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