blob: 2b7397781c99adc44927b26eaddf9da3ce5cb93f [file] [log] [blame]
Thomas Gleixner2874c5f2019-05-27 08:55:01 +02001// SPDX-License-Identifier: GPL-2.0-or-later
Yinghai Lu95f72d12010-07-12 14:36:09 +10002/*
3 * Procedures for maintaining information about logical memory blocks.
4 *
5 * Peter Bergner, IBM Corp. June 2001.
6 * Copyright (C) 2001 Peter Bergner.
Yinghai Lu95f72d12010-07-12 14:36:09 +10007 */
8
9#include <linux/kernel.h>
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -070010#include <linux/slab.h>
Yinghai Lu95f72d12010-07-12 14:36:09 +100011#include <linux/init.h>
12#include <linux/bitops.h>
Benjamin Herrenschmidt449e8df2010-07-06 15:39:07 -070013#include <linux/poison.h>
Benjamin Herrenschmidtc196f762010-07-06 15:39:16 -070014#include <linux/pfn.h>
Benjamin Herrenschmidt6d03b882010-07-06 15:39:19 -070015#include <linux/debugfs.h>
Randy Dunlap514c6032018-04-05 16:25:34 -070016#include <linux/kmemleak.h>
Benjamin Herrenschmidt6d03b882010-07-06 15:39:19 -070017#include <linux/seq_file.h>
Yinghai Lu95f72d12010-07-12 14:36:09 +100018#include <linux/memblock.h>
19
Christoph Hellwigc4c5ad62016-07-28 15:48:06 -070020#include <asm/sections.h>
Santosh Shilimkar26f09e92014-01-21 15:50:19 -080021#include <linux/io.h>
22
23#include "internal.h"
Tang Chen79442ed2013-11-12 15:07:59 -080024
Ard Biesheuvel8a5b403d2019-02-15 13:33:32 +010025#define INIT_MEMBLOCK_REGIONS 128
26#define INIT_PHYSMEM_REGIONS 4
27
28#ifndef INIT_MEMBLOCK_RESERVED_REGIONS
29# define INIT_MEMBLOCK_RESERVED_REGIONS INIT_MEMBLOCK_REGIONS
30#endif
31
Mike Rapoport3e039c52018-06-30 17:55:05 +030032/**
33 * DOC: memblock overview
34 *
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
37 * running.
38 *
39 * Memblock views the system memory as collections of contiguous
40 * regions. There are several types of these collections:
41 *
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
David Hildenbrand77649902020-07-01 16:18:29 +020047 * * ``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.
Mike Rapoport3e039c52018-06-30 17:55:05 +030050 *
Mauro Carvalho Chehab9303c9d2020-09-25 12:01:25 +020051 * Each region is represented by struct memblock_region that
Mike Rapoport3e039c52018-06-30 17:55:05 +030052 * defines the region extents, its attributes and NUMA node id on NUMA
Mauro Carvalho Chehab1bf162e2020-09-28 15:50:33 +020053 * systems. Every memory type is described by the struct memblock_type
54 * which contains an array of memory regions along with
David Hildenbrand77649902020-07-01 16:18:29 +020055 * the allocator metadata. The "memory" and "reserved" types are nicely
Mauro Carvalho Chehab9303c9d2020-09-25 12:01:25 +020056 * wrapped with struct memblock. This structure is statically
David Hildenbrand77649902020-07-01 16:18:29 +020057 * 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.
Cao jin6e5af9a2019-11-30 17:56:21 -080061 * 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.
Mike Rapoport3e039c52018-06-30 17:55:05 +030065 *
66 * The early architecture setup should tell memblock what the physical
Cao jin6e5af9a2019-11-30 17:56:21 -080067 * 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.
Mike Rapoport3e039c52018-06-30 17:55:05 +030073 *
Mike Rapoporta2974132019-03-11 23:30:54 -070074 * Once memblock is setup the memory can be allocated using one of the
75 * API variants:
76 *
Cao jin6e5af9a2019-11-30 17:56:21 -080077 * * 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.
Mike Rapoporta2974132019-03-11 23:30:54 -070081 *
Ethon Pauldf1758d2020-06-04 16:49:16 -070082 * Note, that both API variants use implicit assumptions about allowed
Mike Rapoporta2974132019-03-11 23:30:54 -070083 * memory ranges and the fallback methods. Consult the documentation
Cao jin6e5af9a2019-11-30 17:56:21 -080084 * of memblock_alloc_internal() and memblock_alloc_range_nid()
85 * functions for more elaborate description.
Mike Rapoport3e039c52018-06-30 17:55:05 +030086 *
Cao jin6e5af9a2019-11-30 17:56:21 -080087 * As the system boot progresses, the architecture specific mem_init()
88 * function frees all the memory to the buddy page allocator.
Mike Rapoport3e039c52018-06-30 17:55:05 +030089 *
Cao jin6e5af9a2019-11-30 17:56:21 -080090 * Unless an architecture enables %CONFIG_ARCH_KEEP_MEMBLOCK, the
David Hildenbrand77649902020-07-01 16:18:29 +020091 * memblock data structures (except "physmem") will be discarded after the
92 * system initialization completes.
Mike Rapoport3e039c52018-06-30 17:55:05 +030093 */
94
Mike Rapoporta9ee6cf2021-06-28 19:43:01 -070095#ifndef CONFIG_NUMA
Mike Rapoportbda49a82018-10-30 15:09:40 -070096struct pglist_data __refdata contig_page_data;
97EXPORT_SYMBOL(contig_page_data);
98#endif
99
100unsigned long max_low_pfn;
101unsigned long min_low_pfn;
102unsigned long max_pfn;
103unsigned long long max_possible_pfn;
104
Tejun Heofe091c22011-12-08 10:22:07 -0800105static struct memblock_region memblock_memory_init_regions[INIT_MEMBLOCK_REGIONS] __initdata_memblock;
Ard Biesheuvel8a5b403d2019-02-15 13:33:32 +0100106static struct memblock_region memblock_reserved_init_regions[INIT_MEMBLOCK_RESERVED_REGIONS] __initdata_memblock;
Philipp Hachtmann70210ed2014-01-29 18:16:01 +0100107#ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
David Hildenbrand77649902020-07-01 16:18:29 +0200108static struct memblock_region memblock_physmem_init_regions[INIT_PHYSMEM_REGIONS];
Philipp Hachtmann70210ed2014-01-29 18:16:01 +0100109#endif
Tejun Heofe091c22011-12-08 10:22:07 -0800110
111struct memblock memblock __initdata_memblock = {
112 .memory.regions = memblock_memory_init_regions,
113 .memory.cnt = 1, /* empty dummy entry */
114 .memory.max = INIT_MEMBLOCK_REGIONS,
Heiko Carstens0262d9c2017-02-24 14:55:59 -0800115 .memory.name = "memory",
Tejun Heofe091c22011-12-08 10:22:07 -0800116
117 .reserved.regions = memblock_reserved_init_regions,
118 .reserved.cnt = 1, /* empty dummy entry */
Ard Biesheuvel8a5b403d2019-02-15 13:33:32 +0100119 .reserved.max = INIT_MEMBLOCK_RESERVED_REGIONS,
Heiko Carstens0262d9c2017-02-24 14:55:59 -0800120 .reserved.name = "reserved",
Tejun Heofe091c22011-12-08 10:22:07 -0800121
Tang Chen79442ed2013-11-12 15:07:59 -0800122 .bottom_up = false,
Tejun Heofe091c22011-12-08 10:22:07 -0800123 .current_limit = MEMBLOCK_ALLOC_ANYWHERE,
124};
Yinghai Lu95f72d12010-07-12 14:36:09 +1000125
David Hildenbrand77649902020-07-01 16:18:29 +0200126#ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
127struct memblock_type physmem = {
128 .regions = memblock_physmem_init_regions,
129 .cnt = 1, /* empty dummy entry */
130 .max = INIT_PHYSMEM_REGIONS,
131 .name = "physmem",
132};
133#endif
134
Mike Rapoport9f3d5ea2020-10-13 16:58:25 -0700135/*
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()
140 */
141static __refdata struct memblock_type *memblock_memory = &memblock.memory;
142
Mike Rapoportcd991db2020-10-13 16:57:49 -0700143#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])
147
Mike Rapoport87c55872020-10-13 16:57:54 -0700148#define memblock_dbg(fmt, ...) \
149 do { \
150 if (memblock_debug) \
151 pr_info(fmt, ##__VA_ARGS__); \
152 } while (0)
153
154static int memblock_debug __initdata_memblock;
Tony Lucka3f5baf2015-06-24 16:58:12 -0700155static bool system_has_some_mirror __initdata_memblock = false;
Tejun Heo1aadc052011-12-08 10:22:08 -0800156static int memblock_can_resize __initdata_memblock;
Gavin Shan181eb392012-05-29 15:06:50 -0700157static int memblock_memory_in_slab __initdata_memblock = 0;
158static int memblock_reserved_in_slab __initdata_memblock = 0;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000159
Mike Rapoportc366ea82019-03-11 23:29:46 -0700160static enum memblock_flags __init_memblock choose_memblock_flags(void)
Tony Lucka3f5baf2015-06-24 16:58:12 -0700161{
162 return system_has_some_mirror ? MEMBLOCK_MIRROR : MEMBLOCK_NONE;
163}
164
Tejun Heoeb18f1b2011-12-08 10:22:07 -0800165/* adjust *@size so that (@base + *@size) doesn't overflow, return new size */
166static inline phys_addr_t memblock_cap_size(phys_addr_t base, phys_addr_t *size)
167{
Stefan Agner1c4bc432018-06-07 17:06:15 -0700168 return *size = min(*size, PHYS_ADDR_MAX - base);
Tejun Heoeb18f1b2011-12-08 10:22:07 -0800169}
170
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000171/*
172 * Address comparison utilities
173 */
Yinghai Lu10d06432010-07-28 15:43:02 +1000174static unsigned long __init_memblock memblock_addrs_overlap(phys_addr_t base1, phys_addr_t size1,
Benjamin Herrenschmidt2898cc42010-08-04 13:34:42 +1000175 phys_addr_t base2, phys_addr_t size2)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000176{
177 return ((base1 < (base2 + size2)) && (base2 < (base1 + size1)));
178}
179
Tang Chen95cf82e2015-09-08 15:02:03 -0700180bool __init_memblock memblock_overlaps_region(struct memblock_type *type,
H Hartley Sweeten2d7d3eb2011-10-31 17:09:15 -0700181 phys_addr_t base, phys_addr_t size)
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000182{
183 unsigned long i;
184
Mike Rapoport023accf2021-06-30 09:12:13 +0300185 memblock_cap_size(base, &size);
186
Alexander Kuleshovf14516f2016-01-14 15:20:39 -0800187 for (i = 0; i < type->cnt; i++)
188 if (memblock_addrs_overlap(base, size, type->regions[i].base,
189 type->regions[i].size))
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000190 break;
Tang Chenc5c5c9d2015-09-08 15:02:00 -0700191 return i < type->cnt;
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000192}
193
Mike Rapoport47cec442018-06-30 17:55:02 +0300194/**
Tang Chen79442ed2013-11-12 15:07:59 -0800195 * __memblock_find_range_bottom_up - find free area utility in bottom-up
196 * @start: start of candidate range
Mike Rapoport47cec442018-06-30 17:55:02 +0300197 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_ANYWHERE or
198 * %MEMBLOCK_ALLOC_ACCESSIBLE
Tang Chen79442ed2013-11-12 15:07:59 -0800199 * @size: size of free area to find
200 * @align: alignment of free area to find
Grygorii Strashkob1154232014-01-21 15:50:16 -0800201 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
Tony Luckfc6daaf2015-06-24 16:58:09 -0700202 * @flags: pick from blocks based on memory attributes
Tang Chen79442ed2013-11-12 15:07:59 -0800203 *
204 * Utility called from memblock_find_in_range_node(), find free area bottom-up.
205 *
Mike Rapoport47cec442018-06-30 17:55:02 +0300206 * Return:
Tang Chen79442ed2013-11-12 15:07:59 -0800207 * Found address on success, 0 on failure.
208 */
209static phys_addr_t __init_memblock
210__memblock_find_range_bottom_up(phys_addr_t start, phys_addr_t end,
Tony Luckfc6daaf2015-06-24 16:58:09 -0700211 phys_addr_t size, phys_addr_t align, int nid,
Mike Rapoporte1720fe2018-06-30 17:55:01 +0300212 enum memblock_flags flags)
Tang Chen79442ed2013-11-12 15:07:59 -0800213{
214 phys_addr_t this_start, this_end, cand;
215 u64 i;
216
Tony Luckfc6daaf2015-06-24 16:58:09 -0700217 for_each_free_mem_range(i, nid, flags, &this_start, &this_end, NULL) {
Tang Chen79442ed2013-11-12 15:07:59 -0800218 this_start = clamp(this_start, start, end);
219 this_end = clamp(this_end, start, end);
220
221 cand = round_up(this_start, align);
222 if (cand < this_end && this_end - cand >= size)
223 return cand;
224 }
225
226 return 0;
227}
228
Tejun Heo7bd0b0f2011-12-08 10:22:09 -0800229/**
Tang Chen14028992013-11-12 15:07:57 -0800230 * __memblock_find_range_top_down - find free area utility, in top-down
231 * @start: start of candidate range
Mike Rapoport47cec442018-06-30 17:55:02 +0300232 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_ANYWHERE or
233 * %MEMBLOCK_ALLOC_ACCESSIBLE
Tang Chen14028992013-11-12 15:07:57 -0800234 * @size: size of free area to find
235 * @align: alignment of free area to find
Grygorii Strashkob1154232014-01-21 15:50:16 -0800236 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
Tony Luckfc6daaf2015-06-24 16:58:09 -0700237 * @flags: pick from blocks based on memory attributes
Tang Chen14028992013-11-12 15:07:57 -0800238 *
239 * Utility called from memblock_find_in_range_node(), find free area top-down.
240 *
Mike Rapoport47cec442018-06-30 17:55:02 +0300241 * Return:
Tang Chen79442ed2013-11-12 15:07:59 -0800242 * Found address on success, 0 on failure.
Tang Chen14028992013-11-12 15:07:57 -0800243 */
244static phys_addr_t __init_memblock
245__memblock_find_range_top_down(phys_addr_t start, phys_addr_t end,
Tony Luckfc6daaf2015-06-24 16:58:09 -0700246 phys_addr_t size, phys_addr_t align, int nid,
Mike Rapoporte1720fe2018-06-30 17:55:01 +0300247 enum memblock_flags flags)
Tang Chen14028992013-11-12 15:07:57 -0800248{
249 phys_addr_t this_start, this_end, cand;
250 u64 i;
251
Tony Luckfc6daaf2015-06-24 16:58:09 -0700252 for_each_free_mem_range_reverse(i, nid, flags, &this_start, &this_end,
253 NULL) {
Tang Chen14028992013-11-12 15:07:57 -0800254 this_start = clamp(this_start, start, end);
255 this_end = clamp(this_end, start, end);
256
257 if (this_end < size)
258 continue;
259
260 cand = round_down(this_end - size, align);
261 if (cand >= this_start)
262 return cand;
263 }
264
265 return 0;
266}
267
268/**
Tejun Heo7bd0b0f2011-12-08 10:22:09 -0800269 * memblock_find_in_range_node - find free area in given range and node
Tejun Heo7bd0b0f2011-12-08 10:22:09 -0800270 * @size: size of free area to find
271 * @align: alignment of free area to find
Grygorii Strashko87029ee2014-01-21 15:50:14 -0800272 * @start: start of candidate range
Mike Rapoport47cec442018-06-30 17:55:02 +0300273 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_ANYWHERE or
274 * %MEMBLOCK_ALLOC_ACCESSIBLE
Grygorii Strashkob1154232014-01-21 15:50:16 -0800275 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
Tony Luckfc6daaf2015-06-24 16:58:09 -0700276 * @flags: pick from blocks based on memory attributes
Tejun Heo7bd0b0f2011-12-08 10:22:09 -0800277 *
278 * Find @size free area aligned to @align in the specified range and node.
279 *
Mike Rapoport47cec442018-06-30 17:55:02 +0300280 * Return:
Tang Chen79442ed2013-11-12 15:07:59 -0800281 * Found address on success, 0 on failure.
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000282 */
Mike Rapoportc366ea82019-03-11 23:29:46 -0700283static phys_addr_t __init_memblock memblock_find_in_range_node(phys_addr_t size,
Grygorii Strashko87029ee2014-01-21 15:50:14 -0800284 phys_addr_t align, phys_addr_t start,
Mike Rapoporte1720fe2018-06-30 17:55:01 +0300285 phys_addr_t end, int nid,
286 enum memblock_flags flags)
Tang Chenf7210e62013-02-22 16:33:51 -0800287{
Tang Chenf7210e62013-02-22 16:33:51 -0800288 /* pump up @end */
Qian Caifed84c72018-12-28 00:36:29 -0800289 if (end == MEMBLOCK_ALLOC_ACCESSIBLE ||
290 end == MEMBLOCK_ALLOC_KASAN)
Tang Chenf7210e62013-02-22 16:33:51 -0800291 end = memblock.current_limit;
292
293 /* avoid allocating the first page */
294 start = max_t(phys_addr_t, start, PAGE_SIZE);
295 end = max(start, end);
Tang Chen79442ed2013-11-12 15:07:59 -0800296
Roman Gushchin2dcb3962021-02-04 18:32:36 -0800297 if (memblock_bottom_up())
298 return __memblock_find_range_bottom_up(start, end, size, align,
299 nid, flags);
300 else
301 return __memblock_find_range_top_down(start, end, size, align,
302 nid, flags);
Tang Chenf7210e62013-02-22 16:33:51 -0800303}
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000304
Tejun Heo7bd0b0f2011-12-08 10:22:09 -0800305/**
306 * memblock_find_in_range - find free area in given range
307 * @start: start of candidate range
Mike Rapoport47cec442018-06-30 17:55:02 +0300308 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_ANYWHERE or
309 * %MEMBLOCK_ALLOC_ACCESSIBLE
Tejun Heo7bd0b0f2011-12-08 10:22:09 -0800310 * @size: size of free area to find
311 * @align: alignment of free area to find
312 *
313 * Find @size free area aligned to @align in the specified range.
314 *
Mike Rapoport47cec442018-06-30 17:55:02 +0300315 * Return:
Tang Chen79442ed2013-11-12 15:07:59 -0800316 * Found address on success, 0 on failure.
Tejun Heo7bd0b0f2011-12-08 10:22:09 -0800317 */
Mike Rapoporta7259df2021-09-02 15:00:26 -0700318static phys_addr_t __init_memblock memblock_find_in_range(phys_addr_t start,
Tejun Heo7bd0b0f2011-12-08 10:22:09 -0800319 phys_addr_t end, phys_addr_t size,
320 phys_addr_t align)
321{
Tony Lucka3f5baf2015-06-24 16:58:12 -0700322 phys_addr_t ret;
Mike Rapoporte1720fe2018-06-30 17:55:01 +0300323 enum memblock_flags flags = choose_memblock_flags();
Tony Lucka3f5baf2015-06-24 16:58:12 -0700324
325again:
326 ret = memblock_find_in_range_node(size, align, start, end,
327 NUMA_NO_NODE, flags);
328
329 if (!ret && (flags & MEMBLOCK_MIRROR)) {
330 pr_warn("Could not allocate %pap bytes of mirrored memory\n",
331 &size);
332 flags &= ~MEMBLOCK_MIRROR;
333 goto again;
334 }
335
336 return ret;
Tejun Heo7bd0b0f2011-12-08 10:22:09 -0800337}
338
Yinghai Lu10d06432010-07-28 15:43:02 +1000339static void __init_memblock memblock_remove_region(struct memblock_type *type, unsigned long r)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000340{
Tejun Heo1440c4e2011-12-08 10:22:08 -0800341 type->total_size -= type->regions[r].size;
Tejun Heo7c0caeb2011-07-14 11:43:42 +0200342 memmove(&type->regions[r], &type->regions[r + 1],
343 (type->cnt - (r + 1)) * sizeof(type->regions[r]));
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000344 type->cnt--;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000345
Benjamin Herrenschmidt8f7a6602011-03-22 16:33:43 -0700346 /* Special case for empty arrays */
347 if (type->cnt == 0) {
Tejun Heo1440c4e2011-12-08 10:22:08 -0800348 WARN_ON(type->total_size != 0);
Benjamin Herrenschmidt8f7a6602011-03-22 16:33:43 -0700349 type->cnt = 1;
350 type->regions[0].base = 0;
351 type->regions[0].size = 0;
Tang Chen66a20752014-01-21 15:49:20 -0800352 type->regions[0].flags = 0;
Tejun Heo7c0caeb2011-07-14 11:43:42 +0200353 memblock_set_region_node(&type->regions[0], MAX_NUMNODES);
Benjamin Herrenschmidt8f7a6602011-03-22 16:33:43 -0700354 }
Yinghai Lu95f72d12010-07-12 14:36:09 +1000355}
356
Mike Rapoport350e88b2019-05-13 17:22:59 -0700357#ifndef CONFIG_ARCH_KEEP_MEMBLOCK
Pavel Tatashin3010f872017-08-18 15:16:05 -0700358/**
Mike Rapoport47cec442018-06-30 17:55:02 +0300359 * memblock_discard - discard memory and reserved arrays if they were allocated
Pavel Tatashin3010f872017-08-18 15:16:05 -0700360 */
361void __init memblock_discard(void)
Yinghai Lu29f67382012-07-11 14:02:56 -0700362{
Pavel Tatashin3010f872017-08-18 15:16:05 -0700363 phys_addr_t addr, size;
Yinghai Lu29f67382012-07-11 14:02:56 -0700364
Pavel Tatashin3010f872017-08-18 15:16:05 -0700365 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);
Miaohe Linfd21a0b2022-02-17 22:53:27 +0800369 if (memblock_reserved_in_slab)
370 kfree(memblock.reserved.regions);
371 else
372 __memblock_free_late(addr, size);
Pavel Tatashin3010f872017-08-18 15:16:05 -0700373 }
Yinghai Lu29f67382012-07-11 14:02:56 -0700374
Pavel Tatashin91b540f92017-08-25 15:55:46 -0700375 if (memblock.memory.regions != memblock_memory_init_regions) {
Pavel Tatashin3010f872017-08-18 15:16:05 -0700376 addr = __pa(memblock.memory.regions);
377 size = PAGE_ALIGN(sizeof(struct memblock_region) *
378 memblock.memory.max);
Miaohe Linfd21a0b2022-02-17 22:53:27 +0800379 if (memblock_memory_in_slab)
380 kfree(memblock.memory.regions);
381 else
382 __memblock_free_late(addr, size);
Pavel Tatashin3010f872017-08-18 15:16:05 -0700383 }
Mike Rapoport9f3d5ea2020-10-13 16:58:25 -0700384
385 memblock_memory = NULL;
Yinghai Lu29f67382012-07-11 14:02:56 -0700386}
Philipp Hachtmann5e270e22014-01-23 15:53:11 -0800387#endif
388
Greg Pearson48c3b582012-06-20 12:53:05 -0700389/**
390 * memblock_double_array - double the size of the memblock regions array
391 * @type: memblock type of the regions array being doubled
392 * @new_area_start: starting address of memory range to avoid overlap with
393 * @new_area_size: size of memory range to avoid overlap with
394 *
395 * Double the size of the @type regions array. If memblock is being used to
396 * allocate memory for a new reserved regions array and there is a previously
Mike Rapoport47cec442018-06-30 17:55:02 +0300397 * allocated memory range [@new_area_start, @new_area_start + @new_area_size]
Greg Pearson48c3b582012-06-20 12:53:05 -0700398 * waiting to be reserved, ensure the memory used by the new array does
399 * not overlap.
400 *
Mike Rapoport47cec442018-06-30 17:55:02 +0300401 * Return:
Greg Pearson48c3b582012-06-20 12:53:05 -0700402 * 0 on success, -1 on failure.
403 */
404static int __init_memblock memblock_double_array(struct memblock_type *type,
405 phys_addr_t new_area_start,
406 phys_addr_t new_area_size)
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700407{
408 struct memblock_region *new_array, *old_array;
Yinghai Lu29f67382012-07-11 14:02:56 -0700409 phys_addr_t old_alloc_size, new_alloc_size;
Mike Rapoporta36aab82018-08-17 15:47:17 -0700410 phys_addr_t old_size, new_size, addr, new_end;
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700411 int use_slab = slab_is_available();
Gavin Shan181eb392012-05-29 15:06:50 -0700412 int *in_slab;
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700413
414 /* We don't allow resizing until we know about the reserved regions
415 * of memory that aren't suitable for allocation
416 */
417 if (!memblock_can_resize)
418 return -1;
419
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700420 /* Calculate new doubled size */
421 old_size = type->max * sizeof(struct memblock_region);
422 new_size = old_size << 1;
Yinghai Lu29f67382012-07-11 14:02:56 -0700423 /*
424 * We need to allocated new one align to PAGE_SIZE,
425 * so we can free them completely later.
426 */
427 old_alloc_size = PAGE_ALIGN(old_size);
428 new_alloc_size = PAGE_ALIGN(new_size);
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700429
Gavin Shan181eb392012-05-29 15:06:50 -0700430 /* Retrieve the slab flag */
431 if (type == &memblock.memory)
432 in_slab = &memblock_memory_in_slab;
433 else
434 in_slab = &memblock_reserved_in_slab;
435
Mike Rapoporta2974132019-03-11 23:30:54 -0700436 /* Try to find some space for it */
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700437 if (use_slab) {
438 new_array = kmalloc(new_size, GFP_KERNEL);
Tejun Heo1f5026a2011-07-12 09:58:09 +0200439 addr = new_array ? __pa(new_array) : 0;
Gavin Shan4e2f0772012-05-29 15:06:50 -0700440 } else {
Greg Pearson48c3b582012-06-20 12:53:05 -0700441 /* only exclude range when trying to double reserved.regions */
442 if (type != &memblock.reserved)
443 new_area_start = new_area_size = 0;
444
445 addr = memblock_find_in_range(new_area_start + new_area_size,
446 memblock.current_limit,
Yinghai Lu29f67382012-07-11 14:02:56 -0700447 new_alloc_size, PAGE_SIZE);
Greg Pearson48c3b582012-06-20 12:53:05 -0700448 if (!addr && new_area_size)
449 addr = memblock_find_in_range(0,
Andrew Mortonfd073832012-07-31 16:42:40 -0700450 min(new_area_start, memblock.current_limit),
451 new_alloc_size, PAGE_SIZE);
Greg Pearson48c3b582012-06-20 12:53:05 -0700452
Sachin Kamat15674862012-09-04 13:55:05 +0530453 new_array = addr ? __va(addr) : NULL;
Gavin Shan4e2f0772012-05-29 15:06:50 -0700454 }
Tejun Heo1f5026a2011-07-12 09:58:09 +0200455 if (!addr) {
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700456 pr_err("memblock: Failed to double %s array from %ld to %ld entries !\n",
Heiko Carstens0262d9c2017-02-24 14:55:59 -0800457 type->name, type->max, type->max * 2);
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700458 return -1;
459 }
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700460
Mike Rapoporta36aab82018-08-17 15:47:17 -0700461 new_end = addr + new_size - 1;
462 memblock_dbg("memblock: %s is doubled to %ld at [%pa-%pa]",
463 type->name, type->max * 2, &addr, &new_end);
Yinghai Luea9e4372010-07-28 15:13:22 +1000464
Andrew Mortonfd073832012-07-31 16:42:40 -0700465 /*
466 * Found space, we now need to move the array over before we add the
467 * reserved region since it may be our reserved array itself that is
468 * full.
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700469 */
470 memcpy(new_array, type->regions, old_size);
471 memset(new_array + type->max, 0, old_size);
472 old_array = type->regions;
473 type->regions = new_array;
474 type->max <<= 1;
475
Andrew Mortonfd073832012-07-31 16:42:40 -0700476 /* Free old array. We needn't free it if the array is the static one */
Gavin Shan181eb392012-05-29 15:06:50 -0700477 if (*in_slab)
478 kfree(old_array);
479 else if (old_array != memblock_memory_init_regions &&
480 old_array != memblock_reserved_init_regions)
Linus Torvalds77e02cf2021-09-14 13:23:22 -0700481 memblock_free_ptr(old_array, old_alloc_size);
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700482
Andrew Mortonfd073832012-07-31 16:42:40 -0700483 /*
484 * Reserve the new array if that comes from the memblock. Otherwise, we
485 * needn't do it
Gavin Shan181eb392012-05-29 15:06:50 -0700486 */
487 if (!use_slab)
Yinghai Lu29f67382012-07-11 14:02:56 -0700488 BUG_ON(memblock_reserve(addr, new_alloc_size));
Gavin Shan181eb392012-05-29 15:06:50 -0700489
490 /* Update slab flag */
491 *in_slab = use_slab;
492
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700493 return 0;
494}
495
Tejun Heo784656f92011-07-12 11:15:55 +0200496/**
497 * memblock_merge_regions - merge neighboring compatible regions
498 * @type: memblock type to scan
499 *
500 * Scan @type and merge neighboring compatible regions.
501 */
502static void __init_memblock memblock_merge_regions(struct memblock_type *type)
503{
504 int i = 0;
505
506 /* cnt never goes below 1 */
507 while (i < type->cnt - 1) {
508 struct memblock_region *this = &type->regions[i];
509 struct memblock_region *next = &type->regions[i + 1];
510
Tejun Heo7c0caeb2011-07-14 11:43:42 +0200511 if (this->base + this->size != next->base ||
512 memblock_get_region_node(this) !=
Tang Chen66a20752014-01-21 15:49:20 -0800513 memblock_get_region_node(next) ||
514 this->flags != next->flags) {
Tejun Heo784656f92011-07-12 11:15:55 +0200515 BUG_ON(this->base + this->size > next->base);
516 i++;
517 continue;
518 }
519
520 this->size += next->size;
Lin Fengc0232ae2013-01-11 14:31:44 -0800521 /* move forward from next + 1, index of which is i + 2 */
522 memmove(next, next + 1, (type->cnt - (i + 2)) * sizeof(*next));
Tejun Heo784656f92011-07-12 11:15:55 +0200523 type->cnt--;
524 }
525}
526
527/**
528 * memblock_insert_region - insert new memblock region
Tang Chen209ff862013-04-29 15:08:41 -0700529 * @type: memblock type to insert into
530 * @idx: index for the insertion point
531 * @base: base address of the new region
532 * @size: size of the new region
533 * @nid: node id of the new region
Tang Chen66a20752014-01-21 15:49:20 -0800534 * @flags: flags of the new region
Tejun Heo784656f92011-07-12 11:15:55 +0200535 *
Mike Rapoport47cec442018-06-30 17:55:02 +0300536 * Insert new memblock region [@base, @base + @size) into @type at @idx.
Alexander Kuleshov412d0002016-08-04 15:31:46 -0700537 * @type must already have extra room to accommodate the new region.
Tejun Heo784656f92011-07-12 11:15:55 +0200538 */
539static void __init_memblock memblock_insert_region(struct memblock_type *type,
540 int idx, phys_addr_t base,
Tang Chen66a20752014-01-21 15:49:20 -0800541 phys_addr_t size,
Mike Rapoporte1720fe2018-06-30 17:55:01 +0300542 int nid,
543 enum memblock_flags flags)
Tejun Heo784656f92011-07-12 11:15:55 +0200544{
545 struct memblock_region *rgn = &type->regions[idx];
546
547 BUG_ON(type->cnt >= type->max);
548 memmove(rgn + 1, rgn, (type->cnt - idx) * sizeof(*rgn));
549 rgn->base = base;
550 rgn->size = size;
Tang Chen66a20752014-01-21 15:49:20 -0800551 rgn->flags = flags;
Tejun Heo7c0caeb2011-07-14 11:43:42 +0200552 memblock_set_region_node(rgn, nid);
Tejun Heo784656f92011-07-12 11:15:55 +0200553 type->cnt++;
Tejun Heo1440c4e2011-12-08 10:22:08 -0800554 type->total_size += size;
Tejun Heo784656f92011-07-12 11:15:55 +0200555}
556
557/**
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +0100558 * memblock_add_range - add new memblock region
Tejun Heo784656f92011-07-12 11:15:55 +0200559 * @type: memblock type to add new region into
560 * @base: base address of the new region
561 * @size: size of the new region
Tejun Heo7fb0bc32011-12-08 10:22:08 -0800562 * @nid: nid of the new region
Tang Chen66a20752014-01-21 15:49:20 -0800563 * @flags: flags of the new region
Tejun Heo784656f92011-07-12 11:15:55 +0200564 *
Mike Rapoport47cec442018-06-30 17:55:02 +0300565 * Add new memblock region [@base, @base + @size) into @type. The new region
Tejun Heo784656f92011-07-12 11:15:55 +0200566 * is allowed to overlap with existing ones - overlaps don't affect already
567 * existing regions. @type is guaranteed to be minimal (all neighbouring
568 * compatible regions are merged) after the addition.
569 *
Mike Rapoport47cec442018-06-30 17:55:02 +0300570 * Return:
Tejun Heo784656f92011-07-12 11:15:55 +0200571 * 0 on success, -errno on failure.
572 */
Anshuman Khandual02634a42020-01-30 22:14:20 -0800573static int __init_memblock memblock_add_range(struct memblock_type *type,
Tang Chen66a20752014-01-21 15:49:20 -0800574 phys_addr_t base, phys_addr_t size,
Mike Rapoporte1720fe2018-06-30 17:55:01 +0300575 int nid, enum memblock_flags flags)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000576{
Tejun Heo784656f92011-07-12 11:15:55 +0200577 bool insert = false;
Tejun Heoeb18f1b2011-12-08 10:22:07 -0800578 phys_addr_t obase = base;
579 phys_addr_t end = base + memblock_cap_size(base, &size);
Alexander Kuleshov8c9c1702016-01-14 15:20:42 -0800580 int idx, nr_new;
581 struct memblock_region *rgn;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000582
Tejun Heob3dc6272012-04-20 08:31:34 -0700583 if (!size)
584 return 0;
585
Tejun Heo784656f92011-07-12 11:15:55 +0200586 /* special case for empty array */
587 if (type->regions[0].size == 0) {
Tejun Heo1440c4e2011-12-08 10:22:08 -0800588 WARN_ON(type->cnt != 1 || type->total_size);
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000589 type->regions[0].base = base;
590 type->regions[0].size = size;
Tang Chen66a20752014-01-21 15:49:20 -0800591 type->regions[0].flags = flags;
Tejun Heo7fb0bc32011-12-08 10:22:08 -0800592 memblock_set_region_node(&type->regions[0], nid);
Tejun Heo1440c4e2011-12-08 10:22:08 -0800593 type->total_size = size;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000594 return 0;
595 }
Tejun Heo784656f92011-07-12 11:15:55 +0200596repeat:
597 /*
598 * The following is executed twice. Once with %false @insert and
599 * then with %true. The first counts the number of regions needed
Alexander Kuleshov412d0002016-08-04 15:31:46 -0700600 * to accommodate the new area. The second actually inserts them.
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700601 */
Tejun Heo784656f92011-07-12 11:15:55 +0200602 base = obase;
603 nr_new = 0;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000604
Gioh Kim66e8b432017-11-15 17:33:42 -0800605 for_each_memblock_type(idx, type, rgn) {
Tejun Heo784656f92011-07-12 11:15:55 +0200606 phys_addr_t rbase = rgn->base;
607 phys_addr_t rend = rbase + rgn->size;
608
609 if (rbase >= end)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000610 break;
Tejun Heo784656f92011-07-12 11:15:55 +0200611 if (rend <= base)
612 continue;
613 /*
614 * @rgn overlaps. If it separates the lower part of new
615 * area, insert that portion.
616 */
617 if (rbase > base) {
Mike Rapoporta9ee6cf2021-06-28 19:43:01 -0700618#ifdef CONFIG_NUMA
Wei Yangc0a29492015-09-04 15:47:38 -0700619 WARN_ON(nid != memblock_get_region_node(rgn));
620#endif
Wei Yang4fcab5f2015-09-08 14:59:53 -0700621 WARN_ON(flags != rgn->flags);
Tejun Heo784656f92011-07-12 11:15:55 +0200622 nr_new++;
623 if (insert)
Alexander Kuleshov8c9c1702016-01-14 15:20:42 -0800624 memblock_insert_region(type, idx++, base,
Tang Chen66a20752014-01-21 15:49:20 -0800625 rbase - base, nid,
626 flags);
Yinghai Lu95f72d12010-07-12 14:36:09 +1000627 }
Tejun Heo784656f92011-07-12 11:15:55 +0200628 /* area below @rend is dealt with, forget about it */
629 base = min(rend, end);
Yinghai Lu95f72d12010-07-12 14:36:09 +1000630 }
Yinghai Lu95f72d12010-07-12 14:36:09 +1000631
Tejun Heo784656f92011-07-12 11:15:55 +0200632 /* insert the remaining portion */
633 if (base < end) {
634 nr_new++;
635 if (insert)
Alexander Kuleshov8c9c1702016-01-14 15:20:42 -0800636 memblock_insert_region(type, idx, base, end - base,
Tang Chen66a20752014-01-21 15:49:20 -0800637 nid, flags);
Tejun Heo784656f92011-07-12 11:15:55 +0200638 }
639
nimisoloef3cc4d2016-07-26 15:24:56 -0700640 if (!nr_new)
641 return 0;
642
Tejun Heo784656f92011-07-12 11:15:55 +0200643 /*
644 * If this was the first round, resize array and repeat for actual
645 * insertions; otherwise, merge and return.
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700646 */
Tejun Heo784656f92011-07-12 11:15:55 +0200647 if (!insert) {
648 while (type->cnt + nr_new > type->max)
Greg Pearson48c3b582012-06-20 12:53:05 -0700649 if (memblock_double_array(type, obase, size) < 0)
Tejun Heo784656f92011-07-12 11:15:55 +0200650 return -ENOMEM;
651 insert = true;
652 goto repeat;
653 } else {
654 memblock_merge_regions(type);
655 return 0;
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700656 }
Yinghai Lu95f72d12010-07-12 14:36:09 +1000657}
658
Mike Rapoport48a833c2018-06-30 17:55:03 +0300659/**
660 * memblock_add_node - add new memblock region within a NUMA node
661 * @base: base address of the new region
662 * @size: size of the new region
663 * @nid: nid of the new region
664 *
665 * Add new memblock region [@base, @base + @size) to the "memory"
666 * type. See memblock_add_range() description for mode details
667 *
668 * Return:
669 * 0 on success, -errno on failure.
670 */
Tejun Heo7fb0bc32011-12-08 10:22:08 -0800671int __init_memblock memblock_add_node(phys_addr_t base, phys_addr_t size,
672 int nid)
673{
Geert Uytterhoeven00974b92021-08-11 10:54:36 +0200674 phys_addr_t end = base + size - 1;
675
676 memblock_dbg("%s: [%pa-%pa] nid=%d %pS\n", __func__,
677 &base, &end, nid, (void *)_RET_IP_);
678
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +0100679 return memblock_add_range(&memblock.memory, base, size, nid, 0);
Tejun Heo7fb0bc32011-12-08 10:22:08 -0800680}
681
Mike Rapoport48a833c2018-06-30 17:55:03 +0300682/**
683 * memblock_add - add new memblock region
684 * @base: base address of the new region
685 * @size: size of the new region
686 *
687 * Add new memblock region [@base, @base + @size) to the "memory"
688 * type. See memblock_add_range() description for mode details
689 *
690 * Return:
691 * 0 on success, -errno on failure.
692 */
Alexander Kuleshovf705ac42016-05-20 16:57:35 -0700693int __init_memblock memblock_add(phys_addr_t base, phys_addr_t size)
Alexander Kuleshov6a4055b2015-04-15 16:14:44 -0700694{
Miles Chen5d63f812017-02-22 15:46:42 -0800695 phys_addr_t end = base + size - 1;
696
Anshuman Khanduala090d712020-01-30 22:14:23 -0800697 memblock_dbg("%s: [%pa-%pa] %pS\n", __func__,
Miles Chen5d63f812017-02-22 15:46:42 -0800698 &base, &end, (void *)_RET_IP_);
Alexander Kuleshov6a4055b2015-04-15 16:14:44 -0700699
Alexander Kuleshovf705ac42016-05-20 16:57:35 -0700700 return memblock_add_range(&memblock.memory, base, size, MAX_NUMNODES, 0);
Yinghai Lu95f72d12010-07-12 14:36:09 +1000701}
702
Tejun Heo6a9ceb32011-12-08 10:22:07 -0800703/**
704 * memblock_isolate_range - isolate given range into disjoint memblocks
705 * @type: memblock type to isolate range for
706 * @base: base of range to isolate
707 * @size: size of range to isolate
708 * @start_rgn: out parameter for the start of isolated region
709 * @end_rgn: out parameter for the end of isolated region
710 *
711 * Walk @type and ensure that regions don't cross the boundaries defined by
Mike Rapoport47cec442018-06-30 17:55:02 +0300712 * [@base, @base + @size). Crossing regions are split at the boundaries,
Tejun Heo6a9ceb32011-12-08 10:22:07 -0800713 * which may create at most two more regions. The index of the first
714 * region inside the range is returned in *@start_rgn and end in *@end_rgn.
715 *
Mike Rapoport47cec442018-06-30 17:55:02 +0300716 * Return:
Tejun Heo6a9ceb32011-12-08 10:22:07 -0800717 * 0 on success, -errno on failure.
718 */
719static int __init_memblock memblock_isolate_range(struct memblock_type *type,
720 phys_addr_t base, phys_addr_t size,
721 int *start_rgn, int *end_rgn)
722{
Tejun Heoeb18f1b2011-12-08 10:22:07 -0800723 phys_addr_t end = base + memblock_cap_size(base, &size);
Alexander Kuleshov8c9c1702016-01-14 15:20:42 -0800724 int idx;
725 struct memblock_region *rgn;
Tejun Heo6a9ceb32011-12-08 10:22:07 -0800726
727 *start_rgn = *end_rgn = 0;
728
Tejun Heob3dc6272012-04-20 08:31:34 -0700729 if (!size)
730 return 0;
731
Tejun Heo6a9ceb32011-12-08 10:22:07 -0800732 /* we'll create at most two more regions */
733 while (type->cnt + 2 > type->max)
Greg Pearson48c3b582012-06-20 12:53:05 -0700734 if (memblock_double_array(type, base, size) < 0)
Tejun Heo6a9ceb32011-12-08 10:22:07 -0800735 return -ENOMEM;
736
Gioh Kim66e8b432017-11-15 17:33:42 -0800737 for_each_memblock_type(idx, type, rgn) {
Tejun Heo6a9ceb32011-12-08 10:22:07 -0800738 phys_addr_t rbase = rgn->base;
739 phys_addr_t rend = rbase + rgn->size;
740
741 if (rbase >= end)
742 break;
743 if (rend <= base)
744 continue;
745
746 if (rbase < base) {
747 /*
748 * @rgn intersects from below. Split and continue
749 * to process the next region - the new top half.
750 */
751 rgn->base = base;
Tejun Heo1440c4e2011-12-08 10:22:08 -0800752 rgn->size -= base - rbase;
753 type->total_size -= base - rbase;
Alexander Kuleshov8c9c1702016-01-14 15:20:42 -0800754 memblock_insert_region(type, idx, rbase, base - rbase,
Tang Chen66a20752014-01-21 15:49:20 -0800755 memblock_get_region_node(rgn),
756 rgn->flags);
Tejun Heo6a9ceb32011-12-08 10:22:07 -0800757 } else if (rend > end) {
758 /*
759 * @rgn intersects from above. Split and redo the
760 * current region - the new bottom half.
761 */
762 rgn->base = end;
Tejun Heo1440c4e2011-12-08 10:22:08 -0800763 rgn->size -= end - rbase;
764 type->total_size -= end - rbase;
Alexander Kuleshov8c9c1702016-01-14 15:20:42 -0800765 memblock_insert_region(type, idx--, rbase, end - rbase,
Tang Chen66a20752014-01-21 15:49:20 -0800766 memblock_get_region_node(rgn),
767 rgn->flags);
Tejun Heo6a9ceb32011-12-08 10:22:07 -0800768 } else {
769 /* @rgn is fully contained, record it */
770 if (!*end_rgn)
Alexander Kuleshov8c9c1702016-01-14 15:20:42 -0800771 *start_rgn = idx;
772 *end_rgn = idx + 1;
Tejun Heo6a9ceb32011-12-08 10:22:07 -0800773 }
774 }
775
776 return 0;
777}
Tejun Heo6a9ceb32011-12-08 10:22:07 -0800778
Alexander Kuleshov35bd16a2015-11-05 18:47:00 -0800779static int __init_memblock memblock_remove_range(struct memblock_type *type,
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +0100780 phys_addr_t base, phys_addr_t size)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000781{
Tejun Heo71936182011-12-08 10:22:07 -0800782 int start_rgn, end_rgn;
783 int i, ret;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000784
Tejun Heo71936182011-12-08 10:22:07 -0800785 ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
786 if (ret)
787 return ret;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000788
Tejun Heo71936182011-12-08 10:22:07 -0800789 for (i = end_rgn - 1; i >= start_rgn; i--)
790 memblock_remove_region(type, i);
Benjamin Herrenschmidt8f7a6602011-03-22 16:33:43 -0700791 return 0;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000792}
793
Tejun Heo581adcb2011-12-08 10:22:06 -0800794int __init_memblock memblock_remove(phys_addr_t base, phys_addr_t size)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000795{
Minchan Kim25cf23d2018-06-07 17:07:35 -0700796 phys_addr_t end = base + size - 1;
797
Anshuman Khanduala090d712020-01-30 22:14:23 -0800798 memblock_dbg("%s: [%pa-%pa] %pS\n", __func__,
Minchan Kim25cf23d2018-06-07 17:07:35 -0700799 &base, &end, (void *)_RET_IP_);
800
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +0100801 return memblock_remove_range(&memblock.memory, base, size);
Yinghai Lu95f72d12010-07-12 14:36:09 +1000802}
803
Mike Rapoport4d728682018-12-28 00:35:29 -0800804/**
Linus Torvalds77e02cf2021-09-14 13:23:22 -0700805 * memblock_free_ptr - free boot memory allocation
806 * @ptr: starting address of the boot memory allocation
807 * @size: size of the boot memory block in bytes
808 *
809 * Free boot memory block previously allocated by memblock_alloc_xx() API.
810 * The freeing memory will not be released to the buddy allocator.
811 */
812void __init_memblock memblock_free_ptr(void *ptr, size_t size)
813{
814 if (ptr)
815 memblock_free(__pa(ptr), size);
816}
817
818/**
Mike Rapoport4d728682018-12-28 00:35:29 -0800819 * memblock_free - free boot memory block
820 * @base: phys starting address of the boot memory block
821 * @size: size of the boot memory block in bytes
822 *
823 * Free boot memory block previously allocated by memblock_alloc_xx() API.
824 * The freeing memory will not be released to the buddy allocator.
825 */
Tejun Heo581adcb2011-12-08 10:22:06 -0800826int __init_memblock memblock_free(phys_addr_t base, phys_addr_t size)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000827{
Miles Chen5d63f812017-02-22 15:46:42 -0800828 phys_addr_t end = base + size - 1;
829
Anshuman Khanduala090d712020-01-30 22:14:23 -0800830 memblock_dbg("%s: [%pa-%pa] %pS\n", __func__,
Miles Chen5d63f812017-02-22 15:46:42 -0800831 &base, &end, (void *)_RET_IP_);
Tejun Heo24aa0782011-07-12 11:16:06 +0200832
Catalin Marinas9099dae2016-10-11 13:55:11 -0700833 kmemleak_free_part_phys(base, size);
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +0100834 return memblock_remove_range(&memblock.reserved, base, size);
Yinghai Lu95f72d12010-07-12 14:36:09 +1000835}
836
Alexander Kuleshovf705ac42016-05-20 16:57:35 -0700837int __init_memblock memblock_reserve(phys_addr_t base, phys_addr_t size)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000838{
Miles Chen5d63f812017-02-22 15:46:42 -0800839 phys_addr_t end = base + size - 1;
840
Anshuman Khanduala090d712020-01-30 22:14:23 -0800841 memblock_dbg("%s: [%pa-%pa] %pS\n", __func__,
Miles Chen5d63f812017-02-22 15:46:42 -0800842 &base, &end, (void *)_RET_IP_);
Yinghai Lu95f72d12010-07-12 14:36:09 +1000843
Alexander Kuleshovf705ac42016-05-20 16:57:35 -0700844 return memblock_add_range(&memblock.reserved, base, size, MAX_NUMNODES, 0);
Yinghai Lu95f72d12010-07-12 14:36:09 +1000845}
846
Anshuman Khandual02634a42020-01-30 22:14:20 -0800847#ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
848int __init_memblock memblock_physmem_add(phys_addr_t base, phys_addr_t size)
849{
850 phys_addr_t end = base + size - 1;
851
852 memblock_dbg("%s: [%pa-%pa] %pS\n", __func__,
853 &base, &end, (void *)_RET_IP_);
854
David Hildenbrand77649902020-07-01 16:18:29 +0200855 return memblock_add_range(&physmem, base, size, MAX_NUMNODES, 0);
Anshuman Khandual02634a42020-01-30 22:14:20 -0800856}
857#endif
858
Tejun Heo35fd0802011-07-12 11:15:59 +0200859/**
Mike Rapoport47cec442018-06-30 17:55:02 +0300860 * memblock_setclr_flag - set or clear flag for a memory region
861 * @base: base address of the region
862 * @size: size of the region
863 * @set: set or clear the flag
Haitao Shi8958b242020-12-15 20:47:26 -0800864 * @flag: the flag to update
Tang Chen66b16ed2014-01-21 15:49:23 -0800865 *
Tony Luck4308ce12014-12-12 16:54:59 -0800866 * This function isolates region [@base, @base + @size), and sets/clears flag
Tang Chen66b16ed2014-01-21 15:49:23 -0800867 *
Mike Rapoport47cec442018-06-30 17:55:02 +0300868 * Return: 0 on success, -errno on failure.
Tang Chen66b16ed2014-01-21 15:49:23 -0800869 */
Tony Luck4308ce12014-12-12 16:54:59 -0800870static int __init_memblock memblock_setclr_flag(phys_addr_t base,
871 phys_addr_t size, int set, int flag)
Tang Chen66b16ed2014-01-21 15:49:23 -0800872{
873 struct memblock_type *type = &memblock.memory;
874 int i, ret, start_rgn, end_rgn;
875
876 ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
877 if (ret)
878 return ret;
879
Mike Rapoportfe145122019-03-11 23:30:46 -0700880 for (i = start_rgn; i < end_rgn; i++) {
881 struct memblock_region *r = &type->regions[i];
882
Tony Luck4308ce12014-12-12 16:54:59 -0800883 if (set)
Mike Rapoportfe145122019-03-11 23:30:46 -0700884 r->flags |= flag;
Tony Luck4308ce12014-12-12 16:54:59 -0800885 else
Mike Rapoportfe145122019-03-11 23:30:46 -0700886 r->flags &= ~flag;
887 }
Tang Chen66b16ed2014-01-21 15:49:23 -0800888
889 memblock_merge_regions(type);
890 return 0;
891}
892
893/**
Tony Luck4308ce12014-12-12 16:54:59 -0800894 * memblock_mark_hotplug - Mark hotpluggable memory with flag MEMBLOCK_HOTPLUG.
895 * @base: the base phys addr of the region
896 * @size: the size of the region
897 *
Mike Rapoport47cec442018-06-30 17:55:02 +0300898 * Return: 0 on success, -errno on failure.
Tony Luck4308ce12014-12-12 16:54:59 -0800899 */
900int __init_memblock memblock_mark_hotplug(phys_addr_t base, phys_addr_t size)
901{
902 return memblock_setclr_flag(base, size, 1, MEMBLOCK_HOTPLUG);
903}
904
905/**
Tang Chen66b16ed2014-01-21 15:49:23 -0800906 * memblock_clear_hotplug - Clear flag MEMBLOCK_HOTPLUG for a specified region.
907 * @base: the base phys addr of the region
908 * @size: the size of the region
909 *
Mike Rapoport47cec442018-06-30 17:55:02 +0300910 * Return: 0 on success, -errno on failure.
Tang Chen66b16ed2014-01-21 15:49:23 -0800911 */
912int __init_memblock memblock_clear_hotplug(phys_addr_t base, phys_addr_t size)
913{
Tony Luck4308ce12014-12-12 16:54:59 -0800914 return memblock_setclr_flag(base, size, 0, MEMBLOCK_HOTPLUG);
Tang Chen66b16ed2014-01-21 15:49:23 -0800915}
916
917/**
Tony Lucka3f5baf2015-06-24 16:58:12 -0700918 * memblock_mark_mirror - Mark mirrored memory with flag MEMBLOCK_MIRROR.
919 * @base: the base phys addr of the region
920 * @size: the size of the region
921 *
Mike Rapoport47cec442018-06-30 17:55:02 +0300922 * Return: 0 on success, -errno on failure.
Tony Lucka3f5baf2015-06-24 16:58:12 -0700923 */
924int __init_memblock memblock_mark_mirror(phys_addr_t base, phys_addr_t size)
925{
926 system_has_some_mirror = true;
927
928 return memblock_setclr_flag(base, size, 1, MEMBLOCK_MIRROR);
929}
930
Ard Biesheuvelbf3d3cc2015-11-30 13:28:15 +0100931/**
932 * memblock_mark_nomap - Mark a memory region with flag MEMBLOCK_NOMAP.
933 * @base: the base phys addr of the region
934 * @size: the size of the region
935 *
Mike Rapoport9092d4f2021-06-30 18:51:16 -0700936 * The memory regions marked with %MEMBLOCK_NOMAP will not be added to the
937 * direct mapping of the physical memory. These regions will still be
938 * covered by the memory map. The struct page representing NOMAP memory
939 * frames in the memory map will be PageReserved()
940 *
Mike Rapoport658aafc2021-10-21 10:09:29 +0300941 * Note: if the memory being marked %MEMBLOCK_NOMAP was allocated from
942 * memblock, the caller must inform kmemleak to ignore that memory
943 *
Mike Rapoport47cec442018-06-30 17:55:02 +0300944 * Return: 0 on success, -errno on failure.
Ard Biesheuvelbf3d3cc2015-11-30 13:28:15 +0100945 */
946int __init_memblock memblock_mark_nomap(phys_addr_t base, phys_addr_t size)
947{
Mike Rapoport6c9a5452021-10-21 10:09:28 +0300948 return memblock_setclr_flag(base, size, 1, MEMBLOCK_NOMAP);
Ard Biesheuvelbf3d3cc2015-11-30 13:28:15 +0100949}
Tony Lucka3f5baf2015-06-24 16:58:12 -0700950
951/**
AKASHI Takahiro4c546b82017-04-03 11:23:54 +0900952 * memblock_clear_nomap - Clear flag MEMBLOCK_NOMAP for a specified region.
953 * @base: the base phys addr of the region
954 * @size: the size of the region
955 *
Mike Rapoport47cec442018-06-30 17:55:02 +0300956 * Return: 0 on success, -errno on failure.
AKASHI Takahiro4c546b82017-04-03 11:23:54 +0900957 */
958int __init_memblock memblock_clear_nomap(phys_addr_t base, phys_addr_t size)
959{
960 return memblock_setclr_flag(base, size, 0, MEMBLOCK_NOMAP);
961}
962
Mike Rapoport9f3d5ea2020-10-13 16:58:25 -0700963static bool should_skip_region(struct memblock_type *type,
964 struct memblock_region *m,
965 int nid, int flags)
Mike Rapoportc9a688a2019-03-11 23:30:50 -0700966{
967 int m_nid = memblock_get_region_node(m);
968
Mike Rapoport9f3d5ea2020-10-13 16:58:25 -0700969 /* we never skip regions when iterating memblock.reserved or physmem */
970 if (type != memblock_memory)
971 return false;
972
Mike Rapoportc9a688a2019-03-11 23:30:50 -0700973 /* only memory regions are associated with nodes, check it */
974 if (nid != NUMA_NO_NODE && nid != m_nid)
975 return true;
976
977 /* skip hotpluggable memory regions if needed */
Mike Rapoport79e482e2021-07-23 15:50:26 -0700978 if (movable_node_is_enabled() && memblock_is_hotpluggable(m) &&
979 !(flags & MEMBLOCK_HOTPLUG))
Mike Rapoportc9a688a2019-03-11 23:30:50 -0700980 return true;
981
982 /* if we want mirror memory skip non-mirror memory regions */
983 if ((flags & MEMBLOCK_MIRROR) && !memblock_is_mirror(m))
984 return true;
985
986 /* skip nomap memory unless we were asked for it explicitly */
987 if (!(flags & MEMBLOCK_NOMAP) && memblock_is_nomap(m))
988 return true;
989
990 return false;
991}
992
Robin Holt8e7a7f82015-06-30 14:56:41 -0700993/**
Mike Rapoporta2974132019-03-11 23:30:54 -0700994 * __next_mem_range - next function for for_each_free_mem_range() etc.
Tejun Heo35fd0802011-07-12 11:15:59 +0200995 * @idx: pointer to u64 loop variable
Grygorii Strashkob1154232014-01-21 15:50:16 -0800996 * @nid: node selector, %NUMA_NO_NODE for all nodes
Tony Luckfc6daaf2015-06-24 16:58:09 -0700997 * @flags: pick from blocks based on memory attributes
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +0100998 * @type_a: pointer to memblock_type from where the range is taken
999 * @type_b: pointer to memblock_type which excludes memory from being taken
Wanpeng Lidad75572012-06-20 12:53:01 -07001000 * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL
1001 * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL
1002 * @out_nid: ptr to int for nid of the range, can be %NULL
Tejun Heo35fd0802011-07-12 11:15:59 +02001003 *
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +01001004 * Find the first area from *@idx which matches @nid, fill the out
Tejun Heo35fd0802011-07-12 11:15:59 +02001005 * parameters, and update *@idx for the next iteration. The lower 32bit of
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +01001006 * *@idx contains index into type_a and the upper 32bit indexes the
1007 * areas before each region in type_b. For example, if type_b regions
Tejun Heo35fd0802011-07-12 11:15:59 +02001008 * look like the following,
1009 *
1010 * 0:[0-16), 1:[32-48), 2:[128-130)
1011 *
1012 * The upper 32bit indexes the following regions.
1013 *
1014 * 0:[0-0), 1:[16-32), 2:[48-128), 3:[130-MAX)
1015 *
1016 * As both region arrays are sorted, the function advances the two indices
1017 * in lockstep and returns each intersection.
1018 */
David Hildenbrand77649902020-07-01 16:18:29 +02001019void __next_mem_range(u64 *idx, int nid, enum memblock_flags flags,
1020 struct memblock_type *type_a,
1021 struct memblock_type *type_b, phys_addr_t *out_start,
1022 phys_addr_t *out_end, int *out_nid)
Tejun Heo35fd0802011-07-12 11:15:59 +02001023{
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +01001024 int idx_a = *idx & 0xffffffff;
1025 int idx_b = *idx >> 32;
Grygorii Strashkob1154232014-01-21 15:50:16 -08001026
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +01001027 if (WARN_ONCE(nid == MAX_NUMNODES,
1028 "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
Grygorii Strashko560dca272014-01-21 15:50:55 -08001029 nid = NUMA_NO_NODE;
Tejun Heo35fd0802011-07-12 11:15:59 +02001030
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +01001031 for (; idx_a < type_a->cnt; idx_a++) {
1032 struct memblock_region *m = &type_a->regions[idx_a];
1033
Tejun Heo35fd0802011-07-12 11:15:59 +02001034 phys_addr_t m_start = m->base;
1035 phys_addr_t m_end = m->base + m->size;
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +01001036 int m_nid = memblock_get_region_node(m);
Tejun Heo35fd0802011-07-12 11:15:59 +02001037
Mike Rapoport9f3d5ea2020-10-13 16:58:25 -07001038 if (should_skip_region(type_a, m, nid, flags))
Ard Biesheuvelbf3d3cc2015-11-30 13:28:15 +01001039 continue;
1040
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +01001041 if (!type_b) {
1042 if (out_start)
1043 *out_start = m_start;
1044 if (out_end)
1045 *out_end = m_end;
1046 if (out_nid)
1047 *out_nid = m_nid;
1048 idx_a++;
1049 *idx = (u32)idx_a | (u64)idx_b << 32;
1050 return;
1051 }
Tejun Heo35fd0802011-07-12 11:15:59 +02001052
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +01001053 /* scan areas before each reservation */
1054 for (; idx_b < type_b->cnt + 1; idx_b++) {
1055 struct memblock_region *r;
1056 phys_addr_t r_start;
1057 phys_addr_t r_end;
1058
1059 r = &type_b->regions[idx_b];
1060 r_start = idx_b ? r[-1].base + r[-1].size : 0;
1061 r_end = idx_b < type_b->cnt ?
Stefan Agner1c4bc432018-06-07 17:06:15 -07001062 r->base : PHYS_ADDR_MAX;
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +01001063
1064 /*
1065 * if idx_b advanced past idx_a,
1066 * break out to advance idx_a
1067 */
Tejun Heo35fd0802011-07-12 11:15:59 +02001068 if (r_start >= m_end)
1069 break;
1070 /* if the two regions intersect, we're done */
1071 if (m_start < r_end) {
1072 if (out_start)
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +01001073 *out_start =
1074 max(m_start, r_start);
Tejun Heo35fd0802011-07-12 11:15:59 +02001075 if (out_end)
1076 *out_end = min(m_end, r_end);
1077 if (out_nid)
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +01001078 *out_nid = m_nid;
Tejun Heo35fd0802011-07-12 11:15:59 +02001079 /*
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +01001080 * The region which ends first is
1081 * advanced for the next iteration.
Tejun Heo35fd0802011-07-12 11:15:59 +02001082 */
1083 if (m_end <= r_end)
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +01001084 idx_a++;
Tejun Heo35fd0802011-07-12 11:15:59 +02001085 else
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +01001086 idx_b++;
1087 *idx = (u32)idx_a | (u64)idx_b << 32;
Tejun Heo35fd0802011-07-12 11:15:59 +02001088 return;
1089 }
1090 }
1091 }
1092
1093 /* signal end of iteration */
1094 *idx = ULLONG_MAX;
1095}
1096
Tejun Heo7bd0b0f2011-12-08 10:22:09 -08001097/**
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +01001098 * __next_mem_range_rev - generic next function for for_each_*_range_rev()
1099 *
Tejun Heo7bd0b0f2011-12-08 10:22:09 -08001100 * @idx: pointer to u64 loop variable
Alexander Kuleshovad5ea8c2015-09-08 15:04:22 -07001101 * @nid: node selector, %NUMA_NO_NODE for all nodes
Tony Luckfc6daaf2015-06-24 16:58:09 -07001102 * @flags: pick from blocks based on memory attributes
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +01001103 * @type_a: pointer to memblock_type from where the range is taken
1104 * @type_b: pointer to memblock_type which excludes memory from being taken
Wanpeng Lidad75572012-06-20 12:53:01 -07001105 * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL
1106 * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL
1107 * @out_nid: ptr to int for nid of the range, can be %NULL
Tejun Heo7bd0b0f2011-12-08 10:22:09 -08001108 *
Mike Rapoport47cec442018-06-30 17:55:02 +03001109 * Finds the next range from type_a which is not marked as unsuitable
1110 * in type_b.
1111 *
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +01001112 * Reverse of __next_mem_range().
Tejun Heo7bd0b0f2011-12-08 10:22:09 -08001113 */
Mike Rapoporte1720fe2018-06-30 17:55:01 +03001114void __init_memblock __next_mem_range_rev(u64 *idx, int nid,
1115 enum memblock_flags flags,
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +01001116 struct memblock_type *type_a,
1117 struct memblock_type *type_b,
1118 phys_addr_t *out_start,
1119 phys_addr_t *out_end, int *out_nid)
Tejun Heo7bd0b0f2011-12-08 10:22:09 -08001120{
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +01001121 int idx_a = *idx & 0xffffffff;
1122 int idx_b = *idx >> 32;
Grygorii Strashkob1154232014-01-21 15:50:16 -08001123
Grygorii Strashko560dca272014-01-21 15:50:55 -08001124 if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
1125 nid = NUMA_NO_NODE;
Tejun Heo7bd0b0f2011-12-08 10:22:09 -08001126
1127 if (*idx == (u64)ULLONG_MAX) {
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +01001128 idx_a = type_a->cnt - 1;
zijun_hue47608a2016-08-04 15:32:00 -07001129 if (type_b != NULL)
1130 idx_b = type_b->cnt;
1131 else
1132 idx_b = 0;
Tejun Heo7bd0b0f2011-12-08 10:22:09 -08001133 }
1134
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +01001135 for (; idx_a >= 0; idx_a--) {
1136 struct memblock_region *m = &type_a->regions[idx_a];
1137
Tejun Heo7bd0b0f2011-12-08 10:22:09 -08001138 phys_addr_t m_start = m->base;
1139 phys_addr_t m_end = m->base + m->size;
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +01001140 int m_nid = memblock_get_region_node(m);
Tejun Heo7bd0b0f2011-12-08 10:22:09 -08001141
Mike Rapoport9f3d5ea2020-10-13 16:58:25 -07001142 if (should_skip_region(type_a, m, nid, flags))
Ard Biesheuvelbf3d3cc2015-11-30 13:28:15 +01001143 continue;
1144
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +01001145 if (!type_b) {
1146 if (out_start)
1147 *out_start = m_start;
1148 if (out_end)
1149 *out_end = m_end;
1150 if (out_nid)
1151 *out_nid = m_nid;
zijun_hufb399b42016-07-28 15:48:56 -07001152 idx_a--;
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +01001153 *idx = (u32)idx_a | (u64)idx_b << 32;
1154 return;
1155 }
Tejun Heo7bd0b0f2011-12-08 10:22:09 -08001156
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +01001157 /* scan areas before each reservation */
1158 for (; idx_b >= 0; idx_b--) {
1159 struct memblock_region *r;
1160 phys_addr_t r_start;
1161 phys_addr_t r_end;
1162
1163 r = &type_b->regions[idx_b];
1164 r_start = idx_b ? r[-1].base + r[-1].size : 0;
1165 r_end = idx_b < type_b->cnt ?
Stefan Agner1c4bc432018-06-07 17:06:15 -07001166 r->base : PHYS_ADDR_MAX;
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +01001167 /*
1168 * if idx_b advanced past idx_a,
1169 * break out to advance idx_a
1170 */
1171
Tejun Heo7bd0b0f2011-12-08 10:22:09 -08001172 if (r_end <= m_start)
1173 break;
1174 /* if the two regions intersect, we're done */
1175 if (m_end > r_start) {
1176 if (out_start)
1177 *out_start = max(m_start, r_start);
1178 if (out_end)
1179 *out_end = min(m_end, r_end);
1180 if (out_nid)
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +01001181 *out_nid = m_nid;
Tejun Heo7bd0b0f2011-12-08 10:22:09 -08001182 if (m_start >= r_start)
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +01001183 idx_a--;
Tejun Heo7bd0b0f2011-12-08 10:22:09 -08001184 else
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +01001185 idx_b--;
1186 *idx = (u32)idx_a | (u64)idx_b << 32;
Tejun Heo7bd0b0f2011-12-08 10:22:09 -08001187 return;
1188 }
1189 }
1190 }
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +01001191 /* signal end of iteration */
Tejun Heo7bd0b0f2011-12-08 10:22:09 -08001192 *idx = ULLONG_MAX;
1193}
1194
Tejun Heo7c0caeb2011-07-14 11:43:42 +02001195/*
Chen Chang45e79812018-11-16 15:08:57 -08001196 * Common iterator interface used to define for_each_mem_pfn_range().
Tejun Heo7c0caeb2011-07-14 11:43:42 +02001197 */
1198void __init_memblock __next_mem_pfn_range(int *idx, int nid,
1199 unsigned long *out_start_pfn,
1200 unsigned long *out_end_pfn, int *out_nid)
1201{
1202 struct memblock_type *type = &memblock.memory;
1203 struct memblock_region *r;
Mike Rapoportd622abf2020-06-03 15:56:53 -07001204 int r_nid;
Tejun Heo7c0caeb2011-07-14 11:43:42 +02001205
1206 while (++*idx < type->cnt) {
1207 r = &type->regions[*idx];
Mike Rapoportd622abf2020-06-03 15:56:53 -07001208 r_nid = memblock_get_region_node(r);
Tejun Heo7c0caeb2011-07-14 11:43:42 +02001209
1210 if (PFN_UP(r->base) >= PFN_DOWN(r->base + r->size))
1211 continue;
Mike Rapoportd622abf2020-06-03 15:56:53 -07001212 if (nid == MAX_NUMNODES || nid == r_nid)
Tejun Heo7c0caeb2011-07-14 11:43:42 +02001213 break;
1214 }
1215 if (*idx >= type->cnt) {
1216 *idx = -1;
1217 return;
1218 }
1219
1220 if (out_start_pfn)
1221 *out_start_pfn = PFN_UP(r->base);
1222 if (out_end_pfn)
1223 *out_end_pfn = PFN_DOWN(r->base + r->size);
1224 if (out_nid)
Mike Rapoportd622abf2020-06-03 15:56:53 -07001225 *out_nid = r_nid;
Tejun Heo7c0caeb2011-07-14 11:43:42 +02001226}
1227
1228/**
1229 * memblock_set_node - set node ID on memblock regions
1230 * @base: base of area to set node ID for
1231 * @size: size of area to set node ID for
Tang Chene7e8de52014-01-21 15:49:26 -08001232 * @type: memblock type to set node ID for
Tejun Heo7c0caeb2011-07-14 11:43:42 +02001233 * @nid: node ID to set
1234 *
Mike Rapoport47cec442018-06-30 17:55:02 +03001235 * Set the nid of memblock @type regions in [@base, @base + @size) to @nid.
Tejun Heo7c0caeb2011-07-14 11:43:42 +02001236 * Regions which cross the area boundaries are split as necessary.
1237 *
Mike Rapoport47cec442018-06-30 17:55:02 +03001238 * Return:
Tejun Heo7c0caeb2011-07-14 11:43:42 +02001239 * 0 on success, -errno on failure.
1240 */
1241int __init_memblock memblock_set_node(phys_addr_t base, phys_addr_t size,
Tang Chene7e8de52014-01-21 15:49:26 -08001242 struct memblock_type *type, int nid)
Tejun Heo7c0caeb2011-07-14 11:43:42 +02001243{
Mike Rapoporta9ee6cf2021-06-28 19:43:01 -07001244#ifdef CONFIG_NUMA
Tejun Heo6a9ceb32011-12-08 10:22:07 -08001245 int start_rgn, end_rgn;
1246 int i, ret;
Tejun Heo7c0caeb2011-07-14 11:43:42 +02001247
Tejun Heo6a9ceb32011-12-08 10:22:07 -08001248 ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
1249 if (ret)
1250 return ret;
Tejun Heo7c0caeb2011-07-14 11:43:42 +02001251
Tejun Heo6a9ceb32011-12-08 10:22:07 -08001252 for (i = start_rgn; i < end_rgn; i++)
Wanpeng Lie9d24ad2012-10-08 16:32:21 -07001253 memblock_set_region_node(&type->regions[i], nid);
Tejun Heo7c0caeb2011-07-14 11:43:42 +02001254
1255 memblock_merge_regions(type);
Mike Rapoport3f08a302020-06-03 15:57:02 -07001256#endif
Tejun Heo7c0caeb2011-07-14 11:43:42 +02001257 return 0;
1258}
Mike Rapoport3f08a302020-06-03 15:57:02 -07001259
Alexander Duyck837566e2019-05-13 17:21:17 -07001260#ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1261/**
1262 * __next_mem_pfn_range_in_zone - iterator for for_each_*_range_in_zone()
1263 *
1264 * @idx: pointer to u64 loop variable
1265 * @zone: zone in which all of the memory blocks reside
1266 * @out_spfn: ptr to ulong for start pfn of the range, can be %NULL
1267 * @out_epfn: ptr to ulong for end pfn of the range, can be %NULL
1268 *
1269 * This function is meant to be a zone/pfn specific wrapper for the
1270 * for_each_mem_range type iterators. Specifically they are used in the
1271 * deferred memory init routines and as such we were duplicating much of
1272 * this logic throughout the code. So instead of having it in multiple
1273 * locations it seemed like it would make more sense to centralize this to
1274 * one new iterator that does everything they need.
1275 */
1276void __init_memblock
1277__next_mem_pfn_range_in_zone(u64 *idx, struct zone *zone,
1278 unsigned long *out_spfn, unsigned long *out_epfn)
1279{
1280 int zone_nid = zone_to_nid(zone);
1281 phys_addr_t spa, epa;
1282 int nid;
1283
1284 __next_mem_range(idx, zone_nid, MEMBLOCK_NONE,
1285 &memblock.memory, &memblock.reserved,
1286 &spa, &epa, &nid);
1287
1288 while (*idx != U64_MAX) {
1289 unsigned long epfn = PFN_DOWN(epa);
1290 unsigned long spfn = PFN_UP(spa);
1291
1292 /*
1293 * Verify the end is at least past the start of the zone and
1294 * that we have at least one PFN to initialize.
1295 */
1296 if (zone->zone_start_pfn < epfn && spfn < epfn) {
1297 /* if we went too far just stop searching */
1298 if (zone_end_pfn(zone) <= spfn) {
1299 *idx = U64_MAX;
1300 break;
1301 }
1302
1303 if (out_spfn)
1304 *out_spfn = max(zone->zone_start_pfn, spfn);
1305 if (out_epfn)
1306 *out_epfn = min(zone_end_pfn(zone), epfn);
1307
1308 return;
1309 }
1310
1311 __next_mem_range(idx, zone_nid, MEMBLOCK_NONE,
1312 &memblock.memory, &memblock.reserved,
1313 &spa, &epa, &nid);
1314 }
1315
1316 /* signal end of iteration */
1317 if (out_spfn)
1318 *out_spfn = ULONG_MAX;
1319 if (out_epfn)
1320 *out_epfn = 0;
1321}
1322
1323#endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
Tejun Heo7c0caeb2011-07-14 11:43:42 +02001324
Mike Rapoport92d12f92019-03-11 23:29:41 -07001325/**
1326 * memblock_alloc_range_nid - allocate boot memory block
1327 * @size: size of memory block to be allocated in bytes
1328 * @align: alignment of the region and block's size
1329 * @start: the lower bound of the memory region to allocate (phys address)
1330 * @end: the upper bound of the memory region to allocate (phys address)
1331 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
Yunfeng Ye0ac398b2019-11-30 17:56:27 -08001332 * @exact_nid: control the allocation fall back to other nodes
Mike Rapoport92d12f92019-03-11 23:29:41 -07001333 *
1334 * The allocation is performed from memory region limited by
Cao jin95830662019-11-30 17:56:24 -08001335 * memblock.current_limit if @end == %MEMBLOCK_ALLOC_ACCESSIBLE.
Mike Rapoport92d12f92019-03-11 23:29:41 -07001336 *
Yunfeng Ye0ac398b2019-11-30 17:56:27 -08001337 * If the specified node can not hold the requested memory and @exact_nid
1338 * is false, the allocation falls back to any node in the system.
Mike Rapoport92d12f92019-03-11 23:29:41 -07001339 *
1340 * For systems with memory mirroring, the allocation is attempted first
1341 * from the regions with mirroring enabled and then retried from any
1342 * memory region.
1343 *
1344 * In addition, function sets the min_count to 0 using kmemleak_alloc_phys for
1345 * allocated boot memory block, so that it is never reported as leaks.
1346 *
1347 * Return:
1348 * Physical address of allocated memory block on success, %0 on failure.
1349 */
Aslan Bakirov8676af12020-04-10 14:32:42 -07001350phys_addr_t __init memblock_alloc_range_nid(phys_addr_t size,
Akinobu Mita2bfc2862014-06-04 16:06:53 -07001351 phys_addr_t align, phys_addr_t start,
Yunfeng Ye0ac398b2019-11-30 17:56:27 -08001352 phys_addr_t end, int nid,
1353 bool exact_nid)
Yinghai Lu95f72d12010-07-12 14:36:09 +10001354{
Mike Rapoport92d12f92019-03-11 23:29:41 -07001355 enum memblock_flags flags = choose_memblock_flags();
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +10001356 phys_addr_t found;
Yinghai Lu95f72d12010-07-12 14:36:09 +10001357
Mike Rapoport92d12f92019-03-11 23:29:41 -07001358 if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
1359 nid = NUMA_NO_NODE;
1360
Mike Rapoport2f770802018-10-30 15:10:01 -07001361 if (!align) {
1362 /* Can't use WARNs this early in boot on powerpc */
1363 dump_stack();
1364 align = SMP_CACHE_BYTES;
1365 }
1366
Mike Rapoport92d12f92019-03-11 23:29:41 -07001367again:
Tony Luckfc6daaf2015-06-24 16:58:09 -07001368 found = memblock_find_in_range_node(size, align, start, end, nid,
1369 flags);
Mike Rapoport92d12f92019-03-11 23:29:41 -07001370 if (found && !memblock_reserve(found, size))
1371 goto done;
1372
Yunfeng Ye0ac398b2019-11-30 17:56:27 -08001373 if (nid != NUMA_NO_NODE && !exact_nid) {
Mike Rapoport92d12f92019-03-11 23:29:41 -07001374 found = memblock_find_in_range_node(size, align, start,
1375 end, NUMA_NO_NODE,
1376 flags);
1377 if (found && !memblock_reserve(found, size))
1378 goto done;
1379 }
1380
1381 if (flags & MEMBLOCK_MIRROR) {
1382 flags &= ~MEMBLOCK_MIRROR;
1383 pr_warn("Could not allocate %pap bytes of mirrored memory\n",
1384 &size);
1385 goto again;
1386 }
1387
1388 return 0;
1389
1390done:
1391 /* Skip kmemleak for kasan_init() due to high volume. */
1392 if (end != MEMBLOCK_ALLOC_KASAN)
Catalin Marinasaedf95e2014-06-06 14:38:20 -07001393 /*
Mike Rapoport92d12f92019-03-11 23:29:41 -07001394 * The min_count is set to 0 so that memblock allocated
1395 * blocks are never reported as leaks. This is because many
1396 * of these blocks are only referred via the physical
1397 * address which is not looked up by kmemleak.
Catalin Marinasaedf95e2014-06-06 14:38:20 -07001398 */
Catalin Marinas9099dae2016-10-11 13:55:11 -07001399 kmemleak_alloc_phys(found, size, 0, 0);
Mike Rapoport92d12f92019-03-11 23:29:41 -07001400
1401 return found;
Yinghai Lu95f72d12010-07-12 14:36:09 +10001402}
1403
Mike Rapoporta2974132019-03-11 23:30:54 -07001404/**
1405 * memblock_phys_alloc_range - allocate a memory block inside specified range
1406 * @size: size of memory block to be allocated in bytes
1407 * @align: alignment of the region and block's size
1408 * @start: the lower bound of the memory region to allocate (physical address)
1409 * @end: the upper bound of the memory region to allocate (physical address)
1410 *
1411 * Allocate @size bytes in the between @start and @end.
1412 *
1413 * Return: physical address of the allocated memory block on success,
1414 * %0 on failure.
1415 */
Mike Rapoport8a770c22019-03-11 23:29:16 -07001416phys_addr_t __init memblock_phys_alloc_range(phys_addr_t size,
1417 phys_addr_t align,
1418 phys_addr_t start,
1419 phys_addr_t end)
Akinobu Mita2bfc2862014-06-04 16:06:53 -07001420{
Faiyaz Mohammedb5cf2d62020-11-16 10:14:04 +05301421 memblock_dbg("%s: %llu bytes align=0x%llx from=%pa max_addr=%pa %pS\n",
1422 __func__, (u64)size, (u64)align, &start, &end,
1423 (void *)_RET_IP_);
Yunfeng Ye0ac398b2019-11-30 17:56:27 -08001424 return memblock_alloc_range_nid(size, align, start, end, NUMA_NO_NODE,
1425 false);
Tejun Heo7bd0b0f2011-12-08 10:22:09 -08001426}
1427
Mike Rapoporta2974132019-03-11 23:30:54 -07001428/**
Levi Yun17cbe032021-01-20 21:28:18 +09001429 * memblock_phys_alloc_try_nid - allocate a memory block from specified NUMA node
Mike Rapoporta2974132019-03-11 23:30:54 -07001430 * @size: size of memory block to be allocated in bytes
1431 * @align: alignment of the region and block's size
1432 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1433 *
1434 * Allocates memory block from the specified NUMA node. If the node
1435 * has no available memory, attempts to allocated from any node in the
1436 * system.
1437 *
1438 * Return: physical address of the allocated memory block on success,
1439 * %0 on failure.
1440 */
Mike Rapoport9a8dd702018-10-30 15:07:59 -07001441phys_addr_t __init memblock_phys_alloc_try_nid(phys_addr_t size, phys_addr_t align, int nid)
Benjamin Herrenschmidt9d1e2492010-07-06 15:39:17 -07001442{
Mike Rapoport33755572019-03-11 23:29:21 -07001443 return memblock_alloc_range_nid(size, align, 0,
Yunfeng Ye0ac398b2019-11-30 17:56:27 -08001444 MEMBLOCK_ALLOC_ACCESSIBLE, nid, false);
Yinghai Lu95f72d12010-07-12 14:36:09 +10001445}
1446
Santosh Shilimkar26f09e92014-01-21 15:50:19 -08001447/**
Mike Rapoporteb31d552018-10-30 15:08:04 -07001448 * memblock_alloc_internal - allocate boot memory block
Santosh Shilimkar26f09e92014-01-21 15:50:19 -08001449 * @size: size of memory block to be allocated in bytes
1450 * @align: alignment of the region and block's size
1451 * @min_addr: the lower bound of the memory region to allocate (phys address)
1452 * @max_addr: the upper bound of the memory region to allocate (phys address)
1453 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
Yunfeng Ye0ac398b2019-11-30 17:56:27 -08001454 * @exact_nid: control the allocation fall back to other nodes
Santosh Shilimkar26f09e92014-01-21 15:50:19 -08001455 *
Mike Rapoport92d12f92019-03-11 23:29:41 -07001456 * Allocates memory block using memblock_alloc_range_nid() and
1457 * converts the returned physical address to virtual.
1458 *
Santosh Shilimkar26f09e92014-01-21 15:50:19 -08001459 * The @min_addr limit is dropped if it can not be satisfied and the allocation
Mike Rapoport92d12f92019-03-11 23:29:41 -07001460 * will fall back to memory below @min_addr. Other constraints, such
1461 * as node and mirrored memory will be handled again in
1462 * memblock_alloc_range_nid().
Santosh Shilimkar26f09e92014-01-21 15:50:19 -08001463 *
Mike Rapoport47cec442018-06-30 17:55:02 +03001464 * Return:
Santosh Shilimkar26f09e92014-01-21 15:50:19 -08001465 * Virtual address of allocated memory block on success, NULL on failure.
1466 */
Mike Rapoporteb31d552018-10-30 15:08:04 -07001467static void * __init memblock_alloc_internal(
Santosh Shilimkar26f09e92014-01-21 15:50:19 -08001468 phys_addr_t size, phys_addr_t align,
1469 phys_addr_t min_addr, phys_addr_t max_addr,
Yunfeng Ye0ac398b2019-11-30 17:56:27 -08001470 int nid, bool exact_nid)
Santosh Shilimkar26f09e92014-01-21 15:50:19 -08001471{
1472 phys_addr_t alloc;
Santosh Shilimkar26f09e92014-01-21 15:50:19 -08001473
1474 /*
1475 * Detect any accidental use of these APIs after slab is ready, as at
1476 * this moment memblock may be deinitialized already and its
Mike Rapoportc6ffc5c2018-10-30 15:09:30 -07001477 * internal data may be destroyed (after execution of memblock_free_all)
Santosh Shilimkar26f09e92014-01-21 15:50:19 -08001478 */
1479 if (WARN_ON_ONCE(slab_is_available()))
1480 return kzalloc_node(size, GFP_NOWAIT, nid);
1481
Mike Rapoportf3057ad2019-10-18 20:20:01 -07001482 if (max_addr > memblock.current_limit)
1483 max_addr = memblock.current_limit;
1484
Yunfeng Ye0ac398b2019-11-30 17:56:27 -08001485 alloc = memblock_alloc_range_nid(size, align, min_addr, max_addr, nid,
1486 exact_nid);
Mike Rapoport2f770802018-10-30 15:10:01 -07001487
Mike Rapoport92d12f92019-03-11 23:29:41 -07001488 /* retry allocation without lower limit */
1489 if (!alloc && min_addr)
Yunfeng Ye0ac398b2019-11-30 17:56:27 -08001490 alloc = memblock_alloc_range_nid(size, align, 0, max_addr, nid,
1491 exact_nid);
Santosh Shilimkar26f09e92014-01-21 15:50:19 -08001492
Mike Rapoport92d12f92019-03-11 23:29:41 -07001493 if (!alloc)
1494 return NULL;
Santosh Shilimkar26f09e92014-01-21 15:50:19 -08001495
Mike Rapoport92d12f92019-03-11 23:29:41 -07001496 return phys_to_virt(alloc);
Santosh Shilimkar26f09e92014-01-21 15:50:19 -08001497}
1498
1499/**
Yunfeng Ye0ac398b2019-11-30 17:56:27 -08001500 * memblock_alloc_exact_nid_raw - allocate boot memory block on the exact node
1501 * without zeroing memory
1502 * @size: size of memory block to be allocated in bytes
1503 * @align: alignment of the region and block's size
1504 * @min_addr: the lower bound of the memory region from where the allocation
1505 * is preferred (phys address)
1506 * @max_addr: the upper bound of the memory region from where the allocation
1507 * is preferred (phys address), or %MEMBLOCK_ALLOC_ACCESSIBLE to
1508 * allocate only from memory limited by memblock.current_limit value
1509 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1510 *
1511 * Public function, provides additional debug information (including caller
1512 * info), if enabled. Does not zero allocated memory.
1513 *
1514 * Return:
1515 * Virtual address of allocated memory block on success, NULL on failure.
1516 */
1517void * __init memblock_alloc_exact_nid_raw(
1518 phys_addr_t size, phys_addr_t align,
1519 phys_addr_t min_addr, phys_addr_t max_addr,
1520 int nid)
1521{
Yunfeng Ye0ac398b2019-11-30 17:56:27 -08001522 memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=%pa max_addr=%pa %pS\n",
1523 __func__, (u64)size, (u64)align, nid, &min_addr,
1524 &max_addr, (void *)_RET_IP_);
1525
Mike Rapoport08678802021-09-02 14:58:05 -07001526 return memblock_alloc_internal(size, align, min_addr, max_addr, nid,
1527 true);
Yunfeng Ye0ac398b2019-11-30 17:56:27 -08001528}
1529
1530/**
Mike Rapoporteb31d552018-10-30 15:08:04 -07001531 * memblock_alloc_try_nid_raw - allocate boot memory block without zeroing
Pavel Tatashinea1f5f32017-11-15 17:36:27 -08001532 * memory and without panicking
1533 * @size: size of memory block to be allocated in bytes
1534 * @align: alignment of the region and block's size
1535 * @min_addr: the lower bound of the memory region from where the allocation
1536 * is preferred (phys address)
1537 * @max_addr: the upper bound of the memory region from where the allocation
Mike Rapoport97ad1082018-10-30 15:09:44 -07001538 * is preferred (phys address), or %MEMBLOCK_ALLOC_ACCESSIBLE to
Pavel Tatashinea1f5f32017-11-15 17:36:27 -08001539 * allocate only from memory limited by memblock.current_limit value
1540 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1541 *
1542 * Public function, provides additional debug information (including caller
1543 * info), if enabled. Does not zero allocated memory, does not panic if request
1544 * cannot be satisfied.
1545 *
Mike Rapoport47cec442018-06-30 17:55:02 +03001546 * Return:
Pavel Tatashinea1f5f32017-11-15 17:36:27 -08001547 * Virtual address of allocated memory block on success, NULL on failure.
1548 */
Mike Rapoporteb31d552018-10-30 15:08:04 -07001549void * __init memblock_alloc_try_nid_raw(
Pavel Tatashinea1f5f32017-11-15 17:36:27 -08001550 phys_addr_t size, phys_addr_t align,
1551 phys_addr_t min_addr, phys_addr_t max_addr,
1552 int nid)
1553{
Sakari Ailusd75f7732019-03-25 21:32:28 +02001554 memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=%pa max_addr=%pa %pS\n",
Mike Rapoporta36aab82018-08-17 15:47:17 -07001555 __func__, (u64)size, (u64)align, nid, &min_addr,
1556 &max_addr, (void *)_RET_IP_);
Pavel Tatashinea1f5f32017-11-15 17:36:27 -08001557
Mike Rapoport08678802021-09-02 14:58:05 -07001558 return memblock_alloc_internal(size, align, min_addr, max_addr, nid,
1559 false);
Pavel Tatashinea1f5f32017-11-15 17:36:27 -08001560}
1561
1562/**
Mike Rapoportc0dbe822019-03-11 23:30:37 -07001563 * memblock_alloc_try_nid - allocate boot memory block
Santosh Shilimkar26f09e92014-01-21 15:50:19 -08001564 * @size: size of memory block to be allocated in bytes
1565 * @align: alignment of the region and block's size
1566 * @min_addr: the lower bound of the memory region from where the allocation
1567 * is preferred (phys address)
1568 * @max_addr: the upper bound of the memory region from where the allocation
Mike Rapoport97ad1082018-10-30 15:09:44 -07001569 * is preferred (phys address), or %MEMBLOCK_ALLOC_ACCESSIBLE to
Santosh Shilimkar26f09e92014-01-21 15:50:19 -08001570 * allocate only from memory limited by memblock.current_limit value
1571 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1572 *
Mike Rapoportc0dbe822019-03-11 23:30:37 -07001573 * Public function, provides additional debug information (including caller
1574 * info), if enabled. This function zeroes the allocated memory.
Santosh Shilimkar26f09e92014-01-21 15:50:19 -08001575 *
Mike Rapoport47cec442018-06-30 17:55:02 +03001576 * Return:
Santosh Shilimkar26f09e92014-01-21 15:50:19 -08001577 * Virtual address of allocated memory block on success, NULL on failure.
1578 */
Mike Rapoporteb31d552018-10-30 15:08:04 -07001579void * __init memblock_alloc_try_nid(
Santosh Shilimkar26f09e92014-01-21 15:50:19 -08001580 phys_addr_t size, phys_addr_t align,
1581 phys_addr_t min_addr, phys_addr_t max_addr,
1582 int nid)
1583{
1584 void *ptr;
1585
Sakari Ailusd75f7732019-03-25 21:32:28 +02001586 memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=%pa max_addr=%pa %pS\n",
Mike Rapoporta36aab82018-08-17 15:47:17 -07001587 __func__, (u64)size, (u64)align, nid, &min_addr,
1588 &max_addr, (void *)_RET_IP_);
Mike Rapoporteb31d552018-10-30 15:08:04 -07001589 ptr = memblock_alloc_internal(size, align,
Yunfeng Ye0ac398b2019-11-30 17:56:27 -08001590 min_addr, max_addr, nid, false);
Mike Rapoportc0dbe822019-03-11 23:30:37 -07001591 if (ptr)
Pavel Tatashinea1f5f32017-11-15 17:36:27 -08001592 memset(ptr, 0, size);
Santosh Shilimkar26f09e92014-01-21 15:50:19 -08001593
Mike Rapoportc0dbe822019-03-11 23:30:37 -07001594 return ptr;
Santosh Shilimkar26f09e92014-01-21 15:50:19 -08001595}
1596
1597/**
Mike Rapoporta2974132019-03-11 23:30:54 -07001598 * __memblock_free_late - free pages directly to buddy allocator
Mike Rapoport48a833c2018-06-30 17:55:03 +03001599 * @base: phys starting address of the boot memory block
Santosh Shilimkar26f09e92014-01-21 15:50:19 -08001600 * @size: size of the boot memory block in bytes
1601 *
Mike Rapoporta2974132019-03-11 23:30:54 -07001602 * This is only useful when the memblock allocator has already been torn
Santosh Shilimkar26f09e92014-01-21 15:50:19 -08001603 * down, but we are still initializing the system. Pages are released directly
Mike Rapoporta2974132019-03-11 23:30:54 -07001604 * to the buddy allocator.
Santosh Shilimkar26f09e92014-01-21 15:50:19 -08001605 */
1606void __init __memblock_free_late(phys_addr_t base, phys_addr_t size)
1607{
Mike Rapoporta36aab82018-08-17 15:47:17 -07001608 phys_addr_t cursor, end;
Santosh Shilimkar26f09e92014-01-21 15:50:19 -08001609
Mike Rapoporta36aab82018-08-17 15:47:17 -07001610 end = base + size - 1;
Sakari Ailusd75f7732019-03-25 21:32:28 +02001611 memblock_dbg("%s: [%pa-%pa] %pS\n",
Mike Rapoporta36aab82018-08-17 15:47:17 -07001612 __func__, &base, &end, (void *)_RET_IP_);
Catalin Marinas9099dae2016-10-11 13:55:11 -07001613 kmemleak_free_part_phys(base, size);
Santosh Shilimkar26f09e92014-01-21 15:50:19 -08001614 cursor = PFN_UP(base);
1615 end = PFN_DOWN(base + size);
1616
1617 for (; cursor < end; cursor++) {
Mike Rapoport7c2ee342018-10-30 15:09:36 -07001618 memblock_free_pages(pfn_to_page(cursor), cursor, 0);
Arun KSca79b0c2018-12-28 00:34:29 -08001619 totalram_pages_inc();
Santosh Shilimkar26f09e92014-01-21 15:50:19 -08001620 }
1621}
Benjamin Herrenschmidt9d1e2492010-07-06 15:39:17 -07001622
1623/*
1624 * Remaining API functions
1625 */
1626
David Gibson1f1ffb8a2016-02-05 15:36:19 -08001627phys_addr_t __init_memblock memblock_phys_mem_size(void)
Yinghai Lu95f72d12010-07-12 14:36:09 +10001628{
Tejun Heo1440c4e2011-12-08 10:22:08 -08001629 return memblock.memory.total_size;
Yinghai Lu95f72d12010-07-12 14:36:09 +10001630}
1631
Srikar Dronamraju8907de52016-10-07 16:59:18 -07001632phys_addr_t __init_memblock memblock_reserved_size(void)
1633{
1634 return memblock.reserved.total_size;
1635}
1636
Sam Ravnborg0a93ebe2011-10-31 17:08:16 -07001637/* lowest address */
1638phys_addr_t __init_memblock memblock_start_of_DRAM(void)
1639{
1640 return memblock.memory.regions[0].base;
1641}
1642
Yinghai Lu10d06432010-07-28 15:43:02 +10001643phys_addr_t __init_memblock memblock_end_of_DRAM(void)
Yinghai Lu95f72d12010-07-12 14:36:09 +10001644{
1645 int idx = memblock.memory.cnt - 1;
1646
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +10001647 return (memblock.memory.regions[idx].base + memblock.memory.regions[idx].size);
Yinghai Lu95f72d12010-07-12 14:36:09 +10001648}
1649
Dennis Chena571d4e2016-07-28 15:48:26 -07001650static phys_addr_t __init_memblock __find_max_addr(phys_addr_t limit)
Yinghai Lu95f72d12010-07-12 14:36:09 +10001651{
Stefan Agner1c4bc432018-06-07 17:06:15 -07001652 phys_addr_t max_addr = PHYS_ADDR_MAX;
Emil Medve136199f2014-04-07 15:37:52 -07001653 struct memblock_region *r;
Yinghai Lu95f72d12010-07-12 14:36:09 +10001654
Dennis Chena571d4e2016-07-28 15:48:26 -07001655 /*
1656 * translate the memory @limit size into the max address within one of
1657 * the memory memblock regions, if the @limit exceeds the total size
Stefan Agner1c4bc432018-06-07 17:06:15 -07001658 * of those regions, max_addr will keep original value PHYS_ADDR_MAX
Dennis Chena571d4e2016-07-28 15:48:26 -07001659 */
Mike Rapoportcc6de162020-10-13 16:58:30 -07001660 for_each_mem_region(r) {
Tejun Heoc0ce8fe2011-12-08 10:22:07 -08001661 if (limit <= r->size) {
1662 max_addr = r->base + limit;
1663 break;
1664 }
1665 limit -= r->size;
Yinghai Lu95f72d12010-07-12 14:36:09 +10001666 }
1667
Dennis Chena571d4e2016-07-28 15:48:26 -07001668 return max_addr;
1669}
1670
1671void __init memblock_enforce_memory_limit(phys_addr_t limit)
1672{
Colin Ian King49aef712020-04-01 21:11:01 -07001673 phys_addr_t max_addr;
Dennis Chena571d4e2016-07-28 15:48:26 -07001674
1675 if (!limit)
1676 return;
1677
1678 max_addr = __find_max_addr(limit);
1679
1680 /* @limit exceeds the total size of the memory, do nothing */
Stefan Agner1c4bc432018-06-07 17:06:15 -07001681 if (max_addr == PHYS_ADDR_MAX)
Dennis Chena571d4e2016-07-28 15:48:26 -07001682 return;
1683
Tejun Heoc0ce8fe2011-12-08 10:22:07 -08001684 /* truncate both memory and reserved regions */
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +01001685 memblock_remove_range(&memblock.memory, max_addr,
Stefan Agner1c4bc432018-06-07 17:06:15 -07001686 PHYS_ADDR_MAX);
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +01001687 memblock_remove_range(&memblock.reserved, max_addr,
Stefan Agner1c4bc432018-06-07 17:06:15 -07001688 PHYS_ADDR_MAX);
Yinghai Lu95f72d12010-07-12 14:36:09 +10001689}
1690
AKASHI Takahiroc9ca9b42017-04-03 11:23:55 +09001691void __init memblock_cap_memory_range(phys_addr_t base, phys_addr_t size)
1692{
1693 int start_rgn, end_rgn;
1694 int i, ret;
1695
1696 if (!size)
1697 return;
1698
Peng Fan5173ed72021-10-18 15:15:45 -07001699 if (!memblock_memory->total_size) {
Geert Uytterhoevene888fa72021-08-11 10:55:18 +02001700 pr_warn("%s: No memory registered yet\n", __func__);
1701 return;
1702 }
1703
AKASHI Takahiroc9ca9b42017-04-03 11:23:55 +09001704 ret = memblock_isolate_range(&memblock.memory, base, size,
1705 &start_rgn, &end_rgn);
1706 if (ret)
1707 return;
1708
1709 /* remove all the MAP regions */
1710 for (i = memblock.memory.cnt - 1; i >= end_rgn; i--)
1711 if (!memblock_is_nomap(&memblock.memory.regions[i]))
1712 memblock_remove_region(&memblock.memory, i);
1713
1714 for (i = start_rgn - 1; i >= 0; i--)
1715 if (!memblock_is_nomap(&memblock.memory.regions[i]))
1716 memblock_remove_region(&memblock.memory, i);
1717
1718 /* truncate the reserved regions */
1719 memblock_remove_range(&memblock.reserved, 0, base);
1720 memblock_remove_range(&memblock.reserved,
Stefan Agner1c4bc432018-06-07 17:06:15 -07001721 base + size, PHYS_ADDR_MAX);
AKASHI Takahiroc9ca9b42017-04-03 11:23:55 +09001722}
1723
Dennis Chena571d4e2016-07-28 15:48:26 -07001724void __init memblock_mem_limit_remove_map(phys_addr_t limit)
1725{
Dennis Chena571d4e2016-07-28 15:48:26 -07001726 phys_addr_t max_addr;
Dennis Chena571d4e2016-07-28 15:48:26 -07001727
1728 if (!limit)
1729 return;
1730
1731 max_addr = __find_max_addr(limit);
1732
1733 /* @limit exceeds the total size of the memory, do nothing */
Stefan Agner1c4bc432018-06-07 17:06:15 -07001734 if (max_addr == PHYS_ADDR_MAX)
Dennis Chena571d4e2016-07-28 15:48:26 -07001735 return;
1736
AKASHI Takahiroc9ca9b42017-04-03 11:23:55 +09001737 memblock_cap_memory_range(0, max_addr);
Dennis Chena571d4e2016-07-28 15:48:26 -07001738}
1739
Yinghai Lucd794812010-10-11 12:34:09 -07001740static int __init_memblock memblock_search(struct memblock_type *type, phys_addr_t addr)
Benjamin Herrenschmidt72d4b0b2010-08-04 14:38:47 +10001741{
1742 unsigned int left = 0, right = type->cnt;
1743
1744 do {
1745 unsigned int mid = (right + left) / 2;
1746
1747 if (addr < type->regions[mid].base)
1748 right = mid;
1749 else if (addr >= (type->regions[mid].base +
1750 type->regions[mid].size))
1751 left = mid + 1;
1752 else
1753 return mid;
1754 } while (left < right);
1755 return -1;
1756}
1757
Yueyi Lif5a222d2018-12-14 14:17:06 -08001758bool __init_memblock memblock_is_reserved(phys_addr_t addr)
Yinghai Lu95f72d12010-07-12 14:36:09 +10001759{
Benjamin Herrenschmidt72d4b0b2010-08-04 14:38:47 +10001760 return memblock_search(&memblock.reserved, addr) != -1;
1761}
Yinghai Lu95f72d12010-07-12 14:36:09 +10001762
Yaowei Baib4ad0c72016-01-14 15:18:54 -08001763bool __init_memblock memblock_is_memory(phys_addr_t addr)
Benjamin Herrenschmidt72d4b0b2010-08-04 14:38:47 +10001764{
1765 return memblock_search(&memblock.memory, addr) != -1;
1766}
1767
Yaowei Bai937f0c22018-02-06 15:41:18 -08001768bool __init_memblock memblock_is_map_memory(phys_addr_t addr)
Ard Biesheuvelbf3d3cc2015-11-30 13:28:15 +01001769{
1770 int i = memblock_search(&memblock.memory, addr);
1771
1772 if (i == -1)
1773 return false;
1774 return !memblock_is_nomap(&memblock.memory.regions[i]);
1775}
1776
Yinghai Lue76b63f2013-09-11 14:22:17 -07001777int __init_memblock memblock_search_pfn_nid(unsigned long pfn,
1778 unsigned long *start_pfn, unsigned long *end_pfn)
1779{
1780 struct memblock_type *type = &memblock.memory;
Fabian Frederick16763232014-04-07 15:37:53 -07001781 int mid = memblock_search(type, PFN_PHYS(pfn));
Yinghai Lue76b63f2013-09-11 14:22:17 -07001782
1783 if (mid == -1)
1784 return -1;
1785
Fabian Frederickf7e2f7e2014-06-04 16:07:51 -07001786 *start_pfn = PFN_DOWN(type->regions[mid].base);
1787 *end_pfn = PFN_DOWN(type->regions[mid].base + type->regions[mid].size);
Yinghai Lue76b63f2013-09-11 14:22:17 -07001788
Mike Rapoportd622abf2020-06-03 15:56:53 -07001789 return memblock_get_region_node(&type->regions[mid]);
Yinghai Lue76b63f2013-09-11 14:22:17 -07001790}
Yinghai Lue76b63f2013-09-11 14:22:17 -07001791
Stephen Boydeab30942012-05-24 00:45:21 -07001792/**
1793 * memblock_is_region_memory - check if a region is a subset of memory
1794 * @base: base of region to check
1795 * @size: size of region to check
1796 *
Mike Rapoport47cec442018-06-30 17:55:02 +03001797 * Check if the region [@base, @base + @size) is a subset of a memory block.
Stephen Boydeab30942012-05-24 00:45:21 -07001798 *
Mike Rapoport47cec442018-06-30 17:55:02 +03001799 * Return:
Stephen Boydeab30942012-05-24 00:45:21 -07001800 * 0 if false, non-zero if true
1801 */
Yaowei Bai937f0c22018-02-06 15:41:18 -08001802bool __init_memblock memblock_is_region_memory(phys_addr_t base, phys_addr_t size)
Benjamin Herrenschmidt72d4b0b2010-08-04 14:38:47 +10001803{
Tomi Valkeinenabb65272011-01-20 14:44:20 -08001804 int idx = memblock_search(&memblock.memory, base);
Tejun Heoeb18f1b2011-12-08 10:22:07 -08001805 phys_addr_t end = base + memblock_cap_size(base, &size);
Benjamin Herrenschmidt72d4b0b2010-08-04 14:38:47 +10001806
1807 if (idx == -1)
Yaowei Bai937f0c22018-02-06 15:41:18 -08001808 return false;
Wei Yangef415ef2017-02-22 15:45:04 -08001809 return (memblock.memory.regions[idx].base +
Tejun Heoeb18f1b2011-12-08 10:22:07 -08001810 memblock.memory.regions[idx].size) >= end;
Yinghai Lu95f72d12010-07-12 14:36:09 +10001811}
1812
Stephen Boydeab30942012-05-24 00:45:21 -07001813/**
1814 * memblock_is_region_reserved - check if a region intersects reserved memory
1815 * @base: base of region to check
1816 * @size: size of region to check
1817 *
Mike Rapoport47cec442018-06-30 17:55:02 +03001818 * Check if the region [@base, @base + @size) intersects a reserved
1819 * memory block.
Stephen Boydeab30942012-05-24 00:45:21 -07001820 *
Mike Rapoport47cec442018-06-30 17:55:02 +03001821 * Return:
Tang Chenc5c5c9d2015-09-08 15:02:00 -07001822 * True if they intersect, false if not.
Stephen Boydeab30942012-05-24 00:45:21 -07001823 */
Tang Chenc5c5c9d2015-09-08 15:02:00 -07001824bool __init_memblock memblock_is_region_reserved(phys_addr_t base, phys_addr_t size)
Yinghai Lu95f72d12010-07-12 14:36:09 +10001825{
Tang Chenc5c5c9d2015-09-08 15:02:00 -07001826 return memblock_overlaps_region(&memblock.reserved, base, size);
Yinghai Lu95f72d12010-07-12 14:36:09 +10001827}
1828
Yinghai Lu6ede1fd2012-10-22 16:35:18 -07001829void __init_memblock memblock_trim_memory(phys_addr_t align)
1830{
Yinghai Lu6ede1fd2012-10-22 16:35:18 -07001831 phys_addr_t start, end, orig_start, orig_end;
Emil Medve136199f2014-04-07 15:37:52 -07001832 struct memblock_region *r;
Yinghai Lu6ede1fd2012-10-22 16:35:18 -07001833
Mike Rapoportcc6de162020-10-13 16:58:30 -07001834 for_each_mem_region(r) {
Emil Medve136199f2014-04-07 15:37:52 -07001835 orig_start = r->base;
1836 orig_end = r->base + r->size;
Yinghai Lu6ede1fd2012-10-22 16:35:18 -07001837 start = round_up(orig_start, align);
1838 end = round_down(orig_end, align);
1839
1840 if (start == orig_start && end == orig_end)
1841 continue;
1842
1843 if (start < end) {
Emil Medve136199f2014-04-07 15:37:52 -07001844 r->base = start;
1845 r->size = end - start;
Yinghai Lu6ede1fd2012-10-22 16:35:18 -07001846 } else {
Emil Medve136199f2014-04-07 15:37:52 -07001847 memblock_remove_region(&memblock.memory,
1848 r - memblock.memory.regions);
1849 r--;
Yinghai Lu6ede1fd2012-10-22 16:35:18 -07001850 }
1851 }
1852}
Benjamin Herrenschmidte63075a2010-07-06 15:39:01 -07001853
Yinghai Lu3661ca62010-09-15 13:05:29 -07001854void __init_memblock memblock_set_current_limit(phys_addr_t limit)
Benjamin Herrenschmidte63075a2010-07-06 15:39:01 -07001855{
1856 memblock.current_limit = limit;
1857}
1858
Laura Abbottfec51012014-02-27 01:23:43 +01001859phys_addr_t __init_memblock memblock_get_current_limit(void)
1860{
1861 return memblock.current_limit;
1862}
1863
Heiko Carstens0262d9c2017-02-24 14:55:59 -08001864static void __init_memblock memblock_dump(struct memblock_type *type)
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +10001865{
Miles Chen5d63f812017-02-22 15:46:42 -08001866 phys_addr_t base, end, size;
Mike Rapoporte1720fe2018-06-30 17:55:01 +03001867 enum memblock_flags flags;
Alexander Kuleshov8c9c1702016-01-14 15:20:42 -08001868 int idx;
1869 struct memblock_region *rgn;
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +10001870
Heiko Carstens0262d9c2017-02-24 14:55:59 -08001871 pr_info(" %s.cnt = 0x%lx\n", type->name, type->cnt);
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +10001872
Gioh Kim66e8b432017-11-15 17:33:42 -08001873 for_each_memblock_type(idx, type, rgn) {
Tejun Heo7c0caeb2011-07-14 11:43:42 +02001874 char nid_buf[32] = "";
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +10001875
Tejun Heo7c0caeb2011-07-14 11:43:42 +02001876 base = rgn->base;
1877 size = rgn->size;
Miles Chen5d63f812017-02-22 15:46:42 -08001878 end = base + size - 1;
Tang Chen66a20752014-01-21 15:49:20 -08001879 flags = rgn->flags;
Mike Rapoporta9ee6cf2021-06-28 19:43:01 -07001880#ifdef CONFIG_NUMA
Tejun Heo7c0caeb2011-07-14 11:43:42 +02001881 if (memblock_get_region_node(rgn) != MAX_NUMNODES)
1882 snprintf(nid_buf, sizeof(nid_buf), " on node %d",
1883 memblock_get_region_node(rgn));
1884#endif
Mike Rapoporte1720fe2018-06-30 17:55:01 +03001885 pr_info(" %s[%#x]\t[%pa-%pa], %pa bytes%s flags: %#x\n",
Heiko Carstens0262d9c2017-02-24 14:55:59 -08001886 type->name, idx, &base, &end, &size, nid_buf, flags);
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +10001887 }
1888}
1889
Mike Rapoport87c55872020-10-13 16:57:54 -07001890static void __init_memblock __memblock_dump_all(void)
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +10001891{
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +10001892 pr_info("MEMBLOCK configuration:\n");
Miles Chen5d63f812017-02-22 15:46:42 -08001893 pr_info(" memory size = %pa reserved size = %pa\n",
1894 &memblock.memory.total_size,
1895 &memblock.reserved.total_size);
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +10001896
Heiko Carstens0262d9c2017-02-24 14:55:59 -08001897 memblock_dump(&memblock.memory);
1898 memblock_dump(&memblock.reserved);
Heiko Carstens409efd42017-02-24 14:55:56 -08001899#ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
David Hildenbrand77649902020-07-01 16:18:29 +02001900 memblock_dump(&physmem);
Heiko Carstens409efd42017-02-24 14:55:56 -08001901#endif
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +10001902}
1903
Mike Rapoport87c55872020-10-13 16:57:54 -07001904void __init_memblock memblock_dump_all(void)
1905{
1906 if (memblock_debug)
1907 __memblock_dump_all();
1908}
1909
Tejun Heo1aadc052011-12-08 10:22:08 -08001910void __init memblock_allow_resize(void)
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +10001911{
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -07001912 memblock_can_resize = 1;
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +10001913}
1914
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +10001915static int __init early_memblock(char *p)
1916{
1917 if (p && strstr(p, "debug"))
1918 memblock_debug = 1;
1919 return 0;
1920}
1921early_param("memblock", early_memblock);
1922
Mike Rapoport4f5b0c12020-12-14 19:09:59 -08001923static void __init free_memmap(unsigned long start_pfn, unsigned long end_pfn)
1924{
1925 struct page *start_pg, *end_pg;
1926 phys_addr_t pg, pgend;
1927
1928 /*
1929 * Convert start_pfn/end_pfn to a struct page pointer.
1930 */
1931 start_pg = pfn_to_page(start_pfn - 1) + 1;
1932 end_pg = pfn_to_page(end_pfn - 1) + 1;
1933
1934 /*
1935 * Convert to physical addresses, and round start upwards and end
1936 * downwards.
1937 */
1938 pg = PAGE_ALIGN(__pa(start_pg));
1939 pgend = __pa(end_pg) & PAGE_MASK;
1940
1941 /*
1942 * If there are free pages between these, free the section of the
1943 * memmap array.
1944 */
1945 if (pg < pgend)
1946 memblock_free(pg, pgend - pg);
1947}
1948
1949/*
1950 * The mem_map array can get very big. Free the unused area of the memory map.
1951 */
1952static void __init free_unused_memmap(void)
1953{
1954 unsigned long start, end, prev_end = 0;
1955 int i;
1956
1957 if (!IS_ENABLED(CONFIG_HAVE_ARCH_PFN_VALID) ||
1958 IS_ENABLED(CONFIG_SPARSEMEM_VMEMMAP))
1959 return;
1960
1961 /*
1962 * This relies on each bank being in address order.
1963 * The banks are sorted previously in bootmem_init().
1964 */
1965 for_each_mem_pfn_range(i, MAX_NUMNODES, &start, &end, NULL) {
1966#ifdef CONFIG_SPARSEMEM
1967 /*
1968 * Take care not to free memmap entries that don't exist
1969 * due to SPARSEMEM sections which aren't present.
1970 */
1971 start = min(start, ALIGN(prev_end, PAGES_PER_SECTION));
Mike Rapoport4f5b0c12020-12-14 19:09:59 -08001972#endif
Mike Rapoport4f5b0c12020-12-14 19:09:59 -08001973 /*
Mike Rapoporte2a86802021-05-17 21:15:15 +03001974 * Align down here since many operations in VM subsystem
1975 * presume that there are no holes in the memory map inside
1976 * a pageblock
Mike Rapoport4f5b0c12020-12-14 19:09:59 -08001977 */
Mike Rapoporte2a86802021-05-17 21:15:15 +03001978 start = round_down(start, pageblock_nr_pages);
Mike Rapoport4f5b0c12020-12-14 19:09:59 -08001979
1980 /*
1981 * If we had a previous bank, and there is a space
1982 * between the current bank and the previous, free it.
1983 */
1984 if (prev_end && prev_end < start)
1985 free_memmap(prev_end, start);
1986
1987 /*
Mike Rapoporte2a86802021-05-17 21:15:15 +03001988 * Align up here since many operations in VM subsystem
1989 * presume that there are no holes in the memory map inside
1990 * a pageblock
Mike Rapoport4f5b0c12020-12-14 19:09:59 -08001991 */
Mike Rapoporte2a86802021-05-17 21:15:15 +03001992 prev_end = ALIGN(end, pageblock_nr_pages);
Mike Rapoport4f5b0c12020-12-14 19:09:59 -08001993 }
1994
1995#ifdef CONFIG_SPARSEMEM
Mike Rapoportf921f532021-05-17 21:31:59 +03001996 if (!IS_ALIGNED(prev_end, PAGES_PER_SECTION)) {
1997 prev_end = ALIGN(end, pageblock_nr_pages);
Mike Rapoport4f5b0c12020-12-14 19:09:59 -08001998 free_memmap(prev_end, ALIGN(prev_end, PAGES_PER_SECTION));
Mike Rapoportf921f532021-05-17 21:31:59 +03001999 }
Mike Rapoport4f5b0c12020-12-14 19:09:59 -08002000#endif
2001}
2002
Mike Rapoportbda49a82018-10-30 15:09:40 -07002003static void __init __free_pages_memory(unsigned long start, unsigned long end)
2004{
2005 int order;
2006
2007 while (start < end) {
2008 order = min(MAX_ORDER - 1UL, __ffs(start));
2009
2010 while (start + (1UL << order) > end)
2011 order--;
2012
2013 memblock_free_pages(pfn_to_page(start), start, order);
2014
2015 start += (1UL << order);
2016 }
2017}
2018
2019static unsigned long __init __free_memory_core(phys_addr_t start,
2020 phys_addr_t end)
2021{
2022 unsigned long start_pfn = PFN_UP(start);
2023 unsigned long end_pfn = min_t(unsigned long,
2024 PFN_DOWN(end), max_low_pfn);
2025
2026 if (start_pfn >= end_pfn)
2027 return 0;
2028
2029 __free_pages_memory(start_pfn, end_pfn);
2030
2031 return end_pfn - start_pfn;
2032}
2033
Mike Rapoport9092d4f2021-06-30 18:51:16 -07002034static void __init memmap_init_reserved_pages(void)
2035{
2036 struct memblock_region *region;
2037 phys_addr_t start, end;
2038 u64 i;
2039
2040 /* initialize struct pages for the reserved regions */
2041 for_each_reserved_mem_range(i, &start, &end)
2042 reserve_bootmem_region(start, end);
2043
2044 /* and also treat struct pages for the NOMAP regions as PageReserved */
2045 for_each_mem_region(region) {
2046 if (memblock_is_nomap(region)) {
2047 start = region->base;
2048 end = start + region->size;
2049 reserve_bootmem_region(start, end);
2050 }
2051 }
2052}
2053
Mike Rapoportbda49a82018-10-30 15:09:40 -07002054static unsigned long __init free_low_memory_core_early(void)
2055{
2056 unsigned long count = 0;
2057 phys_addr_t start, end;
2058 u64 i;
2059
2060 memblock_clear_hotplug(0, -1);
2061
Mike Rapoport9092d4f2021-06-30 18:51:16 -07002062 memmap_init_reserved_pages();
Mike Rapoportbda49a82018-10-30 15:09:40 -07002063
2064 /*
2065 * We need to use NUMA_NO_NODE instead of NODE_DATA(0)->node_id
2066 * because in some case like Node0 doesn't have RAM installed
2067 * low ram will be on Node1
2068 */
2069 for_each_free_mem_range(i, NUMA_NO_NODE, MEMBLOCK_NONE, &start, &end,
2070 NULL)
2071 count += __free_memory_core(start, end);
2072
2073 return count;
2074}
2075
2076static int reset_managed_pages_done __initdata;
2077
2078void reset_node_managed_pages(pg_data_t *pgdat)
2079{
2080 struct zone *z;
2081
2082 for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++)
Arun KS9705bea2018-12-28 00:34:24 -08002083 atomic_long_set(&z->managed_pages, 0);
Mike Rapoportbda49a82018-10-30 15:09:40 -07002084}
2085
2086void __init reset_all_zones_managed_pages(void)
2087{
2088 struct pglist_data *pgdat;
2089
2090 if (reset_managed_pages_done)
2091 return;
2092
2093 for_each_online_pgdat(pgdat)
2094 reset_node_managed_pages(pgdat);
2095
2096 reset_managed_pages_done = 1;
2097}
2098
2099/**
2100 * memblock_free_all - release free pages to the buddy allocator
Mike Rapoportbda49a82018-10-30 15:09:40 -07002101 */
Daeseok Youn097d43d2021-01-14 16:08:17 +09002102void __init memblock_free_all(void)
Mike Rapoportbda49a82018-10-30 15:09:40 -07002103{
2104 unsigned long pages;
2105
Mike Rapoport4f5b0c12020-12-14 19:09:59 -08002106 free_unused_memmap();
Mike Rapoportbda49a82018-10-30 15:09:40 -07002107 reset_all_zones_managed_pages();
2108
2109 pages = free_low_memory_core_early();
Arun KSca79b0c2018-12-28 00:34:29 -08002110 totalram_pages_add(pages);
Mike Rapoportbda49a82018-10-30 15:09:40 -07002111}
2112
Mike Rapoport350e88b2019-05-13 17:22:59 -07002113#if defined(CONFIG_DEBUG_FS) && defined(CONFIG_ARCH_KEEP_MEMBLOCK)
Benjamin Herrenschmidt6d03b882010-07-06 15:39:19 -07002114
2115static int memblock_debug_show(struct seq_file *m, void *private)
2116{
2117 struct memblock_type *type = m->private;
2118 struct memblock_region *reg;
2119 int i;
Miles Chen5d63f812017-02-22 15:46:42 -08002120 phys_addr_t end;
Benjamin Herrenschmidt6d03b882010-07-06 15:39:19 -07002121
2122 for (i = 0; i < type->cnt; i++) {
2123 reg = &type->regions[i];
Miles Chen5d63f812017-02-22 15:46:42 -08002124 end = reg->base + reg->size - 1;
Benjamin Herrenschmidt6d03b882010-07-06 15:39:19 -07002125
Miles Chen5d63f812017-02-22 15:46:42 -08002126 seq_printf(m, "%4d: ", i);
2127 seq_printf(m, "%pa..%pa\n", &reg->base, &end);
Benjamin Herrenschmidt6d03b882010-07-06 15:39:19 -07002128 }
2129 return 0;
2130}
Andy Shevchenko5ad35092018-04-05 16:23:16 -07002131DEFINE_SHOW_ATTRIBUTE(memblock_debug);
Benjamin Herrenschmidt6d03b882010-07-06 15:39:19 -07002132
2133static int __init memblock_init_debugfs(void)
2134{
2135 struct dentry *root = debugfs_create_dir("memblock", NULL);
Greg Kroah-Hartmand9f79792019-03-05 15:46:09 -08002136
Joe Perches0825a6f2018-06-14 15:27:58 -07002137 debugfs_create_file("memory", 0444, root,
2138 &memblock.memory, &memblock_debug_fops);
2139 debugfs_create_file("reserved", 0444, root,
2140 &memblock.reserved, &memblock_debug_fops);
Philipp Hachtmann70210ed2014-01-29 18:16:01 +01002141#ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
David Hildenbrand77649902020-07-01 16:18:29 +02002142 debugfs_create_file("physmem", 0444, root, &physmem,
2143 &memblock_debug_fops);
Philipp Hachtmann70210ed2014-01-29 18:16:01 +01002144#endif
Benjamin Herrenschmidt6d03b882010-07-06 15:39:19 -07002145
2146 return 0;
2147}
2148__initcall(memblock_init_debugfs);
2149
2150#endif /* CONFIG_DEBUG_FS */