1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * Procedures for maintaining information about logical memory blocks.
5 * Peter Bergner, IBM Corp. June 2001.
6 * Copyright (C) 2001 Peter Bergner.
9 #include <linux/kernel.h>
10 #include <linux/slab.h>
11 #include <linux/init.h>
12 #include <linux/bitops.h>
13 #include <linux/poison.h>
14 #include <linux/pfn.h>
15 #include <linux/debugfs.h>
16 #include <linux/kmemleak.h>
17 #include <linux/seq_file.h>
18 #include <linux/memblock.h>
20 #include <asm/sections.h>
25 #define INIT_MEMBLOCK_REGIONS 128
26 #define INIT_PHYSMEM_REGIONS 4
28 #ifndef INIT_MEMBLOCK_RESERVED_REGIONS
29 # define INIT_MEMBLOCK_RESERVED_REGIONS INIT_MEMBLOCK_REGIONS
33 * DOC: memblock overview
35 * Memblock is a method of managing memory regions during the early
36 * boot period when the usual kernel memory allocators are not up and
39 * Memblock views the system memory as collections of contiguous
40 * regions. There are several types of these collections:
42 * * ``memory`` - describes the physical memory available to the
43 * kernel; this may differ from the actual physical memory installed
44 * in the system, for instance when the memory is restricted with
45 * ``mem=`` command line parameter
46 * * ``reserved`` - describes the regions that were allocated
47 * * ``physmem`` - describes the actual physical memory available during
48 * boot regardless of the possible restrictions and memory hot(un)plug;
49 * the ``physmem`` type is only available on some architectures.
51 * Each region is represented by struct memblock_region that
52 * defines the region extents, its attributes and NUMA node id on NUMA
53 * systems. Every memory type is described by the struct memblock_type
54 * which contains an array of memory regions along with
55 * the allocator metadata. The "memory" and "reserved" types are nicely
56 * wrapped with struct memblock. This structure is statically
57 * initialized at build time. The region arrays are initially sized to
58 * %INIT_MEMBLOCK_REGIONS for "memory" and %INIT_MEMBLOCK_RESERVED_REGIONS
59 * for "reserved". The region array for "physmem" is initially sized to
60 * %INIT_PHYSMEM_REGIONS.
61 * The memblock_allow_resize() enables automatic resizing of the region
62 * arrays during addition of new regions. This feature should be used
63 * with care so that memory allocated for the region array will not
64 * overlap with areas that should be reserved, for example initrd.
66 * The early architecture setup should tell memblock what the physical
67 * memory layout is by using memblock_add() or memblock_add_node()
68 * functions. The first function does not assign the region to a NUMA
69 * node and it is appropriate for UMA systems. Yet, it is possible to
70 * use it on NUMA systems as well and assign the region to a NUMA node
71 * later in the setup process using memblock_set_node(). The
72 * memblock_add_node() performs such an assignment directly.
74 * Once memblock is setup the memory can be allocated using one of the
77 * * memblock_phys_alloc*() - these functions return the **physical**
78 * address of the allocated memory
79 * * memblock_alloc*() - these functions return the **virtual** address
80 * of the allocated memory.
82 * Note, that both API variants use implicit assumptions about allowed
83 * memory ranges and the fallback methods. Consult the documentation
84 * of memblock_alloc_internal() and memblock_alloc_range_nid()
85 * functions for more elaborate description.
87 * As the system boot progresses, the architecture specific mem_init()
88 * function frees all the memory to the buddy page allocator.
90 * Unless an architecture enables %CONFIG_ARCH_KEEP_MEMBLOCK, the
91 * memblock data structures (except "physmem") will be discarded after the
92 * system initialization completes.
96 struct pglist_data __refdata contig_page_data
;
97 EXPORT_SYMBOL(contig_page_data
);
100 unsigned long max_low_pfn
;
101 unsigned long min_low_pfn
;
102 unsigned long max_pfn
;
103 unsigned long long max_possible_pfn
;
105 static struct memblock_region memblock_memory_init_regions
[INIT_MEMBLOCK_REGIONS
] __initdata_memblock
;
106 static struct memblock_region memblock_reserved_init_regions
[INIT_MEMBLOCK_RESERVED_REGIONS
] __initdata_memblock
;
107 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
108 static struct memblock_region memblock_physmem_init_regions
[INIT_PHYSMEM_REGIONS
];
111 struct memblock memblock __initdata_memblock
= {
112 .memory
.regions
= memblock_memory_init_regions
,
113 .memory
.cnt
= 1, /* empty dummy entry */
114 .memory
.max
= INIT_MEMBLOCK_REGIONS
,
115 .memory
.name
= "memory",
117 .reserved
.regions
= memblock_reserved_init_regions
,
118 .reserved
.cnt
= 1, /* empty dummy entry */
119 .reserved
.max
= INIT_MEMBLOCK_RESERVED_REGIONS
,
120 .reserved
.name
= "reserved",
123 .current_limit
= MEMBLOCK_ALLOC_ANYWHERE
,
126 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
127 struct memblock_type physmem
= {
128 .regions
= memblock_physmem_init_regions
,
129 .cnt
= 1, /* empty dummy entry */
130 .max
= INIT_PHYSMEM_REGIONS
,
136 * keep a pointer to &memblock.memory in the text section to use it in
137 * __next_mem_range() and its helpers.
138 * For architectures that do not keep memblock data after init, this
139 * pointer will be reset to NULL at memblock_discard()
141 static __refdata
struct memblock_type
*memblock_memory
= &memblock
.memory
;
143 #define for_each_memblock_type(i, memblock_type, rgn) \
144 for (i = 0, rgn = &memblock_type->regions[0]; \
145 i < memblock_type->cnt; \
146 i++, rgn = &memblock_type->regions[i])
148 #define memblock_dbg(fmt, ...) \
150 if (memblock_debug) \
151 pr_info(fmt, ##__VA_ARGS__); \
154 static int memblock_debug __initdata_memblock
;
155 static bool system_has_some_mirror __initdata_memblock
= false;
156 static int memblock_can_resize __initdata_memblock
;
157 static int memblock_memory_in_slab __initdata_memblock
= 0;
158 static int memblock_reserved_in_slab __initdata_memblock
= 0;
160 static enum memblock_flags __init_memblock
choose_memblock_flags(void)
162 return system_has_some_mirror
? MEMBLOCK_MIRROR
: MEMBLOCK_NONE
;
165 /* adjust *@size so that (@base + *@size) doesn't overflow, return new size */
166 static inline phys_addr_t
memblock_cap_size(phys_addr_t base
, phys_addr_t
*size
)
168 return *size
= min(*size
, PHYS_ADDR_MAX
- base
);
172 * Address comparison utilities
174 static unsigned long __init_memblock
memblock_addrs_overlap(phys_addr_t base1
, phys_addr_t size1
,
175 phys_addr_t base2
, phys_addr_t size2
)
177 return ((base1
< (base2
+ size2
)) && (base2
< (base1
+ size1
)));
180 bool __init_memblock
memblock_overlaps_region(struct memblock_type
*type
,
181 phys_addr_t base
, phys_addr_t size
)
185 memblock_cap_size(base
, &size
);
187 for (i
= 0; i
< type
->cnt
; i
++)
188 if (memblock_addrs_overlap(base
, size
, type
->regions
[i
].base
,
189 type
->regions
[i
].size
))
191 return i
< type
->cnt
;
195 * __memblock_find_range_bottom_up - find free area utility in bottom-up
196 * @start: start of candidate range
197 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_ANYWHERE or
198 * %MEMBLOCK_ALLOC_ACCESSIBLE
199 * @size: size of free area to find
200 * @align: alignment of free area to find
201 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
202 * @flags: pick from blocks based on memory attributes
204 * Utility called from memblock_find_in_range_node(), find free area bottom-up.
207 * Found address on success, 0 on failure.
209 static phys_addr_t __init_memblock
210 __memblock_find_range_bottom_up(phys_addr_t start
, phys_addr_t end
,
211 phys_addr_t size
, phys_addr_t align
, int nid
,
212 enum memblock_flags flags
)
214 phys_addr_t this_start
, this_end
, cand
;
217 for_each_free_mem_range(i
, nid
, flags
, &this_start
, &this_end
, NULL
) {
218 this_start
= clamp(this_start
, start
, end
);
219 this_end
= clamp(this_end
, start
, end
);
221 cand
= round_up(this_start
, align
);
222 if (cand
< this_end
&& this_end
- cand
>= size
)
230 * __memblock_find_range_top_down - find free area utility, in top-down
231 * @start: start of candidate range
232 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_ANYWHERE or
233 * %MEMBLOCK_ALLOC_ACCESSIBLE
234 * @size: size of free area to find
235 * @align: alignment of free area to find
236 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
237 * @flags: pick from blocks based on memory attributes
239 * Utility called from memblock_find_in_range_node(), find free area top-down.
242 * Found address on success, 0 on failure.
244 static phys_addr_t __init_memblock
245 __memblock_find_range_top_down(phys_addr_t start
, phys_addr_t end
,
246 phys_addr_t size
, phys_addr_t align
, int nid
,
247 enum memblock_flags flags
)
249 phys_addr_t this_start
, this_end
, cand
;
252 for_each_free_mem_range_reverse(i
, nid
, flags
, &this_start
, &this_end
,
254 this_start
= clamp(this_start
, start
, end
);
255 this_end
= clamp(this_end
, start
, end
);
260 cand
= round_down(this_end
- size
, align
);
261 if (cand
>= this_start
)
269 * memblock_find_in_range_node - find free area in given range and node
270 * @size: size of free area to find
271 * @align: alignment of free area to find
272 * @start: start of candidate range
273 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_ANYWHERE or
274 * %MEMBLOCK_ALLOC_ACCESSIBLE
275 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
276 * @flags: pick from blocks based on memory attributes
278 * Find @size free area aligned to @align in the specified range and node.
281 * Found address on success, 0 on failure.
283 static phys_addr_t __init_memblock
memblock_find_in_range_node(phys_addr_t size
,
284 phys_addr_t align
, phys_addr_t start
,
285 phys_addr_t end
, int nid
,
286 enum memblock_flags flags
)
289 if (end
== MEMBLOCK_ALLOC_ACCESSIBLE
||
290 end
== MEMBLOCK_ALLOC_KASAN
)
291 end
= memblock
.current_limit
;
293 /* avoid allocating the first page */
294 start
= max_t(phys_addr_t
, start
, PAGE_SIZE
);
295 end
= max(start
, end
);
297 if (memblock_bottom_up())
298 return __memblock_find_range_bottom_up(start
, end
, size
, align
,
301 return __memblock_find_range_top_down(start
, end
, size
, align
,
306 * memblock_find_in_range - find free area in given range
307 * @start: start of candidate range
308 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_ANYWHERE or
309 * %MEMBLOCK_ALLOC_ACCESSIBLE
310 * @size: size of free area to find
311 * @align: alignment of free area to find
313 * Find @size free area aligned to @align in the specified range.
316 * Found address on success, 0 on failure.
318 phys_addr_t __init_memblock
memblock_find_in_range(phys_addr_t start
,
319 phys_addr_t end
, phys_addr_t size
,
323 enum memblock_flags flags
= choose_memblock_flags();
326 ret
= memblock_find_in_range_node(size
, align
, start
, end
,
327 NUMA_NO_NODE
, flags
);
329 if (!ret
&& (flags
& MEMBLOCK_MIRROR
)) {
330 pr_warn("Could not allocate %pap bytes of mirrored memory\n",
332 flags
&= ~MEMBLOCK_MIRROR
;
339 static void __init_memblock
memblock_remove_region(struct memblock_type
*type
, unsigned long r
)
341 type
->total_size
-= type
->regions
[r
].size
;
342 memmove(&type
->regions
[r
], &type
->regions
[r
+ 1],
343 (type
->cnt
- (r
+ 1)) * sizeof(type
->regions
[r
]));
346 /* Special case for empty arrays */
347 if (type
->cnt
== 0) {
348 WARN_ON(type
->total_size
!= 0);
350 type
->regions
[0].base
= 0;
351 type
->regions
[0].size
= 0;
352 type
->regions
[0].flags
= 0;
353 memblock_set_region_node(&type
->regions
[0], MAX_NUMNODES
);
357 #ifndef CONFIG_ARCH_KEEP_MEMBLOCK
359 * memblock_discard - discard memory and reserved arrays if they were allocated
361 void __init
memblock_discard(void)
363 phys_addr_t addr
, size
;
365 if (memblock
.reserved
.regions
!= memblock_reserved_init_regions
) {
366 addr
= __pa(memblock
.reserved
.regions
);
367 size
= PAGE_ALIGN(sizeof(struct memblock_region
) *
368 memblock
.reserved
.max
);
369 __memblock_free_late(addr
, size
);
372 if (memblock
.memory
.regions
!= memblock_memory_init_regions
) {
373 addr
= __pa(memblock
.memory
.regions
);
374 size
= PAGE_ALIGN(sizeof(struct memblock_region
) *
375 memblock
.memory
.max
);
376 __memblock_free_late(addr
, size
);
379 memblock_memory
= NULL
;
384 * memblock_double_array - double the size of the memblock regions array
385 * @type: memblock type of the regions array being doubled
386 * @new_area_start: starting address of memory range to avoid overlap with
387 * @new_area_size: size of memory range to avoid overlap with
389 * Double the size of the @type regions array. If memblock is being used to
390 * allocate memory for a new reserved regions array and there is a previously
391 * allocated memory range [@new_area_start, @new_area_start + @new_area_size]
392 * waiting to be reserved, ensure the memory used by the new array does
396 * 0 on success, -1 on failure.
398 static int __init_memblock
memblock_double_array(struct memblock_type
*type
,
399 phys_addr_t new_area_start
,
400 phys_addr_t new_area_size
)
402 struct memblock_region
*new_array
, *old_array
;
403 phys_addr_t old_alloc_size
, new_alloc_size
;
404 phys_addr_t old_size
, new_size
, addr
, new_end
;
405 int use_slab
= slab_is_available();
408 /* We don't allow resizing until we know about the reserved regions
409 * of memory that aren't suitable for allocation
411 if (!memblock_can_resize
)
414 /* Calculate new doubled size */
415 old_size
= type
->max
* sizeof(struct memblock_region
);
416 new_size
= old_size
<< 1;
418 * We need to allocated new one align to PAGE_SIZE,
419 * so we can free them completely later.
421 old_alloc_size
= PAGE_ALIGN(old_size
);
422 new_alloc_size
= PAGE_ALIGN(new_size
);
424 /* Retrieve the slab flag */
425 if (type
== &memblock
.memory
)
426 in_slab
= &memblock_memory_in_slab
;
428 in_slab
= &memblock_reserved_in_slab
;
430 /* Try to find some space for it */
432 new_array
= kmalloc(new_size
, GFP_KERNEL
);
433 addr
= new_array
? __pa(new_array
) : 0;
435 /* only exclude range when trying to double reserved.regions */
436 if (type
!= &memblock
.reserved
)
437 new_area_start
= new_area_size
= 0;
439 addr
= memblock_find_in_range(new_area_start
+ new_area_size
,
440 memblock
.current_limit
,
441 new_alloc_size
, PAGE_SIZE
);
442 if (!addr
&& new_area_size
)
443 addr
= memblock_find_in_range(0,
444 min(new_area_start
, memblock
.current_limit
),
445 new_alloc_size
, PAGE_SIZE
);
447 new_array
= addr
? __va(addr
) : NULL
;
450 pr_err("memblock: Failed to double %s array from %ld to %ld entries !\n",
451 type
->name
, type
->max
, type
->max
* 2);
455 new_end
= addr
+ new_size
- 1;
456 memblock_dbg("memblock: %s is doubled to %ld at [%pa-%pa]",
457 type
->name
, type
->max
* 2, &addr
, &new_end
);
460 * Found space, we now need to move the array over before we add the
461 * reserved region since it may be our reserved array itself that is
464 memcpy(new_array
, type
->regions
, old_size
);
465 memset(new_array
+ type
->max
, 0, old_size
);
466 old_array
= type
->regions
;
467 type
->regions
= new_array
;
470 /* Free old array. We needn't free it if the array is the static one */
473 else if (old_array
!= memblock_memory_init_regions
&&
474 old_array
!= memblock_reserved_init_regions
)
475 memblock_free(__pa(old_array
), old_alloc_size
);
478 * Reserve the new array if that comes from the memblock. Otherwise, we
482 BUG_ON(memblock_reserve(addr
, new_alloc_size
));
484 /* Update slab flag */
491 * memblock_merge_regions - merge neighboring compatible regions
492 * @type: memblock type to scan
494 * Scan @type and merge neighboring compatible regions.
496 static void __init_memblock
memblock_merge_regions(struct memblock_type
*type
)
500 /* cnt never goes below 1 */
501 while (i
< type
->cnt
- 1) {
502 struct memblock_region
*this = &type
->regions
[i
];
503 struct memblock_region
*next
= &type
->regions
[i
+ 1];
505 if (this->base
+ this->size
!= next
->base
||
506 memblock_get_region_node(this) !=
507 memblock_get_region_node(next
) ||
508 this->flags
!= next
->flags
) {
509 BUG_ON(this->base
+ this->size
> next
->base
);
514 this->size
+= next
->size
;
515 /* move forward from next + 1, index of which is i + 2 */
516 memmove(next
, next
+ 1, (type
->cnt
- (i
+ 2)) * sizeof(*next
));
522 * memblock_insert_region - insert new memblock region
523 * @type: memblock type to insert into
524 * @idx: index for the insertion point
525 * @base: base address of the new region
526 * @size: size of the new region
527 * @nid: node id of the new region
528 * @flags: flags of the new region
530 * Insert new memblock region [@base, @base + @size) into @type at @idx.
531 * @type must already have extra room to accommodate the new region.
533 static void __init_memblock
memblock_insert_region(struct memblock_type
*type
,
534 int idx
, phys_addr_t base
,
537 enum memblock_flags flags
)
539 struct memblock_region
*rgn
= &type
->regions
[idx
];
541 BUG_ON(type
->cnt
>= type
->max
);
542 memmove(rgn
+ 1, rgn
, (type
->cnt
- idx
) * sizeof(*rgn
));
546 memblock_set_region_node(rgn
, nid
);
548 type
->total_size
+= size
;
552 * memblock_add_range - add new memblock region
553 * @type: memblock type to add new region into
554 * @base: base address of the new region
555 * @size: size of the new region
556 * @nid: nid of the new region
557 * @flags: flags of the new region
559 * Add new memblock region [@base, @base + @size) into @type. The new region
560 * is allowed to overlap with existing ones - overlaps don't affect already
561 * existing regions. @type is guaranteed to be minimal (all neighbouring
562 * compatible regions are merged) after the addition.
565 * 0 on success, -errno on failure.
567 static int __init_memblock
memblock_add_range(struct memblock_type
*type
,
568 phys_addr_t base
, phys_addr_t size
,
569 int nid
, enum memblock_flags flags
)
572 phys_addr_t obase
= base
;
573 phys_addr_t end
= base
+ memblock_cap_size(base
, &size
);
575 struct memblock_region
*rgn
;
580 /* special case for empty array */
581 if (type
->regions
[0].size
== 0) {
582 WARN_ON(type
->cnt
!= 1 || type
->total_size
);
583 type
->regions
[0].base
= base
;
584 type
->regions
[0].size
= size
;
585 type
->regions
[0].flags
= flags
;
586 memblock_set_region_node(&type
->regions
[0], nid
);
587 type
->total_size
= size
;
592 * The following is executed twice. Once with %false @insert and
593 * then with %true. The first counts the number of regions needed
594 * to accommodate the new area. The second actually inserts them.
599 for_each_memblock_type(idx
, type
, rgn
) {
600 phys_addr_t rbase
= rgn
->base
;
601 phys_addr_t rend
= rbase
+ rgn
->size
;
608 * @rgn overlaps. If it separates the lower part of new
609 * area, insert that portion.
613 WARN_ON(nid
!= memblock_get_region_node(rgn
));
615 WARN_ON(flags
!= rgn
->flags
);
618 memblock_insert_region(type
, idx
++, base
,
622 /* area below @rend is dealt with, forget about it */
623 base
= min(rend
, end
);
626 /* insert the remaining portion */
630 memblock_insert_region(type
, idx
, base
, end
- base
,
638 * If this was the first round, resize array and repeat for actual
639 * insertions; otherwise, merge and return.
642 while (type
->cnt
+ nr_new
> type
->max
)
643 if (memblock_double_array(type
, obase
, size
) < 0)
648 memblock_merge_regions(type
);
654 * memblock_add_node - add new memblock region within a NUMA node
655 * @base: base address of the new region
656 * @size: size of the new region
657 * @nid: nid of the new region
659 * Add new memblock region [@base, @base + @size) to the "memory"
660 * type. See memblock_add_range() description for mode details
663 * 0 on success, -errno on failure.
665 int __init_memblock
memblock_add_node(phys_addr_t base
, phys_addr_t size
,
668 return memblock_add_range(&memblock
.memory
, base
, size
, nid
, 0);
672 * memblock_add - add new memblock region
673 * @base: base address of the new region
674 * @size: size of the new region
676 * Add new memblock region [@base, @base + @size) to the "memory"
677 * type. See memblock_add_range() description for mode details
680 * 0 on success, -errno on failure.
682 int __init_memblock
memblock_add(phys_addr_t base
, phys_addr_t size
)
684 phys_addr_t end
= base
+ size
- 1;
686 memblock_dbg("%s: [%pa-%pa] %pS\n", __func__
,
687 &base
, &end
, (void *)_RET_IP_
);
689 return memblock_add_range(&memblock
.memory
, base
, size
, MAX_NUMNODES
, 0);
693 * memblock_isolate_range - isolate given range into disjoint memblocks
694 * @type: memblock type to isolate range for
695 * @base: base of range to isolate
696 * @size: size of range to isolate
697 * @start_rgn: out parameter for the start of isolated region
698 * @end_rgn: out parameter for the end of isolated region
700 * Walk @type and ensure that regions don't cross the boundaries defined by
701 * [@base, @base + @size). Crossing regions are split at the boundaries,
702 * which may create at most two more regions. The index of the first
703 * region inside the range is returned in *@start_rgn and end in *@end_rgn.
706 * 0 on success, -errno on failure.
708 static int __init_memblock
memblock_isolate_range(struct memblock_type
*type
,
709 phys_addr_t base
, phys_addr_t size
,
710 int *start_rgn
, int *end_rgn
)
712 phys_addr_t end
= base
+ memblock_cap_size(base
, &size
);
714 struct memblock_region
*rgn
;
716 *start_rgn
= *end_rgn
= 0;
721 /* we'll create at most two more regions */
722 while (type
->cnt
+ 2 > type
->max
)
723 if (memblock_double_array(type
, base
, size
) < 0)
726 for_each_memblock_type(idx
, type
, rgn
) {
727 phys_addr_t rbase
= rgn
->base
;
728 phys_addr_t rend
= rbase
+ rgn
->size
;
737 * @rgn intersects from below. Split and continue
738 * to process the next region - the new top half.
741 rgn
->size
-= base
- rbase
;
742 type
->total_size
-= base
- rbase
;
743 memblock_insert_region(type
, idx
, rbase
, base
- rbase
,
744 memblock_get_region_node(rgn
),
746 } else if (rend
> end
) {
748 * @rgn intersects from above. Split and redo the
749 * current region - the new bottom half.
752 rgn
->size
-= end
- rbase
;
753 type
->total_size
-= end
- rbase
;
754 memblock_insert_region(type
, idx
--, rbase
, end
- rbase
,
755 memblock_get_region_node(rgn
),
758 /* @rgn is fully contained, record it */
768 static int __init_memblock
memblock_remove_range(struct memblock_type
*type
,
769 phys_addr_t base
, phys_addr_t size
)
771 int start_rgn
, end_rgn
;
774 ret
= memblock_isolate_range(type
, base
, size
, &start_rgn
, &end_rgn
);
778 for (i
= end_rgn
- 1; i
>= start_rgn
; i
--)
779 memblock_remove_region(type
, i
);
783 int __init_memblock
memblock_remove(phys_addr_t base
, phys_addr_t size
)
785 phys_addr_t end
= base
+ size
- 1;
787 memblock_dbg("%s: [%pa-%pa] %pS\n", __func__
,
788 &base
, &end
, (void *)_RET_IP_
);
790 return memblock_remove_range(&memblock
.memory
, base
, size
);
794 * memblock_free - free boot memory block
795 * @base: phys starting address of the boot memory block
796 * @size: size of the boot memory block in bytes
798 * Free boot memory block previously allocated by memblock_alloc_xx() API.
799 * The freeing memory will not be released to the buddy allocator.
801 int __init_memblock
memblock_free(phys_addr_t base
, phys_addr_t size
)
803 phys_addr_t end
= base
+ size
- 1;
805 memblock_dbg("%s: [%pa-%pa] %pS\n", __func__
,
806 &base
, &end
, (void *)_RET_IP_
);
808 kmemleak_free_part_phys(base
, size
);
809 return memblock_remove_range(&memblock
.reserved
, base
, size
);
812 int __init_memblock
memblock_reserve(phys_addr_t base
, phys_addr_t size
)
814 phys_addr_t end
= base
+ size
- 1;
816 memblock_dbg("%s: [%pa-%pa] %pS\n", __func__
,
817 &base
, &end
, (void *)_RET_IP_
);
819 return memblock_add_range(&memblock
.reserved
, base
, size
, MAX_NUMNODES
, 0);
822 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
823 int __init_memblock
memblock_physmem_add(phys_addr_t base
, phys_addr_t size
)
825 phys_addr_t end
= base
+ size
- 1;
827 memblock_dbg("%s: [%pa-%pa] %pS\n", __func__
,
828 &base
, &end
, (void *)_RET_IP_
);
830 return memblock_add_range(&physmem
, base
, size
, MAX_NUMNODES
, 0);
835 * memblock_setclr_flag - set or clear flag for a memory region
836 * @base: base address of the region
837 * @size: size of the region
838 * @set: set or clear the flag
839 * @flag: the flag to update
841 * This function isolates region [@base, @base + @size), and sets/clears flag
843 * Return: 0 on success, -errno on failure.
845 static int __init_memblock
memblock_setclr_flag(phys_addr_t base
,
846 phys_addr_t size
, int set
, int flag
)
848 struct memblock_type
*type
= &memblock
.memory
;
849 int i
, ret
, start_rgn
, end_rgn
;
851 ret
= memblock_isolate_range(type
, base
, size
, &start_rgn
, &end_rgn
);
855 for (i
= start_rgn
; i
< end_rgn
; i
++) {
856 struct memblock_region
*r
= &type
->regions
[i
];
864 memblock_merge_regions(type
);
869 * memblock_mark_hotplug - Mark hotpluggable memory with flag MEMBLOCK_HOTPLUG.
870 * @base: the base phys addr of the region
871 * @size: the size of the region
873 * Return: 0 on success, -errno on failure.
875 int __init_memblock
memblock_mark_hotplug(phys_addr_t base
, phys_addr_t size
)
877 return memblock_setclr_flag(base
, size
, 1, MEMBLOCK_HOTPLUG
);
881 * memblock_clear_hotplug - Clear flag MEMBLOCK_HOTPLUG for a specified region.
882 * @base: the base phys addr of the region
883 * @size: the size of the region
885 * Return: 0 on success, -errno on failure.
887 int __init_memblock
memblock_clear_hotplug(phys_addr_t base
, phys_addr_t size
)
889 return memblock_setclr_flag(base
, size
, 0, MEMBLOCK_HOTPLUG
);
893 * memblock_mark_mirror - Mark mirrored memory with flag MEMBLOCK_MIRROR.
894 * @base: the base phys addr of the region
895 * @size: the size of the region
897 * Return: 0 on success, -errno on failure.
899 int __init_memblock
memblock_mark_mirror(phys_addr_t base
, phys_addr_t size
)
901 system_has_some_mirror
= true;
903 return memblock_setclr_flag(base
, size
, 1, MEMBLOCK_MIRROR
);
907 * memblock_mark_nomap - Mark a memory region with flag MEMBLOCK_NOMAP.
908 * @base: the base phys addr of the region
909 * @size: the size of the region
911 * The memory regions marked with %MEMBLOCK_NOMAP will not be added to the
912 * direct mapping of the physical memory. These regions will still be
913 * covered by the memory map. The struct page representing NOMAP memory
914 * frames in the memory map will be PageReserved()
916 * Return: 0 on success, -errno on failure.
918 int __init_memblock
memblock_mark_nomap(phys_addr_t base
, phys_addr_t size
)
920 return memblock_setclr_flag(base
, size
, 1, MEMBLOCK_NOMAP
);
924 * memblock_clear_nomap - Clear flag MEMBLOCK_NOMAP for a specified region.
925 * @base: the base phys addr of the region
926 * @size: the size of the region
928 * Return: 0 on success, -errno on failure.
930 int __init_memblock
memblock_clear_nomap(phys_addr_t base
, phys_addr_t size
)
932 return memblock_setclr_flag(base
, size
, 0, MEMBLOCK_NOMAP
);
935 static bool should_skip_region(struct memblock_type
*type
,
936 struct memblock_region
*m
,
939 int m_nid
= memblock_get_region_node(m
);
941 /* we never skip regions when iterating memblock.reserved or physmem */
942 if (type
!= memblock_memory
)
945 /* only memory regions are associated with nodes, check it */
946 if (nid
!= NUMA_NO_NODE
&& nid
!= m_nid
)
949 /* skip hotpluggable memory regions if needed */
950 if (movable_node_is_enabled() && memblock_is_hotpluggable(m
))
953 /* if we want mirror memory skip non-mirror memory regions */
954 if ((flags
& MEMBLOCK_MIRROR
) && !memblock_is_mirror(m
))
957 /* skip nomap memory unless we were asked for it explicitly */
958 if (!(flags
& MEMBLOCK_NOMAP
) && memblock_is_nomap(m
))
965 * __next_mem_range - next function for for_each_free_mem_range() etc.
966 * @idx: pointer to u64 loop variable
967 * @nid: node selector, %NUMA_NO_NODE for all nodes
968 * @flags: pick from blocks based on memory attributes
969 * @type_a: pointer to memblock_type from where the range is taken
970 * @type_b: pointer to memblock_type which excludes memory from being taken
971 * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL
972 * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL
973 * @out_nid: ptr to int for nid of the range, can be %NULL
975 * Find the first area from *@idx which matches @nid, fill the out
976 * parameters, and update *@idx for the next iteration. The lower 32bit of
977 * *@idx contains index into type_a and the upper 32bit indexes the
978 * areas before each region in type_b. For example, if type_b regions
979 * look like the following,
981 * 0:[0-16), 1:[32-48), 2:[128-130)
983 * The upper 32bit indexes the following regions.
985 * 0:[0-0), 1:[16-32), 2:[48-128), 3:[130-MAX)
987 * As both region arrays are sorted, the function advances the two indices
988 * in lockstep and returns each intersection.
990 void __next_mem_range(u64
*idx
, int nid
, enum memblock_flags flags
,
991 struct memblock_type
*type_a
,
992 struct memblock_type
*type_b
, phys_addr_t
*out_start
,
993 phys_addr_t
*out_end
, int *out_nid
)
995 int idx_a
= *idx
& 0xffffffff;
996 int idx_b
= *idx
>> 32;
998 if (WARN_ONCE(nid
== MAX_NUMNODES
,
999 "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
1002 for (; idx_a
< type_a
->cnt
; idx_a
++) {
1003 struct memblock_region
*m
= &type_a
->regions
[idx_a
];
1005 phys_addr_t m_start
= m
->base
;
1006 phys_addr_t m_end
= m
->base
+ m
->size
;
1007 int m_nid
= memblock_get_region_node(m
);
1009 if (should_skip_region(type_a
, m
, nid
, flags
))
1014 *out_start
= m_start
;
1020 *idx
= (u32
)idx_a
| (u64
)idx_b
<< 32;
1024 /* scan areas before each reservation */
1025 for (; idx_b
< type_b
->cnt
+ 1; idx_b
++) {
1026 struct memblock_region
*r
;
1027 phys_addr_t r_start
;
1030 r
= &type_b
->regions
[idx_b
];
1031 r_start
= idx_b
? r
[-1].base
+ r
[-1].size
: 0;
1032 r_end
= idx_b
< type_b
->cnt
?
1033 r
->base
: PHYS_ADDR_MAX
;
1036 * if idx_b advanced past idx_a,
1037 * break out to advance idx_a
1039 if (r_start
>= m_end
)
1041 /* if the two regions intersect, we're done */
1042 if (m_start
< r_end
) {
1045 max(m_start
, r_start
);
1047 *out_end
= min(m_end
, r_end
);
1051 * The region which ends first is
1052 * advanced for the next iteration.
1058 *idx
= (u32
)idx_a
| (u64
)idx_b
<< 32;
1064 /* signal end of iteration */
1069 * __next_mem_range_rev - generic next function for for_each_*_range_rev()
1071 * @idx: pointer to u64 loop variable
1072 * @nid: node selector, %NUMA_NO_NODE for all nodes
1073 * @flags: pick from blocks based on memory attributes
1074 * @type_a: pointer to memblock_type from where the range is taken
1075 * @type_b: pointer to memblock_type which excludes memory from being taken
1076 * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL
1077 * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL
1078 * @out_nid: ptr to int for nid of the range, can be %NULL
1080 * Finds the next range from type_a which is not marked as unsuitable
1083 * Reverse of __next_mem_range().
1085 void __init_memblock
__next_mem_range_rev(u64
*idx
, int nid
,
1086 enum memblock_flags flags
,
1087 struct memblock_type
*type_a
,
1088 struct memblock_type
*type_b
,
1089 phys_addr_t
*out_start
,
1090 phys_addr_t
*out_end
, int *out_nid
)
1092 int idx_a
= *idx
& 0xffffffff;
1093 int idx_b
= *idx
>> 32;
1095 if (WARN_ONCE(nid
== MAX_NUMNODES
, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
1098 if (*idx
== (u64
)ULLONG_MAX
) {
1099 idx_a
= type_a
->cnt
- 1;
1101 idx_b
= type_b
->cnt
;
1106 for (; idx_a
>= 0; idx_a
--) {
1107 struct memblock_region
*m
= &type_a
->regions
[idx_a
];
1109 phys_addr_t m_start
= m
->base
;
1110 phys_addr_t m_end
= m
->base
+ m
->size
;
1111 int m_nid
= memblock_get_region_node(m
);
1113 if (should_skip_region(type_a
, m
, nid
, flags
))
1118 *out_start
= m_start
;
1124 *idx
= (u32
)idx_a
| (u64
)idx_b
<< 32;
1128 /* scan areas before each reservation */
1129 for (; idx_b
>= 0; idx_b
--) {
1130 struct memblock_region
*r
;
1131 phys_addr_t r_start
;
1134 r
= &type_b
->regions
[idx_b
];
1135 r_start
= idx_b
? r
[-1].base
+ r
[-1].size
: 0;
1136 r_end
= idx_b
< type_b
->cnt
?
1137 r
->base
: PHYS_ADDR_MAX
;
1139 * if idx_b advanced past idx_a,
1140 * break out to advance idx_a
1143 if (r_end
<= m_start
)
1145 /* if the two regions intersect, we're done */
1146 if (m_end
> r_start
) {
1148 *out_start
= max(m_start
, r_start
);
1150 *out_end
= min(m_end
, r_end
);
1153 if (m_start
>= r_start
)
1157 *idx
= (u32
)idx_a
| (u64
)idx_b
<< 32;
1162 /* signal end of iteration */
1167 * Common iterator interface used to define for_each_mem_pfn_range().
1169 void __init_memblock
__next_mem_pfn_range(int *idx
, int nid
,
1170 unsigned long *out_start_pfn
,
1171 unsigned long *out_end_pfn
, int *out_nid
)
1173 struct memblock_type
*type
= &memblock
.memory
;
1174 struct memblock_region
*r
;
1177 while (++*idx
< type
->cnt
) {
1178 r
= &type
->regions
[*idx
];
1179 r_nid
= memblock_get_region_node(r
);
1181 if (PFN_UP(r
->base
) >= PFN_DOWN(r
->base
+ r
->size
))
1183 if (nid
== MAX_NUMNODES
|| nid
== r_nid
)
1186 if (*idx
>= type
->cnt
) {
1192 *out_start_pfn
= PFN_UP(r
->base
);
1194 *out_end_pfn
= PFN_DOWN(r
->base
+ r
->size
);
1200 * memblock_set_node - set node ID on memblock regions
1201 * @base: base of area to set node ID for
1202 * @size: size of area to set node ID for
1203 * @type: memblock type to set node ID for
1204 * @nid: node ID to set
1206 * Set the nid of memblock @type regions in [@base, @base + @size) to @nid.
1207 * Regions which cross the area boundaries are split as necessary.
1210 * 0 on success, -errno on failure.
1212 int __init_memblock
memblock_set_node(phys_addr_t base
, phys_addr_t size
,
1213 struct memblock_type
*type
, int nid
)
1216 int start_rgn
, end_rgn
;
1219 ret
= memblock_isolate_range(type
, base
, size
, &start_rgn
, &end_rgn
);
1223 for (i
= start_rgn
; i
< end_rgn
; i
++)
1224 memblock_set_region_node(&type
->regions
[i
], nid
);
1226 memblock_merge_regions(type
);
1231 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1233 * __next_mem_pfn_range_in_zone - iterator for for_each_*_range_in_zone()
1235 * @idx: pointer to u64 loop variable
1236 * @zone: zone in which all of the memory blocks reside
1237 * @out_spfn: ptr to ulong for start pfn of the range, can be %NULL
1238 * @out_epfn: ptr to ulong for end pfn of the range, can be %NULL
1240 * This function is meant to be a zone/pfn specific wrapper for the
1241 * for_each_mem_range type iterators. Specifically they are used in the
1242 * deferred memory init routines and as such we were duplicating much of
1243 * this logic throughout the code. So instead of having it in multiple
1244 * locations it seemed like it would make more sense to centralize this to
1245 * one new iterator that does everything they need.
1247 void __init_memblock
1248 __next_mem_pfn_range_in_zone(u64
*idx
, struct zone
*zone
,
1249 unsigned long *out_spfn
, unsigned long *out_epfn
)
1251 int zone_nid
= zone_to_nid(zone
);
1252 phys_addr_t spa
, epa
;
1255 __next_mem_range(idx
, zone_nid
, MEMBLOCK_NONE
,
1256 &memblock
.memory
, &memblock
.reserved
,
1259 while (*idx
!= U64_MAX
) {
1260 unsigned long epfn
= PFN_DOWN(epa
);
1261 unsigned long spfn
= PFN_UP(spa
);
1264 * Verify the end is at least past the start of the zone and
1265 * that we have at least one PFN to initialize.
1267 if (zone
->zone_start_pfn
< epfn
&& spfn
< epfn
) {
1268 /* if we went too far just stop searching */
1269 if (zone_end_pfn(zone
) <= spfn
) {
1275 *out_spfn
= max(zone
->zone_start_pfn
, spfn
);
1277 *out_epfn
= min(zone_end_pfn(zone
), epfn
);
1282 __next_mem_range(idx
, zone_nid
, MEMBLOCK_NONE
,
1283 &memblock
.memory
, &memblock
.reserved
,
1287 /* signal end of iteration */
1289 *out_spfn
= ULONG_MAX
;
1294 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
1297 * memblock_alloc_range_nid - allocate boot memory block
1298 * @size: size of memory block to be allocated in bytes
1299 * @align: alignment of the region and block's size
1300 * @start: the lower bound of the memory region to allocate (phys address)
1301 * @end: the upper bound of the memory region to allocate (phys address)
1302 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1303 * @exact_nid: control the allocation fall back to other nodes
1305 * The allocation is performed from memory region limited by
1306 * memblock.current_limit if @end == %MEMBLOCK_ALLOC_ACCESSIBLE.
1308 * If the specified node can not hold the requested memory and @exact_nid
1309 * is false, the allocation falls back to any node in the system.
1311 * For systems with memory mirroring, the allocation is attempted first
1312 * from the regions with mirroring enabled and then retried from any
1315 * In addition, function sets the min_count to 0 using kmemleak_alloc_phys for
1316 * allocated boot memory block, so that it is never reported as leaks.
1319 * Physical address of allocated memory block on success, %0 on failure.
1321 phys_addr_t __init
memblock_alloc_range_nid(phys_addr_t size
,
1322 phys_addr_t align
, phys_addr_t start
,
1323 phys_addr_t end
, int nid
,
1326 enum memblock_flags flags
= choose_memblock_flags();
1329 if (WARN_ONCE(nid
== MAX_NUMNODES
, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
1333 /* Can't use WARNs this early in boot on powerpc */
1335 align
= SMP_CACHE_BYTES
;
1339 found
= memblock_find_in_range_node(size
, align
, start
, end
, nid
,
1341 if (found
&& !memblock_reserve(found
, size
))
1344 if (nid
!= NUMA_NO_NODE
&& !exact_nid
) {
1345 found
= memblock_find_in_range_node(size
, align
, start
,
1348 if (found
&& !memblock_reserve(found
, size
))
1352 if (flags
& MEMBLOCK_MIRROR
) {
1353 flags
&= ~MEMBLOCK_MIRROR
;
1354 pr_warn("Could not allocate %pap bytes of mirrored memory\n",
1362 /* Skip kmemleak for kasan_init() due to high volume. */
1363 if (end
!= MEMBLOCK_ALLOC_KASAN
)
1365 * The min_count is set to 0 so that memblock allocated
1366 * blocks are never reported as leaks. This is because many
1367 * of these blocks are only referred via the physical
1368 * address which is not looked up by kmemleak.
1370 kmemleak_alloc_phys(found
, size
, 0, 0);
1376 * memblock_phys_alloc_range - allocate a memory block inside specified range
1377 * @size: size of memory block to be allocated in bytes
1378 * @align: alignment of the region and block's size
1379 * @start: the lower bound of the memory region to allocate (physical address)
1380 * @end: the upper bound of the memory region to allocate (physical address)
1382 * Allocate @size bytes in the between @start and @end.
1384 * Return: physical address of the allocated memory block on success,
1387 phys_addr_t __init
memblock_phys_alloc_range(phys_addr_t size
,
1392 memblock_dbg("%s: %llu bytes align=0x%llx from=%pa max_addr=%pa %pS\n",
1393 __func__
, (u64
)size
, (u64
)align
, &start
, &end
,
1395 return memblock_alloc_range_nid(size
, align
, start
, end
, NUMA_NO_NODE
,
1400 * memblock_phys_alloc_try_nid - allocate a memory block from specified NUMA node
1401 * @size: size of memory block to be allocated in bytes
1402 * @align: alignment of the region and block's size
1403 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1405 * Allocates memory block from the specified NUMA node. If the node
1406 * has no available memory, attempts to allocated from any node in the
1409 * Return: physical address of the allocated memory block on success,
1412 phys_addr_t __init
memblock_phys_alloc_try_nid(phys_addr_t size
, phys_addr_t align
, int nid
)
1414 return memblock_alloc_range_nid(size
, align
, 0,
1415 MEMBLOCK_ALLOC_ACCESSIBLE
, nid
, false);
1419 * memblock_alloc_internal - allocate boot memory block
1420 * @size: size of memory block to be allocated in bytes
1421 * @align: alignment of the region and block's size
1422 * @min_addr: the lower bound of the memory region to allocate (phys address)
1423 * @max_addr: the upper bound of the memory region to allocate (phys address)
1424 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1425 * @exact_nid: control the allocation fall back to other nodes
1427 * Allocates memory block using memblock_alloc_range_nid() and
1428 * converts the returned physical address to virtual.
1430 * The @min_addr limit is dropped if it can not be satisfied and the allocation
1431 * will fall back to memory below @min_addr. Other constraints, such
1432 * as node and mirrored memory will be handled again in
1433 * memblock_alloc_range_nid().
1436 * Virtual address of allocated memory block on success, NULL on failure.
1438 static void * __init
memblock_alloc_internal(
1439 phys_addr_t size
, phys_addr_t align
,
1440 phys_addr_t min_addr
, phys_addr_t max_addr
,
1441 int nid
, bool exact_nid
)
1446 * Detect any accidental use of these APIs after slab is ready, as at
1447 * this moment memblock may be deinitialized already and its
1448 * internal data may be destroyed (after execution of memblock_free_all)
1450 if (WARN_ON_ONCE(slab_is_available()))
1451 return kzalloc_node(size
, GFP_NOWAIT
, nid
);
1453 if (max_addr
> memblock
.current_limit
)
1454 max_addr
= memblock
.current_limit
;
1456 alloc
= memblock_alloc_range_nid(size
, align
, min_addr
, max_addr
, nid
,
1459 /* retry allocation without lower limit */
1460 if (!alloc
&& min_addr
)
1461 alloc
= memblock_alloc_range_nid(size
, align
, 0, max_addr
, nid
,
1467 return phys_to_virt(alloc
);
1471 * memblock_alloc_exact_nid_raw - allocate boot memory block on the exact node
1472 * without zeroing memory
1473 * @size: size of memory block to be allocated in bytes
1474 * @align: alignment of the region and block's size
1475 * @min_addr: the lower bound of the memory region from where the allocation
1476 * is preferred (phys address)
1477 * @max_addr: the upper bound of the memory region from where the allocation
1478 * is preferred (phys address), or %MEMBLOCK_ALLOC_ACCESSIBLE to
1479 * allocate only from memory limited by memblock.current_limit value
1480 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1482 * Public function, provides additional debug information (including caller
1483 * info), if enabled. Does not zero allocated memory.
1486 * Virtual address of allocated memory block on success, NULL on failure.
1488 void * __init
memblock_alloc_exact_nid_raw(
1489 phys_addr_t size
, phys_addr_t align
,
1490 phys_addr_t min_addr
, phys_addr_t max_addr
,
1495 memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=%pa max_addr=%pa %pS\n",
1496 __func__
, (u64
)size
, (u64
)align
, nid
, &min_addr
,
1497 &max_addr
, (void *)_RET_IP_
);
1499 ptr
= memblock_alloc_internal(size
, align
,
1500 min_addr
, max_addr
, nid
, true);
1501 if (ptr
&& size
> 0)
1502 page_init_poison(ptr
, size
);
1508 * memblock_alloc_try_nid_raw - allocate boot memory block without zeroing
1509 * memory and without panicking
1510 * @size: size of memory block to be allocated in bytes
1511 * @align: alignment of the region and block's size
1512 * @min_addr: the lower bound of the memory region from where the allocation
1513 * is preferred (phys address)
1514 * @max_addr: the upper bound of the memory region from where the allocation
1515 * is preferred (phys address), or %MEMBLOCK_ALLOC_ACCESSIBLE to
1516 * allocate only from memory limited by memblock.current_limit value
1517 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1519 * Public function, provides additional debug information (including caller
1520 * info), if enabled. Does not zero allocated memory, does not panic if request
1521 * cannot be satisfied.
1524 * Virtual address of allocated memory block on success, NULL on failure.
1526 void * __init
memblock_alloc_try_nid_raw(
1527 phys_addr_t size
, phys_addr_t align
,
1528 phys_addr_t min_addr
, phys_addr_t max_addr
,
1533 memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=%pa max_addr=%pa %pS\n",
1534 __func__
, (u64
)size
, (u64
)align
, nid
, &min_addr
,
1535 &max_addr
, (void *)_RET_IP_
);
1537 ptr
= memblock_alloc_internal(size
, align
,
1538 min_addr
, max_addr
, nid
, false);
1539 if (ptr
&& size
> 0)
1540 page_init_poison(ptr
, size
);
1546 * memblock_alloc_try_nid - allocate boot memory block
1547 * @size: size of memory block to be allocated in bytes
1548 * @align: alignment of the region and block's size
1549 * @min_addr: the lower bound of the memory region from where the allocation
1550 * is preferred (phys address)
1551 * @max_addr: the upper bound of the memory region from where the allocation
1552 * is preferred (phys address), or %MEMBLOCK_ALLOC_ACCESSIBLE to
1553 * allocate only from memory limited by memblock.current_limit value
1554 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1556 * Public function, provides additional debug information (including caller
1557 * info), if enabled. This function zeroes the allocated memory.
1560 * Virtual address of allocated memory block on success, NULL on failure.
1562 void * __init
memblock_alloc_try_nid(
1563 phys_addr_t size
, phys_addr_t align
,
1564 phys_addr_t min_addr
, phys_addr_t max_addr
,
1569 memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=%pa max_addr=%pa %pS\n",
1570 __func__
, (u64
)size
, (u64
)align
, nid
, &min_addr
,
1571 &max_addr
, (void *)_RET_IP_
);
1572 ptr
= memblock_alloc_internal(size
, align
,
1573 min_addr
, max_addr
, nid
, false);
1575 memset(ptr
, 0, size
);
1581 * __memblock_free_late - free pages directly to buddy allocator
1582 * @base: phys starting address of the boot memory block
1583 * @size: size of the boot memory block in bytes
1585 * This is only useful when the memblock allocator has already been torn
1586 * down, but we are still initializing the system. Pages are released directly
1587 * to the buddy allocator.
1589 void __init
__memblock_free_late(phys_addr_t base
, phys_addr_t size
)
1591 phys_addr_t cursor
, end
;
1593 end
= base
+ size
- 1;
1594 memblock_dbg("%s: [%pa-%pa] %pS\n",
1595 __func__
, &base
, &end
, (void *)_RET_IP_
);
1596 kmemleak_free_part_phys(base
, size
);
1597 cursor
= PFN_UP(base
);
1598 end
= PFN_DOWN(base
+ size
);
1600 for (; cursor
< end
; cursor
++) {
1601 memblock_free_pages(pfn_to_page(cursor
), cursor
, 0);
1602 totalram_pages_inc();
1607 * Remaining API functions
1610 phys_addr_t __init_memblock
memblock_phys_mem_size(void)
1612 return memblock
.memory
.total_size
;
1615 phys_addr_t __init_memblock
memblock_reserved_size(void)
1617 return memblock
.reserved
.total_size
;
1620 /* lowest address */
1621 phys_addr_t __init_memblock
memblock_start_of_DRAM(void)
1623 return memblock
.memory
.regions
[0].base
;
1626 phys_addr_t __init_memblock
memblock_end_of_DRAM(void)
1628 int idx
= memblock
.memory
.cnt
- 1;
1630 return (memblock
.memory
.regions
[idx
].base
+ memblock
.memory
.regions
[idx
].size
);
1633 static phys_addr_t __init_memblock
__find_max_addr(phys_addr_t limit
)
1635 phys_addr_t max_addr
= PHYS_ADDR_MAX
;
1636 struct memblock_region
*r
;
1639 * translate the memory @limit size into the max address within one of
1640 * the memory memblock regions, if the @limit exceeds the total size
1641 * of those regions, max_addr will keep original value PHYS_ADDR_MAX
1643 for_each_mem_region(r
) {
1644 if (limit
<= r
->size
) {
1645 max_addr
= r
->base
+ limit
;
1654 void __init
memblock_enforce_memory_limit(phys_addr_t limit
)
1656 phys_addr_t max_addr
;
1661 max_addr
= __find_max_addr(limit
);
1663 /* @limit exceeds the total size of the memory, do nothing */
1664 if (max_addr
== PHYS_ADDR_MAX
)
1667 /* truncate both memory and reserved regions */
1668 memblock_remove_range(&memblock
.memory
, max_addr
,
1670 memblock_remove_range(&memblock
.reserved
, max_addr
,
1674 void __init
memblock_cap_memory_range(phys_addr_t base
, phys_addr_t size
)
1676 int start_rgn
, end_rgn
;
1682 ret
= memblock_isolate_range(&memblock
.memory
, base
, size
,
1683 &start_rgn
, &end_rgn
);
1687 /* remove all the MAP regions */
1688 for (i
= memblock
.memory
.cnt
- 1; i
>= end_rgn
; i
--)
1689 if (!memblock_is_nomap(&memblock
.memory
.regions
[i
]))
1690 memblock_remove_region(&memblock
.memory
, i
);
1692 for (i
= start_rgn
- 1; i
>= 0; i
--)
1693 if (!memblock_is_nomap(&memblock
.memory
.regions
[i
]))
1694 memblock_remove_region(&memblock
.memory
, i
);
1696 /* truncate the reserved regions */
1697 memblock_remove_range(&memblock
.reserved
, 0, base
);
1698 memblock_remove_range(&memblock
.reserved
,
1699 base
+ size
, PHYS_ADDR_MAX
);
1702 void __init
memblock_mem_limit_remove_map(phys_addr_t limit
)
1704 phys_addr_t max_addr
;
1709 max_addr
= __find_max_addr(limit
);
1711 /* @limit exceeds the total size of the memory, do nothing */
1712 if (max_addr
== PHYS_ADDR_MAX
)
1715 memblock_cap_memory_range(0, max_addr
);
1718 static int __init_memblock
memblock_search(struct memblock_type
*type
, phys_addr_t addr
)
1720 unsigned int left
= 0, right
= type
->cnt
;
1723 unsigned int mid
= (right
+ left
) / 2;
1725 if (addr
< type
->regions
[mid
].base
)
1727 else if (addr
>= (type
->regions
[mid
].base
+
1728 type
->regions
[mid
].size
))
1732 } while (left
< right
);
1736 bool __init_memblock
memblock_is_reserved(phys_addr_t addr
)
1738 return memblock_search(&memblock
.reserved
, addr
) != -1;
1741 bool __init_memblock
memblock_is_memory(phys_addr_t addr
)
1743 return memblock_search(&memblock
.memory
, addr
) != -1;
1746 bool __init_memblock
memblock_is_map_memory(phys_addr_t addr
)
1748 int i
= memblock_search(&memblock
.memory
, addr
);
1752 return !memblock_is_nomap(&memblock
.memory
.regions
[i
]);
1755 int __init_memblock
memblock_search_pfn_nid(unsigned long pfn
,
1756 unsigned long *start_pfn
, unsigned long *end_pfn
)
1758 struct memblock_type
*type
= &memblock
.memory
;
1759 int mid
= memblock_search(type
, PFN_PHYS(pfn
));
1764 *start_pfn
= PFN_DOWN(type
->regions
[mid
].base
);
1765 *end_pfn
= PFN_DOWN(type
->regions
[mid
].base
+ type
->regions
[mid
].size
);
1767 return memblock_get_region_node(&type
->regions
[mid
]);
1771 * memblock_is_region_memory - check if a region is a subset of memory
1772 * @base: base of region to check
1773 * @size: size of region to check
1775 * Check if the region [@base, @base + @size) is a subset of a memory block.
1778 * 0 if false, non-zero if true
1780 bool __init_memblock
memblock_is_region_memory(phys_addr_t base
, phys_addr_t size
)
1782 int idx
= memblock_search(&memblock
.memory
, base
);
1783 phys_addr_t end
= base
+ memblock_cap_size(base
, &size
);
1787 return (memblock
.memory
.regions
[idx
].base
+
1788 memblock
.memory
.regions
[idx
].size
) >= end
;
1792 * memblock_is_region_reserved - check if a region intersects reserved memory
1793 * @base: base of region to check
1794 * @size: size of region to check
1796 * Check if the region [@base, @base + @size) intersects a reserved
1800 * True if they intersect, false if not.
1802 bool __init_memblock
memblock_is_region_reserved(phys_addr_t base
, phys_addr_t size
)
1804 return memblock_overlaps_region(&memblock
.reserved
, base
, size
);
1807 void __init_memblock
memblock_trim_memory(phys_addr_t align
)
1809 phys_addr_t start
, end
, orig_start
, orig_end
;
1810 struct memblock_region
*r
;
1812 for_each_mem_region(r
) {
1813 orig_start
= r
->base
;
1814 orig_end
= r
->base
+ r
->size
;
1815 start
= round_up(orig_start
, align
);
1816 end
= round_down(orig_end
, align
);
1818 if (start
== orig_start
&& end
== orig_end
)
1823 r
->size
= end
- start
;
1825 memblock_remove_region(&memblock
.memory
,
1826 r
- memblock
.memory
.regions
);
1832 void __init_memblock
memblock_set_current_limit(phys_addr_t limit
)
1834 memblock
.current_limit
= limit
;
1837 phys_addr_t __init_memblock
memblock_get_current_limit(void)
1839 return memblock
.current_limit
;
1842 static void __init_memblock
memblock_dump(struct memblock_type
*type
)
1844 phys_addr_t base
, end
, size
;
1845 enum memblock_flags flags
;
1847 struct memblock_region
*rgn
;
1849 pr_info(" %s.cnt = 0x%lx\n", type
->name
, type
->cnt
);
1851 for_each_memblock_type(idx
, type
, rgn
) {
1852 char nid_buf
[32] = "";
1856 end
= base
+ size
- 1;
1859 if (memblock_get_region_node(rgn
) != MAX_NUMNODES
)
1860 snprintf(nid_buf
, sizeof(nid_buf
), " on node %d",
1861 memblock_get_region_node(rgn
));
1863 pr_info(" %s[%#x]\t[%pa-%pa], %pa bytes%s flags: %#x\n",
1864 type
->name
, idx
, &base
, &end
, &size
, nid_buf
, flags
);
1868 static void __init_memblock
__memblock_dump_all(void)
1870 pr_info("MEMBLOCK configuration:\n");
1871 pr_info(" memory size = %pa reserved size = %pa\n",
1872 &memblock
.memory
.total_size
,
1873 &memblock
.reserved
.total_size
);
1875 memblock_dump(&memblock
.memory
);
1876 memblock_dump(&memblock
.reserved
);
1877 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
1878 memblock_dump(&physmem
);
1882 void __init_memblock
memblock_dump_all(void)
1885 __memblock_dump_all();
1888 void __init
memblock_allow_resize(void)
1890 memblock_can_resize
= 1;
1893 static int __init
early_memblock(char *p
)
1895 if (p
&& strstr(p
, "debug"))
1899 early_param("memblock", early_memblock
);
1901 static void __init
free_memmap(unsigned long start_pfn
, unsigned long end_pfn
)
1903 struct page
*start_pg
, *end_pg
;
1904 phys_addr_t pg
, pgend
;
1907 * Convert start_pfn/end_pfn to a struct page pointer.
1909 start_pg
= pfn_to_page(start_pfn
- 1) + 1;
1910 end_pg
= pfn_to_page(end_pfn
- 1) + 1;
1913 * Convert to physical addresses, and round start upwards and end
1916 pg
= PAGE_ALIGN(__pa(start_pg
));
1917 pgend
= __pa(end_pg
) & PAGE_MASK
;
1920 * If there are free pages between these, free the section of the
1924 memblock_free(pg
, pgend
- pg
);
1928 * The mem_map array can get very big. Free the unused area of the memory map.
1930 static void __init
free_unused_memmap(void)
1932 unsigned long start
, end
, prev_end
= 0;
1935 if (!IS_ENABLED(CONFIG_HAVE_ARCH_PFN_VALID
) ||
1936 IS_ENABLED(CONFIG_SPARSEMEM_VMEMMAP
))
1940 * This relies on each bank being in address order.
1941 * The banks are sorted previously in bootmem_init().
1943 for_each_mem_pfn_range(i
, MAX_NUMNODES
, &start
, &end
, NULL
) {
1944 #ifdef CONFIG_SPARSEMEM
1946 * Take care not to free memmap entries that don't exist
1947 * due to SPARSEMEM sections which aren't present.
1949 start
= min(start
, ALIGN(prev_end
, PAGES_PER_SECTION
));
1952 * Align down here since many operations in VM subsystem
1953 * presume that there are no holes in the memory map inside
1956 start
= round_down(start
, pageblock_nr_pages
);
1959 * If we had a previous bank, and there is a space
1960 * between the current bank and the previous, free it.
1962 if (prev_end
&& prev_end
< start
)
1963 free_memmap(prev_end
, start
);
1966 * Align up here since many operations in VM subsystem
1967 * presume that there are no holes in the memory map inside
1970 prev_end
= ALIGN(end
, pageblock_nr_pages
);
1973 #ifdef CONFIG_SPARSEMEM
1974 if (!IS_ALIGNED(prev_end
, PAGES_PER_SECTION
)) {
1975 prev_end
= ALIGN(end
, pageblock_nr_pages
);
1976 free_memmap(prev_end
, ALIGN(prev_end
, PAGES_PER_SECTION
));
1981 static void __init
__free_pages_memory(unsigned long start
, unsigned long end
)
1985 while (start
< end
) {
1986 order
= min(MAX_ORDER
- 1UL, __ffs(start
));
1988 while (start
+ (1UL << order
) > end
)
1991 memblock_free_pages(pfn_to_page(start
), start
, order
);
1993 start
+= (1UL << order
);
1997 static unsigned long __init
__free_memory_core(phys_addr_t start
,
2000 unsigned long start_pfn
= PFN_UP(start
);
2001 unsigned long end_pfn
= min_t(unsigned long,
2002 PFN_DOWN(end
), max_low_pfn
);
2004 if (start_pfn
>= end_pfn
)
2007 __free_pages_memory(start_pfn
, end_pfn
);
2009 return end_pfn
- start_pfn
;
2012 static void __init
memmap_init_reserved_pages(void)
2014 struct memblock_region
*region
;
2015 phys_addr_t start
, end
;
2018 /* initialize struct pages for the reserved regions */
2019 for_each_reserved_mem_range(i
, &start
, &end
)
2020 reserve_bootmem_region(start
, end
);
2022 /* and also treat struct pages for the NOMAP regions as PageReserved */
2023 for_each_mem_region(region
) {
2024 if (memblock_is_nomap(region
)) {
2025 start
= region
->base
;
2026 end
= start
+ region
->size
;
2027 reserve_bootmem_region(start
, end
);
2032 static unsigned long __init
free_low_memory_core_early(void)
2034 unsigned long count
= 0;
2035 phys_addr_t start
, end
;
2038 memblock_clear_hotplug(0, -1);
2040 memmap_init_reserved_pages();
2043 * We need to use NUMA_NO_NODE instead of NODE_DATA(0)->node_id
2044 * because in some case like Node0 doesn't have RAM installed
2045 * low ram will be on Node1
2047 for_each_free_mem_range(i
, NUMA_NO_NODE
, MEMBLOCK_NONE
, &start
, &end
,
2049 count
+= __free_memory_core(start
, end
);
2054 static int reset_managed_pages_done __initdata
;
2056 void reset_node_managed_pages(pg_data_t
*pgdat
)
2060 for (z
= pgdat
->node_zones
; z
< pgdat
->node_zones
+ MAX_NR_ZONES
; z
++)
2061 atomic_long_set(&z
->managed_pages
, 0);
2064 void __init
reset_all_zones_managed_pages(void)
2066 struct pglist_data
*pgdat
;
2068 if (reset_managed_pages_done
)
2071 for_each_online_pgdat(pgdat
)
2072 reset_node_managed_pages(pgdat
);
2074 reset_managed_pages_done
= 1;
2078 * memblock_free_all - release free pages to the buddy allocator
2080 void __init
memblock_free_all(void)
2082 unsigned long pages
;
2084 free_unused_memmap();
2085 reset_all_zones_managed_pages();
2087 pages
= free_low_memory_core_early();
2088 totalram_pages_add(pages
);
2091 #if defined(CONFIG_DEBUG_FS) && defined(CONFIG_ARCH_KEEP_MEMBLOCK)
2093 static int memblock_debug_show(struct seq_file
*m
, void *private)
2095 struct memblock_type
*type
= m
->private;
2096 struct memblock_region
*reg
;
2100 for (i
= 0; i
< type
->cnt
; i
++) {
2101 reg
= &type
->regions
[i
];
2102 end
= reg
->base
+ reg
->size
- 1;
2104 seq_printf(m
, "%4d: ", i
);
2105 seq_printf(m
, "%pa..%pa\n", ®
->base
, &end
);
2109 DEFINE_SHOW_ATTRIBUTE(memblock_debug
);
2111 static int __init
memblock_init_debugfs(void)
2113 struct dentry
*root
= debugfs_create_dir("memblock", NULL
);
2115 debugfs_create_file("memory", 0444, root
,
2116 &memblock
.memory
, &memblock_debug_fops
);
2117 debugfs_create_file("reserved", 0444, root
,
2118 &memblock
.reserved
, &memblock_debug_fops
);
2119 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
2120 debugfs_create_file("physmem", 0444, root
, &physmem
,
2121 &memblock_debug_fops
);
2126 __initcall(memblock_init_debugfs
);
2128 #endif /* CONFIG_DEBUG_FS */