blob: e4412a676c88494ec313ebb3e974d2d71ee5c525 [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>
Dave Hansen3947be12005-10-29 18:16:54 -070029
30#include <asm/tlbflush.h>
31
Adrian Bunk1e5ad9a2008-04-28 20:40:08 +030032#include "internal.h"
33
Keith Mannthey45e0b782006-09-30 23:27:09 -070034/* add this memory to iomem resource */
35static struct resource *register_memory_resource(u64 start, u64 size)
36{
37 struct resource *res;
38 res = kzalloc(sizeof(struct resource), GFP_KERNEL);
39 BUG_ON(!res);
40
41 res->name = "System RAM";
42 res->start = start;
43 res->end = start + size - 1;
Yasunori Goto887c3cb2007-11-14 16:59:20 -080044 res->flags = IORESOURCE_MEM | IORESOURCE_BUSY;
Keith Mannthey45e0b782006-09-30 23:27:09 -070045 if (request_resource(&iomem_resource, res) < 0) {
46 printk("System RAM resource %llx - %llx cannot be added\n",
47 (unsigned long long)res->start, (unsigned long long)res->end);
48 kfree(res);
49 res = NULL;
50 }
51 return res;
52}
53
54static void release_memory_resource(struct resource *res)
55{
56 if (!res)
57 return;
58 release_resource(res);
59 kfree(res);
60 return;
61}
62
Keith Mannthey53947022006-09-30 23:27:08 -070063#ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
Yasunori Goto04753272008-04-28 02:13:31 -070064#ifndef CONFIG_SPARSEMEM_VMEMMAP
Yasunori Gotoaf370fb2008-07-23 21:28:17 -070065static void get_page_bootmem(unsigned long info, struct page *page, int type)
Yasunori Goto04753272008-04-28 02:13:31 -070066{
Yasunori Gotoaf370fb2008-07-23 21:28:17 -070067 atomic_set(&page->_mapcount, type);
Yasunori Goto04753272008-04-28 02:13:31 -070068 SetPagePrivate(page);
69 set_page_private(page, info);
70 atomic_inc(&page->_count);
71}
72
73void put_page_bootmem(struct page *page)
74{
Yasunori Gotoaf370fb2008-07-23 21:28:17 -070075 int type;
Yasunori Goto04753272008-04-28 02:13:31 -070076
Yasunori Gotoaf370fb2008-07-23 21:28:17 -070077 type = atomic_read(&page->_mapcount);
78 BUG_ON(type >= -1);
Yasunori Goto04753272008-04-28 02:13:31 -070079
80 if (atomic_dec_return(&page->_count) == 1) {
81 ClearPagePrivate(page);
82 set_page_private(page, 0);
83 reset_page_mapcount(page);
84 __free_pages_bootmem(page, 0);
85 }
86
87}
88
Adrian Bunkd92bc312008-07-23 21:28:12 -070089static void register_page_bootmem_info_section(unsigned long start_pfn)
Yasunori Goto04753272008-04-28 02:13:31 -070090{
91 unsigned long *usemap, mapsize, section_nr, i;
92 struct mem_section *ms;
93 struct page *page, *memmap;
94
95 if (!pfn_valid(start_pfn))
96 return;
97
98 section_nr = pfn_to_section_nr(start_pfn);
99 ms = __nr_to_section(section_nr);
100
101 /* Get section's memmap address */
102 memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
103
104 /*
105 * Get page for the memmap's phys address
106 * XXX: need more consideration for sparse_vmemmap...
107 */
108 page = virt_to_page(memmap);
109 mapsize = sizeof(struct page) * PAGES_PER_SECTION;
110 mapsize = PAGE_ALIGN(mapsize) >> PAGE_SHIFT;
111
112 /* remember memmap's page */
113 for (i = 0; i < mapsize; i++, page++)
114 get_page_bootmem(section_nr, page, SECTION_INFO);
115
116 usemap = __nr_to_section(section_nr)->pageblock_flags;
117 page = virt_to_page(usemap);
118
119 mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT;
120
121 for (i = 0; i < mapsize; i++, page++)
Yasunori Gotoaf370fb2008-07-23 21:28:17 -0700122 get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
Yasunori Goto04753272008-04-28 02:13:31 -0700123
124}
125
126void register_page_bootmem_info_node(struct pglist_data *pgdat)
127{
128 unsigned long i, pfn, end_pfn, nr_pages;
129 int node = pgdat->node_id;
130 struct page *page;
131 struct zone *zone;
132
133 nr_pages = PAGE_ALIGN(sizeof(struct pglist_data)) >> PAGE_SHIFT;
134 page = virt_to_page(pgdat);
135
136 for (i = 0; i < nr_pages; i++, page++)
137 get_page_bootmem(node, page, NODE_INFO);
138
139 zone = &pgdat->node_zones[0];
140 for (; zone < pgdat->node_zones + MAX_NR_ZONES - 1; zone++) {
141 if (zone->wait_table) {
142 nr_pages = zone->wait_table_hash_nr_entries
143 * sizeof(wait_queue_head_t);
144 nr_pages = PAGE_ALIGN(nr_pages) >> PAGE_SHIFT;
145 page = virt_to_page(zone->wait_table);
146
147 for (i = 0; i < nr_pages; i++, page++)
148 get_page_bootmem(node, page, NODE_INFO);
149 }
150 }
151
152 pfn = pgdat->node_start_pfn;
153 end_pfn = pfn + pgdat->node_spanned_pages;
154
155 /* register_section info */
156 for (; pfn < end_pfn; pfn += PAGES_PER_SECTION)
157 register_page_bootmem_info_section(pfn);
158
159}
160#endif /* !CONFIG_SPARSEMEM_VMEMMAP */
161
Heiko Carstens76cdd582008-05-14 16:05:52 -0700162static void grow_zone_span(struct zone *zone, unsigned long start_pfn,
163 unsigned long end_pfn)
164{
165 unsigned long old_zone_end_pfn;
166
167 zone_span_writelock(zone);
168
169 old_zone_end_pfn = zone->zone_start_pfn + zone->spanned_pages;
170 if (start_pfn < zone->zone_start_pfn)
171 zone->zone_start_pfn = start_pfn;
172
173 zone->spanned_pages = max(old_zone_end_pfn, end_pfn) -
174 zone->zone_start_pfn;
175
176 zone_span_writeunlock(zone);
177}
178
179static void grow_pgdat_span(struct pglist_data *pgdat, unsigned long start_pfn,
180 unsigned long end_pfn)
181{
182 unsigned long old_pgdat_end_pfn =
183 pgdat->node_start_pfn + pgdat->node_spanned_pages;
184
185 if (start_pfn < pgdat->node_start_pfn)
186 pgdat->node_start_pfn = start_pfn;
187
188 pgdat->node_spanned_pages = max(old_pgdat_end_pfn, end_pfn) -
189 pgdat->node_start_pfn;
190}
191
Al Viro31168482008-11-22 17:33:24 +0000192static int __meminit __add_zone(struct zone *zone, unsigned long phys_start_pfn)
Dave Hansen3947be12005-10-29 18:16:54 -0700193{
194 struct pglist_data *pgdat = zone->zone_pgdat;
195 int nr_pages = PAGES_PER_SECTION;
196 int nid = pgdat->node_id;
197 int zone_type;
Heiko Carstens76cdd582008-05-14 16:05:52 -0700198 unsigned long flags;
Dave Hansen3947be12005-10-29 18:16:54 -0700199
200 zone_type = zone - pgdat->node_zones;
Heiko Carstens76cdd582008-05-14 16:05:52 -0700201 if (!zone->wait_table) {
202 int ret;
203
204 ret = init_currently_empty_zone(zone, phys_start_pfn,
205 nr_pages, MEMMAP_HOTPLUG);
206 if (ret)
207 return ret;
208 }
209 pgdat_resize_lock(zone->zone_pgdat, &flags);
210 grow_zone_span(zone, phys_start_pfn, phys_start_pfn + nr_pages);
211 grow_pgdat_span(zone->zone_pgdat, phys_start_pfn,
212 phys_start_pfn + nr_pages);
213 pgdat_resize_unlock(zone->zone_pgdat, &flags);
Dave Hansena2f3aa02007-01-10 23:15:30 -0800214 memmap_init_zone(nr_pages, nid, zone_type,
215 phys_start_pfn, MEMMAP_HOTPLUG);
Yasunori Goto718127c2006-06-23 02:03:10 -0700216 return 0;
Dave Hansen3947be12005-10-29 18:16:54 -0700217}
218
Gary Hadec04fc582009-01-06 14:39:14 -0800219static int __meminit __add_section(int nid, struct zone *zone,
220 unsigned long phys_start_pfn)
Dave Hansen3947be12005-10-29 18:16:54 -0700221{
Dave Hansen3947be12005-10-29 18:16:54 -0700222 int nr_pages = PAGES_PER_SECTION;
Dave Hansen3947be12005-10-29 18:16:54 -0700223 int ret;
224
KAMEZAWA Hiroyukiebd15302006-08-05 12:15:06 -0700225 if (pfn_valid(phys_start_pfn))
226 return -EEXIST;
227
Dave Hansen0b0acbe2005-10-29 18:16:55 -0700228 ret = sparse_add_one_section(zone, phys_start_pfn, nr_pages);
Dave Hansen3947be12005-10-29 18:16:54 -0700229
230 if (ret < 0)
231 return ret;
232
Yasunori Goto718127c2006-06-23 02:03:10 -0700233 ret = __add_zone(zone, phys_start_pfn);
234
235 if (ret < 0)
236 return ret;
237
Gary Hadec04fc582009-01-06 14:39:14 -0800238 return register_new_memory(nid, __pfn_to_section(phys_start_pfn));
Dave Hansen3947be12005-10-29 18:16:54 -0700239}
240
Yasunori Goto0c0a4a52008-04-28 02:13:34 -0700241#ifdef CONFIG_SPARSEMEM_VMEMMAP
242static int __remove_section(struct zone *zone, struct mem_section *ms)
243{
244 /*
245 * XXX: Freeing memmap with vmemmap is not implement yet.
246 * This should be removed later.
247 */
248 return -EBUSY;
249}
250#else
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700251static int __remove_section(struct zone *zone, struct mem_section *ms)
252{
253 unsigned long flags;
254 struct pglist_data *pgdat = zone->zone_pgdat;
255 int ret = -EINVAL;
256
257 if (!valid_section(ms))
258 return ret;
259
260 ret = unregister_memory_section(ms);
261 if (ret)
262 return ret;
263
264 pgdat_resize_lock(pgdat, &flags);
265 sparse_remove_one_section(zone, ms);
266 pgdat_resize_unlock(pgdat, &flags);
267 return 0;
268}
Yasunori Goto0c0a4a52008-04-28 02:13:34 -0700269#endif
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700270
Dave Hansen3947be12005-10-29 18:16:54 -0700271/*
272 * Reasonably generic function for adding memory. It is
273 * expected that archs that support memory hotplug will
274 * call this function after deciding the zone to which to
275 * add the new pages.
276 */
Gary Hadec04fc582009-01-06 14:39:14 -0800277int __ref __add_pages(int nid, struct zone *zone, unsigned long phys_start_pfn,
278 unsigned long nr_pages)
Dave Hansen3947be12005-10-29 18:16:54 -0700279{
280 unsigned long i;
281 int err = 0;
KAMEZAWA Hiroyuki6f712712006-08-05 12:14:58 -0700282 int start_sec, end_sec;
283 /* during initialize mem_map, align hot-added range to section */
284 start_sec = pfn_to_section_nr(phys_start_pfn);
285 end_sec = pfn_to_section_nr(phys_start_pfn + nr_pages - 1);
Dave Hansen3947be12005-10-29 18:16:54 -0700286
KAMEZAWA Hiroyuki6f712712006-08-05 12:14:58 -0700287 for (i = start_sec; i <= end_sec; i++) {
Gary Hadec04fc582009-01-06 14:39:14 -0800288 err = __add_section(nid, zone, i << PFN_SECTION_SHIFT);
Dave Hansen3947be12005-10-29 18:16:54 -0700289
KAMEZAWA Hiroyuki6f712712006-08-05 12:14:58 -0700290 /*
Simon Arlott183ff222007-10-20 01:27:18 +0200291 * EEXIST is finally dealt with by ioresource collision
KAMEZAWA Hiroyuki6f712712006-08-05 12:14:58 -0700292 * check. see add_memory() => register_memory_resource()
293 * Warning will be printed if there is collision.
Joel H Schoppbed120c2006-05-01 12:16:11 -0700294 */
295 if (err && (err != -EEXIST))
Dave Hansen3947be12005-10-29 18:16:54 -0700296 break;
KAMEZAWA Hiroyuki6f712712006-08-05 12:14:58 -0700297 err = 0;
Dave Hansen3947be12005-10-29 18:16:54 -0700298 }
299
300 return err;
301}
Joel H Schoppbed120c2006-05-01 12:16:11 -0700302EXPORT_SYMBOL_GPL(__add_pages);
Dave Hansen3947be12005-10-29 18:16:54 -0700303
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700304/**
305 * __remove_pages() - remove sections of pages from a zone
306 * @zone: zone from which pages need to be removed
307 * @phys_start_pfn: starting pageframe (must be aligned to start of a section)
308 * @nr_pages: number of pages to remove (must be multiple of section size)
309 *
310 * Generic helper function to remove section mappings and sysfs entries
311 * for the section of the memory we are removing. Caller needs to make
312 * sure that pages are marked reserved and zones are adjust properly by
313 * calling offline_pages().
314 */
315int __remove_pages(struct zone *zone, unsigned long phys_start_pfn,
316 unsigned long nr_pages)
317{
318 unsigned long i, ret = 0;
319 int sections_to_remove;
320
321 /*
322 * We can only remove entire sections
323 */
324 BUG_ON(phys_start_pfn & ~PAGE_SECTION_MASK);
325 BUG_ON(nr_pages % PAGES_PER_SECTION);
326
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700327 sections_to_remove = nr_pages / PAGES_PER_SECTION;
328 for (i = 0; i < sections_to_remove; i++) {
329 unsigned long pfn = phys_start_pfn + i*PAGES_PER_SECTION;
Nathan Fontenotde7f0cb2008-10-18 20:27:14 -0700330 release_mem_region(pfn << PAGE_SHIFT,
331 PAGES_PER_SECTION << PAGE_SHIFT);
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700332 ret = __remove_section(zone, __pfn_to_section(pfn));
333 if (ret)
334 break;
335 }
336 return ret;
337}
338EXPORT_SYMBOL_GPL(__remove_pages);
339
Jeremy Fitzhardinge180c06e2008-04-28 02:12:03 -0700340void online_page(struct page *page)
341{
342 totalram_pages++;
343 num_physpages++;
344
345#ifdef CONFIG_HIGHMEM
346 if (PageHighMem(page))
347 totalhigh_pages++;
348#endif
349
350#ifdef CONFIG_FLATMEM
351 max_mapnr = max(page_to_pfn(page), max_mapnr);
352#endif
353
354 ClearPageReserved(page);
355 init_page_count(page);
356 __free_page(page);
357}
358
KAMEZAWA Hiroyuki75884fb2007-10-16 01:26:10 -0700359static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages,
360 void *arg)
Dave Hansen3947be12005-10-29 18:16:54 -0700361{
362 unsigned long i;
KAMEZAWA Hiroyuki75884fb2007-10-16 01:26:10 -0700363 unsigned long onlined_pages = *(unsigned long *)arg;
364 struct page *page;
365 if (PageReserved(pfn_to_page(start_pfn)))
366 for (i = 0; i < nr_pages; i++) {
367 page = pfn_to_page(start_pfn + i);
368 online_page(page);
369 onlined_pages++;
370 }
371 *(unsigned long *)arg = onlined_pages;
372 return 0;
373}
374
375
376int online_pages(unsigned long pfn, unsigned long nr_pages)
377{
Dave Hansen3947be12005-10-29 18:16:54 -0700378 unsigned long onlined_pages = 0;
379 struct zone *zone;
Yasunori Goto68113782006-06-23 02:03:11 -0700380 int need_zonelists_rebuild = 0;
Yasunori Goto7b78d332007-10-21 16:41:36 -0700381 int nid;
382 int ret;
383 struct memory_notify arg;
Dave Hansen3947be12005-10-29 18:16:54 -0700384
Yasunori Goto7b78d332007-10-21 16:41:36 -0700385 arg.start_pfn = pfn;
386 arg.nr_pages = nr_pages;
387 arg.status_change_nid = -1;
388
389 nid = page_to_nid(pfn_to_page(pfn));
390 if (node_present_pages(nid) == 0)
391 arg.status_change_nid = nid;
392
393 ret = memory_notify(MEM_GOING_ONLINE, &arg);
394 ret = notifier_to_errno(ret);
395 if (ret) {
396 memory_notify(MEM_CANCEL_ONLINE, &arg);
397 return ret;
398 }
Dave Hansen3947be12005-10-29 18:16:54 -0700399 /*
400 * This doesn't need a lock to do pfn_to_page().
401 * The section can't be removed here because of the
Daniel Walkerda19cbc2008-02-04 23:35:47 -0800402 * memory_block->state_mutex.
Dave Hansen3947be12005-10-29 18:16:54 -0700403 */
404 zone = page_zone(pfn_to_page(pfn));
Yasunori Goto68113782006-06-23 02:03:11 -0700405 /*
406 * If this zone is not populated, then it is not in zonelist.
407 * This means the page allocator ignores this zone.
408 * So, zonelist must be updated after online.
409 */
410 if (!populated_zone(zone))
411 need_zonelists_rebuild = 1;
412
Geoff Levandfd8a4222008-05-14 16:05:50 -0700413 ret = walk_memory_resource(pfn, nr_pages, &onlined_pages,
KAMEZAWA Hiroyuki75884fb2007-10-16 01:26:10 -0700414 online_pages_range);
Geoff Levandfd8a4222008-05-14 16:05:50 -0700415 if (ret) {
416 printk(KERN_DEBUG "online_pages %lx at %lx failed\n",
417 nr_pages, pfn);
418 memory_notify(MEM_CANCEL_ONLINE, &arg);
419 return ret;
420 }
421
Dave Hansen3947be12005-10-29 18:16:54 -0700422 zone->present_pages += onlined_pages;
Yasunori Gotof2937be2006-03-09 17:33:51 -0800423 zone->zone_pgdat->node_present_pages += onlined_pages;
Dave Hansen3947be12005-10-29 18:16:54 -0700424
Minchan Kimbc75d332009-06-16 15:32:48 -0700425 setup_per_zone_wmarks();
Minchan Kimbce73942009-06-16 15:32:50 -0700426 calculate_zone_inactive_ratio(zone);
Christoph Lameter7ea15302007-10-16 01:25:29 -0700427 if (onlined_pages) {
428 kswapd_run(zone_to_nid(zone));
429 node_set_state(zone_to_nid(zone), N_HIGH_MEMORY);
430 }
Dave Hansen61b13992005-10-29 18:16:56 -0700431
Yasunori Goto68113782006-06-23 02:03:11 -0700432 if (need_zonelists_rebuild)
433 build_all_zonelists();
Kent Liu2f7f24e2008-07-23 21:28:18 -0700434 else
435 vm_total_pages = nr_free_pagecache_pages();
436
Chandra Seetharaman2d1d43f2006-09-29 02:01:25 -0700437 writeback_set_ratelimit();
Yasunori Goto7b78d332007-10-21 16:41:36 -0700438
439 if (onlined_pages)
440 memory_notify(MEM_ONLINE, &arg);
441
Dave Hansen3947be12005-10-29 18:16:54 -0700442 return 0;
443}
Keith Mannthey53947022006-09-30 23:27:08 -0700444#endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
Yasunori Gotobc02af92006-06-27 02:53:30 -0700445
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700446static pg_data_t *hotadd_new_pgdat(int nid, u64 start)
447{
448 struct pglist_data *pgdat;
449 unsigned long zones_size[MAX_NR_ZONES] = {0};
450 unsigned long zholes_size[MAX_NR_ZONES] = {0};
451 unsigned long start_pfn = start >> PAGE_SHIFT;
452
453 pgdat = arch_alloc_nodedata(nid);
454 if (!pgdat)
455 return NULL;
456
457 arch_refresh_nodedata(nid, pgdat);
458
459 /* we can use NODE_DATA(nid) from here */
460
461 /* init node's zones as empty zones, we don't have any present pages.*/
Johannes Weiner9109fb72008-07-23 21:27:20 -0700462 free_area_init_node(nid, zones_size, start_pfn, zholes_size);
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700463
464 return pgdat;
465}
466
467static void rollback_node_hotadd(int nid, pg_data_t *pgdat)
468{
469 arch_refresh_nodedata(nid, NULL);
470 arch_free_nodedata(pgdat);
471 return;
472}
473
KAMEZAWA Hiroyuki0a547032006-06-27 02:53:35 -0700474
Al Viro31168482008-11-22 17:33:24 +0000475/* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
476int __ref add_memory(int nid, u64 start, u64 size)
Yasunori Gotobc02af92006-06-27 02:53:30 -0700477{
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700478 pg_data_t *pgdat = NULL;
479 int new_pgdat = 0;
KAMEZAWA Hiroyukiebd15302006-08-05 12:15:06 -0700480 struct resource *res;
Yasunori Gotobc02af92006-06-27 02:53:30 -0700481 int ret;
482
KAMEZAWA Hiroyukiebd15302006-08-05 12:15:06 -0700483 res = register_memory_resource(start, size);
484 if (!res)
485 return -EEXIST;
486
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700487 if (!node_online(nid)) {
488 pgdat = hotadd_new_pgdat(nid, start);
489 if (!pgdat)
490 return -ENOMEM;
491 new_pgdat = 1;
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700492 }
493
Yasunori Gotobc02af92006-06-27 02:53:30 -0700494 /* call arch's memory hotadd */
495 ret = arch_add_memory(nid, start, size);
496
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700497 if (ret < 0)
498 goto error;
499
Yasunori Goto0fc44152006-06-27 02:53:38 -0700500 /* we online node here. we can't roll back from here. */
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700501 node_set_online(nid);
502
Yasunori Goto0fc44152006-06-27 02:53:38 -0700503 if (new_pgdat) {
504 ret = register_one_node(nid);
505 /*
506 * If sysfs file of new node can't create, cpu on the node
507 * can't be hot-added. There is no rollback way now.
508 * So, check by BUG_ON() to catch it reluctantly..
509 */
510 BUG_ON(ret);
511 }
512
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700513 return ret;
514error:
515 /* rollback pgdat allocation and others */
516 if (new_pgdat)
517 rollback_node_hotadd(nid, pgdat);
KAMEZAWA Hiroyukiebd15302006-08-05 12:15:06 -0700518 if (res)
519 release_memory_resource(res);
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700520
Yasunori Gotobc02af92006-06-27 02:53:30 -0700521 return ret;
522}
523EXPORT_SYMBOL_GPL(add_memory);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700524
525#ifdef CONFIG_MEMORY_HOTREMOVE
526/*
Badari Pulavarty5c755e92008-07-23 21:28:19 -0700527 * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy
528 * set and the size of the free page is given by page_order(). Using this,
529 * the function determines if the pageblock contains only free pages.
530 * Due to buddy contraints, a free page at least the size of a pageblock will
531 * be located at the start of the pageblock
532 */
533static inline int pageblock_free(struct page *page)
534{
535 return PageBuddy(page) && page_order(page) >= pageblock_order;
536}
537
538/* Return the start of the next active pageblock after a given page */
539static struct page *next_active_pageblock(struct page *page)
540{
541 int pageblocks_stride;
542
543 /* Ensure the starting page is pageblock-aligned */
544 BUG_ON(page_to_pfn(page) & (pageblock_nr_pages - 1));
545
546 /* Move forward by at least 1 * pageblock_nr_pages */
547 pageblocks_stride = 1;
548
549 /* If the entire pageblock is free, move to the end of free page */
550 if (pageblock_free(page))
551 pageblocks_stride += page_order(page) - pageblock_order;
552
553 return page + (pageblocks_stride * pageblock_nr_pages);
554}
555
556/* Checks if this range of memory is likely to be hot-removable. */
557int is_mem_section_removable(unsigned long start_pfn, unsigned long nr_pages)
558{
559 int type;
560 struct page *page = pfn_to_page(start_pfn);
561 struct page *end_page = page + nr_pages;
562
563 /* Check the starting page of each pageblock within the range */
564 for (; page < end_page; page = next_active_pageblock(page)) {
565 type = get_pageblock_migratetype(page);
566
567 /*
568 * A pageblock containing MOVABLE or free pages is considered
569 * removable
570 */
571 if (type != MIGRATE_MOVABLE && !pageblock_free(page))
572 return 0;
573
574 /*
575 * A pageblock starting with a PageReserved page is not
576 * considered removable.
577 */
578 if (PageReserved(page))
579 return 0;
580 }
581
582 /* All pageblocks in the memory block are likely to be hot-removable */
583 return 1;
584}
585
586/*
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700587 * Confirm all pages in a range [start, end) is belongs to the same zone.
588 */
589static int test_pages_in_a_zone(unsigned long start_pfn, unsigned long end_pfn)
590{
591 unsigned long pfn;
592 struct zone *zone = NULL;
593 struct page *page;
594 int i;
595 for (pfn = start_pfn;
596 pfn < end_pfn;
597 pfn += MAX_ORDER_NR_PAGES) {
598 i = 0;
599 /* This is just a CONFIG_HOLES_IN_ZONE check.*/
600 while ((i < MAX_ORDER_NR_PAGES) && !pfn_valid_within(pfn + i))
601 i++;
602 if (i == MAX_ORDER_NR_PAGES)
603 continue;
604 page = pfn_to_page(pfn + i);
605 if (zone && page_zone(page) != zone)
606 return 0;
607 zone = page_zone(page);
608 }
609 return 1;
610}
611
612/*
613 * Scanning pfn is much easier than scanning lru list.
614 * Scan pfn from start to end and Find LRU page.
615 */
616int scan_lru_pages(unsigned long start, unsigned long end)
617{
618 unsigned long pfn;
619 struct page *page;
620 for (pfn = start; pfn < end; pfn++) {
621 if (pfn_valid(pfn)) {
622 page = pfn_to_page(pfn);
623 if (PageLRU(page))
624 return pfn;
625 }
626 }
627 return 0;
628}
629
630static struct page *
Hugh Dickins3c1d4372009-01-06 14:39:23 -0800631hotremove_migrate_alloc(struct page *page, unsigned long private, int **x)
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700632{
Hugh Dickins3c1d4372009-01-06 14:39:23 -0800633 /* This should be improooooved!! */
634 return alloc_page(GFP_HIGHUSER_MOVABLE);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700635}
636
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700637#define NR_OFFLINE_AT_ONCE_PAGES (256)
638static int
639do_migrate_range(unsigned long start_pfn, unsigned long end_pfn)
640{
641 unsigned long pfn;
642 struct page *page;
643 int move_pages = NR_OFFLINE_AT_ONCE_PAGES;
644 int not_managed = 0;
645 int ret = 0;
646 LIST_HEAD(source);
647
648 for (pfn = start_pfn; pfn < end_pfn && move_pages > 0; pfn++) {
649 if (!pfn_valid(pfn))
650 continue;
651 page = pfn_to_page(pfn);
652 if (!page_count(page))
653 continue;
654 /*
655 * We can skip free pages. And we can only deal with pages on
656 * LRU.
657 */
Nick Piggin62695a82008-10-18 20:26:09 -0700658 ret = isolate_lru_page(page);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700659 if (!ret) { /* Success */
Nick Piggin62695a82008-10-18 20:26:09 -0700660 list_add_tail(&page->lru, &source);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700661 move_pages--;
662 } else {
663 /* Becasue we don't have big zone->lock. we should
664 check this again here. */
665 if (page_count(page))
666 not_managed++;
667#ifdef CONFIG_DEBUG_VM
668 printk(KERN_INFO "removing from LRU failed"
669 " %lx/%d/%lx\n",
670 pfn, page_count(page), page->flags);
671#endif
672 }
673 }
674 ret = -EBUSY;
675 if (not_managed) {
676 if (!list_empty(&source))
677 putback_lru_pages(&source);
678 goto out;
679 }
680 ret = 0;
681 if (list_empty(&source))
682 goto out;
683 /* this function returns # of failed pages */
684 ret = migrate_pages(&source, hotremove_migrate_alloc, 0);
685
686out:
687 return ret;
688}
689
690/*
691 * remove from free_area[] and mark all as Reserved.
692 */
693static int
694offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages,
695 void *data)
696{
697 __offline_isolated_pages(start, start + nr_pages);
698 return 0;
699}
700
701static void
702offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
703{
704 walk_memory_resource(start_pfn, end_pfn - start_pfn, NULL,
705 offline_isolated_pages_cb);
706}
707
708/*
709 * Check all pages in range, recoreded as memory resource, are isolated.
710 */
711static int
712check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages,
713 void *data)
714{
715 int ret;
716 long offlined = *(long *)data;
717 ret = test_pages_isolated(start_pfn, start_pfn + nr_pages);
718 offlined = nr_pages;
719 if (!ret)
720 *(long *)data += offlined;
721 return ret;
722}
723
724static long
725check_pages_isolated(unsigned long start_pfn, unsigned long end_pfn)
726{
727 long offlined = 0;
728 int ret;
729
730 ret = walk_memory_resource(start_pfn, end_pfn - start_pfn, &offlined,
731 check_pages_isolated_cb);
732 if (ret < 0)
733 offlined = (long)ret;
734 return offlined;
735}
736
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700737int offline_pages(unsigned long start_pfn,
738 unsigned long end_pfn, unsigned long timeout)
739{
740 unsigned long pfn, nr_pages, expire;
741 long offlined_pages;
Yasunori Goto7b78d332007-10-21 16:41:36 -0700742 int ret, drain, retry_max, node;
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700743 struct zone *zone;
Yasunori Goto7b78d332007-10-21 16:41:36 -0700744 struct memory_notify arg;
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700745
746 BUG_ON(start_pfn >= end_pfn);
747 /* at least, alignment against pageblock is necessary */
748 if (!IS_ALIGNED(start_pfn, pageblock_nr_pages))
749 return -EINVAL;
750 if (!IS_ALIGNED(end_pfn, pageblock_nr_pages))
751 return -EINVAL;
752 /* This makes hotplug much easier...and readable.
753 we assume this for now. .*/
754 if (!test_pages_in_a_zone(start_pfn, end_pfn))
755 return -EINVAL;
Yasunori Goto7b78d332007-10-21 16:41:36 -0700756
757 zone = page_zone(pfn_to_page(start_pfn));
758 node = zone_to_nid(zone);
759 nr_pages = end_pfn - start_pfn;
760
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700761 /* set above range as isolated */
762 ret = start_isolate_page_range(start_pfn, end_pfn);
763 if (ret)
764 return ret;
Yasunori Goto7b78d332007-10-21 16:41:36 -0700765
766 arg.start_pfn = start_pfn;
767 arg.nr_pages = nr_pages;
768 arg.status_change_nid = -1;
769 if (nr_pages >= node_present_pages(node))
770 arg.status_change_nid = node;
771
772 ret = memory_notify(MEM_GOING_OFFLINE, &arg);
773 ret = notifier_to_errno(ret);
774 if (ret)
775 goto failed_removal;
776
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700777 pfn = start_pfn;
778 expire = jiffies + timeout;
779 drain = 0;
780 retry_max = 5;
781repeat:
782 /* start memory hot removal */
783 ret = -EAGAIN;
784 if (time_after(jiffies, expire))
785 goto failed_removal;
786 ret = -EINTR;
787 if (signal_pending(current))
788 goto failed_removal;
789 ret = 0;
790 if (drain) {
791 lru_add_drain_all();
792 flush_scheduled_work();
793 cond_resched();
Christoph Lameter9f8f2172008-02-04 22:29:11 -0800794 drain_all_pages();
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700795 }
796
797 pfn = scan_lru_pages(start_pfn, end_pfn);
798 if (pfn) { /* We have page on LRU */
799 ret = do_migrate_range(pfn, end_pfn);
800 if (!ret) {
801 drain = 1;
802 goto repeat;
803 } else {
804 if (ret < 0)
805 if (--retry_max == 0)
806 goto failed_removal;
807 yield();
808 drain = 1;
809 goto repeat;
810 }
811 }
812 /* drain all zone's lru pagevec, this is asyncronous... */
813 lru_add_drain_all();
814 flush_scheduled_work();
815 yield();
816 /* drain pcp pages , this is synchrouns. */
Christoph Lameter9f8f2172008-02-04 22:29:11 -0800817 drain_all_pages();
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700818 /* check again */
819 offlined_pages = check_pages_isolated(start_pfn, end_pfn);
820 if (offlined_pages < 0) {
821 ret = -EBUSY;
822 goto failed_removal;
823 }
824 printk(KERN_INFO "Offlined Pages %ld\n", offlined_pages);
825 /* Ok, all of our target is islaoted.
826 We cannot do rollback at this point. */
827 offline_isolated_pages(start_pfn, end_pfn);
KAMEZAWA Hiroyukidbc0e4c2007-11-14 16:59:12 -0800828 /* reset pagetype flags and makes migrate type to be MOVABLE */
829 undo_isolate_page_range(start_pfn, end_pfn);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700830 /* removal success */
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700831 zone->present_pages -= offlined_pages;
832 zone->zone_pgdat->node_present_pages -= offlined_pages;
833 totalram_pages -= offlined_pages;
834 num_physpages -= offlined_pages;
Yasunori Goto7b78d332007-10-21 16:41:36 -0700835
Minchan Kimbce73942009-06-16 15:32:50 -0700836 setup_per_zone_wmarks();
837 calculate_zone_inactive_ratio(zone);
838
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700839 vm_total_pages = nr_free_pagecache_pages();
840 writeback_set_ratelimit();
Yasunori Goto7b78d332007-10-21 16:41:36 -0700841
842 memory_notify(MEM_OFFLINE, &arg);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700843 return 0;
844
845failed_removal:
846 printk(KERN_INFO "memory offlining %lx to %lx failed\n",
847 start_pfn, end_pfn);
Yasunori Goto7b78d332007-10-21 16:41:36 -0700848 memory_notify(MEM_CANCEL_OFFLINE, &arg);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700849 /* pushback to free area */
850 undo_isolate_page_range(start_pfn, end_pfn);
Yasunori Goto7b78d332007-10-21 16:41:36 -0700851
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700852 return ret;
853}
Badari Pulavarty71088782008-10-18 20:25:58 -0700854
855int remove_memory(u64 start, u64 size)
856{
857 unsigned long start_pfn, end_pfn;
858
859 start_pfn = PFN_DOWN(start);
860 end_pfn = start_pfn + PFN_DOWN(size);
861 return offline_pages(start_pfn, end_pfn, 120 * HZ);
862}
KAMEZAWA Hiroyuki48e94192007-10-16 01:26:14 -0700863#else
864int remove_memory(u64 start, u64 size)
865{
866 return -EINVAL;
867}
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700868#endif /* CONFIG_MEMORY_HOTREMOVE */
Badari Pulavarty71088782008-10-18 20:25:58 -0700869EXPORT_SYMBOL_GPL(remove_memory);