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> | |
b95f1b31 | 14 | #include <linux/export.h> |
3947be19 | 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 | ||
9d0ad8ca DK |
37 | /* |
38 | * online_page_callback contains pointer to current page onlining function. | |
39 | * Initially it is generic_online_page(). If it is required it could be | |
40 | * changed by calling set_online_page_callback() for callback registration | |
41 | * and restore_online_page_callback() for generic callback restore. | |
42 | */ | |
43 | ||
44 | static void generic_online_page(struct page *page); | |
45 | ||
46 | static online_page_callback_t online_page_callback = generic_online_page; | |
47 | ||
20d6c96b KM |
48 | DEFINE_MUTEX(mem_hotplug_mutex); |
49 | ||
50 | void lock_memory_hotplug(void) | |
51 | { | |
52 | mutex_lock(&mem_hotplug_mutex); | |
53 | ||
54 | /* for exclusive hibernation if CONFIG_HIBERNATION=y */ | |
55 | lock_system_sleep(); | |
56 | } | |
57 | ||
58 | void unlock_memory_hotplug(void) | |
59 | { | |
60 | unlock_system_sleep(); | |
61 | mutex_unlock(&mem_hotplug_mutex); | |
62 | } | |
63 | ||
64 | ||
45e0b78b KM |
65 | /* add this memory to iomem resource */ |
66 | static struct resource *register_memory_resource(u64 start, u64 size) | |
67 | { | |
68 | struct resource *res; | |
69 | res = kzalloc(sizeof(struct resource), GFP_KERNEL); | |
70 | BUG_ON(!res); | |
71 | ||
72 | res->name = "System RAM"; | |
73 | res->start = start; | |
74 | res->end = start + size - 1; | |
887c3cb1 | 75 | res->flags = IORESOURCE_MEM | IORESOURCE_BUSY; |
45e0b78b | 76 | if (request_resource(&iomem_resource, res) < 0) { |
a62e2f4f | 77 | printk("System RAM resource %pR cannot be added\n", res); |
45e0b78b KM |
78 | kfree(res); |
79 | res = NULL; | |
80 | } | |
81 | return res; | |
82 | } | |
83 | ||
84 | static void release_memory_resource(struct resource *res) | |
85 | { | |
86 | if (!res) | |
87 | return; | |
88 | release_resource(res); | |
89 | kfree(res); | |
90 | return; | |
91 | } | |
92 | ||
53947027 | 93 | #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE |
04753278 | 94 | #ifndef CONFIG_SPARSEMEM_VMEMMAP |
5f24ce5f AA |
95 | static void get_page_bootmem(unsigned long info, struct page *page, |
96 | unsigned long type) | |
04753278 | 97 | { |
5f24ce5f | 98 | page->lru.next = (struct list_head *) type; |
04753278 YG |
99 | SetPagePrivate(page); |
100 | set_page_private(page, info); | |
101 | atomic_inc(&page->_count); | |
102 | } | |
103 | ||
23ce932a RM |
104 | /* reference to __meminit __free_pages_bootmem is valid |
105 | * so use __ref to tell modpost not to generate a warning */ | |
106 | void __ref put_page_bootmem(struct page *page) | |
04753278 | 107 | { |
5f24ce5f | 108 | unsigned long type; |
04753278 | 109 | |
5f24ce5f AA |
110 | type = (unsigned long) page->lru.next; |
111 | BUG_ON(type < MEMORY_HOTPLUG_MIN_BOOTMEM_TYPE || | |
112 | type > MEMORY_HOTPLUG_MAX_BOOTMEM_TYPE); | |
04753278 YG |
113 | |
114 | if (atomic_dec_return(&page->_count) == 1) { | |
115 | ClearPagePrivate(page); | |
116 | set_page_private(page, 0); | |
5f24ce5f | 117 | INIT_LIST_HEAD(&page->lru); |
04753278 YG |
118 | __free_pages_bootmem(page, 0); |
119 | } | |
120 | ||
121 | } | |
122 | ||
d92bc318 | 123 | static void register_page_bootmem_info_section(unsigned long start_pfn) |
04753278 YG |
124 | { |
125 | unsigned long *usemap, mapsize, section_nr, i; | |
126 | struct mem_section *ms; | |
127 | struct page *page, *memmap; | |
128 | ||
04753278 YG |
129 | section_nr = pfn_to_section_nr(start_pfn); |
130 | ms = __nr_to_section(section_nr); | |
131 | ||
132 | /* Get section's memmap address */ | |
133 | memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr); | |
134 | ||
135 | /* | |
136 | * Get page for the memmap's phys address | |
137 | * XXX: need more consideration for sparse_vmemmap... | |
138 | */ | |
139 | page = virt_to_page(memmap); | |
140 | mapsize = sizeof(struct page) * PAGES_PER_SECTION; | |
141 | mapsize = PAGE_ALIGN(mapsize) >> PAGE_SHIFT; | |
142 | ||
143 | /* remember memmap's page */ | |
144 | for (i = 0; i < mapsize; i++, page++) | |
145 | get_page_bootmem(section_nr, page, SECTION_INFO); | |
146 | ||
147 | usemap = __nr_to_section(section_nr)->pageblock_flags; | |
148 | page = virt_to_page(usemap); | |
149 | ||
150 | mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT; | |
151 | ||
152 | for (i = 0; i < mapsize; i++, page++) | |
af370fb8 | 153 | get_page_bootmem(section_nr, page, MIX_SECTION_INFO); |
04753278 YG |
154 | |
155 | } | |
156 | ||
157 | void register_page_bootmem_info_node(struct pglist_data *pgdat) | |
158 | { | |
159 | unsigned long i, pfn, end_pfn, nr_pages; | |
160 | int node = pgdat->node_id; | |
161 | struct page *page; | |
162 | struct zone *zone; | |
163 | ||
164 | nr_pages = PAGE_ALIGN(sizeof(struct pglist_data)) >> PAGE_SHIFT; | |
165 | page = virt_to_page(pgdat); | |
166 | ||
167 | for (i = 0; i < nr_pages; i++, page++) | |
168 | get_page_bootmem(node, page, NODE_INFO); | |
169 | ||
170 | zone = &pgdat->node_zones[0]; | |
171 | for (; zone < pgdat->node_zones + MAX_NR_ZONES - 1; zone++) { | |
172 | if (zone->wait_table) { | |
173 | nr_pages = zone->wait_table_hash_nr_entries | |
174 | * sizeof(wait_queue_head_t); | |
175 | nr_pages = PAGE_ALIGN(nr_pages) >> PAGE_SHIFT; | |
176 | page = virt_to_page(zone->wait_table); | |
177 | ||
178 | for (i = 0; i < nr_pages; i++, page++) | |
179 | get_page_bootmem(node, page, NODE_INFO); | |
180 | } | |
181 | } | |
182 | ||
183 | pfn = pgdat->node_start_pfn; | |
184 | end_pfn = pfn + pgdat->node_spanned_pages; | |
185 | ||
186 | /* register_section info */ | |
f14851af | 187 | for (; pfn < end_pfn; pfn += PAGES_PER_SECTION) { |
188 | /* | |
189 | * Some platforms can assign the same pfn to multiple nodes - on | |
190 | * node0 as well as nodeN. To avoid registering a pfn against | |
191 | * multiple nodes we check that this pfn does not already | |
192 | * reside in some other node. | |
193 | */ | |
194 | if (pfn_valid(pfn) && (pfn_to_nid(pfn) == node)) | |
195 | register_page_bootmem_info_section(pfn); | |
196 | } | |
04753278 YG |
197 | } |
198 | #endif /* !CONFIG_SPARSEMEM_VMEMMAP */ | |
199 | ||
76cdd58e HC |
200 | static void grow_zone_span(struct zone *zone, unsigned long start_pfn, |
201 | unsigned long end_pfn) | |
202 | { | |
203 | unsigned long old_zone_end_pfn; | |
204 | ||
205 | zone_span_writelock(zone); | |
206 | ||
207 | old_zone_end_pfn = zone->zone_start_pfn + zone->spanned_pages; | |
208 | if (start_pfn < zone->zone_start_pfn) | |
209 | zone->zone_start_pfn = start_pfn; | |
210 | ||
211 | zone->spanned_pages = max(old_zone_end_pfn, end_pfn) - | |
212 | zone->zone_start_pfn; | |
213 | ||
214 | zone_span_writeunlock(zone); | |
215 | } | |
216 | ||
217 | static void grow_pgdat_span(struct pglist_data *pgdat, unsigned long start_pfn, | |
218 | unsigned long end_pfn) | |
219 | { | |
220 | unsigned long old_pgdat_end_pfn = | |
221 | pgdat->node_start_pfn + pgdat->node_spanned_pages; | |
222 | ||
223 | if (start_pfn < pgdat->node_start_pfn) | |
224 | pgdat->node_start_pfn = start_pfn; | |
225 | ||
226 | pgdat->node_spanned_pages = max(old_pgdat_end_pfn, end_pfn) - | |
227 | pgdat->node_start_pfn; | |
228 | } | |
229 | ||
31168481 | 230 | static int __meminit __add_zone(struct zone *zone, unsigned long phys_start_pfn) |
3947be19 DH |
231 | { |
232 | struct pglist_data *pgdat = zone->zone_pgdat; | |
233 | int nr_pages = PAGES_PER_SECTION; | |
234 | int nid = pgdat->node_id; | |
235 | int zone_type; | |
76cdd58e | 236 | unsigned long flags; |
3947be19 DH |
237 | |
238 | zone_type = zone - pgdat->node_zones; | |
76cdd58e HC |
239 | if (!zone->wait_table) { |
240 | int ret; | |
241 | ||
242 | ret = init_currently_empty_zone(zone, phys_start_pfn, | |
243 | nr_pages, MEMMAP_HOTPLUG); | |
244 | if (ret) | |
245 | return ret; | |
246 | } | |
247 | pgdat_resize_lock(zone->zone_pgdat, &flags); | |
248 | grow_zone_span(zone, phys_start_pfn, phys_start_pfn + nr_pages); | |
249 | grow_pgdat_span(zone->zone_pgdat, phys_start_pfn, | |
250 | phys_start_pfn + nr_pages); | |
251 | pgdat_resize_unlock(zone->zone_pgdat, &flags); | |
a2f3aa02 DH |
252 | memmap_init_zone(nr_pages, nid, zone_type, |
253 | phys_start_pfn, MEMMAP_HOTPLUG); | |
718127cc | 254 | return 0; |
3947be19 DH |
255 | } |
256 | ||
c04fc586 GH |
257 | static int __meminit __add_section(int nid, struct zone *zone, |
258 | unsigned long phys_start_pfn) | |
3947be19 | 259 | { |
3947be19 | 260 | int nr_pages = PAGES_PER_SECTION; |
3947be19 DH |
261 | int ret; |
262 | ||
ebd15302 KH |
263 | if (pfn_valid(phys_start_pfn)) |
264 | return -EEXIST; | |
265 | ||
0b0acbec | 266 | ret = sparse_add_one_section(zone, phys_start_pfn, nr_pages); |
3947be19 DH |
267 | |
268 | if (ret < 0) | |
269 | return ret; | |
270 | ||
718127cc YG |
271 | ret = __add_zone(zone, phys_start_pfn); |
272 | ||
273 | if (ret < 0) | |
274 | return ret; | |
275 | ||
c04fc586 | 276 | return register_new_memory(nid, __pfn_to_section(phys_start_pfn)); |
3947be19 DH |
277 | } |
278 | ||
0c0a4a51 YG |
279 | #ifdef CONFIG_SPARSEMEM_VMEMMAP |
280 | static int __remove_section(struct zone *zone, struct mem_section *ms) | |
281 | { | |
282 | /* | |
283 | * XXX: Freeing memmap with vmemmap is not implement yet. | |
284 | * This should be removed later. | |
285 | */ | |
286 | return -EBUSY; | |
287 | } | |
288 | #else | |
ea01ea93 BP |
289 | static int __remove_section(struct zone *zone, struct mem_section *ms) |
290 | { | |
291 | unsigned long flags; | |
292 | struct pglist_data *pgdat = zone->zone_pgdat; | |
293 | int ret = -EINVAL; | |
294 | ||
295 | if (!valid_section(ms)) | |
296 | return ret; | |
297 | ||
298 | ret = unregister_memory_section(ms); | |
299 | if (ret) | |
300 | return ret; | |
301 | ||
302 | pgdat_resize_lock(pgdat, &flags); | |
303 | sparse_remove_one_section(zone, ms); | |
304 | pgdat_resize_unlock(pgdat, &flags); | |
305 | return 0; | |
306 | } | |
0c0a4a51 | 307 | #endif |
ea01ea93 | 308 | |
3947be19 DH |
309 | /* |
310 | * Reasonably generic function for adding memory. It is | |
311 | * expected that archs that support memory hotplug will | |
312 | * call this function after deciding the zone to which to | |
313 | * add the new pages. | |
314 | */ | |
c04fc586 GH |
315 | int __ref __add_pages(int nid, struct zone *zone, unsigned long phys_start_pfn, |
316 | unsigned long nr_pages) | |
3947be19 DH |
317 | { |
318 | unsigned long i; | |
319 | int err = 0; | |
6f712711 KH |
320 | int start_sec, end_sec; |
321 | /* during initialize mem_map, align hot-added range to section */ | |
322 | start_sec = pfn_to_section_nr(phys_start_pfn); | |
323 | end_sec = pfn_to_section_nr(phys_start_pfn + nr_pages - 1); | |
3947be19 | 324 | |
6f712711 | 325 | for (i = start_sec; i <= end_sec; i++) { |
c04fc586 | 326 | err = __add_section(nid, zone, i << PFN_SECTION_SHIFT); |
3947be19 | 327 | |
6f712711 | 328 | /* |
183ff22b | 329 | * EEXIST is finally dealt with by ioresource collision |
6f712711 KH |
330 | * check. see add_memory() => register_memory_resource() |
331 | * Warning will be printed if there is collision. | |
bed120c6 JS |
332 | */ |
333 | if (err && (err != -EEXIST)) | |
3947be19 | 334 | break; |
6f712711 | 335 | err = 0; |
3947be19 DH |
336 | } |
337 | ||
338 | return err; | |
339 | } | |
bed120c6 | 340 | EXPORT_SYMBOL_GPL(__add_pages); |
3947be19 | 341 | |
ea01ea93 BP |
342 | /** |
343 | * __remove_pages() - remove sections of pages from a zone | |
344 | * @zone: zone from which pages need to be removed | |
345 | * @phys_start_pfn: starting pageframe (must be aligned to start of a section) | |
346 | * @nr_pages: number of pages to remove (must be multiple of section size) | |
347 | * | |
348 | * Generic helper function to remove section mappings and sysfs entries | |
349 | * for the section of the memory we are removing. Caller needs to make | |
350 | * sure that pages are marked reserved and zones are adjust properly by | |
351 | * calling offline_pages(). | |
352 | */ | |
353 | int __remove_pages(struct zone *zone, unsigned long phys_start_pfn, | |
354 | unsigned long nr_pages) | |
355 | { | |
356 | unsigned long i, ret = 0; | |
357 | int sections_to_remove; | |
358 | ||
359 | /* | |
360 | * We can only remove entire sections | |
361 | */ | |
362 | BUG_ON(phys_start_pfn & ~PAGE_SECTION_MASK); | |
363 | BUG_ON(nr_pages % PAGES_PER_SECTION); | |
364 | ||
ea01ea93 BP |
365 | sections_to_remove = nr_pages / PAGES_PER_SECTION; |
366 | for (i = 0; i < sections_to_remove; i++) { | |
367 | unsigned long pfn = phys_start_pfn + i*PAGES_PER_SECTION; | |
de7f0cba NF |
368 | release_mem_region(pfn << PAGE_SHIFT, |
369 | PAGES_PER_SECTION << PAGE_SHIFT); | |
ea01ea93 BP |
370 | ret = __remove_section(zone, __pfn_to_section(pfn)); |
371 | if (ret) | |
372 | break; | |
373 | } | |
374 | return ret; | |
375 | } | |
376 | EXPORT_SYMBOL_GPL(__remove_pages); | |
377 | ||
9d0ad8ca DK |
378 | int set_online_page_callback(online_page_callback_t callback) |
379 | { | |
380 | int rc = -EINVAL; | |
381 | ||
382 | lock_memory_hotplug(); | |
383 | ||
384 | if (online_page_callback == generic_online_page) { | |
385 | online_page_callback = callback; | |
386 | rc = 0; | |
387 | } | |
388 | ||
389 | unlock_memory_hotplug(); | |
390 | ||
391 | return rc; | |
392 | } | |
393 | EXPORT_SYMBOL_GPL(set_online_page_callback); | |
394 | ||
395 | int restore_online_page_callback(online_page_callback_t callback) | |
396 | { | |
397 | int rc = -EINVAL; | |
398 | ||
399 | lock_memory_hotplug(); | |
400 | ||
401 | if (online_page_callback == callback) { | |
402 | online_page_callback = generic_online_page; | |
403 | rc = 0; | |
404 | } | |
405 | ||
406 | unlock_memory_hotplug(); | |
407 | ||
408 | return rc; | |
409 | } | |
410 | EXPORT_SYMBOL_GPL(restore_online_page_callback); | |
411 | ||
412 | void __online_page_set_limits(struct page *page) | |
180c06ef | 413 | { |
4738e1b9 JB |
414 | unsigned long pfn = page_to_pfn(page); |
415 | ||
4738e1b9 JB |
416 | if (pfn >= num_physpages) |
417 | num_physpages = pfn + 1; | |
9d0ad8ca DK |
418 | } |
419 | EXPORT_SYMBOL_GPL(__online_page_set_limits); | |
420 | ||
421 | void __online_page_increment_counters(struct page *page) | |
422 | { | |
423 | totalram_pages++; | |
180c06ef JF |
424 | |
425 | #ifdef CONFIG_HIGHMEM | |
426 | if (PageHighMem(page)) | |
427 | totalhigh_pages++; | |
428 | #endif | |
9d0ad8ca DK |
429 | } |
430 | EXPORT_SYMBOL_GPL(__online_page_increment_counters); | |
180c06ef | 431 | |
9d0ad8ca DK |
432 | void __online_page_free(struct page *page) |
433 | { | |
180c06ef JF |
434 | ClearPageReserved(page); |
435 | init_page_count(page); | |
436 | __free_page(page); | |
437 | } | |
9d0ad8ca DK |
438 | EXPORT_SYMBOL_GPL(__online_page_free); |
439 | ||
440 | static void generic_online_page(struct page *page) | |
441 | { | |
442 | __online_page_set_limits(page); | |
443 | __online_page_increment_counters(page); | |
444 | __online_page_free(page); | |
445 | } | |
180c06ef | 446 | |
75884fb1 KH |
447 | static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages, |
448 | void *arg) | |
3947be19 DH |
449 | { |
450 | unsigned long i; | |
75884fb1 KH |
451 | unsigned long onlined_pages = *(unsigned long *)arg; |
452 | struct page *page; | |
453 | if (PageReserved(pfn_to_page(start_pfn))) | |
454 | for (i = 0; i < nr_pages; i++) { | |
455 | page = pfn_to_page(start_pfn + i); | |
9d0ad8ca | 456 | (*online_page_callback)(page); |
75884fb1 KH |
457 | onlined_pages++; |
458 | } | |
459 | *(unsigned long *)arg = onlined_pages; | |
460 | return 0; | |
461 | } | |
462 | ||
463 | ||
839a4fcc | 464 | int __ref online_pages(unsigned long pfn, unsigned long nr_pages) |
75884fb1 | 465 | { |
3947be19 DH |
466 | unsigned long onlined_pages = 0; |
467 | struct zone *zone; | |
6811378e | 468 | int need_zonelists_rebuild = 0; |
7b78d335 YG |
469 | int nid; |
470 | int ret; | |
471 | struct memory_notify arg; | |
472 | ||
925268a0 | 473 | lock_memory_hotplug(); |
7b78d335 YG |
474 | arg.start_pfn = pfn; |
475 | arg.nr_pages = nr_pages; | |
476 | arg.status_change_nid = -1; | |
477 | ||
478 | nid = page_to_nid(pfn_to_page(pfn)); | |
479 | if (node_present_pages(nid) == 0) | |
480 | arg.status_change_nid = nid; | |
3947be19 | 481 | |
7b78d335 YG |
482 | ret = memory_notify(MEM_GOING_ONLINE, &arg); |
483 | ret = notifier_to_errno(ret); | |
484 | if (ret) { | |
485 | memory_notify(MEM_CANCEL_ONLINE, &arg); | |
925268a0 | 486 | unlock_memory_hotplug(); |
7b78d335 YG |
487 | return ret; |
488 | } | |
3947be19 DH |
489 | /* |
490 | * This doesn't need a lock to do pfn_to_page(). | |
491 | * The section can't be removed here because of the | |
da19cbcf | 492 | * memory_block->state_mutex. |
3947be19 DH |
493 | */ |
494 | zone = page_zone(pfn_to_page(pfn)); | |
6811378e YG |
495 | /* |
496 | * If this zone is not populated, then it is not in zonelist. | |
497 | * This means the page allocator ignores this zone. | |
498 | * So, zonelist must be updated after online. | |
499 | */ | |
4eaf3f64 | 500 | mutex_lock(&zonelists_mutex); |
6811378e YG |
501 | if (!populated_zone(zone)) |
502 | need_zonelists_rebuild = 1; | |
503 | ||
908eedc6 | 504 | ret = walk_system_ram_range(pfn, nr_pages, &onlined_pages, |
75884fb1 | 505 | online_pages_range); |
fd8a4221 | 506 | if (ret) { |
4eaf3f64 | 507 | mutex_unlock(&zonelists_mutex); |
a62e2f4f BH |
508 | printk(KERN_DEBUG "online_pages [mem %#010llx-%#010llx] failed\n", |
509 | (unsigned long long) pfn << PAGE_SHIFT, | |
510 | (((unsigned long long) pfn + nr_pages) | |
511 | << PAGE_SHIFT) - 1); | |
fd8a4221 | 512 | memory_notify(MEM_CANCEL_ONLINE, &arg); |
925268a0 | 513 | unlock_memory_hotplug(); |
fd8a4221 GL |
514 | return ret; |
515 | } | |
516 | ||
3947be19 | 517 | zone->present_pages += onlined_pages; |
f2937be5 | 518 | zone->zone_pgdat->node_present_pages += onlined_pages; |
08dff7b7 JL |
519 | if (onlined_pages) { |
520 | node_set_state(zone_to_nid(zone), N_HIGH_MEMORY); | |
521 | if (need_zonelists_rebuild) | |
522 | build_all_zonelists(NULL, zone); | |
523 | else | |
524 | zone_pcp_update(zone); | |
525 | } | |
3947be19 | 526 | |
4eaf3f64 | 527 | mutex_unlock(&zonelists_mutex); |
1b79acc9 KM |
528 | |
529 | init_per_zone_wmark_min(); | |
530 | ||
08dff7b7 | 531 | if (onlined_pages) |
7ea1530a | 532 | kswapd_run(zone_to_nid(zone)); |
61b13993 | 533 | |
1f522509 | 534 | vm_total_pages = nr_free_pagecache_pages(); |
2f7f24ec | 535 | |
2d1d43f6 | 536 | writeback_set_ratelimit(); |
7b78d335 YG |
537 | |
538 | if (onlined_pages) | |
539 | memory_notify(MEM_ONLINE, &arg); | |
925268a0 | 540 | unlock_memory_hotplug(); |
7b78d335 | 541 | |
3947be19 DH |
542 | return 0; |
543 | } | |
53947027 | 544 | #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */ |
bc02af93 | 545 | |
e1319331 HS |
546 | /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */ |
547 | static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start) | |
9af3c2de YG |
548 | { |
549 | struct pglist_data *pgdat; | |
550 | unsigned long zones_size[MAX_NR_ZONES] = {0}; | |
551 | unsigned long zholes_size[MAX_NR_ZONES] = {0}; | |
552 | unsigned long start_pfn = start >> PAGE_SHIFT; | |
553 | ||
554 | pgdat = arch_alloc_nodedata(nid); | |
555 | if (!pgdat) | |
556 | return NULL; | |
557 | ||
558 | arch_refresh_nodedata(nid, pgdat); | |
559 | ||
560 | /* we can use NODE_DATA(nid) from here */ | |
561 | ||
562 | /* init node's zones as empty zones, we don't have any present pages.*/ | |
9109fb7b | 563 | free_area_init_node(nid, zones_size, start_pfn, zholes_size); |
9af3c2de | 564 | |
959ecc48 KH |
565 | /* |
566 | * The node we allocated has no zone fallback lists. For avoiding | |
567 | * to access not-initialized zonelist, build here. | |
568 | */ | |
f957db4f | 569 | mutex_lock(&zonelists_mutex); |
9adb62a5 | 570 | build_all_zonelists(pgdat, NULL); |
f957db4f | 571 | mutex_unlock(&zonelists_mutex); |
959ecc48 | 572 | |
9af3c2de YG |
573 | return pgdat; |
574 | } | |
575 | ||
576 | static void rollback_node_hotadd(int nid, pg_data_t *pgdat) | |
577 | { | |
578 | arch_refresh_nodedata(nid, NULL); | |
579 | arch_free_nodedata(pgdat); | |
580 | return; | |
581 | } | |
582 | ||
0a547039 | 583 | |
cf23422b | 584 | /* |
585 | * called by cpu_up() to online a node without onlined memory. | |
586 | */ | |
587 | int mem_online_node(int nid) | |
588 | { | |
589 | pg_data_t *pgdat; | |
590 | int ret; | |
591 | ||
20d6c96b | 592 | lock_memory_hotplug(); |
cf23422b | 593 | pgdat = hotadd_new_pgdat(nid, 0); |
7553e8f2 | 594 | if (!pgdat) { |
cf23422b | 595 | ret = -ENOMEM; |
596 | goto out; | |
597 | } | |
598 | node_set_online(nid); | |
599 | ret = register_one_node(nid); | |
600 | BUG_ON(ret); | |
601 | ||
602 | out: | |
20d6c96b | 603 | unlock_memory_hotplug(); |
cf23422b | 604 | return ret; |
605 | } | |
606 | ||
31168481 AV |
607 | /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */ |
608 | int __ref add_memory(int nid, u64 start, u64 size) | |
bc02af93 | 609 | { |
9af3c2de YG |
610 | pg_data_t *pgdat = NULL; |
611 | int new_pgdat = 0; | |
ebd15302 | 612 | struct resource *res; |
bc02af93 YG |
613 | int ret; |
614 | ||
20d6c96b | 615 | lock_memory_hotplug(); |
6ad696d2 | 616 | |
ebd15302 | 617 | res = register_memory_resource(start, size); |
6ad696d2 | 618 | ret = -EEXIST; |
ebd15302 | 619 | if (!res) |
6ad696d2 | 620 | goto out; |
ebd15302 | 621 | |
9af3c2de YG |
622 | if (!node_online(nid)) { |
623 | pgdat = hotadd_new_pgdat(nid, start); | |
6ad696d2 | 624 | ret = -ENOMEM; |
9af3c2de | 625 | if (!pgdat) |
41b9e2d7 | 626 | goto error; |
9af3c2de | 627 | new_pgdat = 1; |
9af3c2de YG |
628 | } |
629 | ||
bc02af93 YG |
630 | /* call arch's memory hotadd */ |
631 | ret = arch_add_memory(nid, start, size); | |
632 | ||
9af3c2de YG |
633 | if (ret < 0) |
634 | goto error; | |
635 | ||
0fc44159 | 636 | /* we online node here. we can't roll back from here. */ |
9af3c2de YG |
637 | node_set_online(nid); |
638 | ||
0fc44159 YG |
639 | if (new_pgdat) { |
640 | ret = register_one_node(nid); | |
641 | /* | |
642 | * If sysfs file of new node can't create, cpu on the node | |
643 | * can't be hot-added. There is no rollback way now. | |
644 | * So, check by BUG_ON() to catch it reluctantly.. | |
645 | */ | |
646 | BUG_ON(ret); | |
647 | } | |
648 | ||
d96ae530 | 649 | /* create new memmap entry */ |
650 | firmware_map_add_hotplug(start, start + size, "System RAM"); | |
651 | ||
6ad696d2 AK |
652 | goto out; |
653 | ||
9af3c2de YG |
654 | error: |
655 | /* rollback pgdat allocation and others */ | |
656 | if (new_pgdat) | |
657 | rollback_node_hotadd(nid, pgdat); | |
ebd15302 KH |
658 | if (res) |
659 | release_memory_resource(res); | |
9af3c2de | 660 | |
6ad696d2 | 661 | out: |
20d6c96b | 662 | unlock_memory_hotplug(); |
bc02af93 YG |
663 | return ret; |
664 | } | |
665 | EXPORT_SYMBOL_GPL(add_memory); | |
0c0e6195 KH |
666 | |
667 | #ifdef CONFIG_MEMORY_HOTREMOVE | |
5c755e9f BP |
668 | /* |
669 | * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy | |
670 | * set and the size of the free page is given by page_order(). Using this, | |
671 | * the function determines if the pageblock contains only free pages. | |
672 | * Due to buddy contraints, a free page at least the size of a pageblock will | |
673 | * be located at the start of the pageblock | |
674 | */ | |
675 | static inline int pageblock_free(struct page *page) | |
676 | { | |
677 | return PageBuddy(page) && page_order(page) >= pageblock_order; | |
678 | } | |
679 | ||
680 | /* Return the start of the next active pageblock after a given page */ | |
681 | static struct page *next_active_pageblock(struct page *page) | |
682 | { | |
5c755e9f BP |
683 | /* Ensure the starting page is pageblock-aligned */ |
684 | BUG_ON(page_to_pfn(page) & (pageblock_nr_pages - 1)); | |
685 | ||
5c755e9f | 686 | /* If the entire pageblock is free, move to the end of free page */ |
0dcc48c1 KH |
687 | if (pageblock_free(page)) { |
688 | int order; | |
689 | /* be careful. we don't have locks, page_order can be changed.*/ | |
690 | order = page_order(page); | |
691 | if ((order < MAX_ORDER) && (order >= pageblock_order)) | |
692 | return page + (1 << order); | |
693 | } | |
5c755e9f | 694 | |
0dcc48c1 | 695 | return page + pageblock_nr_pages; |
5c755e9f BP |
696 | } |
697 | ||
698 | /* Checks if this range of memory is likely to be hot-removable. */ | |
699 | int is_mem_section_removable(unsigned long start_pfn, unsigned long nr_pages) | |
700 | { | |
5c755e9f BP |
701 | struct page *page = pfn_to_page(start_pfn); |
702 | struct page *end_page = page + nr_pages; | |
703 | ||
704 | /* Check the starting page of each pageblock within the range */ | |
705 | for (; page < end_page; page = next_active_pageblock(page)) { | |
49ac8255 | 706 | if (!is_pageblock_removable_nolock(page)) |
5c755e9f | 707 | return 0; |
49ac8255 | 708 | cond_resched(); |
5c755e9f BP |
709 | } |
710 | ||
711 | /* All pageblocks in the memory block are likely to be hot-removable */ | |
712 | return 1; | |
713 | } | |
714 | ||
0c0e6195 KH |
715 | /* |
716 | * Confirm all pages in a range [start, end) is belongs to the same zone. | |
717 | */ | |
718 | static int test_pages_in_a_zone(unsigned long start_pfn, unsigned long end_pfn) | |
719 | { | |
720 | unsigned long pfn; | |
721 | struct zone *zone = NULL; | |
722 | struct page *page; | |
723 | int i; | |
724 | for (pfn = start_pfn; | |
725 | pfn < end_pfn; | |
726 | pfn += MAX_ORDER_NR_PAGES) { | |
727 | i = 0; | |
728 | /* This is just a CONFIG_HOLES_IN_ZONE check.*/ | |
729 | while ((i < MAX_ORDER_NR_PAGES) && !pfn_valid_within(pfn + i)) | |
730 | i++; | |
731 | if (i == MAX_ORDER_NR_PAGES) | |
732 | continue; | |
733 | page = pfn_to_page(pfn + i); | |
734 | if (zone && page_zone(page) != zone) | |
735 | return 0; | |
736 | zone = page_zone(page); | |
737 | } | |
738 | return 1; | |
739 | } | |
740 | ||
741 | /* | |
742 | * Scanning pfn is much easier than scanning lru list. | |
743 | * Scan pfn from start to end and Find LRU page. | |
744 | */ | |
7bbc0905 | 745 | static unsigned long scan_lru_pages(unsigned long start, unsigned long end) |
0c0e6195 KH |
746 | { |
747 | unsigned long pfn; | |
748 | struct page *page; | |
749 | for (pfn = start; pfn < end; pfn++) { | |
750 | if (pfn_valid(pfn)) { | |
751 | page = pfn_to_page(pfn); | |
752 | if (PageLRU(page)) | |
753 | return pfn; | |
754 | } | |
755 | } | |
756 | return 0; | |
757 | } | |
758 | ||
759 | static struct page * | |
3c1d4378 | 760 | hotremove_migrate_alloc(struct page *page, unsigned long private, int **x) |
0c0e6195 | 761 | { |
3c1d4378 HD |
762 | /* This should be improooooved!! */ |
763 | return alloc_page(GFP_HIGHUSER_MOVABLE); | |
0c0e6195 KH |
764 | } |
765 | ||
0c0e6195 KH |
766 | #define NR_OFFLINE_AT_ONCE_PAGES (256) |
767 | static int | |
768 | do_migrate_range(unsigned long start_pfn, unsigned long end_pfn) | |
769 | { | |
770 | unsigned long pfn; | |
771 | struct page *page; | |
772 | int move_pages = NR_OFFLINE_AT_ONCE_PAGES; | |
773 | int not_managed = 0; | |
774 | int ret = 0; | |
775 | LIST_HEAD(source); | |
776 | ||
777 | for (pfn = start_pfn; pfn < end_pfn && move_pages > 0; pfn++) { | |
778 | if (!pfn_valid(pfn)) | |
779 | continue; | |
780 | page = pfn_to_page(pfn); | |
700c2a46 | 781 | if (!get_page_unless_zero(page)) |
0c0e6195 KH |
782 | continue; |
783 | /* | |
784 | * We can skip free pages. And we can only deal with pages on | |
785 | * LRU. | |
786 | */ | |
62695a84 | 787 | ret = isolate_lru_page(page); |
0c0e6195 | 788 | if (!ret) { /* Success */ |
700c2a46 | 789 | put_page(page); |
62695a84 | 790 | list_add_tail(&page->lru, &source); |
0c0e6195 | 791 | move_pages--; |
6d9c285a KM |
792 | inc_zone_page_state(page, NR_ISOLATED_ANON + |
793 | page_is_file_cache(page)); | |
794 | ||
0c0e6195 | 795 | } else { |
0c0e6195 | 796 | #ifdef CONFIG_DEBUG_VM |
718a3821 WF |
797 | printk(KERN_ALERT "removing pfn %lx from LRU failed\n", |
798 | pfn); | |
799 | dump_page(page); | |
0c0e6195 | 800 | #endif |
700c2a46 | 801 | put_page(page); |
25985edc | 802 | /* Because we don't have big zone->lock. we should |
809c4449 BL |
803 | check this again here. */ |
804 | if (page_count(page)) { | |
805 | not_managed++; | |
f3ab2636 | 806 | ret = -EBUSY; |
809c4449 BL |
807 | break; |
808 | } | |
0c0e6195 KH |
809 | } |
810 | } | |
f3ab2636 BL |
811 | if (!list_empty(&source)) { |
812 | if (not_managed) { | |
813 | putback_lru_pages(&source); | |
814 | goto out; | |
815 | } | |
816 | /* this function returns # of failed pages */ | |
77f1fe6b | 817 | ret = migrate_pages(&source, hotremove_migrate_alloc, 0, |
a6bc32b8 | 818 | true, MIGRATE_SYNC); |
f3ab2636 | 819 | if (ret) |
0c0e6195 | 820 | putback_lru_pages(&source); |
0c0e6195 | 821 | } |
0c0e6195 KH |
822 | out: |
823 | return ret; | |
824 | } | |
825 | ||
826 | /* | |
827 | * remove from free_area[] and mark all as Reserved. | |
828 | */ | |
829 | static int | |
830 | offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages, | |
831 | void *data) | |
832 | { | |
833 | __offline_isolated_pages(start, start + nr_pages); | |
834 | return 0; | |
835 | } | |
836 | ||
837 | static void | |
838 | offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn) | |
839 | { | |
908eedc6 | 840 | walk_system_ram_range(start_pfn, end_pfn - start_pfn, NULL, |
0c0e6195 KH |
841 | offline_isolated_pages_cb); |
842 | } | |
843 | ||
844 | /* | |
845 | * Check all pages in range, recoreded as memory resource, are isolated. | |
846 | */ | |
847 | static int | |
848 | check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages, | |
849 | void *data) | |
850 | { | |
851 | int ret; | |
852 | long offlined = *(long *)data; | |
853 | ret = test_pages_isolated(start_pfn, start_pfn + nr_pages); | |
854 | offlined = nr_pages; | |
855 | if (!ret) | |
856 | *(long *)data += offlined; | |
857 | return ret; | |
858 | } | |
859 | ||
860 | static long | |
861 | check_pages_isolated(unsigned long start_pfn, unsigned long end_pfn) | |
862 | { | |
863 | long offlined = 0; | |
864 | int ret; | |
865 | ||
908eedc6 | 866 | ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn, &offlined, |
0c0e6195 KH |
867 | check_pages_isolated_cb); |
868 | if (ret < 0) | |
869 | offlined = (long)ret; | |
870 | return offlined; | |
871 | } | |
872 | ||
839a4fcc | 873 | static int __ref offline_pages(unsigned long start_pfn, |
0c0e6195 KH |
874 | unsigned long end_pfn, unsigned long timeout) |
875 | { | |
876 | unsigned long pfn, nr_pages, expire; | |
877 | long offlined_pages; | |
7b78d335 | 878 | int ret, drain, retry_max, node; |
0c0e6195 | 879 | struct zone *zone; |
7b78d335 | 880 | struct memory_notify arg; |
0c0e6195 KH |
881 | |
882 | BUG_ON(start_pfn >= end_pfn); | |
883 | /* at least, alignment against pageblock is necessary */ | |
884 | if (!IS_ALIGNED(start_pfn, pageblock_nr_pages)) | |
885 | return -EINVAL; | |
886 | if (!IS_ALIGNED(end_pfn, pageblock_nr_pages)) | |
887 | return -EINVAL; | |
888 | /* This makes hotplug much easier...and readable. | |
889 | we assume this for now. .*/ | |
890 | if (!test_pages_in_a_zone(start_pfn, end_pfn)) | |
891 | return -EINVAL; | |
7b78d335 | 892 | |
20d6c96b | 893 | lock_memory_hotplug(); |
6ad696d2 | 894 | |
7b78d335 YG |
895 | zone = page_zone(pfn_to_page(start_pfn)); |
896 | node = zone_to_nid(zone); | |
897 | nr_pages = end_pfn - start_pfn; | |
898 | ||
0c0e6195 | 899 | /* set above range as isolated */ |
0815f3d8 | 900 | ret = start_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE); |
0c0e6195 | 901 | if (ret) |
6ad696d2 | 902 | goto out; |
7b78d335 YG |
903 | |
904 | arg.start_pfn = start_pfn; | |
905 | arg.nr_pages = nr_pages; | |
906 | arg.status_change_nid = -1; | |
907 | if (nr_pages >= node_present_pages(node)) | |
908 | arg.status_change_nid = node; | |
909 | ||
910 | ret = memory_notify(MEM_GOING_OFFLINE, &arg); | |
911 | ret = notifier_to_errno(ret); | |
912 | if (ret) | |
913 | goto failed_removal; | |
914 | ||
0c0e6195 KH |
915 | pfn = start_pfn; |
916 | expire = jiffies + timeout; | |
917 | drain = 0; | |
918 | retry_max = 5; | |
919 | repeat: | |
920 | /* start memory hot removal */ | |
921 | ret = -EAGAIN; | |
922 | if (time_after(jiffies, expire)) | |
923 | goto failed_removal; | |
924 | ret = -EINTR; | |
925 | if (signal_pending(current)) | |
926 | goto failed_removal; | |
927 | ret = 0; | |
928 | if (drain) { | |
929 | lru_add_drain_all(); | |
0c0e6195 | 930 | cond_resched(); |
9f8f2172 | 931 | drain_all_pages(); |
0c0e6195 KH |
932 | } |
933 | ||
934 | pfn = scan_lru_pages(start_pfn, end_pfn); | |
935 | if (pfn) { /* We have page on LRU */ | |
936 | ret = do_migrate_range(pfn, end_pfn); | |
937 | if (!ret) { | |
938 | drain = 1; | |
939 | goto repeat; | |
940 | } else { | |
941 | if (ret < 0) | |
942 | if (--retry_max == 0) | |
943 | goto failed_removal; | |
944 | yield(); | |
945 | drain = 1; | |
946 | goto repeat; | |
947 | } | |
948 | } | |
949 | /* drain all zone's lru pagevec, this is asyncronous... */ | |
950 | lru_add_drain_all(); | |
0c0e6195 KH |
951 | yield(); |
952 | /* drain pcp pages , this is synchrouns. */ | |
9f8f2172 | 953 | drain_all_pages(); |
0c0e6195 KH |
954 | /* check again */ |
955 | offlined_pages = check_pages_isolated(start_pfn, end_pfn); | |
956 | if (offlined_pages < 0) { | |
957 | ret = -EBUSY; | |
958 | goto failed_removal; | |
959 | } | |
960 | printk(KERN_INFO "Offlined Pages %ld\n", offlined_pages); | |
961 | /* Ok, all of our target is islaoted. | |
962 | We cannot do rollback at this point. */ | |
963 | offline_isolated_pages(start_pfn, end_pfn); | |
dbc0e4ce | 964 | /* reset pagetype flags and makes migrate type to be MOVABLE */ |
0815f3d8 | 965 | undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE); |
0c0e6195 | 966 | /* removal success */ |
0c0e6195 KH |
967 | zone->present_pages -= offlined_pages; |
968 | zone->zone_pgdat->node_present_pages -= offlined_pages; | |
969 | totalram_pages -= offlined_pages; | |
7b78d335 | 970 | |
1b79acc9 KM |
971 | init_per_zone_wmark_min(); |
972 | ||
340175b7 JL |
973 | if (!populated_zone(zone)) |
974 | zone_pcp_reset(zone); | |
975 | ||
8fe23e05 DR |
976 | if (!node_present_pages(node)) { |
977 | node_clear_state(node, N_HIGH_MEMORY); | |
978 | kswapd_stop(node); | |
979 | } | |
bce7394a | 980 | |
0c0e6195 KH |
981 | vm_total_pages = nr_free_pagecache_pages(); |
982 | writeback_set_ratelimit(); | |
7b78d335 YG |
983 | |
984 | memory_notify(MEM_OFFLINE, &arg); | |
20d6c96b | 985 | unlock_memory_hotplug(); |
0c0e6195 KH |
986 | return 0; |
987 | ||
988 | failed_removal: | |
a62e2f4f BH |
989 | printk(KERN_INFO "memory offlining [mem %#010llx-%#010llx] failed\n", |
990 | (unsigned long long) start_pfn << PAGE_SHIFT, | |
991 | ((unsigned long long) end_pfn << PAGE_SHIFT) - 1); | |
7b78d335 | 992 | memory_notify(MEM_CANCEL_OFFLINE, &arg); |
0c0e6195 | 993 | /* pushback to free area */ |
0815f3d8 | 994 | undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE); |
7b78d335 | 995 | |
6ad696d2 | 996 | out: |
20d6c96b | 997 | unlock_memory_hotplug(); |
0c0e6195 KH |
998 | return ret; |
999 | } | |
71088785 BP |
1000 | |
1001 | int remove_memory(u64 start, u64 size) | |
1002 | { | |
1003 | unsigned long start_pfn, end_pfn; | |
1004 | ||
1005 | start_pfn = PFN_DOWN(start); | |
1006 | end_pfn = start_pfn + PFN_DOWN(size); | |
1007 | return offline_pages(start_pfn, end_pfn, 120 * HZ); | |
1008 | } | |
48e94196 KH |
1009 | #else |
1010 | int remove_memory(u64 start, u64 size) | |
1011 | { | |
1012 | return -EINVAL; | |
1013 | } | |
0c0e6195 | 1014 | #endif /* CONFIG_MEMORY_HOTREMOVE */ |
71088785 | 1015 | EXPORT_SYMBOL_GPL(remove_memory); |