blob: be3c62e3fb95c716b0e2e5cc0ca71548e71c2fd9 [file] [log] [blame]
Thomas Gleixner457c8992019-05-19 13:08:55 +01001// SPDX-License-Identifier: GPL-2.0-only
Dave Hansen3947be12005-10-29 18:16:54 -07002/*
3 * linux/mm/memory_hotplug.c
4 *
5 * Copyright (C)
6 */
7
Dave Hansen3947be12005-10-29 18:16:54 -07008#include <linux/stddef.h>
9#include <linux/mm.h>
Ingo Molnar174cd4b2017-02-02 19:15:33 +010010#include <linux/sched/signal.h>
Dave Hansen3947be12005-10-29 18:16:54 -070011#include <linux/swap.h>
12#include <linux/interrupt.h>
13#include <linux/pagemap.h>
Dave Hansen3947be12005-10-29 18:16:54 -070014#include <linux/compiler.h>
Paul Gortmakerb95f1b312011-10-16 02:01:52 -040015#include <linux/export.h>
Dave Hansen3947be12005-10-29 18:16:54 -070016#include <linux/pagevec.h>
Chandra Seetharaman2d1d43f2006-09-29 02:01:25 -070017#include <linux/writeback.h>
Dave Hansen3947be12005-10-29 18:16:54 -070018#include <linux/slab.h>
19#include <linux/sysctl.h>
20#include <linux/cpu.h>
21#include <linux/memory.h>
Dan Williams4b94ffd2016-01-15 16:56:22 -080022#include <linux/memremap.h>
Dave Hansen3947be12005-10-29 18:16:54 -070023#include <linux/memory_hotplug.h>
24#include <linux/highmem.h>
25#include <linux/vmalloc.h>
KAMEZAWA Hiroyuki0a547032006-06-27 02:53:35 -070026#include <linux/ioport.h>
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -070027#include <linux/delay.h>
28#include <linux/migrate.h>
29#include <linux/page-isolation.h>
Badari Pulavarty71088782008-10-18 20:25:58 -070030#include <linux/pfn.h>
Andi Kleen6ad696d2009-11-17 14:06:22 -080031#include <linux/suspend.h>
KOSAKI Motohiro6d9c2852009-12-14 17:58:11 -080032#include <linux/mm_inline.h>
akpm@linux-foundation.orgd96ae532010-03-05 13:41:58 -080033#include <linux/firmware-map.h>
Tang Chen60a5a192013-02-22 16:33:14 -080034#include <linux/stop_machine.h>
Naoya Horiguchic8721bb2013-09-11 14:22:09 -070035#include <linux/hugetlb.h>
Tang Chenc5320922013-11-12 15:08:10 -080036#include <linux/memblock.h>
Vlastimil Babka698b1b32016-03-17 14:18:08 -070037#include <linux/compaction.h>
Michal Hockob15c8722018-12-28 00:38:01 -080038#include <linux/rmap.h>
Dave Hansen3947be12005-10-29 18:16:54 -070039
40#include <asm/tlbflush.h>
41
Adrian Bunk1e5ad9a2008-04-28 20:40:08 +030042#include "internal.h"
Dan Williamse900a912019-05-14 15:41:28 -070043#include "shuffle.h"
Adrian Bunk1e5ad9a2008-04-28 20:40:08 +030044
Daniel Kiper9d0ad8c2011-07-25 17:12:05 -070045/*
46 * online_page_callback contains pointer to current page onlining function.
47 * Initially it is generic_online_page(). If it is required it could be
48 * changed by calling set_online_page_callback() for callback registration
49 * and restore_online_page_callback() for generic callback restore.
50 */
51
Daniel Kiper9d0ad8c2011-07-25 17:12:05 -070052static online_page_callback_t online_page_callback = generic_online_page;
Vladimir Davydovbfc8c902014-06-04 16:07:18 -070053static DEFINE_MUTEX(online_page_callback_lock);
Daniel Kiper9d0ad8c2011-07-25 17:12:05 -070054
Thomas Gleixner3f906ba2017-07-10 15:50:09 -070055DEFINE_STATIC_PERCPU_RWSEM(mem_hotplug_lock);
KOSAKI Motohiro20d6c962010-12-02 14:31:19 -080056
Thomas Gleixner3f906ba2017-07-10 15:50:09 -070057void get_online_mems(void)
58{
59 percpu_down_read(&mem_hotplug_lock);
60}
Vladimir Davydovbfc8c902014-06-04 16:07:18 -070061
Thomas Gleixner3f906ba2017-07-10 15:50:09 -070062void put_online_mems(void)
63{
64 percpu_up_read(&mem_hotplug_lock);
65}
Vladimir Davydovbfc8c902014-06-04 16:07:18 -070066
Michal Hocko49323812017-07-06 15:41:05 -070067bool movable_node_enabled = false;
68
Vitaly Kuznetsov8604d9e2016-05-19 17:13:03 -070069#ifndef CONFIG_MEMORY_HOTPLUG_DEFAULT_ONLINE
David Hildenbrand862919e2020-04-06 20:07:40 -070070int memhp_default_online_type = MMOP_OFFLINE;
Vitaly Kuznetsov8604d9e2016-05-19 17:13:03 -070071#else
David Hildenbrand862919e2020-04-06 20:07:40 -070072int memhp_default_online_type = MMOP_ONLINE;
Vitaly Kuznetsov8604d9e2016-05-19 17:13:03 -070073#endif
Vitaly Kuznetsov31bc3852016-03-15 14:56:48 -070074
Vitaly Kuznetsov86dd9952016-05-19 17:13:06 -070075static int __init setup_memhp_default_state(char *str)
76{
David Hildenbrand5f47adf2020-04-06 20:07:44 -070077 const int online_type = memhp_online_type_from_str(str);
78
79 if (online_type >= 0)
80 memhp_default_online_type = online_type;
Vitaly Kuznetsov86dd9952016-05-19 17:13:06 -070081
82 return 1;
83}
84__setup("memhp_default_state=", setup_memhp_default_state);
85
David Rientjes30467e02015-04-14 15:45:11 -070086void mem_hotplug_begin(void)
Vladimir Davydovbfc8c902014-06-04 16:07:18 -070087{
Thomas Gleixner3f906ba2017-07-10 15:50:09 -070088 cpus_read_lock();
89 percpu_down_write(&mem_hotplug_lock);
Vladimir Davydovbfc8c902014-06-04 16:07:18 -070090}
91
David Rientjes30467e02015-04-14 15:45:11 -070092void mem_hotplug_done(void)
Vladimir Davydovbfc8c902014-06-04 16:07:18 -070093{
Thomas Gleixner3f906ba2017-07-10 15:50:09 -070094 percpu_up_write(&mem_hotplug_lock);
95 cpus_read_unlock();
Vladimir Davydovbfc8c902014-06-04 16:07:18 -070096}
KOSAKI Motohiro20d6c962010-12-02 14:31:19 -080097
Juergen Gross357b4da2019-02-14 11:42:39 +010098u64 max_mem_size = U64_MAX;
99
Keith Mannthey45e0b782006-09-30 23:27:09 -0700100/* add this memory to iomem resource */
David Hildenbrand7b7b2722020-06-04 16:48:41 -0700101static struct resource *register_memory_resource(u64 start, u64 size,
102 const char *resource_name)
Keith Mannthey45e0b782006-09-30 23:27:09 -0700103{
Dave Hansen27941292019-02-25 10:57:36 -0800104 struct resource *res;
105 unsigned long flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
David Hildenbrand7b7b2722020-06-04 16:48:41 -0700106
107 if (strcmp(resource_name, "System RAM"))
108 flags |= IORESOURCE_MEM_DRIVER_MANAGED;
Juergen Gross357b4da2019-02-14 11:42:39 +0100109
Baoquan Hef3cd4c82020-04-06 20:06:50 -0700110 /*
111 * Make sure value parsed from 'mem=' only restricts memory adding
112 * while booting, so that memory hotplug won't be impacted. Please
113 * refer to document of 'mem=' in kernel-parameters.txt for more
114 * details.
115 */
116 if (start + size > max_mem_size && system_state < SYSTEM_RUNNING)
Juergen Gross357b4da2019-02-14 11:42:39 +0100117 return ERR_PTR(-E2BIG);
118
Dave Hansen27941292019-02-25 10:57:36 -0800119 /*
120 * Request ownership of the new memory range. This might be
121 * a child of an existing resource that was present but
122 * not marked as busy.
123 */
124 res = __request_region(&iomem_resource, start, size,
125 resource_name, flags);
Keith Mannthey45e0b782006-09-30 23:27:09 -0700126
Dave Hansen27941292019-02-25 10:57:36 -0800127 if (!res) {
128 pr_debug("Unable to reserve System RAM region: %016llx->%016llx\n",
129 start, start + size);
Vitaly Kuznetsov6f754ba2016-01-14 15:21:55 -0800130 return ERR_PTR(-EEXIST);
Keith Mannthey45e0b782006-09-30 23:27:09 -0700131 }
132 return res;
133}
134
135static void release_memory_resource(struct resource *res)
136{
137 if (!res)
138 return;
139 release_resource(res);
140 kfree(res);
Keith Mannthey45e0b782006-09-30 23:27:09 -0700141}
142
Keith Mannthey53947022006-09-30 23:27:08 -0700143#ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
Yasuaki Ishimatsu46723bf2013-02-22 16:33:00 -0800144void get_page_bootmem(unsigned long info, struct page *page,
145 unsigned long type)
Yasunori Goto04753272008-04-28 02:13:31 -0700146{
Yasuaki Ishimatsuddffe982017-02-22 15:45:13 -0800147 page->freelist = (void *)type;
Yasunori Goto04753272008-04-28 02:13:31 -0700148 SetPagePrivate(page);
149 set_page_private(page, info);
Joonsoo Kimfe896d12016-03-17 14:19:26 -0700150 page_ref_inc(page);
Yasunori Goto04753272008-04-28 02:13:31 -0700151}
152
Jiang Liu170a5a72013-07-03 15:03:17 -0700153void put_page_bootmem(struct page *page)
Yasunori Goto04753272008-04-28 02:13:31 -0700154{
Andrea Arcangeli5f24ce52011-01-13 15:47:00 -0800155 unsigned long type;
Yasunori Goto04753272008-04-28 02:13:31 -0700156
Yasuaki Ishimatsuddffe982017-02-22 15:45:13 -0800157 type = (unsigned long) page->freelist;
Andrea Arcangeli5f24ce52011-01-13 15:47:00 -0800158 BUG_ON(type < MEMORY_HOTPLUG_MIN_BOOTMEM_TYPE ||
159 type > MEMORY_HOTPLUG_MAX_BOOTMEM_TYPE);
Yasunori Goto04753272008-04-28 02:13:31 -0700160
Joonsoo Kimfe896d12016-03-17 14:19:26 -0700161 if (page_ref_dec_return(page) == 1) {
Yasuaki Ishimatsuddffe982017-02-22 15:45:13 -0800162 page->freelist = NULL;
Yasunori Goto04753272008-04-28 02:13:31 -0700163 ClearPagePrivate(page);
164 set_page_private(page, 0);
Andrea Arcangeli5f24ce52011-01-13 15:47:00 -0800165 INIT_LIST_HEAD(&page->lru);
Jiang Liu170a5a72013-07-03 15:03:17 -0700166 free_reserved_page(page);
Yasunori Goto04753272008-04-28 02:13:31 -0700167 }
Yasunori Goto04753272008-04-28 02:13:31 -0700168}
169
Yasuaki Ishimatsu46723bf2013-02-22 16:33:00 -0800170#ifdef CONFIG_HAVE_BOOTMEM_INFO_NODE
171#ifndef CONFIG_SPARSEMEM_VMEMMAP
Adrian Bunkd92bc312008-07-23 21:28:12 -0700172static void register_page_bootmem_info_section(unsigned long start_pfn)
Yasunori Goto04753272008-04-28 02:13:31 -0700173{
Dan Williamsf1eca352019-07-18 15:57:57 -0700174 unsigned long mapsize, section_nr, i;
Yasunori Goto04753272008-04-28 02:13:31 -0700175 struct mem_section *ms;
176 struct page *page, *memmap;
Dan Williamsf1eca352019-07-18 15:57:57 -0700177 struct mem_section_usage *usage;
Yasunori Goto04753272008-04-28 02:13:31 -0700178
Yasunori Goto04753272008-04-28 02:13:31 -0700179 section_nr = pfn_to_section_nr(start_pfn);
180 ms = __nr_to_section(section_nr);
181
182 /* Get section's memmap address */
183 memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
184
185 /*
186 * Get page for the memmap's phys address
187 * XXX: need more consideration for sparse_vmemmap...
188 */
189 page = virt_to_page(memmap);
190 mapsize = sizeof(struct page) * PAGES_PER_SECTION;
191 mapsize = PAGE_ALIGN(mapsize) >> PAGE_SHIFT;
192
193 /* remember memmap's page */
194 for (i = 0; i < mapsize; i++, page++)
195 get_page_bootmem(section_nr, page, SECTION_INFO);
196
Dan Williamsf1eca352019-07-18 15:57:57 -0700197 usage = ms->usage;
198 page = virt_to_page(usage);
Yasunori Goto04753272008-04-28 02:13:31 -0700199
Dan Williamsf1eca352019-07-18 15:57:57 -0700200 mapsize = PAGE_ALIGN(mem_section_usage_size()) >> PAGE_SHIFT;
Yasunori Goto04753272008-04-28 02:13:31 -0700201
202 for (i = 0; i < mapsize; i++, page++)
Yasunori Gotoaf370fb2008-07-23 21:28:17 -0700203 get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
Yasunori Goto04753272008-04-28 02:13:31 -0700204
205}
Yasuaki Ishimatsu46723bf2013-02-22 16:33:00 -0800206#else /* CONFIG_SPARSEMEM_VMEMMAP */
207static void register_page_bootmem_info_section(unsigned long start_pfn)
208{
Dan Williamsf1eca352019-07-18 15:57:57 -0700209 unsigned long mapsize, section_nr, i;
Yasuaki Ishimatsu46723bf2013-02-22 16:33:00 -0800210 struct mem_section *ms;
211 struct page *page, *memmap;
Dan Williamsf1eca352019-07-18 15:57:57 -0700212 struct mem_section_usage *usage;
Yasuaki Ishimatsu46723bf2013-02-22 16:33:00 -0800213
Yasuaki Ishimatsu46723bf2013-02-22 16:33:00 -0800214 section_nr = pfn_to_section_nr(start_pfn);
215 ms = __nr_to_section(section_nr);
216
217 memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
218
219 register_page_bootmem_memmap(section_nr, memmap, PAGES_PER_SECTION);
220
Dan Williamsf1eca352019-07-18 15:57:57 -0700221 usage = ms->usage;
222 page = virt_to_page(usage);
Yasuaki Ishimatsu46723bf2013-02-22 16:33:00 -0800223
Dan Williamsf1eca352019-07-18 15:57:57 -0700224 mapsize = PAGE_ALIGN(mem_section_usage_size()) >> PAGE_SHIFT;
Yasuaki Ishimatsu46723bf2013-02-22 16:33:00 -0800225
226 for (i = 0; i < mapsize; i++, page++)
227 get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
228}
229#endif /* !CONFIG_SPARSEMEM_VMEMMAP */
Yasunori Goto04753272008-04-28 02:13:31 -0700230
Linus Torvalds7ded3842016-05-27 15:23:32 -0700231void __init register_page_bootmem_info_node(struct pglist_data *pgdat)
Yasunori Goto04753272008-04-28 02:13:31 -0700232{
233 unsigned long i, pfn, end_pfn, nr_pages;
234 int node = pgdat->node_id;
235 struct page *page;
Yasunori Goto04753272008-04-28 02:13:31 -0700236
237 nr_pages = PAGE_ALIGN(sizeof(struct pglist_data)) >> PAGE_SHIFT;
238 page = virt_to_page(pgdat);
239
240 for (i = 0; i < nr_pages; i++, page++)
241 get_page_bootmem(node, page, NODE_INFO);
242
Yasunori Goto04753272008-04-28 02:13:31 -0700243 pfn = pgdat->node_start_pfn;
Cody P Schaferc1f19492013-02-22 16:35:32 -0800244 end_pfn = pgdat_end_pfn(pgdat);
Yasunori Goto04753272008-04-28 02:13:31 -0700245
Tang Chen7e9f5eb2013-07-08 16:00:23 -0700246 /* register section info */
qiuxishif14851a2012-09-17 14:09:24 -0700247 for (; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
248 /*
249 * Some platforms can assign the same pfn to multiple nodes - on
250 * node0 as well as nodeN. To avoid registering a pfn against
251 * multiple nodes we check that this pfn does not already
Tang Chen7e9f5eb2013-07-08 16:00:23 -0700252 * reside in some other nodes.
qiuxishif14851a2012-09-17 14:09:24 -0700253 */
Yang Shif65e91df2016-05-27 14:27:32 -0700254 if (pfn_valid(pfn) && (early_pfn_to_nid(pfn) == node))
qiuxishif14851a2012-09-17 14:09:24 -0700255 register_page_bootmem_info_section(pfn);
256 }
Yasunori Goto04753272008-04-28 02:13:31 -0700257}
Yasuaki Ishimatsu46723bf2013-02-22 16:33:00 -0800258#endif /* CONFIG_HAVE_BOOTMEM_INFO_NODE */
Yasunori Goto04753272008-04-28 02:13:31 -0700259
Dan Williams7ea62162019-07-18 15:58:22 -0700260static int check_pfn_span(unsigned long pfn, unsigned long nr_pages,
261 const char *reason)
262{
263 /*
264 * Disallow all operations smaller than a sub-section and only
265 * allow operations smaller than a section for
266 * SPARSEMEM_VMEMMAP. Note that check_hotplug_memory_range()
267 * enforces a larger memory_block_size_bytes() granularity for
268 * memory that will be marked online, so this check should only
269 * fire for direct arch_{add,remove}_memory() users outside of
270 * add_memory_resource().
271 */
272 unsigned long min_align;
273
274 if (IS_ENABLED(CONFIG_SPARSEMEM_VMEMMAP))
275 min_align = PAGES_PER_SUBSECTION;
276 else
277 min_align = PAGES_PER_SECTION;
278 if (!IS_ALIGNED(pfn, min_align)
279 || !IS_ALIGNED(nr_pages, min_align)) {
280 WARN(1, "Misaligned __%s_pages start: %#lx end: #%lx\n",
281 reason, pfn, pfn + nr_pages - 1);
282 return -EINVAL;
283 }
284 return 0;
285}
286
Alastair D'Silvadca44362019-11-30 17:53:48 -0800287static int check_hotplug_memory_addressable(unsigned long pfn,
288 unsigned long nr_pages)
289{
290 const u64 max_addr = PFN_PHYS(pfn + nr_pages) - 1;
291
292 if (max_addr >> MAX_PHYSMEM_BITS) {
293 const u64 max_allowed = (1ull << (MAX_PHYSMEM_BITS + 1)) - 1;
294 WARN(1,
295 "Hotplugged memory exceeds maximum addressable address, range=%#llx-%#llx, maximum=%#llx\n",
296 (u64)PFN_PHYS(pfn), max_addr, max_allowed);
297 return -E2BIG;
298 }
299
300 return 0;
301}
302
David Rientjes4edd7ce2013-04-29 15:08:22 -0700303/*
304 * Reasonably generic function for adding memory. It is
305 * expected that archs that support memory hotplug will
306 * call this function after deciding the zone to which to
307 * add the new pages.
308 */
Dan Williams7ea62162019-07-18 15:58:22 -0700309int __ref __add_pages(int nid, unsigned long pfn, unsigned long nr_pages,
Logan Gunthorpef5637d32020-04-10 14:33:21 -0700310 struct mhp_params *params)
David Rientjes4edd7ce2013-04-29 15:08:22 -0700311{
David Hildenbrand6cdd0b32020-04-06 20:06:56 -0700312 const unsigned long end_pfn = pfn + nr_pages;
313 unsigned long cur_nr_pages;
Dan Williams9a845032019-07-18 15:58:43 -0700314 int err;
Logan Gunthorpef5637d32020-04-10 14:33:21 -0700315 struct vmem_altmap *altmap = params->altmap;
Dan Williams4b94ffd2016-01-15 16:56:22 -0800316
Logan Gunthorpebfeb0222020-04-10 14:33:36 -0700317 if (WARN_ON_ONCE(!params->pgprot.pgprot))
318 return -EINVAL;
319
Alastair D'Silvadca44362019-11-30 17:53:48 -0800320 err = check_hotplug_memory_addressable(pfn, nr_pages);
321 if (err)
322 return err;
323
Dan Williams4b94ffd2016-01-15 16:56:22 -0800324 if (altmap) {
325 /*
326 * Validate altmap is within bounds of the total request
327 */
Dan Williams7ea62162019-07-18 15:58:22 -0700328 if (altmap->base_pfn != pfn
Dan Williams4b94ffd2016-01-15 16:56:22 -0800329 || vmem_altmap_offset(altmap) > nr_pages) {
330 pr_warn_once("memory add fail, invalid altmap\n");
Dan Williams7ea62162019-07-18 15:58:22 -0700331 return -EINVAL;
Dan Williams4b94ffd2016-01-15 16:56:22 -0800332 }
333 altmap->alloc = 0;
334 }
335
Dan Williams7ea62162019-07-18 15:58:22 -0700336 err = check_pfn_span(pfn, nr_pages, "add");
337 if (err)
338 return err;
339
David Hildenbrand6cdd0b32020-04-06 20:06:56 -0700340 for (; pfn < end_pfn; pfn += cur_nr_pages) {
341 /* Select all remaining pages up to the next section boundary */
342 cur_nr_pages = min(end_pfn - pfn,
343 SECTION_ALIGN_UP(pfn + 1) - pfn);
344 err = sparse_add_section(nid, pfn, cur_nr_pages, altmap);
Dan Williamsba72b4c2019-07-18 15:58:26 -0700345 if (err)
346 break;
Michal Hockof64ac5e2017-10-03 16:16:16 -0700347 cond_resched();
David Rientjes4edd7ce2013-04-29 15:08:22 -0700348 }
Zhu Guihuac435a392015-06-24 16:58:42 -0700349 vmemmap_populate_print_last();
David Rientjes4edd7ce2013-04-29 15:08:22 -0700350 return err;
351}
David Rientjes4edd7ce2013-04-29 15:08:22 -0700352
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800353/* find the smallest valid pfn in the range [start_pfn, end_pfn) */
YASUAKI ISHIMATSUd09b0132017-10-03 16:16:32 -0700354static unsigned long find_smallest_section_pfn(int nid, struct zone *zone,
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800355 unsigned long start_pfn,
356 unsigned long end_pfn)
357{
Dan Williams49ba3c62019-07-18 15:58:07 -0700358 for (; start_pfn < end_pfn; start_pfn += PAGES_PER_SUBSECTION) {
David Hildenbrand7ce700b2019-11-21 17:53:56 -0800359 if (unlikely(!pfn_to_online_page(start_pfn)))
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800360 continue;
361
362 if (unlikely(pfn_to_nid(start_pfn) != nid))
363 continue;
364
David Hildenbrand9b051582020-02-03 17:34:12 -0800365 if (zone != page_zone(pfn_to_page(start_pfn)))
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800366 continue;
367
368 return start_pfn;
369 }
370
371 return 0;
372}
373
374/* find the biggest valid pfn in the range [start_pfn, end_pfn). */
YASUAKI ISHIMATSUd09b0132017-10-03 16:16:32 -0700375static unsigned long find_biggest_section_pfn(int nid, struct zone *zone,
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800376 unsigned long start_pfn,
377 unsigned long end_pfn)
378{
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800379 unsigned long pfn;
380
381 /* pfn is the end pfn of a memory section. */
382 pfn = end_pfn - 1;
Dan Williams49ba3c62019-07-18 15:58:07 -0700383 for (; pfn >= start_pfn; pfn -= PAGES_PER_SUBSECTION) {
David Hildenbrand7ce700b2019-11-21 17:53:56 -0800384 if (unlikely(!pfn_to_online_page(pfn)))
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800385 continue;
386
387 if (unlikely(pfn_to_nid(pfn) != nid))
388 continue;
389
David Hildenbrand9b051582020-02-03 17:34:12 -0800390 if (zone != page_zone(pfn_to_page(pfn)))
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800391 continue;
392
393 return pfn;
394 }
395
396 return 0;
397}
398
399static void shrink_zone_span(struct zone *zone, unsigned long start_pfn,
400 unsigned long end_pfn)
401{
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800402 unsigned long pfn;
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800403 int nid = zone_to_nid(zone);
404
405 zone_span_writelock(zone);
David Hildenbrand5d120712020-02-03 17:34:19 -0800406 if (zone->zone_start_pfn == start_pfn) {
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800407 /*
408 * If the section is smallest section in the zone, it need
409 * shrink zone->zone_start_pfn and zone->zone_spanned_pages.
410 * In this case, we find second smallest valid mem_section
411 * for shrinking zone.
412 */
413 pfn = find_smallest_section_pfn(nid, zone, end_pfn,
David Hildenbrand5d120712020-02-03 17:34:19 -0800414 zone_end_pfn(zone));
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800415 if (pfn) {
David Hildenbrand5d120712020-02-03 17:34:19 -0800416 zone->spanned_pages = zone_end_pfn(zone) - pfn;
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800417 zone->zone_start_pfn = pfn;
David Hildenbrand950b68d2020-02-03 17:34:16 -0800418 } else {
419 zone->zone_start_pfn = 0;
420 zone->spanned_pages = 0;
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800421 }
David Hildenbrand5d120712020-02-03 17:34:19 -0800422 } else if (zone_end_pfn(zone) == end_pfn) {
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800423 /*
424 * If the section is biggest section in the zone, it need
425 * shrink zone->spanned_pages.
426 * In this case, we find second biggest valid mem_section for
427 * shrinking zone.
428 */
David Hildenbrand5d120712020-02-03 17:34:19 -0800429 pfn = find_biggest_section_pfn(nid, zone, zone->zone_start_pfn,
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800430 start_pfn);
431 if (pfn)
David Hildenbrand5d120712020-02-03 17:34:19 -0800432 zone->spanned_pages = pfn - zone->zone_start_pfn + 1;
David Hildenbrand950b68d2020-02-03 17:34:16 -0800433 else {
434 zone->zone_start_pfn = 0;
435 zone->spanned_pages = 0;
436 }
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800437 }
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800438 zone_span_writeunlock(zone);
439}
440
David Hildenbrand00d6c012019-10-18 20:19:33 -0700441static void update_pgdat_span(struct pglist_data *pgdat)
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800442{
David Hildenbrand00d6c012019-10-18 20:19:33 -0700443 unsigned long node_start_pfn = 0, node_end_pfn = 0;
444 struct zone *zone;
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800445
David Hildenbrand00d6c012019-10-18 20:19:33 -0700446 for (zone = pgdat->node_zones;
447 zone < pgdat->node_zones + MAX_NR_ZONES; zone++) {
448 unsigned long zone_end_pfn = zone->zone_start_pfn +
449 zone->spanned_pages;
450
451 /* No need to lock the zones, they can't change. */
David Hildenbrand656d5712019-11-05 21:17:10 -0800452 if (!zone->spanned_pages)
453 continue;
454 if (!node_end_pfn) {
455 node_start_pfn = zone->zone_start_pfn;
456 node_end_pfn = zone_end_pfn;
457 continue;
458 }
459
David Hildenbrand00d6c012019-10-18 20:19:33 -0700460 if (zone_end_pfn > node_end_pfn)
461 node_end_pfn = zone_end_pfn;
462 if (zone->zone_start_pfn < node_start_pfn)
463 node_start_pfn = zone->zone_start_pfn;
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800464 }
465
David Hildenbrand00d6c012019-10-18 20:19:33 -0700466 pgdat->node_start_pfn = node_start_pfn;
467 pgdat->node_spanned_pages = node_end_pfn - node_start_pfn;
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800468}
469
David Hildenbrandfeee6b22020-01-04 12:59:33 -0800470void __ref remove_pfn_range_from_zone(struct zone *zone,
471 unsigned long start_pfn,
472 unsigned long nr_pages)
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800473{
Ben Widawskyb7e3deb2020-06-25 20:30:51 -0700474 const unsigned long end_pfn = start_pfn + nr_pages;
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800475 struct pglist_data *pgdat = zone->zone_pgdat;
Ben Widawskyb7e3deb2020-06-25 20:30:51 -0700476 unsigned long pfn, cur_nr_pages, flags;
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800477
David Hildenbrandd33695b2020-02-03 17:34:09 -0800478 /* Poison struct pages because they are now uninitialized again. */
Ben Widawskyb7e3deb2020-06-25 20:30:51 -0700479 for (pfn = start_pfn; pfn < end_pfn; pfn += cur_nr_pages) {
480 cond_resched();
481
482 /* Select all remaining pages up to the next section boundary */
483 cur_nr_pages =
484 min(end_pfn - pfn, SECTION_ALIGN_UP(pfn + 1) - pfn);
485 page_init_poison(pfn_to_page(pfn),
486 sizeof(struct page) * cur_nr_pages);
487 }
David Hildenbrandd33695b2020-02-03 17:34:09 -0800488
David Hildenbrand7ce700b2019-11-21 17:53:56 -0800489#ifdef CONFIG_ZONE_DEVICE
490 /*
491 * Zone shrinking code cannot properly deal with ZONE_DEVICE. So
492 * we will not try to shrink the zones - which is okay as
493 * set_zone_contiguous() cannot deal with ZONE_DEVICE either way.
494 */
495 if (zone_idx(zone) == ZONE_DEVICE)
496 return;
497#endif
498
David Hildenbrandfeee6b22020-01-04 12:59:33 -0800499 clear_zone_contiguous(zone);
500
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800501 pgdat_resize_lock(zone->zone_pgdat, &flags);
502 shrink_zone_span(zone, start_pfn, start_pfn + nr_pages);
David Hildenbrand00d6c012019-10-18 20:19:33 -0700503 update_pgdat_span(pgdat);
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800504 pgdat_resize_unlock(zone->zone_pgdat, &flags);
David Hildenbrandfeee6b22020-01-04 12:59:33 -0800505
506 set_zone_contiguous(zone);
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800507}
508
David Hildenbrandfeee6b22020-01-04 12:59:33 -0800509static void __remove_section(unsigned long pfn, unsigned long nr_pages,
510 unsigned long map_offset,
511 struct vmem_altmap *altmap)
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700512{
chenqiwu10404902020-04-06 20:07:48 -0700513 struct mem_section *ms = __pfn_to_section(pfn);
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700514
David Hildenbrand9d1d8872019-05-13 17:21:41 -0700515 if (WARN_ON_ONCE(!valid_section(ms)))
516 return;
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700517
Dan Williamsba72b4c2019-07-18 15:58:26 -0700518 sparse_remove_section(ms, pfn, nr_pages, map_offset, altmap);
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700519}
520
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700521/**
David Hildenbrandfeee6b22020-01-04 12:59:33 -0800522 * __remove_pages() - remove sections of pages
Dan Williams7ea62162019-07-18 15:58:22 -0700523 * @pfn: starting pageframe (must be aligned to start of a section)
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700524 * @nr_pages: number of pages to remove (must be multiple of section size)
Mike Rapoporte8b098f2018-04-05 16:24:57 -0700525 * @altmap: alternative device page map or %NULL if default memmap is used
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700526 *
527 * Generic helper function to remove section mappings and sysfs entries
528 * for the section of the memory we are removing. Caller needs to make
529 * sure that pages are marked reserved and zones are adjust properly by
530 * calling offline_pages().
531 */
David Hildenbrandfeee6b22020-01-04 12:59:33 -0800532void __remove_pages(unsigned long pfn, unsigned long nr_pages,
533 struct vmem_altmap *altmap)
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700534{
David Hildenbrand52fb87c2020-02-03 17:34:23 -0800535 const unsigned long end_pfn = pfn + nr_pages;
536 unsigned long cur_nr_pages;
Dan Williams4b94ffd2016-01-15 16:56:22 -0800537 unsigned long map_offset = 0;
Dan Williams4b94ffd2016-01-15 16:56:22 -0800538
Dan Williams96da4352019-07-18 15:58:15 -0700539 map_offset = vmem_altmap_offset(altmap);
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700540
Dan Williams7ea62162019-07-18 15:58:22 -0700541 if (check_pfn_span(pfn, nr_pages, "remove"))
542 return;
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700543
David Hildenbrand52fb87c2020-02-03 17:34:23 -0800544 for (; pfn < end_pfn; pfn += cur_nr_pages) {
Michal Hockodd33ad72018-11-02 15:48:46 -0700545 cond_resched();
David Hildenbrand52fb87c2020-02-03 17:34:23 -0800546 /* Select all remaining pages up to the next section boundary */
David Hildenbranda11b9412020-04-06 20:06:53 -0700547 cur_nr_pages = min(end_pfn - pfn,
548 SECTION_ALIGN_UP(pfn + 1) - pfn);
David Hildenbrand52fb87c2020-02-03 17:34:23 -0800549 __remove_section(pfn, cur_nr_pages, map_offset, altmap);
Dan Williams4b94ffd2016-01-15 16:56:22 -0800550 map_offset = 0;
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700551 }
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700552}
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700553
Daniel Kiper9d0ad8c2011-07-25 17:12:05 -0700554int set_online_page_callback(online_page_callback_t callback)
555{
556 int rc = -EINVAL;
557
Vladimir Davydovbfc8c902014-06-04 16:07:18 -0700558 get_online_mems();
559 mutex_lock(&online_page_callback_lock);
Daniel Kiper9d0ad8c2011-07-25 17:12:05 -0700560
561 if (online_page_callback == generic_online_page) {
562 online_page_callback = callback;
563 rc = 0;
564 }
565
Vladimir Davydovbfc8c902014-06-04 16:07:18 -0700566 mutex_unlock(&online_page_callback_lock);
567 put_online_mems();
Daniel Kiper9d0ad8c2011-07-25 17:12:05 -0700568
569 return rc;
570}
571EXPORT_SYMBOL_GPL(set_online_page_callback);
572
573int restore_online_page_callback(online_page_callback_t callback)
574{
575 int rc = -EINVAL;
576
Vladimir Davydovbfc8c902014-06-04 16:07:18 -0700577 get_online_mems();
578 mutex_lock(&online_page_callback_lock);
Daniel Kiper9d0ad8c2011-07-25 17:12:05 -0700579
580 if (online_page_callback == callback) {
581 online_page_callback = generic_online_page;
582 rc = 0;
583 }
584
Vladimir Davydovbfc8c902014-06-04 16:07:18 -0700585 mutex_unlock(&online_page_callback_lock);
586 put_online_mems();
Daniel Kiper9d0ad8c2011-07-25 17:12:05 -0700587
588 return rc;
589}
590EXPORT_SYMBOL_GPL(restore_online_page_callback);
591
David Hildenbrand18db1492019-11-30 17:53:51 -0800592void generic_online_page(struct page *page, unsigned int order)
Daniel Kiper9d0ad8c2011-07-25 17:12:05 -0700593{
Vlastimil Babkac87cbc12020-03-05 22:28:42 -0800594 /*
595 * Freeing the page with debug_pagealloc enabled will try to unmap it,
596 * so we should map it first. This is better than introducing a special
597 * case in page freeing fast path.
598 */
599 if (debug_pagealloc_enabled_static())
600 kernel_map_pages(page, 1 << order, 1);
Arun KSa9cd4102019-03-05 15:42:14 -0800601 __free_pages_core(page, order);
602 totalram_pages_add(1UL << order);
603#ifdef CONFIG_HIGHMEM
604 if (PageHighMem(page))
605 totalhigh_pages_add(1UL << order);
606#endif
607}
David Hildenbrand18db1492019-11-30 17:53:51 -0800608EXPORT_SYMBOL_GPL(generic_online_page);
Arun KSa9cd4102019-03-05 15:42:14 -0800609
KAMEZAWA Hiroyuki75884fb2007-10-16 01:26:10 -0700610static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages,
611 void *arg)
Dave Hansen3947be12005-10-29 18:16:54 -0700612{
David Hildenbrandb2c2ab22019-09-23 15:36:02 -0700613 const unsigned long end_pfn = start_pfn + nr_pages;
614 unsigned long pfn;
615 int order;
Michal Hocko2d070ea2017-07-06 15:37:56 -0700616
David Hildenbrandb2c2ab22019-09-23 15:36:02 -0700617 /*
618 * Online the pages. The callback might decide to keep some pages
619 * PG_reserved (to add them to the buddy later), but we still account
620 * them as being online/belonging to this zone ("present").
621 */
622 for (pfn = start_pfn; pfn < end_pfn; pfn += 1ul << order) {
623 order = min(MAX_ORDER - 1, get_order(PFN_PHYS(end_pfn - pfn)));
David Hildenbrandbd02cc02019-09-23 15:36:05 -0700624 /* __free_pages_core() wants pfns to be aligned to the order */
625 if (WARN_ON_ONCE(!IS_ALIGNED(pfn, 1ul << order)))
626 order = 0;
David Hildenbrandb2c2ab22019-09-23 15:36:02 -0700627 (*online_page_callback)(pfn_to_page(pfn), order);
628 }
Michal Hocko2d070ea2017-07-06 15:37:56 -0700629
David Hildenbrandb2c2ab22019-09-23 15:36:02 -0700630 /* mark all involved sections as online */
631 online_mem_sections(start_pfn, end_pfn);
632
633 *(unsigned long *)arg += nr_pages;
KAMEZAWA Hiroyuki75884fb2007-10-16 01:26:10 -0700634 return 0;
635}
636
Lai Jiangshand9713672012-12-11 16:01:03 -0800637/* check which state of node_states will be changed when online memory */
638static void node_states_check_changes_online(unsigned long nr_pages,
639 struct zone *zone, struct memory_notify *arg)
640{
641 int nid = zone_to_nid(zone);
Lai Jiangshand9713672012-12-11 16:01:03 -0800642
Anshuman Khandual98fa15f2019-03-05 15:42:58 -0800643 arg->status_change_nid = NUMA_NO_NODE;
644 arg->status_change_nid_normal = NUMA_NO_NODE;
645 arg->status_change_nid_high = NUMA_NO_NODE;
Lai Jiangshand9713672012-12-11 16:01:03 -0800646
Lai Jiangshan6715ddf2012-12-12 13:51:49 -0800647 if (!node_state(nid, N_MEMORY))
Lai Jiangshand9713672012-12-11 16:01:03 -0800648 arg->status_change_nid = nid;
Oscar Salvador8efe33f2018-10-26 15:07:34 -0700649 if (zone_idx(zone) <= ZONE_NORMAL && !node_state(nid, N_NORMAL_MEMORY))
650 arg->status_change_nid_normal = nid;
651#ifdef CONFIG_HIGHMEM
Baoquan Hed3ba3ae2019-05-13 17:17:35 -0700652 if (zone_idx(zone) <= ZONE_HIGHMEM && !node_state(nid, N_HIGH_MEMORY))
Oscar Salvador8efe33f2018-10-26 15:07:34 -0700653 arg->status_change_nid_high = nid;
654#endif
Lai Jiangshand9713672012-12-11 16:01:03 -0800655}
656
657static void node_states_set_node(int node, struct memory_notify *arg)
658{
659 if (arg->status_change_nid_normal >= 0)
660 node_set_state(node, N_NORMAL_MEMORY);
661
Lai Jiangshan6715ddf2012-12-12 13:51:49 -0800662 if (arg->status_change_nid_high >= 0)
663 node_set_state(node, N_HIGH_MEMORY);
664
Oscar Salvador83d83612018-10-26 15:07:25 -0700665 if (arg->status_change_nid >= 0)
666 node_set_state(node, N_MEMORY);
Lai Jiangshand9713672012-12-11 16:01:03 -0800667}
668
Michal Hockof1dd2cd2017-07-06 15:38:11 -0700669static void __meminit resize_zone_range(struct zone *zone, unsigned long start_pfn,
670 unsigned long nr_pages)
671{
672 unsigned long old_end_pfn = zone_end_pfn(zone);
673
674 if (zone_is_empty(zone) || start_pfn < zone->zone_start_pfn)
675 zone->zone_start_pfn = start_pfn;
676
677 zone->spanned_pages = max(start_pfn + nr_pages, old_end_pfn) - zone->zone_start_pfn;
678}
679
680static void __meminit resize_pgdat_range(struct pglist_data *pgdat, unsigned long start_pfn,
681 unsigned long nr_pages)
682{
683 unsigned long old_end_pfn = pgdat_end_pfn(pgdat);
684
685 if (!pgdat->node_spanned_pages || start_pfn < pgdat->node_start_pfn)
686 pgdat->node_start_pfn = start_pfn;
687
688 pgdat->node_spanned_pages = max(start_pfn + nr_pages, old_end_pfn) - pgdat->node_start_pfn;
Michal Hockof1dd2cd2017-07-06 15:38:11 -0700689
David Hildenbrand3fccb742019-09-23 15:35:37 -0700690}
691/*
692 * Associate the pfn range with the given zone, initializing the memmaps
693 * and resizing the pgdat/zone data to span the added pages. After this
694 * call, all affected pages are PG_reserved.
695 */
Christoph Hellwiga99583e2017-12-29 08:53:57 +0100696void __ref move_pfn_range_to_zone(struct zone *zone, unsigned long start_pfn,
697 unsigned long nr_pages, struct vmem_altmap *altmap)
Michal Hockof1dd2cd2017-07-06 15:38:11 -0700698{
699 struct pglist_data *pgdat = zone->zone_pgdat;
700 int nid = pgdat->node_id;
701 unsigned long flags;
702
Michal Hockof1dd2cd2017-07-06 15:38:11 -0700703 clear_zone_contiguous(zone);
704
705 /* TODO Huh pgdat is irqsave while zone is not. It used to be like that before */
706 pgdat_resize_lock(pgdat, &flags);
707 zone_span_writelock(zone);
Wei Yangfa004ab2018-12-28 00:37:10 -0800708 if (zone_is_empty(zone))
709 init_currently_empty_zone(zone, start_pfn, nr_pages);
Michal Hockof1dd2cd2017-07-06 15:38:11 -0700710 resize_zone_range(zone, start_pfn, nr_pages);
711 zone_span_writeunlock(zone);
712 resize_pgdat_range(pgdat, start_pfn, nr_pages);
713 pgdat_resize_unlock(pgdat, &flags);
714
715 /*
716 * TODO now we have a visible range of pages which are not associated
717 * with their zone properly. Not nice but set_pfnblock_flags_mask
718 * expects the zone spans the pfn range. All the pages in the range
719 * are reserved so nobody should be touching them so we should be safe
720 */
Christoph Hellwiga99583e2017-12-29 08:53:57 +0100721 memmap_init_zone(nr_pages, nid, zone_idx(zone), start_pfn,
722 MEMMAP_HOTPLUG, altmap);
Michal Hockof1dd2cd2017-07-06 15:38:11 -0700723
724 set_zone_contiguous(zone);
725}
726
727/*
Michal Hockoc246a212017-07-06 15:38:18 -0700728 * Returns a default kernel memory zone for the given pfn range.
729 * If no kernel zone covers this pfn range it will automatically go
730 * to the ZONE_NORMAL.
731 */
Michal Hockoc6f03e22017-09-06 16:19:40 -0700732static struct zone *default_kernel_zone_for_pfn(int nid, unsigned long start_pfn,
Michal Hockoc246a212017-07-06 15:38:18 -0700733 unsigned long nr_pages)
734{
735 struct pglist_data *pgdat = NODE_DATA(nid);
736 int zid;
737
738 for (zid = 0; zid <= ZONE_NORMAL; zid++) {
739 struct zone *zone = &pgdat->node_zones[zid];
740
741 if (zone_intersects(zone, start_pfn, nr_pages))
742 return zone;
743 }
744
745 return &pgdat->node_zones[ZONE_NORMAL];
746}
747
Michal Hockoc6f03e22017-09-06 16:19:40 -0700748static inline struct zone *default_zone_for_pfn(int nid, unsigned long start_pfn,
749 unsigned long nr_pages)
Michal Hockoe5e68932017-09-06 16:19:37 -0700750{
Michal Hockoc6f03e22017-09-06 16:19:40 -0700751 struct zone *kernel_zone = default_kernel_zone_for_pfn(nid, start_pfn,
752 nr_pages);
753 struct zone *movable_zone = &NODE_DATA(nid)->node_zones[ZONE_MOVABLE];
754 bool in_kernel = zone_intersects(kernel_zone, start_pfn, nr_pages);
755 bool in_movable = zone_intersects(movable_zone, start_pfn, nr_pages);
Michal Hockoe5e68932017-09-06 16:19:37 -0700756
757 /*
Michal Hockoc6f03e22017-09-06 16:19:40 -0700758 * We inherit the existing zone in a simple case where zones do not
759 * overlap in the given range
Michal Hockoe5e68932017-09-06 16:19:37 -0700760 */
Michal Hockoc6f03e22017-09-06 16:19:40 -0700761 if (in_kernel ^ in_movable)
762 return (in_kernel) ? kernel_zone : movable_zone;
Michal Hockoe5e68932017-09-06 16:19:37 -0700763
Michal Hockoc6f03e22017-09-06 16:19:40 -0700764 /*
765 * If the range doesn't belong to any zone or two zones overlap in the
766 * given range then we use movable zone only if movable_node is
767 * enabled because we always online to a kernel zone by default.
768 */
769 return movable_node_enabled ? movable_zone : kernel_zone;
Michal Hocko9f123ab2017-07-10 15:48:37 -0700770}
771
Michal Hockoe5e68932017-09-06 16:19:37 -0700772struct zone * zone_for_pfn_range(int online_type, int nid, unsigned start_pfn,
773 unsigned long nr_pages)
Michal Hockof1dd2cd2017-07-06 15:38:11 -0700774{
Michal Hockoc6f03e22017-09-06 16:19:40 -0700775 if (online_type == MMOP_ONLINE_KERNEL)
776 return default_kernel_zone_for_pfn(nid, start_pfn, nr_pages);
Michal Hockof1dd2cd2017-07-06 15:38:11 -0700777
Michal Hockoc6f03e22017-09-06 16:19:40 -0700778 if (online_type == MMOP_ONLINE_MOVABLE)
779 return &NODE_DATA(nid)->node_zones[ZONE_MOVABLE];
Reza Arbabdf429ac2016-07-26 15:22:23 -0700780
Michal Hockoc6f03e22017-09-06 16:19:40 -0700781 return default_zone_for_pfn(nid, start_pfn, nr_pages);
Michal Hockoe5e68932017-09-06 16:19:37 -0700782}
783
David Hildenbrandbd5c2342020-01-30 22:14:54 -0800784int __ref online_pages(unsigned long pfn, unsigned long nr_pages,
785 int online_type, int nid)
KAMEZAWA Hiroyuki75884fb2007-10-16 01:26:10 -0700786{
Cody P Schaferaa472282013-07-03 15:02:10 -0700787 unsigned long flags;
Dave Hansen3947be12005-10-29 18:16:54 -0700788 unsigned long onlined_pages = 0;
789 struct zone *zone;
Yasunori Goto68113782006-06-23 02:03:11 -0700790 int need_zonelists_rebuild = 0;
Yasunori Goto7b78d332007-10-21 16:41:36 -0700791 int ret;
792 struct memory_notify arg;
Dave Hansen3947be12005-10-29 18:16:54 -0700793
David Hildenbrand381eab42018-10-30 15:10:29 -0700794 mem_hotplug_begin();
795
Michal Hockof1dd2cd2017-07-06 15:38:11 -0700796 /* associate pfn range with the zone */
David Hildenbrand3fccb742019-09-23 15:35:37 -0700797 zone = zone_for_pfn_range(online_type, nid, pfn, nr_pages);
798 move_pfn_range_to_zone(zone, pfn, nr_pages, NULL);
Michal Hockof1dd2cd2017-07-06 15:38:11 -0700799
Yasunori Goto7b78d332007-10-21 16:41:36 -0700800 arg.start_pfn = pfn;
801 arg.nr_pages = nr_pages;
Lai Jiangshand9713672012-12-11 16:01:03 -0800802 node_states_check_changes_online(nr_pages, zone, &arg);
Yasunori Goto7b78d332007-10-21 16:41:36 -0700803
Yasunori Goto7b78d332007-10-21 16:41:36 -0700804 ret = memory_notify(MEM_GOING_ONLINE, &arg);
805 ret = notifier_to_errno(ret);
Chen Yuconge33e33b2016-03-17 14:19:35 -0700806 if (ret)
807 goto failed_addition;
808
Dave Hansen3947be12005-10-29 18:16:54 -0700809 /*
Yasunori Goto68113782006-06-23 02:03:11 -0700810 * If this zone is not populated, then it is not in zonelist.
811 * This means the page allocator ignores this zone.
812 * So, zonelist must be updated after online.
813 */
Wen Congyang6dcd73d2012-12-11 16:01:01 -0800814 if (!populated_zone(zone)) {
Yasunori Goto68113782006-06-23 02:03:11 -0700815 need_zonelists_rebuild = 1;
Michal Hocko72675e12017-09-06 16:20:24 -0700816 setup_zone_pageset(zone);
Wen Congyang6dcd73d2012-12-11 16:01:01 -0800817 }
Yasunori Goto68113782006-06-23 02:03:11 -0700818
KAMEZAWA Hiroyuki908eedc2009-09-22 16:45:46 -0700819 ret = walk_system_ram_range(pfn, nr_pages, &onlined_pages,
KAMEZAWA Hiroyuki75884fb2007-10-16 01:26:10 -0700820 online_pages_range);
Geoff Levandfd8a4222008-05-14 16:05:50 -0700821 if (ret) {
David Hildenbrandca9a46f2019-09-23 15:36:08 -0700822 /* not a single memory resource was applicable */
Wen Congyang6dcd73d2012-12-11 16:01:01 -0800823 if (need_zonelists_rebuild)
824 zone_pcp_reset(zone);
Chen Yuconge33e33b2016-03-17 14:19:35 -0700825 goto failed_addition;
Geoff Levandfd8a4222008-05-14 16:05:50 -0700826 }
827
Dave Hansen3947be12005-10-29 18:16:54 -0700828 zone->present_pages += onlined_pages;
Cody P Schaferaa472282013-07-03 15:02:10 -0700829
830 pgdat_resize_lock(zone->zone_pgdat, &flags);
Yasunori Gotof2937be2006-03-09 17:33:51 -0800831 zone->zone_pgdat->node_present_pages += onlined_pages;
Cody P Schaferaa472282013-07-03 15:02:10 -0700832 pgdat_resize_unlock(zone->zone_pgdat, &flags);
833
Dan Williamse900a912019-05-14 15:41:28 -0700834 shuffle_zone(zone);
835
David Hildenbrandca9a46f2019-09-23 15:36:08 -0700836 node_states_set_node(nid, &arg);
837 if (need_zonelists_rebuild)
838 build_all_zonelists(NULL);
839 else
840 zone_pcp_update(zone);
Dave Hansen3947be12005-10-29 18:16:54 -0700841
KOSAKI Motohiro1b79acc2011-05-24 17:11:32 -0700842 init_per_zone_wmark_min();
843
David Hildenbrandca9a46f2019-09-23 15:36:08 -0700844 kswapd_run(nid);
845 kcompactd_run(nid);
Dave Hansen61b13992005-10-29 18:16:56 -0700846
Chandra Seetharaman2d1d43f2006-09-29 02:01:25 -0700847 writeback_set_ratelimit();
Yasunori Goto7b78d332007-10-21 16:41:36 -0700848
David Hildenbrandca9a46f2019-09-23 15:36:08 -0700849 memory_notify(MEM_ONLINE, &arg);
David Hildenbrand381eab42018-10-30 15:10:29 -0700850 mem_hotplug_done();
David Rientjes30467e02015-04-14 15:45:11 -0700851 return 0;
Chen Yuconge33e33b2016-03-17 14:19:35 -0700852
853failed_addition:
854 pr_debug("online_pages [mem %#010llx-%#010llx] failed\n",
855 (unsigned long long) pfn << PAGE_SHIFT,
856 (((unsigned long long) pfn + nr_pages) << PAGE_SHIFT) - 1);
857 memory_notify(MEM_CANCEL_ONLINE, &arg);
David Hildenbrandfeee6b22020-01-04 12:59:33 -0800858 remove_pfn_range_from_zone(zone, pfn, nr_pages);
David Hildenbrand381eab42018-10-30 15:10:29 -0700859 mem_hotplug_done();
Chen Yuconge33e33b2016-03-17 14:19:35 -0700860 return ret;
Dave Hansen3947be12005-10-29 18:16:54 -0700861}
Keith Mannthey53947022006-09-30 23:27:08 -0700862#endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
Yasunori Gotobc02af92006-06-27 02:53:30 -0700863
Tang Chen0bd85422014-11-13 15:19:41 -0800864static void reset_node_present_pages(pg_data_t *pgdat)
865{
866 struct zone *z;
867
868 for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++)
869 z->present_pages = 0;
870
871 pgdat->node_present_pages = 0;
872}
873
Hidetoshi Setoe1319332009-11-17 14:06:18 -0800874/* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
David Hildenbrandc68ab182020-06-04 16:48:35 -0700875static pg_data_t __ref *hotadd_new_pgdat(int nid)
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700876{
877 struct pglist_data *pgdat;
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700878
Tang Chena1e565a2013-02-22 16:33:18 -0800879 pgdat = NODE_DATA(nid);
880 if (!pgdat) {
881 pgdat = arch_alloc_nodedata(nid);
882 if (!pgdat)
883 return NULL;
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700884
Wei Yang33fce012019-09-23 15:35:52 -0700885 pgdat->per_cpu_nodestats =
886 alloc_percpu(struct per_cpu_nodestat);
Tang Chena1e565a2013-02-22 16:33:18 -0800887 arch_refresh_nodedata(nid, pgdat);
Gu Zhengb0dc3a32015-03-25 15:55:20 -0700888 } else {
Wei Yang33fce012019-09-23 15:35:52 -0700889 int cpu;
Mel Gormane716f2e2017-05-03 14:53:45 -0700890 /*
Joonsoo Kim97a225e2020-06-03 15:59:01 -0700891 * Reset the nr_zones, order and highest_zoneidx before reuse.
892 * Note that kswapd will init kswapd_highest_zoneidx properly
Mel Gormane716f2e2017-05-03 14:53:45 -0700893 * when it starts in the near future.
894 */
Gu Zhengb0dc3a32015-03-25 15:55:20 -0700895 pgdat->nr_zones = 0;
Mel Gorman38087d92016-07-28 15:45:49 -0700896 pgdat->kswapd_order = 0;
Joonsoo Kim97a225e2020-06-03 15:59:01 -0700897 pgdat->kswapd_highest_zoneidx = 0;
Wei Yang33fce012019-09-23 15:35:52 -0700898 for_each_online_cpu(cpu) {
899 struct per_cpu_nodestat *p;
900
901 p = per_cpu_ptr(pgdat->per_cpu_nodestats, cpu);
902 memset(p, 0, sizeof(*p));
903 }
Tang Chena1e565a2013-02-22 16:33:18 -0800904 }
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700905
906 /* we can use NODE_DATA(nid) from here */
Oscar Salvador03e85f92018-08-21 21:53:43 -0700907 pgdat->node_id = nid;
David Hildenbrandc68ab182020-06-04 16:48:35 -0700908 pgdat->node_start_pfn = 0;
Oscar Salvador03e85f92018-08-21 21:53:43 -0700909
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700910 /* init node's zones as empty zones, we don't have any present pages.*/
Oscar Salvador03e85f92018-08-21 21:53:43 -0700911 free_area_init_core_hotplug(nid);
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700912
KAMEZAWA Hiroyuki959ecc42011-06-15 15:08:38 -0700913 /*
914 * The node we allocated has no zone fallback lists. For avoiding
915 * to access not-initialized zonelist, build here.
916 */
Michal Hocko72675e12017-09-06 16:20:24 -0700917 build_all_zonelists(pgdat);
KAMEZAWA Hiroyuki959ecc42011-06-15 15:08:38 -0700918
Tang Chenf784a3f2014-11-13 15:19:39 -0800919 /*
Tang Chen0bd85422014-11-13 15:19:41 -0800920 * When memory is hot-added, all the memory is in offline state. So
921 * clear all zones' present_pages because they will be updated in
922 * online_pages() and offline_pages().
923 */
Oscar Salvador03e85f92018-08-21 21:53:43 -0700924 reset_node_managed_pages(pgdat);
Tang Chen0bd85422014-11-13 15:19:41 -0800925 reset_node_present_pages(pgdat);
926
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700927 return pgdat;
928}
929
Oscar Salvadorb9ff0362018-08-17 15:46:15 -0700930static void rollback_node_hotadd(int nid)
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700931{
Oscar Salvadorb9ff0362018-08-17 15:46:15 -0700932 pg_data_t *pgdat = NODE_DATA(nid);
933
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700934 arch_refresh_nodedata(nid, NULL);
Reza Arbab58301692016-08-11 15:33:12 -0700935 free_percpu(pgdat->per_cpu_nodestats);
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700936 arch_free_nodedata(pgdat);
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700937}
938
KAMEZAWA Hiroyuki0a547032006-06-27 02:53:35 -0700939
Toshi Kani01b0f192013-11-12 15:07:25 -0800940/**
941 * try_online_node - online a node if offlined
Mike Rapoporte8b098f2018-04-05 16:24:57 -0700942 * @nid: the node ID
Oscar Salvadorb9ff0362018-08-17 15:46:15 -0700943 * @set_node_online: Whether we want to online the node
minskey guocf234222010-05-24 14:32:41 -0700944 * called by cpu_up() to online a node without onlined memory.
Oscar Salvadorb9ff0362018-08-17 15:46:15 -0700945 *
946 * Returns:
947 * 1 -> a new node has been allocated
948 * 0 -> the node is already online
949 * -ENOMEM -> the node could not be allocated
minskey guocf234222010-05-24 14:32:41 -0700950 */
David Hildenbrandc68ab182020-06-04 16:48:35 -0700951static int __try_online_node(int nid, bool set_node_online)
minskey guocf234222010-05-24 14:32:41 -0700952{
Oscar Salvadorb9ff0362018-08-17 15:46:15 -0700953 pg_data_t *pgdat;
954 int ret = 1;
minskey guocf234222010-05-24 14:32:41 -0700955
Toshi Kani01b0f192013-11-12 15:07:25 -0800956 if (node_online(nid))
957 return 0;
958
David Hildenbrandc68ab182020-06-04 16:48:35 -0700959 pgdat = hotadd_new_pgdat(nid);
David Rientjes7553e8f2011-06-22 18:13:01 -0700960 if (!pgdat) {
Toshi Kani01b0f192013-11-12 15:07:25 -0800961 pr_err("Cannot online node %d due to NULL pgdat\n", nid);
minskey guocf234222010-05-24 14:32:41 -0700962 ret = -ENOMEM;
963 goto out;
964 }
Oscar Salvadorb9ff0362018-08-17 15:46:15 -0700965
966 if (set_node_online) {
967 node_set_online(nid);
968 ret = register_one_node(nid);
969 BUG_ON(ret);
970 }
minskey guocf234222010-05-24 14:32:41 -0700971out:
Oscar Salvadorb9ff0362018-08-17 15:46:15 -0700972 return ret;
973}
974
975/*
976 * Users of this function always want to online/register the node
977 */
978int try_online_node(int nid)
979{
980 int ret;
981
982 mem_hotplug_begin();
David Hildenbrandc68ab182020-06-04 16:48:35 -0700983 ret = __try_online_node(nid, true);
Vladimir Davydovbfc8c902014-06-04 16:07:18 -0700984 mem_hotplug_done();
minskey guocf234222010-05-24 14:32:41 -0700985 return ret;
986}
987
Toshi Kani27356f52013-09-11 14:21:49 -0700988static int check_hotplug_memory_range(u64 start, u64 size)
989{
Pavel Tatashinba325582018-04-05 16:22:39 -0700990 /* memory range must be block size aligned */
David Hildenbrandcec3ebd2019-07-18 15:56:25 -0700991 if (!size || !IS_ALIGNED(start, memory_block_size_bytes()) ||
992 !IS_ALIGNED(size, memory_block_size_bytes())) {
Pavel Tatashinba325582018-04-05 16:22:39 -0700993 pr_err("Block size [%#lx] unaligned hotplug range: start %#llx, size %#llx",
David Hildenbrandcec3ebd2019-07-18 15:56:25 -0700994 memory_block_size_bytes(), start, size);
Toshi Kani27356f52013-09-11 14:21:49 -0700995 return -EINVAL;
996 }
997
998 return 0;
999}
1000
Vitaly Kuznetsov31bc3852016-03-15 14:56:48 -07001001static int online_memory_block(struct memory_block *mem, void *arg)
1002{
David Hildenbrand862919e2020-04-06 20:07:40 -07001003 mem->online_type = memhp_default_online_type;
Nathan Fontenotdc18d702017-02-24 15:00:02 -08001004 return device_online(&mem->dev);
Vitaly Kuznetsov31bc3852016-03-15 14:56:48 -07001005}
1006
David Hildenbrand8df1d0e2018-10-30 15:10:24 -07001007/*
1008 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1009 * and online/offline operations (triggered e.g. by sysfs).
1010 *
1011 * we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG
1012 */
David Hildenbrandf29d8e9c02018-12-28 00:35:36 -08001013int __ref add_memory_resource(int nid, struct resource *res)
Yasunori Gotobc02af92006-06-27 02:53:30 -07001014{
Logan Gunthorpebfeb0222020-04-10 14:33:36 -07001015 struct mhp_params params = { .pgprot = PAGE_KERNEL };
David Vrabel62cedb92015-06-25 16:35:49 +01001016 u64 start, size;
Oscar Salvadorb9ff0362018-08-17 15:46:15 -07001017 bool new_node = false;
Yasunori Gotobc02af92006-06-27 02:53:30 -07001018 int ret;
1019
David Vrabel62cedb92015-06-25 16:35:49 +01001020 start = res->start;
1021 size = resource_size(res);
1022
Toshi Kani27356f52013-09-11 14:21:49 -07001023 ret = check_hotplug_memory_range(start, size);
1024 if (ret)
1025 return ret;
1026
Vishal Vermafa6d9ec2020-06-04 16:48:25 -07001027 if (!node_possible(nid)) {
1028 WARN(1, "node %d was absent from the node_possible_map\n", nid);
1029 return -EINVAL;
1030 }
1031
Vladimir Davydovbfc8c902014-06-04 16:07:18 -07001032 mem_hotplug_begin();
Nathan Zimmerac13c462014-01-23 15:53:26 -08001033
David Hildenbrand52219ae2020-06-04 16:48:38 -07001034 if (IS_ENABLED(CONFIG_ARCH_KEEP_MEMBLOCK))
1035 memblock_add_node(start, size, nid);
Tang Chen7f36e3e2015-09-04 15:42:32 -07001036
David Hildenbrandc68ab182020-06-04 16:48:35 -07001037 ret = __try_online_node(nid, false);
Oscar Salvadorb9ff0362018-08-17 15:46:15 -07001038 if (ret < 0)
1039 goto error;
1040 new_node = ret;
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001041
Yasunori Gotobc02af92006-06-27 02:53:30 -07001042 /* call arch's memory hotadd */
Logan Gunthorpef5637d32020-04-10 14:33:21 -07001043 ret = arch_add_memory(nid, start, size, &params);
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001044 if (ret < 0)
1045 goto error;
1046
David Hildenbranddb051a02019-07-18 15:56:56 -07001047 /* create memory block devices after memory was added */
1048 ret = create_memory_block_devices(start, size);
1049 if (ret) {
1050 arch_remove_memory(nid, start, size, NULL);
1051 goto error;
1052 }
1053
Tang Chena1e565a2013-02-22 16:33:18 -08001054 if (new_node) {
Oscar Salvadord5b6f6a32018-08-17 15:46:18 -07001055 /* If sysfs file of new node can't be created, cpu on the node
Yasunori Goto0fc44152006-06-27 02:53:38 -07001056 * can't be hot-added. There is no rollback way now.
1057 * So, check by BUG_ON() to catch it reluctantly..
Oscar Salvadord5b6f6a32018-08-17 15:46:18 -07001058 * We online node here. We can't roll back from here.
Yasunori Goto0fc44152006-06-27 02:53:38 -07001059 */
Oscar Salvadord5b6f6a32018-08-17 15:46:18 -07001060 node_set_online(nid);
1061 ret = __register_one_node(nid);
Yasunori Goto0fc44152006-06-27 02:53:38 -07001062 BUG_ON(ret);
1063 }
1064
Oscar Salvadord5b6f6a32018-08-17 15:46:18 -07001065 /* link memory sections under this node.*/
Oscar Salvador4fbce632018-08-17 15:46:22 -07001066 ret = link_mem_sections(nid, PFN_DOWN(start), PFN_UP(start + size - 1));
Oscar Salvadord5b6f6a32018-08-17 15:46:18 -07001067 BUG_ON(ret);
1068
akpm@linux-foundation.orgd96ae532010-03-05 13:41:58 -08001069 /* create new memmap entry */
David Hildenbrand7b7b2722020-06-04 16:48:41 -07001070 if (!strcmp(res->name, "System RAM"))
1071 firmware_map_add_hotplug(start, start + size, "System RAM");
akpm@linux-foundation.orgd96ae532010-03-05 13:41:58 -08001072
David Hildenbrand381eab42018-10-30 15:10:29 -07001073 /* device_online() will take the lock when calling online_pages() */
1074 mem_hotplug_done();
1075
Vitaly Kuznetsov31bc3852016-03-15 14:56:48 -07001076 /* online pages if requested */
David Hildenbrand862919e2020-04-06 20:07:40 -07001077 if (memhp_default_online_type != MMOP_OFFLINE)
David Hildenbrandfbcf73c2019-07-18 15:57:46 -07001078 walk_memory_blocks(start, size, NULL, online_memory_block);
Vitaly Kuznetsov31bc3852016-03-15 14:56:48 -07001079
David Hildenbrand381eab42018-10-30 15:10:29 -07001080 return ret;
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001081error:
1082 /* rollback pgdat allocation and others */
Oscar Salvadorb9ff0362018-08-17 15:46:15 -07001083 if (new_node)
1084 rollback_node_hotadd(nid);
David Hildenbrand52219ae2020-06-04 16:48:38 -07001085 if (IS_ENABLED(CONFIG_ARCH_KEEP_MEMBLOCK))
1086 memblock_remove(start, size);
Vladimir Davydovbfc8c902014-06-04 16:07:18 -07001087 mem_hotplug_done();
Yasunori Gotobc02af92006-06-27 02:53:30 -07001088 return ret;
1089}
David Vrabel62cedb92015-06-25 16:35:49 +01001090
David Hildenbrand8df1d0e2018-10-30 15:10:24 -07001091/* requires device_hotplug_lock, see add_memory_resource() */
1092int __ref __add_memory(int nid, u64 start, u64 size)
David Vrabel62cedb92015-06-25 16:35:49 +01001093{
1094 struct resource *res;
1095 int ret;
1096
David Hildenbrand7b7b2722020-06-04 16:48:41 -07001097 res = register_memory_resource(start, size, "System RAM");
Vitaly Kuznetsov6f754ba2016-01-14 15:21:55 -08001098 if (IS_ERR(res))
1099 return PTR_ERR(res);
David Vrabel62cedb92015-06-25 16:35:49 +01001100
David Hildenbrandf29d8e9c02018-12-28 00:35:36 -08001101 ret = add_memory_resource(nid, res);
David Vrabel62cedb92015-06-25 16:35:49 +01001102 if (ret < 0)
1103 release_memory_resource(res);
1104 return ret;
1105}
David Hildenbrand8df1d0e2018-10-30 15:10:24 -07001106
1107int add_memory(int nid, u64 start, u64 size)
1108{
1109 int rc;
1110
1111 lock_device_hotplug();
1112 rc = __add_memory(nid, start, size);
1113 unlock_device_hotplug();
1114
1115 return rc;
1116}
Yasunori Gotobc02af92006-06-27 02:53:30 -07001117EXPORT_SYMBOL_GPL(add_memory);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001118
David Hildenbrand7b7b2722020-06-04 16:48:41 -07001119/*
1120 * Add special, driver-managed memory to the system as system RAM. Such
1121 * memory is not exposed via the raw firmware-provided memmap as system
1122 * RAM, instead, it is detected and added by a driver - during cold boot,
1123 * after a reboot, and after kexec.
1124 *
1125 * Reasons why this memory should not be used for the initial memmap of a
1126 * kexec kernel or for placing kexec images:
1127 * - The booting kernel is in charge of determining how this memory will be
1128 * used (e.g., use persistent memory as system RAM)
1129 * - Coordination with a hypervisor is required before this memory
1130 * can be used (e.g., inaccessible parts).
1131 *
1132 * For this memory, no entries in /sys/firmware/memmap ("raw firmware-provided
1133 * memory map") are created. Also, the created memory resource is flagged
1134 * with IORESOURCE_MEM_DRIVER_MANAGED, so in-kernel users can special-case
1135 * this memory as well (esp., not place kexec images onto it).
1136 *
1137 * The resource_name (visible via /proc/iomem) has to have the format
1138 * "System RAM ($DRIVER)".
1139 */
1140int add_memory_driver_managed(int nid, u64 start, u64 size,
1141 const char *resource_name)
1142{
1143 struct resource *res;
1144 int rc;
1145
1146 if (!resource_name ||
1147 strstr(resource_name, "System RAM (") != resource_name ||
1148 resource_name[strlen(resource_name) - 1] != ')')
1149 return -EINVAL;
1150
1151 lock_device_hotplug();
1152
1153 res = register_memory_resource(start, size, resource_name);
1154 if (IS_ERR(res)) {
1155 rc = PTR_ERR(res);
1156 goto out_unlock;
1157 }
1158
1159 rc = add_memory_resource(nid, res);
1160 if (rc < 0)
1161 release_memory_resource(res);
1162
1163out_unlock:
1164 unlock_device_hotplug();
1165 return rc;
1166}
1167EXPORT_SYMBOL_GPL(add_memory_driver_managed);
1168
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001169#ifdef CONFIG_MEMORY_HOTREMOVE
1170/*
David Hildenbrand92917992020-02-03 17:34:26 -08001171 * Confirm all pages in a range [start, end) belong to the same zone (skipping
1172 * memory holes). When true, return the zone.
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001173 */
David Hildenbrand92917992020-02-03 17:34:26 -08001174struct zone *test_pages_in_a_zone(unsigned long start_pfn,
1175 unsigned long end_pfn)
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001176{
Andrew Banman5f0f2882015-12-29 14:54:25 -08001177 unsigned long pfn, sec_end_pfn;
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001178 struct zone *zone = NULL;
1179 struct page *page;
1180 int i;
Toshi Kanideb88a22017-02-03 13:13:20 -08001181 for (pfn = start_pfn, sec_end_pfn = SECTION_ALIGN_UP(start_pfn + 1);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001182 pfn < end_pfn;
Toshi Kanideb88a22017-02-03 13:13:20 -08001183 pfn = sec_end_pfn, sec_end_pfn += PAGES_PER_SECTION) {
Andrew Banman5f0f2882015-12-29 14:54:25 -08001184 /* Make sure the memory section is present first */
1185 if (!present_section_nr(pfn_to_section_nr(pfn)))
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001186 continue;
Andrew Banman5f0f2882015-12-29 14:54:25 -08001187 for (; pfn < sec_end_pfn && pfn < end_pfn;
1188 pfn += MAX_ORDER_NR_PAGES) {
1189 i = 0;
1190 /* This is just a CONFIG_HOLES_IN_ZONE check.*/
1191 while ((i < MAX_ORDER_NR_PAGES) &&
1192 !pfn_valid_within(pfn + i))
1193 i++;
zhong jiangd6d8c8a2017-02-24 14:59:30 -08001194 if (i == MAX_ORDER_NR_PAGES || pfn + i >= end_pfn)
Andrew Banman5f0f2882015-12-29 14:54:25 -08001195 continue;
Mikhail Zaslonko24feb472019-02-01 14:20:38 -08001196 /* Check if we got outside of the zone */
1197 if (zone && !zone_spans_pfn(zone, pfn + i))
David Hildenbrand92917992020-02-03 17:34:26 -08001198 return NULL;
Andrew Banman5f0f2882015-12-29 14:54:25 -08001199 page = pfn_to_page(pfn + i);
1200 if (zone && page_zone(page) != zone)
David Hildenbrand92917992020-02-03 17:34:26 -08001201 return NULL;
Andrew Banman5f0f2882015-12-29 14:54:25 -08001202 zone = page_zone(page);
1203 }
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001204 }
Toshi Kanideb88a22017-02-03 13:13:20 -08001205
David Hildenbrand92917992020-02-03 17:34:26 -08001206 return zone;
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001207}
1208
1209/*
Yisheng Xie0efadf42017-02-24 14:57:39 -08001210 * Scan pfn range [start,end) to find movable/migratable pages (LRU pages,
David Hildenbrandaa218792020-05-07 16:01:30 +02001211 * non-lru movable pages and hugepages). Will skip over most unmovable
1212 * pages (esp., pages that can be skipped when offlining), but bail out on
1213 * definitely unmovable pages.
1214 *
1215 * Returns:
1216 * 0 in case a movable page is found and movable_pfn was updated.
1217 * -ENOENT in case no movable page was found.
1218 * -EBUSY in case a definitely unmovable page was found.
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001219 */
David Hildenbrandaa218792020-05-07 16:01:30 +02001220static int scan_movable_pages(unsigned long start, unsigned long end,
1221 unsigned long *movable_pfn)
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001222{
1223 unsigned long pfn;
Oscar Salvadoreeb0efd2019-02-01 14:20:47 -08001224
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001225 for (pfn = start; pfn < end; pfn++) {
Oscar Salvadoreeb0efd2019-02-01 14:20:47 -08001226 struct page *page, *head;
1227 unsigned long skip;
1228
1229 if (!pfn_valid(pfn))
1230 continue;
1231 page = pfn_to_page(pfn);
1232 if (PageLRU(page))
David Hildenbrandaa218792020-05-07 16:01:30 +02001233 goto found;
Oscar Salvadoreeb0efd2019-02-01 14:20:47 -08001234 if (__PageMovable(page))
David Hildenbrandaa218792020-05-07 16:01:30 +02001235 goto found;
1236
1237 /*
1238 * PageOffline() pages that are not marked __PageMovable() and
1239 * have a reference count > 0 (after MEM_GOING_OFFLINE) are
1240 * definitely unmovable. If their reference count would be 0,
1241 * they could at least be skipped when offlining memory.
1242 */
1243 if (PageOffline(page) && page_count(page))
1244 return -EBUSY;
Oscar Salvadoreeb0efd2019-02-01 14:20:47 -08001245
1246 if (!PageHuge(page))
1247 continue;
1248 head = compound_head(page);
Oscar Salvador39186cb2019-05-13 17:17:32 -07001249 if (page_huge_active(head))
David Hildenbrandaa218792020-05-07 16:01:30 +02001250 goto found;
Matthew Wilcox (Oracle)d8c65462019-09-23 15:34:30 -07001251 skip = compound_nr(head) - (page - head);
Oscar Salvadoreeb0efd2019-02-01 14:20:47 -08001252 pfn += skip - 1;
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001253 }
David Hildenbrandaa218792020-05-07 16:01:30 +02001254 return -ENOENT;
1255found:
1256 *movable_pfn = pfn;
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001257 return 0;
1258}
1259
Michal Hocko666feb22018-04-10 16:30:03 -07001260static struct page *new_node_page(struct page *page, unsigned long private)
Xishi Qiu394e31d2016-07-28 15:48:53 -07001261{
Xishi Qiu394e31d2016-07-28 15:48:53 -07001262 int nid = page_to_nid(page);
Li Zhong231e97e2016-09-28 15:22:38 -07001263 nodemask_t nmask = node_states[N_MEMORY];
Michal Hocko7f252f22017-07-10 15:48:41 -07001264
1265 /*
1266 * try to allocate from a different node but reuse this node if there
1267 * are no other online nodes to be used (e.g. we are offlining a part
1268 * of the only existing node)
1269 */
1270 node_clear(nid, nmask);
1271 if (nodes_empty(nmask))
1272 node_set(nid, nmask);
Xishi Qiu394e31d2016-07-28 15:48:53 -07001273
Michal Hocko8b913232017-07-10 15:48:47 -07001274 return new_page_nodemask(page, nid, &nmask);
Xishi Qiu394e31d2016-07-28 15:48:53 -07001275}
1276
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001277static int
1278do_migrate_range(unsigned long start_pfn, unsigned long end_pfn)
1279{
1280 unsigned long pfn;
1281 struct page *page;
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001282 int ret = 0;
1283 LIST_HEAD(source);
1284
Michal Hockoa85009c2018-12-28 00:38:29 -08001285 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001286 if (!pfn_valid(pfn))
1287 continue;
1288 page = pfn_to_page(pfn);
Naoya Horiguchic8721bb2013-09-11 14:22:09 -07001289
1290 if (PageHuge(page)) {
1291 struct page *head = compound_head(page);
Matthew Wilcox (Oracle)d8c65462019-09-23 15:34:30 -07001292 pfn = page_to_pfn(head) + compound_nr(head) - 1;
Oscar Salvadordaf35382019-03-05 15:48:53 -08001293 isolate_huge_page(head, &source);
Naoya Horiguchic8721bb2013-09-11 14:22:09 -07001294 continue;
Michal Hocko94723aa2018-04-10 16:30:07 -07001295 } else if (PageTransHuge(page))
Naoya Horiguchi8135d892017-09-08 16:11:15 -07001296 pfn = page_to_pfn(compound_head(page))
1297 + hpage_nr_pages(page) - 1;
Naoya Horiguchic8721bb2013-09-11 14:22:09 -07001298
Michal Hockob15c8722018-12-28 00:38:01 -08001299 /*
1300 * HWPoison pages have elevated reference counts so the migration would
1301 * fail on them. It also doesn't make any sense to migrate them in the
1302 * first place. Still try to unmap such a page in case it is still mapped
1303 * (e.g. current hwpoison implementation doesn't unmap KSM pages but keep
1304 * the unmap as the catch all safety net).
1305 */
1306 if (PageHWPoison(page)) {
1307 if (WARN_ON(PageLRU(page)))
1308 isolate_lru_page(page);
1309 if (page_mapped(page))
1310 try_to_unmap(page, TTU_IGNORE_MLOCK | TTU_IGNORE_ACCESS);
1311 continue;
1312 }
1313
Konstantin Khlebnikov700c2a42011-05-24 17:12:19 -07001314 if (!get_page_unless_zero(page))
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001315 continue;
1316 /*
Yisheng Xie0efadf42017-02-24 14:57:39 -08001317 * We can skip free pages. And we can deal with pages on
1318 * LRU and non-lru movable pages.
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001319 */
Yisheng Xie0efadf42017-02-24 14:57:39 -08001320 if (PageLRU(page))
1321 ret = isolate_lru_page(page);
1322 else
1323 ret = isolate_movable_page(page, ISOLATE_UNEVICTABLE);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001324 if (!ret) { /* Success */
Nick Piggin62695a82008-10-18 20:26:09 -07001325 list_add_tail(&page->lru, &source);
Yisheng Xie0efadf42017-02-24 14:57:39 -08001326 if (!__PageMovable(page))
1327 inc_node_page_state(page, NR_ISOLATED_ANON +
Huang Ying9de4f222020-04-06 20:04:41 -07001328 page_is_file_lru(page));
KOSAKI Motohiro6d9c2852009-12-14 17:58:11 -08001329
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001330 } else {
Michal Hocko2932c8b2018-12-28 00:33:53 -08001331 pr_warn("failed to isolate pfn %lx\n", pfn);
Yisheng Xie0efadf42017-02-24 14:57:39 -08001332 dump_page(page, "isolation failed");
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001333 }
Oscar Salvador17230582019-02-01 14:19:57 -08001334 put_page(page);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001335 }
Bob Liuf3ab2632010-10-26 14:22:10 -07001336 if (!list_empty(&source)) {
Xishi Qiu394e31d2016-07-28 15:48:53 -07001337 /* Allocate a new page from the nearest neighbor node */
1338 ret = migrate_pages(&source, new_node_page, NULL, 0,
Hugh Dickins9c620e22013-02-22 16:35:14 -08001339 MIGRATE_SYNC, MR_MEMORY_HOTPLUG);
Michal Hocko2932c8b2018-12-28 00:33:53 -08001340 if (ret) {
1341 list_for_each_entry(page, &source, lru) {
1342 pr_warn("migrating pfn %lx failed ret:%d ",
1343 page_to_pfn(page), ret);
1344 dump_page(page, "migration failure");
1345 }
Naoya Horiguchic8721bb2013-09-11 14:22:09 -07001346 putback_movable_pages(&source);
Michal Hocko2932c8b2018-12-28 00:33:53 -08001347 }
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001348 }
Oscar Salvador17230582019-02-01 14:19:57 -08001349
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001350 return ret;
1351}
1352
David Hildenbrand0ee5f4f2019-11-30 17:54:03 -08001353/* Mark all sections offline and remove all free pages from the buddy. */
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001354static int
1355offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages,
1356 void *data)
1357{
Michal Hocko5557c762019-05-13 17:21:24 -07001358 unsigned long *offlined_pages = (unsigned long *)data;
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001359
Michal Hocko5557c762019-05-13 17:21:24 -07001360 *offlined_pages += __offline_isolated_pages(start, start + nr_pages);
1361 return 0;
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001362}
1363
1364/*
Ethon Paul52cfc242020-06-04 16:48:58 -07001365 * Check all pages in range, recorded as memory resource, are isolated.
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001366 */
1367static int
1368check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages,
1369 void *data)
1370{
David Hildenbrand756d25b2019-11-30 17:54:07 -08001371 return test_pages_isolated(start_pfn, start_pfn + nr_pages,
1372 MEMORY_OFFLINE);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001373}
1374
Tang Chenc5320922013-11-12 15:08:10 -08001375static int __init cmdline_parse_movable_node(char *p)
1376{
Tang Chen55ac5902014-01-21 15:49:35 -08001377 movable_node_enabled = true;
Tang Chenc5320922013-11-12 15:08:10 -08001378 return 0;
1379}
1380early_param("movable_node", cmdline_parse_movable_node);
1381
Lai Jiangshand9713672012-12-11 16:01:03 -08001382/* check which state of node_states will be changed when offline memory */
1383static void node_states_check_changes_offline(unsigned long nr_pages,
1384 struct zone *zone, struct memory_notify *arg)
1385{
1386 struct pglist_data *pgdat = zone->zone_pgdat;
1387 unsigned long present_pages = 0;
Oscar Salvador86b27be2018-10-26 15:07:38 -07001388 enum zone_type zt;
1389
Anshuman Khandual98fa15f2019-03-05 15:42:58 -08001390 arg->status_change_nid = NUMA_NO_NODE;
1391 arg->status_change_nid_normal = NUMA_NO_NODE;
1392 arg->status_change_nid_high = NUMA_NO_NODE;
Lai Jiangshand9713672012-12-11 16:01:03 -08001393
1394 /*
Oscar Salvador86b27be2018-10-26 15:07:38 -07001395 * Check whether node_states[N_NORMAL_MEMORY] will be changed.
1396 * If the memory to be offline is within the range
1397 * [0..ZONE_NORMAL], and it is the last present memory there,
1398 * the zones in that range will become empty after the offlining,
1399 * thus we can determine that we need to clear the node from
1400 * node_states[N_NORMAL_MEMORY].
Lai Jiangshand9713672012-12-11 16:01:03 -08001401 */
Oscar Salvador86b27be2018-10-26 15:07:38 -07001402 for (zt = 0; zt <= ZONE_NORMAL; zt++)
Lai Jiangshand9713672012-12-11 16:01:03 -08001403 present_pages += pgdat->node_zones[zt].present_pages;
Oscar Salvador86b27be2018-10-26 15:07:38 -07001404 if (zone_idx(zone) <= ZONE_NORMAL && nr_pages >= present_pages)
Lai Jiangshand9713672012-12-11 16:01:03 -08001405 arg->status_change_nid_normal = zone_to_nid(zone);
Lai Jiangshand9713672012-12-11 16:01:03 -08001406
Lai Jiangshan6715ddf2012-12-12 13:51:49 -08001407#ifdef CONFIG_HIGHMEM
1408 /*
Oscar Salvador86b27be2018-10-26 15:07:38 -07001409 * node_states[N_HIGH_MEMORY] contains nodes which
1410 * have normal memory or high memory.
1411 * Here we add the present_pages belonging to ZONE_HIGHMEM.
1412 * If the zone is within the range of [0..ZONE_HIGHMEM), and
1413 * we determine that the zones in that range become empty,
1414 * we need to clear the node for N_HIGH_MEMORY.
Lai Jiangshan6715ddf2012-12-12 13:51:49 -08001415 */
Oscar Salvador86b27be2018-10-26 15:07:38 -07001416 present_pages += pgdat->node_zones[ZONE_HIGHMEM].present_pages;
1417 if (zone_idx(zone) <= ZONE_HIGHMEM && nr_pages >= present_pages)
Lai Jiangshan6715ddf2012-12-12 13:51:49 -08001418 arg->status_change_nid_high = zone_to_nid(zone);
Lai Jiangshan6715ddf2012-12-12 13:51:49 -08001419#endif
1420
Lai Jiangshand9713672012-12-11 16:01:03 -08001421 /*
Oscar Salvador86b27be2018-10-26 15:07:38 -07001422 * We have accounted the pages from [0..ZONE_NORMAL), and
1423 * in case of CONFIG_HIGHMEM the pages from ZONE_HIGHMEM
1424 * as well.
1425 * Here we count the possible pages from ZONE_MOVABLE.
1426 * If after having accounted all the pages, we see that the nr_pages
1427 * to be offlined is over or equal to the accounted pages,
1428 * we know that the node will become empty, and so, we can clear
1429 * it for N_MEMORY as well.
Lai Jiangshand9713672012-12-11 16:01:03 -08001430 */
Oscar Salvador86b27be2018-10-26 15:07:38 -07001431 present_pages += pgdat->node_zones[ZONE_MOVABLE].present_pages;
Lai Jiangshand9713672012-12-11 16:01:03 -08001432
Lai Jiangshand9713672012-12-11 16:01:03 -08001433 if (nr_pages >= present_pages)
1434 arg->status_change_nid = zone_to_nid(zone);
Lai Jiangshand9713672012-12-11 16:01:03 -08001435}
1436
1437static void node_states_clear_node(int node, struct memory_notify *arg)
1438{
1439 if (arg->status_change_nid_normal >= 0)
1440 node_clear_state(node, N_NORMAL_MEMORY);
1441
Oscar Salvadorcf01f6f52018-10-26 15:07:28 -07001442 if (arg->status_change_nid_high >= 0)
Lai Jiangshand9713672012-12-11 16:01:03 -08001443 node_clear_state(node, N_HIGH_MEMORY);
Lai Jiangshan6715ddf2012-12-12 13:51:49 -08001444
Oscar Salvadorcf01f6f52018-10-26 15:07:28 -07001445 if (arg->status_change_nid >= 0)
Lai Jiangshan6715ddf2012-12-12 13:51:49 -08001446 node_clear_state(node, N_MEMORY);
Lai Jiangshand9713672012-12-11 16:01:03 -08001447}
1448
David Hildenbrandc5e79ef2019-11-30 17:54:17 -08001449static int count_system_ram_pages_cb(unsigned long start_pfn,
1450 unsigned long nr_pages, void *data)
1451{
1452 unsigned long *nr_system_ram_pages = data;
1453
1454 *nr_system_ram_pages += nr_pages;
1455 return 0;
1456}
1457
Wen Congyanga16cee12012-10-08 16:33:58 -07001458static int __ref __offline_pages(unsigned long start_pfn,
Michal Hockoecde0f32017-11-15 17:33:38 -08001459 unsigned long end_pfn)
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001460{
David Hildenbrandc5e79ef2019-11-30 17:54:17 -08001461 unsigned long pfn, nr_pages = 0;
Michal Hocko5557c762019-05-13 17:21:24 -07001462 unsigned long offlined_pages = 0;
Qian Cai9b7ea462019-03-28 20:43:34 -07001463 int ret, node, nr_isolate_pageblock;
Cody P Schaferd7029092013-07-03 15:02:11 -07001464 unsigned long flags;
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001465 struct zone *zone;
Yasunori Goto7b78d332007-10-21 16:41:36 -07001466 struct memory_notify arg;
Michal Hocko79605092018-12-28 00:33:49 -08001467 char *reason;
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001468
David Hildenbrand381eab42018-10-30 15:10:29 -07001469 mem_hotplug_begin();
1470
David Hildenbrandc5e79ef2019-11-30 17:54:17 -08001471 /*
1472 * Don't allow to offline memory blocks that contain holes.
1473 * Consequently, memory blocks with holes can never get onlined
1474 * via the hotplug path - online_pages() - as hotplugged memory has
1475 * no holes. This way, we e.g., don't have to worry about marking
1476 * memory holes PG_reserved, don't need pfn_valid() checks, and can
1477 * avoid using walk_system_ram_range() later.
1478 */
1479 walk_system_ram_range(start_pfn, end_pfn - start_pfn, &nr_pages,
1480 count_system_ram_pages_cb);
1481 if (nr_pages != end_pfn - start_pfn) {
1482 ret = -EINVAL;
1483 reason = "memory holes";
1484 goto failed_removal;
1485 }
1486
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001487 /* This makes hotplug much easier...and readable.
1488 we assume this for now. .*/
David Hildenbrand92917992020-02-03 17:34:26 -08001489 zone = test_pages_in_a_zone(start_pfn, end_pfn);
1490 if (!zone) {
Michal Hocko79605092018-12-28 00:33:49 -08001491 ret = -EINVAL;
1492 reason = "multizone range";
1493 goto failed_removal;
David Hildenbrand381eab42018-10-30 15:10:29 -07001494 }
Yasunori Goto7b78d332007-10-21 16:41:36 -07001495 node = zone_to_nid(zone);
Yasunori Goto7b78d332007-10-21 16:41:36 -07001496
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001497 /* set above range as isolated */
Wen Congyangb023f462012-12-11 16:00:45 -08001498 ret = start_isolate_page_range(start_pfn, end_pfn,
Michal Hockod381c542018-12-28 00:33:56 -08001499 MIGRATE_MOVABLE,
David Hildenbrand756d25b2019-11-30 17:54:07 -08001500 MEMORY_OFFLINE | REPORT_FAILURE);
Qian Cai9b7ea462019-03-28 20:43:34 -07001501 if (ret < 0) {
Michal Hocko79605092018-12-28 00:33:49 -08001502 reason = "failure to isolate range";
1503 goto failed_removal;
David Hildenbrand381eab42018-10-30 15:10:29 -07001504 }
Qian Cai9b7ea462019-03-28 20:43:34 -07001505 nr_isolate_pageblock = ret;
Yasunori Goto7b78d332007-10-21 16:41:36 -07001506
1507 arg.start_pfn = start_pfn;
1508 arg.nr_pages = nr_pages;
Lai Jiangshand9713672012-12-11 16:01:03 -08001509 node_states_check_changes_offline(nr_pages, zone, &arg);
Yasunori Goto7b78d332007-10-21 16:41:36 -07001510
1511 ret = memory_notify(MEM_GOING_OFFLINE, &arg);
1512 ret = notifier_to_errno(ret);
Michal Hocko79605092018-12-28 00:33:49 -08001513 if (ret) {
1514 reason = "notifier failure";
1515 goto failed_removal_isolated;
1516 }
Yasunori Goto7b78d332007-10-21 16:41:36 -07001517
Michal Hockobb8965b2018-12-28 00:38:32 -08001518 do {
David Hildenbrandaa218792020-05-07 16:01:30 +02001519 pfn = start_pfn;
1520 do {
Michal Hockobb8965b2018-12-28 00:38:32 -08001521 if (signal_pending(current)) {
1522 ret = -EINTR;
1523 reason = "signal backoff";
1524 goto failed_removal_isolated;
1525 }
Michal Hocko72b39cf2017-11-15 17:33:34 -08001526
Michal Hockobb8965b2018-12-28 00:38:32 -08001527 cond_resched();
1528 lru_add_drain_all();
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001529
David Hildenbrandaa218792020-05-07 16:01:30 +02001530 ret = scan_movable_pages(pfn, end_pfn, &pfn);
1531 if (!ret) {
Michal Hockobb8965b2018-12-28 00:38:32 -08001532 /*
1533 * TODO: fatal migration failures should bail
1534 * out
1535 */
1536 do_migrate_range(pfn, end_pfn);
1537 }
David Hildenbrandaa218792020-05-07 16:01:30 +02001538 } while (!ret);
1539
1540 if (ret != -ENOENT) {
1541 reason = "unmovable page";
1542 goto failed_removal_isolated;
Michal Hockobb8965b2018-12-28 00:38:32 -08001543 }
Michal Hocko72b39cf2017-11-15 17:33:34 -08001544
Michal Hockobb8965b2018-12-28 00:38:32 -08001545 /*
1546 * Dissolve free hugepages in the memory block before doing
1547 * offlining actually in order to make hugetlbfs's object
1548 * counting consistent.
1549 */
1550 ret = dissolve_free_huge_pages(start_pfn, end_pfn);
1551 if (ret) {
1552 reason = "failure to dissolve huge pages";
1553 goto failed_removal_isolated;
1554 }
1555 /* check again */
Michal Hocko5557c762019-05-13 17:21:24 -07001556 ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn,
1557 NULL, check_pages_isolated_cb);
1558 } while (ret);
Michal Hockobb8965b2018-12-28 00:38:32 -08001559
Adam Buchbinderb3834be2012-09-19 21:48:02 -04001560 /* Ok, all of our target is isolated.
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001561 We cannot do rollback at this point. */
Michal Hocko5557c762019-05-13 17:21:24 -07001562 walk_system_ram_range(start_pfn, end_pfn - start_pfn,
1563 &offlined_pages, offline_isolated_pages_cb);
1564 pr_info("Offlined Pages %ld\n", offlined_pages);
Qian Cai9b7ea462019-03-28 20:43:34 -07001565 /*
1566 * Onlining will reset pagetype flags and makes migrate type
1567 * MOVABLE, so just need to decrease the number of isolated
1568 * pageblocks zone counter here.
1569 */
1570 spin_lock_irqsave(&zone->lock, flags);
1571 zone->nr_isolate_pageblock -= nr_isolate_pageblock;
1572 spin_unlock_irqrestore(&zone->lock, flags);
1573
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001574 /* removal success */
Jiang Liu3dcc0572013-07-03 15:03:21 -07001575 adjust_managed_page_count(pfn_to_page(start_pfn), -offlined_pages);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001576 zone->present_pages -= offlined_pages;
Cody P Schaferd7029092013-07-03 15:02:11 -07001577
1578 pgdat_resize_lock(zone->zone_pgdat, &flags);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001579 zone->zone_pgdat->node_present_pages -= offlined_pages;
Cody P Schaferd7029092013-07-03 15:02:11 -07001580 pgdat_resize_unlock(zone->zone_pgdat, &flags);
Yasunori Goto7b78d332007-10-21 16:41:36 -07001581
KOSAKI Motohiro1b79acc2011-05-24 17:11:32 -07001582 init_per_zone_wmark_min();
1583
Xishi Qiu1e8537b2012-10-08 16:31:51 -07001584 if (!populated_zone(zone)) {
Jiang Liu340175b2012-07-31 16:43:32 -07001585 zone_pcp_reset(zone);
Michal Hocko72675e12017-09-06 16:20:24 -07001586 build_all_zonelists(NULL);
Xishi Qiu1e8537b2012-10-08 16:31:51 -07001587 } else
1588 zone_pcp_update(zone);
Jiang Liu340175b2012-07-31 16:43:32 -07001589
Lai Jiangshand9713672012-12-11 16:01:03 -08001590 node_states_clear_node(node, &arg);
Vlastimil Babka698b1b32016-03-17 14:18:08 -07001591 if (arg.status_change_nid >= 0) {
David Rientjes8fe23e02009-12-14 17:58:33 -08001592 kswapd_stop(node);
Vlastimil Babka698b1b32016-03-17 14:18:08 -07001593 kcompactd_stop(node);
1594 }
Minchan Kimbce73942009-06-16 15:32:50 -07001595
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001596 writeback_set_ratelimit();
Yasunori Goto7b78d332007-10-21 16:41:36 -07001597
1598 memory_notify(MEM_OFFLINE, &arg);
David Hildenbrandfeee6b22020-01-04 12:59:33 -08001599 remove_pfn_range_from_zone(zone, start_pfn, nr_pages);
David Hildenbrand381eab42018-10-30 15:10:29 -07001600 mem_hotplug_done();
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001601 return 0;
1602
Michal Hocko79605092018-12-28 00:33:49 -08001603failed_removal_isolated:
1604 undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
Qian Caic4efe482019-03-28 20:44:16 -07001605 memory_notify(MEM_CANCEL_OFFLINE, &arg);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001606failed_removal:
Michal Hocko79605092018-12-28 00:33:49 -08001607 pr_debug("memory offlining [mem %#010llx-%#010llx] failed due to %s\n",
Chen Yuconge33e33b2016-03-17 14:19:35 -07001608 (unsigned long long) start_pfn << PAGE_SHIFT,
Michal Hocko79605092018-12-28 00:33:49 -08001609 ((unsigned long long) end_pfn << PAGE_SHIFT) - 1,
1610 reason);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001611 /* pushback to free area */
David Hildenbrand381eab42018-10-30 15:10:29 -07001612 mem_hotplug_done();
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001613 return ret;
1614}
Badari Pulavarty71088782008-10-18 20:25:58 -07001615
Wen Congyanga16cee12012-10-08 16:33:58 -07001616int offline_pages(unsigned long start_pfn, unsigned long nr_pages)
1617{
Michal Hockoecde0f32017-11-15 17:33:38 -08001618 return __offline_pages(start_pfn, start_pfn + nr_pages);
Wen Congyanga16cee12012-10-08 16:33:58 -07001619}
1620
Xishi Qiud6de9d52013-11-12 15:07:20 -08001621static int check_memblock_offlined_cb(struct memory_block *mem, void *arg)
Wen Congyangbbc76be2013-02-22 16:32:54 -08001622{
1623 int ret = !is_memblock_offlined(mem);
1624
Randy Dunlap349daa02013-04-29 15:08:49 -07001625 if (unlikely(ret)) {
1626 phys_addr_t beginpa, endpa;
1627
1628 beginpa = PFN_PHYS(section_nr_to_pfn(mem->start_section_nr));
David Hildenbrandb6c88d32019-09-23 15:35:49 -07001629 endpa = beginpa + memory_block_size_bytes() - 1;
Joe Perches756a0252016-03-17 14:19:47 -07001630 pr_warn("removing memory fails, because memory [%pa-%pa] is onlined\n",
Randy Dunlap349daa02013-04-29 15:08:49 -07001631 &beginpa, &endpa);
Wen Congyangbbc76be2013-02-22 16:32:54 -08001632
Pavel Tatashineca499a2019-07-16 16:30:31 -07001633 return -EBUSY;
1634 }
1635 return 0;
Wen Congyangbbc76be2013-02-22 16:32:54 -08001636}
1637
Toshi Kani0f1cfe92013-09-11 14:21:50 -07001638static int check_cpu_on_node(pg_data_t *pgdat)
Tang Chen60a5a192013-02-22 16:33:14 -08001639{
Tang Chen60a5a192013-02-22 16:33:14 -08001640 int cpu;
1641
1642 for_each_present_cpu(cpu) {
1643 if (cpu_to_node(cpu) == pgdat->node_id)
1644 /*
1645 * the cpu on this node isn't removed, and we can't
1646 * offline this node.
1647 */
1648 return -EBUSY;
1649 }
1650
1651 return 0;
1652}
1653
David Hildenbrand2c91f8f2019-11-15 17:34:57 -08001654static int check_no_memblock_for_node_cb(struct memory_block *mem, void *arg)
1655{
1656 int nid = *(int *)arg;
1657
1658 /*
1659 * If a memory block belongs to multiple nodes, the stored nid is not
1660 * reliable. However, such blocks are always online (e.g., cannot get
1661 * offlined) and, therefore, are still spanned by the node.
1662 */
1663 return mem->nid == nid ? -EEXIST : 0;
1664}
1665
Toshi Kani0f1cfe92013-09-11 14:21:50 -07001666/**
1667 * try_offline_node
Mike Rapoporte8b098f2018-04-05 16:24:57 -07001668 * @nid: the node ID
Toshi Kani0f1cfe92013-09-11 14:21:50 -07001669 *
1670 * Offline a node if all memory sections and cpus of the node are removed.
1671 *
1672 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1673 * and online/offline operations before this call.
1674 */
Wen Congyang90b30cd2013-02-22 16:33:27 -08001675void try_offline_node(int nid)
Tang Chen60a5a192013-02-22 16:33:14 -08001676{
Wen Congyangd822b862013-02-22 16:33:16 -08001677 pg_data_t *pgdat = NODE_DATA(nid);
David Hildenbrand2c91f8f2019-11-15 17:34:57 -08001678 int rc;
Tang Chen60a5a192013-02-22 16:33:14 -08001679
David Hildenbrand2c91f8f2019-11-15 17:34:57 -08001680 /*
1681 * If the node still spans pages (especially ZONE_DEVICE), don't
1682 * offline it. A node spans memory after move_pfn_range_to_zone(),
1683 * e.g., after the memory block was onlined.
1684 */
1685 if (pgdat->node_spanned_pages)
Tang Chen60a5a192013-02-22 16:33:14 -08001686 return;
David Hildenbrand2c91f8f2019-11-15 17:34:57 -08001687
1688 /*
1689 * Especially offline memory blocks might not be spanned by the
1690 * node. They will get spanned by the node once they get onlined.
1691 * However, they link to the node in sysfs and can get onlined later.
1692 */
1693 rc = for_each_memory_block(&nid, check_no_memblock_for_node_cb);
1694 if (rc)
1695 return;
Tang Chen60a5a192013-02-22 16:33:14 -08001696
Michal Hocko46a36792018-12-28 00:34:13 -08001697 if (check_cpu_on_node(pgdat))
Tang Chen60a5a192013-02-22 16:33:14 -08001698 return;
1699
1700 /*
1701 * all memory/cpu of this node are removed, we can offline this
1702 * node now.
1703 */
1704 node_set_offline(nid);
1705 unregister_one_node(nid);
1706}
Wen Congyang90b30cd2013-02-22 16:33:27 -08001707EXPORT_SYMBOL(try_offline_node);
Tang Chen60a5a192013-02-22 16:33:14 -08001708
David Hildenbrandd9eb1412019-05-13 17:21:33 -07001709static void __release_memory_resource(resource_size_t start,
1710 resource_size_t size)
1711{
1712 int ret;
1713
1714 /*
1715 * When removing memory in the same granularity as it was added,
1716 * this function never fails. It might only fail if resources
1717 * have to be adjusted or split. We'll ignore the error, as
1718 * removing of memory cannot fail.
1719 */
1720 ret = release_mem_region_adjustable(&iomem_resource, start, size);
1721 if (ret) {
1722 resource_size_t endres = start + size - 1;
1723
1724 pr_warn("Unable to release resource <%pa-%pa> (%d)\n",
1725 &start, &endres, ret);
1726 }
1727}
1728
Pavel Tatashineca499a2019-07-16 16:30:31 -07001729static int __ref try_remove_memory(int nid, u64 start, u64 size)
Wen Congyangbbc76be2013-02-22 16:32:54 -08001730{
Pavel Tatashineca499a2019-07-16 16:30:31 -07001731 int rc = 0;
Wen Congyang993c1aa2013-02-22 16:32:50 -08001732
Toshi Kani27356f52013-09-11 14:21:49 -07001733 BUG_ON(check_hotplug_memory_range(start, size));
1734
Yasuaki Ishimatsu6677e3e2013-02-22 16:32:52 -08001735 /*
Rafael J. Wysocki242831e2013-05-27 12:58:46 +02001736 * All memory blocks must be offlined before removing memory. Check
Pavel Tatashineca499a2019-07-16 16:30:31 -07001737 * whether all memory blocks in question are offline and return error
Rafael J. Wysocki242831e2013-05-27 12:58:46 +02001738 * if this is not the case.
Yasuaki Ishimatsu6677e3e2013-02-22 16:32:52 -08001739 */
David Hildenbrandfbcf73c2019-07-18 15:57:46 -07001740 rc = walk_memory_blocks(start, size, NULL, check_memblock_offlined_cb);
Pavel Tatashineca499a2019-07-16 16:30:31 -07001741 if (rc)
1742 goto done;
Yasuaki Ishimatsu6677e3e2013-02-22 16:32:52 -08001743
Yasuaki Ishimatsu46c66c42013-02-22 16:32:56 -08001744 /* remove memmap entry */
1745 firmware_map_remove(start, start + size, "System RAM");
1746
Dan Williamsf1037ec2020-01-30 22:11:17 -08001747 /*
1748 * Memory block device removal under the device_hotplug_lock is
1749 * a barrier against racing online attempts.
1750 */
David Hildenbrand4c4b7f92019-07-18 15:57:06 -07001751 remove_memory_block_devices(start, size);
1752
Dan Williamsf1037ec2020-01-30 22:11:17 -08001753 mem_hotplug_begin();
1754
Oscar Salvador2c2a5af2018-12-28 00:36:22 -08001755 arch_remove_memory(nid, start, size, NULL);
David Hildenbrand52219ae2020-06-04 16:48:38 -07001756
1757 if (IS_ENABLED(CONFIG_ARCH_KEEP_MEMBLOCK)) {
1758 memblock_free(start, size);
1759 memblock_remove(start, size);
1760 }
1761
David Hildenbrandd9eb1412019-05-13 17:21:33 -07001762 __release_memory_resource(start, size);
Wen Congyang24d335c2013-02-22 16:32:58 -08001763
Tang Chen60a5a192013-02-22 16:33:14 -08001764 try_offline_node(nid);
1765
Pavel Tatashineca499a2019-07-16 16:30:31 -07001766done:
Vladimir Davydovbfc8c902014-06-04 16:07:18 -07001767 mem_hotplug_done();
Pavel Tatashineca499a2019-07-16 16:30:31 -07001768 return rc;
Badari Pulavarty71088782008-10-18 20:25:58 -07001769}
David Hildenbrandd15e5922018-10-30 15:10:18 -07001770
Pavel Tatashineca499a2019-07-16 16:30:31 -07001771/**
1772 * remove_memory
1773 * @nid: the node ID
1774 * @start: physical address of the region to remove
1775 * @size: size of the region to remove
1776 *
1777 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1778 * and online/offline operations before this call, as required by
1779 * try_offline_node().
1780 */
1781void __remove_memory(int nid, u64 start, u64 size)
David Hildenbrandd15e5922018-10-30 15:10:18 -07001782{
Pavel Tatashineca499a2019-07-16 16:30:31 -07001783
1784 /*
Souptick Joarder29a90db2019-09-23 15:36:18 -07001785 * trigger BUG() if some memory is not offlined prior to calling this
Pavel Tatashineca499a2019-07-16 16:30:31 -07001786 * function
1787 */
1788 if (try_remove_memory(nid, start, size))
1789 BUG();
1790}
1791
1792/*
1793 * Remove memory if every memory block is offline, otherwise return -EBUSY is
1794 * some memory is not offline
1795 */
1796int remove_memory(int nid, u64 start, u64 size)
1797{
1798 int rc;
1799
David Hildenbrandd15e5922018-10-30 15:10:18 -07001800 lock_device_hotplug();
Pavel Tatashineca499a2019-07-16 16:30:31 -07001801 rc = try_remove_memory(nid, start, size);
David Hildenbrandd15e5922018-10-30 15:10:18 -07001802 unlock_device_hotplug();
Pavel Tatashineca499a2019-07-16 16:30:31 -07001803
1804 return rc;
David Hildenbrandd15e5922018-10-30 15:10:18 -07001805}
Badari Pulavarty71088782008-10-18 20:25:58 -07001806EXPORT_SYMBOL_GPL(remove_memory);
David Hildenbrand08b3acd2020-05-07 16:01:32 +02001807
1808/*
1809 * Try to offline and remove a memory block. Might take a long time to
1810 * finish in case memory is still in use. Primarily useful for memory devices
1811 * that logically unplugged all memory (so it's no longer in use) and want to
1812 * offline + remove the memory block.
1813 */
1814int offline_and_remove_memory(int nid, u64 start, u64 size)
1815{
1816 struct memory_block *mem;
1817 int rc = -EINVAL;
1818
1819 if (!IS_ALIGNED(start, memory_block_size_bytes()) ||
1820 size != memory_block_size_bytes())
1821 return rc;
1822
1823 lock_device_hotplug();
1824 mem = find_memory_block(__pfn_to_section(PFN_DOWN(start)));
1825 if (mem)
1826 rc = device_offline(&mem->dev);
1827 /* Ignore if the device is already offline. */
1828 if (rc > 0)
1829 rc = 0;
1830
1831 /*
1832 * In case we succeeded to offline the memory block, remove it.
1833 * This cannot fail as it cannot get onlined in the meantime.
1834 */
1835 if (!rc) {
1836 rc = try_remove_memory(nid, start, size);
1837 WARN_ON_ONCE(rc);
1838 }
1839 unlock_device_hotplug();
1840
1841 return rc;
1842}
1843EXPORT_SYMBOL_GPL(offline_and_remove_memory);
Rafael J. Wysockiaba6efc2013-06-01 22:24:07 +02001844#endif /* CONFIG_MEMORY_HOTREMOVE */