blob: 6166ba5a15f3e533f66f3e6e1ef2be540cdb4c13 [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
Arun KSa9cd4102019-03-05 15:42:14 -080052static void generic_online_page(struct page *page, unsigned int order);
Daniel Kiper9d0ad8c2011-07-25 17:12:05 -070053
54static online_page_callback_t online_page_callback = generic_online_page;
Vladimir Davydovbfc8c902014-06-04 16:07:18 -070055static DEFINE_MUTEX(online_page_callback_lock);
Daniel Kiper9d0ad8c2011-07-25 17:12:05 -070056
Thomas Gleixner3f906ba2017-07-10 15:50:09 -070057DEFINE_STATIC_PERCPU_RWSEM(mem_hotplug_lock);
KOSAKI Motohiro20d6c962010-12-02 14:31:19 -080058
Thomas Gleixner3f906ba2017-07-10 15:50:09 -070059void get_online_mems(void)
60{
61 percpu_down_read(&mem_hotplug_lock);
62}
Vladimir Davydovbfc8c902014-06-04 16:07:18 -070063
Thomas Gleixner3f906ba2017-07-10 15:50:09 -070064void put_online_mems(void)
65{
66 percpu_up_read(&mem_hotplug_lock);
67}
Vladimir Davydovbfc8c902014-06-04 16:07:18 -070068
Michal Hocko49323812017-07-06 15:41:05 -070069bool movable_node_enabled = false;
70
Vitaly Kuznetsov8604d9e2016-05-19 17:13:03 -070071#ifndef CONFIG_MEMORY_HOTPLUG_DEFAULT_ONLINE
Vitaly Kuznetsov31bc3852016-03-15 14:56:48 -070072bool memhp_auto_online;
Vitaly Kuznetsov8604d9e2016-05-19 17:13:03 -070073#else
74bool memhp_auto_online = true;
75#endif
Vitaly Kuznetsov31bc3852016-03-15 14:56:48 -070076EXPORT_SYMBOL_GPL(memhp_auto_online);
77
Vitaly Kuznetsov86dd9952016-05-19 17:13:06 -070078static int __init setup_memhp_default_state(char *str)
79{
80 if (!strcmp(str, "online"))
81 memhp_auto_online = true;
82 else if (!strcmp(str, "offline"))
83 memhp_auto_online = false;
84
85 return 1;
86}
87__setup("memhp_default_state=", setup_memhp_default_state);
88
David Rientjes30467e02015-04-14 15:45:11 -070089void mem_hotplug_begin(void)
Vladimir Davydovbfc8c902014-06-04 16:07:18 -070090{
Thomas Gleixner3f906ba2017-07-10 15:50:09 -070091 cpus_read_lock();
92 percpu_down_write(&mem_hotplug_lock);
Vladimir Davydovbfc8c902014-06-04 16:07:18 -070093}
94
David Rientjes30467e02015-04-14 15:45:11 -070095void mem_hotplug_done(void)
Vladimir Davydovbfc8c902014-06-04 16:07:18 -070096{
Thomas Gleixner3f906ba2017-07-10 15:50:09 -070097 percpu_up_write(&mem_hotplug_lock);
98 cpus_read_unlock();
Vladimir Davydovbfc8c902014-06-04 16:07:18 -070099}
KOSAKI Motohiro20d6c962010-12-02 14:31:19 -0800100
Juergen Gross357b4da2019-02-14 11:42:39 +0100101u64 max_mem_size = U64_MAX;
102
Keith Mannthey45e0b782006-09-30 23:27:09 -0700103/* add this memory to iomem resource */
104static struct resource *register_memory_resource(u64 start, u64 size)
105{
Dave Hansen27941292019-02-25 10:57:36 -0800106 struct resource *res;
107 unsigned long flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
108 char *resource_name = "System RAM";
Juergen Gross357b4da2019-02-14 11:42:39 +0100109
110 if (start + size > max_mem_size)
111 return ERR_PTR(-E2BIG);
112
Dave Hansen27941292019-02-25 10:57:36 -0800113 /*
114 * Request ownership of the new memory range. This might be
115 * a child of an existing resource that was present but
116 * not marked as busy.
117 */
118 res = __request_region(&iomem_resource, start, size,
119 resource_name, flags);
Keith Mannthey45e0b782006-09-30 23:27:09 -0700120
Dave Hansen27941292019-02-25 10:57:36 -0800121 if (!res) {
122 pr_debug("Unable to reserve System RAM region: %016llx->%016llx\n",
123 start, start + size);
Vitaly Kuznetsov6f754ba2016-01-14 15:21:55 -0800124 return ERR_PTR(-EEXIST);
Keith Mannthey45e0b782006-09-30 23:27:09 -0700125 }
126 return res;
127}
128
129static void release_memory_resource(struct resource *res)
130{
131 if (!res)
132 return;
133 release_resource(res);
134 kfree(res);
135 return;
136}
137
Keith Mannthey53947022006-09-30 23:27:08 -0700138#ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
Yasuaki Ishimatsu46723bf2013-02-22 16:33:00 -0800139void get_page_bootmem(unsigned long info, struct page *page,
140 unsigned long type)
Yasunori Goto04753272008-04-28 02:13:31 -0700141{
Yasuaki Ishimatsuddffe982017-02-22 15:45:13 -0800142 page->freelist = (void *)type;
Yasunori Goto04753272008-04-28 02:13:31 -0700143 SetPagePrivate(page);
144 set_page_private(page, info);
Joonsoo Kimfe896d12016-03-17 14:19:26 -0700145 page_ref_inc(page);
Yasunori Goto04753272008-04-28 02:13:31 -0700146}
147
Jiang Liu170a5a72013-07-03 15:03:17 -0700148void put_page_bootmem(struct page *page)
Yasunori Goto04753272008-04-28 02:13:31 -0700149{
Andrea Arcangeli5f24ce52011-01-13 15:47:00 -0800150 unsigned long type;
Yasunori Goto04753272008-04-28 02:13:31 -0700151
Yasuaki Ishimatsuddffe982017-02-22 15:45:13 -0800152 type = (unsigned long) page->freelist;
Andrea Arcangeli5f24ce52011-01-13 15:47:00 -0800153 BUG_ON(type < MEMORY_HOTPLUG_MIN_BOOTMEM_TYPE ||
154 type > MEMORY_HOTPLUG_MAX_BOOTMEM_TYPE);
Yasunori Goto04753272008-04-28 02:13:31 -0700155
Joonsoo Kimfe896d12016-03-17 14:19:26 -0700156 if (page_ref_dec_return(page) == 1) {
Yasuaki Ishimatsuddffe982017-02-22 15:45:13 -0800157 page->freelist = NULL;
Yasunori Goto04753272008-04-28 02:13:31 -0700158 ClearPagePrivate(page);
159 set_page_private(page, 0);
Andrea Arcangeli5f24ce52011-01-13 15:47:00 -0800160 INIT_LIST_HEAD(&page->lru);
Jiang Liu170a5a72013-07-03 15:03:17 -0700161 free_reserved_page(page);
Yasunori Goto04753272008-04-28 02:13:31 -0700162 }
Yasunori Goto04753272008-04-28 02:13:31 -0700163}
164
Yasuaki Ishimatsu46723bf2013-02-22 16:33:00 -0800165#ifdef CONFIG_HAVE_BOOTMEM_INFO_NODE
166#ifndef CONFIG_SPARSEMEM_VMEMMAP
Adrian Bunkd92bc312008-07-23 21:28:12 -0700167static void register_page_bootmem_info_section(unsigned long start_pfn)
Yasunori Goto04753272008-04-28 02:13:31 -0700168{
169 unsigned long *usemap, mapsize, section_nr, i;
170 struct mem_section *ms;
171 struct page *page, *memmap;
172
Yasunori Goto04753272008-04-28 02:13:31 -0700173 section_nr = pfn_to_section_nr(start_pfn);
174 ms = __nr_to_section(section_nr);
175
176 /* Get section's memmap address */
177 memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
178
179 /*
180 * Get page for the memmap's phys address
181 * XXX: need more consideration for sparse_vmemmap...
182 */
183 page = virt_to_page(memmap);
184 mapsize = sizeof(struct page) * PAGES_PER_SECTION;
185 mapsize = PAGE_ALIGN(mapsize) >> PAGE_SHIFT;
186
187 /* remember memmap's page */
188 for (i = 0; i < mapsize; i++, page++)
189 get_page_bootmem(section_nr, page, SECTION_INFO);
190
Oscar Salvador9ac93222018-01-31 16:17:25 -0800191 usemap = ms->pageblock_flags;
Yasunori Goto04753272008-04-28 02:13:31 -0700192 page = virt_to_page(usemap);
193
194 mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT;
195
196 for (i = 0; i < mapsize; i++, page++)
Yasunori Gotoaf370fb2008-07-23 21:28:17 -0700197 get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
Yasunori Goto04753272008-04-28 02:13:31 -0700198
199}
Yasuaki Ishimatsu46723bf2013-02-22 16:33:00 -0800200#else /* CONFIG_SPARSEMEM_VMEMMAP */
201static void register_page_bootmem_info_section(unsigned long start_pfn)
202{
203 unsigned long *usemap, mapsize, section_nr, i;
204 struct mem_section *ms;
205 struct page *page, *memmap;
206
Yasuaki Ishimatsu46723bf2013-02-22 16:33:00 -0800207 section_nr = pfn_to_section_nr(start_pfn);
208 ms = __nr_to_section(section_nr);
209
210 memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
211
212 register_page_bootmem_memmap(section_nr, memmap, PAGES_PER_SECTION);
213
Oscar Salvador9ac93222018-01-31 16:17:25 -0800214 usemap = ms->pageblock_flags;
Yasuaki Ishimatsu46723bf2013-02-22 16:33:00 -0800215 page = virt_to_page(usemap);
216
217 mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT;
218
219 for (i = 0; i < mapsize; i++, page++)
220 get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
221}
222#endif /* !CONFIG_SPARSEMEM_VMEMMAP */
Yasunori Goto04753272008-04-28 02:13:31 -0700223
Linus Torvalds7ded3842016-05-27 15:23:32 -0700224void __init register_page_bootmem_info_node(struct pglist_data *pgdat)
Yasunori Goto04753272008-04-28 02:13:31 -0700225{
226 unsigned long i, pfn, end_pfn, nr_pages;
227 int node = pgdat->node_id;
228 struct page *page;
Yasunori Goto04753272008-04-28 02:13:31 -0700229
230 nr_pages = PAGE_ALIGN(sizeof(struct pglist_data)) >> PAGE_SHIFT;
231 page = virt_to_page(pgdat);
232
233 for (i = 0; i < nr_pages; i++, page++)
234 get_page_bootmem(node, page, NODE_INFO);
235
Yasunori Goto04753272008-04-28 02:13:31 -0700236 pfn = pgdat->node_start_pfn;
Cody P Schaferc1f19492013-02-22 16:35:32 -0800237 end_pfn = pgdat_end_pfn(pgdat);
Yasunori Goto04753272008-04-28 02:13:31 -0700238
Tang Chen7e9f5eb2013-07-08 16:00:23 -0700239 /* register section info */
qiuxishif14851a2012-09-17 14:09:24 -0700240 for (; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
241 /*
242 * Some platforms can assign the same pfn to multiple nodes - on
243 * node0 as well as nodeN. To avoid registering a pfn against
244 * multiple nodes we check that this pfn does not already
Tang Chen7e9f5eb2013-07-08 16:00:23 -0700245 * reside in some other nodes.
qiuxishif14851a2012-09-17 14:09:24 -0700246 */
Yang Shif65e91df2016-05-27 14:27:32 -0700247 if (pfn_valid(pfn) && (early_pfn_to_nid(pfn) == node))
qiuxishif14851a2012-09-17 14:09:24 -0700248 register_page_bootmem_info_section(pfn);
249 }
Yasunori Goto04753272008-04-28 02:13:31 -0700250}
Yasuaki Ishimatsu46723bf2013-02-22 16:33:00 -0800251#endif /* CONFIG_HAVE_BOOTMEM_INFO_NODE */
Yasunori Goto04753272008-04-28 02:13:31 -0700252
Michal Hockof1dd2cd2017-07-06 15:38:11 -0700253static int __meminit __add_section(int nid, unsigned long phys_start_pfn,
Christoph Hellwig7b73d972017-12-29 08:53:54 +0100254 struct vmem_altmap *altmap, bool want_memblock)
Dave Hansen3947be12005-10-29 18:16:54 -0700255{
Dave Hansen3947be12005-10-29 18:16:54 -0700256 int ret;
257
KAMEZAWA Hiroyukiebd15302006-08-05 12:15:06 -0700258 if (pfn_valid(phys_start_pfn))
259 return -EEXIST;
260
Wei Yang4e0d2e72018-12-28 00:37:06 -0800261 ret = sparse_add_one_section(nid, phys_start_pfn, altmap);
Dave Hansen3947be12005-10-29 18:16:54 -0700262 if (ret < 0)
263 return ret;
264
Michal Hocko1b862ae2017-07-06 15:37:45 -0700265 if (!want_memblock)
266 return 0;
267
Pavel Tatashinfc44f7f2018-04-05 16:22:56 -0700268 return hotplug_memory_register(nid, __pfn_to_section(phys_start_pfn));
Dave Hansen3947be12005-10-29 18:16:54 -0700269}
270
David Rientjes4edd7ce2013-04-29 15:08:22 -0700271/*
272 * Reasonably generic function for adding memory. It is
273 * expected that archs that support memory hotplug will
274 * call this function after deciding the zone to which to
275 * add the new pages.
276 */
Michal Hockof1dd2cd2017-07-06 15:38:11 -0700277int __ref __add_pages(int nid, unsigned long phys_start_pfn,
Michal Hocko940519f2019-05-13 17:21:26 -0700278 unsigned long nr_pages, struct mhp_restrictions *restrictions)
David Rientjes4edd7ce2013-04-29 15:08:22 -0700279{
280 unsigned long i;
281 int err = 0;
282 int start_sec, end_sec;
Michal Hocko940519f2019-05-13 17:21:26 -0700283 struct vmem_altmap *altmap = restrictions->altmap;
Dan Williams4b94ffd2016-01-15 16:56:22 -0800284
David Rientjes4edd7ce2013-04-29 15:08:22 -0700285 /* during initialize mem_map, align hot-added range to section */
286 start_sec = pfn_to_section_nr(phys_start_pfn);
287 end_sec = pfn_to_section_nr(phys_start_pfn + nr_pages - 1);
288
Dan Williams4b94ffd2016-01-15 16:56:22 -0800289 if (altmap) {
290 /*
291 * Validate altmap is within bounds of the total request
292 */
293 if (altmap->base_pfn != phys_start_pfn
294 || vmem_altmap_offset(altmap) > nr_pages) {
295 pr_warn_once("memory add fail, invalid altmap\n");
Joonsoo Kim7cf91a92016-03-15 14:57:51 -0700296 err = -EINVAL;
297 goto out;
Dan Williams4b94ffd2016-01-15 16:56:22 -0800298 }
299 altmap->alloc = 0;
300 }
301
David Rientjes4edd7ce2013-04-29 15:08:22 -0700302 for (i = start_sec; i <= end_sec; i++) {
Christoph Hellwig7b73d972017-12-29 08:53:54 +0100303 err = __add_section(nid, section_nr_to_pfn(i), altmap,
Michal Hocko940519f2019-05-13 17:21:26 -0700304 restrictions->flags & MHP_MEMBLOCK_API);
David Rientjes4edd7ce2013-04-29 15:08:22 -0700305
306 /*
307 * EEXIST is finally dealt with by ioresource collision
308 * check. see add_memory() => register_memory_resource()
309 * Warning will be printed if there is collision.
310 */
311 if (err && (err != -EEXIST))
312 break;
313 err = 0;
Michal Hockof64ac5e2017-10-03 16:16:16 -0700314 cond_resched();
David Rientjes4edd7ce2013-04-29 15:08:22 -0700315 }
Zhu Guihuac435a392015-06-24 16:58:42 -0700316 vmemmap_populate_print_last();
Joonsoo Kim7cf91a92016-03-15 14:57:51 -0700317out:
David Rientjes4edd7ce2013-04-29 15:08:22 -0700318 return err;
319}
David Rientjes4edd7ce2013-04-29 15:08:22 -0700320
321#ifdef CONFIG_MEMORY_HOTREMOVE
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800322/* find the smallest valid pfn in the range [start_pfn, end_pfn) */
YASUAKI ISHIMATSUd09b0132017-10-03 16:16:32 -0700323static unsigned long find_smallest_section_pfn(int nid, struct zone *zone,
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800324 unsigned long start_pfn,
325 unsigned long end_pfn)
326{
327 struct mem_section *ms;
328
329 for (; start_pfn < end_pfn; start_pfn += PAGES_PER_SECTION) {
330 ms = __pfn_to_section(start_pfn);
331
332 if (unlikely(!valid_section(ms)))
333 continue;
334
335 if (unlikely(pfn_to_nid(start_pfn) != nid))
336 continue;
337
338 if (zone && zone != page_zone(pfn_to_page(start_pfn)))
339 continue;
340
341 return start_pfn;
342 }
343
344 return 0;
345}
346
347/* find the biggest valid pfn in the range [start_pfn, end_pfn). */
YASUAKI ISHIMATSUd09b0132017-10-03 16:16:32 -0700348static unsigned long find_biggest_section_pfn(int nid, struct zone *zone,
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800349 unsigned long start_pfn,
350 unsigned long end_pfn)
351{
352 struct mem_section *ms;
353 unsigned long pfn;
354
355 /* pfn is the end pfn of a memory section. */
356 pfn = end_pfn - 1;
357 for (; pfn >= start_pfn; pfn -= PAGES_PER_SECTION) {
358 ms = __pfn_to_section(pfn);
359
360 if (unlikely(!valid_section(ms)))
361 continue;
362
363 if (unlikely(pfn_to_nid(pfn) != nid))
364 continue;
365
366 if (zone && zone != page_zone(pfn_to_page(pfn)))
367 continue;
368
369 return pfn;
370 }
371
372 return 0;
373}
374
375static void shrink_zone_span(struct zone *zone, unsigned long start_pfn,
376 unsigned long end_pfn)
377{
Xishi Qiuc33bc312013-09-11 14:21:44 -0700378 unsigned long zone_start_pfn = zone->zone_start_pfn;
379 unsigned long z = zone_end_pfn(zone); /* zone_end_pfn namespace clash */
380 unsigned long zone_end_pfn = z;
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800381 unsigned long pfn;
382 struct mem_section *ms;
383 int nid = zone_to_nid(zone);
384
385 zone_span_writelock(zone);
386 if (zone_start_pfn == start_pfn) {
387 /*
388 * If the section is smallest section in the zone, it need
389 * shrink zone->zone_start_pfn and zone->zone_spanned_pages.
390 * In this case, we find second smallest valid mem_section
391 * for shrinking zone.
392 */
393 pfn = find_smallest_section_pfn(nid, zone, end_pfn,
394 zone_end_pfn);
395 if (pfn) {
396 zone->zone_start_pfn = pfn;
397 zone->spanned_pages = zone_end_pfn - pfn;
398 }
399 } else if (zone_end_pfn == end_pfn) {
400 /*
401 * If the section is biggest section in the zone, it need
402 * shrink zone->spanned_pages.
403 * In this case, we find second biggest valid mem_section for
404 * shrinking zone.
405 */
406 pfn = find_biggest_section_pfn(nid, zone, zone_start_pfn,
407 start_pfn);
408 if (pfn)
409 zone->spanned_pages = pfn - zone_start_pfn + 1;
410 }
411
412 /*
413 * The section is not biggest or smallest mem_section in the zone, it
414 * only creates a hole in the zone. So in this case, we need not
415 * change the zone. But perhaps, the zone has only hole data. Thus
416 * it check the zone has only hole or not.
417 */
418 pfn = zone_start_pfn;
419 for (; pfn < zone_end_pfn; pfn += PAGES_PER_SECTION) {
420 ms = __pfn_to_section(pfn);
421
422 if (unlikely(!valid_section(ms)))
423 continue;
424
425 if (page_zone(pfn_to_page(pfn)) != zone)
426 continue;
427
428 /* If the section is current section, it continues the loop */
429 if (start_pfn == pfn)
430 continue;
431
432 /* If we find valid section, we have nothing to do */
433 zone_span_writeunlock(zone);
434 return;
435 }
436
437 /* The zone has no valid section */
438 zone->zone_start_pfn = 0;
439 zone->spanned_pages = 0;
440 zone_span_writeunlock(zone);
441}
442
443static void shrink_pgdat_span(struct pglist_data *pgdat,
444 unsigned long start_pfn, unsigned long end_pfn)
445{
Xishi Qiu83285c72013-11-12 15:07:19 -0800446 unsigned long pgdat_start_pfn = pgdat->node_start_pfn;
447 unsigned long p = pgdat_end_pfn(pgdat); /* pgdat_end_pfn namespace clash */
448 unsigned long pgdat_end_pfn = p;
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800449 unsigned long pfn;
450 struct mem_section *ms;
451 int nid = pgdat->node_id;
452
453 if (pgdat_start_pfn == start_pfn) {
454 /*
455 * If the section is smallest section in the pgdat, it need
456 * shrink pgdat->node_start_pfn and pgdat->node_spanned_pages.
457 * In this case, we find second smallest valid mem_section
458 * for shrinking zone.
459 */
460 pfn = find_smallest_section_pfn(nid, NULL, end_pfn,
461 pgdat_end_pfn);
462 if (pfn) {
463 pgdat->node_start_pfn = pfn;
464 pgdat->node_spanned_pages = pgdat_end_pfn - pfn;
465 }
466 } else if (pgdat_end_pfn == end_pfn) {
467 /*
468 * If the section is biggest section in the pgdat, it need
469 * shrink pgdat->node_spanned_pages.
470 * In this case, we find second biggest valid mem_section for
471 * shrinking zone.
472 */
473 pfn = find_biggest_section_pfn(nid, NULL, pgdat_start_pfn,
474 start_pfn);
475 if (pfn)
476 pgdat->node_spanned_pages = pfn - pgdat_start_pfn + 1;
477 }
478
479 /*
480 * If the section is not biggest or smallest mem_section in the pgdat,
481 * it only creates a hole in the pgdat. So in this case, we need not
482 * change the pgdat.
483 * But perhaps, the pgdat has only hole data. Thus it check the pgdat
484 * has only hole or not.
485 */
486 pfn = pgdat_start_pfn;
487 for (; pfn < pgdat_end_pfn; pfn += PAGES_PER_SECTION) {
488 ms = __pfn_to_section(pfn);
489
490 if (unlikely(!valid_section(ms)))
491 continue;
492
493 if (pfn_to_nid(pfn) != nid)
494 continue;
495
496 /* If the section is current section, it continues the loop */
497 if (start_pfn == pfn)
498 continue;
499
500 /* If we find valid section, we have nothing to do */
501 return;
502 }
503
504 /* The pgdat has no valid section */
505 pgdat->node_start_pfn = 0;
506 pgdat->node_spanned_pages = 0;
507}
508
509static void __remove_zone(struct zone *zone, unsigned long start_pfn)
510{
511 struct pglist_data *pgdat = zone->zone_pgdat;
512 int nr_pages = PAGES_PER_SECTION;
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800513 unsigned long flags;
514
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800515 pgdat_resize_lock(zone->zone_pgdat, &flags);
516 shrink_zone_span(zone, start_pfn, start_pfn + nr_pages);
517 shrink_pgdat_span(pgdat, start_pfn, start_pfn + nr_pages);
518 pgdat_resize_unlock(zone->zone_pgdat, &flags);
519}
520
David Hildenbrand9d1d8872019-05-13 17:21:41 -0700521static void __remove_section(struct zone *zone, struct mem_section *ms,
522 unsigned long map_offset,
523 struct vmem_altmap *altmap)
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700524{
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800525 unsigned long start_pfn;
526 int scn_nr;
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700527
David Hildenbrand9d1d8872019-05-13 17:21:41 -0700528 if (WARN_ON_ONCE(!valid_section(ms)))
529 return;
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700530
David Hildenbrandcb7b3a32019-05-13 17:21:37 -0700531 unregister_memory_section(ms);
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700532
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800533 scn_nr = __section_nr(ms);
YASUAKI ISHIMATSU1dd2bfc2017-10-03 16:16:29 -0700534 start_pfn = section_nr_to_pfn((unsigned long)scn_nr);
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800535 __remove_zone(zone, start_pfn);
536
Christoph Hellwig24b6d412017-12-29 08:53:56 +0100537 sparse_remove_one_section(zone, ms, map_offset, altmap);
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700538}
539
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700540/**
541 * __remove_pages() - remove sections of pages from a zone
542 * @zone: zone from which pages need to be removed
543 * @phys_start_pfn: starting pageframe (must be aligned to start of a section)
544 * @nr_pages: number of pages to remove (must be multiple of section size)
Mike Rapoporte8b098f2018-04-05 16:24:57 -0700545 * @altmap: alternative device page map or %NULL if default memmap is used
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700546 *
547 * Generic helper function to remove section mappings and sysfs entries
548 * for the section of the memory we are removing. Caller needs to make
549 * sure that pages are marked reserved and zones are adjust properly by
550 * calling offline_pages().
551 */
David Hildenbrandac5c9422019-05-13 17:21:46 -0700552void __remove_pages(struct zone *zone, unsigned long phys_start_pfn,
553 unsigned long nr_pages, struct vmem_altmap *altmap)
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700554{
Toshi Kanife74ebb2013-04-29 15:08:20 -0700555 unsigned long i;
Dan Williams4b94ffd2016-01-15 16:56:22 -0800556 unsigned long map_offset = 0;
David Hildenbrand9d1d8872019-05-13 17:21:41 -0700557 int sections_to_remove;
Dan Williams4b94ffd2016-01-15 16:56:22 -0800558
559 /* In the ZONE_DEVICE case device driver owns the memory region */
Christoph Hellwig514caf22019-06-26 14:27:13 +0200560 if (is_dev_zone(zone))
561 map_offset = vmem_altmap_offset(altmap);
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700562
Joonsoo Kim7cf91a92016-03-15 14:57:51 -0700563 clear_zone_contiguous(zone);
564
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700565 /*
566 * We can only remove entire sections
567 */
568 BUG_ON(phys_start_pfn & ~PAGE_SECTION_MASK);
569 BUG_ON(nr_pages % PAGES_PER_SECTION);
570
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700571 sections_to_remove = nr_pages / PAGES_PER_SECTION;
572 for (i = 0; i < sections_to_remove; i++) {
573 unsigned long pfn = phys_start_pfn + i*PAGES_PER_SECTION;
Dan Williams4b94ffd2016-01-15 16:56:22 -0800574
Michal Hockodd33ad72018-11-02 15:48:46 -0700575 cond_resched();
David Hildenbrand9d1d8872019-05-13 17:21:41 -0700576 __remove_section(zone, __pfn_to_section(pfn), map_offset,
577 altmap);
Dan Williams4b94ffd2016-01-15 16:56:22 -0800578 map_offset = 0;
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700579 }
Joonsoo Kim7cf91a92016-03-15 14:57:51 -0700580
581 set_zone_contiguous(zone);
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700582}
David Rientjes4edd7ce2013-04-29 15:08:22 -0700583#endif /* CONFIG_MEMORY_HOTREMOVE */
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700584
Daniel Kiper9d0ad8c2011-07-25 17:12:05 -0700585int set_online_page_callback(online_page_callback_t callback)
586{
587 int rc = -EINVAL;
588
Vladimir Davydovbfc8c902014-06-04 16:07:18 -0700589 get_online_mems();
590 mutex_lock(&online_page_callback_lock);
Daniel Kiper9d0ad8c2011-07-25 17:12:05 -0700591
592 if (online_page_callback == generic_online_page) {
593 online_page_callback = callback;
594 rc = 0;
595 }
596
Vladimir Davydovbfc8c902014-06-04 16:07:18 -0700597 mutex_unlock(&online_page_callback_lock);
598 put_online_mems();
Daniel Kiper9d0ad8c2011-07-25 17:12:05 -0700599
600 return rc;
601}
602EXPORT_SYMBOL_GPL(set_online_page_callback);
603
604int restore_online_page_callback(online_page_callback_t callback)
605{
606 int rc = -EINVAL;
607
Vladimir Davydovbfc8c902014-06-04 16:07:18 -0700608 get_online_mems();
609 mutex_lock(&online_page_callback_lock);
Daniel Kiper9d0ad8c2011-07-25 17:12:05 -0700610
611 if (online_page_callback == callback) {
612 online_page_callback = generic_online_page;
613 rc = 0;
614 }
615
Vladimir Davydovbfc8c902014-06-04 16:07:18 -0700616 mutex_unlock(&online_page_callback_lock);
617 put_online_mems();
Daniel Kiper9d0ad8c2011-07-25 17:12:05 -0700618
619 return rc;
620}
621EXPORT_SYMBOL_GPL(restore_online_page_callback);
622
623void __online_page_set_limits(struct page *page)
Jeremy Fitzhardinge180c06e2008-04-28 02:12:03 -0700624{
Daniel Kiper9d0ad8c2011-07-25 17:12:05 -0700625}
626EXPORT_SYMBOL_GPL(__online_page_set_limits);
627
628void __online_page_increment_counters(struct page *page)
629{
Jiang Liu3dcc0572013-07-03 15:03:21 -0700630 adjust_managed_page_count(page, 1);
Daniel Kiper9d0ad8c2011-07-25 17:12:05 -0700631}
632EXPORT_SYMBOL_GPL(__online_page_increment_counters);
Jeremy Fitzhardinge180c06e2008-04-28 02:12:03 -0700633
Daniel Kiper9d0ad8c2011-07-25 17:12:05 -0700634void __online_page_free(struct page *page)
635{
Jiang Liu3dcc0572013-07-03 15:03:21 -0700636 __free_reserved_page(page);
Jeremy Fitzhardinge180c06e2008-04-28 02:12:03 -0700637}
Daniel Kiper9d0ad8c2011-07-25 17:12:05 -0700638EXPORT_SYMBOL_GPL(__online_page_free);
639
Arun KSa9cd4102019-03-05 15:42:14 -0800640static void generic_online_page(struct page *page, unsigned int order)
Daniel Kiper9d0ad8c2011-07-25 17:12:05 -0700641{
Qian Caicd02cf12019-03-05 15:49:57 -0800642 kernel_map_pages(page, 1 << order, 1);
Arun KSa9cd4102019-03-05 15:42:14 -0800643 __free_pages_core(page, order);
644 totalram_pages_add(1UL << order);
645#ifdef CONFIG_HIGHMEM
646 if (PageHighMem(page))
647 totalhigh_pages_add(1UL << order);
648#endif
649}
650
651static int online_pages_blocks(unsigned long start, unsigned long nr_pages)
652{
653 unsigned long end = start + nr_pages;
654 int order, onlined_pages = 0;
655
656 while (start < end) {
657 order = min(MAX_ORDER - 1,
658 get_order(PFN_PHYS(end) - PFN_PHYS(start)));
659 (*online_page_callback)(pfn_to_page(start), order);
660
661 onlined_pages += (1UL << order);
662 start += (1UL << order);
663 }
664 return onlined_pages;
Daniel Kiper9d0ad8c2011-07-25 17:12:05 -0700665}
Jeremy Fitzhardinge180c06e2008-04-28 02:12:03 -0700666
KAMEZAWA Hiroyuki75884fb2007-10-16 01:26:10 -0700667static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages,
668 void *arg)
Dave Hansen3947be12005-10-29 18:16:54 -0700669{
KAMEZAWA Hiroyuki75884fb2007-10-16 01:26:10 -0700670 unsigned long onlined_pages = *(unsigned long *)arg;
Michal Hocko2d070ea2017-07-06 15:37:56 -0700671
KAMEZAWA Hiroyuki75884fb2007-10-16 01:26:10 -0700672 if (PageReserved(pfn_to_page(start_pfn)))
Arun KSa9cd4102019-03-05 15:42:14 -0800673 onlined_pages += online_pages_blocks(start_pfn, nr_pages);
Michal Hocko2d070ea2017-07-06 15:37:56 -0700674
675 online_mem_sections(start_pfn, start_pfn + nr_pages);
676
KAMEZAWA Hiroyuki75884fb2007-10-16 01:26:10 -0700677 *(unsigned long *)arg = onlined_pages;
678 return 0;
679}
680
Lai Jiangshand9713672012-12-11 16:01:03 -0800681/* check which state of node_states will be changed when online memory */
682static void node_states_check_changes_online(unsigned long nr_pages,
683 struct zone *zone, struct memory_notify *arg)
684{
685 int nid = zone_to_nid(zone);
Lai Jiangshand9713672012-12-11 16:01:03 -0800686
Anshuman Khandual98fa15f2019-03-05 15:42:58 -0800687 arg->status_change_nid = NUMA_NO_NODE;
688 arg->status_change_nid_normal = NUMA_NO_NODE;
689 arg->status_change_nid_high = NUMA_NO_NODE;
Lai Jiangshand9713672012-12-11 16:01:03 -0800690
Lai Jiangshan6715ddf2012-12-12 13:51:49 -0800691 if (!node_state(nid, N_MEMORY))
Lai Jiangshand9713672012-12-11 16:01:03 -0800692 arg->status_change_nid = nid;
Oscar Salvador8efe33f2018-10-26 15:07:34 -0700693 if (zone_idx(zone) <= ZONE_NORMAL && !node_state(nid, N_NORMAL_MEMORY))
694 arg->status_change_nid_normal = nid;
695#ifdef CONFIG_HIGHMEM
Baoquan Hed3ba3ae2019-05-13 17:17:35 -0700696 if (zone_idx(zone) <= ZONE_HIGHMEM && !node_state(nid, N_HIGH_MEMORY))
Oscar Salvador8efe33f2018-10-26 15:07:34 -0700697 arg->status_change_nid_high = nid;
698#endif
Lai Jiangshand9713672012-12-11 16:01:03 -0800699}
700
701static void node_states_set_node(int node, struct memory_notify *arg)
702{
703 if (arg->status_change_nid_normal >= 0)
704 node_set_state(node, N_NORMAL_MEMORY);
705
Lai Jiangshan6715ddf2012-12-12 13:51:49 -0800706 if (arg->status_change_nid_high >= 0)
707 node_set_state(node, N_HIGH_MEMORY);
708
Oscar Salvador83d83612018-10-26 15:07:25 -0700709 if (arg->status_change_nid >= 0)
710 node_set_state(node, N_MEMORY);
Lai Jiangshand9713672012-12-11 16:01:03 -0800711}
712
Michal Hockof1dd2cd2017-07-06 15:38:11 -0700713static void __meminit resize_zone_range(struct zone *zone, unsigned long start_pfn,
714 unsigned long nr_pages)
715{
716 unsigned long old_end_pfn = zone_end_pfn(zone);
717
718 if (zone_is_empty(zone) || start_pfn < zone->zone_start_pfn)
719 zone->zone_start_pfn = start_pfn;
720
721 zone->spanned_pages = max(start_pfn + nr_pages, old_end_pfn) - zone->zone_start_pfn;
722}
723
724static void __meminit resize_pgdat_range(struct pglist_data *pgdat, unsigned long start_pfn,
725 unsigned long nr_pages)
726{
727 unsigned long old_end_pfn = pgdat_end_pfn(pgdat);
728
729 if (!pgdat->node_spanned_pages || start_pfn < pgdat->node_start_pfn)
730 pgdat->node_start_pfn = start_pfn;
731
732 pgdat->node_spanned_pages = max(start_pfn + nr_pages, old_end_pfn) - pgdat->node_start_pfn;
733}
734
Christoph Hellwiga99583e2017-12-29 08:53:57 +0100735void __ref move_pfn_range_to_zone(struct zone *zone, unsigned long start_pfn,
736 unsigned long nr_pages, struct vmem_altmap *altmap)
Michal Hockof1dd2cd2017-07-06 15:38:11 -0700737{
738 struct pglist_data *pgdat = zone->zone_pgdat;
739 int nid = pgdat->node_id;
740 unsigned long flags;
741
Michal Hockof1dd2cd2017-07-06 15:38:11 -0700742 clear_zone_contiguous(zone);
743
744 /* TODO Huh pgdat is irqsave while zone is not. It used to be like that before */
745 pgdat_resize_lock(pgdat, &flags);
746 zone_span_writelock(zone);
Wei Yangfa004ab2018-12-28 00:37:10 -0800747 if (zone_is_empty(zone))
748 init_currently_empty_zone(zone, start_pfn, nr_pages);
Michal Hockof1dd2cd2017-07-06 15:38:11 -0700749 resize_zone_range(zone, start_pfn, nr_pages);
750 zone_span_writeunlock(zone);
751 resize_pgdat_range(pgdat, start_pfn, nr_pages);
752 pgdat_resize_unlock(pgdat, &flags);
753
754 /*
755 * TODO now we have a visible range of pages which are not associated
756 * with their zone properly. Not nice but set_pfnblock_flags_mask
757 * expects the zone spans the pfn range. All the pages in the range
758 * are reserved so nobody should be touching them so we should be safe
759 */
Christoph Hellwiga99583e2017-12-29 08:53:57 +0100760 memmap_init_zone(nr_pages, nid, zone_idx(zone), start_pfn,
761 MEMMAP_HOTPLUG, altmap);
Michal Hockof1dd2cd2017-07-06 15:38:11 -0700762
763 set_zone_contiguous(zone);
764}
765
766/*
Michal Hockoc246a212017-07-06 15:38:18 -0700767 * Returns a default kernel memory zone for the given pfn range.
768 * If no kernel zone covers this pfn range it will automatically go
769 * to the ZONE_NORMAL.
770 */
Michal Hockoc6f03e22017-09-06 16:19:40 -0700771static struct zone *default_kernel_zone_for_pfn(int nid, unsigned long start_pfn,
Michal Hockoc246a212017-07-06 15:38:18 -0700772 unsigned long nr_pages)
773{
774 struct pglist_data *pgdat = NODE_DATA(nid);
775 int zid;
776
777 for (zid = 0; zid <= ZONE_NORMAL; zid++) {
778 struct zone *zone = &pgdat->node_zones[zid];
779
780 if (zone_intersects(zone, start_pfn, nr_pages))
781 return zone;
782 }
783
784 return &pgdat->node_zones[ZONE_NORMAL];
785}
786
Michal Hockoc6f03e22017-09-06 16:19:40 -0700787static inline struct zone *default_zone_for_pfn(int nid, unsigned long start_pfn,
788 unsigned long nr_pages)
Michal Hockoe5e68932017-09-06 16:19:37 -0700789{
Michal Hockoc6f03e22017-09-06 16:19:40 -0700790 struct zone *kernel_zone = default_kernel_zone_for_pfn(nid, start_pfn,
791 nr_pages);
792 struct zone *movable_zone = &NODE_DATA(nid)->node_zones[ZONE_MOVABLE];
793 bool in_kernel = zone_intersects(kernel_zone, start_pfn, nr_pages);
794 bool in_movable = zone_intersects(movable_zone, start_pfn, nr_pages);
Michal Hockoe5e68932017-09-06 16:19:37 -0700795
796 /*
Michal Hockoc6f03e22017-09-06 16:19:40 -0700797 * We inherit the existing zone in a simple case where zones do not
798 * overlap in the given range
Michal Hockoe5e68932017-09-06 16:19:37 -0700799 */
Michal Hockoc6f03e22017-09-06 16:19:40 -0700800 if (in_kernel ^ in_movable)
801 return (in_kernel) ? kernel_zone : movable_zone;
Michal Hockoe5e68932017-09-06 16:19:37 -0700802
Michal Hockoc6f03e22017-09-06 16:19:40 -0700803 /*
804 * If the range doesn't belong to any zone or two zones overlap in the
805 * given range then we use movable zone only if movable_node is
806 * enabled because we always online to a kernel zone by default.
807 */
808 return movable_node_enabled ? movable_zone : kernel_zone;
Michal Hocko9f123ab2017-07-10 15:48:37 -0700809}
810
Michal Hockoe5e68932017-09-06 16:19:37 -0700811struct zone * zone_for_pfn_range(int online_type, int nid, unsigned start_pfn,
812 unsigned long nr_pages)
Michal Hockof1dd2cd2017-07-06 15:38:11 -0700813{
Michal Hockoc6f03e22017-09-06 16:19:40 -0700814 if (online_type == MMOP_ONLINE_KERNEL)
815 return default_kernel_zone_for_pfn(nid, start_pfn, nr_pages);
Michal Hockof1dd2cd2017-07-06 15:38:11 -0700816
Michal Hockoc6f03e22017-09-06 16:19:40 -0700817 if (online_type == MMOP_ONLINE_MOVABLE)
818 return &NODE_DATA(nid)->node_zones[ZONE_MOVABLE];
Reza Arbabdf429ac2016-07-26 15:22:23 -0700819
Michal Hockoc6f03e22017-09-06 16:19:40 -0700820 return default_zone_for_pfn(nid, start_pfn, nr_pages);
Michal Hockoe5e68932017-09-06 16:19:37 -0700821}
822
823/*
824 * Associates the given pfn range with the given node and the zone appropriate
825 * for the given online type.
826 */
827static struct zone * __meminit move_pfn_range(int online_type, int nid,
828 unsigned long start_pfn, unsigned long nr_pages)
829{
830 struct zone *zone;
831
832 zone = zone_for_pfn_range(online_type, nid, start_pfn, nr_pages);
Christoph Hellwiga99583e2017-12-29 08:53:57 +0100833 move_pfn_range_to_zone(zone, start_pfn, nr_pages, NULL);
Michal Hockof1dd2cd2017-07-06 15:38:11 -0700834 return zone;
Reza Arbabdf429ac2016-07-26 15:22:23 -0700835}
KAMEZAWA Hiroyuki75884fb2007-10-16 01:26:10 -0700836
Lai Jiangshan511c2ab2012-12-11 16:03:16 -0800837int __ref online_pages(unsigned long pfn, unsigned long nr_pages, int online_type)
KAMEZAWA Hiroyuki75884fb2007-10-16 01:26:10 -0700838{
Cody P Schaferaa472282013-07-03 15:02:10 -0700839 unsigned long flags;
Dave Hansen3947be12005-10-29 18:16:54 -0700840 unsigned long onlined_pages = 0;
841 struct zone *zone;
Yasunori Goto68113782006-06-23 02:03:11 -0700842 int need_zonelists_rebuild = 0;
Yasunori Goto7b78d332007-10-21 16:41:36 -0700843 int nid;
844 int ret;
845 struct memory_notify arg;
Pavel Tatashind0dc12e2018-04-05 16:23:00 -0700846 struct memory_block *mem;
Dave Hansen3947be12005-10-29 18:16:54 -0700847
David Hildenbrand381eab42018-10-30 15:10:29 -0700848 mem_hotplug_begin();
849
Pavel Tatashind0dc12e2018-04-05 16:23:00 -0700850 /*
851 * We can't use pfn_to_nid() because nid might be stored in struct page
852 * which is not yet initialized. Instead, we find nid from memory block.
853 */
854 mem = find_memory_block(__pfn_to_section(pfn));
855 nid = mem->nid;
David Hildenbrand89c02e62019-04-25 22:23:37 -0700856 put_device(&mem->dev);
Pavel Tatashind0dc12e2018-04-05 16:23:00 -0700857
Michal Hockof1dd2cd2017-07-06 15:38:11 -0700858 /* associate pfn range with the zone */
859 zone = move_pfn_range(online_type, nid, pfn, nr_pages);
860
Yasunori Goto7b78d332007-10-21 16:41:36 -0700861 arg.start_pfn = pfn;
862 arg.nr_pages = nr_pages;
Lai Jiangshand9713672012-12-11 16:01:03 -0800863 node_states_check_changes_online(nr_pages, zone, &arg);
Yasunori Goto7b78d332007-10-21 16:41:36 -0700864
Yasunori Goto7b78d332007-10-21 16:41:36 -0700865 ret = memory_notify(MEM_GOING_ONLINE, &arg);
866 ret = notifier_to_errno(ret);
Chen Yuconge33e33b2016-03-17 14:19:35 -0700867 if (ret)
868 goto failed_addition;
869
Dave Hansen3947be12005-10-29 18:16:54 -0700870 /*
Yasunori Goto68113782006-06-23 02:03:11 -0700871 * If this zone is not populated, then it is not in zonelist.
872 * This means the page allocator ignores this zone.
873 * So, zonelist must be updated after online.
874 */
Wen Congyang6dcd73d2012-12-11 16:01:01 -0800875 if (!populated_zone(zone)) {
Yasunori Goto68113782006-06-23 02:03:11 -0700876 need_zonelists_rebuild = 1;
Michal Hocko72675e12017-09-06 16:20:24 -0700877 setup_zone_pageset(zone);
Wen Congyang6dcd73d2012-12-11 16:01:01 -0800878 }
Yasunori Goto68113782006-06-23 02:03:11 -0700879
KAMEZAWA Hiroyuki908eedc2009-09-22 16:45:46 -0700880 ret = walk_system_ram_range(pfn, nr_pages, &onlined_pages,
KAMEZAWA Hiroyuki75884fb2007-10-16 01:26:10 -0700881 online_pages_range);
Geoff Levandfd8a4222008-05-14 16:05:50 -0700882 if (ret) {
Wen Congyang6dcd73d2012-12-11 16:01:01 -0800883 if (need_zonelists_rebuild)
884 zone_pcp_reset(zone);
Chen Yuconge33e33b2016-03-17 14:19:35 -0700885 goto failed_addition;
Geoff Levandfd8a4222008-05-14 16:05:50 -0700886 }
887
Dave Hansen3947be12005-10-29 18:16:54 -0700888 zone->present_pages += onlined_pages;
Cody P Schaferaa472282013-07-03 15:02:10 -0700889
890 pgdat_resize_lock(zone->zone_pgdat, &flags);
Yasunori Gotof2937be2006-03-09 17:33:51 -0800891 zone->zone_pgdat->node_present_pages += onlined_pages;
Cody P Schaferaa472282013-07-03 15:02:10 -0700892 pgdat_resize_unlock(zone->zone_pgdat, &flags);
893
Dan Williamse900a912019-05-14 15:41:28 -0700894 shuffle_zone(zone);
895
Jiang Liu08dff7b2012-07-31 16:43:30 -0700896 if (onlined_pages) {
Vlastimil Babkae888ca32016-03-17 14:18:12 -0700897 node_states_set_node(nid, &arg);
Jiang Liu08dff7b2012-07-31 16:43:30 -0700898 if (need_zonelists_rebuild)
Michal Hocko72675e12017-09-06 16:20:24 -0700899 build_all_zonelists(NULL);
Jiang Liu08dff7b2012-07-31 16:43:30 -0700900 else
901 zone_pcp_update(zone);
902 }
Dave Hansen3947be12005-10-29 18:16:54 -0700903
KOSAKI Motohiro1b79acc2011-05-24 17:11:32 -0700904 init_per_zone_wmark_min();
905
Vlastimil Babka698b1b32016-03-17 14:18:08 -0700906 if (onlined_pages) {
Vlastimil Babkae888ca32016-03-17 14:18:12 -0700907 kswapd_run(nid);
Vlastimil Babka698b1b32016-03-17 14:18:08 -0700908 kcompactd_run(nid);
909 }
Dave Hansen61b13992005-10-29 18:16:56 -0700910
Haicheng Li1f522502010-05-24 14:32:51 -0700911 vm_total_pages = nr_free_pagecache_pages();
Kent Liu2f7f24e2008-07-23 21:28:18 -0700912
Chandra Seetharaman2d1d43f2006-09-29 02:01:25 -0700913 writeback_set_ratelimit();
Yasunori Goto7b78d332007-10-21 16:41:36 -0700914
915 if (onlined_pages)
916 memory_notify(MEM_ONLINE, &arg);
David Hildenbrand381eab42018-10-30 15:10:29 -0700917 mem_hotplug_done();
David Rientjes30467e02015-04-14 15:45:11 -0700918 return 0;
Chen Yuconge33e33b2016-03-17 14:19:35 -0700919
920failed_addition:
921 pr_debug("online_pages [mem %#010llx-%#010llx] failed\n",
922 (unsigned long long) pfn << PAGE_SHIFT,
923 (((unsigned long long) pfn + nr_pages) << PAGE_SHIFT) - 1);
924 memory_notify(MEM_CANCEL_ONLINE, &arg);
David Hildenbrand381eab42018-10-30 15:10:29 -0700925 mem_hotplug_done();
Chen Yuconge33e33b2016-03-17 14:19:35 -0700926 return ret;
Dave Hansen3947be12005-10-29 18:16:54 -0700927}
Keith Mannthey53947022006-09-30 23:27:08 -0700928#endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
Yasunori Gotobc02af92006-06-27 02:53:30 -0700929
Tang Chen0bd85422014-11-13 15:19:41 -0800930static void reset_node_present_pages(pg_data_t *pgdat)
931{
932 struct zone *z;
933
934 for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++)
935 z->present_pages = 0;
936
937 pgdat->node_present_pages = 0;
938}
939
Hidetoshi Setoe1319332009-11-17 14:06:18 -0800940/* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
941static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start)
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700942{
943 struct pglist_data *pgdat;
Fabian Frederickc8e861a2014-06-04 16:07:51 -0700944 unsigned long start_pfn = PFN_DOWN(start);
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700945
Tang Chena1e565a2013-02-22 16:33:18 -0800946 pgdat = NODE_DATA(nid);
947 if (!pgdat) {
948 pgdat = arch_alloc_nodedata(nid);
949 if (!pgdat)
950 return NULL;
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700951
Tang Chena1e565a2013-02-22 16:33:18 -0800952 arch_refresh_nodedata(nid, pgdat);
Gu Zhengb0dc3a32015-03-25 15:55:20 -0700953 } else {
Mel Gormane716f2e2017-05-03 14:53:45 -0700954 /*
955 * Reset the nr_zones, order and classzone_idx before reuse.
956 * Note that kswapd will init kswapd_classzone_idx properly
957 * when it starts in the near future.
958 */
Gu Zhengb0dc3a32015-03-25 15:55:20 -0700959 pgdat->nr_zones = 0;
Mel Gorman38087d92016-07-28 15:45:49 -0700960 pgdat->kswapd_order = 0;
961 pgdat->kswapd_classzone_idx = 0;
Tang Chena1e565a2013-02-22 16:33:18 -0800962 }
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700963
964 /* we can use NODE_DATA(nid) from here */
965
Oscar Salvador03e85f92018-08-21 21:53:43 -0700966 pgdat->node_id = nid;
967 pgdat->node_start_pfn = start_pfn;
968
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700969 /* init node's zones as empty zones, we don't have any present pages.*/
Oscar Salvador03e85f92018-08-21 21:53:43 -0700970 free_area_init_core_hotplug(nid);
Reza Arbab58301692016-08-11 15:33:12 -0700971 pgdat->per_cpu_nodestats = alloc_percpu(struct per_cpu_nodestat);
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700972
KAMEZAWA Hiroyuki959ecc42011-06-15 15:08:38 -0700973 /*
974 * The node we allocated has no zone fallback lists. For avoiding
975 * to access not-initialized zonelist, build here.
976 */
Michal Hocko72675e12017-09-06 16:20:24 -0700977 build_all_zonelists(pgdat);
KAMEZAWA Hiroyuki959ecc42011-06-15 15:08:38 -0700978
Tang Chenf784a3f2014-11-13 15:19:39 -0800979 /*
Tang Chen0bd85422014-11-13 15:19:41 -0800980 * When memory is hot-added, all the memory is in offline state. So
981 * clear all zones' present_pages because they will be updated in
982 * online_pages() and offline_pages().
983 */
Oscar Salvador03e85f92018-08-21 21:53:43 -0700984 reset_node_managed_pages(pgdat);
Tang Chen0bd85422014-11-13 15:19:41 -0800985 reset_node_present_pages(pgdat);
986
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700987 return pgdat;
988}
989
Oscar Salvadorb9ff0362018-08-17 15:46:15 -0700990static void rollback_node_hotadd(int nid)
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700991{
Oscar Salvadorb9ff0362018-08-17 15:46:15 -0700992 pg_data_t *pgdat = NODE_DATA(nid);
993
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700994 arch_refresh_nodedata(nid, NULL);
Reza Arbab58301692016-08-11 15:33:12 -0700995 free_percpu(pgdat->per_cpu_nodestats);
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700996 arch_free_nodedata(pgdat);
997 return;
998}
999
KAMEZAWA Hiroyuki0a547032006-06-27 02:53:35 -07001000
Toshi Kani01b0f192013-11-12 15:07:25 -08001001/**
1002 * try_online_node - online a node if offlined
Mike Rapoporte8b098f2018-04-05 16:24:57 -07001003 * @nid: the node ID
Oscar Salvadorb9ff0362018-08-17 15:46:15 -07001004 * @start: start addr of the node
1005 * @set_node_online: Whether we want to online the node
minskey guocf234222010-05-24 14:32:41 -07001006 * called by cpu_up() to online a node without onlined memory.
Oscar Salvadorb9ff0362018-08-17 15:46:15 -07001007 *
1008 * Returns:
1009 * 1 -> a new node has been allocated
1010 * 0 -> the node is already online
1011 * -ENOMEM -> the node could not be allocated
minskey guocf234222010-05-24 14:32:41 -07001012 */
Oscar Salvadorb9ff0362018-08-17 15:46:15 -07001013static int __try_online_node(int nid, u64 start, bool set_node_online)
minskey guocf234222010-05-24 14:32:41 -07001014{
Oscar Salvadorb9ff0362018-08-17 15:46:15 -07001015 pg_data_t *pgdat;
1016 int ret = 1;
minskey guocf234222010-05-24 14:32:41 -07001017
Toshi Kani01b0f192013-11-12 15:07:25 -08001018 if (node_online(nid))
1019 return 0;
1020
Oscar Salvadorb9ff0362018-08-17 15:46:15 -07001021 pgdat = hotadd_new_pgdat(nid, start);
David Rientjes7553e8f2011-06-22 18:13:01 -07001022 if (!pgdat) {
Toshi Kani01b0f192013-11-12 15:07:25 -08001023 pr_err("Cannot online node %d due to NULL pgdat\n", nid);
minskey guocf234222010-05-24 14:32:41 -07001024 ret = -ENOMEM;
1025 goto out;
1026 }
Oscar Salvadorb9ff0362018-08-17 15:46:15 -07001027
1028 if (set_node_online) {
1029 node_set_online(nid);
1030 ret = register_one_node(nid);
1031 BUG_ON(ret);
1032 }
minskey guocf234222010-05-24 14:32:41 -07001033out:
Oscar Salvadorb9ff0362018-08-17 15:46:15 -07001034 return ret;
1035}
1036
1037/*
1038 * Users of this function always want to online/register the node
1039 */
1040int try_online_node(int nid)
1041{
1042 int ret;
1043
1044 mem_hotplug_begin();
1045 ret = __try_online_node(nid, 0, true);
Vladimir Davydovbfc8c902014-06-04 16:07:18 -07001046 mem_hotplug_done();
minskey guocf234222010-05-24 14:32:41 -07001047 return ret;
1048}
1049
Toshi Kani27356f52013-09-11 14:21:49 -07001050static int check_hotplug_memory_range(u64 start, u64 size)
1051{
Pavel Tatashinba325582018-04-05 16:22:39 -07001052 unsigned long block_sz = memory_block_size_bytes();
1053 u64 block_nr_pages = block_sz >> PAGE_SHIFT;
Toshi Kani27356f52013-09-11 14:21:49 -07001054 u64 nr_pages = size >> PAGE_SHIFT;
Pavel Tatashinba325582018-04-05 16:22:39 -07001055 u64 start_pfn = PFN_DOWN(start);
Toshi Kani27356f52013-09-11 14:21:49 -07001056
Pavel Tatashinba325582018-04-05 16:22:39 -07001057 /* memory range must be block size aligned */
1058 if (!nr_pages || !IS_ALIGNED(start_pfn, block_nr_pages) ||
1059 !IS_ALIGNED(nr_pages, block_nr_pages)) {
1060 pr_err("Block size [%#lx] unaligned hotplug range: start %#llx, size %#llx",
1061 block_sz, start, size);
Toshi Kani27356f52013-09-11 14:21:49 -07001062 return -EINVAL;
1063 }
1064
1065 return 0;
1066}
1067
Vitaly Kuznetsov31bc3852016-03-15 14:56:48 -07001068static int online_memory_block(struct memory_block *mem, void *arg)
1069{
Nathan Fontenotdc18d702017-02-24 15:00:02 -08001070 return device_online(&mem->dev);
Vitaly Kuznetsov31bc3852016-03-15 14:56:48 -07001071}
1072
David Hildenbrand8df1d0e2018-10-30 15:10:24 -07001073/*
1074 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1075 * and online/offline operations (triggered e.g. by sysfs).
1076 *
1077 * we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG
1078 */
David Hildenbrandf29d8e9c02018-12-28 00:35:36 -08001079int __ref add_memory_resource(int nid, struct resource *res)
Yasunori Gotobc02af92006-06-27 02:53:30 -07001080{
Michal Hocko940519f2019-05-13 17:21:26 -07001081 struct mhp_restrictions restrictions = {
1082 .flags = MHP_MEMBLOCK_API,
1083 };
David Vrabel62cedb92015-06-25 16:35:49 +01001084 u64 start, size;
Oscar Salvadorb9ff0362018-08-17 15:46:15 -07001085 bool new_node = false;
Yasunori Gotobc02af92006-06-27 02:53:30 -07001086 int ret;
1087
David Vrabel62cedb92015-06-25 16:35:49 +01001088 start = res->start;
1089 size = resource_size(res);
1090
Toshi Kani27356f52013-09-11 14:21:49 -07001091 ret = check_hotplug_memory_range(start, size);
1092 if (ret)
1093 return ret;
1094
Vladimir Davydovbfc8c902014-06-04 16:07:18 -07001095 mem_hotplug_begin();
Nathan Zimmerac13c462014-01-23 15:53:26 -08001096
Tang Chen7f36e3e2015-09-04 15:42:32 -07001097 /*
1098 * Add new range to memblock so that when hotadd_new_pgdat() is called
1099 * to allocate new pgdat, get_pfn_range_for_nid() will be able to find
1100 * this new range and calculate total pages correctly. The range will
1101 * be removed at hot-remove time.
1102 */
1103 memblock_add_node(start, size, nid);
1104
Oscar Salvadorb9ff0362018-08-17 15:46:15 -07001105 ret = __try_online_node(nid, start, false);
1106 if (ret < 0)
1107 goto error;
1108 new_node = ret;
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001109
Yasunori Gotobc02af92006-06-27 02:53:30 -07001110 /* call arch's memory hotadd */
Michal Hocko940519f2019-05-13 17:21:26 -07001111 ret = arch_add_memory(nid, start, size, &restrictions);
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001112 if (ret < 0)
1113 goto error;
1114
Tang Chena1e565a2013-02-22 16:33:18 -08001115 if (new_node) {
Oscar Salvadord5b6f6a32018-08-17 15:46:18 -07001116 /* If sysfs file of new node can't be created, cpu on the node
Yasunori Goto0fc44152006-06-27 02:53:38 -07001117 * can't be hot-added. There is no rollback way now.
1118 * So, check by BUG_ON() to catch it reluctantly..
Oscar Salvadord5b6f6a32018-08-17 15:46:18 -07001119 * We online node here. We can't roll back from here.
Yasunori Goto0fc44152006-06-27 02:53:38 -07001120 */
Oscar Salvadord5b6f6a32018-08-17 15:46:18 -07001121 node_set_online(nid);
1122 ret = __register_one_node(nid);
Yasunori Goto0fc44152006-06-27 02:53:38 -07001123 BUG_ON(ret);
1124 }
1125
Oscar Salvadord5b6f6a32018-08-17 15:46:18 -07001126 /* link memory sections under this node.*/
Oscar Salvador4fbce632018-08-17 15:46:22 -07001127 ret = link_mem_sections(nid, PFN_DOWN(start), PFN_UP(start + size - 1));
Oscar Salvadord5b6f6a32018-08-17 15:46:18 -07001128 BUG_ON(ret);
1129
akpm@linux-foundation.orgd96ae532010-03-05 13:41:58 -08001130 /* create new memmap entry */
1131 firmware_map_add_hotplug(start, start + size, "System RAM");
1132
David Hildenbrand381eab42018-10-30 15:10:29 -07001133 /* device_online() will take the lock when calling online_pages() */
1134 mem_hotplug_done();
1135
Vitaly Kuznetsov31bc3852016-03-15 14:56:48 -07001136 /* online pages if requested */
David Hildenbrandf29d8e9c02018-12-28 00:35:36 -08001137 if (memhp_auto_online)
Vitaly Kuznetsov31bc3852016-03-15 14:56:48 -07001138 walk_memory_range(PFN_DOWN(start), PFN_UP(start + size - 1),
1139 NULL, online_memory_block);
1140
David Hildenbrand381eab42018-10-30 15:10:29 -07001141 return ret;
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001142error:
1143 /* rollback pgdat allocation and others */
Oscar Salvadorb9ff0362018-08-17 15:46:15 -07001144 if (new_node)
1145 rollback_node_hotadd(nid);
Tang Chen7f36e3e2015-09-04 15:42:32 -07001146 memblock_remove(start, size);
Vladimir Davydovbfc8c902014-06-04 16:07:18 -07001147 mem_hotplug_done();
Yasunori Gotobc02af92006-06-27 02:53:30 -07001148 return ret;
1149}
David Vrabel62cedb92015-06-25 16:35:49 +01001150
David Hildenbrand8df1d0e2018-10-30 15:10:24 -07001151/* requires device_hotplug_lock, see add_memory_resource() */
1152int __ref __add_memory(int nid, u64 start, u64 size)
David Vrabel62cedb92015-06-25 16:35:49 +01001153{
1154 struct resource *res;
1155 int ret;
1156
1157 res = register_memory_resource(start, size);
Vitaly Kuznetsov6f754ba2016-01-14 15:21:55 -08001158 if (IS_ERR(res))
1159 return PTR_ERR(res);
David Vrabel62cedb92015-06-25 16:35:49 +01001160
David Hildenbrandf29d8e9c02018-12-28 00:35:36 -08001161 ret = add_memory_resource(nid, res);
David Vrabel62cedb92015-06-25 16:35:49 +01001162 if (ret < 0)
1163 release_memory_resource(res);
1164 return ret;
1165}
David Hildenbrand8df1d0e2018-10-30 15:10:24 -07001166
1167int add_memory(int nid, u64 start, u64 size)
1168{
1169 int rc;
1170
1171 lock_device_hotplug();
1172 rc = __add_memory(nid, start, size);
1173 unlock_device_hotplug();
1174
1175 return rc;
1176}
Yasunori Gotobc02af92006-06-27 02:53:30 -07001177EXPORT_SYMBOL_GPL(add_memory);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001178
1179#ifdef CONFIG_MEMORY_HOTREMOVE
1180/*
Badari Pulavarty5c755e92008-07-23 21:28:19 -07001181 * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy
1182 * set and the size of the free page is given by page_order(). Using this,
1183 * the function determines if the pageblock contains only free pages.
1184 * Due to buddy contraints, a free page at least the size of a pageblock will
1185 * be located at the start of the pageblock
1186 */
1187static inline int pageblock_free(struct page *page)
1188{
1189 return PageBuddy(page) && page_order(page) >= pageblock_order;
1190}
1191
Michal Hocko891cb2a2019-02-20 22:20:46 -08001192/* Return the pfn of the start of the next active pageblock after a given pfn */
1193static unsigned long next_active_pageblock(unsigned long pfn)
Badari Pulavarty5c755e92008-07-23 21:28:19 -07001194{
Michal Hocko891cb2a2019-02-20 22:20:46 -08001195 struct page *page = pfn_to_page(pfn);
1196
Badari Pulavarty5c755e92008-07-23 21:28:19 -07001197 /* Ensure the starting page is pageblock-aligned */
Michal Hocko891cb2a2019-02-20 22:20:46 -08001198 BUG_ON(pfn & (pageblock_nr_pages - 1));
Badari Pulavarty5c755e92008-07-23 21:28:19 -07001199
Badari Pulavarty5c755e92008-07-23 21:28:19 -07001200 /* If the entire pageblock is free, move to the end of free page */
KAMEZAWA Hiroyuki0dcc48c2010-09-09 16:38:01 -07001201 if (pageblock_free(page)) {
1202 int order;
1203 /* be careful. we don't have locks, page_order can be changed.*/
1204 order = page_order(page);
1205 if ((order < MAX_ORDER) && (order >= pageblock_order))
Michal Hocko891cb2a2019-02-20 22:20:46 -08001206 return pfn + (1 << order);
KAMEZAWA Hiroyuki0dcc48c2010-09-09 16:38:01 -07001207 }
Badari Pulavarty5c755e92008-07-23 21:28:19 -07001208
Michal Hocko891cb2a2019-02-20 22:20:46 -08001209 return pfn + pageblock_nr_pages;
Badari Pulavarty5c755e92008-07-23 21:28:19 -07001210}
1211
Michal Hocko891cb2a2019-02-20 22:20:46 -08001212static bool is_pageblock_removable_nolock(unsigned long pfn)
Mathieu Malaterrefb52bba2018-06-07 17:07:43 -07001213{
Michal Hocko891cb2a2019-02-20 22:20:46 -08001214 struct page *page = pfn_to_page(pfn);
Mathieu Malaterrefb52bba2018-06-07 17:07:43 -07001215 struct zone *zone;
Mathieu Malaterrefb52bba2018-06-07 17:07:43 -07001216
1217 /*
1218 * We have to be careful here because we are iterating over memory
1219 * sections which are not zone aware so we might end up outside of
1220 * the zone but still within the section.
1221 * We have to take care about the node as well. If the node is offline
1222 * its NODE_DATA will be NULL - see page_zone.
1223 */
1224 if (!node_online(page_to_nid(page)))
1225 return false;
1226
1227 zone = page_zone(page);
1228 pfn = page_to_pfn(page);
1229 if (!zone_spans_pfn(zone, pfn))
1230 return false;
1231
Michal Hockod381c542018-12-28 00:33:56 -08001232 return !has_unmovable_pages(zone, page, 0, MIGRATE_MOVABLE, SKIP_HWPOISON);
Mathieu Malaterrefb52bba2018-06-07 17:07:43 -07001233}
1234
Badari Pulavarty5c755e92008-07-23 21:28:19 -07001235/* Checks if this range of memory is likely to be hot-removable. */
Yaowei Baic98940f2016-05-19 17:11:26 -07001236bool is_mem_section_removable(unsigned long start_pfn, unsigned long nr_pages)
Badari Pulavarty5c755e92008-07-23 21:28:19 -07001237{
Michal Hocko891cb2a2019-02-20 22:20:46 -08001238 unsigned long end_pfn, pfn;
1239
1240 end_pfn = min(start_pfn + nr_pages,
1241 zone_end_pfn(page_zone(pfn_to_page(start_pfn))));
Badari Pulavarty5c755e92008-07-23 21:28:19 -07001242
1243 /* Check the starting page of each pageblock within the range */
Michal Hocko891cb2a2019-02-20 22:20:46 -08001244 for (pfn = start_pfn; pfn < end_pfn; pfn = next_active_pageblock(pfn)) {
1245 if (!is_pageblock_removable_nolock(pfn))
Yaowei Baic98940f2016-05-19 17:11:26 -07001246 return false;
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07001247 cond_resched();
Badari Pulavarty5c755e92008-07-23 21:28:19 -07001248 }
1249
1250 /* All pageblocks in the memory block are likely to be hot-removable */
Yaowei Baic98940f2016-05-19 17:11:26 -07001251 return true;
Badari Pulavarty5c755e92008-07-23 21:28:19 -07001252}
1253
1254/*
Toshi Kanideb88a22017-02-03 13:13:20 -08001255 * Confirm all pages in a range [start, end) belong to the same zone.
Toshi Kania96dfdd2017-02-03 13:13:23 -08001256 * When true, return its valid [start, end).
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001257 */
Toshi Kania96dfdd2017-02-03 13:13:23 -08001258int test_pages_in_a_zone(unsigned long start_pfn, unsigned long end_pfn,
1259 unsigned long *valid_start, unsigned long *valid_end)
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001260{
Andrew Banman5f0f2882015-12-29 14:54:25 -08001261 unsigned long pfn, sec_end_pfn;
Toshi Kania96dfdd2017-02-03 13:13:23 -08001262 unsigned long start, end;
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001263 struct zone *zone = NULL;
1264 struct page *page;
1265 int i;
Toshi Kanideb88a22017-02-03 13:13:20 -08001266 for (pfn = start_pfn, sec_end_pfn = SECTION_ALIGN_UP(start_pfn + 1);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001267 pfn < end_pfn;
Toshi Kanideb88a22017-02-03 13:13:20 -08001268 pfn = sec_end_pfn, sec_end_pfn += PAGES_PER_SECTION) {
Andrew Banman5f0f2882015-12-29 14:54:25 -08001269 /* Make sure the memory section is present first */
1270 if (!present_section_nr(pfn_to_section_nr(pfn)))
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001271 continue;
Andrew Banman5f0f2882015-12-29 14:54:25 -08001272 for (; pfn < sec_end_pfn && pfn < end_pfn;
1273 pfn += MAX_ORDER_NR_PAGES) {
1274 i = 0;
1275 /* This is just a CONFIG_HOLES_IN_ZONE check.*/
1276 while ((i < MAX_ORDER_NR_PAGES) &&
1277 !pfn_valid_within(pfn + i))
1278 i++;
zhong jiangd6d8c8a2017-02-24 14:59:30 -08001279 if (i == MAX_ORDER_NR_PAGES || pfn + i >= end_pfn)
Andrew Banman5f0f2882015-12-29 14:54:25 -08001280 continue;
Mikhail Zaslonko24feb472019-02-01 14:20:38 -08001281 /* Check if we got outside of the zone */
1282 if (zone && !zone_spans_pfn(zone, pfn + i))
1283 return 0;
Andrew Banman5f0f2882015-12-29 14:54:25 -08001284 page = pfn_to_page(pfn + i);
1285 if (zone && page_zone(page) != zone)
1286 return 0;
Toshi Kania96dfdd2017-02-03 13:13:23 -08001287 if (!zone)
1288 start = pfn + i;
Andrew Banman5f0f2882015-12-29 14:54:25 -08001289 zone = page_zone(page);
Toshi Kania96dfdd2017-02-03 13:13:23 -08001290 end = pfn + MAX_ORDER_NR_PAGES;
Andrew Banman5f0f2882015-12-29 14:54:25 -08001291 }
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001292 }
Toshi Kanideb88a22017-02-03 13:13:20 -08001293
Toshi Kania96dfdd2017-02-03 13:13:23 -08001294 if (zone) {
1295 *valid_start = start;
zhong jiangd6d8c8a2017-02-24 14:59:30 -08001296 *valid_end = min(end, end_pfn);
Toshi Kanideb88a22017-02-03 13:13:20 -08001297 return 1;
Toshi Kania96dfdd2017-02-03 13:13:23 -08001298 } else {
Toshi Kanideb88a22017-02-03 13:13:20 -08001299 return 0;
Toshi Kania96dfdd2017-02-03 13:13:23 -08001300 }
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001301}
1302
1303/*
Yisheng Xie0efadf42017-02-24 14:57:39 -08001304 * Scan pfn range [start,end) to find movable/migratable pages (LRU pages,
1305 * non-lru movable pages and hugepages). We scan pfn because it's much
1306 * easier than scanning over linked list. This function returns the pfn
1307 * of the first found movable page if it's found, otherwise 0.
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001308 */
Naoya Horiguchic8721bb2013-09-11 14:22:09 -07001309static unsigned long scan_movable_pages(unsigned long start, unsigned long end)
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001310{
1311 unsigned long pfn;
Oscar Salvadoreeb0efd2019-02-01 14:20:47 -08001312
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001313 for (pfn = start; pfn < end; pfn++) {
Oscar Salvadoreeb0efd2019-02-01 14:20:47 -08001314 struct page *page, *head;
1315 unsigned long skip;
1316
1317 if (!pfn_valid(pfn))
1318 continue;
1319 page = pfn_to_page(pfn);
1320 if (PageLRU(page))
1321 return pfn;
1322 if (__PageMovable(page))
1323 return pfn;
1324
1325 if (!PageHuge(page))
1326 continue;
1327 head = compound_head(page);
Oscar Salvador39186cb2019-05-13 17:17:32 -07001328 if (page_huge_active(head))
Oscar Salvadoreeb0efd2019-02-01 14:20:47 -08001329 return pfn;
1330 skip = (1 << compound_order(head)) - (page - head);
1331 pfn += skip - 1;
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001332 }
1333 return 0;
1334}
1335
Michal Hocko666feb22018-04-10 16:30:03 -07001336static struct page *new_node_page(struct page *page, unsigned long private)
Xishi Qiu394e31d2016-07-28 15:48:53 -07001337{
Xishi Qiu394e31d2016-07-28 15:48:53 -07001338 int nid = page_to_nid(page);
Li Zhong231e97e2016-09-28 15:22:38 -07001339 nodemask_t nmask = node_states[N_MEMORY];
Michal Hocko7f252f22017-07-10 15:48:41 -07001340
1341 /*
1342 * try to allocate from a different node but reuse this node if there
1343 * are no other online nodes to be used (e.g. we are offlining a part
1344 * of the only existing node)
1345 */
1346 node_clear(nid, nmask);
1347 if (nodes_empty(nmask))
1348 node_set(nid, nmask);
Xishi Qiu394e31d2016-07-28 15:48:53 -07001349
Michal Hocko8b913232017-07-10 15:48:47 -07001350 return new_page_nodemask(page, nid, &nmask);
Xishi Qiu394e31d2016-07-28 15:48:53 -07001351}
1352
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001353static int
1354do_migrate_range(unsigned long start_pfn, unsigned long end_pfn)
1355{
1356 unsigned long pfn;
1357 struct page *page;
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001358 int ret = 0;
1359 LIST_HEAD(source);
1360
Michal Hockoa85009c2018-12-28 00:38:29 -08001361 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001362 if (!pfn_valid(pfn))
1363 continue;
1364 page = pfn_to_page(pfn);
Naoya Horiguchic8721bb2013-09-11 14:22:09 -07001365
1366 if (PageHuge(page)) {
1367 struct page *head = compound_head(page);
Oscar Salvadordaf35382019-03-05 15:48:53 -08001368 pfn = page_to_pfn(head) + (1<<compound_order(head)) - 1;
1369 isolate_huge_page(head, &source);
Naoya Horiguchic8721bb2013-09-11 14:22:09 -07001370 continue;
Michal Hocko94723aa2018-04-10 16:30:07 -07001371 } else if (PageTransHuge(page))
Naoya Horiguchi8135d892017-09-08 16:11:15 -07001372 pfn = page_to_pfn(compound_head(page))
1373 + hpage_nr_pages(page) - 1;
Naoya Horiguchic8721bb2013-09-11 14:22:09 -07001374
Michal Hockob15c8722018-12-28 00:38:01 -08001375 /*
1376 * HWPoison pages have elevated reference counts so the migration would
1377 * fail on them. It also doesn't make any sense to migrate them in the
1378 * first place. Still try to unmap such a page in case it is still mapped
1379 * (e.g. current hwpoison implementation doesn't unmap KSM pages but keep
1380 * the unmap as the catch all safety net).
1381 */
1382 if (PageHWPoison(page)) {
1383 if (WARN_ON(PageLRU(page)))
1384 isolate_lru_page(page);
1385 if (page_mapped(page))
1386 try_to_unmap(page, TTU_IGNORE_MLOCK | TTU_IGNORE_ACCESS);
1387 continue;
1388 }
1389
Konstantin Khlebnikov700c2a42011-05-24 17:12:19 -07001390 if (!get_page_unless_zero(page))
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001391 continue;
1392 /*
Yisheng Xie0efadf42017-02-24 14:57:39 -08001393 * We can skip free pages. And we can deal with pages on
1394 * LRU and non-lru movable pages.
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001395 */
Yisheng Xie0efadf42017-02-24 14:57:39 -08001396 if (PageLRU(page))
1397 ret = isolate_lru_page(page);
1398 else
1399 ret = isolate_movable_page(page, ISOLATE_UNEVICTABLE);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001400 if (!ret) { /* Success */
Nick Piggin62695a82008-10-18 20:26:09 -07001401 list_add_tail(&page->lru, &source);
Yisheng Xie0efadf42017-02-24 14:57:39 -08001402 if (!__PageMovable(page))
1403 inc_node_page_state(page, NR_ISOLATED_ANON +
1404 page_is_file_cache(page));
KOSAKI Motohiro6d9c2852009-12-14 17:58:11 -08001405
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001406 } else {
Michal Hocko2932c8b2018-12-28 00:33:53 -08001407 pr_warn("failed to isolate pfn %lx\n", pfn);
Yisheng Xie0efadf42017-02-24 14:57:39 -08001408 dump_page(page, "isolation failed");
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001409 }
Oscar Salvador17230582019-02-01 14:19:57 -08001410 put_page(page);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001411 }
Bob Liuf3ab2632010-10-26 14:22:10 -07001412 if (!list_empty(&source)) {
Xishi Qiu394e31d2016-07-28 15:48:53 -07001413 /* Allocate a new page from the nearest neighbor node */
1414 ret = migrate_pages(&source, new_node_page, NULL, 0,
Hugh Dickins9c620e22013-02-22 16:35:14 -08001415 MIGRATE_SYNC, MR_MEMORY_HOTPLUG);
Michal Hocko2932c8b2018-12-28 00:33:53 -08001416 if (ret) {
1417 list_for_each_entry(page, &source, lru) {
1418 pr_warn("migrating pfn %lx failed ret:%d ",
1419 page_to_pfn(page), ret);
1420 dump_page(page, "migration failure");
1421 }
Naoya Horiguchic8721bb2013-09-11 14:22:09 -07001422 putback_movable_pages(&source);
Michal Hocko2932c8b2018-12-28 00:33:53 -08001423 }
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001424 }
Oscar Salvador17230582019-02-01 14:19:57 -08001425
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001426 return ret;
1427}
1428
1429/*
1430 * remove from free_area[] and mark all as Reserved.
1431 */
1432static int
1433offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages,
1434 void *data)
1435{
Michal Hocko5557c762019-05-13 17:21:24 -07001436 unsigned long *offlined_pages = (unsigned long *)data;
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001437
Michal Hocko5557c762019-05-13 17:21:24 -07001438 *offlined_pages += __offline_isolated_pages(start, start + nr_pages);
1439 return 0;
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001440}
1441
1442/*
1443 * Check all pages in range, recoreded as memory resource, are isolated.
1444 */
1445static int
1446check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages,
1447 void *data)
1448{
Michal Hocko5557c762019-05-13 17:21:24 -07001449 return test_pages_isolated(start_pfn, start_pfn + nr_pages, true);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001450}
1451
Tang Chenc5320922013-11-12 15:08:10 -08001452static int __init cmdline_parse_movable_node(char *p)
1453{
Michal Hocko49323812017-07-06 15:41:05 -07001454#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
Tang Chen55ac5902014-01-21 15:49:35 -08001455 movable_node_enabled = true;
Michal Hocko49323812017-07-06 15:41:05 -07001456#else
1457 pr_warn("movable_node parameter depends on CONFIG_HAVE_MEMBLOCK_NODE_MAP to work properly\n");
1458#endif
Tang Chenc5320922013-11-12 15:08:10 -08001459 return 0;
1460}
1461early_param("movable_node", cmdline_parse_movable_node);
1462
Lai Jiangshand9713672012-12-11 16:01:03 -08001463/* check which state of node_states will be changed when offline memory */
1464static void node_states_check_changes_offline(unsigned long nr_pages,
1465 struct zone *zone, struct memory_notify *arg)
1466{
1467 struct pglist_data *pgdat = zone->zone_pgdat;
1468 unsigned long present_pages = 0;
Oscar Salvador86b27be2018-10-26 15:07:38 -07001469 enum zone_type zt;
1470
Anshuman Khandual98fa15f2019-03-05 15:42:58 -08001471 arg->status_change_nid = NUMA_NO_NODE;
1472 arg->status_change_nid_normal = NUMA_NO_NODE;
1473 arg->status_change_nid_high = NUMA_NO_NODE;
Lai Jiangshand9713672012-12-11 16:01:03 -08001474
1475 /*
Oscar Salvador86b27be2018-10-26 15:07:38 -07001476 * Check whether node_states[N_NORMAL_MEMORY] will be changed.
1477 * If the memory to be offline is within the range
1478 * [0..ZONE_NORMAL], and it is the last present memory there,
1479 * the zones in that range will become empty after the offlining,
1480 * thus we can determine that we need to clear the node from
1481 * node_states[N_NORMAL_MEMORY].
Lai Jiangshand9713672012-12-11 16:01:03 -08001482 */
Oscar Salvador86b27be2018-10-26 15:07:38 -07001483 for (zt = 0; zt <= ZONE_NORMAL; zt++)
Lai Jiangshand9713672012-12-11 16:01:03 -08001484 present_pages += pgdat->node_zones[zt].present_pages;
Oscar Salvador86b27be2018-10-26 15:07:38 -07001485 if (zone_idx(zone) <= ZONE_NORMAL && nr_pages >= present_pages)
Lai Jiangshand9713672012-12-11 16:01:03 -08001486 arg->status_change_nid_normal = zone_to_nid(zone);
Lai Jiangshand9713672012-12-11 16:01:03 -08001487
Lai Jiangshan6715ddf2012-12-12 13:51:49 -08001488#ifdef CONFIG_HIGHMEM
1489 /*
Oscar Salvador86b27be2018-10-26 15:07:38 -07001490 * node_states[N_HIGH_MEMORY] contains nodes which
1491 * have normal memory or high memory.
1492 * Here we add the present_pages belonging to ZONE_HIGHMEM.
1493 * If the zone is within the range of [0..ZONE_HIGHMEM), and
1494 * we determine that the zones in that range become empty,
1495 * we need to clear the node for N_HIGH_MEMORY.
Lai Jiangshan6715ddf2012-12-12 13:51:49 -08001496 */
Oscar Salvador86b27be2018-10-26 15:07:38 -07001497 present_pages += pgdat->node_zones[ZONE_HIGHMEM].present_pages;
1498 if (zone_idx(zone) <= ZONE_HIGHMEM && nr_pages >= present_pages)
Lai Jiangshan6715ddf2012-12-12 13:51:49 -08001499 arg->status_change_nid_high = zone_to_nid(zone);
Lai Jiangshan6715ddf2012-12-12 13:51:49 -08001500#endif
1501
Lai Jiangshand9713672012-12-11 16:01:03 -08001502 /*
Oscar Salvador86b27be2018-10-26 15:07:38 -07001503 * We have accounted the pages from [0..ZONE_NORMAL), and
1504 * in case of CONFIG_HIGHMEM the pages from ZONE_HIGHMEM
1505 * as well.
1506 * Here we count the possible pages from ZONE_MOVABLE.
1507 * If after having accounted all the pages, we see that the nr_pages
1508 * to be offlined is over or equal to the accounted pages,
1509 * we know that the node will become empty, and so, we can clear
1510 * it for N_MEMORY as well.
Lai Jiangshand9713672012-12-11 16:01:03 -08001511 */
Oscar Salvador86b27be2018-10-26 15:07:38 -07001512 present_pages += pgdat->node_zones[ZONE_MOVABLE].present_pages;
Lai Jiangshand9713672012-12-11 16:01:03 -08001513
Lai Jiangshand9713672012-12-11 16:01:03 -08001514 if (nr_pages >= present_pages)
1515 arg->status_change_nid = zone_to_nid(zone);
Lai Jiangshand9713672012-12-11 16:01:03 -08001516}
1517
1518static void node_states_clear_node(int node, struct memory_notify *arg)
1519{
1520 if (arg->status_change_nid_normal >= 0)
1521 node_clear_state(node, N_NORMAL_MEMORY);
1522
Oscar Salvadorcf01f6f52018-10-26 15:07:28 -07001523 if (arg->status_change_nid_high >= 0)
Lai Jiangshand9713672012-12-11 16:01:03 -08001524 node_clear_state(node, N_HIGH_MEMORY);
Lai Jiangshan6715ddf2012-12-12 13:51:49 -08001525
Oscar Salvadorcf01f6f52018-10-26 15:07:28 -07001526 if (arg->status_change_nid >= 0)
Lai Jiangshan6715ddf2012-12-12 13:51:49 -08001527 node_clear_state(node, N_MEMORY);
Lai Jiangshand9713672012-12-11 16:01:03 -08001528}
1529
Wen Congyanga16cee12012-10-08 16:33:58 -07001530static int __ref __offline_pages(unsigned long start_pfn,
Michal Hockoecde0f32017-11-15 17:33:38 -08001531 unsigned long end_pfn)
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001532{
Michal Hockoecde0f32017-11-15 17:33:38 -08001533 unsigned long pfn, nr_pages;
Michal Hocko5557c762019-05-13 17:21:24 -07001534 unsigned long offlined_pages = 0;
Qian Cai9b7ea462019-03-28 20:43:34 -07001535 int ret, node, nr_isolate_pageblock;
Cody P Schaferd7029092013-07-03 15:02:11 -07001536 unsigned long flags;
Toshi Kania96dfdd2017-02-03 13:13:23 -08001537 unsigned long valid_start, valid_end;
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001538 struct zone *zone;
Yasunori Goto7b78d332007-10-21 16:41:36 -07001539 struct memory_notify arg;
Michal Hocko79605092018-12-28 00:33:49 -08001540 char *reason;
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001541
David Hildenbrand381eab42018-10-30 15:10:29 -07001542 mem_hotplug_begin();
1543
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001544 /* This makes hotplug much easier...and readable.
1545 we assume this for now. .*/
David Hildenbrand381eab42018-10-30 15:10:29 -07001546 if (!test_pages_in_a_zone(start_pfn, end_pfn, &valid_start,
1547 &valid_end)) {
Michal Hocko79605092018-12-28 00:33:49 -08001548 ret = -EINVAL;
1549 reason = "multizone range";
1550 goto failed_removal;
David Hildenbrand381eab42018-10-30 15:10:29 -07001551 }
Yasunori Goto7b78d332007-10-21 16:41:36 -07001552
Toshi Kania96dfdd2017-02-03 13:13:23 -08001553 zone = page_zone(pfn_to_page(valid_start));
Yasunori Goto7b78d332007-10-21 16:41:36 -07001554 node = zone_to_nid(zone);
1555 nr_pages = end_pfn - start_pfn;
1556
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001557 /* set above range as isolated */
Wen Congyangb023f462012-12-11 16:00:45 -08001558 ret = start_isolate_page_range(start_pfn, end_pfn,
Michal Hockod381c542018-12-28 00:33:56 -08001559 MIGRATE_MOVABLE,
1560 SKIP_HWPOISON | REPORT_FAILURE);
Qian Cai9b7ea462019-03-28 20:43:34 -07001561 if (ret < 0) {
Michal Hocko79605092018-12-28 00:33:49 -08001562 reason = "failure to isolate range";
1563 goto failed_removal;
David Hildenbrand381eab42018-10-30 15:10:29 -07001564 }
Qian Cai9b7ea462019-03-28 20:43:34 -07001565 nr_isolate_pageblock = ret;
Yasunori Goto7b78d332007-10-21 16:41:36 -07001566
1567 arg.start_pfn = start_pfn;
1568 arg.nr_pages = nr_pages;
Lai Jiangshand9713672012-12-11 16:01:03 -08001569 node_states_check_changes_offline(nr_pages, zone, &arg);
Yasunori Goto7b78d332007-10-21 16:41:36 -07001570
1571 ret = memory_notify(MEM_GOING_OFFLINE, &arg);
1572 ret = notifier_to_errno(ret);
Michal Hocko79605092018-12-28 00:33:49 -08001573 if (ret) {
1574 reason = "notifier failure";
1575 goto failed_removal_isolated;
1576 }
Yasunori Goto7b78d332007-10-21 16:41:36 -07001577
Michal Hockobb8965b2018-12-28 00:38:32 -08001578 do {
1579 for (pfn = start_pfn; pfn;) {
1580 if (signal_pending(current)) {
1581 ret = -EINTR;
1582 reason = "signal backoff";
1583 goto failed_removal_isolated;
1584 }
Michal Hocko72b39cf2017-11-15 17:33:34 -08001585
Michal Hockobb8965b2018-12-28 00:38:32 -08001586 cond_resched();
1587 lru_add_drain_all();
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001588
Michal Hockobb8965b2018-12-28 00:38:32 -08001589 pfn = scan_movable_pages(pfn, end_pfn);
1590 if (pfn) {
1591 /*
1592 * TODO: fatal migration failures should bail
1593 * out
1594 */
1595 do_migrate_range(pfn, end_pfn);
1596 }
1597 }
Michal Hocko72b39cf2017-11-15 17:33:34 -08001598
Michal Hockobb8965b2018-12-28 00:38:32 -08001599 /*
1600 * Dissolve free hugepages in the memory block before doing
1601 * offlining actually in order to make hugetlbfs's object
1602 * counting consistent.
1603 */
1604 ret = dissolve_free_huge_pages(start_pfn, end_pfn);
1605 if (ret) {
1606 reason = "failure to dissolve huge pages";
1607 goto failed_removal_isolated;
1608 }
1609 /* check again */
Michal Hocko5557c762019-05-13 17:21:24 -07001610 ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn,
1611 NULL, check_pages_isolated_cb);
1612 } while (ret);
Michal Hockobb8965b2018-12-28 00:38:32 -08001613
Adam Buchbinderb3834be2012-09-19 21:48:02 -04001614 /* Ok, all of our target is isolated.
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001615 We cannot do rollback at this point. */
Michal Hocko5557c762019-05-13 17:21:24 -07001616 walk_system_ram_range(start_pfn, end_pfn - start_pfn,
1617 &offlined_pages, offline_isolated_pages_cb);
1618 pr_info("Offlined Pages %ld\n", offlined_pages);
Qian Cai9b7ea462019-03-28 20:43:34 -07001619 /*
1620 * Onlining will reset pagetype flags and makes migrate type
1621 * MOVABLE, so just need to decrease the number of isolated
1622 * pageblocks zone counter here.
1623 */
1624 spin_lock_irqsave(&zone->lock, flags);
1625 zone->nr_isolate_pageblock -= nr_isolate_pageblock;
1626 spin_unlock_irqrestore(&zone->lock, flags);
1627
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001628 /* removal success */
Jiang Liu3dcc0572013-07-03 15:03:21 -07001629 adjust_managed_page_count(pfn_to_page(start_pfn), -offlined_pages);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001630 zone->present_pages -= offlined_pages;
Cody P Schaferd7029092013-07-03 15:02:11 -07001631
1632 pgdat_resize_lock(zone->zone_pgdat, &flags);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001633 zone->zone_pgdat->node_present_pages -= offlined_pages;
Cody P Schaferd7029092013-07-03 15:02:11 -07001634 pgdat_resize_unlock(zone->zone_pgdat, &flags);
Yasunori Goto7b78d332007-10-21 16:41:36 -07001635
KOSAKI Motohiro1b79acc2011-05-24 17:11:32 -07001636 init_per_zone_wmark_min();
1637
Xishi Qiu1e8537b2012-10-08 16:31:51 -07001638 if (!populated_zone(zone)) {
Jiang Liu340175b2012-07-31 16:43:32 -07001639 zone_pcp_reset(zone);
Michal Hocko72675e12017-09-06 16:20:24 -07001640 build_all_zonelists(NULL);
Xishi Qiu1e8537b2012-10-08 16:31:51 -07001641 } else
1642 zone_pcp_update(zone);
Jiang Liu340175b2012-07-31 16:43:32 -07001643
Lai Jiangshand9713672012-12-11 16:01:03 -08001644 node_states_clear_node(node, &arg);
Vlastimil Babka698b1b32016-03-17 14:18:08 -07001645 if (arg.status_change_nid >= 0) {
David Rientjes8fe23e02009-12-14 17:58:33 -08001646 kswapd_stop(node);
Vlastimil Babka698b1b32016-03-17 14:18:08 -07001647 kcompactd_stop(node);
1648 }
Minchan Kimbce73942009-06-16 15:32:50 -07001649
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001650 vm_total_pages = nr_free_pagecache_pages();
1651 writeback_set_ratelimit();
Yasunori Goto7b78d332007-10-21 16:41:36 -07001652
1653 memory_notify(MEM_OFFLINE, &arg);
David Hildenbrand381eab42018-10-30 15:10:29 -07001654 mem_hotplug_done();
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001655 return 0;
1656
Michal Hocko79605092018-12-28 00:33:49 -08001657failed_removal_isolated:
1658 undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
Qian Caic4efe482019-03-28 20:44:16 -07001659 memory_notify(MEM_CANCEL_OFFLINE, &arg);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001660failed_removal:
Michal Hocko79605092018-12-28 00:33:49 -08001661 pr_debug("memory offlining [mem %#010llx-%#010llx] failed due to %s\n",
Chen Yuconge33e33b2016-03-17 14:19:35 -07001662 (unsigned long long) start_pfn << PAGE_SHIFT,
Michal Hocko79605092018-12-28 00:33:49 -08001663 ((unsigned long long) end_pfn << PAGE_SHIFT) - 1,
1664 reason);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001665 /* pushback to free area */
David Hildenbrand381eab42018-10-30 15:10:29 -07001666 mem_hotplug_done();
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001667 return ret;
1668}
Badari Pulavarty71088782008-10-18 20:25:58 -07001669
Wen Congyanga16cee12012-10-08 16:33:58 -07001670int offline_pages(unsigned long start_pfn, unsigned long nr_pages)
1671{
Michal Hockoecde0f32017-11-15 17:33:38 -08001672 return __offline_pages(start_pfn, start_pfn + nr_pages);
Wen Congyanga16cee12012-10-08 16:33:58 -07001673}
Rafael J. Wysockie2ff3942013-05-08 00:29:49 +02001674#endif /* CONFIG_MEMORY_HOTREMOVE */
Wen Congyanga16cee12012-10-08 16:33:58 -07001675
Wen Congyangbbc76be2013-02-22 16:32:54 -08001676/**
1677 * walk_memory_range - walks through all mem sections in [start_pfn, end_pfn)
1678 * @start_pfn: start pfn of the memory range
Toshi Kanie05c4bb2013-04-29 15:06:16 -07001679 * @end_pfn: end pfn of the memory range
Wen Congyangbbc76be2013-02-22 16:32:54 -08001680 * @arg: argument passed to func
1681 * @func: callback for each memory section walked
1682 *
1683 * This function walks through all present mem sections in range
1684 * [start_pfn, end_pfn) and call func on each mem section.
1685 *
1686 * Returns the return value of func.
1687 */
Rafael J. Wysockie2ff3942013-05-08 00:29:49 +02001688int walk_memory_range(unsigned long start_pfn, unsigned long end_pfn,
Wen Congyangbbc76be2013-02-22 16:32:54 -08001689 void *arg, int (*func)(struct memory_block *, void *))
Badari Pulavarty71088782008-10-18 20:25:58 -07001690{
Wen Congyange90bdb72012-10-08 16:34:01 -07001691 struct memory_block *mem = NULL;
1692 struct mem_section *section;
Wen Congyange90bdb72012-10-08 16:34:01 -07001693 unsigned long pfn, section_nr;
1694 int ret;
Badari Pulavarty71088782008-10-18 20:25:58 -07001695
Wen Congyange90bdb72012-10-08 16:34:01 -07001696 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
1697 section_nr = pfn_to_section_nr(pfn);
1698 if (!present_section_nr(section_nr))
1699 continue;
1700
1701 section = __nr_to_section(section_nr);
1702 /* same memblock? */
1703 if (mem)
1704 if ((section_nr >= mem->start_section_nr) &&
1705 (section_nr <= mem->end_section_nr))
1706 continue;
1707
1708 mem = find_memory_block_hinted(section, mem);
1709 if (!mem)
1710 continue;
1711
Wen Congyangbbc76be2013-02-22 16:32:54 -08001712 ret = func(mem, arg);
Wen Congyange90bdb72012-10-08 16:34:01 -07001713 if (ret) {
Wen Congyangbbc76be2013-02-22 16:32:54 -08001714 kobject_put(&mem->dev.kobj);
1715 return ret;
Wen Congyange90bdb72012-10-08 16:34:01 -07001716 }
1717 }
1718
1719 if (mem)
1720 kobject_put(&mem->dev.kobj);
1721
Wen Congyangbbc76be2013-02-22 16:32:54 -08001722 return 0;
1723}
1724
Rafael J. Wysockie2ff3942013-05-08 00:29:49 +02001725#ifdef CONFIG_MEMORY_HOTREMOVE
Xishi Qiud6de9d52013-11-12 15:07:20 -08001726static int check_memblock_offlined_cb(struct memory_block *mem, void *arg)
Wen Congyangbbc76be2013-02-22 16:32:54 -08001727{
1728 int ret = !is_memblock_offlined(mem);
1729
Randy Dunlap349daa02013-04-29 15:08:49 -07001730 if (unlikely(ret)) {
1731 phys_addr_t beginpa, endpa;
1732
1733 beginpa = PFN_PHYS(section_nr_to_pfn(mem->start_section_nr));
1734 endpa = PFN_PHYS(section_nr_to_pfn(mem->end_section_nr + 1))-1;
Joe Perches756a0252016-03-17 14:19:47 -07001735 pr_warn("removing memory fails, because memory [%pa-%pa] is onlined\n",
Randy Dunlap349daa02013-04-29 15:08:49 -07001736 &beginpa, &endpa);
1737 }
Wen Congyangbbc76be2013-02-22 16:32:54 -08001738
1739 return ret;
1740}
1741
Toshi Kani0f1cfe92013-09-11 14:21:50 -07001742static int check_cpu_on_node(pg_data_t *pgdat)
Tang Chen60a5a192013-02-22 16:33:14 -08001743{
Tang Chen60a5a192013-02-22 16:33:14 -08001744 int cpu;
1745
1746 for_each_present_cpu(cpu) {
1747 if (cpu_to_node(cpu) == pgdat->node_id)
1748 /*
1749 * the cpu on this node isn't removed, and we can't
1750 * offline this node.
1751 */
1752 return -EBUSY;
1753 }
1754
1755 return 0;
1756}
1757
Toshi Kani0f1cfe92013-09-11 14:21:50 -07001758/**
1759 * try_offline_node
Mike Rapoporte8b098f2018-04-05 16:24:57 -07001760 * @nid: the node ID
Toshi Kani0f1cfe92013-09-11 14:21:50 -07001761 *
1762 * Offline a node if all memory sections and cpus of the node are removed.
1763 *
1764 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1765 * and online/offline operations before this call.
1766 */
Wen Congyang90b30cd2013-02-22 16:33:27 -08001767void try_offline_node(int nid)
Tang Chen60a5a192013-02-22 16:33:14 -08001768{
Wen Congyangd822b862013-02-22 16:33:16 -08001769 pg_data_t *pgdat = NODE_DATA(nid);
1770 unsigned long start_pfn = pgdat->node_start_pfn;
1771 unsigned long end_pfn = start_pfn + pgdat->node_spanned_pages;
Tang Chen60a5a192013-02-22 16:33:14 -08001772 unsigned long pfn;
1773
1774 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
1775 unsigned long section_nr = pfn_to_section_nr(pfn);
1776
1777 if (!present_section_nr(section_nr))
1778 continue;
1779
1780 if (pfn_to_nid(pfn) != nid)
1781 continue;
1782
1783 /*
1784 * some memory sections of this node are not removed, and we
1785 * can't offline node now.
1786 */
1787 return;
1788 }
1789
Michal Hocko46a36792018-12-28 00:34:13 -08001790 if (check_cpu_on_node(pgdat))
Tang Chen60a5a192013-02-22 16:33:14 -08001791 return;
1792
1793 /*
1794 * all memory/cpu of this node are removed, we can offline this
1795 * node now.
1796 */
1797 node_set_offline(nid);
1798 unregister_one_node(nid);
1799}
Wen Congyang90b30cd2013-02-22 16:33:27 -08001800EXPORT_SYMBOL(try_offline_node);
Tang Chen60a5a192013-02-22 16:33:14 -08001801
David Hildenbrandd9eb1412019-05-13 17:21:33 -07001802static void __release_memory_resource(resource_size_t start,
1803 resource_size_t size)
1804{
1805 int ret;
1806
1807 /*
1808 * When removing memory in the same granularity as it was added,
1809 * this function never fails. It might only fail if resources
1810 * have to be adjusted or split. We'll ignore the error, as
1811 * removing of memory cannot fail.
1812 */
1813 ret = release_mem_region_adjustable(&iomem_resource, start, size);
1814 if (ret) {
1815 resource_size_t endres = start + size - 1;
1816
1817 pr_warn("Unable to release resource <%pa-%pa> (%d)\n",
1818 &start, &endres, ret);
1819 }
1820}
1821
Toshi Kani0f1cfe92013-09-11 14:21:50 -07001822/**
1823 * remove_memory
Mike Rapoporte8b098f2018-04-05 16:24:57 -07001824 * @nid: the node ID
1825 * @start: physical address of the region to remove
1826 * @size: size of the region to remove
Toshi Kani0f1cfe92013-09-11 14:21:50 -07001827 *
1828 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1829 * and online/offline operations before this call, as required by
1830 * try_offline_node().
1831 */
David Hildenbrandd15e5922018-10-30 15:10:18 -07001832void __ref __remove_memory(int nid, u64 start, u64 size)
Wen Congyangbbc76be2013-02-22 16:32:54 -08001833{
Rafael J. Wysocki242831e2013-05-27 12:58:46 +02001834 int ret;
Wen Congyang993c1aa2013-02-22 16:32:50 -08001835
Toshi Kani27356f52013-09-11 14:21:49 -07001836 BUG_ON(check_hotplug_memory_range(start, size));
1837
Vladimir Davydovbfc8c902014-06-04 16:07:18 -07001838 mem_hotplug_begin();
Yasuaki Ishimatsu6677e3e2013-02-22 16:32:52 -08001839
1840 /*
Rafael J. Wysocki242831e2013-05-27 12:58:46 +02001841 * All memory blocks must be offlined before removing memory. Check
1842 * whether all memory blocks in question are offline and trigger a BUG()
1843 * if this is not the case.
Yasuaki Ishimatsu6677e3e2013-02-22 16:32:52 -08001844 */
Rafael J. Wysocki242831e2013-05-27 12:58:46 +02001845 ret = walk_memory_range(PFN_DOWN(start), PFN_UP(start + size - 1), NULL,
Xishi Qiud6de9d52013-11-12 15:07:20 -08001846 check_memblock_offlined_cb);
Vladimir Davydovbfc8c902014-06-04 16:07:18 -07001847 if (ret)
Rafael J. Wysocki242831e2013-05-27 12:58:46 +02001848 BUG();
Yasuaki Ishimatsu6677e3e2013-02-22 16:32:52 -08001849
Yasuaki Ishimatsu46c66c42013-02-22 16:32:56 -08001850 /* remove memmap entry */
1851 firmware_map_remove(start, start + size, "System RAM");
Xishi Qiuf9126ab2015-08-14 15:35:16 -07001852 memblock_free(start, size);
1853 memblock_remove(start, size);
Yasuaki Ishimatsu46c66c42013-02-22 16:32:56 -08001854
Oscar Salvador2c2a5af2018-12-28 00:36:22 -08001855 arch_remove_memory(nid, start, size, NULL);
David Hildenbrandd9eb1412019-05-13 17:21:33 -07001856 __release_memory_resource(start, size);
Wen Congyang24d335c2013-02-22 16:32:58 -08001857
Tang Chen60a5a192013-02-22 16:33:14 -08001858 try_offline_node(nid);
1859
Vladimir Davydovbfc8c902014-06-04 16:07:18 -07001860 mem_hotplug_done();
Badari Pulavarty71088782008-10-18 20:25:58 -07001861}
David Hildenbrandd15e5922018-10-30 15:10:18 -07001862
1863void remove_memory(int nid, u64 start, u64 size)
1864{
1865 lock_device_hotplug();
1866 __remove_memory(nid, start, size);
1867 unlock_device_hotplug();
1868}
Badari Pulavarty71088782008-10-18 20:25:58 -07001869EXPORT_SYMBOL_GPL(remove_memory);
Rafael J. Wysockiaba6efc2013-06-01 22:24:07 +02001870#endif /* CONFIG_MEMORY_HOTREMOVE */