blob: dd186c1a5d53f9ebd27de4c1bedcaac471d6c93e [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>
12#include <linux/bootmem.h>
13#include <linux/compiler.h>
14#include <linux/module.h>
15#include <linux/pagevec.h>
Chandra Seetharaman2d1d43f2006-09-29 02:01:25 -070016#include <linux/writeback.h>
Dave Hansen3947be12005-10-29 18:16:54 -070017#include <linux/slab.h>
18#include <linux/sysctl.h>
19#include <linux/cpu.h>
20#include <linux/memory.h>
21#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>
Dave Hansen3947be12005-10-29 18:16:54 -070032
33#include <asm/tlbflush.h>
34
Adrian Bunk1e5ad9a2008-04-28 20:40:08 +030035#include "internal.h"
36
Keith Mannthey45e0b782006-09-30 23:27:09 -070037/* add this memory to iomem resource */
38static struct resource *register_memory_resource(u64 start, u64 size)
39{
40 struct resource *res;
41 res = kzalloc(sizeof(struct resource), GFP_KERNEL);
42 BUG_ON(!res);
43
44 res->name = "System RAM";
45 res->start = start;
46 res->end = start + size - 1;
Yasunori Goto887c3cb2007-11-14 16:59:20 -080047 res->flags = IORESOURCE_MEM | IORESOURCE_BUSY;
Keith Mannthey45e0b782006-09-30 23:27:09 -070048 if (request_resource(&iomem_resource, res) < 0) {
49 printk("System RAM resource %llx - %llx cannot be added\n",
50 (unsigned long long)res->start, (unsigned long long)res->end);
51 kfree(res);
52 res = NULL;
53 }
54 return res;
55}
56
57static void release_memory_resource(struct resource *res)
58{
59 if (!res)
60 return;
61 release_resource(res);
62 kfree(res);
63 return;
64}
65
Keith Mannthey53947022006-09-30 23:27:08 -070066#ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
Yasunori Goto04753272008-04-28 02:13:31 -070067#ifndef CONFIG_SPARSEMEM_VMEMMAP
Yasunori Gotoaf370fb2008-07-23 21:28:17 -070068static void get_page_bootmem(unsigned long info, struct page *page, int type)
Yasunori Goto04753272008-04-28 02:13:31 -070069{
Yasunori Gotoaf370fb2008-07-23 21:28:17 -070070 atomic_set(&page->_mapcount, type);
Yasunori Goto04753272008-04-28 02:13:31 -070071 SetPagePrivate(page);
72 set_page_private(page, info);
73 atomic_inc(&page->_count);
74}
75
Rakib Mullick23ce9322009-12-14 17:59:44 -080076/* reference to __meminit __free_pages_bootmem is valid
77 * so use __ref to tell modpost not to generate a warning */
78void __ref put_page_bootmem(struct page *page)
Yasunori Goto04753272008-04-28 02:13:31 -070079{
Yasunori Gotoaf370fb2008-07-23 21:28:17 -070080 int type;
Yasunori Goto04753272008-04-28 02:13:31 -070081
Yasunori Gotoaf370fb2008-07-23 21:28:17 -070082 type = atomic_read(&page->_mapcount);
83 BUG_ON(type >= -1);
Yasunori Goto04753272008-04-28 02:13:31 -070084
85 if (atomic_dec_return(&page->_count) == 1) {
86 ClearPagePrivate(page);
87 set_page_private(page, 0);
88 reset_page_mapcount(page);
89 __free_pages_bootmem(page, 0);
90 }
91
92}
93
Adrian Bunkd92bc312008-07-23 21:28:12 -070094static void register_page_bootmem_info_section(unsigned long start_pfn)
Yasunori Goto04753272008-04-28 02:13:31 -070095{
96 unsigned long *usemap, mapsize, section_nr, i;
97 struct mem_section *ms;
98 struct page *page, *memmap;
99
100 if (!pfn_valid(start_pfn))
101 return;
102
103 section_nr = pfn_to_section_nr(start_pfn);
104 ms = __nr_to_section(section_nr);
105
106 /* Get section's memmap address */
107 memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
108
109 /*
110 * Get page for the memmap's phys address
111 * XXX: need more consideration for sparse_vmemmap...
112 */
113 page = virt_to_page(memmap);
114 mapsize = sizeof(struct page) * PAGES_PER_SECTION;
115 mapsize = PAGE_ALIGN(mapsize) >> PAGE_SHIFT;
116
117 /* remember memmap's page */
118 for (i = 0; i < mapsize; i++, page++)
119 get_page_bootmem(section_nr, page, SECTION_INFO);
120
121 usemap = __nr_to_section(section_nr)->pageblock_flags;
122 page = virt_to_page(usemap);
123
124 mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT;
125
126 for (i = 0; i < mapsize; i++, page++)
Yasunori Gotoaf370fb2008-07-23 21:28:17 -0700127 get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
Yasunori Goto04753272008-04-28 02:13:31 -0700128
129}
130
131void register_page_bootmem_info_node(struct pglist_data *pgdat)
132{
133 unsigned long i, pfn, end_pfn, nr_pages;
134 int node = pgdat->node_id;
135 struct page *page;
136 struct zone *zone;
137
138 nr_pages = PAGE_ALIGN(sizeof(struct pglist_data)) >> PAGE_SHIFT;
139 page = virt_to_page(pgdat);
140
141 for (i = 0; i < nr_pages; i++, page++)
142 get_page_bootmem(node, page, NODE_INFO);
143
144 zone = &pgdat->node_zones[0];
145 for (; zone < pgdat->node_zones + MAX_NR_ZONES - 1; zone++) {
146 if (zone->wait_table) {
147 nr_pages = zone->wait_table_hash_nr_entries
148 * sizeof(wait_queue_head_t);
149 nr_pages = PAGE_ALIGN(nr_pages) >> PAGE_SHIFT;
150 page = virt_to_page(zone->wait_table);
151
152 for (i = 0; i < nr_pages; i++, page++)
153 get_page_bootmem(node, page, NODE_INFO);
154 }
155 }
156
157 pfn = pgdat->node_start_pfn;
158 end_pfn = pfn + pgdat->node_spanned_pages;
159
160 /* register_section info */
161 for (; pfn < end_pfn; pfn += PAGES_PER_SECTION)
162 register_page_bootmem_info_section(pfn);
163
164}
165#endif /* !CONFIG_SPARSEMEM_VMEMMAP */
166
Heiko Carstens76cdd582008-05-14 16:05:52 -0700167static void grow_zone_span(struct zone *zone, unsigned long start_pfn,
168 unsigned long end_pfn)
169{
170 unsigned long old_zone_end_pfn;
171
172 zone_span_writelock(zone);
173
174 old_zone_end_pfn = zone->zone_start_pfn + zone->spanned_pages;
175 if (start_pfn < zone->zone_start_pfn)
176 zone->zone_start_pfn = start_pfn;
177
178 zone->spanned_pages = max(old_zone_end_pfn, end_pfn) -
179 zone->zone_start_pfn;
180
181 zone_span_writeunlock(zone);
182}
183
184static void grow_pgdat_span(struct pglist_data *pgdat, unsigned long start_pfn,
185 unsigned long end_pfn)
186{
187 unsigned long old_pgdat_end_pfn =
188 pgdat->node_start_pfn + pgdat->node_spanned_pages;
189
190 if (start_pfn < pgdat->node_start_pfn)
191 pgdat->node_start_pfn = start_pfn;
192
193 pgdat->node_spanned_pages = max(old_pgdat_end_pfn, end_pfn) -
194 pgdat->node_start_pfn;
195}
196
Al Viro31168482008-11-22 17:33:24 +0000197static int __meminit __add_zone(struct zone *zone, unsigned long phys_start_pfn)
Dave Hansen3947be12005-10-29 18:16:54 -0700198{
199 struct pglist_data *pgdat = zone->zone_pgdat;
200 int nr_pages = PAGES_PER_SECTION;
201 int nid = pgdat->node_id;
202 int zone_type;
Heiko Carstens76cdd582008-05-14 16:05:52 -0700203 unsigned long flags;
Dave Hansen3947be12005-10-29 18:16:54 -0700204
205 zone_type = zone - pgdat->node_zones;
Heiko Carstens76cdd582008-05-14 16:05:52 -0700206 if (!zone->wait_table) {
207 int ret;
208
209 ret = init_currently_empty_zone(zone, phys_start_pfn,
210 nr_pages, MEMMAP_HOTPLUG);
211 if (ret)
212 return ret;
213 }
214 pgdat_resize_lock(zone->zone_pgdat, &flags);
215 grow_zone_span(zone, phys_start_pfn, phys_start_pfn + nr_pages);
216 grow_pgdat_span(zone->zone_pgdat, phys_start_pfn,
217 phys_start_pfn + nr_pages);
218 pgdat_resize_unlock(zone->zone_pgdat, &flags);
Dave Hansena2f3aa022007-01-10 23:15:30 -0800219 memmap_init_zone(nr_pages, nid, zone_type,
220 phys_start_pfn, MEMMAP_HOTPLUG);
Yasunori Goto718127c2006-06-23 02:03:10 -0700221 return 0;
Dave Hansen3947be12005-10-29 18:16:54 -0700222}
223
Gary Hadec04fc582009-01-06 14:39:14 -0800224static int __meminit __add_section(int nid, struct zone *zone,
225 unsigned long phys_start_pfn)
Dave Hansen3947be12005-10-29 18:16:54 -0700226{
Dave Hansen3947be12005-10-29 18:16:54 -0700227 int nr_pages = PAGES_PER_SECTION;
Dave Hansen3947be12005-10-29 18:16:54 -0700228 int ret;
229
KAMEZAWA Hiroyukiebd15302006-08-05 12:15:06 -0700230 if (pfn_valid(phys_start_pfn))
231 return -EEXIST;
232
Dave Hansen0b0acbe2005-10-29 18:16:55 -0700233 ret = sparse_add_one_section(zone, phys_start_pfn, nr_pages);
Dave Hansen3947be12005-10-29 18:16:54 -0700234
235 if (ret < 0)
236 return ret;
237
Yasunori Goto718127c2006-06-23 02:03:10 -0700238 ret = __add_zone(zone, phys_start_pfn);
239
240 if (ret < 0)
241 return ret;
242
Gary Hadec04fc582009-01-06 14:39:14 -0800243 return register_new_memory(nid, __pfn_to_section(phys_start_pfn));
Dave Hansen3947be12005-10-29 18:16:54 -0700244}
245
Yasunori Goto0c0a4a52008-04-28 02:13:34 -0700246#ifdef CONFIG_SPARSEMEM_VMEMMAP
247static int __remove_section(struct zone *zone, struct mem_section *ms)
248{
249 /*
250 * XXX: Freeing memmap with vmemmap is not implement yet.
251 * This should be removed later.
252 */
253 return -EBUSY;
254}
255#else
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700256static int __remove_section(struct zone *zone, struct mem_section *ms)
257{
258 unsigned long flags;
259 struct pglist_data *pgdat = zone->zone_pgdat;
260 int ret = -EINVAL;
261
262 if (!valid_section(ms))
263 return ret;
264
265 ret = unregister_memory_section(ms);
266 if (ret)
267 return ret;
268
269 pgdat_resize_lock(pgdat, &flags);
270 sparse_remove_one_section(zone, ms);
271 pgdat_resize_unlock(pgdat, &flags);
272 return 0;
273}
Yasunori Goto0c0a4a52008-04-28 02:13:34 -0700274#endif
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700275
Dave Hansen3947be12005-10-29 18:16:54 -0700276/*
277 * Reasonably generic function for adding memory. It is
278 * expected that archs that support memory hotplug will
279 * call this function after deciding the zone to which to
280 * add the new pages.
281 */
Gary Hadec04fc582009-01-06 14:39:14 -0800282int __ref __add_pages(int nid, struct zone *zone, unsigned long phys_start_pfn,
283 unsigned long nr_pages)
Dave Hansen3947be12005-10-29 18:16:54 -0700284{
285 unsigned long i;
286 int err = 0;
KAMEZAWA Hiroyuki6f712712006-08-05 12:14:58 -0700287 int start_sec, end_sec;
288 /* during initialize mem_map, align hot-added range to section */
289 start_sec = pfn_to_section_nr(phys_start_pfn);
290 end_sec = pfn_to_section_nr(phys_start_pfn + nr_pages - 1);
Dave Hansen3947be12005-10-29 18:16:54 -0700291
KAMEZAWA Hiroyuki6f712712006-08-05 12:14:58 -0700292 for (i = start_sec; i <= end_sec; i++) {
Gary Hadec04fc582009-01-06 14:39:14 -0800293 err = __add_section(nid, zone, i << PFN_SECTION_SHIFT);
Dave Hansen3947be12005-10-29 18:16:54 -0700294
KAMEZAWA Hiroyuki6f712712006-08-05 12:14:58 -0700295 /*
Simon Arlott183ff222007-10-20 01:27:18 +0200296 * EEXIST is finally dealt with by ioresource collision
KAMEZAWA Hiroyuki6f712712006-08-05 12:14:58 -0700297 * check. see add_memory() => register_memory_resource()
298 * Warning will be printed if there is collision.
Joel H Schoppbed120c2006-05-01 12:16:11 -0700299 */
300 if (err && (err != -EEXIST))
Dave Hansen3947be12005-10-29 18:16:54 -0700301 break;
KAMEZAWA Hiroyuki6f712712006-08-05 12:14:58 -0700302 err = 0;
Dave Hansen3947be12005-10-29 18:16:54 -0700303 }
304
305 return err;
306}
Joel H Schoppbed120c2006-05-01 12:16:11 -0700307EXPORT_SYMBOL_GPL(__add_pages);
Dave Hansen3947be12005-10-29 18:16:54 -0700308
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700309/**
310 * __remove_pages() - remove sections of pages from a zone
311 * @zone: zone from which pages need to be removed
312 * @phys_start_pfn: starting pageframe (must be aligned to start of a section)
313 * @nr_pages: number of pages to remove (must be multiple of section size)
314 *
315 * Generic helper function to remove section mappings and sysfs entries
316 * for the section of the memory we are removing. Caller needs to make
317 * sure that pages are marked reserved and zones are adjust properly by
318 * calling offline_pages().
319 */
320int __remove_pages(struct zone *zone, unsigned long phys_start_pfn,
321 unsigned long nr_pages)
322{
323 unsigned long i, ret = 0;
324 int sections_to_remove;
325
326 /*
327 * We can only remove entire sections
328 */
329 BUG_ON(phys_start_pfn & ~PAGE_SECTION_MASK);
330 BUG_ON(nr_pages % PAGES_PER_SECTION);
331
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700332 sections_to_remove = nr_pages / PAGES_PER_SECTION;
333 for (i = 0; i < sections_to_remove; i++) {
334 unsigned long pfn = phys_start_pfn + i*PAGES_PER_SECTION;
Nathan Fontenotde7f0cb2008-10-18 20:27:14 -0700335 release_mem_region(pfn << PAGE_SHIFT,
336 PAGES_PER_SECTION << PAGE_SHIFT);
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700337 ret = __remove_section(zone, __pfn_to_section(pfn));
338 if (ret)
339 break;
340 }
341 return ret;
342}
343EXPORT_SYMBOL_GPL(__remove_pages);
344
Jeremy Fitzhardinge180c06e2008-04-28 02:12:03 -0700345void online_page(struct page *page)
346{
Jan Beulich4738e1b2009-09-21 17:03:03 -0700347 unsigned long pfn = page_to_pfn(page);
348
Jeremy Fitzhardinge180c06e2008-04-28 02:12:03 -0700349 totalram_pages++;
Jan Beulich4738e1b2009-09-21 17:03:03 -0700350 if (pfn >= num_physpages)
351 num_physpages = pfn + 1;
Jeremy Fitzhardinge180c06e2008-04-28 02:12:03 -0700352
353#ifdef CONFIG_HIGHMEM
354 if (PageHighMem(page))
355 totalhigh_pages++;
356#endif
357
358#ifdef CONFIG_FLATMEM
359 max_mapnr = max(page_to_pfn(page), max_mapnr);
360#endif
361
362 ClearPageReserved(page);
363 init_page_count(page);
364 __free_page(page);
365}
366
KAMEZAWA Hiroyuki75884fb2007-10-16 01:26:10 -0700367static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages,
368 void *arg)
Dave Hansen3947be12005-10-29 18:16:54 -0700369{
370 unsigned long i;
KAMEZAWA Hiroyuki75884fb2007-10-16 01:26:10 -0700371 unsigned long onlined_pages = *(unsigned long *)arg;
372 struct page *page;
373 if (PageReserved(pfn_to_page(start_pfn)))
374 for (i = 0; i < nr_pages; i++) {
375 page = pfn_to_page(start_pfn + i);
376 online_page(page);
377 onlined_pages++;
378 }
379 *(unsigned long *)arg = onlined_pages;
380 return 0;
381}
382
383
384int online_pages(unsigned long pfn, unsigned long nr_pages)
385{
Dave Hansen3947be12005-10-29 18:16:54 -0700386 unsigned long onlined_pages = 0;
387 struct zone *zone;
Yasunori Goto68113782006-06-23 02:03:11 -0700388 int need_zonelists_rebuild = 0;
Yasunori Goto7b78d332007-10-21 16:41:36 -0700389 int nid;
390 int ret;
391 struct memory_notify arg;
Dave Hansen3947be12005-10-29 18:16:54 -0700392
Yasunori Goto7b78d332007-10-21 16:41:36 -0700393 arg.start_pfn = pfn;
394 arg.nr_pages = nr_pages;
395 arg.status_change_nid = -1;
396
397 nid = page_to_nid(pfn_to_page(pfn));
398 if (node_present_pages(nid) == 0)
399 arg.status_change_nid = nid;
400
401 ret = memory_notify(MEM_GOING_ONLINE, &arg);
402 ret = notifier_to_errno(ret);
403 if (ret) {
404 memory_notify(MEM_CANCEL_ONLINE, &arg);
405 return ret;
406 }
Dave Hansen3947be12005-10-29 18:16:54 -0700407 /*
408 * This doesn't need a lock to do pfn_to_page().
409 * The section can't be removed here because of the
Daniel Walkerda19cbc2008-02-04 23:35:47 -0800410 * memory_block->state_mutex.
Dave Hansen3947be12005-10-29 18:16:54 -0700411 */
412 zone = page_zone(pfn_to_page(pfn));
Yasunori Goto68113782006-06-23 02:03:11 -0700413 /*
414 * If this zone is not populated, then it is not in zonelist.
415 * This means the page allocator ignores this zone.
416 * So, zonelist must be updated after online.
417 */
Haicheng Li4eaf3f62010-05-24 14:32:52 -0700418 mutex_lock(&zonelists_mutex);
Yasunori Goto68113782006-06-23 02:03:11 -0700419 if (!populated_zone(zone))
420 need_zonelists_rebuild = 1;
421
KAMEZAWA Hiroyuki908eedc2009-09-22 16:45:46 -0700422 ret = walk_system_ram_range(pfn, nr_pages, &onlined_pages,
KAMEZAWA Hiroyuki75884fb2007-10-16 01:26:10 -0700423 online_pages_range);
Geoff Levandfd8a4222008-05-14 16:05:50 -0700424 if (ret) {
Haicheng Li4eaf3f62010-05-24 14:32:52 -0700425 mutex_unlock(&zonelists_mutex);
Geoff Levandfd8a4222008-05-14 16:05:50 -0700426 printk(KERN_DEBUG "online_pages %lx at %lx failed\n",
427 nr_pages, pfn);
428 memory_notify(MEM_CANCEL_ONLINE, &arg);
429 return ret;
430 }
431
Dave Hansen3947be12005-10-29 18:16:54 -0700432 zone->present_pages += onlined_pages;
Yasunori Gotof2937be2006-03-09 17:33:51 -0800433 zone->zone_pgdat->node_present_pages += onlined_pages;
Haicheng Li1f522502010-05-24 14:32:51 -0700434 if (need_zonelists_rebuild)
435 build_all_zonelists(zone);
436 else
437 zone_pcp_update(zone);
Dave Hansen3947be12005-10-29 18:16:54 -0700438
Haicheng Li4eaf3f62010-05-24 14:32:52 -0700439 mutex_unlock(&zonelists_mutex);
Minchan Kimbc75d332009-06-16 15:32:48 -0700440 setup_per_zone_wmarks();
Minchan Kimbce73942009-06-16 15:32:50 -0700441 calculate_zone_inactive_ratio(zone);
Christoph Lameter7ea15302007-10-16 01:25:29 -0700442 if (onlined_pages) {
443 kswapd_run(zone_to_nid(zone));
444 node_set_state(zone_to_nid(zone), N_HIGH_MEMORY);
445 }
Dave Hansen61b13992005-10-29 18:16:56 -0700446
Haicheng Li1f522502010-05-24 14:32:51 -0700447 vm_total_pages = nr_free_pagecache_pages();
Kent Liu2f7f24e2008-07-23 21:28:18 -0700448
Chandra Seetharaman2d1d43f2006-09-29 02:01:25 -0700449 writeback_set_ratelimit();
Yasunori Goto7b78d332007-10-21 16:41:36 -0700450
451 if (onlined_pages)
452 memory_notify(MEM_ONLINE, &arg);
453
Dave Hansen3947be12005-10-29 18:16:54 -0700454 return 0;
455}
Keith Mannthey53947022006-09-30 23:27:08 -0700456#endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
Yasunori Gotobc02af92006-06-27 02:53:30 -0700457
Hidetoshi Setoe1319332009-11-17 14:06:18 -0800458/* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
459static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start)
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700460{
461 struct pglist_data *pgdat;
462 unsigned long zones_size[MAX_NR_ZONES] = {0};
463 unsigned long zholes_size[MAX_NR_ZONES] = {0};
464 unsigned long start_pfn = start >> PAGE_SHIFT;
465
466 pgdat = arch_alloc_nodedata(nid);
467 if (!pgdat)
468 return NULL;
469
470 arch_refresh_nodedata(nid, pgdat);
471
472 /* we can use NODE_DATA(nid) from here */
473
474 /* init node's zones as empty zones, we don't have any present pages.*/
Johannes Weiner9109fb72008-07-23 21:27:20 -0700475 free_area_init_node(nid, zones_size, start_pfn, zholes_size);
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700476
477 return pgdat;
478}
479
480static void rollback_node_hotadd(int nid, pg_data_t *pgdat)
481{
482 arch_refresh_nodedata(nid, NULL);
483 arch_free_nodedata(pgdat);
484 return;
485}
486
KAMEZAWA Hiroyuki0a547032006-06-27 02:53:35 -0700487
minskey guocf234222010-05-24 14:32:41 -0700488/*
489 * called by cpu_up() to online a node without onlined memory.
490 */
491int mem_online_node(int nid)
492{
493 pg_data_t *pgdat;
494 int ret;
495
496 lock_system_sleep();
497 pgdat = hotadd_new_pgdat(nid, 0);
498 if (pgdat) {
499 ret = -ENOMEM;
500 goto out;
501 }
502 node_set_online(nid);
503 ret = register_one_node(nid);
504 BUG_ON(ret);
505
506out:
507 unlock_system_sleep();
508 return ret;
509}
510
Al Viro31168482008-11-22 17:33:24 +0000511/* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
512int __ref add_memory(int nid, u64 start, u64 size)
Yasunori Gotobc02af92006-06-27 02:53:30 -0700513{
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700514 pg_data_t *pgdat = NULL;
515 int new_pgdat = 0;
KAMEZAWA Hiroyukiebd15302006-08-05 12:15:06 -0700516 struct resource *res;
Yasunori Gotobc02af92006-06-27 02:53:30 -0700517 int ret;
518
Andi Kleen6ad696d2009-11-17 14:06:22 -0800519 lock_system_sleep();
520
KAMEZAWA Hiroyukiebd15302006-08-05 12:15:06 -0700521 res = register_memory_resource(start, size);
Andi Kleen6ad696d2009-11-17 14:06:22 -0800522 ret = -EEXIST;
KAMEZAWA Hiroyukiebd15302006-08-05 12:15:06 -0700523 if (!res)
Andi Kleen6ad696d2009-11-17 14:06:22 -0800524 goto out;
KAMEZAWA Hiroyukiebd15302006-08-05 12:15:06 -0700525
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700526 if (!node_online(nid)) {
527 pgdat = hotadd_new_pgdat(nid, start);
Andi Kleen6ad696d2009-11-17 14:06:22 -0800528 ret = -ENOMEM;
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700529 if (!pgdat)
Andi Kleen6ad696d2009-11-17 14:06:22 -0800530 goto out;
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700531 new_pgdat = 1;
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700532 }
533
Yasunori Gotobc02af92006-06-27 02:53:30 -0700534 /* call arch's memory hotadd */
535 ret = arch_add_memory(nid, start, size);
536
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700537 if (ret < 0)
538 goto error;
539
Yasunori Goto0fc44152006-06-27 02:53:38 -0700540 /* we online node here. we can't roll back from here. */
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700541 node_set_online(nid);
542
Yasunori Goto0fc44152006-06-27 02:53:38 -0700543 if (new_pgdat) {
544 ret = register_one_node(nid);
545 /*
546 * If sysfs file of new node can't create, cpu on the node
547 * can't be hot-added. There is no rollback way now.
548 * So, check by BUG_ON() to catch it reluctantly..
549 */
550 BUG_ON(ret);
551 }
552
akpm@linux-foundation.orgd96ae532010-03-05 13:41:58 -0800553 /* create new memmap entry */
554 firmware_map_add_hotplug(start, start + size, "System RAM");
555
Andi Kleen6ad696d2009-11-17 14:06:22 -0800556 goto out;
557
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700558error:
559 /* rollback pgdat allocation and others */
560 if (new_pgdat)
561 rollback_node_hotadd(nid, pgdat);
KAMEZAWA Hiroyukiebd15302006-08-05 12:15:06 -0700562 if (res)
563 release_memory_resource(res);
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700564
Andi Kleen6ad696d2009-11-17 14:06:22 -0800565out:
566 unlock_system_sleep();
Yasunori Gotobc02af92006-06-27 02:53:30 -0700567 return ret;
568}
569EXPORT_SYMBOL_GPL(add_memory);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700570
571#ifdef CONFIG_MEMORY_HOTREMOVE
572/*
Badari Pulavarty5c755e92008-07-23 21:28:19 -0700573 * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy
574 * set and the size of the free page is given by page_order(). Using this,
575 * the function determines if the pageblock contains only free pages.
576 * Due to buddy contraints, a free page at least the size of a pageblock will
577 * be located at the start of the pageblock
578 */
579static inline int pageblock_free(struct page *page)
580{
581 return PageBuddy(page) && page_order(page) >= pageblock_order;
582}
583
584/* Return the start of the next active pageblock after a given page */
585static struct page *next_active_pageblock(struct page *page)
586{
Badari Pulavarty5c755e92008-07-23 21:28:19 -0700587 /* Ensure the starting page is pageblock-aligned */
588 BUG_ON(page_to_pfn(page) & (pageblock_nr_pages - 1));
589
Badari Pulavarty5c755e92008-07-23 21:28:19 -0700590 /* If the entire pageblock is free, move to the end of free page */
KAMEZAWA Hiroyuki0dcc48c2010-09-09 16:38:01 -0700591 if (pageblock_free(page)) {
592 int order;
593 /* be careful. we don't have locks, page_order can be changed.*/
594 order = page_order(page);
595 if ((order < MAX_ORDER) && (order >= pageblock_order))
596 return page + (1 << order);
597 }
Badari Pulavarty5c755e92008-07-23 21:28:19 -0700598
KAMEZAWA Hiroyuki0dcc48c2010-09-09 16:38:01 -0700599 return page + pageblock_nr_pages;
Badari Pulavarty5c755e92008-07-23 21:28:19 -0700600}
601
602/* Checks if this range of memory is likely to be hot-removable. */
603int is_mem_section_removable(unsigned long start_pfn, unsigned long nr_pages)
604{
605 int type;
606 struct page *page = pfn_to_page(start_pfn);
607 struct page *end_page = page + nr_pages;
608
609 /* Check the starting page of each pageblock within the range */
610 for (; page < end_page; page = next_active_pageblock(page)) {
611 type = get_pageblock_migratetype(page);
612
613 /*
614 * A pageblock containing MOVABLE or free pages is considered
615 * removable
616 */
617 if (type != MIGRATE_MOVABLE && !pageblock_free(page))
618 return 0;
619
620 /*
621 * A pageblock starting with a PageReserved page is not
622 * considered removable.
623 */
624 if (PageReserved(page))
625 return 0;
626 }
627
628 /* All pageblocks in the memory block are likely to be hot-removable */
629 return 1;
630}
631
632/*
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700633 * Confirm all pages in a range [start, end) is belongs to the same zone.
634 */
635static int test_pages_in_a_zone(unsigned long start_pfn, unsigned long end_pfn)
636{
637 unsigned long pfn;
638 struct zone *zone = NULL;
639 struct page *page;
640 int i;
641 for (pfn = start_pfn;
642 pfn < end_pfn;
643 pfn += MAX_ORDER_NR_PAGES) {
644 i = 0;
645 /* This is just a CONFIG_HOLES_IN_ZONE check.*/
646 while ((i < MAX_ORDER_NR_PAGES) && !pfn_valid_within(pfn + i))
647 i++;
648 if (i == MAX_ORDER_NR_PAGES)
649 continue;
650 page = pfn_to_page(pfn + i);
651 if (zone && page_zone(page) != zone)
652 return 0;
653 zone = page_zone(page);
654 }
655 return 1;
656}
657
658/*
659 * Scanning pfn is much easier than scanning lru list.
660 * Scan pfn from start to end and Find LRU page.
661 */
662int scan_lru_pages(unsigned long start, unsigned long end)
663{
664 unsigned long pfn;
665 struct page *page;
666 for (pfn = start; pfn < end; pfn++) {
667 if (pfn_valid(pfn)) {
668 page = pfn_to_page(pfn);
669 if (PageLRU(page))
670 return pfn;
671 }
672 }
673 return 0;
674}
675
676static struct page *
Hugh Dickins3c1d4372009-01-06 14:39:23 -0800677hotremove_migrate_alloc(struct page *page, unsigned long private, int **x)
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700678{
Hugh Dickins3c1d4372009-01-06 14:39:23 -0800679 /* This should be improooooved!! */
680 return alloc_page(GFP_HIGHUSER_MOVABLE);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700681}
682
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700683#define NR_OFFLINE_AT_ONCE_PAGES (256)
684static int
685do_migrate_range(unsigned long start_pfn, unsigned long end_pfn)
686{
687 unsigned long pfn;
688 struct page *page;
689 int move_pages = NR_OFFLINE_AT_ONCE_PAGES;
690 int not_managed = 0;
691 int ret = 0;
692 LIST_HEAD(source);
693
694 for (pfn = start_pfn; pfn < end_pfn && move_pages > 0; pfn++) {
695 if (!pfn_valid(pfn))
696 continue;
697 page = pfn_to_page(pfn);
698 if (!page_count(page))
699 continue;
700 /*
701 * We can skip free pages. And we can only deal with pages on
702 * LRU.
703 */
Nick Piggin62695a82008-10-18 20:26:09 -0700704 ret = isolate_lru_page(page);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700705 if (!ret) { /* Success */
Nick Piggin62695a82008-10-18 20:26:09 -0700706 list_add_tail(&page->lru, &source);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700707 move_pages--;
KOSAKI Motohiro6d9c2852009-12-14 17:58:11 -0800708 inc_zone_page_state(page, NR_ISOLATED_ANON +
709 page_is_file_cache(page));
710
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700711 } else {
712 /* Becasue we don't have big zone->lock. we should
713 check this again here. */
714 if (page_count(page))
715 not_managed++;
716#ifdef CONFIG_DEBUG_VM
Wu Fengguang718a3822010-03-10 15:20:43 -0800717 printk(KERN_ALERT "removing pfn %lx from LRU failed\n",
718 pfn);
719 dump_page(page);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700720#endif
721 }
722 }
723 ret = -EBUSY;
724 if (not_managed) {
725 if (!list_empty(&source))
726 putback_lru_pages(&source);
727 goto out;
728 }
729 ret = 0;
730 if (list_empty(&source))
731 goto out;
732 /* this function returns # of failed pages */
Hugh Dickins62b61f62009-12-14 17:59:33 -0800733 ret = migrate_pages(&source, hotremove_migrate_alloc, 0, 1);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700734
735out:
736 return ret;
737}
738
739/*
740 * remove from free_area[] and mark all as Reserved.
741 */
742static int
743offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages,
744 void *data)
745{
746 __offline_isolated_pages(start, start + nr_pages);
747 return 0;
748}
749
750static void
751offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
752{
KAMEZAWA Hiroyuki908eedc2009-09-22 16:45:46 -0700753 walk_system_ram_range(start_pfn, end_pfn - start_pfn, NULL,
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700754 offline_isolated_pages_cb);
755}
756
757/*
758 * Check all pages in range, recoreded as memory resource, are isolated.
759 */
760static int
761check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages,
762 void *data)
763{
764 int ret;
765 long offlined = *(long *)data;
766 ret = test_pages_isolated(start_pfn, start_pfn + nr_pages);
767 offlined = nr_pages;
768 if (!ret)
769 *(long *)data += offlined;
770 return ret;
771}
772
773static long
774check_pages_isolated(unsigned long start_pfn, unsigned long end_pfn)
775{
776 long offlined = 0;
777 int ret;
778
KAMEZAWA Hiroyuki908eedc2009-09-22 16:45:46 -0700779 ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn, &offlined,
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700780 check_pages_isolated_cb);
781 if (ret < 0)
782 offlined = (long)ret;
783 return offlined;
784}
785
Andrew Mortonb4e655a2009-12-14 17:59:35 -0800786static int offline_pages(unsigned long start_pfn,
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700787 unsigned long end_pfn, unsigned long timeout)
788{
789 unsigned long pfn, nr_pages, expire;
790 long offlined_pages;
Yasunori Goto7b78d332007-10-21 16:41:36 -0700791 int ret, drain, retry_max, node;
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700792 struct zone *zone;
Yasunori Goto7b78d332007-10-21 16:41:36 -0700793 struct memory_notify arg;
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700794
795 BUG_ON(start_pfn >= end_pfn);
796 /* at least, alignment against pageblock is necessary */
797 if (!IS_ALIGNED(start_pfn, pageblock_nr_pages))
798 return -EINVAL;
799 if (!IS_ALIGNED(end_pfn, pageblock_nr_pages))
800 return -EINVAL;
801 /* This makes hotplug much easier...and readable.
802 we assume this for now. .*/
803 if (!test_pages_in_a_zone(start_pfn, end_pfn))
804 return -EINVAL;
Yasunori Goto7b78d332007-10-21 16:41:36 -0700805
Andi Kleen6ad696d2009-11-17 14:06:22 -0800806 lock_system_sleep();
807
Yasunori Goto7b78d332007-10-21 16:41:36 -0700808 zone = page_zone(pfn_to_page(start_pfn));
809 node = zone_to_nid(zone);
810 nr_pages = end_pfn - start_pfn;
811
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700812 /* set above range as isolated */
813 ret = start_isolate_page_range(start_pfn, end_pfn);
814 if (ret)
Andi Kleen6ad696d2009-11-17 14:06:22 -0800815 goto out;
Yasunori Goto7b78d332007-10-21 16:41:36 -0700816
817 arg.start_pfn = start_pfn;
818 arg.nr_pages = nr_pages;
819 arg.status_change_nid = -1;
820 if (nr_pages >= node_present_pages(node))
821 arg.status_change_nid = node;
822
823 ret = memory_notify(MEM_GOING_OFFLINE, &arg);
824 ret = notifier_to_errno(ret);
825 if (ret)
826 goto failed_removal;
827
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700828 pfn = start_pfn;
829 expire = jiffies + timeout;
830 drain = 0;
831 retry_max = 5;
832repeat:
833 /* start memory hot removal */
834 ret = -EAGAIN;
835 if (time_after(jiffies, expire))
836 goto failed_removal;
837 ret = -EINTR;
838 if (signal_pending(current))
839 goto failed_removal;
840 ret = 0;
841 if (drain) {
842 lru_add_drain_all();
843 flush_scheduled_work();
844 cond_resched();
Christoph Lameter9f8f2172008-02-04 22:29:11 -0800845 drain_all_pages();
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700846 }
847
848 pfn = scan_lru_pages(start_pfn, end_pfn);
849 if (pfn) { /* We have page on LRU */
850 ret = do_migrate_range(pfn, end_pfn);
851 if (!ret) {
852 drain = 1;
853 goto repeat;
854 } else {
855 if (ret < 0)
856 if (--retry_max == 0)
857 goto failed_removal;
858 yield();
859 drain = 1;
860 goto repeat;
861 }
862 }
863 /* drain all zone's lru pagevec, this is asyncronous... */
864 lru_add_drain_all();
865 flush_scheduled_work();
866 yield();
867 /* drain pcp pages , this is synchrouns. */
Christoph Lameter9f8f2172008-02-04 22:29:11 -0800868 drain_all_pages();
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700869 /* check again */
870 offlined_pages = check_pages_isolated(start_pfn, end_pfn);
871 if (offlined_pages < 0) {
872 ret = -EBUSY;
873 goto failed_removal;
874 }
875 printk(KERN_INFO "Offlined Pages %ld\n", offlined_pages);
876 /* Ok, all of our target is islaoted.
877 We cannot do rollback at this point. */
878 offline_isolated_pages(start_pfn, end_pfn);
KAMEZAWA Hiroyukidbc0e4c2007-11-14 16:59:12 -0800879 /* reset pagetype flags and makes migrate type to be MOVABLE */
880 undo_isolate_page_range(start_pfn, end_pfn);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700881 /* removal success */
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700882 zone->present_pages -= offlined_pages;
883 zone->zone_pgdat->node_present_pages -= offlined_pages;
884 totalram_pages -= offlined_pages;
Yasunori Goto7b78d332007-10-21 16:41:36 -0700885
Minchan Kimbce73942009-06-16 15:32:50 -0700886 setup_per_zone_wmarks();
887 calculate_zone_inactive_ratio(zone);
David Rientjes8fe23e02009-12-14 17:58:33 -0800888 if (!node_present_pages(node)) {
889 node_clear_state(node, N_HIGH_MEMORY);
890 kswapd_stop(node);
891 }
Minchan Kimbce73942009-06-16 15:32:50 -0700892
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700893 vm_total_pages = nr_free_pagecache_pages();
894 writeback_set_ratelimit();
Yasunori Goto7b78d332007-10-21 16:41:36 -0700895
896 memory_notify(MEM_OFFLINE, &arg);
Andi Kleen6ad696d2009-11-17 14:06:22 -0800897 unlock_system_sleep();
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700898 return 0;
899
900failed_removal:
901 printk(KERN_INFO "memory offlining %lx to %lx failed\n",
902 start_pfn, end_pfn);
Yasunori Goto7b78d332007-10-21 16:41:36 -0700903 memory_notify(MEM_CANCEL_OFFLINE, &arg);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700904 /* pushback to free area */
905 undo_isolate_page_range(start_pfn, end_pfn);
Yasunori Goto7b78d332007-10-21 16:41:36 -0700906
Andi Kleen6ad696d2009-11-17 14:06:22 -0800907out:
908 unlock_system_sleep();
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700909 return ret;
910}
Badari Pulavarty71088782008-10-18 20:25:58 -0700911
912int remove_memory(u64 start, u64 size)
913{
914 unsigned long start_pfn, end_pfn;
915
916 start_pfn = PFN_DOWN(start);
917 end_pfn = start_pfn + PFN_DOWN(size);
918 return offline_pages(start_pfn, end_pfn, 120 * HZ);
919}
KAMEZAWA Hiroyuki48e94192007-10-16 01:26:14 -0700920#else
921int remove_memory(u64 start, u64 size)
922{
923 return -EINVAL;
924}
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700925#endif /* CONFIG_MEMORY_HOTREMOVE */
Badari Pulavarty71088782008-10-18 20:25:58 -0700926EXPORT_SYMBOL_GPL(remove_memory);