blob: b91df5cf54d346c59e856c3dd7fba6dc663ee39f [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)
Linus Torvalds77e02cf2021-09-14 13:23:22 -0700475 memblock_free_ptr(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{
Geert Uytterhoeven00974b92021-08-11 10:54:36 +0200668 phys_addr_t end = base + size - 1;
669
670 memblock_dbg("%s: [%pa-%pa] nid=%d %pS\n", __func__,
671 &base, &end, nid, (void *)_RET_IP_);
672
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +0100673 return memblock_add_range(&memblock.memory, base, size, nid, 0);
Tejun Heo7fb0bc32011-12-08 10:22:08 -0800674}
675
Mike Rapoport48a833c2018-06-30 17:55:03 +0300676/**
677 * memblock_add - add new memblock region
678 * @base: base address of the new region
679 * @size: size of the new region
680 *
681 * Add new memblock region [@base, @base + @size) to the "memory"
682 * type. See memblock_add_range() description for mode details
683 *
684 * Return:
685 * 0 on success, -errno on failure.
686 */
Alexander Kuleshovf705ac42016-05-20 16:57:35 -0700687int __init_memblock memblock_add(phys_addr_t base, phys_addr_t size)
Alexander Kuleshov6a4055b2015-04-15 16:14:44 -0700688{
Miles Chen5d63f812017-02-22 15:46:42 -0800689 phys_addr_t end = base + size - 1;
690
Anshuman Khanduala090d712020-01-30 22:14:23 -0800691 memblock_dbg("%s: [%pa-%pa] %pS\n", __func__,
Miles Chen5d63f812017-02-22 15:46:42 -0800692 &base, &end, (void *)_RET_IP_);
Alexander Kuleshov6a4055b2015-04-15 16:14:44 -0700693
Alexander Kuleshovf705ac42016-05-20 16:57:35 -0700694 return memblock_add_range(&memblock.memory, base, size, MAX_NUMNODES, 0);
Yinghai Lu95f72d12010-07-12 14:36:09 +1000695}
696
Tejun Heo6a9ceb32011-12-08 10:22:07 -0800697/**
698 * memblock_isolate_range - isolate given range into disjoint memblocks
699 * @type: memblock type to isolate range for
700 * @base: base of range to isolate
701 * @size: size of range to isolate
702 * @start_rgn: out parameter for the start of isolated region
703 * @end_rgn: out parameter for the end of isolated region
704 *
705 * Walk @type and ensure that regions don't cross the boundaries defined by
Mike Rapoport47cec442018-06-30 17:55:02 +0300706 * [@base, @base + @size). Crossing regions are split at the boundaries,
Tejun Heo6a9ceb32011-12-08 10:22:07 -0800707 * which may create at most two more regions. The index of the first
708 * region inside the range is returned in *@start_rgn and end in *@end_rgn.
709 *
Mike Rapoport47cec442018-06-30 17:55:02 +0300710 * Return:
Tejun Heo6a9ceb32011-12-08 10:22:07 -0800711 * 0 on success, -errno on failure.
712 */
713static int __init_memblock memblock_isolate_range(struct memblock_type *type,
714 phys_addr_t base, phys_addr_t size,
715 int *start_rgn, int *end_rgn)
716{
Tejun Heoeb18f1b2011-12-08 10:22:07 -0800717 phys_addr_t end = base + memblock_cap_size(base, &size);
Alexander Kuleshov8c9c1702016-01-14 15:20:42 -0800718 int idx;
719 struct memblock_region *rgn;
Tejun Heo6a9ceb32011-12-08 10:22:07 -0800720
721 *start_rgn = *end_rgn = 0;
722
Tejun Heob3dc6272012-04-20 08:31:34 -0700723 if (!size)
724 return 0;
725
Tejun Heo6a9ceb32011-12-08 10:22:07 -0800726 /* we'll create at most two more regions */
727 while (type->cnt + 2 > type->max)
Greg Pearson48c3b582012-06-20 12:53:05 -0700728 if (memblock_double_array(type, base, size) < 0)
Tejun Heo6a9ceb32011-12-08 10:22:07 -0800729 return -ENOMEM;
730
Gioh Kim66e8b432017-11-15 17:33:42 -0800731 for_each_memblock_type(idx, type, rgn) {
Tejun Heo6a9ceb32011-12-08 10:22:07 -0800732 phys_addr_t rbase = rgn->base;
733 phys_addr_t rend = rbase + rgn->size;
734
735 if (rbase >= end)
736 break;
737 if (rend <= base)
738 continue;
739
740 if (rbase < base) {
741 /*
742 * @rgn intersects from below. Split and continue
743 * to process the next region - the new top half.
744 */
745 rgn->base = base;
Tejun Heo1440c4e2011-12-08 10:22:08 -0800746 rgn->size -= base - rbase;
747 type->total_size -= base - rbase;
Alexander Kuleshov8c9c1702016-01-14 15:20:42 -0800748 memblock_insert_region(type, idx, rbase, base - rbase,
Tang Chen66a20752014-01-21 15:49:20 -0800749 memblock_get_region_node(rgn),
750 rgn->flags);
Tejun Heo6a9ceb32011-12-08 10:22:07 -0800751 } else if (rend > end) {
752 /*
753 * @rgn intersects from above. Split and redo the
754 * current region - the new bottom half.
755 */
756 rgn->base = end;
Tejun Heo1440c4e2011-12-08 10:22:08 -0800757 rgn->size -= end - rbase;
758 type->total_size -= end - rbase;
Alexander Kuleshov8c9c1702016-01-14 15:20:42 -0800759 memblock_insert_region(type, idx--, rbase, end - rbase,
Tang Chen66a20752014-01-21 15:49:20 -0800760 memblock_get_region_node(rgn),
761 rgn->flags);
Tejun Heo6a9ceb32011-12-08 10:22:07 -0800762 } else {
763 /* @rgn is fully contained, record it */
764 if (!*end_rgn)
Alexander Kuleshov8c9c1702016-01-14 15:20:42 -0800765 *start_rgn = idx;
766 *end_rgn = idx + 1;
Tejun Heo6a9ceb32011-12-08 10:22:07 -0800767 }
768 }
769
770 return 0;
771}
Tejun Heo6a9ceb32011-12-08 10:22:07 -0800772
Alexander Kuleshov35bd16a2015-11-05 18:47:00 -0800773static int __init_memblock memblock_remove_range(struct memblock_type *type,
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +0100774 phys_addr_t base, phys_addr_t size)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000775{
Tejun Heo71936182011-12-08 10:22:07 -0800776 int start_rgn, end_rgn;
777 int i, ret;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000778
Tejun Heo71936182011-12-08 10:22:07 -0800779 ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
780 if (ret)
781 return ret;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000782
Tejun Heo71936182011-12-08 10:22:07 -0800783 for (i = end_rgn - 1; i >= start_rgn; i--)
784 memblock_remove_region(type, i);
Benjamin Herrenschmidt8f7a6602011-03-22 16:33:43 -0700785 return 0;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000786}
787
Tejun Heo581adcb2011-12-08 10:22:06 -0800788int __init_memblock memblock_remove(phys_addr_t base, phys_addr_t size)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000789{
Minchan Kim25cf23d2018-06-07 17:07:35 -0700790 phys_addr_t end = base + size - 1;
791
Anshuman Khanduala090d712020-01-30 22:14:23 -0800792 memblock_dbg("%s: [%pa-%pa] %pS\n", __func__,
Minchan Kim25cf23d2018-06-07 17:07:35 -0700793 &base, &end, (void *)_RET_IP_);
794
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +0100795 return memblock_remove_range(&memblock.memory, base, size);
Yinghai Lu95f72d12010-07-12 14:36:09 +1000796}
797
Mike Rapoport4d728682018-12-28 00:35:29 -0800798/**
Linus Torvalds77e02cf2021-09-14 13:23:22 -0700799 * memblock_free_ptr - free boot memory allocation
800 * @ptr: starting address of the boot memory allocation
801 * @size: size of the boot memory block in bytes
802 *
803 * Free boot memory block previously allocated by memblock_alloc_xx() API.
804 * The freeing memory will not be released to the buddy allocator.
805 */
806void __init_memblock memblock_free_ptr(void *ptr, size_t size)
807{
808 if (ptr)
809 memblock_free(__pa(ptr), size);
810}
811
812/**
Mike Rapoport4d728682018-12-28 00:35:29 -0800813 * memblock_free - free boot memory block
814 * @base: phys starting address of the boot memory block
815 * @size: size of the boot memory block in bytes
816 *
817 * Free boot memory block previously allocated by memblock_alloc_xx() API.
818 * The freeing memory will not be released to the buddy allocator.
819 */
Tejun Heo581adcb2011-12-08 10:22:06 -0800820int __init_memblock memblock_free(phys_addr_t base, phys_addr_t size)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000821{
Miles Chen5d63f812017-02-22 15:46:42 -0800822 phys_addr_t end = base + size - 1;
823
Anshuman Khanduala090d712020-01-30 22:14:23 -0800824 memblock_dbg("%s: [%pa-%pa] %pS\n", __func__,
Miles Chen5d63f812017-02-22 15:46:42 -0800825 &base, &end, (void *)_RET_IP_);
Tejun Heo24aa0782011-07-12 11:16:06 +0200826
Catalin Marinas9099dae2016-10-11 13:55:11 -0700827 kmemleak_free_part_phys(base, size);
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +0100828 return memblock_remove_range(&memblock.reserved, base, size);
Yinghai Lu95f72d12010-07-12 14:36:09 +1000829}
830
Alexander Kuleshovf705ac42016-05-20 16:57:35 -0700831int __init_memblock memblock_reserve(phys_addr_t base, phys_addr_t size)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000832{
Miles Chen5d63f812017-02-22 15:46:42 -0800833 phys_addr_t end = base + size - 1;
834
Anshuman Khanduala090d712020-01-30 22:14:23 -0800835 memblock_dbg("%s: [%pa-%pa] %pS\n", __func__,
Miles Chen5d63f812017-02-22 15:46:42 -0800836 &base, &end, (void *)_RET_IP_);
Yinghai Lu95f72d12010-07-12 14:36:09 +1000837
Alexander Kuleshovf705ac42016-05-20 16:57:35 -0700838 return memblock_add_range(&memblock.reserved, base, size, MAX_NUMNODES, 0);
Yinghai Lu95f72d12010-07-12 14:36:09 +1000839}
840
Anshuman Khandual02634a42020-01-30 22:14:20 -0800841#ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
842int __init_memblock memblock_physmem_add(phys_addr_t base, phys_addr_t size)
843{
844 phys_addr_t end = base + size - 1;
845
846 memblock_dbg("%s: [%pa-%pa] %pS\n", __func__,
847 &base, &end, (void *)_RET_IP_);
848
David Hildenbrand77649902020-07-01 16:18:29 +0200849 return memblock_add_range(&physmem, base, size, MAX_NUMNODES, 0);
Anshuman Khandual02634a42020-01-30 22:14:20 -0800850}
851#endif
852
Tejun Heo35fd0802011-07-12 11:15:59 +0200853/**
Mike Rapoport47cec442018-06-30 17:55:02 +0300854 * memblock_setclr_flag - set or clear flag for a memory region
855 * @base: base address of the region
856 * @size: size of the region
857 * @set: set or clear the flag
Haitao Shi8958b242020-12-15 20:47:26 -0800858 * @flag: the flag to update
Tang Chen66b16ed2014-01-21 15:49:23 -0800859 *
Tony Luck4308ce12014-12-12 16:54:59 -0800860 * This function isolates region [@base, @base + @size), and sets/clears flag
Tang Chen66b16ed2014-01-21 15:49:23 -0800861 *
Mike Rapoport47cec442018-06-30 17:55:02 +0300862 * Return: 0 on success, -errno on failure.
Tang Chen66b16ed2014-01-21 15:49:23 -0800863 */
Tony Luck4308ce12014-12-12 16:54:59 -0800864static int __init_memblock memblock_setclr_flag(phys_addr_t base,
865 phys_addr_t size, int set, int flag)
Tang Chen66b16ed2014-01-21 15:49:23 -0800866{
867 struct memblock_type *type = &memblock.memory;
868 int i, ret, start_rgn, end_rgn;
869
870 ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
871 if (ret)
872 return ret;
873
Mike Rapoportfe145122019-03-11 23:30:46 -0700874 for (i = start_rgn; i < end_rgn; i++) {
875 struct memblock_region *r = &type->regions[i];
876
Tony Luck4308ce12014-12-12 16:54:59 -0800877 if (set)
Mike Rapoportfe145122019-03-11 23:30:46 -0700878 r->flags |= flag;
Tony Luck4308ce12014-12-12 16:54:59 -0800879 else
Mike Rapoportfe145122019-03-11 23:30:46 -0700880 r->flags &= ~flag;
881 }
Tang Chen66b16ed2014-01-21 15:49:23 -0800882
883 memblock_merge_regions(type);
884 return 0;
885}
886
887/**
Tony Luck4308ce12014-12-12 16:54:59 -0800888 * memblock_mark_hotplug - Mark hotpluggable memory with flag MEMBLOCK_HOTPLUG.
889 * @base: the base phys addr of the region
890 * @size: the size of the region
891 *
Mike Rapoport47cec442018-06-30 17:55:02 +0300892 * Return: 0 on success, -errno on failure.
Tony Luck4308ce12014-12-12 16:54:59 -0800893 */
894int __init_memblock memblock_mark_hotplug(phys_addr_t base, phys_addr_t size)
895{
896 return memblock_setclr_flag(base, size, 1, MEMBLOCK_HOTPLUG);
897}
898
899/**
Tang Chen66b16ed2014-01-21 15:49:23 -0800900 * memblock_clear_hotplug - Clear flag MEMBLOCK_HOTPLUG for a specified region.
901 * @base: the base phys addr of the region
902 * @size: the size of the region
903 *
Mike Rapoport47cec442018-06-30 17:55:02 +0300904 * Return: 0 on success, -errno on failure.
Tang Chen66b16ed2014-01-21 15:49:23 -0800905 */
906int __init_memblock memblock_clear_hotplug(phys_addr_t base, phys_addr_t size)
907{
Tony Luck4308ce12014-12-12 16:54:59 -0800908 return memblock_setclr_flag(base, size, 0, MEMBLOCK_HOTPLUG);
Tang Chen66b16ed2014-01-21 15:49:23 -0800909}
910
911/**
Tony Lucka3f5baf2015-06-24 16:58:12 -0700912 * memblock_mark_mirror - Mark mirrored memory with flag MEMBLOCK_MIRROR.
913 * @base: the base phys addr of the region
914 * @size: the size of the region
915 *
Mike Rapoport47cec442018-06-30 17:55:02 +0300916 * Return: 0 on success, -errno on failure.
Tony Lucka3f5baf2015-06-24 16:58:12 -0700917 */
918int __init_memblock memblock_mark_mirror(phys_addr_t base, phys_addr_t size)
919{
920 system_has_some_mirror = true;
921
922 return memblock_setclr_flag(base, size, 1, MEMBLOCK_MIRROR);
923}
924
Ard Biesheuvelbf3d3cc2015-11-30 13:28:15 +0100925/**
926 * memblock_mark_nomap - Mark a memory region with flag MEMBLOCK_NOMAP.
927 * @base: the base phys addr of the region
928 * @size: the size of the region
929 *
Mike Rapoport9092d4f2021-06-30 18:51:16 -0700930 * The memory regions marked with %MEMBLOCK_NOMAP will not be added to the
931 * direct mapping of the physical memory. These regions will still be
932 * covered by the memory map. The struct page representing NOMAP memory
933 * frames in the memory map will be PageReserved()
934 *
Mike Rapoport47cec442018-06-30 17:55:02 +0300935 * Return: 0 on success, -errno on failure.
Ard Biesheuvelbf3d3cc2015-11-30 13:28:15 +0100936 */
937int __init_memblock memblock_mark_nomap(phys_addr_t base, phys_addr_t size)
938{
Mike Rapoport6e44bd62021-10-13 08:36:59 +0300939 int ret = memblock_setclr_flag(base, size, 1, MEMBLOCK_NOMAP);
940
941 if (!ret)
942 kmemleak_free_part_phys(base, size);
943
944 return ret;
Ard Biesheuvelbf3d3cc2015-11-30 13:28:15 +0100945}
Tony Lucka3f5baf2015-06-24 16:58:12 -0700946
947/**
AKASHI Takahiro4c546b82017-04-03 11:23:54 +0900948 * memblock_clear_nomap - Clear flag MEMBLOCK_NOMAP for a specified region.
949 * @base: the base phys addr of the region
950 * @size: the size of the region
951 *
Mike Rapoport47cec442018-06-30 17:55:02 +0300952 * Return: 0 on success, -errno on failure.
AKASHI Takahiro4c546b82017-04-03 11:23:54 +0900953 */
954int __init_memblock memblock_clear_nomap(phys_addr_t base, phys_addr_t size)
955{
956 return memblock_setclr_flag(base, size, 0, MEMBLOCK_NOMAP);
957}
958
Mike Rapoport9f3d5ea2020-10-13 16:58:25 -0700959static bool should_skip_region(struct memblock_type *type,
960 struct memblock_region *m,
961 int nid, int flags)
Mike Rapoportc9a688a2019-03-11 23:30:50 -0700962{
963 int m_nid = memblock_get_region_node(m);
964
Mike Rapoport9f3d5ea2020-10-13 16:58:25 -0700965 /* we never skip regions when iterating memblock.reserved or physmem */
966 if (type != memblock_memory)
967 return false;
968
Mike Rapoportc9a688a2019-03-11 23:30:50 -0700969 /* only memory regions are associated with nodes, check it */
970 if (nid != NUMA_NO_NODE && nid != m_nid)
971 return true;
972
973 /* skip hotpluggable memory regions if needed */
Mike Rapoport79e482e2021-07-23 15:50:26 -0700974 if (movable_node_is_enabled() && memblock_is_hotpluggable(m) &&
975 !(flags & MEMBLOCK_HOTPLUG))
Mike Rapoportc9a688a2019-03-11 23:30:50 -0700976 return true;
977
978 /* if we want mirror memory skip non-mirror memory regions */
979 if ((flags & MEMBLOCK_MIRROR) && !memblock_is_mirror(m))
980 return true;
981
982 /* skip nomap memory unless we were asked for it explicitly */
983 if (!(flags & MEMBLOCK_NOMAP) && memblock_is_nomap(m))
984 return true;
985
986 return false;
987}
988
Robin Holt8e7a7f82015-06-30 14:56:41 -0700989/**
Mike Rapoporta2974132019-03-11 23:30:54 -0700990 * __next_mem_range - next function for for_each_free_mem_range() etc.
Tejun Heo35fd0802011-07-12 11:15:59 +0200991 * @idx: pointer to u64 loop variable
Grygorii Strashkob1154232014-01-21 15:50:16 -0800992 * @nid: node selector, %NUMA_NO_NODE for all nodes
Tony Luckfc6daaf2015-06-24 16:58:09 -0700993 * @flags: pick from blocks based on memory attributes
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +0100994 * @type_a: pointer to memblock_type from where the range is taken
995 * @type_b: pointer to memblock_type which excludes memory from being taken
Wanpeng Lidad75572012-06-20 12:53:01 -0700996 * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL
997 * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL
998 * @out_nid: ptr to int for nid of the range, can be %NULL
Tejun Heo35fd0802011-07-12 11:15:59 +0200999 *
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +01001000 * Find the first area from *@idx which matches @nid, fill the out
Tejun Heo35fd0802011-07-12 11:15:59 +02001001 * parameters, and update *@idx for the next iteration. The lower 32bit of
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +01001002 * *@idx contains index into type_a and the upper 32bit indexes the
1003 * areas before each region in type_b. For example, if type_b regions
Tejun Heo35fd0802011-07-12 11:15:59 +02001004 * look like the following,
1005 *
1006 * 0:[0-16), 1:[32-48), 2:[128-130)
1007 *
1008 * The upper 32bit indexes the following regions.
1009 *
1010 * 0:[0-0), 1:[16-32), 2:[48-128), 3:[130-MAX)
1011 *
1012 * As both region arrays are sorted, the function advances the two indices
1013 * in lockstep and returns each intersection.
1014 */
David Hildenbrand77649902020-07-01 16:18:29 +02001015void __next_mem_range(u64 *idx, int nid, enum memblock_flags flags,
1016 struct memblock_type *type_a,
1017 struct memblock_type *type_b, phys_addr_t *out_start,
1018 phys_addr_t *out_end, int *out_nid)
Tejun Heo35fd0802011-07-12 11:15:59 +02001019{
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +01001020 int idx_a = *idx & 0xffffffff;
1021 int idx_b = *idx >> 32;
Grygorii Strashkob1154232014-01-21 15:50:16 -08001022
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +01001023 if (WARN_ONCE(nid == MAX_NUMNODES,
1024 "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
Grygorii Strashko560dca272014-01-21 15:50:55 -08001025 nid = NUMA_NO_NODE;
Tejun Heo35fd0802011-07-12 11:15:59 +02001026
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +01001027 for (; idx_a < type_a->cnt; idx_a++) {
1028 struct memblock_region *m = &type_a->regions[idx_a];
1029
Tejun Heo35fd0802011-07-12 11:15:59 +02001030 phys_addr_t m_start = m->base;
1031 phys_addr_t m_end = m->base + m->size;
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +01001032 int m_nid = memblock_get_region_node(m);
Tejun Heo35fd0802011-07-12 11:15:59 +02001033
Mike Rapoport9f3d5ea2020-10-13 16:58:25 -07001034 if (should_skip_region(type_a, m, nid, flags))
Ard Biesheuvelbf3d3cc2015-11-30 13:28:15 +01001035 continue;
1036
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +01001037 if (!type_b) {
1038 if (out_start)
1039 *out_start = m_start;
1040 if (out_end)
1041 *out_end = m_end;
1042 if (out_nid)
1043 *out_nid = m_nid;
1044 idx_a++;
1045 *idx = (u32)idx_a | (u64)idx_b << 32;
1046 return;
1047 }
Tejun Heo35fd0802011-07-12 11:15:59 +02001048
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +01001049 /* scan areas before each reservation */
1050 for (; idx_b < type_b->cnt + 1; idx_b++) {
1051 struct memblock_region *r;
1052 phys_addr_t r_start;
1053 phys_addr_t r_end;
1054
1055 r = &type_b->regions[idx_b];
1056 r_start = idx_b ? r[-1].base + r[-1].size : 0;
1057 r_end = idx_b < type_b->cnt ?
Stefan Agner1c4bc432018-06-07 17:06:15 -07001058 r->base : PHYS_ADDR_MAX;
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +01001059
1060 /*
1061 * if idx_b advanced past idx_a,
1062 * break out to advance idx_a
1063 */
Tejun Heo35fd0802011-07-12 11:15:59 +02001064 if (r_start >= m_end)
1065 break;
1066 /* if the two regions intersect, we're done */
1067 if (m_start < r_end) {
1068 if (out_start)
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +01001069 *out_start =
1070 max(m_start, r_start);
Tejun Heo35fd0802011-07-12 11:15:59 +02001071 if (out_end)
1072 *out_end = min(m_end, r_end);
1073 if (out_nid)
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +01001074 *out_nid = m_nid;
Tejun Heo35fd0802011-07-12 11:15:59 +02001075 /*
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +01001076 * The region which ends first is
1077 * advanced for the next iteration.
Tejun Heo35fd0802011-07-12 11:15:59 +02001078 */
1079 if (m_end <= r_end)
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +01001080 idx_a++;
Tejun Heo35fd0802011-07-12 11:15:59 +02001081 else
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +01001082 idx_b++;
1083 *idx = (u32)idx_a | (u64)idx_b << 32;
Tejun Heo35fd0802011-07-12 11:15:59 +02001084 return;
1085 }
1086 }
1087 }
1088
1089 /* signal end of iteration */
1090 *idx = ULLONG_MAX;
1091}
1092
Tejun Heo7bd0b0f2011-12-08 10:22:09 -08001093/**
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +01001094 * __next_mem_range_rev - generic next function for for_each_*_range_rev()
1095 *
Tejun Heo7bd0b0f2011-12-08 10:22:09 -08001096 * @idx: pointer to u64 loop variable
Alexander Kuleshovad5ea8c2015-09-08 15:04:22 -07001097 * @nid: node selector, %NUMA_NO_NODE for all nodes
Tony Luckfc6daaf2015-06-24 16:58:09 -07001098 * @flags: pick from blocks based on memory attributes
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +01001099 * @type_a: pointer to memblock_type from where the range is taken
1100 * @type_b: pointer to memblock_type which excludes memory from being taken
Wanpeng Lidad75572012-06-20 12:53:01 -07001101 * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL
1102 * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL
1103 * @out_nid: ptr to int for nid of the range, can be %NULL
Tejun Heo7bd0b0f2011-12-08 10:22:09 -08001104 *
Mike Rapoport47cec442018-06-30 17:55:02 +03001105 * Finds the next range from type_a which is not marked as unsuitable
1106 * in type_b.
1107 *
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +01001108 * Reverse of __next_mem_range().
Tejun Heo7bd0b0f2011-12-08 10:22:09 -08001109 */
Mike Rapoporte1720fe2018-06-30 17:55:01 +03001110void __init_memblock __next_mem_range_rev(u64 *idx, int nid,
1111 enum memblock_flags flags,
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +01001112 struct memblock_type *type_a,
1113 struct memblock_type *type_b,
1114 phys_addr_t *out_start,
1115 phys_addr_t *out_end, int *out_nid)
Tejun Heo7bd0b0f2011-12-08 10:22:09 -08001116{
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +01001117 int idx_a = *idx & 0xffffffff;
1118 int idx_b = *idx >> 32;
Grygorii Strashkob1154232014-01-21 15:50:16 -08001119
Grygorii Strashko560dca272014-01-21 15:50:55 -08001120 if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
1121 nid = NUMA_NO_NODE;
Tejun Heo7bd0b0f2011-12-08 10:22:09 -08001122
1123 if (*idx == (u64)ULLONG_MAX) {
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +01001124 idx_a = type_a->cnt - 1;
zijun_hue47608a2016-08-04 15:32:00 -07001125 if (type_b != NULL)
1126 idx_b = type_b->cnt;
1127 else
1128 idx_b = 0;
Tejun Heo7bd0b0f2011-12-08 10:22:09 -08001129 }
1130
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +01001131 for (; idx_a >= 0; idx_a--) {
1132 struct memblock_region *m = &type_a->regions[idx_a];
1133
Tejun Heo7bd0b0f2011-12-08 10:22:09 -08001134 phys_addr_t m_start = m->base;
1135 phys_addr_t m_end = m->base + m->size;
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +01001136 int m_nid = memblock_get_region_node(m);
Tejun Heo7bd0b0f2011-12-08 10:22:09 -08001137
Mike Rapoport9f3d5ea2020-10-13 16:58:25 -07001138 if (should_skip_region(type_a, m, nid, flags))
Ard Biesheuvelbf3d3cc2015-11-30 13:28:15 +01001139 continue;
1140
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +01001141 if (!type_b) {
1142 if (out_start)
1143 *out_start = m_start;
1144 if (out_end)
1145 *out_end = m_end;
1146 if (out_nid)
1147 *out_nid = m_nid;
zijun_hufb399b42016-07-28 15:48:56 -07001148 idx_a--;
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +01001149 *idx = (u32)idx_a | (u64)idx_b << 32;
1150 return;
1151 }
Tejun Heo7bd0b0f2011-12-08 10:22:09 -08001152
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +01001153 /* scan areas before each reservation */
1154 for (; idx_b >= 0; idx_b--) {
1155 struct memblock_region *r;
1156 phys_addr_t r_start;
1157 phys_addr_t r_end;
1158
1159 r = &type_b->regions[idx_b];
1160 r_start = idx_b ? r[-1].base + r[-1].size : 0;
1161 r_end = idx_b < type_b->cnt ?
Stefan Agner1c4bc432018-06-07 17:06:15 -07001162 r->base : PHYS_ADDR_MAX;
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +01001163 /*
1164 * if idx_b advanced past idx_a,
1165 * break out to advance idx_a
1166 */
1167
Tejun Heo7bd0b0f2011-12-08 10:22:09 -08001168 if (r_end <= m_start)
1169 break;
1170 /* if the two regions intersect, we're done */
1171 if (m_end > r_start) {
1172 if (out_start)
1173 *out_start = max(m_start, r_start);
1174 if (out_end)
1175 *out_end = min(m_end, r_end);
1176 if (out_nid)
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +01001177 *out_nid = m_nid;
Tejun Heo7bd0b0f2011-12-08 10:22:09 -08001178 if (m_start >= r_start)
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +01001179 idx_a--;
Tejun Heo7bd0b0f2011-12-08 10:22:09 -08001180 else
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +01001181 idx_b--;
1182 *idx = (u32)idx_a | (u64)idx_b << 32;
Tejun Heo7bd0b0f2011-12-08 10:22:09 -08001183 return;
1184 }
1185 }
1186 }
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +01001187 /* signal end of iteration */
Tejun Heo7bd0b0f2011-12-08 10:22:09 -08001188 *idx = ULLONG_MAX;
1189}
1190
Tejun Heo7c0caeb2011-07-14 11:43:42 +02001191/*
Chen Chang45e79812018-11-16 15:08:57 -08001192 * Common iterator interface used to define for_each_mem_pfn_range().
Tejun Heo7c0caeb2011-07-14 11:43:42 +02001193 */
1194void __init_memblock __next_mem_pfn_range(int *idx, int nid,
1195 unsigned long *out_start_pfn,
1196 unsigned long *out_end_pfn, int *out_nid)
1197{
1198 struct memblock_type *type = &memblock.memory;
1199 struct memblock_region *r;
Mike Rapoportd622abf2020-06-03 15:56:53 -07001200 int r_nid;
Tejun Heo7c0caeb2011-07-14 11:43:42 +02001201
1202 while (++*idx < type->cnt) {
1203 r = &type->regions[*idx];
Mike Rapoportd622abf2020-06-03 15:56:53 -07001204 r_nid = memblock_get_region_node(r);
Tejun Heo7c0caeb2011-07-14 11:43:42 +02001205
1206 if (PFN_UP(r->base) >= PFN_DOWN(r->base + r->size))
1207 continue;
Mike Rapoportd622abf2020-06-03 15:56:53 -07001208 if (nid == MAX_NUMNODES || nid == r_nid)
Tejun Heo7c0caeb2011-07-14 11:43:42 +02001209 break;
1210 }
1211 if (*idx >= type->cnt) {
1212 *idx = -1;
1213 return;
1214 }
1215
1216 if (out_start_pfn)
1217 *out_start_pfn = PFN_UP(r->base);
1218 if (out_end_pfn)
1219 *out_end_pfn = PFN_DOWN(r->base + r->size);
1220 if (out_nid)
Mike Rapoportd622abf2020-06-03 15:56:53 -07001221 *out_nid = r_nid;
Tejun Heo7c0caeb2011-07-14 11:43:42 +02001222}
1223
1224/**
1225 * memblock_set_node - set node ID on memblock regions
1226 * @base: base of area to set node ID for
1227 * @size: size of area to set node ID for
Tang Chene7e8de52014-01-21 15:49:26 -08001228 * @type: memblock type to set node ID for
Tejun Heo7c0caeb2011-07-14 11:43:42 +02001229 * @nid: node ID to set
1230 *
Mike Rapoport47cec442018-06-30 17:55:02 +03001231 * Set the nid of memblock @type regions in [@base, @base + @size) to @nid.
Tejun Heo7c0caeb2011-07-14 11:43:42 +02001232 * Regions which cross the area boundaries are split as necessary.
1233 *
Mike Rapoport47cec442018-06-30 17:55:02 +03001234 * Return:
Tejun Heo7c0caeb2011-07-14 11:43:42 +02001235 * 0 on success, -errno on failure.
1236 */
1237int __init_memblock memblock_set_node(phys_addr_t base, phys_addr_t size,
Tang Chene7e8de52014-01-21 15:49:26 -08001238 struct memblock_type *type, int nid)
Tejun Heo7c0caeb2011-07-14 11:43:42 +02001239{
Mike Rapoporta9ee6cf2021-06-28 19:43:01 -07001240#ifdef CONFIG_NUMA
Tejun Heo6a9ceb32011-12-08 10:22:07 -08001241 int start_rgn, end_rgn;
1242 int i, ret;
Tejun Heo7c0caeb2011-07-14 11:43:42 +02001243
Tejun Heo6a9ceb32011-12-08 10:22:07 -08001244 ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
1245 if (ret)
1246 return ret;
Tejun Heo7c0caeb2011-07-14 11:43:42 +02001247
Tejun Heo6a9ceb32011-12-08 10:22:07 -08001248 for (i = start_rgn; i < end_rgn; i++)
Wanpeng Lie9d24ad2012-10-08 16:32:21 -07001249 memblock_set_region_node(&type->regions[i], nid);
Tejun Heo7c0caeb2011-07-14 11:43:42 +02001250
1251 memblock_merge_regions(type);
Mike Rapoport3f08a302020-06-03 15:57:02 -07001252#endif
Tejun Heo7c0caeb2011-07-14 11:43:42 +02001253 return 0;
1254}
Mike Rapoport3f08a302020-06-03 15:57:02 -07001255
Alexander Duyck837566e2019-05-13 17:21:17 -07001256#ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1257/**
1258 * __next_mem_pfn_range_in_zone - iterator for for_each_*_range_in_zone()
1259 *
1260 * @idx: pointer to u64 loop variable
1261 * @zone: zone in which all of the memory blocks reside
1262 * @out_spfn: ptr to ulong for start pfn of the range, can be %NULL
1263 * @out_epfn: ptr to ulong for end pfn of the range, can be %NULL
1264 *
1265 * This function is meant to be a zone/pfn specific wrapper for the
1266 * for_each_mem_range type iterators. Specifically they are used in the
1267 * deferred memory init routines and as such we were duplicating much of
1268 * this logic throughout the code. So instead of having it in multiple
1269 * locations it seemed like it would make more sense to centralize this to
1270 * one new iterator that does everything they need.
1271 */
1272void __init_memblock
1273__next_mem_pfn_range_in_zone(u64 *idx, struct zone *zone,
1274 unsigned long *out_spfn, unsigned long *out_epfn)
1275{
1276 int zone_nid = zone_to_nid(zone);
1277 phys_addr_t spa, epa;
1278 int nid;
1279
1280 __next_mem_range(idx, zone_nid, MEMBLOCK_NONE,
1281 &memblock.memory, &memblock.reserved,
1282 &spa, &epa, &nid);
1283
1284 while (*idx != U64_MAX) {
1285 unsigned long epfn = PFN_DOWN(epa);
1286 unsigned long spfn = PFN_UP(spa);
1287
1288 /*
1289 * Verify the end is at least past the start of the zone and
1290 * that we have at least one PFN to initialize.
1291 */
1292 if (zone->zone_start_pfn < epfn && spfn < epfn) {
1293 /* if we went too far just stop searching */
1294 if (zone_end_pfn(zone) <= spfn) {
1295 *idx = U64_MAX;
1296 break;
1297 }
1298
1299 if (out_spfn)
1300 *out_spfn = max(zone->zone_start_pfn, spfn);
1301 if (out_epfn)
1302 *out_epfn = min(zone_end_pfn(zone), epfn);
1303
1304 return;
1305 }
1306
1307 __next_mem_range(idx, zone_nid, MEMBLOCK_NONE,
1308 &memblock.memory, &memblock.reserved,
1309 &spa, &epa, &nid);
1310 }
1311
1312 /* signal end of iteration */
1313 if (out_spfn)
1314 *out_spfn = ULONG_MAX;
1315 if (out_epfn)
1316 *out_epfn = 0;
1317}
1318
1319#endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
Tejun Heo7c0caeb2011-07-14 11:43:42 +02001320
Mike Rapoport92d12f92019-03-11 23:29:41 -07001321/**
1322 * memblock_alloc_range_nid - allocate boot memory block
1323 * @size: size of memory block to be allocated in bytes
1324 * @align: alignment of the region and block's size
1325 * @start: the lower bound of the memory region to allocate (phys address)
1326 * @end: the upper bound of the memory region to allocate (phys address)
1327 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
Yunfeng Ye0ac398b2019-11-30 17:56:27 -08001328 * @exact_nid: control the allocation fall back to other nodes
Mike Rapoport92d12f92019-03-11 23:29:41 -07001329 *
1330 * The allocation is performed from memory region limited by
Cao jin95830662019-11-30 17:56:24 -08001331 * memblock.current_limit if @end == %MEMBLOCK_ALLOC_ACCESSIBLE.
Mike Rapoport92d12f92019-03-11 23:29:41 -07001332 *
Yunfeng Ye0ac398b2019-11-30 17:56:27 -08001333 * If the specified node can not hold the requested memory and @exact_nid
1334 * is false, the allocation falls back to any node in the system.
Mike Rapoport92d12f92019-03-11 23:29:41 -07001335 *
1336 * For systems with memory mirroring, the allocation is attempted first
1337 * from the regions with mirroring enabled and then retried from any
1338 * memory region.
1339 *
1340 * In addition, function sets the min_count to 0 using kmemleak_alloc_phys for
1341 * allocated boot memory block, so that it is never reported as leaks.
1342 *
1343 * Return:
1344 * Physical address of allocated memory block on success, %0 on failure.
1345 */
Aslan Bakirov8676af12020-04-10 14:32:42 -07001346phys_addr_t __init memblock_alloc_range_nid(phys_addr_t size,
Akinobu Mita2bfc2862014-06-04 16:06:53 -07001347 phys_addr_t align, phys_addr_t start,
Yunfeng Ye0ac398b2019-11-30 17:56:27 -08001348 phys_addr_t end, int nid,
1349 bool exact_nid)
Yinghai Lu95f72d12010-07-12 14:36:09 +10001350{
Mike Rapoport92d12f92019-03-11 23:29:41 -07001351 enum memblock_flags flags = choose_memblock_flags();
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +10001352 phys_addr_t found;
Yinghai Lu95f72d12010-07-12 14:36:09 +10001353
Mike Rapoport92d12f92019-03-11 23:29:41 -07001354 if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
1355 nid = NUMA_NO_NODE;
1356
Mike Rapoport2f770802018-10-30 15:10:01 -07001357 if (!align) {
1358 /* Can't use WARNs this early in boot on powerpc */
1359 dump_stack();
1360 align = SMP_CACHE_BYTES;
1361 }
1362
Mike Rapoport92d12f92019-03-11 23:29:41 -07001363again:
Tony Luckfc6daaf2015-06-24 16:58:09 -07001364 found = memblock_find_in_range_node(size, align, start, end, nid,
1365 flags);
Mike Rapoport92d12f92019-03-11 23:29:41 -07001366 if (found && !memblock_reserve(found, size))
1367 goto done;
1368
Yunfeng Ye0ac398b2019-11-30 17:56:27 -08001369 if (nid != NUMA_NO_NODE && !exact_nid) {
Mike Rapoport92d12f92019-03-11 23:29:41 -07001370 found = memblock_find_in_range_node(size, align, start,
1371 end, NUMA_NO_NODE,
1372 flags);
1373 if (found && !memblock_reserve(found, size))
1374 goto done;
1375 }
1376
1377 if (flags & MEMBLOCK_MIRROR) {
1378 flags &= ~MEMBLOCK_MIRROR;
1379 pr_warn("Could not allocate %pap bytes of mirrored memory\n",
1380 &size);
1381 goto again;
1382 }
1383
1384 return 0;
1385
1386done:
1387 /* Skip kmemleak for kasan_init() due to high volume. */
1388 if (end != MEMBLOCK_ALLOC_KASAN)
Catalin Marinasaedf95e2014-06-06 14:38:20 -07001389 /*
Mike Rapoport92d12f92019-03-11 23:29:41 -07001390 * The min_count is set to 0 so that memblock allocated
1391 * blocks are never reported as leaks. This is because many
1392 * of these blocks are only referred via the physical
1393 * address which is not looked up by kmemleak.
Catalin Marinasaedf95e2014-06-06 14:38:20 -07001394 */
Catalin Marinas9099dae2016-10-11 13:55:11 -07001395 kmemleak_alloc_phys(found, size, 0, 0);
Mike Rapoport92d12f92019-03-11 23:29:41 -07001396
1397 return found;
Yinghai Lu95f72d12010-07-12 14:36:09 +10001398}
1399
Mike Rapoporta2974132019-03-11 23:30:54 -07001400/**
1401 * memblock_phys_alloc_range - allocate a memory block inside specified range
1402 * @size: size of memory block to be allocated in bytes
1403 * @align: alignment of the region and block's size
1404 * @start: the lower bound of the memory region to allocate (physical address)
1405 * @end: the upper bound of the memory region to allocate (physical address)
1406 *
1407 * Allocate @size bytes in the between @start and @end.
1408 *
1409 * Return: physical address of the allocated memory block on success,
1410 * %0 on failure.
1411 */
Mike Rapoport8a770c22019-03-11 23:29:16 -07001412phys_addr_t __init memblock_phys_alloc_range(phys_addr_t size,
1413 phys_addr_t align,
1414 phys_addr_t start,
1415 phys_addr_t end)
Akinobu Mita2bfc2862014-06-04 16:06:53 -07001416{
Faiyaz Mohammedb5cf2d62020-11-16 10:14:04 +05301417 memblock_dbg("%s: %llu bytes align=0x%llx from=%pa max_addr=%pa %pS\n",
1418 __func__, (u64)size, (u64)align, &start, &end,
1419 (void *)_RET_IP_);
Yunfeng Ye0ac398b2019-11-30 17:56:27 -08001420 return memblock_alloc_range_nid(size, align, start, end, NUMA_NO_NODE,
1421 false);
Tejun Heo7bd0b0f2011-12-08 10:22:09 -08001422}
1423
Mike Rapoporta2974132019-03-11 23:30:54 -07001424/**
Levi Yun17cbe032021-01-20 21:28:18 +09001425 * memblock_phys_alloc_try_nid - allocate a memory block from specified NUMA node
Mike Rapoporta2974132019-03-11 23:30:54 -07001426 * @size: size of memory block to be allocated in bytes
1427 * @align: alignment of the region and block's size
1428 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1429 *
1430 * Allocates memory block from the specified NUMA node. If the node
1431 * has no available memory, attempts to allocated from any node in the
1432 * system.
1433 *
1434 * Return: physical address of the allocated memory block on success,
1435 * %0 on failure.
1436 */
Mike Rapoport9a8dd702018-10-30 15:07:59 -07001437phys_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 -07001438{
Mike Rapoport33755572019-03-11 23:29:21 -07001439 return memblock_alloc_range_nid(size, align, 0,
Yunfeng Ye0ac398b2019-11-30 17:56:27 -08001440 MEMBLOCK_ALLOC_ACCESSIBLE, nid, false);
Yinghai Lu95f72d12010-07-12 14:36:09 +10001441}
1442
Santosh Shilimkar26f09e92014-01-21 15:50:19 -08001443/**
Mike Rapoporteb31d552018-10-30 15:08:04 -07001444 * memblock_alloc_internal - allocate boot memory block
Santosh Shilimkar26f09e92014-01-21 15:50:19 -08001445 * @size: size of memory block to be allocated in bytes
1446 * @align: alignment of the region and block's size
1447 * @min_addr: the lower bound of the memory region to allocate (phys address)
1448 * @max_addr: the upper bound of the memory region to allocate (phys address)
1449 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
Yunfeng Ye0ac398b2019-11-30 17:56:27 -08001450 * @exact_nid: control the allocation fall back to other nodes
Santosh Shilimkar26f09e92014-01-21 15:50:19 -08001451 *
Mike Rapoport92d12f92019-03-11 23:29:41 -07001452 * Allocates memory block using memblock_alloc_range_nid() and
1453 * converts the returned physical address to virtual.
1454 *
Santosh Shilimkar26f09e92014-01-21 15:50:19 -08001455 * The @min_addr limit is dropped if it can not be satisfied and the allocation
Mike Rapoport92d12f92019-03-11 23:29:41 -07001456 * will fall back to memory below @min_addr. Other constraints, such
1457 * as node and mirrored memory will be handled again in
1458 * memblock_alloc_range_nid().
Santosh Shilimkar26f09e92014-01-21 15:50:19 -08001459 *
Mike Rapoport47cec442018-06-30 17:55:02 +03001460 * Return:
Santosh Shilimkar26f09e92014-01-21 15:50:19 -08001461 * Virtual address of allocated memory block on success, NULL on failure.
1462 */
Mike Rapoporteb31d552018-10-30 15:08:04 -07001463static void * __init memblock_alloc_internal(
Santosh Shilimkar26f09e92014-01-21 15:50:19 -08001464 phys_addr_t size, phys_addr_t align,
1465 phys_addr_t min_addr, phys_addr_t max_addr,
Yunfeng Ye0ac398b2019-11-30 17:56:27 -08001466 int nid, bool exact_nid)
Santosh Shilimkar26f09e92014-01-21 15:50:19 -08001467{
1468 phys_addr_t alloc;
Santosh Shilimkar26f09e92014-01-21 15:50:19 -08001469
1470 /*
1471 * Detect any accidental use of these APIs after slab is ready, as at
1472 * this moment memblock may be deinitialized already and its
Mike Rapoportc6ffc5c2018-10-30 15:09:30 -07001473 * internal data may be destroyed (after execution of memblock_free_all)
Santosh Shilimkar26f09e92014-01-21 15:50:19 -08001474 */
1475 if (WARN_ON_ONCE(slab_is_available()))
1476 return kzalloc_node(size, GFP_NOWAIT, nid);
1477
Mike Rapoportf3057ad2019-10-18 20:20:01 -07001478 if (max_addr > memblock.current_limit)
1479 max_addr = memblock.current_limit;
1480
Yunfeng Ye0ac398b2019-11-30 17:56:27 -08001481 alloc = memblock_alloc_range_nid(size, align, min_addr, max_addr, nid,
1482 exact_nid);
Mike Rapoport2f770802018-10-30 15:10:01 -07001483
Mike Rapoport92d12f92019-03-11 23:29:41 -07001484 /* retry allocation without lower limit */
1485 if (!alloc && min_addr)
Yunfeng Ye0ac398b2019-11-30 17:56:27 -08001486 alloc = memblock_alloc_range_nid(size, align, 0, max_addr, nid,
1487 exact_nid);
Santosh Shilimkar26f09e92014-01-21 15:50:19 -08001488
Mike Rapoport92d12f92019-03-11 23:29:41 -07001489 if (!alloc)
1490 return NULL;
Santosh Shilimkar26f09e92014-01-21 15:50:19 -08001491
Mike Rapoport92d12f92019-03-11 23:29:41 -07001492 return phys_to_virt(alloc);
Santosh Shilimkar26f09e92014-01-21 15:50:19 -08001493}
1494
1495/**
Yunfeng Ye0ac398b2019-11-30 17:56:27 -08001496 * memblock_alloc_exact_nid_raw - allocate boot memory block on the exact node
1497 * without zeroing memory
1498 * @size: size of memory block to be allocated in bytes
1499 * @align: alignment of the region and block's size
1500 * @min_addr: the lower bound of the memory region from where the allocation
1501 * is preferred (phys address)
1502 * @max_addr: the upper bound of the memory region from where the allocation
1503 * is preferred (phys address), or %MEMBLOCK_ALLOC_ACCESSIBLE to
1504 * allocate only from memory limited by memblock.current_limit value
1505 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1506 *
1507 * Public function, provides additional debug information (including caller
1508 * info), if enabled. Does not zero allocated memory.
1509 *
1510 * Return:
1511 * Virtual address of allocated memory block on success, NULL on failure.
1512 */
1513void * __init memblock_alloc_exact_nid_raw(
1514 phys_addr_t size, phys_addr_t align,
1515 phys_addr_t min_addr, phys_addr_t max_addr,
1516 int nid)
1517{
Yunfeng Ye0ac398b2019-11-30 17:56:27 -08001518 memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=%pa max_addr=%pa %pS\n",
1519 __func__, (u64)size, (u64)align, nid, &min_addr,
1520 &max_addr, (void *)_RET_IP_);
1521
Mike Rapoport08678802021-09-02 14:58:05 -07001522 return memblock_alloc_internal(size, align, min_addr, max_addr, nid,
1523 true);
Yunfeng Ye0ac398b2019-11-30 17:56:27 -08001524}
1525
1526/**
Mike Rapoporteb31d552018-10-30 15:08:04 -07001527 * memblock_alloc_try_nid_raw - allocate boot memory block without zeroing
Pavel Tatashinea1f5f32017-11-15 17:36:27 -08001528 * memory and without panicking
1529 * @size: size of memory block to be allocated in bytes
1530 * @align: alignment of the region and block's size
1531 * @min_addr: the lower bound of the memory region from where the allocation
1532 * is preferred (phys address)
1533 * @max_addr: the upper bound of the memory region from where the allocation
Mike Rapoport97ad1082018-10-30 15:09:44 -07001534 * is preferred (phys address), or %MEMBLOCK_ALLOC_ACCESSIBLE to
Pavel Tatashinea1f5f32017-11-15 17:36:27 -08001535 * allocate only from memory limited by memblock.current_limit value
1536 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1537 *
1538 * Public function, provides additional debug information (including caller
1539 * info), if enabled. Does not zero allocated memory, does not panic if request
1540 * cannot be satisfied.
1541 *
Mike Rapoport47cec442018-06-30 17:55:02 +03001542 * Return:
Pavel Tatashinea1f5f32017-11-15 17:36:27 -08001543 * Virtual address of allocated memory block on success, NULL on failure.
1544 */
Mike Rapoporteb31d552018-10-30 15:08:04 -07001545void * __init memblock_alloc_try_nid_raw(
Pavel Tatashinea1f5f32017-11-15 17:36:27 -08001546 phys_addr_t size, phys_addr_t align,
1547 phys_addr_t min_addr, phys_addr_t max_addr,
1548 int nid)
1549{
Sakari Ailusd75f7732019-03-25 21:32:28 +02001550 memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=%pa max_addr=%pa %pS\n",
Mike Rapoporta36aab82018-08-17 15:47:17 -07001551 __func__, (u64)size, (u64)align, nid, &min_addr,
1552 &max_addr, (void *)_RET_IP_);
Pavel Tatashinea1f5f32017-11-15 17:36:27 -08001553
Mike Rapoport08678802021-09-02 14:58:05 -07001554 return memblock_alloc_internal(size, align, min_addr, max_addr, nid,
1555 false);
Pavel Tatashinea1f5f32017-11-15 17:36:27 -08001556}
1557
1558/**
Mike Rapoportc0dbe822019-03-11 23:30:37 -07001559 * memblock_alloc_try_nid - allocate boot memory block
Santosh Shilimkar26f09e92014-01-21 15:50:19 -08001560 * @size: size of memory block to be allocated in bytes
1561 * @align: alignment of the region and block's size
1562 * @min_addr: the lower bound of the memory region from where the allocation
1563 * is preferred (phys address)
1564 * @max_addr: the upper bound of the memory region from where the allocation
Mike Rapoport97ad1082018-10-30 15:09:44 -07001565 * is preferred (phys address), or %MEMBLOCK_ALLOC_ACCESSIBLE to
Santosh Shilimkar26f09e92014-01-21 15:50:19 -08001566 * allocate only from memory limited by memblock.current_limit value
1567 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1568 *
Mike Rapoportc0dbe822019-03-11 23:30:37 -07001569 * Public function, provides additional debug information (including caller
1570 * info), if enabled. This function zeroes the allocated memory.
Santosh Shilimkar26f09e92014-01-21 15:50:19 -08001571 *
Mike Rapoport47cec442018-06-30 17:55:02 +03001572 * Return:
Santosh Shilimkar26f09e92014-01-21 15:50:19 -08001573 * Virtual address of allocated memory block on success, NULL on failure.
1574 */
Mike Rapoporteb31d552018-10-30 15:08:04 -07001575void * __init memblock_alloc_try_nid(
Santosh Shilimkar26f09e92014-01-21 15:50:19 -08001576 phys_addr_t size, phys_addr_t align,
1577 phys_addr_t min_addr, phys_addr_t max_addr,
1578 int nid)
1579{
1580 void *ptr;
1581
Sakari Ailusd75f7732019-03-25 21:32:28 +02001582 memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=%pa max_addr=%pa %pS\n",
Mike Rapoporta36aab82018-08-17 15:47:17 -07001583 __func__, (u64)size, (u64)align, nid, &min_addr,
1584 &max_addr, (void *)_RET_IP_);
Mike Rapoporteb31d552018-10-30 15:08:04 -07001585 ptr = memblock_alloc_internal(size, align,
Yunfeng Ye0ac398b2019-11-30 17:56:27 -08001586 min_addr, max_addr, nid, false);
Mike Rapoportc0dbe822019-03-11 23:30:37 -07001587 if (ptr)
Pavel Tatashinea1f5f32017-11-15 17:36:27 -08001588 memset(ptr, 0, size);
Santosh Shilimkar26f09e92014-01-21 15:50:19 -08001589
Mike Rapoportc0dbe822019-03-11 23:30:37 -07001590 return ptr;
Santosh Shilimkar26f09e92014-01-21 15:50:19 -08001591}
1592
1593/**
Mike Rapoporta2974132019-03-11 23:30:54 -07001594 * __memblock_free_late - free pages directly to buddy allocator
Mike Rapoport48a833c2018-06-30 17:55:03 +03001595 * @base: phys starting address of the boot memory block
Santosh Shilimkar26f09e92014-01-21 15:50:19 -08001596 * @size: size of the boot memory block in bytes
1597 *
Mike Rapoporta2974132019-03-11 23:30:54 -07001598 * This is only useful when the memblock allocator has already been torn
Santosh Shilimkar26f09e92014-01-21 15:50:19 -08001599 * down, but we are still initializing the system. Pages are released directly
Mike Rapoporta2974132019-03-11 23:30:54 -07001600 * to the buddy allocator.
Santosh Shilimkar26f09e92014-01-21 15:50:19 -08001601 */
1602void __init __memblock_free_late(phys_addr_t base, phys_addr_t size)
1603{
Mike Rapoporta36aab82018-08-17 15:47:17 -07001604 phys_addr_t cursor, end;
Santosh Shilimkar26f09e92014-01-21 15:50:19 -08001605
Mike Rapoporta36aab82018-08-17 15:47:17 -07001606 end = base + size - 1;
Sakari Ailusd75f7732019-03-25 21:32:28 +02001607 memblock_dbg("%s: [%pa-%pa] %pS\n",
Mike Rapoporta36aab82018-08-17 15:47:17 -07001608 __func__, &base, &end, (void *)_RET_IP_);
Catalin Marinas9099dae2016-10-11 13:55:11 -07001609 kmemleak_free_part_phys(base, size);
Santosh Shilimkar26f09e92014-01-21 15:50:19 -08001610 cursor = PFN_UP(base);
1611 end = PFN_DOWN(base + size);
1612
1613 for (; cursor < end; cursor++) {
Mike Rapoport7c2ee342018-10-30 15:09:36 -07001614 memblock_free_pages(pfn_to_page(cursor), cursor, 0);
Arun KSca79b0c2018-12-28 00:34:29 -08001615 totalram_pages_inc();
Santosh Shilimkar26f09e92014-01-21 15:50:19 -08001616 }
1617}
Benjamin Herrenschmidt9d1e2492010-07-06 15:39:17 -07001618
1619/*
1620 * Remaining API functions
1621 */
1622
David Gibson1f1ffb8a2016-02-05 15:36:19 -08001623phys_addr_t __init_memblock memblock_phys_mem_size(void)
Yinghai Lu95f72d12010-07-12 14:36:09 +10001624{
Tejun Heo1440c4e2011-12-08 10:22:08 -08001625 return memblock.memory.total_size;
Yinghai Lu95f72d12010-07-12 14:36:09 +10001626}
1627
Srikar Dronamraju8907de52016-10-07 16:59:18 -07001628phys_addr_t __init_memblock memblock_reserved_size(void)
1629{
1630 return memblock.reserved.total_size;
1631}
1632
Sam Ravnborg0a93ebe2011-10-31 17:08:16 -07001633/* lowest address */
1634phys_addr_t __init_memblock memblock_start_of_DRAM(void)
1635{
1636 return memblock.memory.regions[0].base;
1637}
1638
Yinghai Lu10d06432010-07-28 15:43:02 +10001639phys_addr_t __init_memblock memblock_end_of_DRAM(void)
Yinghai Lu95f72d12010-07-12 14:36:09 +10001640{
1641 int idx = memblock.memory.cnt - 1;
1642
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +10001643 return (memblock.memory.regions[idx].base + memblock.memory.regions[idx].size);
Yinghai Lu95f72d12010-07-12 14:36:09 +10001644}
1645
Dennis Chena571d4e2016-07-28 15:48:26 -07001646static phys_addr_t __init_memblock __find_max_addr(phys_addr_t limit)
Yinghai Lu95f72d12010-07-12 14:36:09 +10001647{
Stefan Agner1c4bc432018-06-07 17:06:15 -07001648 phys_addr_t max_addr = PHYS_ADDR_MAX;
Emil Medve136199f2014-04-07 15:37:52 -07001649 struct memblock_region *r;
Yinghai Lu95f72d12010-07-12 14:36:09 +10001650
Dennis Chena571d4e2016-07-28 15:48:26 -07001651 /*
1652 * translate the memory @limit size into the max address within one of
1653 * the memory memblock regions, if the @limit exceeds the total size
Stefan Agner1c4bc432018-06-07 17:06:15 -07001654 * of those regions, max_addr will keep original value PHYS_ADDR_MAX
Dennis Chena571d4e2016-07-28 15:48:26 -07001655 */
Mike Rapoportcc6de162020-10-13 16:58:30 -07001656 for_each_mem_region(r) {
Tejun Heoc0ce8fe2011-12-08 10:22:07 -08001657 if (limit <= r->size) {
1658 max_addr = r->base + limit;
1659 break;
1660 }
1661 limit -= r->size;
Yinghai Lu95f72d12010-07-12 14:36:09 +10001662 }
1663
Dennis Chena571d4e2016-07-28 15:48:26 -07001664 return max_addr;
1665}
1666
1667void __init memblock_enforce_memory_limit(phys_addr_t limit)
1668{
Colin Ian King49aef712020-04-01 21:11:01 -07001669 phys_addr_t max_addr;
Dennis Chena571d4e2016-07-28 15:48:26 -07001670
1671 if (!limit)
1672 return;
1673
1674 max_addr = __find_max_addr(limit);
1675
1676 /* @limit exceeds the total size of the memory, do nothing */
Stefan Agner1c4bc432018-06-07 17:06:15 -07001677 if (max_addr == PHYS_ADDR_MAX)
Dennis Chena571d4e2016-07-28 15:48:26 -07001678 return;
1679
Tejun Heoc0ce8fe2011-12-08 10:22:07 -08001680 /* truncate both memory and reserved regions */
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +01001681 memblock_remove_range(&memblock.memory, max_addr,
Stefan Agner1c4bc432018-06-07 17:06:15 -07001682 PHYS_ADDR_MAX);
Philipp Hachtmannf1af9d32014-01-29 18:16:01 +01001683 memblock_remove_range(&memblock.reserved, max_addr,
Stefan Agner1c4bc432018-06-07 17:06:15 -07001684 PHYS_ADDR_MAX);
Yinghai Lu95f72d12010-07-12 14:36:09 +10001685}
1686
AKASHI Takahiroc9ca9b42017-04-03 11:23:55 +09001687void __init memblock_cap_memory_range(phys_addr_t base, phys_addr_t size)
1688{
1689 int start_rgn, end_rgn;
1690 int i, ret;
1691
1692 if (!size)
1693 return;
1694
Peng Fan5173ed72021-10-18 15:15:45 -07001695 if (!memblock_memory->total_size) {
Geert Uytterhoevene888fa72021-08-11 10:55:18 +02001696 pr_warn("%s: No memory registered yet\n", __func__);
1697 return;
1698 }
1699
AKASHI Takahiroc9ca9b42017-04-03 11:23:55 +09001700 ret = memblock_isolate_range(&memblock.memory, base, size,
1701 &start_rgn, &end_rgn);
1702 if (ret)
1703 return;
1704
1705 /* remove all the MAP regions */
1706 for (i = memblock.memory.cnt - 1; i >= end_rgn; i--)
1707 if (!memblock_is_nomap(&memblock.memory.regions[i]))
1708 memblock_remove_region(&memblock.memory, i);
1709
1710 for (i = start_rgn - 1; i >= 0; i--)
1711 if (!memblock_is_nomap(&memblock.memory.regions[i]))
1712 memblock_remove_region(&memblock.memory, i);
1713
1714 /* truncate the reserved regions */
1715 memblock_remove_range(&memblock.reserved, 0, base);
1716 memblock_remove_range(&memblock.reserved,
Stefan Agner1c4bc432018-06-07 17:06:15 -07001717 base + size, PHYS_ADDR_MAX);
AKASHI Takahiroc9ca9b42017-04-03 11:23:55 +09001718}
1719
Dennis Chena571d4e2016-07-28 15:48:26 -07001720void __init memblock_mem_limit_remove_map(phys_addr_t limit)
1721{
Dennis Chena571d4e2016-07-28 15:48:26 -07001722 phys_addr_t max_addr;
Dennis Chena571d4e2016-07-28 15:48:26 -07001723
1724 if (!limit)
1725 return;
1726
1727 max_addr = __find_max_addr(limit);
1728
1729 /* @limit exceeds the total size of the memory, do nothing */
Stefan Agner1c4bc432018-06-07 17:06:15 -07001730 if (max_addr == PHYS_ADDR_MAX)
Dennis Chena571d4e2016-07-28 15:48:26 -07001731 return;
1732
AKASHI Takahiroc9ca9b42017-04-03 11:23:55 +09001733 memblock_cap_memory_range(0, max_addr);
Dennis Chena571d4e2016-07-28 15:48:26 -07001734}
1735
Yinghai Lucd794812010-10-11 12:34:09 -07001736static int __init_memblock memblock_search(struct memblock_type *type, phys_addr_t addr)
Benjamin Herrenschmidt72d4b0b2010-08-04 14:38:47 +10001737{
1738 unsigned int left = 0, right = type->cnt;
1739
1740 do {
1741 unsigned int mid = (right + left) / 2;
1742
1743 if (addr < type->regions[mid].base)
1744 right = mid;
1745 else if (addr >= (type->regions[mid].base +
1746 type->regions[mid].size))
1747 left = mid + 1;
1748 else
1749 return mid;
1750 } while (left < right);
1751 return -1;
1752}
1753
Yueyi Lif5a222d2018-12-14 14:17:06 -08001754bool __init_memblock memblock_is_reserved(phys_addr_t addr)
Yinghai Lu95f72d12010-07-12 14:36:09 +10001755{
Benjamin Herrenschmidt72d4b0b2010-08-04 14:38:47 +10001756 return memblock_search(&memblock.reserved, addr) != -1;
1757}
Yinghai Lu95f72d12010-07-12 14:36:09 +10001758
Yaowei Baib4ad0c72016-01-14 15:18:54 -08001759bool __init_memblock memblock_is_memory(phys_addr_t addr)
Benjamin Herrenschmidt72d4b0b2010-08-04 14:38:47 +10001760{
1761 return memblock_search(&memblock.memory, addr) != -1;
1762}
1763
Yaowei Bai937f0c22018-02-06 15:41:18 -08001764bool __init_memblock memblock_is_map_memory(phys_addr_t addr)
Ard Biesheuvelbf3d3cc2015-11-30 13:28:15 +01001765{
1766 int i = memblock_search(&memblock.memory, addr);
1767
1768 if (i == -1)
1769 return false;
1770 return !memblock_is_nomap(&memblock.memory.regions[i]);
1771}
1772
Yinghai Lue76b63f2013-09-11 14:22:17 -07001773int __init_memblock memblock_search_pfn_nid(unsigned long pfn,
1774 unsigned long *start_pfn, unsigned long *end_pfn)
1775{
1776 struct memblock_type *type = &memblock.memory;
Fabian Frederick16763232014-04-07 15:37:53 -07001777 int mid = memblock_search(type, PFN_PHYS(pfn));
Yinghai Lue76b63f2013-09-11 14:22:17 -07001778
1779 if (mid == -1)
1780 return -1;
1781
Fabian Frederickf7e2f7e2014-06-04 16:07:51 -07001782 *start_pfn = PFN_DOWN(type->regions[mid].base);
1783 *end_pfn = PFN_DOWN(type->regions[mid].base + type->regions[mid].size);
Yinghai Lue76b63f2013-09-11 14:22:17 -07001784
Mike Rapoportd622abf2020-06-03 15:56:53 -07001785 return memblock_get_region_node(&type->regions[mid]);
Yinghai Lue76b63f2013-09-11 14:22:17 -07001786}
Yinghai Lue76b63f2013-09-11 14:22:17 -07001787
Stephen Boydeab30942012-05-24 00:45:21 -07001788/**
1789 * memblock_is_region_memory - check if a region is a subset of memory
1790 * @base: base of region to check
1791 * @size: size of region to check
1792 *
Mike Rapoport47cec442018-06-30 17:55:02 +03001793 * Check if the region [@base, @base + @size) is a subset of a memory block.
Stephen Boydeab30942012-05-24 00:45:21 -07001794 *
Mike Rapoport47cec442018-06-30 17:55:02 +03001795 * Return:
Stephen Boydeab30942012-05-24 00:45:21 -07001796 * 0 if false, non-zero if true
1797 */
Yaowei Bai937f0c22018-02-06 15:41:18 -08001798bool __init_memblock memblock_is_region_memory(phys_addr_t base, phys_addr_t size)
Benjamin Herrenschmidt72d4b0b2010-08-04 14:38:47 +10001799{
Tomi Valkeinenabb65272011-01-20 14:44:20 -08001800 int idx = memblock_search(&memblock.memory, base);
Tejun Heoeb18f1b2011-12-08 10:22:07 -08001801 phys_addr_t end = base + memblock_cap_size(base, &size);
Benjamin Herrenschmidt72d4b0b2010-08-04 14:38:47 +10001802
1803 if (idx == -1)
Yaowei Bai937f0c22018-02-06 15:41:18 -08001804 return false;
Wei Yangef415ef2017-02-22 15:45:04 -08001805 return (memblock.memory.regions[idx].base +
Tejun Heoeb18f1b2011-12-08 10:22:07 -08001806 memblock.memory.regions[idx].size) >= end;
Yinghai Lu95f72d12010-07-12 14:36:09 +10001807}
1808
Stephen Boydeab30942012-05-24 00:45:21 -07001809/**
1810 * memblock_is_region_reserved - check if a region intersects reserved memory
1811 * @base: base of region to check
1812 * @size: size of region to check
1813 *
Mike Rapoport47cec442018-06-30 17:55:02 +03001814 * Check if the region [@base, @base + @size) intersects a reserved
1815 * memory block.
Stephen Boydeab30942012-05-24 00:45:21 -07001816 *
Mike Rapoport47cec442018-06-30 17:55:02 +03001817 * Return:
Tang Chenc5c5c9d2015-09-08 15:02:00 -07001818 * True if they intersect, false if not.
Stephen Boydeab30942012-05-24 00:45:21 -07001819 */
Tang Chenc5c5c9d2015-09-08 15:02:00 -07001820bool __init_memblock memblock_is_region_reserved(phys_addr_t base, phys_addr_t size)
Yinghai Lu95f72d12010-07-12 14:36:09 +10001821{
Tang Chenc5c5c9d2015-09-08 15:02:00 -07001822 return memblock_overlaps_region(&memblock.reserved, base, size);
Yinghai Lu95f72d12010-07-12 14:36:09 +10001823}
1824
Yinghai Lu6ede1fd2012-10-22 16:35:18 -07001825void __init_memblock memblock_trim_memory(phys_addr_t align)
1826{
Yinghai Lu6ede1fd2012-10-22 16:35:18 -07001827 phys_addr_t start, end, orig_start, orig_end;
Emil Medve136199f2014-04-07 15:37:52 -07001828 struct memblock_region *r;
Yinghai Lu6ede1fd2012-10-22 16:35:18 -07001829
Mike Rapoportcc6de162020-10-13 16:58:30 -07001830 for_each_mem_region(r) {
Emil Medve136199f2014-04-07 15:37:52 -07001831 orig_start = r->base;
1832 orig_end = r->base + r->size;
Yinghai Lu6ede1fd2012-10-22 16:35:18 -07001833 start = round_up(orig_start, align);
1834 end = round_down(orig_end, align);
1835
1836 if (start == orig_start && end == orig_end)
1837 continue;
1838
1839 if (start < end) {
Emil Medve136199f2014-04-07 15:37:52 -07001840 r->base = start;
1841 r->size = end - start;
Yinghai Lu6ede1fd2012-10-22 16:35:18 -07001842 } else {
Emil Medve136199f2014-04-07 15:37:52 -07001843 memblock_remove_region(&memblock.memory,
1844 r - memblock.memory.regions);
1845 r--;
Yinghai Lu6ede1fd2012-10-22 16:35:18 -07001846 }
1847 }
1848}
Benjamin Herrenschmidte63075a2010-07-06 15:39:01 -07001849
Yinghai Lu3661ca62010-09-15 13:05:29 -07001850void __init_memblock memblock_set_current_limit(phys_addr_t limit)
Benjamin Herrenschmidte63075a2010-07-06 15:39:01 -07001851{
1852 memblock.current_limit = limit;
1853}
1854
Laura Abbottfec51012014-02-27 01:23:43 +01001855phys_addr_t __init_memblock memblock_get_current_limit(void)
1856{
1857 return memblock.current_limit;
1858}
1859
Heiko Carstens0262d9c2017-02-24 14:55:59 -08001860static void __init_memblock memblock_dump(struct memblock_type *type)
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +10001861{
Miles Chen5d63f812017-02-22 15:46:42 -08001862 phys_addr_t base, end, size;
Mike Rapoporte1720fe2018-06-30 17:55:01 +03001863 enum memblock_flags flags;
Alexander Kuleshov8c9c1702016-01-14 15:20:42 -08001864 int idx;
1865 struct memblock_region *rgn;
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +10001866
Heiko Carstens0262d9c2017-02-24 14:55:59 -08001867 pr_info(" %s.cnt = 0x%lx\n", type->name, type->cnt);
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +10001868
Gioh Kim66e8b432017-11-15 17:33:42 -08001869 for_each_memblock_type(idx, type, rgn) {
Tejun Heo7c0caeb2011-07-14 11:43:42 +02001870 char nid_buf[32] = "";
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +10001871
Tejun Heo7c0caeb2011-07-14 11:43:42 +02001872 base = rgn->base;
1873 size = rgn->size;
Miles Chen5d63f812017-02-22 15:46:42 -08001874 end = base + size - 1;
Tang Chen66a20752014-01-21 15:49:20 -08001875 flags = rgn->flags;
Mike Rapoporta9ee6cf2021-06-28 19:43:01 -07001876#ifdef CONFIG_NUMA
Tejun Heo7c0caeb2011-07-14 11:43:42 +02001877 if (memblock_get_region_node(rgn) != MAX_NUMNODES)
1878 snprintf(nid_buf, sizeof(nid_buf), " on node %d",
1879 memblock_get_region_node(rgn));
1880#endif
Mike Rapoporte1720fe2018-06-30 17:55:01 +03001881 pr_info(" %s[%#x]\t[%pa-%pa], %pa bytes%s flags: %#x\n",
Heiko Carstens0262d9c2017-02-24 14:55:59 -08001882 type->name, idx, &base, &end, &size, nid_buf, flags);
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +10001883 }
1884}
1885
Mike Rapoport87c55872020-10-13 16:57:54 -07001886static void __init_memblock __memblock_dump_all(void)
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +10001887{
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +10001888 pr_info("MEMBLOCK configuration:\n");
Miles Chen5d63f812017-02-22 15:46:42 -08001889 pr_info(" memory size = %pa reserved size = %pa\n",
1890 &memblock.memory.total_size,
1891 &memblock.reserved.total_size);
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +10001892
Heiko Carstens0262d9c2017-02-24 14:55:59 -08001893 memblock_dump(&memblock.memory);
1894 memblock_dump(&memblock.reserved);
Heiko Carstens409efd42017-02-24 14:55:56 -08001895#ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
David Hildenbrand77649902020-07-01 16:18:29 +02001896 memblock_dump(&physmem);
Heiko Carstens409efd42017-02-24 14:55:56 -08001897#endif
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +10001898}
1899
Mike Rapoport87c55872020-10-13 16:57:54 -07001900void __init_memblock memblock_dump_all(void)
1901{
1902 if (memblock_debug)
1903 __memblock_dump_all();
1904}
1905
Tejun Heo1aadc052011-12-08 10:22:08 -08001906void __init memblock_allow_resize(void)
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +10001907{
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -07001908 memblock_can_resize = 1;
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +10001909}
1910
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +10001911static int __init early_memblock(char *p)
1912{
1913 if (p && strstr(p, "debug"))
1914 memblock_debug = 1;
1915 return 0;
1916}
1917early_param("memblock", early_memblock);
1918
Mike Rapoport4f5b0c12020-12-14 19:09:59 -08001919static void __init free_memmap(unsigned long start_pfn, unsigned long end_pfn)
1920{
1921 struct page *start_pg, *end_pg;
1922 phys_addr_t pg, pgend;
1923
1924 /*
1925 * Convert start_pfn/end_pfn to a struct page pointer.
1926 */
1927 start_pg = pfn_to_page(start_pfn - 1) + 1;
1928 end_pg = pfn_to_page(end_pfn - 1) + 1;
1929
1930 /*
1931 * Convert to physical addresses, and round start upwards and end
1932 * downwards.
1933 */
1934 pg = PAGE_ALIGN(__pa(start_pg));
1935 pgend = __pa(end_pg) & PAGE_MASK;
1936
1937 /*
1938 * If there are free pages between these, free the section of the
1939 * memmap array.
1940 */
1941 if (pg < pgend)
1942 memblock_free(pg, pgend - pg);
1943}
1944
1945/*
1946 * The mem_map array can get very big. Free the unused area of the memory map.
1947 */
1948static void __init free_unused_memmap(void)
1949{
1950 unsigned long start, end, prev_end = 0;
1951 int i;
1952
1953 if (!IS_ENABLED(CONFIG_HAVE_ARCH_PFN_VALID) ||
1954 IS_ENABLED(CONFIG_SPARSEMEM_VMEMMAP))
1955 return;
1956
1957 /*
1958 * This relies on each bank being in address order.
1959 * The banks are sorted previously in bootmem_init().
1960 */
1961 for_each_mem_pfn_range(i, MAX_NUMNODES, &start, &end, NULL) {
1962#ifdef CONFIG_SPARSEMEM
1963 /*
1964 * Take care not to free memmap entries that don't exist
1965 * due to SPARSEMEM sections which aren't present.
1966 */
1967 start = min(start, ALIGN(prev_end, PAGES_PER_SECTION));
Mike Rapoport4f5b0c12020-12-14 19:09:59 -08001968#endif
Mike Rapoport4f5b0c12020-12-14 19:09:59 -08001969 /*
Mike Rapoporte2a86802021-05-17 21:15:15 +03001970 * Align down here since many operations in VM subsystem
1971 * presume that there are no holes in the memory map inside
1972 * a pageblock
Mike Rapoport4f5b0c12020-12-14 19:09:59 -08001973 */
Mike Rapoporte2a86802021-05-17 21:15:15 +03001974 start = round_down(start, pageblock_nr_pages);
Mike Rapoport4f5b0c12020-12-14 19:09:59 -08001975
1976 /*
1977 * If we had a previous bank, and there is a space
1978 * between the current bank and the previous, free it.
1979 */
1980 if (prev_end && prev_end < start)
1981 free_memmap(prev_end, start);
1982
1983 /*
Mike Rapoporte2a86802021-05-17 21:15:15 +03001984 * Align up here since many operations in VM subsystem
1985 * presume that there are no holes in the memory map inside
1986 * a pageblock
Mike Rapoport4f5b0c12020-12-14 19:09:59 -08001987 */
Mike Rapoporte2a86802021-05-17 21:15:15 +03001988 prev_end = ALIGN(end, pageblock_nr_pages);
Mike Rapoport4f5b0c12020-12-14 19:09:59 -08001989 }
1990
1991#ifdef CONFIG_SPARSEMEM
Mike Rapoportf921f532021-05-17 21:31:59 +03001992 if (!IS_ALIGNED(prev_end, PAGES_PER_SECTION)) {
1993 prev_end = ALIGN(end, pageblock_nr_pages);
Mike Rapoport4f5b0c12020-12-14 19:09:59 -08001994 free_memmap(prev_end, ALIGN(prev_end, PAGES_PER_SECTION));
Mike Rapoportf921f532021-05-17 21:31:59 +03001995 }
Mike Rapoport4f5b0c12020-12-14 19:09:59 -08001996#endif
1997}
1998
Mike Rapoportbda49a82018-10-30 15:09:40 -07001999static void __init __free_pages_memory(unsigned long start, unsigned long end)
2000{
2001 int order;
2002
2003 while (start < end) {
2004 order = min(MAX_ORDER - 1UL, __ffs(start));
2005
2006 while (start + (1UL << order) > end)
2007 order--;
2008
2009 memblock_free_pages(pfn_to_page(start), start, order);
2010
2011 start += (1UL << order);
2012 }
2013}
2014
2015static unsigned long __init __free_memory_core(phys_addr_t start,
2016 phys_addr_t end)
2017{
2018 unsigned long start_pfn = PFN_UP(start);
2019 unsigned long end_pfn = min_t(unsigned long,
2020 PFN_DOWN(end), max_low_pfn);
2021
2022 if (start_pfn >= end_pfn)
2023 return 0;
2024
2025 __free_pages_memory(start_pfn, end_pfn);
2026
2027 return end_pfn - start_pfn;
2028}
2029
Mike Rapoport9092d4f2021-06-30 18:51:16 -07002030static void __init memmap_init_reserved_pages(void)
2031{
2032 struct memblock_region *region;
2033 phys_addr_t start, end;
2034 u64 i;
2035
2036 /* initialize struct pages for the reserved regions */
2037 for_each_reserved_mem_range(i, &start, &end)
2038 reserve_bootmem_region(start, end);
2039
2040 /* and also treat struct pages for the NOMAP regions as PageReserved */
2041 for_each_mem_region(region) {
2042 if (memblock_is_nomap(region)) {
2043 start = region->base;
2044 end = start + region->size;
2045 reserve_bootmem_region(start, end);
2046 }
2047 }
2048}
2049
Mike Rapoportbda49a82018-10-30 15:09:40 -07002050static unsigned long __init free_low_memory_core_early(void)
2051{
2052 unsigned long count = 0;
2053 phys_addr_t start, end;
2054 u64 i;
2055
2056 memblock_clear_hotplug(0, -1);
2057
Mike Rapoport9092d4f2021-06-30 18:51:16 -07002058 memmap_init_reserved_pages();
Mike Rapoportbda49a82018-10-30 15:09:40 -07002059
2060 /*
2061 * We need to use NUMA_NO_NODE instead of NODE_DATA(0)->node_id
2062 * because in some case like Node0 doesn't have RAM installed
2063 * low ram will be on Node1
2064 */
2065 for_each_free_mem_range(i, NUMA_NO_NODE, MEMBLOCK_NONE, &start, &end,
2066 NULL)
2067 count += __free_memory_core(start, end);
2068
2069 return count;
2070}
2071
2072static int reset_managed_pages_done __initdata;
2073
2074void reset_node_managed_pages(pg_data_t *pgdat)
2075{
2076 struct zone *z;
2077
2078 for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++)
Arun KS9705bea2018-12-28 00:34:24 -08002079 atomic_long_set(&z->managed_pages, 0);
Mike Rapoportbda49a82018-10-30 15:09:40 -07002080}
2081
2082void __init reset_all_zones_managed_pages(void)
2083{
2084 struct pglist_data *pgdat;
2085
2086 if (reset_managed_pages_done)
2087 return;
2088
2089 for_each_online_pgdat(pgdat)
2090 reset_node_managed_pages(pgdat);
2091
2092 reset_managed_pages_done = 1;
2093}
2094
2095/**
2096 * memblock_free_all - release free pages to the buddy allocator
Mike Rapoportbda49a82018-10-30 15:09:40 -07002097 */
Daeseok Youn097d43d2021-01-14 16:08:17 +09002098void __init memblock_free_all(void)
Mike Rapoportbda49a82018-10-30 15:09:40 -07002099{
2100 unsigned long pages;
2101
Mike Rapoport4f5b0c12020-12-14 19:09:59 -08002102 free_unused_memmap();
Mike Rapoportbda49a82018-10-30 15:09:40 -07002103 reset_all_zones_managed_pages();
2104
2105 pages = free_low_memory_core_early();
Arun KSca79b0c2018-12-28 00:34:29 -08002106 totalram_pages_add(pages);
Mike Rapoportbda49a82018-10-30 15:09:40 -07002107}
2108
Mike Rapoport350e88b2019-05-13 17:22:59 -07002109#if defined(CONFIG_DEBUG_FS) && defined(CONFIG_ARCH_KEEP_MEMBLOCK)
Benjamin Herrenschmidt6d03b882010-07-06 15:39:19 -07002110
2111static int memblock_debug_show(struct seq_file *m, void *private)
2112{
2113 struct memblock_type *type = m->private;
2114 struct memblock_region *reg;
2115 int i;
Miles Chen5d63f812017-02-22 15:46:42 -08002116 phys_addr_t end;
Benjamin Herrenschmidt6d03b882010-07-06 15:39:19 -07002117
2118 for (i = 0; i < type->cnt; i++) {
2119 reg = &type->regions[i];
Miles Chen5d63f812017-02-22 15:46:42 -08002120 end = reg->base + reg->size - 1;
Benjamin Herrenschmidt6d03b882010-07-06 15:39:19 -07002121
Miles Chen5d63f812017-02-22 15:46:42 -08002122 seq_printf(m, "%4d: ", i);
2123 seq_printf(m, "%pa..%pa\n", &reg->base, &end);
Benjamin Herrenschmidt6d03b882010-07-06 15:39:19 -07002124 }
2125 return 0;
2126}
Andy Shevchenko5ad35092018-04-05 16:23:16 -07002127DEFINE_SHOW_ATTRIBUTE(memblock_debug);
Benjamin Herrenschmidt6d03b882010-07-06 15:39:19 -07002128
2129static int __init memblock_init_debugfs(void)
2130{
2131 struct dentry *root = debugfs_create_dir("memblock", NULL);
Greg Kroah-Hartmand9f79792019-03-05 15:46:09 -08002132
Joe Perches0825a6f2018-06-14 15:27:58 -07002133 debugfs_create_file("memory", 0444, root,
2134 &memblock.memory, &memblock_debug_fops);
2135 debugfs_create_file("reserved", 0444, root,
2136 &memblock.reserved, &memblock_debug_fops);
Philipp Hachtmann70210ed2014-01-29 18:16:01 +01002137#ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
David Hildenbrand77649902020-07-01 16:18:29 +02002138 debugfs_create_file("physmem", 0444, root, &physmem,
2139 &memblock_debug_fops);
Philipp Hachtmann70210ed2014-01-29 18:16:01 +01002140#endif
Benjamin Herrenschmidt6d03b882010-07-06 15:39:19 -07002141
2142 return 0;
2143}
2144__initcall(memblock_init_debugfs);
2145
2146#endif /* CONFIG_DEBUG_FS */