atm: zatm: Fix an error handling path in 'zatm_init_one()'
[linux-2.6/btrfs-unstable.git] / mm / memblock.c
blob2cb25fe4452c279c5ff6ff74cbbfee64128d820e
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/seq_file.h>
21 #include <linux/memblock.h>
23 #include <asm/sections.h>
24 #include <linux/io.h>
26 #include "internal.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;
32 #endif
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,
38 .memory.name = "memory",
40 .reserved.regions = memblock_reserved_init_regions,
41 .reserved.cnt = 1, /* empty dummy entry */
42 .reserved.max = INIT_MEMBLOCK_REGIONS,
43 .reserved.name = "reserved",
45 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
46 .physmem.regions = memblock_physmem_init_regions,
47 .physmem.cnt = 1, /* empty dummy entry */
48 .physmem.max = INIT_PHYSMEM_REGIONS,
49 .physmem.name = "physmem",
50 #endif
52 .bottom_up = false,
53 .current_limit = MEMBLOCK_ALLOC_ANYWHERE,
56 int memblock_debug __initdata_memblock;
57 static bool system_has_some_mirror __initdata_memblock = false;
58 static int memblock_can_resize __initdata_memblock;
59 static int memblock_memory_in_slab __initdata_memblock = 0;
60 static int memblock_reserved_in_slab __initdata_memblock = 0;
62 ulong __init_memblock choose_memblock_flags(void)
64 return system_has_some_mirror ? MEMBLOCK_MIRROR : MEMBLOCK_NONE;
67 /* adjust *@size so that (@base + *@size) doesn't overflow, return new size */
68 static inline phys_addr_t memblock_cap_size(phys_addr_t base, phys_addr_t *size)
70 return *size = min(*size, (phys_addr_t)ULLONG_MAX - base);
74 * Address comparison utilities
76 static unsigned long __init_memblock memblock_addrs_overlap(phys_addr_t base1, phys_addr_t size1,
77 phys_addr_t base2, phys_addr_t size2)
79 return ((base1 < (base2 + size2)) && (base2 < (base1 + size1)));
82 bool __init_memblock memblock_overlaps_region(struct memblock_type *type,
83 phys_addr_t base, phys_addr_t size)
85 unsigned long i;
87 for (i = 0; i < type->cnt; i++)
88 if (memblock_addrs_overlap(base, size, type->regions[i].base,
89 type->regions[i].size))
90 break;
91 return i < type->cnt;
95 * __memblock_find_range_bottom_up - find free area utility in bottom-up
96 * @start: start of candidate range
97 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_{ANYWHERE|ACCESSIBLE}
98 * @size: size of free area to find
99 * @align: alignment of free area to find
100 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
101 * @flags: pick from blocks based on memory attributes
103 * Utility called from memblock_find_in_range_node(), find free area bottom-up.
105 * RETURNS:
106 * Found address on success, 0 on failure.
108 static phys_addr_t __init_memblock
109 __memblock_find_range_bottom_up(phys_addr_t start, phys_addr_t end,
110 phys_addr_t size, phys_addr_t align, int nid,
111 ulong flags)
113 phys_addr_t this_start, this_end, cand;
114 u64 i;
116 for_each_free_mem_range(i, nid, flags, &this_start, &this_end, NULL) {
117 this_start = clamp(this_start, start, end);
118 this_end = clamp(this_end, start, end);
120 cand = round_up(this_start, align);
121 if (cand < this_end && this_end - cand >= size)
122 return cand;
125 return 0;
129 * __memblock_find_range_top_down - find free area utility, in top-down
130 * @start: start of candidate range
131 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_{ANYWHERE|ACCESSIBLE}
132 * @size: size of free area to find
133 * @align: alignment of free area to find
134 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
135 * @flags: pick from blocks based on memory attributes
137 * Utility called from memblock_find_in_range_node(), find free area top-down.
139 * RETURNS:
140 * Found address on success, 0 on failure.
142 static phys_addr_t __init_memblock
143 __memblock_find_range_top_down(phys_addr_t start, phys_addr_t end,
144 phys_addr_t size, phys_addr_t align, int nid,
145 ulong flags)
147 phys_addr_t this_start, this_end, cand;
148 u64 i;
150 for_each_free_mem_range_reverse(i, nid, flags, &this_start, &this_end,
151 NULL) {
152 this_start = clamp(this_start, start, end);
153 this_end = clamp(this_end, start, end);
155 if (this_end < size)
156 continue;
158 cand = round_down(this_end - size, align);
159 if (cand >= this_start)
160 return cand;
163 return 0;
167 * memblock_find_in_range_node - find free area in given range and node
168 * @size: size of free area to find
169 * @align: alignment of free area to find
170 * @start: start of candidate range
171 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_{ANYWHERE|ACCESSIBLE}
172 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
173 * @flags: pick from blocks based on memory attributes
175 * Find @size free area aligned to @align in the specified range and node.
177 * When allocation direction is bottom-up, the @start should be greater
178 * than the end of the kernel image. Otherwise, it will be trimmed. The
179 * reason is that we want the bottom-up allocation just near the kernel
180 * image so it is highly likely that the allocated memory and the kernel
181 * will reside in the same node.
183 * If bottom-up allocation failed, will try to allocate memory top-down.
185 * RETURNS:
186 * Found address on success, 0 on failure.
188 phys_addr_t __init_memblock memblock_find_in_range_node(phys_addr_t size,
189 phys_addr_t align, phys_addr_t start,
190 phys_addr_t end, int nid, ulong flags)
192 phys_addr_t kernel_end, ret;
194 /* pump up @end */
195 if (end == MEMBLOCK_ALLOC_ACCESSIBLE)
196 end = memblock.current_limit;
198 /* avoid allocating the first page */
199 start = max_t(phys_addr_t, start, PAGE_SIZE);
200 end = max(start, end);
201 kernel_end = __pa_symbol(_end);
204 * try bottom-up allocation only when bottom-up mode
205 * is set and @end is above the kernel image.
207 if (memblock_bottom_up() && end > kernel_end) {
208 phys_addr_t bottom_up_start;
210 /* make sure we will allocate above the kernel */
211 bottom_up_start = max(start, kernel_end);
213 /* ok, try bottom-up allocation first */
214 ret = __memblock_find_range_bottom_up(bottom_up_start, end,
215 size, align, nid, flags);
216 if (ret)
217 return ret;
220 * we always limit bottom-up allocation above the kernel,
221 * but top-down allocation doesn't have the limit, so
222 * retrying top-down allocation may succeed when bottom-up
223 * allocation failed.
225 * bottom-up allocation is expected to be fail very rarely,
226 * so we use WARN_ONCE() here to see the stack trace if
227 * fail happens.
229 WARN_ONCE(1, "memblock: bottom-up allocation failed, memory hotunplug may be affected\n");
232 return __memblock_find_range_top_down(start, end, size, align, nid,
233 flags);
237 * memblock_find_in_range - find free area in given range
238 * @start: start of candidate range
239 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_{ANYWHERE|ACCESSIBLE}
240 * @size: size of free area to find
241 * @align: alignment of free area to find
243 * Find @size free area aligned to @align in the specified range.
245 * RETURNS:
246 * Found address on success, 0 on failure.
248 phys_addr_t __init_memblock memblock_find_in_range(phys_addr_t start,
249 phys_addr_t end, phys_addr_t size,
250 phys_addr_t align)
252 phys_addr_t ret;
253 ulong flags = choose_memblock_flags();
255 again:
256 ret = memblock_find_in_range_node(size, align, start, end,
257 NUMA_NO_NODE, flags);
259 if (!ret && (flags & MEMBLOCK_MIRROR)) {
260 pr_warn("Could not allocate %pap bytes of mirrored memory\n",
261 &size);
262 flags &= ~MEMBLOCK_MIRROR;
263 goto again;
266 return ret;
269 static void __init_memblock memblock_remove_region(struct memblock_type *type, unsigned long r)
271 type->total_size -= type->regions[r].size;
272 memmove(&type->regions[r], &type->regions[r + 1],
273 (type->cnt - (r + 1)) * sizeof(type->regions[r]));
274 type->cnt--;
276 /* Special case for empty arrays */
277 if (type->cnt == 0) {
278 WARN_ON(type->total_size != 0);
279 type->cnt = 1;
280 type->regions[0].base = 0;
281 type->regions[0].size = 0;
282 type->regions[0].flags = 0;
283 memblock_set_region_node(&type->regions[0], MAX_NUMNODES);
287 #ifdef CONFIG_ARCH_DISCARD_MEMBLOCK
289 phys_addr_t __init_memblock get_allocated_memblock_reserved_regions_info(
290 phys_addr_t *addr)
292 if (memblock.reserved.regions == memblock_reserved_init_regions)
293 return 0;
295 *addr = __pa(memblock.reserved.regions);
297 return PAGE_ALIGN(sizeof(struct memblock_region) *
298 memblock.reserved.max);
301 phys_addr_t __init_memblock get_allocated_memblock_memory_regions_info(
302 phys_addr_t *addr)
304 if (memblock.memory.regions == memblock_memory_init_regions)
305 return 0;
307 *addr = __pa(memblock.memory.regions);
309 return PAGE_ALIGN(sizeof(struct memblock_region) *
310 memblock.memory.max);
313 #endif
316 * memblock_double_array - double the size of the memblock regions array
317 * @type: memblock type of the regions array being doubled
318 * @new_area_start: starting address of memory range to avoid overlap with
319 * @new_area_size: size of memory range to avoid overlap with
321 * Double the size of the @type regions array. If memblock is being used to
322 * allocate memory for a new reserved regions array and there is a previously
323 * allocated memory range [@new_area_start,@new_area_start+@new_area_size]
324 * waiting to be reserved, ensure the memory used by the new array does
325 * not overlap.
327 * RETURNS:
328 * 0 on success, -1 on failure.
330 static int __init_memblock memblock_double_array(struct memblock_type *type,
331 phys_addr_t new_area_start,
332 phys_addr_t new_area_size)
334 struct memblock_region *new_array, *old_array;
335 phys_addr_t old_alloc_size, new_alloc_size;
336 phys_addr_t old_size, new_size, addr;
337 int use_slab = slab_is_available();
338 int *in_slab;
340 /* We don't allow resizing until we know about the reserved regions
341 * of memory that aren't suitable for allocation
343 if (!memblock_can_resize)
344 return -1;
346 /* Calculate new doubled size */
347 old_size = type->max * sizeof(struct memblock_region);
348 new_size = old_size << 1;
350 * We need to allocated new one align to PAGE_SIZE,
351 * so we can free them completely later.
353 old_alloc_size = PAGE_ALIGN(old_size);
354 new_alloc_size = PAGE_ALIGN(new_size);
356 /* Retrieve the slab flag */
357 if (type == &memblock.memory)
358 in_slab = &memblock_memory_in_slab;
359 else
360 in_slab = &memblock_reserved_in_slab;
362 /* Try to find some space for it.
364 * WARNING: We assume that either slab_is_available() and we use it or
365 * we use MEMBLOCK for allocations. That means that this is unsafe to
366 * use when bootmem is currently active (unless bootmem itself is
367 * implemented on top of MEMBLOCK which isn't the case yet)
369 * This should however not be an issue for now, as we currently only
370 * call into MEMBLOCK while it's still active, or much later when slab
371 * is active for memory hotplug operations
373 if (use_slab) {
374 new_array = kmalloc(new_size, GFP_KERNEL);
375 addr = new_array ? __pa(new_array) : 0;
376 } else {
377 /* only exclude range when trying to double reserved.regions */
378 if (type != &memblock.reserved)
379 new_area_start = new_area_size = 0;
381 addr = memblock_find_in_range(new_area_start + new_area_size,
382 memblock.current_limit,
383 new_alloc_size, PAGE_SIZE);
384 if (!addr && new_area_size)
385 addr = memblock_find_in_range(0,
386 min(new_area_start, memblock.current_limit),
387 new_alloc_size, PAGE_SIZE);
389 new_array = addr ? __va(addr) : NULL;
391 if (!addr) {
392 pr_err("memblock: Failed to double %s array from %ld to %ld entries !\n",
393 type->name, type->max, type->max * 2);
394 return -1;
397 memblock_dbg("memblock: %s is doubled to %ld at [%#010llx-%#010llx]",
398 type->name, type->max * 2, (u64)addr,
399 (u64)addr + new_size - 1);
402 * Found space, we now need to move the array over before we add the
403 * reserved region since it may be our reserved array itself that is
404 * full.
406 memcpy(new_array, type->regions, old_size);
407 memset(new_array + type->max, 0, old_size);
408 old_array = type->regions;
409 type->regions = new_array;
410 type->max <<= 1;
412 /* Free old array. We needn't free it if the array is the static one */
413 if (*in_slab)
414 kfree(old_array);
415 else if (old_array != memblock_memory_init_regions &&
416 old_array != memblock_reserved_init_regions)
417 memblock_free(__pa(old_array), old_alloc_size);
420 * Reserve the new array if that comes from the memblock. Otherwise, we
421 * needn't do it
423 if (!use_slab)
424 BUG_ON(memblock_reserve(addr, new_alloc_size));
426 /* Update slab flag */
427 *in_slab = use_slab;
429 return 0;
433 * memblock_merge_regions - merge neighboring compatible regions
434 * @type: memblock type to scan
436 * Scan @type and merge neighboring compatible regions.
438 static void __init_memblock memblock_merge_regions(struct memblock_type *type)
440 int i = 0;
442 /* cnt never goes below 1 */
443 while (i < type->cnt - 1) {
444 struct memblock_region *this = &type->regions[i];
445 struct memblock_region *next = &type->regions[i + 1];
447 if (this->base + this->size != next->base ||
448 memblock_get_region_node(this) !=
449 memblock_get_region_node(next) ||
450 this->flags != next->flags) {
451 BUG_ON(this->base + this->size > next->base);
452 i++;
453 continue;
456 this->size += next->size;
457 /* move forward from next + 1, index of which is i + 2 */
458 memmove(next, next + 1, (type->cnt - (i + 2)) * sizeof(*next));
459 type->cnt--;
464 * memblock_insert_region - insert new memblock region
465 * @type: memblock type to insert into
466 * @idx: index for the insertion point
467 * @base: base address of the new region
468 * @size: size of the new region
469 * @nid: node id of the new region
470 * @flags: flags of the new region
472 * Insert new memblock region [@base,@base+@size) into @type at @idx.
473 * @type must already have extra room to accommodate the new region.
475 static void __init_memblock memblock_insert_region(struct memblock_type *type,
476 int idx, phys_addr_t base,
477 phys_addr_t size,
478 int nid, unsigned long flags)
480 struct memblock_region *rgn = &type->regions[idx];
482 BUG_ON(type->cnt >= type->max);
483 memmove(rgn + 1, rgn, (type->cnt - idx) * sizeof(*rgn));
484 rgn->base = base;
485 rgn->size = size;
486 rgn->flags = flags;
487 memblock_set_region_node(rgn, nid);
488 type->cnt++;
489 type->total_size += size;
493 * memblock_add_range - add new memblock region
494 * @type: memblock type to add new region into
495 * @base: base address of the new region
496 * @size: size of the new region
497 * @nid: nid of the new region
498 * @flags: flags of the new region
500 * Add new memblock region [@base,@base+@size) into @type. The new region
501 * is allowed to overlap with existing ones - overlaps don't affect already
502 * existing regions. @type is guaranteed to be minimal (all neighbouring
503 * compatible regions are merged) after the addition.
505 * RETURNS:
506 * 0 on success, -errno on failure.
508 int __init_memblock memblock_add_range(struct memblock_type *type,
509 phys_addr_t base, phys_addr_t size,
510 int nid, unsigned long flags)
512 bool insert = false;
513 phys_addr_t obase = base;
514 phys_addr_t end = base + memblock_cap_size(base, &size);
515 int idx, nr_new;
516 struct memblock_region *rgn;
518 if (!size)
519 return 0;
521 /* special case for empty array */
522 if (type->regions[0].size == 0) {
523 WARN_ON(type->cnt != 1 || type->total_size);
524 type->regions[0].base = base;
525 type->regions[0].size = size;
526 type->regions[0].flags = flags;
527 memblock_set_region_node(&type->regions[0], nid);
528 type->total_size = size;
529 return 0;
531 repeat:
533 * The following is executed twice. Once with %false @insert and
534 * then with %true. The first counts the number of regions needed
535 * to accommodate the new area. The second actually inserts them.
537 base = obase;
538 nr_new = 0;
540 for_each_memblock_type(type, rgn) {
541 phys_addr_t rbase = rgn->base;
542 phys_addr_t rend = rbase + rgn->size;
544 if (rbase >= end)
545 break;
546 if (rend <= base)
547 continue;
549 * @rgn overlaps. If it separates the lower part of new
550 * area, insert that portion.
552 if (rbase > base) {
553 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
554 WARN_ON(nid != memblock_get_region_node(rgn));
555 #endif
556 WARN_ON(flags != rgn->flags);
557 nr_new++;
558 if (insert)
559 memblock_insert_region(type, idx++, base,
560 rbase - base, nid,
561 flags);
563 /* area below @rend is dealt with, forget about it */
564 base = min(rend, end);
567 /* insert the remaining portion */
568 if (base < end) {
569 nr_new++;
570 if (insert)
571 memblock_insert_region(type, idx, base, end - base,
572 nid, flags);
575 if (!nr_new)
576 return 0;
579 * If this was the first round, resize array and repeat for actual
580 * insertions; otherwise, merge and return.
582 if (!insert) {
583 while (type->cnt + nr_new > type->max)
584 if (memblock_double_array(type, obase, size) < 0)
585 return -ENOMEM;
586 insert = true;
587 goto repeat;
588 } else {
589 memblock_merge_regions(type);
590 return 0;
594 int __init_memblock memblock_add_node(phys_addr_t base, phys_addr_t size,
595 int nid)
597 return memblock_add_range(&memblock.memory, base, size, nid, 0);
600 int __init_memblock memblock_add(phys_addr_t base, phys_addr_t size)
602 phys_addr_t end = base + size - 1;
604 memblock_dbg("memblock_add: [%pa-%pa] %pF\n",
605 &base, &end, (void *)_RET_IP_);
607 return memblock_add_range(&memblock.memory, base, size, MAX_NUMNODES, 0);
611 * memblock_isolate_range - isolate given range into disjoint memblocks
612 * @type: memblock type to isolate range for
613 * @base: base of range to isolate
614 * @size: size of range to isolate
615 * @start_rgn: out parameter for the start of isolated region
616 * @end_rgn: out parameter for the end of isolated region
618 * Walk @type and ensure that regions don't cross the boundaries defined by
619 * [@base,@base+@size). Crossing regions are split at the boundaries,
620 * which may create at most two more regions. The index of the first
621 * region inside the range is returned in *@start_rgn and end in *@end_rgn.
623 * RETURNS:
624 * 0 on success, -errno on failure.
626 static int __init_memblock memblock_isolate_range(struct memblock_type *type,
627 phys_addr_t base, phys_addr_t size,
628 int *start_rgn, int *end_rgn)
630 phys_addr_t end = base + memblock_cap_size(base, &size);
631 int idx;
632 struct memblock_region *rgn;
634 *start_rgn = *end_rgn = 0;
636 if (!size)
637 return 0;
639 /* we'll create at most two more regions */
640 while (type->cnt + 2 > type->max)
641 if (memblock_double_array(type, base, size) < 0)
642 return -ENOMEM;
644 for_each_memblock_type(type, rgn) {
645 phys_addr_t rbase = rgn->base;
646 phys_addr_t rend = rbase + rgn->size;
648 if (rbase >= end)
649 break;
650 if (rend <= base)
651 continue;
653 if (rbase < base) {
655 * @rgn intersects from below. Split and continue
656 * to process the next region - the new top half.
658 rgn->base = base;
659 rgn->size -= base - rbase;
660 type->total_size -= base - rbase;
661 memblock_insert_region(type, idx, rbase, base - rbase,
662 memblock_get_region_node(rgn),
663 rgn->flags);
664 } else if (rend > end) {
666 * @rgn intersects from above. Split and redo the
667 * current region - the new bottom half.
669 rgn->base = end;
670 rgn->size -= end - rbase;
671 type->total_size -= end - rbase;
672 memblock_insert_region(type, idx--, rbase, end - rbase,
673 memblock_get_region_node(rgn),
674 rgn->flags);
675 } else {
676 /* @rgn is fully contained, record it */
677 if (!*end_rgn)
678 *start_rgn = idx;
679 *end_rgn = idx + 1;
683 return 0;
686 static int __init_memblock memblock_remove_range(struct memblock_type *type,
687 phys_addr_t base, phys_addr_t size)
689 int start_rgn, end_rgn;
690 int i, ret;
692 ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
693 if (ret)
694 return ret;
696 for (i = end_rgn - 1; i >= start_rgn; i--)
697 memblock_remove_region(type, i);
698 return 0;
701 int __init_memblock memblock_remove(phys_addr_t base, phys_addr_t size)
703 return memblock_remove_range(&memblock.memory, base, size);
707 int __init_memblock memblock_free(phys_addr_t base, phys_addr_t size)
709 phys_addr_t end = base + size - 1;
711 memblock_dbg(" memblock_free: [%pa-%pa] %pF\n",
712 &base, &end, (void *)_RET_IP_);
714 kmemleak_free_part_phys(base, size);
715 return memblock_remove_range(&memblock.reserved, base, size);
718 int __init_memblock memblock_reserve(phys_addr_t base, phys_addr_t size)
720 phys_addr_t end = base + size - 1;
722 memblock_dbg("memblock_reserve: [%pa-%pa] %pF\n",
723 &base, &end, (void *)_RET_IP_);
725 return memblock_add_range(&memblock.reserved, base, size, MAX_NUMNODES, 0);
730 * This function isolates region [@base, @base + @size), and sets/clears flag
732 * Return 0 on success, -errno on failure.
734 static int __init_memblock memblock_setclr_flag(phys_addr_t base,
735 phys_addr_t size, int set, int flag)
737 struct memblock_type *type = &memblock.memory;
738 int i, ret, start_rgn, end_rgn;
740 ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
741 if (ret)
742 return ret;
744 for (i = start_rgn; i < end_rgn; i++)
745 if (set)
746 memblock_set_region_flags(&type->regions[i], flag);
747 else
748 memblock_clear_region_flags(&type->regions[i], flag);
750 memblock_merge_regions(type);
751 return 0;
755 * memblock_mark_hotplug - Mark hotpluggable memory with flag MEMBLOCK_HOTPLUG.
756 * @base: the base phys addr of the region
757 * @size: the size of the region
759 * Return 0 on success, -errno on failure.
761 int __init_memblock memblock_mark_hotplug(phys_addr_t base, phys_addr_t size)
763 return memblock_setclr_flag(base, size, 1, MEMBLOCK_HOTPLUG);
767 * memblock_clear_hotplug - Clear flag MEMBLOCK_HOTPLUG for a specified region.
768 * @base: the base phys addr of the region
769 * @size: the size of the region
771 * Return 0 on success, -errno on failure.
773 int __init_memblock memblock_clear_hotplug(phys_addr_t base, phys_addr_t size)
775 return memblock_setclr_flag(base, size, 0, MEMBLOCK_HOTPLUG);
779 * memblock_mark_mirror - Mark mirrored memory with flag MEMBLOCK_MIRROR.
780 * @base: the base phys addr of the region
781 * @size: the size of the region
783 * Return 0 on success, -errno on failure.
785 int __init_memblock memblock_mark_mirror(phys_addr_t base, phys_addr_t size)
787 system_has_some_mirror = true;
789 return memblock_setclr_flag(base, size, 1, MEMBLOCK_MIRROR);
793 * memblock_mark_nomap - Mark a memory region with flag MEMBLOCK_NOMAP.
794 * @base: the base phys addr of the region
795 * @size: the size of the region
797 * Return 0 on success, -errno on failure.
799 int __init_memblock memblock_mark_nomap(phys_addr_t base, phys_addr_t size)
801 return memblock_setclr_flag(base, size, 1, MEMBLOCK_NOMAP);
805 * memblock_clear_nomap - Clear flag MEMBLOCK_NOMAP for a specified region.
806 * @base: the base phys addr of the region
807 * @size: the size of the region
809 * Return 0 on success, -errno on failure.
811 int __init_memblock memblock_clear_nomap(phys_addr_t base, phys_addr_t size)
813 return memblock_setclr_flag(base, size, 0, MEMBLOCK_NOMAP);
817 * __next_reserved_mem_region - next function for for_each_reserved_region()
818 * @idx: pointer to u64 loop variable
819 * @out_start: ptr to phys_addr_t for start address of the region, can be %NULL
820 * @out_end: ptr to phys_addr_t for end address of the region, can be %NULL
822 * Iterate over all reserved memory regions.
824 void __init_memblock __next_reserved_mem_region(u64 *idx,
825 phys_addr_t *out_start,
826 phys_addr_t *out_end)
828 struct memblock_type *type = &memblock.reserved;
830 if (*idx < type->cnt) {
831 struct memblock_region *r = &type->regions[*idx];
832 phys_addr_t base = r->base;
833 phys_addr_t size = r->size;
835 if (out_start)
836 *out_start = base;
837 if (out_end)
838 *out_end = base + size - 1;
840 *idx += 1;
841 return;
844 /* signal end of iteration */
845 *idx = ULLONG_MAX;
849 * __next__mem_range - next function for for_each_free_mem_range() etc.
850 * @idx: pointer to u64 loop variable
851 * @nid: node selector, %NUMA_NO_NODE for all nodes
852 * @flags: pick from blocks based on memory attributes
853 * @type_a: pointer to memblock_type from where the range is taken
854 * @type_b: pointer to memblock_type which excludes memory from being taken
855 * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL
856 * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL
857 * @out_nid: ptr to int for nid of the range, can be %NULL
859 * Find the first area from *@idx which matches @nid, fill the out
860 * parameters, and update *@idx for the next iteration. The lower 32bit of
861 * *@idx contains index into type_a and the upper 32bit indexes the
862 * areas before each region in type_b. For example, if type_b regions
863 * look like the following,
865 * 0:[0-16), 1:[32-48), 2:[128-130)
867 * The upper 32bit indexes the following regions.
869 * 0:[0-0), 1:[16-32), 2:[48-128), 3:[130-MAX)
871 * As both region arrays are sorted, the function advances the two indices
872 * in lockstep and returns each intersection.
874 void __init_memblock __next_mem_range(u64 *idx, int nid, ulong flags,
875 struct memblock_type *type_a,
876 struct memblock_type *type_b,
877 phys_addr_t *out_start,
878 phys_addr_t *out_end, int *out_nid)
880 int idx_a = *idx & 0xffffffff;
881 int idx_b = *idx >> 32;
883 if (WARN_ONCE(nid == MAX_NUMNODES,
884 "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
885 nid = NUMA_NO_NODE;
887 for (; idx_a < type_a->cnt; idx_a++) {
888 struct memblock_region *m = &type_a->regions[idx_a];
890 phys_addr_t m_start = m->base;
891 phys_addr_t m_end = m->base + m->size;
892 int m_nid = memblock_get_region_node(m);
894 /* only memory regions are associated with nodes, check it */
895 if (nid != NUMA_NO_NODE && nid != m_nid)
896 continue;
898 /* skip hotpluggable memory regions if needed */
899 if (movable_node_is_enabled() && memblock_is_hotpluggable(m))
900 continue;
902 /* if we want mirror memory skip non-mirror memory regions */
903 if ((flags & MEMBLOCK_MIRROR) && !memblock_is_mirror(m))
904 continue;
906 /* skip nomap memory unless we were asked for it explicitly */
907 if (!(flags & MEMBLOCK_NOMAP) && memblock_is_nomap(m))
908 continue;
910 if (!type_b) {
911 if (out_start)
912 *out_start = m_start;
913 if (out_end)
914 *out_end = m_end;
915 if (out_nid)
916 *out_nid = m_nid;
917 idx_a++;
918 *idx = (u32)idx_a | (u64)idx_b << 32;
919 return;
922 /* scan areas before each reservation */
923 for (; idx_b < type_b->cnt + 1; idx_b++) {
924 struct memblock_region *r;
925 phys_addr_t r_start;
926 phys_addr_t r_end;
928 r = &type_b->regions[idx_b];
929 r_start = idx_b ? r[-1].base + r[-1].size : 0;
930 r_end = idx_b < type_b->cnt ?
931 r->base : ULLONG_MAX;
934 * if idx_b advanced past idx_a,
935 * break out to advance idx_a
937 if (r_start >= m_end)
938 break;
939 /* if the two regions intersect, we're done */
940 if (m_start < r_end) {
941 if (out_start)
942 *out_start =
943 max(m_start, r_start);
944 if (out_end)
945 *out_end = min(m_end, r_end);
946 if (out_nid)
947 *out_nid = m_nid;
949 * The region which ends first is
950 * advanced for the next iteration.
952 if (m_end <= r_end)
953 idx_a++;
954 else
955 idx_b++;
956 *idx = (u32)idx_a | (u64)idx_b << 32;
957 return;
962 /* signal end of iteration */
963 *idx = ULLONG_MAX;
967 * __next_mem_range_rev - generic next function for for_each_*_range_rev()
969 * Finds the next range from type_a which is not marked as unsuitable
970 * in type_b.
972 * @idx: pointer to u64 loop variable
973 * @nid: node selector, %NUMA_NO_NODE for all nodes
974 * @flags: pick from blocks based on memory attributes
975 * @type_a: pointer to memblock_type from where the range is taken
976 * @type_b: pointer to memblock_type which excludes memory from being taken
977 * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL
978 * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL
979 * @out_nid: ptr to int for nid of the range, can be %NULL
981 * Reverse of __next_mem_range().
983 void __init_memblock __next_mem_range_rev(u64 *idx, int nid, ulong flags,
984 struct memblock_type *type_a,
985 struct memblock_type *type_b,
986 phys_addr_t *out_start,
987 phys_addr_t *out_end, int *out_nid)
989 int idx_a = *idx & 0xffffffff;
990 int idx_b = *idx >> 32;
992 if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
993 nid = NUMA_NO_NODE;
995 if (*idx == (u64)ULLONG_MAX) {
996 idx_a = type_a->cnt - 1;
997 if (type_b != NULL)
998 idx_b = type_b->cnt;
999 else
1000 idx_b = 0;
1003 for (; idx_a >= 0; idx_a--) {
1004 struct memblock_region *m = &type_a->regions[idx_a];
1006 phys_addr_t m_start = m->base;
1007 phys_addr_t m_end = m->base + m->size;
1008 int m_nid = memblock_get_region_node(m);
1010 /* only memory regions are associated with nodes, check it */
1011 if (nid != NUMA_NO_NODE && nid != m_nid)
1012 continue;
1014 /* skip hotpluggable memory regions if needed */
1015 if (movable_node_is_enabled() && memblock_is_hotpluggable(m))
1016 continue;
1018 /* if we want mirror memory skip non-mirror memory regions */
1019 if ((flags & MEMBLOCK_MIRROR) && !memblock_is_mirror(m))
1020 continue;
1022 /* skip nomap memory unless we were asked for it explicitly */
1023 if (!(flags & MEMBLOCK_NOMAP) && memblock_is_nomap(m))
1024 continue;
1026 if (!type_b) {
1027 if (out_start)
1028 *out_start = m_start;
1029 if (out_end)
1030 *out_end = m_end;
1031 if (out_nid)
1032 *out_nid = m_nid;
1033 idx_a--;
1034 *idx = (u32)idx_a | (u64)idx_b << 32;
1035 return;
1038 /* scan areas before each reservation */
1039 for (; idx_b >= 0; idx_b--) {
1040 struct memblock_region *r;
1041 phys_addr_t r_start;
1042 phys_addr_t r_end;
1044 r = &type_b->regions[idx_b];
1045 r_start = idx_b ? r[-1].base + r[-1].size : 0;
1046 r_end = idx_b < type_b->cnt ?
1047 r->base : ULLONG_MAX;
1049 * if idx_b advanced past idx_a,
1050 * break out to advance idx_a
1053 if (r_end <= m_start)
1054 break;
1055 /* if the two regions intersect, we're done */
1056 if (m_end > r_start) {
1057 if (out_start)
1058 *out_start = max(m_start, r_start);
1059 if (out_end)
1060 *out_end = min(m_end, r_end);
1061 if (out_nid)
1062 *out_nid = m_nid;
1063 if (m_start >= r_start)
1064 idx_a--;
1065 else
1066 idx_b--;
1067 *idx = (u32)idx_a | (u64)idx_b << 32;
1068 return;
1072 /* signal end of iteration */
1073 *idx = ULLONG_MAX;
1076 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
1078 * Common iterator interface used to define for_each_mem_range().
1080 void __init_memblock __next_mem_pfn_range(int *idx, int nid,
1081 unsigned long *out_start_pfn,
1082 unsigned long *out_end_pfn, int *out_nid)
1084 struct memblock_type *type = &memblock.memory;
1085 struct memblock_region *r;
1087 while (++*idx < type->cnt) {
1088 r = &type->regions[*idx];
1090 if (PFN_UP(r->base) >= PFN_DOWN(r->base + r->size))
1091 continue;
1092 if (nid == MAX_NUMNODES || nid == r->nid)
1093 break;
1095 if (*idx >= type->cnt) {
1096 *idx = -1;
1097 return;
1100 if (out_start_pfn)
1101 *out_start_pfn = PFN_UP(r->base);
1102 if (out_end_pfn)
1103 *out_end_pfn = PFN_DOWN(r->base + r->size);
1104 if (out_nid)
1105 *out_nid = r->nid;
1108 unsigned long __init_memblock memblock_next_valid_pfn(unsigned long pfn,
1109 unsigned long max_pfn)
1111 struct memblock_type *type = &memblock.memory;
1112 unsigned int right = type->cnt;
1113 unsigned int mid, left = 0;
1114 phys_addr_t addr = PFN_PHYS(pfn + 1);
1116 do {
1117 mid = (right + left) / 2;
1119 if (addr < type->regions[mid].base)
1120 right = mid;
1121 else if (addr >= (type->regions[mid].base +
1122 type->regions[mid].size))
1123 left = mid + 1;
1124 else {
1125 /* addr is within the region, so pfn + 1 is valid */
1126 return min(pfn + 1, max_pfn);
1128 } while (left < right);
1130 if (right == type->cnt)
1131 return max_pfn;
1132 else
1133 return min(PHYS_PFN(type->regions[right].base), max_pfn);
1137 * memblock_set_node - set node ID on memblock regions
1138 * @base: base of area to set node ID for
1139 * @size: size of area to set node ID for
1140 * @type: memblock type to set node ID for
1141 * @nid: node ID to set
1143 * Set the nid of memblock @type regions in [@base,@base+@size) to @nid.
1144 * Regions which cross the area boundaries are split as necessary.
1146 * RETURNS:
1147 * 0 on success, -errno on failure.
1149 int __init_memblock memblock_set_node(phys_addr_t base, phys_addr_t size,
1150 struct memblock_type *type, int nid)
1152 int start_rgn, end_rgn;
1153 int i, ret;
1155 ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
1156 if (ret)
1157 return ret;
1159 for (i = start_rgn; i < end_rgn; i++)
1160 memblock_set_region_node(&type->regions[i], nid);
1162 memblock_merge_regions(type);
1163 return 0;
1165 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
1167 static phys_addr_t __init memblock_alloc_range_nid(phys_addr_t size,
1168 phys_addr_t align, phys_addr_t start,
1169 phys_addr_t end, int nid, ulong flags)
1171 phys_addr_t found;
1173 if (!align)
1174 align = SMP_CACHE_BYTES;
1176 found = memblock_find_in_range_node(size, align, start, end, nid,
1177 flags);
1178 if (found && !memblock_reserve(found, size)) {
1180 * The min_count is set to 0 so that memblock allocations are
1181 * never reported as leaks.
1183 kmemleak_alloc_phys(found, size, 0, 0);
1184 return found;
1186 return 0;
1189 phys_addr_t __init memblock_alloc_range(phys_addr_t size, phys_addr_t align,
1190 phys_addr_t start, phys_addr_t end,
1191 ulong flags)
1193 return memblock_alloc_range_nid(size, align, start, end, NUMA_NO_NODE,
1194 flags);
1197 static phys_addr_t __init memblock_alloc_base_nid(phys_addr_t size,
1198 phys_addr_t align, phys_addr_t max_addr,
1199 int nid, ulong flags)
1201 return memblock_alloc_range_nid(size, align, 0, max_addr, nid, flags);
1204 phys_addr_t __init memblock_alloc_nid(phys_addr_t size, phys_addr_t align, int nid)
1206 ulong flags = choose_memblock_flags();
1207 phys_addr_t ret;
1209 again:
1210 ret = memblock_alloc_base_nid(size, align, MEMBLOCK_ALLOC_ACCESSIBLE,
1211 nid, flags);
1213 if (!ret && (flags & MEMBLOCK_MIRROR)) {
1214 flags &= ~MEMBLOCK_MIRROR;
1215 goto again;
1217 return ret;
1220 phys_addr_t __init __memblock_alloc_base(phys_addr_t size, phys_addr_t align, phys_addr_t max_addr)
1222 return memblock_alloc_base_nid(size, align, max_addr, NUMA_NO_NODE,
1223 MEMBLOCK_NONE);
1226 phys_addr_t __init memblock_alloc_base(phys_addr_t size, phys_addr_t align, phys_addr_t max_addr)
1228 phys_addr_t alloc;
1230 alloc = __memblock_alloc_base(size, align, max_addr);
1232 if (alloc == 0)
1233 panic("ERROR: Failed to allocate %pa bytes below %pa.\n",
1234 &size, &max_addr);
1236 return alloc;
1239 phys_addr_t __init memblock_alloc(phys_addr_t size, phys_addr_t align)
1241 return memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ACCESSIBLE);
1244 phys_addr_t __init memblock_alloc_try_nid(phys_addr_t size, phys_addr_t align, int nid)
1246 phys_addr_t res = memblock_alloc_nid(size, align, nid);
1248 if (res)
1249 return res;
1250 return memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ACCESSIBLE);
1254 * memblock_virt_alloc_internal - allocate boot memory block
1255 * @size: size of memory block to be allocated in bytes
1256 * @align: alignment of the region and block's size
1257 * @min_addr: the lower bound of the memory region to allocate (phys address)
1258 * @max_addr: the upper bound of the memory region to allocate (phys address)
1259 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1261 * The @min_addr limit is dropped if it can not be satisfied and the allocation
1262 * will fall back to memory below @min_addr. Also, allocation may fall back
1263 * to any node in the system if the specified node can not
1264 * hold the requested memory.
1266 * The allocation is performed from memory region limited by
1267 * memblock.current_limit if @max_addr == %BOOTMEM_ALLOC_ACCESSIBLE.
1269 * The memory block is aligned on SMP_CACHE_BYTES if @align == 0.
1271 * The phys address of allocated boot memory block is converted to virtual and
1272 * allocated memory is reset to 0.
1274 * In addition, function sets the min_count to 0 using kmemleak_alloc for
1275 * allocated boot memory block, so that it is never reported as leaks.
1277 * RETURNS:
1278 * Virtual address of allocated memory block on success, NULL on failure.
1280 static void * __init memblock_virt_alloc_internal(
1281 phys_addr_t size, phys_addr_t align,
1282 phys_addr_t min_addr, phys_addr_t max_addr,
1283 int nid)
1285 phys_addr_t alloc;
1286 void *ptr;
1287 ulong flags = choose_memblock_flags();
1289 if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
1290 nid = NUMA_NO_NODE;
1293 * Detect any accidental use of these APIs after slab is ready, as at
1294 * this moment memblock may be deinitialized already and its
1295 * internal data may be destroyed (after execution of free_all_bootmem)
1297 if (WARN_ON_ONCE(slab_is_available()))
1298 return kzalloc_node(size, GFP_NOWAIT, nid);
1300 if (!align)
1301 align = SMP_CACHE_BYTES;
1303 if (max_addr > memblock.current_limit)
1304 max_addr = memblock.current_limit;
1305 again:
1306 alloc = memblock_find_in_range_node(size, align, min_addr, max_addr,
1307 nid, flags);
1308 if (alloc && !memblock_reserve(alloc, size))
1309 goto done;
1311 if (nid != NUMA_NO_NODE) {
1312 alloc = memblock_find_in_range_node(size, align, min_addr,
1313 max_addr, NUMA_NO_NODE,
1314 flags);
1315 if (alloc && !memblock_reserve(alloc, size))
1316 goto done;
1319 if (min_addr) {
1320 min_addr = 0;
1321 goto again;
1324 if (flags & MEMBLOCK_MIRROR) {
1325 flags &= ~MEMBLOCK_MIRROR;
1326 pr_warn("Could not allocate %pap bytes of mirrored memory\n",
1327 &size);
1328 goto again;
1331 return NULL;
1332 done:
1333 ptr = phys_to_virt(alloc);
1334 memset(ptr, 0, size);
1337 * The min_count is set to 0 so that bootmem allocated blocks
1338 * are never reported as leaks. This is because many of these blocks
1339 * are only referred via the physical address which is not
1340 * looked up by kmemleak.
1342 kmemleak_alloc(ptr, size, 0, 0);
1344 return ptr;
1348 * memblock_virt_alloc_try_nid_nopanic - allocate boot memory block
1349 * @size: size of memory block to be allocated in bytes
1350 * @align: alignment of the region and block's size
1351 * @min_addr: the lower bound of the memory region from where the allocation
1352 * is preferred (phys address)
1353 * @max_addr: the upper bound of the memory region from where the allocation
1354 * is preferred (phys address), or %BOOTMEM_ALLOC_ACCESSIBLE to
1355 * allocate only from memory limited by memblock.current_limit value
1356 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1358 * Public version of _memblock_virt_alloc_try_nid_nopanic() which provides
1359 * additional debug information (including caller info), if enabled.
1361 * RETURNS:
1362 * Virtual address of allocated memory block on success, NULL on failure.
1364 void * __init memblock_virt_alloc_try_nid_nopanic(
1365 phys_addr_t size, phys_addr_t align,
1366 phys_addr_t min_addr, phys_addr_t max_addr,
1367 int nid)
1369 memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=0x%llx max_addr=0x%llx %pF\n",
1370 __func__, (u64)size, (u64)align, nid, (u64)min_addr,
1371 (u64)max_addr, (void *)_RET_IP_);
1372 return memblock_virt_alloc_internal(size, align, min_addr,
1373 max_addr, nid);
1377 * memblock_virt_alloc_try_nid - allocate boot memory block with panicking
1378 * @size: size of memory block to be allocated in bytes
1379 * @align: alignment of the region and block's size
1380 * @min_addr: the lower bound of the memory region from where the allocation
1381 * is preferred (phys address)
1382 * @max_addr: the upper bound of the memory region from where the allocation
1383 * is preferred (phys address), or %BOOTMEM_ALLOC_ACCESSIBLE to
1384 * allocate only from memory limited by memblock.current_limit value
1385 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1387 * Public panicking version of _memblock_virt_alloc_try_nid_nopanic()
1388 * which provides debug information (including caller info), if enabled,
1389 * and panics if the request can not be satisfied.
1391 * RETURNS:
1392 * Virtual address of allocated memory block on success, NULL on failure.
1394 void * __init memblock_virt_alloc_try_nid(
1395 phys_addr_t size, phys_addr_t align,
1396 phys_addr_t min_addr, phys_addr_t max_addr,
1397 int nid)
1399 void *ptr;
1401 memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=0x%llx max_addr=0x%llx %pF\n",
1402 __func__, (u64)size, (u64)align, nid, (u64)min_addr,
1403 (u64)max_addr, (void *)_RET_IP_);
1404 ptr = memblock_virt_alloc_internal(size, align,
1405 min_addr, max_addr, nid);
1406 if (ptr)
1407 return ptr;
1409 panic("%s: Failed to allocate %llu bytes align=0x%llx nid=%d from=0x%llx max_addr=0x%llx\n",
1410 __func__, (u64)size, (u64)align, nid, (u64)min_addr,
1411 (u64)max_addr);
1412 return NULL;
1416 * __memblock_free_early - free boot memory block
1417 * @base: phys starting address of the boot memory block
1418 * @size: size of the boot memory block in bytes
1420 * Free boot memory block previously allocated by memblock_virt_alloc_xx() API.
1421 * The freeing memory will not be released to the buddy allocator.
1423 void __init __memblock_free_early(phys_addr_t base, phys_addr_t size)
1425 memblock_dbg("%s: [%#016llx-%#016llx] %pF\n",
1426 __func__, (u64)base, (u64)base + size - 1,
1427 (void *)_RET_IP_);
1428 kmemleak_free_part_phys(base, size);
1429 memblock_remove_range(&memblock.reserved, base, size);
1433 * __memblock_free_late - free bootmem block pages directly to buddy allocator
1434 * @addr: phys starting address of the boot memory block
1435 * @size: size of the boot memory block in bytes
1437 * This is only useful when the bootmem allocator has already been torn
1438 * down, but we are still initializing the system. Pages are released directly
1439 * to the buddy allocator, no bootmem metadata is updated because it is gone.
1441 void __init __memblock_free_late(phys_addr_t base, phys_addr_t size)
1443 u64 cursor, end;
1445 memblock_dbg("%s: [%#016llx-%#016llx] %pF\n",
1446 __func__, (u64)base, (u64)base + size - 1,
1447 (void *)_RET_IP_);
1448 kmemleak_free_part_phys(base, size);
1449 cursor = PFN_UP(base);
1450 end = PFN_DOWN(base + size);
1452 for (; cursor < end; cursor++) {
1453 __free_pages_bootmem(pfn_to_page(cursor), cursor, 0);
1454 totalram_pages++;
1459 * Remaining API functions
1462 phys_addr_t __init_memblock memblock_phys_mem_size(void)
1464 return memblock.memory.total_size;
1467 phys_addr_t __init_memblock memblock_reserved_size(void)
1469 return memblock.reserved.total_size;
1472 phys_addr_t __init memblock_mem_size(unsigned long limit_pfn)
1474 unsigned long pages = 0;
1475 struct memblock_region *r;
1476 unsigned long start_pfn, end_pfn;
1478 for_each_memblock(memory, r) {
1479 start_pfn = memblock_region_memory_base_pfn(r);
1480 end_pfn = memblock_region_memory_end_pfn(r);
1481 start_pfn = min_t(unsigned long, start_pfn, limit_pfn);
1482 end_pfn = min_t(unsigned long, end_pfn, limit_pfn);
1483 pages += end_pfn - start_pfn;
1486 return PFN_PHYS(pages);
1489 /* lowest address */
1490 phys_addr_t __init_memblock memblock_start_of_DRAM(void)
1492 return memblock.memory.regions[0].base;
1495 phys_addr_t __init_memblock memblock_end_of_DRAM(void)
1497 int idx = memblock.memory.cnt - 1;
1499 return (memblock.memory.regions[idx].base + memblock.memory.regions[idx].size);
1502 static phys_addr_t __init_memblock __find_max_addr(phys_addr_t limit)
1504 phys_addr_t max_addr = (phys_addr_t)ULLONG_MAX;
1505 struct memblock_region *r;
1508 * translate the memory @limit size into the max address within one of
1509 * the memory memblock regions, if the @limit exceeds the total size
1510 * of those regions, max_addr will keep original value ULLONG_MAX
1512 for_each_memblock(memory, r) {
1513 if (limit <= r->size) {
1514 max_addr = r->base + limit;
1515 break;
1517 limit -= r->size;
1520 return max_addr;
1523 void __init memblock_enforce_memory_limit(phys_addr_t limit)
1525 phys_addr_t max_addr = (phys_addr_t)ULLONG_MAX;
1527 if (!limit)
1528 return;
1530 max_addr = __find_max_addr(limit);
1532 /* @limit exceeds the total size of the memory, do nothing */
1533 if (max_addr == (phys_addr_t)ULLONG_MAX)
1534 return;
1536 /* truncate both memory and reserved regions */
1537 memblock_remove_range(&memblock.memory, max_addr,
1538 (phys_addr_t)ULLONG_MAX);
1539 memblock_remove_range(&memblock.reserved, max_addr,
1540 (phys_addr_t)ULLONG_MAX);
1543 void __init memblock_cap_memory_range(phys_addr_t base, phys_addr_t size)
1545 int start_rgn, end_rgn;
1546 int i, ret;
1548 if (!size)
1549 return;
1551 ret = memblock_isolate_range(&memblock.memory, base, size,
1552 &start_rgn, &end_rgn);
1553 if (ret)
1554 return;
1556 /* remove all the MAP regions */
1557 for (i = memblock.memory.cnt - 1; i >= end_rgn; i--)
1558 if (!memblock_is_nomap(&memblock.memory.regions[i]))
1559 memblock_remove_region(&memblock.memory, i);
1561 for (i = start_rgn - 1; i >= 0; i--)
1562 if (!memblock_is_nomap(&memblock.memory.regions[i]))
1563 memblock_remove_region(&memblock.memory, i);
1565 /* truncate the reserved regions */
1566 memblock_remove_range(&memblock.reserved, 0, base);
1567 memblock_remove_range(&memblock.reserved,
1568 base + size, (phys_addr_t)ULLONG_MAX);
1571 void __init memblock_mem_limit_remove_map(phys_addr_t limit)
1573 phys_addr_t max_addr;
1575 if (!limit)
1576 return;
1578 max_addr = __find_max_addr(limit);
1580 /* @limit exceeds the total size of the memory, do nothing */
1581 if (max_addr == (phys_addr_t)ULLONG_MAX)
1582 return;
1584 memblock_cap_memory_range(0, max_addr);
1587 static int __init_memblock memblock_search(struct memblock_type *type, phys_addr_t addr)
1589 unsigned int left = 0, right = type->cnt;
1591 do {
1592 unsigned int mid = (right + left) / 2;
1594 if (addr < type->regions[mid].base)
1595 right = mid;
1596 else if (addr >= (type->regions[mid].base +
1597 type->regions[mid].size))
1598 left = mid + 1;
1599 else
1600 return mid;
1601 } while (left < right);
1602 return -1;
1605 bool __init memblock_is_reserved(phys_addr_t addr)
1607 return memblock_search(&memblock.reserved, addr) != -1;
1610 bool __init_memblock memblock_is_memory(phys_addr_t addr)
1612 return memblock_search(&memblock.memory, addr) != -1;
1615 int __init_memblock memblock_is_map_memory(phys_addr_t addr)
1617 int i = memblock_search(&memblock.memory, addr);
1619 if (i == -1)
1620 return false;
1621 return !memblock_is_nomap(&memblock.memory.regions[i]);
1624 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
1625 int __init_memblock memblock_search_pfn_nid(unsigned long pfn,
1626 unsigned long *start_pfn, unsigned long *end_pfn)
1628 struct memblock_type *type = &memblock.memory;
1629 int mid = memblock_search(type, PFN_PHYS(pfn));
1631 if (mid == -1)
1632 return -1;
1634 *start_pfn = PFN_DOWN(type->regions[mid].base);
1635 *end_pfn = PFN_DOWN(type->regions[mid].base + type->regions[mid].size);
1637 return type->regions[mid].nid;
1639 #endif
1642 * memblock_is_region_memory - check if a region is a subset of memory
1643 * @base: base of region to check
1644 * @size: size of region to check
1646 * Check if the region [@base, @base+@size) is a subset of a memory block.
1648 * RETURNS:
1649 * 0 if false, non-zero if true
1651 int __init_memblock memblock_is_region_memory(phys_addr_t base, phys_addr_t size)
1653 int idx = memblock_search(&memblock.memory, base);
1654 phys_addr_t end = base + memblock_cap_size(base, &size);
1656 if (idx == -1)
1657 return 0;
1658 return (memblock.memory.regions[idx].base +
1659 memblock.memory.regions[idx].size) >= end;
1663 * memblock_is_region_reserved - check if a region intersects reserved memory
1664 * @base: base of region to check
1665 * @size: size of region to check
1667 * Check if the region [@base, @base+@size) intersects a reserved memory block.
1669 * RETURNS:
1670 * True if they intersect, false if not.
1672 bool __init_memblock memblock_is_region_reserved(phys_addr_t base, phys_addr_t size)
1674 memblock_cap_size(base, &size);
1675 return memblock_overlaps_region(&memblock.reserved, base, size);
1678 void __init_memblock memblock_trim_memory(phys_addr_t align)
1680 phys_addr_t start, end, orig_start, orig_end;
1681 struct memblock_region *r;
1683 for_each_memblock(memory, r) {
1684 orig_start = r->base;
1685 orig_end = r->base + r->size;
1686 start = round_up(orig_start, align);
1687 end = round_down(orig_end, align);
1689 if (start == orig_start && end == orig_end)
1690 continue;
1692 if (start < end) {
1693 r->base = start;
1694 r->size = end - start;
1695 } else {
1696 memblock_remove_region(&memblock.memory,
1697 r - memblock.memory.regions);
1698 r--;
1703 void __init_memblock memblock_set_current_limit(phys_addr_t limit)
1705 memblock.current_limit = limit;
1708 phys_addr_t __init_memblock memblock_get_current_limit(void)
1710 return memblock.current_limit;
1713 static void __init_memblock memblock_dump(struct memblock_type *type)
1715 phys_addr_t base, end, size;
1716 unsigned long flags;
1717 int idx;
1718 struct memblock_region *rgn;
1720 pr_info(" %s.cnt = 0x%lx\n", type->name, type->cnt);
1722 for_each_memblock_type(type, rgn) {
1723 char nid_buf[32] = "";
1725 base = rgn->base;
1726 size = rgn->size;
1727 end = base + size - 1;
1728 flags = rgn->flags;
1729 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
1730 if (memblock_get_region_node(rgn) != MAX_NUMNODES)
1731 snprintf(nid_buf, sizeof(nid_buf), " on node %d",
1732 memblock_get_region_node(rgn));
1733 #endif
1734 pr_info(" %s[%#x]\t[%pa-%pa], %pa bytes%s flags: %#lx\n",
1735 type->name, idx, &base, &end, &size, nid_buf, flags);
1739 extern unsigned long __init_memblock
1740 memblock_reserved_memory_within(phys_addr_t start_addr, phys_addr_t end_addr)
1742 struct memblock_region *rgn;
1743 unsigned long size = 0;
1744 int idx;
1746 for_each_memblock_type((&memblock.reserved), rgn) {
1747 phys_addr_t start, end;
1749 if (rgn->base + rgn->size < start_addr)
1750 continue;
1751 if (rgn->base > end_addr)
1752 continue;
1754 start = rgn->base;
1755 end = start + rgn->size;
1756 size += end - start;
1759 return size;
1762 void __init_memblock __memblock_dump_all(void)
1764 pr_info("MEMBLOCK configuration:\n");
1765 pr_info(" memory size = %pa reserved size = %pa\n",
1766 &memblock.memory.total_size,
1767 &memblock.reserved.total_size);
1769 memblock_dump(&memblock.memory);
1770 memblock_dump(&memblock.reserved);
1771 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
1772 memblock_dump(&memblock.physmem);
1773 #endif
1776 void __init memblock_allow_resize(void)
1778 memblock_can_resize = 1;
1781 static int __init early_memblock(char *p)
1783 if (p && strstr(p, "debug"))
1784 memblock_debug = 1;
1785 return 0;
1787 early_param("memblock", early_memblock);
1789 #if defined(CONFIG_DEBUG_FS) && !defined(CONFIG_ARCH_DISCARD_MEMBLOCK)
1791 static int memblock_debug_show(struct seq_file *m, void *private)
1793 struct memblock_type *type = m->private;
1794 struct memblock_region *reg;
1795 int i;
1796 phys_addr_t end;
1798 for (i = 0; i < type->cnt; i++) {
1799 reg = &type->regions[i];
1800 end = reg->base + reg->size - 1;
1802 seq_printf(m, "%4d: ", i);
1803 seq_printf(m, "%pa..%pa\n", &reg->base, &end);
1805 return 0;
1808 static int memblock_debug_open(struct inode *inode, struct file *file)
1810 return single_open(file, memblock_debug_show, inode->i_private);
1813 static const struct file_operations memblock_debug_fops = {
1814 .open = memblock_debug_open,
1815 .read = seq_read,
1816 .llseek = seq_lseek,
1817 .release = single_release,
1820 static int __init memblock_init_debugfs(void)
1822 struct dentry *root = debugfs_create_dir("memblock", NULL);
1823 if (!root)
1824 return -ENXIO;
1825 debugfs_create_file("memory", S_IRUGO, root, &memblock.memory, &memblock_debug_fops);
1826 debugfs_create_file("reserved", S_IRUGO, root, &memblock.reserved, &memblock_debug_fops);
1827 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
1828 debugfs_create_file("physmem", S_IRUGO, root, &memblock.physmem, &memblock_debug_fops);
1829 #endif
1831 return 0;
1833 __initcall(memblock_init_debugfs);
1835 #endif /* CONFIG_DEBUG_FS */