blob: aa34431c3f31769b05a43820aca0a75c6975e8a3 [file] [log] [blame]
Dave Hansen3947be12005-10-29 18:16:54 -07001/*
2 * linux/mm/memory_hotplug.c
3 *
4 * Copyright (C)
5 */
6
Dave Hansen3947be12005-10-29 18:16:54 -07007#include <linux/stddef.h>
8#include <linux/mm.h>
9#include <linux/swap.h>
10#include <linux/interrupt.h>
11#include <linux/pagemap.h>
Dave Hansen3947be12005-10-29 18:16:54 -070012#include <linux/compiler.h>
Paul Gortmakerb95f1b312011-10-16 02:01:52 -040013#include <linux/export.h>
Dave Hansen3947be12005-10-29 18:16:54 -070014#include <linux/pagevec.h>
Chandra Seetharaman2d1d43f2006-09-29 02:01:25 -070015#include <linux/writeback.h>
Dave Hansen3947be12005-10-29 18:16:54 -070016#include <linux/slab.h>
17#include <linux/sysctl.h>
18#include <linux/cpu.h>
19#include <linux/memory.h>
Dan Williams4b94ffd2016-01-15 16:56:22 -080020#include <linux/memremap.h>
Dave Hansen3947be12005-10-29 18:16:54 -070021#include <linux/memory_hotplug.h>
22#include <linux/highmem.h>
23#include <linux/vmalloc.h>
KAMEZAWA Hiroyuki0a547032006-06-27 02:53:35 -070024#include <linux/ioport.h>
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -070025#include <linux/delay.h>
26#include <linux/migrate.h>
27#include <linux/page-isolation.h>
Badari Pulavarty71088782008-10-18 20:25:58 -070028#include <linux/pfn.h>
Andi Kleen6ad696d2009-11-17 14:06:22 -080029#include <linux/suspend.h>
KOSAKI Motohiro6d9c2852009-12-14 17:58:11 -080030#include <linux/mm_inline.h>
akpm@linux-foundation.orgd96ae532010-03-05 13:41:58 -080031#include <linux/firmware-map.h>
Tang Chen60a5a192013-02-22 16:33:14 -080032#include <linux/stop_machine.h>
Naoya Horiguchic8721bb2013-09-11 14:22:09 -070033#include <linux/hugetlb.h>
Tang Chenc5320922013-11-12 15:08:10 -080034#include <linux/memblock.h>
Tang Chenf784a3f2014-11-13 15:19:39 -080035#include <linux/bootmem.h>
Vlastimil Babka698b1b32016-03-17 14:18:08 -070036#include <linux/compaction.h>
Dave Hansen3947be12005-10-29 18:16:54 -070037
38#include <asm/tlbflush.h>
39
Adrian Bunk1e5ad9a2008-04-28 20:40:08 +030040#include "internal.h"
41
Daniel Kiper9d0ad8c2011-07-25 17:12:05 -070042/*
43 * online_page_callback contains pointer to current page onlining function.
44 * Initially it is generic_online_page(). If it is required it could be
45 * changed by calling set_online_page_callback() for callback registration
46 * and restore_online_page_callback() for generic callback restore.
47 */
48
49static void generic_online_page(struct page *page);
50
51static online_page_callback_t online_page_callback = generic_online_page;
Vladimir Davydovbfc8c902014-06-04 16:07:18 -070052static DEFINE_MUTEX(online_page_callback_lock);
Daniel Kiper9d0ad8c2011-07-25 17:12:05 -070053
Vladimir Davydovbfc8c902014-06-04 16:07:18 -070054/* The same as the cpu_hotplug lock, but for memory hotplug. */
55static struct {
56 struct task_struct *active_writer;
57 struct mutex lock; /* Synchronizes accesses to refcount, */
58 /*
59 * Also blocks the new readers during
60 * an ongoing mem hotplug operation.
61 */
62 int refcount;
KOSAKI Motohiro20d6c962010-12-02 14:31:19 -080063
Vladimir Davydovbfc8c902014-06-04 16:07:18 -070064#ifdef CONFIG_DEBUG_LOCK_ALLOC
65 struct lockdep_map dep_map;
66#endif
67} mem_hotplug = {
68 .active_writer = NULL,
69 .lock = __MUTEX_INITIALIZER(mem_hotplug.lock),
70 .refcount = 0,
71#ifdef CONFIG_DEBUG_LOCK_ALLOC
72 .dep_map = {.name = "mem_hotplug.lock" },
73#endif
74};
75
76/* Lockdep annotations for get/put_online_mems() and mem_hotplug_begin/end() */
77#define memhp_lock_acquire_read() lock_map_acquire_read(&mem_hotplug.dep_map)
78#define memhp_lock_acquire() lock_map_acquire(&mem_hotplug.dep_map)
79#define memhp_lock_release() lock_map_release(&mem_hotplug.dep_map)
80
Vitaly Kuznetsov31bc3852016-03-15 14:56:48 -070081bool memhp_auto_online;
82EXPORT_SYMBOL_GPL(memhp_auto_online);
83
Vladimir Davydovbfc8c902014-06-04 16:07:18 -070084void get_online_mems(void)
KOSAKI Motohiro20d6c962010-12-02 14:31:19 -080085{
Vladimir Davydovbfc8c902014-06-04 16:07:18 -070086 might_sleep();
87 if (mem_hotplug.active_writer == current)
88 return;
89 memhp_lock_acquire_read();
90 mutex_lock(&mem_hotplug.lock);
91 mem_hotplug.refcount++;
92 mutex_unlock(&mem_hotplug.lock);
93
KOSAKI Motohiro20d6c962010-12-02 14:31:19 -080094}
95
Vladimir Davydovbfc8c902014-06-04 16:07:18 -070096void put_online_mems(void)
KOSAKI Motohiro20d6c962010-12-02 14:31:19 -080097{
Vladimir Davydovbfc8c902014-06-04 16:07:18 -070098 if (mem_hotplug.active_writer == current)
99 return;
100 mutex_lock(&mem_hotplug.lock);
101
102 if (WARN_ON(!mem_hotplug.refcount))
103 mem_hotplug.refcount++; /* try to fix things up */
104
105 if (!--mem_hotplug.refcount && unlikely(mem_hotplug.active_writer))
106 wake_up_process(mem_hotplug.active_writer);
107 mutex_unlock(&mem_hotplug.lock);
108 memhp_lock_release();
109
KOSAKI Motohiro20d6c962010-12-02 14:31:19 -0800110}
111
David Rientjes30467e02015-04-14 15:45:11 -0700112void mem_hotplug_begin(void)
Vladimir Davydovbfc8c902014-06-04 16:07:18 -0700113{
114 mem_hotplug.active_writer = current;
115
116 memhp_lock_acquire();
117 for (;;) {
118 mutex_lock(&mem_hotplug.lock);
119 if (likely(!mem_hotplug.refcount))
120 break;
121 __set_current_state(TASK_UNINTERRUPTIBLE);
122 mutex_unlock(&mem_hotplug.lock);
123 schedule();
124 }
125}
126
David Rientjes30467e02015-04-14 15:45:11 -0700127void mem_hotplug_done(void)
Vladimir Davydovbfc8c902014-06-04 16:07:18 -0700128{
129 mem_hotplug.active_writer = NULL;
130 mutex_unlock(&mem_hotplug.lock);
131 memhp_lock_release();
132}
KOSAKI Motohiro20d6c962010-12-02 14:31:19 -0800133
Keith Mannthey45e0b782006-09-30 23:27:09 -0700134/* add this memory to iomem resource */
135static struct resource *register_memory_resource(u64 start, u64 size)
136{
137 struct resource *res;
138 res = kzalloc(sizeof(struct resource), GFP_KERNEL);
Vitaly Kuznetsov6f754ba2016-01-14 15:21:55 -0800139 if (!res)
140 return ERR_PTR(-ENOMEM);
Keith Mannthey45e0b782006-09-30 23:27:09 -0700141
142 res->name = "System RAM";
143 res->start = start;
144 res->end = start + size - 1;
Toshi Kani782b8662016-01-26 21:57:24 +0100145 res->flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
Keith Mannthey45e0b782006-09-30 23:27:09 -0700146 if (request_resource(&iomem_resource, res) < 0) {
Toshi Kani4996eed2013-07-03 15:02:39 -0700147 pr_debug("System RAM resource %pR cannot be added\n", res);
Keith Mannthey45e0b782006-09-30 23:27:09 -0700148 kfree(res);
Vitaly Kuznetsov6f754ba2016-01-14 15:21:55 -0800149 return ERR_PTR(-EEXIST);
Keith Mannthey45e0b782006-09-30 23:27:09 -0700150 }
151 return res;
152}
153
154static void release_memory_resource(struct resource *res)
155{
156 if (!res)
157 return;
158 release_resource(res);
159 kfree(res);
160 return;
161}
162
Keith Mannthey53947022006-09-30 23:27:08 -0700163#ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
Yasuaki Ishimatsu46723bf2013-02-22 16:33:00 -0800164void get_page_bootmem(unsigned long info, struct page *page,
165 unsigned long type)
Yasunori Goto04753272008-04-28 02:13:31 -0700166{
Andrea Arcangeli5f24ce52011-01-13 15:47:00 -0800167 page->lru.next = (struct list_head *) type;
Yasunori Goto04753272008-04-28 02:13:31 -0700168 SetPagePrivate(page);
169 set_page_private(page, info);
Joonsoo Kimfe896d12016-03-17 14:19:26 -0700170 page_ref_inc(page);
Yasunori Goto04753272008-04-28 02:13:31 -0700171}
172
Jiang Liu170a5a72013-07-03 15:03:17 -0700173void put_page_bootmem(struct page *page)
Yasunori Goto04753272008-04-28 02:13:31 -0700174{
Andrea Arcangeli5f24ce52011-01-13 15:47:00 -0800175 unsigned long type;
Yasunori Goto04753272008-04-28 02:13:31 -0700176
Andrea Arcangeli5f24ce52011-01-13 15:47:00 -0800177 type = (unsigned long) page->lru.next;
178 BUG_ON(type < MEMORY_HOTPLUG_MIN_BOOTMEM_TYPE ||
179 type > MEMORY_HOTPLUG_MAX_BOOTMEM_TYPE);
Yasunori Goto04753272008-04-28 02:13:31 -0700180
Joonsoo Kimfe896d12016-03-17 14:19:26 -0700181 if (page_ref_dec_return(page) == 1) {
Yasunori Goto04753272008-04-28 02:13:31 -0700182 ClearPagePrivate(page);
183 set_page_private(page, 0);
Andrea Arcangeli5f24ce52011-01-13 15:47:00 -0800184 INIT_LIST_HEAD(&page->lru);
Jiang Liu170a5a72013-07-03 15:03:17 -0700185 free_reserved_page(page);
Yasunori Goto04753272008-04-28 02:13:31 -0700186 }
Yasunori Goto04753272008-04-28 02:13:31 -0700187}
188
Yasuaki Ishimatsu46723bf2013-02-22 16:33:00 -0800189#ifdef CONFIG_HAVE_BOOTMEM_INFO_NODE
190#ifndef CONFIG_SPARSEMEM_VMEMMAP
Adrian Bunkd92bc312008-07-23 21:28:12 -0700191static void register_page_bootmem_info_section(unsigned long start_pfn)
Yasunori Goto04753272008-04-28 02:13:31 -0700192{
193 unsigned long *usemap, mapsize, section_nr, i;
194 struct mem_section *ms;
195 struct page *page, *memmap;
196
Yasunori Goto04753272008-04-28 02:13:31 -0700197 section_nr = pfn_to_section_nr(start_pfn);
198 ms = __nr_to_section(section_nr);
199
200 /* Get section's memmap address */
201 memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
202
203 /*
204 * Get page for the memmap's phys address
205 * XXX: need more consideration for sparse_vmemmap...
206 */
207 page = virt_to_page(memmap);
208 mapsize = sizeof(struct page) * PAGES_PER_SECTION;
209 mapsize = PAGE_ALIGN(mapsize) >> PAGE_SHIFT;
210
211 /* remember memmap's page */
212 for (i = 0; i < mapsize; i++, page++)
213 get_page_bootmem(section_nr, page, SECTION_INFO);
214
215 usemap = __nr_to_section(section_nr)->pageblock_flags;
216 page = virt_to_page(usemap);
217
218 mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT;
219
220 for (i = 0; i < mapsize; i++, page++)
Yasunori Gotoaf370fb2008-07-23 21:28:17 -0700221 get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
Yasunori Goto04753272008-04-28 02:13:31 -0700222
223}
Yasuaki Ishimatsu46723bf2013-02-22 16:33:00 -0800224#else /* CONFIG_SPARSEMEM_VMEMMAP */
225static void register_page_bootmem_info_section(unsigned long start_pfn)
226{
227 unsigned long *usemap, mapsize, section_nr, i;
228 struct mem_section *ms;
229 struct page *page, *memmap;
230
231 if (!pfn_valid(start_pfn))
232 return;
233
234 section_nr = pfn_to_section_nr(start_pfn);
235 ms = __nr_to_section(section_nr);
236
237 memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
238
239 register_page_bootmem_memmap(section_nr, memmap, PAGES_PER_SECTION);
240
241 usemap = __nr_to_section(section_nr)->pageblock_flags;
242 page = virt_to_page(usemap);
243
244 mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT;
245
246 for (i = 0; i < mapsize; i++, page++)
247 get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
248}
249#endif /* !CONFIG_SPARSEMEM_VMEMMAP */
Yasunori Goto04753272008-04-28 02:13:31 -0700250
251void register_page_bootmem_info_node(struct pglist_data *pgdat)
252{
253 unsigned long i, pfn, end_pfn, nr_pages;
254 int node = pgdat->node_id;
255 struct page *page;
256 struct zone *zone;
257
258 nr_pages = PAGE_ALIGN(sizeof(struct pglist_data)) >> PAGE_SHIFT;
259 page = virt_to_page(pgdat);
260
261 for (i = 0; i < nr_pages; i++, page++)
262 get_page_bootmem(node, page, NODE_INFO);
263
264 zone = &pgdat->node_zones[0];
265 for (; zone < pgdat->node_zones + MAX_NR_ZONES - 1; zone++) {
Xishi Qiu139c2d72013-09-11 14:21:46 -0700266 if (zone_is_initialized(zone)) {
Yasunori Goto04753272008-04-28 02:13:31 -0700267 nr_pages = zone->wait_table_hash_nr_entries
268 * sizeof(wait_queue_head_t);
269 nr_pages = PAGE_ALIGN(nr_pages) >> PAGE_SHIFT;
270 page = virt_to_page(zone->wait_table);
271
272 for (i = 0; i < nr_pages; i++, page++)
273 get_page_bootmem(node, page, NODE_INFO);
274 }
275 }
276
277 pfn = pgdat->node_start_pfn;
Cody P Schaferc1f19492013-02-22 16:35:32 -0800278 end_pfn = pgdat_end_pfn(pgdat);
Yasunori Goto04753272008-04-28 02:13:31 -0700279
Tang Chen7e9f5eb2013-07-08 16:00:23 -0700280 /* register section info */
qiuxishif14851a2012-09-17 14:09:24 -0700281 for (; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
282 /*
283 * Some platforms can assign the same pfn to multiple nodes - on
284 * node0 as well as nodeN. To avoid registering a pfn against
285 * multiple nodes we check that this pfn does not already
Tang Chen7e9f5eb2013-07-08 16:00:23 -0700286 * reside in some other nodes.
qiuxishif14851a2012-09-17 14:09:24 -0700287 */
288 if (pfn_valid(pfn) && (pfn_to_nid(pfn) == node))
289 register_page_bootmem_info_section(pfn);
290 }
Yasunori Goto04753272008-04-28 02:13:31 -0700291}
Yasuaki Ishimatsu46723bf2013-02-22 16:33:00 -0800292#endif /* CONFIG_HAVE_BOOTMEM_INFO_NODE */
Yasunori Goto04753272008-04-28 02:13:31 -0700293
Fabian Frederickf2765402014-08-06 16:04:57 -0700294static void __meminit grow_zone_span(struct zone *zone, unsigned long start_pfn,
295 unsigned long end_pfn)
Heiko Carstens76cdd582008-05-14 16:05:52 -0700296{
297 unsigned long old_zone_end_pfn;
298
299 zone_span_writelock(zone);
300
Xishi Qiuc33bc312013-09-11 14:21:44 -0700301 old_zone_end_pfn = zone_end_pfn(zone);
Xishi Qiu8080fc02013-09-11 14:21:45 -0700302 if (zone_is_empty(zone) || start_pfn < zone->zone_start_pfn)
Heiko Carstens76cdd582008-05-14 16:05:52 -0700303 zone->zone_start_pfn = start_pfn;
304
305 zone->spanned_pages = max(old_zone_end_pfn, end_pfn) -
306 zone->zone_start_pfn;
307
308 zone_span_writeunlock(zone);
309}
310
Lai Jiangshan511c2ab2012-12-11 16:03:16 -0800311static void resize_zone(struct zone *zone, unsigned long start_pfn,
312 unsigned long end_pfn)
313{
314 zone_span_writelock(zone);
315
Lai Jiangshane455a9b2012-12-11 16:03:20 -0800316 if (end_pfn - start_pfn) {
317 zone->zone_start_pfn = start_pfn;
318 zone->spanned_pages = end_pfn - start_pfn;
319 } else {
320 /*
321 * make it consist as free_area_init_core(),
322 * if spanned_pages = 0, then keep start_pfn = 0
323 */
324 zone->zone_start_pfn = 0;
325 zone->spanned_pages = 0;
326 }
Lai Jiangshan511c2ab2012-12-11 16:03:16 -0800327
328 zone_span_writeunlock(zone);
329}
330
331static void fix_zone_id(struct zone *zone, unsigned long start_pfn,
332 unsigned long end_pfn)
333{
334 enum zone_type zid = zone_idx(zone);
335 int nid = zone->zone_pgdat->node_id;
336 unsigned long pfn;
337
338 for (pfn = start_pfn; pfn < end_pfn; pfn++)
339 set_page_links(pfn_to_page(pfn), zid, nid, pfn);
340}
341
Cody P Schaferf6bbb782013-02-22 16:35:30 -0800342/* Can fail with -ENOMEM from allocating a wait table with vmalloc() or
Santosh Shilimkar9e43aa22014-01-21 15:50:43 -0800343 * alloc_bootmem_node_nopanic()/memblock_virt_alloc_node_nopanic() */
Cody P Schaferf6bbb782013-02-22 16:35:30 -0800344static int __ref ensure_zone_is_initialized(struct zone *zone,
345 unsigned long start_pfn, unsigned long num_pages)
346{
347 if (!zone_is_initialized(zone))
Yaowei Baib171e402015-11-05 18:47:06 -0800348 return init_currently_empty_zone(zone, start_pfn, num_pages);
349
Cody P Schaferf6bbb782013-02-22 16:35:30 -0800350 return 0;
351}
352
Lai Jiangshane455a9b2012-12-11 16:03:20 -0800353static int __meminit move_pfn_range_left(struct zone *z1, struct zone *z2,
Lai Jiangshan511c2ab2012-12-11 16:03:16 -0800354 unsigned long start_pfn, unsigned long end_pfn)
355{
Lai Jiangshane455a9b2012-12-11 16:03:20 -0800356 int ret;
Lai Jiangshan511c2ab2012-12-11 16:03:16 -0800357 unsigned long flags;
Lai Jiangshane455a9b2012-12-11 16:03:20 -0800358 unsigned long z1_start_pfn;
359
Cody P Schafer64dd1b22013-02-22 16:35:31 -0800360 ret = ensure_zone_is_initialized(z1, start_pfn, end_pfn - start_pfn);
361 if (ret)
362 return ret;
Lai Jiangshan511c2ab2012-12-11 16:03:16 -0800363
364 pgdat_resize_lock(z1->zone_pgdat, &flags);
365
366 /* can't move pfns which are higher than @z2 */
Cody P Schafer108bcc92013-02-22 16:35:23 -0800367 if (end_pfn > zone_end_pfn(z2))
Lai Jiangshan511c2ab2012-12-11 16:03:16 -0800368 goto out_fail;
Jiang Liu834405c2013-07-03 15:03:04 -0700369 /* the move out part must be at the left most of @z2 */
Lai Jiangshan511c2ab2012-12-11 16:03:16 -0800370 if (start_pfn > z2->zone_start_pfn)
371 goto out_fail;
372 /* must included/overlap */
373 if (end_pfn <= z2->zone_start_pfn)
374 goto out_fail;
375
Lai Jiangshane455a9b2012-12-11 16:03:20 -0800376 /* use start_pfn for z1's start_pfn if z1 is empty */
Xishi Qiu8080fc02013-09-11 14:21:45 -0700377 if (!zone_is_empty(z1))
Lai Jiangshane455a9b2012-12-11 16:03:20 -0800378 z1_start_pfn = z1->zone_start_pfn;
379 else
380 z1_start_pfn = start_pfn;
381
382 resize_zone(z1, z1_start_pfn, end_pfn);
Cody P Schafer108bcc92013-02-22 16:35:23 -0800383 resize_zone(z2, end_pfn, zone_end_pfn(z2));
Lai Jiangshan511c2ab2012-12-11 16:03:16 -0800384
385 pgdat_resize_unlock(z1->zone_pgdat, &flags);
386
387 fix_zone_id(z1, start_pfn, end_pfn);
388
389 return 0;
390out_fail:
391 pgdat_resize_unlock(z1->zone_pgdat, &flags);
392 return -1;
393}
394
Lai Jiangshane455a9b2012-12-11 16:03:20 -0800395static int __meminit move_pfn_range_right(struct zone *z1, struct zone *z2,
Lai Jiangshan511c2ab2012-12-11 16:03:16 -0800396 unsigned long start_pfn, unsigned long end_pfn)
397{
Lai Jiangshane455a9b2012-12-11 16:03:20 -0800398 int ret;
Lai Jiangshan511c2ab2012-12-11 16:03:16 -0800399 unsigned long flags;
Lai Jiangshane455a9b2012-12-11 16:03:20 -0800400 unsigned long z2_end_pfn;
401
Cody P Schafer64dd1b22013-02-22 16:35:31 -0800402 ret = ensure_zone_is_initialized(z2, start_pfn, end_pfn - start_pfn);
403 if (ret)
404 return ret;
Lai Jiangshan511c2ab2012-12-11 16:03:16 -0800405
406 pgdat_resize_lock(z1->zone_pgdat, &flags);
407
408 /* can't move pfns which are lower than @z1 */
409 if (z1->zone_start_pfn > start_pfn)
410 goto out_fail;
411 /* the move out part mast at the right most of @z1 */
Cody P Schafer108bcc92013-02-22 16:35:23 -0800412 if (zone_end_pfn(z1) > end_pfn)
Lai Jiangshan511c2ab2012-12-11 16:03:16 -0800413 goto out_fail;
414 /* must included/overlap */
Cody P Schafer108bcc92013-02-22 16:35:23 -0800415 if (start_pfn >= zone_end_pfn(z1))
Lai Jiangshan511c2ab2012-12-11 16:03:16 -0800416 goto out_fail;
417
Lai Jiangshane455a9b2012-12-11 16:03:20 -0800418 /* use end_pfn for z2's end_pfn if z2 is empty */
Xishi Qiu8080fc02013-09-11 14:21:45 -0700419 if (!zone_is_empty(z2))
Cody P Schafer108bcc92013-02-22 16:35:23 -0800420 z2_end_pfn = zone_end_pfn(z2);
Lai Jiangshane455a9b2012-12-11 16:03:20 -0800421 else
422 z2_end_pfn = end_pfn;
423
Lai Jiangshan511c2ab2012-12-11 16:03:16 -0800424 resize_zone(z1, z1->zone_start_pfn, start_pfn);
Lai Jiangshane455a9b2012-12-11 16:03:20 -0800425 resize_zone(z2, start_pfn, z2_end_pfn);
Lai Jiangshan511c2ab2012-12-11 16:03:16 -0800426
427 pgdat_resize_unlock(z1->zone_pgdat, &flags);
428
429 fix_zone_id(z2, start_pfn, end_pfn);
430
431 return 0;
432out_fail:
433 pgdat_resize_unlock(z1->zone_pgdat, &flags);
434 return -1;
435}
436
Fabian Frederickf2765402014-08-06 16:04:57 -0700437static void __meminit grow_pgdat_span(struct pglist_data *pgdat, unsigned long start_pfn,
438 unsigned long end_pfn)
Heiko Carstens76cdd582008-05-14 16:05:52 -0700439{
Xishi Qiu83285c72013-11-12 15:07:19 -0800440 unsigned long old_pgdat_end_pfn = pgdat_end_pfn(pgdat);
Heiko Carstens76cdd582008-05-14 16:05:52 -0700441
Tang Chen712cd382012-12-11 16:01:07 -0800442 if (!pgdat->node_spanned_pages || start_pfn < pgdat->node_start_pfn)
Heiko Carstens76cdd582008-05-14 16:05:52 -0700443 pgdat->node_start_pfn = start_pfn;
444
445 pgdat->node_spanned_pages = max(old_pgdat_end_pfn, end_pfn) -
446 pgdat->node_start_pfn;
447}
448
Al Viro31168482008-11-22 17:33:24 +0000449static int __meminit __add_zone(struct zone *zone, unsigned long phys_start_pfn)
Dave Hansen3947be12005-10-29 18:16:54 -0700450{
451 struct pglist_data *pgdat = zone->zone_pgdat;
452 int nr_pages = PAGES_PER_SECTION;
453 int nid = pgdat->node_id;
454 int zone_type;
Mel Gormane298ff72015-08-06 15:46:51 -0700455 unsigned long flags, pfn;
Cody P Schafer64dd1b22013-02-22 16:35:31 -0800456 int ret;
Dave Hansen3947be12005-10-29 18:16:54 -0700457
458 zone_type = zone - pgdat->node_zones;
Cody P Schafer64dd1b22013-02-22 16:35:31 -0800459 ret = ensure_zone_is_initialized(zone, phys_start_pfn, nr_pages);
460 if (ret)
461 return ret;
Heiko Carstens76cdd582008-05-14 16:05:52 -0700462
Heiko Carstens76cdd582008-05-14 16:05:52 -0700463 pgdat_resize_lock(zone->zone_pgdat, &flags);
464 grow_zone_span(zone, phys_start_pfn, phys_start_pfn + nr_pages);
465 grow_pgdat_span(zone->zone_pgdat, phys_start_pfn,
466 phys_start_pfn + nr_pages);
467 pgdat_resize_unlock(zone->zone_pgdat, &flags);
Dave Hansena2f3aa022007-01-10 23:15:30 -0800468 memmap_init_zone(nr_pages, nid, zone_type,
469 phys_start_pfn, MEMMAP_HOTPLUG);
Mel Gormane298ff72015-08-06 15:46:51 -0700470
471 /* online_page_range is called later and expects pages reserved */
472 for (pfn = phys_start_pfn; pfn < phys_start_pfn + nr_pages; pfn++) {
473 if (!pfn_valid(pfn))
474 continue;
475
476 SetPageReserved(pfn_to_page(pfn));
477 }
Yasunori Goto718127c2006-06-23 02:03:10 -0700478 return 0;
Dave Hansen3947be12005-10-29 18:16:54 -0700479}
480
Gary Hadec04fc582009-01-06 14:39:14 -0800481static int __meminit __add_section(int nid, struct zone *zone,
482 unsigned long phys_start_pfn)
Dave Hansen3947be12005-10-29 18:16:54 -0700483{
Dave Hansen3947be12005-10-29 18:16:54 -0700484 int ret;
485
KAMEZAWA Hiroyukiebd15302006-08-05 12:15:06 -0700486 if (pfn_valid(phys_start_pfn))
487 return -EEXIST;
488
Zhang Yanfei85b35fe2013-11-12 15:07:42 -0800489 ret = sparse_add_one_section(zone, phys_start_pfn);
Dave Hansen3947be12005-10-29 18:16:54 -0700490
491 if (ret < 0)
492 return ret;
493
Yasunori Goto718127c2006-06-23 02:03:10 -0700494 ret = __add_zone(zone, phys_start_pfn);
495
496 if (ret < 0)
497 return ret;
498
Gary Hadec04fc582009-01-06 14:39:14 -0800499 return register_new_memory(nid, __pfn_to_section(phys_start_pfn));
Dave Hansen3947be12005-10-29 18:16:54 -0700500}
501
David Rientjes4edd7ce2013-04-29 15:08:22 -0700502/*
503 * Reasonably generic function for adding memory. It is
504 * expected that archs that support memory hotplug will
505 * call this function after deciding the zone to which to
506 * add the new pages.
507 */
508int __ref __add_pages(int nid, struct zone *zone, unsigned long phys_start_pfn,
509 unsigned long nr_pages)
510{
511 unsigned long i;
512 int err = 0;
513 int start_sec, end_sec;
Dan Williams4b94ffd2016-01-15 16:56:22 -0800514 struct vmem_altmap *altmap;
515
Joonsoo Kim7cf91a92016-03-15 14:57:51 -0700516 clear_zone_contiguous(zone);
517
David Rientjes4edd7ce2013-04-29 15:08:22 -0700518 /* during initialize mem_map, align hot-added range to section */
519 start_sec = pfn_to_section_nr(phys_start_pfn);
520 end_sec = pfn_to_section_nr(phys_start_pfn + nr_pages - 1);
521
Dan Williams4b94ffd2016-01-15 16:56:22 -0800522 altmap = to_vmem_altmap((unsigned long) pfn_to_page(phys_start_pfn));
523 if (altmap) {
524 /*
525 * Validate altmap is within bounds of the total request
526 */
527 if (altmap->base_pfn != phys_start_pfn
528 || vmem_altmap_offset(altmap) > nr_pages) {
529 pr_warn_once("memory add fail, invalid altmap\n");
Joonsoo Kim7cf91a92016-03-15 14:57:51 -0700530 err = -EINVAL;
531 goto out;
Dan Williams4b94ffd2016-01-15 16:56:22 -0800532 }
533 altmap->alloc = 0;
534 }
535
David Rientjes4edd7ce2013-04-29 15:08:22 -0700536 for (i = start_sec; i <= end_sec; i++) {
Sheng Yong19c07d52015-04-14 15:44:54 -0700537 err = __add_section(nid, zone, section_nr_to_pfn(i));
David Rientjes4edd7ce2013-04-29 15:08:22 -0700538
539 /*
540 * EEXIST is finally dealt with by ioresource collision
541 * check. see add_memory() => register_memory_resource()
542 * Warning will be printed if there is collision.
543 */
544 if (err && (err != -EEXIST))
545 break;
546 err = 0;
547 }
Zhu Guihuac435a392015-06-24 16:58:42 -0700548 vmemmap_populate_print_last();
Joonsoo Kim7cf91a92016-03-15 14:57:51 -0700549out:
550 set_zone_contiguous(zone);
David Rientjes4edd7ce2013-04-29 15:08:22 -0700551 return err;
552}
553EXPORT_SYMBOL_GPL(__add_pages);
554
555#ifdef CONFIG_MEMORY_HOTREMOVE
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800556/* find the smallest valid pfn in the range [start_pfn, end_pfn) */
557static int find_smallest_section_pfn(int nid, struct zone *zone,
558 unsigned long start_pfn,
559 unsigned long end_pfn)
560{
561 struct mem_section *ms;
562
563 for (; start_pfn < end_pfn; start_pfn += PAGES_PER_SECTION) {
564 ms = __pfn_to_section(start_pfn);
565
566 if (unlikely(!valid_section(ms)))
567 continue;
568
569 if (unlikely(pfn_to_nid(start_pfn) != nid))
570 continue;
571
572 if (zone && zone != page_zone(pfn_to_page(start_pfn)))
573 continue;
574
575 return start_pfn;
576 }
577
578 return 0;
579}
580
581/* find the biggest valid pfn in the range [start_pfn, end_pfn). */
582static int find_biggest_section_pfn(int nid, struct zone *zone,
583 unsigned long start_pfn,
584 unsigned long end_pfn)
585{
586 struct mem_section *ms;
587 unsigned long pfn;
588
589 /* pfn is the end pfn of a memory section. */
590 pfn = end_pfn - 1;
591 for (; pfn >= start_pfn; pfn -= PAGES_PER_SECTION) {
592 ms = __pfn_to_section(pfn);
593
594 if (unlikely(!valid_section(ms)))
595 continue;
596
597 if (unlikely(pfn_to_nid(pfn) != nid))
598 continue;
599
600 if (zone && zone != page_zone(pfn_to_page(pfn)))
601 continue;
602
603 return pfn;
604 }
605
606 return 0;
607}
608
609static void shrink_zone_span(struct zone *zone, unsigned long start_pfn,
610 unsigned long end_pfn)
611{
Xishi Qiuc33bc312013-09-11 14:21:44 -0700612 unsigned long zone_start_pfn = zone->zone_start_pfn;
613 unsigned long z = zone_end_pfn(zone); /* zone_end_pfn namespace clash */
614 unsigned long zone_end_pfn = z;
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800615 unsigned long pfn;
616 struct mem_section *ms;
617 int nid = zone_to_nid(zone);
618
619 zone_span_writelock(zone);
620 if (zone_start_pfn == start_pfn) {
621 /*
622 * If the section is smallest section in the zone, it need
623 * shrink zone->zone_start_pfn and zone->zone_spanned_pages.
624 * In this case, we find second smallest valid mem_section
625 * for shrinking zone.
626 */
627 pfn = find_smallest_section_pfn(nid, zone, end_pfn,
628 zone_end_pfn);
629 if (pfn) {
630 zone->zone_start_pfn = pfn;
631 zone->spanned_pages = zone_end_pfn - pfn;
632 }
633 } else if (zone_end_pfn == end_pfn) {
634 /*
635 * If the section is biggest section in the zone, it need
636 * shrink zone->spanned_pages.
637 * In this case, we find second biggest valid mem_section for
638 * shrinking zone.
639 */
640 pfn = find_biggest_section_pfn(nid, zone, zone_start_pfn,
641 start_pfn);
642 if (pfn)
643 zone->spanned_pages = pfn - zone_start_pfn + 1;
644 }
645
646 /*
647 * The section is not biggest or smallest mem_section in the zone, it
648 * only creates a hole in the zone. So in this case, we need not
649 * change the zone. But perhaps, the zone has only hole data. Thus
650 * it check the zone has only hole or not.
651 */
652 pfn = zone_start_pfn;
653 for (; pfn < zone_end_pfn; pfn += PAGES_PER_SECTION) {
654 ms = __pfn_to_section(pfn);
655
656 if (unlikely(!valid_section(ms)))
657 continue;
658
659 if (page_zone(pfn_to_page(pfn)) != zone)
660 continue;
661
662 /* If the section is current section, it continues the loop */
663 if (start_pfn == pfn)
664 continue;
665
666 /* If we find valid section, we have nothing to do */
667 zone_span_writeunlock(zone);
668 return;
669 }
670
671 /* The zone has no valid section */
672 zone->zone_start_pfn = 0;
673 zone->spanned_pages = 0;
674 zone_span_writeunlock(zone);
675}
676
677static void shrink_pgdat_span(struct pglist_data *pgdat,
678 unsigned long start_pfn, unsigned long end_pfn)
679{
Xishi Qiu83285c72013-11-12 15:07:19 -0800680 unsigned long pgdat_start_pfn = pgdat->node_start_pfn;
681 unsigned long p = pgdat_end_pfn(pgdat); /* pgdat_end_pfn namespace clash */
682 unsigned long pgdat_end_pfn = p;
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800683 unsigned long pfn;
684 struct mem_section *ms;
685 int nid = pgdat->node_id;
686
687 if (pgdat_start_pfn == start_pfn) {
688 /*
689 * If the section is smallest section in the pgdat, it need
690 * shrink pgdat->node_start_pfn and pgdat->node_spanned_pages.
691 * In this case, we find second smallest valid mem_section
692 * for shrinking zone.
693 */
694 pfn = find_smallest_section_pfn(nid, NULL, end_pfn,
695 pgdat_end_pfn);
696 if (pfn) {
697 pgdat->node_start_pfn = pfn;
698 pgdat->node_spanned_pages = pgdat_end_pfn - pfn;
699 }
700 } else if (pgdat_end_pfn == end_pfn) {
701 /*
702 * If the section is biggest section in the pgdat, it need
703 * shrink pgdat->node_spanned_pages.
704 * In this case, we find second biggest valid mem_section for
705 * shrinking zone.
706 */
707 pfn = find_biggest_section_pfn(nid, NULL, pgdat_start_pfn,
708 start_pfn);
709 if (pfn)
710 pgdat->node_spanned_pages = pfn - pgdat_start_pfn + 1;
711 }
712
713 /*
714 * If the section is not biggest or smallest mem_section in the pgdat,
715 * it only creates a hole in the pgdat. So in this case, we need not
716 * change the pgdat.
717 * But perhaps, the pgdat has only hole data. Thus it check the pgdat
718 * has only hole or not.
719 */
720 pfn = pgdat_start_pfn;
721 for (; pfn < pgdat_end_pfn; pfn += PAGES_PER_SECTION) {
722 ms = __pfn_to_section(pfn);
723
724 if (unlikely(!valid_section(ms)))
725 continue;
726
727 if (pfn_to_nid(pfn) != nid)
728 continue;
729
730 /* If the section is current section, it continues the loop */
731 if (start_pfn == pfn)
732 continue;
733
734 /* If we find valid section, we have nothing to do */
735 return;
736 }
737
738 /* The pgdat has no valid section */
739 pgdat->node_start_pfn = 0;
740 pgdat->node_spanned_pages = 0;
741}
742
743static void __remove_zone(struct zone *zone, unsigned long start_pfn)
744{
745 struct pglist_data *pgdat = zone->zone_pgdat;
746 int nr_pages = PAGES_PER_SECTION;
747 int zone_type;
748 unsigned long flags;
749
750 zone_type = zone - pgdat->node_zones;
751
752 pgdat_resize_lock(zone->zone_pgdat, &flags);
753 shrink_zone_span(zone, start_pfn, start_pfn + nr_pages);
754 shrink_pgdat_span(pgdat, start_pfn, start_pfn + nr_pages);
755 pgdat_resize_unlock(zone->zone_pgdat, &flags);
756}
757
Dan Williams4b94ffd2016-01-15 16:56:22 -0800758static int __remove_section(struct zone *zone, struct mem_section *ms,
759 unsigned long map_offset)
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700760{
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800761 unsigned long start_pfn;
762 int scn_nr;
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700763 int ret = -EINVAL;
764
765 if (!valid_section(ms))
766 return ret;
767
768 ret = unregister_memory_section(ms);
769 if (ret)
770 return ret;
771
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800772 scn_nr = __section_nr(ms);
773 start_pfn = section_nr_to_pfn(scn_nr);
774 __remove_zone(zone, start_pfn);
775
Dan Williams4b94ffd2016-01-15 16:56:22 -0800776 sparse_remove_one_section(zone, ms, map_offset);
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700777 return 0;
778}
779
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700780/**
781 * __remove_pages() - remove sections of pages from a zone
782 * @zone: zone from which pages need to be removed
783 * @phys_start_pfn: starting pageframe (must be aligned to start of a section)
784 * @nr_pages: number of pages to remove (must be multiple of section size)
785 *
786 * Generic helper function to remove section mappings and sysfs entries
787 * for the section of the memory we are removing. Caller needs to make
788 * sure that pages are marked reserved and zones are adjust properly by
789 * calling offline_pages().
790 */
791int __remove_pages(struct zone *zone, unsigned long phys_start_pfn,
792 unsigned long nr_pages)
793{
Toshi Kanife74ebb2013-04-29 15:08:20 -0700794 unsigned long i;
Dan Williams4b94ffd2016-01-15 16:56:22 -0800795 unsigned long map_offset = 0;
796 int sections_to_remove, ret = 0;
797
798 /* In the ZONE_DEVICE case device driver owns the memory region */
799 if (is_dev_zone(zone)) {
800 struct page *page = pfn_to_page(phys_start_pfn);
801 struct vmem_altmap *altmap;
802
803 altmap = to_vmem_altmap((unsigned long) page);
804 if (altmap)
805 map_offset = vmem_altmap_offset(altmap);
806 } else {
807 resource_size_t start, size;
808
809 start = phys_start_pfn << PAGE_SHIFT;
810 size = nr_pages * PAGE_SIZE;
811
812 ret = release_mem_region_adjustable(&iomem_resource, start,
813 size);
814 if (ret) {
815 resource_size_t endres = start + size - 1;
816
817 pr_warn("Unable to release resource <%pa-%pa> (%d)\n",
818 &start, &endres, ret);
819 }
820 }
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700821
Joonsoo Kim7cf91a92016-03-15 14:57:51 -0700822 clear_zone_contiguous(zone);
823
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700824 /*
825 * We can only remove entire sections
826 */
827 BUG_ON(phys_start_pfn & ~PAGE_SECTION_MASK);
828 BUG_ON(nr_pages % PAGES_PER_SECTION);
829
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700830 sections_to_remove = nr_pages / PAGES_PER_SECTION;
831 for (i = 0; i < sections_to_remove; i++) {
832 unsigned long pfn = phys_start_pfn + i*PAGES_PER_SECTION;
Dan Williams4b94ffd2016-01-15 16:56:22 -0800833
834 ret = __remove_section(zone, __pfn_to_section(pfn), map_offset);
835 map_offset = 0;
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700836 if (ret)
837 break;
838 }
Joonsoo Kim7cf91a92016-03-15 14:57:51 -0700839
840 set_zone_contiguous(zone);
841
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700842 return ret;
843}
844EXPORT_SYMBOL_GPL(__remove_pages);
David Rientjes4edd7ce2013-04-29 15:08:22 -0700845#endif /* CONFIG_MEMORY_HOTREMOVE */
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700846
Daniel Kiper9d0ad8c2011-07-25 17:12:05 -0700847int set_online_page_callback(online_page_callback_t callback)
848{
849 int rc = -EINVAL;
850
Vladimir Davydovbfc8c902014-06-04 16:07:18 -0700851 get_online_mems();
852 mutex_lock(&online_page_callback_lock);
Daniel Kiper9d0ad8c2011-07-25 17:12:05 -0700853
854 if (online_page_callback == generic_online_page) {
855 online_page_callback = callback;
856 rc = 0;
857 }
858
Vladimir Davydovbfc8c902014-06-04 16:07:18 -0700859 mutex_unlock(&online_page_callback_lock);
860 put_online_mems();
Daniel Kiper9d0ad8c2011-07-25 17:12:05 -0700861
862 return rc;
863}
864EXPORT_SYMBOL_GPL(set_online_page_callback);
865
866int restore_online_page_callback(online_page_callback_t callback)
867{
868 int rc = -EINVAL;
869
Vladimir Davydovbfc8c902014-06-04 16:07:18 -0700870 get_online_mems();
871 mutex_lock(&online_page_callback_lock);
Daniel Kiper9d0ad8c2011-07-25 17:12:05 -0700872
873 if (online_page_callback == callback) {
874 online_page_callback = generic_online_page;
875 rc = 0;
876 }
877
Vladimir Davydovbfc8c902014-06-04 16:07:18 -0700878 mutex_unlock(&online_page_callback_lock);
879 put_online_mems();
Daniel Kiper9d0ad8c2011-07-25 17:12:05 -0700880
881 return rc;
882}
883EXPORT_SYMBOL_GPL(restore_online_page_callback);
884
885void __online_page_set_limits(struct page *page)
Jeremy Fitzhardinge180c06e2008-04-28 02:12:03 -0700886{
Daniel Kiper9d0ad8c2011-07-25 17:12:05 -0700887}
888EXPORT_SYMBOL_GPL(__online_page_set_limits);
889
890void __online_page_increment_counters(struct page *page)
891{
Jiang Liu3dcc0572013-07-03 15:03:21 -0700892 adjust_managed_page_count(page, 1);
Daniel Kiper9d0ad8c2011-07-25 17:12:05 -0700893}
894EXPORT_SYMBOL_GPL(__online_page_increment_counters);
Jeremy Fitzhardinge180c06e2008-04-28 02:12:03 -0700895
Daniel Kiper9d0ad8c2011-07-25 17:12:05 -0700896void __online_page_free(struct page *page)
897{
Jiang Liu3dcc0572013-07-03 15:03:21 -0700898 __free_reserved_page(page);
Jeremy Fitzhardinge180c06e2008-04-28 02:12:03 -0700899}
Daniel Kiper9d0ad8c2011-07-25 17:12:05 -0700900EXPORT_SYMBOL_GPL(__online_page_free);
901
902static void generic_online_page(struct page *page)
903{
904 __online_page_set_limits(page);
905 __online_page_increment_counters(page);
906 __online_page_free(page);
907}
Jeremy Fitzhardinge180c06e2008-04-28 02:12:03 -0700908
KAMEZAWA Hiroyuki75884fb2007-10-16 01:26:10 -0700909static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages,
910 void *arg)
Dave Hansen3947be12005-10-29 18:16:54 -0700911{
912 unsigned long i;
KAMEZAWA Hiroyuki75884fb2007-10-16 01:26:10 -0700913 unsigned long onlined_pages = *(unsigned long *)arg;
914 struct page *page;
915 if (PageReserved(pfn_to_page(start_pfn)))
916 for (i = 0; i < nr_pages; i++) {
917 page = pfn_to_page(start_pfn + i);
Daniel Kiper9d0ad8c2011-07-25 17:12:05 -0700918 (*online_page_callback)(page);
KAMEZAWA Hiroyuki75884fb2007-10-16 01:26:10 -0700919 onlined_pages++;
920 }
921 *(unsigned long *)arg = onlined_pages;
922 return 0;
923}
924
Lai Jiangshan09285af2012-12-12 13:52:04 -0800925#ifdef CONFIG_MOVABLE_NODE
Tang Chen79a4dce2012-12-18 14:23:24 -0800926/*
927 * When CONFIG_MOVABLE_NODE, we permit onlining of a node which doesn't have
928 * normal memory.
929 */
Lai Jiangshan09285af2012-12-12 13:52:04 -0800930static bool can_online_high_movable(struct zone *zone)
931{
932 return true;
933}
Tang Chen79a4dce2012-12-18 14:23:24 -0800934#else /* CONFIG_MOVABLE_NODE */
Lai Jiangshan74d42d82012-12-11 16:03:23 -0800935/* ensure every online node has NORMAL memory */
936static bool can_online_high_movable(struct zone *zone)
937{
938 return node_state(zone_to_nid(zone), N_NORMAL_MEMORY);
939}
Tang Chen79a4dce2012-12-18 14:23:24 -0800940#endif /* CONFIG_MOVABLE_NODE */
Lai Jiangshan74d42d82012-12-11 16:03:23 -0800941
Lai Jiangshand9713672012-12-11 16:01:03 -0800942/* check which state of node_states will be changed when online memory */
943static void node_states_check_changes_online(unsigned long nr_pages,
944 struct zone *zone, struct memory_notify *arg)
945{
946 int nid = zone_to_nid(zone);
947 enum zone_type zone_last = ZONE_NORMAL;
948
949 /*
Lai Jiangshan6715ddf2012-12-12 13:51:49 -0800950 * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY]
951 * contains nodes which have zones of 0...ZONE_NORMAL,
952 * set zone_last to ZONE_NORMAL.
Lai Jiangshand9713672012-12-11 16:01:03 -0800953 *
Lai Jiangshan6715ddf2012-12-12 13:51:49 -0800954 * If we don't have HIGHMEM nor movable node,
955 * node_states[N_NORMAL_MEMORY] contains nodes which have zones of
956 * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE.
Lai Jiangshand9713672012-12-11 16:01:03 -0800957 */
Lai Jiangshan6715ddf2012-12-12 13:51:49 -0800958 if (N_MEMORY == N_NORMAL_MEMORY)
Lai Jiangshand9713672012-12-11 16:01:03 -0800959 zone_last = ZONE_MOVABLE;
960
961 /*
962 * if the memory to be online is in a zone of 0...zone_last, and
963 * the zones of 0...zone_last don't have memory before online, we will
964 * need to set the node to node_states[N_NORMAL_MEMORY] after
965 * the memory is online.
966 */
967 if (zone_idx(zone) <= zone_last && !node_state(nid, N_NORMAL_MEMORY))
968 arg->status_change_nid_normal = nid;
969 else
970 arg->status_change_nid_normal = -1;
971
Lai Jiangshan6715ddf2012-12-12 13:51:49 -0800972#ifdef CONFIG_HIGHMEM
973 /*
974 * If we have movable node, node_states[N_HIGH_MEMORY]
975 * contains nodes which have zones of 0...ZONE_HIGHMEM,
976 * set zone_last to ZONE_HIGHMEM.
977 *
978 * If we don't have movable node, node_states[N_NORMAL_MEMORY]
979 * contains nodes which have zones of 0...ZONE_MOVABLE,
980 * set zone_last to ZONE_MOVABLE.
981 */
982 zone_last = ZONE_HIGHMEM;
983 if (N_MEMORY == N_HIGH_MEMORY)
984 zone_last = ZONE_MOVABLE;
985
986 if (zone_idx(zone) <= zone_last && !node_state(nid, N_HIGH_MEMORY))
987 arg->status_change_nid_high = nid;
988 else
989 arg->status_change_nid_high = -1;
990#else
991 arg->status_change_nid_high = arg->status_change_nid_normal;
992#endif
993
Lai Jiangshand9713672012-12-11 16:01:03 -0800994 /*
995 * if the node don't have memory befor online, we will need to
Lai Jiangshan6715ddf2012-12-12 13:51:49 -0800996 * set the node to node_states[N_MEMORY] after the memory
Lai Jiangshand9713672012-12-11 16:01:03 -0800997 * is online.
998 */
Lai Jiangshan6715ddf2012-12-12 13:51:49 -0800999 if (!node_state(nid, N_MEMORY))
Lai Jiangshand9713672012-12-11 16:01:03 -08001000 arg->status_change_nid = nid;
1001 else
1002 arg->status_change_nid = -1;
1003}
1004
1005static void node_states_set_node(int node, struct memory_notify *arg)
1006{
1007 if (arg->status_change_nid_normal >= 0)
1008 node_set_state(node, N_NORMAL_MEMORY);
1009
Lai Jiangshan6715ddf2012-12-12 13:51:49 -08001010 if (arg->status_change_nid_high >= 0)
1011 node_set_state(node, N_HIGH_MEMORY);
1012
1013 node_set_state(node, N_MEMORY);
Lai Jiangshand9713672012-12-11 16:01:03 -08001014}
1015
KAMEZAWA Hiroyuki75884fb2007-10-16 01:26:10 -07001016
David Rientjes30467e02015-04-14 15:45:11 -07001017/* Must be protected by mem_hotplug_begin() */
Lai Jiangshan511c2ab2012-12-11 16:03:16 -08001018int __ref online_pages(unsigned long pfn, unsigned long nr_pages, int online_type)
KAMEZAWA Hiroyuki75884fb2007-10-16 01:26:10 -07001019{
Cody P Schaferaa472282013-07-03 15:02:10 -07001020 unsigned long flags;
Dave Hansen3947be12005-10-29 18:16:54 -07001021 unsigned long onlined_pages = 0;
1022 struct zone *zone;
Yasunori Goto68113782006-06-23 02:03:11 -07001023 int need_zonelists_rebuild = 0;
Yasunori Goto7b78d332007-10-21 16:41:36 -07001024 int nid;
1025 int ret;
1026 struct memory_notify arg;
Dave Hansen3947be12005-10-29 18:16:54 -07001027
Lai Jiangshand9713672012-12-11 16:01:03 -08001028 /*
1029 * This doesn't need a lock to do pfn_to_page().
1030 * The section can't be removed here because of the
1031 * memory_block->state_mutex.
1032 */
1033 zone = page_zone(pfn_to_page(pfn));
1034
Tang Chen4f7c6b42014-08-06 16:05:13 -07001035 if ((zone_idx(zone) > ZONE_NORMAL ||
1036 online_type == MMOP_ONLINE_MOVABLE) &&
Vladimir Davydovbfc8c902014-06-04 16:07:18 -07001037 !can_online_high_movable(zone))
David Rientjes30467e02015-04-14 15:45:11 -07001038 return -EINVAL;
Lai Jiangshan74d42d82012-12-11 16:03:23 -08001039
Tang Chen4f7c6b42014-08-06 16:05:13 -07001040 if (online_type == MMOP_ONLINE_KERNEL &&
1041 zone_idx(zone) == ZONE_MOVABLE) {
Vladimir Davydovbfc8c902014-06-04 16:07:18 -07001042 if (move_pfn_range_left(zone - 1, zone, pfn, pfn + nr_pages))
David Rientjes30467e02015-04-14 15:45:11 -07001043 return -EINVAL;
Lai Jiangshan511c2ab2012-12-11 16:03:16 -08001044 }
Tang Chen4f7c6b42014-08-06 16:05:13 -07001045 if (online_type == MMOP_ONLINE_MOVABLE &&
1046 zone_idx(zone) == ZONE_MOVABLE - 1) {
Vladimir Davydovbfc8c902014-06-04 16:07:18 -07001047 if (move_pfn_range_right(zone, zone + 1, pfn, pfn + nr_pages))
David Rientjes30467e02015-04-14 15:45:11 -07001048 return -EINVAL;
Lai Jiangshan511c2ab2012-12-11 16:03:16 -08001049 }
1050
1051 /* Previous code may changed the zone of the pfn range */
1052 zone = page_zone(pfn_to_page(pfn));
1053
Yasunori Goto7b78d332007-10-21 16:41:36 -07001054 arg.start_pfn = pfn;
1055 arg.nr_pages = nr_pages;
Lai Jiangshand9713672012-12-11 16:01:03 -08001056 node_states_check_changes_online(nr_pages, zone, &arg);
Yasunori Goto7b78d332007-10-21 16:41:36 -07001057
Vlastimil Babkae888ca32016-03-17 14:18:12 -07001058 nid = zone_to_nid(zone);
Yasunori Goto7b78d332007-10-21 16:41:36 -07001059
1060 ret = memory_notify(MEM_GOING_ONLINE, &arg);
1061 ret = notifier_to_errno(ret);
Chen Yuconge33e33b2016-03-17 14:19:35 -07001062 if (ret)
1063 goto failed_addition;
1064
Dave Hansen3947be12005-10-29 18:16:54 -07001065 /*
Yasunori Goto68113782006-06-23 02:03:11 -07001066 * If this zone is not populated, then it is not in zonelist.
1067 * This means the page allocator ignores this zone.
1068 * So, zonelist must be updated after online.
1069 */
Haicheng Li4eaf3f62010-05-24 14:32:52 -07001070 mutex_lock(&zonelists_mutex);
Wen Congyang6dcd73d2012-12-11 16:01:01 -08001071 if (!populated_zone(zone)) {
Yasunori Goto68113782006-06-23 02:03:11 -07001072 need_zonelists_rebuild = 1;
Wen Congyang6dcd73d2012-12-11 16:01:01 -08001073 build_all_zonelists(NULL, zone);
1074 }
Yasunori Goto68113782006-06-23 02:03:11 -07001075
KAMEZAWA Hiroyuki908eedc2009-09-22 16:45:46 -07001076 ret = walk_system_ram_range(pfn, nr_pages, &onlined_pages,
KAMEZAWA Hiroyuki75884fb2007-10-16 01:26:10 -07001077 online_pages_range);
Geoff Levandfd8a4222008-05-14 16:05:50 -07001078 if (ret) {
Wen Congyang6dcd73d2012-12-11 16:01:01 -08001079 if (need_zonelists_rebuild)
1080 zone_pcp_reset(zone);
Haicheng Li4eaf3f62010-05-24 14:32:52 -07001081 mutex_unlock(&zonelists_mutex);
Chen Yuconge33e33b2016-03-17 14:19:35 -07001082 goto failed_addition;
Geoff Levandfd8a4222008-05-14 16:05:50 -07001083 }
1084
Dave Hansen3947be12005-10-29 18:16:54 -07001085 zone->present_pages += onlined_pages;
Cody P Schaferaa472282013-07-03 15:02:10 -07001086
1087 pgdat_resize_lock(zone->zone_pgdat, &flags);
Yasunori Gotof2937be2006-03-09 17:33:51 -08001088 zone->zone_pgdat->node_present_pages += onlined_pages;
Cody P Schaferaa472282013-07-03 15:02:10 -07001089 pgdat_resize_unlock(zone->zone_pgdat, &flags);
1090
Jiang Liu08dff7b2012-07-31 16:43:30 -07001091 if (onlined_pages) {
Vlastimil Babkae888ca32016-03-17 14:18:12 -07001092 node_states_set_node(nid, &arg);
Jiang Liu08dff7b2012-07-31 16:43:30 -07001093 if (need_zonelists_rebuild)
Wen Congyang6dcd73d2012-12-11 16:01:01 -08001094 build_all_zonelists(NULL, NULL);
Jiang Liu08dff7b2012-07-31 16:43:30 -07001095 else
1096 zone_pcp_update(zone);
1097 }
Dave Hansen3947be12005-10-29 18:16:54 -07001098
Haicheng Li4eaf3f62010-05-24 14:32:52 -07001099 mutex_unlock(&zonelists_mutex);
KOSAKI Motohiro1b79acc2011-05-24 17:11:32 -07001100
1101 init_per_zone_wmark_min();
1102
Vlastimil Babka698b1b32016-03-17 14:18:08 -07001103 if (onlined_pages) {
Vlastimil Babkae888ca32016-03-17 14:18:12 -07001104 kswapd_run(nid);
Vlastimil Babka698b1b32016-03-17 14:18:08 -07001105 kcompactd_run(nid);
1106 }
Dave Hansen61b13992005-10-29 18:16:56 -07001107
Haicheng Li1f522502010-05-24 14:32:51 -07001108 vm_total_pages = nr_free_pagecache_pages();
Kent Liu2f7f24e2008-07-23 21:28:18 -07001109
Chandra Seetharaman2d1d43f2006-09-29 02:01:25 -07001110 writeback_set_ratelimit();
Yasunori Goto7b78d332007-10-21 16:41:36 -07001111
1112 if (onlined_pages)
1113 memory_notify(MEM_ONLINE, &arg);
David Rientjes30467e02015-04-14 15:45:11 -07001114 return 0;
Chen Yuconge33e33b2016-03-17 14:19:35 -07001115
1116failed_addition:
1117 pr_debug("online_pages [mem %#010llx-%#010llx] failed\n",
1118 (unsigned long long) pfn << PAGE_SHIFT,
1119 (((unsigned long long) pfn + nr_pages) << PAGE_SHIFT) - 1);
1120 memory_notify(MEM_CANCEL_ONLINE, &arg);
1121 return ret;
Dave Hansen3947be12005-10-29 18:16:54 -07001122}
Keith Mannthey53947022006-09-30 23:27:08 -07001123#endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
Yasunori Gotobc02af92006-06-27 02:53:30 -07001124
Tang Chen0bd85422014-11-13 15:19:41 -08001125static void reset_node_present_pages(pg_data_t *pgdat)
1126{
1127 struct zone *z;
1128
1129 for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++)
1130 z->present_pages = 0;
1131
1132 pgdat->node_present_pages = 0;
1133}
1134
Hidetoshi Setoe1319332009-11-17 14:06:18 -08001135/* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
1136static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start)
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001137{
1138 struct pglist_data *pgdat;
1139 unsigned long zones_size[MAX_NR_ZONES] = {0};
1140 unsigned long zholes_size[MAX_NR_ZONES] = {0};
Fabian Frederickc8e861a2014-06-04 16:07:51 -07001141 unsigned long start_pfn = PFN_DOWN(start);
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001142
Tang Chena1e565a2013-02-22 16:33:18 -08001143 pgdat = NODE_DATA(nid);
1144 if (!pgdat) {
1145 pgdat = arch_alloc_nodedata(nid);
1146 if (!pgdat)
1147 return NULL;
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001148
Tang Chena1e565a2013-02-22 16:33:18 -08001149 arch_refresh_nodedata(nid, pgdat);
Gu Zhengb0dc3a32015-03-25 15:55:20 -07001150 } else {
1151 /* Reset the nr_zones and classzone_idx to 0 before reuse */
1152 pgdat->nr_zones = 0;
1153 pgdat->classzone_idx = 0;
Tang Chena1e565a2013-02-22 16:33:18 -08001154 }
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001155
1156 /* we can use NODE_DATA(nid) from here */
1157
1158 /* init node's zones as empty zones, we don't have any present pages.*/
Johannes Weiner9109fb72008-07-23 21:27:20 -07001159 free_area_init_node(nid, zones_size, start_pfn, zholes_size);
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001160
KAMEZAWA Hiroyuki959ecc42011-06-15 15:08:38 -07001161 /*
1162 * The node we allocated has no zone fallback lists. For avoiding
1163 * to access not-initialized zonelist, build here.
1164 */
David Rientjesf957db42011-06-22 18:13:04 -07001165 mutex_lock(&zonelists_mutex);
Jiang Liu9adb62a2012-07-31 16:43:28 -07001166 build_all_zonelists(pgdat, NULL);
David Rientjesf957db42011-06-22 18:13:04 -07001167 mutex_unlock(&zonelists_mutex);
KAMEZAWA Hiroyuki959ecc42011-06-15 15:08:38 -07001168
Tang Chenf784a3f2014-11-13 15:19:39 -08001169 /*
1170 * zone->managed_pages is set to an approximate value in
1171 * free_area_init_core(), which will cause
1172 * /sys/device/system/node/nodeX/meminfo has wrong data.
1173 * So reset it to 0 before any memory is onlined.
1174 */
1175 reset_node_managed_pages(pgdat);
1176
Tang Chen0bd85422014-11-13 15:19:41 -08001177 /*
1178 * When memory is hot-added, all the memory is in offline state. So
1179 * clear all zones' present_pages because they will be updated in
1180 * online_pages() and offline_pages().
1181 */
1182 reset_node_present_pages(pgdat);
1183
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001184 return pgdat;
1185}
1186
1187static void rollback_node_hotadd(int nid, pg_data_t *pgdat)
1188{
1189 arch_refresh_nodedata(nid, NULL);
1190 arch_free_nodedata(pgdat);
1191 return;
1192}
1193
KAMEZAWA Hiroyuki0a547032006-06-27 02:53:35 -07001194
Toshi Kani01b0f192013-11-12 15:07:25 -08001195/**
1196 * try_online_node - online a node if offlined
1197 *
minskey guocf234222010-05-24 14:32:41 -07001198 * called by cpu_up() to online a node without onlined memory.
1199 */
Toshi Kani01b0f192013-11-12 15:07:25 -08001200int try_online_node(int nid)
minskey guocf234222010-05-24 14:32:41 -07001201{
1202 pg_data_t *pgdat;
1203 int ret;
1204
Toshi Kani01b0f192013-11-12 15:07:25 -08001205 if (node_online(nid))
1206 return 0;
1207
Vladimir Davydovbfc8c902014-06-04 16:07:18 -07001208 mem_hotplug_begin();
minskey guocf234222010-05-24 14:32:41 -07001209 pgdat = hotadd_new_pgdat(nid, 0);
David Rientjes7553e8f2011-06-22 18:13:01 -07001210 if (!pgdat) {
Toshi Kani01b0f192013-11-12 15:07:25 -08001211 pr_err("Cannot online node %d due to NULL pgdat\n", nid);
minskey guocf234222010-05-24 14:32:41 -07001212 ret = -ENOMEM;
1213 goto out;
1214 }
1215 node_set_online(nid);
1216 ret = register_one_node(nid);
1217 BUG_ON(ret);
1218
Toshi Kani01b0f192013-11-12 15:07:25 -08001219 if (pgdat->node_zonelists->_zonerefs->zone == NULL) {
1220 mutex_lock(&zonelists_mutex);
1221 build_all_zonelists(NULL, NULL);
1222 mutex_unlock(&zonelists_mutex);
1223 }
1224
minskey guocf234222010-05-24 14:32:41 -07001225out:
Vladimir Davydovbfc8c902014-06-04 16:07:18 -07001226 mem_hotplug_done();
minskey guocf234222010-05-24 14:32:41 -07001227 return ret;
1228}
1229
Toshi Kani27356f52013-09-11 14:21:49 -07001230static int check_hotplug_memory_range(u64 start, u64 size)
1231{
Fabian Frederickc8e861a2014-06-04 16:07:51 -07001232 u64 start_pfn = PFN_DOWN(start);
Toshi Kani27356f52013-09-11 14:21:49 -07001233 u64 nr_pages = size >> PAGE_SHIFT;
1234
1235 /* Memory range must be aligned with section */
1236 if ((start_pfn & ~PAGE_SECTION_MASK) ||
1237 (nr_pages % PAGES_PER_SECTION) || (!nr_pages)) {
1238 pr_err("Section-unaligned hotplug range: start 0x%llx, size 0x%llx\n",
1239 (unsigned long long)start,
1240 (unsigned long long)size);
1241 return -EINVAL;
1242 }
1243
1244 return 0;
1245}
1246
Wang Nan63264402014-08-06 16:07:36 -07001247/*
1248 * If movable zone has already been setup, newly added memory should be check.
1249 * If its address is higher than movable zone, it should be added as movable.
1250 * Without this check, movable zone may overlap with other zone.
1251 */
1252static int should_add_memory_movable(int nid, u64 start, u64 size)
1253{
1254 unsigned long start_pfn = start >> PAGE_SHIFT;
1255 pg_data_t *pgdat = NODE_DATA(nid);
1256 struct zone *movable_zone = pgdat->node_zones + ZONE_MOVABLE;
1257
1258 if (zone_is_empty(movable_zone))
1259 return 0;
1260
1261 if (movable_zone->zone_start_pfn <= start_pfn)
1262 return 1;
1263
1264 return 0;
1265}
1266
Dan Williams033fbae2015-08-09 15:29:06 -04001267int zone_for_memory(int nid, u64 start, u64 size, int zone_default,
1268 bool for_device)
Wang Nan63264402014-08-06 16:07:36 -07001269{
Dan Williams033fbae2015-08-09 15:29:06 -04001270#ifdef CONFIG_ZONE_DEVICE
1271 if (for_device)
1272 return ZONE_DEVICE;
1273#endif
Wang Nan63264402014-08-06 16:07:36 -07001274 if (should_add_memory_movable(nid, start, size))
1275 return ZONE_MOVABLE;
1276
1277 return zone_default;
1278}
1279
Vitaly Kuznetsov31bc3852016-03-15 14:56:48 -07001280static int online_memory_block(struct memory_block *mem, void *arg)
1281{
1282 return memory_block_change_state(mem, MEM_ONLINE, MEM_OFFLINE);
1283}
1284
Al Viro31168482008-11-22 17:33:24 +00001285/* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
Vitaly Kuznetsov31bc3852016-03-15 14:56:48 -07001286int __ref add_memory_resource(int nid, struct resource *res, bool online)
Yasunori Gotobc02af92006-06-27 02:53:30 -07001287{
David Vrabel62cedb92015-06-25 16:35:49 +01001288 u64 start, size;
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001289 pg_data_t *pgdat = NULL;
Tang Chena1e565a2013-02-22 16:33:18 -08001290 bool new_pgdat;
1291 bool new_node;
Yasunori Gotobc02af92006-06-27 02:53:30 -07001292 int ret;
1293
David Vrabel62cedb92015-06-25 16:35:49 +01001294 start = res->start;
1295 size = resource_size(res);
1296
Toshi Kani27356f52013-09-11 14:21:49 -07001297 ret = check_hotplug_memory_range(start, size);
1298 if (ret)
1299 return ret;
1300
Tang Chena1e565a2013-02-22 16:33:18 -08001301 { /* Stupid hack to suppress address-never-null warning */
1302 void *p = NODE_DATA(nid);
1303 new_pgdat = !p;
1304 }
Nathan Zimmerac13c462014-01-23 15:53:26 -08001305
Vladimir Davydovbfc8c902014-06-04 16:07:18 -07001306 mem_hotplug_begin();
Nathan Zimmerac13c462014-01-23 15:53:26 -08001307
Tang Chen7f36e3e2015-09-04 15:42:32 -07001308 /*
1309 * Add new range to memblock so that when hotadd_new_pgdat() is called
1310 * to allocate new pgdat, get_pfn_range_for_nid() will be able to find
1311 * this new range and calculate total pages correctly. The range will
1312 * be removed at hot-remove time.
1313 */
1314 memblock_add_node(start, size, nid);
1315
Tang Chena1e565a2013-02-22 16:33:18 -08001316 new_node = !node_online(nid);
1317 if (new_node) {
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001318 pgdat = hotadd_new_pgdat(nid, start);
Andi Kleen6ad696d2009-11-17 14:06:22 -08001319 ret = -ENOMEM;
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001320 if (!pgdat)
Wen Congyang41b9e2d2012-07-11 14:02:31 -07001321 goto error;
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001322 }
1323
Yasunori Gotobc02af92006-06-27 02:53:30 -07001324 /* call arch's memory hotadd */
Dan Williams033fbae2015-08-09 15:29:06 -04001325 ret = arch_add_memory(nid, start, size, false);
Yasunori Gotobc02af92006-06-27 02:53:30 -07001326
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001327 if (ret < 0)
1328 goto error;
1329
Yasunori Goto0fc44152006-06-27 02:53:38 -07001330 /* we online node here. we can't roll back from here. */
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001331 node_set_online(nid);
1332
Tang Chena1e565a2013-02-22 16:33:18 -08001333 if (new_node) {
Yasunori Goto0fc44152006-06-27 02:53:38 -07001334 ret = register_one_node(nid);
1335 /*
1336 * If sysfs file of new node can't create, cpu on the node
1337 * can't be hot-added. There is no rollback way now.
1338 * So, check by BUG_ON() to catch it reluctantly..
1339 */
1340 BUG_ON(ret);
1341 }
1342
akpm@linux-foundation.orgd96ae532010-03-05 13:41:58 -08001343 /* create new memmap entry */
1344 firmware_map_add_hotplug(start, start + size, "System RAM");
1345
Vitaly Kuznetsov31bc3852016-03-15 14:56:48 -07001346 /* online pages if requested */
1347 if (online)
1348 walk_memory_range(PFN_DOWN(start), PFN_UP(start + size - 1),
1349 NULL, online_memory_block);
1350
Andi Kleen6ad696d2009-11-17 14:06:22 -08001351 goto out;
1352
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001353error:
1354 /* rollback pgdat allocation and others */
1355 if (new_pgdat)
1356 rollback_node_hotadd(nid, pgdat);
Tang Chen7f36e3e2015-09-04 15:42:32 -07001357 memblock_remove(start, size);
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001358
Andi Kleen6ad696d2009-11-17 14:06:22 -08001359out:
Vladimir Davydovbfc8c902014-06-04 16:07:18 -07001360 mem_hotplug_done();
Yasunori Gotobc02af92006-06-27 02:53:30 -07001361 return ret;
1362}
David Vrabel62cedb92015-06-25 16:35:49 +01001363EXPORT_SYMBOL_GPL(add_memory_resource);
1364
1365int __ref add_memory(int nid, u64 start, u64 size)
1366{
1367 struct resource *res;
1368 int ret;
1369
1370 res = register_memory_resource(start, size);
Vitaly Kuznetsov6f754ba2016-01-14 15:21:55 -08001371 if (IS_ERR(res))
1372 return PTR_ERR(res);
David Vrabel62cedb92015-06-25 16:35:49 +01001373
Vitaly Kuznetsov31bc3852016-03-15 14:56:48 -07001374 ret = add_memory_resource(nid, res, memhp_auto_online);
David Vrabel62cedb92015-06-25 16:35:49 +01001375 if (ret < 0)
1376 release_memory_resource(res);
1377 return ret;
1378}
Yasunori Gotobc02af92006-06-27 02:53:30 -07001379EXPORT_SYMBOL_GPL(add_memory);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001380
1381#ifdef CONFIG_MEMORY_HOTREMOVE
1382/*
Badari Pulavarty5c755e92008-07-23 21:28:19 -07001383 * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy
1384 * set and the size of the free page is given by page_order(). Using this,
1385 * the function determines if the pageblock contains only free pages.
1386 * Due to buddy contraints, a free page at least the size of a pageblock will
1387 * be located at the start of the pageblock
1388 */
1389static inline int pageblock_free(struct page *page)
1390{
1391 return PageBuddy(page) && page_order(page) >= pageblock_order;
1392}
1393
1394/* Return the start of the next active pageblock after a given page */
1395static struct page *next_active_pageblock(struct page *page)
1396{
Badari Pulavarty5c755e92008-07-23 21:28:19 -07001397 /* Ensure the starting page is pageblock-aligned */
1398 BUG_ON(page_to_pfn(page) & (pageblock_nr_pages - 1));
1399
Badari Pulavarty5c755e92008-07-23 21:28:19 -07001400 /* If the entire pageblock is free, move to the end of free page */
KAMEZAWA Hiroyuki0dcc48c2010-09-09 16:38:01 -07001401 if (pageblock_free(page)) {
1402 int order;
1403 /* be careful. we don't have locks, page_order can be changed.*/
1404 order = page_order(page);
1405 if ((order < MAX_ORDER) && (order >= pageblock_order))
1406 return page + (1 << order);
1407 }
Badari Pulavarty5c755e92008-07-23 21:28:19 -07001408
KAMEZAWA Hiroyuki0dcc48c2010-09-09 16:38:01 -07001409 return page + pageblock_nr_pages;
Badari Pulavarty5c755e92008-07-23 21:28:19 -07001410}
1411
1412/* Checks if this range of memory is likely to be hot-removable. */
1413int is_mem_section_removable(unsigned long start_pfn, unsigned long nr_pages)
1414{
Badari Pulavarty5c755e92008-07-23 21:28:19 -07001415 struct page *page = pfn_to_page(start_pfn);
1416 struct page *end_page = page + nr_pages;
1417
1418 /* Check the starting page of each pageblock within the range */
1419 for (; page < end_page; page = next_active_pageblock(page)) {
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07001420 if (!is_pageblock_removable_nolock(page))
Badari Pulavarty5c755e92008-07-23 21:28:19 -07001421 return 0;
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07001422 cond_resched();
Badari Pulavarty5c755e92008-07-23 21:28:19 -07001423 }
1424
1425 /* All pageblocks in the memory block are likely to be hot-removable */
1426 return 1;
1427}
1428
1429/*
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001430 * Confirm all pages in a range [start, end) is belongs to the same zone.
1431 */
Zhang Zhened2f2402014-10-09 15:26:31 -07001432int test_pages_in_a_zone(unsigned long start_pfn, unsigned long end_pfn)
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001433{
Andrew Banman5f0f2882015-12-29 14:54:25 -08001434 unsigned long pfn, sec_end_pfn;
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001435 struct zone *zone = NULL;
1436 struct page *page;
1437 int i;
Andrew Banman5f0f2882015-12-29 14:54:25 -08001438 for (pfn = start_pfn, sec_end_pfn = SECTION_ALIGN_UP(start_pfn);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001439 pfn < end_pfn;
Andrew Banman5f0f2882015-12-29 14:54:25 -08001440 pfn = sec_end_pfn + 1, sec_end_pfn += PAGES_PER_SECTION) {
1441 /* Make sure the memory section is present first */
1442 if (!present_section_nr(pfn_to_section_nr(pfn)))
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001443 continue;
Andrew Banman5f0f2882015-12-29 14:54:25 -08001444 for (; pfn < sec_end_pfn && pfn < end_pfn;
1445 pfn += MAX_ORDER_NR_PAGES) {
1446 i = 0;
1447 /* This is just a CONFIG_HOLES_IN_ZONE check.*/
1448 while ((i < MAX_ORDER_NR_PAGES) &&
1449 !pfn_valid_within(pfn + i))
1450 i++;
1451 if (i == MAX_ORDER_NR_PAGES)
1452 continue;
1453 page = pfn_to_page(pfn + i);
1454 if (zone && page_zone(page) != zone)
1455 return 0;
1456 zone = page_zone(page);
1457 }
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001458 }
1459 return 1;
1460}
1461
1462/*
Naoya Horiguchic8721bb2013-09-11 14:22:09 -07001463 * Scan pfn range [start,end) to find movable/migratable pages (LRU pages
1464 * and hugepages). We scan pfn because it's much easier than scanning over
1465 * linked list. This function returns the pfn of the first found movable
1466 * page if it's found, otherwise 0.
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001467 */
Naoya Horiguchic8721bb2013-09-11 14:22:09 -07001468static unsigned long scan_movable_pages(unsigned long start, unsigned long end)
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001469{
1470 unsigned long pfn;
1471 struct page *page;
1472 for (pfn = start; pfn < end; pfn++) {
1473 if (pfn_valid(pfn)) {
1474 page = pfn_to_page(pfn);
1475 if (PageLRU(page))
1476 return pfn;
Naoya Horiguchic8721bb2013-09-11 14:22:09 -07001477 if (PageHuge(page)) {
Naoya Horiguchi7e1f0492015-04-15 16:14:41 -07001478 if (page_huge_active(page))
Naoya Horiguchic8721bb2013-09-11 14:22:09 -07001479 return pfn;
1480 else
1481 pfn = round_up(pfn + 1,
1482 1 << compound_order(page)) - 1;
1483 }
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001484 }
1485 }
1486 return 0;
1487}
1488
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001489#define NR_OFFLINE_AT_ONCE_PAGES (256)
1490static int
1491do_migrate_range(unsigned long start_pfn, unsigned long end_pfn)
1492{
1493 unsigned long pfn;
1494 struct page *page;
1495 int move_pages = NR_OFFLINE_AT_ONCE_PAGES;
1496 int not_managed = 0;
1497 int ret = 0;
1498 LIST_HEAD(source);
1499
1500 for (pfn = start_pfn; pfn < end_pfn && move_pages > 0; pfn++) {
1501 if (!pfn_valid(pfn))
1502 continue;
1503 page = pfn_to_page(pfn);
Naoya Horiguchic8721bb2013-09-11 14:22:09 -07001504
1505 if (PageHuge(page)) {
1506 struct page *head = compound_head(page);
1507 pfn = page_to_pfn(head) + (1<<compound_order(head)) - 1;
1508 if (compound_order(head) > PFN_SECTION_SHIFT) {
1509 ret = -EBUSY;
1510 break;
1511 }
1512 if (isolate_huge_page(page, &source))
1513 move_pages -= 1 << compound_order(head);
1514 continue;
1515 }
1516
Konstantin Khlebnikov700c2a42011-05-24 17:12:19 -07001517 if (!get_page_unless_zero(page))
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001518 continue;
1519 /*
1520 * We can skip free pages. And we can only deal with pages on
1521 * LRU.
1522 */
Nick Piggin62695a82008-10-18 20:26:09 -07001523 ret = isolate_lru_page(page);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001524 if (!ret) { /* Success */
Konstantin Khlebnikov700c2a42011-05-24 17:12:19 -07001525 put_page(page);
Nick Piggin62695a82008-10-18 20:26:09 -07001526 list_add_tail(&page->lru, &source);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001527 move_pages--;
KOSAKI Motohiro6d9c2852009-12-14 17:58:11 -08001528 inc_zone_page_state(page, NR_ISOLATED_ANON +
1529 page_is_file_cache(page));
1530
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001531 } else {
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001532#ifdef CONFIG_DEBUG_VM
Chen Yuconge33e33b2016-03-17 14:19:35 -07001533 pr_alert("removing pfn %lx from LRU failed\n", pfn);
Dave Hansenf0b791a2014-01-23 15:52:49 -08001534 dump_page(page, "failed to remove from LRU");
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001535#endif
Konstantin Khlebnikov700c2a42011-05-24 17:12:19 -07001536 put_page(page);
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001537 /* Because we don't have big zone->lock. we should
Bob Liu809c4442010-10-26 14:22:10 -07001538 check this again here. */
1539 if (page_count(page)) {
1540 not_managed++;
Bob Liuf3ab2632010-10-26 14:22:10 -07001541 ret = -EBUSY;
Bob Liu809c4442010-10-26 14:22:10 -07001542 break;
1543 }
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001544 }
1545 }
Bob Liuf3ab2632010-10-26 14:22:10 -07001546 if (!list_empty(&source)) {
1547 if (not_managed) {
Naoya Horiguchic8721bb2013-09-11 14:22:09 -07001548 putback_movable_pages(&source);
Bob Liuf3ab2632010-10-26 14:22:10 -07001549 goto out;
1550 }
Minchan Kim74c08f92012-10-08 16:32:54 -07001551
1552 /*
1553 * alloc_migrate_target should be improooooved!!
1554 * migrate_pages returns # of failed pages.
1555 */
David Rientjes68711a72014-06-04 16:08:25 -07001556 ret = migrate_pages(&source, alloc_migrate_target, NULL, 0,
Hugh Dickins9c620e22013-02-22 16:35:14 -08001557 MIGRATE_SYNC, MR_MEMORY_HOTPLUG);
Bob Liuf3ab2632010-10-26 14:22:10 -07001558 if (ret)
Naoya Horiguchic8721bb2013-09-11 14:22:09 -07001559 putback_movable_pages(&source);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001560 }
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001561out:
1562 return ret;
1563}
1564
1565/*
1566 * remove from free_area[] and mark all as Reserved.
1567 */
1568static int
1569offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages,
1570 void *data)
1571{
1572 __offline_isolated_pages(start, start + nr_pages);
1573 return 0;
1574}
1575
1576static void
1577offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
1578{
KAMEZAWA Hiroyuki908eedc2009-09-22 16:45:46 -07001579 walk_system_ram_range(start_pfn, end_pfn - start_pfn, NULL,
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001580 offline_isolated_pages_cb);
1581}
1582
1583/*
1584 * Check all pages in range, recoreded as memory resource, are isolated.
1585 */
1586static int
1587check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages,
1588 void *data)
1589{
1590 int ret;
1591 long offlined = *(long *)data;
Wen Congyangb023f462012-12-11 16:00:45 -08001592 ret = test_pages_isolated(start_pfn, start_pfn + nr_pages, true);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001593 offlined = nr_pages;
1594 if (!ret)
1595 *(long *)data += offlined;
1596 return ret;
1597}
1598
1599static long
1600check_pages_isolated(unsigned long start_pfn, unsigned long end_pfn)
1601{
1602 long offlined = 0;
1603 int ret;
1604
KAMEZAWA Hiroyuki908eedc2009-09-22 16:45:46 -07001605 ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn, &offlined,
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001606 check_pages_isolated_cb);
1607 if (ret < 0)
1608 offlined = (long)ret;
1609 return offlined;
1610}
1611
Lai Jiangshan09285af2012-12-12 13:52:04 -08001612#ifdef CONFIG_MOVABLE_NODE
Tang Chen79a4dce2012-12-18 14:23:24 -08001613/*
1614 * When CONFIG_MOVABLE_NODE, we permit offlining of a node which doesn't have
1615 * normal memory.
1616 */
Lai Jiangshan09285af2012-12-12 13:52:04 -08001617static bool can_offline_normal(struct zone *zone, unsigned long nr_pages)
1618{
1619 return true;
1620}
Tang Chen79a4dce2012-12-18 14:23:24 -08001621#else /* CONFIG_MOVABLE_NODE */
Lai Jiangshan74d42d82012-12-11 16:03:23 -08001622/* ensure the node has NORMAL memory if it is still online */
1623static bool can_offline_normal(struct zone *zone, unsigned long nr_pages)
1624{
1625 struct pglist_data *pgdat = zone->zone_pgdat;
1626 unsigned long present_pages = 0;
1627 enum zone_type zt;
1628
1629 for (zt = 0; zt <= ZONE_NORMAL; zt++)
1630 present_pages += pgdat->node_zones[zt].present_pages;
1631
1632 if (present_pages > nr_pages)
1633 return true;
1634
1635 present_pages = 0;
1636 for (; zt <= ZONE_MOVABLE; zt++)
1637 present_pages += pgdat->node_zones[zt].present_pages;
1638
1639 /*
1640 * we can't offline the last normal memory until all
1641 * higher memory is offlined.
1642 */
1643 return present_pages == 0;
1644}
Tang Chen79a4dce2012-12-18 14:23:24 -08001645#endif /* CONFIG_MOVABLE_NODE */
Lai Jiangshan74d42d82012-12-11 16:03:23 -08001646
Tang Chenc5320922013-11-12 15:08:10 -08001647static int __init cmdline_parse_movable_node(char *p)
1648{
1649#ifdef CONFIG_MOVABLE_NODE
1650 /*
1651 * Memory used by the kernel cannot be hot-removed because Linux
1652 * cannot migrate the kernel pages. When memory hotplug is
1653 * enabled, we should prevent memblock from allocating memory
1654 * for the kernel.
1655 *
1656 * ACPI SRAT records all hotpluggable memory ranges. But before
1657 * SRAT is parsed, we don't know about it.
1658 *
1659 * The kernel image is loaded into memory at very early time. We
1660 * cannot prevent this anyway. So on NUMA system, we set any
1661 * node the kernel resides in as un-hotpluggable.
1662 *
1663 * Since on modern servers, one node could have double-digit
1664 * gigabytes memory, we can assume the memory around the kernel
1665 * image is also un-hotpluggable. So before SRAT is parsed, just
1666 * allocate memory near the kernel image to try the best to keep
1667 * the kernel away from hotpluggable memory.
1668 */
1669 memblock_set_bottom_up(true);
Tang Chen55ac5902014-01-21 15:49:35 -08001670 movable_node_enabled = true;
Tang Chenc5320922013-11-12 15:08:10 -08001671#else
1672 pr_warn("movable_node option not supported\n");
1673#endif
1674 return 0;
1675}
1676early_param("movable_node", cmdline_parse_movable_node);
1677
Lai Jiangshand9713672012-12-11 16:01:03 -08001678/* check which state of node_states will be changed when offline memory */
1679static void node_states_check_changes_offline(unsigned long nr_pages,
1680 struct zone *zone, struct memory_notify *arg)
1681{
1682 struct pglist_data *pgdat = zone->zone_pgdat;
1683 unsigned long present_pages = 0;
1684 enum zone_type zt, zone_last = ZONE_NORMAL;
1685
1686 /*
Lai Jiangshan6715ddf2012-12-12 13:51:49 -08001687 * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY]
1688 * contains nodes which have zones of 0...ZONE_NORMAL,
1689 * set zone_last to ZONE_NORMAL.
Lai Jiangshand9713672012-12-11 16:01:03 -08001690 *
Lai Jiangshan6715ddf2012-12-12 13:51:49 -08001691 * If we don't have HIGHMEM nor movable node,
1692 * node_states[N_NORMAL_MEMORY] contains nodes which have zones of
1693 * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE.
Lai Jiangshand9713672012-12-11 16:01:03 -08001694 */
Lai Jiangshan6715ddf2012-12-12 13:51:49 -08001695 if (N_MEMORY == N_NORMAL_MEMORY)
Lai Jiangshand9713672012-12-11 16:01:03 -08001696 zone_last = ZONE_MOVABLE;
1697
1698 /*
1699 * check whether node_states[N_NORMAL_MEMORY] will be changed.
1700 * If the memory to be offline is in a zone of 0...zone_last,
1701 * and it is the last present memory, 0...zone_last will
1702 * become empty after offline , thus we can determind we will
1703 * need to clear the node from node_states[N_NORMAL_MEMORY].
1704 */
1705 for (zt = 0; zt <= zone_last; zt++)
1706 present_pages += pgdat->node_zones[zt].present_pages;
1707 if (zone_idx(zone) <= zone_last && nr_pages >= present_pages)
1708 arg->status_change_nid_normal = zone_to_nid(zone);
1709 else
1710 arg->status_change_nid_normal = -1;
1711
Lai Jiangshan6715ddf2012-12-12 13:51:49 -08001712#ifdef CONFIG_HIGHMEM
1713 /*
1714 * If we have movable node, node_states[N_HIGH_MEMORY]
1715 * contains nodes which have zones of 0...ZONE_HIGHMEM,
1716 * set zone_last to ZONE_HIGHMEM.
1717 *
1718 * If we don't have movable node, node_states[N_NORMAL_MEMORY]
1719 * contains nodes which have zones of 0...ZONE_MOVABLE,
1720 * set zone_last to ZONE_MOVABLE.
1721 */
1722 zone_last = ZONE_HIGHMEM;
1723 if (N_MEMORY == N_HIGH_MEMORY)
1724 zone_last = ZONE_MOVABLE;
1725
1726 for (; zt <= zone_last; zt++)
1727 present_pages += pgdat->node_zones[zt].present_pages;
1728 if (zone_idx(zone) <= zone_last && nr_pages >= present_pages)
1729 arg->status_change_nid_high = zone_to_nid(zone);
1730 else
1731 arg->status_change_nid_high = -1;
1732#else
1733 arg->status_change_nid_high = arg->status_change_nid_normal;
1734#endif
1735
Lai Jiangshand9713672012-12-11 16:01:03 -08001736 /*
1737 * node_states[N_HIGH_MEMORY] contains nodes which have 0...ZONE_MOVABLE
1738 */
1739 zone_last = ZONE_MOVABLE;
1740
1741 /*
1742 * check whether node_states[N_HIGH_MEMORY] will be changed
1743 * If we try to offline the last present @nr_pages from the node,
1744 * we can determind we will need to clear the node from
1745 * node_states[N_HIGH_MEMORY].
1746 */
1747 for (; zt <= zone_last; zt++)
1748 present_pages += pgdat->node_zones[zt].present_pages;
1749 if (nr_pages >= present_pages)
1750 arg->status_change_nid = zone_to_nid(zone);
1751 else
1752 arg->status_change_nid = -1;
1753}
1754
1755static void node_states_clear_node(int node, struct memory_notify *arg)
1756{
1757 if (arg->status_change_nid_normal >= 0)
1758 node_clear_state(node, N_NORMAL_MEMORY);
1759
Lai Jiangshan6715ddf2012-12-12 13:51:49 -08001760 if ((N_MEMORY != N_NORMAL_MEMORY) &&
1761 (arg->status_change_nid_high >= 0))
Lai Jiangshand9713672012-12-11 16:01:03 -08001762 node_clear_state(node, N_HIGH_MEMORY);
Lai Jiangshan6715ddf2012-12-12 13:51:49 -08001763
1764 if ((N_MEMORY != N_HIGH_MEMORY) &&
1765 (arg->status_change_nid >= 0))
1766 node_clear_state(node, N_MEMORY);
Lai Jiangshand9713672012-12-11 16:01:03 -08001767}
1768
Wen Congyanga16cee12012-10-08 16:33:58 -07001769static int __ref __offline_pages(unsigned long start_pfn,
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001770 unsigned long end_pfn, unsigned long timeout)
1771{
1772 unsigned long pfn, nr_pages, expire;
1773 long offlined_pages;
Yasunori Goto7b78d332007-10-21 16:41:36 -07001774 int ret, drain, retry_max, node;
Cody P Schaferd7029092013-07-03 15:02:11 -07001775 unsigned long flags;
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001776 struct zone *zone;
Yasunori Goto7b78d332007-10-21 16:41:36 -07001777 struct memory_notify arg;
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001778
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001779 /* at least, alignment against pageblock is necessary */
1780 if (!IS_ALIGNED(start_pfn, pageblock_nr_pages))
1781 return -EINVAL;
1782 if (!IS_ALIGNED(end_pfn, pageblock_nr_pages))
1783 return -EINVAL;
1784 /* This makes hotplug much easier...and readable.
1785 we assume this for now. .*/
1786 if (!test_pages_in_a_zone(start_pfn, end_pfn))
1787 return -EINVAL;
Yasunori Goto7b78d332007-10-21 16:41:36 -07001788
1789 zone = page_zone(pfn_to_page(start_pfn));
1790 node = zone_to_nid(zone);
1791 nr_pages = end_pfn - start_pfn;
1792
Lai Jiangshan74d42d82012-12-11 16:03:23 -08001793 if (zone_idx(zone) <= ZONE_NORMAL && !can_offline_normal(zone, nr_pages))
David Rientjes30467e02015-04-14 15:45:11 -07001794 return -EINVAL;
Lai Jiangshan74d42d82012-12-11 16:03:23 -08001795
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001796 /* set above range as isolated */
Wen Congyangb023f462012-12-11 16:00:45 -08001797 ret = start_isolate_page_range(start_pfn, end_pfn,
1798 MIGRATE_MOVABLE, true);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001799 if (ret)
David Rientjes30467e02015-04-14 15:45:11 -07001800 return ret;
Yasunori Goto7b78d332007-10-21 16:41:36 -07001801
1802 arg.start_pfn = start_pfn;
1803 arg.nr_pages = nr_pages;
Lai Jiangshand9713672012-12-11 16:01:03 -08001804 node_states_check_changes_offline(nr_pages, zone, &arg);
Yasunori Goto7b78d332007-10-21 16:41:36 -07001805
1806 ret = memory_notify(MEM_GOING_OFFLINE, &arg);
1807 ret = notifier_to_errno(ret);
1808 if (ret)
1809 goto failed_removal;
1810
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001811 pfn = start_pfn;
1812 expire = jiffies + timeout;
1813 drain = 0;
1814 retry_max = 5;
1815repeat:
1816 /* start memory hot removal */
1817 ret = -EAGAIN;
1818 if (time_after(jiffies, expire))
1819 goto failed_removal;
1820 ret = -EINTR;
1821 if (signal_pending(current))
1822 goto failed_removal;
1823 ret = 0;
1824 if (drain) {
1825 lru_add_drain_all();
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001826 cond_resched();
Vlastimil Babkac0554322014-12-10 15:43:10 -08001827 drain_all_pages(zone);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001828 }
1829
Naoya Horiguchic8721bb2013-09-11 14:22:09 -07001830 pfn = scan_movable_pages(start_pfn, end_pfn);
1831 if (pfn) { /* We have movable pages */
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001832 ret = do_migrate_range(pfn, end_pfn);
1833 if (!ret) {
1834 drain = 1;
1835 goto repeat;
1836 } else {
1837 if (ret < 0)
1838 if (--retry_max == 0)
1839 goto failed_removal;
1840 yield();
1841 drain = 1;
1842 goto repeat;
1843 }
1844 }
Adam Buchbinderb3834be2012-09-19 21:48:02 -04001845 /* drain all zone's lru pagevec, this is asynchronous... */
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001846 lru_add_drain_all();
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001847 yield();
Adam Buchbinderb3834be2012-09-19 21:48:02 -04001848 /* drain pcp pages, this is synchronous. */
Vlastimil Babkac0554322014-12-10 15:43:10 -08001849 drain_all_pages(zone);
Naoya Horiguchic8721bb2013-09-11 14:22:09 -07001850 /*
1851 * dissolve free hugepages in the memory block before doing offlining
1852 * actually in order to make hugetlbfs's object counting consistent.
1853 */
1854 dissolve_free_huge_pages(start_pfn, end_pfn);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001855 /* check again */
1856 offlined_pages = check_pages_isolated(start_pfn, end_pfn);
1857 if (offlined_pages < 0) {
1858 ret = -EBUSY;
1859 goto failed_removal;
1860 }
Chen Yuconge33e33b2016-03-17 14:19:35 -07001861 pr_info("Offlined Pages %ld\n", offlined_pages);
Adam Buchbinderb3834be2012-09-19 21:48:02 -04001862 /* Ok, all of our target is isolated.
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001863 We cannot do rollback at this point. */
1864 offline_isolated_pages(start_pfn, end_pfn);
KAMEZAWA Hiroyukidbc0e4c2007-11-14 16:59:12 -08001865 /* reset pagetype flags and makes migrate type to be MOVABLE */
Michal Nazarewicz0815f3d2012-04-03 15:06:15 +02001866 undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001867 /* removal success */
Jiang Liu3dcc0572013-07-03 15:03:21 -07001868 adjust_managed_page_count(pfn_to_page(start_pfn), -offlined_pages);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001869 zone->present_pages -= offlined_pages;
Cody P Schaferd7029092013-07-03 15:02:11 -07001870
1871 pgdat_resize_lock(zone->zone_pgdat, &flags);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001872 zone->zone_pgdat->node_present_pages -= offlined_pages;
Cody P Schaferd7029092013-07-03 15:02:11 -07001873 pgdat_resize_unlock(zone->zone_pgdat, &flags);
Yasunori Goto7b78d332007-10-21 16:41:36 -07001874
KOSAKI Motohiro1b79acc2011-05-24 17:11:32 -07001875 init_per_zone_wmark_min();
1876
Xishi Qiu1e8537b2012-10-08 16:31:51 -07001877 if (!populated_zone(zone)) {
Jiang Liu340175b2012-07-31 16:43:32 -07001878 zone_pcp_reset(zone);
Xishi Qiu1e8537b2012-10-08 16:31:51 -07001879 mutex_lock(&zonelists_mutex);
1880 build_all_zonelists(NULL, NULL);
1881 mutex_unlock(&zonelists_mutex);
1882 } else
1883 zone_pcp_update(zone);
Jiang Liu340175b2012-07-31 16:43:32 -07001884
Lai Jiangshand9713672012-12-11 16:01:03 -08001885 node_states_clear_node(node, &arg);
Vlastimil Babka698b1b32016-03-17 14:18:08 -07001886 if (arg.status_change_nid >= 0) {
David Rientjes8fe23e02009-12-14 17:58:33 -08001887 kswapd_stop(node);
Vlastimil Babka698b1b32016-03-17 14:18:08 -07001888 kcompactd_stop(node);
1889 }
Minchan Kimbce73942009-06-16 15:32:50 -07001890
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001891 vm_total_pages = nr_free_pagecache_pages();
1892 writeback_set_ratelimit();
Yasunori Goto7b78d332007-10-21 16:41:36 -07001893
1894 memory_notify(MEM_OFFLINE, &arg);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001895 return 0;
1896
1897failed_removal:
Chen Yuconge33e33b2016-03-17 14:19:35 -07001898 pr_debug("memory offlining [mem %#010llx-%#010llx] failed\n",
1899 (unsigned long long) start_pfn << PAGE_SHIFT,
1900 ((unsigned long long) end_pfn << PAGE_SHIFT) - 1);
Yasunori Goto7b78d332007-10-21 16:41:36 -07001901 memory_notify(MEM_CANCEL_OFFLINE, &arg);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001902 /* pushback to free area */
Michal Nazarewicz0815f3d2012-04-03 15:06:15 +02001903 undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001904 return ret;
1905}
Badari Pulavarty71088782008-10-18 20:25:58 -07001906
David Rientjes30467e02015-04-14 15:45:11 -07001907/* Must be protected by mem_hotplug_begin() */
Wen Congyanga16cee12012-10-08 16:33:58 -07001908int offline_pages(unsigned long start_pfn, unsigned long nr_pages)
1909{
1910 return __offline_pages(start_pfn, start_pfn + nr_pages, 120 * HZ);
1911}
Rafael J. Wysockie2ff3942013-05-08 00:29:49 +02001912#endif /* CONFIG_MEMORY_HOTREMOVE */
Wen Congyanga16cee12012-10-08 16:33:58 -07001913
Wen Congyangbbc76be2013-02-22 16:32:54 -08001914/**
1915 * walk_memory_range - walks through all mem sections in [start_pfn, end_pfn)
1916 * @start_pfn: start pfn of the memory range
Toshi Kanie05c4bb2013-04-29 15:06:16 -07001917 * @end_pfn: end pfn of the memory range
Wen Congyangbbc76be2013-02-22 16:32:54 -08001918 * @arg: argument passed to func
1919 * @func: callback for each memory section walked
1920 *
1921 * This function walks through all present mem sections in range
1922 * [start_pfn, end_pfn) and call func on each mem section.
1923 *
1924 * Returns the return value of func.
1925 */
Rafael J. Wysockie2ff3942013-05-08 00:29:49 +02001926int walk_memory_range(unsigned long start_pfn, unsigned long end_pfn,
Wen Congyangbbc76be2013-02-22 16:32:54 -08001927 void *arg, int (*func)(struct memory_block *, void *))
Badari Pulavarty71088782008-10-18 20:25:58 -07001928{
Wen Congyange90bdb72012-10-08 16:34:01 -07001929 struct memory_block *mem = NULL;
1930 struct mem_section *section;
Wen Congyange90bdb72012-10-08 16:34:01 -07001931 unsigned long pfn, section_nr;
1932 int ret;
Badari Pulavarty71088782008-10-18 20:25:58 -07001933
Wen Congyange90bdb72012-10-08 16:34:01 -07001934 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
1935 section_nr = pfn_to_section_nr(pfn);
1936 if (!present_section_nr(section_nr))
1937 continue;
1938
1939 section = __nr_to_section(section_nr);
1940 /* same memblock? */
1941 if (mem)
1942 if ((section_nr >= mem->start_section_nr) &&
1943 (section_nr <= mem->end_section_nr))
1944 continue;
1945
1946 mem = find_memory_block_hinted(section, mem);
1947 if (!mem)
1948 continue;
1949
Wen Congyangbbc76be2013-02-22 16:32:54 -08001950 ret = func(mem, arg);
Wen Congyange90bdb72012-10-08 16:34:01 -07001951 if (ret) {
Wen Congyangbbc76be2013-02-22 16:32:54 -08001952 kobject_put(&mem->dev.kobj);
1953 return ret;
Wen Congyange90bdb72012-10-08 16:34:01 -07001954 }
1955 }
1956
1957 if (mem)
1958 kobject_put(&mem->dev.kobj);
1959
Wen Congyangbbc76be2013-02-22 16:32:54 -08001960 return 0;
1961}
1962
Rafael J. Wysockie2ff3942013-05-08 00:29:49 +02001963#ifdef CONFIG_MEMORY_HOTREMOVE
Xishi Qiud6de9d52013-11-12 15:07:20 -08001964static int check_memblock_offlined_cb(struct memory_block *mem, void *arg)
Wen Congyangbbc76be2013-02-22 16:32:54 -08001965{
1966 int ret = !is_memblock_offlined(mem);
1967
Randy Dunlap349daa02013-04-29 15:08:49 -07001968 if (unlikely(ret)) {
1969 phys_addr_t beginpa, endpa;
1970
1971 beginpa = PFN_PHYS(section_nr_to_pfn(mem->start_section_nr));
1972 endpa = PFN_PHYS(section_nr_to_pfn(mem->end_section_nr + 1))-1;
Joe Perches756a0252016-03-17 14:19:47 -07001973 pr_warn("removing memory fails, because memory [%pa-%pa] is onlined\n",
Randy Dunlap349daa02013-04-29 15:08:49 -07001974 &beginpa, &endpa);
1975 }
Wen Congyangbbc76be2013-02-22 16:32:54 -08001976
1977 return ret;
1978}
1979
Toshi Kani0f1cfe92013-09-11 14:21:50 -07001980static int check_cpu_on_node(pg_data_t *pgdat)
Tang Chen60a5a192013-02-22 16:33:14 -08001981{
Tang Chen60a5a192013-02-22 16:33:14 -08001982 int cpu;
1983
1984 for_each_present_cpu(cpu) {
1985 if (cpu_to_node(cpu) == pgdat->node_id)
1986 /*
1987 * the cpu on this node isn't removed, and we can't
1988 * offline this node.
1989 */
1990 return -EBUSY;
1991 }
1992
1993 return 0;
1994}
1995
Toshi Kani0f1cfe92013-09-11 14:21:50 -07001996static void unmap_cpu_on_node(pg_data_t *pgdat)
Wen Congyange13fe862013-02-22 16:33:31 -08001997{
1998#ifdef CONFIG_ACPI_NUMA
Wen Congyange13fe862013-02-22 16:33:31 -08001999 int cpu;
2000
2001 for_each_possible_cpu(cpu)
2002 if (cpu_to_node(cpu) == pgdat->node_id)
2003 numa_clear_node(cpu);
2004#endif
2005}
2006
Toshi Kani0f1cfe92013-09-11 14:21:50 -07002007static int check_and_unmap_cpu_on_node(pg_data_t *pgdat)
Wen Congyange13fe862013-02-22 16:33:31 -08002008{
Toshi Kani0f1cfe92013-09-11 14:21:50 -07002009 int ret;
Wen Congyange13fe862013-02-22 16:33:31 -08002010
Toshi Kani0f1cfe92013-09-11 14:21:50 -07002011 ret = check_cpu_on_node(pgdat);
Wen Congyange13fe862013-02-22 16:33:31 -08002012 if (ret)
2013 return ret;
2014
2015 /*
2016 * the node will be offlined when we come here, so we can clear
2017 * the cpu_to_node() now.
2018 */
2019
Toshi Kani0f1cfe92013-09-11 14:21:50 -07002020 unmap_cpu_on_node(pgdat);
Wen Congyange13fe862013-02-22 16:33:31 -08002021 return 0;
2022}
2023
Toshi Kani0f1cfe92013-09-11 14:21:50 -07002024/**
2025 * try_offline_node
2026 *
2027 * Offline a node if all memory sections and cpus of the node are removed.
2028 *
2029 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
2030 * and online/offline operations before this call.
2031 */
Wen Congyang90b30cd2013-02-22 16:33:27 -08002032void try_offline_node(int nid)
Tang Chen60a5a192013-02-22 16:33:14 -08002033{
Wen Congyangd822b862013-02-22 16:33:16 -08002034 pg_data_t *pgdat = NODE_DATA(nid);
2035 unsigned long start_pfn = pgdat->node_start_pfn;
2036 unsigned long end_pfn = start_pfn + pgdat->node_spanned_pages;
Tang Chen60a5a192013-02-22 16:33:14 -08002037 unsigned long pfn;
Wen Congyangd822b862013-02-22 16:33:16 -08002038 int i;
Tang Chen60a5a192013-02-22 16:33:14 -08002039
2040 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
2041 unsigned long section_nr = pfn_to_section_nr(pfn);
2042
2043 if (!present_section_nr(section_nr))
2044 continue;
2045
2046 if (pfn_to_nid(pfn) != nid)
2047 continue;
2048
2049 /*
2050 * some memory sections of this node are not removed, and we
2051 * can't offline node now.
2052 */
2053 return;
2054 }
2055
Toshi Kani0f1cfe92013-09-11 14:21:50 -07002056 if (check_and_unmap_cpu_on_node(pgdat))
Tang Chen60a5a192013-02-22 16:33:14 -08002057 return;
2058
2059 /*
2060 * all memory/cpu of this node are removed, we can offline this
2061 * node now.
2062 */
2063 node_set_offline(nid);
2064 unregister_one_node(nid);
Wen Congyangd822b862013-02-22 16:33:16 -08002065
Wen Congyangd822b862013-02-22 16:33:16 -08002066 /* free waittable in each zone */
2067 for (i = 0; i < MAX_NR_ZONES; i++) {
2068 struct zone *zone = pgdat->node_zones + i;
2069
Jianguo Wuca4b3f32013-03-22 15:04:50 -07002070 /*
2071 * wait_table may be allocated from boot memory,
2072 * here only free if it's allocated by vmalloc.
2073 */
Gu Zheng85bd8392015-06-10 11:14:43 -07002074 if (is_vmalloc_addr(zone->wait_table)) {
Wen Congyangd822b862013-02-22 16:33:16 -08002075 vfree(zone->wait_table);
Gu Zheng85bd8392015-06-10 11:14:43 -07002076 zone->wait_table = NULL;
2077 }
Wen Congyangd822b862013-02-22 16:33:16 -08002078 }
Tang Chen60a5a192013-02-22 16:33:14 -08002079}
Wen Congyang90b30cd2013-02-22 16:33:27 -08002080EXPORT_SYMBOL(try_offline_node);
Tang Chen60a5a192013-02-22 16:33:14 -08002081
Toshi Kani0f1cfe92013-09-11 14:21:50 -07002082/**
2083 * remove_memory
2084 *
2085 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
2086 * and online/offline operations before this call, as required by
2087 * try_offline_node().
2088 */
Rafael J. Wysocki242831e2013-05-27 12:58:46 +02002089void __ref remove_memory(int nid, u64 start, u64 size)
Wen Congyangbbc76be2013-02-22 16:32:54 -08002090{
Rafael J. Wysocki242831e2013-05-27 12:58:46 +02002091 int ret;
Wen Congyang993c1aa2013-02-22 16:32:50 -08002092
Toshi Kani27356f52013-09-11 14:21:49 -07002093 BUG_ON(check_hotplug_memory_range(start, size));
2094
Vladimir Davydovbfc8c902014-06-04 16:07:18 -07002095 mem_hotplug_begin();
Yasuaki Ishimatsu6677e3e2013-02-22 16:32:52 -08002096
2097 /*
Rafael J. Wysocki242831e2013-05-27 12:58:46 +02002098 * All memory blocks must be offlined before removing memory. Check
2099 * whether all memory blocks in question are offline and trigger a BUG()
2100 * if this is not the case.
Yasuaki Ishimatsu6677e3e2013-02-22 16:32:52 -08002101 */
Rafael J. Wysocki242831e2013-05-27 12:58:46 +02002102 ret = walk_memory_range(PFN_DOWN(start), PFN_UP(start + size - 1), NULL,
Xishi Qiud6de9d52013-11-12 15:07:20 -08002103 check_memblock_offlined_cb);
Vladimir Davydovbfc8c902014-06-04 16:07:18 -07002104 if (ret)
Rafael J. Wysocki242831e2013-05-27 12:58:46 +02002105 BUG();
Yasuaki Ishimatsu6677e3e2013-02-22 16:32:52 -08002106
Yasuaki Ishimatsu46c66c42013-02-22 16:32:56 -08002107 /* remove memmap entry */
2108 firmware_map_remove(start, start + size, "System RAM");
Xishi Qiuf9126ab2015-08-14 15:35:16 -07002109 memblock_free(start, size);
2110 memblock_remove(start, size);
Yasuaki Ishimatsu46c66c42013-02-22 16:32:56 -08002111
Wen Congyang24d335c2013-02-22 16:32:58 -08002112 arch_remove_memory(start, size);
2113
Tang Chen60a5a192013-02-22 16:33:14 -08002114 try_offline_node(nid);
2115
Vladimir Davydovbfc8c902014-06-04 16:07:18 -07002116 mem_hotplug_done();
Badari Pulavarty71088782008-10-18 20:25:58 -07002117}
Badari Pulavarty71088782008-10-18 20:25:58 -07002118EXPORT_SYMBOL_GPL(remove_memory);
Rafael J. Wysockiaba6efc2013-06-01 22:24:07 +02002119#endif /* CONFIG_MEMORY_HOTREMOVE */