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
32 #ifndef INIT_MEMBLOCK_MEMORY_REGIONS
33 #define INIT_MEMBLOCK_MEMORY_REGIONS INIT_MEMBLOCK_REGIONS
37 * DOC: memblock overview
39 * Memblock is a method of managing memory regions during the early
40 * boot period when the usual kernel memory allocators are not up and
43 * Memblock views the system memory as collections of contiguous
44 * regions. There are several types of these collections:
46 * * ``memory`` - describes the physical memory available to the
47 * kernel; this may differ from the actual physical memory installed
48 * in the system, for instance when the memory is restricted with
49 * ``mem=`` command line parameter
50 * * ``reserved`` - describes the regions that were allocated
51 * * ``physmem`` - describes the actual physical memory available during
52 * boot regardless of the possible restrictions and memory hot(un)plug;
53 * the ``physmem`` type is only available on some architectures.
55 * Each region is represented by struct memblock_region that
56 * defines the region extents, its attributes and NUMA node id on NUMA
57 * systems. Every memory type is described by the struct memblock_type
58 * which contains an array of memory regions along with
59 * the allocator metadata. The "memory" and "reserved" types are nicely
60 * wrapped with struct memblock. This structure is statically
61 * initialized at build time. The region arrays are initially sized to
62 * %INIT_MEMBLOCK_MEMORY_REGIONS for "memory" and
63 * %INIT_MEMBLOCK_RESERVED_REGIONS for "reserved". The region array
64 * for "physmem" is initially sized to %INIT_PHYSMEM_REGIONS.
65 * The memblock_allow_resize() enables automatic resizing of the region
66 * arrays during addition of new regions. This feature should be used
67 * with care so that memory allocated for the region array will not
68 * overlap with areas that should be reserved, for example initrd.
70 * The early architecture setup should tell memblock what the physical
71 * memory layout is by using memblock_add() or memblock_add_node()
72 * functions. The first function does not assign the region to a NUMA
73 * node and it is appropriate for UMA systems. Yet, it is possible to
74 * use it on NUMA systems as well and assign the region to a NUMA node
75 * later in the setup process using memblock_set_node(). The
76 * memblock_add_node() performs such an assignment directly.
78 * Once memblock is setup the memory can be allocated using one of the
81 * * memblock_phys_alloc*() - these functions return the **physical**
82 * address of the allocated memory
83 * * memblock_alloc*() - these functions return the **virtual** address
84 * of the allocated memory.
86 * Note, that both API variants use implicit assumptions about allowed
87 * memory ranges and the fallback methods. Consult the documentation
88 * of memblock_alloc_internal() and memblock_alloc_range_nid()
89 * functions for more elaborate description.
91 * As the system boot progresses, the architecture specific mem_init()
92 * function frees all the memory to the buddy page allocator.
94 * Unless an architecture enables %CONFIG_ARCH_KEEP_MEMBLOCK, the
95 * memblock data structures (except "physmem") will be discarded after the
96 * system initialization completes.
100 struct pglist_data __refdata contig_page_data
;
101 EXPORT_SYMBOL(contig_page_data
);
104 unsigned long max_low_pfn
;
105 unsigned long min_low_pfn
;
106 unsigned long max_pfn
;
107 unsigned long long max_possible_pfn
;
109 static struct memblock_region memblock_memory_init_regions
[INIT_MEMBLOCK_MEMORY_REGIONS
] __initdata_memblock
;
110 static struct memblock_region memblock_reserved_init_regions
[INIT_MEMBLOCK_RESERVED_REGIONS
] __initdata_memblock
;
111 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
112 static struct memblock_region memblock_physmem_init_regions
[INIT_PHYSMEM_REGIONS
];
115 struct memblock memblock __initdata_memblock
= {
116 .memory
.regions
= memblock_memory_init_regions
,
117 .memory
.cnt
= 1, /* empty dummy entry */
118 .memory
.max
= INIT_MEMBLOCK_MEMORY_REGIONS
,
119 .memory
.name
= "memory",
121 .reserved
.regions
= memblock_reserved_init_regions
,
122 .reserved
.cnt
= 1, /* empty dummy entry */
123 .reserved
.max
= INIT_MEMBLOCK_RESERVED_REGIONS
,
124 .reserved
.name
= "reserved",
127 .current_limit
= MEMBLOCK_ALLOC_ANYWHERE
,
130 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
131 struct memblock_type physmem
= {
132 .regions
= memblock_physmem_init_regions
,
133 .cnt
= 1, /* empty dummy entry */
134 .max
= INIT_PHYSMEM_REGIONS
,
140 * keep a pointer to &memblock.memory in the text section to use it in
141 * __next_mem_range() and its helpers.
142 * For architectures that do not keep memblock data after init, this
143 * pointer will be reset to NULL at memblock_discard()
145 static __refdata
struct memblock_type
*memblock_memory
= &memblock
.memory
;
147 #define for_each_memblock_type(i, memblock_type, rgn) \
148 for (i = 0, rgn = &memblock_type->regions[0]; \
149 i < memblock_type->cnt; \
150 i++, rgn = &memblock_type->regions[i])
152 #define memblock_dbg(fmt, ...) \
154 if (memblock_debug) \
155 pr_info(fmt, ##__VA_ARGS__); \
158 static int memblock_debug __initdata_memblock
;
159 static bool system_has_some_mirror __initdata_memblock
;
160 static int memblock_can_resize __initdata_memblock
;
161 static int memblock_memory_in_slab __initdata_memblock
;
162 static int memblock_reserved_in_slab __initdata_memblock
;
164 bool __init_memblock
memblock_has_mirror(void)
166 return system_has_some_mirror
;
169 static enum memblock_flags __init_memblock
choose_memblock_flags(void)
171 return system_has_some_mirror
? MEMBLOCK_MIRROR
: MEMBLOCK_NONE
;
174 /* adjust *@size so that (@base + *@size) doesn't overflow, return new size */
175 static inline phys_addr_t
memblock_cap_size(phys_addr_t base
, phys_addr_t
*size
)
177 return *size
= min(*size
, PHYS_ADDR_MAX
- base
);
181 * Address comparison utilities
183 static unsigned long __init_memblock
memblock_addrs_overlap(phys_addr_t base1
, phys_addr_t size1
,
184 phys_addr_t base2
, phys_addr_t size2
)
186 return ((base1
< (base2
+ size2
)) && (base2
< (base1
+ size1
)));
189 bool __init_memblock
memblock_overlaps_region(struct memblock_type
*type
,
190 phys_addr_t base
, phys_addr_t size
)
194 memblock_cap_size(base
, &size
);
196 for (i
= 0; i
< type
->cnt
; i
++)
197 if (memblock_addrs_overlap(base
, size
, type
->regions
[i
].base
,
198 type
->regions
[i
].size
))
200 return i
< type
->cnt
;
204 * __memblock_find_range_bottom_up - find free area utility in bottom-up
205 * @start: start of candidate range
206 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_ANYWHERE or
207 * %MEMBLOCK_ALLOC_ACCESSIBLE
208 * @size: size of free area to find
209 * @align: alignment of free area to find
210 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
211 * @flags: pick from blocks based on memory attributes
213 * Utility called from memblock_find_in_range_node(), find free area bottom-up.
216 * Found address on success, 0 on failure.
218 static phys_addr_t __init_memblock
219 __memblock_find_range_bottom_up(phys_addr_t start
, phys_addr_t end
,
220 phys_addr_t size
, phys_addr_t align
, int nid
,
221 enum memblock_flags flags
)
223 phys_addr_t this_start
, this_end
, cand
;
226 for_each_free_mem_range(i
, nid
, flags
, &this_start
, &this_end
, NULL
) {
227 this_start
= clamp(this_start
, start
, end
);
228 this_end
= clamp(this_end
, start
, end
);
230 cand
= round_up(this_start
, align
);
231 if (cand
< this_end
&& this_end
- cand
>= size
)
239 * __memblock_find_range_top_down - find free area utility, in top-down
240 * @start: start of candidate range
241 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_ANYWHERE or
242 * %MEMBLOCK_ALLOC_ACCESSIBLE
243 * @size: size of free area to find
244 * @align: alignment of free area to find
245 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
246 * @flags: pick from blocks based on memory attributes
248 * Utility called from memblock_find_in_range_node(), find free area top-down.
251 * Found address on success, 0 on failure.
253 static phys_addr_t __init_memblock
254 __memblock_find_range_top_down(phys_addr_t start
, phys_addr_t end
,
255 phys_addr_t size
, phys_addr_t align
, int nid
,
256 enum memblock_flags flags
)
258 phys_addr_t this_start
, this_end
, cand
;
261 for_each_free_mem_range_reverse(i
, nid
, flags
, &this_start
, &this_end
,
263 this_start
= clamp(this_start
, start
, end
);
264 this_end
= clamp(this_end
, start
, end
);
269 cand
= round_down(this_end
- size
, align
);
270 if (cand
>= this_start
)
278 * memblock_find_in_range_node - find free area in given range and node
279 * @size: size of free area to find
280 * @align: alignment of free area to find
281 * @start: start of candidate range
282 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_ANYWHERE or
283 * %MEMBLOCK_ALLOC_ACCESSIBLE
284 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
285 * @flags: pick from blocks based on memory attributes
287 * Find @size free area aligned to @align in the specified range and node.
290 * Found address on success, 0 on failure.
292 static phys_addr_t __init_memblock
memblock_find_in_range_node(phys_addr_t size
,
293 phys_addr_t align
, phys_addr_t start
,
294 phys_addr_t end
, int nid
,
295 enum memblock_flags flags
)
298 if (end
== MEMBLOCK_ALLOC_ACCESSIBLE
||
299 end
== MEMBLOCK_ALLOC_NOLEAKTRACE
)
300 end
= memblock
.current_limit
;
302 /* avoid allocating the first page */
303 start
= max_t(phys_addr_t
, start
, PAGE_SIZE
);
304 end
= max(start
, end
);
306 if (memblock_bottom_up())
307 return __memblock_find_range_bottom_up(start
, end
, size
, align
,
310 return __memblock_find_range_top_down(start
, end
, size
, align
,
315 * memblock_find_in_range - find free area in given range
316 * @start: start of candidate range
317 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_ANYWHERE or
318 * %MEMBLOCK_ALLOC_ACCESSIBLE
319 * @size: size of free area to find
320 * @align: alignment of free area to find
322 * Find @size free area aligned to @align in the specified range.
325 * Found address on success, 0 on failure.
327 static phys_addr_t __init_memblock
memblock_find_in_range(phys_addr_t start
,
328 phys_addr_t end
, phys_addr_t size
,
332 enum memblock_flags flags
= choose_memblock_flags();
335 ret
= memblock_find_in_range_node(size
, align
, start
, end
,
336 NUMA_NO_NODE
, flags
);
338 if (!ret
&& (flags
& MEMBLOCK_MIRROR
)) {
339 pr_warn_ratelimited("Could not allocate %pap bytes of mirrored memory\n",
341 flags
&= ~MEMBLOCK_MIRROR
;
348 static void __init_memblock
memblock_remove_region(struct memblock_type
*type
, unsigned long r
)
350 type
->total_size
-= type
->regions
[r
].size
;
351 memmove(&type
->regions
[r
], &type
->regions
[r
+ 1],
352 (type
->cnt
- (r
+ 1)) * sizeof(type
->regions
[r
]));
355 /* Special case for empty arrays */
356 if (type
->cnt
== 0) {
357 WARN_ON(type
->total_size
!= 0);
359 type
->regions
[0].base
= 0;
360 type
->regions
[0].size
= 0;
361 type
->regions
[0].flags
= 0;
362 memblock_set_region_node(&type
->regions
[0], MAX_NUMNODES
);
366 #ifndef CONFIG_ARCH_KEEP_MEMBLOCK
368 * memblock_discard - discard memory and reserved arrays if they were allocated
370 void __init
memblock_discard(void)
372 phys_addr_t addr
, size
;
374 if (memblock
.reserved
.regions
!= memblock_reserved_init_regions
) {
375 addr
= __pa(memblock
.reserved
.regions
);
376 size
= PAGE_ALIGN(sizeof(struct memblock_region
) *
377 memblock
.reserved
.max
);
378 if (memblock_reserved_in_slab
)
379 kfree(memblock
.reserved
.regions
);
381 memblock_free_late(addr
, size
);
384 if (memblock
.memory
.regions
!= memblock_memory_init_regions
) {
385 addr
= __pa(memblock
.memory
.regions
);
386 size
= PAGE_ALIGN(sizeof(struct memblock_region
) *
387 memblock
.memory
.max
);
388 if (memblock_memory_in_slab
)
389 kfree(memblock
.memory
.regions
);
391 memblock_free_late(addr
, size
);
394 memblock_memory
= NULL
;
399 * memblock_double_array - double the size of the memblock regions array
400 * @type: memblock type of the regions array being doubled
401 * @new_area_start: starting address of memory range to avoid overlap with
402 * @new_area_size: size of memory range to avoid overlap with
404 * Double the size of the @type regions array. If memblock is being used to
405 * allocate memory for a new reserved regions array and there is a previously
406 * allocated memory range [@new_area_start, @new_area_start + @new_area_size]
407 * waiting to be reserved, ensure the memory used by the new array does
411 * 0 on success, -1 on failure.
413 static int __init_memblock
memblock_double_array(struct memblock_type
*type
,
414 phys_addr_t new_area_start
,
415 phys_addr_t new_area_size
)
417 struct memblock_region
*new_array
, *old_array
;
418 phys_addr_t old_alloc_size
, new_alloc_size
;
419 phys_addr_t old_size
, new_size
, addr
, new_end
;
420 int use_slab
= slab_is_available();
423 /* We don't allow resizing until we know about the reserved regions
424 * of memory that aren't suitable for allocation
426 if (!memblock_can_resize
)
427 panic("memblock: cannot resize %s array\n", type
->name
);
429 /* Calculate new doubled size */
430 old_size
= type
->max
* sizeof(struct memblock_region
);
431 new_size
= old_size
<< 1;
433 * We need to allocated new one align to PAGE_SIZE,
434 * so we can free them completely later.
436 old_alloc_size
= PAGE_ALIGN(old_size
);
437 new_alloc_size
= PAGE_ALIGN(new_size
);
439 /* Retrieve the slab flag */
440 if (type
== &memblock
.memory
)
441 in_slab
= &memblock_memory_in_slab
;
443 in_slab
= &memblock_reserved_in_slab
;
445 /* Try to find some space for it */
447 new_array
= kmalloc(new_size
, GFP_KERNEL
);
448 addr
= new_array
? __pa(new_array
) : 0;
450 /* only exclude range when trying to double reserved.regions */
451 if (type
!= &memblock
.reserved
)
452 new_area_start
= new_area_size
= 0;
454 addr
= memblock_find_in_range(new_area_start
+ new_area_size
,
455 memblock
.current_limit
,
456 new_alloc_size
, PAGE_SIZE
);
457 if (!addr
&& new_area_size
)
458 addr
= memblock_find_in_range(0,
459 min(new_area_start
, memblock
.current_limit
),
460 new_alloc_size
, PAGE_SIZE
);
462 new_array
= addr
? __va(addr
) : NULL
;
465 pr_err("memblock: Failed to double %s array from %ld to %ld entries !\n",
466 type
->name
, type
->max
, type
->max
* 2);
470 new_end
= addr
+ new_size
- 1;
471 memblock_dbg("memblock: %s is doubled to %ld at [%pa-%pa]",
472 type
->name
, type
->max
* 2, &addr
, &new_end
);
475 * Found space, we now need to move the array over before we add the
476 * reserved region since it may be our reserved array itself that is
479 memcpy(new_array
, type
->regions
, old_size
);
480 memset(new_array
+ type
->max
, 0, old_size
);
481 old_array
= type
->regions
;
482 type
->regions
= new_array
;
485 /* Free old array. We needn't free it if the array is the static one */
488 else if (old_array
!= memblock_memory_init_regions
&&
489 old_array
!= memblock_reserved_init_regions
)
490 memblock_free(old_array
, old_alloc_size
);
493 * Reserve the new array if that comes from the memblock. Otherwise, we
497 BUG_ON(memblock_reserve(addr
, new_alloc_size
));
499 /* Update slab flag */
506 * memblock_merge_regions - merge neighboring compatible regions
507 * @type: memblock type to scan
508 * @start_rgn: start scanning from (@start_rgn - 1)
509 * @end_rgn: end scanning at (@end_rgn - 1)
510 * Scan @type and merge neighboring compatible regions in [@start_rgn - 1, @end_rgn)
512 static void __init_memblock
memblock_merge_regions(struct memblock_type
*type
,
513 unsigned long start_rgn
,
514 unsigned long end_rgn
)
519 end_rgn
= min(end_rgn
, type
->cnt
- 1);
520 while (i
< end_rgn
) {
521 struct memblock_region
*this = &type
->regions
[i
];
522 struct memblock_region
*next
= &type
->regions
[i
+ 1];
524 if (this->base
+ this->size
!= next
->base
||
525 memblock_get_region_node(this) !=
526 memblock_get_region_node(next
) ||
527 this->flags
!= next
->flags
) {
528 BUG_ON(this->base
+ this->size
> next
->base
);
533 this->size
+= next
->size
;
534 /* move forward from next + 1, index of which is i + 2 */
535 memmove(next
, next
+ 1, (type
->cnt
- (i
+ 2)) * sizeof(*next
));
542 * memblock_insert_region - insert new memblock region
543 * @type: memblock type to insert into
544 * @idx: index for the insertion point
545 * @base: base address of the new region
546 * @size: size of the new region
547 * @nid: node id of the new region
548 * @flags: flags of the new region
550 * Insert new memblock region [@base, @base + @size) into @type at @idx.
551 * @type must already have extra room to accommodate the new region.
553 static void __init_memblock
memblock_insert_region(struct memblock_type
*type
,
554 int idx
, phys_addr_t base
,
557 enum memblock_flags flags
)
559 struct memblock_region
*rgn
= &type
->regions
[idx
];
561 BUG_ON(type
->cnt
>= type
->max
);
562 memmove(rgn
+ 1, rgn
, (type
->cnt
- idx
) * sizeof(*rgn
));
566 memblock_set_region_node(rgn
, nid
);
568 type
->total_size
+= size
;
572 * memblock_add_range - add new memblock region
573 * @type: memblock type to add new region into
574 * @base: base address of the new region
575 * @size: size of the new region
576 * @nid: nid of the new region
577 * @flags: flags of the new region
579 * Add new memblock region [@base, @base + @size) into @type. The new region
580 * is allowed to overlap with existing ones - overlaps don't affect already
581 * existing regions. @type is guaranteed to be minimal (all neighbouring
582 * compatible regions are merged) after the addition.
585 * 0 on success, -errno on failure.
587 static int __init_memblock
memblock_add_range(struct memblock_type
*type
,
588 phys_addr_t base
, phys_addr_t size
,
589 int nid
, enum memblock_flags flags
)
592 phys_addr_t obase
= base
;
593 phys_addr_t end
= base
+ memblock_cap_size(base
, &size
);
594 int idx
, nr_new
, start_rgn
= -1, end_rgn
;
595 struct memblock_region
*rgn
;
600 /* special case for empty array */
601 if (type
->regions
[0].size
== 0) {
602 WARN_ON(type
->cnt
!= 1 || type
->total_size
);
603 type
->regions
[0].base
= base
;
604 type
->regions
[0].size
= size
;
605 type
->regions
[0].flags
= flags
;
606 memblock_set_region_node(&type
->regions
[0], nid
);
607 type
->total_size
= size
;
612 * The worst case is when new range overlaps all existing regions,
613 * then we'll need type->cnt + 1 empty regions in @type. So if
614 * type->cnt * 2 + 1 is less than or equal to type->max, we know
615 * that there is enough empty regions in @type, and we can insert
618 if (type
->cnt
* 2 + 1 <= type
->max
)
623 * The following is executed twice. Once with %false @insert and
624 * then with %true. The first counts the number of regions needed
625 * to accommodate the new area. The second actually inserts them.
630 for_each_memblock_type(idx
, type
, rgn
) {
631 phys_addr_t rbase
= rgn
->base
;
632 phys_addr_t rend
= rbase
+ rgn
->size
;
639 * @rgn overlaps. If it separates the lower part of new
640 * area, insert that portion.
644 WARN_ON(nid
!= memblock_get_region_node(rgn
));
646 WARN_ON(flags
!= rgn
->flags
);
652 memblock_insert_region(type
, idx
++, base
,
657 /* area below @rend is dealt with, forget about it */
658 base
= min(rend
, end
);
661 /* insert the remaining portion */
668 memblock_insert_region(type
, idx
, base
, end
- base
,
677 * If this was the first round, resize array and repeat for actual
678 * insertions; otherwise, merge and return.
681 while (type
->cnt
+ nr_new
> type
->max
)
682 if (memblock_double_array(type
, obase
, size
) < 0)
687 memblock_merge_regions(type
, start_rgn
, end_rgn
);
693 * memblock_add_node - add new memblock region within a NUMA node
694 * @base: base address of the new region
695 * @size: size of the new region
696 * @nid: nid of the new region
697 * @flags: flags of the new region
699 * Add new memblock region [@base, @base + @size) to the "memory"
700 * type. See memblock_add_range() description for mode details
703 * 0 on success, -errno on failure.
705 int __init_memblock
memblock_add_node(phys_addr_t base
, phys_addr_t size
,
706 int nid
, enum memblock_flags flags
)
708 phys_addr_t end
= base
+ size
- 1;
710 memblock_dbg("%s: [%pa-%pa] nid=%d flags=%x %pS\n", __func__
,
711 &base
, &end
, nid
, flags
, (void *)_RET_IP_
);
713 return memblock_add_range(&memblock
.memory
, base
, size
, nid
, flags
);
717 * memblock_add - add new memblock region
718 * @base: base address of the new region
719 * @size: size of the new region
721 * Add new memblock region [@base, @base + @size) to the "memory"
722 * type. See memblock_add_range() description for mode details
725 * 0 on success, -errno on failure.
727 int __init_memblock
memblock_add(phys_addr_t base
, phys_addr_t size
)
729 phys_addr_t end
= base
+ size
- 1;
731 memblock_dbg("%s: [%pa-%pa] %pS\n", __func__
,
732 &base
, &end
, (void *)_RET_IP_
);
734 return memblock_add_range(&memblock
.memory
, base
, size
, MAX_NUMNODES
, 0);
738 * memblock_isolate_range - isolate given range into disjoint memblocks
739 * @type: memblock type to isolate range for
740 * @base: base of range to isolate
741 * @size: size of range to isolate
742 * @start_rgn: out parameter for the start of isolated region
743 * @end_rgn: out parameter for the end of isolated region
745 * Walk @type and ensure that regions don't cross the boundaries defined by
746 * [@base, @base + @size). Crossing regions are split at the boundaries,
747 * which may create at most two more regions. The index of the first
748 * region inside the range is returned in *@start_rgn and end in *@end_rgn.
751 * 0 on success, -errno on failure.
753 static int __init_memblock
memblock_isolate_range(struct memblock_type
*type
,
754 phys_addr_t base
, phys_addr_t size
,
755 int *start_rgn
, int *end_rgn
)
757 phys_addr_t end
= base
+ memblock_cap_size(base
, &size
);
759 struct memblock_region
*rgn
;
761 *start_rgn
= *end_rgn
= 0;
766 /* we'll create at most two more regions */
767 while (type
->cnt
+ 2 > type
->max
)
768 if (memblock_double_array(type
, base
, size
) < 0)
771 for_each_memblock_type(idx
, type
, rgn
) {
772 phys_addr_t rbase
= rgn
->base
;
773 phys_addr_t rend
= rbase
+ rgn
->size
;
782 * @rgn intersects from below. Split and continue
783 * to process the next region - the new top half.
786 rgn
->size
-= base
- rbase
;
787 type
->total_size
-= base
- rbase
;
788 memblock_insert_region(type
, idx
, rbase
, base
- rbase
,
789 memblock_get_region_node(rgn
),
791 } else if (rend
> end
) {
793 * @rgn intersects from above. Split and redo the
794 * current region - the new bottom half.
797 rgn
->size
-= end
- rbase
;
798 type
->total_size
-= end
- rbase
;
799 memblock_insert_region(type
, idx
--, rbase
, end
- rbase
,
800 memblock_get_region_node(rgn
),
803 /* @rgn is fully contained, record it */
813 static int __init_memblock
memblock_remove_range(struct memblock_type
*type
,
814 phys_addr_t base
, phys_addr_t size
)
816 int start_rgn
, end_rgn
;
819 ret
= memblock_isolate_range(type
, base
, size
, &start_rgn
, &end_rgn
);
823 for (i
= end_rgn
- 1; i
>= start_rgn
; i
--)
824 memblock_remove_region(type
, i
);
828 int __init_memblock
memblock_remove(phys_addr_t base
, phys_addr_t size
)
830 phys_addr_t end
= base
+ size
- 1;
832 memblock_dbg("%s: [%pa-%pa] %pS\n", __func__
,
833 &base
, &end
, (void *)_RET_IP_
);
835 return memblock_remove_range(&memblock
.memory
, base
, size
);
839 * memblock_free - free boot memory allocation
840 * @ptr: starting address of the boot memory allocation
841 * @size: size of the boot memory block in bytes
843 * Free boot memory block previously allocated by memblock_alloc_xx() API.
844 * The freeing memory will not be released to the buddy allocator.
846 void __init_memblock
memblock_free(void *ptr
, size_t size
)
849 memblock_phys_free(__pa(ptr
), size
);
853 * memblock_phys_free - free boot memory block
854 * @base: phys starting address of the boot memory block
855 * @size: size of the boot memory block in bytes
857 * Free boot memory block previously allocated by memblock_phys_alloc_xx() API.
858 * The freeing memory will not be released to the buddy allocator.
860 int __init_memblock
memblock_phys_free(phys_addr_t base
, phys_addr_t size
)
862 phys_addr_t end
= base
+ size
- 1;
864 memblock_dbg("%s: [%pa-%pa] %pS\n", __func__
,
865 &base
, &end
, (void *)_RET_IP_
);
867 kmemleak_free_part_phys(base
, size
);
868 return memblock_remove_range(&memblock
.reserved
, base
, size
);
871 int __init_memblock
memblock_reserve(phys_addr_t base
, phys_addr_t size
)
873 phys_addr_t end
= base
+ size
- 1;
875 memblock_dbg("%s: [%pa-%pa] %pS\n", __func__
,
876 &base
, &end
, (void *)_RET_IP_
);
878 return memblock_add_range(&memblock
.reserved
, base
, size
, MAX_NUMNODES
, 0);
881 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
882 int __init_memblock
memblock_physmem_add(phys_addr_t base
, phys_addr_t size
)
884 phys_addr_t end
= base
+ size
- 1;
886 memblock_dbg("%s: [%pa-%pa] %pS\n", __func__
,
887 &base
, &end
, (void *)_RET_IP_
);
889 return memblock_add_range(&physmem
, base
, size
, MAX_NUMNODES
, 0);
894 * memblock_setclr_flag - set or clear flag for a memory region
895 * @type: memblock type to set/clear flag for
896 * @base: base address of the region
897 * @size: size of the region
898 * @set: set or clear the flag
899 * @flag: the flag to update
901 * This function isolates region [@base, @base + @size), and sets/clears flag
903 * Return: 0 on success, -errno on failure.
905 static int __init_memblock
memblock_setclr_flag(struct memblock_type
*type
,
906 phys_addr_t base
, phys_addr_t size
, int set
, int flag
)
908 int i
, ret
, start_rgn
, end_rgn
;
910 ret
= memblock_isolate_range(type
, base
, size
, &start_rgn
, &end_rgn
);
914 for (i
= start_rgn
; i
< end_rgn
; i
++) {
915 struct memblock_region
*r
= &type
->regions
[i
];
923 memblock_merge_regions(type
, start_rgn
, end_rgn
);
928 * memblock_mark_hotplug - Mark hotpluggable memory with flag MEMBLOCK_HOTPLUG.
929 * @base: the base phys addr of the region
930 * @size: the size of the region
932 * Return: 0 on success, -errno on failure.
934 int __init_memblock
memblock_mark_hotplug(phys_addr_t base
, phys_addr_t size
)
936 return memblock_setclr_flag(&memblock
.memory
, base
, size
, 1, MEMBLOCK_HOTPLUG
);
940 * memblock_clear_hotplug - Clear flag MEMBLOCK_HOTPLUG for a specified region.
941 * @base: the base phys addr of the region
942 * @size: the size of the region
944 * Return: 0 on success, -errno on failure.
946 int __init_memblock
memblock_clear_hotplug(phys_addr_t base
, phys_addr_t size
)
948 return memblock_setclr_flag(&memblock
.memory
, base
, size
, 0, MEMBLOCK_HOTPLUG
);
952 * memblock_mark_mirror - Mark mirrored memory with flag MEMBLOCK_MIRROR.
953 * @base: the base phys addr of the region
954 * @size: the size of the region
956 * Return: 0 on success, -errno on failure.
958 int __init_memblock
memblock_mark_mirror(phys_addr_t base
, phys_addr_t size
)
960 if (!mirrored_kernelcore
)
963 system_has_some_mirror
= true;
965 return memblock_setclr_flag(&memblock
.memory
, base
, size
, 1, MEMBLOCK_MIRROR
);
969 * memblock_mark_nomap - Mark a memory region with flag MEMBLOCK_NOMAP.
970 * @base: the base phys addr of the region
971 * @size: the size of the region
973 * The memory regions marked with %MEMBLOCK_NOMAP will not be added to the
974 * direct mapping of the physical memory. These regions will still be
975 * covered by the memory map. The struct page representing NOMAP memory
976 * frames in the memory map will be PageReserved()
978 * Note: if the memory being marked %MEMBLOCK_NOMAP was allocated from
979 * memblock, the caller must inform kmemleak to ignore that memory
981 * Return: 0 on success, -errno on failure.
983 int __init_memblock
memblock_mark_nomap(phys_addr_t base
, phys_addr_t size
)
985 return memblock_setclr_flag(&memblock
.memory
, base
, size
, 1, MEMBLOCK_NOMAP
);
989 * memblock_clear_nomap - Clear flag MEMBLOCK_NOMAP for a specified region.
990 * @base: the base phys addr of the region
991 * @size: the size of the region
993 * Return: 0 on success, -errno on failure.
995 int __init_memblock
memblock_clear_nomap(phys_addr_t base
, phys_addr_t size
)
997 return memblock_setclr_flag(&memblock
.memory
, base
, size
, 0, MEMBLOCK_NOMAP
);
1001 * memblock_reserved_mark_noinit - Mark a reserved memory region with flag
1002 * MEMBLOCK_RSRV_NOINIT which results in the struct pages not being initialized
1004 * @base: the base phys addr of the region
1005 * @size: the size of the region
1007 * struct pages will not be initialized for reserved memory regions marked with
1008 * %MEMBLOCK_RSRV_NOINIT.
1010 * Return: 0 on success, -errno on failure.
1012 int __init_memblock
memblock_reserved_mark_noinit(phys_addr_t base
, phys_addr_t size
)
1014 return memblock_setclr_flag(&memblock
.reserved
, base
, size
, 1,
1015 MEMBLOCK_RSRV_NOINIT
);
1018 static bool should_skip_region(struct memblock_type
*type
,
1019 struct memblock_region
*m
,
1022 int m_nid
= memblock_get_region_node(m
);
1024 /* we never skip regions when iterating memblock.reserved or physmem */
1025 if (type
!= memblock_memory
)
1028 /* only memory regions are associated with nodes, check it */
1029 if (nid
!= NUMA_NO_NODE
&& nid
!= m_nid
)
1032 /* skip hotpluggable memory regions if needed */
1033 if (movable_node_is_enabled() && memblock_is_hotpluggable(m
) &&
1034 !(flags
& MEMBLOCK_HOTPLUG
))
1037 /* if we want mirror memory skip non-mirror memory regions */
1038 if ((flags
& MEMBLOCK_MIRROR
) && !memblock_is_mirror(m
))
1041 /* skip nomap memory unless we were asked for it explicitly */
1042 if (!(flags
& MEMBLOCK_NOMAP
) && memblock_is_nomap(m
))
1045 /* skip driver-managed memory unless we were asked for it explicitly */
1046 if (!(flags
& MEMBLOCK_DRIVER_MANAGED
) && memblock_is_driver_managed(m
))
1053 * __next_mem_range - next function for for_each_free_mem_range() etc.
1054 * @idx: pointer to u64 loop variable
1055 * @nid: node selector, %NUMA_NO_NODE for all nodes
1056 * @flags: pick from blocks based on memory attributes
1057 * @type_a: pointer to memblock_type from where the range is taken
1058 * @type_b: pointer to memblock_type which excludes memory from being taken
1059 * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL
1060 * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL
1061 * @out_nid: ptr to int for nid of the range, can be %NULL
1063 * Find the first area from *@idx which matches @nid, fill the out
1064 * parameters, and update *@idx for the next iteration. The lower 32bit of
1065 * *@idx contains index into type_a and the upper 32bit indexes the
1066 * areas before each region in type_b. For example, if type_b regions
1067 * look like the following,
1069 * 0:[0-16), 1:[32-48), 2:[128-130)
1071 * The upper 32bit indexes the following regions.
1073 * 0:[0-0), 1:[16-32), 2:[48-128), 3:[130-MAX)
1075 * As both region arrays are sorted, the function advances the two indices
1076 * in lockstep and returns each intersection.
1078 void __next_mem_range(u64
*idx
, int nid
, enum memblock_flags flags
,
1079 struct memblock_type
*type_a
,
1080 struct memblock_type
*type_b
, phys_addr_t
*out_start
,
1081 phys_addr_t
*out_end
, int *out_nid
)
1083 int idx_a
= *idx
& 0xffffffff;
1084 int idx_b
= *idx
>> 32;
1086 if (WARN_ONCE(nid
== MAX_NUMNODES
,
1087 "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
1090 for (; idx_a
< type_a
->cnt
; idx_a
++) {
1091 struct memblock_region
*m
= &type_a
->regions
[idx_a
];
1093 phys_addr_t m_start
= m
->base
;
1094 phys_addr_t m_end
= m
->base
+ m
->size
;
1095 int m_nid
= memblock_get_region_node(m
);
1097 if (should_skip_region(type_a
, m
, nid
, flags
))
1102 *out_start
= m_start
;
1108 *idx
= (u32
)idx_a
| (u64
)idx_b
<< 32;
1112 /* scan areas before each reservation */
1113 for (; idx_b
< type_b
->cnt
+ 1; idx_b
++) {
1114 struct memblock_region
*r
;
1115 phys_addr_t r_start
;
1118 r
= &type_b
->regions
[idx_b
];
1119 r_start
= idx_b
? r
[-1].base
+ r
[-1].size
: 0;
1120 r_end
= idx_b
< type_b
->cnt
?
1121 r
->base
: PHYS_ADDR_MAX
;
1124 * if idx_b advanced past idx_a,
1125 * break out to advance idx_a
1127 if (r_start
>= m_end
)
1129 /* if the two regions intersect, we're done */
1130 if (m_start
< r_end
) {
1133 max(m_start
, r_start
);
1135 *out_end
= min(m_end
, r_end
);
1139 * The region which ends first is
1140 * advanced for the next iteration.
1146 *idx
= (u32
)idx_a
| (u64
)idx_b
<< 32;
1152 /* signal end of iteration */
1157 * __next_mem_range_rev - generic next function for for_each_*_range_rev()
1159 * @idx: pointer to u64 loop variable
1160 * @nid: node selector, %NUMA_NO_NODE for all nodes
1161 * @flags: pick from blocks based on memory attributes
1162 * @type_a: pointer to memblock_type from where the range is taken
1163 * @type_b: pointer to memblock_type which excludes memory from being taken
1164 * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL
1165 * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL
1166 * @out_nid: ptr to int for nid of the range, can be %NULL
1168 * Finds the next range from type_a which is not marked as unsuitable
1171 * Reverse of __next_mem_range().
1173 void __init_memblock
__next_mem_range_rev(u64
*idx
, int nid
,
1174 enum memblock_flags flags
,
1175 struct memblock_type
*type_a
,
1176 struct memblock_type
*type_b
,
1177 phys_addr_t
*out_start
,
1178 phys_addr_t
*out_end
, int *out_nid
)
1180 int idx_a
= *idx
& 0xffffffff;
1181 int idx_b
= *idx
>> 32;
1183 if (WARN_ONCE(nid
== MAX_NUMNODES
, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
1186 if (*idx
== (u64
)ULLONG_MAX
) {
1187 idx_a
= type_a
->cnt
- 1;
1189 idx_b
= type_b
->cnt
;
1194 for (; idx_a
>= 0; idx_a
--) {
1195 struct memblock_region
*m
= &type_a
->regions
[idx_a
];
1197 phys_addr_t m_start
= m
->base
;
1198 phys_addr_t m_end
= m
->base
+ m
->size
;
1199 int m_nid
= memblock_get_region_node(m
);
1201 if (should_skip_region(type_a
, m
, nid
, flags
))
1206 *out_start
= m_start
;
1212 *idx
= (u32
)idx_a
| (u64
)idx_b
<< 32;
1216 /* scan areas before each reservation */
1217 for (; idx_b
>= 0; idx_b
--) {
1218 struct memblock_region
*r
;
1219 phys_addr_t r_start
;
1222 r
= &type_b
->regions
[idx_b
];
1223 r_start
= idx_b
? r
[-1].base
+ r
[-1].size
: 0;
1224 r_end
= idx_b
< type_b
->cnt
?
1225 r
->base
: PHYS_ADDR_MAX
;
1227 * if idx_b advanced past idx_a,
1228 * break out to advance idx_a
1231 if (r_end
<= m_start
)
1233 /* if the two regions intersect, we're done */
1234 if (m_end
> r_start
) {
1236 *out_start
= max(m_start
, r_start
);
1238 *out_end
= min(m_end
, r_end
);
1241 if (m_start
>= r_start
)
1245 *idx
= (u32
)idx_a
| (u64
)idx_b
<< 32;
1250 /* signal end of iteration */
1255 * Common iterator interface used to define for_each_mem_pfn_range().
1257 void __init_memblock
__next_mem_pfn_range(int *idx
, int nid
,
1258 unsigned long *out_start_pfn
,
1259 unsigned long *out_end_pfn
, int *out_nid
)
1261 struct memblock_type
*type
= &memblock
.memory
;
1262 struct memblock_region
*r
;
1265 while (++*idx
< type
->cnt
) {
1266 r
= &type
->regions
[*idx
];
1267 r_nid
= memblock_get_region_node(r
);
1269 if (PFN_UP(r
->base
) >= PFN_DOWN(r
->base
+ r
->size
))
1271 if (nid
== MAX_NUMNODES
|| nid
== r_nid
)
1274 if (*idx
>= type
->cnt
) {
1280 *out_start_pfn
= PFN_UP(r
->base
);
1282 *out_end_pfn
= PFN_DOWN(r
->base
+ r
->size
);
1288 * memblock_set_node - set node ID on memblock regions
1289 * @base: base of area to set node ID for
1290 * @size: size of area to set node ID for
1291 * @type: memblock type to set node ID for
1292 * @nid: node ID to set
1294 * Set the nid of memblock @type regions in [@base, @base + @size) to @nid.
1295 * Regions which cross the area boundaries are split as necessary.
1298 * 0 on success, -errno on failure.
1300 int __init_memblock
memblock_set_node(phys_addr_t base
, phys_addr_t size
,
1301 struct memblock_type
*type
, int nid
)
1304 int start_rgn
, end_rgn
;
1307 ret
= memblock_isolate_range(type
, base
, size
, &start_rgn
, &end_rgn
);
1311 for (i
= start_rgn
; i
< end_rgn
; i
++)
1312 memblock_set_region_node(&type
->regions
[i
], nid
);
1314 memblock_merge_regions(type
, start_rgn
, end_rgn
);
1319 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1321 * __next_mem_pfn_range_in_zone - iterator for for_each_*_range_in_zone()
1323 * @idx: pointer to u64 loop variable
1324 * @zone: zone in which all of the memory blocks reside
1325 * @out_spfn: ptr to ulong for start pfn of the range, can be %NULL
1326 * @out_epfn: ptr to ulong for end pfn of the range, can be %NULL
1328 * This function is meant to be a zone/pfn specific wrapper for the
1329 * for_each_mem_range type iterators. Specifically they are used in the
1330 * deferred memory init routines and as such we were duplicating much of
1331 * this logic throughout the code. So instead of having it in multiple
1332 * locations it seemed like it would make more sense to centralize this to
1333 * one new iterator that does everything they need.
1335 void __init_memblock
1336 __next_mem_pfn_range_in_zone(u64
*idx
, struct zone
*zone
,
1337 unsigned long *out_spfn
, unsigned long *out_epfn
)
1339 int zone_nid
= zone_to_nid(zone
);
1340 phys_addr_t spa
, epa
;
1342 __next_mem_range(idx
, zone_nid
, MEMBLOCK_NONE
,
1343 &memblock
.memory
, &memblock
.reserved
,
1346 while (*idx
!= U64_MAX
) {
1347 unsigned long epfn
= PFN_DOWN(epa
);
1348 unsigned long spfn
= PFN_UP(spa
);
1351 * Verify the end is at least past the start of the zone and
1352 * that we have at least one PFN to initialize.
1354 if (zone
->zone_start_pfn
< epfn
&& spfn
< epfn
) {
1355 /* if we went too far just stop searching */
1356 if (zone_end_pfn(zone
) <= spfn
) {
1362 *out_spfn
= max(zone
->zone_start_pfn
, spfn
);
1364 *out_epfn
= min(zone_end_pfn(zone
), epfn
);
1369 __next_mem_range(idx
, zone_nid
, MEMBLOCK_NONE
,
1370 &memblock
.memory
, &memblock
.reserved
,
1374 /* signal end of iteration */
1376 *out_spfn
= ULONG_MAX
;
1381 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
1384 * memblock_alloc_range_nid - allocate boot memory block
1385 * @size: size of memory block to be allocated in bytes
1386 * @align: alignment of the region and block's size
1387 * @start: the lower bound of the memory region to allocate (phys address)
1388 * @end: the upper bound of the memory region to allocate (phys address)
1389 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1390 * @exact_nid: control the allocation fall back to other nodes
1392 * The allocation is performed from memory region limited by
1393 * memblock.current_limit if @end == %MEMBLOCK_ALLOC_ACCESSIBLE.
1395 * If the specified node can not hold the requested memory and @exact_nid
1396 * is false, the allocation falls back to any node in the system.
1398 * For systems with memory mirroring, the allocation is attempted first
1399 * from the regions with mirroring enabled and then retried from any
1402 * In addition, function using kmemleak_alloc_phys for allocated boot
1403 * memory block, it is never reported as leaks.
1406 * Physical address of allocated memory block on success, %0 on failure.
1408 phys_addr_t __init
memblock_alloc_range_nid(phys_addr_t size
,
1409 phys_addr_t align
, phys_addr_t start
,
1410 phys_addr_t end
, int nid
,
1413 enum memblock_flags flags
= choose_memblock_flags();
1416 if (WARN_ONCE(nid
== MAX_NUMNODES
, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
1420 /* Can't use WARNs this early in boot on powerpc */
1422 align
= SMP_CACHE_BYTES
;
1426 found
= memblock_find_in_range_node(size
, align
, start
, end
, nid
,
1428 if (found
&& !memblock_reserve(found
, size
))
1431 if (nid
!= NUMA_NO_NODE
&& !exact_nid
) {
1432 found
= memblock_find_in_range_node(size
, align
, start
,
1435 if (found
&& !memblock_reserve(found
, size
))
1439 if (flags
& MEMBLOCK_MIRROR
) {
1440 flags
&= ~MEMBLOCK_MIRROR
;
1441 pr_warn_ratelimited("Could not allocate %pap bytes of mirrored memory\n",
1450 * Skip kmemleak for those places like kasan_init() and
1451 * early_pgtable_alloc() due to high volume.
1453 if (end
!= MEMBLOCK_ALLOC_NOLEAKTRACE
)
1455 * Memblock allocated blocks are never reported as
1456 * leaks. This is because many of these blocks are
1457 * only referred via the physical address which is
1458 * not looked up by kmemleak.
1460 kmemleak_alloc_phys(found
, size
, 0);
1463 * Some Virtual Machine platforms, such as Intel TDX or AMD SEV-SNP,
1464 * require memory to be accepted before it can be used by the
1467 * Accept the memory of the allocated buffer.
1469 accept_memory(found
, found
+ size
);
1475 * memblock_phys_alloc_range - allocate a memory block inside specified range
1476 * @size: size of memory block to be allocated in bytes
1477 * @align: alignment of the region and block's size
1478 * @start: the lower bound of the memory region to allocate (physical address)
1479 * @end: the upper bound of the memory region to allocate (physical address)
1481 * Allocate @size bytes in the between @start and @end.
1483 * Return: physical address of the allocated memory block on success,
1486 phys_addr_t __init
memblock_phys_alloc_range(phys_addr_t size
,
1491 memblock_dbg("%s: %llu bytes align=0x%llx from=%pa max_addr=%pa %pS\n",
1492 __func__
, (u64
)size
, (u64
)align
, &start
, &end
,
1494 return memblock_alloc_range_nid(size
, align
, start
, end
, NUMA_NO_NODE
,
1499 * memblock_phys_alloc_try_nid - allocate a memory block from specified NUMA node
1500 * @size: size of memory block to be allocated in bytes
1501 * @align: alignment of the region and block's size
1502 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1504 * Allocates memory block from the specified NUMA node. If the node
1505 * has no available memory, attempts to allocated from any node in the
1508 * Return: physical address of the allocated memory block on success,
1511 phys_addr_t __init
memblock_phys_alloc_try_nid(phys_addr_t size
, phys_addr_t align
, int nid
)
1513 return memblock_alloc_range_nid(size
, align
, 0,
1514 MEMBLOCK_ALLOC_ACCESSIBLE
, nid
, false);
1518 * memblock_alloc_internal - allocate boot memory block
1519 * @size: size of memory block to be allocated in bytes
1520 * @align: alignment of the region and block's size
1521 * @min_addr: the lower bound of the memory region to allocate (phys address)
1522 * @max_addr: the upper bound of the memory region to allocate (phys address)
1523 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1524 * @exact_nid: control the allocation fall back to other nodes
1526 * Allocates memory block using memblock_alloc_range_nid() and
1527 * converts the returned physical address to virtual.
1529 * The @min_addr limit is dropped if it can not be satisfied and the allocation
1530 * will fall back to memory below @min_addr. Other constraints, such
1531 * as node and mirrored memory will be handled again in
1532 * memblock_alloc_range_nid().
1535 * Virtual address of allocated memory block on success, NULL on failure.
1537 static void * __init
memblock_alloc_internal(
1538 phys_addr_t size
, phys_addr_t align
,
1539 phys_addr_t min_addr
, phys_addr_t max_addr
,
1540 int nid
, bool exact_nid
)
1545 * Detect any accidental use of these APIs after slab is ready, as at
1546 * this moment memblock may be deinitialized already and its
1547 * internal data may be destroyed (after execution of memblock_free_all)
1549 if (WARN_ON_ONCE(slab_is_available()))
1550 return kzalloc_node(size
, GFP_NOWAIT
, nid
);
1552 if (max_addr
> memblock
.current_limit
)
1553 max_addr
= memblock
.current_limit
;
1555 alloc
= memblock_alloc_range_nid(size
, align
, min_addr
, max_addr
, nid
,
1558 /* retry allocation without lower limit */
1559 if (!alloc
&& min_addr
)
1560 alloc
= memblock_alloc_range_nid(size
, align
, 0, max_addr
, nid
,
1566 return phys_to_virt(alloc
);
1570 * memblock_alloc_exact_nid_raw - allocate boot memory block on the exact node
1571 * without zeroing memory
1572 * @size: size of memory block to be allocated in bytes
1573 * @align: alignment of the region and block's size
1574 * @min_addr: the lower bound of the memory region from where the allocation
1575 * is preferred (phys address)
1576 * @max_addr: the upper bound of the memory region from where the allocation
1577 * is preferred (phys address), or %MEMBLOCK_ALLOC_ACCESSIBLE to
1578 * allocate only from memory limited by memblock.current_limit value
1579 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1581 * Public function, provides additional debug information (including caller
1582 * info), if enabled. Does not zero allocated memory.
1585 * Virtual address of allocated memory block on success, NULL on failure.
1587 void * __init
memblock_alloc_exact_nid_raw(
1588 phys_addr_t size
, phys_addr_t align
,
1589 phys_addr_t min_addr
, phys_addr_t max_addr
,
1592 memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=%pa max_addr=%pa %pS\n",
1593 __func__
, (u64
)size
, (u64
)align
, nid
, &min_addr
,
1594 &max_addr
, (void *)_RET_IP_
);
1596 return memblock_alloc_internal(size
, align
, min_addr
, max_addr
, nid
,
1601 * memblock_alloc_try_nid_raw - allocate boot memory block without zeroing
1602 * memory and without panicking
1603 * @size: size of memory block to be allocated in bytes
1604 * @align: alignment of the region and block's size
1605 * @min_addr: the lower bound of the memory region from where the allocation
1606 * is preferred (phys address)
1607 * @max_addr: the upper bound of the memory region from where the allocation
1608 * is preferred (phys address), or %MEMBLOCK_ALLOC_ACCESSIBLE to
1609 * allocate only from memory limited by memblock.current_limit value
1610 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1612 * Public function, provides additional debug information (including caller
1613 * info), if enabled. Does not zero allocated memory, does not panic if request
1614 * cannot be satisfied.
1617 * Virtual address of allocated memory block on success, NULL on failure.
1619 void * __init
memblock_alloc_try_nid_raw(
1620 phys_addr_t size
, phys_addr_t align
,
1621 phys_addr_t min_addr
, phys_addr_t max_addr
,
1624 memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=%pa max_addr=%pa %pS\n",
1625 __func__
, (u64
)size
, (u64
)align
, nid
, &min_addr
,
1626 &max_addr
, (void *)_RET_IP_
);
1628 return memblock_alloc_internal(size
, align
, min_addr
, max_addr
, nid
,
1633 * memblock_alloc_try_nid - allocate boot memory block
1634 * @size: size of memory block to be allocated in bytes
1635 * @align: alignment of the region and block's size
1636 * @min_addr: the lower bound of the memory region from where the allocation
1637 * is preferred (phys address)
1638 * @max_addr: the upper bound of the memory region from where the allocation
1639 * is preferred (phys address), or %MEMBLOCK_ALLOC_ACCESSIBLE to
1640 * allocate only from memory limited by memblock.current_limit value
1641 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1643 * Public function, provides additional debug information (including caller
1644 * info), if enabled. This function zeroes the allocated memory.
1647 * Virtual address of allocated memory block on success, NULL on failure.
1649 void * __init
memblock_alloc_try_nid(
1650 phys_addr_t size
, phys_addr_t align
,
1651 phys_addr_t min_addr
, phys_addr_t max_addr
,
1656 memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=%pa max_addr=%pa %pS\n",
1657 __func__
, (u64
)size
, (u64
)align
, nid
, &min_addr
,
1658 &max_addr
, (void *)_RET_IP_
);
1659 ptr
= memblock_alloc_internal(size
, align
,
1660 min_addr
, max_addr
, nid
, false);
1662 memset(ptr
, 0, size
);
1668 * memblock_free_late - free pages directly to buddy allocator
1669 * @base: phys starting address of the boot memory block
1670 * @size: size of the boot memory block in bytes
1672 * This is only useful when the memblock allocator has already been torn
1673 * down, but we are still initializing the system. Pages are released directly
1674 * to the buddy allocator.
1676 void __init
memblock_free_late(phys_addr_t base
, phys_addr_t size
)
1678 phys_addr_t cursor
, end
;
1680 end
= base
+ size
- 1;
1681 memblock_dbg("%s: [%pa-%pa] %pS\n",
1682 __func__
, &base
, &end
, (void *)_RET_IP_
);
1683 kmemleak_free_part_phys(base
, size
);
1684 cursor
= PFN_UP(base
);
1685 end
= PFN_DOWN(base
+ size
);
1687 for (; cursor
< end
; cursor
++) {
1688 memblock_free_pages(pfn_to_page(cursor
), cursor
, 0);
1689 totalram_pages_inc();
1694 * Remaining API functions
1697 phys_addr_t __init_memblock
memblock_phys_mem_size(void)
1699 return memblock
.memory
.total_size
;
1702 phys_addr_t __init_memblock
memblock_reserved_size(void)
1704 return memblock
.reserved
.total_size
;
1707 /* lowest address */
1708 phys_addr_t __init_memblock
memblock_start_of_DRAM(void)
1710 return memblock
.memory
.regions
[0].base
;
1713 phys_addr_t __init_memblock
memblock_end_of_DRAM(void)
1715 int idx
= memblock
.memory
.cnt
- 1;
1717 return (memblock
.memory
.regions
[idx
].base
+ memblock
.memory
.regions
[idx
].size
);
1720 static phys_addr_t __init_memblock
__find_max_addr(phys_addr_t limit
)
1722 phys_addr_t max_addr
= PHYS_ADDR_MAX
;
1723 struct memblock_region
*r
;
1726 * translate the memory @limit size into the max address within one of
1727 * the memory memblock regions, if the @limit exceeds the total size
1728 * of those regions, max_addr will keep original value PHYS_ADDR_MAX
1730 for_each_mem_region(r
) {
1731 if (limit
<= r
->size
) {
1732 max_addr
= r
->base
+ limit
;
1741 void __init
memblock_enforce_memory_limit(phys_addr_t limit
)
1743 phys_addr_t max_addr
;
1748 max_addr
= __find_max_addr(limit
);
1750 /* @limit exceeds the total size of the memory, do nothing */
1751 if (max_addr
== PHYS_ADDR_MAX
)
1754 /* truncate both memory and reserved regions */
1755 memblock_remove_range(&memblock
.memory
, max_addr
,
1757 memblock_remove_range(&memblock
.reserved
, max_addr
,
1761 void __init
memblock_cap_memory_range(phys_addr_t base
, phys_addr_t size
)
1763 int start_rgn
, end_rgn
;
1769 if (!memblock_memory
->total_size
) {
1770 pr_warn("%s: No memory registered yet\n", __func__
);
1774 ret
= memblock_isolate_range(&memblock
.memory
, base
, size
,
1775 &start_rgn
, &end_rgn
);
1779 /* remove all the MAP regions */
1780 for (i
= memblock
.memory
.cnt
- 1; i
>= end_rgn
; i
--)
1781 if (!memblock_is_nomap(&memblock
.memory
.regions
[i
]))
1782 memblock_remove_region(&memblock
.memory
, i
);
1784 for (i
= start_rgn
- 1; i
>= 0; i
--)
1785 if (!memblock_is_nomap(&memblock
.memory
.regions
[i
]))
1786 memblock_remove_region(&memblock
.memory
, i
);
1788 /* truncate the reserved regions */
1789 memblock_remove_range(&memblock
.reserved
, 0, base
);
1790 memblock_remove_range(&memblock
.reserved
,
1791 base
+ size
, PHYS_ADDR_MAX
);
1794 void __init
memblock_mem_limit_remove_map(phys_addr_t limit
)
1796 phys_addr_t max_addr
;
1801 max_addr
= __find_max_addr(limit
);
1803 /* @limit exceeds the total size of the memory, do nothing */
1804 if (max_addr
== PHYS_ADDR_MAX
)
1807 memblock_cap_memory_range(0, max_addr
);
1810 static int __init_memblock
memblock_search(struct memblock_type
*type
, phys_addr_t addr
)
1812 unsigned int left
= 0, right
= type
->cnt
;
1815 unsigned int mid
= (right
+ left
) / 2;
1817 if (addr
< type
->regions
[mid
].base
)
1819 else if (addr
>= (type
->regions
[mid
].base
+
1820 type
->regions
[mid
].size
))
1824 } while (left
< right
);
1828 bool __init_memblock
memblock_is_reserved(phys_addr_t addr
)
1830 return memblock_search(&memblock
.reserved
, addr
) != -1;
1833 bool __init_memblock
memblock_is_memory(phys_addr_t addr
)
1835 return memblock_search(&memblock
.memory
, addr
) != -1;
1838 bool __init_memblock
memblock_is_map_memory(phys_addr_t addr
)
1840 int i
= memblock_search(&memblock
.memory
, addr
);
1844 return !memblock_is_nomap(&memblock
.memory
.regions
[i
]);
1847 int __init_memblock
memblock_search_pfn_nid(unsigned long pfn
,
1848 unsigned long *start_pfn
, unsigned long *end_pfn
)
1850 struct memblock_type
*type
= &memblock
.memory
;
1851 int mid
= memblock_search(type
, PFN_PHYS(pfn
));
1856 *start_pfn
= PFN_DOWN(type
->regions
[mid
].base
);
1857 *end_pfn
= PFN_DOWN(type
->regions
[mid
].base
+ type
->regions
[mid
].size
);
1859 return memblock_get_region_node(&type
->regions
[mid
]);
1863 * memblock_is_region_memory - check if a region is a subset of memory
1864 * @base: base of region to check
1865 * @size: size of region to check
1867 * Check if the region [@base, @base + @size) is a subset of a memory block.
1870 * 0 if false, non-zero if true
1872 bool __init_memblock
memblock_is_region_memory(phys_addr_t base
, phys_addr_t size
)
1874 int idx
= memblock_search(&memblock
.memory
, base
);
1875 phys_addr_t end
= base
+ memblock_cap_size(base
, &size
);
1879 return (memblock
.memory
.regions
[idx
].base
+
1880 memblock
.memory
.regions
[idx
].size
) >= end
;
1884 * memblock_is_region_reserved - check if a region intersects reserved memory
1885 * @base: base of region to check
1886 * @size: size of region to check
1888 * Check if the region [@base, @base + @size) intersects a reserved
1892 * True if they intersect, false if not.
1894 bool __init_memblock
memblock_is_region_reserved(phys_addr_t base
, phys_addr_t size
)
1896 return memblock_overlaps_region(&memblock
.reserved
, base
, size
);
1899 void __init_memblock
memblock_trim_memory(phys_addr_t align
)
1901 phys_addr_t start
, end
, orig_start
, orig_end
;
1902 struct memblock_region
*r
;
1904 for_each_mem_region(r
) {
1905 orig_start
= r
->base
;
1906 orig_end
= r
->base
+ r
->size
;
1907 start
= round_up(orig_start
, align
);
1908 end
= round_down(orig_end
, align
);
1910 if (start
== orig_start
&& end
== orig_end
)
1915 r
->size
= end
- start
;
1917 memblock_remove_region(&memblock
.memory
,
1918 r
- memblock
.memory
.regions
);
1924 void __init_memblock
memblock_set_current_limit(phys_addr_t limit
)
1926 memblock
.current_limit
= limit
;
1929 phys_addr_t __init_memblock
memblock_get_current_limit(void)
1931 return memblock
.current_limit
;
1934 static void __init_memblock
memblock_dump(struct memblock_type
*type
)
1936 phys_addr_t base
, end
, size
;
1937 enum memblock_flags flags
;
1939 struct memblock_region
*rgn
;
1941 pr_info(" %s.cnt = 0x%lx\n", type
->name
, type
->cnt
);
1943 for_each_memblock_type(idx
, type
, rgn
) {
1944 char nid_buf
[32] = "";
1948 end
= base
+ size
- 1;
1951 if (memblock_get_region_node(rgn
) != MAX_NUMNODES
)
1952 snprintf(nid_buf
, sizeof(nid_buf
), " on node %d",
1953 memblock_get_region_node(rgn
));
1955 pr_info(" %s[%#x]\t[%pa-%pa], %pa bytes%s flags: %#x\n",
1956 type
->name
, idx
, &base
, &end
, &size
, nid_buf
, flags
);
1960 static void __init_memblock
__memblock_dump_all(void)
1962 pr_info("MEMBLOCK configuration:\n");
1963 pr_info(" memory size = %pa reserved size = %pa\n",
1964 &memblock
.memory
.total_size
,
1965 &memblock
.reserved
.total_size
);
1967 memblock_dump(&memblock
.memory
);
1968 memblock_dump(&memblock
.reserved
);
1969 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
1970 memblock_dump(&physmem
);
1974 void __init_memblock
memblock_dump_all(void)
1977 __memblock_dump_all();
1980 void __init
memblock_allow_resize(void)
1982 memblock_can_resize
= 1;
1985 static int __init
early_memblock(char *p
)
1987 if (p
&& strstr(p
, "debug"))
1991 early_param("memblock", early_memblock
);
1993 static void __init
free_memmap(unsigned long start_pfn
, unsigned long end_pfn
)
1995 struct page
*start_pg
, *end_pg
;
1996 phys_addr_t pg
, pgend
;
1999 * Convert start_pfn/end_pfn to a struct page pointer.
2001 start_pg
= pfn_to_page(start_pfn
- 1) + 1;
2002 end_pg
= pfn_to_page(end_pfn
- 1) + 1;
2005 * Convert to physical addresses, and round start upwards and end
2008 pg
= PAGE_ALIGN(__pa(start_pg
));
2009 pgend
= __pa(end_pg
) & PAGE_MASK
;
2012 * If there are free pages between these, free the section of the
2016 memblock_phys_free(pg
, pgend
- pg
);
2020 * The mem_map array can get very big. Free the unused area of the memory map.
2022 static void __init
free_unused_memmap(void)
2024 unsigned long start
, end
, prev_end
= 0;
2027 if (!IS_ENABLED(CONFIG_HAVE_ARCH_PFN_VALID
) ||
2028 IS_ENABLED(CONFIG_SPARSEMEM_VMEMMAP
))
2032 * This relies on each bank being in address order.
2033 * The banks are sorted previously in bootmem_init().
2035 for_each_mem_pfn_range(i
, MAX_NUMNODES
, &start
, &end
, NULL
) {
2036 #ifdef CONFIG_SPARSEMEM
2038 * Take care not to free memmap entries that don't exist
2039 * due to SPARSEMEM sections which aren't present.
2041 start
= min(start
, ALIGN(prev_end
, PAGES_PER_SECTION
));
2044 * Align down here since many operations in VM subsystem
2045 * presume that there are no holes in the memory map inside
2048 start
= pageblock_start_pfn(start
);
2051 * If we had a previous bank, and there is a space
2052 * between the current bank and the previous, free it.
2054 if (prev_end
&& prev_end
< start
)
2055 free_memmap(prev_end
, start
);
2058 * Align up here since many operations in VM subsystem
2059 * presume that there are no holes in the memory map inside
2062 prev_end
= pageblock_align(end
);
2065 #ifdef CONFIG_SPARSEMEM
2066 if (!IS_ALIGNED(prev_end
, PAGES_PER_SECTION
)) {
2067 prev_end
= pageblock_align(end
);
2068 free_memmap(prev_end
, ALIGN(prev_end
, PAGES_PER_SECTION
));
2073 static void __init
__free_pages_memory(unsigned long start
, unsigned long end
)
2077 while (start
< end
) {
2079 * Free the pages in the largest chunks alignment allows.
2081 * __ffs() behaviour is undefined for 0. start == 0 is
2082 * MAX_ORDER-aligned, set order to MAX_ORDER for the case.
2085 order
= min_t(int, MAX_ORDER
, __ffs(start
));
2089 while (start
+ (1UL << order
) > end
)
2092 memblock_free_pages(pfn_to_page(start
), start
, order
);
2094 start
+= (1UL << order
);
2098 static unsigned long __init
__free_memory_core(phys_addr_t start
,
2101 unsigned long start_pfn
= PFN_UP(start
);
2102 unsigned long end_pfn
= min_t(unsigned long,
2103 PFN_DOWN(end
), max_low_pfn
);
2105 if (start_pfn
>= end_pfn
)
2108 __free_pages_memory(start_pfn
, end_pfn
);
2110 return end_pfn
- start_pfn
;
2113 static void __init
memmap_init_reserved_pages(void)
2115 struct memblock_region
*region
;
2116 phys_addr_t start
, end
;
2120 * set nid on all reserved pages and also treat struct
2121 * pages for the NOMAP regions as PageReserved
2123 for_each_mem_region(region
) {
2124 nid
= memblock_get_region_node(region
);
2125 start
= region
->base
;
2126 end
= start
+ region
->size
;
2128 if (memblock_is_nomap(region
))
2129 reserve_bootmem_region(start
, end
, nid
);
2131 memblock_set_node(start
, end
, &memblock
.reserved
, nid
);
2135 * initialize struct pages for reserved regions that don't have
2136 * the MEMBLOCK_RSRV_NOINIT flag set
2138 for_each_reserved_mem_region(region
) {
2139 if (!memblock_is_reserved_noinit(region
)) {
2140 nid
= memblock_get_region_node(region
);
2141 start
= region
->base
;
2142 end
= start
+ region
->size
;
2144 reserve_bootmem_region(start
, end
, nid
);
2149 static unsigned long __init
free_low_memory_core_early(void)
2151 unsigned long count
= 0;
2152 phys_addr_t start
, end
;
2155 memblock_clear_hotplug(0, -1);
2157 memmap_init_reserved_pages();
2160 * We need to use NUMA_NO_NODE instead of NODE_DATA(0)->node_id
2161 * because in some case like Node0 doesn't have RAM installed
2162 * low ram will be on Node1
2164 for_each_free_mem_range(i
, NUMA_NO_NODE
, MEMBLOCK_NONE
, &start
, &end
,
2166 count
+= __free_memory_core(start
, end
);
2171 static int reset_managed_pages_done __initdata
;
2173 static void __init
reset_node_managed_pages(pg_data_t
*pgdat
)
2177 for (z
= pgdat
->node_zones
; z
< pgdat
->node_zones
+ MAX_NR_ZONES
; z
++)
2178 atomic_long_set(&z
->managed_pages
, 0);
2181 void __init
reset_all_zones_managed_pages(void)
2183 struct pglist_data
*pgdat
;
2185 if (reset_managed_pages_done
)
2188 for_each_online_pgdat(pgdat
)
2189 reset_node_managed_pages(pgdat
);
2191 reset_managed_pages_done
= 1;
2195 * memblock_free_all - release free pages to the buddy allocator
2197 void __init
memblock_free_all(void)
2199 unsigned long pages
;
2201 free_unused_memmap();
2202 reset_all_zones_managed_pages();
2204 pages
= free_low_memory_core_early();
2205 totalram_pages_add(pages
);
2208 #if defined(CONFIG_DEBUG_FS) && defined(CONFIG_ARCH_KEEP_MEMBLOCK)
2209 static const char * const flagname
[] = {
2210 [ilog2(MEMBLOCK_HOTPLUG
)] = "HOTPLUG",
2211 [ilog2(MEMBLOCK_MIRROR
)] = "MIRROR",
2212 [ilog2(MEMBLOCK_NOMAP
)] = "NOMAP",
2213 [ilog2(MEMBLOCK_DRIVER_MANAGED
)] = "DRV_MNG",
2216 static int memblock_debug_show(struct seq_file
*m
, void *private)
2218 struct memblock_type
*type
= m
->private;
2219 struct memblock_region
*reg
;
2221 unsigned int count
= ARRAY_SIZE(flagname
);
2224 for (i
= 0; i
< type
->cnt
; i
++) {
2225 reg
= &type
->regions
[i
];
2226 end
= reg
->base
+ reg
->size
- 1;
2227 nid
= memblock_get_region_node(reg
);
2229 seq_printf(m
, "%4d: ", i
);
2230 seq_printf(m
, "%pa..%pa ", ®
->base
, &end
);
2231 if (nid
!= MAX_NUMNODES
)
2232 seq_printf(m
, "%4d ", nid
);
2234 seq_printf(m
, "%4c ", 'x');
2236 for (j
= 0; j
< count
; j
++) {
2237 if (reg
->flags
& (1U << j
)) {
2238 seq_printf(m
, "%s\n", flagname
[j
]);
2243 seq_printf(m
, "%s\n", "UNKNOWN");
2245 seq_printf(m
, "%s\n", "NONE");
2250 DEFINE_SHOW_ATTRIBUTE(memblock_debug
);
2252 static int __init
memblock_init_debugfs(void)
2254 struct dentry
*root
= debugfs_create_dir("memblock", NULL
);
2256 debugfs_create_file("memory", 0444, root
,
2257 &memblock
.memory
, &memblock_debug_fops
);
2258 debugfs_create_file("reserved", 0444, root
,
2259 &memblock
.reserved
, &memblock_debug_fops
);
2260 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
2261 debugfs_create_file("physmem", 0444, root
, &physmem
,
2262 &memblock_debug_fops
);
2267 __initcall(memblock_init_debugfs
);
2269 #endif /* CONFIG_DEBUG_FS */