2 * Procedures for maintaining information about logical memory blocks.
4 * Peter Bergner, IBM Corp. June 2001.
5 * Copyright (C) 2001 Peter Bergner.
7 * This program is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU General Public License
9 * as published by the Free Software Foundation; either version
10 * 2 of the License, or (at your option) any later version.
13 #include <linux/kernel.h>
14 #include <linux/slab.h>
15 #include <linux/init.h>
16 #include <linux/bitops.h>
17 #include <linux/poison.h>
18 #include <linux/pfn.h>
19 #include <linux/debugfs.h>
20 #include <linux/seq_file.h>
21 #include <linux/memblock.h>
23 #include <asm-generic/sections.h>
28 static struct memblock_region memblock_memory_init_regions
[INIT_MEMBLOCK_REGIONS
] __initdata_memblock
;
29 static struct memblock_region memblock_reserved_init_regions
[INIT_MEMBLOCK_REGIONS
] __initdata_memblock
;
31 struct memblock memblock __initdata_memblock
= {
32 .memory
.regions
= memblock_memory_init_regions
,
33 .memory
.cnt
= 1, /* empty dummy entry */
34 .memory
.max
= INIT_MEMBLOCK_REGIONS
,
36 .reserved
.regions
= memblock_reserved_init_regions
,
37 .reserved
.cnt
= 1, /* empty dummy entry */
38 .reserved
.max
= INIT_MEMBLOCK_REGIONS
,
41 .current_limit
= MEMBLOCK_ALLOC_ANYWHERE
,
44 int memblock_debug __initdata_memblock
;
45 #ifdef CONFIG_MOVABLE_NODE
46 bool movable_node_enabled __initdata_memblock
= false;
48 static int memblock_can_resize __initdata_memblock
;
49 static int memblock_memory_in_slab __initdata_memblock
= 0;
50 static int memblock_reserved_in_slab __initdata_memblock
= 0;
52 /* inline so we don't get a warning when pr_debug is compiled out */
53 static __init_memblock
const char *
54 memblock_type_name(struct memblock_type
*type
)
56 if (type
== &memblock
.memory
)
58 else if (type
== &memblock
.reserved
)
64 /* adjust *@size so that (@base + *@size) doesn't overflow, return new size */
65 static inline phys_addr_t
memblock_cap_size(phys_addr_t base
, phys_addr_t
*size
)
67 return *size
= min(*size
, (phys_addr_t
)ULLONG_MAX
- base
);
71 * Address comparison utilities
73 static unsigned long __init_memblock
memblock_addrs_overlap(phys_addr_t base1
, phys_addr_t size1
,
74 phys_addr_t base2
, phys_addr_t size2
)
76 return ((base1
< (base2
+ size2
)) && (base2
< (base1
+ size1
)));
79 static long __init_memblock
memblock_overlaps_region(struct memblock_type
*type
,
80 phys_addr_t base
, phys_addr_t size
)
84 for (i
= 0; i
< type
->cnt
; i
++) {
85 phys_addr_t rgnbase
= type
->regions
[i
].base
;
86 phys_addr_t rgnsize
= type
->regions
[i
].size
;
87 if (memblock_addrs_overlap(base
, size
, rgnbase
, rgnsize
))
91 return (i
< type
->cnt
) ? i
: -1;
95 * __memblock_find_range_bottom_up - find free area utility in bottom-up
96 * @start: start of candidate range
97 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_{ANYWHERE|ACCESSIBLE}
98 * @size: size of free area to find
99 * @align: alignment of free area to find
100 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
102 * Utility called from memblock_find_in_range_node(), find free area bottom-up.
105 * Found address on success, 0 on failure.
107 static phys_addr_t __init_memblock
108 __memblock_find_range_bottom_up(phys_addr_t start
, phys_addr_t end
,
109 phys_addr_t size
, phys_addr_t align
, int nid
)
111 phys_addr_t this_start
, this_end
, cand
;
114 for_each_free_mem_range(i
, nid
, &this_start
, &this_end
, NULL
) {
115 this_start
= clamp(this_start
, start
, end
);
116 this_end
= clamp(this_end
, start
, end
);
118 cand
= round_up(this_start
, align
);
119 if (cand
< this_end
&& this_end
- cand
>= size
)
127 * __memblock_find_range_top_down - find free area utility, in top-down
128 * @start: start of candidate range
129 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_{ANYWHERE|ACCESSIBLE}
130 * @size: size of free area to find
131 * @align: alignment of free area to find
132 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
134 * Utility called from memblock_find_in_range_node(), find free area top-down.
137 * Found address on success, 0 on failure.
139 static phys_addr_t __init_memblock
140 __memblock_find_range_top_down(phys_addr_t start
, phys_addr_t end
,
141 phys_addr_t size
, phys_addr_t align
, int nid
)
143 phys_addr_t this_start
, this_end
, cand
;
146 for_each_free_mem_range_reverse(i
, nid
, &this_start
, &this_end
, NULL
) {
147 this_start
= clamp(this_start
, start
, end
);
148 this_end
= clamp(this_end
, start
, end
);
153 cand
= round_down(this_end
- size
, align
);
154 if (cand
>= this_start
)
162 * memblock_find_in_range_node - find free area in given range and node
163 * @size: size of free area to find
164 * @align: alignment of free area to find
165 * @start: start of candidate range
166 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_{ANYWHERE|ACCESSIBLE}
167 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
169 * Find @size free area aligned to @align in the specified range and node.
171 * When allocation direction is bottom-up, the @start should be greater
172 * than the end of the kernel image. Otherwise, it will be trimmed. The
173 * reason is that we want the bottom-up allocation just near the kernel
174 * image so it is highly likely that the allocated memory and the kernel
175 * will reside in the same node.
177 * If bottom-up allocation failed, will try to allocate memory top-down.
180 * Found address on success, 0 on failure.
182 phys_addr_t __init_memblock
memblock_find_in_range_node(phys_addr_t size
,
183 phys_addr_t align
, phys_addr_t start
,
184 phys_addr_t end
, int nid
)
187 phys_addr_t kernel_end
;
190 if (end
== MEMBLOCK_ALLOC_ACCESSIBLE
)
191 end
= memblock
.current_limit
;
193 /* avoid allocating the first page */
194 start
= max_t(phys_addr_t
, start
, PAGE_SIZE
);
195 end
= max(start
, end
);
196 kernel_end
= __pa_symbol(_end
);
199 * try bottom-up allocation only when bottom-up mode
200 * is set and @end is above the kernel image.
202 if (memblock_bottom_up() && end
> kernel_end
) {
203 phys_addr_t bottom_up_start
;
205 /* make sure we will allocate above the kernel */
206 bottom_up_start
= max(start
, kernel_end
);
208 /* ok, try bottom-up allocation first */
209 ret
= __memblock_find_range_bottom_up(bottom_up_start
, end
,
215 * we always limit bottom-up allocation above the kernel,
216 * but top-down allocation doesn't have the limit, so
217 * retrying top-down allocation may succeed when bottom-up
220 * bottom-up allocation is expected to be fail very rarely,
221 * so we use WARN_ONCE() here to see the stack trace if
224 WARN_ONCE(1, "memblock: bottom-up allocation failed, "
225 "memory hotunplug may be affected\n");
228 return __memblock_find_range_top_down(start
, end
, size
, align
, nid
);
232 * memblock_find_in_range - find free area in given range
233 * @start: start of candidate range
234 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_{ANYWHERE|ACCESSIBLE}
235 * @size: size of free area to find
236 * @align: alignment of free area to find
238 * Find @size free area aligned to @align in the specified range.
241 * Found address on success, 0 on failure.
243 phys_addr_t __init_memblock
memblock_find_in_range(phys_addr_t start
,
244 phys_addr_t end
, phys_addr_t size
,
247 return memblock_find_in_range_node(size
, align
, start
, end
,
251 static void __init_memblock
memblock_remove_region(struct memblock_type
*type
, unsigned long r
)
253 type
->total_size
-= type
->regions
[r
].size
;
254 memmove(&type
->regions
[r
], &type
->regions
[r
+ 1],
255 (type
->cnt
- (r
+ 1)) * sizeof(type
->regions
[r
]));
258 /* Special case for empty arrays */
259 if (type
->cnt
== 0) {
260 WARN_ON(type
->total_size
!= 0);
262 type
->regions
[0].base
= 0;
263 type
->regions
[0].size
= 0;
264 type
->regions
[0].flags
= 0;
265 memblock_set_region_node(&type
->regions
[0], MAX_NUMNODES
);
269 phys_addr_t __init_memblock
get_allocated_memblock_reserved_regions_info(
272 if (memblock
.reserved
.regions
== memblock_reserved_init_regions
)
276 * Don't allow nobootmem allocator to free reserved memory regions
278 * - CONFIG_DEBUG_FS is enabled;
279 * - CONFIG_ARCH_DISCARD_MEMBLOCK is not enabled;
280 * - reserved memory regions array have been resized during boot.
281 * Otherwise debug_fs entry "sys/kernel/debug/memblock/reserved"
282 * will show garbage instead of state of memory reservations.
284 if (IS_ENABLED(CONFIG_DEBUG_FS
) &&
285 !IS_ENABLED(CONFIG_ARCH_DISCARD_MEMBLOCK
))
288 *addr
= __pa(memblock
.reserved
.regions
);
290 return PAGE_ALIGN(sizeof(struct memblock_region
) *
291 memblock
.reserved
.max
);
295 * memblock_double_array - double the size of the memblock regions array
296 * @type: memblock type of the regions array being doubled
297 * @new_area_start: starting address of memory range to avoid overlap with
298 * @new_area_size: size of memory range to avoid overlap with
300 * Double the size of the @type regions array. If memblock is being used to
301 * allocate memory for a new reserved regions array and there is a previously
302 * allocated memory range [@new_area_start,@new_area_start+@new_area_size]
303 * waiting to be reserved, ensure the memory used by the new array does
307 * 0 on success, -1 on failure.
309 static int __init_memblock
memblock_double_array(struct memblock_type
*type
,
310 phys_addr_t new_area_start
,
311 phys_addr_t new_area_size
)
313 struct memblock_region
*new_array
, *old_array
;
314 phys_addr_t old_alloc_size
, new_alloc_size
;
315 phys_addr_t old_size
, new_size
, addr
;
316 int use_slab
= slab_is_available();
319 /* We don't allow resizing until we know about the reserved regions
320 * of memory that aren't suitable for allocation
322 if (!memblock_can_resize
)
325 /* Calculate new doubled size */
326 old_size
= type
->max
* sizeof(struct memblock_region
);
327 new_size
= old_size
<< 1;
329 * We need to allocated new one align to PAGE_SIZE,
330 * so we can free them completely later.
332 old_alloc_size
= PAGE_ALIGN(old_size
);
333 new_alloc_size
= PAGE_ALIGN(new_size
);
335 /* Retrieve the slab flag */
336 if (type
== &memblock
.memory
)
337 in_slab
= &memblock_memory_in_slab
;
339 in_slab
= &memblock_reserved_in_slab
;
341 /* Try to find some space for it.
343 * WARNING: We assume that either slab_is_available() and we use it or
344 * we use MEMBLOCK for allocations. That means that this is unsafe to
345 * use when bootmem is currently active (unless bootmem itself is
346 * implemented on top of MEMBLOCK which isn't the case yet)
348 * This should however not be an issue for now, as we currently only
349 * call into MEMBLOCK while it's still active, or much later when slab
350 * is active for memory hotplug operations
353 new_array
= kmalloc(new_size
, GFP_KERNEL
);
354 addr
= new_array
? __pa(new_array
) : 0;
356 /* only exclude range when trying to double reserved.regions */
357 if (type
!= &memblock
.reserved
)
358 new_area_start
= new_area_size
= 0;
360 addr
= memblock_find_in_range(new_area_start
+ new_area_size
,
361 memblock
.current_limit
,
362 new_alloc_size
, PAGE_SIZE
);
363 if (!addr
&& new_area_size
)
364 addr
= memblock_find_in_range(0,
365 min(new_area_start
, memblock
.current_limit
),
366 new_alloc_size
, PAGE_SIZE
);
368 new_array
= addr
? __va(addr
) : NULL
;
371 pr_err("memblock: Failed to double %s array from %ld to %ld entries !\n",
372 memblock_type_name(type
), type
->max
, type
->max
* 2);
376 memblock_dbg("memblock: %s is doubled to %ld at [%#010llx-%#010llx]",
377 memblock_type_name(type
), type
->max
* 2, (u64
)addr
,
378 (u64
)addr
+ new_size
- 1);
381 * Found space, we now need to move the array over before we add the
382 * reserved region since it may be our reserved array itself that is
385 memcpy(new_array
, type
->regions
, old_size
);
386 memset(new_array
+ type
->max
, 0, old_size
);
387 old_array
= type
->regions
;
388 type
->regions
= new_array
;
391 /* Free old array. We needn't free it if the array is the static one */
394 else if (old_array
!= memblock_memory_init_regions
&&
395 old_array
!= memblock_reserved_init_regions
)
396 memblock_free(__pa(old_array
), old_alloc_size
);
399 * Reserve the new array if that comes from the memblock. Otherwise, we
403 BUG_ON(memblock_reserve(addr
, new_alloc_size
));
405 /* Update slab flag */
412 * memblock_merge_regions - merge neighboring compatible regions
413 * @type: memblock type to scan
415 * Scan @type and merge neighboring compatible regions.
417 static void __init_memblock
memblock_merge_regions(struct memblock_type
*type
)
421 /* cnt never goes below 1 */
422 while (i
< type
->cnt
- 1) {
423 struct memblock_region
*this = &type
->regions
[i
];
424 struct memblock_region
*next
= &type
->regions
[i
+ 1];
426 if (this->base
+ this->size
!= next
->base
||
427 memblock_get_region_node(this) !=
428 memblock_get_region_node(next
) ||
429 this->flags
!= next
->flags
) {
430 BUG_ON(this->base
+ this->size
> next
->base
);
435 this->size
+= next
->size
;
436 /* move forward from next + 1, index of which is i + 2 */
437 memmove(next
, next
+ 1, (type
->cnt
- (i
+ 2)) * sizeof(*next
));
443 * memblock_insert_region - insert new memblock region
444 * @type: memblock type to insert into
445 * @idx: index for the insertion point
446 * @base: base address of the new region
447 * @size: size of the new region
448 * @nid: node id of the new region
449 * @flags: flags of the new region
451 * Insert new memblock region [@base,@base+@size) into @type at @idx.
452 * @type must already have extra room to accomodate the new region.
454 static void __init_memblock
memblock_insert_region(struct memblock_type
*type
,
455 int idx
, phys_addr_t base
,
457 int nid
, unsigned long flags
)
459 struct memblock_region
*rgn
= &type
->regions
[idx
];
461 BUG_ON(type
->cnt
>= type
->max
);
462 memmove(rgn
+ 1, rgn
, (type
->cnt
- idx
) * sizeof(*rgn
));
466 memblock_set_region_node(rgn
, nid
);
468 type
->total_size
+= size
;
472 * memblock_add_region - add new memblock region
473 * @type: memblock type to add new region into
474 * @base: base address of the new region
475 * @size: size of the new region
476 * @nid: nid of the new region
477 * @flags: flags of the new region
479 * Add new memblock region [@base,@base+@size) into @type. The new region
480 * is allowed to overlap with existing ones - overlaps don't affect already
481 * existing regions. @type is guaranteed to be minimal (all neighbouring
482 * compatible regions are merged) after the addition.
485 * 0 on success, -errno on failure.
487 static int __init_memblock
memblock_add_region(struct memblock_type
*type
,
488 phys_addr_t base
, phys_addr_t size
,
489 int nid
, unsigned long flags
)
492 phys_addr_t obase
= base
;
493 phys_addr_t end
= base
+ memblock_cap_size(base
, &size
);
499 /* special case for empty array */
500 if (type
->regions
[0].size
== 0) {
501 WARN_ON(type
->cnt
!= 1 || type
->total_size
);
502 type
->regions
[0].base
= base
;
503 type
->regions
[0].size
= size
;
504 type
->regions
[0].flags
= flags
;
505 memblock_set_region_node(&type
->regions
[0], nid
);
506 type
->total_size
= size
;
511 * The following is executed twice. Once with %false @insert and
512 * then with %true. The first counts the number of regions needed
513 * to accomodate the new area. The second actually inserts them.
518 for (i
= 0; i
< type
->cnt
; i
++) {
519 struct memblock_region
*rgn
= &type
->regions
[i
];
520 phys_addr_t rbase
= rgn
->base
;
521 phys_addr_t rend
= rbase
+ rgn
->size
;
528 * @rgn overlaps. If it separates the lower part of new
529 * area, insert that portion.
534 memblock_insert_region(type
, i
++, base
,
538 /* area below @rend is dealt with, forget about it */
539 base
= min(rend
, end
);
542 /* insert the remaining portion */
546 memblock_insert_region(type
, i
, base
, end
- base
,
551 * If this was the first round, resize array and repeat for actual
552 * insertions; otherwise, merge and return.
555 while (type
->cnt
+ nr_new
> type
->max
)
556 if (memblock_double_array(type
, obase
, size
) < 0)
561 memblock_merge_regions(type
);
566 int __init_memblock
memblock_add_node(phys_addr_t base
, phys_addr_t size
,
569 return memblock_add_region(&memblock
.memory
, base
, size
, nid
, 0);
572 int __init_memblock
memblock_add(phys_addr_t base
, phys_addr_t size
)
574 return memblock_add_region(&memblock
.memory
, base
, size
,
579 * memblock_isolate_range - isolate given range into disjoint memblocks
580 * @type: memblock type to isolate range for
581 * @base: base of range to isolate
582 * @size: size of range to isolate
583 * @start_rgn: out parameter for the start of isolated region
584 * @end_rgn: out parameter for the end of isolated region
586 * Walk @type and ensure that regions don't cross the boundaries defined by
587 * [@base,@base+@size). Crossing regions are split at the boundaries,
588 * which may create at most two more regions. The index of the first
589 * region inside the range is returned in *@start_rgn and end in *@end_rgn.
592 * 0 on success, -errno on failure.
594 static int __init_memblock
memblock_isolate_range(struct memblock_type
*type
,
595 phys_addr_t base
, phys_addr_t size
,
596 int *start_rgn
, int *end_rgn
)
598 phys_addr_t end
= base
+ memblock_cap_size(base
, &size
);
601 *start_rgn
= *end_rgn
= 0;
606 /* we'll create at most two more regions */
607 while (type
->cnt
+ 2 > type
->max
)
608 if (memblock_double_array(type
, base
, size
) < 0)
611 for (i
= 0; i
< type
->cnt
; i
++) {
612 struct memblock_region
*rgn
= &type
->regions
[i
];
613 phys_addr_t rbase
= rgn
->base
;
614 phys_addr_t rend
= rbase
+ rgn
->size
;
623 * @rgn intersects from below. Split and continue
624 * to process the next region - the new top half.
627 rgn
->size
-= base
- rbase
;
628 type
->total_size
-= base
- rbase
;
629 memblock_insert_region(type
, i
, rbase
, base
- rbase
,
630 memblock_get_region_node(rgn
),
632 } else if (rend
> end
) {
634 * @rgn intersects from above. Split and redo the
635 * current region - the new bottom half.
638 rgn
->size
-= end
- rbase
;
639 type
->total_size
-= end
- rbase
;
640 memblock_insert_region(type
, i
--, rbase
, end
- rbase
,
641 memblock_get_region_node(rgn
),
644 /* @rgn is fully contained, record it */
654 static int __init_memblock
__memblock_remove(struct memblock_type
*type
,
655 phys_addr_t base
, phys_addr_t size
)
657 int start_rgn
, end_rgn
;
660 ret
= memblock_isolate_range(type
, base
, size
, &start_rgn
, &end_rgn
);
664 for (i
= end_rgn
- 1; i
>= start_rgn
; i
--)
665 memblock_remove_region(type
, i
);
669 int __init_memblock
memblock_remove(phys_addr_t base
, phys_addr_t size
)
671 return __memblock_remove(&memblock
.memory
, base
, size
);
674 int __init_memblock
memblock_free(phys_addr_t base
, phys_addr_t size
)
676 memblock_dbg(" memblock_free: [%#016llx-%#016llx] %pF\n",
677 (unsigned long long)base
,
678 (unsigned long long)base
+ size
- 1,
681 return __memblock_remove(&memblock
.reserved
, base
, size
);
684 static int __init_memblock
memblock_reserve_region(phys_addr_t base
,
689 struct memblock_type
*_rgn
= &memblock
.reserved
;
691 memblock_dbg("memblock_reserve: [%#016llx-%#016llx] flags %#02lx %pF\n",
692 (unsigned long long)base
,
693 (unsigned long long)base
+ size
- 1,
694 flags
, (void *)_RET_IP_
);
696 return memblock_add_region(_rgn
, base
, size
, nid
, flags
);
699 int __init_memblock
memblock_reserve(phys_addr_t base
, phys_addr_t size
)
701 return memblock_reserve_region(base
, size
, MAX_NUMNODES
, 0);
705 * memblock_mark_hotplug - Mark hotpluggable memory with flag MEMBLOCK_HOTPLUG.
706 * @base: the base phys addr of the region
707 * @size: the size of the region
709 * This function isolates region [@base, @base + @size), and mark it with flag
712 * Return 0 on succees, -errno on failure.
714 int __init_memblock
memblock_mark_hotplug(phys_addr_t base
, phys_addr_t size
)
716 struct memblock_type
*type
= &memblock
.memory
;
717 int i
, ret
, start_rgn
, end_rgn
;
719 ret
= memblock_isolate_range(type
, base
, size
, &start_rgn
, &end_rgn
);
723 for (i
= start_rgn
; i
< end_rgn
; i
++)
724 memblock_set_region_flags(&type
->regions
[i
], MEMBLOCK_HOTPLUG
);
726 memblock_merge_regions(type
);
731 * memblock_clear_hotplug - Clear flag MEMBLOCK_HOTPLUG for a specified region.
732 * @base: the base phys addr of the region
733 * @size: the size of the region
735 * This function isolates region [@base, @base + @size), and clear flag
736 * MEMBLOCK_HOTPLUG for the isolated regions.
738 * Return 0 on succees, -errno on failure.
740 int __init_memblock
memblock_clear_hotplug(phys_addr_t base
, phys_addr_t size
)
742 struct memblock_type
*type
= &memblock
.memory
;
743 int i
, ret
, start_rgn
, end_rgn
;
745 ret
= memblock_isolate_range(type
, base
, size
, &start_rgn
, &end_rgn
);
749 for (i
= start_rgn
; i
< end_rgn
; i
++)
750 memblock_clear_region_flags(&type
->regions
[i
],
753 memblock_merge_regions(type
);
758 * __next_free_mem_range - next function for for_each_free_mem_range()
759 * @idx: pointer to u64 loop variable
760 * @nid: node selector, %NUMA_NO_NODE for all nodes
761 * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL
762 * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL
763 * @out_nid: ptr to int for nid of the range, can be %NULL
765 * Find the first free area from *@idx which matches @nid, fill the out
766 * parameters, and update *@idx for the next iteration. The lower 32bit of
767 * *@idx contains index into memory region and the upper 32bit indexes the
768 * areas before each reserved region. For example, if reserved regions
769 * look like the following,
771 * 0:[0-16), 1:[32-48), 2:[128-130)
773 * The upper 32bit indexes the following regions.
775 * 0:[0-0), 1:[16-32), 2:[48-128), 3:[130-MAX)
777 * As both region arrays are sorted, the function advances the two indices
778 * in lockstep and returns each intersection.
780 void __init_memblock
__next_free_mem_range(u64
*idx
, int nid
,
781 phys_addr_t
*out_start
,
782 phys_addr_t
*out_end
, int *out_nid
)
784 struct memblock_type
*mem
= &memblock
.memory
;
785 struct memblock_type
*rsv
= &memblock
.reserved
;
786 int mi
= *idx
& 0xffffffff;
789 if (WARN_ONCE(nid
== MAX_NUMNODES
, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
792 for ( ; mi
< mem
->cnt
; mi
++) {
793 struct memblock_region
*m
= &mem
->regions
[mi
];
794 phys_addr_t m_start
= m
->base
;
795 phys_addr_t m_end
= m
->base
+ m
->size
;
797 /* only memory regions are associated with nodes, check it */
798 if (nid
!= NUMA_NO_NODE
&& nid
!= memblock_get_region_node(m
))
801 /* scan areas before each reservation for intersection */
802 for ( ; ri
< rsv
->cnt
+ 1; ri
++) {
803 struct memblock_region
*r
= &rsv
->regions
[ri
];
804 phys_addr_t r_start
= ri
? r
[-1].base
+ r
[-1].size
: 0;
805 phys_addr_t r_end
= ri
< rsv
->cnt
? r
->base
: ULLONG_MAX
;
807 /* if ri advanced past mi, break out to advance mi */
808 if (r_start
>= m_end
)
810 /* if the two regions intersect, we're done */
811 if (m_start
< r_end
) {
813 *out_start
= max(m_start
, r_start
);
815 *out_end
= min(m_end
, r_end
);
817 *out_nid
= memblock_get_region_node(m
);
819 * The region which ends first is advanced
820 * for the next iteration.
826 *idx
= (u32
)mi
| (u64
)ri
<< 32;
832 /* signal end of iteration */
837 * __next_free_mem_range_rev - next function for for_each_free_mem_range_reverse()
838 * @idx: pointer to u64 loop variable
839 * @nid: nid: node selector, %NUMA_NO_NODE for all nodes
840 * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL
841 * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL
842 * @out_nid: ptr to int for nid of the range, can be %NULL
844 * Reverse of __next_free_mem_range().
846 * Linux kernel cannot migrate pages used by itself. Memory hotplug users won't
847 * be able to hot-remove hotpluggable memory used by the kernel. So this
848 * function skip hotpluggable regions if needed when allocating memory for the
851 void __init_memblock
__next_free_mem_range_rev(u64
*idx
, int nid
,
852 phys_addr_t
*out_start
,
853 phys_addr_t
*out_end
, int *out_nid
)
855 struct memblock_type
*mem
= &memblock
.memory
;
856 struct memblock_type
*rsv
= &memblock
.reserved
;
857 int mi
= *idx
& 0xffffffff;
860 if (WARN_ONCE(nid
== MAX_NUMNODES
, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
863 if (*idx
== (u64
)ULLONG_MAX
) {
868 for ( ; mi
>= 0; mi
--) {
869 struct memblock_region
*m
= &mem
->regions
[mi
];
870 phys_addr_t m_start
= m
->base
;
871 phys_addr_t m_end
= m
->base
+ m
->size
;
873 /* only memory regions are associated with nodes, check it */
874 if (nid
!= NUMA_NO_NODE
&& nid
!= memblock_get_region_node(m
))
877 /* skip hotpluggable memory regions if needed */
878 if (movable_node_is_enabled() && memblock_is_hotpluggable(m
))
881 /* scan areas before each reservation for intersection */
882 for ( ; ri
>= 0; ri
--) {
883 struct memblock_region
*r
= &rsv
->regions
[ri
];
884 phys_addr_t r_start
= ri
? r
[-1].base
+ r
[-1].size
: 0;
885 phys_addr_t r_end
= ri
< rsv
->cnt
? r
->base
: ULLONG_MAX
;
887 /* if ri advanced past mi, break out to advance mi */
888 if (r_end
<= m_start
)
890 /* if the two regions intersect, we're done */
891 if (m_end
> r_start
) {
893 *out_start
= max(m_start
, r_start
);
895 *out_end
= min(m_end
, r_end
);
897 *out_nid
= memblock_get_region_node(m
);
899 if (m_start
>= r_start
)
903 *idx
= (u32
)mi
| (u64
)ri
<< 32;
912 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
914 * Common iterator interface used to define for_each_mem_range().
916 void __init_memblock
__next_mem_pfn_range(int *idx
, int nid
,
917 unsigned long *out_start_pfn
,
918 unsigned long *out_end_pfn
, int *out_nid
)
920 struct memblock_type
*type
= &memblock
.memory
;
921 struct memblock_region
*r
;
923 while (++*idx
< type
->cnt
) {
924 r
= &type
->regions
[*idx
];
926 if (PFN_UP(r
->base
) >= PFN_DOWN(r
->base
+ r
->size
))
928 if (nid
== MAX_NUMNODES
|| nid
== r
->nid
)
931 if (*idx
>= type
->cnt
) {
937 *out_start_pfn
= PFN_UP(r
->base
);
939 *out_end_pfn
= PFN_DOWN(r
->base
+ r
->size
);
945 * memblock_set_node - set node ID on memblock regions
946 * @base: base of area to set node ID for
947 * @size: size of area to set node ID for
948 * @type: memblock type to set node ID for
949 * @nid: node ID to set
951 * Set the nid of memblock @type regions in [@base,@base+@size) to @nid.
952 * Regions which cross the area boundaries are split as necessary.
955 * 0 on success, -errno on failure.
957 int __init_memblock
memblock_set_node(phys_addr_t base
, phys_addr_t size
,
958 struct memblock_type
*type
, int nid
)
960 int start_rgn
, end_rgn
;
963 ret
= memblock_isolate_range(type
, base
, size
, &start_rgn
, &end_rgn
);
967 for (i
= start_rgn
; i
< end_rgn
; i
++)
968 memblock_set_region_node(&type
->regions
[i
], nid
);
970 memblock_merge_regions(type
);
973 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
975 static phys_addr_t __init
memblock_alloc_base_nid(phys_addr_t size
,
976 phys_addr_t align
, phys_addr_t max_addr
,
982 align
= SMP_CACHE_BYTES
;
984 /* align @size to avoid excessive fragmentation on reserved array */
985 size
= round_up(size
, align
);
987 found
= memblock_find_in_range_node(size
, align
, 0, max_addr
, nid
);
988 if (found
&& !memblock_reserve(found
, size
))
994 phys_addr_t __init
memblock_alloc_nid(phys_addr_t size
, phys_addr_t align
, int nid
)
996 return memblock_alloc_base_nid(size
, align
, MEMBLOCK_ALLOC_ACCESSIBLE
, nid
);
999 phys_addr_t __init
__memblock_alloc_base(phys_addr_t size
, phys_addr_t align
, phys_addr_t max_addr
)
1001 return memblock_alloc_base_nid(size
, align
, max_addr
, NUMA_NO_NODE
);
1004 phys_addr_t __init
memblock_alloc_base(phys_addr_t size
, phys_addr_t align
, phys_addr_t max_addr
)
1008 alloc
= __memblock_alloc_base(size
, align
, max_addr
);
1011 panic("ERROR: Failed to allocate 0x%llx bytes below 0x%llx.\n",
1012 (unsigned long long) size
, (unsigned long long) max_addr
);
1017 phys_addr_t __init
memblock_alloc(phys_addr_t size
, phys_addr_t align
)
1019 return memblock_alloc_base(size
, align
, MEMBLOCK_ALLOC_ACCESSIBLE
);
1022 phys_addr_t __init
memblock_alloc_try_nid(phys_addr_t size
, phys_addr_t align
, int nid
)
1024 phys_addr_t res
= memblock_alloc_nid(size
, align
, nid
);
1028 return memblock_alloc_base(size
, align
, MEMBLOCK_ALLOC_ACCESSIBLE
);
1032 * memblock_virt_alloc_internal - allocate boot memory block
1033 * @size: size of memory block to be allocated in bytes
1034 * @align: alignment of the region and block's size
1035 * @min_addr: the lower bound of the memory region to allocate (phys address)
1036 * @max_addr: the upper bound of the memory region to allocate (phys address)
1037 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1039 * The @min_addr limit is dropped if it can not be satisfied and the allocation
1040 * will fall back to memory below @min_addr. Also, allocation may fall back
1041 * to any node in the system if the specified node can not
1042 * hold the requested memory.
1044 * The allocation is performed from memory region limited by
1045 * memblock.current_limit if @max_addr == %BOOTMEM_ALLOC_ACCESSIBLE.
1047 * The memory block is aligned on SMP_CACHE_BYTES if @align == 0.
1049 * The phys address of allocated boot memory block is converted to virtual and
1050 * allocated memory is reset to 0.
1052 * In addition, function sets the min_count to 0 using kmemleak_alloc for
1053 * allocated boot memory block, so that it is never reported as leaks.
1056 * Virtual address of allocated memory block on success, NULL on failure.
1058 static void * __init
memblock_virt_alloc_internal(
1059 phys_addr_t size
, phys_addr_t align
,
1060 phys_addr_t min_addr
, phys_addr_t max_addr
,
1066 if (WARN_ONCE(nid
== MAX_NUMNODES
, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
1070 * Detect any accidental use of these APIs after slab is ready, as at
1071 * this moment memblock may be deinitialized already and its
1072 * internal data may be destroyed (after execution of free_all_bootmem)
1074 if (WARN_ON_ONCE(slab_is_available()))
1075 return kzalloc_node(size
, GFP_NOWAIT
, nid
);
1078 align
= SMP_CACHE_BYTES
;
1080 /* align @size to avoid excessive fragmentation on reserved array */
1081 size
= round_up(size
, align
);
1084 alloc
= memblock_find_in_range_node(size
, align
, min_addr
, max_addr
,
1089 if (nid
!= NUMA_NO_NODE
) {
1090 alloc
= memblock_find_in_range_node(size
, align
, min_addr
,
1091 max_addr
, NUMA_NO_NODE
);
1104 memblock_reserve(alloc
, size
);
1105 ptr
= phys_to_virt(alloc
);
1106 memset(ptr
, 0, size
);
1109 * The min_count is set to 0 so that bootmem allocated blocks
1110 * are never reported as leaks. This is because many of these blocks
1111 * are only referred via the physical address which is not
1112 * looked up by kmemleak.
1114 kmemleak_alloc(ptr
, size
, 0, 0);
1123 * memblock_virt_alloc_try_nid_nopanic - allocate boot memory block
1124 * @size: size of memory block to be allocated in bytes
1125 * @align: alignment of the region and block's size
1126 * @min_addr: the lower bound of the memory region from where the allocation
1127 * is preferred (phys address)
1128 * @max_addr: the upper bound of the memory region from where the allocation
1129 * is preferred (phys address), or %BOOTMEM_ALLOC_ACCESSIBLE to
1130 * allocate only from memory limited by memblock.current_limit value
1131 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1133 * Public version of _memblock_virt_alloc_try_nid_nopanic() which provides
1134 * additional debug information (including caller info), if enabled.
1137 * Virtual address of allocated memory block on success, NULL on failure.
1139 void * __init
memblock_virt_alloc_try_nid_nopanic(
1140 phys_addr_t size
, phys_addr_t align
,
1141 phys_addr_t min_addr
, phys_addr_t max_addr
,
1144 memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=0x%llx max_addr=0x%llx %pF\n",
1145 __func__
, (u64
)size
, (u64
)align
, nid
, (u64
)min_addr
,
1146 (u64
)max_addr
, (void *)_RET_IP_
);
1147 return memblock_virt_alloc_internal(size
, align
, min_addr
,
1152 * memblock_virt_alloc_try_nid - allocate boot memory block with panicking
1153 * @size: size of memory block to be allocated in bytes
1154 * @align: alignment of the region and block's size
1155 * @min_addr: the lower bound of the memory region from where the allocation
1156 * is preferred (phys address)
1157 * @max_addr: the upper bound of the memory region from where the allocation
1158 * is preferred (phys address), or %BOOTMEM_ALLOC_ACCESSIBLE to
1159 * allocate only from memory limited by memblock.current_limit value
1160 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1162 * Public panicking version of _memblock_virt_alloc_try_nid_nopanic()
1163 * which provides debug information (including caller info), if enabled,
1164 * and panics if the request can not be satisfied.
1167 * Virtual address of allocated memory block on success, NULL on failure.
1169 void * __init
memblock_virt_alloc_try_nid(
1170 phys_addr_t size
, phys_addr_t align
,
1171 phys_addr_t min_addr
, phys_addr_t max_addr
,
1176 memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=0x%llx max_addr=0x%llx %pF\n",
1177 __func__
, (u64
)size
, (u64
)align
, nid
, (u64
)min_addr
,
1178 (u64
)max_addr
, (void *)_RET_IP_
);
1179 ptr
= memblock_virt_alloc_internal(size
, align
,
1180 min_addr
, max_addr
, nid
);
1184 panic("%s: Failed to allocate %llu bytes align=0x%llx nid=%d from=0x%llx max_addr=0x%llx\n",
1185 __func__
, (u64
)size
, (u64
)align
, nid
, (u64
)min_addr
,
1191 * __memblock_free_early - free boot memory block
1192 * @base: phys starting address of the boot memory block
1193 * @size: size of the boot memory block in bytes
1195 * Free boot memory block previously allocated by memblock_virt_alloc_xx() API.
1196 * The freeing memory will not be released to the buddy allocator.
1198 void __init
__memblock_free_early(phys_addr_t base
, phys_addr_t size
)
1200 memblock_dbg("%s: [%#016llx-%#016llx] %pF\n",
1201 __func__
, (u64
)base
, (u64
)base
+ size
- 1,
1203 kmemleak_free_part(__va(base
), size
);
1204 __memblock_remove(&memblock
.reserved
, base
, size
);
1208 * __memblock_free_late - free bootmem block pages directly to buddy allocator
1209 * @addr: phys starting address of the boot memory block
1210 * @size: size of the boot memory block in bytes
1212 * This is only useful when the bootmem allocator has already been torn
1213 * down, but we are still initializing the system. Pages are released directly
1214 * to the buddy allocator, no bootmem metadata is updated because it is gone.
1216 void __init
__memblock_free_late(phys_addr_t base
, phys_addr_t size
)
1220 memblock_dbg("%s: [%#016llx-%#016llx] %pF\n",
1221 __func__
, (u64
)base
, (u64
)base
+ size
- 1,
1223 kmemleak_free_part(__va(base
), size
);
1224 cursor
= PFN_UP(base
);
1225 end
= PFN_DOWN(base
+ size
);
1227 for (; cursor
< end
; cursor
++) {
1228 __free_pages_bootmem(pfn_to_page(cursor
), 0);
1234 * Remaining API functions
1237 phys_addr_t __init
memblock_phys_mem_size(void)
1239 return memblock
.memory
.total_size
;
1242 phys_addr_t __init
memblock_mem_size(unsigned long limit_pfn
)
1244 unsigned long pages
= 0;
1245 struct memblock_region
*r
;
1246 unsigned long start_pfn
, end_pfn
;
1248 for_each_memblock(memory
, r
) {
1249 start_pfn
= memblock_region_memory_base_pfn(r
);
1250 end_pfn
= memblock_region_memory_end_pfn(r
);
1251 start_pfn
= min_t(unsigned long, start_pfn
, limit_pfn
);
1252 end_pfn
= min_t(unsigned long, end_pfn
, limit_pfn
);
1253 pages
+= end_pfn
- start_pfn
;
1256 return (phys_addr_t
)pages
<< PAGE_SHIFT
;
1259 /* lowest address */
1260 phys_addr_t __init_memblock
memblock_start_of_DRAM(void)
1262 return memblock
.memory
.regions
[0].base
;
1265 phys_addr_t __init_memblock
memblock_end_of_DRAM(void)
1267 int idx
= memblock
.memory
.cnt
- 1;
1269 return (memblock
.memory
.regions
[idx
].base
+ memblock
.memory
.regions
[idx
].size
);
1272 void __init
memblock_enforce_memory_limit(phys_addr_t limit
)
1275 phys_addr_t max_addr
= (phys_addr_t
)ULLONG_MAX
;
1280 /* find out max address */
1281 for (i
= 0; i
< memblock
.memory
.cnt
; i
++) {
1282 struct memblock_region
*r
= &memblock
.memory
.regions
[i
];
1284 if (limit
<= r
->size
) {
1285 max_addr
= r
->base
+ limit
;
1291 /* truncate both memory and reserved regions */
1292 __memblock_remove(&memblock
.memory
, max_addr
, (phys_addr_t
)ULLONG_MAX
);
1293 __memblock_remove(&memblock
.reserved
, max_addr
, (phys_addr_t
)ULLONG_MAX
);
1296 static int __init_memblock
memblock_search(struct memblock_type
*type
, phys_addr_t addr
)
1298 unsigned int left
= 0, right
= type
->cnt
;
1301 unsigned int mid
= (right
+ left
) / 2;
1303 if (addr
< type
->regions
[mid
].base
)
1305 else if (addr
>= (type
->regions
[mid
].base
+
1306 type
->regions
[mid
].size
))
1310 } while (left
< right
);
1314 int __init
memblock_is_reserved(phys_addr_t addr
)
1316 return memblock_search(&memblock
.reserved
, addr
) != -1;
1319 int __init_memblock
memblock_is_memory(phys_addr_t addr
)
1321 return memblock_search(&memblock
.memory
, addr
) != -1;
1324 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
1325 int __init_memblock
memblock_search_pfn_nid(unsigned long pfn
,
1326 unsigned long *start_pfn
, unsigned long *end_pfn
)
1328 struct memblock_type
*type
= &memblock
.memory
;
1329 int mid
= memblock_search(type
, (phys_addr_t
)pfn
<< PAGE_SHIFT
);
1334 *start_pfn
= type
->regions
[mid
].base
>> PAGE_SHIFT
;
1335 *end_pfn
= (type
->regions
[mid
].base
+ type
->regions
[mid
].size
)
1338 return type
->regions
[mid
].nid
;
1343 * memblock_is_region_memory - check if a region is a subset of memory
1344 * @base: base of region to check
1345 * @size: size of region to check
1347 * Check if the region [@base, @base+@size) is a subset of a memory block.
1350 * 0 if false, non-zero if true
1352 int __init_memblock
memblock_is_region_memory(phys_addr_t base
, phys_addr_t size
)
1354 int idx
= memblock_search(&memblock
.memory
, base
);
1355 phys_addr_t end
= base
+ memblock_cap_size(base
, &size
);
1359 return memblock
.memory
.regions
[idx
].base
<= base
&&
1360 (memblock
.memory
.regions
[idx
].base
+
1361 memblock
.memory
.regions
[idx
].size
) >= end
;
1365 * memblock_is_region_reserved - check if a region intersects reserved memory
1366 * @base: base of region to check
1367 * @size: size of region to check
1369 * Check if the region [@base, @base+@size) intersects a reserved memory block.
1372 * 0 if false, non-zero if true
1374 int __init_memblock
memblock_is_region_reserved(phys_addr_t base
, phys_addr_t size
)
1376 memblock_cap_size(base
, &size
);
1377 return memblock_overlaps_region(&memblock
.reserved
, base
, size
) >= 0;
1380 void __init_memblock
memblock_trim_memory(phys_addr_t align
)
1383 phys_addr_t start
, end
, orig_start
, orig_end
;
1384 struct memblock_type
*mem
= &memblock
.memory
;
1386 for (i
= 0; i
< mem
->cnt
; i
++) {
1387 orig_start
= mem
->regions
[i
].base
;
1388 orig_end
= mem
->regions
[i
].base
+ mem
->regions
[i
].size
;
1389 start
= round_up(orig_start
, align
);
1390 end
= round_down(orig_end
, align
);
1392 if (start
== orig_start
&& end
== orig_end
)
1396 mem
->regions
[i
].base
= start
;
1397 mem
->regions
[i
].size
= end
- start
;
1399 memblock_remove_region(mem
, i
);
1405 void __init_memblock
memblock_set_current_limit(phys_addr_t limit
)
1407 memblock
.current_limit
= limit
;
1410 static void __init_memblock
memblock_dump(struct memblock_type
*type
, char *name
)
1412 unsigned long long base
, size
;
1413 unsigned long flags
;
1416 pr_info(" %s.cnt = 0x%lx\n", name
, type
->cnt
);
1418 for (i
= 0; i
< type
->cnt
; i
++) {
1419 struct memblock_region
*rgn
= &type
->regions
[i
];
1420 char nid_buf
[32] = "";
1425 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
1426 if (memblock_get_region_node(rgn
) != MAX_NUMNODES
)
1427 snprintf(nid_buf
, sizeof(nid_buf
), " on node %d",
1428 memblock_get_region_node(rgn
));
1430 pr_info(" %s[%#x]\t[%#016llx-%#016llx], %#llx bytes%s flags: %#lx\n",
1431 name
, i
, base
, base
+ size
- 1, size
, nid_buf
, flags
);
1435 void __init_memblock
__memblock_dump_all(void)
1437 pr_info("MEMBLOCK configuration:\n");
1438 pr_info(" memory size = %#llx reserved size = %#llx\n",
1439 (unsigned long long)memblock
.memory
.total_size
,
1440 (unsigned long long)memblock
.reserved
.total_size
);
1442 memblock_dump(&memblock
.memory
, "memory");
1443 memblock_dump(&memblock
.reserved
, "reserved");
1446 void __init
memblock_allow_resize(void)
1448 memblock_can_resize
= 1;
1451 static int __init
early_memblock(char *p
)
1453 if (p
&& strstr(p
, "debug"))
1457 early_param("memblock", early_memblock
);
1459 #if defined(CONFIG_DEBUG_FS) && !defined(CONFIG_ARCH_DISCARD_MEMBLOCK)
1461 static int memblock_debug_show(struct seq_file
*m
, void *private)
1463 struct memblock_type
*type
= m
->private;
1464 struct memblock_region
*reg
;
1467 for (i
= 0; i
< type
->cnt
; i
++) {
1468 reg
= &type
->regions
[i
];
1469 seq_printf(m
, "%4d: ", i
);
1470 if (sizeof(phys_addr_t
) == 4)
1471 seq_printf(m
, "0x%08lx..0x%08lx\n",
1472 (unsigned long)reg
->base
,
1473 (unsigned long)(reg
->base
+ reg
->size
- 1));
1475 seq_printf(m
, "0x%016llx..0x%016llx\n",
1476 (unsigned long long)reg
->base
,
1477 (unsigned long long)(reg
->base
+ reg
->size
- 1));
1483 static int memblock_debug_open(struct inode
*inode
, struct file
*file
)
1485 return single_open(file
, memblock_debug_show
, inode
->i_private
);
1488 static const struct file_operations memblock_debug_fops
= {
1489 .open
= memblock_debug_open
,
1491 .llseek
= seq_lseek
,
1492 .release
= single_release
,
1495 static int __init
memblock_init_debugfs(void)
1497 struct dentry
*root
= debugfs_create_dir("memblock", NULL
);
1500 debugfs_create_file("memory", S_IRUGO
, root
, &memblock
.memory
, &memblock_debug_fops
);
1501 debugfs_create_file("reserved", S_IRUGO
, root
, &memblock
.reserved
, &memblock_debug_fops
);
1505 __initcall(memblock_init_debugfs
);
1507 #endif /* CONFIG_DEBUG_FS */