blob: d397af38f9cefee15141e191ceffe3871eeb9d05 [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"))
David Hildenbrand7cf603d2020-10-15 20:08:33 -0700108 flags |= IORESOURCE_SYSRAM_DRIVER_MANAGED;
Juergen Gross357b4da2019-02-14 11:42:39 +0100109
Baoquan Hef3cd4c862020-04-06 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
Jia Hed622ece2020-08-11 18:32:16 -0700353#ifdef CONFIG_NUMA
354int __weak memory_add_physaddr_to_nid(u64 start)
355{
Dan Williamsa035b6b2020-10-13 16:49:23 -0700356 pr_info_once("Unknown online node for memory at 0x%llx, assuming node 0\n",
Jia Hed622ece2020-08-11 18:32:16 -0700357 start);
358 return 0;
359}
360EXPORT_SYMBOL_GPL(memory_add_physaddr_to_nid);
Dan Williamsa035b6b2020-10-13 16:49:23 -0700361
362int __weak phys_to_target_node(u64 start)
363{
364 pr_info_once("Unknown target node for memory at 0x%llx, assuming node 0\n",
365 start);
366 return 0;
367}
368EXPORT_SYMBOL_GPL(phys_to_target_node);
Jia Hed622ece2020-08-11 18:32:16 -0700369#endif
370
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800371/* find the smallest valid pfn in the range [start_pfn, end_pfn) */
YASUAKI ISHIMATSUd09b0132017-10-03 16:16:32 -0700372static unsigned long find_smallest_section_pfn(int nid, struct zone *zone,
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800373 unsigned long start_pfn,
374 unsigned long end_pfn)
375{
Dan Williams49ba3c62019-07-18 15:58:07 -0700376 for (; start_pfn < end_pfn; start_pfn += PAGES_PER_SUBSECTION) {
David Hildenbrand7ce700b2019-11-21 17:53:56 -0800377 if (unlikely(!pfn_to_online_page(start_pfn)))
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800378 continue;
379
380 if (unlikely(pfn_to_nid(start_pfn) != nid))
381 continue;
382
David Hildenbrand9b051582020-02-03 17:34:12 -0800383 if (zone != page_zone(pfn_to_page(start_pfn)))
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800384 continue;
385
386 return start_pfn;
387 }
388
389 return 0;
390}
391
392/* find the biggest valid pfn in the range [start_pfn, end_pfn). */
YASUAKI ISHIMATSUd09b0132017-10-03 16:16:32 -0700393static unsigned long find_biggest_section_pfn(int nid, struct zone *zone,
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800394 unsigned long start_pfn,
395 unsigned long end_pfn)
396{
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800397 unsigned long pfn;
398
399 /* pfn is the end pfn of a memory section. */
400 pfn = end_pfn - 1;
Dan Williams49ba3c62019-07-18 15:58:07 -0700401 for (; pfn >= start_pfn; pfn -= PAGES_PER_SUBSECTION) {
David Hildenbrand7ce700b2019-11-21 17:53:56 -0800402 if (unlikely(!pfn_to_online_page(pfn)))
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800403 continue;
404
405 if (unlikely(pfn_to_nid(pfn) != nid))
406 continue;
407
David Hildenbrand9b051582020-02-03 17:34:12 -0800408 if (zone != page_zone(pfn_to_page(pfn)))
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800409 continue;
410
411 return pfn;
412 }
413
414 return 0;
415}
416
417static void shrink_zone_span(struct zone *zone, unsigned long start_pfn,
418 unsigned long end_pfn)
419{
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800420 unsigned long pfn;
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800421 int nid = zone_to_nid(zone);
422
423 zone_span_writelock(zone);
David Hildenbrand5d120712020-02-03 17:34:19 -0800424 if (zone->zone_start_pfn == start_pfn) {
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800425 /*
426 * If the section is smallest section in the zone, it need
427 * shrink zone->zone_start_pfn and zone->zone_spanned_pages.
428 * In this case, we find second smallest valid mem_section
429 * for shrinking zone.
430 */
431 pfn = find_smallest_section_pfn(nid, zone, end_pfn,
David Hildenbrand5d120712020-02-03 17:34:19 -0800432 zone_end_pfn(zone));
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800433 if (pfn) {
David Hildenbrand5d120712020-02-03 17:34:19 -0800434 zone->spanned_pages = zone_end_pfn(zone) - pfn;
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800435 zone->zone_start_pfn = pfn;
David Hildenbrand950b68d2020-02-03 17:34:16 -0800436 } else {
437 zone->zone_start_pfn = 0;
438 zone->spanned_pages = 0;
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800439 }
David Hildenbrand5d120712020-02-03 17:34:19 -0800440 } else if (zone_end_pfn(zone) == end_pfn) {
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800441 /*
442 * If the section is biggest section in the zone, it need
443 * shrink zone->spanned_pages.
444 * In this case, we find second biggest valid mem_section for
445 * shrinking zone.
446 */
David Hildenbrand5d120712020-02-03 17:34:19 -0800447 pfn = find_biggest_section_pfn(nid, zone, zone->zone_start_pfn,
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800448 start_pfn);
449 if (pfn)
David Hildenbrand5d120712020-02-03 17:34:19 -0800450 zone->spanned_pages = pfn - zone->zone_start_pfn + 1;
David Hildenbrand950b68d2020-02-03 17:34:16 -0800451 else {
452 zone->zone_start_pfn = 0;
453 zone->spanned_pages = 0;
454 }
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800455 }
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800456 zone_span_writeunlock(zone);
457}
458
David Hildenbrand00d6c012019-10-18 20:19:33 -0700459static void update_pgdat_span(struct pglist_data *pgdat)
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800460{
David Hildenbrand00d6c012019-10-18 20:19:33 -0700461 unsigned long node_start_pfn = 0, node_end_pfn = 0;
462 struct zone *zone;
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800463
David Hildenbrand00d6c012019-10-18 20:19:33 -0700464 for (zone = pgdat->node_zones;
465 zone < pgdat->node_zones + MAX_NR_ZONES; zone++) {
466 unsigned long zone_end_pfn = zone->zone_start_pfn +
467 zone->spanned_pages;
468
469 /* No need to lock the zones, they can't change. */
David Hildenbrand656d5712019-11-05 21:17:10 -0800470 if (!zone->spanned_pages)
471 continue;
472 if (!node_end_pfn) {
473 node_start_pfn = zone->zone_start_pfn;
474 node_end_pfn = zone_end_pfn;
475 continue;
476 }
477
David Hildenbrand00d6c012019-10-18 20:19:33 -0700478 if (zone_end_pfn > node_end_pfn)
479 node_end_pfn = zone_end_pfn;
480 if (zone->zone_start_pfn < node_start_pfn)
481 node_start_pfn = zone->zone_start_pfn;
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800482 }
483
David Hildenbrand00d6c012019-10-18 20:19:33 -0700484 pgdat->node_start_pfn = node_start_pfn;
485 pgdat->node_spanned_pages = node_end_pfn - node_start_pfn;
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800486}
487
David Hildenbrandfeee6b22020-01-04 12:59:33 -0800488void __ref remove_pfn_range_from_zone(struct zone *zone,
489 unsigned long start_pfn,
490 unsigned long nr_pages)
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800491{
Ben Widawskyb7e3deb2020-06-25 20:30:51 -0700492 const unsigned long end_pfn = start_pfn + nr_pages;
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800493 struct pglist_data *pgdat = zone->zone_pgdat;
Ben Widawskyb7e3deb2020-06-25 20:30:51 -0700494 unsigned long pfn, cur_nr_pages, flags;
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800495
David Hildenbrandd33695b2020-02-03 17:34:09 -0800496 /* Poison struct pages because they are now uninitialized again. */
Ben Widawskyb7e3deb2020-06-25 20:30:51 -0700497 for (pfn = start_pfn; pfn < end_pfn; pfn += cur_nr_pages) {
498 cond_resched();
499
500 /* Select all remaining pages up to the next section boundary */
501 cur_nr_pages =
502 min(end_pfn - pfn, SECTION_ALIGN_UP(pfn + 1) - pfn);
503 page_init_poison(pfn_to_page(pfn),
504 sizeof(struct page) * cur_nr_pages);
505 }
David Hildenbrandd33695b2020-02-03 17:34:09 -0800506
David Hildenbrand7ce700b2019-11-21 17:53:56 -0800507#ifdef CONFIG_ZONE_DEVICE
508 /*
509 * Zone shrinking code cannot properly deal with ZONE_DEVICE. So
510 * we will not try to shrink the zones - which is okay as
511 * set_zone_contiguous() cannot deal with ZONE_DEVICE either way.
512 */
513 if (zone_idx(zone) == ZONE_DEVICE)
514 return;
515#endif
516
David Hildenbrandfeee6b22020-01-04 12:59:33 -0800517 clear_zone_contiguous(zone);
518
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800519 pgdat_resize_lock(zone->zone_pgdat, &flags);
520 shrink_zone_span(zone, start_pfn, start_pfn + nr_pages);
David Hildenbrand00d6c012019-10-18 20:19:33 -0700521 update_pgdat_span(pgdat);
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800522 pgdat_resize_unlock(zone->zone_pgdat, &flags);
David Hildenbrandfeee6b22020-01-04 12:59:33 -0800523
524 set_zone_contiguous(zone);
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800525}
526
David Hildenbrandfeee6b22020-01-04 12:59:33 -0800527static void __remove_section(unsigned long pfn, unsigned long nr_pages,
528 unsigned long map_offset,
529 struct vmem_altmap *altmap)
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700530{
chenqiwu10404902020-04-06 20:07:48 -0700531 struct mem_section *ms = __pfn_to_section(pfn);
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700532
David Hildenbrand9d1d8872019-05-13 17:21:41 -0700533 if (WARN_ON_ONCE(!valid_section(ms)))
534 return;
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700535
Dan Williamsba72b4c2019-07-18 15:58:26 -0700536 sparse_remove_section(ms, pfn, nr_pages, map_offset, altmap);
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700537}
538
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700539/**
David Hildenbrandfeee6b22020-01-04 12:59:33 -0800540 * __remove_pages() - remove sections of pages
Dan Williams7ea62162019-07-18 15:58:22 -0700541 * @pfn: starting pageframe (must be aligned to start of a section)
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700542 * @nr_pages: number of pages to remove (must be multiple of section size)
Mike Rapoporte8b098f2018-04-05 16:24:57 -0700543 * @altmap: alternative device page map or %NULL if default memmap is used
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700544 *
545 * Generic helper function to remove section mappings and sysfs entries
546 * for the section of the memory we are removing. Caller needs to make
547 * sure that pages are marked reserved and zones are adjust properly by
548 * calling offline_pages().
549 */
David Hildenbrandfeee6b22020-01-04 12:59:33 -0800550void __remove_pages(unsigned long pfn, unsigned long nr_pages,
551 struct vmem_altmap *altmap)
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700552{
David Hildenbrand52fb87c2020-02-03 17:34:23 -0800553 const unsigned long end_pfn = pfn + nr_pages;
554 unsigned long cur_nr_pages;
Dan Williams4b94ffd2016-01-15 16:56:22 -0800555 unsigned long map_offset = 0;
Dan Williams4b94ffd2016-01-15 16:56:22 -0800556
Dan Williams96da4352019-07-18 15:58:15 -0700557 map_offset = vmem_altmap_offset(altmap);
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700558
Dan Williams7ea62162019-07-18 15:58:22 -0700559 if (check_pfn_span(pfn, nr_pages, "remove"))
560 return;
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700561
David Hildenbrand52fb87c2020-02-03 17:34:23 -0800562 for (; pfn < end_pfn; pfn += cur_nr_pages) {
Michal Hockodd33ad72018-11-02 15:48:46 -0700563 cond_resched();
David Hildenbrand52fb87c2020-02-03 17:34:23 -0800564 /* Select all remaining pages up to the next section boundary */
David Hildenbranda11b9412020-04-06 20:06:53 -0700565 cur_nr_pages = min(end_pfn - pfn,
566 SECTION_ALIGN_UP(pfn + 1) - pfn);
David Hildenbrand52fb87c2020-02-03 17:34:23 -0800567 __remove_section(pfn, cur_nr_pages, map_offset, altmap);
Dan Williams4b94ffd2016-01-15 16:56:22 -0800568 map_offset = 0;
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700569 }
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700570}
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700571
Daniel Kiper9d0ad8c2011-07-25 17:12:05 -0700572int set_online_page_callback(online_page_callback_t callback)
573{
574 int rc = -EINVAL;
575
Vladimir Davydovbfc8c902014-06-04 16:07:18 -0700576 get_online_mems();
577 mutex_lock(&online_page_callback_lock);
Daniel Kiper9d0ad8c2011-07-25 17:12:05 -0700578
579 if (online_page_callback == generic_online_page) {
580 online_page_callback = callback;
581 rc = 0;
582 }
583
Vladimir Davydovbfc8c902014-06-04 16:07:18 -0700584 mutex_unlock(&online_page_callback_lock);
585 put_online_mems();
Daniel Kiper9d0ad8c2011-07-25 17:12:05 -0700586
587 return rc;
588}
589EXPORT_SYMBOL_GPL(set_online_page_callback);
590
591int restore_online_page_callback(online_page_callback_t callback)
592{
593 int rc = -EINVAL;
594
Vladimir Davydovbfc8c902014-06-04 16:07:18 -0700595 get_online_mems();
596 mutex_lock(&online_page_callback_lock);
Daniel Kiper9d0ad8c2011-07-25 17:12:05 -0700597
598 if (online_page_callback == callback) {
599 online_page_callback = generic_online_page;
600 rc = 0;
601 }
602
Vladimir Davydovbfc8c902014-06-04 16:07:18 -0700603 mutex_unlock(&online_page_callback_lock);
604 put_online_mems();
Daniel Kiper9d0ad8c2011-07-25 17:12:05 -0700605
606 return rc;
607}
608EXPORT_SYMBOL_GPL(restore_online_page_callback);
609
David Hildenbrand18db1492019-11-30 17:53:51 -0800610void generic_online_page(struct page *page, unsigned int order)
Daniel Kiper9d0ad8c2011-07-25 17:12:05 -0700611{
Vlastimil Babkac87cbc12020-03-05 22:28:42 -0800612 /*
613 * Freeing the page with debug_pagealloc enabled will try to unmap it,
614 * so we should map it first. This is better than introducing a special
615 * case in page freeing fast path.
616 */
617 if (debug_pagealloc_enabled_static())
618 kernel_map_pages(page, 1 << order, 1);
Arun KSa9cd4102019-03-05 15:42:14 -0800619 __free_pages_core(page, order);
620 totalram_pages_add(1UL << order);
621#ifdef CONFIG_HIGHMEM
622 if (PageHighMem(page))
623 totalhigh_pages_add(1UL << order);
624#endif
625}
David Hildenbrand18db1492019-11-30 17:53:51 -0800626EXPORT_SYMBOL_GPL(generic_online_page);
Arun KSa9cd4102019-03-05 15:42:14 -0800627
David Hildenbrandaac65322020-10-15 20:08:11 -0700628static void online_pages_range(unsigned long start_pfn, unsigned long nr_pages)
Dave Hansen3947be12005-10-29 18:16:54 -0700629{
David Hildenbrandb2c2ab22019-09-23 15:36:02 -0700630 const unsigned long end_pfn = start_pfn + nr_pages;
631 unsigned long pfn;
Michal Hocko2d070ea2017-07-06 15:37:56 -0700632
David Hildenbrandb2c2ab22019-09-23 15:36:02 -0700633 /*
David Hildenbrandaac65322020-10-15 20:08:11 -0700634 * Online the pages in MAX_ORDER - 1 aligned chunks. The callback might
635 * decide to not expose all pages to the buddy (e.g., expose them
636 * later). We account all pages as being online and belonging to this
637 * zone ("present").
David Hildenbrandb2c2ab22019-09-23 15:36:02 -0700638 */
David Hildenbrandaac65322020-10-15 20:08:11 -0700639 for (pfn = start_pfn; pfn < end_pfn; pfn += MAX_ORDER_NR_PAGES)
640 (*online_page_callback)(pfn_to_page(pfn), MAX_ORDER - 1);
Michal Hocko2d070ea2017-07-06 15:37:56 -0700641
David Hildenbrandb2c2ab22019-09-23 15:36:02 -0700642 /* mark all involved sections as online */
643 online_mem_sections(start_pfn, end_pfn);
KAMEZAWA Hiroyuki75884fb2007-10-16 01:26:10 -0700644}
645
Lai Jiangshand9713672012-12-11 16:01:03 -0800646/* check which state of node_states will be changed when online memory */
647static void node_states_check_changes_online(unsigned long nr_pages,
648 struct zone *zone, struct memory_notify *arg)
649{
650 int nid = zone_to_nid(zone);
Lai Jiangshand9713672012-12-11 16:01:03 -0800651
Anshuman Khandual98fa15f2019-03-05 15:42:58 -0800652 arg->status_change_nid = NUMA_NO_NODE;
653 arg->status_change_nid_normal = NUMA_NO_NODE;
654 arg->status_change_nid_high = NUMA_NO_NODE;
Lai Jiangshand9713672012-12-11 16:01:03 -0800655
Lai Jiangshan6715ddf2012-12-12 13:51:49 -0800656 if (!node_state(nid, N_MEMORY))
Lai Jiangshand9713672012-12-11 16:01:03 -0800657 arg->status_change_nid = nid;
Oscar Salvador8efe33f2018-10-26 15:07:34 -0700658 if (zone_idx(zone) <= ZONE_NORMAL && !node_state(nid, N_NORMAL_MEMORY))
659 arg->status_change_nid_normal = nid;
660#ifdef CONFIG_HIGHMEM
Baoquan Hed3ba3ae2019-05-13 17:17:35 -0700661 if (zone_idx(zone) <= ZONE_HIGHMEM && !node_state(nid, N_HIGH_MEMORY))
Oscar Salvador8efe33f2018-10-26 15:07:34 -0700662 arg->status_change_nid_high = nid;
663#endif
Lai Jiangshand9713672012-12-11 16:01:03 -0800664}
665
666static void node_states_set_node(int node, struct memory_notify *arg)
667{
668 if (arg->status_change_nid_normal >= 0)
669 node_set_state(node, N_NORMAL_MEMORY);
670
Lai Jiangshan6715ddf2012-12-12 13:51:49 -0800671 if (arg->status_change_nid_high >= 0)
672 node_set_state(node, N_HIGH_MEMORY);
673
Oscar Salvador83d83612018-10-26 15:07:25 -0700674 if (arg->status_change_nid >= 0)
675 node_set_state(node, N_MEMORY);
Lai Jiangshand9713672012-12-11 16:01:03 -0800676}
677
Michal Hockof1dd2cd2017-07-06 15:38:11 -0700678static void __meminit resize_zone_range(struct zone *zone, unsigned long start_pfn,
679 unsigned long nr_pages)
680{
681 unsigned long old_end_pfn = zone_end_pfn(zone);
682
683 if (zone_is_empty(zone) || start_pfn < zone->zone_start_pfn)
684 zone->zone_start_pfn = start_pfn;
685
686 zone->spanned_pages = max(start_pfn + nr_pages, old_end_pfn) - zone->zone_start_pfn;
687}
688
689static void __meminit resize_pgdat_range(struct pglist_data *pgdat, unsigned long start_pfn,
690 unsigned long nr_pages)
691{
692 unsigned long old_end_pfn = pgdat_end_pfn(pgdat);
693
694 if (!pgdat->node_spanned_pages || start_pfn < pgdat->node_start_pfn)
695 pgdat->node_start_pfn = start_pfn;
696
697 pgdat->node_spanned_pages = max(start_pfn + nr_pages, old_end_pfn) - pgdat->node_start_pfn;
Michal Hockof1dd2cd2017-07-06 15:38:11 -0700698
David Hildenbrand3fccb742019-09-23 15:35:37 -0700699}
700/*
701 * Associate the pfn range with the given zone, initializing the memmaps
702 * and resizing the pgdat/zone data to span the added pages. After this
703 * call, all affected pages are PG_reserved.
David Hildenbrandd882c002020-10-15 20:08:19 -0700704 *
705 * All aligned pageblocks are initialized to the specified migratetype
706 * (usually MIGRATE_MOVABLE). Besides setting the migratetype, no related
707 * zone stats (e.g., nr_isolate_pageblock) are touched.
David Hildenbrand3fccb742019-09-23 15:35:37 -0700708 */
Christoph Hellwiga99583e2017-12-29 08:53:57 +0100709void __ref move_pfn_range_to_zone(struct zone *zone, unsigned long start_pfn,
David Hildenbrandd882c002020-10-15 20:08:19 -0700710 unsigned long nr_pages,
711 struct vmem_altmap *altmap, int migratetype)
Michal Hockof1dd2cd2017-07-06 15:38:11 -0700712{
713 struct pglist_data *pgdat = zone->zone_pgdat;
714 int nid = pgdat->node_id;
715 unsigned long flags;
716
Michal Hockof1dd2cd2017-07-06 15:38:11 -0700717 clear_zone_contiguous(zone);
718
719 /* TODO Huh pgdat is irqsave while zone is not. It used to be like that before */
720 pgdat_resize_lock(pgdat, &flags);
721 zone_span_writelock(zone);
Wei Yangfa004ab2018-12-28 00:37:10 -0800722 if (zone_is_empty(zone))
723 init_currently_empty_zone(zone, start_pfn, nr_pages);
Michal Hockof1dd2cd2017-07-06 15:38:11 -0700724 resize_zone_range(zone, start_pfn, nr_pages);
725 zone_span_writeunlock(zone);
726 resize_pgdat_range(pgdat, start_pfn, nr_pages);
727 pgdat_resize_unlock(pgdat, &flags);
728
729 /*
730 * TODO now we have a visible range of pages which are not associated
731 * with their zone properly. Not nice but set_pfnblock_flags_mask
732 * expects the zone spans the pfn range. All the pages in the range
733 * are reserved so nobody should be touching them so we should be safe
734 */
Christoph Hellwiga99583e2017-12-29 08:53:57 +0100735 memmap_init_zone(nr_pages, nid, zone_idx(zone), start_pfn,
David Hildenbrandd882c002020-10-15 20:08:19 -0700736 MEMINIT_HOTPLUG, altmap, migratetype);
Michal Hockof1dd2cd2017-07-06 15:38:11 -0700737
738 set_zone_contiguous(zone);
739}
740
741/*
Michal Hockoc246a212017-07-06 15:38:18 -0700742 * Returns a default kernel memory zone for the given pfn range.
743 * If no kernel zone covers this pfn range it will automatically go
744 * to the ZONE_NORMAL.
745 */
Michal Hockoc6f03e22017-09-06 16:19:40 -0700746static struct zone *default_kernel_zone_for_pfn(int nid, unsigned long start_pfn,
Michal Hockoc246a212017-07-06 15:38:18 -0700747 unsigned long nr_pages)
748{
749 struct pglist_data *pgdat = NODE_DATA(nid);
750 int zid;
751
752 for (zid = 0; zid <= ZONE_NORMAL; zid++) {
753 struct zone *zone = &pgdat->node_zones[zid];
754
755 if (zone_intersects(zone, start_pfn, nr_pages))
756 return zone;
757 }
758
759 return &pgdat->node_zones[ZONE_NORMAL];
760}
761
Michal Hockoc6f03e22017-09-06 16:19:40 -0700762static inline struct zone *default_zone_for_pfn(int nid, unsigned long start_pfn,
763 unsigned long nr_pages)
Michal Hockoe5e68932017-09-06 16:19:37 -0700764{
Michal Hockoc6f03e22017-09-06 16:19:40 -0700765 struct zone *kernel_zone = default_kernel_zone_for_pfn(nid, start_pfn,
766 nr_pages);
767 struct zone *movable_zone = &NODE_DATA(nid)->node_zones[ZONE_MOVABLE];
768 bool in_kernel = zone_intersects(kernel_zone, start_pfn, nr_pages);
769 bool in_movable = zone_intersects(movable_zone, start_pfn, nr_pages);
Michal Hockoe5e68932017-09-06 16:19:37 -0700770
771 /*
Michal Hockoc6f03e22017-09-06 16:19:40 -0700772 * We inherit the existing zone in a simple case where zones do not
773 * overlap in the given range
Michal Hockoe5e68932017-09-06 16:19:37 -0700774 */
Michal Hockoc6f03e22017-09-06 16:19:40 -0700775 if (in_kernel ^ in_movable)
776 return (in_kernel) ? kernel_zone : movable_zone;
Michal Hockoe5e68932017-09-06 16:19:37 -0700777
Michal Hockoc6f03e22017-09-06 16:19:40 -0700778 /*
779 * If the range doesn't belong to any zone or two zones overlap in the
780 * given range then we use movable zone only if movable_node is
781 * enabled because we always online to a kernel zone by default.
782 */
783 return movable_node_enabled ? movable_zone : kernel_zone;
Michal Hocko9f123ab2017-07-10 15:48:37 -0700784}
785
Michal Hockoe5e68932017-09-06 16:19:37 -0700786struct zone * zone_for_pfn_range(int online_type, int nid, unsigned start_pfn,
787 unsigned long nr_pages)
Michal Hockof1dd2cd2017-07-06 15:38:11 -0700788{
Michal Hockoc6f03e22017-09-06 16:19:40 -0700789 if (online_type == MMOP_ONLINE_KERNEL)
790 return default_kernel_zone_for_pfn(nid, start_pfn, nr_pages);
Michal Hockof1dd2cd2017-07-06 15:38:11 -0700791
Michal Hockoc6f03e22017-09-06 16:19:40 -0700792 if (online_type == MMOP_ONLINE_MOVABLE)
793 return &NODE_DATA(nid)->node_zones[ZONE_MOVABLE];
Reza Arbabdf429ac2016-07-26 15:22:23 -0700794
Michal Hockoc6f03e22017-09-06 16:19:40 -0700795 return default_zone_for_pfn(nid, start_pfn, nr_pages);
Michal Hockoe5e68932017-09-06 16:19:37 -0700796}
797
David Hildenbrandbd5c2342020-01-30 22:14:54 -0800798int __ref online_pages(unsigned long pfn, unsigned long nr_pages,
799 int online_type, int nid)
KAMEZAWA Hiroyuki75884fb2007-10-16 01:26:10 -0700800{
Cody P Schaferaa472282013-07-03 15:02:10 -0700801 unsigned long flags;
Dave Hansen3947be12005-10-29 18:16:54 -0700802 struct zone *zone;
Yasunori Goto68113782006-06-23 02:03:11 -0700803 int need_zonelists_rebuild = 0;
Yasunori Goto7b78d332007-10-21 16:41:36 -0700804 int ret;
805 struct memory_notify arg;
Dave Hansen3947be12005-10-29 18:16:54 -0700806
David Hildenbrand4986fac2020-10-15 20:07:50 -0700807 /* We can only online full sections (e.g., SECTION_IS_ONLINE) */
808 if (WARN_ON_ONCE(!nr_pages ||
809 !IS_ALIGNED(pfn | nr_pages, PAGES_PER_SECTION)))
810 return -EINVAL;
811
David Hildenbrand381eab42018-10-30 15:10:29 -0700812 mem_hotplug_begin();
813
Michal Hockof1dd2cd2017-07-06 15:38:11 -0700814 /* associate pfn range with the zone */
David Hildenbrand3fccb742019-09-23 15:35:37 -0700815 zone = zone_for_pfn_range(online_type, nid, pfn, nr_pages);
David Hildenbrandb30c5922020-10-15 20:08:23 -0700816 move_pfn_range_to_zone(zone, pfn, nr_pages, NULL, MIGRATE_ISOLATE);
Michal Hockof1dd2cd2017-07-06 15:38:11 -0700817
Yasunori Goto7b78d332007-10-21 16:41:36 -0700818 arg.start_pfn = pfn;
819 arg.nr_pages = nr_pages;
Lai Jiangshand9713672012-12-11 16:01:03 -0800820 node_states_check_changes_online(nr_pages, zone, &arg);
Yasunori Goto7b78d332007-10-21 16:41:36 -0700821
Yasunori Goto7b78d332007-10-21 16:41:36 -0700822 ret = memory_notify(MEM_GOING_ONLINE, &arg);
823 ret = notifier_to_errno(ret);
Chen Yuconge33e33b2016-03-17 14:19:35 -0700824 if (ret)
825 goto failed_addition;
826
Dave Hansen3947be12005-10-29 18:16:54 -0700827 /*
David Hildenbrandb30c5922020-10-15 20:08:23 -0700828 * Fixup the number of isolated pageblocks before marking the sections
829 * onlining, such that undo_isolate_page_range() works correctly.
830 */
831 spin_lock_irqsave(&zone->lock, flags);
832 zone->nr_isolate_pageblock += nr_pages / pageblock_nr_pages;
833 spin_unlock_irqrestore(&zone->lock, flags);
834
835 /*
Yasunori Goto68113782006-06-23 02:03:11 -0700836 * If this zone is not populated, then it is not in zonelist.
837 * This means the page allocator ignores this zone.
838 * So, zonelist must be updated after online.
839 */
Wen Congyang6dcd73d2012-12-11 16:01:01 -0800840 if (!populated_zone(zone)) {
Yasunori Goto68113782006-06-23 02:03:11 -0700841 need_zonelists_rebuild = 1;
Michal Hocko72675e12017-09-06 16:20:24 -0700842 setup_zone_pageset(zone);
Wen Congyang6dcd73d2012-12-11 16:01:01 -0800843 }
Yasunori Goto68113782006-06-23 02:03:11 -0700844
David Hildenbrandaac65322020-10-15 20:08:11 -0700845 online_pages_range(pfn, nr_pages);
846 zone->present_pages += nr_pages;
Cody P Schaferaa472282013-07-03 15:02:10 -0700847
848 pgdat_resize_lock(zone->zone_pgdat, &flags);
David Hildenbrandaac65322020-10-15 20:08:11 -0700849 zone->zone_pgdat->node_present_pages += nr_pages;
Cody P Schaferaa472282013-07-03 15:02:10 -0700850 pgdat_resize_unlock(zone->zone_pgdat, &flags);
851
David Hildenbrandb30c5922020-10-15 20:08:23 -0700852 node_states_set_node(nid, &arg);
853 if (need_zonelists_rebuild)
854 build_all_zonelists(NULL);
855 zone_pcp_update(zone);
856
857 /* Basic onlining is complete, allow allocation of onlined pages. */
858 undo_isolate_page_range(pfn, pfn + nr_pages, MIGRATE_MOVABLE);
859
David Hildenbrand93146d92020-08-06 23:25:35 -0700860 /*
861 * When exposing larger, physically contiguous memory areas to the
862 * buddy, shuffling in the buddy (when freeing onlined pages, putting
863 * them either to the head or the tail of the freelist) is only helpful
864 * for maintaining the shuffle, but not for creating the initial
865 * shuffle. Shuffle the whole zone to make sure the just onlined pages
David Hildenbrandb30c5922020-10-15 20:08:23 -0700866 * are properly distributed across the whole freelist. Make sure to
867 * shuffle once pageblocks are no longer isolated.
David Hildenbrand93146d92020-08-06 23:25:35 -0700868 */
Dan Williamse900a912019-05-14 15:41:28 -0700869 shuffle_zone(zone);
870
KOSAKI Motohiro1b79acc2011-05-24 17:11:32 -0700871 init_per_zone_wmark_min();
872
David Hildenbrandca9a46f2019-09-23 15:36:08 -0700873 kswapd_run(nid);
874 kcompactd_run(nid);
Dave Hansen61b13992005-10-29 18:16:56 -0700875
Chandra Seetharaman2d1d43f2006-09-29 02:01:25 -0700876 writeback_set_ratelimit();
Yasunori Goto7b78d332007-10-21 16:41:36 -0700877
David Hildenbrandca9a46f2019-09-23 15:36:08 -0700878 memory_notify(MEM_ONLINE, &arg);
David Hildenbrand381eab42018-10-30 15:10:29 -0700879 mem_hotplug_done();
David Rientjes30467e02015-04-14 15:45:11 -0700880 return 0;
Chen Yuconge33e33b2016-03-17 14:19:35 -0700881
882failed_addition:
883 pr_debug("online_pages [mem %#010llx-%#010llx] failed\n",
884 (unsigned long long) pfn << PAGE_SHIFT,
885 (((unsigned long long) pfn + nr_pages) << PAGE_SHIFT) - 1);
886 memory_notify(MEM_CANCEL_ONLINE, &arg);
David Hildenbrandfeee6b22020-01-04 12:59:33 -0800887 remove_pfn_range_from_zone(zone, pfn, nr_pages);
David Hildenbrand381eab42018-10-30 15:10:29 -0700888 mem_hotplug_done();
Chen Yuconge33e33b2016-03-17 14:19:35 -0700889 return ret;
Dave Hansen3947be12005-10-29 18:16:54 -0700890}
Keith Mannthey53947022006-09-30 23:27:08 -0700891#endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
Yasunori Gotobc02af92006-06-27 02:53:30 -0700892
Tang Chen0bd85422014-11-13 15:19:41 -0800893static void reset_node_present_pages(pg_data_t *pgdat)
894{
895 struct zone *z;
896
897 for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++)
898 z->present_pages = 0;
899
900 pgdat->node_present_pages = 0;
901}
902
Hidetoshi Setoe1319332009-11-17 14:06:18 -0800903/* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
David Hildenbrandc68ab182020-06-04 16:48:35 -0700904static pg_data_t __ref *hotadd_new_pgdat(int nid)
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700905{
906 struct pglist_data *pgdat;
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700907
Tang Chena1e565a2013-02-22 16:33:18 -0800908 pgdat = NODE_DATA(nid);
909 if (!pgdat) {
910 pgdat = arch_alloc_nodedata(nid);
911 if (!pgdat)
912 return NULL;
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700913
Wei Yang33fce012019-09-23 15:35:52 -0700914 pgdat->per_cpu_nodestats =
915 alloc_percpu(struct per_cpu_nodestat);
Tang Chena1e565a2013-02-22 16:33:18 -0800916 arch_refresh_nodedata(nid, pgdat);
Gu Zhengb0dc3a32015-03-25 15:55:20 -0700917 } else {
Wei Yang33fce012019-09-23 15:35:52 -0700918 int cpu;
Mel Gormane716f2e2017-05-03 14:53:45 -0700919 /*
Joonsoo Kim97a225e2020-06-03 15:59:01 -0700920 * Reset the nr_zones, order and highest_zoneidx before reuse.
921 * Note that kswapd will init kswapd_highest_zoneidx properly
Mel Gormane716f2e2017-05-03 14:53:45 -0700922 * when it starts in the near future.
923 */
Gu Zhengb0dc3a32015-03-25 15:55:20 -0700924 pgdat->nr_zones = 0;
Mel Gorman38087d92016-07-28 15:45:49 -0700925 pgdat->kswapd_order = 0;
Joonsoo Kim97a225e2020-06-03 15:59:01 -0700926 pgdat->kswapd_highest_zoneidx = 0;
Wei Yang33fce012019-09-23 15:35:52 -0700927 for_each_online_cpu(cpu) {
928 struct per_cpu_nodestat *p;
929
930 p = per_cpu_ptr(pgdat->per_cpu_nodestats, cpu);
931 memset(p, 0, sizeof(*p));
932 }
Tang Chena1e565a2013-02-22 16:33:18 -0800933 }
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700934
935 /* we can use NODE_DATA(nid) from here */
Oscar Salvador03e85f92018-08-21 21:53:43 -0700936 pgdat->node_id = nid;
David Hildenbrandc68ab182020-06-04 16:48:35 -0700937 pgdat->node_start_pfn = 0;
Oscar Salvador03e85f92018-08-21 21:53:43 -0700938
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700939 /* init node's zones as empty zones, we don't have any present pages.*/
Oscar Salvador03e85f92018-08-21 21:53:43 -0700940 free_area_init_core_hotplug(nid);
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700941
KAMEZAWA Hiroyuki959ecc42011-06-15 15:08:38 -0700942 /*
943 * The node we allocated has no zone fallback lists. For avoiding
944 * to access not-initialized zonelist, build here.
945 */
Michal Hocko72675e12017-09-06 16:20:24 -0700946 build_all_zonelists(pgdat);
KAMEZAWA Hiroyuki959ecc42011-06-15 15:08:38 -0700947
Tang Chenf784a3f2014-11-13 15:19:39 -0800948 /*
Tang Chen0bd85422014-11-13 15:19:41 -0800949 * When memory is hot-added, all the memory is in offline state. So
950 * clear all zones' present_pages because they will be updated in
951 * online_pages() and offline_pages().
952 */
Oscar Salvador03e85f92018-08-21 21:53:43 -0700953 reset_node_managed_pages(pgdat);
Tang Chen0bd85422014-11-13 15:19:41 -0800954 reset_node_present_pages(pgdat);
955
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700956 return pgdat;
957}
958
Oscar Salvadorb9ff0362018-08-17 15:46:15 -0700959static void rollback_node_hotadd(int nid)
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700960{
Oscar Salvadorb9ff0362018-08-17 15:46:15 -0700961 pg_data_t *pgdat = NODE_DATA(nid);
962
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700963 arch_refresh_nodedata(nid, NULL);
Reza Arbab58301692016-08-11 15:33:12 -0700964 free_percpu(pgdat->per_cpu_nodestats);
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700965 arch_free_nodedata(pgdat);
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700966}
967
KAMEZAWA Hiroyuki0a547032006-06-27 02:53:35 -0700968
Toshi Kani01b0f192013-11-12 15:07:25 -0800969/**
970 * try_online_node - online a node if offlined
Mike Rapoporte8b098f2018-04-05 16:24:57 -0700971 * @nid: the node ID
Oscar Salvadorb9ff0362018-08-17 15:46:15 -0700972 * @set_node_online: Whether we want to online the node
minskey guocf234222010-05-24 14:32:41 -0700973 * called by cpu_up() to online a node without onlined memory.
Oscar Salvadorb9ff0362018-08-17 15:46:15 -0700974 *
975 * Returns:
976 * 1 -> a new node has been allocated
977 * 0 -> the node is already online
978 * -ENOMEM -> the node could not be allocated
minskey guocf234222010-05-24 14:32:41 -0700979 */
David Hildenbrandc68ab182020-06-04 16:48:35 -0700980static int __try_online_node(int nid, bool set_node_online)
minskey guocf234222010-05-24 14:32:41 -0700981{
Oscar Salvadorb9ff0362018-08-17 15:46:15 -0700982 pg_data_t *pgdat;
983 int ret = 1;
minskey guocf234222010-05-24 14:32:41 -0700984
Toshi Kani01b0f192013-11-12 15:07:25 -0800985 if (node_online(nid))
986 return 0;
987
David Hildenbrandc68ab182020-06-04 16:48:35 -0700988 pgdat = hotadd_new_pgdat(nid);
David Rientjes7553e8f2011-06-22 18:13:01 -0700989 if (!pgdat) {
Toshi Kani01b0f192013-11-12 15:07:25 -0800990 pr_err("Cannot online node %d due to NULL pgdat\n", nid);
minskey guocf234222010-05-24 14:32:41 -0700991 ret = -ENOMEM;
992 goto out;
993 }
Oscar Salvadorb9ff0362018-08-17 15:46:15 -0700994
995 if (set_node_online) {
996 node_set_online(nid);
997 ret = register_one_node(nid);
998 BUG_ON(ret);
999 }
minskey guocf234222010-05-24 14:32:41 -07001000out:
Oscar Salvadorb9ff0362018-08-17 15:46:15 -07001001 return ret;
1002}
1003
1004/*
1005 * Users of this function always want to online/register the node
1006 */
1007int try_online_node(int nid)
1008{
1009 int ret;
1010
1011 mem_hotplug_begin();
David Hildenbrandc68ab182020-06-04 16:48:35 -07001012 ret = __try_online_node(nid, true);
Vladimir Davydovbfc8c902014-06-04 16:07:18 -07001013 mem_hotplug_done();
minskey guocf234222010-05-24 14:32:41 -07001014 return ret;
1015}
1016
Toshi Kani27356f52013-09-11 14:21:49 -07001017static int check_hotplug_memory_range(u64 start, u64 size)
1018{
Pavel Tatashinba325582018-04-05 16:22:39 -07001019 /* memory range must be block size aligned */
David Hildenbrandcec3ebd2019-07-18 15:56:25 -07001020 if (!size || !IS_ALIGNED(start, memory_block_size_bytes()) ||
1021 !IS_ALIGNED(size, memory_block_size_bytes())) {
Pavel Tatashinba325582018-04-05 16:22:39 -07001022 pr_err("Block size [%#lx] unaligned hotplug range: start %#llx, size %#llx",
David Hildenbrandcec3ebd2019-07-18 15:56:25 -07001023 memory_block_size_bytes(), start, size);
Toshi Kani27356f52013-09-11 14:21:49 -07001024 return -EINVAL;
1025 }
1026
1027 return 0;
1028}
1029
Vitaly Kuznetsov31bc3852016-03-15 14:56:48 -07001030static int online_memory_block(struct memory_block *mem, void *arg)
1031{
David Hildenbrand862919e2020-04-06 20:07:40 -07001032 mem->online_type = memhp_default_online_type;
Nathan Fontenotdc18d702017-02-24 15:00:02 -08001033 return device_online(&mem->dev);
Vitaly Kuznetsov31bc3852016-03-15 14:56:48 -07001034}
1035
David Hildenbrand8df1d0e2018-10-30 15:10:24 -07001036/*
1037 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1038 * and online/offline operations (triggered e.g. by sysfs).
1039 *
1040 * we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG
1041 */
David Hildenbrandb6117192020-10-15 20:08:44 -07001042int __ref add_memory_resource(int nid, struct resource *res, mhp_t mhp_flags)
Yasunori Gotobc02af92006-06-27 02:53:30 -07001043{
Logan Gunthorpebfeb0222020-04-10 14:33:36 -07001044 struct mhp_params params = { .pgprot = PAGE_KERNEL };
David Vrabel62cedb92015-06-25 16:35:49 +01001045 u64 start, size;
Oscar Salvadorb9ff0362018-08-17 15:46:15 -07001046 bool new_node = false;
Yasunori Gotobc02af92006-06-27 02:53:30 -07001047 int ret;
1048
David Vrabel62cedb92015-06-25 16:35:49 +01001049 start = res->start;
1050 size = resource_size(res);
1051
Toshi Kani27356f52013-09-11 14:21:49 -07001052 ret = check_hotplug_memory_range(start, size);
1053 if (ret)
1054 return ret;
1055
Vishal Vermafa6d9ec2020-06-04 16:48:25 -07001056 if (!node_possible(nid)) {
1057 WARN(1, "node %d was absent from the node_possible_map\n", nid);
1058 return -EINVAL;
1059 }
1060
Vladimir Davydovbfc8c902014-06-04 16:07:18 -07001061 mem_hotplug_begin();
Nathan Zimmerac13c462014-01-23 15:53:26 -08001062
David Hildenbrand52219ae2020-06-04 16:48:38 -07001063 if (IS_ENABLED(CONFIG_ARCH_KEEP_MEMBLOCK))
1064 memblock_add_node(start, size, nid);
Tang Chen7f36e3e2015-09-04 15:42:32 -07001065
David Hildenbrandc68ab182020-06-04 16:48:35 -07001066 ret = __try_online_node(nid, false);
Oscar Salvadorb9ff0362018-08-17 15:46:15 -07001067 if (ret < 0)
1068 goto error;
1069 new_node = ret;
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001070
Yasunori Gotobc02af92006-06-27 02:53:30 -07001071 /* call arch's memory hotadd */
Logan Gunthorpef5637d32020-04-10 14:33:21 -07001072 ret = arch_add_memory(nid, start, size, &params);
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001073 if (ret < 0)
1074 goto error;
1075
David Hildenbranddb051a02019-07-18 15:56:56 -07001076 /* create memory block devices after memory was added */
1077 ret = create_memory_block_devices(start, size);
1078 if (ret) {
1079 arch_remove_memory(nid, start, size, NULL);
1080 goto error;
1081 }
1082
Tang Chena1e565a2013-02-22 16:33:18 -08001083 if (new_node) {
Oscar Salvadord5b6f6a32018-08-17 15:46:18 -07001084 /* If sysfs file of new node can't be created, cpu on the node
Yasunori Goto0fc44152006-06-27 02:53:38 -07001085 * can't be hot-added. There is no rollback way now.
1086 * So, check by BUG_ON() to catch it reluctantly..
Oscar Salvadord5b6f6a32018-08-17 15:46:18 -07001087 * We online node here. We can't roll back from here.
Yasunori Goto0fc44152006-06-27 02:53:38 -07001088 */
Oscar Salvadord5b6f6a32018-08-17 15:46:18 -07001089 node_set_online(nid);
1090 ret = __register_one_node(nid);
Yasunori Goto0fc44152006-06-27 02:53:38 -07001091 BUG_ON(ret);
1092 }
1093
Oscar Salvadord5b6f6a32018-08-17 15:46:18 -07001094 /* link memory sections under this node.*/
Laurent Dufour90c7eae2020-10-15 20:09:15 -07001095 link_mem_sections(nid, PFN_DOWN(start), PFN_UP(start + size - 1),
1096 MEMINIT_HOTPLUG);
Oscar Salvadord5b6f6a32018-08-17 15:46:18 -07001097
akpm@linux-foundation.orgd96ae532010-03-05 13:41:58 -08001098 /* create new memmap entry */
David Hildenbrand7b7b2722020-06-04 16:48:41 -07001099 if (!strcmp(res->name, "System RAM"))
1100 firmware_map_add_hotplug(start, start + size, "System RAM");
akpm@linux-foundation.orgd96ae532010-03-05 13:41:58 -08001101
David Hildenbrand381eab42018-10-30 15:10:29 -07001102 /* device_online() will take the lock when calling online_pages() */
1103 mem_hotplug_done();
1104
David Hildenbrand9ca65512020-10-15 20:08:49 -07001105 /*
1106 * In case we're allowed to merge the resource, flag it and trigger
1107 * merging now that adding succeeded.
1108 */
1109 if (mhp_flags & MEMHP_MERGE_RESOURCE)
1110 merge_system_ram_resource(res);
1111
Vitaly Kuznetsov31bc3852016-03-15 14:56:48 -07001112 /* online pages if requested */
David Hildenbrand862919e2020-04-06 20:07:40 -07001113 if (memhp_default_online_type != MMOP_OFFLINE)
David Hildenbrandfbcf73c2019-07-18 15:57:46 -07001114 walk_memory_blocks(start, size, NULL, online_memory_block);
Vitaly Kuznetsov31bc3852016-03-15 14:56:48 -07001115
David Hildenbrand381eab42018-10-30 15:10:29 -07001116 return ret;
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001117error:
1118 /* rollback pgdat allocation and others */
Oscar Salvadorb9ff0362018-08-17 15:46:15 -07001119 if (new_node)
1120 rollback_node_hotadd(nid);
David Hildenbrand52219ae2020-06-04 16:48:38 -07001121 if (IS_ENABLED(CONFIG_ARCH_KEEP_MEMBLOCK))
1122 memblock_remove(start, size);
Vladimir Davydovbfc8c902014-06-04 16:07:18 -07001123 mem_hotplug_done();
Yasunori Gotobc02af92006-06-27 02:53:30 -07001124 return ret;
1125}
David Vrabel62cedb92015-06-25 16:35:49 +01001126
David Hildenbrand8df1d0e2018-10-30 15:10:24 -07001127/* requires device_hotplug_lock, see add_memory_resource() */
David Hildenbrandb6117192020-10-15 20:08:44 -07001128int __ref __add_memory(int nid, u64 start, u64 size, mhp_t mhp_flags)
David Vrabel62cedb92015-06-25 16:35:49 +01001129{
1130 struct resource *res;
1131 int ret;
1132
David Hildenbrand7b7b2722020-06-04 16:48:41 -07001133 res = register_memory_resource(start, size, "System RAM");
Vitaly Kuznetsov6f754ba2016-01-14 15:21:55 -08001134 if (IS_ERR(res))
1135 return PTR_ERR(res);
David Vrabel62cedb92015-06-25 16:35:49 +01001136
David Hildenbrandb6117192020-10-15 20:08:44 -07001137 ret = add_memory_resource(nid, res, mhp_flags);
David Vrabel62cedb92015-06-25 16:35:49 +01001138 if (ret < 0)
1139 release_memory_resource(res);
1140 return ret;
1141}
David Hildenbrand8df1d0e2018-10-30 15:10:24 -07001142
David Hildenbrandb6117192020-10-15 20:08:44 -07001143int add_memory(int nid, u64 start, u64 size, mhp_t mhp_flags)
David Hildenbrand8df1d0e2018-10-30 15:10:24 -07001144{
1145 int rc;
1146
1147 lock_device_hotplug();
David Hildenbrandb6117192020-10-15 20:08:44 -07001148 rc = __add_memory(nid, start, size, mhp_flags);
David Hildenbrand8df1d0e2018-10-30 15:10:24 -07001149 unlock_device_hotplug();
1150
1151 return rc;
1152}
Yasunori Gotobc02af92006-06-27 02:53:30 -07001153EXPORT_SYMBOL_GPL(add_memory);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001154
David Hildenbrand7b7b2722020-06-04 16:48:41 -07001155/*
1156 * Add special, driver-managed memory to the system as system RAM. Such
1157 * memory is not exposed via the raw firmware-provided memmap as system
1158 * RAM, instead, it is detected and added by a driver - during cold boot,
1159 * after a reboot, and after kexec.
1160 *
1161 * Reasons why this memory should not be used for the initial memmap of a
1162 * kexec kernel or for placing kexec images:
1163 * - The booting kernel is in charge of determining how this memory will be
1164 * used (e.g., use persistent memory as system RAM)
1165 * - Coordination with a hypervisor is required before this memory
1166 * can be used (e.g., inaccessible parts).
1167 *
1168 * For this memory, no entries in /sys/firmware/memmap ("raw firmware-provided
1169 * memory map") are created. Also, the created memory resource is flagged
David Hildenbrand7cf603d2020-10-15 20:08:33 -07001170 * with IORESOURCE_SYSRAM_DRIVER_MANAGED, so in-kernel users can special-case
David Hildenbrand7b7b2722020-06-04 16:48:41 -07001171 * this memory as well (esp., not place kexec images onto it).
1172 *
1173 * The resource_name (visible via /proc/iomem) has to have the format
1174 * "System RAM ($DRIVER)".
1175 */
1176int add_memory_driver_managed(int nid, u64 start, u64 size,
David Hildenbrandb6117192020-10-15 20:08:44 -07001177 const char *resource_name, mhp_t mhp_flags)
David Hildenbrand7b7b2722020-06-04 16:48:41 -07001178{
1179 struct resource *res;
1180 int rc;
1181
1182 if (!resource_name ||
1183 strstr(resource_name, "System RAM (") != resource_name ||
1184 resource_name[strlen(resource_name) - 1] != ')')
1185 return -EINVAL;
1186
1187 lock_device_hotplug();
1188
1189 res = register_memory_resource(start, size, resource_name);
1190 if (IS_ERR(res)) {
1191 rc = PTR_ERR(res);
1192 goto out_unlock;
1193 }
1194
David Hildenbrandb6117192020-10-15 20:08:44 -07001195 rc = add_memory_resource(nid, res, mhp_flags);
David Hildenbrand7b7b2722020-06-04 16:48:41 -07001196 if (rc < 0)
1197 release_memory_resource(res);
1198
1199out_unlock:
1200 unlock_device_hotplug();
1201 return rc;
1202}
1203EXPORT_SYMBOL_GPL(add_memory_driver_managed);
1204
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001205#ifdef CONFIG_MEMORY_HOTREMOVE
1206/*
David Hildenbrand92917992020-02-03 17:34:26 -08001207 * Confirm all pages in a range [start, end) belong to the same zone (skipping
1208 * memory holes). When true, return the zone.
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001209 */
David Hildenbrand92917992020-02-03 17:34:26 -08001210struct zone *test_pages_in_a_zone(unsigned long start_pfn,
1211 unsigned long end_pfn)
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001212{
Andrew Banman5f0f2882015-12-29 14:54:25 -08001213 unsigned long pfn, sec_end_pfn;
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001214 struct zone *zone = NULL;
1215 struct page *page;
1216 int i;
Toshi Kanideb88a22017-02-03 13:13:20 -08001217 for (pfn = start_pfn, sec_end_pfn = SECTION_ALIGN_UP(start_pfn + 1);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001218 pfn < end_pfn;
Toshi Kanideb88a22017-02-03 13:13:20 -08001219 pfn = sec_end_pfn, sec_end_pfn += PAGES_PER_SECTION) {
Andrew Banman5f0f2882015-12-29 14:54:25 -08001220 /* Make sure the memory section is present first */
1221 if (!present_section_nr(pfn_to_section_nr(pfn)))
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001222 continue;
Andrew Banman5f0f2882015-12-29 14:54:25 -08001223 for (; pfn < sec_end_pfn && pfn < end_pfn;
1224 pfn += MAX_ORDER_NR_PAGES) {
1225 i = 0;
1226 /* This is just a CONFIG_HOLES_IN_ZONE check.*/
1227 while ((i < MAX_ORDER_NR_PAGES) &&
1228 !pfn_valid_within(pfn + i))
1229 i++;
zhong jiangd6d8c8a2017-02-24 14:59:30 -08001230 if (i == MAX_ORDER_NR_PAGES || pfn + i >= end_pfn)
Andrew Banman5f0f2882015-12-29 14:54:25 -08001231 continue;
Mikhail Zaslonko24feb472019-02-01 14:20:38 -08001232 /* Check if we got outside of the zone */
1233 if (zone && !zone_spans_pfn(zone, pfn + i))
David Hildenbrand92917992020-02-03 17:34:26 -08001234 return NULL;
Andrew Banman5f0f2882015-12-29 14:54:25 -08001235 page = pfn_to_page(pfn + i);
1236 if (zone && page_zone(page) != zone)
David Hildenbrand92917992020-02-03 17:34:26 -08001237 return NULL;
Andrew Banman5f0f2882015-12-29 14:54:25 -08001238 zone = page_zone(page);
1239 }
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001240 }
Toshi Kanideb88a22017-02-03 13:13:20 -08001241
David Hildenbrand92917992020-02-03 17:34:26 -08001242 return zone;
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001243}
1244
1245/*
Yisheng Xie0efadf42017-02-24 14:57:39 -08001246 * Scan pfn range [start,end) to find movable/migratable pages (LRU pages,
David Hildenbrandaa218792020-05-07 16:01:30 +02001247 * non-lru movable pages and hugepages). Will skip over most unmovable
1248 * pages (esp., pages that can be skipped when offlining), but bail out on
1249 * definitely unmovable pages.
1250 *
1251 * Returns:
1252 * 0 in case a movable page is found and movable_pfn was updated.
1253 * -ENOENT in case no movable page was found.
1254 * -EBUSY in case a definitely unmovable page was found.
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001255 */
David Hildenbrandaa218792020-05-07 16:01:30 +02001256static int scan_movable_pages(unsigned long start, unsigned long end,
1257 unsigned long *movable_pfn)
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001258{
1259 unsigned long pfn;
Oscar Salvadoreeb0efd2019-02-01 14:20:47 -08001260
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001261 for (pfn = start; pfn < end; pfn++) {
Oscar Salvadoreeb0efd2019-02-01 14:20:47 -08001262 struct page *page, *head;
1263 unsigned long skip;
1264
1265 if (!pfn_valid(pfn))
1266 continue;
1267 page = pfn_to_page(pfn);
1268 if (PageLRU(page))
David Hildenbrandaa218792020-05-07 16:01:30 +02001269 goto found;
Oscar Salvadoreeb0efd2019-02-01 14:20:47 -08001270 if (__PageMovable(page))
David Hildenbrandaa218792020-05-07 16:01:30 +02001271 goto found;
1272
1273 /*
1274 * PageOffline() pages that are not marked __PageMovable() and
1275 * have a reference count > 0 (after MEM_GOING_OFFLINE) are
1276 * definitely unmovable. If their reference count would be 0,
1277 * they could at least be skipped when offlining memory.
1278 */
1279 if (PageOffline(page) && page_count(page))
1280 return -EBUSY;
Oscar Salvadoreeb0efd2019-02-01 14:20:47 -08001281
1282 if (!PageHuge(page))
1283 continue;
1284 head = compound_head(page);
Oscar Salvador39186cb2019-05-13 17:17:32 -07001285 if (page_huge_active(head))
David Hildenbrandaa218792020-05-07 16:01:30 +02001286 goto found;
Matthew Wilcox (Oracle)d8c65462019-09-23 15:34:30 -07001287 skip = compound_nr(head) - (page - head);
Oscar Salvadoreeb0efd2019-02-01 14:20:47 -08001288 pfn += skip - 1;
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001289 }
David Hildenbrandaa218792020-05-07 16:01:30 +02001290 return -ENOENT;
1291found:
1292 *movable_pfn = pfn;
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001293 return 0;
1294}
1295
Michal Hocko666feb22018-04-10 16:30:03 -07001296static struct page *new_node_page(struct page *page, unsigned long private)
Xishi Qiu394e31d2016-07-28 15:48:53 -07001297{
Li Zhong231e97e2016-09-28 15:22:38 -07001298 nodemask_t nmask = node_states[N_MEMORY];
Joonsoo Kim19fc7be2020-08-11 18:37:25 -07001299 struct migration_target_control mtc = {
1300 .nid = page_to_nid(page),
1301 .nmask = &nmask,
1302 .gfp_mask = GFP_USER | __GFP_MOVABLE | __GFP_RETRY_MAYFAIL,
1303 };
Michal Hocko7f252f22017-07-10 15:48:41 -07001304
1305 /*
1306 * try to allocate from a different node but reuse this node if there
1307 * are no other online nodes to be used (e.g. we are offlining a part
1308 * of the only existing node)
1309 */
Joonsoo Kim19fc7be2020-08-11 18:37:25 -07001310 node_clear(mtc.nid, nmask);
Michal Hocko7f252f22017-07-10 15:48:41 -07001311 if (nodes_empty(nmask))
Joonsoo Kim19fc7be2020-08-11 18:37:25 -07001312 node_set(mtc.nid, nmask);
Xishi Qiu394e31d2016-07-28 15:48:53 -07001313
Joonsoo Kim19fc7be2020-08-11 18:37:25 -07001314 return alloc_migration_target(page, (unsigned long)&mtc);
Xishi Qiu394e31d2016-07-28 15:48:53 -07001315}
1316
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001317static int
1318do_migrate_range(unsigned long start_pfn, unsigned long end_pfn)
1319{
1320 unsigned long pfn;
Matthew Wilcox (Oracle)6c357842020-08-14 17:30:37 -07001321 struct page *page, *head;
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001322 int ret = 0;
1323 LIST_HEAD(source);
1324
Michal Hockoa85009c2018-12-28 00:38:29 -08001325 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001326 if (!pfn_valid(pfn))
1327 continue;
1328 page = pfn_to_page(pfn);
Matthew Wilcox (Oracle)6c357842020-08-14 17:30:37 -07001329 head = compound_head(page);
Naoya Horiguchic8721bb2013-09-11 14:22:09 -07001330
1331 if (PageHuge(page)) {
Matthew Wilcox (Oracle)d8c65462019-09-23 15:34:30 -07001332 pfn = page_to_pfn(head) + compound_nr(head) - 1;
Oscar Salvadordaf35382019-03-05 15:48:53 -08001333 isolate_huge_page(head, &source);
Naoya Horiguchic8721bb2013-09-11 14:22:09 -07001334 continue;
Michal Hocko94723aa2018-04-10 16:30:07 -07001335 } else if (PageTransHuge(page))
Matthew Wilcox (Oracle)6c357842020-08-14 17:30:37 -07001336 pfn = page_to_pfn(head) + thp_nr_pages(page) - 1;
Naoya Horiguchic8721bb2013-09-11 14:22:09 -07001337
Michal Hockob15c8722018-12-28 00:38:01 -08001338 /*
1339 * HWPoison pages have elevated reference counts so the migration would
1340 * fail on them. It also doesn't make any sense to migrate them in the
1341 * first place. Still try to unmap such a page in case it is still mapped
1342 * (e.g. current hwpoison implementation doesn't unmap KSM pages but keep
1343 * the unmap as the catch all safety net).
1344 */
1345 if (PageHWPoison(page)) {
1346 if (WARN_ON(PageLRU(page)))
1347 isolate_lru_page(page);
1348 if (page_mapped(page))
1349 try_to_unmap(page, TTU_IGNORE_MLOCK | TTU_IGNORE_ACCESS);
1350 continue;
1351 }
1352
Konstantin Khlebnikov700c2a42011-05-24 17:12:19 -07001353 if (!get_page_unless_zero(page))
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001354 continue;
1355 /*
Yisheng Xie0efadf42017-02-24 14:57:39 -08001356 * We can skip free pages. And we can deal with pages on
1357 * LRU and non-lru movable pages.
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001358 */
Yisheng Xie0efadf42017-02-24 14:57:39 -08001359 if (PageLRU(page))
1360 ret = isolate_lru_page(page);
1361 else
1362 ret = isolate_movable_page(page, ISOLATE_UNEVICTABLE);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001363 if (!ret) { /* Success */
Nick Piggin62695a82008-10-18 20:26:09 -07001364 list_add_tail(&page->lru, &source);
Yisheng Xie0efadf42017-02-24 14:57:39 -08001365 if (!__PageMovable(page))
1366 inc_node_page_state(page, NR_ISOLATED_ANON +
Huang Ying9de4f222020-04-06 20:04:41 -07001367 page_is_file_lru(page));
KOSAKI Motohiro6d9c2852009-12-14 17:58:11 -08001368
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001369 } else {
Michal Hocko2932c8b2018-12-28 00:33:53 -08001370 pr_warn("failed to isolate pfn %lx\n", pfn);
Yisheng Xie0efadf42017-02-24 14:57:39 -08001371 dump_page(page, "isolation failed");
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001372 }
Oscar Salvador17230582019-02-01 14:19:57 -08001373 put_page(page);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001374 }
Bob Liuf3ab2632010-10-26 14:22:10 -07001375 if (!list_empty(&source)) {
Xishi Qiu394e31d2016-07-28 15:48:53 -07001376 /* Allocate a new page from the nearest neighbor node */
1377 ret = migrate_pages(&source, new_node_page, NULL, 0,
Hugh Dickins9c620e22013-02-22 16:35:14 -08001378 MIGRATE_SYNC, MR_MEMORY_HOTPLUG);
Michal Hocko2932c8b2018-12-28 00:33:53 -08001379 if (ret) {
1380 list_for_each_entry(page, &source, lru) {
1381 pr_warn("migrating pfn %lx failed ret:%d ",
1382 page_to_pfn(page), ret);
1383 dump_page(page, "migration failure");
1384 }
Naoya Horiguchic8721bb2013-09-11 14:22:09 -07001385 putback_movable_pages(&source);
Michal Hocko2932c8b2018-12-28 00:33:53 -08001386 }
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001387 }
Oscar Salvador17230582019-02-01 14:19:57 -08001388
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001389 return ret;
1390}
1391
Tang Chenc5320922013-11-12 15:08:10 -08001392static int __init cmdline_parse_movable_node(char *p)
1393{
Tang Chen55ac5902014-01-21 15:49:35 -08001394 movable_node_enabled = true;
Tang Chenc5320922013-11-12 15:08:10 -08001395 return 0;
1396}
1397early_param("movable_node", cmdline_parse_movable_node);
1398
Lai Jiangshand9713672012-12-11 16:01:03 -08001399/* check which state of node_states will be changed when offline memory */
1400static void node_states_check_changes_offline(unsigned long nr_pages,
1401 struct zone *zone, struct memory_notify *arg)
1402{
1403 struct pglist_data *pgdat = zone->zone_pgdat;
1404 unsigned long present_pages = 0;
Oscar Salvador86b27be2018-10-26 15:07:38 -07001405 enum zone_type zt;
1406
Anshuman Khandual98fa15f2019-03-05 15:42:58 -08001407 arg->status_change_nid = NUMA_NO_NODE;
1408 arg->status_change_nid_normal = NUMA_NO_NODE;
1409 arg->status_change_nid_high = NUMA_NO_NODE;
Lai Jiangshand9713672012-12-11 16:01:03 -08001410
1411 /*
Oscar Salvador86b27be2018-10-26 15:07:38 -07001412 * Check whether node_states[N_NORMAL_MEMORY] will be changed.
1413 * If the memory to be offline is within the range
1414 * [0..ZONE_NORMAL], and it is the last present memory there,
1415 * the zones in that range will become empty after the offlining,
1416 * thus we can determine that we need to clear the node from
1417 * node_states[N_NORMAL_MEMORY].
Lai Jiangshand9713672012-12-11 16:01:03 -08001418 */
Oscar Salvador86b27be2018-10-26 15:07:38 -07001419 for (zt = 0; zt <= ZONE_NORMAL; zt++)
Lai Jiangshand9713672012-12-11 16:01:03 -08001420 present_pages += pgdat->node_zones[zt].present_pages;
Oscar Salvador86b27be2018-10-26 15:07:38 -07001421 if (zone_idx(zone) <= ZONE_NORMAL && nr_pages >= present_pages)
Lai Jiangshand9713672012-12-11 16:01:03 -08001422 arg->status_change_nid_normal = zone_to_nid(zone);
Lai Jiangshand9713672012-12-11 16:01:03 -08001423
Lai Jiangshan6715ddf2012-12-12 13:51:49 -08001424#ifdef CONFIG_HIGHMEM
1425 /*
Oscar Salvador86b27be2018-10-26 15:07:38 -07001426 * node_states[N_HIGH_MEMORY] contains nodes which
1427 * have normal memory or high memory.
1428 * Here we add the present_pages belonging to ZONE_HIGHMEM.
1429 * If the zone is within the range of [0..ZONE_HIGHMEM), and
1430 * we determine that the zones in that range become empty,
1431 * we need to clear the node for N_HIGH_MEMORY.
Lai Jiangshan6715ddf2012-12-12 13:51:49 -08001432 */
Oscar Salvador86b27be2018-10-26 15:07:38 -07001433 present_pages += pgdat->node_zones[ZONE_HIGHMEM].present_pages;
1434 if (zone_idx(zone) <= ZONE_HIGHMEM && nr_pages >= present_pages)
Lai Jiangshan6715ddf2012-12-12 13:51:49 -08001435 arg->status_change_nid_high = zone_to_nid(zone);
Lai Jiangshan6715ddf2012-12-12 13:51:49 -08001436#endif
1437
Lai Jiangshand9713672012-12-11 16:01:03 -08001438 /*
Oscar Salvador86b27be2018-10-26 15:07:38 -07001439 * We have accounted the pages from [0..ZONE_NORMAL), and
1440 * in case of CONFIG_HIGHMEM the pages from ZONE_HIGHMEM
1441 * as well.
1442 * Here we count the possible pages from ZONE_MOVABLE.
1443 * If after having accounted all the pages, we see that the nr_pages
1444 * to be offlined is over or equal to the accounted pages,
1445 * we know that the node will become empty, and so, we can clear
1446 * it for N_MEMORY as well.
Lai Jiangshand9713672012-12-11 16:01:03 -08001447 */
Oscar Salvador86b27be2018-10-26 15:07:38 -07001448 present_pages += pgdat->node_zones[ZONE_MOVABLE].present_pages;
Lai Jiangshand9713672012-12-11 16:01:03 -08001449
Lai Jiangshand9713672012-12-11 16:01:03 -08001450 if (nr_pages >= present_pages)
1451 arg->status_change_nid = zone_to_nid(zone);
Lai Jiangshand9713672012-12-11 16:01:03 -08001452}
1453
1454static void node_states_clear_node(int node, struct memory_notify *arg)
1455{
1456 if (arg->status_change_nid_normal >= 0)
1457 node_clear_state(node, N_NORMAL_MEMORY);
1458
Oscar Salvadorcf01f6f52018-10-26 15:07:28 -07001459 if (arg->status_change_nid_high >= 0)
Lai Jiangshand9713672012-12-11 16:01:03 -08001460 node_clear_state(node, N_HIGH_MEMORY);
Lai Jiangshan6715ddf2012-12-12 13:51:49 -08001461
Oscar Salvadorcf01f6f52018-10-26 15:07:28 -07001462 if (arg->status_change_nid >= 0)
Lai Jiangshan6715ddf2012-12-12 13:51:49 -08001463 node_clear_state(node, N_MEMORY);
Lai Jiangshand9713672012-12-11 16:01:03 -08001464}
1465
David Hildenbrandc5e79ef2019-11-30 17:54:17 -08001466static int count_system_ram_pages_cb(unsigned long start_pfn,
1467 unsigned long nr_pages, void *data)
1468{
1469 unsigned long *nr_system_ram_pages = data;
1470
1471 *nr_system_ram_pages += nr_pages;
1472 return 0;
1473}
1474
David Hildenbrand73a11c92020-10-15 20:07:46 -07001475int __ref offline_pages(unsigned long start_pfn, unsigned long nr_pages)
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001476{
David Hildenbrand73a11c92020-10-15 20:07:46 -07001477 const unsigned long end_pfn = start_pfn + nr_pages;
David Hildenbrand0a1a9a02020-10-15 20:07:54 -07001478 unsigned long pfn, system_ram_pages = 0;
Cody P Schaferd7029092013-07-03 15:02:11 -07001479 unsigned long flags;
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001480 struct zone *zone;
Yasunori Goto7b78d332007-10-21 16:41:36 -07001481 struct memory_notify arg;
David Hildenbrandea151532020-10-15 20:08:03 -07001482 int ret, node;
Michal Hocko79605092018-12-28 00:33:49 -08001483 char *reason;
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001484
David Hildenbrand4986fac2020-10-15 20:07:50 -07001485 /* We can only offline full sections (e.g., SECTION_IS_ONLINE) */
1486 if (WARN_ON_ONCE(!nr_pages ||
1487 !IS_ALIGNED(start_pfn | nr_pages, PAGES_PER_SECTION)))
1488 return -EINVAL;
1489
David Hildenbrand381eab42018-10-30 15:10:29 -07001490 mem_hotplug_begin();
1491
David Hildenbrandc5e79ef2019-11-30 17:54:17 -08001492 /*
1493 * Don't allow to offline memory blocks that contain holes.
1494 * Consequently, memory blocks with holes can never get onlined
1495 * via the hotplug path - online_pages() - as hotplugged memory has
1496 * no holes. This way, we e.g., don't have to worry about marking
1497 * memory holes PG_reserved, don't need pfn_valid() checks, and can
1498 * avoid using walk_system_ram_range() later.
1499 */
David Hildenbrand73a11c92020-10-15 20:07:46 -07001500 walk_system_ram_range(start_pfn, nr_pages, &system_ram_pages,
David Hildenbrandc5e79ef2019-11-30 17:54:17 -08001501 count_system_ram_pages_cb);
David Hildenbrand73a11c92020-10-15 20:07:46 -07001502 if (system_ram_pages != nr_pages) {
David Hildenbrandc5e79ef2019-11-30 17:54:17 -08001503 ret = -EINVAL;
1504 reason = "memory holes";
1505 goto failed_removal;
1506 }
1507
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001508 /* This makes hotplug much easier...and readable.
1509 we assume this for now. .*/
David Hildenbrand92917992020-02-03 17:34:26 -08001510 zone = test_pages_in_a_zone(start_pfn, end_pfn);
1511 if (!zone) {
Michal Hocko79605092018-12-28 00:33:49 -08001512 ret = -EINVAL;
1513 reason = "multizone range";
1514 goto failed_removal;
David Hildenbrand381eab42018-10-30 15:10:29 -07001515 }
Yasunori Goto7b78d332007-10-21 16:41:36 -07001516 node = zone_to_nid(zone);
Yasunori Goto7b78d332007-10-21 16:41:36 -07001517
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001518 /* set above range as isolated */
Wen Congyangb023f462012-12-11 16:00:45 -08001519 ret = start_isolate_page_range(start_pfn, end_pfn,
Michal Hockod381c542018-12-28 00:33:56 -08001520 MIGRATE_MOVABLE,
David Hildenbrand756d25be2019-11-30 17:54:07 -08001521 MEMORY_OFFLINE | REPORT_FAILURE);
David Hildenbrand3fa0c7c2020-10-15 20:08:07 -07001522 if (ret) {
Michal Hocko79605092018-12-28 00:33:49 -08001523 reason = "failure to isolate range";
1524 goto failed_removal;
David Hildenbrand381eab42018-10-30 15:10:29 -07001525 }
Yasunori Goto7b78d332007-10-21 16:41:36 -07001526
1527 arg.start_pfn = start_pfn;
1528 arg.nr_pages = nr_pages;
Lai Jiangshand9713672012-12-11 16:01:03 -08001529 node_states_check_changes_offline(nr_pages, zone, &arg);
Yasunori Goto7b78d332007-10-21 16:41:36 -07001530
1531 ret = memory_notify(MEM_GOING_OFFLINE, &arg);
1532 ret = notifier_to_errno(ret);
Michal Hocko79605092018-12-28 00:33:49 -08001533 if (ret) {
1534 reason = "notifier failure";
1535 goto failed_removal_isolated;
1536 }
Yasunori Goto7b78d332007-10-21 16:41:36 -07001537
Michal Hockobb8965b2018-12-28 00:38:32 -08001538 do {
David Hildenbrandaa218792020-05-07 16:01:30 +02001539 pfn = start_pfn;
1540 do {
Michal Hockobb8965b2018-12-28 00:38:32 -08001541 if (signal_pending(current)) {
1542 ret = -EINTR;
1543 reason = "signal backoff";
1544 goto failed_removal_isolated;
1545 }
Michal Hocko72b39cf2017-11-15 17:33:34 -08001546
Michal Hockobb8965b2018-12-28 00:38:32 -08001547 cond_resched();
1548 lru_add_drain_all();
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001549
David Hildenbrandaa218792020-05-07 16:01:30 +02001550 ret = scan_movable_pages(pfn, end_pfn, &pfn);
1551 if (!ret) {
Michal Hockobb8965b2018-12-28 00:38:32 -08001552 /*
1553 * TODO: fatal migration failures should bail
1554 * out
1555 */
1556 do_migrate_range(pfn, end_pfn);
1557 }
David Hildenbrandaa218792020-05-07 16:01:30 +02001558 } while (!ret);
1559
1560 if (ret != -ENOENT) {
1561 reason = "unmovable page";
1562 goto failed_removal_isolated;
Michal Hockobb8965b2018-12-28 00:38:32 -08001563 }
Michal Hocko72b39cf2017-11-15 17:33:34 -08001564
Michal Hockobb8965b2018-12-28 00:38:32 -08001565 /*
1566 * Dissolve free hugepages in the memory block before doing
1567 * offlining actually in order to make hugetlbfs's object
1568 * counting consistent.
1569 */
1570 ret = dissolve_free_huge_pages(start_pfn, end_pfn);
1571 if (ret) {
1572 reason = "failure to dissolve huge pages";
1573 goto failed_removal_isolated;
1574 }
David Hildenbrand0a1a9a02020-10-15 20:07:54 -07001575
Pavel Tatashin96831822020-09-18 21:20:31 -07001576 /*
1577 * per-cpu pages are drained in start_isolate_page_range, but if
1578 * there are still pages that are not free, make sure that we
1579 * drain again, because when we isolated range we might
1580 * have raced with another thread that was adding pages to pcp
1581 * list.
1582 *
1583 * Forward progress should be still guaranteed because
1584 * pages on the pcp list can only belong to MOVABLE_ZONE
1585 * because has_unmovable_pages explicitly checks for
1586 * PageBuddy on freed pages on other zones.
1587 */
David Hildenbrand0a1a9a02020-10-15 20:07:54 -07001588 ret = test_pages_isolated(start_pfn, end_pfn, MEMORY_OFFLINE);
Pavel Tatashin96831822020-09-18 21:20:31 -07001589 if (ret)
1590 drain_all_pages(zone);
Michal Hocko5557c762019-05-13 17:21:24 -07001591 } while (ret);
Michal Hockobb8965b2018-12-28 00:38:32 -08001592
David Hildenbrand0a1a9a02020-10-15 20:07:54 -07001593 /* Mark all sections offline and remove free pages from the buddy. */
1594 __offline_isolated_pages(start_pfn, end_pfn);
1595 pr_info("Offlined Pages %ld\n", nr_pages);
1596
Qian Cai9b7ea462019-03-28 20:43:34 -07001597 /*
David Hildenbrandb30c5922020-10-15 20:08:23 -07001598 * The memory sections are marked offline, and the pageblock flags
1599 * effectively stale; nobody should be touching them. Fixup the number
1600 * of isolated pageblocks, memory onlining will properly revert this.
Qian Cai9b7ea462019-03-28 20:43:34 -07001601 */
1602 spin_lock_irqsave(&zone->lock, flags);
David Hildenbrandea151532020-10-15 20:08:03 -07001603 zone->nr_isolate_pageblock -= nr_pages / pageblock_nr_pages;
Qian Cai9b7ea462019-03-28 20:43:34 -07001604 spin_unlock_irqrestore(&zone->lock, flags);
1605
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001606 /* removal success */
David Hildenbrand0a1a9a02020-10-15 20:07:54 -07001607 adjust_managed_page_count(pfn_to_page(start_pfn), -nr_pages);
1608 zone->present_pages -= nr_pages;
Cody P Schaferd7029092013-07-03 15:02:11 -07001609
1610 pgdat_resize_lock(zone->zone_pgdat, &flags);
David Hildenbrand0a1a9a02020-10-15 20:07:54 -07001611 zone->zone_pgdat->node_present_pages -= nr_pages;
Cody P Schaferd7029092013-07-03 15:02:11 -07001612 pgdat_resize_unlock(zone->zone_pgdat, &flags);
Yasunori Goto7b78d332007-10-21 16:41:36 -07001613
KOSAKI Motohiro1b79acc2011-05-24 17:11:32 -07001614 init_per_zone_wmark_min();
1615
Xishi Qiu1e8537b2012-10-08 16:31:51 -07001616 if (!populated_zone(zone)) {
Jiang Liu340175b2012-07-31 16:43:32 -07001617 zone_pcp_reset(zone);
Michal Hocko72675e12017-09-06 16:20:24 -07001618 build_all_zonelists(NULL);
Xishi Qiu1e8537b2012-10-08 16:31:51 -07001619 } else
1620 zone_pcp_update(zone);
Jiang Liu340175b2012-07-31 16:43:32 -07001621
Lai Jiangshand9713672012-12-11 16:01:03 -08001622 node_states_clear_node(node, &arg);
Vlastimil Babka698b1b32016-03-17 14:18:08 -07001623 if (arg.status_change_nid >= 0) {
David Rientjes8fe23e02009-12-14 17:58:33 -08001624 kswapd_stop(node);
Vlastimil Babka698b1b32016-03-17 14:18:08 -07001625 kcompactd_stop(node);
1626 }
Minchan Kimbce73942009-06-16 15:32:50 -07001627
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001628 writeback_set_ratelimit();
Yasunori Goto7b78d332007-10-21 16:41:36 -07001629
1630 memory_notify(MEM_OFFLINE, &arg);
David Hildenbrandfeee6b22020-01-04 12:59:33 -08001631 remove_pfn_range_from_zone(zone, start_pfn, nr_pages);
David Hildenbrand381eab42018-10-30 15:10:29 -07001632 mem_hotplug_done();
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001633 return 0;
1634
Michal Hocko79605092018-12-28 00:33:49 -08001635failed_removal_isolated:
1636 undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
Qian Caic4efe482019-03-28 20:44:16 -07001637 memory_notify(MEM_CANCEL_OFFLINE, &arg);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001638failed_removal:
Michal Hocko79605092018-12-28 00:33:49 -08001639 pr_debug("memory offlining [mem %#010llx-%#010llx] failed due to %s\n",
Chen Yuconge33e33b2016-03-17 14:19:35 -07001640 (unsigned long long) start_pfn << PAGE_SHIFT,
Michal Hocko79605092018-12-28 00:33:49 -08001641 ((unsigned long long) end_pfn << PAGE_SHIFT) - 1,
1642 reason);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001643 /* pushback to free area */
David Hildenbrand381eab42018-10-30 15:10:29 -07001644 mem_hotplug_done();
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001645 return ret;
1646}
Badari Pulavarty71088782008-10-18 20:25:58 -07001647
Xishi Qiud6de9d52013-11-12 15:07:20 -08001648static int check_memblock_offlined_cb(struct memory_block *mem, void *arg)
Wen Congyangbbc76be2013-02-22 16:32:54 -08001649{
1650 int ret = !is_memblock_offlined(mem);
1651
Randy Dunlap349daa02013-04-29 15:08:49 -07001652 if (unlikely(ret)) {
1653 phys_addr_t beginpa, endpa;
1654
1655 beginpa = PFN_PHYS(section_nr_to_pfn(mem->start_section_nr));
David Hildenbrandb6c88d32019-09-23 15:35:49 -07001656 endpa = beginpa + memory_block_size_bytes() - 1;
Joe Perches756a0252016-03-17 14:19:47 -07001657 pr_warn("removing memory fails, because memory [%pa-%pa] is onlined\n",
Randy Dunlap349daa02013-04-29 15:08:49 -07001658 &beginpa, &endpa);
Wen Congyangbbc76be2013-02-22 16:32:54 -08001659
Pavel Tatashineca499a2019-07-16 16:30:31 -07001660 return -EBUSY;
1661 }
1662 return 0;
Wen Congyangbbc76be2013-02-22 16:32:54 -08001663}
1664
Toshi Kani0f1cfe92013-09-11 14:21:50 -07001665static int check_cpu_on_node(pg_data_t *pgdat)
Tang Chen60a5a192013-02-22 16:33:14 -08001666{
Tang Chen60a5a192013-02-22 16:33:14 -08001667 int cpu;
1668
1669 for_each_present_cpu(cpu) {
1670 if (cpu_to_node(cpu) == pgdat->node_id)
1671 /*
1672 * the cpu on this node isn't removed, and we can't
1673 * offline this node.
1674 */
1675 return -EBUSY;
1676 }
1677
1678 return 0;
1679}
1680
David Hildenbrand2c91f8f2019-11-15 17:34:57 -08001681static int check_no_memblock_for_node_cb(struct memory_block *mem, void *arg)
1682{
1683 int nid = *(int *)arg;
1684
1685 /*
1686 * If a memory block belongs to multiple nodes, the stored nid is not
1687 * reliable. However, such blocks are always online (e.g., cannot get
1688 * offlined) and, therefore, are still spanned by the node.
1689 */
1690 return mem->nid == nid ? -EEXIST : 0;
1691}
1692
Toshi Kani0f1cfe92013-09-11 14:21:50 -07001693/**
1694 * try_offline_node
Mike Rapoporte8b098f2018-04-05 16:24:57 -07001695 * @nid: the node ID
Toshi Kani0f1cfe92013-09-11 14:21:50 -07001696 *
1697 * Offline a node if all memory sections and cpus of the node are removed.
1698 *
1699 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1700 * and online/offline operations before this call.
1701 */
Wen Congyang90b30cd2013-02-22 16:33:27 -08001702void try_offline_node(int nid)
Tang Chen60a5a192013-02-22 16:33:14 -08001703{
Wen Congyangd822b862013-02-22 16:33:16 -08001704 pg_data_t *pgdat = NODE_DATA(nid);
David Hildenbrand2c91f8f2019-11-15 17:34:57 -08001705 int rc;
Tang Chen60a5a192013-02-22 16:33:14 -08001706
David Hildenbrand2c91f8f2019-11-15 17:34:57 -08001707 /*
1708 * If the node still spans pages (especially ZONE_DEVICE), don't
1709 * offline it. A node spans memory after move_pfn_range_to_zone(),
1710 * e.g., after the memory block was onlined.
1711 */
1712 if (pgdat->node_spanned_pages)
Tang Chen60a5a192013-02-22 16:33:14 -08001713 return;
David Hildenbrand2c91f8f2019-11-15 17:34:57 -08001714
1715 /*
1716 * Especially offline memory blocks might not be spanned by the
1717 * node. They will get spanned by the node once they get onlined.
1718 * However, they link to the node in sysfs and can get onlined later.
1719 */
1720 rc = for_each_memory_block(&nid, check_no_memblock_for_node_cb);
1721 if (rc)
1722 return;
Tang Chen60a5a192013-02-22 16:33:14 -08001723
Michal Hocko46a36792018-12-28 00:34:13 -08001724 if (check_cpu_on_node(pgdat))
Tang Chen60a5a192013-02-22 16:33:14 -08001725 return;
1726
1727 /*
1728 * all memory/cpu of this node are removed, we can offline this
1729 * node now.
1730 */
1731 node_set_offline(nid);
1732 unregister_one_node(nid);
1733}
Wen Congyang90b30cd2013-02-22 16:33:27 -08001734EXPORT_SYMBOL(try_offline_node);
Tang Chen60a5a192013-02-22 16:33:14 -08001735
Pavel Tatashineca499a2019-07-16 16:30:31 -07001736static int __ref try_remove_memory(int nid, u64 start, u64 size)
Wen Congyangbbc76be2013-02-22 16:32:54 -08001737{
Pavel Tatashineca499a2019-07-16 16:30:31 -07001738 int rc = 0;
Wen Congyang993c1aa2013-02-22 16:32:50 -08001739
Toshi Kani27356f52013-09-11 14:21:49 -07001740 BUG_ON(check_hotplug_memory_range(start, size));
1741
Yasuaki Ishimatsu6677e3e2013-02-22 16:32:52 -08001742 /*
Rafael J. Wysocki242831e2013-05-27 12:58:46 +02001743 * All memory blocks must be offlined before removing memory. Check
Pavel Tatashineca499a2019-07-16 16:30:31 -07001744 * whether all memory blocks in question are offline and return error
Rafael J. Wysocki242831e2013-05-27 12:58:46 +02001745 * if this is not the case.
Yasuaki Ishimatsu6677e3e2013-02-22 16:32:52 -08001746 */
David Hildenbrandfbcf73c2019-07-18 15:57:46 -07001747 rc = walk_memory_blocks(start, size, NULL, check_memblock_offlined_cb);
Pavel Tatashineca499a2019-07-16 16:30:31 -07001748 if (rc)
Jia Heb4223a52020-08-11 18:32:20 -07001749 return rc;
Yasuaki Ishimatsu6677e3e2013-02-22 16:32:52 -08001750
Yasuaki Ishimatsu46c66c42013-02-22 16:32:56 -08001751 /* remove memmap entry */
1752 firmware_map_remove(start, start + size, "System RAM");
1753
Dan Williamsf1037ec2020-01-30 22:11:17 -08001754 /*
1755 * Memory block device removal under the device_hotplug_lock is
1756 * a barrier against racing online attempts.
1757 */
David Hildenbrand4c4b7f92019-07-18 15:57:06 -07001758 remove_memory_block_devices(start, size);
1759
Dan Williamsf1037ec2020-01-30 22:11:17 -08001760 mem_hotplug_begin();
1761
Oscar Salvador2c2a5af2018-12-28 00:36:22 -08001762 arch_remove_memory(nid, start, size, NULL);
David Hildenbrand52219ae2020-06-04 16:48:38 -07001763
1764 if (IS_ENABLED(CONFIG_ARCH_KEEP_MEMBLOCK)) {
1765 memblock_free(start, size);
1766 memblock_remove(start, size);
1767 }
1768
David Hildenbrandcb8e3c82020-10-15 20:09:12 -07001769 release_mem_region_adjustable(start, size);
Wen Congyang24d335c2013-02-22 16:32:58 -08001770
Tang Chen60a5a192013-02-22 16:33:14 -08001771 try_offline_node(nid);
1772
Vladimir Davydovbfc8c902014-06-04 16:07:18 -07001773 mem_hotplug_done();
Jia Heb4223a52020-08-11 18:32:20 -07001774 return 0;
Badari Pulavarty71088782008-10-18 20:25:58 -07001775}
David Hildenbrandd15e5922018-10-30 15:10:18 -07001776
Pavel Tatashineca499a2019-07-16 16:30:31 -07001777/**
1778 * remove_memory
1779 * @nid: the node ID
1780 * @start: physical address of the region to remove
1781 * @size: size of the region to remove
1782 *
1783 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1784 * and online/offline operations before this call, as required by
1785 * try_offline_node().
1786 */
1787void __remove_memory(int nid, u64 start, u64 size)
David Hildenbrandd15e5922018-10-30 15:10:18 -07001788{
Pavel Tatashineca499a2019-07-16 16:30:31 -07001789
1790 /*
Souptick Joarder29a90db2019-09-23 15:36:18 -07001791 * trigger BUG() if some memory is not offlined prior to calling this
Pavel Tatashineca499a2019-07-16 16:30:31 -07001792 * function
1793 */
1794 if (try_remove_memory(nid, start, size))
1795 BUG();
1796}
1797
1798/*
1799 * Remove memory if every memory block is offline, otherwise return -EBUSY is
1800 * some memory is not offline
1801 */
1802int remove_memory(int nid, u64 start, u64 size)
1803{
1804 int rc;
1805
David Hildenbrandd15e5922018-10-30 15:10:18 -07001806 lock_device_hotplug();
Pavel Tatashineca499a2019-07-16 16:30:31 -07001807 rc = try_remove_memory(nid, start, size);
David Hildenbrandd15e5922018-10-30 15:10:18 -07001808 unlock_device_hotplug();
Pavel Tatashineca499a2019-07-16 16:30:31 -07001809
1810 return rc;
David Hildenbrandd15e5922018-10-30 15:10:18 -07001811}
Badari Pulavarty71088782008-10-18 20:25:58 -07001812EXPORT_SYMBOL_GPL(remove_memory);
David Hildenbrand08b3acd2020-05-07 16:01:32 +02001813
1814/*
1815 * Try to offline and remove a memory block. Might take a long time to
1816 * finish in case memory is still in use. Primarily useful for memory devices
1817 * that logically unplugged all memory (so it's no longer in use) and want to
1818 * offline + remove the memory block.
1819 */
1820int offline_and_remove_memory(int nid, u64 start, u64 size)
1821{
1822 struct memory_block *mem;
1823 int rc = -EINVAL;
1824
1825 if (!IS_ALIGNED(start, memory_block_size_bytes()) ||
1826 size != memory_block_size_bytes())
1827 return rc;
1828
1829 lock_device_hotplug();
1830 mem = find_memory_block(__pfn_to_section(PFN_DOWN(start)));
1831 if (mem)
1832 rc = device_offline(&mem->dev);
1833 /* Ignore if the device is already offline. */
1834 if (rc > 0)
1835 rc = 0;
1836
1837 /*
1838 * In case we succeeded to offline the memory block, remove it.
1839 * This cannot fail as it cannot get onlined in the meantime.
1840 */
1841 if (!rc) {
1842 rc = try_remove_memory(nid, start, size);
1843 WARN_ON_ONCE(rc);
1844 }
1845 unlock_device_hotplug();
1846
1847 return rc;
1848}
1849EXPORT_SYMBOL_GPL(offline_and_remove_memory);
Rafael J. Wysockiaba6efc2013-06-01 22:24:07 +02001850#endif /* CONFIG_MEMORY_HOTREMOVE */