blob: 962927309b6e963fa05a0acf09608c2ec6cef761 [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 Kuznetsov8604d9e2016-05-19 17:13:03 -070081#ifndef CONFIG_MEMORY_HOTPLUG_DEFAULT_ONLINE
Vitaly Kuznetsov31bc3852016-03-15 14:56:48 -070082bool memhp_auto_online;
Vitaly Kuznetsov8604d9e2016-05-19 17:13:03 -070083#else
84bool memhp_auto_online = true;
85#endif
Vitaly Kuznetsov31bc3852016-03-15 14:56:48 -070086EXPORT_SYMBOL_GPL(memhp_auto_online);
87
Vitaly Kuznetsov86dd9952016-05-19 17:13:06 -070088static int __init setup_memhp_default_state(char *str)
89{
90 if (!strcmp(str, "online"))
91 memhp_auto_online = true;
92 else if (!strcmp(str, "offline"))
93 memhp_auto_online = false;
94
95 return 1;
96}
97__setup("memhp_default_state=", setup_memhp_default_state);
98
Vladimir Davydovbfc8c902014-06-04 16:07:18 -070099void get_online_mems(void)
KOSAKI Motohiro20d6c962010-12-02 14:31:19 -0800100{
Vladimir Davydovbfc8c902014-06-04 16:07:18 -0700101 might_sleep();
102 if (mem_hotplug.active_writer == current)
103 return;
104 memhp_lock_acquire_read();
105 mutex_lock(&mem_hotplug.lock);
106 mem_hotplug.refcount++;
107 mutex_unlock(&mem_hotplug.lock);
108
KOSAKI Motohiro20d6c962010-12-02 14:31:19 -0800109}
110
Vladimir Davydovbfc8c902014-06-04 16:07:18 -0700111void put_online_mems(void)
KOSAKI Motohiro20d6c962010-12-02 14:31:19 -0800112{
Vladimir Davydovbfc8c902014-06-04 16:07:18 -0700113 if (mem_hotplug.active_writer == current)
114 return;
115 mutex_lock(&mem_hotplug.lock);
116
117 if (WARN_ON(!mem_hotplug.refcount))
118 mem_hotplug.refcount++; /* try to fix things up */
119
120 if (!--mem_hotplug.refcount && unlikely(mem_hotplug.active_writer))
121 wake_up_process(mem_hotplug.active_writer);
122 mutex_unlock(&mem_hotplug.lock);
123 memhp_lock_release();
124
KOSAKI Motohiro20d6c962010-12-02 14:31:19 -0800125}
126
David Rientjes30467e02015-04-14 15:45:11 -0700127void mem_hotplug_begin(void)
Vladimir Davydovbfc8c902014-06-04 16:07:18 -0700128{
129 mem_hotplug.active_writer = current;
130
131 memhp_lock_acquire();
132 for (;;) {
133 mutex_lock(&mem_hotplug.lock);
134 if (likely(!mem_hotplug.refcount))
135 break;
136 __set_current_state(TASK_UNINTERRUPTIBLE);
137 mutex_unlock(&mem_hotplug.lock);
138 schedule();
139 }
140}
141
David Rientjes30467e02015-04-14 15:45:11 -0700142void mem_hotplug_done(void)
Vladimir Davydovbfc8c902014-06-04 16:07:18 -0700143{
144 mem_hotplug.active_writer = NULL;
145 mutex_unlock(&mem_hotplug.lock);
146 memhp_lock_release();
147}
KOSAKI Motohiro20d6c962010-12-02 14:31:19 -0800148
Keith Mannthey45e0b782006-09-30 23:27:09 -0700149/* add this memory to iomem resource */
150static struct resource *register_memory_resource(u64 start, u64 size)
151{
152 struct resource *res;
153 res = kzalloc(sizeof(struct resource), GFP_KERNEL);
Vitaly Kuznetsov6f754ba2016-01-14 15:21:55 -0800154 if (!res)
155 return ERR_PTR(-ENOMEM);
Keith Mannthey45e0b782006-09-30 23:27:09 -0700156
157 res->name = "System RAM";
158 res->start = start;
159 res->end = start + size - 1;
Toshi Kani782b8662016-01-26 21:57:24 +0100160 res->flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
Keith Mannthey45e0b782006-09-30 23:27:09 -0700161 if (request_resource(&iomem_resource, res) < 0) {
Toshi Kani4996eed2013-07-03 15:02:39 -0700162 pr_debug("System RAM resource %pR cannot be added\n", res);
Keith Mannthey45e0b782006-09-30 23:27:09 -0700163 kfree(res);
Vitaly Kuznetsov6f754ba2016-01-14 15:21:55 -0800164 return ERR_PTR(-EEXIST);
Keith Mannthey45e0b782006-09-30 23:27:09 -0700165 }
166 return res;
167}
168
169static void release_memory_resource(struct resource *res)
170{
171 if (!res)
172 return;
173 release_resource(res);
174 kfree(res);
175 return;
176}
177
Keith Mannthey53947022006-09-30 23:27:08 -0700178#ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
Yasuaki Ishimatsu46723bf2013-02-22 16:33:00 -0800179void get_page_bootmem(unsigned long info, struct page *page,
180 unsigned long type)
Yasunori Goto04753272008-04-28 02:13:31 -0700181{
Andrea Arcangeli5f24ce52011-01-13 15:47:00 -0800182 page->lru.next = (struct list_head *) type;
Yasunori Goto04753272008-04-28 02:13:31 -0700183 SetPagePrivate(page);
184 set_page_private(page, info);
Joonsoo Kimfe896d12016-03-17 14:19:26 -0700185 page_ref_inc(page);
Yasunori Goto04753272008-04-28 02:13:31 -0700186}
187
Jiang Liu170a5a72013-07-03 15:03:17 -0700188void put_page_bootmem(struct page *page)
Yasunori Goto04753272008-04-28 02:13:31 -0700189{
Andrea Arcangeli5f24ce52011-01-13 15:47:00 -0800190 unsigned long type;
Yasunori Goto04753272008-04-28 02:13:31 -0700191
Andrea Arcangeli5f24ce52011-01-13 15:47:00 -0800192 type = (unsigned long) page->lru.next;
193 BUG_ON(type < MEMORY_HOTPLUG_MIN_BOOTMEM_TYPE ||
194 type > MEMORY_HOTPLUG_MAX_BOOTMEM_TYPE);
Yasunori Goto04753272008-04-28 02:13:31 -0700195
Joonsoo Kimfe896d12016-03-17 14:19:26 -0700196 if (page_ref_dec_return(page) == 1) {
Yasunori Goto04753272008-04-28 02:13:31 -0700197 ClearPagePrivate(page);
198 set_page_private(page, 0);
Andrea Arcangeli5f24ce52011-01-13 15:47:00 -0800199 INIT_LIST_HEAD(&page->lru);
Jiang Liu170a5a72013-07-03 15:03:17 -0700200 free_reserved_page(page);
Yasunori Goto04753272008-04-28 02:13:31 -0700201 }
Yasunori Goto04753272008-04-28 02:13:31 -0700202}
203
Yasuaki Ishimatsu46723bf2013-02-22 16:33:00 -0800204#ifdef CONFIG_HAVE_BOOTMEM_INFO_NODE
205#ifndef CONFIG_SPARSEMEM_VMEMMAP
Adrian Bunkd92bc312008-07-23 21:28:12 -0700206static void register_page_bootmem_info_section(unsigned long start_pfn)
Yasunori Goto04753272008-04-28 02:13:31 -0700207{
208 unsigned long *usemap, mapsize, section_nr, i;
209 struct mem_section *ms;
210 struct page *page, *memmap;
211
Yasunori Goto04753272008-04-28 02:13:31 -0700212 section_nr = pfn_to_section_nr(start_pfn);
213 ms = __nr_to_section(section_nr);
214
215 /* Get section's memmap address */
216 memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
217
218 /*
219 * Get page for the memmap's phys address
220 * XXX: need more consideration for sparse_vmemmap...
221 */
222 page = virt_to_page(memmap);
223 mapsize = sizeof(struct page) * PAGES_PER_SECTION;
224 mapsize = PAGE_ALIGN(mapsize) >> PAGE_SHIFT;
225
226 /* remember memmap's page */
227 for (i = 0; i < mapsize; i++, page++)
228 get_page_bootmem(section_nr, page, SECTION_INFO);
229
230 usemap = __nr_to_section(section_nr)->pageblock_flags;
231 page = virt_to_page(usemap);
232
233 mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT;
234
235 for (i = 0; i < mapsize; i++, page++)
Yasunori Gotoaf370fb2008-07-23 21:28:17 -0700236 get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
Yasunori Goto04753272008-04-28 02:13:31 -0700237
238}
Yasuaki Ishimatsu46723bf2013-02-22 16:33:00 -0800239#else /* CONFIG_SPARSEMEM_VMEMMAP */
240static void register_page_bootmem_info_section(unsigned long start_pfn)
241{
242 unsigned long *usemap, mapsize, section_nr, i;
243 struct mem_section *ms;
244 struct page *page, *memmap;
245
246 if (!pfn_valid(start_pfn))
247 return;
248
249 section_nr = pfn_to_section_nr(start_pfn);
250 ms = __nr_to_section(section_nr);
251
252 memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
253
254 register_page_bootmem_memmap(section_nr, memmap, PAGES_PER_SECTION);
255
256 usemap = __nr_to_section(section_nr)->pageblock_flags;
257 page = virt_to_page(usemap);
258
259 mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT;
260
261 for (i = 0; i < mapsize; i++, page++)
262 get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
263}
264#endif /* !CONFIG_SPARSEMEM_VMEMMAP */
Yasunori Goto04753272008-04-28 02:13:31 -0700265
Linus Torvalds7ded3842016-05-27 15:23:32 -0700266void __init register_page_bootmem_info_node(struct pglist_data *pgdat)
Yasunori Goto04753272008-04-28 02:13:31 -0700267{
268 unsigned long i, pfn, end_pfn, nr_pages;
269 int node = pgdat->node_id;
270 struct page *page;
271 struct zone *zone;
272
273 nr_pages = PAGE_ALIGN(sizeof(struct pglist_data)) >> PAGE_SHIFT;
274 page = virt_to_page(pgdat);
275
276 for (i = 0; i < nr_pages; i++, page++)
277 get_page_bootmem(node, page, NODE_INFO);
278
279 zone = &pgdat->node_zones[0];
280 for (; zone < pgdat->node_zones + MAX_NR_ZONES - 1; zone++) {
Xishi Qiu139c2d72013-09-11 14:21:46 -0700281 if (zone_is_initialized(zone)) {
Yasunori Goto04753272008-04-28 02:13:31 -0700282 nr_pages = zone->wait_table_hash_nr_entries
283 * sizeof(wait_queue_head_t);
284 nr_pages = PAGE_ALIGN(nr_pages) >> PAGE_SHIFT;
285 page = virt_to_page(zone->wait_table);
286
287 for (i = 0; i < nr_pages; i++, page++)
288 get_page_bootmem(node, page, NODE_INFO);
289 }
290 }
291
292 pfn = pgdat->node_start_pfn;
Cody P Schaferc1f19492013-02-22 16:35:32 -0800293 end_pfn = pgdat_end_pfn(pgdat);
Yasunori Goto04753272008-04-28 02:13:31 -0700294
Tang Chen7e9f5eb2013-07-08 16:00:23 -0700295 /* register section info */
qiuxishif14851a2012-09-17 14:09:24 -0700296 for (; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
297 /*
298 * Some platforms can assign the same pfn to multiple nodes - on
299 * node0 as well as nodeN. To avoid registering a pfn against
300 * multiple nodes we check that this pfn does not already
Tang Chen7e9f5eb2013-07-08 16:00:23 -0700301 * reside in some other nodes.
qiuxishif14851a2012-09-17 14:09:24 -0700302 */
Yang Shif65e91df2016-05-27 14:27:32 -0700303 if (pfn_valid(pfn) && (early_pfn_to_nid(pfn) == node))
qiuxishif14851a2012-09-17 14:09:24 -0700304 register_page_bootmem_info_section(pfn);
305 }
Yasunori Goto04753272008-04-28 02:13:31 -0700306}
Yasuaki Ishimatsu46723bf2013-02-22 16:33:00 -0800307#endif /* CONFIG_HAVE_BOOTMEM_INFO_NODE */
Yasunori Goto04753272008-04-28 02:13:31 -0700308
Fabian Frederickf2765402014-08-06 16:04:57 -0700309static void __meminit grow_zone_span(struct zone *zone, unsigned long start_pfn,
310 unsigned long end_pfn)
Heiko Carstens76cdd582008-05-14 16:05:52 -0700311{
312 unsigned long old_zone_end_pfn;
313
314 zone_span_writelock(zone);
315
Xishi Qiuc33bc312013-09-11 14:21:44 -0700316 old_zone_end_pfn = zone_end_pfn(zone);
Xishi Qiu8080fc02013-09-11 14:21:45 -0700317 if (zone_is_empty(zone) || start_pfn < zone->zone_start_pfn)
Heiko Carstens76cdd582008-05-14 16:05:52 -0700318 zone->zone_start_pfn = start_pfn;
319
320 zone->spanned_pages = max(old_zone_end_pfn, end_pfn) -
321 zone->zone_start_pfn;
322
323 zone_span_writeunlock(zone);
324}
325
Lai Jiangshan511c2ab2012-12-11 16:03:16 -0800326static void resize_zone(struct zone *zone, unsigned long start_pfn,
327 unsigned long end_pfn)
328{
329 zone_span_writelock(zone);
330
Lai Jiangshane455a9b2012-12-11 16:03:20 -0800331 if (end_pfn - start_pfn) {
332 zone->zone_start_pfn = start_pfn;
333 zone->spanned_pages = end_pfn - start_pfn;
334 } else {
335 /*
336 * make it consist as free_area_init_core(),
337 * if spanned_pages = 0, then keep start_pfn = 0
338 */
339 zone->zone_start_pfn = 0;
340 zone->spanned_pages = 0;
341 }
Lai Jiangshan511c2ab2012-12-11 16:03:16 -0800342
343 zone_span_writeunlock(zone);
344}
345
346static void fix_zone_id(struct zone *zone, unsigned long start_pfn,
347 unsigned long end_pfn)
348{
349 enum zone_type zid = zone_idx(zone);
350 int nid = zone->zone_pgdat->node_id;
351 unsigned long pfn;
352
353 for (pfn = start_pfn; pfn < end_pfn; pfn++)
354 set_page_links(pfn_to_page(pfn), zid, nid, pfn);
355}
356
Cody P Schaferf6bbb782013-02-22 16:35:30 -0800357/* Can fail with -ENOMEM from allocating a wait table with vmalloc() or
Santosh Shilimkar9e43aa22014-01-21 15:50:43 -0800358 * alloc_bootmem_node_nopanic()/memblock_virt_alloc_node_nopanic() */
Cody P Schaferf6bbb782013-02-22 16:35:30 -0800359static int __ref ensure_zone_is_initialized(struct zone *zone,
360 unsigned long start_pfn, unsigned long num_pages)
361{
362 if (!zone_is_initialized(zone))
Yaowei Baib171e402015-11-05 18:47:06 -0800363 return init_currently_empty_zone(zone, start_pfn, num_pages);
364
Cody P Schaferf6bbb782013-02-22 16:35:30 -0800365 return 0;
366}
367
Lai Jiangshane455a9b2012-12-11 16:03:20 -0800368static int __meminit move_pfn_range_left(struct zone *z1, struct zone *z2,
Lai Jiangshan511c2ab2012-12-11 16:03:16 -0800369 unsigned long start_pfn, unsigned long end_pfn)
370{
Lai Jiangshane455a9b2012-12-11 16:03:20 -0800371 int ret;
Lai Jiangshan511c2ab2012-12-11 16:03:16 -0800372 unsigned long flags;
Lai Jiangshane455a9b2012-12-11 16:03:20 -0800373 unsigned long z1_start_pfn;
374
Cody P Schafer64dd1b22013-02-22 16:35:31 -0800375 ret = ensure_zone_is_initialized(z1, start_pfn, end_pfn - start_pfn);
376 if (ret)
377 return ret;
Lai Jiangshan511c2ab2012-12-11 16:03:16 -0800378
379 pgdat_resize_lock(z1->zone_pgdat, &flags);
380
381 /* can't move pfns which are higher than @z2 */
Cody P Schafer108bcc92013-02-22 16:35:23 -0800382 if (end_pfn > zone_end_pfn(z2))
Lai Jiangshan511c2ab2012-12-11 16:03:16 -0800383 goto out_fail;
Jiang Liu834405c2013-07-03 15:03:04 -0700384 /* the move out part must be at the left most of @z2 */
Lai Jiangshan511c2ab2012-12-11 16:03:16 -0800385 if (start_pfn > z2->zone_start_pfn)
386 goto out_fail;
387 /* must included/overlap */
388 if (end_pfn <= z2->zone_start_pfn)
389 goto out_fail;
390
Lai Jiangshane455a9b2012-12-11 16:03:20 -0800391 /* use start_pfn for z1's start_pfn if z1 is empty */
Xishi Qiu8080fc02013-09-11 14:21:45 -0700392 if (!zone_is_empty(z1))
Lai Jiangshane455a9b2012-12-11 16:03:20 -0800393 z1_start_pfn = z1->zone_start_pfn;
394 else
395 z1_start_pfn = start_pfn;
396
397 resize_zone(z1, z1_start_pfn, end_pfn);
Cody P Schafer108bcc92013-02-22 16:35:23 -0800398 resize_zone(z2, end_pfn, zone_end_pfn(z2));
Lai Jiangshan511c2ab2012-12-11 16:03:16 -0800399
400 pgdat_resize_unlock(z1->zone_pgdat, &flags);
401
402 fix_zone_id(z1, start_pfn, end_pfn);
403
404 return 0;
405out_fail:
406 pgdat_resize_unlock(z1->zone_pgdat, &flags);
407 return -1;
408}
409
Lai Jiangshane455a9b2012-12-11 16:03:20 -0800410static int __meminit move_pfn_range_right(struct zone *z1, struct zone *z2,
Lai Jiangshan511c2ab2012-12-11 16:03:16 -0800411 unsigned long start_pfn, unsigned long end_pfn)
412{
Lai Jiangshane455a9b2012-12-11 16:03:20 -0800413 int ret;
Lai Jiangshan511c2ab2012-12-11 16:03:16 -0800414 unsigned long flags;
Lai Jiangshane455a9b2012-12-11 16:03:20 -0800415 unsigned long z2_end_pfn;
416
Cody P Schafer64dd1b22013-02-22 16:35:31 -0800417 ret = ensure_zone_is_initialized(z2, start_pfn, end_pfn - start_pfn);
418 if (ret)
419 return ret;
Lai Jiangshan511c2ab2012-12-11 16:03:16 -0800420
421 pgdat_resize_lock(z1->zone_pgdat, &flags);
422
423 /* can't move pfns which are lower than @z1 */
424 if (z1->zone_start_pfn > start_pfn)
425 goto out_fail;
426 /* the move out part mast at the right most of @z1 */
Cody P Schafer108bcc92013-02-22 16:35:23 -0800427 if (zone_end_pfn(z1) > end_pfn)
Lai Jiangshan511c2ab2012-12-11 16:03:16 -0800428 goto out_fail;
429 /* must included/overlap */
Cody P Schafer108bcc92013-02-22 16:35:23 -0800430 if (start_pfn >= zone_end_pfn(z1))
Lai Jiangshan511c2ab2012-12-11 16:03:16 -0800431 goto out_fail;
432
Lai Jiangshane455a9b2012-12-11 16:03:20 -0800433 /* use end_pfn for z2's end_pfn if z2 is empty */
Xishi Qiu8080fc02013-09-11 14:21:45 -0700434 if (!zone_is_empty(z2))
Cody P Schafer108bcc92013-02-22 16:35:23 -0800435 z2_end_pfn = zone_end_pfn(z2);
Lai Jiangshane455a9b2012-12-11 16:03:20 -0800436 else
437 z2_end_pfn = end_pfn;
438
Lai Jiangshan511c2ab2012-12-11 16:03:16 -0800439 resize_zone(z1, z1->zone_start_pfn, start_pfn);
Lai Jiangshane455a9b2012-12-11 16:03:20 -0800440 resize_zone(z2, start_pfn, z2_end_pfn);
Lai Jiangshan511c2ab2012-12-11 16:03:16 -0800441
442 pgdat_resize_unlock(z1->zone_pgdat, &flags);
443
444 fix_zone_id(z2, start_pfn, end_pfn);
445
446 return 0;
447out_fail:
448 pgdat_resize_unlock(z1->zone_pgdat, &flags);
449 return -1;
450}
451
Reza Arbabe51e6c82016-07-26 15:22:20 -0700452static struct zone * __meminit move_pfn_range(int zone_shift,
453 unsigned long start_pfn, unsigned long end_pfn)
454{
455 struct zone *zone = page_zone(pfn_to_page(start_pfn));
456 int ret = 0;
457
458 if (zone_shift < 0)
459 ret = move_pfn_range_left(zone + zone_shift, zone,
460 start_pfn, end_pfn);
461 else if (zone_shift)
462 ret = move_pfn_range_right(zone, zone + zone_shift,
463 start_pfn, end_pfn);
464
465 if (ret)
466 return NULL;
467
468 return zone + zone_shift;
469}
470
Fabian Frederickf2765402014-08-06 16:04:57 -0700471static void __meminit grow_pgdat_span(struct pglist_data *pgdat, unsigned long start_pfn,
472 unsigned long end_pfn)
Heiko Carstens76cdd582008-05-14 16:05:52 -0700473{
Xishi Qiu83285c72013-11-12 15:07:19 -0800474 unsigned long old_pgdat_end_pfn = pgdat_end_pfn(pgdat);
Heiko Carstens76cdd582008-05-14 16:05:52 -0700475
Tang Chen712cd382012-12-11 16:01:07 -0800476 if (!pgdat->node_spanned_pages || start_pfn < pgdat->node_start_pfn)
Heiko Carstens76cdd582008-05-14 16:05:52 -0700477 pgdat->node_start_pfn = start_pfn;
478
479 pgdat->node_spanned_pages = max(old_pgdat_end_pfn, end_pfn) -
480 pgdat->node_start_pfn;
481}
482
Al Viro31168482008-11-22 17:33:24 +0000483static int __meminit __add_zone(struct zone *zone, unsigned long phys_start_pfn)
Dave Hansen3947be12005-10-29 18:16:54 -0700484{
485 struct pglist_data *pgdat = zone->zone_pgdat;
486 int nr_pages = PAGES_PER_SECTION;
487 int nid = pgdat->node_id;
488 int zone_type;
Mel Gormane298ff72015-08-06 15:46:51 -0700489 unsigned long flags, pfn;
Cody P Schafer64dd1b22013-02-22 16:35:31 -0800490 int ret;
Dave Hansen3947be12005-10-29 18:16:54 -0700491
492 zone_type = zone - pgdat->node_zones;
Cody P Schafer64dd1b22013-02-22 16:35:31 -0800493 ret = ensure_zone_is_initialized(zone, phys_start_pfn, nr_pages);
494 if (ret)
495 return ret;
Heiko Carstens76cdd582008-05-14 16:05:52 -0700496
Heiko Carstens76cdd582008-05-14 16:05:52 -0700497 pgdat_resize_lock(zone->zone_pgdat, &flags);
498 grow_zone_span(zone, phys_start_pfn, phys_start_pfn + nr_pages);
499 grow_pgdat_span(zone->zone_pgdat, phys_start_pfn,
500 phys_start_pfn + nr_pages);
501 pgdat_resize_unlock(zone->zone_pgdat, &flags);
Dave Hansena2f3aa022007-01-10 23:15:30 -0800502 memmap_init_zone(nr_pages, nid, zone_type,
503 phys_start_pfn, MEMMAP_HOTPLUG);
Mel Gormane298ff72015-08-06 15:46:51 -0700504
505 /* online_page_range is called later and expects pages reserved */
506 for (pfn = phys_start_pfn; pfn < phys_start_pfn + nr_pages; pfn++) {
507 if (!pfn_valid(pfn))
508 continue;
509
510 SetPageReserved(pfn_to_page(pfn));
511 }
Yasunori Goto718127c2006-06-23 02:03:10 -0700512 return 0;
Dave Hansen3947be12005-10-29 18:16:54 -0700513}
514
Gary Hadec04fc582009-01-06 14:39:14 -0800515static int __meminit __add_section(int nid, struct zone *zone,
516 unsigned long phys_start_pfn)
Dave Hansen3947be12005-10-29 18:16:54 -0700517{
Dave Hansen3947be12005-10-29 18:16:54 -0700518 int ret;
519
KAMEZAWA Hiroyukiebd15302006-08-05 12:15:06 -0700520 if (pfn_valid(phys_start_pfn))
521 return -EEXIST;
522
Zhang Yanfei85b35fe2013-11-12 15:07:42 -0800523 ret = sparse_add_one_section(zone, phys_start_pfn);
Dave Hansen3947be12005-10-29 18:16:54 -0700524
525 if (ret < 0)
526 return ret;
527
Yasunori Goto718127c2006-06-23 02:03:10 -0700528 ret = __add_zone(zone, phys_start_pfn);
529
530 if (ret < 0)
531 return ret;
532
Gary Hadec04fc582009-01-06 14:39:14 -0800533 return register_new_memory(nid, __pfn_to_section(phys_start_pfn));
Dave Hansen3947be12005-10-29 18:16:54 -0700534}
535
David Rientjes4edd7ce2013-04-29 15:08:22 -0700536/*
537 * Reasonably generic function for adding memory. It is
538 * expected that archs that support memory hotplug will
539 * call this function after deciding the zone to which to
540 * add the new pages.
541 */
542int __ref __add_pages(int nid, struct zone *zone, unsigned long phys_start_pfn,
543 unsigned long nr_pages)
544{
545 unsigned long i;
546 int err = 0;
547 int start_sec, end_sec;
Dan Williams4b94ffd2016-01-15 16:56:22 -0800548 struct vmem_altmap *altmap;
549
Joonsoo Kim7cf91a92016-03-15 14:57:51 -0700550 clear_zone_contiguous(zone);
551
David Rientjes4edd7ce2013-04-29 15:08:22 -0700552 /* during initialize mem_map, align hot-added range to section */
553 start_sec = pfn_to_section_nr(phys_start_pfn);
554 end_sec = pfn_to_section_nr(phys_start_pfn + nr_pages - 1);
555
Dan Williams4b94ffd2016-01-15 16:56:22 -0800556 altmap = to_vmem_altmap((unsigned long) pfn_to_page(phys_start_pfn));
557 if (altmap) {
558 /*
559 * Validate altmap is within bounds of the total request
560 */
561 if (altmap->base_pfn != phys_start_pfn
562 || vmem_altmap_offset(altmap) > nr_pages) {
563 pr_warn_once("memory add fail, invalid altmap\n");
Joonsoo Kim7cf91a92016-03-15 14:57:51 -0700564 err = -EINVAL;
565 goto out;
Dan Williams4b94ffd2016-01-15 16:56:22 -0800566 }
567 altmap->alloc = 0;
568 }
569
David Rientjes4edd7ce2013-04-29 15:08:22 -0700570 for (i = start_sec; i <= end_sec; i++) {
Sheng Yong19c07d52015-04-14 15:44:54 -0700571 err = __add_section(nid, zone, section_nr_to_pfn(i));
David Rientjes4edd7ce2013-04-29 15:08:22 -0700572
573 /*
574 * EEXIST is finally dealt with by ioresource collision
575 * check. see add_memory() => register_memory_resource()
576 * Warning will be printed if there is collision.
577 */
578 if (err && (err != -EEXIST))
579 break;
580 err = 0;
581 }
Zhu Guihuac435a392015-06-24 16:58:42 -0700582 vmemmap_populate_print_last();
Joonsoo Kim7cf91a92016-03-15 14:57:51 -0700583out:
584 set_zone_contiguous(zone);
David Rientjes4edd7ce2013-04-29 15:08:22 -0700585 return err;
586}
587EXPORT_SYMBOL_GPL(__add_pages);
588
589#ifdef CONFIG_MEMORY_HOTREMOVE
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800590/* find the smallest valid pfn in the range [start_pfn, end_pfn) */
591static int find_smallest_section_pfn(int nid, struct zone *zone,
592 unsigned long start_pfn,
593 unsigned long end_pfn)
594{
595 struct mem_section *ms;
596
597 for (; start_pfn < end_pfn; start_pfn += PAGES_PER_SECTION) {
598 ms = __pfn_to_section(start_pfn);
599
600 if (unlikely(!valid_section(ms)))
601 continue;
602
603 if (unlikely(pfn_to_nid(start_pfn) != nid))
604 continue;
605
606 if (zone && zone != page_zone(pfn_to_page(start_pfn)))
607 continue;
608
609 return start_pfn;
610 }
611
612 return 0;
613}
614
615/* find the biggest valid pfn in the range [start_pfn, end_pfn). */
616static int find_biggest_section_pfn(int nid, struct zone *zone,
617 unsigned long start_pfn,
618 unsigned long end_pfn)
619{
620 struct mem_section *ms;
621 unsigned long pfn;
622
623 /* pfn is the end pfn of a memory section. */
624 pfn = end_pfn - 1;
625 for (; pfn >= start_pfn; pfn -= PAGES_PER_SECTION) {
626 ms = __pfn_to_section(pfn);
627
628 if (unlikely(!valid_section(ms)))
629 continue;
630
631 if (unlikely(pfn_to_nid(pfn) != nid))
632 continue;
633
634 if (zone && zone != page_zone(pfn_to_page(pfn)))
635 continue;
636
637 return pfn;
638 }
639
640 return 0;
641}
642
643static void shrink_zone_span(struct zone *zone, unsigned long start_pfn,
644 unsigned long end_pfn)
645{
Xishi Qiuc33bc312013-09-11 14:21:44 -0700646 unsigned long zone_start_pfn = zone->zone_start_pfn;
647 unsigned long z = zone_end_pfn(zone); /* zone_end_pfn namespace clash */
648 unsigned long zone_end_pfn = z;
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800649 unsigned long pfn;
650 struct mem_section *ms;
651 int nid = zone_to_nid(zone);
652
653 zone_span_writelock(zone);
654 if (zone_start_pfn == start_pfn) {
655 /*
656 * If the section is smallest section in the zone, it need
657 * shrink zone->zone_start_pfn and zone->zone_spanned_pages.
658 * In this case, we find second smallest valid mem_section
659 * for shrinking zone.
660 */
661 pfn = find_smallest_section_pfn(nid, zone, end_pfn,
662 zone_end_pfn);
663 if (pfn) {
664 zone->zone_start_pfn = pfn;
665 zone->spanned_pages = zone_end_pfn - pfn;
666 }
667 } else if (zone_end_pfn == end_pfn) {
668 /*
669 * If the section is biggest section in the zone, it need
670 * shrink zone->spanned_pages.
671 * In this case, we find second biggest valid mem_section for
672 * shrinking zone.
673 */
674 pfn = find_biggest_section_pfn(nid, zone, zone_start_pfn,
675 start_pfn);
676 if (pfn)
677 zone->spanned_pages = pfn - zone_start_pfn + 1;
678 }
679
680 /*
681 * The section is not biggest or smallest mem_section in the zone, it
682 * only creates a hole in the zone. So in this case, we need not
683 * change the zone. But perhaps, the zone has only hole data. Thus
684 * it check the zone has only hole or not.
685 */
686 pfn = zone_start_pfn;
687 for (; pfn < zone_end_pfn; pfn += PAGES_PER_SECTION) {
688 ms = __pfn_to_section(pfn);
689
690 if (unlikely(!valid_section(ms)))
691 continue;
692
693 if (page_zone(pfn_to_page(pfn)) != zone)
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 zone_span_writeunlock(zone);
702 return;
703 }
704
705 /* The zone has no valid section */
706 zone->zone_start_pfn = 0;
707 zone->spanned_pages = 0;
708 zone_span_writeunlock(zone);
709}
710
711static void shrink_pgdat_span(struct pglist_data *pgdat,
712 unsigned long start_pfn, unsigned long end_pfn)
713{
Xishi Qiu83285c72013-11-12 15:07:19 -0800714 unsigned long pgdat_start_pfn = pgdat->node_start_pfn;
715 unsigned long p = pgdat_end_pfn(pgdat); /* pgdat_end_pfn namespace clash */
716 unsigned long pgdat_end_pfn = p;
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800717 unsigned long pfn;
718 struct mem_section *ms;
719 int nid = pgdat->node_id;
720
721 if (pgdat_start_pfn == start_pfn) {
722 /*
723 * If the section is smallest section in the pgdat, it need
724 * shrink pgdat->node_start_pfn and pgdat->node_spanned_pages.
725 * In this case, we find second smallest valid mem_section
726 * for shrinking zone.
727 */
728 pfn = find_smallest_section_pfn(nid, NULL, end_pfn,
729 pgdat_end_pfn);
730 if (pfn) {
731 pgdat->node_start_pfn = pfn;
732 pgdat->node_spanned_pages = pgdat_end_pfn - pfn;
733 }
734 } else if (pgdat_end_pfn == end_pfn) {
735 /*
736 * If the section is biggest section in the pgdat, it need
737 * shrink pgdat->node_spanned_pages.
738 * In this case, we find second biggest valid mem_section for
739 * shrinking zone.
740 */
741 pfn = find_biggest_section_pfn(nid, NULL, pgdat_start_pfn,
742 start_pfn);
743 if (pfn)
744 pgdat->node_spanned_pages = pfn - pgdat_start_pfn + 1;
745 }
746
747 /*
748 * If the section is not biggest or smallest mem_section in the pgdat,
749 * it only creates a hole in the pgdat. So in this case, we need not
750 * change the pgdat.
751 * But perhaps, the pgdat has only hole data. Thus it check the pgdat
752 * has only hole or not.
753 */
754 pfn = pgdat_start_pfn;
755 for (; pfn < pgdat_end_pfn; pfn += PAGES_PER_SECTION) {
756 ms = __pfn_to_section(pfn);
757
758 if (unlikely(!valid_section(ms)))
759 continue;
760
761 if (pfn_to_nid(pfn) != nid)
762 continue;
763
764 /* If the section is current section, it continues the loop */
765 if (start_pfn == pfn)
766 continue;
767
768 /* If we find valid section, we have nothing to do */
769 return;
770 }
771
772 /* The pgdat has no valid section */
773 pgdat->node_start_pfn = 0;
774 pgdat->node_spanned_pages = 0;
775}
776
777static void __remove_zone(struct zone *zone, unsigned long start_pfn)
778{
779 struct pglist_data *pgdat = zone->zone_pgdat;
780 int nr_pages = PAGES_PER_SECTION;
781 int zone_type;
782 unsigned long flags;
783
784 zone_type = zone - pgdat->node_zones;
785
786 pgdat_resize_lock(zone->zone_pgdat, &flags);
787 shrink_zone_span(zone, start_pfn, start_pfn + nr_pages);
788 shrink_pgdat_span(pgdat, start_pfn, start_pfn + nr_pages);
789 pgdat_resize_unlock(zone->zone_pgdat, &flags);
790}
791
Dan Williams4b94ffd2016-01-15 16:56:22 -0800792static int __remove_section(struct zone *zone, struct mem_section *ms,
793 unsigned long map_offset)
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700794{
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800795 unsigned long start_pfn;
796 int scn_nr;
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700797 int ret = -EINVAL;
798
799 if (!valid_section(ms))
800 return ret;
801
802 ret = unregister_memory_section(ms);
803 if (ret)
804 return ret;
805
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800806 scn_nr = __section_nr(ms);
807 start_pfn = section_nr_to_pfn(scn_nr);
808 __remove_zone(zone, start_pfn);
809
Dan Williams4b94ffd2016-01-15 16:56:22 -0800810 sparse_remove_one_section(zone, ms, map_offset);
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700811 return 0;
812}
813
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700814/**
815 * __remove_pages() - remove sections of pages from a zone
816 * @zone: zone from which pages need to be removed
817 * @phys_start_pfn: starting pageframe (must be aligned to start of a section)
818 * @nr_pages: number of pages to remove (must be multiple of section size)
819 *
820 * Generic helper function to remove section mappings and sysfs entries
821 * for the section of the memory we are removing. Caller needs to make
822 * sure that pages are marked reserved and zones are adjust properly by
823 * calling offline_pages().
824 */
825int __remove_pages(struct zone *zone, unsigned long phys_start_pfn,
826 unsigned long nr_pages)
827{
Toshi Kanife74ebb2013-04-29 15:08:20 -0700828 unsigned long i;
Dan Williams4b94ffd2016-01-15 16:56:22 -0800829 unsigned long map_offset = 0;
830 int sections_to_remove, ret = 0;
831
832 /* In the ZONE_DEVICE case device driver owns the memory region */
833 if (is_dev_zone(zone)) {
834 struct page *page = pfn_to_page(phys_start_pfn);
835 struct vmem_altmap *altmap;
836
837 altmap = to_vmem_altmap((unsigned long) page);
838 if (altmap)
839 map_offset = vmem_altmap_offset(altmap);
840 } else {
841 resource_size_t start, size;
842
843 start = phys_start_pfn << PAGE_SHIFT;
844 size = nr_pages * PAGE_SIZE;
845
846 ret = release_mem_region_adjustable(&iomem_resource, start,
847 size);
848 if (ret) {
849 resource_size_t endres = start + size - 1;
850
851 pr_warn("Unable to release resource <%pa-%pa> (%d)\n",
852 &start, &endres, ret);
853 }
854 }
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700855
Joonsoo Kim7cf91a92016-03-15 14:57:51 -0700856 clear_zone_contiguous(zone);
857
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700858 /*
859 * We can only remove entire sections
860 */
861 BUG_ON(phys_start_pfn & ~PAGE_SECTION_MASK);
862 BUG_ON(nr_pages % PAGES_PER_SECTION);
863
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700864 sections_to_remove = nr_pages / PAGES_PER_SECTION;
865 for (i = 0; i < sections_to_remove; i++) {
866 unsigned long pfn = phys_start_pfn + i*PAGES_PER_SECTION;
Dan Williams4b94ffd2016-01-15 16:56:22 -0800867
868 ret = __remove_section(zone, __pfn_to_section(pfn), map_offset);
869 map_offset = 0;
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700870 if (ret)
871 break;
872 }
Joonsoo Kim7cf91a92016-03-15 14:57:51 -0700873
874 set_zone_contiguous(zone);
875
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700876 return ret;
877}
878EXPORT_SYMBOL_GPL(__remove_pages);
David Rientjes4edd7ce2013-04-29 15:08:22 -0700879#endif /* CONFIG_MEMORY_HOTREMOVE */
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700880
Daniel Kiper9d0ad8c2011-07-25 17:12:05 -0700881int set_online_page_callback(online_page_callback_t callback)
882{
883 int rc = -EINVAL;
884
Vladimir Davydovbfc8c902014-06-04 16:07:18 -0700885 get_online_mems();
886 mutex_lock(&online_page_callback_lock);
Daniel Kiper9d0ad8c2011-07-25 17:12:05 -0700887
888 if (online_page_callback == generic_online_page) {
889 online_page_callback = callback;
890 rc = 0;
891 }
892
Vladimir Davydovbfc8c902014-06-04 16:07:18 -0700893 mutex_unlock(&online_page_callback_lock);
894 put_online_mems();
Daniel Kiper9d0ad8c2011-07-25 17:12:05 -0700895
896 return rc;
897}
898EXPORT_SYMBOL_GPL(set_online_page_callback);
899
900int restore_online_page_callback(online_page_callback_t callback)
901{
902 int rc = -EINVAL;
903
Vladimir Davydovbfc8c902014-06-04 16:07:18 -0700904 get_online_mems();
905 mutex_lock(&online_page_callback_lock);
Daniel Kiper9d0ad8c2011-07-25 17:12:05 -0700906
907 if (online_page_callback == callback) {
908 online_page_callback = generic_online_page;
909 rc = 0;
910 }
911
Vladimir Davydovbfc8c902014-06-04 16:07:18 -0700912 mutex_unlock(&online_page_callback_lock);
913 put_online_mems();
Daniel Kiper9d0ad8c2011-07-25 17:12:05 -0700914
915 return rc;
916}
917EXPORT_SYMBOL_GPL(restore_online_page_callback);
918
919void __online_page_set_limits(struct page *page)
Jeremy Fitzhardinge180c06e2008-04-28 02:12:03 -0700920{
Daniel Kiper9d0ad8c2011-07-25 17:12:05 -0700921}
922EXPORT_SYMBOL_GPL(__online_page_set_limits);
923
924void __online_page_increment_counters(struct page *page)
925{
Jiang Liu3dcc0572013-07-03 15:03:21 -0700926 adjust_managed_page_count(page, 1);
Daniel Kiper9d0ad8c2011-07-25 17:12:05 -0700927}
928EXPORT_SYMBOL_GPL(__online_page_increment_counters);
Jeremy Fitzhardinge180c06e2008-04-28 02:12:03 -0700929
Daniel Kiper9d0ad8c2011-07-25 17:12:05 -0700930void __online_page_free(struct page *page)
931{
Jiang Liu3dcc0572013-07-03 15:03:21 -0700932 __free_reserved_page(page);
Jeremy Fitzhardinge180c06e2008-04-28 02:12:03 -0700933}
Daniel Kiper9d0ad8c2011-07-25 17:12:05 -0700934EXPORT_SYMBOL_GPL(__online_page_free);
935
936static void generic_online_page(struct page *page)
937{
938 __online_page_set_limits(page);
939 __online_page_increment_counters(page);
940 __online_page_free(page);
941}
Jeremy Fitzhardinge180c06e2008-04-28 02:12:03 -0700942
KAMEZAWA Hiroyuki75884fb2007-10-16 01:26:10 -0700943static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages,
944 void *arg)
Dave Hansen3947be12005-10-29 18:16:54 -0700945{
946 unsigned long i;
KAMEZAWA Hiroyuki75884fb2007-10-16 01:26:10 -0700947 unsigned long onlined_pages = *(unsigned long *)arg;
948 struct page *page;
949 if (PageReserved(pfn_to_page(start_pfn)))
950 for (i = 0; i < nr_pages; i++) {
951 page = pfn_to_page(start_pfn + i);
Daniel Kiper9d0ad8c2011-07-25 17:12:05 -0700952 (*online_page_callback)(page);
KAMEZAWA Hiroyuki75884fb2007-10-16 01:26:10 -0700953 onlined_pages++;
954 }
955 *(unsigned long *)arg = onlined_pages;
956 return 0;
957}
958
Lai Jiangshan09285af2012-12-12 13:52:04 -0800959#ifdef CONFIG_MOVABLE_NODE
Tang Chen79a4dce2012-12-18 14:23:24 -0800960/*
961 * When CONFIG_MOVABLE_NODE, we permit onlining of a node which doesn't have
962 * normal memory.
963 */
Lai Jiangshan09285af2012-12-12 13:52:04 -0800964static bool can_online_high_movable(struct zone *zone)
965{
966 return true;
967}
Tang Chen79a4dce2012-12-18 14:23:24 -0800968#else /* CONFIG_MOVABLE_NODE */
Lai Jiangshan74d42d82012-12-11 16:03:23 -0800969/* ensure every online node has NORMAL memory */
970static bool can_online_high_movable(struct zone *zone)
971{
972 return node_state(zone_to_nid(zone), N_NORMAL_MEMORY);
973}
Tang Chen79a4dce2012-12-18 14:23:24 -0800974#endif /* CONFIG_MOVABLE_NODE */
Lai Jiangshan74d42d82012-12-11 16:03:23 -0800975
Lai Jiangshand9713672012-12-11 16:01:03 -0800976/* check which state of node_states will be changed when online memory */
977static void node_states_check_changes_online(unsigned long nr_pages,
978 struct zone *zone, struct memory_notify *arg)
979{
980 int nid = zone_to_nid(zone);
981 enum zone_type zone_last = ZONE_NORMAL;
982
983 /*
Lai Jiangshan6715ddf2012-12-12 13:51:49 -0800984 * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY]
985 * contains nodes which have zones of 0...ZONE_NORMAL,
986 * set zone_last to ZONE_NORMAL.
Lai Jiangshand9713672012-12-11 16:01:03 -0800987 *
Lai Jiangshan6715ddf2012-12-12 13:51:49 -0800988 * If we don't have HIGHMEM nor movable node,
989 * node_states[N_NORMAL_MEMORY] contains nodes which have zones of
990 * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE.
Lai Jiangshand9713672012-12-11 16:01:03 -0800991 */
Lai Jiangshan6715ddf2012-12-12 13:51:49 -0800992 if (N_MEMORY == N_NORMAL_MEMORY)
Lai Jiangshand9713672012-12-11 16:01:03 -0800993 zone_last = ZONE_MOVABLE;
994
995 /*
996 * if the memory to be online is in a zone of 0...zone_last, and
997 * the zones of 0...zone_last don't have memory before online, we will
998 * need to set the node to node_states[N_NORMAL_MEMORY] after
999 * the memory is online.
1000 */
1001 if (zone_idx(zone) <= zone_last && !node_state(nid, N_NORMAL_MEMORY))
1002 arg->status_change_nid_normal = nid;
1003 else
1004 arg->status_change_nid_normal = -1;
1005
Lai Jiangshan6715ddf2012-12-12 13:51:49 -08001006#ifdef CONFIG_HIGHMEM
1007 /*
1008 * If we have movable node, node_states[N_HIGH_MEMORY]
1009 * contains nodes which have zones of 0...ZONE_HIGHMEM,
1010 * set zone_last to ZONE_HIGHMEM.
1011 *
1012 * If we don't have movable node, node_states[N_NORMAL_MEMORY]
1013 * contains nodes which have zones of 0...ZONE_MOVABLE,
1014 * set zone_last to ZONE_MOVABLE.
1015 */
1016 zone_last = ZONE_HIGHMEM;
1017 if (N_MEMORY == N_HIGH_MEMORY)
1018 zone_last = ZONE_MOVABLE;
1019
1020 if (zone_idx(zone) <= zone_last && !node_state(nid, N_HIGH_MEMORY))
1021 arg->status_change_nid_high = nid;
1022 else
1023 arg->status_change_nid_high = -1;
1024#else
1025 arg->status_change_nid_high = arg->status_change_nid_normal;
1026#endif
1027
Lai Jiangshand9713672012-12-11 16:01:03 -08001028 /*
1029 * if the node don't have memory befor online, we will need to
Lai Jiangshan6715ddf2012-12-12 13:51:49 -08001030 * set the node to node_states[N_MEMORY] after the memory
Lai Jiangshand9713672012-12-11 16:01:03 -08001031 * is online.
1032 */
Lai Jiangshan6715ddf2012-12-12 13:51:49 -08001033 if (!node_state(nid, N_MEMORY))
Lai Jiangshand9713672012-12-11 16:01:03 -08001034 arg->status_change_nid = nid;
1035 else
1036 arg->status_change_nid = -1;
1037}
1038
1039static void node_states_set_node(int node, struct memory_notify *arg)
1040{
1041 if (arg->status_change_nid_normal >= 0)
1042 node_set_state(node, N_NORMAL_MEMORY);
1043
Lai Jiangshan6715ddf2012-12-12 13:51:49 -08001044 if (arg->status_change_nid_high >= 0)
1045 node_set_state(node, N_HIGH_MEMORY);
1046
1047 node_set_state(node, N_MEMORY);
Lai Jiangshand9713672012-12-11 16:01:03 -08001048}
1049
Reza Arbabdf429ac2016-07-26 15:22:23 -07001050int zone_can_shift(unsigned long pfn, unsigned long nr_pages,
1051 enum zone_type target)
1052{
1053 struct zone *zone = page_zone(pfn_to_page(pfn));
1054 enum zone_type idx = zone_idx(zone);
1055 int i;
1056
1057 if (idx < target) {
1058 /* pages must be at end of current zone */
1059 if (pfn + nr_pages != zone_end_pfn(zone))
1060 return 0;
1061
1062 /* no zones in use between current zone and target */
1063 for (i = idx + 1; i < target; i++)
1064 if (zone_is_initialized(zone - idx + i))
1065 return 0;
1066 }
1067
1068 if (target < idx) {
1069 /* pages must be at beginning of current zone */
1070 if (pfn != zone->zone_start_pfn)
1071 return 0;
1072
1073 /* no zones in use between current zone and target */
1074 for (i = target + 1; i < idx; i++)
1075 if (zone_is_initialized(zone - idx + i))
1076 return 0;
1077 }
1078
1079 return target - idx;
1080}
KAMEZAWA Hiroyuki75884fb2007-10-16 01:26:10 -07001081
David Rientjes30467e02015-04-14 15:45:11 -07001082/* Must be protected by mem_hotplug_begin() */
Lai Jiangshan511c2ab2012-12-11 16:03:16 -08001083int __ref online_pages(unsigned long pfn, unsigned long nr_pages, int online_type)
KAMEZAWA Hiroyuki75884fb2007-10-16 01:26:10 -07001084{
Cody P Schaferaa472282013-07-03 15:02:10 -07001085 unsigned long flags;
Dave Hansen3947be12005-10-29 18:16:54 -07001086 unsigned long onlined_pages = 0;
1087 struct zone *zone;
Yasunori Goto68113782006-06-23 02:03:11 -07001088 int need_zonelists_rebuild = 0;
Yasunori Goto7b78d332007-10-21 16:41:36 -07001089 int nid;
1090 int ret;
1091 struct memory_notify arg;
Reza Arbabe51e6c82016-07-26 15:22:20 -07001092 int zone_shift = 0;
Dave Hansen3947be12005-10-29 18:16:54 -07001093
Lai Jiangshand9713672012-12-11 16:01:03 -08001094 /*
1095 * This doesn't need a lock to do pfn_to_page().
1096 * The section can't be removed here because of the
1097 * memory_block->state_mutex.
1098 */
1099 zone = page_zone(pfn_to_page(pfn));
1100
Tang Chen4f7c6b42014-08-06 16:05:13 -07001101 if ((zone_idx(zone) > ZONE_NORMAL ||
1102 online_type == MMOP_ONLINE_MOVABLE) &&
Vladimir Davydovbfc8c902014-06-04 16:07:18 -07001103 !can_online_high_movable(zone))
David Rientjes30467e02015-04-14 15:45:11 -07001104 return -EINVAL;
Lai Jiangshan74d42d82012-12-11 16:03:23 -08001105
Reza Arbabdf429ac2016-07-26 15:22:23 -07001106 if (online_type == MMOP_ONLINE_KERNEL)
1107 zone_shift = zone_can_shift(pfn, nr_pages, ZONE_NORMAL);
1108 else if (online_type == MMOP_ONLINE_MOVABLE)
1109 zone_shift = zone_can_shift(pfn, nr_pages, ZONE_MOVABLE);
Reza Arbabe51e6c82016-07-26 15:22:20 -07001110
1111 zone = move_pfn_range(zone_shift, pfn, pfn + nr_pages);
1112 if (!zone)
1113 return -EINVAL;
Lai Jiangshan511c2ab2012-12-11 16:03:16 -08001114
Yasunori Goto7b78d332007-10-21 16:41:36 -07001115 arg.start_pfn = pfn;
1116 arg.nr_pages = nr_pages;
Lai Jiangshand9713672012-12-11 16:01:03 -08001117 node_states_check_changes_online(nr_pages, zone, &arg);
Yasunori Goto7b78d332007-10-21 16:41:36 -07001118
Vlastimil Babkae888ca32016-03-17 14:18:12 -07001119 nid = zone_to_nid(zone);
Yasunori Goto7b78d332007-10-21 16:41:36 -07001120
1121 ret = memory_notify(MEM_GOING_ONLINE, &arg);
1122 ret = notifier_to_errno(ret);
Chen Yuconge33e33b2016-03-17 14:19:35 -07001123 if (ret)
1124 goto failed_addition;
1125
Dave Hansen3947be12005-10-29 18:16:54 -07001126 /*
Yasunori Goto68113782006-06-23 02:03:11 -07001127 * If this zone is not populated, then it is not in zonelist.
1128 * This means the page allocator ignores this zone.
1129 * So, zonelist must be updated after online.
1130 */
Haicheng Li4eaf3f62010-05-24 14:32:52 -07001131 mutex_lock(&zonelists_mutex);
Wen Congyang6dcd73d2012-12-11 16:01:01 -08001132 if (!populated_zone(zone)) {
Yasunori Goto68113782006-06-23 02:03:11 -07001133 need_zonelists_rebuild = 1;
Wen Congyang6dcd73d2012-12-11 16:01:01 -08001134 build_all_zonelists(NULL, zone);
1135 }
Yasunori Goto68113782006-06-23 02:03:11 -07001136
KAMEZAWA Hiroyuki908eedc2009-09-22 16:45:46 -07001137 ret = walk_system_ram_range(pfn, nr_pages, &onlined_pages,
KAMEZAWA Hiroyuki75884fb2007-10-16 01:26:10 -07001138 online_pages_range);
Geoff Levandfd8a4222008-05-14 16:05:50 -07001139 if (ret) {
Wen Congyang6dcd73d2012-12-11 16:01:01 -08001140 if (need_zonelists_rebuild)
1141 zone_pcp_reset(zone);
Haicheng Li4eaf3f62010-05-24 14:32:52 -07001142 mutex_unlock(&zonelists_mutex);
Chen Yuconge33e33b2016-03-17 14:19:35 -07001143 goto failed_addition;
Geoff Levandfd8a4222008-05-14 16:05:50 -07001144 }
1145
Dave Hansen3947be12005-10-29 18:16:54 -07001146 zone->present_pages += onlined_pages;
Cody P Schaferaa472282013-07-03 15:02:10 -07001147
1148 pgdat_resize_lock(zone->zone_pgdat, &flags);
Yasunori Gotof2937be2006-03-09 17:33:51 -08001149 zone->zone_pgdat->node_present_pages += onlined_pages;
Cody P Schaferaa472282013-07-03 15:02:10 -07001150 pgdat_resize_unlock(zone->zone_pgdat, &flags);
1151
Jiang Liu08dff7b2012-07-31 16:43:30 -07001152 if (onlined_pages) {
Vlastimil Babkae888ca32016-03-17 14:18:12 -07001153 node_states_set_node(nid, &arg);
Jiang Liu08dff7b2012-07-31 16:43:30 -07001154 if (need_zonelists_rebuild)
Wen Congyang6dcd73d2012-12-11 16:01:01 -08001155 build_all_zonelists(NULL, NULL);
Jiang Liu08dff7b2012-07-31 16:43:30 -07001156 else
1157 zone_pcp_update(zone);
1158 }
Dave Hansen3947be12005-10-29 18:16:54 -07001159
Haicheng Li4eaf3f62010-05-24 14:32:52 -07001160 mutex_unlock(&zonelists_mutex);
KOSAKI Motohiro1b79acc2011-05-24 17:11:32 -07001161
1162 init_per_zone_wmark_min();
1163
Vlastimil Babka698b1b32016-03-17 14:18:08 -07001164 if (onlined_pages) {
Vlastimil Babkae888ca32016-03-17 14:18:12 -07001165 kswapd_run(nid);
Vlastimil Babka698b1b32016-03-17 14:18:08 -07001166 kcompactd_run(nid);
1167 }
Dave Hansen61b13992005-10-29 18:16:56 -07001168
Haicheng Li1f522502010-05-24 14:32:51 -07001169 vm_total_pages = nr_free_pagecache_pages();
Kent Liu2f7f24e2008-07-23 21:28:18 -07001170
Chandra Seetharaman2d1d43f2006-09-29 02:01:25 -07001171 writeback_set_ratelimit();
Yasunori Goto7b78d332007-10-21 16:41:36 -07001172
1173 if (onlined_pages)
1174 memory_notify(MEM_ONLINE, &arg);
David Rientjes30467e02015-04-14 15:45:11 -07001175 return 0;
Chen Yuconge33e33b2016-03-17 14:19:35 -07001176
1177failed_addition:
1178 pr_debug("online_pages [mem %#010llx-%#010llx] failed\n",
1179 (unsigned long long) pfn << PAGE_SHIFT,
1180 (((unsigned long long) pfn + nr_pages) << PAGE_SHIFT) - 1);
1181 memory_notify(MEM_CANCEL_ONLINE, &arg);
1182 return ret;
Dave Hansen3947be12005-10-29 18:16:54 -07001183}
Keith Mannthey53947022006-09-30 23:27:08 -07001184#endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
Yasunori Gotobc02af92006-06-27 02:53:30 -07001185
Tang Chen0bd85422014-11-13 15:19:41 -08001186static void reset_node_present_pages(pg_data_t *pgdat)
1187{
1188 struct zone *z;
1189
1190 for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++)
1191 z->present_pages = 0;
1192
1193 pgdat->node_present_pages = 0;
1194}
1195
Hidetoshi Setoe1319332009-11-17 14:06:18 -08001196/* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
1197static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start)
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001198{
1199 struct pglist_data *pgdat;
1200 unsigned long zones_size[MAX_NR_ZONES] = {0};
1201 unsigned long zholes_size[MAX_NR_ZONES] = {0};
Fabian Frederickc8e861a2014-06-04 16:07:51 -07001202 unsigned long start_pfn = PFN_DOWN(start);
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001203
Tang Chena1e565a2013-02-22 16:33:18 -08001204 pgdat = NODE_DATA(nid);
1205 if (!pgdat) {
1206 pgdat = arch_alloc_nodedata(nid);
1207 if (!pgdat)
1208 return NULL;
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001209
Tang Chena1e565a2013-02-22 16:33:18 -08001210 arch_refresh_nodedata(nid, pgdat);
Gu Zhengb0dc3a32015-03-25 15:55:20 -07001211 } else {
Mel Gorman38087d92016-07-28 15:45:49 -07001212 /* Reset the nr_zones, order and classzone_idx before reuse */
Gu Zhengb0dc3a32015-03-25 15:55:20 -07001213 pgdat->nr_zones = 0;
Mel Gorman38087d92016-07-28 15:45:49 -07001214 pgdat->kswapd_order = 0;
1215 pgdat->kswapd_classzone_idx = 0;
Tang Chena1e565a2013-02-22 16:33:18 -08001216 }
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001217
1218 /* we can use NODE_DATA(nid) from here */
1219
1220 /* init node's zones as empty zones, we don't have any present pages.*/
Johannes Weiner9109fb72008-07-23 21:27:20 -07001221 free_area_init_node(nid, zones_size, start_pfn, zholes_size);
Reza Arbab58301692016-08-11 15:33:12 -07001222 pgdat->per_cpu_nodestats = alloc_percpu(struct per_cpu_nodestat);
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001223
KAMEZAWA Hiroyuki959ecc42011-06-15 15:08:38 -07001224 /*
1225 * The node we allocated has no zone fallback lists. For avoiding
1226 * to access not-initialized zonelist, build here.
1227 */
David Rientjesf957db42011-06-22 18:13:04 -07001228 mutex_lock(&zonelists_mutex);
Jiang Liu9adb62a2012-07-31 16:43:28 -07001229 build_all_zonelists(pgdat, NULL);
David Rientjesf957db42011-06-22 18:13:04 -07001230 mutex_unlock(&zonelists_mutex);
KAMEZAWA Hiroyuki959ecc42011-06-15 15:08:38 -07001231
Tang Chenf784a3f2014-11-13 15:19:39 -08001232 /*
1233 * zone->managed_pages is set to an approximate value in
1234 * free_area_init_core(), which will cause
1235 * /sys/device/system/node/nodeX/meminfo has wrong data.
1236 * So reset it to 0 before any memory is onlined.
1237 */
1238 reset_node_managed_pages(pgdat);
1239
Tang Chen0bd85422014-11-13 15:19:41 -08001240 /*
1241 * When memory is hot-added, all the memory is in offline state. So
1242 * clear all zones' present_pages because they will be updated in
1243 * online_pages() and offline_pages().
1244 */
1245 reset_node_present_pages(pgdat);
1246
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001247 return pgdat;
1248}
1249
1250static void rollback_node_hotadd(int nid, pg_data_t *pgdat)
1251{
1252 arch_refresh_nodedata(nid, NULL);
Reza Arbab58301692016-08-11 15:33:12 -07001253 free_percpu(pgdat->per_cpu_nodestats);
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001254 arch_free_nodedata(pgdat);
1255 return;
1256}
1257
KAMEZAWA Hiroyuki0a547032006-06-27 02:53:35 -07001258
Toshi Kani01b0f192013-11-12 15:07:25 -08001259/**
1260 * try_online_node - online a node if offlined
1261 *
minskey guocf234222010-05-24 14:32:41 -07001262 * called by cpu_up() to online a node without onlined memory.
1263 */
Toshi Kani01b0f192013-11-12 15:07:25 -08001264int try_online_node(int nid)
minskey guocf234222010-05-24 14:32:41 -07001265{
1266 pg_data_t *pgdat;
1267 int ret;
1268
Toshi Kani01b0f192013-11-12 15:07:25 -08001269 if (node_online(nid))
1270 return 0;
1271
Vladimir Davydovbfc8c902014-06-04 16:07:18 -07001272 mem_hotplug_begin();
minskey guocf234222010-05-24 14:32:41 -07001273 pgdat = hotadd_new_pgdat(nid, 0);
David Rientjes7553e8f2011-06-22 18:13:01 -07001274 if (!pgdat) {
Toshi Kani01b0f192013-11-12 15:07:25 -08001275 pr_err("Cannot online node %d due to NULL pgdat\n", nid);
minskey guocf234222010-05-24 14:32:41 -07001276 ret = -ENOMEM;
1277 goto out;
1278 }
1279 node_set_online(nid);
1280 ret = register_one_node(nid);
1281 BUG_ON(ret);
1282
Toshi Kani01b0f192013-11-12 15:07:25 -08001283 if (pgdat->node_zonelists->_zonerefs->zone == NULL) {
1284 mutex_lock(&zonelists_mutex);
1285 build_all_zonelists(NULL, NULL);
1286 mutex_unlock(&zonelists_mutex);
1287 }
1288
minskey guocf234222010-05-24 14:32:41 -07001289out:
Vladimir Davydovbfc8c902014-06-04 16:07:18 -07001290 mem_hotplug_done();
minskey guocf234222010-05-24 14:32:41 -07001291 return ret;
1292}
1293
Toshi Kani27356f52013-09-11 14:21:49 -07001294static int check_hotplug_memory_range(u64 start, u64 size)
1295{
Fabian Frederickc8e861a2014-06-04 16:07:51 -07001296 u64 start_pfn = PFN_DOWN(start);
Toshi Kani27356f52013-09-11 14:21:49 -07001297 u64 nr_pages = size >> PAGE_SHIFT;
1298
1299 /* Memory range must be aligned with section */
1300 if ((start_pfn & ~PAGE_SECTION_MASK) ||
1301 (nr_pages % PAGES_PER_SECTION) || (!nr_pages)) {
1302 pr_err("Section-unaligned hotplug range: start 0x%llx, size 0x%llx\n",
1303 (unsigned long long)start,
1304 (unsigned long long)size);
1305 return -EINVAL;
1306 }
1307
1308 return 0;
1309}
1310
Wang Nan63264402014-08-06 16:07:36 -07001311/*
1312 * If movable zone has already been setup, newly added memory should be check.
1313 * If its address is higher than movable zone, it should be added as movable.
1314 * Without this check, movable zone may overlap with other zone.
1315 */
1316static int should_add_memory_movable(int nid, u64 start, u64 size)
1317{
1318 unsigned long start_pfn = start >> PAGE_SHIFT;
1319 pg_data_t *pgdat = NODE_DATA(nid);
1320 struct zone *movable_zone = pgdat->node_zones + ZONE_MOVABLE;
1321
1322 if (zone_is_empty(movable_zone))
1323 return 0;
1324
1325 if (movable_zone->zone_start_pfn <= start_pfn)
1326 return 1;
1327
1328 return 0;
1329}
1330
Dan Williams033fbae2015-08-09 15:29:06 -04001331int zone_for_memory(int nid, u64 start, u64 size, int zone_default,
1332 bool for_device)
Wang Nan63264402014-08-06 16:07:36 -07001333{
Dan Williams033fbae2015-08-09 15:29:06 -04001334#ifdef CONFIG_ZONE_DEVICE
1335 if (for_device)
1336 return ZONE_DEVICE;
1337#endif
Wang Nan63264402014-08-06 16:07:36 -07001338 if (should_add_memory_movable(nid, start, size))
1339 return ZONE_MOVABLE;
1340
1341 return zone_default;
1342}
1343
Vitaly Kuznetsov31bc3852016-03-15 14:56:48 -07001344static int online_memory_block(struct memory_block *mem, void *arg)
1345{
1346 return memory_block_change_state(mem, MEM_ONLINE, MEM_OFFLINE);
1347}
1348
Al Viro31168482008-11-22 17:33:24 +00001349/* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
Vitaly Kuznetsov31bc3852016-03-15 14:56:48 -07001350int __ref add_memory_resource(int nid, struct resource *res, bool online)
Yasunori Gotobc02af92006-06-27 02:53:30 -07001351{
David Vrabel62cedb92015-06-25 16:35:49 +01001352 u64 start, size;
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001353 pg_data_t *pgdat = NULL;
Tang Chena1e565a2013-02-22 16:33:18 -08001354 bool new_pgdat;
1355 bool new_node;
Yasunori Gotobc02af92006-06-27 02:53:30 -07001356 int ret;
1357
David Vrabel62cedb92015-06-25 16:35:49 +01001358 start = res->start;
1359 size = resource_size(res);
1360
Toshi Kani27356f52013-09-11 14:21:49 -07001361 ret = check_hotplug_memory_range(start, size);
1362 if (ret)
1363 return ret;
1364
Tang Chena1e565a2013-02-22 16:33:18 -08001365 { /* Stupid hack to suppress address-never-null warning */
1366 void *p = NODE_DATA(nid);
1367 new_pgdat = !p;
1368 }
Nathan Zimmerac13c462014-01-23 15:53:26 -08001369
Vladimir Davydovbfc8c902014-06-04 16:07:18 -07001370 mem_hotplug_begin();
Nathan Zimmerac13c462014-01-23 15:53:26 -08001371
Tang Chen7f36e3e2015-09-04 15:42:32 -07001372 /*
1373 * Add new range to memblock so that when hotadd_new_pgdat() is called
1374 * to allocate new pgdat, get_pfn_range_for_nid() will be able to find
1375 * this new range and calculate total pages correctly. The range will
1376 * be removed at hot-remove time.
1377 */
1378 memblock_add_node(start, size, nid);
1379
Tang Chena1e565a2013-02-22 16:33:18 -08001380 new_node = !node_online(nid);
1381 if (new_node) {
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001382 pgdat = hotadd_new_pgdat(nid, start);
Andi Kleen6ad696d2009-11-17 14:06:22 -08001383 ret = -ENOMEM;
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001384 if (!pgdat)
Wen Congyang41b9e2d2012-07-11 14:02:31 -07001385 goto error;
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001386 }
1387
Yasunori Gotobc02af92006-06-27 02:53:30 -07001388 /* call arch's memory hotadd */
Dan Williams033fbae2015-08-09 15:29:06 -04001389 ret = arch_add_memory(nid, start, size, false);
Yasunori Gotobc02af92006-06-27 02:53:30 -07001390
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001391 if (ret < 0)
1392 goto error;
1393
Yasunori Goto0fc44152006-06-27 02:53:38 -07001394 /* we online node here. we can't roll back from here. */
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001395 node_set_online(nid);
1396
Tang Chena1e565a2013-02-22 16:33:18 -08001397 if (new_node) {
Yasunori Goto0fc44152006-06-27 02:53:38 -07001398 ret = register_one_node(nid);
1399 /*
1400 * If sysfs file of new node can't create, cpu on the node
1401 * can't be hot-added. There is no rollback way now.
1402 * So, check by BUG_ON() to catch it reluctantly..
1403 */
1404 BUG_ON(ret);
1405 }
1406
akpm@linux-foundation.orgd96ae532010-03-05 13:41:58 -08001407 /* create new memmap entry */
1408 firmware_map_add_hotplug(start, start + size, "System RAM");
1409
Vitaly Kuznetsov31bc3852016-03-15 14:56:48 -07001410 /* online pages if requested */
1411 if (online)
1412 walk_memory_range(PFN_DOWN(start), PFN_UP(start + size - 1),
1413 NULL, online_memory_block);
1414
Andi Kleen6ad696d2009-11-17 14:06:22 -08001415 goto out;
1416
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001417error:
1418 /* rollback pgdat allocation and others */
1419 if (new_pgdat)
1420 rollback_node_hotadd(nid, pgdat);
Tang Chen7f36e3e2015-09-04 15:42:32 -07001421 memblock_remove(start, size);
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001422
Andi Kleen6ad696d2009-11-17 14:06:22 -08001423out:
Vladimir Davydovbfc8c902014-06-04 16:07:18 -07001424 mem_hotplug_done();
Yasunori Gotobc02af92006-06-27 02:53:30 -07001425 return ret;
1426}
David Vrabel62cedb92015-06-25 16:35:49 +01001427EXPORT_SYMBOL_GPL(add_memory_resource);
1428
1429int __ref add_memory(int nid, u64 start, u64 size)
1430{
1431 struct resource *res;
1432 int ret;
1433
1434 res = register_memory_resource(start, size);
Vitaly Kuznetsov6f754ba2016-01-14 15:21:55 -08001435 if (IS_ERR(res))
1436 return PTR_ERR(res);
David Vrabel62cedb92015-06-25 16:35:49 +01001437
Vitaly Kuznetsov31bc3852016-03-15 14:56:48 -07001438 ret = add_memory_resource(nid, res, memhp_auto_online);
David Vrabel62cedb92015-06-25 16:35:49 +01001439 if (ret < 0)
1440 release_memory_resource(res);
1441 return ret;
1442}
Yasunori Gotobc02af92006-06-27 02:53:30 -07001443EXPORT_SYMBOL_GPL(add_memory);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001444
1445#ifdef CONFIG_MEMORY_HOTREMOVE
1446/*
Badari Pulavarty5c755e92008-07-23 21:28:19 -07001447 * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy
1448 * set and the size of the free page is given by page_order(). Using this,
1449 * the function determines if the pageblock contains only free pages.
1450 * Due to buddy contraints, a free page at least the size of a pageblock will
1451 * be located at the start of the pageblock
1452 */
1453static inline int pageblock_free(struct page *page)
1454{
1455 return PageBuddy(page) && page_order(page) >= pageblock_order;
1456}
1457
1458/* Return the start of the next active pageblock after a given page */
1459static struct page *next_active_pageblock(struct page *page)
1460{
Badari Pulavarty5c755e92008-07-23 21:28:19 -07001461 /* Ensure the starting page is pageblock-aligned */
1462 BUG_ON(page_to_pfn(page) & (pageblock_nr_pages - 1));
1463
Badari Pulavarty5c755e92008-07-23 21:28:19 -07001464 /* If the entire pageblock is free, move to the end of free page */
KAMEZAWA Hiroyuki0dcc48c2010-09-09 16:38:01 -07001465 if (pageblock_free(page)) {
1466 int order;
1467 /* be careful. we don't have locks, page_order can be changed.*/
1468 order = page_order(page);
1469 if ((order < MAX_ORDER) && (order >= pageblock_order))
1470 return page + (1 << order);
1471 }
Badari Pulavarty5c755e92008-07-23 21:28:19 -07001472
KAMEZAWA Hiroyuki0dcc48c2010-09-09 16:38:01 -07001473 return page + pageblock_nr_pages;
Badari Pulavarty5c755e92008-07-23 21:28:19 -07001474}
1475
1476/* Checks if this range of memory is likely to be hot-removable. */
Yaowei Baic98940f2016-05-19 17:11:26 -07001477bool is_mem_section_removable(unsigned long start_pfn, unsigned long nr_pages)
Badari Pulavarty5c755e92008-07-23 21:28:19 -07001478{
Badari Pulavarty5c755e92008-07-23 21:28:19 -07001479 struct page *page = pfn_to_page(start_pfn);
1480 struct page *end_page = page + nr_pages;
1481
1482 /* Check the starting page of each pageblock within the range */
1483 for (; page < end_page; page = next_active_pageblock(page)) {
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07001484 if (!is_pageblock_removable_nolock(page))
Yaowei Baic98940f2016-05-19 17:11:26 -07001485 return false;
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07001486 cond_resched();
Badari Pulavarty5c755e92008-07-23 21:28:19 -07001487 }
1488
1489 /* All pageblocks in the memory block are likely to be hot-removable */
Yaowei Baic98940f2016-05-19 17:11:26 -07001490 return true;
Badari Pulavarty5c755e92008-07-23 21:28:19 -07001491}
1492
1493/*
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001494 * Confirm all pages in a range [start, end) is belongs to the same zone.
1495 */
Zhang Zhened2f2402014-10-09 15:26:31 -07001496int test_pages_in_a_zone(unsigned long start_pfn, unsigned long end_pfn)
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001497{
Andrew Banman5f0f2882015-12-29 14:54:25 -08001498 unsigned long pfn, sec_end_pfn;
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001499 struct zone *zone = NULL;
1500 struct page *page;
1501 int i;
Andrew Banman5f0f2882015-12-29 14:54:25 -08001502 for (pfn = start_pfn, sec_end_pfn = SECTION_ALIGN_UP(start_pfn);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001503 pfn < end_pfn;
Andrew Banman5f0f2882015-12-29 14:54:25 -08001504 pfn = sec_end_pfn + 1, sec_end_pfn += PAGES_PER_SECTION) {
1505 /* Make sure the memory section is present first */
1506 if (!present_section_nr(pfn_to_section_nr(pfn)))
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001507 continue;
Andrew Banman5f0f2882015-12-29 14:54:25 -08001508 for (; pfn < sec_end_pfn && pfn < end_pfn;
1509 pfn += MAX_ORDER_NR_PAGES) {
1510 i = 0;
1511 /* This is just a CONFIG_HOLES_IN_ZONE check.*/
1512 while ((i < MAX_ORDER_NR_PAGES) &&
1513 !pfn_valid_within(pfn + i))
1514 i++;
1515 if (i == MAX_ORDER_NR_PAGES)
1516 continue;
1517 page = pfn_to_page(pfn + i);
1518 if (zone && page_zone(page) != zone)
1519 return 0;
1520 zone = page_zone(page);
1521 }
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001522 }
1523 return 1;
1524}
1525
1526/*
Naoya Horiguchic8721bb2013-09-11 14:22:09 -07001527 * Scan pfn range [start,end) to find movable/migratable pages (LRU pages
1528 * and hugepages). We scan pfn because it's much easier than scanning over
1529 * linked list. This function returns the pfn of the first found movable
1530 * page if it's found, otherwise 0.
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001531 */
Naoya Horiguchic8721bb2013-09-11 14:22:09 -07001532static unsigned long scan_movable_pages(unsigned long start, unsigned long end)
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001533{
1534 unsigned long pfn;
1535 struct page *page;
1536 for (pfn = start; pfn < end; pfn++) {
1537 if (pfn_valid(pfn)) {
1538 page = pfn_to_page(pfn);
1539 if (PageLRU(page))
1540 return pfn;
Naoya Horiguchic8721bb2013-09-11 14:22:09 -07001541 if (PageHuge(page)) {
Naoya Horiguchi7e1f0492015-04-15 16:14:41 -07001542 if (page_huge_active(page))
Naoya Horiguchic8721bb2013-09-11 14:22:09 -07001543 return pfn;
1544 else
1545 pfn = round_up(pfn + 1,
1546 1 << compound_order(page)) - 1;
1547 }
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001548 }
1549 }
1550 return 0;
1551}
1552
Xishi Qiu394e31d2016-07-28 15:48:53 -07001553static struct page *new_node_page(struct page *page, unsigned long private,
1554 int **result)
1555{
1556 gfp_t gfp_mask = GFP_USER | __GFP_MOVABLE;
1557 int nid = page_to_nid(page);
Li Zhong231e97e2016-09-28 15:22:38 -07001558 nodemask_t nmask = node_states[N_MEMORY];
1559 struct page *new_page = NULL;
Xishi Qiu394e31d2016-07-28 15:48:53 -07001560
1561 /*
1562 * TODO: allocate a destination hugepage from a nearest neighbor node,
1563 * accordance with memory policy of the user process if possible. For
1564 * now as a simple work-around, we use the next node for destination.
1565 */
1566 if (PageHuge(page))
1567 return alloc_huge_page_node(page_hstate(compound_head(page)),
1568 next_node_in(nid, nmask));
1569
Li Zhong231e97e2016-09-28 15:22:38 -07001570 node_clear(nid, nmask);
Li Zhong9bb627b2016-09-19 14:43:52 -07001571
Xishi Qiu394e31d2016-07-28 15:48:53 -07001572 if (PageHighMem(page)
1573 || (zone_idx(page_zone(page)) == ZONE_MOVABLE))
1574 gfp_mask |= __GFP_HIGHMEM;
1575
Li Zhong231e97e2016-09-28 15:22:38 -07001576 if (!nodes_empty(nmask))
1577 new_page = __alloc_pages_nodemask(gfp_mask, 0,
Xishi Qiu394e31d2016-07-28 15:48:53 -07001578 node_zonelist(nid, gfp_mask), &nmask);
1579 if (!new_page)
1580 new_page = __alloc_pages(gfp_mask, 0,
1581 node_zonelist(nid, gfp_mask));
1582
1583 return new_page;
1584}
1585
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001586#define NR_OFFLINE_AT_ONCE_PAGES (256)
1587static int
1588do_migrate_range(unsigned long start_pfn, unsigned long end_pfn)
1589{
1590 unsigned long pfn;
1591 struct page *page;
1592 int move_pages = NR_OFFLINE_AT_ONCE_PAGES;
1593 int not_managed = 0;
1594 int ret = 0;
1595 LIST_HEAD(source);
1596
1597 for (pfn = start_pfn; pfn < end_pfn && move_pages > 0; pfn++) {
1598 if (!pfn_valid(pfn))
1599 continue;
1600 page = pfn_to_page(pfn);
Naoya Horiguchic8721bb2013-09-11 14:22:09 -07001601
1602 if (PageHuge(page)) {
1603 struct page *head = compound_head(page);
1604 pfn = page_to_pfn(head) + (1<<compound_order(head)) - 1;
1605 if (compound_order(head) > PFN_SECTION_SHIFT) {
1606 ret = -EBUSY;
1607 break;
1608 }
1609 if (isolate_huge_page(page, &source))
1610 move_pages -= 1 << compound_order(head);
1611 continue;
1612 }
1613
Konstantin Khlebnikov700c2a42011-05-24 17:12:19 -07001614 if (!get_page_unless_zero(page))
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001615 continue;
1616 /*
1617 * We can skip free pages. And we can only deal with pages on
1618 * LRU.
1619 */
Nick Piggin62695a82008-10-18 20:26:09 -07001620 ret = isolate_lru_page(page);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001621 if (!ret) { /* Success */
Konstantin Khlebnikov700c2a42011-05-24 17:12:19 -07001622 put_page(page);
Nick Piggin62695a82008-10-18 20:26:09 -07001623 list_add_tail(&page->lru, &source);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001624 move_pages--;
Mel Gorman599d0c92016-07-28 15:45:31 -07001625 inc_node_page_state(page, NR_ISOLATED_ANON +
KOSAKI Motohiro6d9c2852009-12-14 17:58:11 -08001626 page_is_file_cache(page));
1627
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001628 } else {
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001629#ifdef CONFIG_DEBUG_VM
Chen Yuconge33e33b2016-03-17 14:19:35 -07001630 pr_alert("removing pfn %lx from LRU failed\n", pfn);
Dave Hansenf0b791a2014-01-23 15:52:49 -08001631 dump_page(page, "failed to remove from LRU");
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001632#endif
Konstantin Khlebnikov700c2a42011-05-24 17:12:19 -07001633 put_page(page);
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001634 /* Because we don't have big zone->lock. we should
Bob Liu809c4442010-10-26 14:22:10 -07001635 check this again here. */
1636 if (page_count(page)) {
1637 not_managed++;
Bob Liuf3ab2632010-10-26 14:22:10 -07001638 ret = -EBUSY;
Bob Liu809c4442010-10-26 14:22:10 -07001639 break;
1640 }
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001641 }
1642 }
Bob Liuf3ab2632010-10-26 14:22:10 -07001643 if (!list_empty(&source)) {
1644 if (not_managed) {
Naoya Horiguchic8721bb2013-09-11 14:22:09 -07001645 putback_movable_pages(&source);
Bob Liuf3ab2632010-10-26 14:22:10 -07001646 goto out;
1647 }
Minchan Kim74c08f92012-10-08 16:32:54 -07001648
Xishi Qiu394e31d2016-07-28 15:48:53 -07001649 /* Allocate a new page from the nearest neighbor node */
1650 ret = migrate_pages(&source, new_node_page, NULL, 0,
Hugh Dickins9c620e22013-02-22 16:35:14 -08001651 MIGRATE_SYNC, MR_MEMORY_HOTPLUG);
Bob Liuf3ab2632010-10-26 14:22:10 -07001652 if (ret)
Naoya Horiguchic8721bb2013-09-11 14:22:09 -07001653 putback_movable_pages(&source);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001654 }
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001655out:
1656 return ret;
1657}
1658
1659/*
1660 * remove from free_area[] and mark all as Reserved.
1661 */
1662static int
1663offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages,
1664 void *data)
1665{
1666 __offline_isolated_pages(start, start + nr_pages);
1667 return 0;
1668}
1669
1670static void
1671offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
1672{
KAMEZAWA Hiroyuki908eedc2009-09-22 16:45:46 -07001673 walk_system_ram_range(start_pfn, end_pfn - start_pfn, NULL,
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001674 offline_isolated_pages_cb);
1675}
1676
1677/*
1678 * Check all pages in range, recoreded as memory resource, are isolated.
1679 */
1680static int
1681check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages,
1682 void *data)
1683{
1684 int ret;
1685 long offlined = *(long *)data;
Wen Congyangb023f462012-12-11 16:00:45 -08001686 ret = test_pages_isolated(start_pfn, start_pfn + nr_pages, true);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001687 offlined = nr_pages;
1688 if (!ret)
1689 *(long *)data += offlined;
1690 return ret;
1691}
1692
1693static long
1694check_pages_isolated(unsigned long start_pfn, unsigned long end_pfn)
1695{
1696 long offlined = 0;
1697 int ret;
1698
KAMEZAWA Hiroyuki908eedc2009-09-22 16:45:46 -07001699 ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn, &offlined,
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001700 check_pages_isolated_cb);
1701 if (ret < 0)
1702 offlined = (long)ret;
1703 return offlined;
1704}
1705
Lai Jiangshan09285af2012-12-12 13:52:04 -08001706#ifdef CONFIG_MOVABLE_NODE
Tang Chen79a4dce2012-12-18 14:23:24 -08001707/*
1708 * When CONFIG_MOVABLE_NODE, we permit offlining of a node which doesn't have
1709 * normal memory.
1710 */
Lai Jiangshan09285af2012-12-12 13:52:04 -08001711static bool can_offline_normal(struct zone *zone, unsigned long nr_pages)
1712{
1713 return true;
1714}
Tang Chen79a4dce2012-12-18 14:23:24 -08001715#else /* CONFIG_MOVABLE_NODE */
Lai Jiangshan74d42d82012-12-11 16:03:23 -08001716/* ensure the node has NORMAL memory if it is still online */
1717static bool can_offline_normal(struct zone *zone, unsigned long nr_pages)
1718{
1719 struct pglist_data *pgdat = zone->zone_pgdat;
1720 unsigned long present_pages = 0;
1721 enum zone_type zt;
1722
1723 for (zt = 0; zt <= ZONE_NORMAL; zt++)
1724 present_pages += pgdat->node_zones[zt].present_pages;
1725
1726 if (present_pages > nr_pages)
1727 return true;
1728
1729 present_pages = 0;
1730 for (; zt <= ZONE_MOVABLE; zt++)
1731 present_pages += pgdat->node_zones[zt].present_pages;
1732
1733 /*
1734 * we can't offline the last normal memory until all
1735 * higher memory is offlined.
1736 */
1737 return present_pages == 0;
1738}
Tang Chen79a4dce2012-12-18 14:23:24 -08001739#endif /* CONFIG_MOVABLE_NODE */
Lai Jiangshan74d42d82012-12-11 16:03:23 -08001740
Tang Chenc5320922013-11-12 15:08:10 -08001741static int __init cmdline_parse_movable_node(char *p)
1742{
1743#ifdef CONFIG_MOVABLE_NODE
1744 /*
1745 * Memory used by the kernel cannot be hot-removed because Linux
1746 * cannot migrate the kernel pages. When memory hotplug is
1747 * enabled, we should prevent memblock from allocating memory
1748 * for the kernel.
1749 *
1750 * ACPI SRAT records all hotpluggable memory ranges. But before
1751 * SRAT is parsed, we don't know about it.
1752 *
1753 * The kernel image is loaded into memory at very early time. We
1754 * cannot prevent this anyway. So on NUMA system, we set any
1755 * node the kernel resides in as un-hotpluggable.
1756 *
1757 * Since on modern servers, one node could have double-digit
1758 * gigabytes memory, we can assume the memory around the kernel
1759 * image is also un-hotpluggable. So before SRAT is parsed, just
1760 * allocate memory near the kernel image to try the best to keep
1761 * the kernel away from hotpluggable memory.
1762 */
1763 memblock_set_bottom_up(true);
Tang Chen55ac5902014-01-21 15:49:35 -08001764 movable_node_enabled = true;
Tang Chenc5320922013-11-12 15:08:10 -08001765#else
1766 pr_warn("movable_node option not supported\n");
1767#endif
1768 return 0;
1769}
1770early_param("movable_node", cmdline_parse_movable_node);
1771
Lai Jiangshand9713672012-12-11 16:01:03 -08001772/* check which state of node_states will be changed when offline memory */
1773static void node_states_check_changes_offline(unsigned long nr_pages,
1774 struct zone *zone, struct memory_notify *arg)
1775{
1776 struct pglist_data *pgdat = zone->zone_pgdat;
1777 unsigned long present_pages = 0;
1778 enum zone_type zt, zone_last = ZONE_NORMAL;
1779
1780 /*
Lai Jiangshan6715ddf2012-12-12 13:51:49 -08001781 * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY]
1782 * contains nodes which have zones of 0...ZONE_NORMAL,
1783 * set zone_last to ZONE_NORMAL.
Lai Jiangshand9713672012-12-11 16:01:03 -08001784 *
Lai Jiangshan6715ddf2012-12-12 13:51:49 -08001785 * If we don't have HIGHMEM nor movable node,
1786 * node_states[N_NORMAL_MEMORY] contains nodes which have zones of
1787 * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE.
Lai Jiangshand9713672012-12-11 16:01:03 -08001788 */
Lai Jiangshan6715ddf2012-12-12 13:51:49 -08001789 if (N_MEMORY == N_NORMAL_MEMORY)
Lai Jiangshand9713672012-12-11 16:01:03 -08001790 zone_last = ZONE_MOVABLE;
1791
1792 /*
1793 * check whether node_states[N_NORMAL_MEMORY] will be changed.
1794 * If the memory to be offline is in a zone of 0...zone_last,
1795 * and it is the last present memory, 0...zone_last will
1796 * become empty after offline , thus we can determind we will
1797 * need to clear the node from node_states[N_NORMAL_MEMORY].
1798 */
1799 for (zt = 0; zt <= zone_last; zt++)
1800 present_pages += pgdat->node_zones[zt].present_pages;
1801 if (zone_idx(zone) <= zone_last && nr_pages >= present_pages)
1802 arg->status_change_nid_normal = zone_to_nid(zone);
1803 else
1804 arg->status_change_nid_normal = -1;
1805
Lai Jiangshan6715ddf2012-12-12 13:51:49 -08001806#ifdef CONFIG_HIGHMEM
1807 /*
1808 * If we have movable node, node_states[N_HIGH_MEMORY]
1809 * contains nodes which have zones of 0...ZONE_HIGHMEM,
1810 * set zone_last to ZONE_HIGHMEM.
1811 *
1812 * If we don't have movable node, node_states[N_NORMAL_MEMORY]
1813 * contains nodes which have zones of 0...ZONE_MOVABLE,
1814 * set zone_last to ZONE_MOVABLE.
1815 */
1816 zone_last = ZONE_HIGHMEM;
1817 if (N_MEMORY == N_HIGH_MEMORY)
1818 zone_last = ZONE_MOVABLE;
1819
1820 for (; zt <= zone_last; zt++)
1821 present_pages += pgdat->node_zones[zt].present_pages;
1822 if (zone_idx(zone) <= zone_last && nr_pages >= present_pages)
1823 arg->status_change_nid_high = zone_to_nid(zone);
1824 else
1825 arg->status_change_nid_high = -1;
1826#else
1827 arg->status_change_nid_high = arg->status_change_nid_normal;
1828#endif
1829
Lai Jiangshand9713672012-12-11 16:01:03 -08001830 /*
1831 * node_states[N_HIGH_MEMORY] contains nodes which have 0...ZONE_MOVABLE
1832 */
1833 zone_last = ZONE_MOVABLE;
1834
1835 /*
1836 * check whether node_states[N_HIGH_MEMORY] will be changed
1837 * If we try to offline the last present @nr_pages from the node,
1838 * we can determind we will need to clear the node from
1839 * node_states[N_HIGH_MEMORY].
1840 */
1841 for (; zt <= zone_last; zt++)
1842 present_pages += pgdat->node_zones[zt].present_pages;
1843 if (nr_pages >= present_pages)
1844 arg->status_change_nid = zone_to_nid(zone);
1845 else
1846 arg->status_change_nid = -1;
1847}
1848
1849static void node_states_clear_node(int node, struct memory_notify *arg)
1850{
1851 if (arg->status_change_nid_normal >= 0)
1852 node_clear_state(node, N_NORMAL_MEMORY);
1853
Lai Jiangshan6715ddf2012-12-12 13:51:49 -08001854 if ((N_MEMORY != N_NORMAL_MEMORY) &&
1855 (arg->status_change_nid_high >= 0))
Lai Jiangshand9713672012-12-11 16:01:03 -08001856 node_clear_state(node, N_HIGH_MEMORY);
Lai Jiangshan6715ddf2012-12-12 13:51:49 -08001857
1858 if ((N_MEMORY != N_HIGH_MEMORY) &&
1859 (arg->status_change_nid >= 0))
1860 node_clear_state(node, N_MEMORY);
Lai Jiangshand9713672012-12-11 16:01:03 -08001861}
1862
Wen Congyanga16cee12012-10-08 16:33:58 -07001863static int __ref __offline_pages(unsigned long start_pfn,
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001864 unsigned long end_pfn, unsigned long timeout)
1865{
1866 unsigned long pfn, nr_pages, expire;
1867 long offlined_pages;
Yasunori Goto7b78d332007-10-21 16:41:36 -07001868 int ret, drain, retry_max, node;
Cody P Schaferd7029092013-07-03 15:02:11 -07001869 unsigned long flags;
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001870 struct zone *zone;
Yasunori Goto7b78d332007-10-21 16:41:36 -07001871 struct memory_notify arg;
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001872
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001873 /* at least, alignment against pageblock is necessary */
1874 if (!IS_ALIGNED(start_pfn, pageblock_nr_pages))
1875 return -EINVAL;
1876 if (!IS_ALIGNED(end_pfn, pageblock_nr_pages))
1877 return -EINVAL;
1878 /* This makes hotplug much easier...and readable.
1879 we assume this for now. .*/
1880 if (!test_pages_in_a_zone(start_pfn, end_pfn))
1881 return -EINVAL;
Yasunori Goto7b78d332007-10-21 16:41:36 -07001882
1883 zone = page_zone(pfn_to_page(start_pfn));
1884 node = zone_to_nid(zone);
1885 nr_pages = end_pfn - start_pfn;
1886
Lai Jiangshan74d42d82012-12-11 16:03:23 -08001887 if (zone_idx(zone) <= ZONE_NORMAL && !can_offline_normal(zone, nr_pages))
David Rientjes30467e02015-04-14 15:45:11 -07001888 return -EINVAL;
Lai Jiangshan74d42d82012-12-11 16:03:23 -08001889
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001890 /* set above range as isolated */
Wen Congyangb023f462012-12-11 16:00:45 -08001891 ret = start_isolate_page_range(start_pfn, end_pfn,
1892 MIGRATE_MOVABLE, true);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001893 if (ret)
David Rientjes30467e02015-04-14 15:45:11 -07001894 return ret;
Yasunori Goto7b78d332007-10-21 16:41:36 -07001895
1896 arg.start_pfn = start_pfn;
1897 arg.nr_pages = nr_pages;
Lai Jiangshand9713672012-12-11 16:01:03 -08001898 node_states_check_changes_offline(nr_pages, zone, &arg);
Yasunori Goto7b78d332007-10-21 16:41:36 -07001899
1900 ret = memory_notify(MEM_GOING_OFFLINE, &arg);
1901 ret = notifier_to_errno(ret);
1902 if (ret)
1903 goto failed_removal;
1904
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001905 pfn = start_pfn;
1906 expire = jiffies + timeout;
1907 drain = 0;
1908 retry_max = 5;
1909repeat:
1910 /* start memory hot removal */
1911 ret = -EAGAIN;
1912 if (time_after(jiffies, expire))
1913 goto failed_removal;
1914 ret = -EINTR;
1915 if (signal_pending(current))
1916 goto failed_removal;
1917 ret = 0;
1918 if (drain) {
1919 lru_add_drain_all();
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001920 cond_resched();
Vlastimil Babkac0554322014-12-10 15:43:10 -08001921 drain_all_pages(zone);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001922 }
1923
Naoya Horiguchic8721bb2013-09-11 14:22:09 -07001924 pfn = scan_movable_pages(start_pfn, end_pfn);
1925 if (pfn) { /* We have movable pages */
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001926 ret = do_migrate_range(pfn, end_pfn);
1927 if (!ret) {
1928 drain = 1;
1929 goto repeat;
1930 } else {
1931 if (ret < 0)
1932 if (--retry_max == 0)
1933 goto failed_removal;
1934 yield();
1935 drain = 1;
1936 goto repeat;
1937 }
1938 }
Adam Buchbinderb3834be2012-09-19 21:48:02 -04001939 /* drain all zone's lru pagevec, this is asynchronous... */
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001940 lru_add_drain_all();
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001941 yield();
Adam Buchbinderb3834be2012-09-19 21:48:02 -04001942 /* drain pcp pages, this is synchronous. */
Vlastimil Babkac0554322014-12-10 15:43:10 -08001943 drain_all_pages(zone);
Naoya Horiguchic8721bb2013-09-11 14:22:09 -07001944 /*
1945 * dissolve free hugepages in the memory block before doing offlining
1946 * actually in order to make hugetlbfs's object counting consistent.
1947 */
Gerald Schaefer082d5b62016-10-07 17:01:10 -07001948 ret = dissolve_free_huge_pages(start_pfn, end_pfn);
1949 if (ret)
1950 goto failed_removal;
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001951 /* check again */
1952 offlined_pages = check_pages_isolated(start_pfn, end_pfn);
1953 if (offlined_pages < 0) {
1954 ret = -EBUSY;
1955 goto failed_removal;
1956 }
Chen Yuconge33e33b2016-03-17 14:19:35 -07001957 pr_info("Offlined Pages %ld\n", offlined_pages);
Adam Buchbinderb3834be2012-09-19 21:48:02 -04001958 /* Ok, all of our target is isolated.
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001959 We cannot do rollback at this point. */
1960 offline_isolated_pages(start_pfn, end_pfn);
KAMEZAWA Hiroyukidbc0e4c2007-11-14 16:59:12 -08001961 /* reset pagetype flags and makes migrate type to be MOVABLE */
Michal Nazarewicz0815f3d2012-04-03 15:06:15 +02001962 undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001963 /* removal success */
Jiang Liu3dcc0572013-07-03 15:03:21 -07001964 adjust_managed_page_count(pfn_to_page(start_pfn), -offlined_pages);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001965 zone->present_pages -= offlined_pages;
Cody P Schaferd7029092013-07-03 15:02:11 -07001966
1967 pgdat_resize_lock(zone->zone_pgdat, &flags);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001968 zone->zone_pgdat->node_present_pages -= offlined_pages;
Cody P Schaferd7029092013-07-03 15:02:11 -07001969 pgdat_resize_unlock(zone->zone_pgdat, &flags);
Yasunori Goto7b78d332007-10-21 16:41:36 -07001970
KOSAKI Motohiro1b79acc2011-05-24 17:11:32 -07001971 init_per_zone_wmark_min();
1972
Xishi Qiu1e8537b2012-10-08 16:31:51 -07001973 if (!populated_zone(zone)) {
Jiang Liu340175b2012-07-31 16:43:32 -07001974 zone_pcp_reset(zone);
Xishi Qiu1e8537b2012-10-08 16:31:51 -07001975 mutex_lock(&zonelists_mutex);
1976 build_all_zonelists(NULL, NULL);
1977 mutex_unlock(&zonelists_mutex);
1978 } else
1979 zone_pcp_update(zone);
Jiang Liu340175b2012-07-31 16:43:32 -07001980
Lai Jiangshand9713672012-12-11 16:01:03 -08001981 node_states_clear_node(node, &arg);
Vlastimil Babka698b1b32016-03-17 14:18:08 -07001982 if (arg.status_change_nid >= 0) {
David Rientjes8fe23e02009-12-14 17:58:33 -08001983 kswapd_stop(node);
Vlastimil Babka698b1b32016-03-17 14:18:08 -07001984 kcompactd_stop(node);
1985 }
Minchan Kimbce73942009-06-16 15:32:50 -07001986
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001987 vm_total_pages = nr_free_pagecache_pages();
1988 writeback_set_ratelimit();
Yasunori Goto7b78d332007-10-21 16:41:36 -07001989
1990 memory_notify(MEM_OFFLINE, &arg);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001991 return 0;
1992
1993failed_removal:
Chen Yuconge33e33b2016-03-17 14:19:35 -07001994 pr_debug("memory offlining [mem %#010llx-%#010llx] failed\n",
1995 (unsigned long long) start_pfn << PAGE_SHIFT,
1996 ((unsigned long long) end_pfn << PAGE_SHIFT) - 1);
Yasunori Goto7b78d332007-10-21 16:41:36 -07001997 memory_notify(MEM_CANCEL_OFFLINE, &arg);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001998 /* pushback to free area */
Michal Nazarewicz0815f3d2012-04-03 15:06:15 +02001999 undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07002000 return ret;
2001}
Badari Pulavarty71088782008-10-18 20:25:58 -07002002
David Rientjes30467e02015-04-14 15:45:11 -07002003/* Must be protected by mem_hotplug_begin() */
Wen Congyanga16cee12012-10-08 16:33:58 -07002004int offline_pages(unsigned long start_pfn, unsigned long nr_pages)
2005{
2006 return __offline_pages(start_pfn, start_pfn + nr_pages, 120 * HZ);
2007}
Rafael J. Wysockie2ff3942013-05-08 00:29:49 +02002008#endif /* CONFIG_MEMORY_HOTREMOVE */
Wen Congyanga16cee12012-10-08 16:33:58 -07002009
Wen Congyangbbc76be2013-02-22 16:32:54 -08002010/**
2011 * walk_memory_range - walks through all mem sections in [start_pfn, end_pfn)
2012 * @start_pfn: start pfn of the memory range
Toshi Kanie05c4bb2013-04-29 15:06:16 -07002013 * @end_pfn: end pfn of the memory range
Wen Congyangbbc76be2013-02-22 16:32:54 -08002014 * @arg: argument passed to func
2015 * @func: callback for each memory section walked
2016 *
2017 * This function walks through all present mem sections in range
2018 * [start_pfn, end_pfn) and call func on each mem section.
2019 *
2020 * Returns the return value of func.
2021 */
Rafael J. Wysockie2ff3942013-05-08 00:29:49 +02002022int walk_memory_range(unsigned long start_pfn, unsigned long end_pfn,
Wen Congyangbbc76be2013-02-22 16:32:54 -08002023 void *arg, int (*func)(struct memory_block *, void *))
Badari Pulavarty71088782008-10-18 20:25:58 -07002024{
Wen Congyange90bdb72012-10-08 16:34:01 -07002025 struct memory_block *mem = NULL;
2026 struct mem_section *section;
Wen Congyange90bdb72012-10-08 16:34:01 -07002027 unsigned long pfn, section_nr;
2028 int ret;
Badari Pulavarty71088782008-10-18 20:25:58 -07002029
Wen Congyange90bdb72012-10-08 16:34:01 -07002030 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
2031 section_nr = pfn_to_section_nr(pfn);
2032 if (!present_section_nr(section_nr))
2033 continue;
2034
2035 section = __nr_to_section(section_nr);
2036 /* same memblock? */
2037 if (mem)
2038 if ((section_nr >= mem->start_section_nr) &&
2039 (section_nr <= mem->end_section_nr))
2040 continue;
2041
2042 mem = find_memory_block_hinted(section, mem);
2043 if (!mem)
2044 continue;
2045
Wen Congyangbbc76be2013-02-22 16:32:54 -08002046 ret = func(mem, arg);
Wen Congyange90bdb72012-10-08 16:34:01 -07002047 if (ret) {
Wen Congyangbbc76be2013-02-22 16:32:54 -08002048 kobject_put(&mem->dev.kobj);
2049 return ret;
Wen Congyange90bdb72012-10-08 16:34:01 -07002050 }
2051 }
2052
2053 if (mem)
2054 kobject_put(&mem->dev.kobj);
2055
Wen Congyangbbc76be2013-02-22 16:32:54 -08002056 return 0;
2057}
2058
Rafael J. Wysockie2ff3942013-05-08 00:29:49 +02002059#ifdef CONFIG_MEMORY_HOTREMOVE
Xishi Qiud6de9d52013-11-12 15:07:20 -08002060static int check_memblock_offlined_cb(struct memory_block *mem, void *arg)
Wen Congyangbbc76be2013-02-22 16:32:54 -08002061{
2062 int ret = !is_memblock_offlined(mem);
2063
Randy Dunlap349daa02013-04-29 15:08:49 -07002064 if (unlikely(ret)) {
2065 phys_addr_t beginpa, endpa;
2066
2067 beginpa = PFN_PHYS(section_nr_to_pfn(mem->start_section_nr));
2068 endpa = PFN_PHYS(section_nr_to_pfn(mem->end_section_nr + 1))-1;
Joe Perches756a025f02016-03-17 14:19:47 -07002069 pr_warn("removing memory fails, because memory [%pa-%pa] is onlined\n",
Randy Dunlap349daa02013-04-29 15:08:49 -07002070 &beginpa, &endpa);
2071 }
Wen Congyangbbc76be2013-02-22 16:32:54 -08002072
2073 return ret;
2074}
2075
Toshi Kani0f1cfe92013-09-11 14:21:50 -07002076static int check_cpu_on_node(pg_data_t *pgdat)
Tang Chen60a5a192013-02-22 16:33:14 -08002077{
Tang Chen60a5a192013-02-22 16:33:14 -08002078 int cpu;
2079
2080 for_each_present_cpu(cpu) {
2081 if (cpu_to_node(cpu) == pgdat->node_id)
2082 /*
2083 * the cpu on this node isn't removed, and we can't
2084 * offline this node.
2085 */
2086 return -EBUSY;
2087 }
2088
2089 return 0;
2090}
2091
Toshi Kani0f1cfe92013-09-11 14:21:50 -07002092static void unmap_cpu_on_node(pg_data_t *pgdat)
Wen Congyange13fe862013-02-22 16:33:31 -08002093{
2094#ifdef CONFIG_ACPI_NUMA
Wen Congyange13fe862013-02-22 16:33:31 -08002095 int cpu;
2096
2097 for_each_possible_cpu(cpu)
2098 if (cpu_to_node(cpu) == pgdat->node_id)
2099 numa_clear_node(cpu);
2100#endif
2101}
2102
Toshi Kani0f1cfe92013-09-11 14:21:50 -07002103static int check_and_unmap_cpu_on_node(pg_data_t *pgdat)
Wen Congyange13fe862013-02-22 16:33:31 -08002104{
Toshi Kani0f1cfe92013-09-11 14:21:50 -07002105 int ret;
Wen Congyange13fe862013-02-22 16:33:31 -08002106
Toshi Kani0f1cfe92013-09-11 14:21:50 -07002107 ret = check_cpu_on_node(pgdat);
Wen Congyange13fe862013-02-22 16:33:31 -08002108 if (ret)
2109 return ret;
2110
2111 /*
2112 * the node will be offlined when we come here, so we can clear
2113 * the cpu_to_node() now.
2114 */
2115
Toshi Kani0f1cfe92013-09-11 14:21:50 -07002116 unmap_cpu_on_node(pgdat);
Wen Congyange13fe862013-02-22 16:33:31 -08002117 return 0;
2118}
2119
Toshi Kani0f1cfe92013-09-11 14:21:50 -07002120/**
2121 * try_offline_node
2122 *
2123 * Offline a node if all memory sections and cpus of the node are removed.
2124 *
2125 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
2126 * and online/offline operations before this call.
2127 */
Wen Congyang90b30cd2013-02-22 16:33:27 -08002128void try_offline_node(int nid)
Tang Chen60a5a192013-02-22 16:33:14 -08002129{
Wen Congyangd822b862013-02-22 16:33:16 -08002130 pg_data_t *pgdat = NODE_DATA(nid);
2131 unsigned long start_pfn = pgdat->node_start_pfn;
2132 unsigned long end_pfn = start_pfn + pgdat->node_spanned_pages;
Tang Chen60a5a192013-02-22 16:33:14 -08002133 unsigned long pfn;
Wen Congyangd822b862013-02-22 16:33:16 -08002134 int i;
Tang Chen60a5a192013-02-22 16:33:14 -08002135
2136 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
2137 unsigned long section_nr = pfn_to_section_nr(pfn);
2138
2139 if (!present_section_nr(section_nr))
2140 continue;
2141
2142 if (pfn_to_nid(pfn) != nid)
2143 continue;
2144
2145 /*
2146 * some memory sections of this node are not removed, and we
2147 * can't offline node now.
2148 */
2149 return;
2150 }
2151
Toshi Kani0f1cfe92013-09-11 14:21:50 -07002152 if (check_and_unmap_cpu_on_node(pgdat))
Tang Chen60a5a192013-02-22 16:33:14 -08002153 return;
2154
2155 /*
2156 * all memory/cpu of this node are removed, we can offline this
2157 * node now.
2158 */
2159 node_set_offline(nid);
2160 unregister_one_node(nid);
Wen Congyangd822b862013-02-22 16:33:16 -08002161
Wen Congyangd822b862013-02-22 16:33:16 -08002162 /* free waittable in each zone */
2163 for (i = 0; i < MAX_NR_ZONES; i++) {
2164 struct zone *zone = pgdat->node_zones + i;
2165
Jianguo Wuca4b3f32013-03-22 15:04:50 -07002166 /*
2167 * wait_table may be allocated from boot memory,
2168 * here only free if it's allocated by vmalloc.
2169 */
Gu Zheng85bd8392015-06-10 11:14:43 -07002170 if (is_vmalloc_addr(zone->wait_table)) {
Wen Congyangd822b862013-02-22 16:33:16 -08002171 vfree(zone->wait_table);
Gu Zheng85bd8392015-06-10 11:14:43 -07002172 zone->wait_table = NULL;
2173 }
Wen Congyangd822b862013-02-22 16:33:16 -08002174 }
Tang Chen60a5a192013-02-22 16:33:14 -08002175}
Wen Congyang90b30cd2013-02-22 16:33:27 -08002176EXPORT_SYMBOL(try_offline_node);
Tang Chen60a5a192013-02-22 16:33:14 -08002177
Toshi Kani0f1cfe92013-09-11 14:21:50 -07002178/**
2179 * remove_memory
2180 *
2181 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
2182 * and online/offline operations before this call, as required by
2183 * try_offline_node().
2184 */
Rafael J. Wysocki242831e2013-05-27 12:58:46 +02002185void __ref remove_memory(int nid, u64 start, u64 size)
Wen Congyangbbc76be2013-02-22 16:32:54 -08002186{
Rafael J. Wysocki242831e2013-05-27 12:58:46 +02002187 int ret;
Wen Congyang993c1aa2013-02-22 16:32:50 -08002188
Toshi Kani27356f52013-09-11 14:21:49 -07002189 BUG_ON(check_hotplug_memory_range(start, size));
2190
Vladimir Davydovbfc8c902014-06-04 16:07:18 -07002191 mem_hotplug_begin();
Yasuaki Ishimatsu6677e3e2013-02-22 16:32:52 -08002192
2193 /*
Rafael J. Wysocki242831e2013-05-27 12:58:46 +02002194 * All memory blocks must be offlined before removing memory. Check
2195 * whether all memory blocks in question are offline and trigger a BUG()
2196 * if this is not the case.
Yasuaki Ishimatsu6677e3e2013-02-22 16:32:52 -08002197 */
Rafael J. Wysocki242831e2013-05-27 12:58:46 +02002198 ret = walk_memory_range(PFN_DOWN(start), PFN_UP(start + size - 1), NULL,
Xishi Qiud6de9d52013-11-12 15:07:20 -08002199 check_memblock_offlined_cb);
Vladimir Davydovbfc8c902014-06-04 16:07:18 -07002200 if (ret)
Rafael J. Wysocki242831e2013-05-27 12:58:46 +02002201 BUG();
Yasuaki Ishimatsu6677e3e2013-02-22 16:32:52 -08002202
Yasuaki Ishimatsu46c66c42013-02-22 16:32:56 -08002203 /* remove memmap entry */
2204 firmware_map_remove(start, start + size, "System RAM");
Xishi Qiuf9126ab2015-08-14 15:35:16 -07002205 memblock_free(start, size);
2206 memblock_remove(start, size);
Yasuaki Ishimatsu46c66c42013-02-22 16:32:56 -08002207
Wen Congyang24d335c2013-02-22 16:32:58 -08002208 arch_remove_memory(start, size);
2209
Tang Chen60a5a192013-02-22 16:33:14 -08002210 try_offline_node(nid);
2211
Vladimir Davydovbfc8c902014-06-04 16:07:18 -07002212 mem_hotplug_done();
Badari Pulavarty71088782008-10-18 20:25:58 -07002213}
Badari Pulavarty71088782008-10-18 20:25:58 -07002214EXPORT_SYMBOL_GPL(remove_memory);
Rafael J. Wysockiaba6efc2013-06-01 22:24:07 +02002215#endif /* CONFIG_MEMORY_HOTREMOVE */