]>
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 | ||
e455a9b9 | 285 | static int __meminit move_pfn_range_left(struct zone *z1, struct zone *z2, |
511c2aba LJ |
286 | unsigned long start_pfn, unsigned long end_pfn) |
287 | { | |
e455a9b9 | 288 | int ret; |
511c2aba | 289 | unsigned long flags; |
e455a9b9 LJ |
290 | unsigned long z1_start_pfn; |
291 | ||
292 | if (!z1->wait_table) { | |
293 | ret = init_currently_empty_zone(z1, start_pfn, | |
294 | end_pfn - start_pfn, MEMMAP_HOTPLUG); | |
295 | if (ret) | |
296 | return ret; | |
297 | } | |
511c2aba LJ |
298 | |
299 | pgdat_resize_lock(z1->zone_pgdat, &flags); | |
300 | ||
301 | /* can't move pfns which are higher than @z2 */ | |
302 | if (end_pfn > z2->zone_start_pfn + z2->spanned_pages) | |
303 | goto out_fail; | |
304 | /* the move out part mast at the left most of @z2 */ | |
305 | if (start_pfn > z2->zone_start_pfn) | |
306 | goto out_fail; | |
307 | /* must included/overlap */ | |
308 | if (end_pfn <= z2->zone_start_pfn) | |
309 | goto out_fail; | |
310 | ||
e455a9b9 LJ |
311 | /* use start_pfn for z1's start_pfn if z1 is empty */ |
312 | if (z1->spanned_pages) | |
313 | z1_start_pfn = z1->zone_start_pfn; | |
314 | else | |
315 | z1_start_pfn = start_pfn; | |
316 | ||
317 | resize_zone(z1, z1_start_pfn, end_pfn); | |
511c2aba LJ |
318 | resize_zone(z2, end_pfn, z2->zone_start_pfn + z2->spanned_pages); |
319 | ||
320 | pgdat_resize_unlock(z1->zone_pgdat, &flags); | |
321 | ||
322 | fix_zone_id(z1, start_pfn, end_pfn); | |
323 | ||
324 | return 0; | |
325 | out_fail: | |
326 | pgdat_resize_unlock(z1->zone_pgdat, &flags); | |
327 | return -1; | |
328 | } | |
329 | ||
e455a9b9 | 330 | static int __meminit move_pfn_range_right(struct zone *z1, struct zone *z2, |
511c2aba LJ |
331 | unsigned long start_pfn, unsigned long end_pfn) |
332 | { | |
e455a9b9 | 333 | int ret; |
511c2aba | 334 | unsigned long flags; |
e455a9b9 LJ |
335 | unsigned long z2_end_pfn; |
336 | ||
337 | if (!z2->wait_table) { | |
338 | ret = init_currently_empty_zone(z2, start_pfn, | |
339 | end_pfn - start_pfn, MEMMAP_HOTPLUG); | |
340 | if (ret) | |
341 | return ret; | |
342 | } | |
511c2aba LJ |
343 | |
344 | pgdat_resize_lock(z1->zone_pgdat, &flags); | |
345 | ||
346 | /* can't move pfns which are lower than @z1 */ | |
347 | if (z1->zone_start_pfn > start_pfn) | |
348 | goto out_fail; | |
349 | /* the move out part mast at the right most of @z1 */ | |
350 | if (z1->zone_start_pfn + z1->spanned_pages > end_pfn) | |
351 | goto out_fail; | |
352 | /* must included/overlap */ | |
353 | if (start_pfn >= z1->zone_start_pfn + z1->spanned_pages) | |
354 | goto out_fail; | |
355 | ||
e455a9b9 LJ |
356 | /* use end_pfn for z2's end_pfn if z2 is empty */ |
357 | if (z2->spanned_pages) | |
358 | z2_end_pfn = z2->zone_start_pfn + z2->spanned_pages; | |
359 | else | |
360 | z2_end_pfn = end_pfn; | |
361 | ||
511c2aba | 362 | resize_zone(z1, z1->zone_start_pfn, start_pfn); |
e455a9b9 | 363 | resize_zone(z2, start_pfn, z2_end_pfn); |
511c2aba LJ |
364 | |
365 | pgdat_resize_unlock(z1->zone_pgdat, &flags); | |
366 | ||
367 | fix_zone_id(z2, start_pfn, end_pfn); | |
368 | ||
369 | return 0; | |
370 | out_fail: | |
371 | pgdat_resize_unlock(z1->zone_pgdat, &flags); | |
372 | return -1; | |
373 | } | |
374 | ||
76cdd58e HC |
375 | static void grow_pgdat_span(struct pglist_data *pgdat, unsigned long start_pfn, |
376 | unsigned long end_pfn) | |
377 | { | |
378 | unsigned long old_pgdat_end_pfn = | |
379 | pgdat->node_start_pfn + pgdat->node_spanned_pages; | |
380 | ||
712cd386 | 381 | if (!pgdat->node_spanned_pages || start_pfn < pgdat->node_start_pfn) |
76cdd58e HC |
382 | pgdat->node_start_pfn = start_pfn; |
383 | ||
384 | pgdat->node_spanned_pages = max(old_pgdat_end_pfn, end_pfn) - | |
385 | pgdat->node_start_pfn; | |
386 | } | |
387 | ||
31168481 | 388 | static int __meminit __add_zone(struct zone *zone, unsigned long phys_start_pfn) |
3947be19 DH |
389 | { |
390 | struct pglist_data *pgdat = zone->zone_pgdat; | |
391 | int nr_pages = PAGES_PER_SECTION; | |
392 | int nid = pgdat->node_id; | |
393 | int zone_type; | |
76cdd58e | 394 | unsigned long flags; |
3947be19 DH |
395 | |
396 | zone_type = zone - pgdat->node_zones; | |
76cdd58e HC |
397 | if (!zone->wait_table) { |
398 | int ret; | |
399 | ||
400 | ret = init_currently_empty_zone(zone, phys_start_pfn, | |
401 | nr_pages, MEMMAP_HOTPLUG); | |
402 | if (ret) | |
403 | return ret; | |
404 | } | |
405 | pgdat_resize_lock(zone->zone_pgdat, &flags); | |
406 | grow_zone_span(zone, phys_start_pfn, phys_start_pfn + nr_pages); | |
407 | grow_pgdat_span(zone->zone_pgdat, phys_start_pfn, | |
408 | phys_start_pfn + nr_pages); | |
409 | pgdat_resize_unlock(zone->zone_pgdat, &flags); | |
a2f3aa02 DH |
410 | memmap_init_zone(nr_pages, nid, zone_type, |
411 | phys_start_pfn, MEMMAP_HOTPLUG); | |
718127cc | 412 | return 0; |
3947be19 DH |
413 | } |
414 | ||
c04fc586 GH |
415 | static int __meminit __add_section(int nid, struct zone *zone, |
416 | unsigned long phys_start_pfn) | |
3947be19 | 417 | { |
3947be19 | 418 | int nr_pages = PAGES_PER_SECTION; |
3947be19 DH |
419 | int ret; |
420 | ||
ebd15302 KH |
421 | if (pfn_valid(phys_start_pfn)) |
422 | return -EEXIST; | |
423 | ||
0b0acbec | 424 | ret = sparse_add_one_section(zone, phys_start_pfn, nr_pages); |
3947be19 DH |
425 | |
426 | if (ret < 0) | |
427 | return ret; | |
428 | ||
718127cc YG |
429 | ret = __add_zone(zone, phys_start_pfn); |
430 | ||
431 | if (ret < 0) | |
432 | return ret; | |
433 | ||
c04fc586 | 434 | return register_new_memory(nid, __pfn_to_section(phys_start_pfn)); |
3947be19 DH |
435 | } |
436 | ||
815121d2 YI |
437 | /* find the smallest valid pfn in the range [start_pfn, end_pfn) */ |
438 | static int find_smallest_section_pfn(int nid, struct zone *zone, | |
439 | unsigned long start_pfn, | |
440 | unsigned long end_pfn) | |
441 | { | |
442 | struct mem_section *ms; | |
443 | ||
444 | for (; start_pfn < end_pfn; start_pfn += PAGES_PER_SECTION) { | |
445 | ms = __pfn_to_section(start_pfn); | |
446 | ||
447 | if (unlikely(!valid_section(ms))) | |
448 | continue; | |
449 | ||
450 | if (unlikely(pfn_to_nid(start_pfn) != nid)) | |
451 | continue; | |
452 | ||
453 | if (zone && zone != page_zone(pfn_to_page(start_pfn))) | |
454 | continue; | |
455 | ||
456 | return start_pfn; | |
457 | } | |
458 | ||
459 | return 0; | |
460 | } | |
461 | ||
462 | /* find the biggest valid pfn in the range [start_pfn, end_pfn). */ | |
463 | static int find_biggest_section_pfn(int nid, struct zone *zone, | |
464 | unsigned long start_pfn, | |
465 | unsigned long end_pfn) | |
466 | { | |
467 | struct mem_section *ms; | |
468 | unsigned long pfn; | |
469 | ||
470 | /* pfn is the end pfn of a memory section. */ | |
471 | pfn = end_pfn - 1; | |
472 | for (; pfn >= start_pfn; pfn -= PAGES_PER_SECTION) { | |
473 | ms = __pfn_to_section(pfn); | |
474 | ||
475 | if (unlikely(!valid_section(ms))) | |
476 | continue; | |
477 | ||
478 | if (unlikely(pfn_to_nid(pfn) != nid)) | |
479 | continue; | |
480 | ||
481 | if (zone && zone != page_zone(pfn_to_page(pfn))) | |
482 | continue; | |
483 | ||
484 | return pfn; | |
485 | } | |
486 | ||
487 | return 0; | |
488 | } | |
489 | ||
490 | static void shrink_zone_span(struct zone *zone, unsigned long start_pfn, | |
491 | unsigned long end_pfn) | |
492 | { | |
493 | unsigned long zone_start_pfn = zone->zone_start_pfn; | |
494 | unsigned long zone_end_pfn = zone->zone_start_pfn + zone->spanned_pages; | |
495 | unsigned long pfn; | |
496 | struct mem_section *ms; | |
497 | int nid = zone_to_nid(zone); | |
498 | ||
499 | zone_span_writelock(zone); | |
500 | if (zone_start_pfn == start_pfn) { | |
501 | /* | |
502 | * If the section is smallest section in the zone, it need | |
503 | * shrink zone->zone_start_pfn and zone->zone_spanned_pages. | |
504 | * In this case, we find second smallest valid mem_section | |
505 | * for shrinking zone. | |
506 | */ | |
507 | pfn = find_smallest_section_pfn(nid, zone, end_pfn, | |
508 | zone_end_pfn); | |
509 | if (pfn) { | |
510 | zone->zone_start_pfn = pfn; | |
511 | zone->spanned_pages = zone_end_pfn - pfn; | |
512 | } | |
513 | } else if (zone_end_pfn == end_pfn) { | |
514 | /* | |
515 | * If the section is biggest section in the zone, it need | |
516 | * shrink zone->spanned_pages. | |
517 | * In this case, we find second biggest valid mem_section for | |
518 | * shrinking zone. | |
519 | */ | |
520 | pfn = find_biggest_section_pfn(nid, zone, zone_start_pfn, | |
521 | start_pfn); | |
522 | if (pfn) | |
523 | zone->spanned_pages = pfn - zone_start_pfn + 1; | |
524 | } | |
525 | ||
526 | /* | |
527 | * The section is not biggest or smallest mem_section in the zone, it | |
528 | * only creates a hole in the zone. So in this case, we need not | |
529 | * change the zone. But perhaps, the zone has only hole data. Thus | |
530 | * it check the zone has only hole or not. | |
531 | */ | |
532 | pfn = zone_start_pfn; | |
533 | for (; pfn < zone_end_pfn; pfn += PAGES_PER_SECTION) { | |
534 | ms = __pfn_to_section(pfn); | |
535 | ||
536 | if (unlikely(!valid_section(ms))) | |
537 | continue; | |
538 | ||
539 | if (page_zone(pfn_to_page(pfn)) != zone) | |
540 | continue; | |
541 | ||
542 | /* If the section is current section, it continues the loop */ | |
543 | if (start_pfn == pfn) | |
544 | continue; | |
545 | ||
546 | /* If we find valid section, we have nothing to do */ | |
547 | zone_span_writeunlock(zone); | |
548 | return; | |
549 | } | |
550 | ||
551 | /* The zone has no valid section */ | |
552 | zone->zone_start_pfn = 0; | |
553 | zone->spanned_pages = 0; | |
554 | zone_span_writeunlock(zone); | |
555 | } | |
556 | ||
557 | static void shrink_pgdat_span(struct pglist_data *pgdat, | |
558 | unsigned long start_pfn, unsigned long end_pfn) | |
559 | { | |
560 | unsigned long pgdat_start_pfn = pgdat->node_start_pfn; | |
561 | unsigned long pgdat_end_pfn = | |
562 | pgdat->node_start_pfn + pgdat->node_spanned_pages; | |
563 | unsigned long pfn; | |
564 | struct mem_section *ms; | |
565 | int nid = pgdat->node_id; | |
566 | ||
567 | if (pgdat_start_pfn == start_pfn) { | |
568 | /* | |
569 | * If the section is smallest section in the pgdat, it need | |
570 | * shrink pgdat->node_start_pfn and pgdat->node_spanned_pages. | |
571 | * In this case, we find second smallest valid mem_section | |
572 | * for shrinking zone. | |
573 | */ | |
574 | pfn = find_smallest_section_pfn(nid, NULL, end_pfn, | |
575 | pgdat_end_pfn); | |
576 | if (pfn) { | |
577 | pgdat->node_start_pfn = pfn; | |
578 | pgdat->node_spanned_pages = pgdat_end_pfn - pfn; | |
579 | } | |
580 | } else if (pgdat_end_pfn == end_pfn) { | |
581 | /* | |
582 | * If the section is biggest section in the pgdat, it need | |
583 | * shrink pgdat->node_spanned_pages. | |
584 | * In this case, we find second biggest valid mem_section for | |
585 | * shrinking zone. | |
586 | */ | |
587 | pfn = find_biggest_section_pfn(nid, NULL, pgdat_start_pfn, | |
588 | start_pfn); | |
589 | if (pfn) | |
590 | pgdat->node_spanned_pages = pfn - pgdat_start_pfn + 1; | |
591 | } | |
592 | ||
593 | /* | |
594 | * If the section is not biggest or smallest mem_section in the pgdat, | |
595 | * it only creates a hole in the pgdat. So in this case, we need not | |
596 | * change the pgdat. | |
597 | * But perhaps, the pgdat has only hole data. Thus it check the pgdat | |
598 | * has only hole or not. | |
599 | */ | |
600 | pfn = pgdat_start_pfn; | |
601 | for (; pfn < pgdat_end_pfn; pfn += PAGES_PER_SECTION) { | |
602 | ms = __pfn_to_section(pfn); | |
603 | ||
604 | if (unlikely(!valid_section(ms))) | |
605 | continue; | |
606 | ||
607 | if (pfn_to_nid(pfn) != nid) | |
608 | continue; | |
609 | ||
610 | /* If the section is current section, it continues the loop */ | |
611 | if (start_pfn == pfn) | |
612 | continue; | |
613 | ||
614 | /* If we find valid section, we have nothing to do */ | |
615 | return; | |
616 | } | |
617 | ||
618 | /* The pgdat has no valid section */ | |
619 | pgdat->node_start_pfn = 0; | |
620 | pgdat->node_spanned_pages = 0; | |
621 | } | |
622 | ||
623 | static void __remove_zone(struct zone *zone, unsigned long start_pfn) | |
624 | { | |
625 | struct pglist_data *pgdat = zone->zone_pgdat; | |
626 | int nr_pages = PAGES_PER_SECTION; | |
627 | int zone_type; | |
628 | unsigned long flags; | |
629 | ||
630 | zone_type = zone - pgdat->node_zones; | |
631 | ||
632 | pgdat_resize_lock(zone->zone_pgdat, &flags); | |
633 | shrink_zone_span(zone, start_pfn, start_pfn + nr_pages); | |
634 | shrink_pgdat_span(pgdat, start_pfn, start_pfn + nr_pages); | |
635 | pgdat_resize_unlock(zone->zone_pgdat, &flags); | |
636 | } | |
637 | ||
ea01ea93 BP |
638 | static int __remove_section(struct zone *zone, struct mem_section *ms) |
639 | { | |
815121d2 YI |
640 | unsigned long start_pfn; |
641 | int scn_nr; | |
ea01ea93 BP |
642 | int ret = -EINVAL; |
643 | ||
644 | if (!valid_section(ms)) | |
645 | return ret; | |
646 | ||
647 | ret = unregister_memory_section(ms); | |
648 | if (ret) | |
649 | return ret; | |
650 | ||
815121d2 YI |
651 | scn_nr = __section_nr(ms); |
652 | start_pfn = section_nr_to_pfn(scn_nr); | |
653 | __remove_zone(zone, start_pfn); | |
654 | ||
ea01ea93 | 655 | sparse_remove_one_section(zone, ms); |
ea01ea93 BP |
656 | return 0; |
657 | } | |
658 | ||
3947be19 DH |
659 | /* |
660 | * Reasonably generic function for adding memory. It is | |
661 | * expected that archs that support memory hotplug will | |
662 | * call this function after deciding the zone to which to | |
663 | * add the new pages. | |
664 | */ | |
c04fc586 GH |
665 | int __ref __add_pages(int nid, struct zone *zone, unsigned long phys_start_pfn, |
666 | unsigned long nr_pages) | |
3947be19 DH |
667 | { |
668 | unsigned long i; | |
669 | int err = 0; | |
6f712711 KH |
670 | int start_sec, end_sec; |
671 | /* during initialize mem_map, align hot-added range to section */ | |
672 | start_sec = pfn_to_section_nr(phys_start_pfn); | |
673 | end_sec = pfn_to_section_nr(phys_start_pfn + nr_pages - 1); | |
3947be19 | 674 | |
6f712711 | 675 | for (i = start_sec; i <= end_sec; i++) { |
c04fc586 | 676 | err = __add_section(nid, zone, i << PFN_SECTION_SHIFT); |
3947be19 | 677 | |
6f712711 | 678 | /* |
183ff22b | 679 | * EEXIST is finally dealt with by ioresource collision |
6f712711 KH |
680 | * check. see add_memory() => register_memory_resource() |
681 | * Warning will be printed if there is collision. | |
bed120c6 JS |
682 | */ |
683 | if (err && (err != -EEXIST)) | |
3947be19 | 684 | break; |
6f712711 | 685 | err = 0; |
3947be19 DH |
686 | } |
687 | ||
688 | return err; | |
689 | } | |
bed120c6 | 690 | EXPORT_SYMBOL_GPL(__add_pages); |
3947be19 | 691 | |
ea01ea93 BP |
692 | /** |
693 | * __remove_pages() - remove sections of pages from a zone | |
694 | * @zone: zone from which pages need to be removed | |
695 | * @phys_start_pfn: starting pageframe (must be aligned to start of a section) | |
696 | * @nr_pages: number of pages to remove (must be multiple of section size) | |
697 | * | |
698 | * Generic helper function to remove section mappings and sysfs entries | |
699 | * for the section of the memory we are removing. Caller needs to make | |
700 | * sure that pages are marked reserved and zones are adjust properly by | |
701 | * calling offline_pages(). | |
702 | */ | |
703 | int __remove_pages(struct zone *zone, unsigned long phys_start_pfn, | |
704 | unsigned long nr_pages) | |
705 | { | |
706 | unsigned long i, ret = 0; | |
707 | int sections_to_remove; | |
708 | ||
709 | /* | |
710 | * We can only remove entire sections | |
711 | */ | |
712 | BUG_ON(phys_start_pfn & ~PAGE_SECTION_MASK); | |
713 | BUG_ON(nr_pages % PAGES_PER_SECTION); | |
714 | ||
d760afd4 YI |
715 | release_mem_region(phys_start_pfn << PAGE_SHIFT, nr_pages * PAGE_SIZE); |
716 | ||
ea01ea93 BP |
717 | sections_to_remove = nr_pages / PAGES_PER_SECTION; |
718 | for (i = 0; i < sections_to_remove; i++) { | |
719 | unsigned long pfn = phys_start_pfn + i*PAGES_PER_SECTION; | |
720 | ret = __remove_section(zone, __pfn_to_section(pfn)); | |
721 | if (ret) | |
722 | break; | |
723 | } | |
724 | return ret; | |
725 | } | |
726 | EXPORT_SYMBOL_GPL(__remove_pages); | |
727 | ||
9d0ad8ca DK |
728 | int set_online_page_callback(online_page_callback_t callback) |
729 | { | |
730 | int rc = -EINVAL; | |
731 | ||
732 | lock_memory_hotplug(); | |
733 | ||
734 | if (online_page_callback == generic_online_page) { | |
735 | online_page_callback = callback; | |
736 | rc = 0; | |
737 | } | |
738 | ||
739 | unlock_memory_hotplug(); | |
740 | ||
741 | return rc; | |
742 | } | |
743 | EXPORT_SYMBOL_GPL(set_online_page_callback); | |
744 | ||
745 | int restore_online_page_callback(online_page_callback_t callback) | |
746 | { | |
747 | int rc = -EINVAL; | |
748 | ||
749 | lock_memory_hotplug(); | |
750 | ||
751 | if (online_page_callback == callback) { | |
752 | online_page_callback = generic_online_page; | |
753 | rc = 0; | |
754 | } | |
755 | ||
756 | unlock_memory_hotplug(); | |
757 | ||
758 | return rc; | |
759 | } | |
760 | EXPORT_SYMBOL_GPL(restore_online_page_callback); | |
761 | ||
762 | void __online_page_set_limits(struct page *page) | |
180c06ef | 763 | { |
4738e1b9 JB |
764 | unsigned long pfn = page_to_pfn(page); |
765 | ||
4738e1b9 JB |
766 | if (pfn >= num_physpages) |
767 | num_physpages = pfn + 1; | |
9d0ad8ca DK |
768 | } |
769 | EXPORT_SYMBOL_GPL(__online_page_set_limits); | |
770 | ||
771 | void __online_page_increment_counters(struct page *page) | |
772 | { | |
773 | totalram_pages++; | |
180c06ef JF |
774 | |
775 | #ifdef CONFIG_HIGHMEM | |
776 | if (PageHighMem(page)) | |
777 | totalhigh_pages++; | |
778 | #endif | |
9d0ad8ca DK |
779 | } |
780 | EXPORT_SYMBOL_GPL(__online_page_increment_counters); | |
180c06ef | 781 | |
9d0ad8ca DK |
782 | void __online_page_free(struct page *page) |
783 | { | |
180c06ef JF |
784 | ClearPageReserved(page); |
785 | init_page_count(page); | |
786 | __free_page(page); | |
787 | } | |
9d0ad8ca DK |
788 | EXPORT_SYMBOL_GPL(__online_page_free); |
789 | ||
790 | static void generic_online_page(struct page *page) | |
791 | { | |
792 | __online_page_set_limits(page); | |
793 | __online_page_increment_counters(page); | |
794 | __online_page_free(page); | |
795 | } | |
180c06ef | 796 | |
75884fb1 KH |
797 | static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages, |
798 | void *arg) | |
3947be19 DH |
799 | { |
800 | unsigned long i; | |
75884fb1 KH |
801 | unsigned long onlined_pages = *(unsigned long *)arg; |
802 | struct page *page; | |
803 | if (PageReserved(pfn_to_page(start_pfn))) | |
804 | for (i = 0; i < nr_pages; i++) { | |
805 | page = pfn_to_page(start_pfn + i); | |
9d0ad8ca | 806 | (*online_page_callback)(page); |
75884fb1 KH |
807 | onlined_pages++; |
808 | } | |
809 | *(unsigned long *)arg = onlined_pages; | |
810 | return 0; | |
811 | } | |
812 | ||
09285af7 | 813 | #ifdef CONFIG_MOVABLE_NODE |
79a4dcef TC |
814 | /* |
815 | * When CONFIG_MOVABLE_NODE, we permit onlining of a node which doesn't have | |
816 | * normal memory. | |
817 | */ | |
09285af7 LJ |
818 | static bool can_online_high_movable(struct zone *zone) |
819 | { | |
820 | return true; | |
821 | } | |
79a4dcef | 822 | #else /* CONFIG_MOVABLE_NODE */ |
74d42d8f LJ |
823 | /* ensure every online node has NORMAL memory */ |
824 | static bool can_online_high_movable(struct zone *zone) | |
825 | { | |
826 | return node_state(zone_to_nid(zone), N_NORMAL_MEMORY); | |
827 | } | |
79a4dcef | 828 | #endif /* CONFIG_MOVABLE_NODE */ |
74d42d8f | 829 | |
d9713679 LJ |
830 | /* check which state of node_states will be changed when online memory */ |
831 | static void node_states_check_changes_online(unsigned long nr_pages, | |
832 | struct zone *zone, struct memory_notify *arg) | |
833 | { | |
834 | int nid = zone_to_nid(zone); | |
835 | enum zone_type zone_last = ZONE_NORMAL; | |
836 | ||
837 | /* | |
6715ddf9 LJ |
838 | * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY] |
839 | * contains nodes which have zones of 0...ZONE_NORMAL, | |
840 | * set zone_last to ZONE_NORMAL. | |
d9713679 | 841 | * |
6715ddf9 LJ |
842 | * If we don't have HIGHMEM nor movable node, |
843 | * node_states[N_NORMAL_MEMORY] contains nodes which have zones of | |
844 | * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE. | |
d9713679 | 845 | */ |
6715ddf9 | 846 | if (N_MEMORY == N_NORMAL_MEMORY) |
d9713679 LJ |
847 | zone_last = ZONE_MOVABLE; |
848 | ||
849 | /* | |
850 | * if the memory to be online is in a zone of 0...zone_last, and | |
851 | * the zones of 0...zone_last don't have memory before online, we will | |
852 | * need to set the node to node_states[N_NORMAL_MEMORY] after | |
853 | * the memory is online. | |
854 | */ | |
855 | if (zone_idx(zone) <= zone_last && !node_state(nid, N_NORMAL_MEMORY)) | |
856 | arg->status_change_nid_normal = nid; | |
857 | else | |
858 | arg->status_change_nid_normal = -1; | |
859 | ||
6715ddf9 LJ |
860 | #ifdef CONFIG_HIGHMEM |
861 | /* | |
862 | * If we have movable node, node_states[N_HIGH_MEMORY] | |
863 | * contains nodes which have zones of 0...ZONE_HIGHMEM, | |
864 | * set zone_last to ZONE_HIGHMEM. | |
865 | * | |
866 | * If we don't have movable node, node_states[N_NORMAL_MEMORY] | |
867 | * contains nodes which have zones of 0...ZONE_MOVABLE, | |
868 | * set zone_last to ZONE_MOVABLE. | |
869 | */ | |
870 | zone_last = ZONE_HIGHMEM; | |
871 | if (N_MEMORY == N_HIGH_MEMORY) | |
872 | zone_last = ZONE_MOVABLE; | |
873 | ||
874 | if (zone_idx(zone) <= zone_last && !node_state(nid, N_HIGH_MEMORY)) | |
875 | arg->status_change_nid_high = nid; | |
876 | else | |
877 | arg->status_change_nid_high = -1; | |
878 | #else | |
879 | arg->status_change_nid_high = arg->status_change_nid_normal; | |
880 | #endif | |
881 | ||
d9713679 LJ |
882 | /* |
883 | * if the node don't have memory befor online, we will need to | |
6715ddf9 | 884 | * set the node to node_states[N_MEMORY] after the memory |
d9713679 LJ |
885 | * is online. |
886 | */ | |
6715ddf9 | 887 | if (!node_state(nid, N_MEMORY)) |
d9713679 LJ |
888 | arg->status_change_nid = nid; |
889 | else | |
890 | arg->status_change_nid = -1; | |
891 | } | |
892 | ||
893 | static void node_states_set_node(int node, struct memory_notify *arg) | |
894 | { | |
895 | if (arg->status_change_nid_normal >= 0) | |
896 | node_set_state(node, N_NORMAL_MEMORY); | |
897 | ||
6715ddf9 LJ |
898 | if (arg->status_change_nid_high >= 0) |
899 | node_set_state(node, N_HIGH_MEMORY); | |
900 | ||
901 | node_set_state(node, N_MEMORY); | |
d9713679 LJ |
902 | } |
903 | ||
75884fb1 | 904 | |
511c2aba | 905 | int __ref online_pages(unsigned long pfn, unsigned long nr_pages, int online_type) |
75884fb1 | 906 | { |
3947be19 DH |
907 | unsigned long onlined_pages = 0; |
908 | struct zone *zone; | |
6811378e | 909 | int need_zonelists_rebuild = 0; |
7b78d335 YG |
910 | int nid; |
911 | int ret; | |
912 | struct memory_notify arg; | |
913 | ||
925268a0 | 914 | lock_memory_hotplug(); |
d9713679 LJ |
915 | /* |
916 | * This doesn't need a lock to do pfn_to_page(). | |
917 | * The section can't be removed here because of the | |
918 | * memory_block->state_mutex. | |
919 | */ | |
920 | zone = page_zone(pfn_to_page(pfn)); | |
921 | ||
74d42d8f LJ |
922 | if ((zone_idx(zone) > ZONE_NORMAL || online_type == ONLINE_MOVABLE) && |
923 | !can_online_high_movable(zone)) { | |
924 | unlock_memory_hotplug(); | |
925 | return -1; | |
926 | } | |
927 | ||
511c2aba LJ |
928 | if (online_type == ONLINE_KERNEL && zone_idx(zone) == ZONE_MOVABLE) { |
929 | if (move_pfn_range_left(zone - 1, zone, pfn, pfn + nr_pages)) { | |
930 | unlock_memory_hotplug(); | |
931 | return -1; | |
932 | } | |
933 | } | |
934 | if (online_type == ONLINE_MOVABLE && zone_idx(zone) == ZONE_MOVABLE - 1) { | |
935 | if (move_pfn_range_right(zone, zone + 1, pfn, pfn + nr_pages)) { | |
936 | unlock_memory_hotplug(); | |
937 | return -1; | |
938 | } | |
939 | } | |
940 | ||
941 | /* Previous code may changed the zone of the pfn range */ | |
942 | zone = page_zone(pfn_to_page(pfn)); | |
943 | ||
7b78d335 YG |
944 | arg.start_pfn = pfn; |
945 | arg.nr_pages = nr_pages; | |
d9713679 | 946 | node_states_check_changes_online(nr_pages, zone, &arg); |
7b78d335 YG |
947 | |
948 | nid = page_to_nid(pfn_to_page(pfn)); | |
3947be19 | 949 | |
7b78d335 YG |
950 | ret = memory_notify(MEM_GOING_ONLINE, &arg); |
951 | ret = notifier_to_errno(ret); | |
952 | if (ret) { | |
953 | memory_notify(MEM_CANCEL_ONLINE, &arg); | |
925268a0 | 954 | unlock_memory_hotplug(); |
7b78d335 YG |
955 | return ret; |
956 | } | |
6811378e YG |
957 | /* |
958 | * If this zone is not populated, then it is not in zonelist. | |
959 | * This means the page allocator ignores this zone. | |
960 | * So, zonelist must be updated after online. | |
961 | */ | |
4eaf3f64 | 962 | mutex_lock(&zonelists_mutex); |
6dcd73d7 | 963 | if (!populated_zone(zone)) { |
6811378e | 964 | need_zonelists_rebuild = 1; |
6dcd73d7 WC |
965 | build_all_zonelists(NULL, zone); |
966 | } | |
6811378e | 967 | |
908eedc6 | 968 | ret = walk_system_ram_range(pfn, nr_pages, &onlined_pages, |
75884fb1 | 969 | online_pages_range); |
fd8a4221 | 970 | if (ret) { |
6dcd73d7 WC |
971 | if (need_zonelists_rebuild) |
972 | zone_pcp_reset(zone); | |
4eaf3f64 | 973 | mutex_unlock(&zonelists_mutex); |
a62e2f4f BH |
974 | printk(KERN_DEBUG "online_pages [mem %#010llx-%#010llx] failed\n", |
975 | (unsigned long long) pfn << PAGE_SHIFT, | |
976 | (((unsigned long long) pfn + nr_pages) | |
977 | << PAGE_SHIFT) - 1); | |
fd8a4221 | 978 | memory_notify(MEM_CANCEL_ONLINE, &arg); |
925268a0 | 979 | unlock_memory_hotplug(); |
fd8a4221 GL |
980 | return ret; |
981 | } | |
982 | ||
9feedc9d | 983 | zone->managed_pages += onlined_pages; |
3947be19 | 984 | zone->present_pages += onlined_pages; |
f2937be5 | 985 | zone->zone_pgdat->node_present_pages += onlined_pages; |
08dff7b7 | 986 | if (onlined_pages) { |
d9713679 | 987 | node_states_set_node(zone_to_nid(zone), &arg); |
08dff7b7 | 988 | if (need_zonelists_rebuild) |
6dcd73d7 | 989 | build_all_zonelists(NULL, NULL); |
08dff7b7 JL |
990 | else |
991 | zone_pcp_update(zone); | |
992 | } | |
3947be19 | 993 | |
4eaf3f64 | 994 | mutex_unlock(&zonelists_mutex); |
1b79acc9 KM |
995 | |
996 | init_per_zone_wmark_min(); | |
997 | ||
08dff7b7 | 998 | if (onlined_pages) |
7ea1530a | 999 | kswapd_run(zone_to_nid(zone)); |
61b13993 | 1000 | |
1f522509 | 1001 | vm_total_pages = nr_free_pagecache_pages(); |
2f7f24ec | 1002 | |
2d1d43f6 | 1003 | writeback_set_ratelimit(); |
7b78d335 YG |
1004 | |
1005 | if (onlined_pages) | |
1006 | memory_notify(MEM_ONLINE, &arg); | |
925268a0 | 1007 | unlock_memory_hotplug(); |
7b78d335 | 1008 | |
3947be19 DH |
1009 | return 0; |
1010 | } | |
53947027 | 1011 | #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */ |
bc02af93 | 1012 | |
e1319331 HS |
1013 | /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */ |
1014 | static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start) | |
9af3c2de YG |
1015 | { |
1016 | struct pglist_data *pgdat; | |
1017 | unsigned long zones_size[MAX_NR_ZONES] = {0}; | |
1018 | unsigned long zholes_size[MAX_NR_ZONES] = {0}; | |
1019 | unsigned long start_pfn = start >> PAGE_SHIFT; | |
1020 | ||
a1e565aa TC |
1021 | pgdat = NODE_DATA(nid); |
1022 | if (!pgdat) { | |
1023 | pgdat = arch_alloc_nodedata(nid); | |
1024 | if (!pgdat) | |
1025 | return NULL; | |
9af3c2de | 1026 | |
a1e565aa TC |
1027 | arch_refresh_nodedata(nid, pgdat); |
1028 | } | |
9af3c2de YG |
1029 | |
1030 | /* we can use NODE_DATA(nid) from here */ | |
1031 | ||
1032 | /* init node's zones as empty zones, we don't have any present pages.*/ | |
9109fb7b | 1033 | free_area_init_node(nid, zones_size, start_pfn, zholes_size); |
9af3c2de | 1034 | |
959ecc48 KH |
1035 | /* |
1036 | * The node we allocated has no zone fallback lists. For avoiding | |
1037 | * to access not-initialized zonelist, build here. | |
1038 | */ | |
f957db4f | 1039 | mutex_lock(&zonelists_mutex); |
9adb62a5 | 1040 | build_all_zonelists(pgdat, NULL); |
f957db4f | 1041 | mutex_unlock(&zonelists_mutex); |
959ecc48 | 1042 | |
9af3c2de YG |
1043 | return pgdat; |
1044 | } | |
1045 | ||
1046 | static void rollback_node_hotadd(int nid, pg_data_t *pgdat) | |
1047 | { | |
1048 | arch_refresh_nodedata(nid, NULL); | |
1049 | arch_free_nodedata(pgdat); | |
1050 | return; | |
1051 | } | |
1052 | ||
0a547039 | 1053 | |
cf23422b | 1054 | /* |
1055 | * called by cpu_up() to online a node without onlined memory. | |
1056 | */ | |
1057 | int mem_online_node(int nid) | |
1058 | { | |
1059 | pg_data_t *pgdat; | |
1060 | int ret; | |
1061 | ||
20d6c96b | 1062 | lock_memory_hotplug(); |
cf23422b | 1063 | pgdat = hotadd_new_pgdat(nid, 0); |
7553e8f2 | 1064 | if (!pgdat) { |
cf23422b | 1065 | ret = -ENOMEM; |
1066 | goto out; | |
1067 | } | |
1068 | node_set_online(nid); | |
1069 | ret = register_one_node(nid); | |
1070 | BUG_ON(ret); | |
1071 | ||
1072 | out: | |
20d6c96b | 1073 | unlock_memory_hotplug(); |
cf23422b | 1074 | return ret; |
1075 | } | |
1076 | ||
31168481 AV |
1077 | /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */ |
1078 | int __ref add_memory(int nid, u64 start, u64 size) | |
bc02af93 | 1079 | { |
9af3c2de | 1080 | pg_data_t *pgdat = NULL; |
a1e565aa TC |
1081 | bool new_pgdat; |
1082 | bool new_node; | |
ebd15302 | 1083 | struct resource *res; |
bc02af93 YG |
1084 | int ret; |
1085 | ||
20d6c96b | 1086 | lock_memory_hotplug(); |
6ad696d2 | 1087 | |
ebd15302 | 1088 | res = register_memory_resource(start, size); |
6ad696d2 | 1089 | ret = -EEXIST; |
ebd15302 | 1090 | if (!res) |
6ad696d2 | 1091 | goto out; |
ebd15302 | 1092 | |
a1e565aa TC |
1093 | { /* Stupid hack to suppress address-never-null warning */ |
1094 | void *p = NODE_DATA(nid); | |
1095 | new_pgdat = !p; | |
1096 | } | |
1097 | new_node = !node_online(nid); | |
1098 | if (new_node) { | |
9af3c2de | 1099 | pgdat = hotadd_new_pgdat(nid, start); |
6ad696d2 | 1100 | ret = -ENOMEM; |
9af3c2de | 1101 | if (!pgdat) |
41b9e2d7 | 1102 | goto error; |
9af3c2de YG |
1103 | } |
1104 | ||
bc02af93 YG |
1105 | /* call arch's memory hotadd */ |
1106 | ret = arch_add_memory(nid, start, size); | |
1107 | ||
9af3c2de YG |
1108 | if (ret < 0) |
1109 | goto error; | |
1110 | ||
0fc44159 | 1111 | /* we online node here. we can't roll back from here. */ |
9af3c2de YG |
1112 | node_set_online(nid); |
1113 | ||
a1e565aa | 1114 | if (new_node) { |
0fc44159 YG |
1115 | ret = register_one_node(nid); |
1116 | /* | |
1117 | * If sysfs file of new node can't create, cpu on the node | |
1118 | * can't be hot-added. There is no rollback way now. | |
1119 | * So, check by BUG_ON() to catch it reluctantly.. | |
1120 | */ | |
1121 | BUG_ON(ret); | |
1122 | } | |
1123 | ||
d96ae530 AM |
1124 | /* create new memmap entry */ |
1125 | firmware_map_add_hotplug(start, start + size, "System RAM"); | |
1126 | ||
6ad696d2 AK |
1127 | goto out; |
1128 | ||
9af3c2de YG |
1129 | error: |
1130 | /* rollback pgdat allocation and others */ | |
1131 | if (new_pgdat) | |
1132 | rollback_node_hotadd(nid, pgdat); | |
a864b9d0 | 1133 | release_memory_resource(res); |
9af3c2de | 1134 | |
6ad696d2 | 1135 | out: |
20d6c96b | 1136 | unlock_memory_hotplug(); |
bc02af93 YG |
1137 | return ret; |
1138 | } | |
1139 | EXPORT_SYMBOL_GPL(add_memory); | |
0c0e6195 KH |
1140 | |
1141 | #ifdef CONFIG_MEMORY_HOTREMOVE | |
5c755e9f BP |
1142 | /* |
1143 | * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy | |
1144 | * set and the size of the free page is given by page_order(). Using this, | |
1145 | * the function determines if the pageblock contains only free pages. | |
1146 | * Due to buddy contraints, a free page at least the size of a pageblock will | |
1147 | * be located at the start of the pageblock | |
1148 | */ | |
1149 | static inline int pageblock_free(struct page *page) | |
1150 | { | |
1151 | return PageBuddy(page) && page_order(page) >= pageblock_order; | |
1152 | } | |
1153 | ||
1154 | /* Return the start of the next active pageblock after a given page */ | |
1155 | static struct page *next_active_pageblock(struct page *page) | |
1156 | { | |
5c755e9f BP |
1157 | /* Ensure the starting page is pageblock-aligned */ |
1158 | BUG_ON(page_to_pfn(page) & (pageblock_nr_pages - 1)); | |
1159 | ||
5c755e9f | 1160 | /* If the entire pageblock is free, move to the end of free page */ |
0dcc48c1 KH |
1161 | if (pageblock_free(page)) { |
1162 | int order; | |
1163 | /* be careful. we don't have locks, page_order can be changed.*/ | |
1164 | order = page_order(page); | |
1165 | if ((order < MAX_ORDER) && (order >= pageblock_order)) | |
1166 | return page + (1 << order); | |
1167 | } | |
5c755e9f | 1168 | |
0dcc48c1 | 1169 | return page + pageblock_nr_pages; |
5c755e9f BP |
1170 | } |
1171 | ||
1172 | /* Checks if this range of memory is likely to be hot-removable. */ | |
1173 | int is_mem_section_removable(unsigned long start_pfn, unsigned long nr_pages) | |
1174 | { | |
5c755e9f BP |
1175 | struct page *page = pfn_to_page(start_pfn); |
1176 | struct page *end_page = page + nr_pages; | |
1177 | ||
1178 | /* Check the starting page of each pageblock within the range */ | |
1179 | for (; page < end_page; page = next_active_pageblock(page)) { | |
49ac8255 | 1180 | if (!is_pageblock_removable_nolock(page)) |
5c755e9f | 1181 | return 0; |
49ac8255 | 1182 | cond_resched(); |
5c755e9f BP |
1183 | } |
1184 | ||
1185 | /* All pageblocks in the memory block are likely to be hot-removable */ | |
1186 | return 1; | |
1187 | } | |
1188 | ||
0c0e6195 KH |
1189 | /* |
1190 | * Confirm all pages in a range [start, end) is belongs to the same zone. | |
1191 | */ | |
1192 | static int test_pages_in_a_zone(unsigned long start_pfn, unsigned long end_pfn) | |
1193 | { | |
1194 | unsigned long pfn; | |
1195 | struct zone *zone = NULL; | |
1196 | struct page *page; | |
1197 | int i; | |
1198 | for (pfn = start_pfn; | |
1199 | pfn < end_pfn; | |
1200 | pfn += MAX_ORDER_NR_PAGES) { | |
1201 | i = 0; | |
1202 | /* This is just a CONFIG_HOLES_IN_ZONE check.*/ | |
1203 | while ((i < MAX_ORDER_NR_PAGES) && !pfn_valid_within(pfn + i)) | |
1204 | i++; | |
1205 | if (i == MAX_ORDER_NR_PAGES) | |
1206 | continue; | |
1207 | page = pfn_to_page(pfn + i); | |
1208 | if (zone && page_zone(page) != zone) | |
1209 | return 0; | |
1210 | zone = page_zone(page); | |
1211 | } | |
1212 | return 1; | |
1213 | } | |
1214 | ||
1215 | /* | |
1216 | * Scanning pfn is much easier than scanning lru list. | |
1217 | * Scan pfn from start to end and Find LRU page. | |
1218 | */ | |
7bbc0905 | 1219 | static unsigned long scan_lru_pages(unsigned long start, unsigned long end) |
0c0e6195 KH |
1220 | { |
1221 | unsigned long pfn; | |
1222 | struct page *page; | |
1223 | for (pfn = start; pfn < end; pfn++) { | |
1224 | if (pfn_valid(pfn)) { | |
1225 | page = pfn_to_page(pfn); | |
1226 | if (PageLRU(page)) | |
1227 | return pfn; | |
1228 | } | |
1229 | } | |
1230 | return 0; | |
1231 | } | |
1232 | ||
0c0e6195 KH |
1233 | #define NR_OFFLINE_AT_ONCE_PAGES (256) |
1234 | static int | |
1235 | do_migrate_range(unsigned long start_pfn, unsigned long end_pfn) | |
1236 | { | |
1237 | unsigned long pfn; | |
1238 | struct page *page; | |
1239 | int move_pages = NR_OFFLINE_AT_ONCE_PAGES; | |
1240 | int not_managed = 0; | |
1241 | int ret = 0; | |
1242 | LIST_HEAD(source); | |
1243 | ||
1244 | for (pfn = start_pfn; pfn < end_pfn && move_pages > 0; pfn++) { | |
1245 | if (!pfn_valid(pfn)) | |
1246 | continue; | |
1247 | page = pfn_to_page(pfn); | |
700c2a46 | 1248 | if (!get_page_unless_zero(page)) |
0c0e6195 KH |
1249 | continue; |
1250 | /* | |
1251 | * We can skip free pages. And we can only deal with pages on | |
1252 | * LRU. | |
1253 | */ | |
62695a84 | 1254 | ret = isolate_lru_page(page); |
0c0e6195 | 1255 | if (!ret) { /* Success */ |
700c2a46 | 1256 | put_page(page); |
62695a84 | 1257 | list_add_tail(&page->lru, &source); |
0c0e6195 | 1258 | move_pages--; |
6d9c285a KM |
1259 | inc_zone_page_state(page, NR_ISOLATED_ANON + |
1260 | page_is_file_cache(page)); | |
1261 | ||
0c0e6195 | 1262 | } else { |
0c0e6195 | 1263 | #ifdef CONFIG_DEBUG_VM |
718a3821 WF |
1264 | printk(KERN_ALERT "removing pfn %lx from LRU failed\n", |
1265 | pfn); | |
1266 | dump_page(page); | |
0c0e6195 | 1267 | #endif |
700c2a46 | 1268 | put_page(page); |
25985edc | 1269 | /* Because we don't have big zone->lock. we should |
809c4449 BL |
1270 | check this again here. */ |
1271 | if (page_count(page)) { | |
1272 | not_managed++; | |
f3ab2636 | 1273 | ret = -EBUSY; |
809c4449 BL |
1274 | break; |
1275 | } | |
0c0e6195 KH |
1276 | } |
1277 | } | |
f3ab2636 BL |
1278 | if (!list_empty(&source)) { |
1279 | if (not_managed) { | |
1280 | putback_lru_pages(&source); | |
1281 | goto out; | |
1282 | } | |
74c08f98 MK |
1283 | |
1284 | /* | |
1285 | * alloc_migrate_target should be improooooved!! | |
1286 | * migrate_pages returns # of failed pages. | |
1287 | */ | |
1288 | ret = migrate_pages(&source, alloc_migrate_target, 0, | |
7b2a2d4a MG |
1289 | true, MIGRATE_SYNC, |
1290 | MR_MEMORY_HOTPLUG); | |
f3ab2636 | 1291 | if (ret) |
0c0e6195 | 1292 | putback_lru_pages(&source); |
0c0e6195 | 1293 | } |
0c0e6195 KH |
1294 | out: |
1295 | return ret; | |
1296 | } | |
1297 | ||
1298 | /* | |
1299 | * remove from free_area[] and mark all as Reserved. | |
1300 | */ | |
1301 | static int | |
1302 | offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages, | |
1303 | void *data) | |
1304 | { | |
1305 | __offline_isolated_pages(start, start + nr_pages); | |
1306 | return 0; | |
1307 | } | |
1308 | ||
1309 | static void | |
1310 | offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn) | |
1311 | { | |
908eedc6 | 1312 | walk_system_ram_range(start_pfn, end_pfn - start_pfn, NULL, |
0c0e6195 KH |
1313 | offline_isolated_pages_cb); |
1314 | } | |
1315 | ||
1316 | /* | |
1317 | * Check all pages in range, recoreded as memory resource, are isolated. | |
1318 | */ | |
1319 | static int | |
1320 | check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages, | |
1321 | void *data) | |
1322 | { | |
1323 | int ret; | |
1324 | long offlined = *(long *)data; | |
b023f468 | 1325 | ret = test_pages_isolated(start_pfn, start_pfn + nr_pages, true); |
0c0e6195 KH |
1326 | offlined = nr_pages; |
1327 | if (!ret) | |
1328 | *(long *)data += offlined; | |
1329 | return ret; | |
1330 | } | |
1331 | ||
1332 | static long | |
1333 | check_pages_isolated(unsigned long start_pfn, unsigned long end_pfn) | |
1334 | { | |
1335 | long offlined = 0; | |
1336 | int ret; | |
1337 | ||
908eedc6 | 1338 | ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn, &offlined, |
0c0e6195 KH |
1339 | check_pages_isolated_cb); |
1340 | if (ret < 0) | |
1341 | offlined = (long)ret; | |
1342 | return offlined; | |
1343 | } | |
1344 | ||
09285af7 | 1345 | #ifdef CONFIG_MOVABLE_NODE |
79a4dcef TC |
1346 | /* |
1347 | * When CONFIG_MOVABLE_NODE, we permit offlining of a node which doesn't have | |
1348 | * normal memory. | |
1349 | */ | |
09285af7 LJ |
1350 | static bool can_offline_normal(struct zone *zone, unsigned long nr_pages) |
1351 | { | |
1352 | return true; | |
1353 | } | |
79a4dcef | 1354 | #else /* CONFIG_MOVABLE_NODE */ |
74d42d8f LJ |
1355 | /* ensure the node has NORMAL memory if it is still online */ |
1356 | static bool can_offline_normal(struct zone *zone, unsigned long nr_pages) | |
1357 | { | |
1358 | struct pglist_data *pgdat = zone->zone_pgdat; | |
1359 | unsigned long present_pages = 0; | |
1360 | enum zone_type zt; | |
1361 | ||
1362 | for (zt = 0; zt <= ZONE_NORMAL; zt++) | |
1363 | present_pages += pgdat->node_zones[zt].present_pages; | |
1364 | ||
1365 | if (present_pages > nr_pages) | |
1366 | return true; | |
1367 | ||
1368 | present_pages = 0; | |
1369 | for (; zt <= ZONE_MOVABLE; zt++) | |
1370 | present_pages += pgdat->node_zones[zt].present_pages; | |
1371 | ||
1372 | /* | |
1373 | * we can't offline the last normal memory until all | |
1374 | * higher memory is offlined. | |
1375 | */ | |
1376 | return present_pages == 0; | |
1377 | } | |
79a4dcef | 1378 | #endif /* CONFIG_MOVABLE_NODE */ |
74d42d8f | 1379 | |
d9713679 LJ |
1380 | /* check which state of node_states will be changed when offline memory */ |
1381 | static void node_states_check_changes_offline(unsigned long nr_pages, | |
1382 | struct zone *zone, struct memory_notify *arg) | |
1383 | { | |
1384 | struct pglist_data *pgdat = zone->zone_pgdat; | |
1385 | unsigned long present_pages = 0; | |
1386 | enum zone_type zt, zone_last = ZONE_NORMAL; | |
1387 | ||
1388 | /* | |
6715ddf9 LJ |
1389 | * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY] |
1390 | * contains nodes which have zones of 0...ZONE_NORMAL, | |
1391 | * set zone_last to ZONE_NORMAL. | |
d9713679 | 1392 | * |
6715ddf9 LJ |
1393 | * If we don't have HIGHMEM nor movable node, |
1394 | * node_states[N_NORMAL_MEMORY] contains nodes which have zones of | |
1395 | * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE. | |
d9713679 | 1396 | */ |
6715ddf9 | 1397 | if (N_MEMORY == N_NORMAL_MEMORY) |
d9713679 LJ |
1398 | zone_last = ZONE_MOVABLE; |
1399 | ||
1400 | /* | |
1401 | * check whether node_states[N_NORMAL_MEMORY] will be changed. | |
1402 | * If the memory to be offline is in a zone of 0...zone_last, | |
1403 | * and it is the last present memory, 0...zone_last will | |
1404 | * become empty after offline , thus we can determind we will | |
1405 | * need to clear the node from node_states[N_NORMAL_MEMORY]. | |
1406 | */ | |
1407 | for (zt = 0; zt <= zone_last; zt++) | |
1408 | present_pages += pgdat->node_zones[zt].present_pages; | |
1409 | if (zone_idx(zone) <= zone_last && nr_pages >= present_pages) | |
1410 | arg->status_change_nid_normal = zone_to_nid(zone); | |
1411 | else | |
1412 | arg->status_change_nid_normal = -1; | |
1413 | ||
6715ddf9 LJ |
1414 | #ifdef CONFIG_HIGHMEM |
1415 | /* | |
1416 | * If we have movable node, node_states[N_HIGH_MEMORY] | |
1417 | * contains nodes which have zones of 0...ZONE_HIGHMEM, | |
1418 | * set zone_last to ZONE_HIGHMEM. | |
1419 | * | |
1420 | * If we don't have movable node, node_states[N_NORMAL_MEMORY] | |
1421 | * contains nodes which have zones of 0...ZONE_MOVABLE, | |
1422 | * set zone_last to ZONE_MOVABLE. | |
1423 | */ | |
1424 | zone_last = ZONE_HIGHMEM; | |
1425 | if (N_MEMORY == N_HIGH_MEMORY) | |
1426 | zone_last = ZONE_MOVABLE; | |
1427 | ||
1428 | for (; zt <= zone_last; zt++) | |
1429 | present_pages += pgdat->node_zones[zt].present_pages; | |
1430 | if (zone_idx(zone) <= zone_last && nr_pages >= present_pages) | |
1431 | arg->status_change_nid_high = zone_to_nid(zone); | |
1432 | else | |
1433 | arg->status_change_nid_high = -1; | |
1434 | #else | |
1435 | arg->status_change_nid_high = arg->status_change_nid_normal; | |
1436 | #endif | |
1437 | ||
d9713679 LJ |
1438 | /* |
1439 | * node_states[N_HIGH_MEMORY] contains nodes which have 0...ZONE_MOVABLE | |
1440 | */ | |
1441 | zone_last = ZONE_MOVABLE; | |
1442 | ||
1443 | /* | |
1444 | * check whether node_states[N_HIGH_MEMORY] will be changed | |
1445 | * If we try to offline the last present @nr_pages from the node, | |
1446 | * we can determind we will need to clear the node from | |
1447 | * node_states[N_HIGH_MEMORY]. | |
1448 | */ | |
1449 | for (; zt <= zone_last; zt++) | |
1450 | present_pages += pgdat->node_zones[zt].present_pages; | |
1451 | if (nr_pages >= present_pages) | |
1452 | arg->status_change_nid = zone_to_nid(zone); | |
1453 | else | |
1454 | arg->status_change_nid = -1; | |
1455 | } | |
1456 | ||
1457 | static void node_states_clear_node(int node, struct memory_notify *arg) | |
1458 | { | |
1459 | if (arg->status_change_nid_normal >= 0) | |
1460 | node_clear_state(node, N_NORMAL_MEMORY); | |
1461 | ||
6715ddf9 LJ |
1462 | if ((N_MEMORY != N_NORMAL_MEMORY) && |
1463 | (arg->status_change_nid_high >= 0)) | |
d9713679 | 1464 | node_clear_state(node, N_HIGH_MEMORY); |
6715ddf9 LJ |
1465 | |
1466 | if ((N_MEMORY != N_HIGH_MEMORY) && | |
1467 | (arg->status_change_nid >= 0)) | |
1468 | node_clear_state(node, N_MEMORY); | |
d9713679 LJ |
1469 | } |
1470 | ||
a16cee10 | 1471 | static int __ref __offline_pages(unsigned long start_pfn, |
0c0e6195 KH |
1472 | unsigned long end_pfn, unsigned long timeout) |
1473 | { | |
1474 | unsigned long pfn, nr_pages, expire; | |
1475 | long offlined_pages; | |
7b78d335 | 1476 | int ret, drain, retry_max, node; |
0c0e6195 | 1477 | struct zone *zone; |
7b78d335 | 1478 | struct memory_notify arg; |
0c0e6195 KH |
1479 | |
1480 | BUG_ON(start_pfn >= end_pfn); | |
1481 | /* at least, alignment against pageblock is necessary */ | |
1482 | if (!IS_ALIGNED(start_pfn, pageblock_nr_pages)) | |
1483 | return -EINVAL; | |
1484 | if (!IS_ALIGNED(end_pfn, pageblock_nr_pages)) | |
1485 | return -EINVAL; | |
1486 | /* This makes hotplug much easier...and readable. | |
1487 | we assume this for now. .*/ | |
1488 | if (!test_pages_in_a_zone(start_pfn, end_pfn)) | |
1489 | return -EINVAL; | |
7b78d335 | 1490 | |
20d6c96b | 1491 | lock_memory_hotplug(); |
6ad696d2 | 1492 | |
7b78d335 YG |
1493 | zone = page_zone(pfn_to_page(start_pfn)); |
1494 | node = zone_to_nid(zone); | |
1495 | nr_pages = end_pfn - start_pfn; | |
1496 | ||
74d42d8f LJ |
1497 | ret = -EINVAL; |
1498 | if (zone_idx(zone) <= ZONE_NORMAL && !can_offline_normal(zone, nr_pages)) | |
1499 | goto out; | |
1500 | ||
0c0e6195 | 1501 | /* set above range as isolated */ |
b023f468 WC |
1502 | ret = start_isolate_page_range(start_pfn, end_pfn, |
1503 | MIGRATE_MOVABLE, true); | |
0c0e6195 | 1504 | if (ret) |
6ad696d2 | 1505 | goto out; |
7b78d335 YG |
1506 | |
1507 | arg.start_pfn = start_pfn; | |
1508 | arg.nr_pages = nr_pages; | |
d9713679 | 1509 | node_states_check_changes_offline(nr_pages, zone, &arg); |
7b78d335 YG |
1510 | |
1511 | ret = memory_notify(MEM_GOING_OFFLINE, &arg); | |
1512 | ret = notifier_to_errno(ret); | |
1513 | if (ret) | |
1514 | goto failed_removal; | |
1515 | ||
0c0e6195 KH |
1516 | pfn = start_pfn; |
1517 | expire = jiffies + timeout; | |
1518 | drain = 0; | |
1519 | retry_max = 5; | |
1520 | repeat: | |
1521 | /* start memory hot removal */ | |
1522 | ret = -EAGAIN; | |
1523 | if (time_after(jiffies, expire)) | |
1524 | goto failed_removal; | |
1525 | ret = -EINTR; | |
1526 | if (signal_pending(current)) | |
1527 | goto failed_removal; | |
1528 | ret = 0; | |
1529 | if (drain) { | |
1530 | lru_add_drain_all(); | |
0c0e6195 | 1531 | cond_resched(); |
9f8f2172 | 1532 | drain_all_pages(); |
0c0e6195 KH |
1533 | } |
1534 | ||
1535 | pfn = scan_lru_pages(start_pfn, end_pfn); | |
1536 | if (pfn) { /* We have page on LRU */ | |
1537 | ret = do_migrate_range(pfn, end_pfn); | |
1538 | if (!ret) { | |
1539 | drain = 1; | |
1540 | goto repeat; | |
1541 | } else { | |
1542 | if (ret < 0) | |
1543 | if (--retry_max == 0) | |
1544 | goto failed_removal; | |
1545 | yield(); | |
1546 | drain = 1; | |
1547 | goto repeat; | |
1548 | } | |
1549 | } | |
b3834be5 | 1550 | /* drain all zone's lru pagevec, this is asynchronous... */ |
0c0e6195 | 1551 | lru_add_drain_all(); |
0c0e6195 | 1552 | yield(); |
b3834be5 | 1553 | /* drain pcp pages, this is synchronous. */ |
9f8f2172 | 1554 | drain_all_pages(); |
0c0e6195 KH |
1555 | /* check again */ |
1556 | offlined_pages = check_pages_isolated(start_pfn, end_pfn); | |
1557 | if (offlined_pages < 0) { | |
1558 | ret = -EBUSY; | |
1559 | goto failed_removal; | |
1560 | } | |
1561 | printk(KERN_INFO "Offlined Pages %ld\n", offlined_pages); | |
b3834be5 | 1562 | /* Ok, all of our target is isolated. |
0c0e6195 KH |
1563 | We cannot do rollback at this point. */ |
1564 | offline_isolated_pages(start_pfn, end_pfn); | |
dbc0e4ce | 1565 | /* reset pagetype flags and makes migrate type to be MOVABLE */ |
0815f3d8 | 1566 | undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE); |
0c0e6195 | 1567 | /* removal success */ |
9feedc9d | 1568 | zone->managed_pages -= offlined_pages; |
0c0e6195 KH |
1569 | zone->present_pages -= offlined_pages; |
1570 | zone->zone_pgdat->node_present_pages -= offlined_pages; | |
1571 | totalram_pages -= offlined_pages; | |
7b78d335 | 1572 | |
1b79acc9 KM |
1573 | init_per_zone_wmark_min(); |
1574 | ||
1e8537ba | 1575 | if (!populated_zone(zone)) { |
340175b7 | 1576 | zone_pcp_reset(zone); |
1e8537ba XQ |
1577 | mutex_lock(&zonelists_mutex); |
1578 | build_all_zonelists(NULL, NULL); | |
1579 | mutex_unlock(&zonelists_mutex); | |
1580 | } else | |
1581 | zone_pcp_update(zone); | |
340175b7 | 1582 | |
d9713679 LJ |
1583 | node_states_clear_node(node, &arg); |
1584 | if (arg.status_change_nid >= 0) | |
8fe23e05 | 1585 | kswapd_stop(node); |
bce7394a | 1586 | |
0c0e6195 KH |
1587 | vm_total_pages = nr_free_pagecache_pages(); |
1588 | writeback_set_ratelimit(); | |
7b78d335 YG |
1589 | |
1590 | memory_notify(MEM_OFFLINE, &arg); | |
20d6c96b | 1591 | unlock_memory_hotplug(); |
0c0e6195 KH |
1592 | return 0; |
1593 | ||
1594 | failed_removal: | |
a62e2f4f BH |
1595 | printk(KERN_INFO "memory offlining [mem %#010llx-%#010llx] failed\n", |
1596 | (unsigned long long) start_pfn << PAGE_SHIFT, | |
1597 | ((unsigned long long) end_pfn << PAGE_SHIFT) - 1); | |
7b78d335 | 1598 | memory_notify(MEM_CANCEL_OFFLINE, &arg); |
0c0e6195 | 1599 | /* pushback to free area */ |
0815f3d8 | 1600 | undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE); |
7b78d335 | 1601 | |
6ad696d2 | 1602 | out: |
20d6c96b | 1603 | unlock_memory_hotplug(); |
0c0e6195 KH |
1604 | return ret; |
1605 | } | |
71088785 | 1606 | |
a16cee10 WC |
1607 | int offline_pages(unsigned long start_pfn, unsigned long nr_pages) |
1608 | { | |
1609 | return __offline_pages(start_pfn, start_pfn + nr_pages, 120 * HZ); | |
1610 | } | |
1611 | ||
bbc76be6 WC |
1612 | /** |
1613 | * walk_memory_range - walks through all mem sections in [start_pfn, end_pfn) | |
1614 | * @start_pfn: start pfn of the memory range | |
1615 | * @end_pfn: end pft of the memory range | |
1616 | * @arg: argument passed to func | |
1617 | * @func: callback for each memory section walked | |
1618 | * | |
1619 | * This function walks through all present mem sections in range | |
1620 | * [start_pfn, end_pfn) and call func on each mem section. | |
1621 | * | |
1622 | * Returns the return value of func. | |
1623 | */ | |
1624 | static int walk_memory_range(unsigned long start_pfn, unsigned long end_pfn, | |
1625 | void *arg, int (*func)(struct memory_block *, void *)) | |
71088785 | 1626 | { |
e90bdb7f WC |
1627 | struct memory_block *mem = NULL; |
1628 | struct mem_section *section; | |
e90bdb7f WC |
1629 | unsigned long pfn, section_nr; |
1630 | int ret; | |
e90bdb7f WC |
1631 | |
1632 | for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) { | |
1633 | section_nr = pfn_to_section_nr(pfn); | |
1634 | if (!present_section_nr(section_nr)) | |
1635 | continue; | |
1636 | ||
1637 | section = __nr_to_section(section_nr); | |
1638 | /* same memblock? */ | |
1639 | if (mem) | |
1640 | if ((section_nr >= mem->start_section_nr) && | |
1641 | (section_nr <= mem->end_section_nr)) | |
1642 | continue; | |
1643 | ||
1644 | mem = find_memory_block_hinted(section, mem); | |
1645 | if (!mem) | |
1646 | continue; | |
1647 | ||
bbc76be6 | 1648 | ret = func(mem, arg); |
e90bdb7f | 1649 | if (ret) { |
bbc76be6 WC |
1650 | kobject_put(&mem->dev.kobj); |
1651 | return ret; | |
e90bdb7f WC |
1652 | } |
1653 | } | |
1654 | ||
1655 | if (mem) | |
1656 | kobject_put(&mem->dev.kobj); | |
1657 | ||
bbc76be6 WC |
1658 | return 0; |
1659 | } | |
1660 | ||
1661 | /** | |
1662 | * offline_memory_block_cb - callback function for offlining memory block | |
1663 | * @mem: the memory block to be offlined | |
1664 | * @arg: buffer to hold error msg | |
1665 | * | |
1666 | * Always return 0, and put the error msg in arg if any. | |
1667 | */ | |
1668 | static int offline_memory_block_cb(struct memory_block *mem, void *arg) | |
1669 | { | |
1670 | int *ret = arg; | |
1671 | int error = offline_memory_block(mem); | |
1672 | ||
1673 | if (error != 0 && *ret == 0) | |
1674 | *ret = error; | |
1675 | ||
1676 | return 0; | |
1677 | } | |
1678 | ||
1679 | static int is_memblock_offlined_cb(struct memory_block *mem, void *arg) | |
1680 | { | |
1681 | int ret = !is_memblock_offlined(mem); | |
1682 | ||
1683 | if (unlikely(ret)) | |
1684 | pr_warn("removing memory fails, because memory " | |
1685 | "[%#010llx-%#010llx] is onlined\n", | |
1686 | PFN_PHYS(section_nr_to_pfn(mem->start_section_nr)), | |
1687 | PFN_PHYS(section_nr_to_pfn(mem->end_section_nr + 1))-1); | |
1688 | ||
1689 | return ret; | |
1690 | } | |
1691 | ||
60a5a19e TC |
1692 | static int check_cpu_on_node(void *data) |
1693 | { | |
1694 | struct pglist_data *pgdat = data; | |
1695 | int cpu; | |
1696 | ||
1697 | for_each_present_cpu(cpu) { | |
1698 | if (cpu_to_node(cpu) == pgdat->node_id) | |
1699 | /* | |
1700 | * the cpu on this node isn't removed, and we can't | |
1701 | * offline this node. | |
1702 | */ | |
1703 | return -EBUSY; | |
1704 | } | |
1705 | ||
1706 | return 0; | |
1707 | } | |
1708 | ||
e13fe869 WC |
1709 | static void unmap_cpu_on_node(void *data) |
1710 | { | |
1711 | #ifdef CONFIG_ACPI_NUMA | |
1712 | struct pglist_data *pgdat = data; | |
1713 | int cpu; | |
1714 | ||
1715 | for_each_possible_cpu(cpu) | |
1716 | if (cpu_to_node(cpu) == pgdat->node_id) | |
1717 | numa_clear_node(cpu); | |
1718 | #endif | |
1719 | } | |
1720 | ||
1721 | static int check_and_unmap_cpu_on_node(void *data) | |
1722 | { | |
1723 | int ret = check_cpu_on_node(data); | |
1724 | ||
1725 | if (ret) | |
1726 | return ret; | |
1727 | ||
1728 | /* | |
1729 | * the node will be offlined when we come here, so we can clear | |
1730 | * the cpu_to_node() now. | |
1731 | */ | |
1732 | ||
1733 | unmap_cpu_on_node(data); | |
1734 | return 0; | |
1735 | } | |
1736 | ||
60a5a19e | 1737 | /* offline the node if all memory sections of this node are removed */ |
90b30cdc | 1738 | void try_offline_node(int nid) |
60a5a19e | 1739 | { |
d822b86a WC |
1740 | pg_data_t *pgdat = NODE_DATA(nid); |
1741 | unsigned long start_pfn = pgdat->node_start_pfn; | |
1742 | unsigned long end_pfn = start_pfn + pgdat->node_spanned_pages; | |
60a5a19e | 1743 | unsigned long pfn; |
d822b86a WC |
1744 | struct page *pgdat_page = virt_to_page(pgdat); |
1745 | int i; | |
60a5a19e TC |
1746 | |
1747 | for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) { | |
1748 | unsigned long section_nr = pfn_to_section_nr(pfn); | |
1749 | ||
1750 | if (!present_section_nr(section_nr)) | |
1751 | continue; | |
1752 | ||
1753 | if (pfn_to_nid(pfn) != nid) | |
1754 | continue; | |
1755 | ||
1756 | /* | |
1757 | * some memory sections of this node are not removed, and we | |
1758 | * can't offline node now. | |
1759 | */ | |
1760 | return; | |
1761 | } | |
1762 | ||
e13fe869 | 1763 | if (stop_machine(check_and_unmap_cpu_on_node, pgdat, NULL)) |
60a5a19e TC |
1764 | return; |
1765 | ||
1766 | /* | |
1767 | * all memory/cpu of this node are removed, we can offline this | |
1768 | * node now. | |
1769 | */ | |
1770 | node_set_offline(nid); | |
1771 | unregister_one_node(nid); | |
d822b86a WC |
1772 | |
1773 | if (!PageSlab(pgdat_page) && !PageCompound(pgdat_page)) | |
1774 | /* node data is allocated from boot memory */ | |
1775 | return; | |
1776 | ||
1777 | /* free waittable in each zone */ | |
1778 | for (i = 0; i < MAX_NR_ZONES; i++) { | |
1779 | struct zone *zone = pgdat->node_zones + i; | |
1780 | ||
1781 | if (zone->wait_table) | |
1782 | vfree(zone->wait_table); | |
1783 | } | |
1784 | ||
1785 | /* | |
1786 | * Since there is no way to guarentee the address of pgdat/zone is not | |
1787 | * on stack of any kernel threads or used by other kernel objects | |
1788 | * without reference counting or other symchronizing method, do not | |
1789 | * reset node_data and free pgdat here. Just reset it to 0 and reuse | |
1790 | * the memory when the node is online again. | |
1791 | */ | |
1792 | memset(pgdat, 0, sizeof(*pgdat)); | |
60a5a19e | 1793 | } |
90b30cdc | 1794 | EXPORT_SYMBOL(try_offline_node); |
60a5a19e TC |
1795 | |
1796 | int __ref remove_memory(int nid, u64 start, u64 size) | |
bbc76be6 WC |
1797 | { |
1798 | unsigned long start_pfn, end_pfn; | |
1799 | int ret = 0; | |
1800 | int retry = 1; | |
1801 | ||
1802 | start_pfn = PFN_DOWN(start); | |
1803 | end_pfn = start_pfn + PFN_DOWN(size); | |
1804 | ||
1805 | /* | |
1806 | * When CONFIG_MEMCG is on, one memory block may be used by other | |
1807 | * blocks to store page cgroup when onlining pages. But we don't know | |
1808 | * in what order pages are onlined. So we iterate twice to offline | |
1809 | * memory: | |
1810 | * 1st iterate: offline every non primary memory block. | |
1811 | * 2nd iterate: offline primary (i.e. first added) memory block. | |
1812 | */ | |
1813 | repeat: | |
1814 | walk_memory_range(start_pfn, end_pfn, &ret, | |
1815 | offline_memory_block_cb); | |
1816 | if (ret) { | |
1817 | if (!retry) | |
1818 | return ret; | |
1819 | ||
1820 | retry = 0; | |
1821 | ret = 0; | |
993c1aad WC |
1822 | goto repeat; |
1823 | } | |
1824 | ||
6677e3ea YI |
1825 | lock_memory_hotplug(); |
1826 | ||
1827 | /* | |
1828 | * we have offlined all memory blocks like this: | |
1829 | * 1. lock memory hotplug | |
1830 | * 2. offline a memory block | |
1831 | * 3. unlock memory hotplug | |
1832 | * | |
1833 | * repeat step1-3 to offline the memory block. All memory blocks | |
1834 | * must be offlined before removing memory. But we don't hold the | |
1835 | * lock in the whole operation. So we should check whether all | |
1836 | * memory blocks are offlined. | |
1837 | */ | |
1838 | ||
bbc76be6 WC |
1839 | ret = walk_memory_range(start_pfn, end_pfn, NULL, |
1840 | is_memblock_offlined_cb); | |
1841 | if (ret) { | |
1842 | unlock_memory_hotplug(); | |
1843 | return ret; | |
6677e3ea YI |
1844 | } |
1845 | ||
46c66c4b YI |
1846 | /* remove memmap entry */ |
1847 | firmware_map_remove(start, start + size, "System RAM"); | |
1848 | ||
24d335ca WC |
1849 | arch_remove_memory(start, size); |
1850 | ||
60a5a19e TC |
1851 | try_offline_node(nid); |
1852 | ||
6677e3ea YI |
1853 | unlock_memory_hotplug(); |
1854 | ||
e90bdb7f | 1855 | return 0; |
71088785 | 1856 | } |
48e94196 | 1857 | #else |
a16cee10 WC |
1858 | int offline_pages(unsigned long start_pfn, unsigned long nr_pages) |
1859 | { | |
1860 | return -EINVAL; | |
1861 | } | |
60a5a19e | 1862 | int remove_memory(int nid, u64 start, u64 size) |
48e94196 KH |
1863 | { |
1864 | return -EINVAL; | |
1865 | } | |
0c0e6195 | 1866 | #endif /* CONFIG_MEMORY_HOTREMOVE */ |
71088785 | 1867 | EXPORT_SYMBOL_GPL(remove_memory); |