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; | |
712cd386 | 208 | if (!zone->spanned_pages || start_pfn < zone->zone_start_pfn) |
76cdd58e HC |
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 | ||
511c2aba LJ |
217 | static void resize_zone(struct zone *zone, unsigned long start_pfn, |
218 | unsigned long end_pfn) | |
219 | { | |
220 | zone_span_writelock(zone); | |
221 | ||
e455a9b9 LJ |
222 | if (end_pfn - start_pfn) { |
223 | zone->zone_start_pfn = start_pfn; | |
224 | zone->spanned_pages = end_pfn - start_pfn; | |
225 | } else { | |
226 | /* | |
227 | * make it consist as free_area_init_core(), | |
228 | * if spanned_pages = 0, then keep start_pfn = 0 | |
229 | */ | |
230 | zone->zone_start_pfn = 0; | |
231 | zone->spanned_pages = 0; | |
232 | } | |
511c2aba LJ |
233 | |
234 | zone_span_writeunlock(zone); | |
235 | } | |
236 | ||
237 | static void fix_zone_id(struct zone *zone, unsigned long start_pfn, | |
238 | unsigned long end_pfn) | |
239 | { | |
240 | enum zone_type zid = zone_idx(zone); | |
241 | int nid = zone->zone_pgdat->node_id; | |
242 | unsigned long pfn; | |
243 | ||
244 | for (pfn = start_pfn; pfn < end_pfn; pfn++) | |
245 | set_page_links(pfn_to_page(pfn), zid, nid, pfn); | |
246 | } | |
247 | ||
e455a9b9 | 248 | static int __meminit move_pfn_range_left(struct zone *z1, struct zone *z2, |
511c2aba LJ |
249 | unsigned long start_pfn, unsigned long end_pfn) |
250 | { | |
e455a9b9 | 251 | int ret; |
511c2aba | 252 | unsigned long flags; |
e455a9b9 LJ |
253 | unsigned long z1_start_pfn; |
254 | ||
255 | if (!z1->wait_table) { | |
256 | ret = init_currently_empty_zone(z1, start_pfn, | |
257 | end_pfn - start_pfn, MEMMAP_HOTPLUG); | |
258 | if (ret) | |
259 | return ret; | |
260 | } | |
511c2aba LJ |
261 | |
262 | pgdat_resize_lock(z1->zone_pgdat, &flags); | |
263 | ||
264 | /* can't move pfns which are higher than @z2 */ | |
265 | if (end_pfn > z2->zone_start_pfn + z2->spanned_pages) | |
266 | goto out_fail; | |
267 | /* the move out part mast at the left most of @z2 */ | |
268 | if (start_pfn > z2->zone_start_pfn) | |
269 | goto out_fail; | |
270 | /* must included/overlap */ | |
271 | if (end_pfn <= z2->zone_start_pfn) | |
272 | goto out_fail; | |
273 | ||
e455a9b9 LJ |
274 | /* use start_pfn for z1's start_pfn if z1 is empty */ |
275 | if (z1->spanned_pages) | |
276 | z1_start_pfn = z1->zone_start_pfn; | |
277 | else | |
278 | z1_start_pfn = start_pfn; | |
279 | ||
280 | resize_zone(z1, z1_start_pfn, end_pfn); | |
511c2aba LJ |
281 | resize_zone(z2, end_pfn, z2->zone_start_pfn + z2->spanned_pages); |
282 | ||
283 | pgdat_resize_unlock(z1->zone_pgdat, &flags); | |
284 | ||
285 | fix_zone_id(z1, start_pfn, end_pfn); | |
286 | ||
287 | return 0; | |
288 | out_fail: | |
289 | pgdat_resize_unlock(z1->zone_pgdat, &flags); | |
290 | return -1; | |
291 | } | |
292 | ||
e455a9b9 | 293 | static int __meminit move_pfn_range_right(struct zone *z1, struct zone *z2, |
511c2aba LJ |
294 | unsigned long start_pfn, unsigned long end_pfn) |
295 | { | |
e455a9b9 | 296 | int ret; |
511c2aba | 297 | unsigned long flags; |
e455a9b9 LJ |
298 | unsigned long z2_end_pfn; |
299 | ||
300 | if (!z2->wait_table) { | |
301 | ret = init_currently_empty_zone(z2, start_pfn, | |
302 | end_pfn - start_pfn, MEMMAP_HOTPLUG); | |
303 | if (ret) | |
304 | return ret; | |
305 | } | |
511c2aba LJ |
306 | |
307 | pgdat_resize_lock(z1->zone_pgdat, &flags); | |
308 | ||
309 | /* can't move pfns which are lower than @z1 */ | |
310 | if (z1->zone_start_pfn > start_pfn) | |
311 | goto out_fail; | |
312 | /* the move out part mast at the right most of @z1 */ | |
313 | if (z1->zone_start_pfn + z1->spanned_pages > end_pfn) | |
314 | goto out_fail; | |
315 | /* must included/overlap */ | |
316 | if (start_pfn >= z1->zone_start_pfn + z1->spanned_pages) | |
317 | goto out_fail; | |
318 | ||
e455a9b9 LJ |
319 | /* use end_pfn for z2's end_pfn if z2 is empty */ |
320 | if (z2->spanned_pages) | |
321 | z2_end_pfn = z2->zone_start_pfn + z2->spanned_pages; | |
322 | else | |
323 | z2_end_pfn = end_pfn; | |
324 | ||
511c2aba | 325 | resize_zone(z1, z1->zone_start_pfn, start_pfn); |
e455a9b9 | 326 | resize_zone(z2, start_pfn, z2_end_pfn); |
511c2aba LJ |
327 | |
328 | pgdat_resize_unlock(z1->zone_pgdat, &flags); | |
329 | ||
330 | fix_zone_id(z2, start_pfn, end_pfn); | |
331 | ||
332 | return 0; | |
333 | out_fail: | |
334 | pgdat_resize_unlock(z1->zone_pgdat, &flags); | |
335 | return -1; | |
336 | } | |
337 | ||
76cdd58e HC |
338 | static void grow_pgdat_span(struct pglist_data *pgdat, unsigned long start_pfn, |
339 | unsigned long end_pfn) | |
340 | { | |
341 | unsigned long old_pgdat_end_pfn = | |
342 | pgdat->node_start_pfn + pgdat->node_spanned_pages; | |
343 | ||
712cd386 | 344 | if (!pgdat->node_spanned_pages || start_pfn < pgdat->node_start_pfn) |
76cdd58e HC |
345 | pgdat->node_start_pfn = start_pfn; |
346 | ||
347 | pgdat->node_spanned_pages = max(old_pgdat_end_pfn, end_pfn) - | |
348 | pgdat->node_start_pfn; | |
349 | } | |
350 | ||
31168481 | 351 | static int __meminit __add_zone(struct zone *zone, unsigned long phys_start_pfn) |
3947be19 DH |
352 | { |
353 | struct pglist_data *pgdat = zone->zone_pgdat; | |
354 | int nr_pages = PAGES_PER_SECTION; | |
355 | int nid = pgdat->node_id; | |
356 | int zone_type; | |
76cdd58e | 357 | unsigned long flags; |
3947be19 DH |
358 | |
359 | zone_type = zone - pgdat->node_zones; | |
76cdd58e HC |
360 | if (!zone->wait_table) { |
361 | int ret; | |
362 | ||
363 | ret = init_currently_empty_zone(zone, phys_start_pfn, | |
364 | nr_pages, MEMMAP_HOTPLUG); | |
365 | if (ret) | |
366 | return ret; | |
367 | } | |
368 | pgdat_resize_lock(zone->zone_pgdat, &flags); | |
369 | grow_zone_span(zone, phys_start_pfn, phys_start_pfn + nr_pages); | |
370 | grow_pgdat_span(zone->zone_pgdat, phys_start_pfn, | |
371 | phys_start_pfn + nr_pages); | |
372 | pgdat_resize_unlock(zone->zone_pgdat, &flags); | |
a2f3aa02 DH |
373 | memmap_init_zone(nr_pages, nid, zone_type, |
374 | phys_start_pfn, MEMMAP_HOTPLUG); | |
718127cc | 375 | return 0; |
3947be19 DH |
376 | } |
377 | ||
c04fc586 GH |
378 | static int __meminit __add_section(int nid, struct zone *zone, |
379 | unsigned long phys_start_pfn) | |
3947be19 | 380 | { |
3947be19 | 381 | int nr_pages = PAGES_PER_SECTION; |
3947be19 DH |
382 | int ret; |
383 | ||
ebd15302 KH |
384 | if (pfn_valid(phys_start_pfn)) |
385 | return -EEXIST; | |
386 | ||
0b0acbec | 387 | ret = sparse_add_one_section(zone, phys_start_pfn, nr_pages); |
3947be19 DH |
388 | |
389 | if (ret < 0) | |
390 | return ret; | |
391 | ||
718127cc YG |
392 | ret = __add_zone(zone, phys_start_pfn); |
393 | ||
394 | if (ret < 0) | |
395 | return ret; | |
396 | ||
c04fc586 | 397 | return register_new_memory(nid, __pfn_to_section(phys_start_pfn)); |
3947be19 DH |
398 | } |
399 | ||
0c0a4a51 YG |
400 | #ifdef CONFIG_SPARSEMEM_VMEMMAP |
401 | static int __remove_section(struct zone *zone, struct mem_section *ms) | |
402 | { | |
403 | /* | |
404 | * XXX: Freeing memmap with vmemmap is not implement yet. | |
405 | * This should be removed later. | |
406 | */ | |
407 | return -EBUSY; | |
408 | } | |
409 | #else | |
ea01ea93 BP |
410 | static int __remove_section(struct zone *zone, struct mem_section *ms) |
411 | { | |
412 | unsigned long flags; | |
413 | struct pglist_data *pgdat = zone->zone_pgdat; | |
414 | int ret = -EINVAL; | |
415 | ||
416 | if (!valid_section(ms)) | |
417 | return ret; | |
418 | ||
419 | ret = unregister_memory_section(ms); | |
420 | if (ret) | |
421 | return ret; | |
422 | ||
423 | pgdat_resize_lock(pgdat, &flags); | |
424 | sparse_remove_one_section(zone, ms); | |
425 | pgdat_resize_unlock(pgdat, &flags); | |
426 | return 0; | |
427 | } | |
0c0a4a51 | 428 | #endif |
ea01ea93 | 429 | |
3947be19 DH |
430 | /* |
431 | * Reasonably generic function for adding memory. It is | |
432 | * expected that archs that support memory hotplug will | |
433 | * call this function after deciding the zone to which to | |
434 | * add the new pages. | |
435 | */ | |
c04fc586 GH |
436 | int __ref __add_pages(int nid, struct zone *zone, unsigned long phys_start_pfn, |
437 | unsigned long nr_pages) | |
3947be19 DH |
438 | { |
439 | unsigned long i; | |
440 | int err = 0; | |
6f712711 KH |
441 | int start_sec, end_sec; |
442 | /* during initialize mem_map, align hot-added range to section */ | |
443 | start_sec = pfn_to_section_nr(phys_start_pfn); | |
444 | end_sec = pfn_to_section_nr(phys_start_pfn + nr_pages - 1); | |
3947be19 | 445 | |
6f712711 | 446 | for (i = start_sec; i <= end_sec; i++) { |
c04fc586 | 447 | err = __add_section(nid, zone, i << PFN_SECTION_SHIFT); |
3947be19 | 448 | |
6f712711 | 449 | /* |
183ff22b | 450 | * EEXIST is finally dealt with by ioresource collision |
6f712711 KH |
451 | * check. see add_memory() => register_memory_resource() |
452 | * Warning will be printed if there is collision. | |
bed120c6 JS |
453 | */ |
454 | if (err && (err != -EEXIST)) | |
3947be19 | 455 | break; |
6f712711 | 456 | err = 0; |
3947be19 DH |
457 | } |
458 | ||
459 | return err; | |
460 | } | |
bed120c6 | 461 | EXPORT_SYMBOL_GPL(__add_pages); |
3947be19 | 462 | |
ea01ea93 BP |
463 | /** |
464 | * __remove_pages() - remove sections of pages from a zone | |
465 | * @zone: zone from which pages need to be removed | |
466 | * @phys_start_pfn: starting pageframe (must be aligned to start of a section) | |
467 | * @nr_pages: number of pages to remove (must be multiple of section size) | |
468 | * | |
469 | * Generic helper function to remove section mappings and sysfs entries | |
470 | * for the section of the memory we are removing. Caller needs to make | |
471 | * sure that pages are marked reserved and zones are adjust properly by | |
472 | * calling offline_pages(). | |
473 | */ | |
474 | int __remove_pages(struct zone *zone, unsigned long phys_start_pfn, | |
475 | unsigned long nr_pages) | |
476 | { | |
477 | unsigned long i, ret = 0; | |
478 | int sections_to_remove; | |
479 | ||
480 | /* | |
481 | * We can only remove entire sections | |
482 | */ | |
483 | BUG_ON(phys_start_pfn & ~PAGE_SECTION_MASK); | |
484 | BUG_ON(nr_pages % PAGES_PER_SECTION); | |
485 | ||
d760afd4 YI |
486 | release_mem_region(phys_start_pfn << PAGE_SHIFT, nr_pages * PAGE_SIZE); |
487 | ||
ea01ea93 BP |
488 | sections_to_remove = nr_pages / PAGES_PER_SECTION; |
489 | for (i = 0; i < sections_to_remove; i++) { | |
490 | unsigned long pfn = phys_start_pfn + i*PAGES_PER_SECTION; | |
491 | ret = __remove_section(zone, __pfn_to_section(pfn)); | |
492 | if (ret) | |
493 | break; | |
494 | } | |
495 | return ret; | |
496 | } | |
497 | EXPORT_SYMBOL_GPL(__remove_pages); | |
498 | ||
9d0ad8ca DK |
499 | int set_online_page_callback(online_page_callback_t callback) |
500 | { | |
501 | int rc = -EINVAL; | |
502 | ||
503 | lock_memory_hotplug(); | |
504 | ||
505 | if (online_page_callback == generic_online_page) { | |
506 | online_page_callback = callback; | |
507 | rc = 0; | |
508 | } | |
509 | ||
510 | unlock_memory_hotplug(); | |
511 | ||
512 | return rc; | |
513 | } | |
514 | EXPORT_SYMBOL_GPL(set_online_page_callback); | |
515 | ||
516 | int restore_online_page_callback(online_page_callback_t callback) | |
517 | { | |
518 | int rc = -EINVAL; | |
519 | ||
520 | lock_memory_hotplug(); | |
521 | ||
522 | if (online_page_callback == callback) { | |
523 | online_page_callback = generic_online_page; | |
524 | rc = 0; | |
525 | } | |
526 | ||
527 | unlock_memory_hotplug(); | |
528 | ||
529 | return rc; | |
530 | } | |
531 | EXPORT_SYMBOL_GPL(restore_online_page_callback); | |
532 | ||
533 | void __online_page_set_limits(struct page *page) | |
180c06ef | 534 | { |
4738e1b9 JB |
535 | unsigned long pfn = page_to_pfn(page); |
536 | ||
4738e1b9 JB |
537 | if (pfn >= num_physpages) |
538 | num_physpages = pfn + 1; | |
9d0ad8ca DK |
539 | } |
540 | EXPORT_SYMBOL_GPL(__online_page_set_limits); | |
541 | ||
542 | void __online_page_increment_counters(struct page *page) | |
543 | { | |
544 | totalram_pages++; | |
180c06ef JF |
545 | |
546 | #ifdef CONFIG_HIGHMEM | |
547 | if (PageHighMem(page)) | |
548 | totalhigh_pages++; | |
549 | #endif | |
9d0ad8ca DK |
550 | } |
551 | EXPORT_SYMBOL_GPL(__online_page_increment_counters); | |
180c06ef | 552 | |
9d0ad8ca DK |
553 | void __online_page_free(struct page *page) |
554 | { | |
180c06ef JF |
555 | ClearPageReserved(page); |
556 | init_page_count(page); | |
557 | __free_page(page); | |
558 | } | |
9d0ad8ca DK |
559 | EXPORT_SYMBOL_GPL(__online_page_free); |
560 | ||
561 | static void generic_online_page(struct page *page) | |
562 | { | |
563 | __online_page_set_limits(page); | |
564 | __online_page_increment_counters(page); | |
565 | __online_page_free(page); | |
566 | } | |
180c06ef | 567 | |
75884fb1 KH |
568 | static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages, |
569 | void *arg) | |
3947be19 DH |
570 | { |
571 | unsigned long i; | |
75884fb1 KH |
572 | unsigned long onlined_pages = *(unsigned long *)arg; |
573 | struct page *page; | |
574 | if (PageReserved(pfn_to_page(start_pfn))) | |
575 | for (i = 0; i < nr_pages; i++) { | |
576 | page = pfn_to_page(start_pfn + i); | |
9d0ad8ca | 577 | (*online_page_callback)(page); |
75884fb1 KH |
578 | onlined_pages++; |
579 | } | |
580 | *(unsigned long *)arg = onlined_pages; | |
581 | return 0; | |
582 | } | |
583 | ||
74d42d8f LJ |
584 | /* ensure every online node has NORMAL memory */ |
585 | static bool can_online_high_movable(struct zone *zone) | |
586 | { | |
587 | return node_state(zone_to_nid(zone), N_NORMAL_MEMORY); | |
588 | } | |
589 | ||
d9713679 LJ |
590 | /* check which state of node_states will be changed when online memory */ |
591 | static void node_states_check_changes_online(unsigned long nr_pages, | |
592 | struct zone *zone, struct memory_notify *arg) | |
593 | { | |
594 | int nid = zone_to_nid(zone); | |
595 | enum zone_type zone_last = ZONE_NORMAL; | |
596 | ||
597 | /* | |
598 | * If we have HIGHMEM, node_states[N_NORMAL_MEMORY] contains nodes | |
599 | * which have 0...ZONE_NORMAL, set zone_last to ZONE_NORMAL. | |
600 | * | |
601 | * If we don't have HIGHMEM, node_states[N_NORMAL_MEMORY] contains nodes | |
602 | * which have 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE. | |
603 | */ | |
604 | if (N_HIGH_MEMORY == N_NORMAL_MEMORY) | |
605 | zone_last = ZONE_MOVABLE; | |
606 | ||
607 | /* | |
608 | * if the memory to be online is in a zone of 0...zone_last, and | |
609 | * the zones of 0...zone_last don't have memory before online, we will | |
610 | * need to set the node to node_states[N_NORMAL_MEMORY] after | |
611 | * the memory is online. | |
612 | */ | |
613 | if (zone_idx(zone) <= zone_last && !node_state(nid, N_NORMAL_MEMORY)) | |
614 | arg->status_change_nid_normal = nid; | |
615 | else | |
616 | arg->status_change_nid_normal = -1; | |
617 | ||
618 | /* | |
619 | * if the node don't have memory befor online, we will need to | |
620 | * set the node to node_states[N_HIGH_MEMORY] after the memory | |
621 | * is online. | |
622 | */ | |
623 | if (!node_state(nid, N_HIGH_MEMORY)) | |
624 | arg->status_change_nid = nid; | |
625 | else | |
626 | arg->status_change_nid = -1; | |
627 | } | |
628 | ||
629 | static void node_states_set_node(int node, struct memory_notify *arg) | |
630 | { | |
631 | if (arg->status_change_nid_normal >= 0) | |
632 | node_set_state(node, N_NORMAL_MEMORY); | |
633 | ||
634 | node_set_state(node, N_HIGH_MEMORY); | |
635 | } | |
636 | ||
75884fb1 | 637 | |
511c2aba | 638 | int __ref online_pages(unsigned long pfn, unsigned long nr_pages, int online_type) |
75884fb1 | 639 | { |
3947be19 DH |
640 | unsigned long onlined_pages = 0; |
641 | struct zone *zone; | |
6811378e | 642 | int need_zonelists_rebuild = 0; |
7b78d335 YG |
643 | int nid; |
644 | int ret; | |
645 | struct memory_notify arg; | |
646 | ||
925268a0 | 647 | lock_memory_hotplug(); |
d9713679 LJ |
648 | /* |
649 | * This doesn't need a lock to do pfn_to_page(). | |
650 | * The section can't be removed here because of the | |
651 | * memory_block->state_mutex. | |
652 | */ | |
653 | zone = page_zone(pfn_to_page(pfn)); | |
654 | ||
74d42d8f LJ |
655 | if ((zone_idx(zone) > ZONE_NORMAL || online_type == ONLINE_MOVABLE) && |
656 | !can_online_high_movable(zone)) { | |
657 | unlock_memory_hotplug(); | |
658 | return -1; | |
659 | } | |
660 | ||
511c2aba LJ |
661 | if (online_type == ONLINE_KERNEL && zone_idx(zone) == ZONE_MOVABLE) { |
662 | if (move_pfn_range_left(zone - 1, zone, pfn, pfn + nr_pages)) { | |
663 | unlock_memory_hotplug(); | |
664 | return -1; | |
665 | } | |
666 | } | |
667 | if (online_type == ONLINE_MOVABLE && zone_idx(zone) == ZONE_MOVABLE - 1) { | |
668 | if (move_pfn_range_right(zone, zone + 1, pfn, pfn + nr_pages)) { | |
669 | unlock_memory_hotplug(); | |
670 | return -1; | |
671 | } | |
672 | } | |
673 | ||
674 | /* Previous code may changed the zone of the pfn range */ | |
675 | zone = page_zone(pfn_to_page(pfn)); | |
676 | ||
7b78d335 YG |
677 | arg.start_pfn = pfn; |
678 | arg.nr_pages = nr_pages; | |
d9713679 | 679 | node_states_check_changes_online(nr_pages, zone, &arg); |
7b78d335 YG |
680 | |
681 | nid = page_to_nid(pfn_to_page(pfn)); | |
3947be19 | 682 | |
7b78d335 YG |
683 | ret = memory_notify(MEM_GOING_ONLINE, &arg); |
684 | ret = notifier_to_errno(ret); | |
685 | if (ret) { | |
686 | memory_notify(MEM_CANCEL_ONLINE, &arg); | |
925268a0 | 687 | unlock_memory_hotplug(); |
7b78d335 YG |
688 | return ret; |
689 | } | |
6811378e YG |
690 | /* |
691 | * If this zone is not populated, then it is not in zonelist. | |
692 | * This means the page allocator ignores this zone. | |
693 | * So, zonelist must be updated after online. | |
694 | */ | |
4eaf3f64 | 695 | mutex_lock(&zonelists_mutex); |
6dcd73d7 | 696 | if (!populated_zone(zone)) { |
6811378e | 697 | need_zonelists_rebuild = 1; |
6dcd73d7 WC |
698 | build_all_zonelists(NULL, zone); |
699 | } | |
6811378e | 700 | |
908eedc6 | 701 | ret = walk_system_ram_range(pfn, nr_pages, &onlined_pages, |
75884fb1 | 702 | online_pages_range); |
fd8a4221 | 703 | if (ret) { |
6dcd73d7 WC |
704 | if (need_zonelists_rebuild) |
705 | zone_pcp_reset(zone); | |
4eaf3f64 | 706 | mutex_unlock(&zonelists_mutex); |
a62e2f4f BH |
707 | printk(KERN_DEBUG "online_pages [mem %#010llx-%#010llx] failed\n", |
708 | (unsigned long long) pfn << PAGE_SHIFT, | |
709 | (((unsigned long long) pfn + nr_pages) | |
710 | << PAGE_SHIFT) - 1); | |
fd8a4221 | 711 | memory_notify(MEM_CANCEL_ONLINE, &arg); |
925268a0 | 712 | unlock_memory_hotplug(); |
fd8a4221 GL |
713 | return ret; |
714 | } | |
715 | ||
3947be19 | 716 | zone->present_pages += onlined_pages; |
f2937be5 | 717 | zone->zone_pgdat->node_present_pages += onlined_pages; |
08dff7b7 | 718 | if (onlined_pages) { |
d9713679 | 719 | node_states_set_node(zone_to_nid(zone), &arg); |
08dff7b7 | 720 | if (need_zonelists_rebuild) |
6dcd73d7 | 721 | build_all_zonelists(NULL, NULL); |
08dff7b7 JL |
722 | else |
723 | zone_pcp_update(zone); | |
724 | } | |
3947be19 | 725 | |
4eaf3f64 | 726 | mutex_unlock(&zonelists_mutex); |
1b79acc9 KM |
727 | |
728 | init_per_zone_wmark_min(); | |
729 | ||
08dff7b7 | 730 | if (onlined_pages) |
7ea1530a | 731 | kswapd_run(zone_to_nid(zone)); |
61b13993 | 732 | |
1f522509 | 733 | vm_total_pages = nr_free_pagecache_pages(); |
2f7f24ec | 734 | |
2d1d43f6 | 735 | writeback_set_ratelimit(); |
7b78d335 YG |
736 | |
737 | if (onlined_pages) | |
738 | memory_notify(MEM_ONLINE, &arg); | |
925268a0 | 739 | unlock_memory_hotplug(); |
7b78d335 | 740 | |
3947be19 DH |
741 | return 0; |
742 | } | |
53947027 | 743 | #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */ |
bc02af93 | 744 | |
e1319331 HS |
745 | /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */ |
746 | static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start) | |
9af3c2de YG |
747 | { |
748 | struct pglist_data *pgdat; | |
749 | unsigned long zones_size[MAX_NR_ZONES] = {0}; | |
750 | unsigned long zholes_size[MAX_NR_ZONES] = {0}; | |
751 | unsigned long start_pfn = start >> PAGE_SHIFT; | |
752 | ||
753 | pgdat = arch_alloc_nodedata(nid); | |
754 | if (!pgdat) | |
755 | return NULL; | |
756 | ||
757 | arch_refresh_nodedata(nid, pgdat); | |
758 | ||
759 | /* we can use NODE_DATA(nid) from here */ | |
760 | ||
761 | /* init node's zones as empty zones, we don't have any present pages.*/ | |
9109fb7b | 762 | free_area_init_node(nid, zones_size, start_pfn, zholes_size); |
9af3c2de | 763 | |
959ecc48 KH |
764 | /* |
765 | * The node we allocated has no zone fallback lists. For avoiding | |
766 | * to access not-initialized zonelist, build here. | |
767 | */ | |
f957db4f | 768 | mutex_lock(&zonelists_mutex); |
9adb62a5 | 769 | build_all_zonelists(pgdat, NULL); |
f957db4f | 770 | mutex_unlock(&zonelists_mutex); |
959ecc48 | 771 | |
9af3c2de YG |
772 | return pgdat; |
773 | } | |
774 | ||
775 | static void rollback_node_hotadd(int nid, pg_data_t *pgdat) | |
776 | { | |
777 | arch_refresh_nodedata(nid, NULL); | |
778 | arch_free_nodedata(pgdat); | |
779 | return; | |
780 | } | |
781 | ||
0a547039 | 782 | |
cf23422b | 783 | /* |
784 | * called by cpu_up() to online a node without onlined memory. | |
785 | */ | |
786 | int mem_online_node(int nid) | |
787 | { | |
788 | pg_data_t *pgdat; | |
789 | int ret; | |
790 | ||
20d6c96b | 791 | lock_memory_hotplug(); |
cf23422b | 792 | pgdat = hotadd_new_pgdat(nid, 0); |
7553e8f2 | 793 | if (!pgdat) { |
cf23422b | 794 | ret = -ENOMEM; |
795 | goto out; | |
796 | } | |
797 | node_set_online(nid); | |
798 | ret = register_one_node(nid); | |
799 | BUG_ON(ret); | |
800 | ||
801 | out: | |
20d6c96b | 802 | unlock_memory_hotplug(); |
cf23422b | 803 | return ret; |
804 | } | |
805 | ||
31168481 AV |
806 | /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */ |
807 | int __ref add_memory(int nid, u64 start, u64 size) | |
bc02af93 | 808 | { |
9af3c2de YG |
809 | pg_data_t *pgdat = NULL; |
810 | int new_pgdat = 0; | |
ebd15302 | 811 | struct resource *res; |
bc02af93 YG |
812 | int ret; |
813 | ||
20d6c96b | 814 | lock_memory_hotplug(); |
6ad696d2 | 815 | |
ebd15302 | 816 | res = register_memory_resource(start, size); |
6ad696d2 | 817 | ret = -EEXIST; |
ebd15302 | 818 | if (!res) |
6ad696d2 | 819 | goto out; |
ebd15302 | 820 | |
9af3c2de YG |
821 | if (!node_online(nid)) { |
822 | pgdat = hotadd_new_pgdat(nid, start); | |
6ad696d2 | 823 | ret = -ENOMEM; |
9af3c2de | 824 | if (!pgdat) |
41b9e2d7 | 825 | goto error; |
9af3c2de | 826 | new_pgdat = 1; |
9af3c2de YG |
827 | } |
828 | ||
bc02af93 YG |
829 | /* call arch's memory hotadd */ |
830 | ret = arch_add_memory(nid, start, size); | |
831 | ||
9af3c2de YG |
832 | if (ret < 0) |
833 | goto error; | |
834 | ||
0fc44159 | 835 | /* we online node here. we can't roll back from here. */ |
9af3c2de YG |
836 | node_set_online(nid); |
837 | ||
0fc44159 YG |
838 | if (new_pgdat) { |
839 | ret = register_one_node(nid); | |
840 | /* | |
841 | * If sysfs file of new node can't create, cpu on the node | |
842 | * can't be hot-added. There is no rollback way now. | |
843 | * So, check by BUG_ON() to catch it reluctantly.. | |
844 | */ | |
845 | BUG_ON(ret); | |
846 | } | |
847 | ||
d96ae530 | 848 | /* create new memmap entry */ |
849 | firmware_map_add_hotplug(start, start + size, "System RAM"); | |
850 | ||
6ad696d2 AK |
851 | goto out; |
852 | ||
9af3c2de YG |
853 | error: |
854 | /* rollback pgdat allocation and others */ | |
855 | if (new_pgdat) | |
856 | rollback_node_hotadd(nid, pgdat); | |
ebd15302 KH |
857 | if (res) |
858 | release_memory_resource(res); | |
9af3c2de | 859 | |
6ad696d2 | 860 | out: |
20d6c96b | 861 | unlock_memory_hotplug(); |
bc02af93 YG |
862 | return ret; |
863 | } | |
864 | EXPORT_SYMBOL_GPL(add_memory); | |
0c0e6195 KH |
865 | |
866 | #ifdef CONFIG_MEMORY_HOTREMOVE | |
5c755e9f BP |
867 | /* |
868 | * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy | |
869 | * set and the size of the free page is given by page_order(). Using this, | |
870 | * the function determines if the pageblock contains only free pages. | |
871 | * Due to buddy contraints, a free page at least the size of a pageblock will | |
872 | * be located at the start of the pageblock | |
873 | */ | |
874 | static inline int pageblock_free(struct page *page) | |
875 | { | |
876 | return PageBuddy(page) && page_order(page) >= pageblock_order; | |
877 | } | |
878 | ||
879 | /* Return the start of the next active pageblock after a given page */ | |
880 | static struct page *next_active_pageblock(struct page *page) | |
881 | { | |
5c755e9f BP |
882 | /* Ensure the starting page is pageblock-aligned */ |
883 | BUG_ON(page_to_pfn(page) & (pageblock_nr_pages - 1)); | |
884 | ||
5c755e9f | 885 | /* If the entire pageblock is free, move to the end of free page */ |
0dcc48c1 KH |
886 | if (pageblock_free(page)) { |
887 | int order; | |
888 | /* be careful. we don't have locks, page_order can be changed.*/ | |
889 | order = page_order(page); | |
890 | if ((order < MAX_ORDER) && (order >= pageblock_order)) | |
891 | return page + (1 << order); | |
892 | } | |
5c755e9f | 893 | |
0dcc48c1 | 894 | return page + pageblock_nr_pages; |
5c755e9f BP |
895 | } |
896 | ||
897 | /* Checks if this range of memory is likely to be hot-removable. */ | |
898 | int is_mem_section_removable(unsigned long start_pfn, unsigned long nr_pages) | |
899 | { | |
5c755e9f BP |
900 | struct page *page = pfn_to_page(start_pfn); |
901 | struct page *end_page = page + nr_pages; | |
902 | ||
903 | /* Check the starting page of each pageblock within the range */ | |
904 | for (; page < end_page; page = next_active_pageblock(page)) { | |
49ac8255 | 905 | if (!is_pageblock_removable_nolock(page)) |
5c755e9f | 906 | return 0; |
49ac8255 | 907 | cond_resched(); |
5c755e9f BP |
908 | } |
909 | ||
910 | /* All pageblocks in the memory block are likely to be hot-removable */ | |
911 | return 1; | |
912 | } | |
913 | ||
0c0e6195 KH |
914 | /* |
915 | * Confirm all pages in a range [start, end) is belongs to the same zone. | |
916 | */ | |
917 | static int test_pages_in_a_zone(unsigned long start_pfn, unsigned long end_pfn) | |
918 | { | |
919 | unsigned long pfn; | |
920 | struct zone *zone = NULL; | |
921 | struct page *page; | |
922 | int i; | |
923 | for (pfn = start_pfn; | |
924 | pfn < end_pfn; | |
925 | pfn += MAX_ORDER_NR_PAGES) { | |
926 | i = 0; | |
927 | /* This is just a CONFIG_HOLES_IN_ZONE check.*/ | |
928 | while ((i < MAX_ORDER_NR_PAGES) && !pfn_valid_within(pfn + i)) | |
929 | i++; | |
930 | if (i == MAX_ORDER_NR_PAGES) | |
931 | continue; | |
932 | page = pfn_to_page(pfn + i); | |
933 | if (zone && page_zone(page) != zone) | |
934 | return 0; | |
935 | zone = page_zone(page); | |
936 | } | |
937 | return 1; | |
938 | } | |
939 | ||
940 | /* | |
941 | * Scanning pfn is much easier than scanning lru list. | |
942 | * Scan pfn from start to end and Find LRU page. | |
943 | */ | |
7bbc0905 | 944 | static unsigned long scan_lru_pages(unsigned long start, unsigned long end) |
0c0e6195 KH |
945 | { |
946 | unsigned long pfn; | |
947 | struct page *page; | |
948 | for (pfn = start; pfn < end; pfn++) { | |
949 | if (pfn_valid(pfn)) { | |
950 | page = pfn_to_page(pfn); | |
951 | if (PageLRU(page)) | |
952 | return pfn; | |
953 | } | |
954 | } | |
955 | return 0; | |
956 | } | |
957 | ||
0c0e6195 KH |
958 | #define NR_OFFLINE_AT_ONCE_PAGES (256) |
959 | static int | |
960 | do_migrate_range(unsigned long start_pfn, unsigned long end_pfn) | |
961 | { | |
962 | unsigned long pfn; | |
963 | struct page *page; | |
964 | int move_pages = NR_OFFLINE_AT_ONCE_PAGES; | |
965 | int not_managed = 0; | |
966 | int ret = 0; | |
967 | LIST_HEAD(source); | |
968 | ||
969 | for (pfn = start_pfn; pfn < end_pfn && move_pages > 0; pfn++) { | |
970 | if (!pfn_valid(pfn)) | |
971 | continue; | |
972 | page = pfn_to_page(pfn); | |
700c2a46 | 973 | if (!get_page_unless_zero(page)) |
0c0e6195 KH |
974 | continue; |
975 | /* | |
976 | * We can skip free pages. And we can only deal with pages on | |
977 | * LRU. | |
978 | */ | |
62695a84 | 979 | ret = isolate_lru_page(page); |
0c0e6195 | 980 | if (!ret) { /* Success */ |
700c2a46 | 981 | put_page(page); |
62695a84 | 982 | list_add_tail(&page->lru, &source); |
0c0e6195 | 983 | move_pages--; |
6d9c285a KM |
984 | inc_zone_page_state(page, NR_ISOLATED_ANON + |
985 | page_is_file_cache(page)); | |
986 | ||
0c0e6195 | 987 | } else { |
0c0e6195 | 988 | #ifdef CONFIG_DEBUG_VM |
718a3821 WF |
989 | printk(KERN_ALERT "removing pfn %lx from LRU failed\n", |
990 | pfn); | |
991 | dump_page(page); | |
0c0e6195 | 992 | #endif |
700c2a46 | 993 | put_page(page); |
25985edc | 994 | /* Because we don't have big zone->lock. we should |
809c4449 BL |
995 | check this again here. */ |
996 | if (page_count(page)) { | |
997 | not_managed++; | |
f3ab2636 | 998 | ret = -EBUSY; |
809c4449 BL |
999 | break; |
1000 | } | |
0c0e6195 KH |
1001 | } |
1002 | } | |
f3ab2636 BL |
1003 | if (!list_empty(&source)) { |
1004 | if (not_managed) { | |
1005 | putback_lru_pages(&source); | |
1006 | goto out; | |
1007 | } | |
74c08f98 MK |
1008 | |
1009 | /* | |
1010 | * alloc_migrate_target should be improooooved!! | |
1011 | * migrate_pages returns # of failed pages. | |
1012 | */ | |
1013 | ret = migrate_pages(&source, alloc_migrate_target, 0, | |
a6bc32b8 | 1014 | true, MIGRATE_SYNC); |
f3ab2636 | 1015 | if (ret) |
0c0e6195 | 1016 | putback_lru_pages(&source); |
0c0e6195 | 1017 | } |
0c0e6195 KH |
1018 | out: |
1019 | return ret; | |
1020 | } | |
1021 | ||
1022 | /* | |
1023 | * remove from free_area[] and mark all as Reserved. | |
1024 | */ | |
1025 | static int | |
1026 | offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages, | |
1027 | void *data) | |
1028 | { | |
1029 | __offline_isolated_pages(start, start + nr_pages); | |
1030 | return 0; | |
1031 | } | |
1032 | ||
1033 | static void | |
1034 | offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn) | |
1035 | { | |
908eedc6 | 1036 | walk_system_ram_range(start_pfn, end_pfn - start_pfn, NULL, |
0c0e6195 KH |
1037 | offline_isolated_pages_cb); |
1038 | } | |
1039 | ||
1040 | /* | |
1041 | * Check all pages in range, recoreded as memory resource, are isolated. | |
1042 | */ | |
1043 | static int | |
1044 | check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages, | |
1045 | void *data) | |
1046 | { | |
1047 | int ret; | |
1048 | long offlined = *(long *)data; | |
b023f468 | 1049 | ret = test_pages_isolated(start_pfn, start_pfn + nr_pages, true); |
0c0e6195 KH |
1050 | offlined = nr_pages; |
1051 | if (!ret) | |
1052 | *(long *)data += offlined; | |
1053 | return ret; | |
1054 | } | |
1055 | ||
1056 | static long | |
1057 | check_pages_isolated(unsigned long start_pfn, unsigned long end_pfn) | |
1058 | { | |
1059 | long offlined = 0; | |
1060 | int ret; | |
1061 | ||
908eedc6 | 1062 | ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn, &offlined, |
0c0e6195 KH |
1063 | check_pages_isolated_cb); |
1064 | if (ret < 0) | |
1065 | offlined = (long)ret; | |
1066 | return offlined; | |
1067 | } | |
1068 | ||
74d42d8f LJ |
1069 | /* ensure the node has NORMAL memory if it is still online */ |
1070 | static bool can_offline_normal(struct zone *zone, unsigned long nr_pages) | |
1071 | { | |
1072 | struct pglist_data *pgdat = zone->zone_pgdat; | |
1073 | unsigned long present_pages = 0; | |
1074 | enum zone_type zt; | |
1075 | ||
1076 | for (zt = 0; zt <= ZONE_NORMAL; zt++) | |
1077 | present_pages += pgdat->node_zones[zt].present_pages; | |
1078 | ||
1079 | if (present_pages > nr_pages) | |
1080 | return true; | |
1081 | ||
1082 | present_pages = 0; | |
1083 | for (; zt <= ZONE_MOVABLE; zt++) | |
1084 | present_pages += pgdat->node_zones[zt].present_pages; | |
1085 | ||
1086 | /* | |
1087 | * we can't offline the last normal memory until all | |
1088 | * higher memory is offlined. | |
1089 | */ | |
1090 | return present_pages == 0; | |
1091 | } | |
1092 | ||
d9713679 LJ |
1093 | /* check which state of node_states will be changed when offline memory */ |
1094 | static void node_states_check_changes_offline(unsigned long nr_pages, | |
1095 | struct zone *zone, struct memory_notify *arg) | |
1096 | { | |
1097 | struct pglist_data *pgdat = zone->zone_pgdat; | |
1098 | unsigned long present_pages = 0; | |
1099 | enum zone_type zt, zone_last = ZONE_NORMAL; | |
1100 | ||
1101 | /* | |
1102 | * If we have HIGHMEM, node_states[N_NORMAL_MEMORY] contains nodes | |
1103 | * which have 0...ZONE_NORMAL, set zone_last to ZONE_NORMAL. | |
1104 | * | |
1105 | * If we don't have HIGHMEM, node_states[N_NORMAL_MEMORY] contains nodes | |
1106 | * which have 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE. | |
1107 | */ | |
1108 | if (N_HIGH_MEMORY == N_NORMAL_MEMORY) | |
1109 | zone_last = ZONE_MOVABLE; | |
1110 | ||
1111 | /* | |
1112 | * check whether node_states[N_NORMAL_MEMORY] will be changed. | |
1113 | * If the memory to be offline is in a zone of 0...zone_last, | |
1114 | * and it is the last present memory, 0...zone_last will | |
1115 | * become empty after offline , thus we can determind we will | |
1116 | * need to clear the node from node_states[N_NORMAL_MEMORY]. | |
1117 | */ | |
1118 | for (zt = 0; zt <= zone_last; zt++) | |
1119 | present_pages += pgdat->node_zones[zt].present_pages; | |
1120 | if (zone_idx(zone) <= zone_last && nr_pages >= present_pages) | |
1121 | arg->status_change_nid_normal = zone_to_nid(zone); | |
1122 | else | |
1123 | arg->status_change_nid_normal = -1; | |
1124 | ||
1125 | /* | |
1126 | * node_states[N_HIGH_MEMORY] contains nodes which have 0...ZONE_MOVABLE | |
1127 | */ | |
1128 | zone_last = ZONE_MOVABLE; | |
1129 | ||
1130 | /* | |
1131 | * check whether node_states[N_HIGH_MEMORY] will be changed | |
1132 | * If we try to offline the last present @nr_pages from the node, | |
1133 | * we can determind we will need to clear the node from | |
1134 | * node_states[N_HIGH_MEMORY]. | |
1135 | */ | |
1136 | for (; zt <= zone_last; zt++) | |
1137 | present_pages += pgdat->node_zones[zt].present_pages; | |
1138 | if (nr_pages >= present_pages) | |
1139 | arg->status_change_nid = zone_to_nid(zone); | |
1140 | else | |
1141 | arg->status_change_nid = -1; | |
1142 | } | |
1143 | ||
1144 | static void node_states_clear_node(int node, struct memory_notify *arg) | |
1145 | { | |
1146 | if (arg->status_change_nid_normal >= 0) | |
1147 | node_clear_state(node, N_NORMAL_MEMORY); | |
1148 | ||
1149 | if ((N_HIGH_MEMORY != N_NORMAL_MEMORY) && | |
1150 | (arg->status_change_nid >= 0)) | |
1151 | node_clear_state(node, N_HIGH_MEMORY); | |
1152 | } | |
1153 | ||
a16cee10 | 1154 | static int __ref __offline_pages(unsigned long start_pfn, |
0c0e6195 KH |
1155 | unsigned long end_pfn, unsigned long timeout) |
1156 | { | |
1157 | unsigned long pfn, nr_pages, expire; | |
1158 | long offlined_pages; | |
7b78d335 | 1159 | int ret, drain, retry_max, node; |
0c0e6195 | 1160 | struct zone *zone; |
7b78d335 | 1161 | struct memory_notify arg; |
0c0e6195 KH |
1162 | |
1163 | BUG_ON(start_pfn >= end_pfn); | |
1164 | /* at least, alignment against pageblock is necessary */ | |
1165 | if (!IS_ALIGNED(start_pfn, pageblock_nr_pages)) | |
1166 | return -EINVAL; | |
1167 | if (!IS_ALIGNED(end_pfn, pageblock_nr_pages)) | |
1168 | return -EINVAL; | |
1169 | /* This makes hotplug much easier...and readable. | |
1170 | we assume this for now. .*/ | |
1171 | if (!test_pages_in_a_zone(start_pfn, end_pfn)) | |
1172 | return -EINVAL; | |
7b78d335 | 1173 | |
20d6c96b | 1174 | lock_memory_hotplug(); |
6ad696d2 | 1175 | |
7b78d335 YG |
1176 | zone = page_zone(pfn_to_page(start_pfn)); |
1177 | node = zone_to_nid(zone); | |
1178 | nr_pages = end_pfn - start_pfn; | |
1179 | ||
74d42d8f LJ |
1180 | ret = -EINVAL; |
1181 | if (zone_idx(zone) <= ZONE_NORMAL && !can_offline_normal(zone, nr_pages)) | |
1182 | goto out; | |
1183 | ||
0c0e6195 | 1184 | /* set above range as isolated */ |
b023f468 WC |
1185 | ret = start_isolate_page_range(start_pfn, end_pfn, |
1186 | MIGRATE_MOVABLE, true); | |
0c0e6195 | 1187 | if (ret) |
6ad696d2 | 1188 | goto out; |
7b78d335 YG |
1189 | |
1190 | arg.start_pfn = start_pfn; | |
1191 | arg.nr_pages = nr_pages; | |
d9713679 | 1192 | node_states_check_changes_offline(nr_pages, zone, &arg); |
7b78d335 YG |
1193 | |
1194 | ret = memory_notify(MEM_GOING_OFFLINE, &arg); | |
1195 | ret = notifier_to_errno(ret); | |
1196 | if (ret) | |
1197 | goto failed_removal; | |
1198 | ||
0c0e6195 KH |
1199 | pfn = start_pfn; |
1200 | expire = jiffies + timeout; | |
1201 | drain = 0; | |
1202 | retry_max = 5; | |
1203 | repeat: | |
1204 | /* start memory hot removal */ | |
1205 | ret = -EAGAIN; | |
1206 | if (time_after(jiffies, expire)) | |
1207 | goto failed_removal; | |
1208 | ret = -EINTR; | |
1209 | if (signal_pending(current)) | |
1210 | goto failed_removal; | |
1211 | ret = 0; | |
1212 | if (drain) { | |
1213 | lru_add_drain_all(); | |
0c0e6195 | 1214 | cond_resched(); |
9f8f2172 | 1215 | drain_all_pages(); |
0c0e6195 KH |
1216 | } |
1217 | ||
1218 | pfn = scan_lru_pages(start_pfn, end_pfn); | |
1219 | if (pfn) { /* We have page on LRU */ | |
1220 | ret = do_migrate_range(pfn, end_pfn); | |
1221 | if (!ret) { | |
1222 | drain = 1; | |
1223 | goto repeat; | |
1224 | } else { | |
1225 | if (ret < 0) | |
1226 | if (--retry_max == 0) | |
1227 | goto failed_removal; | |
1228 | yield(); | |
1229 | drain = 1; | |
1230 | goto repeat; | |
1231 | } | |
1232 | } | |
1233 | /* drain all zone's lru pagevec, this is asyncronous... */ | |
1234 | lru_add_drain_all(); | |
0c0e6195 KH |
1235 | yield(); |
1236 | /* drain pcp pages , this is synchrouns. */ | |
9f8f2172 | 1237 | drain_all_pages(); |
0c0e6195 KH |
1238 | /* check again */ |
1239 | offlined_pages = check_pages_isolated(start_pfn, end_pfn); | |
1240 | if (offlined_pages < 0) { | |
1241 | ret = -EBUSY; | |
1242 | goto failed_removal; | |
1243 | } | |
1244 | printk(KERN_INFO "Offlined Pages %ld\n", offlined_pages); | |
1245 | /* Ok, all of our target is islaoted. | |
1246 | We cannot do rollback at this point. */ | |
1247 | offline_isolated_pages(start_pfn, end_pfn); | |
dbc0e4ce | 1248 | /* reset pagetype flags and makes migrate type to be MOVABLE */ |
0815f3d8 | 1249 | undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE); |
0c0e6195 | 1250 | /* removal success */ |
0c0e6195 KH |
1251 | zone->present_pages -= offlined_pages; |
1252 | zone->zone_pgdat->node_present_pages -= offlined_pages; | |
1253 | totalram_pages -= offlined_pages; | |
7b78d335 | 1254 | |
1b79acc9 KM |
1255 | init_per_zone_wmark_min(); |
1256 | ||
1e8537ba | 1257 | if (!populated_zone(zone)) { |
340175b7 | 1258 | zone_pcp_reset(zone); |
1e8537ba XQ |
1259 | mutex_lock(&zonelists_mutex); |
1260 | build_all_zonelists(NULL, NULL); | |
1261 | mutex_unlock(&zonelists_mutex); | |
1262 | } else | |
1263 | zone_pcp_update(zone); | |
340175b7 | 1264 | |
d9713679 LJ |
1265 | node_states_clear_node(node, &arg); |
1266 | if (arg.status_change_nid >= 0) | |
8fe23e05 | 1267 | kswapd_stop(node); |
bce7394a | 1268 | |
0c0e6195 KH |
1269 | vm_total_pages = nr_free_pagecache_pages(); |
1270 | writeback_set_ratelimit(); | |
7b78d335 YG |
1271 | |
1272 | memory_notify(MEM_OFFLINE, &arg); | |
20d6c96b | 1273 | unlock_memory_hotplug(); |
0c0e6195 KH |
1274 | return 0; |
1275 | ||
1276 | failed_removal: | |
a62e2f4f BH |
1277 | printk(KERN_INFO "memory offlining [mem %#010llx-%#010llx] failed\n", |
1278 | (unsigned long long) start_pfn << PAGE_SHIFT, | |
1279 | ((unsigned long long) end_pfn << PAGE_SHIFT) - 1); | |
7b78d335 | 1280 | memory_notify(MEM_CANCEL_OFFLINE, &arg); |
0c0e6195 | 1281 | /* pushback to free area */ |
0815f3d8 | 1282 | undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE); |
7b78d335 | 1283 | |
6ad696d2 | 1284 | out: |
20d6c96b | 1285 | unlock_memory_hotplug(); |
0c0e6195 KH |
1286 | return ret; |
1287 | } | |
71088785 | 1288 | |
a16cee10 WC |
1289 | int offline_pages(unsigned long start_pfn, unsigned long nr_pages) |
1290 | { | |
1291 | return __offline_pages(start_pfn, start_pfn + nr_pages, 120 * HZ); | |
1292 | } | |
1293 | ||
71088785 BP |
1294 | int remove_memory(u64 start, u64 size) |
1295 | { | |
e90bdb7f WC |
1296 | struct memory_block *mem = NULL; |
1297 | struct mem_section *section; | |
71088785 | 1298 | unsigned long start_pfn, end_pfn; |
e90bdb7f WC |
1299 | unsigned long pfn, section_nr; |
1300 | int ret; | |
71088785 BP |
1301 | |
1302 | start_pfn = PFN_DOWN(start); | |
1303 | end_pfn = start_pfn + PFN_DOWN(size); | |
e90bdb7f WC |
1304 | |
1305 | for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) { | |
1306 | section_nr = pfn_to_section_nr(pfn); | |
1307 | if (!present_section_nr(section_nr)) | |
1308 | continue; | |
1309 | ||
1310 | section = __nr_to_section(section_nr); | |
1311 | /* same memblock? */ | |
1312 | if (mem) | |
1313 | if ((section_nr >= mem->start_section_nr) && | |
1314 | (section_nr <= mem->end_section_nr)) | |
1315 | continue; | |
1316 | ||
1317 | mem = find_memory_block_hinted(section, mem); | |
1318 | if (!mem) | |
1319 | continue; | |
1320 | ||
1321 | ret = offline_memory_block(mem); | |
1322 | if (ret) { | |
1323 | kobject_put(&mem->dev.kobj); | |
1324 | return ret; | |
1325 | } | |
1326 | } | |
1327 | ||
1328 | if (mem) | |
1329 | kobject_put(&mem->dev.kobj); | |
1330 | ||
1331 | return 0; | |
71088785 | 1332 | } |
48e94196 | 1333 | #else |
a16cee10 WC |
1334 | int offline_pages(unsigned long start_pfn, unsigned long nr_pages) |
1335 | { | |
1336 | return -EINVAL; | |
1337 | } | |
48e94196 KH |
1338 | int remove_memory(u64 start, u64 size) |
1339 | { | |
1340 | return -EINVAL; | |
1341 | } | |
0c0e6195 | 1342 | #endif /* CONFIG_MEMORY_HOTREMOVE */ |
71088785 | 1343 | EXPORT_SYMBOL_GPL(remove_memory); |