blob: caf2a14c37ada40b4d690eaae18e6d72cd274d9b [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
266void register_page_bootmem_info_node(struct pglist_data *pgdat)
267{
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 */
303 if (pfn_valid(pfn) && (pfn_to_nid(pfn) == node))
304 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
Fabian Frederickf2765402014-08-06 16:04:57 -0700452static void __meminit grow_pgdat_span(struct pglist_data *pgdat, unsigned long start_pfn,
453 unsigned long end_pfn)
Heiko Carstens76cdd582008-05-14 16:05:52 -0700454{
Xishi Qiu83285c72013-11-12 15:07:19 -0800455 unsigned long old_pgdat_end_pfn = pgdat_end_pfn(pgdat);
Heiko Carstens76cdd582008-05-14 16:05:52 -0700456
Tang Chen712cd382012-12-11 16:01:07 -0800457 if (!pgdat->node_spanned_pages || start_pfn < pgdat->node_start_pfn)
Heiko Carstens76cdd582008-05-14 16:05:52 -0700458 pgdat->node_start_pfn = start_pfn;
459
460 pgdat->node_spanned_pages = max(old_pgdat_end_pfn, end_pfn) -
461 pgdat->node_start_pfn;
462}
463
Al Viro31168482008-11-22 17:33:24 +0000464static int __meminit __add_zone(struct zone *zone, unsigned long phys_start_pfn)
Dave Hansen3947be12005-10-29 18:16:54 -0700465{
466 struct pglist_data *pgdat = zone->zone_pgdat;
467 int nr_pages = PAGES_PER_SECTION;
468 int nid = pgdat->node_id;
469 int zone_type;
Mel Gormane298ff72015-08-06 15:46:51 -0700470 unsigned long flags, pfn;
Cody P Schafer64dd1b22013-02-22 16:35:31 -0800471 int ret;
Dave Hansen3947be12005-10-29 18:16:54 -0700472
473 zone_type = zone - pgdat->node_zones;
Cody P Schafer64dd1b22013-02-22 16:35:31 -0800474 ret = ensure_zone_is_initialized(zone, phys_start_pfn, nr_pages);
475 if (ret)
476 return ret;
Heiko Carstens76cdd582008-05-14 16:05:52 -0700477
Heiko Carstens76cdd582008-05-14 16:05:52 -0700478 pgdat_resize_lock(zone->zone_pgdat, &flags);
479 grow_zone_span(zone, phys_start_pfn, phys_start_pfn + nr_pages);
480 grow_pgdat_span(zone->zone_pgdat, phys_start_pfn,
481 phys_start_pfn + nr_pages);
482 pgdat_resize_unlock(zone->zone_pgdat, &flags);
Dave Hansena2f3aa022007-01-10 23:15:30 -0800483 memmap_init_zone(nr_pages, nid, zone_type,
484 phys_start_pfn, MEMMAP_HOTPLUG);
Mel Gormane298ff72015-08-06 15:46:51 -0700485
486 /* online_page_range is called later and expects pages reserved */
487 for (pfn = phys_start_pfn; pfn < phys_start_pfn + nr_pages; pfn++) {
488 if (!pfn_valid(pfn))
489 continue;
490
491 SetPageReserved(pfn_to_page(pfn));
492 }
Yasunori Goto718127c2006-06-23 02:03:10 -0700493 return 0;
Dave Hansen3947be12005-10-29 18:16:54 -0700494}
495
Gary Hadec04fc582009-01-06 14:39:14 -0800496static int __meminit __add_section(int nid, struct zone *zone,
497 unsigned long phys_start_pfn)
Dave Hansen3947be12005-10-29 18:16:54 -0700498{
Dave Hansen3947be12005-10-29 18:16:54 -0700499 int ret;
500
KAMEZAWA Hiroyukiebd15302006-08-05 12:15:06 -0700501 if (pfn_valid(phys_start_pfn))
502 return -EEXIST;
503
Zhang Yanfei85b35fe2013-11-12 15:07:42 -0800504 ret = sparse_add_one_section(zone, phys_start_pfn);
Dave Hansen3947be12005-10-29 18:16:54 -0700505
506 if (ret < 0)
507 return ret;
508
Yasunori Goto718127c2006-06-23 02:03:10 -0700509 ret = __add_zone(zone, phys_start_pfn);
510
511 if (ret < 0)
512 return ret;
513
Gary Hadec04fc582009-01-06 14:39:14 -0800514 return register_new_memory(nid, __pfn_to_section(phys_start_pfn));
Dave Hansen3947be12005-10-29 18:16:54 -0700515}
516
David Rientjes4edd7ce2013-04-29 15:08:22 -0700517/*
518 * Reasonably generic function for adding memory. It is
519 * expected that archs that support memory hotplug will
520 * call this function after deciding the zone to which to
521 * add the new pages.
522 */
523int __ref __add_pages(int nid, struct zone *zone, unsigned long phys_start_pfn,
524 unsigned long nr_pages)
525{
526 unsigned long i;
527 int err = 0;
528 int start_sec, end_sec;
Dan Williams4b94ffd2016-01-15 16:56:22 -0800529 struct vmem_altmap *altmap;
530
Joonsoo Kim7cf91a92016-03-15 14:57:51 -0700531 clear_zone_contiguous(zone);
532
David Rientjes4edd7ce2013-04-29 15:08:22 -0700533 /* during initialize mem_map, align hot-added range to section */
534 start_sec = pfn_to_section_nr(phys_start_pfn);
535 end_sec = pfn_to_section_nr(phys_start_pfn + nr_pages - 1);
536
Dan Williams4b94ffd2016-01-15 16:56:22 -0800537 altmap = to_vmem_altmap((unsigned long) pfn_to_page(phys_start_pfn));
538 if (altmap) {
539 /*
540 * Validate altmap is within bounds of the total request
541 */
542 if (altmap->base_pfn != phys_start_pfn
543 || vmem_altmap_offset(altmap) > nr_pages) {
544 pr_warn_once("memory add fail, invalid altmap\n");
Joonsoo Kim7cf91a92016-03-15 14:57:51 -0700545 err = -EINVAL;
546 goto out;
Dan Williams4b94ffd2016-01-15 16:56:22 -0800547 }
548 altmap->alloc = 0;
549 }
550
David Rientjes4edd7ce2013-04-29 15:08:22 -0700551 for (i = start_sec; i <= end_sec; i++) {
Sheng Yong19c07d52015-04-14 15:44:54 -0700552 err = __add_section(nid, zone, section_nr_to_pfn(i));
David Rientjes4edd7ce2013-04-29 15:08:22 -0700553
554 /*
555 * EEXIST is finally dealt with by ioresource collision
556 * check. see add_memory() => register_memory_resource()
557 * Warning will be printed if there is collision.
558 */
559 if (err && (err != -EEXIST))
560 break;
561 err = 0;
562 }
Zhu Guihuac435a392015-06-24 16:58:42 -0700563 vmemmap_populate_print_last();
Joonsoo Kim7cf91a92016-03-15 14:57:51 -0700564out:
565 set_zone_contiguous(zone);
David Rientjes4edd7ce2013-04-29 15:08:22 -0700566 return err;
567}
568EXPORT_SYMBOL_GPL(__add_pages);
569
570#ifdef CONFIG_MEMORY_HOTREMOVE
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800571/* find the smallest valid pfn in the range [start_pfn, end_pfn) */
572static int find_smallest_section_pfn(int nid, struct zone *zone,
573 unsigned long start_pfn,
574 unsigned long end_pfn)
575{
576 struct mem_section *ms;
577
578 for (; start_pfn < end_pfn; start_pfn += PAGES_PER_SECTION) {
579 ms = __pfn_to_section(start_pfn);
580
581 if (unlikely(!valid_section(ms)))
582 continue;
583
584 if (unlikely(pfn_to_nid(start_pfn) != nid))
585 continue;
586
587 if (zone && zone != page_zone(pfn_to_page(start_pfn)))
588 continue;
589
590 return start_pfn;
591 }
592
593 return 0;
594}
595
596/* find the biggest valid pfn in the range [start_pfn, end_pfn). */
597static int find_biggest_section_pfn(int nid, struct zone *zone,
598 unsigned long start_pfn,
599 unsigned long end_pfn)
600{
601 struct mem_section *ms;
602 unsigned long pfn;
603
604 /* pfn is the end pfn of a memory section. */
605 pfn = end_pfn - 1;
606 for (; pfn >= start_pfn; pfn -= PAGES_PER_SECTION) {
607 ms = __pfn_to_section(pfn);
608
609 if (unlikely(!valid_section(ms)))
610 continue;
611
612 if (unlikely(pfn_to_nid(pfn) != nid))
613 continue;
614
615 if (zone && zone != page_zone(pfn_to_page(pfn)))
616 continue;
617
618 return pfn;
619 }
620
621 return 0;
622}
623
624static void shrink_zone_span(struct zone *zone, unsigned long start_pfn,
625 unsigned long end_pfn)
626{
Xishi Qiuc33bc312013-09-11 14:21:44 -0700627 unsigned long zone_start_pfn = zone->zone_start_pfn;
628 unsigned long z = zone_end_pfn(zone); /* zone_end_pfn namespace clash */
629 unsigned long zone_end_pfn = z;
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800630 unsigned long pfn;
631 struct mem_section *ms;
632 int nid = zone_to_nid(zone);
633
634 zone_span_writelock(zone);
635 if (zone_start_pfn == start_pfn) {
636 /*
637 * If the section is smallest section in the zone, it need
638 * shrink zone->zone_start_pfn and zone->zone_spanned_pages.
639 * In this case, we find second smallest valid mem_section
640 * for shrinking zone.
641 */
642 pfn = find_smallest_section_pfn(nid, zone, end_pfn,
643 zone_end_pfn);
644 if (pfn) {
645 zone->zone_start_pfn = pfn;
646 zone->spanned_pages = zone_end_pfn - pfn;
647 }
648 } else if (zone_end_pfn == end_pfn) {
649 /*
650 * If the section is biggest section in the zone, it need
651 * shrink zone->spanned_pages.
652 * In this case, we find second biggest valid mem_section for
653 * shrinking zone.
654 */
655 pfn = find_biggest_section_pfn(nid, zone, zone_start_pfn,
656 start_pfn);
657 if (pfn)
658 zone->spanned_pages = pfn - zone_start_pfn + 1;
659 }
660
661 /*
662 * The section is not biggest or smallest mem_section in the zone, it
663 * only creates a hole in the zone. So in this case, we need not
664 * change the zone. But perhaps, the zone has only hole data. Thus
665 * it check the zone has only hole or not.
666 */
667 pfn = zone_start_pfn;
668 for (; pfn < zone_end_pfn; pfn += PAGES_PER_SECTION) {
669 ms = __pfn_to_section(pfn);
670
671 if (unlikely(!valid_section(ms)))
672 continue;
673
674 if (page_zone(pfn_to_page(pfn)) != zone)
675 continue;
676
677 /* If the section is current section, it continues the loop */
678 if (start_pfn == pfn)
679 continue;
680
681 /* If we find valid section, we have nothing to do */
682 zone_span_writeunlock(zone);
683 return;
684 }
685
686 /* The zone has no valid section */
687 zone->zone_start_pfn = 0;
688 zone->spanned_pages = 0;
689 zone_span_writeunlock(zone);
690}
691
692static void shrink_pgdat_span(struct pglist_data *pgdat,
693 unsigned long start_pfn, unsigned long end_pfn)
694{
Xishi Qiu83285c72013-11-12 15:07:19 -0800695 unsigned long pgdat_start_pfn = pgdat->node_start_pfn;
696 unsigned long p = pgdat_end_pfn(pgdat); /* pgdat_end_pfn namespace clash */
697 unsigned long pgdat_end_pfn = p;
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800698 unsigned long pfn;
699 struct mem_section *ms;
700 int nid = pgdat->node_id;
701
702 if (pgdat_start_pfn == start_pfn) {
703 /*
704 * If the section is smallest section in the pgdat, it need
705 * shrink pgdat->node_start_pfn and pgdat->node_spanned_pages.
706 * In this case, we find second smallest valid mem_section
707 * for shrinking zone.
708 */
709 pfn = find_smallest_section_pfn(nid, NULL, end_pfn,
710 pgdat_end_pfn);
711 if (pfn) {
712 pgdat->node_start_pfn = pfn;
713 pgdat->node_spanned_pages = pgdat_end_pfn - pfn;
714 }
715 } else if (pgdat_end_pfn == end_pfn) {
716 /*
717 * If the section is biggest section in the pgdat, it need
718 * shrink pgdat->node_spanned_pages.
719 * In this case, we find second biggest valid mem_section for
720 * shrinking zone.
721 */
722 pfn = find_biggest_section_pfn(nid, NULL, pgdat_start_pfn,
723 start_pfn);
724 if (pfn)
725 pgdat->node_spanned_pages = pfn - pgdat_start_pfn + 1;
726 }
727
728 /*
729 * If the section is not biggest or smallest mem_section in the pgdat,
730 * it only creates a hole in the pgdat. So in this case, we need not
731 * change the pgdat.
732 * But perhaps, the pgdat has only hole data. Thus it check the pgdat
733 * has only hole or not.
734 */
735 pfn = pgdat_start_pfn;
736 for (; pfn < pgdat_end_pfn; pfn += PAGES_PER_SECTION) {
737 ms = __pfn_to_section(pfn);
738
739 if (unlikely(!valid_section(ms)))
740 continue;
741
742 if (pfn_to_nid(pfn) != nid)
743 continue;
744
745 /* If the section is current section, it continues the loop */
746 if (start_pfn == pfn)
747 continue;
748
749 /* If we find valid section, we have nothing to do */
750 return;
751 }
752
753 /* The pgdat has no valid section */
754 pgdat->node_start_pfn = 0;
755 pgdat->node_spanned_pages = 0;
756}
757
758static void __remove_zone(struct zone *zone, unsigned long start_pfn)
759{
760 struct pglist_data *pgdat = zone->zone_pgdat;
761 int nr_pages = PAGES_PER_SECTION;
762 int zone_type;
763 unsigned long flags;
764
765 zone_type = zone - pgdat->node_zones;
766
767 pgdat_resize_lock(zone->zone_pgdat, &flags);
768 shrink_zone_span(zone, start_pfn, start_pfn + nr_pages);
769 shrink_pgdat_span(pgdat, start_pfn, start_pfn + nr_pages);
770 pgdat_resize_unlock(zone->zone_pgdat, &flags);
771}
772
Dan Williams4b94ffd2016-01-15 16:56:22 -0800773static int __remove_section(struct zone *zone, struct mem_section *ms,
774 unsigned long map_offset)
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700775{
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800776 unsigned long start_pfn;
777 int scn_nr;
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700778 int ret = -EINVAL;
779
780 if (!valid_section(ms))
781 return ret;
782
783 ret = unregister_memory_section(ms);
784 if (ret)
785 return ret;
786
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800787 scn_nr = __section_nr(ms);
788 start_pfn = section_nr_to_pfn(scn_nr);
789 __remove_zone(zone, start_pfn);
790
Dan Williams4b94ffd2016-01-15 16:56:22 -0800791 sparse_remove_one_section(zone, ms, map_offset);
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700792 return 0;
793}
794
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700795/**
796 * __remove_pages() - remove sections of pages from a zone
797 * @zone: zone from which pages need to be removed
798 * @phys_start_pfn: starting pageframe (must be aligned to start of a section)
799 * @nr_pages: number of pages to remove (must be multiple of section size)
800 *
801 * Generic helper function to remove section mappings and sysfs entries
802 * for the section of the memory we are removing. Caller needs to make
803 * sure that pages are marked reserved and zones are adjust properly by
804 * calling offline_pages().
805 */
806int __remove_pages(struct zone *zone, unsigned long phys_start_pfn,
807 unsigned long nr_pages)
808{
Toshi Kanife74ebb2013-04-29 15:08:20 -0700809 unsigned long i;
Dan Williams4b94ffd2016-01-15 16:56:22 -0800810 unsigned long map_offset = 0;
811 int sections_to_remove, ret = 0;
812
813 /* In the ZONE_DEVICE case device driver owns the memory region */
814 if (is_dev_zone(zone)) {
815 struct page *page = pfn_to_page(phys_start_pfn);
816 struct vmem_altmap *altmap;
817
818 altmap = to_vmem_altmap((unsigned long) page);
819 if (altmap)
820 map_offset = vmem_altmap_offset(altmap);
821 } else {
822 resource_size_t start, size;
823
824 start = phys_start_pfn << PAGE_SHIFT;
825 size = nr_pages * PAGE_SIZE;
826
827 ret = release_mem_region_adjustable(&iomem_resource, start,
828 size);
829 if (ret) {
830 resource_size_t endres = start + size - 1;
831
832 pr_warn("Unable to release resource <%pa-%pa> (%d)\n",
833 &start, &endres, ret);
834 }
835 }
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700836
Joonsoo Kim7cf91a92016-03-15 14:57:51 -0700837 clear_zone_contiguous(zone);
838
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700839 /*
840 * We can only remove entire sections
841 */
842 BUG_ON(phys_start_pfn & ~PAGE_SECTION_MASK);
843 BUG_ON(nr_pages % PAGES_PER_SECTION);
844
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700845 sections_to_remove = nr_pages / PAGES_PER_SECTION;
846 for (i = 0; i < sections_to_remove; i++) {
847 unsigned long pfn = phys_start_pfn + i*PAGES_PER_SECTION;
Dan Williams4b94ffd2016-01-15 16:56:22 -0800848
849 ret = __remove_section(zone, __pfn_to_section(pfn), map_offset);
850 map_offset = 0;
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700851 if (ret)
852 break;
853 }
Joonsoo Kim7cf91a92016-03-15 14:57:51 -0700854
855 set_zone_contiguous(zone);
856
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700857 return ret;
858}
859EXPORT_SYMBOL_GPL(__remove_pages);
David Rientjes4edd7ce2013-04-29 15:08:22 -0700860#endif /* CONFIG_MEMORY_HOTREMOVE */
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700861
Daniel Kiper9d0ad8c2011-07-25 17:12:05 -0700862int set_online_page_callback(online_page_callback_t callback)
863{
864 int rc = -EINVAL;
865
Vladimir Davydovbfc8c902014-06-04 16:07:18 -0700866 get_online_mems();
867 mutex_lock(&online_page_callback_lock);
Daniel Kiper9d0ad8c2011-07-25 17:12:05 -0700868
869 if (online_page_callback == generic_online_page) {
870 online_page_callback = callback;
871 rc = 0;
872 }
873
Vladimir Davydovbfc8c902014-06-04 16:07:18 -0700874 mutex_unlock(&online_page_callback_lock);
875 put_online_mems();
Daniel Kiper9d0ad8c2011-07-25 17:12:05 -0700876
877 return rc;
878}
879EXPORT_SYMBOL_GPL(set_online_page_callback);
880
881int restore_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 == callback) {
889 online_page_callback = generic_online_page;
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(restore_online_page_callback);
899
900void __online_page_set_limits(struct page *page)
Jeremy Fitzhardinge180c06e2008-04-28 02:12:03 -0700901{
Daniel Kiper9d0ad8c2011-07-25 17:12:05 -0700902}
903EXPORT_SYMBOL_GPL(__online_page_set_limits);
904
905void __online_page_increment_counters(struct page *page)
906{
Jiang Liu3dcc0572013-07-03 15:03:21 -0700907 adjust_managed_page_count(page, 1);
Daniel Kiper9d0ad8c2011-07-25 17:12:05 -0700908}
909EXPORT_SYMBOL_GPL(__online_page_increment_counters);
Jeremy Fitzhardinge180c06e2008-04-28 02:12:03 -0700910
Daniel Kiper9d0ad8c2011-07-25 17:12:05 -0700911void __online_page_free(struct page *page)
912{
Jiang Liu3dcc0572013-07-03 15:03:21 -0700913 __free_reserved_page(page);
Jeremy Fitzhardinge180c06e2008-04-28 02:12:03 -0700914}
Daniel Kiper9d0ad8c2011-07-25 17:12:05 -0700915EXPORT_SYMBOL_GPL(__online_page_free);
916
917static void generic_online_page(struct page *page)
918{
919 __online_page_set_limits(page);
920 __online_page_increment_counters(page);
921 __online_page_free(page);
922}
Jeremy Fitzhardinge180c06e2008-04-28 02:12:03 -0700923
KAMEZAWA Hiroyuki75884fb2007-10-16 01:26:10 -0700924static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages,
925 void *arg)
Dave Hansen3947be12005-10-29 18:16:54 -0700926{
927 unsigned long i;
KAMEZAWA Hiroyuki75884fb2007-10-16 01:26:10 -0700928 unsigned long onlined_pages = *(unsigned long *)arg;
929 struct page *page;
930 if (PageReserved(pfn_to_page(start_pfn)))
931 for (i = 0; i < nr_pages; i++) {
932 page = pfn_to_page(start_pfn + i);
Daniel Kiper9d0ad8c2011-07-25 17:12:05 -0700933 (*online_page_callback)(page);
KAMEZAWA Hiroyuki75884fb2007-10-16 01:26:10 -0700934 onlined_pages++;
935 }
936 *(unsigned long *)arg = onlined_pages;
937 return 0;
938}
939
Lai Jiangshan09285af2012-12-12 13:52:04 -0800940#ifdef CONFIG_MOVABLE_NODE
Tang Chen79a4dce2012-12-18 14:23:24 -0800941/*
942 * When CONFIG_MOVABLE_NODE, we permit onlining of a node which doesn't have
943 * normal memory.
944 */
Lai Jiangshan09285af2012-12-12 13:52:04 -0800945static bool can_online_high_movable(struct zone *zone)
946{
947 return true;
948}
Tang Chen79a4dce2012-12-18 14:23:24 -0800949#else /* CONFIG_MOVABLE_NODE */
Lai Jiangshan74d42d82012-12-11 16:03:23 -0800950/* ensure every online node has NORMAL memory */
951static bool can_online_high_movable(struct zone *zone)
952{
953 return node_state(zone_to_nid(zone), N_NORMAL_MEMORY);
954}
Tang Chen79a4dce2012-12-18 14:23:24 -0800955#endif /* CONFIG_MOVABLE_NODE */
Lai Jiangshan74d42d82012-12-11 16:03:23 -0800956
Lai Jiangshand9713672012-12-11 16:01:03 -0800957/* check which state of node_states will be changed when online memory */
958static void node_states_check_changes_online(unsigned long nr_pages,
959 struct zone *zone, struct memory_notify *arg)
960{
961 int nid = zone_to_nid(zone);
962 enum zone_type zone_last = ZONE_NORMAL;
963
964 /*
Lai Jiangshan6715ddf2012-12-12 13:51:49 -0800965 * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY]
966 * contains nodes which have zones of 0...ZONE_NORMAL,
967 * set zone_last to ZONE_NORMAL.
Lai Jiangshand9713672012-12-11 16:01:03 -0800968 *
Lai Jiangshan6715ddf2012-12-12 13:51:49 -0800969 * If we don't have HIGHMEM nor movable node,
970 * node_states[N_NORMAL_MEMORY] contains nodes which have zones of
971 * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE.
Lai Jiangshand9713672012-12-11 16:01:03 -0800972 */
Lai Jiangshan6715ddf2012-12-12 13:51:49 -0800973 if (N_MEMORY == N_NORMAL_MEMORY)
Lai Jiangshand9713672012-12-11 16:01:03 -0800974 zone_last = ZONE_MOVABLE;
975
976 /*
977 * if the memory to be online is in a zone of 0...zone_last, and
978 * the zones of 0...zone_last don't have memory before online, we will
979 * need to set the node to node_states[N_NORMAL_MEMORY] after
980 * the memory is online.
981 */
982 if (zone_idx(zone) <= zone_last && !node_state(nid, N_NORMAL_MEMORY))
983 arg->status_change_nid_normal = nid;
984 else
985 arg->status_change_nid_normal = -1;
986
Lai Jiangshan6715ddf2012-12-12 13:51:49 -0800987#ifdef CONFIG_HIGHMEM
988 /*
989 * If we have movable node, node_states[N_HIGH_MEMORY]
990 * contains nodes which have zones of 0...ZONE_HIGHMEM,
991 * set zone_last to ZONE_HIGHMEM.
992 *
993 * If we don't have movable node, node_states[N_NORMAL_MEMORY]
994 * contains nodes which have zones of 0...ZONE_MOVABLE,
995 * set zone_last to ZONE_MOVABLE.
996 */
997 zone_last = ZONE_HIGHMEM;
998 if (N_MEMORY == N_HIGH_MEMORY)
999 zone_last = ZONE_MOVABLE;
1000
1001 if (zone_idx(zone) <= zone_last && !node_state(nid, N_HIGH_MEMORY))
1002 arg->status_change_nid_high = nid;
1003 else
1004 arg->status_change_nid_high = -1;
1005#else
1006 arg->status_change_nid_high = arg->status_change_nid_normal;
1007#endif
1008
Lai Jiangshand9713672012-12-11 16:01:03 -08001009 /*
1010 * if the node don't have memory befor online, we will need to
Lai Jiangshan6715ddf2012-12-12 13:51:49 -08001011 * set the node to node_states[N_MEMORY] after the memory
Lai Jiangshand9713672012-12-11 16:01:03 -08001012 * is online.
1013 */
Lai Jiangshan6715ddf2012-12-12 13:51:49 -08001014 if (!node_state(nid, N_MEMORY))
Lai Jiangshand9713672012-12-11 16:01:03 -08001015 arg->status_change_nid = nid;
1016 else
1017 arg->status_change_nid = -1;
1018}
1019
1020static void node_states_set_node(int node, struct memory_notify *arg)
1021{
1022 if (arg->status_change_nid_normal >= 0)
1023 node_set_state(node, N_NORMAL_MEMORY);
1024
Lai Jiangshan6715ddf2012-12-12 13:51:49 -08001025 if (arg->status_change_nid_high >= 0)
1026 node_set_state(node, N_HIGH_MEMORY);
1027
1028 node_set_state(node, N_MEMORY);
Lai Jiangshand9713672012-12-11 16:01:03 -08001029}
1030
KAMEZAWA Hiroyuki75884fb2007-10-16 01:26:10 -07001031
David Rientjes30467e02015-04-14 15:45:11 -07001032/* Must be protected by mem_hotplug_begin() */
Lai Jiangshan511c2ab2012-12-11 16:03:16 -08001033int __ref online_pages(unsigned long pfn, unsigned long nr_pages, int online_type)
KAMEZAWA Hiroyuki75884fb2007-10-16 01:26:10 -07001034{
Cody P Schaferaa472282013-07-03 15:02:10 -07001035 unsigned long flags;
Dave Hansen3947be12005-10-29 18:16:54 -07001036 unsigned long onlined_pages = 0;
1037 struct zone *zone;
Yasunori Goto68113782006-06-23 02:03:11 -07001038 int need_zonelists_rebuild = 0;
Yasunori Goto7b78d332007-10-21 16:41:36 -07001039 int nid;
1040 int ret;
1041 struct memory_notify arg;
Dave Hansen3947be12005-10-29 18:16:54 -07001042
Lai Jiangshand9713672012-12-11 16:01:03 -08001043 /*
1044 * This doesn't need a lock to do pfn_to_page().
1045 * The section can't be removed here because of the
1046 * memory_block->state_mutex.
1047 */
1048 zone = page_zone(pfn_to_page(pfn));
1049
Tang Chen4f7c6b42014-08-06 16:05:13 -07001050 if ((zone_idx(zone) > ZONE_NORMAL ||
1051 online_type == MMOP_ONLINE_MOVABLE) &&
Vladimir Davydovbfc8c902014-06-04 16:07:18 -07001052 !can_online_high_movable(zone))
David Rientjes30467e02015-04-14 15:45:11 -07001053 return -EINVAL;
Lai Jiangshan74d42d82012-12-11 16:03:23 -08001054
Tang Chen4f7c6b42014-08-06 16:05:13 -07001055 if (online_type == MMOP_ONLINE_KERNEL &&
1056 zone_idx(zone) == ZONE_MOVABLE) {
Vladimir Davydovbfc8c902014-06-04 16:07:18 -07001057 if (move_pfn_range_left(zone - 1, zone, pfn, pfn + nr_pages))
David Rientjes30467e02015-04-14 15:45:11 -07001058 return -EINVAL;
Lai Jiangshan511c2ab2012-12-11 16:03:16 -08001059 }
Tang Chen4f7c6b42014-08-06 16:05:13 -07001060 if (online_type == MMOP_ONLINE_MOVABLE &&
1061 zone_idx(zone) == ZONE_MOVABLE - 1) {
Vladimir Davydovbfc8c902014-06-04 16:07:18 -07001062 if (move_pfn_range_right(zone, zone + 1, pfn, pfn + nr_pages))
David Rientjes30467e02015-04-14 15:45:11 -07001063 return -EINVAL;
Lai Jiangshan511c2ab2012-12-11 16:03:16 -08001064 }
1065
1066 /* Previous code may changed the zone of the pfn range */
1067 zone = page_zone(pfn_to_page(pfn));
1068
Yasunori Goto7b78d332007-10-21 16:41:36 -07001069 arg.start_pfn = pfn;
1070 arg.nr_pages = nr_pages;
Lai Jiangshand9713672012-12-11 16:01:03 -08001071 node_states_check_changes_online(nr_pages, zone, &arg);
Yasunori Goto7b78d332007-10-21 16:41:36 -07001072
Vlastimil Babkae888ca32016-03-17 14:18:12 -07001073 nid = zone_to_nid(zone);
Yasunori Goto7b78d332007-10-21 16:41:36 -07001074
1075 ret = memory_notify(MEM_GOING_ONLINE, &arg);
1076 ret = notifier_to_errno(ret);
Chen Yuconge33e33b2016-03-17 14:19:35 -07001077 if (ret)
1078 goto failed_addition;
1079
Dave Hansen3947be12005-10-29 18:16:54 -07001080 /*
Yasunori Goto68113782006-06-23 02:03:11 -07001081 * If this zone is not populated, then it is not in zonelist.
1082 * This means the page allocator ignores this zone.
1083 * So, zonelist must be updated after online.
1084 */
Haicheng Li4eaf3f62010-05-24 14:32:52 -07001085 mutex_lock(&zonelists_mutex);
Wen Congyang6dcd73d2012-12-11 16:01:01 -08001086 if (!populated_zone(zone)) {
Yasunori Goto68113782006-06-23 02:03:11 -07001087 need_zonelists_rebuild = 1;
Wen Congyang6dcd73d2012-12-11 16:01:01 -08001088 build_all_zonelists(NULL, zone);
1089 }
Yasunori Goto68113782006-06-23 02:03:11 -07001090
KAMEZAWA Hiroyuki908eedc2009-09-22 16:45:46 -07001091 ret = walk_system_ram_range(pfn, nr_pages, &onlined_pages,
KAMEZAWA Hiroyuki75884fb2007-10-16 01:26:10 -07001092 online_pages_range);
Geoff Levandfd8a4222008-05-14 16:05:50 -07001093 if (ret) {
Wen Congyang6dcd73d2012-12-11 16:01:01 -08001094 if (need_zonelists_rebuild)
1095 zone_pcp_reset(zone);
Haicheng Li4eaf3f62010-05-24 14:32:52 -07001096 mutex_unlock(&zonelists_mutex);
Chen Yuconge33e33b2016-03-17 14:19:35 -07001097 goto failed_addition;
Geoff Levandfd8a4222008-05-14 16:05:50 -07001098 }
1099
Dave Hansen3947be12005-10-29 18:16:54 -07001100 zone->present_pages += onlined_pages;
Cody P Schaferaa472282013-07-03 15:02:10 -07001101
1102 pgdat_resize_lock(zone->zone_pgdat, &flags);
Yasunori Gotof2937be2006-03-09 17:33:51 -08001103 zone->zone_pgdat->node_present_pages += onlined_pages;
Cody P Schaferaa472282013-07-03 15:02:10 -07001104 pgdat_resize_unlock(zone->zone_pgdat, &flags);
1105
Jiang Liu08dff7b2012-07-31 16:43:30 -07001106 if (onlined_pages) {
Vlastimil Babkae888ca32016-03-17 14:18:12 -07001107 node_states_set_node(nid, &arg);
Jiang Liu08dff7b2012-07-31 16:43:30 -07001108 if (need_zonelists_rebuild)
Wen Congyang6dcd73d2012-12-11 16:01:01 -08001109 build_all_zonelists(NULL, NULL);
Jiang Liu08dff7b2012-07-31 16:43:30 -07001110 else
1111 zone_pcp_update(zone);
1112 }
Dave Hansen3947be12005-10-29 18:16:54 -07001113
Haicheng Li4eaf3f62010-05-24 14:32:52 -07001114 mutex_unlock(&zonelists_mutex);
KOSAKI Motohiro1b79acc2011-05-24 17:11:32 -07001115
1116 init_per_zone_wmark_min();
1117
Vlastimil Babka698b1b32016-03-17 14:18:08 -07001118 if (onlined_pages) {
Vlastimil Babkae888ca32016-03-17 14:18:12 -07001119 kswapd_run(nid);
Vlastimil Babka698b1b32016-03-17 14:18:08 -07001120 kcompactd_run(nid);
1121 }
Dave Hansen61b13992005-10-29 18:16:56 -07001122
Haicheng Li1f522502010-05-24 14:32:51 -07001123 vm_total_pages = nr_free_pagecache_pages();
Kent Liu2f7f24e2008-07-23 21:28:18 -07001124
Chandra Seetharaman2d1d43f2006-09-29 02:01:25 -07001125 writeback_set_ratelimit();
Yasunori Goto7b78d332007-10-21 16:41:36 -07001126
1127 if (onlined_pages)
1128 memory_notify(MEM_ONLINE, &arg);
David Rientjes30467e02015-04-14 15:45:11 -07001129 return 0;
Chen Yuconge33e33b2016-03-17 14:19:35 -07001130
1131failed_addition:
1132 pr_debug("online_pages [mem %#010llx-%#010llx] failed\n",
1133 (unsigned long long) pfn << PAGE_SHIFT,
1134 (((unsigned long long) pfn + nr_pages) << PAGE_SHIFT) - 1);
1135 memory_notify(MEM_CANCEL_ONLINE, &arg);
1136 return ret;
Dave Hansen3947be12005-10-29 18:16:54 -07001137}
Keith Mannthey53947022006-09-30 23:27:08 -07001138#endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
Yasunori Gotobc02af92006-06-27 02:53:30 -07001139
Tang Chen0bd85422014-11-13 15:19:41 -08001140static void reset_node_present_pages(pg_data_t *pgdat)
1141{
1142 struct zone *z;
1143
1144 for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++)
1145 z->present_pages = 0;
1146
1147 pgdat->node_present_pages = 0;
1148}
1149
Hidetoshi Setoe1319332009-11-17 14:06:18 -08001150/* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
1151static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start)
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001152{
1153 struct pglist_data *pgdat;
1154 unsigned long zones_size[MAX_NR_ZONES] = {0};
1155 unsigned long zholes_size[MAX_NR_ZONES] = {0};
Fabian Frederickc8e861a2014-06-04 16:07:51 -07001156 unsigned long start_pfn = PFN_DOWN(start);
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001157
Tang Chena1e565a2013-02-22 16:33:18 -08001158 pgdat = NODE_DATA(nid);
1159 if (!pgdat) {
1160 pgdat = arch_alloc_nodedata(nid);
1161 if (!pgdat)
1162 return NULL;
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001163
Tang Chena1e565a2013-02-22 16:33:18 -08001164 arch_refresh_nodedata(nid, pgdat);
Gu Zhengb0dc3a32015-03-25 15:55:20 -07001165 } else {
1166 /* Reset the nr_zones and classzone_idx to 0 before reuse */
1167 pgdat->nr_zones = 0;
1168 pgdat->classzone_idx = 0;
Tang Chena1e565a2013-02-22 16:33:18 -08001169 }
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001170
1171 /* we can use NODE_DATA(nid) from here */
1172
1173 /* init node's zones as empty zones, we don't have any present pages.*/
Johannes Weiner9109fb72008-07-23 21:27:20 -07001174 free_area_init_node(nid, zones_size, start_pfn, zholes_size);
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001175
KAMEZAWA Hiroyuki959ecc42011-06-15 15:08:38 -07001176 /*
1177 * The node we allocated has no zone fallback lists. For avoiding
1178 * to access not-initialized zonelist, build here.
1179 */
David Rientjesf957db42011-06-22 18:13:04 -07001180 mutex_lock(&zonelists_mutex);
Jiang Liu9adb62a2012-07-31 16:43:28 -07001181 build_all_zonelists(pgdat, NULL);
David Rientjesf957db42011-06-22 18:13:04 -07001182 mutex_unlock(&zonelists_mutex);
KAMEZAWA Hiroyuki959ecc42011-06-15 15:08:38 -07001183
Tang Chenf784a3f2014-11-13 15:19:39 -08001184 /*
1185 * zone->managed_pages is set to an approximate value in
1186 * free_area_init_core(), which will cause
1187 * /sys/device/system/node/nodeX/meminfo has wrong data.
1188 * So reset it to 0 before any memory is onlined.
1189 */
1190 reset_node_managed_pages(pgdat);
1191
Tang Chen0bd85422014-11-13 15:19:41 -08001192 /*
1193 * When memory is hot-added, all the memory is in offline state. So
1194 * clear all zones' present_pages because they will be updated in
1195 * online_pages() and offline_pages().
1196 */
1197 reset_node_present_pages(pgdat);
1198
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001199 return pgdat;
1200}
1201
1202static void rollback_node_hotadd(int nid, pg_data_t *pgdat)
1203{
1204 arch_refresh_nodedata(nid, NULL);
1205 arch_free_nodedata(pgdat);
1206 return;
1207}
1208
KAMEZAWA Hiroyuki0a547032006-06-27 02:53:35 -07001209
Toshi Kani01b0f192013-11-12 15:07:25 -08001210/**
1211 * try_online_node - online a node if offlined
1212 *
minskey guocf234222010-05-24 14:32:41 -07001213 * called by cpu_up() to online a node without onlined memory.
1214 */
Toshi Kani01b0f192013-11-12 15:07:25 -08001215int try_online_node(int nid)
minskey guocf234222010-05-24 14:32:41 -07001216{
1217 pg_data_t *pgdat;
1218 int ret;
1219
Toshi Kani01b0f192013-11-12 15:07:25 -08001220 if (node_online(nid))
1221 return 0;
1222
Vladimir Davydovbfc8c902014-06-04 16:07:18 -07001223 mem_hotplug_begin();
minskey guocf234222010-05-24 14:32:41 -07001224 pgdat = hotadd_new_pgdat(nid, 0);
David Rientjes7553e8f2011-06-22 18:13:01 -07001225 if (!pgdat) {
Toshi Kani01b0f192013-11-12 15:07:25 -08001226 pr_err("Cannot online node %d due to NULL pgdat\n", nid);
minskey guocf234222010-05-24 14:32:41 -07001227 ret = -ENOMEM;
1228 goto out;
1229 }
1230 node_set_online(nid);
1231 ret = register_one_node(nid);
1232 BUG_ON(ret);
1233
Toshi Kani01b0f192013-11-12 15:07:25 -08001234 if (pgdat->node_zonelists->_zonerefs->zone == NULL) {
1235 mutex_lock(&zonelists_mutex);
1236 build_all_zonelists(NULL, NULL);
1237 mutex_unlock(&zonelists_mutex);
1238 }
1239
minskey guocf234222010-05-24 14:32:41 -07001240out:
Vladimir Davydovbfc8c902014-06-04 16:07:18 -07001241 mem_hotplug_done();
minskey guocf234222010-05-24 14:32:41 -07001242 return ret;
1243}
1244
Toshi Kani27356f52013-09-11 14:21:49 -07001245static int check_hotplug_memory_range(u64 start, u64 size)
1246{
Fabian Frederickc8e861a2014-06-04 16:07:51 -07001247 u64 start_pfn = PFN_DOWN(start);
Toshi Kani27356f52013-09-11 14:21:49 -07001248 u64 nr_pages = size >> PAGE_SHIFT;
1249
1250 /* Memory range must be aligned with section */
1251 if ((start_pfn & ~PAGE_SECTION_MASK) ||
1252 (nr_pages % PAGES_PER_SECTION) || (!nr_pages)) {
1253 pr_err("Section-unaligned hotplug range: start 0x%llx, size 0x%llx\n",
1254 (unsigned long long)start,
1255 (unsigned long long)size);
1256 return -EINVAL;
1257 }
1258
1259 return 0;
1260}
1261
Wang Nan63264402014-08-06 16:07:36 -07001262/*
1263 * If movable zone has already been setup, newly added memory should be check.
1264 * If its address is higher than movable zone, it should be added as movable.
1265 * Without this check, movable zone may overlap with other zone.
1266 */
1267static int should_add_memory_movable(int nid, u64 start, u64 size)
1268{
1269 unsigned long start_pfn = start >> PAGE_SHIFT;
1270 pg_data_t *pgdat = NODE_DATA(nid);
1271 struct zone *movable_zone = pgdat->node_zones + ZONE_MOVABLE;
1272
1273 if (zone_is_empty(movable_zone))
1274 return 0;
1275
1276 if (movable_zone->zone_start_pfn <= start_pfn)
1277 return 1;
1278
1279 return 0;
1280}
1281
Dan Williams033fbae2015-08-09 15:29:06 -04001282int zone_for_memory(int nid, u64 start, u64 size, int zone_default,
1283 bool for_device)
Wang Nan63264402014-08-06 16:07:36 -07001284{
Dan Williams033fbae2015-08-09 15:29:06 -04001285#ifdef CONFIG_ZONE_DEVICE
1286 if (for_device)
1287 return ZONE_DEVICE;
1288#endif
Wang Nan63264402014-08-06 16:07:36 -07001289 if (should_add_memory_movable(nid, start, size))
1290 return ZONE_MOVABLE;
1291
1292 return zone_default;
1293}
1294
Vitaly Kuznetsov31bc3852016-03-15 14:56:48 -07001295static int online_memory_block(struct memory_block *mem, void *arg)
1296{
1297 return memory_block_change_state(mem, MEM_ONLINE, MEM_OFFLINE);
1298}
1299
Al Viro31168482008-11-22 17:33:24 +00001300/* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
Vitaly Kuznetsov31bc3852016-03-15 14:56:48 -07001301int __ref add_memory_resource(int nid, struct resource *res, bool online)
Yasunori Gotobc02af92006-06-27 02:53:30 -07001302{
David Vrabel62cedb92015-06-25 16:35:49 +01001303 u64 start, size;
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001304 pg_data_t *pgdat = NULL;
Tang Chena1e565a2013-02-22 16:33:18 -08001305 bool new_pgdat;
1306 bool new_node;
Yasunori Gotobc02af92006-06-27 02:53:30 -07001307 int ret;
1308
David Vrabel62cedb92015-06-25 16:35:49 +01001309 start = res->start;
1310 size = resource_size(res);
1311
Toshi Kani27356f52013-09-11 14:21:49 -07001312 ret = check_hotplug_memory_range(start, size);
1313 if (ret)
1314 return ret;
1315
Tang Chena1e565a2013-02-22 16:33:18 -08001316 { /* Stupid hack to suppress address-never-null warning */
1317 void *p = NODE_DATA(nid);
1318 new_pgdat = !p;
1319 }
Nathan Zimmerac13c462014-01-23 15:53:26 -08001320
Vladimir Davydovbfc8c902014-06-04 16:07:18 -07001321 mem_hotplug_begin();
Nathan Zimmerac13c462014-01-23 15:53:26 -08001322
Tang Chen7f36e3e2015-09-04 15:42:32 -07001323 /*
1324 * Add new range to memblock so that when hotadd_new_pgdat() is called
1325 * to allocate new pgdat, get_pfn_range_for_nid() will be able to find
1326 * this new range and calculate total pages correctly. The range will
1327 * be removed at hot-remove time.
1328 */
1329 memblock_add_node(start, size, nid);
1330
Tang Chena1e565a2013-02-22 16:33:18 -08001331 new_node = !node_online(nid);
1332 if (new_node) {
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001333 pgdat = hotadd_new_pgdat(nid, start);
Andi Kleen6ad696d2009-11-17 14:06:22 -08001334 ret = -ENOMEM;
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001335 if (!pgdat)
Wen Congyang41b9e2d2012-07-11 14:02:31 -07001336 goto error;
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001337 }
1338
Yasunori Gotobc02af92006-06-27 02:53:30 -07001339 /* call arch's memory hotadd */
Dan Williams033fbae2015-08-09 15:29:06 -04001340 ret = arch_add_memory(nid, start, size, false);
Yasunori Gotobc02af92006-06-27 02:53:30 -07001341
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001342 if (ret < 0)
1343 goto error;
1344
Yasunori Goto0fc44152006-06-27 02:53:38 -07001345 /* we online node here. we can't roll back from here. */
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001346 node_set_online(nid);
1347
Tang Chena1e565a2013-02-22 16:33:18 -08001348 if (new_node) {
Yasunori Goto0fc44152006-06-27 02:53:38 -07001349 ret = register_one_node(nid);
1350 /*
1351 * If sysfs file of new node can't create, cpu on the node
1352 * can't be hot-added. There is no rollback way now.
1353 * So, check by BUG_ON() to catch it reluctantly..
1354 */
1355 BUG_ON(ret);
1356 }
1357
akpm@linux-foundation.orgd96ae532010-03-05 13:41:58 -08001358 /* create new memmap entry */
1359 firmware_map_add_hotplug(start, start + size, "System RAM");
1360
Vitaly Kuznetsov31bc3852016-03-15 14:56:48 -07001361 /* online pages if requested */
1362 if (online)
1363 walk_memory_range(PFN_DOWN(start), PFN_UP(start + size - 1),
1364 NULL, online_memory_block);
1365
Andi Kleen6ad696d2009-11-17 14:06:22 -08001366 goto out;
1367
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001368error:
1369 /* rollback pgdat allocation and others */
1370 if (new_pgdat)
1371 rollback_node_hotadd(nid, pgdat);
Tang Chen7f36e3e2015-09-04 15:42:32 -07001372 memblock_remove(start, size);
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001373
Andi Kleen6ad696d2009-11-17 14:06:22 -08001374out:
Vladimir Davydovbfc8c902014-06-04 16:07:18 -07001375 mem_hotplug_done();
Yasunori Gotobc02af92006-06-27 02:53:30 -07001376 return ret;
1377}
David Vrabel62cedb92015-06-25 16:35:49 +01001378EXPORT_SYMBOL_GPL(add_memory_resource);
1379
1380int __ref add_memory(int nid, u64 start, u64 size)
1381{
1382 struct resource *res;
1383 int ret;
1384
1385 res = register_memory_resource(start, size);
Vitaly Kuznetsov6f754ba2016-01-14 15:21:55 -08001386 if (IS_ERR(res))
1387 return PTR_ERR(res);
David Vrabel62cedb92015-06-25 16:35:49 +01001388
Vitaly Kuznetsov31bc3852016-03-15 14:56:48 -07001389 ret = add_memory_resource(nid, res, memhp_auto_online);
David Vrabel62cedb92015-06-25 16:35:49 +01001390 if (ret < 0)
1391 release_memory_resource(res);
1392 return ret;
1393}
Yasunori Gotobc02af92006-06-27 02:53:30 -07001394EXPORT_SYMBOL_GPL(add_memory);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001395
1396#ifdef CONFIG_MEMORY_HOTREMOVE
1397/*
Badari Pulavarty5c755e92008-07-23 21:28:19 -07001398 * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy
1399 * set and the size of the free page is given by page_order(). Using this,
1400 * the function determines if the pageblock contains only free pages.
1401 * Due to buddy contraints, a free page at least the size of a pageblock will
1402 * be located at the start of the pageblock
1403 */
1404static inline int pageblock_free(struct page *page)
1405{
1406 return PageBuddy(page) && page_order(page) >= pageblock_order;
1407}
1408
1409/* Return the start of the next active pageblock after a given page */
1410static struct page *next_active_pageblock(struct page *page)
1411{
Badari Pulavarty5c755e92008-07-23 21:28:19 -07001412 /* Ensure the starting page is pageblock-aligned */
1413 BUG_ON(page_to_pfn(page) & (pageblock_nr_pages - 1));
1414
Badari Pulavarty5c755e92008-07-23 21:28:19 -07001415 /* If the entire pageblock is free, move to the end of free page */
KAMEZAWA Hiroyuki0dcc48c2010-09-09 16:38:01 -07001416 if (pageblock_free(page)) {
1417 int order;
1418 /* be careful. we don't have locks, page_order can be changed.*/
1419 order = page_order(page);
1420 if ((order < MAX_ORDER) && (order >= pageblock_order))
1421 return page + (1 << order);
1422 }
Badari Pulavarty5c755e92008-07-23 21:28:19 -07001423
KAMEZAWA Hiroyuki0dcc48c2010-09-09 16:38:01 -07001424 return page + pageblock_nr_pages;
Badari Pulavarty5c755e92008-07-23 21:28:19 -07001425}
1426
1427/* Checks if this range of memory is likely to be hot-removable. */
Yaowei Baic98940f2016-05-19 17:11:26 -07001428bool is_mem_section_removable(unsigned long start_pfn, unsigned long nr_pages)
Badari Pulavarty5c755e92008-07-23 21:28:19 -07001429{
Badari Pulavarty5c755e92008-07-23 21:28:19 -07001430 struct page *page = pfn_to_page(start_pfn);
1431 struct page *end_page = page + nr_pages;
1432
1433 /* Check the starting page of each pageblock within the range */
1434 for (; page < end_page; page = next_active_pageblock(page)) {
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07001435 if (!is_pageblock_removable_nolock(page))
Yaowei Baic98940f2016-05-19 17:11:26 -07001436 return false;
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07001437 cond_resched();
Badari Pulavarty5c755e92008-07-23 21:28:19 -07001438 }
1439
1440 /* All pageblocks in the memory block are likely to be hot-removable */
Yaowei Baic98940f2016-05-19 17:11:26 -07001441 return true;
Badari Pulavarty5c755e92008-07-23 21:28:19 -07001442}
1443
1444/*
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001445 * Confirm all pages in a range [start, end) is belongs to the same zone.
1446 */
Zhang Zhened2f2402014-10-09 15:26:31 -07001447int test_pages_in_a_zone(unsigned long start_pfn, unsigned long end_pfn)
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001448{
Andrew Banman5f0f2882015-12-29 14:54:25 -08001449 unsigned long pfn, sec_end_pfn;
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001450 struct zone *zone = NULL;
1451 struct page *page;
1452 int i;
Andrew Banman5f0f2882015-12-29 14:54:25 -08001453 for (pfn = start_pfn, sec_end_pfn = SECTION_ALIGN_UP(start_pfn);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001454 pfn < end_pfn;
Andrew Banman5f0f2882015-12-29 14:54:25 -08001455 pfn = sec_end_pfn + 1, sec_end_pfn += PAGES_PER_SECTION) {
1456 /* Make sure the memory section is present first */
1457 if (!present_section_nr(pfn_to_section_nr(pfn)))
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001458 continue;
Andrew Banman5f0f2882015-12-29 14:54:25 -08001459 for (; pfn < sec_end_pfn && pfn < end_pfn;
1460 pfn += MAX_ORDER_NR_PAGES) {
1461 i = 0;
1462 /* This is just a CONFIG_HOLES_IN_ZONE check.*/
1463 while ((i < MAX_ORDER_NR_PAGES) &&
1464 !pfn_valid_within(pfn + i))
1465 i++;
1466 if (i == MAX_ORDER_NR_PAGES)
1467 continue;
1468 page = pfn_to_page(pfn + i);
1469 if (zone && page_zone(page) != zone)
1470 return 0;
1471 zone = page_zone(page);
1472 }
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001473 }
1474 return 1;
1475}
1476
1477/*
Naoya Horiguchic8721bb2013-09-11 14:22:09 -07001478 * Scan pfn range [start,end) to find movable/migratable pages (LRU pages
1479 * and hugepages). We scan pfn because it's much easier than scanning over
1480 * linked list. This function returns the pfn of the first found movable
1481 * page if it's found, otherwise 0.
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001482 */
Naoya Horiguchic8721bb2013-09-11 14:22:09 -07001483static unsigned long scan_movable_pages(unsigned long start, unsigned long end)
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001484{
1485 unsigned long pfn;
1486 struct page *page;
1487 for (pfn = start; pfn < end; pfn++) {
1488 if (pfn_valid(pfn)) {
1489 page = pfn_to_page(pfn);
1490 if (PageLRU(page))
1491 return pfn;
Naoya Horiguchic8721bb2013-09-11 14:22:09 -07001492 if (PageHuge(page)) {
Naoya Horiguchi7e1f0492015-04-15 16:14:41 -07001493 if (page_huge_active(page))
Naoya Horiguchic8721bb2013-09-11 14:22:09 -07001494 return pfn;
1495 else
1496 pfn = round_up(pfn + 1,
1497 1 << compound_order(page)) - 1;
1498 }
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001499 }
1500 }
1501 return 0;
1502}
1503
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001504#define NR_OFFLINE_AT_ONCE_PAGES (256)
1505static int
1506do_migrate_range(unsigned long start_pfn, unsigned long end_pfn)
1507{
1508 unsigned long pfn;
1509 struct page *page;
1510 int move_pages = NR_OFFLINE_AT_ONCE_PAGES;
1511 int not_managed = 0;
1512 int ret = 0;
1513 LIST_HEAD(source);
1514
1515 for (pfn = start_pfn; pfn < end_pfn && move_pages > 0; pfn++) {
1516 if (!pfn_valid(pfn))
1517 continue;
1518 page = pfn_to_page(pfn);
Naoya Horiguchic8721bb2013-09-11 14:22:09 -07001519
1520 if (PageHuge(page)) {
1521 struct page *head = compound_head(page);
1522 pfn = page_to_pfn(head) + (1<<compound_order(head)) - 1;
1523 if (compound_order(head) > PFN_SECTION_SHIFT) {
1524 ret = -EBUSY;
1525 break;
1526 }
1527 if (isolate_huge_page(page, &source))
1528 move_pages -= 1 << compound_order(head);
1529 continue;
1530 }
1531
Konstantin Khlebnikov700c2a42011-05-24 17:12:19 -07001532 if (!get_page_unless_zero(page))
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001533 continue;
1534 /*
1535 * We can skip free pages. And we can only deal with pages on
1536 * LRU.
1537 */
Nick Piggin62695a82008-10-18 20:26:09 -07001538 ret = isolate_lru_page(page);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001539 if (!ret) { /* Success */
Konstantin Khlebnikov700c2a42011-05-24 17:12:19 -07001540 put_page(page);
Nick Piggin62695a82008-10-18 20:26:09 -07001541 list_add_tail(&page->lru, &source);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001542 move_pages--;
KOSAKI Motohiro6d9c2852009-12-14 17:58:11 -08001543 inc_zone_page_state(page, NR_ISOLATED_ANON +
1544 page_is_file_cache(page));
1545
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001546 } else {
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001547#ifdef CONFIG_DEBUG_VM
Chen Yuconge33e33b2016-03-17 14:19:35 -07001548 pr_alert("removing pfn %lx from LRU failed\n", pfn);
Dave Hansenf0b791a2014-01-23 15:52:49 -08001549 dump_page(page, "failed to remove from LRU");
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001550#endif
Konstantin Khlebnikov700c2a42011-05-24 17:12:19 -07001551 put_page(page);
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001552 /* Because we don't have big zone->lock. we should
Bob Liu809c4442010-10-26 14:22:10 -07001553 check this again here. */
1554 if (page_count(page)) {
1555 not_managed++;
Bob Liuf3ab2632010-10-26 14:22:10 -07001556 ret = -EBUSY;
Bob Liu809c4442010-10-26 14:22:10 -07001557 break;
1558 }
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001559 }
1560 }
Bob Liuf3ab2632010-10-26 14:22:10 -07001561 if (!list_empty(&source)) {
1562 if (not_managed) {
Naoya Horiguchic8721bb2013-09-11 14:22:09 -07001563 putback_movable_pages(&source);
Bob Liuf3ab2632010-10-26 14:22:10 -07001564 goto out;
1565 }
Minchan Kim74c08f92012-10-08 16:32:54 -07001566
1567 /*
1568 * alloc_migrate_target should be improooooved!!
1569 * migrate_pages returns # of failed pages.
1570 */
David Rientjes68711a72014-06-04 16:08:25 -07001571 ret = migrate_pages(&source, alloc_migrate_target, NULL, 0,
Hugh Dickins9c620e22013-02-22 16:35:14 -08001572 MIGRATE_SYNC, MR_MEMORY_HOTPLUG);
Bob Liuf3ab2632010-10-26 14:22:10 -07001573 if (ret)
Naoya Horiguchic8721bb2013-09-11 14:22:09 -07001574 putback_movable_pages(&source);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001575 }
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001576out:
1577 return ret;
1578}
1579
1580/*
1581 * remove from free_area[] and mark all as Reserved.
1582 */
1583static int
1584offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages,
1585 void *data)
1586{
1587 __offline_isolated_pages(start, start + nr_pages);
1588 return 0;
1589}
1590
1591static void
1592offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
1593{
KAMEZAWA Hiroyuki908eedc2009-09-22 16:45:46 -07001594 walk_system_ram_range(start_pfn, end_pfn - start_pfn, NULL,
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001595 offline_isolated_pages_cb);
1596}
1597
1598/*
1599 * Check all pages in range, recoreded as memory resource, are isolated.
1600 */
1601static int
1602check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages,
1603 void *data)
1604{
1605 int ret;
1606 long offlined = *(long *)data;
Wen Congyangb023f462012-12-11 16:00:45 -08001607 ret = test_pages_isolated(start_pfn, start_pfn + nr_pages, true);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001608 offlined = nr_pages;
1609 if (!ret)
1610 *(long *)data += offlined;
1611 return ret;
1612}
1613
1614static long
1615check_pages_isolated(unsigned long start_pfn, unsigned long end_pfn)
1616{
1617 long offlined = 0;
1618 int ret;
1619
KAMEZAWA Hiroyuki908eedc2009-09-22 16:45:46 -07001620 ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn, &offlined,
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001621 check_pages_isolated_cb);
1622 if (ret < 0)
1623 offlined = (long)ret;
1624 return offlined;
1625}
1626
Lai Jiangshan09285af2012-12-12 13:52:04 -08001627#ifdef CONFIG_MOVABLE_NODE
Tang Chen79a4dce2012-12-18 14:23:24 -08001628/*
1629 * When CONFIG_MOVABLE_NODE, we permit offlining of a node which doesn't have
1630 * normal memory.
1631 */
Lai Jiangshan09285af2012-12-12 13:52:04 -08001632static bool can_offline_normal(struct zone *zone, unsigned long nr_pages)
1633{
1634 return true;
1635}
Tang Chen79a4dce2012-12-18 14:23:24 -08001636#else /* CONFIG_MOVABLE_NODE */
Lai Jiangshan74d42d82012-12-11 16:03:23 -08001637/* ensure the node has NORMAL memory if it is still online */
1638static bool can_offline_normal(struct zone *zone, unsigned long nr_pages)
1639{
1640 struct pglist_data *pgdat = zone->zone_pgdat;
1641 unsigned long present_pages = 0;
1642 enum zone_type zt;
1643
1644 for (zt = 0; zt <= ZONE_NORMAL; zt++)
1645 present_pages += pgdat->node_zones[zt].present_pages;
1646
1647 if (present_pages > nr_pages)
1648 return true;
1649
1650 present_pages = 0;
1651 for (; zt <= ZONE_MOVABLE; zt++)
1652 present_pages += pgdat->node_zones[zt].present_pages;
1653
1654 /*
1655 * we can't offline the last normal memory until all
1656 * higher memory is offlined.
1657 */
1658 return present_pages == 0;
1659}
Tang Chen79a4dce2012-12-18 14:23:24 -08001660#endif /* CONFIG_MOVABLE_NODE */
Lai Jiangshan74d42d82012-12-11 16:03:23 -08001661
Tang Chenc5320922013-11-12 15:08:10 -08001662static int __init cmdline_parse_movable_node(char *p)
1663{
1664#ifdef CONFIG_MOVABLE_NODE
1665 /*
1666 * Memory used by the kernel cannot be hot-removed because Linux
1667 * cannot migrate the kernel pages. When memory hotplug is
1668 * enabled, we should prevent memblock from allocating memory
1669 * for the kernel.
1670 *
1671 * ACPI SRAT records all hotpluggable memory ranges. But before
1672 * SRAT is parsed, we don't know about it.
1673 *
1674 * The kernel image is loaded into memory at very early time. We
1675 * cannot prevent this anyway. So on NUMA system, we set any
1676 * node the kernel resides in as un-hotpluggable.
1677 *
1678 * Since on modern servers, one node could have double-digit
1679 * gigabytes memory, we can assume the memory around the kernel
1680 * image is also un-hotpluggable. So before SRAT is parsed, just
1681 * allocate memory near the kernel image to try the best to keep
1682 * the kernel away from hotpluggable memory.
1683 */
1684 memblock_set_bottom_up(true);
Tang Chen55ac5902014-01-21 15:49:35 -08001685 movable_node_enabled = true;
Tang Chenc5320922013-11-12 15:08:10 -08001686#else
1687 pr_warn("movable_node option not supported\n");
1688#endif
1689 return 0;
1690}
1691early_param("movable_node", cmdline_parse_movable_node);
1692
Lai Jiangshand9713672012-12-11 16:01:03 -08001693/* check which state of node_states will be changed when offline memory */
1694static void node_states_check_changes_offline(unsigned long nr_pages,
1695 struct zone *zone, struct memory_notify *arg)
1696{
1697 struct pglist_data *pgdat = zone->zone_pgdat;
1698 unsigned long present_pages = 0;
1699 enum zone_type zt, zone_last = ZONE_NORMAL;
1700
1701 /*
Lai Jiangshan6715ddf2012-12-12 13:51:49 -08001702 * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY]
1703 * contains nodes which have zones of 0...ZONE_NORMAL,
1704 * set zone_last to ZONE_NORMAL.
Lai Jiangshand9713672012-12-11 16:01:03 -08001705 *
Lai Jiangshan6715ddf2012-12-12 13:51:49 -08001706 * If we don't have HIGHMEM nor movable node,
1707 * node_states[N_NORMAL_MEMORY] contains nodes which have zones of
1708 * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE.
Lai Jiangshand9713672012-12-11 16:01:03 -08001709 */
Lai Jiangshan6715ddf2012-12-12 13:51:49 -08001710 if (N_MEMORY == N_NORMAL_MEMORY)
Lai Jiangshand9713672012-12-11 16:01:03 -08001711 zone_last = ZONE_MOVABLE;
1712
1713 /*
1714 * check whether node_states[N_NORMAL_MEMORY] will be changed.
1715 * If the memory to be offline is in a zone of 0...zone_last,
1716 * and it is the last present memory, 0...zone_last will
1717 * become empty after offline , thus we can determind we will
1718 * need to clear the node from node_states[N_NORMAL_MEMORY].
1719 */
1720 for (zt = 0; zt <= zone_last; zt++)
1721 present_pages += pgdat->node_zones[zt].present_pages;
1722 if (zone_idx(zone) <= zone_last && nr_pages >= present_pages)
1723 arg->status_change_nid_normal = zone_to_nid(zone);
1724 else
1725 arg->status_change_nid_normal = -1;
1726
Lai Jiangshan6715ddf2012-12-12 13:51:49 -08001727#ifdef CONFIG_HIGHMEM
1728 /*
1729 * If we have movable node, node_states[N_HIGH_MEMORY]
1730 * contains nodes which have zones of 0...ZONE_HIGHMEM,
1731 * set zone_last to ZONE_HIGHMEM.
1732 *
1733 * If we don't have movable node, node_states[N_NORMAL_MEMORY]
1734 * contains nodes which have zones of 0...ZONE_MOVABLE,
1735 * set zone_last to ZONE_MOVABLE.
1736 */
1737 zone_last = ZONE_HIGHMEM;
1738 if (N_MEMORY == N_HIGH_MEMORY)
1739 zone_last = ZONE_MOVABLE;
1740
1741 for (; zt <= zone_last; zt++)
1742 present_pages += pgdat->node_zones[zt].present_pages;
1743 if (zone_idx(zone) <= zone_last && nr_pages >= present_pages)
1744 arg->status_change_nid_high = zone_to_nid(zone);
1745 else
1746 arg->status_change_nid_high = -1;
1747#else
1748 arg->status_change_nid_high = arg->status_change_nid_normal;
1749#endif
1750
Lai Jiangshand9713672012-12-11 16:01:03 -08001751 /*
1752 * node_states[N_HIGH_MEMORY] contains nodes which have 0...ZONE_MOVABLE
1753 */
1754 zone_last = ZONE_MOVABLE;
1755
1756 /*
1757 * check whether node_states[N_HIGH_MEMORY] will be changed
1758 * If we try to offline the last present @nr_pages from the node,
1759 * we can determind we will need to clear the node from
1760 * node_states[N_HIGH_MEMORY].
1761 */
1762 for (; zt <= zone_last; zt++)
1763 present_pages += pgdat->node_zones[zt].present_pages;
1764 if (nr_pages >= present_pages)
1765 arg->status_change_nid = zone_to_nid(zone);
1766 else
1767 arg->status_change_nid = -1;
1768}
1769
1770static void node_states_clear_node(int node, struct memory_notify *arg)
1771{
1772 if (arg->status_change_nid_normal >= 0)
1773 node_clear_state(node, N_NORMAL_MEMORY);
1774
Lai Jiangshan6715ddf2012-12-12 13:51:49 -08001775 if ((N_MEMORY != N_NORMAL_MEMORY) &&
1776 (arg->status_change_nid_high >= 0))
Lai Jiangshand9713672012-12-11 16:01:03 -08001777 node_clear_state(node, N_HIGH_MEMORY);
Lai Jiangshan6715ddf2012-12-12 13:51:49 -08001778
1779 if ((N_MEMORY != N_HIGH_MEMORY) &&
1780 (arg->status_change_nid >= 0))
1781 node_clear_state(node, N_MEMORY);
Lai Jiangshand9713672012-12-11 16:01:03 -08001782}
1783
Wen Congyanga16cee12012-10-08 16:33:58 -07001784static int __ref __offline_pages(unsigned long start_pfn,
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001785 unsigned long end_pfn, unsigned long timeout)
1786{
1787 unsigned long pfn, nr_pages, expire;
1788 long offlined_pages;
Yasunori Goto7b78d332007-10-21 16:41:36 -07001789 int ret, drain, retry_max, node;
Cody P Schaferd7029092013-07-03 15:02:11 -07001790 unsigned long flags;
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001791 struct zone *zone;
Yasunori Goto7b78d332007-10-21 16:41:36 -07001792 struct memory_notify arg;
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001793
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001794 /* at least, alignment against pageblock is necessary */
1795 if (!IS_ALIGNED(start_pfn, pageblock_nr_pages))
1796 return -EINVAL;
1797 if (!IS_ALIGNED(end_pfn, pageblock_nr_pages))
1798 return -EINVAL;
1799 /* This makes hotplug much easier...and readable.
1800 we assume this for now. .*/
1801 if (!test_pages_in_a_zone(start_pfn, end_pfn))
1802 return -EINVAL;
Yasunori Goto7b78d332007-10-21 16:41:36 -07001803
1804 zone = page_zone(pfn_to_page(start_pfn));
1805 node = zone_to_nid(zone);
1806 nr_pages = end_pfn - start_pfn;
1807
Lai Jiangshan74d42d82012-12-11 16:03:23 -08001808 if (zone_idx(zone) <= ZONE_NORMAL && !can_offline_normal(zone, nr_pages))
David Rientjes30467e02015-04-14 15:45:11 -07001809 return -EINVAL;
Lai Jiangshan74d42d82012-12-11 16:03:23 -08001810
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001811 /* set above range as isolated */
Wen Congyangb023f462012-12-11 16:00:45 -08001812 ret = start_isolate_page_range(start_pfn, end_pfn,
1813 MIGRATE_MOVABLE, true);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001814 if (ret)
David Rientjes30467e02015-04-14 15:45:11 -07001815 return ret;
Yasunori Goto7b78d332007-10-21 16:41:36 -07001816
1817 arg.start_pfn = start_pfn;
1818 arg.nr_pages = nr_pages;
Lai Jiangshand9713672012-12-11 16:01:03 -08001819 node_states_check_changes_offline(nr_pages, zone, &arg);
Yasunori Goto7b78d332007-10-21 16:41:36 -07001820
1821 ret = memory_notify(MEM_GOING_OFFLINE, &arg);
1822 ret = notifier_to_errno(ret);
1823 if (ret)
1824 goto failed_removal;
1825
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001826 pfn = start_pfn;
1827 expire = jiffies + timeout;
1828 drain = 0;
1829 retry_max = 5;
1830repeat:
1831 /* start memory hot removal */
1832 ret = -EAGAIN;
1833 if (time_after(jiffies, expire))
1834 goto failed_removal;
1835 ret = -EINTR;
1836 if (signal_pending(current))
1837 goto failed_removal;
1838 ret = 0;
1839 if (drain) {
1840 lru_add_drain_all();
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001841 cond_resched();
Vlastimil Babkac0554322014-12-10 15:43:10 -08001842 drain_all_pages(zone);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001843 }
1844
Naoya Horiguchic8721bb2013-09-11 14:22:09 -07001845 pfn = scan_movable_pages(start_pfn, end_pfn);
1846 if (pfn) { /* We have movable pages */
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001847 ret = do_migrate_range(pfn, end_pfn);
1848 if (!ret) {
1849 drain = 1;
1850 goto repeat;
1851 } else {
1852 if (ret < 0)
1853 if (--retry_max == 0)
1854 goto failed_removal;
1855 yield();
1856 drain = 1;
1857 goto repeat;
1858 }
1859 }
Adam Buchbinderb3834be2012-09-19 21:48:02 -04001860 /* drain all zone's lru pagevec, this is asynchronous... */
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001861 lru_add_drain_all();
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001862 yield();
Adam Buchbinderb3834be2012-09-19 21:48:02 -04001863 /* drain pcp pages, this is synchronous. */
Vlastimil Babkac0554322014-12-10 15:43:10 -08001864 drain_all_pages(zone);
Naoya Horiguchic8721bb2013-09-11 14:22:09 -07001865 /*
1866 * dissolve free hugepages in the memory block before doing offlining
1867 * actually in order to make hugetlbfs's object counting consistent.
1868 */
1869 dissolve_free_huge_pages(start_pfn, end_pfn);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001870 /* check again */
1871 offlined_pages = check_pages_isolated(start_pfn, end_pfn);
1872 if (offlined_pages < 0) {
1873 ret = -EBUSY;
1874 goto failed_removal;
1875 }
Chen Yuconge33e33b2016-03-17 14:19:35 -07001876 pr_info("Offlined Pages %ld\n", offlined_pages);
Adam Buchbinderb3834be2012-09-19 21:48:02 -04001877 /* Ok, all of our target is isolated.
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001878 We cannot do rollback at this point. */
1879 offline_isolated_pages(start_pfn, end_pfn);
KAMEZAWA Hiroyukidbc0e4c2007-11-14 16:59:12 -08001880 /* reset pagetype flags and makes migrate type to be MOVABLE */
Michal Nazarewicz0815f3d2012-04-03 15:06:15 +02001881 undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001882 /* removal success */
Jiang Liu3dcc0572013-07-03 15:03:21 -07001883 adjust_managed_page_count(pfn_to_page(start_pfn), -offlined_pages);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001884 zone->present_pages -= offlined_pages;
Cody P Schaferd7029092013-07-03 15:02:11 -07001885
1886 pgdat_resize_lock(zone->zone_pgdat, &flags);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001887 zone->zone_pgdat->node_present_pages -= offlined_pages;
Cody P Schaferd7029092013-07-03 15:02:11 -07001888 pgdat_resize_unlock(zone->zone_pgdat, &flags);
Yasunori Goto7b78d332007-10-21 16:41:36 -07001889
KOSAKI Motohiro1b79acc2011-05-24 17:11:32 -07001890 init_per_zone_wmark_min();
1891
Xishi Qiu1e8537b2012-10-08 16:31:51 -07001892 if (!populated_zone(zone)) {
Jiang Liu340175b2012-07-31 16:43:32 -07001893 zone_pcp_reset(zone);
Xishi Qiu1e8537b2012-10-08 16:31:51 -07001894 mutex_lock(&zonelists_mutex);
1895 build_all_zonelists(NULL, NULL);
1896 mutex_unlock(&zonelists_mutex);
1897 } else
1898 zone_pcp_update(zone);
Jiang Liu340175b2012-07-31 16:43:32 -07001899
Lai Jiangshand9713672012-12-11 16:01:03 -08001900 node_states_clear_node(node, &arg);
Vlastimil Babka698b1b32016-03-17 14:18:08 -07001901 if (arg.status_change_nid >= 0) {
David Rientjes8fe23e02009-12-14 17:58:33 -08001902 kswapd_stop(node);
Vlastimil Babka698b1b32016-03-17 14:18:08 -07001903 kcompactd_stop(node);
1904 }
Minchan Kimbce73942009-06-16 15:32:50 -07001905
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001906 vm_total_pages = nr_free_pagecache_pages();
1907 writeback_set_ratelimit();
Yasunori Goto7b78d332007-10-21 16:41:36 -07001908
1909 memory_notify(MEM_OFFLINE, &arg);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001910 return 0;
1911
1912failed_removal:
Chen Yuconge33e33b2016-03-17 14:19:35 -07001913 pr_debug("memory offlining [mem %#010llx-%#010llx] failed\n",
1914 (unsigned long long) start_pfn << PAGE_SHIFT,
1915 ((unsigned long long) end_pfn << PAGE_SHIFT) - 1);
Yasunori Goto7b78d332007-10-21 16:41:36 -07001916 memory_notify(MEM_CANCEL_OFFLINE, &arg);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001917 /* pushback to free area */
Michal Nazarewicz0815f3d2012-04-03 15:06:15 +02001918 undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001919 return ret;
1920}
Badari Pulavarty71088782008-10-18 20:25:58 -07001921
David Rientjes30467e02015-04-14 15:45:11 -07001922/* Must be protected by mem_hotplug_begin() */
Wen Congyanga16cee12012-10-08 16:33:58 -07001923int offline_pages(unsigned long start_pfn, unsigned long nr_pages)
1924{
1925 return __offline_pages(start_pfn, start_pfn + nr_pages, 120 * HZ);
1926}
Rafael J. Wysockie2ff3942013-05-08 00:29:49 +02001927#endif /* CONFIG_MEMORY_HOTREMOVE */
Wen Congyanga16cee12012-10-08 16:33:58 -07001928
Wen Congyangbbc76be2013-02-22 16:32:54 -08001929/**
1930 * walk_memory_range - walks through all mem sections in [start_pfn, end_pfn)
1931 * @start_pfn: start pfn of the memory range
Toshi Kanie05c4bb2013-04-29 15:06:16 -07001932 * @end_pfn: end pfn of the memory range
Wen Congyangbbc76be2013-02-22 16:32:54 -08001933 * @arg: argument passed to func
1934 * @func: callback for each memory section walked
1935 *
1936 * This function walks through all present mem sections in range
1937 * [start_pfn, end_pfn) and call func on each mem section.
1938 *
1939 * Returns the return value of func.
1940 */
Rafael J. Wysockie2ff3942013-05-08 00:29:49 +02001941int walk_memory_range(unsigned long start_pfn, unsigned long end_pfn,
Wen Congyangbbc76be2013-02-22 16:32:54 -08001942 void *arg, int (*func)(struct memory_block *, void *))
Badari Pulavarty71088782008-10-18 20:25:58 -07001943{
Wen Congyange90bdb72012-10-08 16:34:01 -07001944 struct memory_block *mem = NULL;
1945 struct mem_section *section;
Wen Congyange90bdb72012-10-08 16:34:01 -07001946 unsigned long pfn, section_nr;
1947 int ret;
Badari Pulavarty71088782008-10-18 20:25:58 -07001948
Wen Congyange90bdb72012-10-08 16:34:01 -07001949 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
1950 section_nr = pfn_to_section_nr(pfn);
1951 if (!present_section_nr(section_nr))
1952 continue;
1953
1954 section = __nr_to_section(section_nr);
1955 /* same memblock? */
1956 if (mem)
1957 if ((section_nr >= mem->start_section_nr) &&
1958 (section_nr <= mem->end_section_nr))
1959 continue;
1960
1961 mem = find_memory_block_hinted(section, mem);
1962 if (!mem)
1963 continue;
1964
Wen Congyangbbc76be2013-02-22 16:32:54 -08001965 ret = func(mem, arg);
Wen Congyange90bdb72012-10-08 16:34:01 -07001966 if (ret) {
Wen Congyangbbc76be2013-02-22 16:32:54 -08001967 kobject_put(&mem->dev.kobj);
1968 return ret;
Wen Congyange90bdb72012-10-08 16:34:01 -07001969 }
1970 }
1971
1972 if (mem)
1973 kobject_put(&mem->dev.kobj);
1974
Wen Congyangbbc76be2013-02-22 16:32:54 -08001975 return 0;
1976}
1977
Rafael J. Wysockie2ff3942013-05-08 00:29:49 +02001978#ifdef CONFIG_MEMORY_HOTREMOVE
Xishi Qiud6de9d52013-11-12 15:07:20 -08001979static int check_memblock_offlined_cb(struct memory_block *mem, void *arg)
Wen Congyangbbc76be2013-02-22 16:32:54 -08001980{
1981 int ret = !is_memblock_offlined(mem);
1982
Randy Dunlap349daa02013-04-29 15:08:49 -07001983 if (unlikely(ret)) {
1984 phys_addr_t beginpa, endpa;
1985
1986 beginpa = PFN_PHYS(section_nr_to_pfn(mem->start_section_nr));
1987 endpa = PFN_PHYS(section_nr_to_pfn(mem->end_section_nr + 1))-1;
Joe Perches756a0252016-03-17 14:19:47 -07001988 pr_warn("removing memory fails, because memory [%pa-%pa] is onlined\n",
Randy Dunlap349daa02013-04-29 15:08:49 -07001989 &beginpa, &endpa);
1990 }
Wen Congyangbbc76be2013-02-22 16:32:54 -08001991
1992 return ret;
1993}
1994
Toshi Kani0f1cfe92013-09-11 14:21:50 -07001995static int check_cpu_on_node(pg_data_t *pgdat)
Tang Chen60a5a192013-02-22 16:33:14 -08001996{
Tang Chen60a5a192013-02-22 16:33:14 -08001997 int cpu;
1998
1999 for_each_present_cpu(cpu) {
2000 if (cpu_to_node(cpu) == pgdat->node_id)
2001 /*
2002 * the cpu on this node isn't removed, and we can't
2003 * offline this node.
2004 */
2005 return -EBUSY;
2006 }
2007
2008 return 0;
2009}
2010
Toshi Kani0f1cfe92013-09-11 14:21:50 -07002011static void unmap_cpu_on_node(pg_data_t *pgdat)
Wen Congyange13fe862013-02-22 16:33:31 -08002012{
2013#ifdef CONFIG_ACPI_NUMA
Wen Congyange13fe862013-02-22 16:33:31 -08002014 int cpu;
2015
2016 for_each_possible_cpu(cpu)
2017 if (cpu_to_node(cpu) == pgdat->node_id)
2018 numa_clear_node(cpu);
2019#endif
2020}
2021
Toshi Kani0f1cfe92013-09-11 14:21:50 -07002022static int check_and_unmap_cpu_on_node(pg_data_t *pgdat)
Wen Congyange13fe862013-02-22 16:33:31 -08002023{
Toshi Kani0f1cfe92013-09-11 14:21:50 -07002024 int ret;
Wen Congyange13fe862013-02-22 16:33:31 -08002025
Toshi Kani0f1cfe92013-09-11 14:21:50 -07002026 ret = check_cpu_on_node(pgdat);
Wen Congyange13fe862013-02-22 16:33:31 -08002027 if (ret)
2028 return ret;
2029
2030 /*
2031 * the node will be offlined when we come here, so we can clear
2032 * the cpu_to_node() now.
2033 */
2034
Toshi Kani0f1cfe92013-09-11 14:21:50 -07002035 unmap_cpu_on_node(pgdat);
Wen Congyange13fe862013-02-22 16:33:31 -08002036 return 0;
2037}
2038
Toshi Kani0f1cfe92013-09-11 14:21:50 -07002039/**
2040 * try_offline_node
2041 *
2042 * Offline a node if all memory sections and cpus of the node are removed.
2043 *
2044 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
2045 * and online/offline operations before this call.
2046 */
Wen Congyang90b30cd2013-02-22 16:33:27 -08002047void try_offline_node(int nid)
Tang Chen60a5a192013-02-22 16:33:14 -08002048{
Wen Congyangd822b862013-02-22 16:33:16 -08002049 pg_data_t *pgdat = NODE_DATA(nid);
2050 unsigned long start_pfn = pgdat->node_start_pfn;
2051 unsigned long end_pfn = start_pfn + pgdat->node_spanned_pages;
Tang Chen60a5a192013-02-22 16:33:14 -08002052 unsigned long pfn;
Wen Congyangd822b862013-02-22 16:33:16 -08002053 int i;
Tang Chen60a5a192013-02-22 16:33:14 -08002054
2055 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
2056 unsigned long section_nr = pfn_to_section_nr(pfn);
2057
2058 if (!present_section_nr(section_nr))
2059 continue;
2060
2061 if (pfn_to_nid(pfn) != nid)
2062 continue;
2063
2064 /*
2065 * some memory sections of this node are not removed, and we
2066 * can't offline node now.
2067 */
2068 return;
2069 }
2070
Toshi Kani0f1cfe92013-09-11 14:21:50 -07002071 if (check_and_unmap_cpu_on_node(pgdat))
Tang Chen60a5a192013-02-22 16:33:14 -08002072 return;
2073
2074 /*
2075 * all memory/cpu of this node are removed, we can offline this
2076 * node now.
2077 */
2078 node_set_offline(nid);
2079 unregister_one_node(nid);
Wen Congyangd822b862013-02-22 16:33:16 -08002080
Wen Congyangd822b862013-02-22 16:33:16 -08002081 /* free waittable in each zone */
2082 for (i = 0; i < MAX_NR_ZONES; i++) {
2083 struct zone *zone = pgdat->node_zones + i;
2084
Jianguo Wuca4b3f32013-03-22 15:04:50 -07002085 /*
2086 * wait_table may be allocated from boot memory,
2087 * here only free if it's allocated by vmalloc.
2088 */
Gu Zheng85bd8392015-06-10 11:14:43 -07002089 if (is_vmalloc_addr(zone->wait_table)) {
Wen Congyangd822b862013-02-22 16:33:16 -08002090 vfree(zone->wait_table);
Gu Zheng85bd8392015-06-10 11:14:43 -07002091 zone->wait_table = NULL;
2092 }
Wen Congyangd822b862013-02-22 16:33:16 -08002093 }
Tang Chen60a5a192013-02-22 16:33:14 -08002094}
Wen Congyang90b30cd2013-02-22 16:33:27 -08002095EXPORT_SYMBOL(try_offline_node);
Tang Chen60a5a192013-02-22 16:33:14 -08002096
Toshi Kani0f1cfe92013-09-11 14:21:50 -07002097/**
2098 * remove_memory
2099 *
2100 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
2101 * and online/offline operations before this call, as required by
2102 * try_offline_node().
2103 */
Rafael J. Wysocki242831e2013-05-27 12:58:46 +02002104void __ref remove_memory(int nid, u64 start, u64 size)
Wen Congyangbbc76be2013-02-22 16:32:54 -08002105{
Rafael J. Wysocki242831e2013-05-27 12:58:46 +02002106 int ret;
Wen Congyang993c1aa2013-02-22 16:32:50 -08002107
Toshi Kani27356f52013-09-11 14:21:49 -07002108 BUG_ON(check_hotplug_memory_range(start, size));
2109
Vladimir Davydovbfc8c902014-06-04 16:07:18 -07002110 mem_hotplug_begin();
Yasuaki Ishimatsu6677e3e2013-02-22 16:32:52 -08002111
2112 /*
Rafael J. Wysocki242831e2013-05-27 12:58:46 +02002113 * All memory blocks must be offlined before removing memory. Check
2114 * whether all memory blocks in question are offline and trigger a BUG()
2115 * if this is not the case.
Yasuaki Ishimatsu6677e3e2013-02-22 16:32:52 -08002116 */
Rafael J. Wysocki242831e2013-05-27 12:58:46 +02002117 ret = walk_memory_range(PFN_DOWN(start), PFN_UP(start + size - 1), NULL,
Xishi Qiud6de9d52013-11-12 15:07:20 -08002118 check_memblock_offlined_cb);
Vladimir Davydovbfc8c902014-06-04 16:07:18 -07002119 if (ret)
Rafael J. Wysocki242831e2013-05-27 12:58:46 +02002120 BUG();
Yasuaki Ishimatsu6677e3e2013-02-22 16:32:52 -08002121
Yasuaki Ishimatsu46c66c42013-02-22 16:32:56 -08002122 /* remove memmap entry */
2123 firmware_map_remove(start, start + size, "System RAM");
Xishi Qiuf9126ab2015-08-14 15:35:16 -07002124 memblock_free(start, size);
2125 memblock_remove(start, size);
Yasuaki Ishimatsu46c66c42013-02-22 16:32:56 -08002126
Wen Congyang24d335c2013-02-22 16:32:58 -08002127 arch_remove_memory(start, size);
2128
Tang Chen60a5a192013-02-22 16:33:14 -08002129 try_offline_node(nid);
2130
Vladimir Davydovbfc8c902014-06-04 16:07:18 -07002131 mem_hotplug_done();
Badari Pulavarty71088782008-10-18 20:25:58 -07002132}
Badari Pulavarty71088782008-10-18 20:25:58 -07002133EXPORT_SYMBOL_GPL(remove_memory);
Rafael J. Wysockiaba6efc2013-06-01 22:24:07 +02002134#endif /* CONFIG_MEMORY_HOTREMOVE */