock_free_late(addr, size);
304 if (memblock.memory.regions != memblock_memory_init_regions) {
305 addr = __pa(memblock.memory.regions);
306 size = PAGE_ALIGN(sizeof(struct memblock_region) *
307 memblock.memory.max);
308 __memblock_free_late(addr, size);
311 #endif
314 * memblock_double_array - double the size of the memblock regions array
315 * @type: memblock type of the regions array being doubled
316 * @new_area_start: starting address of memory range to avoid overlap with
317 * @new_area_size: size of memory range to avoid overlap with
319 * Double the size of the @type regions array. If memblock is being used to
320 * allocate memory for a new reserved regions array and there is a previously
321 * allocated memory range [@new_area_start,@new_area_start+@new_area_size]
322 * waiting to be reserved, ensure the memory used by the new array does
323 * not overlap.
325 * RETURNS:
326 * 0 on success, -1 on failure.
328 static int __init_memblock memblock_double_array(struct memblock_type *type,
329 phys_addr_t new_area_start,
330 phys_addr_t new_area_size)
332 struct memblock_region *new_array, *old_array;
333 phys_addr_t old_alloc_size, new_alloc_size;
334 phys_addr_t old_size, new_size, addr;
335 int use_slab = slab_is_available();
336 int *in_slab;
338 /* We don't allow resizing until we know about the reserved regions
339 * of memory that aren't suitable for allocation
341 if (!memblock_can_resize)
342 return -1;
344 /* Calculate new doubled size */
345 old_size = type->max * sizeof(struct memblock_region);
346 new_size = old_size << 1;
348 * We need to allocated new one align to PAGE_SIZE,
349 * so we can free them completely later.
351 old_alloc_size = PAGE_ALIGN(old_size);
352 new_alloc_size = PAGE_ALIGN(new_size);
354 /* Retrieve the slab flag */
355 if (type == &memblock.memory)
356 in_slab = &memblock_memory_in_slab;
357 else
358 in_slab = &memblock_reserved_in_slab;
360 /* Try to find some space for it.
362 * WARNING: We assume that either slab_is_available() and we use it or
363 * we use MEMBLOCK for allocations. That means that this is unsafe to
364 * use when bootmem is currently active (unless bootmem itself is
365 * implemented on top of MEMBLOCK which isn't the case yet)
367 * This should however not be an issue for now, as we currently only
368 * call into MEMBLOCK while it's still active, or much later when slab
369 * is active for memory hotplug operations
371 if (use_slab) {
372 new_array = kmalloc(new_size, GFP_KERNEL);
373 addr = new_array ? __pa(new_array) : 0;
374 } else {
375 /* only exclude range when trying to double reserved.regions */
376 if (type != &memblock.reserved)
377 new_area_start = new_area_size = 0;
379 addr = memblock_find_in_range(new_area_start + new_area_size,
380 memblock.current_limit,
381 new_alloc_size, PAGE_SIZE);
382 if (!addr && new_area_size)
383 addr = memblock_find_in_range(0,
384 min(new_area_start, memblock.current_limit),
385 new_alloc_size, PAGE_SIZE);
387 new_array = addr ? __va(addr) : NULL;
389 if (!addr) {
390 pr_err("memblock: Failed to double %s array from %ld to %ld entries !\n",
391 type->name, type->max, type->max * 2);
392 return -1;
395 memblock_dbg("memblock: %s is doubled to %ld at [%#010llx-%#010llx]",
396 type->name, type->max * 2, (u64)addr,
397 (u64)addr + new_size - 1);
400 * Found space, we now need to move the array over before we add the
401 * reserved region since it may be our reserved array itself that is
402 * full.
404 memcpy(new_array, type->regions, old_size);
405 memset(new_array + type->max, 0, old_size);
406 old_array = type->regions;
407 type->regions = new_array;
408 type->max <<= 1;
410 /* Free old array. We needn't free it if the array is the static one */
411 if (*in_slab)
412 kfree(old_array);
413 else if (old_array != memblock_memory_init_regions &&
414 old_array != memblock_reserved_init_regions)
415 memblock_free(__pa(old_array), old_alloc_size);
418 * Reserve the new array if that comes from the memblock. Otherwise, we
419 * needn't do it
421 if (!use_slab)
422 BUG_ON(memblock_reserve(addr, new_alloc_size));
424 /* Update slab flag */
425 *in_slab = use_slab;
427 return 0;
431 * memblock_merge_regions - merge neighboring compatible regions
432 * @type: memblock type to scan
434 * Scan @type and merge neighboring compatible regions.
436 static void __init_memblock memblock_merge_regions(struct memblock_type *type)
438 int i = 0;
440 /* cnt never goes below 1 */
441 while (i < type->cnt - 1) {
442 struct memblock_region *this = &type->regions[i];
443 struct memblock_region *next = &type->regions[i + 1];
445 if (this->base + this->size != next->base ||
446 memblock_get_region_node(this) !=
447 memblock_get_region_node(next) ||
448 this->flags != next->flags) {
449 BUG_ON(this->base + this->size > next->base);
450 i++;
451 continue;
454 this->size += next->size;
455 /* move forward from next + 1, index of which is i + 2 */
456 memmove(next, next + 1, (type->cnt - (i + 2)) * sizeof(*next));
457 type->cnt--;
462 * memblock_insert_region - insert new memblock region
463 * @type: memblock type to insert into
464 * @idx: index for the insertion point
465 * @base: base address of the new region
466 * @size: size of the new region
467 * @nid: node id of the new region
468 * @flags: flags of the new region
470 * Insert new memblock region [@base,@base+@size) into @type at @idx.
471 * @type must already have extra room to accommodate the new region.
473 static void __init_memblock memblock_insert_region(struct memblock_type *type,
474 int idx, phys_addr_t base,
475 phys_addr_t size,
476 int nid, unsigned long flags)
478 struct memblock_region *rgn = &type->regions[idx];
480 BUG_ON(type->cnt >= type->max);
481 memmove(rgn + 1, rgn, (type->cnt - idx) * sizeof(*rgn));
482 rgn->base = base;
483 rgn->size = size;
484 rgn->flags = flags;
485 memblock_set_region_node(rgn, nid);
486 type->cnt++;
487 type->total_size += size;
491 * memblock_add_range - add new memblock region
492 * @type: memblock type to add new region into
493 * @base: base address of the new region
494 * @size: size of the new region
495 * @nid: nid of the new region
496 * @flags: flags of the new region
498 * Add new memblock region [@base,@base+@size) into @type. The new region
499 * is allowed to overlap with existing ones - overlaps don't affect already
500 * existing regions. @type is guaranteed to be minimal (all neighbouring
501 * compatible regions are merged) after the addition.
503 * RETURNS:
504 * 0 on success, -errno on failure.
506 int __init_memblock memblock_add_range(struct memblock_type *type,
507 phys_addr_t base, phys_addr_t size,
508 int nid, unsigned long flags)
510 bool insert = false;
511 phys_addr_t obase = base;
512 phys_addr_t end = base + memblock_cap_size(base, &size);
513 int idx, nr_new;
514 struct memblock_region *rgn;
516 if (!size)
517 return 0;
519 /* special case for empty array */
520 if (type->regions[0].size == 0) {
521 WARN_ON(type->cnt != 1 || type->total_size);
522 type->regions[0].base = base;
523 type->regions[0].size = size;
524 type->regions[0].flags = flags;
525 memblock_set_region_node(&type->regions[0], nid);
526 type->total_size = size;
527 return 0;
529 repeat:
531 * The following is executed twice. Once with %false @insert and
532 * then with %true. The first counts the number of regions needed
533 * to accommodate the new area. The second actually inserts them.
535 base = obase;
536 nr_new = 0;
538 for_each_memblock_type(idx, type, rgn) {
539 phys_addr_t rbase = rgn->base;
540 phys_addr_t rend = rbase + rgn->size;
542 if (rbase >= end)
543 break;
544 if (rend <= base)
545 continue;
547 * @rgn overlaps. If it separates the lower part of new
548 * area, insert that portion.
550 if (rbase > base) {
551 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
552 WARN_ON(nid != memblock_get_region_node(rgn));
553 #endif
554 WARN_ON(flags != rgn->flags);
555 nr_new++;
556 if (insert)
557 memblock_insert_region(type, idx++, base,
558 rbase - base, nid,
559 flags);
561 /* area below @rend is dealt with, forget about it */
562 base = min(rend, end);
565 /* insert the remaining portion */
566 if (base < end) {
567 nr_new++;
568 if (insert)
569 memblock_insert_region(type, idx, base, end - base,
570 nid, flags);
573 if (!nr_new)
574 return 0;
577 * If this was the first round, resize array and repeat for actual
578 * insertions; otherwise, merge and return.
580 if (!insert) {
581 while (type->cnt + nr_new > type->max)
582 if (memblock_double_array(type, obase, size) < 0)
583 return -ENOMEM;
584 insert = true;
585 goto repeat;
586 } else {
587 memblock_merge_regions(type);
588 return 0;
592 int __init_memblock memblock_add_node(phys_addr_t base, phys_addr_t size,
593 int nid)
595 return memblock_add_range(&memblock.memory, base, size, nid, 0);
598 int __init_memblock memblock_add(phys_addr_t base, phys_addr_t size)
600 phys_addr_t end = base + size - 1;
602 memblock_dbg("memblock_add: [%pa-%pa] %pF\n",
603 &base, &end, (void *)_RET_IP_);
605 return memblock_add_range(&memblock.memory, base, size, MAX_NUMNODES, 0);
609 * memblock_isolate_range - isolate given range into disjoint memblocks
610 * @type: memblock type to isolate range for
611 * @base: base of range to isolate
612 * @size: size of range to isolate
613 * @start_rgn: out parameter for the start of isolated region
614 * @end_rgn: out parameter for the end of isolated region
616 * Walk @type and ensure that regions don't cross the boundaries defined by
617 * [@base,@base+@size). Crossing regions are split at the boundaries,
618 * which may create at most two more regions. The index of the first
619 * region inside the range is returned in *@start_rgn and end in *@end_rgn.
621 * RETURNS:
622 * 0 on success, -errno on failure.
624 static int __init_memblock memblock_isolate_range(struct memblock_type *type,
625 phys_addr_t base, phys_addr_t size,
626 int *start_rgn, int *end_rgn)
628 phys_addr_t end = base + memblock_cap_size(base, &size);
629 int idx;
630 struct memblock_region *rgn;
632 *start_rgn = *end_rgn = 0;
634 if (!size)
635 return 0;
637 /* we'll create at most two more regions */
638 while (type->cnt + 2 > type->max)
639 if (memblock_double_array(type, base, size) < 0)
640 return -ENOMEM;
642 for_each_memblock_type(idx, type, rgn) {
643 phys_addr_t rbase = rgn->base;
644 phys_addr_t rend = rbase + rgn->size;
646 if (rbase >= end)
647 break;
648 if (rend <= base)
649 continue;
651 if (rbase < base) {
653 * @rgn intersects from below. Split and continue
654 * to process the next region - the new top half.
656 rgn->base = base;
657 rgn->size -= base - rbase;
658 type->total_size -= base - rbase;
659 memblock_insert_region(type, idx, rbase, base - rbase,
660 memblock_get_region_node(rgn),
661 rgn->flags);
662 } else if (rend > end) {
664 * @rgn intersects from above. Split and redo the
665 * current region - the new bottom half.
667 rgn->base = end;
668 rgn->size -= end - rbase;
669 type->total_size -= end - rbase;
670 memblock_insert_region(type, idx--, rbase, end - rbase,
671 memblock_get_region_node(rgn),
672 rgn->flags);
673 } else {
674 /* @rgn is fully contained, record it */
675 if (!*end_rgn)
676 *start_rgn = idx;
677 *end_rgn = idx + 1;
681 return 0;
684 static int __init_memblock memblock_remove_range(struct memblock_type *type,
685 phys_addr_t base, phys_addr_t size)
687 int start_rgn, end_rgn;
688 int i, ret;
690 ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
691 if (ret)
692 return ret;
694 for (i = end_rgn - 1; i >= start_rgn; i--)
695 memblock_remove_region(type, i);
696 return 0;
699 int __init_memblock memblock_remove(phys_addr_t base, phys_addr_t size)
701 phys_addr_t end = base + size - 1;
703 memblock_dbg("memblock_remove: [%pa-%pa] %pS\n",
704 &base, &end, (void *)_RET_IP_);
706 return memblock_remove_range(&memblock.memory, base, size);
710 int __init_memblock memblock_free(phys_addr_t base, phys_addr_t size)
712 phys_addr_t end = base + size - 1;
714 memblock_dbg(" memblock_free: [%pa-%pa] %pF\n",
715 &base, &end, (void *)_RET_IP_);
717 kmemleak_free_part_phys(base, size);
718 return memblock_remove_range(&memblock.reserved, base, size);
721 int __init_memblock memblock_reserve(phys_addr_t base, phys_addr_t size)
723 phys_addr_t end = base + size - 1;
725 memblock_dbg("memblock_reserve: [%pa-%pa] %pF\n",
726 &base, &end, (void *)_RET_IP_);
728 return memblock_add_range(&memblock.reserved, base, size, MAX_NUMNODES, 0);
733 * This function isolates region [@base, @base + @size), and sets/clears flag
735 * Return 0 on success, -errno on failure.
737 static int __init_memblock memblock_setclr_flag(phys_addr_t base,
738 phys_addr_t size, int set, int flag)
740 struct memblock_type *type = &memblock.memory;
741 int i, ret, start_rgn, end_rgn;
743 ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
744 if (ret)
745 return ret;
747 for (i = start_rgn; i < end_rgn; i++)
748 if (set)
749 memblock_set_region_flags(&type->regions[i], flag);
750 else
751 memblock_clear_region_flags(&type->regions[i], flag);
753 memblock_merge_regions(type);
754 return 0;
758 * memblock_mark_hotplug - Mark hotpluggable memory with flag MEMBLOCK_HOTPLUG.
759 * @base: the base phys addr of the region
760 * @size: the size of the region
762 * Return 0 on success, -errno on failure.
764 int __init_memblock memblock_mark_hotplug(phys_addr_t base, phys_addr_t size)
766 return memblock_setclr_flag(base, size, 1, MEMBLOCK_HOTPLUG);
770 * memblock_clear_hotplug - Clear flag MEMBLOCK_HOTPLUG for a specified region.
771 * @base: the base phys addr of the region
772 * @size: the size of the region
774 * Return 0 on success, -errno on failure.
776 int __init_memblock memblock_clear_hotplug(phys_addr_t base, phys_addr_t size)
778 return memblock_setclr_flag(base, size, 0, MEMBLOCK_HOTPLUG);
782 * memblock_mark_mirror - Mark mirrored memory with flag MEMBLOCK_MIRROR.
783 * @base: the base phys addr of the region
784 * @size: the size of the region
786 * Return 0 on success, -errno on failure.
788 int __init_memblock memblock_mark_mirror(phys_addr_t base, phys_addr_t size)
790 system_has_some_mirror = true;
792 return memblock_setclr_flag(base, size, 1, MEMBLOCK_MIRROR);
796 * memblock_mark_nomap - Mark a memory region with flag MEMBLOCK_NOMAP.
797 * @base: the base phys addr of the region
798 * @size: the size of the region
800 * Return 0 on success, -errno on failure.
802 int __init_memblock memblock_mark_nomap(phys_addr_t base, phys_addr_t size)
804 return memblock_setclr_flag(base, size, 1, MEMBLOCK_NOMAP);
808 * memblock_clear_nomap - Clear flag MEMBLOCK_NOMAP for a specified region.
809 * @base: the base phys addr of the region
810 * @size: the size of the region
812 * Return 0 on success, -errno on failure.
814 int __init_memblock memblock_clear_nomap(phys_addr_t base, phys_addr_t size)
816 return memblock_setclr_flag(base, size, 0, MEMBLOCK_NOMAP);
820 * __next_reserved_mem_region - next function for for_each_reserved_region()
821 * @idx: pointer to u64 loop variable
822 * @out_start: ptr to phys_addr_t for start address of the region, can be %NULL
823 * @out_end: ptr to phys_addr_t for end address of the region, can be %NULL
825 * Iterate over all reserved memory regions.
827 void __init_memblock __next_reserved_mem_region(u64 *idx,
828 phys_addr_t *out_start,
829 phys_addr_t *out_end)
831 struct memblock_type *type = &memblock.reserved;
833 if (*idx < type->cnt) {
834 struct memblock_region *r = &type->regions[*idx];
835 phys_addr_t base = r->base;
836 phys_addr_t size = r->size;
838 if (out_start)
839 *out_start = base;
840 if (out_end)
841 *out_end = base + size - 1;
843 *idx += 1;
844 return;
847 /* signal end of iteration */
848 *idx = ULLONG_MAX;
852 * __next__mem_range - next function for for_each_free_mem_range() etc.
853 * @idx: pointer to u64 loop variable
854 * @nid: node selector, %NUMA_NO_NODE for all nodes
855 * @flags: pick from blocks based on memory attributes
856 * @type_a: pointer to memblock_type from where the range is taken
857 * @type_b: pointer to memblock_type which excludes memory from being taken
858 * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL
859 * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL
860 * @out_nid: ptr to int for nid of the range, can be %NULL
862 * Find the first area from *@idx which matches @nid, fill the out
863 * parameters, and update *@idx for the next iteration. The lower 32bit of
864 * *@idx contains index into type_a and the upper 32bit indexes the
865 * areas before each region in type_b. For example, if type_b regions
866 * look like the following,
868 * 0:[0-16), 1:[32-48), 2:[128-130)
870 * The upper 32bit indexes the following regions.
872 * 0:[0-0), 1:[16-32), 2:[48-128), 3:[130-MAX)
874 * As both region arrays are sorted, the function advances the two indices
875 * in lockstep and returns each intersection.
877 void __init_memblock __next_mem_range(u64 *idx, int nid, ulong flags,
878 struct memblock_type *type_a,
879 struct memblock_type *type_b,
880 phys_addr_t *out_start,
881 phys_addr_t *out_end, int *out_nid)
883 int idx_a = *idx & 0xffffffff;
884 int idx_b = *idx >> 32;
886 if (WARN_ONCE(nid == MAX_NUMNODES,
887 "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
888 nid = NUMA_NO_NODE;
890 for (; idx_a < type_a->cnt; idx_a++) {
891 struct memblock_region *m = &type_a->regions[idx_a];
893 phys_addr_t m_start = m->base;
894 phys_addr_t m_end = m->base + m->size;
895 int m_nid = memblock_get_region_node(m);
897 /* only memory regions are associated with nodes, check it */
898 if (nid != NUMA_NO_NODE && nid != m_nid)
899 continue;
901 /* skip hotpluggable memory regions if needed */
902 if (movable_node_is_enabled() && memblock_is_hotpluggable(m))
903 continue;
905 /* if we want mirror memory skip non-mirror memory regions */
906 if ((flags & MEMBLOCK_MIRROR) && !memblock_is_mirror(m))
907 continue;
909 /* skip nomap memory unless we were asked for it explicitly */
910 if (!(flags & MEMBLOCK_NOMAP) && memblock_is_nomap(m))
911 continue;
913 if (!type_b) {
914 if (out_start)
915 *out_start = m_start;
916 if (out_end)
917 *out_end = m_end;
918 if (out_nid)
919 *out_nid = m_nid;
920 idx_a++;
921 *idx = (u32)idx_a | (u64)idx_b << 32;
922 return;
925 /* scan areas before each reservation */
926 for (; idx_b < type_b->cnt + 1; idx_b++) {
927 struct memblock_region *r;
928 phys_addr_t r_start;
929 phys_addr_t r_end;
931 r = &type_b->regions[idx_b];
932 r_start = idx_b ? r[-1].base + r[-1].size : 0;
933 r_end = idx_b < type_b->cnt ?
934 r->base : PHYS_ADDR_MAX;
937 * if idx_b advanced past idx_a,
938 * break out to advance idx_a
940 if (r_start >= m_end)
941 break;
942 /* if the two regions intersect, we're done */
943 if (m_start < r_end) {
944 if (out_start)
945 *out_start =
946 max(m_start, r_start);
947 if (out_end)
948 *out_end = min(m_end, r_end);
949 if (out_nid)
950 *out_nid = m_nid;
952 * The region which ends first is
953 * advanced for the next iteration.
955 if (m_end <= r_end)
956 idx_a++;
957 else
958 idx_b++;
959 *idx = (u32)idx_a | (u64)idx_b << 32;
960 return;
965 /* signal end of iteration */
966 *idx = ULLONG_MAX;
970 * __next_mem_range_rev - generic next function for for_each_*_range_rev()
972 * Finds the next range from type_a which is not marked as unsuitable
973 * in type_b.
975 * @idx: pointer to u64 loop variable
976 * @nid: node selector, %NUMA_NO_NODE for all nodes
977 * @flags: pick from blocks based on memory attributes
978 * @type_a: pointer to memblock_type from where the range is taken
979 * @type_b: pointer to memblock_type which excludes memory from being taken
980 * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL
981 * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL
982 * @out_nid: ptr to int for nid of the range, can be %NULL
984 * Reverse of __next_mem_range().
986 void __init_memblock __next_mem_range_rev(u64 *idx, int nid, ulong flags,
987 struct memblock_type *type_a,
988 struct memblock_type *type_b,
989 phys_addr_t *out_start,
990 phys_addr_t *out_end, int *out_nid)
992 int idx_a = *idx & 0xffffffff;
993 int idx_b = *idx >> 32;
995 if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
996 nid = NUMA_NO_NODE;
998 if (*idx == (u64)ULLONG_MAX) {
999 idx_a = type_a->cnt - 1;
1000 if (type_b != NULL)
1001 idx_b = type_b->cnt;
1002 else
1003 idx_b = 0;
1006 for (; idx_a >= 0; idx_a--) {
1007 struct memblock_region *m = &type_a->regions[idx_a];
1009 phys_addr_t m_start = m->base;
1010 phys_addr_t m_end = m->base + m->size;
1011 int m_nid = memblock_get_region_node(m);
1013 /* only memory regions are associated with nodes, check it */
1014 if (nid != NUMA_NO_NODE && nid != m_nid)
1015 continue;
1017 /* skip hotpluggable memory regions if needed */
1018 if (movable_node_is_enabled() && memblock_is_hotpluggable(m))
1019 continue;
1021 /* if we want mirror memory skip non-mirror memory regions */
1022 if ((flags & MEMBLOCK_MIRROR) && !memblock_is_mirror(m))
1023 continue;
1025 /* skip nomap memory unless we were asked for it explicitly */
1026 if (!(flags & MEMBLOCK_NOMAP) && memblock_is_nomap(m))
1027 continue;
1029 if (!type_b) {
1030 if (out_start)
1031 *out_start = m_start;
1032 if (out_end)
1033 *out_end = m_end;
1034 if (out_nid)
1035 *out_nid = m_nid;
1036 idx_a--;
1037 *idx = (u32)idx_a | (u64)idx_b << 32;
1038 return;
1041 /* scan areas before each reservation */
1042 for (; idx_b >= 0; idx_b--) {
1043 struct memblock_region *r;
1044 phys_addr_t r_start;
1045 phys_addr_t r_end;
1047 r = &type_b->regions[idx_b];
1048 r_start = idx_b ? r[-1].base + r[-1].size : 0;
1049 r_end = idx_b < type_b->cnt ?
1050 r->base : PHYS_ADDR_MAX;
1052 * if idx_b advanced past idx_a,
1053 * break out to advance idx_a
1056 if (r_end <= m_start)
1057 break;
1058 /* if the two regions intersect, we're done */
1059 if (m_end > r_start) {
1060 if (out_start)
1061 *out_start = max(m_start, r_start);
1062 if (out_end)
1063 *out_end = min(m_end, r_end);
1064 if (out_nid)
1065 *out_nid = m_nid;
1066 if (m_start >= r_start)
1067 idx_a--;
1068 else
1069 idx_b--;
1070 *idx = (u32)idx_a | (u64)idx_b << 32;
1071 return;
1075 /* signal end of iteration */
1076 *idx = ULLONG_MAX;
1079 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
1081 * Common iterator interface used to define for_each_mem_range().
1083 void __init_memblock __next_mem_pfn_range(int *idx, int nid,
1084 unsigned long *out_start_pfn,
1085 unsigned long *out_end_pfn, int *out_nid)
1087 struct memblock_type *type = &memblock.memory;
1088 struct memblock_region *r;
1090 while (++*idx < type->cnt) {
1091 r = &type->regions[*idx];
1093 if (PFN_UP(r->base) >= PFN_DOWN(r->base + r->size))
1094 continue;
1095 if (nid == MAX_NUMNODES || nid == r->nid)
1096 break;
1098 if (*idx >= type->cnt) {
1099 *idx = -1;
1100 return;
1103 if (out_start_pfn)
1104 *out_start_pfn = PFN_UP(r->base);
1105 if (out_end_pfn)
1106 *out_end_pfn = PFN_DOWN(r->base + r->size);
1107 if (out_nid)
1108 *out_nid = r->nid;
1112 * memblock_set_node - set node ID on memblock regions
1113 * @base: base of area to set node ID for
1114 * @size: size of area to set node ID for
1115 * @type: memblock type to set node ID for
1116 * @nid: node ID to set
1118 * Set the nid of memblock @type regions in [@base,@base+@size) to @nid.
1119 * Regions which cross the area boundaries are split as necessary.
1121 * RETURNS:
1122 * 0 on success, -errno on failure.
1124 int __init_memblock memblock_set_node(phys_addr_t base, phys_addr_t size,
1125 struct memblock_type *type, int nid)
1127 int start_rgn, end_rgn;
1128 int i, ret;
1130 ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
1131 if (ret)
1132 return ret;
1134 for (i = start_rgn; i < end_rgn; i++)
1135 memblock_set_region_node(&type->regions[i], nid);
1137 memblock_merge_regions(type);
1138 return 0;
1140 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
1142 static phys_addr_t __init memblock_alloc_range_nid(phys_addr_t size,
1143 phys_addr_t align, phys_addr_t start,
1144 phys_addr_t end, int nid, ulong flags)
1146 phys_addr_t found;
1148 if (!align)
1149 align = SMP_CACHE_BYTES;
1151 found = memblock_find_in_range_node(size, align, start, end, nid,
1152 flags);
1153 if (found && !memblock_reserve(found, size)) {
1155 * The min_count is set to 0 so that memblock allocations are
1156 * never reported as leaks.
1158 kmemleak_alloc_phys(found, size, 0, 0);
1159 return found;
1161 return 0;
1164 phys_addr_t __init memblock_alloc_range(phys_addr_t size, phys_addr_t align,
1165 phys_addr_t start, phys_addr_t end,
1166 ulong flags)
1168 return memblock_alloc_range_nid(size, align, start, end, NUMA_NO_NODE,
1169 flags);
1172 phys_addr_t __init memblock_alloc_base_nid(phys_addr_t size,
1173 phys_addr_t align, phys_addr_t max_addr,
1174 int nid, ulong flags)
1176 return memblock_alloc_range_nid(size, align, 0, max_addr, nid, flags);
1179 phys_addr_t __init memblock_alloc_nid(phys_addr_t size, phys_addr_t align, int nid)
1181 ulong flags = choose_memblock_flags();
1182 phys_addr_t ret;
1184 again:
1185 ret = memblock_alloc_base_nid(size, align, MEMBLOCK_ALLOC_ACCESSIBLE,
1186 nid, flags);
1188 if (!ret && (flags & MEMBLOCK_MIRROR)) {
1189 flags &= ~MEMBLOCK_MIRROR;
1190 goto again;
1192 return ret;
1195 phys_addr_t __init __memblock_alloc_base(phys_addr_t size, phys_addr_t align, phys_addr_t max_addr)
1197 return memblock_alloc_base_nid(size, align, max_addr, NUMA_NO_NODE,
1198 MEMBLOCK_NONE);
1201 phys_addr_t __init memblock_alloc_base(phys_addr_t size, phys_addr_t align, phys_addr_t max_addr)
1203 phys_addr_t alloc;
1205 alloc = __memblock_alloc_base(size, align, max_addr);
1207 if (alloc == 0)
1208 panic("ERROR: Failed to allocate %pa bytes below %pa.\n",
1209 &size, &max_addr);
1211 return alloc;
1214 phys_addr_t __init memblock_alloc(phys_addr_t size, phys_addr_t align)
1216 return memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ACCESSIBLE);
1219 phys_addr_t __init memblock_alloc_try_nid(phys_addr_t size, phys_addr_t align, int nid)
1221 phys_addr_t res = memblock_alloc_nid(size, align, nid);
1223 if (res)
1224 return res;
1225 return memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ACCESSIBLE);
1229 * memblock_virt_alloc_internal - allocate boot memory block
1230 * @size: size of memory block to be allocated in bytes
1231 * @align: alignment of the region and block's size
1232 * @min_addr: the lower bound of the memory region to allocate (phys address)
1233 * @max_addr: the upper bound of the memory region to allocate (phys address)
1234 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1236 * The @min_addr limit is dropped if it can not be satisfied and the allocation
1237 * will fall back to memory below @min_addr. Also, allocation may fall back
1238 * to any node in the system if the specified node can not
1239 * hold the requested memory.
1241 * The allocation is performed from memory region limited by
1242 * memblock.current_limit if @max_addr == %BOOTMEM_ALLOC_ACCESSIBLE.
1244 * The memory block is aligned on SMP_CACHE_BYTES if @align == 0.
1246 * The phys address of allocated boot memory block is converted to virtual and
1247 * allocated memory is reset to 0.
1249 * In addition, function sets the min_count to 0 using kmemleak_alloc for
1250 * allocated boot memory block, so that it is never reported as leaks.
1252 * RETURNS:
1253 * Virtual address of allocated memory block on success, NULL on failure.
1255 static void * __init memblock_virt_alloc_internal(
1256 phys_addr_t size, phys_addr_t align,
1257 phys_addr_t min_addr, phys_addr_t max_addr,
1258 int nid)
1260 phys_addr_t alloc;
1261 void *ptr;
1262 ulong flags = choose_memblock_flags();
1264 if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
1265 nid = NUMA_NO_NODE;
1268 * Detect any accidental use of these APIs after slab is ready, as at
1269 * this moment memblock may be deinitialized already and its
1270 * internal data may be destroyed (after execution of free_all_bootmem)
1272 if (WARN_ON_ONCE(slab_is_available()))
1273 return kzalloc_node(size, GFP_NOWAIT, nid);
1275 if (!align)
1276 align = SMP_CACHE_BYTES;
1278 if (max_addr > memblock.current_limit)
1279 max_addr = memblock.current_limit;
1280 again:
1281 alloc = memblock_find_in_range_node(size, align, min_addr, max_addr,
1282 nid, flags);
1283 if (alloc && !memblock_reserve(alloc, size))
1284 goto done;
1286 if (nid != NUMA_NO_NODE) {
1287 alloc = memblock_find_in_range_node(size, align, min_addr,
1288 max_addr, NUMA_NO_NODE,
1289 flags);
1290 if (alloc && !memblock_reserve(alloc, size))
1291 goto done;
1294 if (min_addr) {
1295 min_addr = 0;
1296 goto again;
1299 if (flags & MEMBLOCK_MIRROR) {
1300 flags &= ~MEMBLOCK_MIRROR;
1301 pr_warn("Could not allocate %pap bytes of mirrored memory\n",
1302 &size);
1303 goto again;
1306 return NULL;
1307 done:
1308 ptr = phys_to_virt(alloc);
1311 * The min_count is set to 0 so that bootmem allocated blocks
1312 * are never reported as leaks. This is because many of these blocks
1313 * are only referred via the physical address which is not
1314 * looked up by kmemleak.
1316 kmemleak_alloc(ptr, size, 0, 0);
1318 return ptr;
1322 * memblock_virt_alloc_try_nid_raw - allocate boot memory block without zeroing
1323 * memory and without panicking
1324 * @size: size of memory block to be allocated in bytes
1325 * @align: alignment of the region and block's size
1326 * @min_addr: the lower bound of the memory region from where the allocation
1327 * is preferred (phys address)
1328 * @max_addr: the upper bound of the memory region from where the allocation
1329 * is preferred (phys address), or %BOOTMEM_ALLOC_ACCESSIBLE to
1330 * allocate only from memory limited by memblock.current_limit value
1331 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1333 * Public function, provides additional debug information (including caller
1334 * info), if enabled. Does not zero allocated memory, does not panic if request
1335 * cannot be satisfied.
1337 * RETURNS:
1338 * Virtual address of allocated memory block on success, NULL on failure.
1340 void * __init memblock_virt_alloc_try_nid_raw(
1341 phys_addr_t size, phys_addr_t align,
1342 phys_addr_t min_addr, phys_addr_t max_addr,
1343 int nid)
1345 void *ptr;
1347 memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=0x%llx max_addr=0x%llx %pF\n",
1348 __func__, (u64)size, (u64)align, nid, (u64)min_addr,
1349 (u64)max_addr, (void *)_RET_IP_);
1351 ptr = memblock_virt_alloc_internal(size, align,
1352 min_addr, max_addr, nid);
1353 #ifdef CONFIG_DEBUG_VM
1354 if (ptr && size > 0)
1355 memset(ptr, PAGE_POISON_PATTERN, size);
1356 #endif
1357 return ptr;
1361 * memblock_virt_alloc_try_nid_nopanic - allocate boot memory block
1362 * @size: size of memory block to be allocated in bytes
1363 * @align: alignment of the region and block's size
1364 * @min_addr: the lower bound of the memory region from where the allocation
1365 * is preferred (phys address)
1366 * @max_addr: the upper bound of the memory region from where the allocation
1367 * is preferred (phys address), or %BOOTMEM_ALLOC_ACCESSIBLE to
1368 * allocate only from memory limited by memblock.current_limit value
1369 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1371 * Public function, provides additional debug information (including caller
1372 * info), if enabled. This function zeroes the allocated memory.
1374 * RETURNS:
1375 * Virtual address of allocated memory block on success, NULL on failure.
1377 void * __init memblock_virt_alloc_try_nid_nopanic(
1378 phys_addr_t size, phys_addr_t align,
1379 phys_addr_t min_addr, phys_addr_t max_addr,
1380 int nid)
1382 void *ptr;
1384 memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=0x%llx max_addr=0x%llx %pF\n",
1385 __func__, (u64)size, (u64)align, nid, (u64)min_addr,
1386 (u64)max_addr, (void *)_RET_IP_);
1388 ptr = memblock_virt_alloc_internal(size, align,
1389 min_addr, max_addr, nid);
1390 if (ptr)
1391 memset(ptr, 0, size);
1392 return ptr;
1396 * memblock_virt_alloc_try_nid - allocate boot memory block with panicking
1397 * @size: size of memory block to be allocated in bytes
1398 * @align: alignment of the region and block's size
1399 * @min_addr: the lower bound of the memory region from where the allocation
1400 * is preferred (phys address)
1401 * @max_addr: the upper bound of the memory region from where the allocation
1402 * is preferred (phys address), or %BOOTMEM_ALLOC_ACCESSIBLE to
1403 * allocate only from memory limited by memblock.current_limit value
1404 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1406 * Public panicking version of memblock_virt_alloc_try_nid_nopanic()
1407 * which provides debug information (including caller info), if enabled,
1408 * and panics if the request can not be satisfied.
1410 * RETURNS:
1411 * Virtual address of allocated memory block on success, NULL on failure.
1413 void * __init memblock_virt_alloc_try_nid(
1414 phys_addr_t size, phys_addr_t align,
1415 phys_addr_t min_addr, phys_addr_t max_addr,
1416 int nid)
1418 void *ptr;
1420 memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=0x%llx max_addr=0x%llx %pF\n",
1421 __func__, (u64)size, (u64)align, nid, (u64)min_addr,
1422 (u64)max_addr, (void *)_RET_IP_);
1423 ptr = memblock_virt_alloc_internal(size, align,
1424 min_addr, max_addr, nid);
1425 if (ptr) {
1426 memset(ptr, 0, size);
1427 return ptr;
1430 panic("%s: Failed to allocate %llu bytes align=0x%llx nid=%d from=0x%llx max_addr=0x%llx\n",
1431 __func__, (u64)size, (u64)align, nid, (u64)min_addr,
1432 (u64)max_addr);
1433 return NULL;
1437 * __memblock_free_early - free boot memory block
1438 * @base: phys starting address of the boot memory block
1439 * @size: size of the boot memory block in bytes
1441 * Free boot memory block previously allocated by memblock_virt_alloc_xx() API.
1442 * The freeing memory will not be released to the buddy allocator.
1444 void __init __memblock_free_early(phys_addr_t base, phys_addr_t size)
1446 memblock_dbg("%s: [%#016llx-%#016llx] %pF\n",
1447 __func__, (u64)base, (u64)base + size - 1,
1448 (void *)_RET_IP_);
1449 kmemleak_free_part_phys(base, size);
1450 memblock_remove_range(&memblock.reserved, base, size);
1454 * __memblock_free_late - free bootmem block pages directly to buddy allocator
1455 * @addr: phys starting address of the boot memory block
1456 * @size: size of the boot memory block in bytes
1458 * This is only useful when the bootmem allocator has already been torn
1459 * down, but we are still initializing the system. Pages are released directly
1460 * to the buddy allocator, no bootmem metadata is updated because it is gone.
1462 void __init __memblock_free_late(phys_addr_t base, phys_addr_t size)
1464 u64 cursor, end;
1466 memblock_dbg("%s: [%#016llx-%#016llx] %pF\n",
1467 __func__, (u64)base, (u64)base + size - 1,
1468 (void *)_RET_IP_);
1469 kmemleak_free_part_phys(base, size);
1470 cursor = PFN_UP(base);
1471 end = PFN_DOWN(base + size);
1473 for (; cursor < end; cursor++) {
1474 __free_pages_bootmem(pfn_to_page(cursor), cursor, 0);
1475 totalram_pages++;
1480 * Remaining API functions
1483 phys_addr_t __init_memblock memblock_phys_mem_size(void)
1485 return memblock.memory.total_size;
1488 phys_addr_t __init_memblock memblock_reserved_size(void)
1490 return memblock.reserved.total_size;
1493 phys_addr_t __init memblock_mem_size(unsigned long limit_pfn)
1495 unsigned long pages = 0;
1496 struct memblock_region *r;
1497 unsigned long start_pfn, end_pfn;
1499 for_each_memblock(memory, r) {
1500 start_pfn = memblock_region_memory_base_pfn(r);
1501 end_pfn = memblock_region_memory_end_pfn(r);
1502 start_pfn = min_t(unsigned long, start_pfn, limit_pfn);
1503 end_pfn = min_t(unsigned long, end_pfn, limit_pfn);
1504 pages += end_pfn - start_pfn;
1507 return PFN_PHYS(pages);
1510 /* lowest address */
1511 phys_addr_t __init_memblock memblock_start_of_DRAM(void)
1513 return memblock.memory.regions[0].base;
1516 phys_addr_t __init_memblock memblock_end_of_DRAM(void)
1518 int idx = memblock.memory.cnt - 1;
1520 return (memblock.memory.regions[idx].base + memblock.memory.regions[idx].size);
1523 static phys_addr_t __init_memblock __find_max_addr(phys_addr_t limit)
1525 phys_addr_t max_addr = PHYS_ADDR_MAX;
1526 struct memblock_region *r;
1529 * translate the memory @limit size into the max address within one of
1530 * the memory memblock regions, if the @limit exceeds the total size
1531 * of those regions, max_addr will keep original value PHYS_ADDR_MAX
1533 for_each_memblock(memory, r) {
1534 if (limit <= r->size) {
1535 max_addr = r->base + limit;
1536 break;
1538 limit -= r->size;
1541 return max_addr;
1544 void __init memblock_enforce_memory_limit(phys_addr_t limit)
1546 phys_addr_t max_addr = PHYS_ADDR_MAX;
1548 if (!limit)
1549 return;
1551 max_addr = __find_max_addr(limit);
1553 /* @limit exceeds the total size of the memory, do nothing */
1554 if (max_addr == PHYS_ADDR_MAX)
1555 return;
1557 /* truncate both memory and reserved regions */
1558 memblock_remove_range(&memblock.memory, max_addr,
1559 PHYS_ADDR_MAX);
1560 memblock_remove_range(&memblock.reserved, max_addr,
1561 PHYS_ADDR_MAX);
1564 void __init memblock_cap_memory_range(phys_addr_t base, phys_addr_t size)
1566 int start_rgn, end_rgn;
1567 int i, ret;
1569 if (!size)
1570 return;
1572 ret = memblock_isolate_range(&memblock.memory, base, size,
1573 &start_rgn, &end_rgn);
1574 if (ret)
1575 return;
1577 /* remove all the MAP regions */
1578 for (i = memblock.memory.cnt - 1; i >= end_rgn; i--)
1579 if (!memblock_is_nomap(&memblock.memory.regions[i]))
1580 memblock_remove_region(&memblock.memory, i);
1582 for (i = start_rgn - 1; i >= 0; i--)
1583 if (!memblock_is_nomap(&memblock.memory.regions[i]))
1584 memblock_remove_region(&memblock.memory, i);
1586 /* truncate the reserved regions */
1587 memblock_remove_range(&memblock.reserved, 0, base);
1588 memblock_remove_range(&memblock.reserved,
1589 base + size, PHYS_ADDR_MAX);
1592 void __init memblock_mem_limit_remove_map(phys_addr_t limit)
1594 phys_addr_t max_addr;
1596 if (!limit)
1597 return;
1599 max_addr = __find_max_addr(limit);
1601 /* @limit exceeds the total size of the memory, do nothing */
1602 if (max_addr == PHYS_ADDR_MAX)
1603 return;
1605 memblock_cap_memory_range(0, max_addr);
1608 static int __init_memblock memblock_search(struct memblock_type *type, phys_addr_t addr)
1610 unsigned int left = 0, right = type->cnt;
1612 do {
1613 unsigned int mid = (right + left) / 2;
1615 if (addr < type->regions[mid].base)
1616 right = mid;
1617 else if (addr >= (type->regions[mid].base +
1618 type->regions[mid].size))
1619 left = mid + 1;
1620 else
1621 return mid;
1622 } while (left < right);
1623 return -1;
1626 bool __init memblock_is_reserved(phys_addr_t addr)
1628 return memblock_search(&memblock.reserved, addr) != -1;
1631 bool __init_memblock memblock_is_memory(phys_addr_t addr)
1633 return memblock_search(&memblock.memory, addr) != -1;
1636 bool __init_memblock memblock_is_map_memory(phys_addr_t addr)
1638 int i = memblock_search(&memblock.memory, addr);
1640 if (i == -1)
1641 return false;
1642 return !memblock_is_nomap(&memblock.memory.regions[i]);
1645 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
1646 int __init_memblock memblock_search_pfn_nid(unsigned long pfn,
1647 unsigned long *start_pfn, unsigned long *end_pfn)
1649 struct memblock_type *type = &memblock.memory;
1650 int mid = memblock_search(type, PFN_PHYS(pfn));
1652 if (mid == -1)
1653 return -1;
1655 *start_pfn = PFN_DOWN(type->regions[mid].base);
1656 *end_pfn = PFN_DOWN(type->regions[mid].base + type->regions[mid].size);
1658 return type->regions[mid].nid;
1660 #endif
1663 * memblock_is_region_memory - check if a region is a subset of memory
1664 * @base: base of region to check
1665 * @size: size of region to check
1667 * Check if the region [@base, @base+@size) is a subset of a memory block.
1669 * RETURNS:
1670 * 0 if false, non-zero if true
1672 bool __init_memblock memblock_is_region_memory(phys_addr_t base, phys_addr_t size)
1674 int idx = memblock_search(&memblock.memory, base);
1675 phys_addr_t end = base + memblock_cap_size(base, &size);
1677 if (idx == -1)
1678 return false;
1679 return (memblock.memory.regions[idx].base +
1680 memblock.memory.regions[idx].size) >= end;
1684 * memblock_is_region_reserved - check if a region intersects reserved memory
1685 * @base: base of region to check
1686 * @size: size of region to check
1688 * Check if the region [@base, @base+@size) intersects a reserved memory block.
1690 * RETURNS:
1691 * True if they intersect, false if not.
1693 bool __init_memblock memblock_is_region_reserved(phys_addr_t base, phys_addr_t size)
1695 memblock_cap_size(base, &size);
1696 return memblock_overlaps_region(&memblock.reserved, base, size);
1699 void __init_memblock memblock_trim_memory(phys_addr_t align)
1701 phys_addr_t start, end, orig_start, orig_end;
1702 struct memblock_region *r;
1704 for_each_memblock(memory, r) {
1705 orig_start = r->base;
1706 orig_end = r->base + r->size;
1707 start = round_up(orig_start, align);
1708 end = round_down(orig_end, align);
1710 if (start == orig_start && end == orig_end)
1711 continue;
1713 if (start < end) {
1714 r->base = start;
1715 r->size = end - start;
1716 } else {
1717 memblock_remove_region(&memblock.memory,
1718 r - memblock.memory.regions);
1719 r--;
1724 void __init_memblock memblock_set_current_limit(phys_addr_t limit)
1726 memblock.current_limit = limit;
1729 phys_addr_t __init_memblock memblock_get_current_limit(void)
1731 return memblock.current_limit;
1734 static void __init_memblock memblock_dump(struct memblock_type *type)
1736 phys_addr_t base, end, size;
1737 unsigned long flags;
1738 int idx;
1739 struct memblock_region *rgn;
1741 pr_info(" %s.cnt = 0x%lx\n", type->name, type->cnt);
1743 for_each_memblock_type(idx, type, rgn) {
1744 char nid_buf[32] = "";
1746 base = rgn->base;
1747 size = rgn->size;
1748 end = base + size - 1;
1749 flags = rgn->flags;
1750 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
1751 if (memblock_get_region_node(rgn) != MAX_NUMNODES)
1752 snprintf(nid_buf, sizeof(nid_buf), " on node %d",
1753 memblock_get_region_node(rgn));
1754 #endif
1755 pr_info(" %s[%#x]\t[%pa-%pa], %pa bytes%s flags: %#lx\n",
1756 type->name, idx, &base, &end, &size, nid_buf, flags);
1760 void __init_memblock __memblock_dump_all(void)
1762 pr_info("MEMBLOCK configuration:\n");
1763 pr_info(" memory size = %pa reserved size = %pa\n",
1764 &memblock.memory.total_size,
1765 &memblock.reserved.total_size);
1767 memblock_dump(&memblock.memory);
1768 memblock_dump(&memblock.reserved);
1769 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
1770 memblock_dump(&memblock.physmem);
1771 #endif
1774 void __init memblock_allow_resize(void)
1776 memblock_can_resize = 1;
1779 static int __init early_memblock(char *p)
1781 if (p && strstr(p, "debug"))
1782 memblock_debug = 1;
1783 return 0;
1785 early_param("memblock", early_memblock);
1787 #if defined(CONFIG_DEBUG_FS) && !defined(CONFIG_ARCH_DISCARD_MEMBLOCK)
1789 static int memblock_debug_show(struct seq_file *m, void *private)
1791 struct memblock_type *type = m->private;
1792 struct memblock_region *reg;
1793 int i;
1794 phys_addr_t end;
1796 for (i = 0; i < type->cnt; i++) {
1797 reg = &type->regions[i];
1798 end = reg->base + reg->size - 1;
1800 seq_printf(m, "%4d: ", i);
1801 seq_printf(m, "%pa..%pa\n", &reg->base, &end);
1803 return 0;
1805 DEFINE_SHOW_ATTRIBUTE(memblock_debug);
1807 static int __init memblock_init_debugfs(void)
1809 struct dentry *root = debugfs_create_dir("memblock", NULL);
1810 if (!root)
1811 return -ENXIO;
1812 debugfs_create_file("memory", 0444, root,
1813 &memblock.memory, &memblock_debug_fops);
1814 debugfs_create_file("reserved", 0444, root,
1815 &memblock.reserved, &memblock_debug_fops);
1816 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
1817 debugfs_create_file("physmem", 0444, root,
1818 &memblock.physmem, &memblock_debug_fops);
1819 #endif
1821 return 0;
1823 __initcall(memblock_init_debugfs);
1825 #endif /* CONFIG_DEBUG_FS */