blob: 1bf4807cb21e49ccbd1bb4232574507e8d41384c [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>
20#include <linux/memory_hotplug.h>
21#include <linux/highmem.h>
22#include <linux/vmalloc.h>
KAMEZAWA Hiroyuki0a547032006-06-27 02:53:35 -070023#include <linux/ioport.h>
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -070024#include <linux/delay.h>
25#include <linux/migrate.h>
26#include <linux/page-isolation.h>
Badari Pulavarty71088782008-10-18 20:25:58 -070027#include <linux/pfn.h>
Andi Kleen6ad696d2009-11-17 14:06:22 -080028#include <linux/suspend.h>
KOSAKI Motohiro6d9c2852009-12-14 17:58:11 -080029#include <linux/mm_inline.h>
akpm@linux-foundation.orgd96ae532010-03-05 13:41:58 -080030#include <linux/firmware-map.h>
Tang Chen60a5a192013-02-22 16:33:14 -080031#include <linux/stop_machine.h>
Naoya Horiguchic8721bb2013-09-11 14:22:09 -070032#include <linux/hugetlb.h>
Tang Chenc5320922013-11-12 15:08:10 -080033#include <linux/memblock.h>
Tang Chenf784a3f2014-11-13 15:19:39 -080034#include <linux/bootmem.h>
Dave Hansen3947be12005-10-29 18:16:54 -070035
36#include <asm/tlbflush.h>
37
Adrian Bunk1e5ad9a2008-04-28 20:40:08 +030038#include "internal.h"
39
Daniel Kiper9d0ad8c2011-07-25 17:12:05 -070040/*
41 * online_page_callback contains pointer to current page onlining function.
42 * Initially it is generic_online_page(). If it is required it could be
43 * changed by calling set_online_page_callback() for callback registration
44 * and restore_online_page_callback() for generic callback restore.
45 */
46
47static void generic_online_page(struct page *page);
48
49static online_page_callback_t online_page_callback = generic_online_page;
Vladimir Davydovbfc8c902014-06-04 16:07:18 -070050static DEFINE_MUTEX(online_page_callback_lock);
Daniel Kiper9d0ad8c2011-07-25 17:12:05 -070051
Vladimir Davydovbfc8c902014-06-04 16:07:18 -070052/* The same as the cpu_hotplug lock, but for memory hotplug. */
53static struct {
54 struct task_struct *active_writer;
55 struct mutex lock; /* Synchronizes accesses to refcount, */
56 /*
57 * Also blocks the new readers during
58 * an ongoing mem hotplug operation.
59 */
60 int refcount;
KOSAKI Motohiro20d6c962010-12-02 14:31:19 -080061
Vladimir Davydovbfc8c902014-06-04 16:07:18 -070062#ifdef CONFIG_DEBUG_LOCK_ALLOC
63 struct lockdep_map dep_map;
64#endif
65} mem_hotplug = {
66 .active_writer = NULL,
67 .lock = __MUTEX_INITIALIZER(mem_hotplug.lock),
68 .refcount = 0,
69#ifdef CONFIG_DEBUG_LOCK_ALLOC
70 .dep_map = {.name = "mem_hotplug.lock" },
71#endif
72};
73
74/* Lockdep annotations for get/put_online_mems() and mem_hotplug_begin/end() */
75#define memhp_lock_acquire_read() lock_map_acquire_read(&mem_hotplug.dep_map)
76#define memhp_lock_acquire() lock_map_acquire(&mem_hotplug.dep_map)
77#define memhp_lock_release() lock_map_release(&mem_hotplug.dep_map)
78
79void get_online_mems(void)
KOSAKI Motohiro20d6c962010-12-02 14:31:19 -080080{
Vladimir Davydovbfc8c902014-06-04 16:07:18 -070081 might_sleep();
82 if (mem_hotplug.active_writer == current)
83 return;
84 memhp_lock_acquire_read();
85 mutex_lock(&mem_hotplug.lock);
86 mem_hotplug.refcount++;
87 mutex_unlock(&mem_hotplug.lock);
88
KOSAKI Motohiro20d6c962010-12-02 14:31:19 -080089}
90
Vladimir Davydovbfc8c902014-06-04 16:07:18 -070091void put_online_mems(void)
KOSAKI Motohiro20d6c962010-12-02 14:31:19 -080092{
Vladimir Davydovbfc8c902014-06-04 16:07:18 -070093 if (mem_hotplug.active_writer == current)
94 return;
95 mutex_lock(&mem_hotplug.lock);
96
97 if (WARN_ON(!mem_hotplug.refcount))
98 mem_hotplug.refcount++; /* try to fix things up */
99
100 if (!--mem_hotplug.refcount && unlikely(mem_hotplug.active_writer))
101 wake_up_process(mem_hotplug.active_writer);
102 mutex_unlock(&mem_hotplug.lock);
103 memhp_lock_release();
104
KOSAKI Motohiro20d6c962010-12-02 14:31:19 -0800105}
106
Vladimir Davydovbfc8c902014-06-04 16:07:18 -0700107static void mem_hotplug_begin(void)
108{
109 mem_hotplug.active_writer = current;
110
111 memhp_lock_acquire();
112 for (;;) {
113 mutex_lock(&mem_hotplug.lock);
114 if (likely(!mem_hotplug.refcount))
115 break;
116 __set_current_state(TASK_UNINTERRUPTIBLE);
117 mutex_unlock(&mem_hotplug.lock);
118 schedule();
119 }
120}
121
122static void mem_hotplug_done(void)
123{
124 mem_hotplug.active_writer = NULL;
125 mutex_unlock(&mem_hotplug.lock);
126 memhp_lock_release();
127}
KOSAKI Motohiro20d6c962010-12-02 14:31:19 -0800128
Keith Mannthey45e0b782006-09-30 23:27:09 -0700129/* add this memory to iomem resource */
130static struct resource *register_memory_resource(u64 start, u64 size)
131{
132 struct resource *res;
133 res = kzalloc(sizeof(struct resource), GFP_KERNEL);
134 BUG_ON(!res);
135
136 res->name = "System RAM";
137 res->start = start;
138 res->end = start + size - 1;
Yasunori Goto887c3cb2007-11-14 16:59:20 -0800139 res->flags = IORESOURCE_MEM | IORESOURCE_BUSY;
Keith Mannthey45e0b782006-09-30 23:27:09 -0700140 if (request_resource(&iomem_resource, res) < 0) {
Toshi Kani4996eed2013-07-03 15:02:39 -0700141 pr_debug("System RAM resource %pR cannot be added\n", res);
Keith Mannthey45e0b782006-09-30 23:27:09 -0700142 kfree(res);
143 res = NULL;
144 }
145 return res;
146}
147
148static void release_memory_resource(struct resource *res)
149{
150 if (!res)
151 return;
152 release_resource(res);
153 kfree(res);
154 return;
155}
156
Keith Mannthey53947022006-09-30 23:27:08 -0700157#ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
Yasuaki Ishimatsu46723bf2013-02-22 16:33:00 -0800158void get_page_bootmem(unsigned long info, struct page *page,
159 unsigned long type)
Yasunori Goto04753272008-04-28 02:13:31 -0700160{
Andrea Arcangeli5f24ce52011-01-13 15:47:00 -0800161 page->lru.next = (struct list_head *) type;
Yasunori Goto04753272008-04-28 02:13:31 -0700162 SetPagePrivate(page);
163 set_page_private(page, info);
164 atomic_inc(&page->_count);
165}
166
Jiang Liu170a5a72013-07-03 15:03:17 -0700167void put_page_bootmem(struct page *page)
Yasunori Goto04753272008-04-28 02:13:31 -0700168{
Andrea Arcangeli5f24ce52011-01-13 15:47:00 -0800169 unsigned long type;
Yasunori Goto04753272008-04-28 02:13:31 -0700170
Andrea Arcangeli5f24ce52011-01-13 15:47:00 -0800171 type = (unsigned long) page->lru.next;
172 BUG_ON(type < MEMORY_HOTPLUG_MIN_BOOTMEM_TYPE ||
173 type > MEMORY_HOTPLUG_MAX_BOOTMEM_TYPE);
Yasunori Goto04753272008-04-28 02:13:31 -0700174
175 if (atomic_dec_return(&page->_count) == 1) {
176 ClearPagePrivate(page);
177 set_page_private(page, 0);
Andrea Arcangeli5f24ce52011-01-13 15:47:00 -0800178 INIT_LIST_HEAD(&page->lru);
Jiang Liu170a5a72013-07-03 15:03:17 -0700179 free_reserved_page(page);
Yasunori Goto04753272008-04-28 02:13:31 -0700180 }
Yasunori Goto04753272008-04-28 02:13:31 -0700181}
182
Yasuaki Ishimatsu46723bf2013-02-22 16:33:00 -0800183#ifdef CONFIG_HAVE_BOOTMEM_INFO_NODE
184#ifndef CONFIG_SPARSEMEM_VMEMMAP
Adrian Bunkd92bc312008-07-23 21:28:12 -0700185static void register_page_bootmem_info_section(unsigned long start_pfn)
Yasunori Goto04753272008-04-28 02:13:31 -0700186{
187 unsigned long *usemap, mapsize, section_nr, i;
188 struct mem_section *ms;
189 struct page *page, *memmap;
190
Yasunori Goto04753272008-04-28 02:13:31 -0700191 section_nr = pfn_to_section_nr(start_pfn);
192 ms = __nr_to_section(section_nr);
193
194 /* Get section's memmap address */
195 memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
196
197 /*
198 * Get page for the memmap's phys address
199 * XXX: need more consideration for sparse_vmemmap...
200 */
201 page = virt_to_page(memmap);
202 mapsize = sizeof(struct page) * PAGES_PER_SECTION;
203 mapsize = PAGE_ALIGN(mapsize) >> PAGE_SHIFT;
204
205 /* remember memmap's page */
206 for (i = 0; i < mapsize; i++, page++)
207 get_page_bootmem(section_nr, page, SECTION_INFO);
208
209 usemap = __nr_to_section(section_nr)->pageblock_flags;
210 page = virt_to_page(usemap);
211
212 mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT;
213
214 for (i = 0; i < mapsize; i++, page++)
Yasunori Gotoaf370fb2008-07-23 21:28:17 -0700215 get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
Yasunori Goto04753272008-04-28 02:13:31 -0700216
217}
Yasuaki Ishimatsu46723bf2013-02-22 16:33:00 -0800218#else /* CONFIG_SPARSEMEM_VMEMMAP */
219static void register_page_bootmem_info_section(unsigned long start_pfn)
220{
221 unsigned long *usemap, mapsize, section_nr, i;
222 struct mem_section *ms;
223 struct page *page, *memmap;
224
225 if (!pfn_valid(start_pfn))
226 return;
227
228 section_nr = pfn_to_section_nr(start_pfn);
229 ms = __nr_to_section(section_nr);
230
231 memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
232
233 register_page_bootmem_memmap(section_nr, memmap, PAGES_PER_SECTION);
234
235 usemap = __nr_to_section(section_nr)->pageblock_flags;
236 page = virt_to_page(usemap);
237
238 mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT;
239
240 for (i = 0; i < mapsize; i++, page++)
241 get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
242}
243#endif /* !CONFIG_SPARSEMEM_VMEMMAP */
Yasunori Goto04753272008-04-28 02:13:31 -0700244
245void register_page_bootmem_info_node(struct pglist_data *pgdat)
246{
247 unsigned long i, pfn, end_pfn, nr_pages;
248 int node = pgdat->node_id;
249 struct page *page;
250 struct zone *zone;
251
252 nr_pages = PAGE_ALIGN(sizeof(struct pglist_data)) >> PAGE_SHIFT;
253 page = virt_to_page(pgdat);
254
255 for (i = 0; i < nr_pages; i++, page++)
256 get_page_bootmem(node, page, NODE_INFO);
257
258 zone = &pgdat->node_zones[0];
259 for (; zone < pgdat->node_zones + MAX_NR_ZONES - 1; zone++) {
Xishi Qiu139c2d72013-09-11 14:21:46 -0700260 if (zone_is_initialized(zone)) {
Yasunori Goto04753272008-04-28 02:13:31 -0700261 nr_pages = zone->wait_table_hash_nr_entries
262 * sizeof(wait_queue_head_t);
263 nr_pages = PAGE_ALIGN(nr_pages) >> PAGE_SHIFT;
264 page = virt_to_page(zone->wait_table);
265
266 for (i = 0; i < nr_pages; i++, page++)
267 get_page_bootmem(node, page, NODE_INFO);
268 }
269 }
270
271 pfn = pgdat->node_start_pfn;
Cody P Schaferc1f19492013-02-22 16:35:32 -0800272 end_pfn = pgdat_end_pfn(pgdat);
Yasunori Goto04753272008-04-28 02:13:31 -0700273
Tang Chen7e9f5eb2013-07-08 16:00:23 -0700274 /* register section info */
qiuxishif14851a2012-09-17 14:09:24 -0700275 for (; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
276 /*
277 * Some platforms can assign the same pfn to multiple nodes - on
278 * node0 as well as nodeN. To avoid registering a pfn against
279 * multiple nodes we check that this pfn does not already
Tang Chen7e9f5eb2013-07-08 16:00:23 -0700280 * reside in some other nodes.
qiuxishif14851a2012-09-17 14:09:24 -0700281 */
282 if (pfn_valid(pfn) && (pfn_to_nid(pfn) == node))
283 register_page_bootmem_info_section(pfn);
284 }
Yasunori Goto04753272008-04-28 02:13:31 -0700285}
Yasuaki Ishimatsu46723bf2013-02-22 16:33:00 -0800286#endif /* CONFIG_HAVE_BOOTMEM_INFO_NODE */
Yasunori Goto04753272008-04-28 02:13:31 -0700287
Fabian Frederickf2765402014-08-06 16:04:57 -0700288static void __meminit grow_zone_span(struct zone *zone, unsigned long start_pfn,
289 unsigned long end_pfn)
Heiko Carstens76cdd582008-05-14 16:05:52 -0700290{
291 unsigned long old_zone_end_pfn;
292
293 zone_span_writelock(zone);
294
Xishi Qiuc33bc312013-09-11 14:21:44 -0700295 old_zone_end_pfn = zone_end_pfn(zone);
Xishi Qiu8080fc02013-09-11 14:21:45 -0700296 if (zone_is_empty(zone) || start_pfn < zone->zone_start_pfn)
Heiko Carstens76cdd582008-05-14 16:05:52 -0700297 zone->zone_start_pfn = start_pfn;
298
299 zone->spanned_pages = max(old_zone_end_pfn, end_pfn) -
300 zone->zone_start_pfn;
301
302 zone_span_writeunlock(zone);
303}
304
Lai Jiangshan511c2ab2012-12-11 16:03:16 -0800305static void resize_zone(struct zone *zone, unsigned long start_pfn,
306 unsigned long end_pfn)
307{
308 zone_span_writelock(zone);
309
Lai Jiangshane455a9b2012-12-11 16:03:20 -0800310 if (end_pfn - start_pfn) {
311 zone->zone_start_pfn = start_pfn;
312 zone->spanned_pages = end_pfn - start_pfn;
313 } else {
314 /*
315 * make it consist as free_area_init_core(),
316 * if spanned_pages = 0, then keep start_pfn = 0
317 */
318 zone->zone_start_pfn = 0;
319 zone->spanned_pages = 0;
320 }
Lai Jiangshan511c2ab2012-12-11 16:03:16 -0800321
322 zone_span_writeunlock(zone);
323}
324
325static void fix_zone_id(struct zone *zone, unsigned long start_pfn,
326 unsigned long end_pfn)
327{
328 enum zone_type zid = zone_idx(zone);
329 int nid = zone->zone_pgdat->node_id;
330 unsigned long pfn;
331
332 for (pfn = start_pfn; pfn < end_pfn; pfn++)
333 set_page_links(pfn_to_page(pfn), zid, nid, pfn);
334}
335
Cody P Schaferf6bbb782013-02-22 16:35:30 -0800336/* Can fail with -ENOMEM from allocating a wait table with vmalloc() or
Santosh Shilimkar9e43aa22014-01-21 15:50:43 -0800337 * alloc_bootmem_node_nopanic()/memblock_virt_alloc_node_nopanic() */
Cody P Schaferf6bbb782013-02-22 16:35:30 -0800338static int __ref ensure_zone_is_initialized(struct zone *zone,
339 unsigned long start_pfn, unsigned long num_pages)
340{
341 if (!zone_is_initialized(zone))
342 return init_currently_empty_zone(zone, start_pfn, num_pages,
343 MEMMAP_HOTPLUG);
344 return 0;
345}
346
Lai Jiangshane455a9b2012-12-11 16:03:20 -0800347static int __meminit move_pfn_range_left(struct zone *z1, struct zone *z2,
Lai Jiangshan511c2ab2012-12-11 16:03:16 -0800348 unsigned long start_pfn, unsigned long end_pfn)
349{
Lai Jiangshane455a9b2012-12-11 16:03:20 -0800350 int ret;
Lai Jiangshan511c2ab2012-12-11 16:03:16 -0800351 unsigned long flags;
Lai Jiangshane455a9b2012-12-11 16:03:20 -0800352 unsigned long z1_start_pfn;
353
Cody P Schafer64dd1b22013-02-22 16:35:31 -0800354 ret = ensure_zone_is_initialized(z1, start_pfn, end_pfn - start_pfn);
355 if (ret)
356 return ret;
Lai Jiangshan511c2ab2012-12-11 16:03:16 -0800357
358 pgdat_resize_lock(z1->zone_pgdat, &flags);
359
360 /* can't move pfns which are higher than @z2 */
Cody P Schafer108bcc92013-02-22 16:35:23 -0800361 if (end_pfn > zone_end_pfn(z2))
Lai Jiangshan511c2ab2012-12-11 16:03:16 -0800362 goto out_fail;
Jiang Liu834405c2013-07-03 15:03:04 -0700363 /* the move out part must be at the left most of @z2 */
Lai Jiangshan511c2ab2012-12-11 16:03:16 -0800364 if (start_pfn > z2->zone_start_pfn)
365 goto out_fail;
366 /* must included/overlap */
367 if (end_pfn <= z2->zone_start_pfn)
368 goto out_fail;
369
Lai Jiangshane455a9b2012-12-11 16:03:20 -0800370 /* use start_pfn for z1's start_pfn if z1 is empty */
Xishi Qiu8080fc02013-09-11 14:21:45 -0700371 if (!zone_is_empty(z1))
Lai Jiangshane455a9b2012-12-11 16:03:20 -0800372 z1_start_pfn = z1->zone_start_pfn;
373 else
374 z1_start_pfn = start_pfn;
375
376 resize_zone(z1, z1_start_pfn, end_pfn);
Cody P Schafer108bcc92013-02-22 16:35:23 -0800377 resize_zone(z2, end_pfn, zone_end_pfn(z2));
Lai Jiangshan511c2ab2012-12-11 16:03:16 -0800378
379 pgdat_resize_unlock(z1->zone_pgdat, &flags);
380
381 fix_zone_id(z1, start_pfn, end_pfn);
382
383 return 0;
384out_fail:
385 pgdat_resize_unlock(z1->zone_pgdat, &flags);
386 return -1;
387}
388
Lai Jiangshane455a9b2012-12-11 16:03:20 -0800389static int __meminit move_pfn_range_right(struct zone *z1, struct zone *z2,
Lai Jiangshan511c2ab2012-12-11 16:03:16 -0800390 unsigned long start_pfn, unsigned long end_pfn)
391{
Lai Jiangshane455a9b2012-12-11 16:03:20 -0800392 int ret;
Lai Jiangshan511c2ab2012-12-11 16:03:16 -0800393 unsigned long flags;
Lai Jiangshane455a9b2012-12-11 16:03:20 -0800394 unsigned long z2_end_pfn;
395
Cody P Schafer64dd1b22013-02-22 16:35:31 -0800396 ret = ensure_zone_is_initialized(z2, start_pfn, end_pfn - start_pfn);
397 if (ret)
398 return ret;
Lai Jiangshan511c2ab2012-12-11 16:03:16 -0800399
400 pgdat_resize_lock(z1->zone_pgdat, &flags);
401
402 /* can't move pfns which are lower than @z1 */
403 if (z1->zone_start_pfn > start_pfn)
404 goto out_fail;
405 /* the move out part mast at the right most of @z1 */
Cody P Schafer108bcc92013-02-22 16:35:23 -0800406 if (zone_end_pfn(z1) > end_pfn)
Lai Jiangshan511c2ab2012-12-11 16:03:16 -0800407 goto out_fail;
408 /* must included/overlap */
Cody P Schafer108bcc92013-02-22 16:35:23 -0800409 if (start_pfn >= zone_end_pfn(z1))
Lai Jiangshan511c2ab2012-12-11 16:03:16 -0800410 goto out_fail;
411
Lai Jiangshane455a9b2012-12-11 16:03:20 -0800412 /* use end_pfn for z2's end_pfn if z2 is empty */
Xishi Qiu8080fc02013-09-11 14:21:45 -0700413 if (!zone_is_empty(z2))
Cody P Schafer108bcc92013-02-22 16:35:23 -0800414 z2_end_pfn = zone_end_pfn(z2);
Lai Jiangshane455a9b2012-12-11 16:03:20 -0800415 else
416 z2_end_pfn = end_pfn;
417
Lai Jiangshan511c2ab2012-12-11 16:03:16 -0800418 resize_zone(z1, z1->zone_start_pfn, start_pfn);
Lai Jiangshane455a9b2012-12-11 16:03:20 -0800419 resize_zone(z2, start_pfn, z2_end_pfn);
Lai Jiangshan511c2ab2012-12-11 16:03:16 -0800420
421 pgdat_resize_unlock(z1->zone_pgdat, &flags);
422
423 fix_zone_id(z2, start_pfn, end_pfn);
424
425 return 0;
426out_fail:
427 pgdat_resize_unlock(z1->zone_pgdat, &flags);
428 return -1;
429}
430
Fabian Frederickf2765402014-08-06 16:04:57 -0700431static void __meminit grow_pgdat_span(struct pglist_data *pgdat, unsigned long start_pfn,
432 unsigned long end_pfn)
Heiko Carstens76cdd582008-05-14 16:05:52 -0700433{
Xishi Qiu83285c72013-11-12 15:07:19 -0800434 unsigned long old_pgdat_end_pfn = pgdat_end_pfn(pgdat);
Heiko Carstens76cdd582008-05-14 16:05:52 -0700435
Tang Chen712cd382012-12-11 16:01:07 -0800436 if (!pgdat->node_spanned_pages || start_pfn < pgdat->node_start_pfn)
Heiko Carstens76cdd582008-05-14 16:05:52 -0700437 pgdat->node_start_pfn = start_pfn;
438
439 pgdat->node_spanned_pages = max(old_pgdat_end_pfn, end_pfn) -
440 pgdat->node_start_pfn;
441}
442
Al Viro31168482008-11-22 17:33:24 +0000443static int __meminit __add_zone(struct zone *zone, unsigned long phys_start_pfn)
Dave Hansen3947be12005-10-29 18:16:54 -0700444{
445 struct pglist_data *pgdat = zone->zone_pgdat;
446 int nr_pages = PAGES_PER_SECTION;
447 int nid = pgdat->node_id;
448 int zone_type;
Heiko Carstens76cdd582008-05-14 16:05:52 -0700449 unsigned long flags;
Cody P Schafer64dd1b22013-02-22 16:35:31 -0800450 int ret;
Dave Hansen3947be12005-10-29 18:16:54 -0700451
452 zone_type = zone - pgdat->node_zones;
Cody P Schafer64dd1b22013-02-22 16:35:31 -0800453 ret = ensure_zone_is_initialized(zone, phys_start_pfn, nr_pages);
454 if (ret)
455 return ret;
Heiko Carstens76cdd582008-05-14 16:05:52 -0700456
Heiko Carstens76cdd582008-05-14 16:05:52 -0700457 pgdat_resize_lock(zone->zone_pgdat, &flags);
458 grow_zone_span(zone, phys_start_pfn, phys_start_pfn + nr_pages);
459 grow_pgdat_span(zone->zone_pgdat, phys_start_pfn,
460 phys_start_pfn + nr_pages);
461 pgdat_resize_unlock(zone->zone_pgdat, &flags);
Dave Hansena2f3aa022007-01-10 23:15:30 -0800462 memmap_init_zone(nr_pages, nid, zone_type,
463 phys_start_pfn, MEMMAP_HOTPLUG);
Yasunori Goto718127c2006-06-23 02:03:10 -0700464 return 0;
Dave Hansen3947be12005-10-29 18:16:54 -0700465}
466
Gary Hadec04fc582009-01-06 14:39:14 -0800467static int __meminit __add_section(int nid, struct zone *zone,
468 unsigned long phys_start_pfn)
Dave Hansen3947be12005-10-29 18:16:54 -0700469{
Dave Hansen3947be12005-10-29 18:16:54 -0700470 int ret;
471
KAMEZAWA Hiroyukiebd15302006-08-05 12:15:06 -0700472 if (pfn_valid(phys_start_pfn))
473 return -EEXIST;
474
Zhang Yanfei85b35fe2013-11-12 15:07:42 -0800475 ret = sparse_add_one_section(zone, phys_start_pfn);
Dave Hansen3947be12005-10-29 18:16:54 -0700476
477 if (ret < 0)
478 return ret;
479
Yasunori Goto718127c2006-06-23 02:03:10 -0700480 ret = __add_zone(zone, phys_start_pfn);
481
482 if (ret < 0)
483 return ret;
484
Gary Hadec04fc582009-01-06 14:39:14 -0800485 return register_new_memory(nid, __pfn_to_section(phys_start_pfn));
Dave Hansen3947be12005-10-29 18:16:54 -0700486}
487
David Rientjes4edd7ce2013-04-29 15:08:22 -0700488/*
489 * Reasonably generic function for adding memory. It is
490 * expected that archs that support memory hotplug will
491 * call this function after deciding the zone to which to
492 * add the new pages.
493 */
494int __ref __add_pages(int nid, struct zone *zone, unsigned long phys_start_pfn,
495 unsigned long nr_pages)
496{
497 unsigned long i;
498 int err = 0;
499 int start_sec, end_sec;
500 /* during initialize mem_map, align hot-added range to section */
501 start_sec = pfn_to_section_nr(phys_start_pfn);
502 end_sec = pfn_to_section_nr(phys_start_pfn + nr_pages - 1);
503
504 for (i = start_sec; i <= end_sec; i++) {
505 err = __add_section(nid, zone, i << PFN_SECTION_SHIFT);
506
507 /*
508 * EEXIST is finally dealt with by ioresource collision
509 * check. see add_memory() => register_memory_resource()
510 * Warning will be printed if there is collision.
511 */
512 if (err && (err != -EEXIST))
513 break;
514 err = 0;
515 }
516
517 return err;
518}
519EXPORT_SYMBOL_GPL(__add_pages);
520
521#ifdef CONFIG_MEMORY_HOTREMOVE
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800522/* find the smallest valid pfn in the range [start_pfn, end_pfn) */
523static int find_smallest_section_pfn(int nid, struct zone *zone,
524 unsigned long start_pfn,
525 unsigned long end_pfn)
526{
527 struct mem_section *ms;
528
529 for (; start_pfn < end_pfn; start_pfn += PAGES_PER_SECTION) {
530 ms = __pfn_to_section(start_pfn);
531
532 if (unlikely(!valid_section(ms)))
533 continue;
534
535 if (unlikely(pfn_to_nid(start_pfn) != nid))
536 continue;
537
538 if (zone && zone != page_zone(pfn_to_page(start_pfn)))
539 continue;
540
541 return start_pfn;
542 }
543
544 return 0;
545}
546
547/* find the biggest valid pfn in the range [start_pfn, end_pfn). */
548static int find_biggest_section_pfn(int nid, struct zone *zone,
549 unsigned long start_pfn,
550 unsigned long end_pfn)
551{
552 struct mem_section *ms;
553 unsigned long pfn;
554
555 /* pfn is the end pfn of a memory section. */
556 pfn = end_pfn - 1;
557 for (; pfn >= start_pfn; pfn -= PAGES_PER_SECTION) {
558 ms = __pfn_to_section(pfn);
559
560 if (unlikely(!valid_section(ms)))
561 continue;
562
563 if (unlikely(pfn_to_nid(pfn) != nid))
564 continue;
565
566 if (zone && zone != page_zone(pfn_to_page(pfn)))
567 continue;
568
569 return pfn;
570 }
571
572 return 0;
573}
574
575static void shrink_zone_span(struct zone *zone, unsigned long start_pfn,
576 unsigned long end_pfn)
577{
Xishi Qiuc33bc312013-09-11 14:21:44 -0700578 unsigned long zone_start_pfn = zone->zone_start_pfn;
579 unsigned long z = zone_end_pfn(zone); /* zone_end_pfn namespace clash */
580 unsigned long zone_end_pfn = z;
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800581 unsigned long pfn;
582 struct mem_section *ms;
583 int nid = zone_to_nid(zone);
584
585 zone_span_writelock(zone);
586 if (zone_start_pfn == start_pfn) {
587 /*
588 * If the section is smallest section in the zone, it need
589 * shrink zone->zone_start_pfn and zone->zone_spanned_pages.
590 * In this case, we find second smallest valid mem_section
591 * for shrinking zone.
592 */
593 pfn = find_smallest_section_pfn(nid, zone, end_pfn,
594 zone_end_pfn);
595 if (pfn) {
596 zone->zone_start_pfn = pfn;
597 zone->spanned_pages = zone_end_pfn - pfn;
598 }
599 } else if (zone_end_pfn == end_pfn) {
600 /*
601 * If the section is biggest section in the zone, it need
602 * shrink zone->spanned_pages.
603 * In this case, we find second biggest valid mem_section for
604 * shrinking zone.
605 */
606 pfn = find_biggest_section_pfn(nid, zone, zone_start_pfn,
607 start_pfn);
608 if (pfn)
609 zone->spanned_pages = pfn - zone_start_pfn + 1;
610 }
611
612 /*
613 * The section is not biggest or smallest mem_section in the zone, it
614 * only creates a hole in the zone. So in this case, we need not
615 * change the zone. But perhaps, the zone has only hole data. Thus
616 * it check the zone has only hole or not.
617 */
618 pfn = zone_start_pfn;
619 for (; pfn < zone_end_pfn; pfn += PAGES_PER_SECTION) {
620 ms = __pfn_to_section(pfn);
621
622 if (unlikely(!valid_section(ms)))
623 continue;
624
625 if (page_zone(pfn_to_page(pfn)) != zone)
626 continue;
627
628 /* If the section is current section, it continues the loop */
629 if (start_pfn == pfn)
630 continue;
631
632 /* If we find valid section, we have nothing to do */
633 zone_span_writeunlock(zone);
634 return;
635 }
636
637 /* The zone has no valid section */
638 zone->zone_start_pfn = 0;
639 zone->spanned_pages = 0;
640 zone_span_writeunlock(zone);
641}
642
643static void shrink_pgdat_span(struct pglist_data *pgdat,
644 unsigned long start_pfn, unsigned long end_pfn)
645{
Xishi Qiu83285c72013-11-12 15:07:19 -0800646 unsigned long pgdat_start_pfn = pgdat->node_start_pfn;
647 unsigned long p = pgdat_end_pfn(pgdat); /* pgdat_end_pfn namespace clash */
648 unsigned long pgdat_end_pfn = p;
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800649 unsigned long pfn;
650 struct mem_section *ms;
651 int nid = pgdat->node_id;
652
653 if (pgdat_start_pfn == start_pfn) {
654 /*
655 * If the section is smallest section in the pgdat, it need
656 * shrink pgdat->node_start_pfn and pgdat->node_spanned_pages.
657 * In this case, we find second smallest valid mem_section
658 * for shrinking zone.
659 */
660 pfn = find_smallest_section_pfn(nid, NULL, end_pfn,
661 pgdat_end_pfn);
662 if (pfn) {
663 pgdat->node_start_pfn = pfn;
664 pgdat->node_spanned_pages = pgdat_end_pfn - pfn;
665 }
666 } else if (pgdat_end_pfn == end_pfn) {
667 /*
668 * If the section is biggest section in the pgdat, it need
669 * shrink pgdat->node_spanned_pages.
670 * In this case, we find second biggest valid mem_section for
671 * shrinking zone.
672 */
673 pfn = find_biggest_section_pfn(nid, NULL, pgdat_start_pfn,
674 start_pfn);
675 if (pfn)
676 pgdat->node_spanned_pages = pfn - pgdat_start_pfn + 1;
677 }
678
679 /*
680 * If the section is not biggest or smallest mem_section in the pgdat,
681 * it only creates a hole in the pgdat. So in this case, we need not
682 * change the pgdat.
683 * But perhaps, the pgdat has only hole data. Thus it check the pgdat
684 * has only hole or not.
685 */
686 pfn = pgdat_start_pfn;
687 for (; pfn < pgdat_end_pfn; pfn += PAGES_PER_SECTION) {
688 ms = __pfn_to_section(pfn);
689
690 if (unlikely(!valid_section(ms)))
691 continue;
692
693 if (pfn_to_nid(pfn) != nid)
694 continue;
695
696 /* If the section is current section, it continues the loop */
697 if (start_pfn == pfn)
698 continue;
699
700 /* If we find valid section, we have nothing to do */
701 return;
702 }
703
704 /* The pgdat has no valid section */
705 pgdat->node_start_pfn = 0;
706 pgdat->node_spanned_pages = 0;
707}
708
709static void __remove_zone(struct zone *zone, unsigned long start_pfn)
710{
711 struct pglist_data *pgdat = zone->zone_pgdat;
712 int nr_pages = PAGES_PER_SECTION;
713 int zone_type;
714 unsigned long flags;
715
716 zone_type = zone - pgdat->node_zones;
717
718 pgdat_resize_lock(zone->zone_pgdat, &flags);
719 shrink_zone_span(zone, start_pfn, start_pfn + nr_pages);
720 shrink_pgdat_span(pgdat, start_pfn, start_pfn + nr_pages);
721 pgdat_resize_unlock(zone->zone_pgdat, &flags);
722}
723
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700724static int __remove_section(struct zone *zone, struct mem_section *ms)
725{
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800726 unsigned long start_pfn;
727 int scn_nr;
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700728 int ret = -EINVAL;
729
730 if (!valid_section(ms))
731 return ret;
732
733 ret = unregister_memory_section(ms);
734 if (ret)
735 return ret;
736
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800737 scn_nr = __section_nr(ms);
738 start_pfn = section_nr_to_pfn(scn_nr);
739 __remove_zone(zone, start_pfn);
740
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700741 sparse_remove_one_section(zone, ms);
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700742 return 0;
743}
744
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700745/**
746 * __remove_pages() - remove sections of pages from a zone
747 * @zone: zone from which pages need to be removed
748 * @phys_start_pfn: starting pageframe (must be aligned to start of a section)
749 * @nr_pages: number of pages to remove (must be multiple of section size)
750 *
751 * Generic helper function to remove section mappings and sysfs entries
752 * for the section of the memory we are removing. Caller needs to make
753 * sure that pages are marked reserved and zones are adjust properly by
754 * calling offline_pages().
755 */
756int __remove_pages(struct zone *zone, unsigned long phys_start_pfn,
757 unsigned long nr_pages)
758{
Toshi Kanife74ebb2013-04-29 15:08:20 -0700759 unsigned long i;
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700760 int sections_to_remove;
Toshi Kanife74ebb2013-04-29 15:08:20 -0700761 resource_size_t start, size;
762 int ret = 0;
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700763
764 /*
765 * We can only remove entire sections
766 */
767 BUG_ON(phys_start_pfn & ~PAGE_SECTION_MASK);
768 BUG_ON(nr_pages % PAGES_PER_SECTION);
769
Toshi Kanife74ebb2013-04-29 15:08:20 -0700770 start = phys_start_pfn << PAGE_SHIFT;
771 size = nr_pages * PAGE_SIZE;
772 ret = release_mem_region_adjustable(&iomem_resource, start, size);
Randy Dunlap348f9f02013-05-24 15:55:30 -0700773 if (ret) {
774 resource_size_t endres = start + size - 1;
775
776 pr_warn("Unable to release resource <%pa-%pa> (%d)\n",
777 &start, &endres, ret);
778 }
Yasuaki Ishimatsud760afd2012-10-08 16:34:14 -0700779
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700780 sections_to_remove = nr_pages / PAGES_PER_SECTION;
781 for (i = 0; i < sections_to_remove; i++) {
782 unsigned long pfn = phys_start_pfn + i*PAGES_PER_SECTION;
783 ret = __remove_section(zone, __pfn_to_section(pfn));
784 if (ret)
785 break;
786 }
787 return ret;
788}
789EXPORT_SYMBOL_GPL(__remove_pages);
David Rientjes4edd7ce2013-04-29 15:08:22 -0700790#endif /* CONFIG_MEMORY_HOTREMOVE */
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700791
Daniel Kiper9d0ad8c2011-07-25 17:12:05 -0700792int set_online_page_callback(online_page_callback_t callback)
793{
794 int rc = -EINVAL;
795
Vladimir Davydovbfc8c902014-06-04 16:07:18 -0700796 get_online_mems();
797 mutex_lock(&online_page_callback_lock);
Daniel Kiper9d0ad8c2011-07-25 17:12:05 -0700798
799 if (online_page_callback == generic_online_page) {
800 online_page_callback = callback;
801 rc = 0;
802 }
803
Vladimir Davydovbfc8c902014-06-04 16:07:18 -0700804 mutex_unlock(&online_page_callback_lock);
805 put_online_mems();
Daniel Kiper9d0ad8c2011-07-25 17:12:05 -0700806
807 return rc;
808}
809EXPORT_SYMBOL_GPL(set_online_page_callback);
810
811int restore_online_page_callback(online_page_callback_t callback)
812{
813 int rc = -EINVAL;
814
Vladimir Davydovbfc8c902014-06-04 16:07:18 -0700815 get_online_mems();
816 mutex_lock(&online_page_callback_lock);
Daniel Kiper9d0ad8c2011-07-25 17:12:05 -0700817
818 if (online_page_callback == callback) {
819 online_page_callback = generic_online_page;
820 rc = 0;
821 }
822
Vladimir Davydovbfc8c902014-06-04 16:07:18 -0700823 mutex_unlock(&online_page_callback_lock);
824 put_online_mems();
Daniel Kiper9d0ad8c2011-07-25 17:12:05 -0700825
826 return rc;
827}
828EXPORT_SYMBOL_GPL(restore_online_page_callback);
829
830void __online_page_set_limits(struct page *page)
Jeremy Fitzhardinge180c06e2008-04-28 02:12:03 -0700831{
Daniel Kiper9d0ad8c2011-07-25 17:12:05 -0700832}
833EXPORT_SYMBOL_GPL(__online_page_set_limits);
834
835void __online_page_increment_counters(struct page *page)
836{
Jiang Liu3dcc0572013-07-03 15:03:21 -0700837 adjust_managed_page_count(page, 1);
Daniel Kiper9d0ad8c2011-07-25 17:12:05 -0700838}
839EXPORT_SYMBOL_GPL(__online_page_increment_counters);
Jeremy Fitzhardinge180c06e2008-04-28 02:12:03 -0700840
Daniel Kiper9d0ad8c2011-07-25 17:12:05 -0700841void __online_page_free(struct page *page)
842{
Jiang Liu3dcc0572013-07-03 15:03:21 -0700843 __free_reserved_page(page);
Jeremy Fitzhardinge180c06e2008-04-28 02:12:03 -0700844}
Daniel Kiper9d0ad8c2011-07-25 17:12:05 -0700845EXPORT_SYMBOL_GPL(__online_page_free);
846
847static void generic_online_page(struct page *page)
848{
849 __online_page_set_limits(page);
850 __online_page_increment_counters(page);
851 __online_page_free(page);
852}
Jeremy Fitzhardinge180c06e2008-04-28 02:12:03 -0700853
KAMEZAWA Hiroyuki75884fb2007-10-16 01:26:10 -0700854static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages,
855 void *arg)
Dave Hansen3947be12005-10-29 18:16:54 -0700856{
857 unsigned long i;
KAMEZAWA Hiroyuki75884fb2007-10-16 01:26:10 -0700858 unsigned long onlined_pages = *(unsigned long *)arg;
859 struct page *page;
860 if (PageReserved(pfn_to_page(start_pfn)))
861 for (i = 0; i < nr_pages; i++) {
862 page = pfn_to_page(start_pfn + i);
Daniel Kiper9d0ad8c2011-07-25 17:12:05 -0700863 (*online_page_callback)(page);
KAMEZAWA Hiroyuki75884fb2007-10-16 01:26:10 -0700864 onlined_pages++;
865 }
866 *(unsigned long *)arg = onlined_pages;
867 return 0;
868}
869
Lai Jiangshan09285af2012-12-12 13:52:04 -0800870#ifdef CONFIG_MOVABLE_NODE
Tang Chen79a4dce2012-12-18 14:23:24 -0800871/*
872 * When CONFIG_MOVABLE_NODE, we permit onlining of a node which doesn't have
873 * normal memory.
874 */
Lai Jiangshan09285af2012-12-12 13:52:04 -0800875static bool can_online_high_movable(struct zone *zone)
876{
877 return true;
878}
Tang Chen79a4dce2012-12-18 14:23:24 -0800879#else /* CONFIG_MOVABLE_NODE */
Lai Jiangshan74d42d82012-12-11 16:03:23 -0800880/* ensure every online node has NORMAL memory */
881static bool can_online_high_movable(struct zone *zone)
882{
883 return node_state(zone_to_nid(zone), N_NORMAL_MEMORY);
884}
Tang Chen79a4dce2012-12-18 14:23:24 -0800885#endif /* CONFIG_MOVABLE_NODE */
Lai Jiangshan74d42d82012-12-11 16:03:23 -0800886
Lai Jiangshand9713672012-12-11 16:01:03 -0800887/* check which state of node_states will be changed when online memory */
888static void node_states_check_changes_online(unsigned long nr_pages,
889 struct zone *zone, struct memory_notify *arg)
890{
891 int nid = zone_to_nid(zone);
892 enum zone_type zone_last = ZONE_NORMAL;
893
894 /*
Lai Jiangshan6715ddf2012-12-12 13:51:49 -0800895 * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY]
896 * contains nodes which have zones of 0...ZONE_NORMAL,
897 * set zone_last to ZONE_NORMAL.
Lai Jiangshand9713672012-12-11 16:01:03 -0800898 *
Lai Jiangshan6715ddf2012-12-12 13:51:49 -0800899 * If we don't have HIGHMEM nor movable node,
900 * node_states[N_NORMAL_MEMORY] contains nodes which have zones of
901 * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE.
Lai Jiangshand9713672012-12-11 16:01:03 -0800902 */
Lai Jiangshan6715ddf2012-12-12 13:51:49 -0800903 if (N_MEMORY == N_NORMAL_MEMORY)
Lai Jiangshand9713672012-12-11 16:01:03 -0800904 zone_last = ZONE_MOVABLE;
905
906 /*
907 * if the memory to be online is in a zone of 0...zone_last, and
908 * the zones of 0...zone_last don't have memory before online, we will
909 * need to set the node to node_states[N_NORMAL_MEMORY] after
910 * the memory is online.
911 */
912 if (zone_idx(zone) <= zone_last && !node_state(nid, N_NORMAL_MEMORY))
913 arg->status_change_nid_normal = nid;
914 else
915 arg->status_change_nid_normal = -1;
916
Lai Jiangshan6715ddf2012-12-12 13:51:49 -0800917#ifdef CONFIG_HIGHMEM
918 /*
919 * If we have movable node, node_states[N_HIGH_MEMORY]
920 * contains nodes which have zones of 0...ZONE_HIGHMEM,
921 * set zone_last to ZONE_HIGHMEM.
922 *
923 * If we don't have movable node, node_states[N_NORMAL_MEMORY]
924 * contains nodes which have zones of 0...ZONE_MOVABLE,
925 * set zone_last to ZONE_MOVABLE.
926 */
927 zone_last = ZONE_HIGHMEM;
928 if (N_MEMORY == N_HIGH_MEMORY)
929 zone_last = ZONE_MOVABLE;
930
931 if (zone_idx(zone) <= zone_last && !node_state(nid, N_HIGH_MEMORY))
932 arg->status_change_nid_high = nid;
933 else
934 arg->status_change_nid_high = -1;
935#else
936 arg->status_change_nid_high = arg->status_change_nid_normal;
937#endif
938
Lai Jiangshand9713672012-12-11 16:01:03 -0800939 /*
940 * if the node don't have memory befor online, we will need to
Lai Jiangshan6715ddf2012-12-12 13:51:49 -0800941 * set the node to node_states[N_MEMORY] after the memory
Lai Jiangshand9713672012-12-11 16:01:03 -0800942 * is online.
943 */
Lai Jiangshan6715ddf2012-12-12 13:51:49 -0800944 if (!node_state(nid, N_MEMORY))
Lai Jiangshand9713672012-12-11 16:01:03 -0800945 arg->status_change_nid = nid;
946 else
947 arg->status_change_nid = -1;
948}
949
950static void node_states_set_node(int node, struct memory_notify *arg)
951{
952 if (arg->status_change_nid_normal >= 0)
953 node_set_state(node, N_NORMAL_MEMORY);
954
Lai Jiangshan6715ddf2012-12-12 13:51:49 -0800955 if (arg->status_change_nid_high >= 0)
956 node_set_state(node, N_HIGH_MEMORY);
957
958 node_set_state(node, N_MEMORY);
Lai Jiangshand9713672012-12-11 16:01:03 -0800959}
960
KAMEZAWA Hiroyuki75884fb2007-10-16 01:26:10 -0700961
Lai Jiangshan511c2ab2012-12-11 16:03:16 -0800962int __ref online_pages(unsigned long pfn, unsigned long nr_pages, int online_type)
KAMEZAWA Hiroyuki75884fb2007-10-16 01:26:10 -0700963{
Cody P Schaferaa472282013-07-03 15:02:10 -0700964 unsigned long flags;
Dave Hansen3947be12005-10-29 18:16:54 -0700965 unsigned long onlined_pages = 0;
966 struct zone *zone;
Yasunori Goto68113782006-06-23 02:03:11 -0700967 int need_zonelists_rebuild = 0;
Yasunori Goto7b78d332007-10-21 16:41:36 -0700968 int nid;
969 int ret;
970 struct memory_notify arg;
Dave Hansen3947be12005-10-29 18:16:54 -0700971
Vladimir Davydovbfc8c902014-06-04 16:07:18 -0700972 mem_hotplug_begin();
Lai Jiangshand9713672012-12-11 16:01:03 -0800973 /*
974 * This doesn't need a lock to do pfn_to_page().
975 * The section can't be removed here because of the
976 * memory_block->state_mutex.
977 */
978 zone = page_zone(pfn_to_page(pfn));
979
Vladimir Davydovbfc8c902014-06-04 16:07:18 -0700980 ret = -EINVAL;
Tang Chen4f7c6b42014-08-06 16:05:13 -0700981 if ((zone_idx(zone) > ZONE_NORMAL ||
982 online_type == MMOP_ONLINE_MOVABLE) &&
Vladimir Davydovbfc8c902014-06-04 16:07:18 -0700983 !can_online_high_movable(zone))
984 goto out;
Lai Jiangshan74d42d82012-12-11 16:03:23 -0800985
Tang Chen4f7c6b42014-08-06 16:05:13 -0700986 if (online_type == MMOP_ONLINE_KERNEL &&
987 zone_idx(zone) == ZONE_MOVABLE) {
Vladimir Davydovbfc8c902014-06-04 16:07:18 -0700988 if (move_pfn_range_left(zone - 1, zone, pfn, pfn + nr_pages))
989 goto out;
Lai Jiangshan511c2ab2012-12-11 16:03:16 -0800990 }
Tang Chen4f7c6b42014-08-06 16:05:13 -0700991 if (online_type == MMOP_ONLINE_MOVABLE &&
992 zone_idx(zone) == ZONE_MOVABLE - 1) {
Vladimir Davydovbfc8c902014-06-04 16:07:18 -0700993 if (move_pfn_range_right(zone, zone + 1, pfn, pfn + nr_pages))
994 goto out;
Lai Jiangshan511c2ab2012-12-11 16:03:16 -0800995 }
996
997 /* Previous code may changed the zone of the pfn range */
998 zone = page_zone(pfn_to_page(pfn));
999
Yasunori Goto7b78d332007-10-21 16:41:36 -07001000 arg.start_pfn = pfn;
1001 arg.nr_pages = nr_pages;
Lai Jiangshand9713672012-12-11 16:01:03 -08001002 node_states_check_changes_online(nr_pages, zone, &arg);
Yasunori Goto7b78d332007-10-21 16:41:36 -07001003
Xishi Qiu9c2606b2013-11-12 15:07:21 -08001004 nid = pfn_to_nid(pfn);
Yasunori Goto7b78d332007-10-21 16:41:36 -07001005
1006 ret = memory_notify(MEM_GOING_ONLINE, &arg);
1007 ret = notifier_to_errno(ret);
1008 if (ret) {
1009 memory_notify(MEM_CANCEL_ONLINE, &arg);
Vladimir Davydovbfc8c902014-06-04 16:07:18 -07001010 goto out;
Yasunori Goto7b78d332007-10-21 16:41:36 -07001011 }
Dave Hansen3947be12005-10-29 18:16:54 -07001012 /*
Yasunori Goto68113782006-06-23 02:03:11 -07001013 * If this zone is not populated, then it is not in zonelist.
1014 * This means the page allocator ignores this zone.
1015 * So, zonelist must be updated after online.
1016 */
Haicheng Li4eaf3f62010-05-24 14:32:52 -07001017 mutex_lock(&zonelists_mutex);
Wen Congyang6dcd73d2012-12-11 16:01:01 -08001018 if (!populated_zone(zone)) {
Yasunori Goto68113782006-06-23 02:03:11 -07001019 need_zonelists_rebuild = 1;
Wen Congyang6dcd73d2012-12-11 16:01:01 -08001020 build_all_zonelists(NULL, zone);
1021 }
Yasunori Goto68113782006-06-23 02:03:11 -07001022
KAMEZAWA Hiroyuki908eedc2009-09-22 16:45:46 -07001023 ret = walk_system_ram_range(pfn, nr_pages, &onlined_pages,
KAMEZAWA Hiroyuki75884fb2007-10-16 01:26:10 -07001024 online_pages_range);
Geoff Levandfd8a4222008-05-14 16:05:50 -07001025 if (ret) {
Wen Congyang6dcd73d2012-12-11 16:01:01 -08001026 if (need_zonelists_rebuild)
1027 zone_pcp_reset(zone);
Haicheng Li4eaf3f62010-05-24 14:32:52 -07001028 mutex_unlock(&zonelists_mutex);
Bjorn Helgaasa62e2f42012-05-29 15:06:30 -07001029 printk(KERN_DEBUG "online_pages [mem %#010llx-%#010llx] failed\n",
1030 (unsigned long long) pfn << PAGE_SHIFT,
1031 (((unsigned long long) pfn + nr_pages)
1032 << PAGE_SHIFT) - 1);
Geoff Levandfd8a4222008-05-14 16:05:50 -07001033 memory_notify(MEM_CANCEL_ONLINE, &arg);
Vladimir Davydovbfc8c902014-06-04 16:07:18 -07001034 goto out;
Geoff Levandfd8a4222008-05-14 16:05:50 -07001035 }
1036
Dave Hansen3947be12005-10-29 18:16:54 -07001037 zone->present_pages += onlined_pages;
Cody P Schaferaa472282013-07-03 15:02:10 -07001038
1039 pgdat_resize_lock(zone->zone_pgdat, &flags);
Yasunori Gotof2937be2006-03-09 17:33:51 -08001040 zone->zone_pgdat->node_present_pages += onlined_pages;
Cody P Schaferaa472282013-07-03 15:02:10 -07001041 pgdat_resize_unlock(zone->zone_pgdat, &flags);
1042
Jiang Liu08dff7b2012-07-31 16:43:30 -07001043 if (onlined_pages) {
Lai Jiangshand9713672012-12-11 16:01:03 -08001044 node_states_set_node(zone_to_nid(zone), &arg);
Jiang Liu08dff7b2012-07-31 16:43:30 -07001045 if (need_zonelists_rebuild)
Wen Congyang6dcd73d2012-12-11 16:01:01 -08001046 build_all_zonelists(NULL, NULL);
Jiang Liu08dff7b2012-07-31 16:43:30 -07001047 else
1048 zone_pcp_update(zone);
1049 }
Dave Hansen3947be12005-10-29 18:16:54 -07001050
Haicheng Li4eaf3f62010-05-24 14:32:52 -07001051 mutex_unlock(&zonelists_mutex);
KOSAKI Motohiro1b79acc2011-05-24 17:11:32 -07001052
1053 init_per_zone_wmark_min();
1054
Jiang Liu08dff7b2012-07-31 16:43:30 -07001055 if (onlined_pages)
Christoph Lameter7ea15302007-10-16 01:25:29 -07001056 kswapd_run(zone_to_nid(zone));
Dave Hansen61b13992005-10-29 18:16:56 -07001057
Haicheng Li1f522502010-05-24 14:32:51 -07001058 vm_total_pages = nr_free_pagecache_pages();
Kent Liu2f7f24e2008-07-23 21:28:18 -07001059
Chandra Seetharaman2d1d43f2006-09-29 02:01:25 -07001060 writeback_set_ratelimit();
Yasunori Goto7b78d332007-10-21 16:41:36 -07001061
1062 if (onlined_pages)
1063 memory_notify(MEM_ONLINE, &arg);
Vladimir Davydovbfc8c902014-06-04 16:07:18 -07001064out:
1065 mem_hotplug_done();
1066 return ret;
Dave Hansen3947be12005-10-29 18:16:54 -07001067}
Keith Mannthey53947022006-09-30 23:27:08 -07001068#endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
Yasunori Gotobc02af92006-06-27 02:53:30 -07001069
Tang Chen0bd85422014-11-13 15:19:41 -08001070static void reset_node_present_pages(pg_data_t *pgdat)
1071{
1072 struct zone *z;
1073
1074 for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++)
1075 z->present_pages = 0;
1076
1077 pgdat->node_present_pages = 0;
1078}
1079
Hidetoshi Setoe1319332009-11-17 14:06:18 -08001080/* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
1081static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start)
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001082{
1083 struct pglist_data *pgdat;
1084 unsigned long zones_size[MAX_NR_ZONES] = {0};
1085 unsigned long zholes_size[MAX_NR_ZONES] = {0};
Fabian Frederickc8e861a2014-06-04 16:07:51 -07001086 unsigned long start_pfn = PFN_DOWN(start);
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001087
Tang Chena1e565a2013-02-22 16:33:18 -08001088 pgdat = NODE_DATA(nid);
1089 if (!pgdat) {
1090 pgdat = arch_alloc_nodedata(nid);
1091 if (!pgdat)
1092 return NULL;
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001093
Tang Chena1e565a2013-02-22 16:33:18 -08001094 arch_refresh_nodedata(nid, pgdat);
1095 }
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001096
1097 /* we can use NODE_DATA(nid) from here */
1098
1099 /* init node's zones as empty zones, we don't have any present pages.*/
Johannes Weiner9109fb72008-07-23 21:27:20 -07001100 free_area_init_node(nid, zones_size, start_pfn, zholes_size);
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001101
KAMEZAWA Hiroyuki959ecc42011-06-15 15:08:38 -07001102 /*
1103 * The node we allocated has no zone fallback lists. For avoiding
1104 * to access not-initialized zonelist, build here.
1105 */
David Rientjesf957db42011-06-22 18:13:04 -07001106 mutex_lock(&zonelists_mutex);
Jiang Liu9adb62a2012-07-31 16:43:28 -07001107 build_all_zonelists(pgdat, NULL);
David Rientjesf957db42011-06-22 18:13:04 -07001108 mutex_unlock(&zonelists_mutex);
KAMEZAWA Hiroyuki959ecc42011-06-15 15:08:38 -07001109
Tang Chenf784a3f2014-11-13 15:19:39 -08001110 /*
1111 * zone->managed_pages is set to an approximate value in
1112 * free_area_init_core(), which will cause
1113 * /sys/device/system/node/nodeX/meminfo has wrong data.
1114 * So reset it to 0 before any memory is onlined.
1115 */
1116 reset_node_managed_pages(pgdat);
1117
Tang Chen0bd85422014-11-13 15:19:41 -08001118 /*
1119 * When memory is hot-added, all the memory is in offline state. So
1120 * clear all zones' present_pages because they will be updated in
1121 * online_pages() and offline_pages().
1122 */
1123 reset_node_present_pages(pgdat);
1124
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001125 return pgdat;
1126}
1127
1128static void rollback_node_hotadd(int nid, pg_data_t *pgdat)
1129{
1130 arch_refresh_nodedata(nid, NULL);
1131 arch_free_nodedata(pgdat);
1132 return;
1133}
1134
KAMEZAWA Hiroyuki0a547032006-06-27 02:53:35 -07001135
Toshi Kani01b0f192013-11-12 15:07:25 -08001136/**
1137 * try_online_node - online a node if offlined
1138 *
minskey guocf234222010-05-24 14:32:41 -07001139 * called by cpu_up() to online a node without onlined memory.
1140 */
Toshi Kani01b0f192013-11-12 15:07:25 -08001141int try_online_node(int nid)
minskey guocf234222010-05-24 14:32:41 -07001142{
1143 pg_data_t *pgdat;
1144 int ret;
1145
Toshi Kani01b0f192013-11-12 15:07:25 -08001146 if (node_online(nid))
1147 return 0;
1148
Vladimir Davydovbfc8c902014-06-04 16:07:18 -07001149 mem_hotplug_begin();
minskey guocf234222010-05-24 14:32:41 -07001150 pgdat = hotadd_new_pgdat(nid, 0);
David Rientjes7553e8f2011-06-22 18:13:01 -07001151 if (!pgdat) {
Toshi Kani01b0f192013-11-12 15:07:25 -08001152 pr_err("Cannot online node %d due to NULL pgdat\n", nid);
minskey guocf234222010-05-24 14:32:41 -07001153 ret = -ENOMEM;
1154 goto out;
1155 }
1156 node_set_online(nid);
1157 ret = register_one_node(nid);
1158 BUG_ON(ret);
1159
Toshi Kani01b0f192013-11-12 15:07:25 -08001160 if (pgdat->node_zonelists->_zonerefs->zone == NULL) {
1161 mutex_lock(&zonelists_mutex);
1162 build_all_zonelists(NULL, NULL);
1163 mutex_unlock(&zonelists_mutex);
1164 }
1165
minskey guocf234222010-05-24 14:32:41 -07001166out:
Vladimir Davydovbfc8c902014-06-04 16:07:18 -07001167 mem_hotplug_done();
minskey guocf234222010-05-24 14:32:41 -07001168 return ret;
1169}
1170
Toshi Kani27356f52013-09-11 14:21:49 -07001171static int check_hotplug_memory_range(u64 start, u64 size)
1172{
Fabian Frederickc8e861a2014-06-04 16:07:51 -07001173 u64 start_pfn = PFN_DOWN(start);
Toshi Kani27356f52013-09-11 14:21:49 -07001174 u64 nr_pages = size >> PAGE_SHIFT;
1175
1176 /* Memory range must be aligned with section */
1177 if ((start_pfn & ~PAGE_SECTION_MASK) ||
1178 (nr_pages % PAGES_PER_SECTION) || (!nr_pages)) {
1179 pr_err("Section-unaligned hotplug range: start 0x%llx, size 0x%llx\n",
1180 (unsigned long long)start,
1181 (unsigned long long)size);
1182 return -EINVAL;
1183 }
1184
1185 return 0;
1186}
1187
Wang Nan63264402014-08-06 16:07:36 -07001188/*
1189 * If movable zone has already been setup, newly added memory should be check.
1190 * If its address is higher than movable zone, it should be added as movable.
1191 * Without this check, movable zone may overlap with other zone.
1192 */
1193static int should_add_memory_movable(int nid, u64 start, u64 size)
1194{
1195 unsigned long start_pfn = start >> PAGE_SHIFT;
1196 pg_data_t *pgdat = NODE_DATA(nid);
1197 struct zone *movable_zone = pgdat->node_zones + ZONE_MOVABLE;
1198
1199 if (zone_is_empty(movable_zone))
1200 return 0;
1201
1202 if (movable_zone->zone_start_pfn <= start_pfn)
1203 return 1;
1204
1205 return 0;
1206}
1207
1208int zone_for_memory(int nid, u64 start, u64 size, int zone_default)
1209{
1210 if (should_add_memory_movable(nid, start, size))
1211 return ZONE_MOVABLE;
1212
1213 return zone_default;
1214}
1215
Al Viro31168482008-11-22 17:33:24 +00001216/* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
1217int __ref add_memory(int nid, u64 start, u64 size)
Yasunori Gotobc02af92006-06-27 02:53:30 -07001218{
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001219 pg_data_t *pgdat = NULL;
Tang Chena1e565a2013-02-22 16:33:18 -08001220 bool new_pgdat;
1221 bool new_node;
KAMEZAWA Hiroyukiebd15302006-08-05 12:15:06 -07001222 struct resource *res;
Yasunori Gotobc02af92006-06-27 02:53:30 -07001223 int ret;
1224
Toshi Kani27356f52013-09-11 14:21:49 -07001225 ret = check_hotplug_memory_range(start, size);
1226 if (ret)
1227 return ret;
1228
KAMEZAWA Hiroyukiebd15302006-08-05 12:15:06 -07001229 res = register_memory_resource(start, size);
Andi Kleen6ad696d2009-11-17 14:06:22 -08001230 ret = -EEXIST;
KAMEZAWA Hiroyukiebd15302006-08-05 12:15:06 -07001231 if (!res)
Nathan Zimmerac13c462014-01-23 15:53:26 -08001232 return ret;
KAMEZAWA Hiroyukiebd15302006-08-05 12:15:06 -07001233
Tang Chena1e565a2013-02-22 16:33:18 -08001234 { /* Stupid hack to suppress address-never-null warning */
1235 void *p = NODE_DATA(nid);
1236 new_pgdat = !p;
1237 }
Nathan Zimmerac13c462014-01-23 15:53:26 -08001238
Vladimir Davydovbfc8c902014-06-04 16:07:18 -07001239 mem_hotplug_begin();
Nathan Zimmerac13c462014-01-23 15:53:26 -08001240
Tang Chena1e565a2013-02-22 16:33:18 -08001241 new_node = !node_online(nid);
1242 if (new_node) {
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001243 pgdat = hotadd_new_pgdat(nid, start);
Andi Kleen6ad696d2009-11-17 14:06:22 -08001244 ret = -ENOMEM;
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001245 if (!pgdat)
Wen Congyang41b9e2d2012-07-11 14:02:31 -07001246 goto error;
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001247 }
1248
Yasunori Gotobc02af92006-06-27 02:53:30 -07001249 /* call arch's memory hotadd */
1250 ret = arch_add_memory(nid, start, size);
1251
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001252 if (ret < 0)
1253 goto error;
1254
Yasunori Goto0fc44152006-06-27 02:53:38 -07001255 /* we online node here. we can't roll back from here. */
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001256 node_set_online(nid);
1257
Tang Chena1e565a2013-02-22 16:33:18 -08001258 if (new_node) {
Yasunori Goto0fc44152006-06-27 02:53:38 -07001259 ret = register_one_node(nid);
1260 /*
1261 * If sysfs file of new node can't create, cpu on the node
1262 * can't be hot-added. There is no rollback way now.
1263 * So, check by BUG_ON() to catch it reluctantly..
1264 */
1265 BUG_ON(ret);
1266 }
1267
akpm@linux-foundation.orgd96ae532010-03-05 13:41:58 -08001268 /* create new memmap entry */
1269 firmware_map_add_hotplug(start, start + size, "System RAM");
1270
Andi Kleen6ad696d2009-11-17 14:06:22 -08001271 goto out;
1272
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001273error:
1274 /* rollback pgdat allocation and others */
1275 if (new_pgdat)
1276 rollback_node_hotadd(nid, pgdat);
Sasha Levina864b9d2013-02-22 16:32:48 -08001277 release_memory_resource(res);
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001278
Andi Kleen6ad696d2009-11-17 14:06:22 -08001279out:
Vladimir Davydovbfc8c902014-06-04 16:07:18 -07001280 mem_hotplug_done();
Yasunori Gotobc02af92006-06-27 02:53:30 -07001281 return ret;
1282}
1283EXPORT_SYMBOL_GPL(add_memory);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001284
1285#ifdef CONFIG_MEMORY_HOTREMOVE
1286/*
Badari Pulavarty5c755e92008-07-23 21:28:19 -07001287 * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy
1288 * set and the size of the free page is given by page_order(). Using this,
1289 * the function determines if the pageblock contains only free pages.
1290 * Due to buddy contraints, a free page at least the size of a pageblock will
1291 * be located at the start of the pageblock
1292 */
1293static inline int pageblock_free(struct page *page)
1294{
1295 return PageBuddy(page) && page_order(page) >= pageblock_order;
1296}
1297
1298/* Return the start of the next active pageblock after a given page */
1299static struct page *next_active_pageblock(struct page *page)
1300{
Badari Pulavarty5c755e92008-07-23 21:28:19 -07001301 /* Ensure the starting page is pageblock-aligned */
1302 BUG_ON(page_to_pfn(page) & (pageblock_nr_pages - 1));
1303
Badari Pulavarty5c755e92008-07-23 21:28:19 -07001304 /* If the entire pageblock is free, move to the end of free page */
KAMEZAWA Hiroyuki0dcc48c2010-09-09 16:38:01 -07001305 if (pageblock_free(page)) {
1306 int order;
1307 /* be careful. we don't have locks, page_order can be changed.*/
1308 order = page_order(page);
1309 if ((order < MAX_ORDER) && (order >= pageblock_order))
1310 return page + (1 << order);
1311 }
Badari Pulavarty5c755e92008-07-23 21:28:19 -07001312
KAMEZAWA Hiroyuki0dcc48c2010-09-09 16:38:01 -07001313 return page + pageblock_nr_pages;
Badari Pulavarty5c755e92008-07-23 21:28:19 -07001314}
1315
1316/* Checks if this range of memory is likely to be hot-removable. */
1317int is_mem_section_removable(unsigned long start_pfn, unsigned long nr_pages)
1318{
Badari Pulavarty5c755e92008-07-23 21:28:19 -07001319 struct page *page = pfn_to_page(start_pfn);
1320 struct page *end_page = page + nr_pages;
1321
1322 /* Check the starting page of each pageblock within the range */
1323 for (; page < end_page; page = next_active_pageblock(page)) {
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07001324 if (!is_pageblock_removable_nolock(page))
Badari Pulavarty5c755e92008-07-23 21:28:19 -07001325 return 0;
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07001326 cond_resched();
Badari Pulavarty5c755e92008-07-23 21:28:19 -07001327 }
1328
1329 /* All pageblocks in the memory block are likely to be hot-removable */
1330 return 1;
1331}
1332
1333/*
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001334 * Confirm all pages in a range [start, end) is belongs to the same zone.
1335 */
Zhang Zhened2f2402014-10-09 15:26:31 -07001336int test_pages_in_a_zone(unsigned long start_pfn, unsigned long end_pfn)
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001337{
1338 unsigned long pfn;
1339 struct zone *zone = NULL;
1340 struct page *page;
1341 int i;
1342 for (pfn = start_pfn;
1343 pfn < end_pfn;
1344 pfn += MAX_ORDER_NR_PAGES) {
1345 i = 0;
1346 /* This is just a CONFIG_HOLES_IN_ZONE check.*/
1347 while ((i < MAX_ORDER_NR_PAGES) && !pfn_valid_within(pfn + i))
1348 i++;
1349 if (i == MAX_ORDER_NR_PAGES)
1350 continue;
1351 page = pfn_to_page(pfn + i);
1352 if (zone && page_zone(page) != zone)
1353 return 0;
1354 zone = page_zone(page);
1355 }
1356 return 1;
1357}
1358
1359/*
Naoya Horiguchic8721bb2013-09-11 14:22:09 -07001360 * Scan pfn range [start,end) to find movable/migratable pages (LRU pages
1361 * and hugepages). We scan pfn because it's much easier than scanning over
1362 * linked list. This function returns the pfn of the first found movable
1363 * page if it's found, otherwise 0.
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001364 */
Naoya Horiguchic8721bb2013-09-11 14:22:09 -07001365static unsigned long scan_movable_pages(unsigned long start, unsigned long end)
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001366{
1367 unsigned long pfn;
1368 struct page *page;
1369 for (pfn = start; pfn < end; pfn++) {
1370 if (pfn_valid(pfn)) {
1371 page = pfn_to_page(pfn);
1372 if (PageLRU(page))
1373 return pfn;
Naoya Horiguchic8721bb2013-09-11 14:22:09 -07001374 if (PageHuge(page)) {
1375 if (is_hugepage_active(page))
1376 return pfn;
1377 else
1378 pfn = round_up(pfn + 1,
1379 1 << compound_order(page)) - 1;
1380 }
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001381 }
1382 }
1383 return 0;
1384}
1385
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001386#define NR_OFFLINE_AT_ONCE_PAGES (256)
1387static int
1388do_migrate_range(unsigned long start_pfn, unsigned long end_pfn)
1389{
1390 unsigned long pfn;
1391 struct page *page;
1392 int move_pages = NR_OFFLINE_AT_ONCE_PAGES;
1393 int not_managed = 0;
1394 int ret = 0;
1395 LIST_HEAD(source);
1396
1397 for (pfn = start_pfn; pfn < end_pfn && move_pages > 0; pfn++) {
1398 if (!pfn_valid(pfn))
1399 continue;
1400 page = pfn_to_page(pfn);
Naoya Horiguchic8721bb2013-09-11 14:22:09 -07001401
1402 if (PageHuge(page)) {
1403 struct page *head = compound_head(page);
1404 pfn = page_to_pfn(head) + (1<<compound_order(head)) - 1;
1405 if (compound_order(head) > PFN_SECTION_SHIFT) {
1406 ret = -EBUSY;
1407 break;
1408 }
1409 if (isolate_huge_page(page, &source))
1410 move_pages -= 1 << compound_order(head);
1411 continue;
1412 }
1413
Konstantin Khlebnikov700c2a42011-05-24 17:12:19 -07001414 if (!get_page_unless_zero(page))
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001415 continue;
1416 /*
1417 * We can skip free pages. And we can only deal with pages on
1418 * LRU.
1419 */
Nick Piggin62695a82008-10-18 20:26:09 -07001420 ret = isolate_lru_page(page);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001421 if (!ret) { /* Success */
Konstantin Khlebnikov700c2a42011-05-24 17:12:19 -07001422 put_page(page);
Nick Piggin62695a82008-10-18 20:26:09 -07001423 list_add_tail(&page->lru, &source);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001424 move_pages--;
KOSAKI Motohiro6d9c2852009-12-14 17:58:11 -08001425 inc_zone_page_state(page, NR_ISOLATED_ANON +
1426 page_is_file_cache(page));
1427
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001428 } else {
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001429#ifdef CONFIG_DEBUG_VM
Wu Fengguang718a3822010-03-10 15:20:43 -08001430 printk(KERN_ALERT "removing pfn %lx from LRU failed\n",
1431 pfn);
Dave Hansenf0b791a2014-01-23 15:52:49 -08001432 dump_page(page, "failed to remove from LRU");
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001433#endif
Konstantin Khlebnikov700c2a42011-05-24 17:12:19 -07001434 put_page(page);
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001435 /* Because we don't have big zone->lock. we should
Bob Liu809c4442010-10-26 14:22:10 -07001436 check this again here. */
1437 if (page_count(page)) {
1438 not_managed++;
Bob Liuf3ab2632010-10-26 14:22:10 -07001439 ret = -EBUSY;
Bob Liu809c4442010-10-26 14:22:10 -07001440 break;
1441 }
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001442 }
1443 }
Bob Liuf3ab2632010-10-26 14:22:10 -07001444 if (!list_empty(&source)) {
1445 if (not_managed) {
Naoya Horiguchic8721bb2013-09-11 14:22:09 -07001446 putback_movable_pages(&source);
Bob Liuf3ab2632010-10-26 14:22:10 -07001447 goto out;
1448 }
Minchan Kim74c08f92012-10-08 16:32:54 -07001449
1450 /*
1451 * alloc_migrate_target should be improooooved!!
1452 * migrate_pages returns # of failed pages.
1453 */
David Rientjes68711a72014-06-04 16:08:25 -07001454 ret = migrate_pages(&source, alloc_migrate_target, NULL, 0,
Hugh Dickins9c620e22013-02-22 16:35:14 -08001455 MIGRATE_SYNC, MR_MEMORY_HOTPLUG);
Bob Liuf3ab2632010-10-26 14:22:10 -07001456 if (ret)
Naoya Horiguchic8721bb2013-09-11 14:22:09 -07001457 putback_movable_pages(&source);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001458 }
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001459out:
1460 return ret;
1461}
1462
1463/*
1464 * remove from free_area[] and mark all as Reserved.
1465 */
1466static int
1467offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages,
1468 void *data)
1469{
1470 __offline_isolated_pages(start, start + nr_pages);
1471 return 0;
1472}
1473
1474static void
1475offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
1476{
KAMEZAWA Hiroyuki908eedc2009-09-22 16:45:46 -07001477 walk_system_ram_range(start_pfn, end_pfn - start_pfn, NULL,
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001478 offline_isolated_pages_cb);
1479}
1480
1481/*
1482 * Check all pages in range, recoreded as memory resource, are isolated.
1483 */
1484static int
1485check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages,
1486 void *data)
1487{
1488 int ret;
1489 long offlined = *(long *)data;
Wen Congyangb023f462012-12-11 16:00:45 -08001490 ret = test_pages_isolated(start_pfn, start_pfn + nr_pages, true);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001491 offlined = nr_pages;
1492 if (!ret)
1493 *(long *)data += offlined;
1494 return ret;
1495}
1496
1497static long
1498check_pages_isolated(unsigned long start_pfn, unsigned long end_pfn)
1499{
1500 long offlined = 0;
1501 int ret;
1502
KAMEZAWA Hiroyuki908eedc2009-09-22 16:45:46 -07001503 ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn, &offlined,
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001504 check_pages_isolated_cb);
1505 if (ret < 0)
1506 offlined = (long)ret;
1507 return offlined;
1508}
1509
Lai Jiangshan09285af2012-12-12 13:52:04 -08001510#ifdef CONFIG_MOVABLE_NODE
Tang Chen79a4dce2012-12-18 14:23:24 -08001511/*
1512 * When CONFIG_MOVABLE_NODE, we permit offlining of a node which doesn't have
1513 * normal memory.
1514 */
Lai Jiangshan09285af2012-12-12 13:52:04 -08001515static bool can_offline_normal(struct zone *zone, unsigned long nr_pages)
1516{
1517 return true;
1518}
Tang Chen79a4dce2012-12-18 14:23:24 -08001519#else /* CONFIG_MOVABLE_NODE */
Lai Jiangshan74d42d82012-12-11 16:03:23 -08001520/* ensure the node has NORMAL memory if it is still online */
1521static bool can_offline_normal(struct zone *zone, unsigned long nr_pages)
1522{
1523 struct pglist_data *pgdat = zone->zone_pgdat;
1524 unsigned long present_pages = 0;
1525 enum zone_type zt;
1526
1527 for (zt = 0; zt <= ZONE_NORMAL; zt++)
1528 present_pages += pgdat->node_zones[zt].present_pages;
1529
1530 if (present_pages > nr_pages)
1531 return true;
1532
1533 present_pages = 0;
1534 for (; zt <= ZONE_MOVABLE; zt++)
1535 present_pages += pgdat->node_zones[zt].present_pages;
1536
1537 /*
1538 * we can't offline the last normal memory until all
1539 * higher memory is offlined.
1540 */
1541 return present_pages == 0;
1542}
Tang Chen79a4dce2012-12-18 14:23:24 -08001543#endif /* CONFIG_MOVABLE_NODE */
Lai Jiangshan74d42d82012-12-11 16:03:23 -08001544
Tang Chenc5320922013-11-12 15:08:10 -08001545static int __init cmdline_parse_movable_node(char *p)
1546{
1547#ifdef CONFIG_MOVABLE_NODE
1548 /*
1549 * Memory used by the kernel cannot be hot-removed because Linux
1550 * cannot migrate the kernel pages. When memory hotplug is
1551 * enabled, we should prevent memblock from allocating memory
1552 * for the kernel.
1553 *
1554 * ACPI SRAT records all hotpluggable memory ranges. But before
1555 * SRAT is parsed, we don't know about it.
1556 *
1557 * The kernel image is loaded into memory at very early time. We
1558 * cannot prevent this anyway. So on NUMA system, we set any
1559 * node the kernel resides in as un-hotpluggable.
1560 *
1561 * Since on modern servers, one node could have double-digit
1562 * gigabytes memory, we can assume the memory around the kernel
1563 * image is also un-hotpluggable. So before SRAT is parsed, just
1564 * allocate memory near the kernel image to try the best to keep
1565 * the kernel away from hotpluggable memory.
1566 */
1567 memblock_set_bottom_up(true);
Tang Chen55ac5902014-01-21 15:49:35 -08001568 movable_node_enabled = true;
Tang Chenc5320922013-11-12 15:08:10 -08001569#else
1570 pr_warn("movable_node option not supported\n");
1571#endif
1572 return 0;
1573}
1574early_param("movable_node", cmdline_parse_movable_node);
1575
Lai Jiangshand9713672012-12-11 16:01:03 -08001576/* check which state of node_states will be changed when offline memory */
1577static void node_states_check_changes_offline(unsigned long nr_pages,
1578 struct zone *zone, struct memory_notify *arg)
1579{
1580 struct pglist_data *pgdat = zone->zone_pgdat;
1581 unsigned long present_pages = 0;
1582 enum zone_type zt, zone_last = ZONE_NORMAL;
1583
1584 /*
Lai Jiangshan6715ddf2012-12-12 13:51:49 -08001585 * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY]
1586 * contains nodes which have zones of 0...ZONE_NORMAL,
1587 * set zone_last to ZONE_NORMAL.
Lai Jiangshand9713672012-12-11 16:01:03 -08001588 *
Lai Jiangshan6715ddf2012-12-12 13:51:49 -08001589 * If we don't have HIGHMEM nor movable node,
1590 * node_states[N_NORMAL_MEMORY] contains nodes which have zones of
1591 * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE.
Lai Jiangshand9713672012-12-11 16:01:03 -08001592 */
Lai Jiangshan6715ddf2012-12-12 13:51:49 -08001593 if (N_MEMORY == N_NORMAL_MEMORY)
Lai Jiangshand9713672012-12-11 16:01:03 -08001594 zone_last = ZONE_MOVABLE;
1595
1596 /*
1597 * check whether node_states[N_NORMAL_MEMORY] will be changed.
1598 * If the memory to be offline is in a zone of 0...zone_last,
1599 * and it is the last present memory, 0...zone_last will
1600 * become empty after offline , thus we can determind we will
1601 * need to clear the node from node_states[N_NORMAL_MEMORY].
1602 */
1603 for (zt = 0; zt <= zone_last; zt++)
1604 present_pages += pgdat->node_zones[zt].present_pages;
1605 if (zone_idx(zone) <= zone_last && nr_pages >= present_pages)
1606 arg->status_change_nid_normal = zone_to_nid(zone);
1607 else
1608 arg->status_change_nid_normal = -1;
1609
Lai Jiangshan6715ddf2012-12-12 13:51:49 -08001610#ifdef CONFIG_HIGHMEM
1611 /*
1612 * If we have movable node, node_states[N_HIGH_MEMORY]
1613 * contains nodes which have zones of 0...ZONE_HIGHMEM,
1614 * set zone_last to ZONE_HIGHMEM.
1615 *
1616 * If we don't have movable node, node_states[N_NORMAL_MEMORY]
1617 * contains nodes which have zones of 0...ZONE_MOVABLE,
1618 * set zone_last to ZONE_MOVABLE.
1619 */
1620 zone_last = ZONE_HIGHMEM;
1621 if (N_MEMORY == N_HIGH_MEMORY)
1622 zone_last = ZONE_MOVABLE;
1623
1624 for (; zt <= zone_last; zt++)
1625 present_pages += pgdat->node_zones[zt].present_pages;
1626 if (zone_idx(zone) <= zone_last && nr_pages >= present_pages)
1627 arg->status_change_nid_high = zone_to_nid(zone);
1628 else
1629 arg->status_change_nid_high = -1;
1630#else
1631 arg->status_change_nid_high = arg->status_change_nid_normal;
1632#endif
1633
Lai Jiangshand9713672012-12-11 16:01:03 -08001634 /*
1635 * node_states[N_HIGH_MEMORY] contains nodes which have 0...ZONE_MOVABLE
1636 */
1637 zone_last = ZONE_MOVABLE;
1638
1639 /*
1640 * check whether node_states[N_HIGH_MEMORY] will be changed
1641 * If we try to offline the last present @nr_pages from the node,
1642 * we can determind we will need to clear the node from
1643 * node_states[N_HIGH_MEMORY].
1644 */
1645 for (; zt <= zone_last; zt++)
1646 present_pages += pgdat->node_zones[zt].present_pages;
1647 if (nr_pages >= present_pages)
1648 arg->status_change_nid = zone_to_nid(zone);
1649 else
1650 arg->status_change_nid = -1;
1651}
1652
1653static void node_states_clear_node(int node, struct memory_notify *arg)
1654{
1655 if (arg->status_change_nid_normal >= 0)
1656 node_clear_state(node, N_NORMAL_MEMORY);
1657
Lai Jiangshan6715ddf2012-12-12 13:51:49 -08001658 if ((N_MEMORY != N_NORMAL_MEMORY) &&
1659 (arg->status_change_nid_high >= 0))
Lai Jiangshand9713672012-12-11 16:01:03 -08001660 node_clear_state(node, N_HIGH_MEMORY);
Lai Jiangshan6715ddf2012-12-12 13:51:49 -08001661
1662 if ((N_MEMORY != N_HIGH_MEMORY) &&
1663 (arg->status_change_nid >= 0))
1664 node_clear_state(node, N_MEMORY);
Lai Jiangshand9713672012-12-11 16:01:03 -08001665}
1666
Wen Congyanga16cee12012-10-08 16:33:58 -07001667static int __ref __offline_pages(unsigned long start_pfn,
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001668 unsigned long end_pfn, unsigned long timeout)
1669{
1670 unsigned long pfn, nr_pages, expire;
1671 long offlined_pages;
Yasunori Goto7b78d332007-10-21 16:41:36 -07001672 int ret, drain, retry_max, node;
Cody P Schaferd7029092013-07-03 15:02:11 -07001673 unsigned long flags;
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001674 struct zone *zone;
Yasunori Goto7b78d332007-10-21 16:41:36 -07001675 struct memory_notify arg;
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001676
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001677 /* at least, alignment against pageblock is necessary */
1678 if (!IS_ALIGNED(start_pfn, pageblock_nr_pages))
1679 return -EINVAL;
1680 if (!IS_ALIGNED(end_pfn, pageblock_nr_pages))
1681 return -EINVAL;
1682 /* This makes hotplug much easier...and readable.
1683 we assume this for now. .*/
1684 if (!test_pages_in_a_zone(start_pfn, end_pfn))
1685 return -EINVAL;
Yasunori Goto7b78d332007-10-21 16:41:36 -07001686
Vladimir Davydovbfc8c902014-06-04 16:07:18 -07001687 mem_hotplug_begin();
Andi Kleen6ad696d2009-11-17 14:06:22 -08001688
Yasunori Goto7b78d332007-10-21 16:41:36 -07001689 zone = page_zone(pfn_to_page(start_pfn));
1690 node = zone_to_nid(zone);
1691 nr_pages = end_pfn - start_pfn;
1692
Lai Jiangshan74d42d82012-12-11 16:03:23 -08001693 ret = -EINVAL;
1694 if (zone_idx(zone) <= ZONE_NORMAL && !can_offline_normal(zone, nr_pages))
1695 goto out;
1696
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001697 /* set above range as isolated */
Wen Congyangb023f462012-12-11 16:00:45 -08001698 ret = start_isolate_page_range(start_pfn, end_pfn,
1699 MIGRATE_MOVABLE, true);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001700 if (ret)
Andi Kleen6ad696d2009-11-17 14:06:22 -08001701 goto out;
Yasunori Goto7b78d332007-10-21 16:41:36 -07001702
1703 arg.start_pfn = start_pfn;
1704 arg.nr_pages = nr_pages;
Lai Jiangshand9713672012-12-11 16:01:03 -08001705 node_states_check_changes_offline(nr_pages, zone, &arg);
Yasunori Goto7b78d332007-10-21 16:41:36 -07001706
1707 ret = memory_notify(MEM_GOING_OFFLINE, &arg);
1708 ret = notifier_to_errno(ret);
1709 if (ret)
1710 goto failed_removal;
1711
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001712 pfn = start_pfn;
1713 expire = jiffies + timeout;
1714 drain = 0;
1715 retry_max = 5;
1716repeat:
1717 /* start memory hot removal */
1718 ret = -EAGAIN;
1719 if (time_after(jiffies, expire))
1720 goto failed_removal;
1721 ret = -EINTR;
1722 if (signal_pending(current))
1723 goto failed_removal;
1724 ret = 0;
1725 if (drain) {
1726 lru_add_drain_all();
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001727 cond_resched();
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001728 drain_all_pages();
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001729 }
1730
Naoya Horiguchic8721bb2013-09-11 14:22:09 -07001731 pfn = scan_movable_pages(start_pfn, end_pfn);
1732 if (pfn) { /* We have movable pages */
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001733 ret = do_migrate_range(pfn, end_pfn);
1734 if (!ret) {
1735 drain = 1;
1736 goto repeat;
1737 } else {
1738 if (ret < 0)
1739 if (--retry_max == 0)
1740 goto failed_removal;
1741 yield();
1742 drain = 1;
1743 goto repeat;
1744 }
1745 }
Adam Buchbinderb3834be2012-09-19 21:48:02 -04001746 /* drain all zone's lru pagevec, this is asynchronous... */
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001747 lru_add_drain_all();
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001748 yield();
Adam Buchbinderb3834be2012-09-19 21:48:02 -04001749 /* drain pcp pages, this is synchronous. */
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001750 drain_all_pages();
Naoya Horiguchic8721bb2013-09-11 14:22:09 -07001751 /*
1752 * dissolve free hugepages in the memory block before doing offlining
1753 * actually in order to make hugetlbfs's object counting consistent.
1754 */
1755 dissolve_free_huge_pages(start_pfn, end_pfn);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001756 /* check again */
1757 offlined_pages = check_pages_isolated(start_pfn, end_pfn);
1758 if (offlined_pages < 0) {
1759 ret = -EBUSY;
1760 goto failed_removal;
1761 }
1762 printk(KERN_INFO "Offlined Pages %ld\n", offlined_pages);
Adam Buchbinderb3834be2012-09-19 21:48:02 -04001763 /* Ok, all of our target is isolated.
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001764 We cannot do rollback at this point. */
1765 offline_isolated_pages(start_pfn, end_pfn);
KAMEZAWA Hiroyukidbc0e4c2007-11-14 16:59:12 -08001766 /* reset pagetype flags and makes migrate type to be MOVABLE */
Michal Nazarewicz0815f3d2012-04-03 15:06:15 +02001767 undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001768 /* removal success */
Jiang Liu3dcc0572013-07-03 15:03:21 -07001769 adjust_managed_page_count(pfn_to_page(start_pfn), -offlined_pages);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001770 zone->present_pages -= offlined_pages;
Cody P Schaferd7029092013-07-03 15:02:11 -07001771
1772 pgdat_resize_lock(zone->zone_pgdat, &flags);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001773 zone->zone_pgdat->node_present_pages -= offlined_pages;
Cody P Schaferd7029092013-07-03 15:02:11 -07001774 pgdat_resize_unlock(zone->zone_pgdat, &flags);
Yasunori Goto7b78d332007-10-21 16:41:36 -07001775
KOSAKI Motohiro1b79acc2011-05-24 17:11:32 -07001776 init_per_zone_wmark_min();
1777
Xishi Qiu1e8537b2012-10-08 16:31:51 -07001778 if (!populated_zone(zone)) {
Jiang Liu340175b2012-07-31 16:43:32 -07001779 zone_pcp_reset(zone);
Xishi Qiu1e8537b2012-10-08 16:31:51 -07001780 mutex_lock(&zonelists_mutex);
1781 build_all_zonelists(NULL, NULL);
1782 mutex_unlock(&zonelists_mutex);
1783 } else
1784 zone_pcp_update(zone);
Jiang Liu340175b2012-07-31 16:43:32 -07001785
Lai Jiangshand9713672012-12-11 16:01:03 -08001786 node_states_clear_node(node, &arg);
1787 if (arg.status_change_nid >= 0)
David Rientjes8fe23e02009-12-14 17:58:33 -08001788 kswapd_stop(node);
Minchan Kimbce73942009-06-16 15:32:50 -07001789
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001790 vm_total_pages = nr_free_pagecache_pages();
1791 writeback_set_ratelimit();
Yasunori Goto7b78d332007-10-21 16:41:36 -07001792
1793 memory_notify(MEM_OFFLINE, &arg);
Vladimir Davydovbfc8c902014-06-04 16:07:18 -07001794 mem_hotplug_done();
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001795 return 0;
1796
1797failed_removal:
Bjorn Helgaasa62e2f42012-05-29 15:06:30 -07001798 printk(KERN_INFO "memory offlining [mem %#010llx-%#010llx] failed\n",
1799 (unsigned long long) start_pfn << PAGE_SHIFT,
1800 ((unsigned long long) end_pfn << PAGE_SHIFT) - 1);
Yasunori Goto7b78d332007-10-21 16:41:36 -07001801 memory_notify(MEM_CANCEL_OFFLINE, &arg);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001802 /* pushback to free area */
Michal Nazarewicz0815f3d2012-04-03 15:06:15 +02001803 undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
Yasunori Goto7b78d332007-10-21 16:41:36 -07001804
Andi Kleen6ad696d2009-11-17 14:06:22 -08001805out:
Vladimir Davydovbfc8c902014-06-04 16:07:18 -07001806 mem_hotplug_done();
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001807 return ret;
1808}
Badari Pulavarty71088782008-10-18 20:25:58 -07001809
Wen Congyanga16cee12012-10-08 16:33:58 -07001810int offline_pages(unsigned long start_pfn, unsigned long nr_pages)
1811{
1812 return __offline_pages(start_pfn, start_pfn + nr_pages, 120 * HZ);
1813}
Rafael J. Wysockie2ff3942013-05-08 00:29:49 +02001814#endif /* CONFIG_MEMORY_HOTREMOVE */
Wen Congyanga16cee12012-10-08 16:33:58 -07001815
Wen Congyangbbc76be2013-02-22 16:32:54 -08001816/**
1817 * walk_memory_range - walks through all mem sections in [start_pfn, end_pfn)
1818 * @start_pfn: start pfn of the memory range
Toshi Kanie05c4bb2013-04-29 15:06:16 -07001819 * @end_pfn: end pfn of the memory range
Wen Congyangbbc76be2013-02-22 16:32:54 -08001820 * @arg: argument passed to func
1821 * @func: callback for each memory section walked
1822 *
1823 * This function walks through all present mem sections in range
1824 * [start_pfn, end_pfn) and call func on each mem section.
1825 *
1826 * Returns the return value of func.
1827 */
Rafael J. Wysockie2ff3942013-05-08 00:29:49 +02001828int walk_memory_range(unsigned long start_pfn, unsigned long end_pfn,
Wen Congyangbbc76be2013-02-22 16:32:54 -08001829 void *arg, int (*func)(struct memory_block *, void *))
Badari Pulavarty71088782008-10-18 20:25:58 -07001830{
Wen Congyange90bdb72012-10-08 16:34:01 -07001831 struct memory_block *mem = NULL;
1832 struct mem_section *section;
Wen Congyange90bdb72012-10-08 16:34:01 -07001833 unsigned long pfn, section_nr;
1834 int ret;
Badari Pulavarty71088782008-10-18 20:25:58 -07001835
Wen Congyange90bdb72012-10-08 16:34:01 -07001836 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
1837 section_nr = pfn_to_section_nr(pfn);
1838 if (!present_section_nr(section_nr))
1839 continue;
1840
1841 section = __nr_to_section(section_nr);
1842 /* same memblock? */
1843 if (mem)
1844 if ((section_nr >= mem->start_section_nr) &&
1845 (section_nr <= mem->end_section_nr))
1846 continue;
1847
1848 mem = find_memory_block_hinted(section, mem);
1849 if (!mem)
1850 continue;
1851
Wen Congyangbbc76be2013-02-22 16:32:54 -08001852 ret = func(mem, arg);
Wen Congyange90bdb72012-10-08 16:34:01 -07001853 if (ret) {
Wen Congyangbbc76be2013-02-22 16:32:54 -08001854 kobject_put(&mem->dev.kobj);
1855 return ret;
Wen Congyange90bdb72012-10-08 16:34:01 -07001856 }
1857 }
1858
1859 if (mem)
1860 kobject_put(&mem->dev.kobj);
1861
Wen Congyangbbc76be2013-02-22 16:32:54 -08001862 return 0;
1863}
1864
Rafael J. Wysockie2ff3942013-05-08 00:29:49 +02001865#ifdef CONFIG_MEMORY_HOTREMOVE
Xishi Qiud6de9d52013-11-12 15:07:20 -08001866static int check_memblock_offlined_cb(struct memory_block *mem, void *arg)
Wen Congyangbbc76be2013-02-22 16:32:54 -08001867{
1868 int ret = !is_memblock_offlined(mem);
1869
Randy Dunlap349daa02013-04-29 15:08:49 -07001870 if (unlikely(ret)) {
1871 phys_addr_t beginpa, endpa;
1872
1873 beginpa = PFN_PHYS(section_nr_to_pfn(mem->start_section_nr));
1874 endpa = PFN_PHYS(section_nr_to_pfn(mem->end_section_nr + 1))-1;
Wen Congyangbbc76be2013-02-22 16:32:54 -08001875 pr_warn("removing memory fails, because memory "
Randy Dunlap349daa02013-04-29 15:08:49 -07001876 "[%pa-%pa] is onlined\n",
1877 &beginpa, &endpa);
1878 }
Wen Congyangbbc76be2013-02-22 16:32:54 -08001879
1880 return ret;
1881}
1882
Toshi Kani0f1cfe92013-09-11 14:21:50 -07001883static int check_cpu_on_node(pg_data_t *pgdat)
Tang Chen60a5a192013-02-22 16:33:14 -08001884{
Tang Chen60a5a192013-02-22 16:33:14 -08001885 int cpu;
1886
1887 for_each_present_cpu(cpu) {
1888 if (cpu_to_node(cpu) == pgdat->node_id)
1889 /*
1890 * the cpu on this node isn't removed, and we can't
1891 * offline this node.
1892 */
1893 return -EBUSY;
1894 }
1895
1896 return 0;
1897}
1898
Toshi Kani0f1cfe92013-09-11 14:21:50 -07001899static void unmap_cpu_on_node(pg_data_t *pgdat)
Wen Congyange13fe862013-02-22 16:33:31 -08001900{
1901#ifdef CONFIG_ACPI_NUMA
Wen Congyange13fe862013-02-22 16:33:31 -08001902 int cpu;
1903
1904 for_each_possible_cpu(cpu)
1905 if (cpu_to_node(cpu) == pgdat->node_id)
1906 numa_clear_node(cpu);
1907#endif
1908}
1909
Toshi Kani0f1cfe92013-09-11 14:21:50 -07001910static int check_and_unmap_cpu_on_node(pg_data_t *pgdat)
Wen Congyange13fe862013-02-22 16:33:31 -08001911{
Toshi Kani0f1cfe92013-09-11 14:21:50 -07001912 int ret;
Wen Congyange13fe862013-02-22 16:33:31 -08001913
Toshi Kani0f1cfe92013-09-11 14:21:50 -07001914 ret = check_cpu_on_node(pgdat);
Wen Congyange13fe862013-02-22 16:33:31 -08001915 if (ret)
1916 return ret;
1917
1918 /*
1919 * the node will be offlined when we come here, so we can clear
1920 * the cpu_to_node() now.
1921 */
1922
Toshi Kani0f1cfe92013-09-11 14:21:50 -07001923 unmap_cpu_on_node(pgdat);
Wen Congyange13fe862013-02-22 16:33:31 -08001924 return 0;
1925}
1926
Toshi Kani0f1cfe92013-09-11 14:21:50 -07001927/**
1928 * try_offline_node
1929 *
1930 * Offline a node if all memory sections and cpus of the node are removed.
1931 *
1932 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1933 * and online/offline operations before this call.
1934 */
Wen Congyang90b30cd2013-02-22 16:33:27 -08001935void try_offline_node(int nid)
Tang Chen60a5a192013-02-22 16:33:14 -08001936{
Wen Congyangd822b862013-02-22 16:33:16 -08001937 pg_data_t *pgdat = NODE_DATA(nid);
1938 unsigned long start_pfn = pgdat->node_start_pfn;
1939 unsigned long end_pfn = start_pfn + pgdat->node_spanned_pages;
Tang Chen60a5a192013-02-22 16:33:14 -08001940 unsigned long pfn;
Wen Congyangd822b862013-02-22 16:33:16 -08001941 int i;
Tang Chen60a5a192013-02-22 16:33:14 -08001942
1943 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
1944 unsigned long section_nr = pfn_to_section_nr(pfn);
1945
1946 if (!present_section_nr(section_nr))
1947 continue;
1948
1949 if (pfn_to_nid(pfn) != nid)
1950 continue;
1951
1952 /*
1953 * some memory sections of this node are not removed, and we
1954 * can't offline node now.
1955 */
1956 return;
1957 }
1958
Toshi Kani0f1cfe92013-09-11 14:21:50 -07001959 if (check_and_unmap_cpu_on_node(pgdat))
Tang Chen60a5a192013-02-22 16:33:14 -08001960 return;
1961
1962 /*
1963 * all memory/cpu of this node are removed, we can offline this
1964 * node now.
1965 */
1966 node_set_offline(nid);
1967 unregister_one_node(nid);
Wen Congyangd822b862013-02-22 16:33:16 -08001968
Wen Congyangd822b862013-02-22 16:33:16 -08001969 /* free waittable in each zone */
1970 for (i = 0; i < MAX_NR_ZONES; i++) {
1971 struct zone *zone = pgdat->node_zones + i;
1972
Jianguo Wuca4b3f32013-03-22 15:04:50 -07001973 /*
1974 * wait_table may be allocated from boot memory,
1975 * here only free if it's allocated by vmalloc.
1976 */
1977 if (is_vmalloc_addr(zone->wait_table))
Wen Congyangd822b862013-02-22 16:33:16 -08001978 vfree(zone->wait_table);
1979 }
1980
1981 /*
1982 * Since there is no way to guarentee the address of pgdat/zone is not
1983 * on stack of any kernel threads or used by other kernel objects
1984 * without reference counting or other symchronizing method, do not
1985 * reset node_data and free pgdat here. Just reset it to 0 and reuse
1986 * the memory when the node is online again.
1987 */
1988 memset(pgdat, 0, sizeof(*pgdat));
Tang Chen60a5a192013-02-22 16:33:14 -08001989}
Wen Congyang90b30cd2013-02-22 16:33:27 -08001990EXPORT_SYMBOL(try_offline_node);
Tang Chen60a5a192013-02-22 16:33:14 -08001991
Toshi Kani0f1cfe92013-09-11 14:21:50 -07001992/**
1993 * remove_memory
1994 *
1995 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1996 * and online/offline operations before this call, as required by
1997 * try_offline_node().
1998 */
Rafael J. Wysocki242831e2013-05-27 12:58:46 +02001999void __ref remove_memory(int nid, u64 start, u64 size)
Wen Congyangbbc76be2013-02-22 16:32:54 -08002000{
Rafael J. Wysocki242831e2013-05-27 12:58:46 +02002001 int ret;
Wen Congyang993c1aa2013-02-22 16:32:50 -08002002
Toshi Kani27356f52013-09-11 14:21:49 -07002003 BUG_ON(check_hotplug_memory_range(start, size));
2004
Vladimir Davydovbfc8c902014-06-04 16:07:18 -07002005 mem_hotplug_begin();
Yasuaki Ishimatsu6677e3e2013-02-22 16:32:52 -08002006
2007 /*
Rafael J. Wysocki242831e2013-05-27 12:58:46 +02002008 * All memory blocks must be offlined before removing memory. Check
2009 * whether all memory blocks in question are offline and trigger a BUG()
2010 * if this is not the case.
Yasuaki Ishimatsu6677e3e2013-02-22 16:32:52 -08002011 */
Rafael J. Wysocki242831e2013-05-27 12:58:46 +02002012 ret = walk_memory_range(PFN_DOWN(start), PFN_UP(start + size - 1), NULL,
Xishi Qiud6de9d52013-11-12 15:07:20 -08002013 check_memblock_offlined_cb);
Vladimir Davydovbfc8c902014-06-04 16:07:18 -07002014 if (ret)
Rafael J. Wysocki242831e2013-05-27 12:58:46 +02002015 BUG();
Yasuaki Ishimatsu6677e3e2013-02-22 16:32:52 -08002016
Yasuaki Ishimatsu46c66c42013-02-22 16:32:56 -08002017 /* remove memmap entry */
2018 firmware_map_remove(start, start + size, "System RAM");
2019
Wen Congyang24d335c2013-02-22 16:32:58 -08002020 arch_remove_memory(start, size);
2021
Tang Chen60a5a192013-02-22 16:33:14 -08002022 try_offline_node(nid);
2023
Vladimir Davydovbfc8c902014-06-04 16:07:18 -07002024 mem_hotplug_done();
Badari Pulavarty71088782008-10-18 20:25:58 -07002025}
Badari Pulavarty71088782008-10-18 20:25:58 -07002026EXPORT_SYMBOL_GPL(remove_memory);
Rafael J. Wysockiaba6efc2013-06-01 22:24:07 +02002027#endif /* CONFIG_MEMORY_HOTREMOVE */