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> |
60a5a19e | 32 | #include <linux/stop_machine.h> |
3947be19 DH |
33 | |
34 | #include <asm/tlbflush.h> | |
35 | ||
1e5ad9a3 AB |
36 | #include "internal.h" |
37 | ||
9d0ad8ca DK |
38 | /* |
39 | * online_page_callback contains pointer to current page onlining function. | |
40 | * Initially it is generic_online_page(). If it is required it could be | |
41 | * changed by calling set_online_page_callback() for callback registration | |
42 | * and restore_online_page_callback() for generic callback restore. | |
43 | */ | |
44 | ||
45 | static void generic_online_page(struct page *page); | |
46 | ||
47 | static online_page_callback_t online_page_callback = generic_online_page; | |
48 | ||
20d6c96b KM |
49 | DEFINE_MUTEX(mem_hotplug_mutex); |
50 | ||
51 | void lock_memory_hotplug(void) | |
52 | { | |
53 | mutex_lock(&mem_hotplug_mutex); | |
54 | ||
55 | /* for exclusive hibernation if CONFIG_HIBERNATION=y */ | |
56 | lock_system_sleep(); | |
57 | } | |
58 | ||
59 | void unlock_memory_hotplug(void) | |
60 | { | |
61 | unlock_system_sleep(); | |
62 | mutex_unlock(&mem_hotplug_mutex); | |
63 | } | |
64 | ||
65 | ||
45e0b78b KM |
66 | /* add this memory to iomem resource */ |
67 | static struct resource *register_memory_resource(u64 start, u64 size) | |
68 | { | |
69 | struct resource *res; | |
70 | res = kzalloc(sizeof(struct resource), GFP_KERNEL); | |
71 | BUG_ON(!res); | |
72 | ||
73 | res->name = "System RAM"; | |
74 | res->start = start; | |
75 | res->end = start + size - 1; | |
887c3cb1 | 76 | res->flags = IORESOURCE_MEM | IORESOURCE_BUSY; |
45e0b78b | 77 | if (request_resource(&iomem_resource, res) < 0) { |
a62e2f4f | 78 | printk("System RAM resource %pR cannot be added\n", res); |
45e0b78b KM |
79 | kfree(res); |
80 | res = NULL; | |
81 | } | |
82 | return res; | |
83 | } | |
84 | ||
85 | static void release_memory_resource(struct resource *res) | |
86 | { | |
87 | if (!res) | |
88 | return; | |
89 | release_resource(res); | |
90 | kfree(res); | |
91 | return; | |
92 | } | |
93 | ||
53947027 | 94 | #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE |
46723bfa YI |
95 | 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; |
9feedc9d | 109 | static DEFINE_MUTEX(ppb_lock); |
04753278 | 110 | |
5f24ce5f AA |
111 | type = (unsigned long) page->lru.next; |
112 | BUG_ON(type < MEMORY_HOTPLUG_MIN_BOOTMEM_TYPE || | |
113 | type > MEMORY_HOTPLUG_MAX_BOOTMEM_TYPE); | |
04753278 YG |
114 | |
115 | if (atomic_dec_return(&page->_count) == 1) { | |
116 | ClearPagePrivate(page); | |
117 | set_page_private(page, 0); | |
5f24ce5f | 118 | INIT_LIST_HEAD(&page->lru); |
9feedc9d JL |
119 | |
120 | /* | |
121 | * Please refer to comment for __free_pages_bootmem() | |
122 | * for why we serialize here. | |
123 | */ | |
124 | mutex_lock(&ppb_lock); | |
04753278 | 125 | __free_pages_bootmem(page, 0); |
9feedc9d | 126 | mutex_unlock(&ppb_lock); |
c60514b6 | 127 | totalram_pages++; |
04753278 YG |
128 | } |
129 | ||
130 | } | |
131 | ||
46723bfa YI |
132 | #ifdef CONFIG_HAVE_BOOTMEM_INFO_NODE |
133 | #ifndef CONFIG_SPARSEMEM_VMEMMAP | |
d92bc318 | 134 | static void register_page_bootmem_info_section(unsigned long start_pfn) |
04753278 YG |
135 | { |
136 | unsigned long *usemap, mapsize, section_nr, i; | |
137 | struct mem_section *ms; | |
138 | struct page *page, *memmap; | |
139 | ||
04753278 YG |
140 | section_nr = pfn_to_section_nr(start_pfn); |
141 | ms = __nr_to_section(section_nr); | |
142 | ||
143 | /* Get section's memmap address */ | |
144 | memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr); | |
145 | ||
146 | /* | |
147 | * Get page for the memmap's phys address | |
148 | * XXX: need more consideration for sparse_vmemmap... | |
149 | */ | |
150 | page = virt_to_page(memmap); | |
151 | mapsize = sizeof(struct page) * PAGES_PER_SECTION; | |
152 | mapsize = PAGE_ALIGN(mapsize) >> PAGE_SHIFT; | |
153 | ||
154 | /* remember memmap's page */ | |
155 | for (i = 0; i < mapsize; i++, page++) | |
156 | get_page_bootmem(section_nr, page, SECTION_INFO); | |
157 | ||
158 | usemap = __nr_to_section(section_nr)->pageblock_flags; | |
159 | page = virt_to_page(usemap); | |
160 | ||
161 | mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT; | |
162 | ||
163 | for (i = 0; i < mapsize; i++, page++) | |
af370fb8 | 164 | get_page_bootmem(section_nr, page, MIX_SECTION_INFO); |
04753278 YG |
165 | |
166 | } | |
46723bfa YI |
167 | #else /* CONFIG_SPARSEMEM_VMEMMAP */ |
168 | static void register_page_bootmem_info_section(unsigned long start_pfn) | |
169 | { | |
170 | unsigned long *usemap, mapsize, section_nr, i; | |
171 | struct mem_section *ms; | |
172 | struct page *page, *memmap; | |
173 | ||
174 | if (!pfn_valid(start_pfn)) | |
175 | return; | |
176 | ||
177 | section_nr = pfn_to_section_nr(start_pfn); | |
178 | ms = __nr_to_section(section_nr); | |
179 | ||
180 | memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr); | |
181 | ||
182 | register_page_bootmem_memmap(section_nr, memmap, PAGES_PER_SECTION); | |
183 | ||
184 | usemap = __nr_to_section(section_nr)->pageblock_flags; | |
185 | page = virt_to_page(usemap); | |
186 | ||
187 | mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT; | |
188 | ||
189 | for (i = 0; i < mapsize; i++, page++) | |
190 | get_page_bootmem(section_nr, page, MIX_SECTION_INFO); | |
191 | } | |
192 | #endif /* !CONFIG_SPARSEMEM_VMEMMAP */ | |
04753278 YG |
193 | |
194 | void register_page_bootmem_info_node(struct pglist_data *pgdat) | |
195 | { | |
196 | unsigned long i, pfn, end_pfn, nr_pages; | |
197 | int node = pgdat->node_id; | |
198 | struct page *page; | |
199 | struct zone *zone; | |
200 | ||
201 | nr_pages = PAGE_ALIGN(sizeof(struct pglist_data)) >> PAGE_SHIFT; | |
202 | page = virt_to_page(pgdat); | |
203 | ||
204 | for (i = 0; i < nr_pages; i++, page++) | |
205 | get_page_bootmem(node, page, NODE_INFO); | |
206 | ||
207 | zone = &pgdat->node_zones[0]; | |
208 | for (; zone < pgdat->node_zones + MAX_NR_ZONES - 1; zone++) { | |
209 | if (zone->wait_table) { | |
210 | nr_pages = zone->wait_table_hash_nr_entries | |
211 | * sizeof(wait_queue_head_t); | |
212 | nr_pages = PAGE_ALIGN(nr_pages) >> PAGE_SHIFT; | |
213 | page = virt_to_page(zone->wait_table); | |
214 | ||
215 | for (i = 0; i < nr_pages; i++, page++) | |
216 | get_page_bootmem(node, page, NODE_INFO); | |
217 | } | |
218 | } | |
219 | ||
220 | pfn = pgdat->node_start_pfn; | |
221 | end_pfn = pfn + pgdat->node_spanned_pages; | |
222 | ||
223 | /* register_section info */ | |
f14851af | 224 | for (; pfn < end_pfn; pfn += PAGES_PER_SECTION) { |
225 | /* | |
226 | * Some platforms can assign the same pfn to multiple nodes - on | |
227 | * node0 as well as nodeN. To avoid registering a pfn against | |
228 | * multiple nodes we check that this pfn does not already | |
229 | * reside in some other node. | |
230 | */ | |
231 | if (pfn_valid(pfn) && (pfn_to_nid(pfn) == node)) | |
232 | register_page_bootmem_info_section(pfn); | |
233 | } | |
04753278 | 234 | } |
46723bfa | 235 | #endif /* CONFIG_HAVE_BOOTMEM_INFO_NODE */ |
04753278 | 236 | |
76cdd58e HC |
237 | static void grow_zone_span(struct zone *zone, unsigned long start_pfn, |
238 | unsigned long end_pfn) | |
239 | { | |
240 | unsigned long old_zone_end_pfn; | |
241 | ||
242 | zone_span_writelock(zone); | |
243 | ||
244 | old_zone_end_pfn = zone->zone_start_pfn + zone->spanned_pages; | |
712cd386 | 245 | if (!zone->spanned_pages || start_pfn < zone->zone_start_pfn) |
76cdd58e HC |
246 | zone->zone_start_pfn = start_pfn; |
247 | ||
248 | zone->spanned_pages = max(old_zone_end_pfn, end_pfn) - | |
249 | zone->zone_start_pfn; | |
250 | ||
251 | zone_span_writeunlock(zone); | |
252 | } | |
253 | ||
511c2aba LJ |
254 | static void resize_zone(struct zone *zone, unsigned long start_pfn, |
255 | unsigned long end_pfn) | |
256 | { | |
257 | zone_span_writelock(zone); | |
258 | ||
e455a9b9 LJ |
259 | if (end_pfn - start_pfn) { |
260 | zone->zone_start_pfn = start_pfn; | |
261 | zone->spanned_pages = end_pfn - start_pfn; | |
262 | } else { | |
263 | /* | |
264 | * make it consist as free_area_init_core(), | |
265 | * if spanned_pages = 0, then keep start_pfn = 0 | |
266 | */ | |
267 | zone->zone_start_pfn = 0; | |
268 | zone->spanned_pages = 0; | |
269 | } | |
511c2aba LJ |
270 | |
271 | zone_span_writeunlock(zone); | |
272 | } | |
273 | ||
274 | static void fix_zone_id(struct zone *zone, unsigned long start_pfn, | |
275 | unsigned long end_pfn) | |
276 | { | |
277 | enum zone_type zid = zone_idx(zone); | |
278 | int nid = zone->zone_pgdat->node_id; | |
279 | unsigned long pfn; | |
280 | ||
281 | for (pfn = start_pfn; pfn < end_pfn; pfn++) | |
282 | set_page_links(pfn_to_page(pfn), zid, nid, pfn); | |
283 | } | |
284 | ||
f6bbb78e CS |
285 | /* Can fail with -ENOMEM from allocating a wait table with vmalloc() or |
286 | * alloc_bootmem_node_nopanic() */ | |
287 | static int __ref ensure_zone_is_initialized(struct zone *zone, | |
288 | unsigned long start_pfn, unsigned long num_pages) | |
289 | { | |
290 | if (!zone_is_initialized(zone)) | |
291 | return init_currently_empty_zone(zone, start_pfn, num_pages, | |
292 | MEMMAP_HOTPLUG); | |
293 | return 0; | |
294 | } | |
295 | ||
e455a9b9 | 296 | static int __meminit move_pfn_range_left(struct zone *z1, struct zone *z2, |
511c2aba LJ |
297 | unsigned long start_pfn, unsigned long end_pfn) |
298 | { | |
e455a9b9 | 299 | int ret; |
511c2aba | 300 | unsigned long flags; |
e455a9b9 LJ |
301 | unsigned long z1_start_pfn; |
302 | ||
303 | if (!z1->wait_table) { | |
304 | ret = init_currently_empty_zone(z1, start_pfn, | |
305 | end_pfn - start_pfn, MEMMAP_HOTPLUG); | |
306 | if (ret) | |
307 | return ret; | |
308 | } | |
511c2aba LJ |
309 | |
310 | pgdat_resize_lock(z1->zone_pgdat, &flags); | |
311 | ||
312 | /* can't move pfns which are higher than @z2 */ | |
108bcc96 | 313 | if (end_pfn > zone_end_pfn(z2)) |
511c2aba LJ |
314 | goto out_fail; |
315 | /* the move out part mast at the left most of @z2 */ | |
316 | if (start_pfn > z2->zone_start_pfn) | |
317 | goto out_fail; | |
318 | /* must included/overlap */ | |
319 | if (end_pfn <= z2->zone_start_pfn) | |
320 | goto out_fail; | |
321 | ||
e455a9b9 LJ |
322 | /* use start_pfn for z1's start_pfn if z1 is empty */ |
323 | if (z1->spanned_pages) | |
324 | z1_start_pfn = z1->zone_start_pfn; | |
325 | else | |
326 | z1_start_pfn = start_pfn; | |
327 | ||
328 | resize_zone(z1, z1_start_pfn, end_pfn); | |
108bcc96 | 329 | resize_zone(z2, end_pfn, zone_end_pfn(z2)); |
511c2aba LJ |
330 | |
331 | pgdat_resize_unlock(z1->zone_pgdat, &flags); | |
332 | ||
333 | fix_zone_id(z1, start_pfn, end_pfn); | |
334 | ||
335 | return 0; | |
336 | out_fail: | |
337 | pgdat_resize_unlock(z1->zone_pgdat, &flags); | |
338 | return -1; | |
339 | } | |
340 | ||
e455a9b9 | 341 | static int __meminit move_pfn_range_right(struct zone *z1, struct zone *z2, |
511c2aba LJ |
342 | unsigned long start_pfn, unsigned long end_pfn) |
343 | { | |
e455a9b9 | 344 | int ret; |
511c2aba | 345 | unsigned long flags; |
e455a9b9 LJ |
346 | unsigned long z2_end_pfn; |
347 | ||
348 | if (!z2->wait_table) { | |
349 | ret = init_currently_empty_zone(z2, start_pfn, | |
350 | end_pfn - start_pfn, MEMMAP_HOTPLUG); | |
351 | if (ret) | |
352 | return ret; | |
353 | } | |
511c2aba LJ |
354 | |
355 | pgdat_resize_lock(z1->zone_pgdat, &flags); | |
356 | ||
357 | /* can't move pfns which are lower than @z1 */ | |
358 | if (z1->zone_start_pfn > start_pfn) | |
359 | goto out_fail; | |
360 | /* the move out part mast at the right most of @z1 */ | |
108bcc96 | 361 | if (zone_end_pfn(z1) > end_pfn) |
511c2aba LJ |
362 | goto out_fail; |
363 | /* must included/overlap */ | |
108bcc96 | 364 | if (start_pfn >= zone_end_pfn(z1)) |
511c2aba LJ |
365 | goto out_fail; |
366 | ||
e455a9b9 LJ |
367 | /* use end_pfn for z2's end_pfn if z2 is empty */ |
368 | if (z2->spanned_pages) | |
108bcc96 | 369 | z2_end_pfn = zone_end_pfn(z2); |
e455a9b9 LJ |
370 | else |
371 | z2_end_pfn = end_pfn; | |
372 | ||
511c2aba | 373 | resize_zone(z1, z1->zone_start_pfn, start_pfn); |
e455a9b9 | 374 | resize_zone(z2, start_pfn, z2_end_pfn); |
511c2aba LJ |
375 | |
376 | pgdat_resize_unlock(z1->zone_pgdat, &flags); | |
377 | ||
378 | fix_zone_id(z2, start_pfn, end_pfn); | |
379 | ||
380 | return 0; | |
381 | out_fail: | |
382 | pgdat_resize_unlock(z1->zone_pgdat, &flags); | |
383 | return -1; | |
384 | } | |
385 | ||
76cdd58e HC |
386 | static void grow_pgdat_span(struct pglist_data *pgdat, unsigned long start_pfn, |
387 | unsigned long end_pfn) | |
388 | { | |
389 | unsigned long old_pgdat_end_pfn = | |
390 | pgdat->node_start_pfn + pgdat->node_spanned_pages; | |
391 | ||
712cd386 | 392 | if (!pgdat->node_spanned_pages || start_pfn < pgdat->node_start_pfn) |
76cdd58e HC |
393 | pgdat->node_start_pfn = start_pfn; |
394 | ||
395 | pgdat->node_spanned_pages = max(old_pgdat_end_pfn, end_pfn) - | |
396 | pgdat->node_start_pfn; | |
397 | } | |
398 | ||
31168481 | 399 | static int __meminit __add_zone(struct zone *zone, unsigned long phys_start_pfn) |
3947be19 DH |
400 | { |
401 | struct pglist_data *pgdat = zone->zone_pgdat; | |
402 | int nr_pages = PAGES_PER_SECTION; | |
403 | int nid = pgdat->node_id; | |
404 | int zone_type; | |
76cdd58e | 405 | unsigned long flags; |
3947be19 DH |
406 | |
407 | zone_type = zone - pgdat->node_zones; | |
76cdd58e HC |
408 | if (!zone->wait_table) { |
409 | int ret; | |
410 | ||
411 | ret = init_currently_empty_zone(zone, phys_start_pfn, | |
412 | nr_pages, MEMMAP_HOTPLUG); | |
413 | if (ret) | |
414 | return ret; | |
415 | } | |
416 | pgdat_resize_lock(zone->zone_pgdat, &flags); | |
417 | grow_zone_span(zone, phys_start_pfn, phys_start_pfn + nr_pages); | |
418 | grow_pgdat_span(zone->zone_pgdat, phys_start_pfn, | |
419 | phys_start_pfn + nr_pages); | |
420 | pgdat_resize_unlock(zone->zone_pgdat, &flags); | |
a2f3aa02 DH |
421 | memmap_init_zone(nr_pages, nid, zone_type, |
422 | phys_start_pfn, MEMMAP_HOTPLUG); | |
718127cc | 423 | return 0; |
3947be19 DH |
424 | } |
425 | ||
c04fc586 GH |
426 | static int __meminit __add_section(int nid, struct zone *zone, |
427 | unsigned long phys_start_pfn) | |
3947be19 | 428 | { |
3947be19 | 429 | int nr_pages = PAGES_PER_SECTION; |
3947be19 DH |
430 | int ret; |
431 | ||
ebd15302 KH |
432 | if (pfn_valid(phys_start_pfn)) |
433 | return -EEXIST; | |
434 | ||
0b0acbec | 435 | ret = sparse_add_one_section(zone, phys_start_pfn, nr_pages); |
3947be19 DH |
436 | |
437 | if (ret < 0) | |
438 | return ret; | |
439 | ||
718127cc YG |
440 | ret = __add_zone(zone, phys_start_pfn); |
441 | ||
442 | if (ret < 0) | |
443 | return ret; | |
444 | ||
c04fc586 | 445 | return register_new_memory(nid, __pfn_to_section(phys_start_pfn)); |
3947be19 DH |
446 | } |
447 | ||
815121d2 YI |
448 | /* find the smallest valid pfn in the range [start_pfn, end_pfn) */ |
449 | static int find_smallest_section_pfn(int nid, struct zone *zone, | |
450 | unsigned long start_pfn, | |
451 | unsigned long end_pfn) | |
452 | { | |
453 | struct mem_section *ms; | |
454 | ||
455 | for (; start_pfn < end_pfn; start_pfn += PAGES_PER_SECTION) { | |
456 | ms = __pfn_to_section(start_pfn); | |
457 | ||
458 | if (unlikely(!valid_section(ms))) | |
459 | continue; | |
460 | ||
461 | if (unlikely(pfn_to_nid(start_pfn) != nid)) | |
462 | continue; | |
463 | ||
464 | if (zone && zone != page_zone(pfn_to_page(start_pfn))) | |
465 | continue; | |
466 | ||
467 | return start_pfn; | |
468 | } | |
469 | ||
470 | return 0; | |
471 | } | |
472 | ||
473 | /* find the biggest valid pfn in the range [start_pfn, end_pfn). */ | |
474 | static int find_biggest_section_pfn(int nid, struct zone *zone, | |
475 | unsigned long start_pfn, | |
476 | unsigned long end_pfn) | |
477 | { | |
478 | struct mem_section *ms; | |
479 | unsigned long pfn; | |
480 | ||
481 | /* pfn is the end pfn of a memory section. */ | |
482 | pfn = end_pfn - 1; | |
483 | for (; pfn >= start_pfn; pfn -= PAGES_PER_SECTION) { | |
484 | ms = __pfn_to_section(pfn); | |
485 | ||
486 | if (unlikely(!valid_section(ms))) | |
487 | continue; | |
488 | ||
489 | if (unlikely(pfn_to_nid(pfn) != nid)) | |
490 | continue; | |
491 | ||
492 | if (zone && zone != page_zone(pfn_to_page(pfn))) | |
493 | continue; | |
494 | ||
495 | return pfn; | |
496 | } | |
497 | ||
498 | return 0; | |
499 | } | |
500 | ||
501 | static void shrink_zone_span(struct zone *zone, unsigned long start_pfn, | |
502 | unsigned long end_pfn) | |
503 | { | |
504 | unsigned long zone_start_pfn = zone->zone_start_pfn; | |
505 | unsigned long zone_end_pfn = zone->zone_start_pfn + zone->spanned_pages; | |
506 | unsigned long pfn; | |
507 | struct mem_section *ms; | |
508 | int nid = zone_to_nid(zone); | |
509 | ||
510 | zone_span_writelock(zone); | |
511 | if (zone_start_pfn == start_pfn) { | |
512 | /* | |
513 | * If the section is smallest section in the zone, it need | |
514 | * shrink zone->zone_start_pfn and zone->zone_spanned_pages. | |
515 | * In this case, we find second smallest valid mem_section | |
516 | * for shrinking zone. | |
517 | */ | |
518 | pfn = find_smallest_section_pfn(nid, zone, end_pfn, | |
519 | zone_end_pfn); | |
520 | if (pfn) { | |
521 | zone->zone_start_pfn = pfn; | |
522 | zone->spanned_pages = zone_end_pfn - pfn; | |
523 | } | |
524 | } else if (zone_end_pfn == end_pfn) { | |
525 | /* | |
526 | * If the section is biggest section in the zone, it need | |
527 | * shrink zone->spanned_pages. | |
528 | * In this case, we find second biggest valid mem_section for | |
529 | * shrinking zone. | |
530 | */ | |
531 | pfn = find_biggest_section_pfn(nid, zone, zone_start_pfn, | |
532 | start_pfn); | |
533 | if (pfn) | |
534 | zone->spanned_pages = pfn - zone_start_pfn + 1; | |
535 | } | |
536 | ||
537 | /* | |
538 | * The section is not biggest or smallest mem_section in the zone, it | |
539 | * only creates a hole in the zone. So in this case, we need not | |
540 | * change the zone. But perhaps, the zone has only hole data. Thus | |
541 | * it check the zone has only hole or not. | |
542 | */ | |
543 | pfn = zone_start_pfn; | |
544 | for (; pfn < zone_end_pfn; pfn += PAGES_PER_SECTION) { | |
545 | ms = __pfn_to_section(pfn); | |
546 | ||
547 | if (unlikely(!valid_section(ms))) | |
548 | continue; | |
549 | ||
550 | if (page_zone(pfn_to_page(pfn)) != zone) | |
551 | continue; | |
552 | ||
553 | /* If the section is current section, it continues the loop */ | |
554 | if (start_pfn == pfn) | |
555 | continue; | |
556 | ||
557 | /* If we find valid section, we have nothing to do */ | |
558 | zone_span_writeunlock(zone); | |
559 | return; | |
560 | } | |
561 | ||
562 | /* The zone has no valid section */ | |
563 | zone->zone_start_pfn = 0; | |
564 | zone->spanned_pages = 0; | |
565 | zone_span_writeunlock(zone); | |
566 | } | |
567 | ||
568 | static void shrink_pgdat_span(struct pglist_data *pgdat, | |
569 | unsigned long start_pfn, unsigned long end_pfn) | |
570 | { | |
571 | unsigned long pgdat_start_pfn = pgdat->node_start_pfn; | |
572 | unsigned long pgdat_end_pfn = | |
573 | pgdat->node_start_pfn + pgdat->node_spanned_pages; | |
574 | unsigned long pfn; | |
575 | struct mem_section *ms; | |
576 | int nid = pgdat->node_id; | |
577 | ||
578 | if (pgdat_start_pfn == start_pfn) { | |
579 | /* | |
580 | * If the section is smallest section in the pgdat, it need | |
581 | * shrink pgdat->node_start_pfn and pgdat->node_spanned_pages. | |
582 | * In this case, we find second smallest valid mem_section | |
583 | * for shrinking zone. | |
584 | */ | |
585 | pfn = find_smallest_section_pfn(nid, NULL, end_pfn, | |
586 | pgdat_end_pfn); | |
587 | if (pfn) { | |
588 | pgdat->node_start_pfn = pfn; | |
589 | pgdat->node_spanned_pages = pgdat_end_pfn - pfn; | |
590 | } | |
591 | } else if (pgdat_end_pfn == end_pfn) { | |
592 | /* | |
593 | * If the section is biggest section in the pgdat, it need | |
594 | * shrink pgdat->node_spanned_pages. | |
595 | * In this case, we find second biggest valid mem_section for | |
596 | * shrinking zone. | |
597 | */ | |
598 | pfn = find_biggest_section_pfn(nid, NULL, pgdat_start_pfn, | |
599 | start_pfn); | |
600 | if (pfn) | |
601 | pgdat->node_spanned_pages = pfn - pgdat_start_pfn + 1; | |
602 | } | |
603 | ||
604 | /* | |
605 | * If the section is not biggest or smallest mem_section in the pgdat, | |
606 | * it only creates a hole in the pgdat. So in this case, we need not | |
607 | * change the pgdat. | |
608 | * But perhaps, the pgdat has only hole data. Thus it check the pgdat | |
609 | * has only hole or not. | |
610 | */ | |
611 | pfn = pgdat_start_pfn; | |
612 | for (; pfn < pgdat_end_pfn; pfn += PAGES_PER_SECTION) { | |
613 | ms = __pfn_to_section(pfn); | |
614 | ||
615 | if (unlikely(!valid_section(ms))) | |
616 | continue; | |
617 | ||
618 | if (pfn_to_nid(pfn) != nid) | |
619 | continue; | |
620 | ||
621 | /* If the section is current section, it continues the loop */ | |
622 | if (start_pfn == pfn) | |
623 | continue; | |
624 | ||
625 | /* If we find valid section, we have nothing to do */ | |
626 | return; | |
627 | } | |
628 | ||
629 | /* The pgdat has no valid section */ | |
630 | pgdat->node_start_pfn = 0; | |
631 | pgdat->node_spanned_pages = 0; | |
632 | } | |
633 | ||
634 | static void __remove_zone(struct zone *zone, unsigned long start_pfn) | |
635 | { | |
636 | struct pglist_data *pgdat = zone->zone_pgdat; | |
637 | int nr_pages = PAGES_PER_SECTION; | |
638 | int zone_type; | |
639 | unsigned long flags; | |
640 | ||
641 | zone_type = zone - pgdat->node_zones; | |
642 | ||
643 | pgdat_resize_lock(zone->zone_pgdat, &flags); | |
644 | shrink_zone_span(zone, start_pfn, start_pfn + nr_pages); | |
645 | shrink_pgdat_span(pgdat, start_pfn, start_pfn + nr_pages); | |
646 | pgdat_resize_unlock(zone->zone_pgdat, &flags); | |
647 | } | |
648 | ||
ea01ea93 BP |
649 | static int __remove_section(struct zone *zone, struct mem_section *ms) |
650 | { | |
815121d2 YI |
651 | unsigned long start_pfn; |
652 | int scn_nr; | |
ea01ea93 BP |
653 | int ret = -EINVAL; |
654 | ||
655 | if (!valid_section(ms)) | |
656 | return ret; | |
657 | ||
658 | ret = unregister_memory_section(ms); | |
659 | if (ret) | |
660 | return ret; | |
661 | ||
815121d2 YI |
662 | scn_nr = __section_nr(ms); |
663 | start_pfn = section_nr_to_pfn(scn_nr); | |
664 | __remove_zone(zone, start_pfn); | |
665 | ||
ea01ea93 | 666 | sparse_remove_one_section(zone, ms); |
ea01ea93 BP |
667 | return 0; |
668 | } | |
669 | ||
3947be19 DH |
670 | /* |
671 | * Reasonably generic function for adding memory. It is | |
672 | * expected that archs that support memory hotplug will | |
673 | * call this function after deciding the zone to which to | |
674 | * add the new pages. | |
675 | */ | |
c04fc586 GH |
676 | int __ref __add_pages(int nid, struct zone *zone, unsigned long phys_start_pfn, |
677 | unsigned long nr_pages) | |
3947be19 DH |
678 | { |
679 | unsigned long i; | |
680 | int err = 0; | |
6f712711 KH |
681 | int start_sec, end_sec; |
682 | /* during initialize mem_map, align hot-added range to section */ | |
683 | start_sec = pfn_to_section_nr(phys_start_pfn); | |
684 | end_sec = pfn_to_section_nr(phys_start_pfn + nr_pages - 1); | |
3947be19 | 685 | |
6f712711 | 686 | for (i = start_sec; i <= end_sec; i++) { |
c04fc586 | 687 | err = __add_section(nid, zone, i << PFN_SECTION_SHIFT); |
3947be19 | 688 | |
6f712711 | 689 | /* |
183ff22b | 690 | * EEXIST is finally dealt with by ioresource collision |
6f712711 KH |
691 | * check. see add_memory() => register_memory_resource() |
692 | * Warning will be printed if there is collision. | |
bed120c6 JS |
693 | */ |
694 | if (err && (err != -EEXIST)) | |
3947be19 | 695 | break; |
6f712711 | 696 | err = 0; |
3947be19 DH |
697 | } |
698 | ||
699 | return err; | |
700 | } | |
bed120c6 | 701 | EXPORT_SYMBOL_GPL(__add_pages); |
3947be19 | 702 | |
ea01ea93 BP |
703 | /** |
704 | * __remove_pages() - remove sections of pages from a zone | |
705 | * @zone: zone from which pages need to be removed | |
706 | * @phys_start_pfn: starting pageframe (must be aligned to start of a section) | |
707 | * @nr_pages: number of pages to remove (must be multiple of section size) | |
708 | * | |
709 | * Generic helper function to remove section mappings and sysfs entries | |
710 | * for the section of the memory we are removing. Caller needs to make | |
711 | * sure that pages are marked reserved and zones are adjust properly by | |
712 | * calling offline_pages(). | |
713 | */ | |
714 | int __remove_pages(struct zone *zone, unsigned long phys_start_pfn, | |
715 | unsigned long nr_pages) | |
716 | { | |
717 | unsigned long i, ret = 0; | |
718 | int sections_to_remove; | |
719 | ||
720 | /* | |
721 | * We can only remove entire sections | |
722 | */ | |
723 | BUG_ON(phys_start_pfn & ~PAGE_SECTION_MASK); | |
724 | BUG_ON(nr_pages % PAGES_PER_SECTION); | |
725 | ||
d760afd4 YI |
726 | release_mem_region(phys_start_pfn << PAGE_SHIFT, nr_pages * PAGE_SIZE); |
727 | ||
ea01ea93 BP |
728 | sections_to_remove = nr_pages / PAGES_PER_SECTION; |
729 | for (i = 0; i < sections_to_remove; i++) { | |
730 | unsigned long pfn = phys_start_pfn + i*PAGES_PER_SECTION; | |
731 | ret = __remove_section(zone, __pfn_to_section(pfn)); | |
732 | if (ret) | |
733 | break; | |
734 | } | |
735 | return ret; | |
736 | } | |
737 | EXPORT_SYMBOL_GPL(__remove_pages); | |
738 | ||
9d0ad8ca DK |
739 | int set_online_page_callback(online_page_callback_t callback) |
740 | { | |
741 | int rc = -EINVAL; | |
742 | ||
743 | lock_memory_hotplug(); | |
744 | ||
745 | if (online_page_callback == generic_online_page) { | |
746 | online_page_callback = callback; | |
747 | rc = 0; | |
748 | } | |
749 | ||
750 | unlock_memory_hotplug(); | |
751 | ||
752 | return rc; | |
753 | } | |
754 | EXPORT_SYMBOL_GPL(set_online_page_callback); | |
755 | ||
756 | int restore_online_page_callback(online_page_callback_t callback) | |
757 | { | |
758 | int rc = -EINVAL; | |
759 | ||
760 | lock_memory_hotplug(); | |
761 | ||
762 | if (online_page_callback == callback) { | |
763 | online_page_callback = generic_online_page; | |
764 | rc = 0; | |
765 | } | |
766 | ||
767 | unlock_memory_hotplug(); | |
768 | ||
769 | return rc; | |
770 | } | |
771 | EXPORT_SYMBOL_GPL(restore_online_page_callback); | |
772 | ||
773 | void __online_page_set_limits(struct page *page) | |
180c06ef | 774 | { |
4738e1b9 JB |
775 | unsigned long pfn = page_to_pfn(page); |
776 | ||
4738e1b9 JB |
777 | if (pfn >= num_physpages) |
778 | num_physpages = pfn + 1; | |
9d0ad8ca DK |
779 | } |
780 | EXPORT_SYMBOL_GPL(__online_page_set_limits); | |
781 | ||
782 | void __online_page_increment_counters(struct page *page) | |
783 | { | |
784 | totalram_pages++; | |
180c06ef JF |
785 | |
786 | #ifdef CONFIG_HIGHMEM | |
787 | if (PageHighMem(page)) | |
788 | totalhigh_pages++; | |
789 | #endif | |
9d0ad8ca DK |
790 | } |
791 | EXPORT_SYMBOL_GPL(__online_page_increment_counters); | |
180c06ef | 792 | |
9d0ad8ca DK |
793 | void __online_page_free(struct page *page) |
794 | { | |
180c06ef JF |
795 | ClearPageReserved(page); |
796 | init_page_count(page); | |
797 | __free_page(page); | |
798 | } | |
9d0ad8ca DK |
799 | EXPORT_SYMBOL_GPL(__online_page_free); |
800 | ||
801 | static void generic_online_page(struct page *page) | |
802 | { | |
803 | __online_page_set_limits(page); | |
804 | __online_page_increment_counters(page); | |
805 | __online_page_free(page); | |
806 | } | |
180c06ef | 807 | |
75884fb1 KH |
808 | static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages, |
809 | void *arg) | |
3947be19 DH |
810 | { |
811 | unsigned long i; | |
75884fb1 KH |
812 | unsigned long onlined_pages = *(unsigned long *)arg; |
813 | struct page *page; | |
814 | if (PageReserved(pfn_to_page(start_pfn))) | |
815 | for (i = 0; i < nr_pages; i++) { | |
816 | page = pfn_to_page(start_pfn + i); | |
9d0ad8ca | 817 | (*online_page_callback)(page); |
75884fb1 KH |
818 | onlined_pages++; |
819 | } | |
820 | *(unsigned long *)arg = onlined_pages; | |
821 | return 0; | |
822 | } | |
823 | ||
09285af7 | 824 | #ifdef CONFIG_MOVABLE_NODE |
79a4dcef TC |
825 | /* |
826 | * When CONFIG_MOVABLE_NODE, we permit onlining of a node which doesn't have | |
827 | * normal memory. | |
828 | */ | |
09285af7 LJ |
829 | static bool can_online_high_movable(struct zone *zone) |
830 | { | |
831 | return true; | |
832 | } | |
79a4dcef | 833 | #else /* CONFIG_MOVABLE_NODE */ |
74d42d8f LJ |
834 | /* ensure every online node has NORMAL memory */ |
835 | static bool can_online_high_movable(struct zone *zone) | |
836 | { | |
837 | return node_state(zone_to_nid(zone), N_NORMAL_MEMORY); | |
838 | } | |
79a4dcef | 839 | #endif /* CONFIG_MOVABLE_NODE */ |
74d42d8f | 840 | |
d9713679 LJ |
841 | /* check which state of node_states will be changed when online memory */ |
842 | static void node_states_check_changes_online(unsigned long nr_pages, | |
843 | struct zone *zone, struct memory_notify *arg) | |
844 | { | |
845 | int nid = zone_to_nid(zone); | |
846 | enum zone_type zone_last = ZONE_NORMAL; | |
847 | ||
848 | /* | |
6715ddf9 LJ |
849 | * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY] |
850 | * contains nodes which have zones of 0...ZONE_NORMAL, | |
851 | * set zone_last to ZONE_NORMAL. | |
d9713679 | 852 | * |
6715ddf9 LJ |
853 | * If we don't have HIGHMEM nor movable node, |
854 | * node_states[N_NORMAL_MEMORY] contains nodes which have zones of | |
855 | * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE. | |
d9713679 | 856 | */ |
6715ddf9 | 857 | if (N_MEMORY == N_NORMAL_MEMORY) |
d9713679 LJ |
858 | zone_last = ZONE_MOVABLE; |
859 | ||
860 | /* | |
861 | * if the memory to be online is in a zone of 0...zone_last, and | |
862 | * the zones of 0...zone_last don't have memory before online, we will | |
863 | * need to set the node to node_states[N_NORMAL_MEMORY] after | |
864 | * the memory is online. | |
865 | */ | |
866 | if (zone_idx(zone) <= zone_last && !node_state(nid, N_NORMAL_MEMORY)) | |
867 | arg->status_change_nid_normal = nid; | |
868 | else | |
869 | arg->status_change_nid_normal = -1; | |
870 | ||
6715ddf9 LJ |
871 | #ifdef CONFIG_HIGHMEM |
872 | /* | |
873 | * If we have movable node, node_states[N_HIGH_MEMORY] | |
874 | * contains nodes which have zones of 0...ZONE_HIGHMEM, | |
875 | * set zone_last to ZONE_HIGHMEM. | |
876 | * | |
877 | * If we don't have movable node, node_states[N_NORMAL_MEMORY] | |
878 | * contains nodes which have zones of 0...ZONE_MOVABLE, | |
879 | * set zone_last to ZONE_MOVABLE. | |
880 | */ | |
881 | zone_last = ZONE_HIGHMEM; | |
882 | if (N_MEMORY == N_HIGH_MEMORY) | |
883 | zone_last = ZONE_MOVABLE; | |
884 | ||
885 | if (zone_idx(zone) <= zone_last && !node_state(nid, N_HIGH_MEMORY)) | |
886 | arg->status_change_nid_high = nid; | |
887 | else | |
888 | arg->status_change_nid_high = -1; | |
889 | #else | |
890 | arg->status_change_nid_high = arg->status_change_nid_normal; | |
891 | #endif | |
892 | ||
d9713679 LJ |
893 | /* |
894 | * if the node don't have memory befor online, we will need to | |
6715ddf9 | 895 | * set the node to node_states[N_MEMORY] after the memory |
d9713679 LJ |
896 | * is online. |
897 | */ | |
6715ddf9 | 898 | if (!node_state(nid, N_MEMORY)) |
d9713679 LJ |
899 | arg->status_change_nid = nid; |
900 | else | |
901 | arg->status_change_nid = -1; | |
902 | } | |
903 | ||
904 | static void node_states_set_node(int node, struct memory_notify *arg) | |
905 | { | |
906 | if (arg->status_change_nid_normal >= 0) | |
907 | node_set_state(node, N_NORMAL_MEMORY); | |
908 | ||
6715ddf9 LJ |
909 | if (arg->status_change_nid_high >= 0) |
910 | node_set_state(node, N_HIGH_MEMORY); | |
911 | ||
912 | node_set_state(node, N_MEMORY); | |
d9713679 LJ |
913 | } |
914 | ||
75884fb1 | 915 | |
511c2aba | 916 | int __ref online_pages(unsigned long pfn, unsigned long nr_pages, int online_type) |
75884fb1 | 917 | { |
3947be19 DH |
918 | unsigned long onlined_pages = 0; |
919 | struct zone *zone; | |
6811378e | 920 | int need_zonelists_rebuild = 0; |
7b78d335 YG |
921 | int nid; |
922 | int ret; | |
923 | struct memory_notify arg; | |
924 | ||
925268a0 | 925 | lock_memory_hotplug(); |
d9713679 LJ |
926 | /* |
927 | * This doesn't need a lock to do pfn_to_page(). | |
928 | * The section can't be removed here because of the | |
929 | * memory_block->state_mutex. | |
930 | */ | |
931 | zone = page_zone(pfn_to_page(pfn)); | |
932 | ||
74d42d8f LJ |
933 | if ((zone_idx(zone) > ZONE_NORMAL || online_type == ONLINE_MOVABLE) && |
934 | !can_online_high_movable(zone)) { | |
935 | unlock_memory_hotplug(); | |
936 | return -1; | |
937 | } | |
938 | ||
511c2aba LJ |
939 | if (online_type == ONLINE_KERNEL && zone_idx(zone) == ZONE_MOVABLE) { |
940 | if (move_pfn_range_left(zone - 1, zone, pfn, pfn + nr_pages)) { | |
941 | unlock_memory_hotplug(); | |
942 | return -1; | |
943 | } | |
944 | } | |
945 | if (online_type == ONLINE_MOVABLE && zone_idx(zone) == ZONE_MOVABLE - 1) { | |
946 | if (move_pfn_range_right(zone, zone + 1, pfn, pfn + nr_pages)) { | |
947 | unlock_memory_hotplug(); | |
948 | return -1; | |
949 | } | |
950 | } | |
951 | ||
952 | /* Previous code may changed the zone of the pfn range */ | |
953 | zone = page_zone(pfn_to_page(pfn)); | |
954 | ||
7b78d335 YG |
955 | arg.start_pfn = pfn; |
956 | arg.nr_pages = nr_pages; | |
d9713679 | 957 | node_states_check_changes_online(nr_pages, zone, &arg); |
7b78d335 YG |
958 | |
959 | nid = page_to_nid(pfn_to_page(pfn)); | |
3947be19 | 960 | |
7b78d335 YG |
961 | ret = memory_notify(MEM_GOING_ONLINE, &arg); |
962 | ret = notifier_to_errno(ret); | |
963 | if (ret) { | |
964 | memory_notify(MEM_CANCEL_ONLINE, &arg); | |
925268a0 | 965 | unlock_memory_hotplug(); |
7b78d335 YG |
966 | return ret; |
967 | } | |
6811378e YG |
968 | /* |
969 | * If this zone is not populated, then it is not in zonelist. | |
970 | * This means the page allocator ignores this zone. | |
971 | * So, zonelist must be updated after online. | |
972 | */ | |
4eaf3f64 | 973 | mutex_lock(&zonelists_mutex); |
6dcd73d7 | 974 | if (!populated_zone(zone)) { |
6811378e | 975 | need_zonelists_rebuild = 1; |
6dcd73d7 WC |
976 | build_all_zonelists(NULL, zone); |
977 | } | |
6811378e | 978 | |
908eedc6 | 979 | ret = walk_system_ram_range(pfn, nr_pages, &onlined_pages, |
75884fb1 | 980 | online_pages_range); |
fd8a4221 | 981 | if (ret) { |
6dcd73d7 WC |
982 | if (need_zonelists_rebuild) |
983 | zone_pcp_reset(zone); | |
4eaf3f64 | 984 | mutex_unlock(&zonelists_mutex); |
a62e2f4f BH |
985 | printk(KERN_DEBUG "online_pages [mem %#010llx-%#010llx] failed\n", |
986 | (unsigned long long) pfn << PAGE_SHIFT, | |
987 | (((unsigned long long) pfn + nr_pages) | |
988 | << PAGE_SHIFT) - 1); | |
fd8a4221 | 989 | memory_notify(MEM_CANCEL_ONLINE, &arg); |
925268a0 | 990 | unlock_memory_hotplug(); |
fd8a4221 GL |
991 | return ret; |
992 | } | |
993 | ||
9feedc9d | 994 | zone->managed_pages += onlined_pages; |
3947be19 | 995 | zone->present_pages += onlined_pages; |
f2937be5 | 996 | zone->zone_pgdat->node_present_pages += onlined_pages; |
08dff7b7 | 997 | if (onlined_pages) { |
d9713679 | 998 | node_states_set_node(zone_to_nid(zone), &arg); |
08dff7b7 | 999 | if (need_zonelists_rebuild) |
6dcd73d7 | 1000 | build_all_zonelists(NULL, NULL); |
08dff7b7 JL |
1001 | else |
1002 | zone_pcp_update(zone); | |
1003 | } | |
3947be19 | 1004 | |
4eaf3f64 | 1005 | mutex_unlock(&zonelists_mutex); |
1b79acc9 KM |
1006 | |
1007 | init_per_zone_wmark_min(); | |
1008 | ||
08dff7b7 | 1009 | if (onlined_pages) |
7ea1530a | 1010 | kswapd_run(zone_to_nid(zone)); |
61b13993 | 1011 | |
1f522509 | 1012 | vm_total_pages = nr_free_pagecache_pages(); |
2f7f24ec | 1013 | |
2d1d43f6 | 1014 | writeback_set_ratelimit(); |
7b78d335 YG |
1015 | |
1016 | if (onlined_pages) | |
1017 | memory_notify(MEM_ONLINE, &arg); | |
925268a0 | 1018 | unlock_memory_hotplug(); |
7b78d335 | 1019 | |
3947be19 DH |
1020 | return 0; |
1021 | } | |
53947027 | 1022 | #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */ |
bc02af93 | 1023 | |
e1319331 HS |
1024 | /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */ |
1025 | static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start) | |
9af3c2de YG |
1026 | { |
1027 | struct pglist_data *pgdat; | |
1028 | unsigned long zones_size[MAX_NR_ZONES] = {0}; | |
1029 | unsigned long zholes_size[MAX_NR_ZONES] = {0}; | |
1030 | unsigned long start_pfn = start >> PAGE_SHIFT; | |
1031 | ||
a1e565aa TC |
1032 | pgdat = NODE_DATA(nid); |
1033 | if (!pgdat) { | |
1034 | pgdat = arch_alloc_nodedata(nid); | |
1035 | if (!pgdat) | |
1036 | return NULL; | |
9af3c2de | 1037 | |
a1e565aa TC |
1038 | arch_refresh_nodedata(nid, pgdat); |
1039 | } | |
9af3c2de YG |
1040 | |
1041 | /* we can use NODE_DATA(nid) from here */ | |
1042 | ||
1043 | /* init node's zones as empty zones, we don't have any present pages.*/ | |
9109fb7b | 1044 | free_area_init_node(nid, zones_size, start_pfn, zholes_size); |
9af3c2de | 1045 | |
959ecc48 KH |
1046 | /* |
1047 | * The node we allocated has no zone fallback lists. For avoiding | |
1048 | * to access not-initialized zonelist, build here. | |
1049 | */ | |
f957db4f | 1050 | mutex_lock(&zonelists_mutex); |
9adb62a5 | 1051 | build_all_zonelists(pgdat, NULL); |
f957db4f | 1052 | mutex_unlock(&zonelists_mutex); |
959ecc48 | 1053 | |
9af3c2de YG |
1054 | return pgdat; |
1055 | } | |
1056 | ||
1057 | static void rollback_node_hotadd(int nid, pg_data_t *pgdat) | |
1058 | { | |
1059 | arch_refresh_nodedata(nid, NULL); | |
1060 | arch_free_nodedata(pgdat); | |
1061 | return; | |
1062 | } | |
1063 | ||
0a547039 | 1064 | |
cf23422b | 1065 | /* |
1066 | * called by cpu_up() to online a node without onlined memory. | |
1067 | */ | |
1068 | int mem_online_node(int nid) | |
1069 | { | |
1070 | pg_data_t *pgdat; | |
1071 | int ret; | |
1072 | ||
20d6c96b | 1073 | lock_memory_hotplug(); |
cf23422b | 1074 | pgdat = hotadd_new_pgdat(nid, 0); |
7553e8f2 | 1075 | if (!pgdat) { |
cf23422b | 1076 | ret = -ENOMEM; |
1077 | goto out; | |
1078 | } | |
1079 | node_set_online(nid); | |
1080 | ret = register_one_node(nid); | |
1081 | BUG_ON(ret); | |
1082 | ||
1083 | out: | |
20d6c96b | 1084 | unlock_memory_hotplug(); |
cf23422b | 1085 | return ret; |
1086 | } | |
1087 | ||
31168481 AV |
1088 | /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */ |
1089 | int __ref add_memory(int nid, u64 start, u64 size) | |
bc02af93 | 1090 | { |
9af3c2de | 1091 | pg_data_t *pgdat = NULL; |
a1e565aa TC |
1092 | bool new_pgdat; |
1093 | bool new_node; | |
ebd15302 | 1094 | struct resource *res; |
bc02af93 YG |
1095 | int ret; |
1096 | ||
20d6c96b | 1097 | lock_memory_hotplug(); |
6ad696d2 | 1098 | |
ebd15302 | 1099 | res = register_memory_resource(start, size); |
6ad696d2 | 1100 | ret = -EEXIST; |
ebd15302 | 1101 | if (!res) |
6ad696d2 | 1102 | goto out; |
ebd15302 | 1103 | |
a1e565aa TC |
1104 | { /* Stupid hack to suppress address-never-null warning */ |
1105 | void *p = NODE_DATA(nid); | |
1106 | new_pgdat = !p; | |
1107 | } | |
1108 | new_node = !node_online(nid); | |
1109 | if (new_node) { | |
9af3c2de | 1110 | pgdat = hotadd_new_pgdat(nid, start); |
6ad696d2 | 1111 | ret = -ENOMEM; |
9af3c2de | 1112 | if (!pgdat) |
41b9e2d7 | 1113 | goto error; |
9af3c2de YG |
1114 | } |
1115 | ||
bc02af93 YG |
1116 | /* call arch's memory hotadd */ |
1117 | ret = arch_add_memory(nid, start, size); | |
1118 | ||
9af3c2de YG |
1119 | if (ret < 0) |
1120 | goto error; | |
1121 | ||
0fc44159 | 1122 | /* we online node here. we can't roll back from here. */ |
9af3c2de YG |
1123 | node_set_online(nid); |
1124 | ||
a1e565aa | 1125 | if (new_node) { |
0fc44159 YG |
1126 | ret = register_one_node(nid); |
1127 | /* | |
1128 | * If sysfs file of new node can't create, cpu on the node | |
1129 | * can't be hot-added. There is no rollback way now. | |
1130 | * So, check by BUG_ON() to catch it reluctantly.. | |
1131 | */ | |
1132 | BUG_ON(ret); | |
1133 | } | |
1134 | ||
d96ae530 | 1135 | /* create new memmap entry */ |
1136 | firmware_map_add_hotplug(start, start + size, "System RAM"); | |
1137 | ||
6ad696d2 AK |
1138 | goto out; |
1139 | ||
9af3c2de YG |
1140 | error: |
1141 | /* rollback pgdat allocation and others */ | |
1142 | if (new_pgdat) | |
1143 | rollback_node_hotadd(nid, pgdat); | |
a864b9d0 | 1144 | release_memory_resource(res); |
9af3c2de | 1145 | |
6ad696d2 | 1146 | out: |
20d6c96b | 1147 | unlock_memory_hotplug(); |
bc02af93 YG |
1148 | return ret; |
1149 | } | |
1150 | EXPORT_SYMBOL_GPL(add_memory); | |
0c0e6195 KH |
1151 | |
1152 | #ifdef CONFIG_MEMORY_HOTREMOVE | |
5c755e9f BP |
1153 | /* |
1154 | * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy | |
1155 | * set and the size of the free page is given by page_order(). Using this, | |
1156 | * the function determines if the pageblock contains only free pages. | |
1157 | * Due to buddy contraints, a free page at least the size of a pageblock will | |
1158 | * be located at the start of the pageblock | |
1159 | */ | |
1160 | static inline int pageblock_free(struct page *page) | |
1161 | { | |
1162 | return PageBuddy(page) && page_order(page) >= pageblock_order; | |
1163 | } | |
1164 | ||
1165 | /* Return the start of the next active pageblock after a given page */ | |
1166 | static struct page *next_active_pageblock(struct page *page) | |
1167 | { | |
5c755e9f BP |
1168 | /* Ensure the starting page is pageblock-aligned */ |
1169 | BUG_ON(page_to_pfn(page) & (pageblock_nr_pages - 1)); | |
1170 | ||
5c755e9f | 1171 | /* If the entire pageblock is free, move to the end of free page */ |
0dcc48c1 KH |
1172 | if (pageblock_free(page)) { |
1173 | int order; | |
1174 | /* be careful. we don't have locks, page_order can be changed.*/ | |
1175 | order = page_order(page); | |
1176 | if ((order < MAX_ORDER) && (order >= pageblock_order)) | |
1177 | return page + (1 << order); | |
1178 | } | |
5c755e9f | 1179 | |
0dcc48c1 | 1180 | return page + pageblock_nr_pages; |
5c755e9f BP |
1181 | } |
1182 | ||
1183 | /* Checks if this range of memory is likely to be hot-removable. */ | |
1184 | int is_mem_section_removable(unsigned long start_pfn, unsigned long nr_pages) | |
1185 | { | |
5c755e9f BP |
1186 | struct page *page = pfn_to_page(start_pfn); |
1187 | struct page *end_page = page + nr_pages; | |
1188 | ||
1189 | /* Check the starting page of each pageblock within the range */ | |
1190 | for (; page < end_page; page = next_active_pageblock(page)) { | |
49ac8255 | 1191 | if (!is_pageblock_removable_nolock(page)) |
5c755e9f | 1192 | return 0; |
49ac8255 | 1193 | cond_resched(); |
5c755e9f BP |
1194 | } |
1195 | ||
1196 | /* All pageblocks in the memory block are likely to be hot-removable */ | |
1197 | return 1; | |
1198 | } | |
1199 | ||
0c0e6195 KH |
1200 | /* |
1201 | * Confirm all pages in a range [start, end) is belongs to the same zone. | |
1202 | */ | |
1203 | static int test_pages_in_a_zone(unsigned long start_pfn, unsigned long end_pfn) | |
1204 | { | |
1205 | unsigned long pfn; | |
1206 | struct zone *zone = NULL; | |
1207 | struct page *page; | |
1208 | int i; | |
1209 | for (pfn = start_pfn; | |
1210 | pfn < end_pfn; | |
1211 | pfn += MAX_ORDER_NR_PAGES) { | |
1212 | i = 0; | |
1213 | /* This is just a CONFIG_HOLES_IN_ZONE check.*/ | |
1214 | while ((i < MAX_ORDER_NR_PAGES) && !pfn_valid_within(pfn + i)) | |
1215 | i++; | |
1216 | if (i == MAX_ORDER_NR_PAGES) | |
1217 | continue; | |
1218 | page = pfn_to_page(pfn + i); | |
1219 | if (zone && page_zone(page) != zone) | |
1220 | return 0; | |
1221 | zone = page_zone(page); | |
1222 | } | |
1223 | return 1; | |
1224 | } | |
1225 | ||
1226 | /* | |
1227 | * Scanning pfn is much easier than scanning lru list. | |
1228 | * Scan pfn from start to end and Find LRU page. | |
1229 | */ | |
7bbc0905 | 1230 | static unsigned long scan_lru_pages(unsigned long start, unsigned long end) |
0c0e6195 KH |
1231 | { |
1232 | unsigned long pfn; | |
1233 | struct page *page; | |
1234 | for (pfn = start; pfn < end; pfn++) { | |
1235 | if (pfn_valid(pfn)) { | |
1236 | page = pfn_to_page(pfn); | |
1237 | if (PageLRU(page)) | |
1238 | return pfn; | |
1239 | } | |
1240 | } | |
1241 | return 0; | |
1242 | } | |
1243 | ||
0c0e6195 KH |
1244 | #define NR_OFFLINE_AT_ONCE_PAGES (256) |
1245 | static int | |
1246 | do_migrate_range(unsigned long start_pfn, unsigned long end_pfn) | |
1247 | { | |
1248 | unsigned long pfn; | |
1249 | struct page *page; | |
1250 | int move_pages = NR_OFFLINE_AT_ONCE_PAGES; | |
1251 | int not_managed = 0; | |
1252 | int ret = 0; | |
1253 | LIST_HEAD(source); | |
1254 | ||
1255 | for (pfn = start_pfn; pfn < end_pfn && move_pages > 0; pfn++) { | |
1256 | if (!pfn_valid(pfn)) | |
1257 | continue; | |
1258 | page = pfn_to_page(pfn); | |
700c2a46 | 1259 | if (!get_page_unless_zero(page)) |
0c0e6195 KH |
1260 | continue; |
1261 | /* | |
1262 | * We can skip free pages. And we can only deal with pages on | |
1263 | * LRU. | |
1264 | */ | |
62695a84 | 1265 | ret = isolate_lru_page(page); |
0c0e6195 | 1266 | if (!ret) { /* Success */ |
700c2a46 | 1267 | put_page(page); |
62695a84 | 1268 | list_add_tail(&page->lru, &source); |
0c0e6195 | 1269 | move_pages--; |
6d9c285a KM |
1270 | inc_zone_page_state(page, NR_ISOLATED_ANON + |
1271 | page_is_file_cache(page)); | |
1272 | ||
0c0e6195 | 1273 | } else { |
0c0e6195 | 1274 | #ifdef CONFIG_DEBUG_VM |
718a3821 WF |
1275 | printk(KERN_ALERT "removing pfn %lx from LRU failed\n", |
1276 | pfn); | |
1277 | dump_page(page); | |
0c0e6195 | 1278 | #endif |
700c2a46 | 1279 | put_page(page); |
25985edc | 1280 | /* Because we don't have big zone->lock. we should |
809c4449 BL |
1281 | check this again here. */ |
1282 | if (page_count(page)) { | |
1283 | not_managed++; | |
f3ab2636 | 1284 | ret = -EBUSY; |
809c4449 BL |
1285 | break; |
1286 | } | |
0c0e6195 KH |
1287 | } |
1288 | } | |
f3ab2636 BL |
1289 | if (!list_empty(&source)) { |
1290 | if (not_managed) { | |
1291 | putback_lru_pages(&source); | |
1292 | goto out; | |
1293 | } | |
74c08f98 MK |
1294 | |
1295 | /* | |
1296 | * alloc_migrate_target should be improooooved!! | |
1297 | * migrate_pages returns # of failed pages. | |
1298 | */ | |
1299 | ret = migrate_pages(&source, alloc_migrate_target, 0, | |
9c620e2b | 1300 | MIGRATE_SYNC, MR_MEMORY_HOTPLUG); |
f3ab2636 | 1301 | if (ret) |
0c0e6195 | 1302 | putback_lru_pages(&source); |
0c0e6195 | 1303 | } |
0c0e6195 KH |
1304 | out: |
1305 | return ret; | |
1306 | } | |
1307 | ||
1308 | /* | |
1309 | * remove from free_area[] and mark all as Reserved. | |
1310 | */ | |
1311 | static int | |
1312 | offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages, | |
1313 | void *data) | |
1314 | { | |
1315 | __offline_isolated_pages(start, start + nr_pages); | |
1316 | return 0; | |
1317 | } | |
1318 | ||
1319 | static void | |
1320 | offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn) | |
1321 | { | |
908eedc6 | 1322 | walk_system_ram_range(start_pfn, end_pfn - start_pfn, NULL, |
0c0e6195 KH |
1323 | offline_isolated_pages_cb); |
1324 | } | |
1325 | ||
1326 | /* | |
1327 | * Check all pages in range, recoreded as memory resource, are isolated. | |
1328 | */ | |
1329 | static int | |
1330 | check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages, | |
1331 | void *data) | |
1332 | { | |
1333 | int ret; | |
1334 | long offlined = *(long *)data; | |
b023f468 | 1335 | ret = test_pages_isolated(start_pfn, start_pfn + nr_pages, true); |
0c0e6195 KH |
1336 | offlined = nr_pages; |
1337 | if (!ret) | |
1338 | *(long *)data += offlined; | |
1339 | return ret; | |
1340 | } | |
1341 | ||
1342 | static long | |
1343 | check_pages_isolated(unsigned long start_pfn, unsigned long end_pfn) | |
1344 | { | |
1345 | long offlined = 0; | |
1346 | int ret; | |
1347 | ||
908eedc6 | 1348 | ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn, &offlined, |
0c0e6195 KH |
1349 | check_pages_isolated_cb); |
1350 | if (ret < 0) | |
1351 | offlined = (long)ret; | |
1352 | return offlined; | |
1353 | } | |
1354 | ||
09285af7 | 1355 | #ifdef CONFIG_MOVABLE_NODE |
79a4dcef TC |
1356 | /* |
1357 | * When CONFIG_MOVABLE_NODE, we permit offlining of a node which doesn't have | |
1358 | * normal memory. | |
1359 | */ | |
09285af7 LJ |
1360 | static bool can_offline_normal(struct zone *zone, unsigned long nr_pages) |
1361 | { | |
1362 | return true; | |
1363 | } | |
79a4dcef | 1364 | #else /* CONFIG_MOVABLE_NODE */ |
74d42d8f LJ |
1365 | /* ensure the node has NORMAL memory if it is still online */ |
1366 | static bool can_offline_normal(struct zone *zone, unsigned long nr_pages) | |
1367 | { | |
1368 | struct pglist_data *pgdat = zone->zone_pgdat; | |
1369 | unsigned long present_pages = 0; | |
1370 | enum zone_type zt; | |
1371 | ||
1372 | for (zt = 0; zt <= ZONE_NORMAL; zt++) | |
1373 | present_pages += pgdat->node_zones[zt].present_pages; | |
1374 | ||
1375 | if (present_pages > nr_pages) | |
1376 | return true; | |
1377 | ||
1378 | present_pages = 0; | |
1379 | for (; zt <= ZONE_MOVABLE; zt++) | |
1380 | present_pages += pgdat->node_zones[zt].present_pages; | |
1381 | ||
1382 | /* | |
1383 | * we can't offline the last normal memory until all | |
1384 | * higher memory is offlined. | |
1385 | */ | |
1386 | return present_pages == 0; | |
1387 | } | |
79a4dcef | 1388 | #endif /* CONFIG_MOVABLE_NODE */ |
74d42d8f | 1389 | |
d9713679 LJ |
1390 | /* check which state of node_states will be changed when offline memory */ |
1391 | static void node_states_check_changes_offline(unsigned long nr_pages, | |
1392 | struct zone *zone, struct memory_notify *arg) | |
1393 | { | |
1394 | struct pglist_data *pgdat = zone->zone_pgdat; | |
1395 | unsigned long present_pages = 0; | |
1396 | enum zone_type zt, zone_last = ZONE_NORMAL; | |
1397 | ||
1398 | /* | |
6715ddf9 LJ |
1399 | * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY] |
1400 | * contains nodes which have zones of 0...ZONE_NORMAL, | |
1401 | * set zone_last to ZONE_NORMAL. | |
d9713679 | 1402 | * |
6715ddf9 LJ |
1403 | * If we don't have HIGHMEM nor movable node, |
1404 | * node_states[N_NORMAL_MEMORY] contains nodes which have zones of | |
1405 | * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE. | |
d9713679 | 1406 | */ |
6715ddf9 | 1407 | if (N_MEMORY == N_NORMAL_MEMORY) |
d9713679 LJ |
1408 | zone_last = ZONE_MOVABLE; |
1409 | ||
1410 | /* | |
1411 | * check whether node_states[N_NORMAL_MEMORY] will be changed. | |
1412 | * If the memory to be offline is in a zone of 0...zone_last, | |
1413 | * and it is the last present memory, 0...zone_last will | |
1414 | * become empty after offline , thus we can determind we will | |
1415 | * need to clear the node from node_states[N_NORMAL_MEMORY]. | |
1416 | */ | |
1417 | for (zt = 0; zt <= zone_last; zt++) | |
1418 | present_pages += pgdat->node_zones[zt].present_pages; | |
1419 | if (zone_idx(zone) <= zone_last && nr_pages >= present_pages) | |
1420 | arg->status_change_nid_normal = zone_to_nid(zone); | |
1421 | else | |
1422 | arg->status_change_nid_normal = -1; | |
1423 | ||
6715ddf9 LJ |
1424 | #ifdef CONFIG_HIGHMEM |
1425 | /* | |
1426 | * If we have movable node, node_states[N_HIGH_MEMORY] | |
1427 | * contains nodes which have zones of 0...ZONE_HIGHMEM, | |
1428 | * set zone_last to ZONE_HIGHMEM. | |
1429 | * | |
1430 | * If we don't have movable node, node_states[N_NORMAL_MEMORY] | |
1431 | * contains nodes which have zones of 0...ZONE_MOVABLE, | |
1432 | * set zone_last to ZONE_MOVABLE. | |
1433 | */ | |
1434 | zone_last = ZONE_HIGHMEM; | |
1435 | if (N_MEMORY == N_HIGH_MEMORY) | |
1436 | zone_last = ZONE_MOVABLE; | |
1437 | ||
1438 | for (; zt <= zone_last; zt++) | |
1439 | present_pages += pgdat->node_zones[zt].present_pages; | |
1440 | if (zone_idx(zone) <= zone_last && nr_pages >= present_pages) | |
1441 | arg->status_change_nid_high = zone_to_nid(zone); | |
1442 | else | |
1443 | arg->status_change_nid_high = -1; | |
1444 | #else | |
1445 | arg->status_change_nid_high = arg->status_change_nid_normal; | |
1446 | #endif | |
1447 | ||
d9713679 LJ |
1448 | /* |
1449 | * node_states[N_HIGH_MEMORY] contains nodes which have 0...ZONE_MOVABLE | |
1450 | */ | |
1451 | zone_last = ZONE_MOVABLE; | |
1452 | ||
1453 | /* | |
1454 | * check whether node_states[N_HIGH_MEMORY] will be changed | |
1455 | * If we try to offline the last present @nr_pages from the node, | |
1456 | * we can determind we will need to clear the node from | |
1457 | * node_states[N_HIGH_MEMORY]. | |
1458 | */ | |
1459 | for (; zt <= zone_last; zt++) | |
1460 | present_pages += pgdat->node_zones[zt].present_pages; | |
1461 | if (nr_pages >= present_pages) | |
1462 | arg->status_change_nid = zone_to_nid(zone); | |
1463 | else | |
1464 | arg->status_change_nid = -1; | |
1465 | } | |
1466 | ||
1467 | static void node_states_clear_node(int node, struct memory_notify *arg) | |
1468 | { | |
1469 | if (arg->status_change_nid_normal >= 0) | |
1470 | node_clear_state(node, N_NORMAL_MEMORY); | |
1471 | ||
6715ddf9 LJ |
1472 | if ((N_MEMORY != N_NORMAL_MEMORY) && |
1473 | (arg->status_change_nid_high >= 0)) | |
d9713679 | 1474 | node_clear_state(node, N_HIGH_MEMORY); |
6715ddf9 LJ |
1475 | |
1476 | if ((N_MEMORY != N_HIGH_MEMORY) && | |
1477 | (arg->status_change_nid >= 0)) | |
1478 | node_clear_state(node, N_MEMORY); | |
d9713679 LJ |
1479 | } |
1480 | ||
a16cee10 | 1481 | static int __ref __offline_pages(unsigned long start_pfn, |
0c0e6195 KH |
1482 | unsigned long end_pfn, unsigned long timeout) |
1483 | { | |
1484 | unsigned long pfn, nr_pages, expire; | |
1485 | long offlined_pages; | |
7b78d335 | 1486 | int ret, drain, retry_max, node; |
0c0e6195 | 1487 | struct zone *zone; |
7b78d335 | 1488 | struct memory_notify arg; |
0c0e6195 KH |
1489 | |
1490 | BUG_ON(start_pfn >= end_pfn); | |
1491 | /* at least, alignment against pageblock is necessary */ | |
1492 | if (!IS_ALIGNED(start_pfn, pageblock_nr_pages)) | |
1493 | return -EINVAL; | |
1494 | if (!IS_ALIGNED(end_pfn, pageblock_nr_pages)) | |
1495 | return -EINVAL; | |
1496 | /* This makes hotplug much easier...and readable. | |
1497 | we assume this for now. .*/ | |
1498 | if (!test_pages_in_a_zone(start_pfn, end_pfn)) | |
1499 | return -EINVAL; | |
7b78d335 | 1500 | |
20d6c96b | 1501 | lock_memory_hotplug(); |
6ad696d2 | 1502 | |
7b78d335 YG |
1503 | zone = page_zone(pfn_to_page(start_pfn)); |
1504 | node = zone_to_nid(zone); | |
1505 | nr_pages = end_pfn - start_pfn; | |
1506 | ||
74d42d8f LJ |
1507 | ret = -EINVAL; |
1508 | if (zone_idx(zone) <= ZONE_NORMAL && !can_offline_normal(zone, nr_pages)) | |
1509 | goto out; | |
1510 | ||
0c0e6195 | 1511 | /* set above range as isolated */ |
b023f468 WC |
1512 | ret = start_isolate_page_range(start_pfn, end_pfn, |
1513 | MIGRATE_MOVABLE, true); | |
0c0e6195 | 1514 | if (ret) |
6ad696d2 | 1515 | goto out; |
7b78d335 YG |
1516 | |
1517 | arg.start_pfn = start_pfn; | |
1518 | arg.nr_pages = nr_pages; | |
d9713679 | 1519 | node_states_check_changes_offline(nr_pages, zone, &arg); |
7b78d335 YG |
1520 | |
1521 | ret = memory_notify(MEM_GOING_OFFLINE, &arg); | |
1522 | ret = notifier_to_errno(ret); | |
1523 | if (ret) | |
1524 | goto failed_removal; | |
1525 | ||
0c0e6195 KH |
1526 | pfn = start_pfn; |
1527 | expire = jiffies + timeout; | |
1528 | drain = 0; | |
1529 | retry_max = 5; | |
1530 | repeat: | |
1531 | /* start memory hot removal */ | |
1532 | ret = -EAGAIN; | |
1533 | if (time_after(jiffies, expire)) | |
1534 | goto failed_removal; | |
1535 | ret = -EINTR; | |
1536 | if (signal_pending(current)) | |
1537 | goto failed_removal; | |
1538 | ret = 0; | |
1539 | if (drain) { | |
1540 | lru_add_drain_all(); | |
0c0e6195 | 1541 | cond_resched(); |
9f8f2172 | 1542 | drain_all_pages(); |
0c0e6195 KH |
1543 | } |
1544 | ||
1545 | pfn = scan_lru_pages(start_pfn, end_pfn); | |
1546 | if (pfn) { /* We have page on LRU */ | |
1547 | ret = do_migrate_range(pfn, end_pfn); | |
1548 | if (!ret) { | |
1549 | drain = 1; | |
1550 | goto repeat; | |
1551 | } else { | |
1552 | if (ret < 0) | |
1553 | if (--retry_max == 0) | |
1554 | goto failed_removal; | |
1555 | yield(); | |
1556 | drain = 1; | |
1557 | goto repeat; | |
1558 | } | |
1559 | } | |
b3834be5 | 1560 | /* drain all zone's lru pagevec, this is asynchronous... */ |
0c0e6195 | 1561 | lru_add_drain_all(); |
0c0e6195 | 1562 | yield(); |
b3834be5 | 1563 | /* drain pcp pages, this is synchronous. */ |
9f8f2172 | 1564 | drain_all_pages(); |
0c0e6195 KH |
1565 | /* check again */ |
1566 | offlined_pages = check_pages_isolated(start_pfn, end_pfn); | |
1567 | if (offlined_pages < 0) { | |
1568 | ret = -EBUSY; | |
1569 | goto failed_removal; | |
1570 | } | |
1571 | printk(KERN_INFO "Offlined Pages %ld\n", offlined_pages); | |
b3834be5 | 1572 | /* Ok, all of our target is isolated. |
0c0e6195 KH |
1573 | We cannot do rollback at this point. */ |
1574 | offline_isolated_pages(start_pfn, end_pfn); | |
dbc0e4ce | 1575 | /* reset pagetype flags and makes migrate type to be MOVABLE */ |
0815f3d8 | 1576 | undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE); |
0c0e6195 | 1577 | /* removal success */ |
9feedc9d | 1578 | zone->managed_pages -= offlined_pages; |
0c0e6195 KH |
1579 | zone->present_pages -= offlined_pages; |
1580 | zone->zone_pgdat->node_present_pages -= offlined_pages; | |
1581 | totalram_pages -= offlined_pages; | |
7b78d335 | 1582 | |
1b79acc9 KM |
1583 | init_per_zone_wmark_min(); |
1584 | ||
1e8537ba | 1585 | if (!populated_zone(zone)) { |
340175b7 | 1586 | zone_pcp_reset(zone); |
1e8537ba XQ |
1587 | mutex_lock(&zonelists_mutex); |
1588 | build_all_zonelists(NULL, NULL); | |
1589 | mutex_unlock(&zonelists_mutex); | |
1590 | } else | |
1591 | zone_pcp_update(zone); | |
340175b7 | 1592 | |
d9713679 LJ |
1593 | node_states_clear_node(node, &arg); |
1594 | if (arg.status_change_nid >= 0) | |
8fe23e05 | 1595 | kswapd_stop(node); |
bce7394a | 1596 | |
0c0e6195 KH |
1597 | vm_total_pages = nr_free_pagecache_pages(); |
1598 | writeback_set_ratelimit(); | |
7b78d335 YG |
1599 | |
1600 | memory_notify(MEM_OFFLINE, &arg); | |
20d6c96b | 1601 | unlock_memory_hotplug(); |
0c0e6195 KH |
1602 | return 0; |
1603 | ||
1604 | failed_removal: | |
a62e2f4f BH |
1605 | printk(KERN_INFO "memory offlining [mem %#010llx-%#010llx] failed\n", |
1606 | (unsigned long long) start_pfn << PAGE_SHIFT, | |
1607 | ((unsigned long long) end_pfn << PAGE_SHIFT) - 1); | |
7b78d335 | 1608 | memory_notify(MEM_CANCEL_OFFLINE, &arg); |
0c0e6195 | 1609 | /* pushback to free area */ |
0815f3d8 | 1610 | undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE); |
7b78d335 | 1611 | |
6ad696d2 | 1612 | out: |
20d6c96b | 1613 | unlock_memory_hotplug(); |
0c0e6195 KH |
1614 | return ret; |
1615 | } | |
71088785 | 1616 | |
a16cee10 WC |
1617 | int offline_pages(unsigned long start_pfn, unsigned long nr_pages) |
1618 | { | |
1619 | return __offline_pages(start_pfn, start_pfn + nr_pages, 120 * HZ); | |
1620 | } | |
1621 | ||
bbc76be6 WC |
1622 | /** |
1623 | * walk_memory_range - walks through all mem sections in [start_pfn, end_pfn) | |
1624 | * @start_pfn: start pfn of the memory range | |
1625 | * @end_pfn: end pft of the memory range | |
1626 | * @arg: argument passed to func | |
1627 | * @func: callback for each memory section walked | |
1628 | * | |
1629 | * This function walks through all present mem sections in range | |
1630 | * [start_pfn, end_pfn) and call func on each mem section. | |
1631 | * | |
1632 | * Returns the return value of func. | |
1633 | */ | |
1634 | static int walk_memory_range(unsigned long start_pfn, unsigned long end_pfn, | |
1635 | void *arg, int (*func)(struct memory_block *, void *)) | |
71088785 | 1636 | { |
e90bdb7f WC |
1637 | struct memory_block *mem = NULL; |
1638 | struct mem_section *section; | |
e90bdb7f WC |
1639 | unsigned long pfn, section_nr; |
1640 | int ret; | |
e90bdb7f WC |
1641 | |
1642 | for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) { | |
1643 | section_nr = pfn_to_section_nr(pfn); | |
1644 | if (!present_section_nr(section_nr)) | |
1645 | continue; | |
1646 | ||
1647 | section = __nr_to_section(section_nr); | |
1648 | /* same memblock? */ | |
1649 | if (mem) | |
1650 | if ((section_nr >= mem->start_section_nr) && | |
1651 | (section_nr <= mem->end_section_nr)) | |
1652 | continue; | |
1653 | ||
1654 | mem = find_memory_block_hinted(section, mem); | |
1655 | if (!mem) | |
1656 | continue; | |
1657 | ||
bbc76be6 | 1658 | ret = func(mem, arg); |
e90bdb7f | 1659 | if (ret) { |
bbc76be6 WC |
1660 | kobject_put(&mem->dev.kobj); |
1661 | return ret; | |
e90bdb7f WC |
1662 | } |
1663 | } | |
1664 | ||
1665 | if (mem) | |
1666 | kobject_put(&mem->dev.kobj); | |
1667 | ||
bbc76be6 WC |
1668 | return 0; |
1669 | } | |
1670 | ||
1671 | /** | |
1672 | * offline_memory_block_cb - callback function for offlining memory block | |
1673 | * @mem: the memory block to be offlined | |
1674 | * @arg: buffer to hold error msg | |
1675 | * | |
1676 | * Always return 0, and put the error msg in arg if any. | |
1677 | */ | |
1678 | static int offline_memory_block_cb(struct memory_block *mem, void *arg) | |
1679 | { | |
1680 | int *ret = arg; | |
1681 | int error = offline_memory_block(mem); | |
1682 | ||
1683 | if (error != 0 && *ret == 0) | |
1684 | *ret = error; | |
1685 | ||
1686 | return 0; | |
1687 | } | |
1688 | ||
1689 | static int is_memblock_offlined_cb(struct memory_block *mem, void *arg) | |
1690 | { | |
1691 | int ret = !is_memblock_offlined(mem); | |
1692 | ||
1693 | if (unlikely(ret)) | |
1694 | pr_warn("removing memory fails, because memory " | |
1695 | "[%#010llx-%#010llx] is onlined\n", | |
1696 | PFN_PHYS(section_nr_to_pfn(mem->start_section_nr)), | |
1697 | PFN_PHYS(section_nr_to_pfn(mem->end_section_nr + 1))-1); | |
1698 | ||
1699 | return ret; | |
1700 | } | |
1701 | ||
60a5a19e TC |
1702 | static int check_cpu_on_node(void *data) |
1703 | { | |
1704 | struct pglist_data *pgdat = data; | |
1705 | int cpu; | |
1706 | ||
1707 | for_each_present_cpu(cpu) { | |
1708 | if (cpu_to_node(cpu) == pgdat->node_id) | |
1709 | /* | |
1710 | * the cpu on this node isn't removed, and we can't | |
1711 | * offline this node. | |
1712 | */ | |
1713 | return -EBUSY; | |
1714 | } | |
1715 | ||
1716 | return 0; | |
1717 | } | |
1718 | ||
e13fe869 WC |
1719 | static void unmap_cpu_on_node(void *data) |
1720 | { | |
1721 | #ifdef CONFIG_ACPI_NUMA | |
1722 | struct pglist_data *pgdat = data; | |
1723 | int cpu; | |
1724 | ||
1725 | for_each_possible_cpu(cpu) | |
1726 | if (cpu_to_node(cpu) == pgdat->node_id) | |
1727 | numa_clear_node(cpu); | |
1728 | #endif | |
1729 | } | |
1730 | ||
1731 | static int check_and_unmap_cpu_on_node(void *data) | |
1732 | { | |
1733 | int ret = check_cpu_on_node(data); | |
1734 | ||
1735 | if (ret) | |
1736 | return ret; | |
1737 | ||
1738 | /* | |
1739 | * the node will be offlined when we come here, so we can clear | |
1740 | * the cpu_to_node() now. | |
1741 | */ | |
1742 | ||
1743 | unmap_cpu_on_node(data); | |
1744 | return 0; | |
1745 | } | |
1746 | ||
60a5a19e | 1747 | /* offline the node if all memory sections of this node are removed */ |
90b30cdc | 1748 | void try_offline_node(int nid) |
60a5a19e | 1749 | { |
d822b86a WC |
1750 | pg_data_t *pgdat = NODE_DATA(nid); |
1751 | unsigned long start_pfn = pgdat->node_start_pfn; | |
1752 | unsigned long end_pfn = start_pfn + pgdat->node_spanned_pages; | |
60a5a19e | 1753 | unsigned long pfn; |
d822b86a WC |
1754 | struct page *pgdat_page = virt_to_page(pgdat); |
1755 | int i; | |
60a5a19e TC |
1756 | |
1757 | for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) { | |
1758 | unsigned long section_nr = pfn_to_section_nr(pfn); | |
1759 | ||
1760 | if (!present_section_nr(section_nr)) | |
1761 | continue; | |
1762 | ||
1763 | if (pfn_to_nid(pfn) != nid) | |
1764 | continue; | |
1765 | ||
1766 | /* | |
1767 | * some memory sections of this node are not removed, and we | |
1768 | * can't offline node now. | |
1769 | */ | |
1770 | return; | |
1771 | } | |
1772 | ||
e13fe869 | 1773 | if (stop_machine(check_and_unmap_cpu_on_node, pgdat, NULL)) |
60a5a19e TC |
1774 | return; |
1775 | ||
1776 | /* | |
1777 | * all memory/cpu of this node are removed, we can offline this | |
1778 | * node now. | |
1779 | */ | |
1780 | node_set_offline(nid); | |
1781 | unregister_one_node(nid); | |
d822b86a WC |
1782 | |
1783 | if (!PageSlab(pgdat_page) && !PageCompound(pgdat_page)) | |
1784 | /* node data is allocated from boot memory */ | |
1785 | return; | |
1786 | ||
1787 | /* free waittable in each zone */ | |
1788 | for (i = 0; i < MAX_NR_ZONES; i++) { | |
1789 | struct zone *zone = pgdat->node_zones + i; | |
1790 | ||
1791 | if (zone->wait_table) | |
1792 | vfree(zone->wait_table); | |
1793 | } | |
1794 | ||
1795 | /* | |
1796 | * Since there is no way to guarentee the address of pgdat/zone is not | |
1797 | * on stack of any kernel threads or used by other kernel objects | |
1798 | * without reference counting or other symchronizing method, do not | |
1799 | * reset node_data and free pgdat here. Just reset it to 0 and reuse | |
1800 | * the memory when the node is online again. | |
1801 | */ | |
1802 | memset(pgdat, 0, sizeof(*pgdat)); | |
60a5a19e | 1803 | } |
90b30cdc | 1804 | EXPORT_SYMBOL(try_offline_node); |
60a5a19e TC |
1805 | |
1806 | int __ref remove_memory(int nid, u64 start, u64 size) | |
bbc76be6 WC |
1807 | { |
1808 | unsigned long start_pfn, end_pfn; | |
1809 | int ret = 0; | |
1810 | int retry = 1; | |
1811 | ||
1812 | start_pfn = PFN_DOWN(start); | |
1813 | end_pfn = start_pfn + PFN_DOWN(size); | |
1814 | ||
1815 | /* | |
1816 | * When CONFIG_MEMCG is on, one memory block may be used by other | |
1817 | * blocks to store page cgroup when onlining pages. But we don't know | |
1818 | * in what order pages are onlined. So we iterate twice to offline | |
1819 | * memory: | |
1820 | * 1st iterate: offline every non primary memory block. | |
1821 | * 2nd iterate: offline primary (i.e. first added) memory block. | |
1822 | */ | |
1823 | repeat: | |
1824 | walk_memory_range(start_pfn, end_pfn, &ret, | |
1825 | offline_memory_block_cb); | |
1826 | if (ret) { | |
1827 | if (!retry) | |
1828 | return ret; | |
1829 | ||
1830 | retry = 0; | |
1831 | ret = 0; | |
993c1aad WC |
1832 | goto repeat; |
1833 | } | |
1834 | ||
6677e3ea YI |
1835 | lock_memory_hotplug(); |
1836 | ||
1837 | /* | |
1838 | * we have offlined all memory blocks like this: | |
1839 | * 1. lock memory hotplug | |
1840 | * 2. offline a memory block | |
1841 | * 3. unlock memory hotplug | |
1842 | * | |
1843 | * repeat step1-3 to offline the memory block. All memory blocks | |
1844 | * must be offlined before removing memory. But we don't hold the | |
1845 | * lock in the whole operation. So we should check whether all | |
1846 | * memory blocks are offlined. | |
1847 | */ | |
1848 | ||
bbc76be6 WC |
1849 | ret = walk_memory_range(start_pfn, end_pfn, NULL, |
1850 | is_memblock_offlined_cb); | |
1851 | if (ret) { | |
1852 | unlock_memory_hotplug(); | |
1853 | return ret; | |
6677e3ea YI |
1854 | } |
1855 | ||
46c66c4b YI |
1856 | /* remove memmap entry */ |
1857 | firmware_map_remove(start, start + size, "System RAM"); | |
1858 | ||
24d335ca WC |
1859 | arch_remove_memory(start, size); |
1860 | ||
60a5a19e TC |
1861 | try_offline_node(nid); |
1862 | ||
6677e3ea YI |
1863 | unlock_memory_hotplug(); |
1864 | ||
e90bdb7f | 1865 | return 0; |
71088785 | 1866 | } |
48e94196 | 1867 | #else |
a16cee10 WC |
1868 | int offline_pages(unsigned long start_pfn, unsigned long nr_pages) |
1869 | { | |
1870 | return -EINVAL; | |
1871 | } | |
60a5a19e | 1872 | int remove_memory(int nid, u64 start, u64 size) |
48e94196 KH |
1873 | { |
1874 | return -EINVAL; | |
1875 | } | |
0c0e6195 | 1876 | #endif /* CONFIG_MEMORY_HOTREMOVE */ |
71088785 | 1877 | EXPORT_SYMBOL_GPL(remove_memory); |