Commit | Line | Data |
---|---|---|
3947be19 DH |
1 | /* |
2 | * linux/mm/memory_hotplug.c | |
3 | * | |
4 | * Copyright (C) | |
5 | */ | |
6 | ||
3947be19 DH |
7 | #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> | |
2d1d43f6 | 16 | #include <linux/writeback.h> |
3947be19 DH |
17 | #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> | |
0a547039 | 24 | #include <linux/ioport.h> |
0c0e6195 KH |
25 | #include <linux/delay.h> |
26 | #include <linux/migrate.h> | |
27 | #include <linux/page-isolation.h> | |
71088785 | 28 | #include <linux/pfn.h> |
6ad696d2 | 29 | #include <linux/suspend.h> |
6d9c285a | 30 | #include <linux/mm_inline.h> |
d96ae530 | 31 | #include <linux/firmware-map.h> |
3947be19 DH |
32 | |
33 | #include <asm/tlbflush.h> | |
34 | ||
1e5ad9a3 AB |
35 | #include "internal.h" |
36 | ||
20d6c96b KM |
37 | DEFINE_MUTEX(mem_hotplug_mutex); |
38 | ||
39 | void lock_memory_hotplug(void) | |
40 | { | |
41 | mutex_lock(&mem_hotplug_mutex); | |
42 | ||
43 | /* for exclusive hibernation if CONFIG_HIBERNATION=y */ | |
44 | lock_system_sleep(); | |
45 | } | |
46 | ||
47 | void unlock_memory_hotplug(void) | |
48 | { | |
49 | unlock_system_sleep(); | |
50 | mutex_unlock(&mem_hotplug_mutex); | |
51 | } | |
52 | ||
53 | ||
45e0b78b KM |
54 | /* add this memory to iomem resource */ |
55 | static struct resource *register_memory_resource(u64 start, u64 size) | |
56 | { | |
57 | struct resource *res; | |
58 | res = kzalloc(sizeof(struct resource), GFP_KERNEL); | |
59 | BUG_ON(!res); | |
60 | ||
61 | res->name = "System RAM"; | |
62 | res->start = start; | |
63 | res->end = start + size - 1; | |
887c3cb1 | 64 | res->flags = IORESOURCE_MEM | IORESOURCE_BUSY; |
45e0b78b KM |
65 | if (request_resource(&iomem_resource, res) < 0) { |
66 | printk("System RAM resource %llx - %llx cannot be added\n", | |
67 | (unsigned long long)res->start, (unsigned long long)res->end); | |
68 | kfree(res); | |
69 | res = NULL; | |
70 | } | |
71 | return res; | |
72 | } | |
73 | ||
74 | static void release_memory_resource(struct resource *res) | |
75 | { | |
76 | if (!res) | |
77 | return; | |
78 | release_resource(res); | |
79 | kfree(res); | |
80 | return; | |
81 | } | |
82 | ||
53947027 | 83 | #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE |
04753278 | 84 | #ifndef CONFIG_SPARSEMEM_VMEMMAP |
af370fb8 | 85 | static void get_page_bootmem(unsigned long info, struct page *page, int type) |
04753278 | 86 | { |
af370fb8 | 87 | atomic_set(&page->_mapcount, type); |
04753278 YG |
88 | SetPagePrivate(page); |
89 | set_page_private(page, info); | |
90 | atomic_inc(&page->_count); | |
91 | } | |
92 | ||
23ce932a RM |
93 | /* reference to __meminit __free_pages_bootmem is valid |
94 | * so use __ref to tell modpost not to generate a warning */ | |
95 | void __ref put_page_bootmem(struct page *page) | |
04753278 | 96 | { |
af370fb8 | 97 | int type; |
04753278 | 98 | |
af370fb8 YG |
99 | type = atomic_read(&page->_mapcount); |
100 | BUG_ON(type >= -1); | |
04753278 YG |
101 | |
102 | if (atomic_dec_return(&page->_count) == 1) { | |
103 | ClearPagePrivate(page); | |
104 | set_page_private(page, 0); | |
105 | reset_page_mapcount(page); | |
106 | __free_pages_bootmem(page, 0); | |
107 | } | |
108 | ||
109 | } | |
110 | ||
d92bc318 | 111 | static void register_page_bootmem_info_section(unsigned long start_pfn) |
04753278 YG |
112 | { |
113 | unsigned long *usemap, mapsize, section_nr, i; | |
114 | struct mem_section *ms; | |
115 | struct page *page, *memmap; | |
116 | ||
117 | if (!pfn_valid(start_pfn)) | |
118 | return; | |
119 | ||
120 | section_nr = pfn_to_section_nr(start_pfn); | |
121 | ms = __nr_to_section(section_nr); | |
122 | ||
123 | /* Get section's memmap address */ | |
124 | memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr); | |
125 | ||
126 | /* | |
127 | * Get page for the memmap's phys address | |
128 | * XXX: need more consideration for sparse_vmemmap... | |
129 | */ | |
130 | page = virt_to_page(memmap); | |
131 | mapsize = sizeof(struct page) * PAGES_PER_SECTION; | |
132 | mapsize = PAGE_ALIGN(mapsize) >> PAGE_SHIFT; | |
133 | ||
134 | /* remember memmap's page */ | |
135 | for (i = 0; i < mapsize; i++, page++) | |
136 | get_page_bootmem(section_nr, page, SECTION_INFO); | |
137 | ||
138 | usemap = __nr_to_section(section_nr)->pageblock_flags; | |
139 | page = virt_to_page(usemap); | |
140 | ||
141 | mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT; | |
142 | ||
143 | for (i = 0; i < mapsize; i++, page++) | |
af370fb8 | 144 | get_page_bootmem(section_nr, page, MIX_SECTION_INFO); |
04753278 YG |
145 | |
146 | } | |
147 | ||
148 | void register_page_bootmem_info_node(struct pglist_data *pgdat) | |
149 | { | |
150 | unsigned long i, pfn, end_pfn, nr_pages; | |
151 | int node = pgdat->node_id; | |
152 | struct page *page; | |
153 | struct zone *zone; | |
154 | ||
155 | nr_pages = PAGE_ALIGN(sizeof(struct pglist_data)) >> PAGE_SHIFT; | |
156 | page = virt_to_page(pgdat); | |
157 | ||
158 | for (i = 0; i < nr_pages; i++, page++) | |
159 | get_page_bootmem(node, page, NODE_INFO); | |
160 | ||
161 | zone = &pgdat->node_zones[0]; | |
162 | for (; zone < pgdat->node_zones + MAX_NR_ZONES - 1; zone++) { | |
163 | if (zone->wait_table) { | |
164 | nr_pages = zone->wait_table_hash_nr_entries | |
165 | * sizeof(wait_queue_head_t); | |
166 | nr_pages = PAGE_ALIGN(nr_pages) >> PAGE_SHIFT; | |
167 | page = virt_to_page(zone->wait_table); | |
168 | ||
169 | for (i = 0; i < nr_pages; i++, page++) | |
170 | get_page_bootmem(node, page, NODE_INFO); | |
171 | } | |
172 | } | |
173 | ||
174 | pfn = pgdat->node_start_pfn; | |
175 | end_pfn = pfn + pgdat->node_spanned_pages; | |
176 | ||
177 | /* register_section info */ | |
178 | for (; pfn < end_pfn; pfn += PAGES_PER_SECTION) | |
179 | register_page_bootmem_info_section(pfn); | |
180 | ||
181 | } | |
182 | #endif /* !CONFIG_SPARSEMEM_VMEMMAP */ | |
183 | ||
76cdd58e HC |
184 | static void grow_zone_span(struct zone *zone, unsigned long start_pfn, |
185 | unsigned long end_pfn) | |
186 | { | |
187 | unsigned long old_zone_end_pfn; | |
188 | ||
189 | zone_span_writelock(zone); | |
190 | ||
191 | old_zone_end_pfn = zone->zone_start_pfn + zone->spanned_pages; | |
192 | if (start_pfn < zone->zone_start_pfn) | |
193 | zone->zone_start_pfn = start_pfn; | |
194 | ||
195 | zone->spanned_pages = max(old_zone_end_pfn, end_pfn) - | |
196 | zone->zone_start_pfn; | |
197 | ||
198 | zone_span_writeunlock(zone); | |
199 | } | |
200 | ||
201 | static void grow_pgdat_span(struct pglist_data *pgdat, unsigned long start_pfn, | |
202 | unsigned long end_pfn) | |
203 | { | |
204 | unsigned long old_pgdat_end_pfn = | |
205 | pgdat->node_start_pfn + pgdat->node_spanned_pages; | |
206 | ||
207 | if (start_pfn < pgdat->node_start_pfn) | |
208 | pgdat->node_start_pfn = start_pfn; | |
209 | ||
210 | pgdat->node_spanned_pages = max(old_pgdat_end_pfn, end_pfn) - | |
211 | pgdat->node_start_pfn; | |
212 | } | |
213 | ||
31168481 | 214 | static int __meminit __add_zone(struct zone *zone, unsigned long phys_start_pfn) |
3947be19 DH |
215 | { |
216 | struct pglist_data *pgdat = zone->zone_pgdat; | |
217 | int nr_pages = PAGES_PER_SECTION; | |
218 | int nid = pgdat->node_id; | |
219 | int zone_type; | |
76cdd58e | 220 | unsigned long flags; |
3947be19 DH |
221 | |
222 | zone_type = zone - pgdat->node_zones; | |
76cdd58e HC |
223 | if (!zone->wait_table) { |
224 | int ret; | |
225 | ||
226 | ret = init_currently_empty_zone(zone, phys_start_pfn, | |
227 | nr_pages, MEMMAP_HOTPLUG); | |
228 | if (ret) | |
229 | return ret; | |
230 | } | |
231 | pgdat_resize_lock(zone->zone_pgdat, &flags); | |
232 | grow_zone_span(zone, phys_start_pfn, phys_start_pfn + nr_pages); | |
233 | grow_pgdat_span(zone->zone_pgdat, phys_start_pfn, | |
234 | phys_start_pfn + nr_pages); | |
235 | pgdat_resize_unlock(zone->zone_pgdat, &flags); | |
a2f3aa02 DH |
236 | memmap_init_zone(nr_pages, nid, zone_type, |
237 | phys_start_pfn, MEMMAP_HOTPLUG); | |
718127cc | 238 | return 0; |
3947be19 DH |
239 | } |
240 | ||
c04fc586 GH |
241 | static int __meminit __add_section(int nid, struct zone *zone, |
242 | unsigned long phys_start_pfn) | |
3947be19 | 243 | { |
3947be19 | 244 | int nr_pages = PAGES_PER_SECTION; |
3947be19 DH |
245 | int ret; |
246 | ||
ebd15302 KH |
247 | if (pfn_valid(phys_start_pfn)) |
248 | return -EEXIST; | |
249 | ||
0b0acbec | 250 | ret = sparse_add_one_section(zone, phys_start_pfn, nr_pages); |
3947be19 DH |
251 | |
252 | if (ret < 0) | |
253 | return ret; | |
254 | ||
718127cc YG |
255 | ret = __add_zone(zone, phys_start_pfn); |
256 | ||
257 | if (ret < 0) | |
258 | return ret; | |
259 | ||
c04fc586 | 260 | return register_new_memory(nid, __pfn_to_section(phys_start_pfn)); |
3947be19 DH |
261 | } |
262 | ||
0c0a4a51 YG |
263 | #ifdef CONFIG_SPARSEMEM_VMEMMAP |
264 | static int __remove_section(struct zone *zone, struct mem_section *ms) | |
265 | { | |
266 | /* | |
267 | * XXX: Freeing memmap with vmemmap is not implement yet. | |
268 | * This should be removed later. | |
269 | */ | |
270 | return -EBUSY; | |
271 | } | |
272 | #else | |
ea01ea93 BP |
273 | static int __remove_section(struct zone *zone, struct mem_section *ms) |
274 | { | |
275 | unsigned long flags; | |
276 | struct pglist_data *pgdat = zone->zone_pgdat; | |
277 | int ret = -EINVAL; | |
278 | ||
279 | if (!valid_section(ms)) | |
280 | return ret; | |
281 | ||
282 | ret = unregister_memory_section(ms); | |
283 | if (ret) | |
284 | return ret; | |
285 | ||
286 | pgdat_resize_lock(pgdat, &flags); | |
287 | sparse_remove_one_section(zone, ms); | |
288 | pgdat_resize_unlock(pgdat, &flags); | |
289 | return 0; | |
290 | } | |
0c0a4a51 | 291 | #endif |
ea01ea93 | 292 | |
3947be19 DH |
293 | /* |
294 | * Reasonably generic function for adding memory. It is | |
295 | * expected that archs that support memory hotplug will | |
296 | * call this function after deciding the zone to which to | |
297 | * add the new pages. | |
298 | */ | |
c04fc586 GH |
299 | int __ref __add_pages(int nid, struct zone *zone, unsigned long phys_start_pfn, |
300 | unsigned long nr_pages) | |
3947be19 DH |
301 | { |
302 | unsigned long i; | |
303 | int err = 0; | |
6f712711 KH |
304 | int start_sec, end_sec; |
305 | /* during initialize mem_map, align hot-added range to section */ | |
306 | start_sec = pfn_to_section_nr(phys_start_pfn); | |
307 | end_sec = pfn_to_section_nr(phys_start_pfn + nr_pages - 1); | |
3947be19 | 308 | |
6f712711 | 309 | for (i = start_sec; i <= end_sec; i++) { |
c04fc586 | 310 | err = __add_section(nid, zone, i << PFN_SECTION_SHIFT); |
3947be19 | 311 | |
6f712711 | 312 | /* |
183ff22b | 313 | * EEXIST is finally dealt with by ioresource collision |
6f712711 KH |
314 | * check. see add_memory() => register_memory_resource() |
315 | * Warning will be printed if there is collision. | |
bed120c6 JS |
316 | */ |
317 | if (err && (err != -EEXIST)) | |
3947be19 | 318 | break; |
6f712711 | 319 | err = 0; |
3947be19 DH |
320 | } |
321 | ||
322 | return err; | |
323 | } | |
bed120c6 | 324 | EXPORT_SYMBOL_GPL(__add_pages); |
3947be19 | 325 | |
ea01ea93 BP |
326 | /** |
327 | * __remove_pages() - remove sections of pages from a zone | |
328 | * @zone: zone from which pages need to be removed | |
329 | * @phys_start_pfn: starting pageframe (must be aligned to start of a section) | |
330 | * @nr_pages: number of pages to remove (must be multiple of section size) | |
331 | * | |
332 | * Generic helper function to remove section mappings and sysfs entries | |
333 | * for the section of the memory we are removing. Caller needs to make | |
334 | * sure that pages are marked reserved and zones are adjust properly by | |
335 | * calling offline_pages(). | |
336 | */ | |
337 | int __remove_pages(struct zone *zone, unsigned long phys_start_pfn, | |
338 | unsigned long nr_pages) | |
339 | { | |
340 | unsigned long i, ret = 0; | |
341 | int sections_to_remove; | |
342 | ||
343 | /* | |
344 | * We can only remove entire sections | |
345 | */ | |
346 | BUG_ON(phys_start_pfn & ~PAGE_SECTION_MASK); | |
347 | BUG_ON(nr_pages % PAGES_PER_SECTION); | |
348 | ||
ea01ea93 BP |
349 | sections_to_remove = nr_pages / PAGES_PER_SECTION; |
350 | for (i = 0; i < sections_to_remove; i++) { | |
351 | unsigned long pfn = phys_start_pfn + i*PAGES_PER_SECTION; | |
de7f0cba NF |
352 | release_mem_region(pfn << PAGE_SHIFT, |
353 | PAGES_PER_SECTION << PAGE_SHIFT); | |
ea01ea93 BP |
354 | ret = __remove_section(zone, __pfn_to_section(pfn)); |
355 | if (ret) | |
356 | break; | |
357 | } | |
358 | return ret; | |
359 | } | |
360 | EXPORT_SYMBOL_GPL(__remove_pages); | |
361 | ||
180c06ef JF |
362 | void online_page(struct page *page) |
363 | { | |
4738e1b9 JB |
364 | unsigned long pfn = page_to_pfn(page); |
365 | ||
180c06ef | 366 | totalram_pages++; |
4738e1b9 JB |
367 | if (pfn >= num_physpages) |
368 | num_physpages = pfn + 1; | |
180c06ef JF |
369 | |
370 | #ifdef CONFIG_HIGHMEM | |
371 | if (PageHighMem(page)) | |
372 | totalhigh_pages++; | |
373 | #endif | |
374 | ||
375 | #ifdef CONFIG_FLATMEM | |
376 | max_mapnr = max(page_to_pfn(page), max_mapnr); | |
377 | #endif | |
378 | ||
379 | ClearPageReserved(page); | |
380 | init_page_count(page); | |
381 | __free_page(page); | |
382 | } | |
383 | ||
75884fb1 KH |
384 | static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages, |
385 | void *arg) | |
3947be19 DH |
386 | { |
387 | unsigned long i; | |
75884fb1 KH |
388 | unsigned long onlined_pages = *(unsigned long *)arg; |
389 | struct page *page; | |
390 | if (PageReserved(pfn_to_page(start_pfn))) | |
391 | for (i = 0; i < nr_pages; i++) { | |
392 | page = pfn_to_page(start_pfn + i); | |
393 | online_page(page); | |
394 | onlined_pages++; | |
395 | } | |
396 | *(unsigned long *)arg = onlined_pages; | |
397 | return 0; | |
398 | } | |
399 | ||
400 | ||
401 | int online_pages(unsigned long pfn, unsigned long nr_pages) | |
402 | { | |
3947be19 DH |
403 | unsigned long onlined_pages = 0; |
404 | struct zone *zone; | |
6811378e | 405 | int need_zonelists_rebuild = 0; |
7b78d335 YG |
406 | int nid; |
407 | int ret; | |
408 | struct memory_notify arg; | |
409 | ||
925268a0 | 410 | lock_memory_hotplug(); |
7b78d335 YG |
411 | arg.start_pfn = pfn; |
412 | arg.nr_pages = nr_pages; | |
413 | arg.status_change_nid = -1; | |
414 | ||
415 | nid = page_to_nid(pfn_to_page(pfn)); | |
416 | if (node_present_pages(nid) == 0) | |
417 | arg.status_change_nid = nid; | |
3947be19 | 418 | |
7b78d335 YG |
419 | ret = memory_notify(MEM_GOING_ONLINE, &arg); |
420 | ret = notifier_to_errno(ret); | |
421 | if (ret) { | |
422 | memory_notify(MEM_CANCEL_ONLINE, &arg); | |
925268a0 | 423 | unlock_memory_hotplug(); |
7b78d335 YG |
424 | return ret; |
425 | } | |
3947be19 DH |
426 | /* |
427 | * This doesn't need a lock to do pfn_to_page(). | |
428 | * The section can't be removed here because of the | |
da19cbcf | 429 | * memory_block->state_mutex. |
3947be19 DH |
430 | */ |
431 | zone = page_zone(pfn_to_page(pfn)); | |
6811378e YG |
432 | /* |
433 | * If this zone is not populated, then it is not in zonelist. | |
434 | * This means the page allocator ignores this zone. | |
435 | * So, zonelist must be updated after online. | |
436 | */ | |
4eaf3f64 | 437 | mutex_lock(&zonelists_mutex); |
6811378e YG |
438 | if (!populated_zone(zone)) |
439 | need_zonelists_rebuild = 1; | |
440 | ||
908eedc6 | 441 | ret = walk_system_ram_range(pfn, nr_pages, &onlined_pages, |
75884fb1 | 442 | online_pages_range); |
fd8a4221 | 443 | if (ret) { |
4eaf3f64 | 444 | mutex_unlock(&zonelists_mutex); |
fd8a4221 GL |
445 | printk(KERN_DEBUG "online_pages %lx at %lx failed\n", |
446 | nr_pages, pfn); | |
447 | memory_notify(MEM_CANCEL_ONLINE, &arg); | |
925268a0 | 448 | unlock_memory_hotplug(); |
fd8a4221 GL |
449 | return ret; |
450 | } | |
451 | ||
3947be19 | 452 | zone->present_pages += onlined_pages; |
f2937be5 | 453 | zone->zone_pgdat->node_present_pages += onlined_pages; |
1f522509 HL |
454 | if (need_zonelists_rebuild) |
455 | build_all_zonelists(zone); | |
456 | else | |
457 | zone_pcp_update(zone); | |
3947be19 | 458 | |
4eaf3f64 | 459 | mutex_unlock(&zonelists_mutex); |
bc75d33f | 460 | setup_per_zone_wmarks(); |
bce7394a | 461 | calculate_zone_inactive_ratio(zone); |
7ea1530a CL |
462 | if (onlined_pages) { |
463 | kswapd_run(zone_to_nid(zone)); | |
464 | node_set_state(zone_to_nid(zone), N_HIGH_MEMORY); | |
465 | } | |
61b13993 | 466 | |
1f522509 | 467 | vm_total_pages = nr_free_pagecache_pages(); |
2f7f24ec | 468 | |
2d1d43f6 | 469 | writeback_set_ratelimit(); |
7b78d335 YG |
470 | |
471 | if (onlined_pages) | |
472 | memory_notify(MEM_ONLINE, &arg); | |
925268a0 | 473 | unlock_memory_hotplug(); |
7b78d335 | 474 | |
3947be19 DH |
475 | return 0; |
476 | } | |
53947027 | 477 | #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */ |
bc02af93 | 478 | |
e1319331 HS |
479 | /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */ |
480 | static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start) | |
9af3c2de YG |
481 | { |
482 | struct pglist_data *pgdat; | |
483 | unsigned long zones_size[MAX_NR_ZONES] = {0}; | |
484 | unsigned long zholes_size[MAX_NR_ZONES] = {0}; | |
485 | unsigned long start_pfn = start >> PAGE_SHIFT; | |
486 | ||
487 | pgdat = arch_alloc_nodedata(nid); | |
488 | if (!pgdat) | |
489 | return NULL; | |
490 | ||
491 | arch_refresh_nodedata(nid, pgdat); | |
492 | ||
493 | /* we can use NODE_DATA(nid) from here */ | |
494 | ||
495 | /* init node's zones as empty zones, we don't have any present pages.*/ | |
9109fb7b | 496 | free_area_init_node(nid, zones_size, start_pfn, zholes_size); |
9af3c2de YG |
497 | |
498 | return pgdat; | |
499 | } | |
500 | ||
501 | static void rollback_node_hotadd(int nid, pg_data_t *pgdat) | |
502 | { | |
503 | arch_refresh_nodedata(nid, NULL); | |
504 | arch_free_nodedata(pgdat); | |
505 | return; | |
506 | } | |
507 | ||
0a547039 | 508 | |
cf23422b | 509 | /* |
510 | * called by cpu_up() to online a node without onlined memory. | |
511 | */ | |
512 | int mem_online_node(int nid) | |
513 | { | |
514 | pg_data_t *pgdat; | |
515 | int ret; | |
516 | ||
20d6c96b | 517 | lock_memory_hotplug(); |
cf23422b | 518 | pgdat = hotadd_new_pgdat(nid, 0); |
519 | if (pgdat) { | |
520 | ret = -ENOMEM; | |
521 | goto out; | |
522 | } | |
523 | node_set_online(nid); | |
524 | ret = register_one_node(nid); | |
525 | BUG_ON(ret); | |
526 | ||
527 | out: | |
20d6c96b | 528 | unlock_memory_hotplug(); |
cf23422b | 529 | return ret; |
530 | } | |
531 | ||
31168481 AV |
532 | /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */ |
533 | int __ref add_memory(int nid, u64 start, u64 size) | |
bc02af93 | 534 | { |
9af3c2de YG |
535 | pg_data_t *pgdat = NULL; |
536 | int new_pgdat = 0; | |
ebd15302 | 537 | struct resource *res; |
bc02af93 YG |
538 | int ret; |
539 | ||
20d6c96b | 540 | lock_memory_hotplug(); |
6ad696d2 | 541 | |
ebd15302 | 542 | res = register_memory_resource(start, size); |
6ad696d2 | 543 | ret = -EEXIST; |
ebd15302 | 544 | if (!res) |
6ad696d2 | 545 | goto out; |
ebd15302 | 546 | |
9af3c2de YG |
547 | if (!node_online(nid)) { |
548 | pgdat = hotadd_new_pgdat(nid, start); | |
6ad696d2 | 549 | ret = -ENOMEM; |
9af3c2de | 550 | if (!pgdat) |
6ad696d2 | 551 | goto out; |
9af3c2de | 552 | new_pgdat = 1; |
9af3c2de YG |
553 | } |
554 | ||
bc02af93 YG |
555 | /* call arch's memory hotadd */ |
556 | ret = arch_add_memory(nid, start, size); | |
557 | ||
9af3c2de YG |
558 | if (ret < 0) |
559 | goto error; | |
560 | ||
0fc44159 | 561 | /* we online node here. we can't roll back from here. */ |
9af3c2de YG |
562 | node_set_online(nid); |
563 | ||
0fc44159 YG |
564 | if (new_pgdat) { |
565 | ret = register_one_node(nid); | |
566 | /* | |
567 | * If sysfs file of new node can't create, cpu on the node | |
568 | * can't be hot-added. There is no rollback way now. | |
569 | * So, check by BUG_ON() to catch it reluctantly.. | |
570 | */ | |
571 | BUG_ON(ret); | |
572 | } | |
573 | ||
d96ae530 | 574 | /* create new memmap entry */ |
575 | firmware_map_add_hotplug(start, start + size, "System RAM"); | |
576 | ||
6ad696d2 AK |
577 | goto out; |
578 | ||
9af3c2de YG |
579 | error: |
580 | /* rollback pgdat allocation and others */ | |
581 | if (new_pgdat) | |
582 | rollback_node_hotadd(nid, pgdat); | |
ebd15302 KH |
583 | if (res) |
584 | release_memory_resource(res); | |
9af3c2de | 585 | |
6ad696d2 | 586 | out: |
20d6c96b | 587 | unlock_memory_hotplug(); |
bc02af93 YG |
588 | return ret; |
589 | } | |
590 | EXPORT_SYMBOL_GPL(add_memory); | |
0c0e6195 KH |
591 | |
592 | #ifdef CONFIG_MEMORY_HOTREMOVE | |
5c755e9f BP |
593 | /* |
594 | * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy | |
595 | * set and the size of the free page is given by page_order(). Using this, | |
596 | * the function determines if the pageblock contains only free pages. | |
597 | * Due to buddy contraints, a free page at least the size of a pageblock will | |
598 | * be located at the start of the pageblock | |
599 | */ | |
600 | static inline int pageblock_free(struct page *page) | |
601 | { | |
602 | return PageBuddy(page) && page_order(page) >= pageblock_order; | |
603 | } | |
604 | ||
605 | /* Return the start of the next active pageblock after a given page */ | |
606 | static struct page *next_active_pageblock(struct page *page) | |
607 | { | |
5c755e9f BP |
608 | /* Ensure the starting page is pageblock-aligned */ |
609 | BUG_ON(page_to_pfn(page) & (pageblock_nr_pages - 1)); | |
610 | ||
5c755e9f | 611 | /* If the entire pageblock is free, move to the end of free page */ |
0dcc48c1 KH |
612 | if (pageblock_free(page)) { |
613 | int order; | |
614 | /* be careful. we don't have locks, page_order can be changed.*/ | |
615 | order = page_order(page); | |
616 | if ((order < MAX_ORDER) && (order >= pageblock_order)) | |
617 | return page + (1 << order); | |
618 | } | |
5c755e9f | 619 | |
0dcc48c1 | 620 | return page + pageblock_nr_pages; |
5c755e9f BP |
621 | } |
622 | ||
623 | /* Checks if this range of memory is likely to be hot-removable. */ | |
624 | int is_mem_section_removable(unsigned long start_pfn, unsigned long nr_pages) | |
625 | { | |
5c755e9f BP |
626 | struct page *page = pfn_to_page(start_pfn); |
627 | struct page *end_page = page + nr_pages; | |
628 | ||
629 | /* Check the starting page of each pageblock within the range */ | |
630 | for (; page < end_page; page = next_active_pageblock(page)) { | |
49ac8255 | 631 | if (!is_pageblock_removable_nolock(page)) |
5c755e9f | 632 | return 0; |
49ac8255 | 633 | cond_resched(); |
5c755e9f BP |
634 | } |
635 | ||
636 | /* All pageblocks in the memory block are likely to be hot-removable */ | |
637 | return 1; | |
638 | } | |
639 | ||
0c0e6195 KH |
640 | /* |
641 | * Confirm all pages in a range [start, end) is belongs to the same zone. | |
642 | */ | |
643 | static int test_pages_in_a_zone(unsigned long start_pfn, unsigned long end_pfn) | |
644 | { | |
645 | unsigned long pfn; | |
646 | struct zone *zone = NULL; | |
647 | struct page *page; | |
648 | int i; | |
649 | for (pfn = start_pfn; | |
650 | pfn < end_pfn; | |
651 | pfn += MAX_ORDER_NR_PAGES) { | |
652 | i = 0; | |
653 | /* This is just a CONFIG_HOLES_IN_ZONE check.*/ | |
654 | while ((i < MAX_ORDER_NR_PAGES) && !pfn_valid_within(pfn + i)) | |
655 | i++; | |
656 | if (i == MAX_ORDER_NR_PAGES) | |
657 | continue; | |
658 | page = pfn_to_page(pfn + i); | |
659 | if (zone && page_zone(page) != zone) | |
660 | return 0; | |
661 | zone = page_zone(page); | |
662 | } | |
663 | return 1; | |
664 | } | |
665 | ||
666 | /* | |
667 | * Scanning pfn is much easier than scanning lru list. | |
668 | * Scan pfn from start to end and Find LRU page. | |
669 | */ | |
7bbc0905 | 670 | static unsigned long scan_lru_pages(unsigned long start, unsigned long end) |
0c0e6195 KH |
671 | { |
672 | unsigned long pfn; | |
673 | struct page *page; | |
674 | for (pfn = start; pfn < end; pfn++) { | |
675 | if (pfn_valid(pfn)) { | |
676 | page = pfn_to_page(pfn); | |
677 | if (PageLRU(page)) | |
678 | return pfn; | |
679 | } | |
680 | } | |
681 | return 0; | |
682 | } | |
683 | ||
684 | static struct page * | |
3c1d4378 | 685 | hotremove_migrate_alloc(struct page *page, unsigned long private, int **x) |
0c0e6195 | 686 | { |
3c1d4378 HD |
687 | /* This should be improooooved!! */ |
688 | return alloc_page(GFP_HIGHUSER_MOVABLE); | |
0c0e6195 KH |
689 | } |
690 | ||
0c0e6195 KH |
691 | #define NR_OFFLINE_AT_ONCE_PAGES (256) |
692 | static int | |
693 | do_migrate_range(unsigned long start_pfn, unsigned long end_pfn) | |
694 | { | |
695 | unsigned long pfn; | |
696 | struct page *page; | |
697 | int move_pages = NR_OFFLINE_AT_ONCE_PAGES; | |
698 | int not_managed = 0; | |
699 | int ret = 0; | |
700 | LIST_HEAD(source); | |
701 | ||
702 | for (pfn = start_pfn; pfn < end_pfn && move_pages > 0; pfn++) { | |
703 | if (!pfn_valid(pfn)) | |
704 | continue; | |
705 | page = pfn_to_page(pfn); | |
706 | if (!page_count(page)) | |
707 | continue; | |
708 | /* | |
709 | * We can skip free pages. And we can only deal with pages on | |
710 | * LRU. | |
711 | */ | |
62695a84 | 712 | ret = isolate_lru_page(page); |
0c0e6195 | 713 | if (!ret) { /* Success */ |
62695a84 | 714 | list_add_tail(&page->lru, &source); |
0c0e6195 | 715 | move_pages--; |
6d9c285a KM |
716 | inc_zone_page_state(page, NR_ISOLATED_ANON + |
717 | page_is_file_cache(page)); | |
718 | ||
0c0e6195 | 719 | } else { |
0c0e6195 | 720 | #ifdef CONFIG_DEBUG_VM |
718a3821 WF |
721 | printk(KERN_ALERT "removing pfn %lx from LRU failed\n", |
722 | pfn); | |
723 | dump_page(page); | |
0c0e6195 | 724 | #endif |
809c4449 BL |
725 | /* Becasue we don't have big zone->lock. we should |
726 | check this again here. */ | |
727 | if (page_count(page)) { | |
728 | not_managed++; | |
f3ab2636 | 729 | ret = -EBUSY; |
809c4449 BL |
730 | break; |
731 | } | |
0c0e6195 KH |
732 | } |
733 | } | |
f3ab2636 BL |
734 | if (!list_empty(&source)) { |
735 | if (not_managed) { | |
736 | putback_lru_pages(&source); | |
737 | goto out; | |
738 | } | |
739 | /* this function returns # of failed pages */ | |
740 | ret = migrate_pages(&source, hotremove_migrate_alloc, 0, 1); | |
741 | if (ret) | |
0c0e6195 | 742 | putback_lru_pages(&source); |
0c0e6195 | 743 | } |
0c0e6195 KH |
744 | out: |
745 | return ret; | |
746 | } | |
747 | ||
748 | /* | |
749 | * remove from free_area[] and mark all as Reserved. | |
750 | */ | |
751 | static int | |
752 | offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages, | |
753 | void *data) | |
754 | { | |
755 | __offline_isolated_pages(start, start + nr_pages); | |
756 | return 0; | |
757 | } | |
758 | ||
759 | static void | |
760 | offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn) | |
761 | { | |
908eedc6 | 762 | walk_system_ram_range(start_pfn, end_pfn - start_pfn, NULL, |
0c0e6195 KH |
763 | offline_isolated_pages_cb); |
764 | } | |
765 | ||
766 | /* | |
767 | * Check all pages in range, recoreded as memory resource, are isolated. | |
768 | */ | |
769 | static int | |
770 | check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages, | |
771 | void *data) | |
772 | { | |
773 | int ret; | |
774 | long offlined = *(long *)data; | |
775 | ret = test_pages_isolated(start_pfn, start_pfn + nr_pages); | |
776 | offlined = nr_pages; | |
777 | if (!ret) | |
778 | *(long *)data += offlined; | |
779 | return ret; | |
780 | } | |
781 | ||
782 | static long | |
783 | check_pages_isolated(unsigned long start_pfn, unsigned long end_pfn) | |
784 | { | |
785 | long offlined = 0; | |
786 | int ret; | |
787 | ||
908eedc6 | 788 | ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn, &offlined, |
0c0e6195 KH |
789 | check_pages_isolated_cb); |
790 | if (ret < 0) | |
791 | offlined = (long)ret; | |
792 | return offlined; | |
793 | } | |
794 | ||
b4e655a4 | 795 | static int offline_pages(unsigned long start_pfn, |
0c0e6195 KH |
796 | unsigned long end_pfn, unsigned long timeout) |
797 | { | |
798 | unsigned long pfn, nr_pages, expire; | |
799 | long offlined_pages; | |
7b78d335 | 800 | int ret, drain, retry_max, node; |
0c0e6195 | 801 | struct zone *zone; |
7b78d335 | 802 | struct memory_notify arg; |
0c0e6195 KH |
803 | |
804 | BUG_ON(start_pfn >= end_pfn); | |
805 | /* at least, alignment against pageblock is necessary */ | |
806 | if (!IS_ALIGNED(start_pfn, pageblock_nr_pages)) | |
807 | return -EINVAL; | |
808 | if (!IS_ALIGNED(end_pfn, pageblock_nr_pages)) | |
809 | return -EINVAL; | |
810 | /* This makes hotplug much easier...and readable. | |
811 | we assume this for now. .*/ | |
812 | if (!test_pages_in_a_zone(start_pfn, end_pfn)) | |
813 | return -EINVAL; | |
7b78d335 | 814 | |
20d6c96b | 815 | lock_memory_hotplug(); |
6ad696d2 | 816 | |
7b78d335 YG |
817 | zone = page_zone(pfn_to_page(start_pfn)); |
818 | node = zone_to_nid(zone); | |
819 | nr_pages = end_pfn - start_pfn; | |
820 | ||
0c0e6195 KH |
821 | /* set above range as isolated */ |
822 | ret = start_isolate_page_range(start_pfn, end_pfn); | |
823 | if (ret) | |
6ad696d2 | 824 | goto out; |
7b78d335 YG |
825 | |
826 | arg.start_pfn = start_pfn; | |
827 | arg.nr_pages = nr_pages; | |
828 | arg.status_change_nid = -1; | |
829 | if (nr_pages >= node_present_pages(node)) | |
830 | arg.status_change_nid = node; | |
831 | ||
832 | ret = memory_notify(MEM_GOING_OFFLINE, &arg); | |
833 | ret = notifier_to_errno(ret); | |
834 | if (ret) | |
835 | goto failed_removal; | |
836 | ||
0c0e6195 KH |
837 | pfn = start_pfn; |
838 | expire = jiffies + timeout; | |
839 | drain = 0; | |
840 | retry_max = 5; | |
841 | repeat: | |
842 | /* start memory hot removal */ | |
843 | ret = -EAGAIN; | |
844 | if (time_after(jiffies, expire)) | |
845 | goto failed_removal; | |
846 | ret = -EINTR; | |
847 | if (signal_pending(current)) | |
848 | goto failed_removal; | |
849 | ret = 0; | |
850 | if (drain) { | |
851 | lru_add_drain_all(); | |
0c0e6195 | 852 | cond_resched(); |
9f8f2172 | 853 | drain_all_pages(); |
0c0e6195 KH |
854 | } |
855 | ||
856 | pfn = scan_lru_pages(start_pfn, end_pfn); | |
857 | if (pfn) { /* We have page on LRU */ | |
858 | ret = do_migrate_range(pfn, end_pfn); | |
859 | if (!ret) { | |
860 | drain = 1; | |
861 | goto repeat; | |
862 | } else { | |
863 | if (ret < 0) | |
864 | if (--retry_max == 0) | |
865 | goto failed_removal; | |
866 | yield(); | |
867 | drain = 1; | |
868 | goto repeat; | |
869 | } | |
870 | } | |
871 | /* drain all zone's lru pagevec, this is asyncronous... */ | |
872 | lru_add_drain_all(); | |
0c0e6195 KH |
873 | yield(); |
874 | /* drain pcp pages , this is synchrouns. */ | |
9f8f2172 | 875 | drain_all_pages(); |
0c0e6195 KH |
876 | /* check again */ |
877 | offlined_pages = check_pages_isolated(start_pfn, end_pfn); | |
878 | if (offlined_pages < 0) { | |
879 | ret = -EBUSY; | |
880 | goto failed_removal; | |
881 | } | |
882 | printk(KERN_INFO "Offlined Pages %ld\n", offlined_pages); | |
883 | /* Ok, all of our target is islaoted. | |
884 | We cannot do rollback at this point. */ | |
885 | offline_isolated_pages(start_pfn, end_pfn); | |
dbc0e4ce KH |
886 | /* reset pagetype flags and makes migrate type to be MOVABLE */ |
887 | undo_isolate_page_range(start_pfn, end_pfn); | |
0c0e6195 | 888 | /* removal success */ |
0c0e6195 KH |
889 | zone->present_pages -= offlined_pages; |
890 | zone->zone_pgdat->node_present_pages -= offlined_pages; | |
891 | totalram_pages -= offlined_pages; | |
7b78d335 | 892 | |
bce7394a MK |
893 | setup_per_zone_wmarks(); |
894 | calculate_zone_inactive_ratio(zone); | |
8fe23e05 DR |
895 | if (!node_present_pages(node)) { |
896 | node_clear_state(node, N_HIGH_MEMORY); | |
897 | kswapd_stop(node); | |
898 | } | |
bce7394a | 899 | |
0c0e6195 KH |
900 | vm_total_pages = nr_free_pagecache_pages(); |
901 | writeback_set_ratelimit(); | |
7b78d335 YG |
902 | |
903 | memory_notify(MEM_OFFLINE, &arg); | |
20d6c96b | 904 | unlock_memory_hotplug(); |
0c0e6195 KH |
905 | return 0; |
906 | ||
907 | failed_removal: | |
908 | printk(KERN_INFO "memory offlining %lx to %lx failed\n", | |
909 | start_pfn, end_pfn); | |
7b78d335 | 910 | memory_notify(MEM_CANCEL_OFFLINE, &arg); |
0c0e6195 KH |
911 | /* pushback to free area */ |
912 | undo_isolate_page_range(start_pfn, end_pfn); | |
7b78d335 | 913 | |
6ad696d2 | 914 | out: |
20d6c96b | 915 | unlock_memory_hotplug(); |
0c0e6195 KH |
916 | return ret; |
917 | } | |
71088785 BP |
918 | |
919 | int remove_memory(u64 start, u64 size) | |
920 | { | |
921 | unsigned long start_pfn, end_pfn; | |
922 | ||
923 | start_pfn = PFN_DOWN(start); | |
924 | end_pfn = start_pfn + PFN_DOWN(size); | |
925 | return offline_pages(start_pfn, end_pfn, 120 * HZ); | |
926 | } | |
48e94196 KH |
927 | #else |
928 | int remove_memory(u64 start, u64 size) | |
929 | { | |
930 | return -EINVAL; | |
931 | } | |
0c0e6195 | 932 | #endif /* CONFIG_MEMORY_HOTREMOVE */ |
71088785 | 933 | EXPORT_SYMBOL_GPL(remove_memory); |