blob: e6b4654f9dfda2dd25b6daa6c2ffa753190e5f06 [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);
369 __memblock_free_late(addr, size);
370 }
Yinghai Lu29f67382012-07-11 14:02:56 -0700371
Pavel Tatashin91b540f92017-08-25 15:55:46 -0700372 if (memblock.memory.regions != memblock_memory_init_regions) {
Pavel Tatashin3010f872017-08-18 15:16:05 -0700373 addr = __pa(memblock.memory.regions);
374 size = PAGE_ALIGN(sizeof(struct memblock_region) *
375 memblock.memory.max);
376 __memblock_free_late(addr, size);
377 }
Mike Rapoport9f3d5ea2020-10-13 16:58:25 -0700378
379 memblock_memory = NULL;
Yinghai Lu29f67382012-07-11 14:02:56 -0700380}
Philipp Hachtmann5e270e22014-01-23 15:53:11 -0800381#endif
382
Greg Pearson48c3b582012-06-20 12:53:05 -0700383/**
384 * memblock_double_array - double the size of the memblock regions array
385 * @type: memblock type of the regions array being doubled
386 * @new_area_start: starting address of memory range to avoid overlap with
387 * @new_area_size: size of memory range to avoid overlap with
388 *
389 * Double the size of the @type regions array. If memblock is being used to
390 * allocate memory for a new reserved regions array and there is a previously
Mike Rapoport47cec442018-06-30 17:55:02 +0300391 * allocated memory range [@new_area_start, @new_area_start + @new_area_size]
Greg Pearson48c3b582012-06-20 12:53:05 -0700392 * waiting to be reserved, ensure the memory used by the new array does
393 * not overlap.
394 *
Mike Rapoport47cec442018-06-30 17:55:02 +0300395 * Return:
Greg Pearson48c3b582012-06-20 12:53:05 -0700396 * 0 on success, -1 on failure.
397 */
398static int __init_memblock memblock_double_array(struct memblock_type *type,
399 phys_addr_t new_area_start,
400 phys_addr_t new_area_size)
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700401{
402 struct memblock_region *new_array, *old_array;
Yinghai Lu29f67382012-07-11 14:02:56 -0700403 phys_addr_t old_alloc_size, new_alloc_size;
Mike Rapoporta36aab82018-08-17 15:47:17 -0700404 phys_addr_t old_size, new_size, addr, new_end;
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700405 int use_slab = slab_is_available();
Gavin Shan181eb392012-05-29 15:06:50 -0700406 int *in_slab;
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700407
408 /* We don't allow resizing until we know about the reserved regions
409 * of memory that aren't suitable for allocation
410 */
411 if (!memblock_can_resize)
412 return -1;
413
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700414 /* Calculate new doubled size */
415 old_size = type->max * sizeof(struct memblock_region);
416 new_size = old_size << 1;
Yinghai Lu29f67382012-07-11 14:02:56 -0700417 /*
418 * We need to allocated new one align to PAGE_SIZE,
419 * so we can free them completely later.
420 */
421 old_alloc_size = PAGE_ALIGN(old_size);
422 new_alloc_size = PAGE_ALIGN(new_size);
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700423
Gavin Shan181eb392012-05-29 15:06:50 -0700424 /* Retrieve the slab flag */
425 if (type == &memblock.memory)
426 in_slab = &memblock_memory_in_slab;
427 else
428 in_slab = &memblock_reserved_in_slab;
429
Mike Rapoporta2974132019-03-11 23:30:54 -0700430 /* Try to find some space for it */
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700431 if (use_slab) {
432 new_array = kmalloc(new_size, GFP_KERNEL);
Tejun Heo1f5026a2011-07-12 09:58:09 +0200433 addr = new_array ? __pa(new_array) : 0;
Gavin Shan4e2f0772012-05-29 15:06:50 -0700434 } else {
Greg Pearson48c3b582012-06-20 12:53:05 -0700435 /* only exclude range when trying to double reserved.regions */
436 if (type != &memblock.reserved)
437 new_area_start = new_area_size = 0;
438
439 addr = memblock_find_in_range(new_area_start + new_area_size,
440 memblock.current_limit,
Yinghai Lu29f67382012-07-11 14:02:56 -0700441 new_alloc_size, PAGE_SIZE);
Greg Pearson48c3b582012-06-20 12:53:05 -0700442 if (!addr && new_area_size)
443 addr = memblock_find_in_range(0,
Andrew Mortonfd073832012-07-31 16:42:40 -0700444 min(new_area_start, memblock.current_limit),
445 new_alloc_size, PAGE_SIZE);
Greg Pearson48c3b582012-06-20 12:53:05 -0700446
Sachin Kamat15674862012-09-04 13:55:05 +0530447 new_array = addr ? __va(addr) : NULL;
Gavin Shan4e2f0772012-05-29 15:06:50 -0700448 }
Tejun Heo1f5026a2011-07-12 09:58:09 +0200449 if (!addr) {
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700450 pr_err("memblock: Failed to double %s array from %ld to %ld entries !\n",
Heiko Carstens0262d9c2017-02-24 14:55:59 -0800451 type->name, type->max, type->max * 2);
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700452 return -1;
453 }
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700454
Mike Rapoporta36aab82018-08-17 15:47:17 -0700455 new_end = addr + new_size - 1;
456 memblock_dbg("memblock: %s is doubled to %ld at [%pa-%pa]",
457 type->name, type->max * 2, &addr, &new_end);
Yinghai Luea9e4372010-07-28 15:13:22 +1000458
Andrew Mortonfd073832012-07-31 16:42:40 -0700459 /*
460 * Found space, we now need to move the array over before we add the
461 * reserved region since it may be our reserved array itself that is
462 * full.
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700463 */
464 memcpy(new_array, type->regions, old_size);
465 memset(new_array + type->max, 0, old_size);
466 old_array = type->regions;
467 type->regions = new_array;
468 type->max <<= 1;
469
Andrew Mortonfd073832012-07-31 16:42:40 -0700470 /* Free old array. We needn't free it if the array is the static one */
Gavin Shan181eb392012-05-29 15:06:50 -0700471 if (*in_slab)
472 kfree(old_array);
473 else if (old_array != memblock_memory_init_regions &&
474 old_array != memblock_reserved_init_regions)
Yinghai Lu29f67382012-07-11 14:02:56 -0700475 memblock_free(__pa(old_array), old_alloc_size);
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700476
Andrew Mortonfd073832012-07-31 16:42:40 -0700477 /*
478 * Reserve the new array if that comes from the memblock. Otherwise, we
479 * needn't do it
Gavin Shan181eb392012-05-29 15:06:50 -0700480 */
481 if (!use_slab)
Yinghai Lu29f67382012-07-11 14:02:56 -0700482 BUG_ON(memblock_reserve(addr, new_alloc_size));
Gavin Shan181eb392012-05-29 15:06:50 -0700483
484 /* Update slab flag */
485 *in_slab = use_slab;
486
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700487 return 0;
488}
489
Tejun Heo784656f92011-07-12 11:15:55 +0200490/**
491 * memblock_merge_regions - merge neighboring compatible regions
492 * @type: memblock type to scan
493 *
494 * Scan @type and merge neighboring compatible regions.
495 */
496static void __init_memblock memblock_merge_regions(struct memblock_type *type)
497{
498 int i = 0;
499
500 /* cnt never goes below 1 */
501 while (i < type->cnt - 1) {
502 struct memblock_region *this = &type->regions[i];
503 struct memblock_region *next = &type->regions[i + 1];
504
Tejun Heo7c0caeb2011-07-14 11:43:42 +0200505 if (this->base + this->size != next->base ||
506 memblock_get_region_node(this) !=
Tang Chen66a20752014-01-21 15:49:20 -0800507 memblock_get_region_node(next) ||
508 this->flags != next->flags) {
Tejun Heo784656f92011-07-12 11:15:55 +0200509 BUG_ON(this->base + this->size > next->base);
510 i++;
511 continue;
512 }
513
514 this->size += next->size;
Lin Fengc0232ae2013-01-11 14:31:44 -0800515 /* move forward from next + 1, index of which is i + 2 */
516 memmove(next, next + 1, (type->cnt - (i + 2)) * sizeof(*next));
Tejun Heo784656f92011-07-12 11:15:55 +0200517 type->cnt--;
518 }
519}
520
521/**
522 * memblock_insert_region - insert new memblock region
Tang Chen209ff862013-04-29 15:08:41 -0700523 * @type: memblock type to insert into
524 * @idx: index for the insertion point
525 * @base: base address of the new region
526 * @size: size of the new region
527 * @nid: node id of the new region
Tang Chen66a20752014-01-21 15:49:20 -0800528 * @flags: flags of the new region
Tejun Heo784656f92011-07-12 11:15:55 +0200529 *
Mike Rapoport47cec442018-06-30 17:55:02 +0300530 * Insert new memblock region [@base, @base + @size) into @type at @idx.
Alexander Kuleshov412d0002016-08-04 15:31:46 -0700531 * @type must already have extra room to accommodate the new region.
Tejun Heo784656f92011-07-12 11:15:55 +0200532 */
533static void __init_memblock memblock_insert_region(struct memblock_type *type,
534 int idx, phys_addr_t base,
Tang Chen66a20752014-01-21 15:49:20 -0800535 phys_addr_t size,
Mike Rapoporte1720fe2018-06-30 17:55:01 +0300536 int nid,
537 enum memblock_flags flags)
Tejun Heo784656f92011-07-12 11:15:55 +0200538{
539 struct memblock_region *rgn = &type->regions[idx];
540
541 BUG_ON(type->cnt >= type->max);
542 memmove(rgn + 1, rgn, (type->cnt - idx) * sizeof(*rgn));
543 rgn->base = base;
544 rgn->size = size;
Tang Chen66a20752014-01-21 15:49:20 -0800545 rgn->flags = flags;
Tejun Heo7c0caeb2011-07-14 11:43:42 +0200546 memblock_set_region_node(rgn, nid);
Tejun Heo784656f92011-07-12 11:15:55 +0200547 type->cnt++;
Tejun Heo1440c4e2011-12-08 10:22:08 -0800548 type->total_size += size;
Tejun Heo784656f92011-07-12 11:15:55 +0200549}
550
551/**
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +0100552 * memblock_add_range - add new memblock region
Tejun Heo784656f92011-07-12 11:15:55 +0200553 * @type: memblock type to add new region into
554 * @base: base address of the new region
555 * @size: size of the new region
Tejun Heo7fb0bc32011-12-08 10:22:08 -0800556 * @nid: nid of the new region
Tang Chen66a20752014-01-21 15:49:20 -0800557 * @flags: flags of the new region
Tejun Heo784656f92011-07-12 11:15:55 +0200558 *
Mike Rapoport47cec442018-06-30 17:55:02 +0300559 * Add new memblock region [@base, @base + @size) into @type. The new region
Tejun Heo784656f92011-07-12 11:15:55 +0200560 * is allowed to overlap with existing ones - overlaps don't affect already
561 * existing regions. @type is guaranteed to be minimal (all neighbouring
562 * compatible regions are merged) after the addition.
563 *
Mike Rapoport47cec442018-06-30 17:55:02 +0300564 * Return:
Tejun Heo784656f92011-07-12 11:15:55 +0200565 * 0 on success, -errno on failure.
566 */
Anshuman Khandual02634a42020-01-30 22:14:20 -0800567static int __init_memblock memblock_add_range(struct memblock_type *type,
Tang Chen66a20752014-01-21 15:49:20 -0800568 phys_addr_t base, phys_addr_t size,
Mike Rapoporte1720fe2018-06-30 17:55:01 +0300569 int nid, enum memblock_flags flags)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000570{
Tejun Heo784656f92011-07-12 11:15:55 +0200571 bool insert = false;
Tejun Heoeb18f1b2011-12-08 10:22:07 -0800572 phys_addr_t obase = base;
573 phys_addr_t end = base + memblock_cap_size(base, &size);
Alexander Kuleshov8c9c1702016-01-14 15:20:42 -0800574 int idx, nr_new;
575 struct memblock_region *rgn;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000576
Tejun Heob3dc6272012-04-20 08:31:34 -0700577 if (!size)
578 return 0;
579
Tejun Heo784656f92011-07-12 11:15:55 +0200580 /* special case for empty array */
581 if (type->regions[0].size == 0) {
Tejun Heo1440c4e2011-12-08 10:22:08 -0800582 WARN_ON(type->cnt != 1 || type->total_size);
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000583 type->regions[0].base = base;
584 type->regions[0].size = size;
Tang Chen66a20752014-01-21 15:49:20 -0800585 type->regions[0].flags = flags;
Tejun Heo7fb0bc32011-12-08 10:22:08 -0800586 memblock_set_region_node(&type->regions[0], nid);
Tejun Heo1440c4e2011-12-08 10:22:08 -0800587 type->total_size = size;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000588 return 0;
589 }
Tejun Heo784656f92011-07-12 11:15:55 +0200590repeat:
591 /*
592 * The following is executed twice. Once with %false @insert and
593 * then with %true. The first counts the number of regions needed
Alexander Kuleshov412d0002016-08-04 15:31:46 -0700594 * to accommodate the new area. The second actually inserts them.
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700595 */
Tejun Heo784656f92011-07-12 11:15:55 +0200596 base = obase;
597 nr_new = 0;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000598
Gioh Kim66e8b432017-11-15 17:33:42 -0800599 for_each_memblock_type(idx, type, rgn) {
Tejun Heo784656f92011-07-12 11:15:55 +0200600 phys_addr_t rbase = rgn->base;
601 phys_addr_t rend = rbase + rgn->size;
602
603 if (rbase >= end)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000604 break;
Tejun Heo784656f92011-07-12 11:15:55 +0200605 if (rend <= base)
606 continue;
607 /*
608 * @rgn overlaps. If it separates the lower part of new
609 * area, insert that portion.
610 */
611 if (rbase > base) {
Mike Rapoporta9ee6cf2021-06-28 19:43:01 -0700612#ifdef CONFIG_NUMA
Wei Yangc0a29492015-09-04 15:47:38 -0700613 WARN_ON(nid != memblock_get_region_node(rgn));
614#endif
Wei Yang4fcab5f2015-09-08 14:59:53 -0700615 WARN_ON(flags != rgn->flags);
Tejun Heo784656f92011-07-12 11:15:55 +0200616 nr_new++;
617 if (insert)
Alexander Kuleshov8c9c1702016-01-14 15:20:42 -0800618 memblock_insert_region(type, idx++, base,
Tang Chen66a20752014-01-21 15:49:20 -0800619 rbase - base, nid,
620 flags);
Yinghai Lu95f72d12010-07-12 14:36:09 +1000621 }
Tejun Heo784656f92011-07-12 11:15:55 +0200622 /* area below @rend is dealt with, forget about it */
623 base = min(rend, end);
Yinghai Lu95f72d12010-07-12 14:36:09 +1000624 }
Yinghai Lu95f72d12010-07-12 14:36:09 +1000625
Tejun Heo784656f92011-07-12 11:15:55 +0200626 /* insert the remaining portion */
627 if (base < end) {
628 nr_new++;
629 if (insert)
Alexander Kuleshov8c9c1702016-01-14 15:20:42 -0800630 memblock_insert_region(type, idx, base, end - base,
Tang Chen66a20752014-01-21 15:49:20 -0800631 nid, flags);
Tejun Heo784656f92011-07-12 11:15:55 +0200632 }
633
nimisoloef3cc4d2016-07-26 15:24:56 -0700634 if (!nr_new)
635 return 0;
636
Tejun Heo784656f92011-07-12 11:15:55 +0200637 /*
638 * If this was the first round, resize array and repeat for actual
639 * insertions; otherwise, merge and return.
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700640 */
Tejun Heo784656f92011-07-12 11:15:55 +0200641 if (!insert) {
642 while (type->cnt + nr_new > type->max)
Greg Pearson48c3b582012-06-20 12:53:05 -0700643 if (memblock_double_array(type, obase, size) < 0)
Tejun Heo784656f92011-07-12 11:15:55 +0200644 return -ENOMEM;
645 insert = true;
646 goto repeat;
647 } else {
648 memblock_merge_regions(type);
649 return 0;
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700650 }
Yinghai Lu95f72d12010-07-12 14:36:09 +1000651}
652
Mike Rapoport48a833c2018-06-30 17:55:03 +0300653/**
654 * memblock_add_node - add new memblock region within a NUMA node
655 * @base: base address of the new region
656 * @size: size of the new region
657 * @nid: nid of the new region
658 *
659 * Add new memblock region [@base, @base + @size) to the "memory"
660 * type. See memblock_add_range() description for mode details
661 *
662 * Return:
663 * 0 on success, -errno on failure.
664 */
Tejun Heo7fb0bc32011-12-08 10:22:08 -0800665int __init_memblock memblock_add_node(phys_addr_t base, phys_addr_t size,
666 int nid)
667{
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +0100668 return memblock_add_range(&memblock.memory, base, size, nid, 0);
Tejun Heo7fb0bc32011-12-08 10:22:08 -0800669}
670
Mike Rapoport48a833c2018-06-30 17:55:03 +0300671/**
672 * memblock_add - add new memblock region
673 * @base: base address of the new region
674 * @size: size of the new region
675 *
676 * Add new memblock region [@base, @base + @size) to the "memory"
677 * type. See memblock_add_range() description for mode details
678 *
679 * Return:
680 * 0 on success, -errno on failure.
681 */
Alexander Kuleshovf705ac42016-05-20 16:57:35 -0700682int __init_memblock memblock_add(phys_addr_t base, phys_addr_t size)
Alexander Kuleshov6a4055b2015-04-15 16:14:44 -0700683{
Miles Chen5d63f812017-02-22 15:46:42 -0800684 phys_addr_t end = base + size - 1;
685
Anshuman Khanduala090d712020-01-30 22:14:23 -0800686 memblock_dbg("%s: [%pa-%pa] %pS\n", __func__,
Miles Chen5d63f812017-02-22 15:46:42 -0800687 &base, &end, (void *)_RET_IP_);
Alexander Kuleshov6a4055b2015-04-15 16:14:44 -0700688
Alexander Kuleshovf705ac42016-05-20 16:57:35 -0700689 return memblock_add_range(&memblock.memory, base, size, MAX_NUMNODES, 0);
Yinghai Lu95f72d12010-07-12 14:36:09 +1000690}
691
Tejun Heo6a9ceb32011-12-08 10:22:07 -0800692/**
693 * memblock_isolate_range - isolate given range into disjoint memblocks
694 * @type: memblock type to isolate range for
695 * @base: base of range to isolate
696 * @size: size of range to isolate
697 * @start_rgn: out parameter for the start of isolated region
698 * @end_rgn: out parameter for the end of isolated region
699 *
700 * Walk @type and ensure that regions don't cross the boundaries defined by
Mike Rapoport47cec442018-06-30 17:55:02 +0300701 * [@base, @base + @size). Crossing regions are split at the boundaries,
Tejun Heo6a9ceb32011-12-08 10:22:07 -0800702 * which may create at most two more regions. The index of the first
703 * region inside the range is returned in *@start_rgn and end in *@end_rgn.
704 *
Mike Rapoport47cec442018-06-30 17:55:02 +0300705 * Return:
Tejun Heo6a9ceb32011-12-08 10:22:07 -0800706 * 0 on success, -errno on failure.
707 */
708static int __init_memblock memblock_isolate_range(struct memblock_type *type,
709 phys_addr_t base, phys_addr_t size,
710 int *start_rgn, int *end_rgn)
711{
Tejun Heoeb18f1b2011-12-08 10:22:07 -0800712 phys_addr_t end = base + memblock_cap_size(base, &size);
Alexander Kuleshov8c9c1702016-01-14 15:20:42 -0800713 int idx;
714 struct memblock_region *rgn;
Tejun Heo6a9ceb32011-12-08 10:22:07 -0800715
716 *start_rgn = *end_rgn = 0;
717
Tejun Heob3dc6272012-04-20 08:31:34 -0700718 if (!size)
719 return 0;
720
Tejun Heo6a9ceb32011-12-08 10:22:07 -0800721 /* we'll create at most two more regions */
722 while (type->cnt + 2 > type->max)
Greg Pearson48c3b582012-06-20 12:53:05 -0700723 if (memblock_double_array(type, base, size) < 0)
Tejun Heo6a9ceb32011-12-08 10:22:07 -0800724 return -ENOMEM;
725
Gioh Kim66e8b432017-11-15 17:33:42 -0800726 for_each_memblock_type(idx, type, rgn) {
Tejun Heo6a9ceb32011-12-08 10:22:07 -0800727 phys_addr_t rbase = rgn->base;
728 phys_addr_t rend = rbase + rgn->size;
729
730 if (rbase >= end)
731 break;
732 if (rend <= base)
733 continue;
734
735 if (rbase < base) {
736 /*
737 * @rgn intersects from below. Split and continue
738 * to process the next region - the new top half.
739 */
740 rgn->base = base;
Tejun Heo1440c4e2011-12-08 10:22:08 -0800741 rgn->size -= base - rbase;
742 type->total_size -= base - rbase;
Alexander Kuleshov8c9c1702016-01-14 15:20:42 -0800743 memblock_insert_region(type, idx, rbase, base - rbase,
Tang Chen66a20752014-01-21 15:49:20 -0800744 memblock_get_region_node(rgn),
745 rgn->flags);
Tejun Heo6a9ceb32011-12-08 10:22:07 -0800746 } else if (rend > end) {
747 /*
748 * @rgn intersects from above. Split and redo the
749 * current region - the new bottom half.
750 */
751 rgn->base = end;
Tejun Heo1440c4e2011-12-08 10:22:08 -0800752 rgn->size -= end - rbase;
753 type->total_size -= end - rbase;
Alexander Kuleshov8c9c1702016-01-14 15:20:42 -0800754 memblock_insert_region(type, idx--, rbase, end - 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 {
758 /* @rgn is fully contained, record it */
759 if (!*end_rgn)
Alexander Kuleshov8c9c1702016-01-14 15:20:42 -0800760 *start_rgn = idx;
761 *end_rgn = idx + 1;
Tejun Heo6a9ceb32011-12-08 10:22:07 -0800762 }
763 }
764
765 return 0;
766}
Tejun Heo6a9ceb32011-12-08 10:22:07 -0800767
Alexander Kuleshov35bd16a2015-11-05 18:47:00 -0800768static int __init_memblock memblock_remove_range(struct memblock_type *type,
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +0100769 phys_addr_t base, phys_addr_t size)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000770{
Tejun Heo71936182011-12-08 10:22:07 -0800771 int start_rgn, end_rgn;
772 int i, ret;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000773
Tejun Heo71936182011-12-08 10:22:07 -0800774 ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
775 if (ret)
776 return ret;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000777
Tejun Heo71936182011-12-08 10:22:07 -0800778 for (i = end_rgn - 1; i >= start_rgn; i--)
779 memblock_remove_region(type, i);
Benjamin Herrenschmidt8f7a6602011-03-22 16:33:43 -0700780 return 0;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000781}
782
Tejun Heo581adcb2011-12-08 10:22:06 -0800783int __init_memblock memblock_remove(phys_addr_t base, phys_addr_t size)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000784{
Minchan Kim25cf23d2018-06-07 17:07:35 -0700785 phys_addr_t end = base + size - 1;
786
Anshuman Khanduala090d712020-01-30 22:14:23 -0800787 memblock_dbg("%s: [%pa-%pa] %pS\n", __func__,
Minchan Kim25cf23d2018-06-07 17:07:35 -0700788 &base, &end, (void *)_RET_IP_);
789
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +0100790 return memblock_remove_range(&memblock.memory, base, size);
Yinghai Lu95f72d12010-07-12 14:36:09 +1000791}
792
Mike Rapoport4d728682018-12-28 00:35:29 -0800793/**
794 * memblock_free - free boot memory block
795 * @base: phys starting address of the boot memory block
796 * @size: size of the boot memory block in bytes
797 *
798 * Free boot memory block previously allocated by memblock_alloc_xx() API.
799 * The freeing memory will not be released to the buddy allocator.
800 */
Tejun Heo581adcb2011-12-08 10:22:06 -0800801int __init_memblock memblock_free(phys_addr_t base, phys_addr_t size)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000802{
Miles Chen5d63f812017-02-22 15:46:42 -0800803 phys_addr_t end = base + size - 1;
804
Anshuman Khanduala090d712020-01-30 22:14:23 -0800805 memblock_dbg("%s: [%pa-%pa] %pS\n", __func__,
Miles Chen5d63f812017-02-22 15:46:42 -0800806 &base, &end, (void *)_RET_IP_);
Tejun Heo24aa0782011-07-12 11:16:06 +0200807
Catalin Marinas9099dae2016-10-11 13:55:11 -0700808 kmemleak_free_part_phys(base, size);
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +0100809 return memblock_remove_range(&memblock.reserved, base, size);
Yinghai Lu95f72d12010-07-12 14:36:09 +1000810}
811
Alexander Kuleshovf705ac42016-05-20 16:57:35 -0700812int __init_memblock memblock_reserve(phys_addr_t base, phys_addr_t size)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000813{
Miles Chen5d63f812017-02-22 15:46:42 -0800814 phys_addr_t end = base + size - 1;
815
Anshuman Khanduala090d712020-01-30 22:14:23 -0800816 memblock_dbg("%s: [%pa-%pa] %pS\n", __func__,
Miles Chen5d63f812017-02-22 15:46:42 -0800817 &base, &end, (void *)_RET_IP_);
Yinghai Lu95f72d12010-07-12 14:36:09 +1000818
Alexander Kuleshovf705ac42016-05-20 16:57:35 -0700819 return memblock_add_range(&memblock.reserved, base, size, MAX_NUMNODES, 0);
Yinghai Lu95f72d12010-07-12 14:36:09 +1000820}
821
Anshuman Khandual02634a42020-01-30 22:14:20 -0800822#ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
823int __init_memblock memblock_physmem_add(phys_addr_t base, phys_addr_t size)
824{
825 phys_addr_t end = base + size - 1;
826
827 memblock_dbg("%s: [%pa-%pa] %pS\n", __func__,
828 &base, &end, (void *)_RET_IP_);
829
David Hildenbrand77649902020-07-01 16:18:29 +0200830 return memblock_add_range(&physmem, base, size, MAX_NUMNODES, 0);
Anshuman Khandual02634a42020-01-30 22:14:20 -0800831}
832#endif
833
Tejun Heo35fd0802011-07-12 11:15:59 +0200834/**
Mike Rapoport47cec442018-06-30 17:55:02 +0300835 * memblock_setclr_flag - set or clear flag for a memory region
836 * @base: base address of the region
837 * @size: size of the region
838 * @set: set or clear the flag
Haitao Shi8958b242020-12-15 20:47:26 -0800839 * @flag: the flag to update
Tang Chen66b16ed2014-01-21 15:49:23 -0800840 *
Tony Luck4308ce12014-12-12 16:54:59 -0800841 * This function isolates region [@base, @base + @size), and sets/clears flag
Tang Chen66b16ed2014-01-21 15:49:23 -0800842 *
Mike Rapoport47cec442018-06-30 17:55:02 +0300843 * Return: 0 on success, -errno on failure.
Tang Chen66b16ed2014-01-21 15:49:23 -0800844 */
Tony Luck4308ce12014-12-12 16:54:59 -0800845static int __init_memblock memblock_setclr_flag(phys_addr_t base,
846 phys_addr_t size, int set, int flag)
Tang Chen66b16ed2014-01-21 15:49:23 -0800847{
848 struct memblock_type *type = &memblock.memory;
849 int i, ret, start_rgn, end_rgn;
850
851 ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
852 if (ret)
853 return ret;
854
Mike Rapoportfe145122019-03-11 23:30:46 -0700855 for (i = start_rgn; i < end_rgn; i++) {
856 struct memblock_region *r = &type->regions[i];
857
Tony Luck4308ce12014-12-12 16:54:59 -0800858 if (set)
Mike Rapoportfe145122019-03-11 23:30:46 -0700859 r->flags |= flag;
Tony Luck4308ce12014-12-12 16:54:59 -0800860 else
Mike Rapoportfe145122019-03-11 23:30:46 -0700861 r->flags &= ~flag;
862 }
Tang Chen66b16ed2014-01-21 15:49:23 -0800863
864 memblock_merge_regions(type);
865 return 0;
866}
867
868/**
Tony Luck4308ce12014-12-12 16:54:59 -0800869 * memblock_mark_hotplug - Mark hotpluggable memory with flag MEMBLOCK_HOTPLUG.
870 * @base: the base phys addr of the region
871 * @size: the size of the region
872 *
Mike Rapoport47cec442018-06-30 17:55:02 +0300873 * Return: 0 on success, -errno on failure.
Tony Luck4308ce12014-12-12 16:54:59 -0800874 */
875int __init_memblock memblock_mark_hotplug(phys_addr_t base, phys_addr_t size)
876{
877 return memblock_setclr_flag(base, size, 1, MEMBLOCK_HOTPLUG);
878}
879
880/**
Tang Chen66b16ed2014-01-21 15:49:23 -0800881 * memblock_clear_hotplug - Clear flag MEMBLOCK_HOTPLUG for a specified region.
882 * @base: the base phys addr of the region
883 * @size: the size of the region
884 *
Mike Rapoport47cec442018-06-30 17:55:02 +0300885 * Return: 0 on success, -errno on failure.
Tang Chen66b16ed2014-01-21 15:49:23 -0800886 */
887int __init_memblock memblock_clear_hotplug(phys_addr_t base, phys_addr_t size)
888{
Tony Luck4308ce12014-12-12 16:54:59 -0800889 return memblock_setclr_flag(base, size, 0, MEMBLOCK_HOTPLUG);
Tang Chen66b16ed2014-01-21 15:49:23 -0800890}
891
892/**
Tony Lucka3f5baf2015-06-24 16:58:12 -0700893 * memblock_mark_mirror - Mark mirrored memory with flag MEMBLOCK_MIRROR.
894 * @base: the base phys addr of the region
895 * @size: the size of the region
896 *
Mike Rapoport47cec442018-06-30 17:55:02 +0300897 * Return: 0 on success, -errno on failure.
Tony Lucka3f5baf2015-06-24 16:58:12 -0700898 */
899int __init_memblock memblock_mark_mirror(phys_addr_t base, phys_addr_t size)
900{
901 system_has_some_mirror = true;
902
903 return memblock_setclr_flag(base, size, 1, MEMBLOCK_MIRROR);
904}
905
Ard Biesheuvelbf3d3cc2015-11-30 13:28:15 +0100906/**
907 * memblock_mark_nomap - Mark a memory region with flag MEMBLOCK_NOMAP.
908 * @base: the base phys addr of the region
909 * @size: the size of the region
910 *
Mike Rapoport9092d4f2021-06-30 18:51:16 -0700911 * The memory regions marked with %MEMBLOCK_NOMAP will not be added to the
912 * direct mapping of the physical memory. These regions will still be
913 * covered by the memory map. The struct page representing NOMAP memory
914 * frames in the memory map will be PageReserved()
915 *
Mike Rapoport47cec442018-06-30 17:55:02 +0300916 * Return: 0 on success, -errno on failure.
Ard Biesheuvelbf3d3cc2015-11-30 13:28:15 +0100917 */
918int __init_memblock memblock_mark_nomap(phys_addr_t base, phys_addr_t size)
919{
920 return memblock_setclr_flag(base, size, 1, MEMBLOCK_NOMAP);
921}
Tony Lucka3f5baf2015-06-24 16:58:12 -0700922
923/**
AKASHI Takahiro4c546b82017-04-03 11:23:54 +0900924 * memblock_clear_nomap - Clear flag MEMBLOCK_NOMAP for a specified region.
925 * @base: the base phys addr of the region
926 * @size: the size of the region
927 *
Mike Rapoport47cec442018-06-30 17:55:02 +0300928 * Return: 0 on success, -errno on failure.
AKASHI Takahiro4c546b82017-04-03 11:23:54 +0900929 */
930int __init_memblock memblock_clear_nomap(phys_addr_t base, phys_addr_t size)
931{
932 return memblock_setclr_flag(base, size, 0, MEMBLOCK_NOMAP);
933}
934
Mike Rapoport9f3d5ea2020-10-13 16:58:25 -0700935static bool should_skip_region(struct memblock_type *type,
936 struct memblock_region *m,
937 int nid, int flags)
Mike Rapoportc9a688a2019-03-11 23:30:50 -0700938{
939 int m_nid = memblock_get_region_node(m);
940
Mike Rapoport9f3d5ea2020-10-13 16:58:25 -0700941 /* we never skip regions when iterating memblock.reserved or physmem */
942 if (type != memblock_memory)
943 return false;
944
Mike Rapoportc9a688a2019-03-11 23:30:50 -0700945 /* only memory regions are associated with nodes, check it */
946 if (nid != NUMA_NO_NODE && nid != m_nid)
947 return true;
948
949 /* skip hotpluggable memory regions if needed */
Mike Rapoport79e482e2021-07-23 15:50:26 -0700950 if (movable_node_is_enabled() && memblock_is_hotpluggable(m) &&
951 !(flags & MEMBLOCK_HOTPLUG))
Mike Rapoportc9a688a2019-03-11 23:30:50 -0700952 return true;
953
954 /* if we want mirror memory skip non-mirror memory regions */
955 if ((flags & MEMBLOCK_MIRROR) && !memblock_is_mirror(m))
956 return true;
957
958 /* skip nomap memory unless we were asked for it explicitly */
959 if (!(flags & MEMBLOCK_NOMAP) && memblock_is_nomap(m))
960 return true;
961
962 return false;
963}
964
Robin Holt8e7a7f82015-06-30 14:56:41 -0700965/**
Mike Rapoporta2974132019-03-11 23:30:54 -0700966 * __next_mem_range - next function for for_each_free_mem_range() etc.
Tejun Heo35fd0802011-07-12 11:15:59 +0200967 * @idx: pointer to u64 loop variable
Grygorii Strashkob1154232014-01-21 15:50:16 -0800968 * @nid: node selector, %NUMA_NO_NODE for all nodes
Tony Luckfc6daaf2015-06-24 16:58:09 -0700969 * @flags: pick from blocks based on memory attributes
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +0100970 * @type_a: pointer to memblock_type from where the range is taken
971 * @type_b: pointer to memblock_type which excludes memory from being taken
Wanpeng Lidad75572012-06-20 12:53:01 -0700972 * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL
973 * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL
974 * @out_nid: ptr to int for nid of the range, can be %NULL
Tejun Heo35fd0802011-07-12 11:15:59 +0200975 *
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +0100976 * Find the first area from *@idx which matches @nid, fill the out
Tejun Heo35fd0802011-07-12 11:15:59 +0200977 * parameters, and update *@idx for the next iteration. The lower 32bit of
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +0100978 * *@idx contains index into type_a and the upper 32bit indexes the
979 * areas before each region in type_b. For example, if type_b regions
Tejun Heo35fd0802011-07-12 11:15:59 +0200980 * look like the following,
981 *
982 * 0:[0-16), 1:[32-48), 2:[128-130)
983 *
984 * The upper 32bit indexes the following regions.
985 *
986 * 0:[0-0), 1:[16-32), 2:[48-128), 3:[130-MAX)
987 *
988 * As both region arrays are sorted, the function advances the two indices
989 * in lockstep and returns each intersection.
990 */
David Hildenbrand77649902020-07-01 16:18:29 +0200991void __next_mem_range(u64 *idx, int nid, enum memblock_flags flags,
992 struct memblock_type *type_a,
993 struct memblock_type *type_b, phys_addr_t *out_start,
994 phys_addr_t *out_end, int *out_nid)
Tejun Heo35fd0802011-07-12 11:15:59 +0200995{
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +0100996 int idx_a = *idx & 0xffffffff;
997 int idx_b = *idx >> 32;
Grygorii Strashkob1154232014-01-21 15:50:16 -0800998
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +0100999 if (WARN_ONCE(nid == MAX_NUMNODES,
1000 "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
Grygorii Strashko560dca272014-01-21 15:50:55 -08001001 nid = NUMA_NO_NODE;
Tejun Heo35fd0802011-07-12 11:15:59 +02001002
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +01001003 for (; idx_a < type_a->cnt; idx_a++) {
1004 struct memblock_region *m = &type_a->regions[idx_a];
1005
Tejun Heo35fd0802011-07-12 11:15:59 +02001006 phys_addr_t m_start = m->base;
1007 phys_addr_t m_end = m->base + m->size;
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +01001008 int m_nid = memblock_get_region_node(m);
Tejun Heo35fd0802011-07-12 11:15:59 +02001009
Mike Rapoport9f3d5ea2020-10-13 16:58:25 -07001010 if (should_skip_region(type_a, m, nid, flags))
Ard Biesheuvelbf3d3cc2015-11-30 13:28:15 +01001011 continue;
1012
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +01001013 if (!type_b) {
1014 if (out_start)
1015 *out_start = m_start;
1016 if (out_end)
1017 *out_end = m_end;
1018 if (out_nid)
1019 *out_nid = m_nid;
1020 idx_a++;
1021 *idx = (u32)idx_a | (u64)idx_b << 32;
1022 return;
1023 }
Tejun Heo35fd0802011-07-12 11:15:59 +02001024
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +01001025 /* scan areas before each reservation */
1026 for (; idx_b < type_b->cnt + 1; idx_b++) {
1027 struct memblock_region *r;
1028 phys_addr_t r_start;
1029 phys_addr_t r_end;
1030
1031 r = &type_b->regions[idx_b];
1032 r_start = idx_b ? r[-1].base + r[-1].size : 0;
1033 r_end = idx_b < type_b->cnt ?
Stefan Agner1c4bc432018-06-07 17:06:15 -07001034 r->base : PHYS_ADDR_MAX;
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +01001035
1036 /*
1037 * if idx_b advanced past idx_a,
1038 * break out to advance idx_a
1039 */
Tejun Heo35fd0802011-07-12 11:15:59 +02001040 if (r_start >= m_end)
1041 break;
1042 /* if the two regions intersect, we're done */
1043 if (m_start < r_end) {
1044 if (out_start)
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +01001045 *out_start =
1046 max(m_start, r_start);
Tejun Heo35fd0802011-07-12 11:15:59 +02001047 if (out_end)
1048 *out_end = min(m_end, r_end);
1049 if (out_nid)
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +01001050 *out_nid = m_nid;
Tejun Heo35fd0802011-07-12 11:15:59 +02001051 /*
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +01001052 * The region which ends first is
1053 * advanced for the next iteration.
Tejun Heo35fd0802011-07-12 11:15:59 +02001054 */
1055 if (m_end <= r_end)
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +01001056 idx_a++;
Tejun Heo35fd0802011-07-12 11:15:59 +02001057 else
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +01001058 idx_b++;
1059 *idx = (u32)idx_a | (u64)idx_b << 32;
Tejun Heo35fd0802011-07-12 11:15:59 +02001060 return;
1061 }
1062 }
1063 }
1064
1065 /* signal end of iteration */
1066 *idx = ULLONG_MAX;
1067}
1068
Tejun Heo7bd0b0f2011-12-08 10:22:09 -08001069/**
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +01001070 * __next_mem_range_rev - generic next function for for_each_*_range_rev()
1071 *
Tejun Heo7bd0b0f2011-12-08 10:22:09 -08001072 * @idx: pointer to u64 loop variable
Alexander Kuleshovad5ea8c2015-09-08 15:04:22 -07001073 * @nid: node selector, %NUMA_NO_NODE for all nodes
Tony Luckfc6daaf2015-06-24 16:58:09 -07001074 * @flags: pick from blocks based on memory attributes
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +01001075 * @type_a: pointer to memblock_type from where the range is taken
1076 * @type_b: pointer to memblock_type which excludes memory from being taken
Wanpeng Lidad75572012-06-20 12:53:01 -07001077 * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL
1078 * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL
1079 * @out_nid: ptr to int for nid of the range, can be %NULL
Tejun Heo7bd0b0f2011-12-08 10:22:09 -08001080 *
Mike Rapoport47cec442018-06-30 17:55:02 +03001081 * Finds the next range from type_a which is not marked as unsuitable
1082 * in type_b.
1083 *
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +01001084 * Reverse of __next_mem_range().
Tejun Heo7bd0b0f2011-12-08 10:22:09 -08001085 */
Mike Rapoporte1720fe2018-06-30 17:55:01 +03001086void __init_memblock __next_mem_range_rev(u64 *idx, int nid,
1087 enum memblock_flags flags,
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +01001088 struct memblock_type *type_a,
1089 struct memblock_type *type_b,
1090 phys_addr_t *out_start,
1091 phys_addr_t *out_end, int *out_nid)
Tejun Heo7bd0b0f2011-12-08 10:22:09 -08001092{
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +01001093 int idx_a = *idx & 0xffffffff;
1094 int idx_b = *idx >> 32;
Grygorii Strashkob1154232014-01-21 15:50:16 -08001095
Grygorii Strashko560dca272014-01-21 15:50:55 -08001096 if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
1097 nid = NUMA_NO_NODE;
Tejun Heo7bd0b0f2011-12-08 10:22:09 -08001098
1099 if (*idx == (u64)ULLONG_MAX) {
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +01001100 idx_a = type_a->cnt - 1;
zijun_hue47608a2016-08-04 15:32:00 -07001101 if (type_b != NULL)
1102 idx_b = type_b->cnt;
1103 else
1104 idx_b = 0;
Tejun Heo7bd0b0f2011-12-08 10:22:09 -08001105 }
1106
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +01001107 for (; idx_a >= 0; idx_a--) {
1108 struct memblock_region *m = &type_a->regions[idx_a];
1109
Tejun Heo7bd0b0f2011-12-08 10:22:09 -08001110 phys_addr_t m_start = m->base;
1111 phys_addr_t m_end = m->base + m->size;
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +01001112 int m_nid = memblock_get_region_node(m);
Tejun Heo7bd0b0f2011-12-08 10:22:09 -08001113
Mike Rapoport9f3d5ea2020-10-13 16:58:25 -07001114 if (should_skip_region(type_a, m, nid, flags))
Ard Biesheuvelbf3d3cc2015-11-30 13:28:15 +01001115 continue;
1116
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +01001117 if (!type_b) {
1118 if (out_start)
1119 *out_start = m_start;
1120 if (out_end)
1121 *out_end = m_end;
1122 if (out_nid)
1123 *out_nid = m_nid;
zijun_hufb399b42016-07-28 15:48:56 -07001124 idx_a--;
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +01001125 *idx = (u32)idx_a | (u64)idx_b << 32;
1126 return;
1127 }
Tejun Heo7bd0b0f2011-12-08 10:22:09 -08001128
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +01001129 /* scan areas before each reservation */
1130 for (; idx_b >= 0; idx_b--) {
1131 struct memblock_region *r;
1132 phys_addr_t r_start;
1133 phys_addr_t r_end;
1134
1135 r = &type_b->regions[idx_b];
1136 r_start = idx_b ? r[-1].base + r[-1].size : 0;
1137 r_end = idx_b < type_b->cnt ?
Stefan Agner1c4bc432018-06-07 17:06:15 -07001138 r->base : PHYS_ADDR_MAX;
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +01001139 /*
1140 * if idx_b advanced past idx_a,
1141 * break out to advance idx_a
1142 */
1143
Tejun Heo7bd0b0f2011-12-08 10:22:09 -08001144 if (r_end <= m_start)
1145 break;
1146 /* if the two regions intersect, we're done */
1147 if (m_end > r_start) {
1148 if (out_start)
1149 *out_start = max(m_start, r_start);
1150 if (out_end)
1151 *out_end = min(m_end, r_end);
1152 if (out_nid)
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +01001153 *out_nid = m_nid;
Tejun Heo7bd0b0f2011-12-08 10:22:09 -08001154 if (m_start >= r_start)
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +01001155 idx_a--;
Tejun Heo7bd0b0f2011-12-08 10:22:09 -08001156 else
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +01001157 idx_b--;
1158 *idx = (u32)idx_a | (u64)idx_b << 32;
Tejun Heo7bd0b0f2011-12-08 10:22:09 -08001159 return;
1160 }
1161 }
1162 }
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +01001163 /* signal end of iteration */
Tejun Heo7bd0b0f2011-12-08 10:22:09 -08001164 *idx = ULLONG_MAX;
1165}
1166
Tejun Heo7c0caeb2011-07-14 11:43:42 +02001167/*
Chen Chang45e79812018-11-16 15:08:57 -08001168 * Common iterator interface used to define for_each_mem_pfn_range().
Tejun Heo7c0caeb2011-07-14 11:43:42 +02001169 */
1170void __init_memblock __next_mem_pfn_range(int *idx, int nid,
1171 unsigned long *out_start_pfn,
1172 unsigned long *out_end_pfn, int *out_nid)
1173{
1174 struct memblock_type *type = &memblock.memory;
1175 struct memblock_region *r;
Mike Rapoportd622abf2020-06-03 15:56:53 -07001176 int r_nid;
Tejun Heo7c0caeb2011-07-14 11:43:42 +02001177
1178 while (++*idx < type->cnt) {
1179 r = &type->regions[*idx];
Mike Rapoportd622abf2020-06-03 15:56:53 -07001180 r_nid = memblock_get_region_node(r);
Tejun Heo7c0caeb2011-07-14 11:43:42 +02001181
1182 if (PFN_UP(r->base) >= PFN_DOWN(r->base + r->size))
1183 continue;
Mike Rapoportd622abf2020-06-03 15:56:53 -07001184 if (nid == MAX_NUMNODES || nid == r_nid)
Tejun Heo7c0caeb2011-07-14 11:43:42 +02001185 break;
1186 }
1187 if (*idx >= type->cnt) {
1188 *idx = -1;
1189 return;
1190 }
1191
1192 if (out_start_pfn)
1193 *out_start_pfn = PFN_UP(r->base);
1194 if (out_end_pfn)
1195 *out_end_pfn = PFN_DOWN(r->base + r->size);
1196 if (out_nid)
Mike Rapoportd622abf2020-06-03 15:56:53 -07001197 *out_nid = r_nid;
Tejun Heo7c0caeb2011-07-14 11:43:42 +02001198}
1199
1200/**
1201 * memblock_set_node - set node ID on memblock regions
1202 * @base: base of area to set node ID for
1203 * @size: size of area to set node ID for
Tang Chene7e8de52014-01-21 15:49:26 -08001204 * @type: memblock type to set node ID for
Tejun Heo7c0caeb2011-07-14 11:43:42 +02001205 * @nid: node ID to set
1206 *
Mike Rapoport47cec442018-06-30 17:55:02 +03001207 * Set the nid of memblock @type regions in [@base, @base + @size) to @nid.
Tejun Heo7c0caeb2011-07-14 11:43:42 +02001208 * Regions which cross the area boundaries are split as necessary.
1209 *
Mike Rapoport47cec442018-06-30 17:55:02 +03001210 * Return:
Tejun Heo7c0caeb2011-07-14 11:43:42 +02001211 * 0 on success, -errno on failure.
1212 */
1213int __init_memblock memblock_set_node(phys_addr_t base, phys_addr_t size,
Tang Chene7e8de52014-01-21 15:49:26 -08001214 struct memblock_type *type, int nid)
Tejun Heo7c0caeb2011-07-14 11:43:42 +02001215{
Mike Rapoporta9ee6cf2021-06-28 19:43:01 -07001216#ifdef CONFIG_NUMA
Tejun Heo6a9ceb32011-12-08 10:22:07 -08001217 int start_rgn, end_rgn;
1218 int i, ret;
Tejun Heo7c0caeb2011-07-14 11:43:42 +02001219
Tejun Heo6a9ceb32011-12-08 10:22:07 -08001220 ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
1221 if (ret)
1222 return ret;
Tejun Heo7c0caeb2011-07-14 11:43:42 +02001223
Tejun Heo6a9ceb32011-12-08 10:22:07 -08001224 for (i = start_rgn; i < end_rgn; i++)
Wanpeng Lie9d24ad2012-10-08 16:32:21 -07001225 memblock_set_region_node(&type->regions[i], nid);
Tejun Heo7c0caeb2011-07-14 11:43:42 +02001226
1227 memblock_merge_regions(type);
Mike Rapoport3f08a302020-06-03 15:57:02 -07001228#endif
Tejun Heo7c0caeb2011-07-14 11:43:42 +02001229 return 0;
1230}
Mike Rapoport3f08a302020-06-03 15:57:02 -07001231
Alexander Duyck837566e2019-05-13 17:21:17 -07001232#ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1233/**
1234 * __next_mem_pfn_range_in_zone - iterator for for_each_*_range_in_zone()
1235 *
1236 * @idx: pointer to u64 loop variable
1237 * @zone: zone in which all of the memory blocks reside
1238 * @out_spfn: ptr to ulong for start pfn of the range, can be %NULL
1239 * @out_epfn: ptr to ulong for end pfn of the range, can be %NULL
1240 *
1241 * This function is meant to be a zone/pfn specific wrapper for the
1242 * for_each_mem_range type iterators. Specifically they are used in the
1243 * deferred memory init routines and as such we were duplicating much of
1244 * this logic throughout the code. So instead of having it in multiple
1245 * locations it seemed like it would make more sense to centralize this to
1246 * one new iterator that does everything they need.
1247 */
1248void __init_memblock
1249__next_mem_pfn_range_in_zone(u64 *idx, struct zone *zone,
1250 unsigned long *out_spfn, unsigned long *out_epfn)
1251{
1252 int zone_nid = zone_to_nid(zone);
1253 phys_addr_t spa, epa;
1254 int nid;
1255
1256 __next_mem_range(idx, zone_nid, MEMBLOCK_NONE,
1257 &memblock.memory, &memblock.reserved,
1258 &spa, &epa, &nid);
1259
1260 while (*idx != U64_MAX) {
1261 unsigned long epfn = PFN_DOWN(epa);
1262 unsigned long spfn = PFN_UP(spa);
1263
1264 /*
1265 * Verify the end is at least past the start of the zone and
1266 * that we have at least one PFN to initialize.
1267 */
1268 if (zone->zone_start_pfn < epfn && spfn < epfn) {
1269 /* if we went too far just stop searching */
1270 if (zone_end_pfn(zone) <= spfn) {
1271 *idx = U64_MAX;
1272 break;
1273 }
1274
1275 if (out_spfn)
1276 *out_spfn = max(zone->zone_start_pfn, spfn);
1277 if (out_epfn)
1278 *out_epfn = min(zone_end_pfn(zone), epfn);
1279
1280 return;
1281 }
1282
1283 __next_mem_range(idx, zone_nid, MEMBLOCK_NONE,
1284 &memblock.memory, &memblock.reserved,
1285 &spa, &epa, &nid);
1286 }
1287
1288 /* signal end of iteration */
1289 if (out_spfn)
1290 *out_spfn = ULONG_MAX;
1291 if (out_epfn)
1292 *out_epfn = 0;
1293}
1294
1295#endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
Tejun Heo7c0caeb2011-07-14 11:43:42 +02001296
Mike Rapoport92d12f92019-03-11 23:29:41 -07001297/**
1298 * memblock_alloc_range_nid - allocate boot memory block
1299 * @size: size of memory block to be allocated in bytes
1300 * @align: alignment of the region and block's size
1301 * @start: the lower bound of the memory region to allocate (phys address)
1302 * @end: the upper bound of the memory region to allocate (phys address)
1303 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
Yunfeng Ye0ac398b2019-11-30 17:56:27 -08001304 * @exact_nid: control the allocation fall back to other nodes
Mike Rapoport92d12f92019-03-11 23:29:41 -07001305 *
1306 * The allocation is performed from memory region limited by
Cao jin95830662019-11-30 17:56:24 -08001307 * memblock.current_limit if @end == %MEMBLOCK_ALLOC_ACCESSIBLE.
Mike Rapoport92d12f92019-03-11 23:29:41 -07001308 *
Yunfeng Ye0ac398b2019-11-30 17:56:27 -08001309 * If the specified node can not hold the requested memory and @exact_nid
1310 * is false, the allocation falls back to any node in the system.
Mike Rapoport92d12f92019-03-11 23:29:41 -07001311 *
1312 * For systems with memory mirroring, the allocation is attempted first
1313 * from the regions with mirroring enabled and then retried from any
1314 * memory region.
1315 *
1316 * In addition, function sets the min_count to 0 using kmemleak_alloc_phys for
1317 * allocated boot memory block, so that it is never reported as leaks.
1318 *
1319 * Return:
1320 * Physical address of allocated memory block on success, %0 on failure.
1321 */
Aslan Bakirov8676af12020-04-10 14:32:42 -07001322phys_addr_t __init memblock_alloc_range_nid(phys_addr_t size,
Akinobu Mita2bfc2862014-06-04 16:06:53 -07001323 phys_addr_t align, phys_addr_t start,
Yunfeng Ye0ac398b2019-11-30 17:56:27 -08001324 phys_addr_t end, int nid,
1325 bool exact_nid)
Yinghai Lu95f72d12010-07-12 14:36:09 +10001326{
Mike Rapoport92d12f92019-03-11 23:29:41 -07001327 enum memblock_flags flags = choose_memblock_flags();
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +10001328 phys_addr_t found;
Yinghai Lu95f72d12010-07-12 14:36:09 +10001329
Mike Rapoport92d12f92019-03-11 23:29:41 -07001330 if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
1331 nid = NUMA_NO_NODE;
1332
Mike Rapoport2f770802018-10-30 15:10:01 -07001333 if (!align) {
1334 /* Can't use WARNs this early in boot on powerpc */
1335 dump_stack();
1336 align = SMP_CACHE_BYTES;
1337 }
1338
Mike Rapoport92d12f92019-03-11 23:29:41 -07001339again:
Tony Luckfc6daaf2015-06-24 16:58:09 -07001340 found = memblock_find_in_range_node(size, align, start, end, nid,
1341 flags);
Mike Rapoport92d12f92019-03-11 23:29:41 -07001342 if (found && !memblock_reserve(found, size))
1343 goto done;
1344
Yunfeng Ye0ac398b2019-11-30 17:56:27 -08001345 if (nid != NUMA_NO_NODE && !exact_nid) {
Mike Rapoport92d12f92019-03-11 23:29:41 -07001346 found = memblock_find_in_range_node(size, align, start,
1347 end, NUMA_NO_NODE,
1348 flags);
1349 if (found && !memblock_reserve(found, size))
1350 goto done;
1351 }
1352
1353 if (flags & MEMBLOCK_MIRROR) {
1354 flags &= ~MEMBLOCK_MIRROR;
1355 pr_warn("Could not allocate %pap bytes of mirrored memory\n",
1356 &size);
1357 goto again;
1358 }
1359
1360 return 0;
1361
1362done:
1363 /* Skip kmemleak for kasan_init() due to high volume. */
1364 if (end != MEMBLOCK_ALLOC_KASAN)
Catalin Marinasaedf95e2014-06-06 14:38:20 -07001365 /*
Mike Rapoport92d12f92019-03-11 23:29:41 -07001366 * The min_count is set to 0 so that memblock allocated
1367 * blocks are never reported as leaks. This is because many
1368 * of these blocks are only referred via the physical
1369 * address which is not looked up by kmemleak.
Catalin Marinasaedf95e2014-06-06 14:38:20 -07001370 */
Catalin Marinas9099dae2016-10-11 13:55:11 -07001371 kmemleak_alloc_phys(found, size, 0, 0);
Mike Rapoport92d12f92019-03-11 23:29:41 -07001372
1373 return found;
Yinghai Lu95f72d12010-07-12 14:36:09 +10001374}
1375
Mike Rapoporta2974132019-03-11 23:30:54 -07001376/**
1377 * memblock_phys_alloc_range - allocate a memory block inside specified range
1378 * @size: size of memory block to be allocated in bytes
1379 * @align: alignment of the region and block's size
1380 * @start: the lower bound of the memory region to allocate (physical address)
1381 * @end: the upper bound of the memory region to allocate (physical address)
1382 *
1383 * Allocate @size bytes in the between @start and @end.
1384 *
1385 * Return: physical address of the allocated memory block on success,
1386 * %0 on failure.
1387 */
Mike Rapoport8a770c22019-03-11 23:29:16 -07001388phys_addr_t __init memblock_phys_alloc_range(phys_addr_t size,
1389 phys_addr_t align,
1390 phys_addr_t start,
1391 phys_addr_t end)
Akinobu Mita2bfc2862014-06-04 16:06:53 -07001392{
Faiyaz Mohammedb5cf2d62020-11-16 10:14:04 +05301393 memblock_dbg("%s: %llu bytes align=0x%llx from=%pa max_addr=%pa %pS\n",
1394 __func__, (u64)size, (u64)align, &start, &end,
1395 (void *)_RET_IP_);
Yunfeng Ye0ac398b2019-11-30 17:56:27 -08001396 return memblock_alloc_range_nid(size, align, start, end, NUMA_NO_NODE,
1397 false);
Tejun Heo7bd0b0f2011-12-08 10:22:09 -08001398}
1399
Mike Rapoporta2974132019-03-11 23:30:54 -07001400/**
Levi Yun17cbe032021-01-20 21:28:18 +09001401 * memblock_phys_alloc_try_nid - allocate a memory block from specified NUMA node
Mike Rapoporta2974132019-03-11 23:30:54 -07001402 * @size: size of memory block to be allocated in bytes
1403 * @align: alignment of the region and block's size
1404 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1405 *
1406 * Allocates memory block from the specified NUMA node. If the node
1407 * has no available memory, attempts to allocated from any node in the
1408 * system.
1409 *
1410 * Return: physical address of the allocated memory block on success,
1411 * %0 on failure.
1412 */
Mike Rapoport9a8dd702018-10-30 15:07:59 -07001413phys_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 -07001414{
Mike Rapoport33755572019-03-11 23:29:21 -07001415 return memblock_alloc_range_nid(size, align, 0,
Yunfeng Ye0ac398b2019-11-30 17:56:27 -08001416 MEMBLOCK_ALLOC_ACCESSIBLE, nid, false);
Yinghai Lu95f72d12010-07-12 14:36:09 +10001417}
1418
Santosh Shilimkar26f09e92014-01-21 15:50:19 -08001419/**
Mike Rapoporteb31d552018-10-30 15:08:04 -07001420 * memblock_alloc_internal - allocate boot memory block
Santosh Shilimkar26f09e92014-01-21 15:50:19 -08001421 * @size: size of memory block to be allocated in bytes
1422 * @align: alignment of the region and block's size
1423 * @min_addr: the lower bound of the memory region to allocate (phys address)
1424 * @max_addr: the upper bound of the memory region to allocate (phys address)
1425 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
Yunfeng Ye0ac398b2019-11-30 17:56:27 -08001426 * @exact_nid: control the allocation fall back to other nodes
Santosh Shilimkar26f09e92014-01-21 15:50:19 -08001427 *
Mike Rapoport92d12f92019-03-11 23:29:41 -07001428 * Allocates memory block using memblock_alloc_range_nid() and
1429 * converts the returned physical address to virtual.
1430 *
Santosh Shilimkar26f09e92014-01-21 15:50:19 -08001431 * The @min_addr limit is dropped if it can not be satisfied and the allocation
Mike Rapoport92d12f92019-03-11 23:29:41 -07001432 * will fall back to memory below @min_addr. Other constraints, such
1433 * as node and mirrored memory will be handled again in
1434 * memblock_alloc_range_nid().
Santosh Shilimkar26f09e92014-01-21 15:50:19 -08001435 *
Mike Rapoport47cec442018-06-30 17:55:02 +03001436 * Return:
Santosh Shilimkar26f09e92014-01-21 15:50:19 -08001437 * Virtual address of allocated memory block on success, NULL on failure.
1438 */
Mike Rapoporteb31d552018-10-30 15:08:04 -07001439static void * __init memblock_alloc_internal(
Santosh Shilimkar26f09e92014-01-21 15:50:19 -08001440 phys_addr_t size, phys_addr_t align,
1441 phys_addr_t min_addr, phys_addr_t max_addr,
Yunfeng Ye0ac398b2019-11-30 17:56:27 -08001442 int nid, bool exact_nid)
Santosh Shilimkar26f09e92014-01-21 15:50:19 -08001443{
1444 phys_addr_t alloc;
Santosh Shilimkar26f09e92014-01-21 15:50:19 -08001445
1446 /*
1447 * Detect any accidental use of these APIs after slab is ready, as at
1448 * this moment memblock may be deinitialized already and its
Mike Rapoportc6ffc5c2018-10-30 15:09:30 -07001449 * internal data may be destroyed (after execution of memblock_free_all)
Santosh Shilimkar26f09e92014-01-21 15:50:19 -08001450 */
1451 if (WARN_ON_ONCE(slab_is_available()))
1452 return kzalloc_node(size, GFP_NOWAIT, nid);
1453
Mike Rapoportf3057ad2019-10-18 20:20:01 -07001454 if (max_addr > memblock.current_limit)
1455 max_addr = memblock.current_limit;
1456
Yunfeng Ye0ac398b2019-11-30 17:56:27 -08001457 alloc = memblock_alloc_range_nid(size, align, min_addr, max_addr, nid,
1458 exact_nid);
Mike Rapoport2f770802018-10-30 15:10:01 -07001459
Mike Rapoport92d12f92019-03-11 23:29:41 -07001460 /* retry allocation without lower limit */
1461 if (!alloc && min_addr)
Yunfeng Ye0ac398b2019-11-30 17:56:27 -08001462 alloc = memblock_alloc_range_nid(size, align, 0, max_addr, nid,
1463 exact_nid);
Santosh Shilimkar26f09e92014-01-21 15:50:19 -08001464
Mike Rapoport92d12f92019-03-11 23:29:41 -07001465 if (!alloc)
1466 return NULL;
Santosh Shilimkar26f09e92014-01-21 15:50:19 -08001467
Mike Rapoport92d12f92019-03-11 23:29:41 -07001468 return phys_to_virt(alloc);
Santosh Shilimkar26f09e92014-01-21 15:50:19 -08001469}
1470
1471/**
Yunfeng Ye0ac398b2019-11-30 17:56:27 -08001472 * memblock_alloc_exact_nid_raw - allocate boot memory block on the exact node
1473 * without zeroing memory
1474 * @size: size of memory block to be allocated in bytes
1475 * @align: alignment of the region and block's size
1476 * @min_addr: the lower bound of the memory region from where the allocation
1477 * is preferred (phys address)
1478 * @max_addr: the upper bound of the memory region from where the allocation
1479 * is preferred (phys address), or %MEMBLOCK_ALLOC_ACCESSIBLE to
1480 * allocate only from memory limited by memblock.current_limit value
1481 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1482 *
1483 * Public function, provides additional debug information (including caller
1484 * info), if enabled. Does not zero allocated memory.
1485 *
1486 * Return:
1487 * Virtual address of allocated memory block on success, NULL on failure.
1488 */
1489void * __init memblock_alloc_exact_nid_raw(
1490 phys_addr_t size, phys_addr_t align,
1491 phys_addr_t min_addr, phys_addr_t max_addr,
1492 int nid)
1493{
Yunfeng Ye0ac398b2019-11-30 17:56:27 -08001494 memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=%pa max_addr=%pa %pS\n",
1495 __func__, (u64)size, (u64)align, nid, &min_addr,
1496 &max_addr, (void *)_RET_IP_);
1497
Mike Rapoport08678802021-09-02 14:58:05 -07001498 return memblock_alloc_internal(size, align, min_addr, max_addr, nid,
1499 true);
Yunfeng Ye0ac398b2019-11-30 17:56:27 -08001500}
1501
1502/**
Mike Rapoporteb31d552018-10-30 15:08:04 -07001503 * memblock_alloc_try_nid_raw - allocate boot memory block without zeroing
Pavel Tatashinea1f5f32017-11-15 17:36:27 -08001504 * memory and without panicking
1505 * @size: size of memory block to be allocated in bytes
1506 * @align: alignment of the region and block's size
1507 * @min_addr: the lower bound of the memory region from where the allocation
1508 * is preferred (phys address)
1509 * @max_addr: the upper bound of the memory region from where the allocation
Mike Rapoport97ad1082018-10-30 15:09:44 -07001510 * is preferred (phys address), or %MEMBLOCK_ALLOC_ACCESSIBLE to
Pavel Tatashinea1f5f32017-11-15 17:36:27 -08001511 * allocate only from memory limited by memblock.current_limit value
1512 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1513 *
1514 * Public function, provides additional debug information (including caller
1515 * info), if enabled. Does not zero allocated memory, does not panic if request
1516 * cannot be satisfied.
1517 *
Mike Rapoport47cec442018-06-30 17:55:02 +03001518 * Return:
Pavel Tatashinea1f5f32017-11-15 17:36:27 -08001519 * Virtual address of allocated memory block on success, NULL on failure.
1520 */
Mike Rapoporteb31d552018-10-30 15:08:04 -07001521void * __init memblock_alloc_try_nid_raw(
Pavel Tatashinea1f5f32017-11-15 17:36:27 -08001522 phys_addr_t size, phys_addr_t align,
1523 phys_addr_t min_addr, phys_addr_t max_addr,
1524 int nid)
1525{
Sakari Ailusd75f7732019-03-25 21:32:28 +02001526 memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=%pa max_addr=%pa %pS\n",
Mike Rapoporta36aab82018-08-17 15:47:17 -07001527 __func__, (u64)size, (u64)align, nid, &min_addr,
1528 &max_addr, (void *)_RET_IP_);
Pavel Tatashinea1f5f32017-11-15 17:36:27 -08001529
Mike Rapoport08678802021-09-02 14:58:05 -07001530 return memblock_alloc_internal(size, align, min_addr, max_addr, nid,
1531 false);
Pavel Tatashinea1f5f32017-11-15 17:36:27 -08001532}
1533
1534/**
Mike Rapoportc0dbe822019-03-11 23:30:37 -07001535 * memblock_alloc_try_nid - allocate boot memory block
Santosh Shilimkar26f09e92014-01-21 15:50:19 -08001536 * @size: size of memory block to be allocated in bytes
1537 * @align: alignment of the region and block's size
1538 * @min_addr: the lower bound of the memory region from where the allocation
1539 * is preferred (phys address)
1540 * @max_addr: the upper bound of the memory region from where the allocation
Mike Rapoport97ad1082018-10-30 15:09:44 -07001541 * is preferred (phys address), or %MEMBLOCK_ALLOC_ACCESSIBLE to
Santosh Shilimkar26f09e92014-01-21 15:50:19 -08001542 * allocate only from memory limited by memblock.current_limit value
1543 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1544 *
Mike Rapoportc0dbe822019-03-11 23:30:37 -07001545 * Public function, provides additional debug information (including caller
1546 * info), if enabled. This function zeroes the allocated memory.
Santosh Shilimkar26f09e92014-01-21 15:50:19 -08001547 *
Mike Rapoport47cec442018-06-30 17:55:02 +03001548 * Return:
Santosh Shilimkar26f09e92014-01-21 15:50:19 -08001549 * Virtual address of allocated memory block on success, NULL on failure.
1550 */
Mike Rapoporteb31d552018-10-30 15:08:04 -07001551void * __init memblock_alloc_try_nid(
Santosh Shilimkar26f09e92014-01-21 15:50:19 -08001552 phys_addr_t size, phys_addr_t align,
1553 phys_addr_t min_addr, phys_addr_t max_addr,
1554 int nid)
1555{
1556 void *ptr;
1557
Sakari Ailusd75f7732019-03-25 21:32:28 +02001558 memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=%pa max_addr=%pa %pS\n",
Mike Rapoporta36aab82018-08-17 15:47:17 -07001559 __func__, (u64)size, (u64)align, nid, &min_addr,
1560 &max_addr, (void *)_RET_IP_);
Mike Rapoporteb31d552018-10-30 15:08:04 -07001561 ptr = memblock_alloc_internal(size, align,
Yunfeng Ye0ac398b2019-11-30 17:56:27 -08001562 min_addr, max_addr, nid, false);
Mike Rapoportc0dbe822019-03-11 23:30:37 -07001563 if (ptr)
Pavel Tatashinea1f5f32017-11-15 17:36:27 -08001564 memset(ptr, 0, size);
Santosh Shilimkar26f09e92014-01-21 15:50:19 -08001565
Mike Rapoportc0dbe822019-03-11 23:30:37 -07001566 return ptr;
Santosh Shilimkar26f09e92014-01-21 15:50:19 -08001567}
1568
1569/**
Mike Rapoporta2974132019-03-11 23:30:54 -07001570 * __memblock_free_late - free pages directly to buddy allocator
Mike Rapoport48a833c2018-06-30 17:55:03 +03001571 * @base: phys starting address of the boot memory block
Santosh Shilimkar26f09e92014-01-21 15:50:19 -08001572 * @size: size of the boot memory block in bytes
1573 *
Mike Rapoporta2974132019-03-11 23:30:54 -07001574 * This is only useful when the memblock allocator has already been torn
Santosh Shilimkar26f09e92014-01-21 15:50:19 -08001575 * down, but we are still initializing the system. Pages are released directly
Mike Rapoporta2974132019-03-11 23:30:54 -07001576 * to the buddy allocator.
Santosh Shilimkar26f09e92014-01-21 15:50:19 -08001577 */
1578void __init __memblock_free_late(phys_addr_t base, phys_addr_t size)
1579{
Mike Rapoporta36aab82018-08-17 15:47:17 -07001580 phys_addr_t cursor, end;
Santosh Shilimkar26f09e92014-01-21 15:50:19 -08001581
Mike Rapoporta36aab82018-08-17 15:47:17 -07001582 end = base + size - 1;
Sakari Ailusd75f7732019-03-25 21:32:28 +02001583 memblock_dbg("%s: [%pa-%pa] %pS\n",
Mike Rapoporta36aab82018-08-17 15:47:17 -07001584 __func__, &base, &end, (void *)_RET_IP_);
Catalin Marinas9099dae2016-10-11 13:55:11 -07001585 kmemleak_free_part_phys(base, size);
Santosh Shilimkar26f09e92014-01-21 15:50:19 -08001586 cursor = PFN_UP(base);
1587 end = PFN_DOWN(base + size);
1588
1589 for (; cursor < end; cursor++) {
Mike Rapoport7c2ee342018-10-30 15:09:36 -07001590 memblock_free_pages(pfn_to_page(cursor), cursor, 0);
Arun KSca79b0c2018-12-28 00:34:29 -08001591 totalram_pages_inc();
Santosh Shilimkar26f09e92014-01-21 15:50:19 -08001592 }
1593}
Benjamin Herrenschmidt9d1e2492010-07-06 15:39:17 -07001594
1595/*
1596 * Remaining API functions
1597 */
1598
David Gibson1f1ffb8a2016-02-05 15:36:19 -08001599phys_addr_t __init_memblock memblock_phys_mem_size(void)
Yinghai Lu95f72d12010-07-12 14:36:09 +10001600{
Tejun Heo1440c4e2011-12-08 10:22:08 -08001601 return memblock.memory.total_size;
Yinghai Lu95f72d12010-07-12 14:36:09 +10001602}
1603
Srikar Dronamraju8907de52016-10-07 16:59:18 -07001604phys_addr_t __init_memblock memblock_reserved_size(void)
1605{
1606 return memblock.reserved.total_size;
1607}
1608
Sam Ravnborg0a93ebe2011-10-31 17:08:16 -07001609/* lowest address */
1610phys_addr_t __init_memblock memblock_start_of_DRAM(void)
1611{
1612 return memblock.memory.regions[0].base;
1613}
1614
Yinghai Lu10d06432010-07-28 15:43:02 +10001615phys_addr_t __init_memblock memblock_end_of_DRAM(void)
Yinghai Lu95f72d12010-07-12 14:36:09 +10001616{
1617 int idx = memblock.memory.cnt - 1;
1618
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +10001619 return (memblock.memory.regions[idx].base + memblock.memory.regions[idx].size);
Yinghai Lu95f72d12010-07-12 14:36:09 +10001620}
1621
Dennis Chena571d4e2016-07-28 15:48:26 -07001622static phys_addr_t __init_memblock __find_max_addr(phys_addr_t limit)
Yinghai Lu95f72d12010-07-12 14:36:09 +10001623{
Stefan Agner1c4bc432018-06-07 17:06:15 -07001624 phys_addr_t max_addr = PHYS_ADDR_MAX;
Emil Medve136199f2014-04-07 15:37:52 -07001625 struct memblock_region *r;
Yinghai Lu95f72d12010-07-12 14:36:09 +10001626
Dennis Chena571d4e2016-07-28 15:48:26 -07001627 /*
1628 * translate the memory @limit size into the max address within one of
1629 * the memory memblock regions, if the @limit exceeds the total size
Stefan Agner1c4bc432018-06-07 17:06:15 -07001630 * of those regions, max_addr will keep original value PHYS_ADDR_MAX
Dennis Chena571d4e2016-07-28 15:48:26 -07001631 */
Mike Rapoportcc6de162020-10-13 16:58:30 -07001632 for_each_mem_region(r) {
Tejun Heoc0ce8fe2011-12-08 10:22:07 -08001633 if (limit <= r->size) {
1634 max_addr = r->base + limit;
1635 break;
1636 }
1637 limit -= r->size;
Yinghai Lu95f72d12010-07-12 14:36:09 +10001638 }
1639
Dennis Chena571d4e2016-07-28 15:48:26 -07001640 return max_addr;
1641}
1642
1643void __init memblock_enforce_memory_limit(phys_addr_t limit)
1644{
Colin Ian King49aef712020-04-01 21:11:01 -07001645 phys_addr_t max_addr;
Dennis Chena571d4e2016-07-28 15:48:26 -07001646
1647 if (!limit)
1648 return;
1649
1650 max_addr = __find_max_addr(limit);
1651
1652 /* @limit exceeds the total size of the memory, do nothing */
Stefan Agner1c4bc432018-06-07 17:06:15 -07001653 if (max_addr == PHYS_ADDR_MAX)
Dennis Chena571d4e2016-07-28 15:48:26 -07001654 return;
1655
Tejun Heoc0ce8fe2011-12-08 10:22:07 -08001656 /* truncate both memory and reserved regions */
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +01001657 memblock_remove_range(&memblock.memory, max_addr,
Stefan Agner1c4bc432018-06-07 17:06:15 -07001658 PHYS_ADDR_MAX);
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +01001659 memblock_remove_range(&memblock.reserved, max_addr,
Stefan Agner1c4bc432018-06-07 17:06:15 -07001660 PHYS_ADDR_MAX);
Yinghai Lu95f72d12010-07-12 14:36:09 +10001661}
1662
AKASHI Takahiroc9ca9b42017-04-03 11:23:55 +09001663void __init memblock_cap_memory_range(phys_addr_t base, phys_addr_t size)
1664{
1665 int start_rgn, end_rgn;
1666 int i, ret;
1667
1668 if (!size)
1669 return;
1670
1671 ret = memblock_isolate_range(&memblock.memory, base, size,
1672 &start_rgn, &end_rgn);
1673 if (ret)
1674 return;
1675
1676 /* remove all the MAP regions */
1677 for (i = memblock.memory.cnt - 1; i >= end_rgn; i--)
1678 if (!memblock_is_nomap(&memblock.memory.regions[i]))
1679 memblock_remove_region(&memblock.memory, i);
1680
1681 for (i = start_rgn - 1; i >= 0; i--)
1682 if (!memblock_is_nomap(&memblock.memory.regions[i]))
1683 memblock_remove_region(&memblock.memory, i);
1684
1685 /* truncate the reserved regions */
1686 memblock_remove_range(&memblock.reserved, 0, base);
1687 memblock_remove_range(&memblock.reserved,
Stefan Agner1c4bc432018-06-07 17:06:15 -07001688 base + size, PHYS_ADDR_MAX);
AKASHI Takahiroc9ca9b42017-04-03 11:23:55 +09001689}
1690
Dennis Chena571d4e2016-07-28 15:48:26 -07001691void __init memblock_mem_limit_remove_map(phys_addr_t limit)
1692{
Dennis Chena571d4e2016-07-28 15:48:26 -07001693 phys_addr_t max_addr;
Dennis Chena571d4e2016-07-28 15:48:26 -07001694
1695 if (!limit)
1696 return;
1697
1698 max_addr = __find_max_addr(limit);
1699
1700 /* @limit exceeds the total size of the memory, do nothing */
Stefan Agner1c4bc432018-06-07 17:06:15 -07001701 if (max_addr == PHYS_ADDR_MAX)
Dennis Chena571d4e2016-07-28 15:48:26 -07001702 return;
1703
AKASHI Takahiroc9ca9b42017-04-03 11:23:55 +09001704 memblock_cap_memory_range(0, max_addr);
Dennis Chena571d4e2016-07-28 15:48:26 -07001705}
1706
Yinghai Lucd794812010-10-11 12:34:09 -07001707static int __init_memblock memblock_search(struct memblock_type *type, phys_addr_t addr)
Benjamin Herrenschmidt72d4b0b2010-08-04 14:38:47 +10001708{
1709 unsigned int left = 0, right = type->cnt;
1710
1711 do {
1712 unsigned int mid = (right + left) / 2;
1713
1714 if (addr < type->regions[mid].base)
1715 right = mid;
1716 else if (addr >= (type->regions[mid].base +
1717 type->regions[mid].size))
1718 left = mid + 1;
1719 else
1720 return mid;
1721 } while (left < right);
1722 return -1;
1723}
1724
Yueyi Lif5a222d2018-12-14 14:17:06 -08001725bool __init_memblock memblock_is_reserved(phys_addr_t addr)
Yinghai Lu95f72d12010-07-12 14:36:09 +10001726{
Benjamin Herrenschmidt72d4b0b2010-08-04 14:38:47 +10001727 return memblock_search(&memblock.reserved, addr) != -1;
1728}
Yinghai Lu95f72d12010-07-12 14:36:09 +10001729
Yaowei Baib4ad0c72016-01-14 15:18:54 -08001730bool __init_memblock memblock_is_memory(phys_addr_t addr)
Benjamin Herrenschmidt72d4b0b2010-08-04 14:38:47 +10001731{
1732 return memblock_search(&memblock.memory, addr) != -1;
1733}
1734
Yaowei Bai937f0c22018-02-06 15:41:18 -08001735bool __init_memblock memblock_is_map_memory(phys_addr_t addr)
Ard Biesheuvelbf3d3cc2015-11-30 13:28:15 +01001736{
1737 int i = memblock_search(&memblock.memory, addr);
1738
1739 if (i == -1)
1740 return false;
1741 return !memblock_is_nomap(&memblock.memory.regions[i]);
1742}
1743
Yinghai Lue76b63f2013-09-11 14:22:17 -07001744int __init_memblock memblock_search_pfn_nid(unsigned long pfn,
1745 unsigned long *start_pfn, unsigned long *end_pfn)
1746{
1747 struct memblock_type *type = &memblock.memory;
Fabian Frederick16763232014-04-07 15:37:53 -07001748 int mid = memblock_search(type, PFN_PHYS(pfn));
Yinghai Lue76b63f2013-09-11 14:22:17 -07001749
1750 if (mid == -1)
1751 return -1;
1752
Fabian Frederickf7e2f7e2014-06-04 16:07:51 -07001753 *start_pfn = PFN_DOWN(type->regions[mid].base);
1754 *end_pfn = PFN_DOWN(type->regions[mid].base + type->regions[mid].size);
Yinghai Lue76b63f2013-09-11 14:22:17 -07001755
Mike Rapoportd622abf2020-06-03 15:56:53 -07001756 return memblock_get_region_node(&type->regions[mid]);
Yinghai Lue76b63f2013-09-11 14:22:17 -07001757}
Yinghai Lue76b63f2013-09-11 14:22:17 -07001758
Stephen Boydeab30942012-05-24 00:45:21 -07001759/**
1760 * memblock_is_region_memory - check if a region is a subset of memory
1761 * @base: base of region to check
1762 * @size: size of region to check
1763 *
Mike Rapoport47cec442018-06-30 17:55:02 +03001764 * Check if the region [@base, @base + @size) is a subset of a memory block.
Stephen Boydeab30942012-05-24 00:45:21 -07001765 *
Mike Rapoport47cec442018-06-30 17:55:02 +03001766 * Return:
Stephen Boydeab30942012-05-24 00:45:21 -07001767 * 0 if false, non-zero if true
1768 */
Yaowei Bai937f0c22018-02-06 15:41:18 -08001769bool __init_memblock memblock_is_region_memory(phys_addr_t base, phys_addr_t size)
Benjamin Herrenschmidt72d4b0b2010-08-04 14:38:47 +10001770{
Tomi Valkeinenabb65272011-01-20 14:44:20 -08001771 int idx = memblock_search(&memblock.memory, base);
Tejun Heoeb18f1b2011-12-08 10:22:07 -08001772 phys_addr_t end = base + memblock_cap_size(base, &size);
Benjamin Herrenschmidt72d4b0b2010-08-04 14:38:47 +10001773
1774 if (idx == -1)
Yaowei Bai937f0c22018-02-06 15:41:18 -08001775 return false;
Wei Yangef415ef2017-02-22 15:45:04 -08001776 return (memblock.memory.regions[idx].base +
Tejun Heoeb18f1b2011-12-08 10:22:07 -08001777 memblock.memory.regions[idx].size) >= end;
Yinghai Lu95f72d12010-07-12 14:36:09 +10001778}
1779
Stephen Boydeab30942012-05-24 00:45:21 -07001780/**
1781 * memblock_is_region_reserved - check if a region intersects reserved memory
1782 * @base: base of region to check
1783 * @size: size of region to check
1784 *
Mike Rapoport47cec442018-06-30 17:55:02 +03001785 * Check if the region [@base, @base + @size) intersects a reserved
1786 * memory block.
Stephen Boydeab30942012-05-24 00:45:21 -07001787 *
Mike Rapoport47cec442018-06-30 17:55:02 +03001788 * Return:
Tang Chenc5c5c9d2015-09-08 15:02:00 -07001789 * True if they intersect, false if not.
Stephen Boydeab30942012-05-24 00:45:21 -07001790 */
Tang Chenc5c5c9d2015-09-08 15:02:00 -07001791bool __init_memblock memblock_is_region_reserved(phys_addr_t base, phys_addr_t size)
Yinghai Lu95f72d12010-07-12 14:36:09 +10001792{
Tang Chenc5c5c9d2015-09-08 15:02:00 -07001793 return memblock_overlaps_region(&memblock.reserved, base, size);
Yinghai Lu95f72d12010-07-12 14:36:09 +10001794}
1795
Yinghai Lu6ede1fd2012-10-22 16:35:18 -07001796void __init_memblock memblock_trim_memory(phys_addr_t align)
1797{
Yinghai Lu6ede1fd2012-10-22 16:35:18 -07001798 phys_addr_t start, end, orig_start, orig_end;
Emil Medve136199f2014-04-07 15:37:52 -07001799 struct memblock_region *r;
Yinghai Lu6ede1fd2012-10-22 16:35:18 -07001800
Mike Rapoportcc6de162020-10-13 16:58:30 -07001801 for_each_mem_region(r) {
Emil Medve136199f2014-04-07 15:37:52 -07001802 orig_start = r->base;
1803 orig_end = r->base + r->size;
Yinghai Lu6ede1fd2012-10-22 16:35:18 -07001804 start = round_up(orig_start, align);
1805 end = round_down(orig_end, align);
1806
1807 if (start == orig_start && end == orig_end)
1808 continue;
1809
1810 if (start < end) {
Emil Medve136199f2014-04-07 15:37:52 -07001811 r->base = start;
1812 r->size = end - start;
Yinghai Lu6ede1fd2012-10-22 16:35:18 -07001813 } else {
Emil Medve136199f2014-04-07 15:37:52 -07001814 memblock_remove_region(&memblock.memory,
1815 r - memblock.memory.regions);
1816 r--;
Yinghai Lu6ede1fd2012-10-22 16:35:18 -07001817 }
1818 }
1819}
Benjamin Herrenschmidte63075a2010-07-06 15:39:01 -07001820
Yinghai Lu3661ca62010-09-15 13:05:29 -07001821void __init_memblock memblock_set_current_limit(phys_addr_t limit)
Benjamin Herrenschmidte63075a2010-07-06 15:39:01 -07001822{
1823 memblock.current_limit = limit;
1824}
1825
Laura Abbottfec51012014-02-27 01:23:43 +01001826phys_addr_t __init_memblock memblock_get_current_limit(void)
1827{
1828 return memblock.current_limit;
1829}
1830
Heiko Carstens0262d9c2017-02-24 14:55:59 -08001831static void __init_memblock memblock_dump(struct memblock_type *type)
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +10001832{
Miles Chen5d63f812017-02-22 15:46:42 -08001833 phys_addr_t base, end, size;
Mike Rapoporte1720fe2018-06-30 17:55:01 +03001834 enum memblock_flags flags;
Alexander Kuleshov8c9c1702016-01-14 15:20:42 -08001835 int idx;
1836 struct memblock_region *rgn;
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +10001837
Heiko Carstens0262d9c2017-02-24 14:55:59 -08001838 pr_info(" %s.cnt = 0x%lx\n", type->name, type->cnt);
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +10001839
Gioh Kim66e8b432017-11-15 17:33:42 -08001840 for_each_memblock_type(idx, type, rgn) {
Tejun Heo7c0caeb2011-07-14 11:43:42 +02001841 char nid_buf[32] = "";
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +10001842
Tejun Heo7c0caeb2011-07-14 11:43:42 +02001843 base = rgn->base;
1844 size = rgn->size;
Miles Chen5d63f812017-02-22 15:46:42 -08001845 end = base + size - 1;
Tang Chen66a20752014-01-21 15:49:20 -08001846 flags = rgn->flags;
Mike Rapoporta9ee6cf2021-06-28 19:43:01 -07001847#ifdef CONFIG_NUMA
Tejun Heo7c0caeb2011-07-14 11:43:42 +02001848 if (memblock_get_region_node(rgn) != MAX_NUMNODES)
1849 snprintf(nid_buf, sizeof(nid_buf), " on node %d",
1850 memblock_get_region_node(rgn));
1851#endif
Mike Rapoporte1720fe2018-06-30 17:55:01 +03001852 pr_info(" %s[%#x]\t[%pa-%pa], %pa bytes%s flags: %#x\n",
Heiko Carstens0262d9c2017-02-24 14:55:59 -08001853 type->name, idx, &base, &end, &size, nid_buf, flags);
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +10001854 }
1855}
1856
Mike Rapoport87c55872020-10-13 16:57:54 -07001857static void __init_memblock __memblock_dump_all(void)
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +10001858{
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +10001859 pr_info("MEMBLOCK configuration:\n");
Miles Chen5d63f812017-02-22 15:46:42 -08001860 pr_info(" memory size = %pa reserved size = %pa\n",
1861 &memblock.memory.total_size,
1862 &memblock.reserved.total_size);
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +10001863
Heiko Carstens0262d9c2017-02-24 14:55:59 -08001864 memblock_dump(&memblock.memory);
1865 memblock_dump(&memblock.reserved);
Heiko Carstens409efd42017-02-24 14:55:56 -08001866#ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
David Hildenbrand77649902020-07-01 16:18:29 +02001867 memblock_dump(&physmem);
Heiko Carstens409efd42017-02-24 14:55:56 -08001868#endif
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +10001869}
1870
Mike Rapoport87c55872020-10-13 16:57:54 -07001871void __init_memblock memblock_dump_all(void)
1872{
1873 if (memblock_debug)
1874 __memblock_dump_all();
1875}
1876
Tejun Heo1aadc052011-12-08 10:22:08 -08001877void __init memblock_allow_resize(void)
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +10001878{
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -07001879 memblock_can_resize = 1;
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +10001880}
1881
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +10001882static int __init early_memblock(char *p)
1883{
1884 if (p && strstr(p, "debug"))
1885 memblock_debug = 1;
1886 return 0;
1887}
1888early_param("memblock", early_memblock);
1889
Mike Rapoport4f5b0c12020-12-14 19:09:59 -08001890static void __init free_memmap(unsigned long start_pfn, unsigned long end_pfn)
1891{
1892 struct page *start_pg, *end_pg;
1893 phys_addr_t pg, pgend;
1894
1895 /*
1896 * Convert start_pfn/end_pfn to a struct page pointer.
1897 */
1898 start_pg = pfn_to_page(start_pfn - 1) + 1;
1899 end_pg = pfn_to_page(end_pfn - 1) + 1;
1900
1901 /*
1902 * Convert to physical addresses, and round start upwards and end
1903 * downwards.
1904 */
1905 pg = PAGE_ALIGN(__pa(start_pg));
1906 pgend = __pa(end_pg) & PAGE_MASK;
1907
1908 /*
1909 * If there are free pages between these, free the section of the
1910 * memmap array.
1911 */
1912 if (pg < pgend)
1913 memblock_free(pg, pgend - pg);
1914}
1915
1916/*
1917 * The mem_map array can get very big. Free the unused area of the memory map.
1918 */
1919static void __init free_unused_memmap(void)
1920{
1921 unsigned long start, end, prev_end = 0;
1922 int i;
1923
1924 if (!IS_ENABLED(CONFIG_HAVE_ARCH_PFN_VALID) ||
1925 IS_ENABLED(CONFIG_SPARSEMEM_VMEMMAP))
1926 return;
1927
1928 /*
1929 * This relies on each bank being in address order.
1930 * The banks are sorted previously in bootmem_init().
1931 */
1932 for_each_mem_pfn_range(i, MAX_NUMNODES, &start, &end, NULL) {
1933#ifdef CONFIG_SPARSEMEM
1934 /*
1935 * Take care not to free memmap entries that don't exist
1936 * due to SPARSEMEM sections which aren't present.
1937 */
1938 start = min(start, ALIGN(prev_end, PAGES_PER_SECTION));
Mike Rapoport4f5b0c12020-12-14 19:09:59 -08001939#endif
Mike Rapoport4f5b0c12020-12-14 19:09:59 -08001940 /*
Mike Rapoporte2a86802021-05-17 21:15:15 +03001941 * Align down here since many operations in VM subsystem
1942 * presume that there are no holes in the memory map inside
1943 * a pageblock
Mike Rapoport4f5b0c12020-12-14 19:09:59 -08001944 */
Mike Rapoporte2a86802021-05-17 21:15:15 +03001945 start = round_down(start, pageblock_nr_pages);
Mike Rapoport4f5b0c12020-12-14 19:09:59 -08001946
1947 /*
1948 * If we had a previous bank, and there is a space
1949 * between the current bank and the previous, free it.
1950 */
1951 if (prev_end && prev_end < start)
1952 free_memmap(prev_end, start);
1953
1954 /*
Mike Rapoporte2a86802021-05-17 21:15:15 +03001955 * Align up here since many operations in VM subsystem
1956 * presume that there are no holes in the memory map inside
1957 * a pageblock
Mike Rapoport4f5b0c12020-12-14 19:09:59 -08001958 */
Mike Rapoporte2a86802021-05-17 21:15:15 +03001959 prev_end = ALIGN(end, pageblock_nr_pages);
Mike Rapoport4f5b0c12020-12-14 19:09:59 -08001960 }
1961
1962#ifdef CONFIG_SPARSEMEM
Mike Rapoportf921f532021-05-17 21:31:59 +03001963 if (!IS_ALIGNED(prev_end, PAGES_PER_SECTION)) {
1964 prev_end = ALIGN(end, pageblock_nr_pages);
Mike Rapoport4f5b0c12020-12-14 19:09:59 -08001965 free_memmap(prev_end, ALIGN(prev_end, PAGES_PER_SECTION));
Mike Rapoportf921f532021-05-17 21:31:59 +03001966 }
Mike Rapoport4f5b0c12020-12-14 19:09:59 -08001967#endif
1968}
1969
Mike Rapoportbda49a82018-10-30 15:09:40 -07001970static void __init __free_pages_memory(unsigned long start, unsigned long end)
1971{
1972 int order;
1973
1974 while (start < end) {
1975 order = min(MAX_ORDER - 1UL, __ffs(start));
1976
1977 while (start + (1UL << order) > end)
1978 order--;
1979
1980 memblock_free_pages(pfn_to_page(start), start, order);
1981
1982 start += (1UL << order);
1983 }
1984}
1985
1986static unsigned long __init __free_memory_core(phys_addr_t start,
1987 phys_addr_t end)
1988{
1989 unsigned long start_pfn = PFN_UP(start);
1990 unsigned long end_pfn = min_t(unsigned long,
1991 PFN_DOWN(end), max_low_pfn);
1992
1993 if (start_pfn >= end_pfn)
1994 return 0;
1995
1996 __free_pages_memory(start_pfn, end_pfn);
1997
1998 return end_pfn - start_pfn;
1999}
2000
Mike Rapoport9092d4f2021-06-30 18:51:16 -07002001static void __init memmap_init_reserved_pages(void)
2002{
2003 struct memblock_region *region;
2004 phys_addr_t start, end;
2005 u64 i;
2006
2007 /* initialize struct pages for the reserved regions */
2008 for_each_reserved_mem_range(i, &start, &end)
2009 reserve_bootmem_region(start, end);
2010
2011 /* and also treat struct pages for the NOMAP regions as PageReserved */
2012 for_each_mem_region(region) {
2013 if (memblock_is_nomap(region)) {
2014 start = region->base;
2015 end = start + region->size;
2016 reserve_bootmem_region(start, end);
2017 }
2018 }
2019}
2020
Mike Rapoportbda49a82018-10-30 15:09:40 -07002021static unsigned long __init free_low_memory_core_early(void)
2022{
2023 unsigned long count = 0;
2024 phys_addr_t start, end;
2025 u64 i;
2026
2027 memblock_clear_hotplug(0, -1);
2028
Mike Rapoport9092d4f2021-06-30 18:51:16 -07002029 memmap_init_reserved_pages();
Mike Rapoportbda49a82018-10-30 15:09:40 -07002030
2031 /*
2032 * We need to use NUMA_NO_NODE instead of NODE_DATA(0)->node_id
2033 * because in some case like Node0 doesn't have RAM installed
2034 * low ram will be on Node1
2035 */
2036 for_each_free_mem_range(i, NUMA_NO_NODE, MEMBLOCK_NONE, &start, &end,
2037 NULL)
2038 count += __free_memory_core(start, end);
2039
2040 return count;
2041}
2042
2043static int reset_managed_pages_done __initdata;
2044
2045void reset_node_managed_pages(pg_data_t *pgdat)
2046{
2047 struct zone *z;
2048
2049 for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++)
Arun KS9705bea2018-12-28 00:34:24 -08002050 atomic_long_set(&z->managed_pages, 0);
Mike Rapoportbda49a82018-10-30 15:09:40 -07002051}
2052
2053void __init reset_all_zones_managed_pages(void)
2054{
2055 struct pglist_data *pgdat;
2056
2057 if (reset_managed_pages_done)
2058 return;
2059
2060 for_each_online_pgdat(pgdat)
2061 reset_node_managed_pages(pgdat);
2062
2063 reset_managed_pages_done = 1;
2064}
2065
2066/**
2067 * memblock_free_all - release free pages to the buddy allocator
Mike Rapoportbda49a82018-10-30 15:09:40 -07002068 */
Daeseok Youn097d43d2021-01-14 16:08:17 +09002069void __init memblock_free_all(void)
Mike Rapoportbda49a82018-10-30 15:09:40 -07002070{
2071 unsigned long pages;
2072
Mike Rapoport4f5b0c12020-12-14 19:09:59 -08002073 free_unused_memmap();
Mike Rapoportbda49a82018-10-30 15:09:40 -07002074 reset_all_zones_managed_pages();
2075
2076 pages = free_low_memory_core_early();
Arun KSca79b0c2018-12-28 00:34:29 -08002077 totalram_pages_add(pages);
Mike Rapoportbda49a82018-10-30 15:09:40 -07002078}
2079
Mike Rapoport350e88b2019-05-13 17:22:59 -07002080#if defined(CONFIG_DEBUG_FS) && defined(CONFIG_ARCH_KEEP_MEMBLOCK)
Benjamin Herrenschmidt6d03b882010-07-06 15:39:19 -07002081
2082static int memblock_debug_show(struct seq_file *m, void *private)
2083{
2084 struct memblock_type *type = m->private;
2085 struct memblock_region *reg;
2086 int i;
Miles Chen5d63f812017-02-22 15:46:42 -08002087 phys_addr_t end;
Benjamin Herrenschmidt6d03b882010-07-06 15:39:19 -07002088
2089 for (i = 0; i < type->cnt; i++) {
2090 reg = &type->regions[i];
Miles Chen5d63f812017-02-22 15:46:42 -08002091 end = reg->base + reg->size - 1;
Benjamin Herrenschmidt6d03b882010-07-06 15:39:19 -07002092
Miles Chen5d63f812017-02-22 15:46:42 -08002093 seq_printf(m, "%4d: ", i);
2094 seq_printf(m, "%pa..%pa\n", &reg->base, &end);
Benjamin Herrenschmidt6d03b882010-07-06 15:39:19 -07002095 }
2096 return 0;
2097}
Andy Shevchenko5ad35092018-04-05 16:23:16 -07002098DEFINE_SHOW_ATTRIBUTE(memblock_debug);
Benjamin Herrenschmidt6d03b882010-07-06 15:39:19 -07002099
2100static int __init memblock_init_debugfs(void)
2101{
2102 struct dentry *root = debugfs_create_dir("memblock", NULL);
Greg Kroah-Hartmand9f79792019-03-05 15:46:09 -08002103
Joe Perches0825a6f2018-06-14 15:27:58 -07002104 debugfs_create_file("memory", 0444, root,
2105 &memblock.memory, &memblock_debug_fops);
2106 debugfs_create_file("reserved", 0444, root,
2107 &memblock.reserved, &memblock_debug_fops);
Philipp Hachtmann70210ed2014-01-29 18:16:01 +01002108#ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
David Hildenbrand77649902020-07-01 16:18:29 +02002109 debugfs_create_file("physmem", 0444, root, &physmem,
2110 &memblock_debug_fops);
Philipp Hachtmann70210ed2014-01-29 18:16:01 +01002111#endif
Benjamin Herrenschmidt6d03b882010-07-06 15:39:19 -07002112
2113 return 0;
2114}
2115__initcall(memblock_init_debugfs);
2116
2117#endif /* CONFIG_DEBUG_FS */