]>
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); |
04753278 YG |
127 | } |
128 | ||
129 | } | |
130 | ||
46723bfa YI |
131 | #ifdef CONFIG_HAVE_BOOTMEM_INFO_NODE |
132 | #ifndef CONFIG_SPARSEMEM_VMEMMAP | |
d92bc318 | 133 | static void register_page_bootmem_info_section(unsigned long start_pfn) |
04753278 YG |
134 | { |
135 | unsigned long *usemap, mapsize, section_nr, i; | |
136 | struct mem_section *ms; | |
137 | struct page *page, *memmap; | |
138 | ||
04753278 YG |
139 | section_nr = pfn_to_section_nr(start_pfn); |
140 | ms = __nr_to_section(section_nr); | |
141 | ||
142 | /* Get section's memmap address */ | |
143 | memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr); | |
144 | ||
145 | /* | |
146 | * Get page for the memmap's phys address | |
147 | * XXX: need more consideration for sparse_vmemmap... | |
148 | */ | |
149 | page = virt_to_page(memmap); | |
150 | mapsize = sizeof(struct page) * PAGES_PER_SECTION; | |
151 | mapsize = PAGE_ALIGN(mapsize) >> PAGE_SHIFT; | |
152 | ||
153 | /* remember memmap's page */ | |
154 | for (i = 0; i < mapsize; i++, page++) | |
155 | get_page_bootmem(section_nr, page, SECTION_INFO); | |
156 | ||
157 | usemap = __nr_to_section(section_nr)->pageblock_flags; | |
158 | page = virt_to_page(usemap); | |
159 | ||
160 | mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT; | |
161 | ||
162 | for (i = 0; i < mapsize; i++, page++) | |
af370fb8 | 163 | get_page_bootmem(section_nr, page, MIX_SECTION_INFO); |
04753278 YG |
164 | |
165 | } | |
46723bfa YI |
166 | #else /* CONFIG_SPARSEMEM_VMEMMAP */ |
167 | static void register_page_bootmem_info_section(unsigned long start_pfn) | |
168 | { | |
169 | unsigned long *usemap, mapsize, section_nr, i; | |
170 | struct mem_section *ms; | |
171 | struct page *page, *memmap; | |
172 | ||
173 | if (!pfn_valid(start_pfn)) | |
174 | return; | |
175 | ||
176 | section_nr = pfn_to_section_nr(start_pfn); | |
177 | ms = __nr_to_section(section_nr); | |
178 | ||
179 | memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr); | |
180 | ||
181 | register_page_bootmem_memmap(section_nr, memmap, PAGES_PER_SECTION); | |
182 | ||
183 | usemap = __nr_to_section(section_nr)->pageblock_flags; | |
184 | page = virt_to_page(usemap); | |
185 | ||
186 | mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT; | |
187 | ||
188 | for (i = 0; i < mapsize; i++, page++) | |
189 | get_page_bootmem(section_nr, page, MIX_SECTION_INFO); | |
190 | } | |
191 | #endif /* !CONFIG_SPARSEMEM_VMEMMAP */ | |
04753278 YG |
192 | |
193 | void register_page_bootmem_info_node(struct pglist_data *pgdat) | |
194 | { | |
195 | unsigned long i, pfn, end_pfn, nr_pages; | |
196 | int node = pgdat->node_id; | |
197 | struct page *page; | |
198 | struct zone *zone; | |
199 | ||
200 | nr_pages = PAGE_ALIGN(sizeof(struct pglist_data)) >> PAGE_SHIFT; | |
201 | page = virt_to_page(pgdat); | |
202 | ||
203 | for (i = 0; i < nr_pages; i++, page++) | |
204 | get_page_bootmem(node, page, NODE_INFO); | |
205 | ||
206 | zone = &pgdat->node_zones[0]; | |
207 | for (; zone < pgdat->node_zones + MAX_NR_ZONES - 1; zone++) { | |
208 | if (zone->wait_table) { | |
209 | nr_pages = zone->wait_table_hash_nr_entries | |
210 | * sizeof(wait_queue_head_t); | |
211 | nr_pages = PAGE_ALIGN(nr_pages) >> PAGE_SHIFT; | |
212 | page = virt_to_page(zone->wait_table); | |
213 | ||
214 | for (i = 0; i < nr_pages; i++, page++) | |
215 | get_page_bootmem(node, page, NODE_INFO); | |
216 | } | |
217 | } | |
218 | ||
219 | pfn = pgdat->node_start_pfn; | |
220 | end_pfn = pfn + pgdat->node_spanned_pages; | |
221 | ||
222 | /* register_section info */ | |
f14851af | 223 | for (; pfn < end_pfn; pfn += PAGES_PER_SECTION) { |
224 | /* | |
225 | * Some platforms can assign the same pfn to multiple nodes - on | |
226 | * node0 as well as nodeN. To avoid registering a pfn against | |
227 | * multiple nodes we check that this pfn does not already | |
228 | * reside in some other node. | |
229 | */ | |
230 | if (pfn_valid(pfn) && (pfn_to_nid(pfn) == node)) | |
231 | register_page_bootmem_info_section(pfn); | |
232 | } | |
04753278 | 233 | } |
46723bfa | 234 | #endif /* CONFIG_HAVE_BOOTMEM_INFO_NODE */ |
04753278 | 235 | |
76cdd58e HC |
236 | static void grow_zone_span(struct zone *zone, unsigned long start_pfn, |
237 | unsigned long end_pfn) | |
238 | { | |
239 | unsigned long old_zone_end_pfn; | |
240 | ||
241 | zone_span_writelock(zone); | |
242 | ||
243 | old_zone_end_pfn = zone->zone_start_pfn + zone->spanned_pages; | |
712cd386 | 244 | if (!zone->spanned_pages || start_pfn < zone->zone_start_pfn) |
76cdd58e HC |
245 | zone->zone_start_pfn = start_pfn; |
246 | ||
247 | zone->spanned_pages = max(old_zone_end_pfn, end_pfn) - | |
248 | zone->zone_start_pfn; | |
249 | ||
250 | zone_span_writeunlock(zone); | |
251 | } | |
252 | ||
511c2aba LJ |
253 | static void resize_zone(struct zone *zone, unsigned long start_pfn, |
254 | unsigned long end_pfn) | |
255 | { | |
256 | zone_span_writelock(zone); | |
257 | ||
e455a9b9 LJ |
258 | if (end_pfn - start_pfn) { |
259 | zone->zone_start_pfn = start_pfn; | |
260 | zone->spanned_pages = end_pfn - start_pfn; | |
261 | } else { | |
262 | /* | |
263 | * make it consist as free_area_init_core(), | |
264 | * if spanned_pages = 0, then keep start_pfn = 0 | |
265 | */ | |
266 | zone->zone_start_pfn = 0; | |
267 | zone->spanned_pages = 0; | |
268 | } | |
511c2aba LJ |
269 | |
270 | zone_span_writeunlock(zone); | |
271 | } | |
272 | ||
273 | static void fix_zone_id(struct zone *zone, unsigned long start_pfn, | |
274 | unsigned long end_pfn) | |
275 | { | |
276 | enum zone_type zid = zone_idx(zone); | |
277 | int nid = zone->zone_pgdat->node_id; | |
278 | unsigned long pfn; | |
279 | ||
280 | for (pfn = start_pfn; pfn < end_pfn; pfn++) | |
281 | set_page_links(pfn_to_page(pfn), zid, nid, pfn); | |
282 | } | |
283 | ||
e455a9b9 | 284 | static int __meminit move_pfn_range_left(struct zone *z1, struct zone *z2, |
511c2aba LJ |
285 | unsigned long start_pfn, unsigned long end_pfn) |
286 | { | |
e455a9b9 | 287 | int ret; |
511c2aba | 288 | unsigned long flags; |
e455a9b9 LJ |
289 | unsigned long z1_start_pfn; |
290 | ||
291 | if (!z1->wait_table) { | |
292 | ret = init_currently_empty_zone(z1, start_pfn, | |
293 | end_pfn - start_pfn, MEMMAP_HOTPLUG); | |
294 | if (ret) | |
295 | return ret; | |
296 | } | |
511c2aba LJ |
297 | |
298 | pgdat_resize_lock(z1->zone_pgdat, &flags); | |
299 | ||
300 | /* can't move pfns which are higher than @z2 */ | |
301 | if (end_pfn > z2->zone_start_pfn + z2->spanned_pages) | |
302 | goto out_fail; | |
303 | /* the move out part mast at the left most of @z2 */ | |
304 | if (start_pfn > z2->zone_start_pfn) | |
305 | goto out_fail; | |
306 | /* must included/overlap */ | |
307 | if (end_pfn <= z2->zone_start_pfn) | |
308 | goto out_fail; | |
309 | ||
e455a9b9 LJ |
310 | /* use start_pfn for z1's start_pfn if z1 is empty */ |
311 | if (z1->spanned_pages) | |
312 | z1_start_pfn = z1->zone_start_pfn; | |
313 | else | |
314 | z1_start_pfn = start_pfn; | |
315 | ||
316 | resize_zone(z1, z1_start_pfn, end_pfn); | |
511c2aba LJ |
317 | resize_zone(z2, end_pfn, z2->zone_start_pfn + z2->spanned_pages); |
318 | ||
319 | pgdat_resize_unlock(z1->zone_pgdat, &flags); | |
320 | ||
321 | fix_zone_id(z1, start_pfn, end_pfn); | |
322 | ||
323 | return 0; | |
324 | out_fail: | |
325 | pgdat_resize_unlock(z1->zone_pgdat, &flags); | |
326 | return -1; | |
327 | } | |
328 | ||
e455a9b9 | 329 | static int __meminit move_pfn_range_right(struct zone *z1, struct zone *z2, |
511c2aba LJ |
330 | unsigned long start_pfn, unsigned long end_pfn) |
331 | { | |
e455a9b9 | 332 | int ret; |
511c2aba | 333 | unsigned long flags; |
e455a9b9 LJ |
334 | unsigned long z2_end_pfn; |
335 | ||
336 | if (!z2->wait_table) { | |
337 | ret = init_currently_empty_zone(z2, start_pfn, | |
338 | end_pfn - start_pfn, MEMMAP_HOTPLUG); | |
339 | if (ret) | |
340 | return ret; | |
341 | } | |
511c2aba LJ |
342 | |
343 | pgdat_resize_lock(z1->zone_pgdat, &flags); | |
344 | ||
345 | /* can't move pfns which are lower than @z1 */ | |
346 | if (z1->zone_start_pfn > start_pfn) | |
347 | goto out_fail; | |
348 | /* the move out part mast at the right most of @z1 */ | |
349 | if (z1->zone_start_pfn + z1->spanned_pages > end_pfn) | |
350 | goto out_fail; | |
351 | /* must included/overlap */ | |
352 | if (start_pfn >= z1->zone_start_pfn + z1->spanned_pages) | |
353 | goto out_fail; | |
354 | ||
e455a9b9 LJ |
355 | /* use end_pfn for z2's end_pfn if z2 is empty */ |
356 | if (z2->spanned_pages) | |
357 | z2_end_pfn = z2->zone_start_pfn + z2->spanned_pages; | |
358 | else | |
359 | z2_end_pfn = end_pfn; | |
360 | ||
511c2aba | 361 | resize_zone(z1, z1->zone_start_pfn, start_pfn); |
e455a9b9 | 362 | resize_zone(z2, start_pfn, z2_end_pfn); |
511c2aba LJ |
363 | |
364 | pgdat_resize_unlock(z1->zone_pgdat, &flags); | |
365 | ||
366 | fix_zone_id(z2, start_pfn, end_pfn); | |
367 | ||
368 | return 0; | |
369 | out_fail: | |
370 | pgdat_resize_unlock(z1->zone_pgdat, &flags); | |
371 | return -1; | |
372 | } | |
373 | ||
76cdd58e HC |
374 | static void grow_pgdat_span(struct pglist_data *pgdat, unsigned long start_pfn, |
375 | unsigned long end_pfn) | |
376 | { | |
377 | unsigned long old_pgdat_end_pfn = | |
378 | pgdat->node_start_pfn + pgdat->node_spanned_pages; | |
379 | ||
712cd386 | 380 | if (!pgdat->node_spanned_pages || start_pfn < pgdat->node_start_pfn) |
76cdd58e HC |
381 | pgdat->node_start_pfn = start_pfn; |
382 | ||
383 | pgdat->node_spanned_pages = max(old_pgdat_end_pfn, end_pfn) - | |
384 | pgdat->node_start_pfn; | |
385 | } | |
386 | ||
31168481 | 387 | static int __meminit __add_zone(struct zone *zone, unsigned long phys_start_pfn) |
3947be19 DH |
388 | { |
389 | struct pglist_data *pgdat = zone->zone_pgdat; | |
390 | int nr_pages = PAGES_PER_SECTION; | |
391 | int nid = pgdat->node_id; | |
392 | int zone_type; | |
76cdd58e | 393 | unsigned long flags; |
3947be19 DH |
394 | |
395 | zone_type = zone - pgdat->node_zones; | |
76cdd58e HC |
396 | if (!zone->wait_table) { |
397 | int ret; | |
398 | ||
399 | ret = init_currently_empty_zone(zone, phys_start_pfn, | |
400 | nr_pages, MEMMAP_HOTPLUG); | |
401 | if (ret) | |
402 | return ret; | |
403 | } | |
404 | pgdat_resize_lock(zone->zone_pgdat, &flags); | |
405 | grow_zone_span(zone, phys_start_pfn, phys_start_pfn + nr_pages); | |
406 | grow_pgdat_span(zone->zone_pgdat, phys_start_pfn, | |
407 | phys_start_pfn + nr_pages); | |
408 | pgdat_resize_unlock(zone->zone_pgdat, &flags); | |
a2f3aa02 DH |
409 | memmap_init_zone(nr_pages, nid, zone_type, |
410 | phys_start_pfn, MEMMAP_HOTPLUG); | |
718127cc | 411 | return 0; |
3947be19 DH |
412 | } |
413 | ||
c04fc586 GH |
414 | static int __meminit __add_section(int nid, struct zone *zone, |
415 | unsigned long phys_start_pfn) | |
3947be19 | 416 | { |
3947be19 | 417 | int nr_pages = PAGES_PER_SECTION; |
3947be19 DH |
418 | int ret; |
419 | ||
ebd15302 KH |
420 | if (pfn_valid(phys_start_pfn)) |
421 | return -EEXIST; | |
422 | ||
0b0acbec | 423 | ret = sparse_add_one_section(zone, phys_start_pfn, nr_pages); |
3947be19 DH |
424 | |
425 | if (ret < 0) | |
426 | return ret; | |
427 | ||
718127cc YG |
428 | ret = __add_zone(zone, phys_start_pfn); |
429 | ||
430 | if (ret < 0) | |
431 | return ret; | |
432 | ||
c04fc586 | 433 | return register_new_memory(nid, __pfn_to_section(phys_start_pfn)); |
3947be19 DH |
434 | } |
435 | ||
815121d2 YI |
436 | /* find the smallest valid pfn in the range [start_pfn, end_pfn) */ |
437 | static int find_smallest_section_pfn(int nid, struct zone *zone, | |
438 | unsigned long start_pfn, | |
439 | unsigned long end_pfn) | |
440 | { | |
441 | struct mem_section *ms; | |
442 | ||
443 | for (; start_pfn < end_pfn; start_pfn += PAGES_PER_SECTION) { | |
444 | ms = __pfn_to_section(start_pfn); | |
445 | ||
446 | if (unlikely(!valid_section(ms))) | |
447 | continue; | |
448 | ||
449 | if (unlikely(pfn_to_nid(start_pfn) != nid)) | |
450 | continue; | |
451 | ||
452 | if (zone && zone != page_zone(pfn_to_page(start_pfn))) | |
453 | continue; | |
454 | ||
455 | return start_pfn; | |
456 | } | |
457 | ||
458 | return 0; | |
459 | } | |
460 | ||
461 | /* find the biggest valid pfn in the range [start_pfn, end_pfn). */ | |
462 | static int find_biggest_section_pfn(int nid, struct zone *zone, | |
463 | unsigned long start_pfn, | |
464 | unsigned long end_pfn) | |
465 | { | |
466 | struct mem_section *ms; | |
467 | unsigned long pfn; | |
468 | ||
469 | /* pfn is the end pfn of a memory section. */ | |
470 | pfn = end_pfn - 1; | |
471 | for (; pfn >= start_pfn; pfn -= PAGES_PER_SECTION) { | |
472 | ms = __pfn_to_section(pfn); | |
473 | ||
474 | if (unlikely(!valid_section(ms))) | |
475 | continue; | |
476 | ||
477 | if (unlikely(pfn_to_nid(pfn) != nid)) | |
478 | continue; | |
479 | ||
480 | if (zone && zone != page_zone(pfn_to_page(pfn))) | |
481 | continue; | |
482 | ||
483 | return pfn; | |
484 | } | |
485 | ||
486 | return 0; | |
487 | } | |
488 | ||
489 | static void shrink_zone_span(struct zone *zone, unsigned long start_pfn, | |
490 | unsigned long end_pfn) | |
491 | { | |
492 | unsigned long zone_start_pfn = zone->zone_start_pfn; | |
493 | unsigned long zone_end_pfn = zone->zone_start_pfn + zone->spanned_pages; | |
494 | unsigned long pfn; | |
495 | struct mem_section *ms; | |
496 | int nid = zone_to_nid(zone); | |
497 | ||
498 | zone_span_writelock(zone); | |
499 | if (zone_start_pfn == start_pfn) { | |
500 | /* | |
501 | * If the section is smallest section in the zone, it need | |
502 | * shrink zone->zone_start_pfn and zone->zone_spanned_pages. | |
503 | * In this case, we find second smallest valid mem_section | |
504 | * for shrinking zone. | |
505 | */ | |
506 | pfn = find_smallest_section_pfn(nid, zone, end_pfn, | |
507 | zone_end_pfn); | |
508 | if (pfn) { | |
509 | zone->zone_start_pfn = pfn; | |
510 | zone->spanned_pages = zone_end_pfn - pfn; | |
511 | } | |
512 | } else if (zone_end_pfn == end_pfn) { | |
513 | /* | |
514 | * If the section is biggest section in the zone, it need | |
515 | * shrink zone->spanned_pages. | |
516 | * In this case, we find second biggest valid mem_section for | |
517 | * shrinking zone. | |
518 | */ | |
519 | pfn = find_biggest_section_pfn(nid, zone, zone_start_pfn, | |
520 | start_pfn); | |
521 | if (pfn) | |
522 | zone->spanned_pages = pfn - zone_start_pfn + 1; | |
523 | } | |
524 | ||
525 | /* | |
526 | * The section is not biggest or smallest mem_section in the zone, it | |
527 | * only creates a hole in the zone. So in this case, we need not | |
528 | * change the zone. But perhaps, the zone has only hole data. Thus | |
529 | * it check the zone has only hole or not. | |
530 | */ | |
531 | pfn = zone_start_pfn; | |
532 | for (; pfn < zone_end_pfn; pfn += PAGES_PER_SECTION) { | |
533 | ms = __pfn_to_section(pfn); | |
534 | ||
535 | if (unlikely(!valid_section(ms))) | |
536 | continue; | |
537 | ||
538 | if (page_zone(pfn_to_page(pfn)) != zone) | |
539 | continue; | |
540 | ||
541 | /* If the section is current section, it continues the loop */ | |
542 | if (start_pfn == pfn) | |
543 | continue; | |
544 | ||
545 | /* If we find valid section, we have nothing to do */ | |
546 | zone_span_writeunlock(zone); | |
547 | return; | |
548 | } | |
549 | ||
550 | /* The zone has no valid section */ | |
551 | zone->zone_start_pfn = 0; | |
552 | zone->spanned_pages = 0; | |
553 | zone_span_writeunlock(zone); | |
554 | } | |
555 | ||
556 | static void shrink_pgdat_span(struct pglist_data *pgdat, | |
557 | unsigned long start_pfn, unsigned long end_pfn) | |
558 | { | |
559 | unsigned long pgdat_start_pfn = pgdat->node_start_pfn; | |
560 | unsigned long pgdat_end_pfn = | |
561 | pgdat->node_start_pfn + pgdat->node_spanned_pages; | |
562 | unsigned long pfn; | |
563 | struct mem_section *ms; | |
564 | int nid = pgdat->node_id; | |
565 | ||
566 | if (pgdat_start_pfn == start_pfn) { | |
567 | /* | |
568 | * If the section is smallest section in the pgdat, it need | |
569 | * shrink pgdat->node_start_pfn and pgdat->node_spanned_pages. | |
570 | * In this case, we find second smallest valid mem_section | |
571 | * for shrinking zone. | |
572 | */ | |
573 | pfn = find_smallest_section_pfn(nid, NULL, end_pfn, | |
574 | pgdat_end_pfn); | |
575 | if (pfn) { | |
576 | pgdat->node_start_pfn = pfn; | |
577 | pgdat->node_spanned_pages = pgdat_end_pfn - pfn; | |
578 | } | |
579 | } else if (pgdat_end_pfn == end_pfn) { | |
580 | /* | |
581 | * If the section is biggest section in the pgdat, it need | |
582 | * shrink pgdat->node_spanned_pages. | |
583 | * In this case, we find second biggest valid mem_section for | |
584 | * shrinking zone. | |
585 | */ | |
586 | pfn = find_biggest_section_pfn(nid, NULL, pgdat_start_pfn, | |
587 | start_pfn); | |
588 | if (pfn) | |
589 | pgdat->node_spanned_pages = pfn - pgdat_start_pfn + 1; | |
590 | } | |
591 | ||
592 | /* | |
593 | * If the section is not biggest or smallest mem_section in the pgdat, | |
594 | * it only creates a hole in the pgdat. So in this case, we need not | |
595 | * change the pgdat. | |
596 | * But perhaps, the pgdat has only hole data. Thus it check the pgdat | |
597 | * has only hole or not. | |
598 | */ | |
599 | pfn = pgdat_start_pfn; | |
600 | for (; pfn < pgdat_end_pfn; pfn += PAGES_PER_SECTION) { | |
601 | ms = __pfn_to_section(pfn); | |
602 | ||
603 | if (unlikely(!valid_section(ms))) | |
604 | continue; | |
605 | ||
606 | if (pfn_to_nid(pfn) != nid) | |
607 | continue; | |
608 | ||
609 | /* If the section is current section, it continues the loop */ | |
610 | if (start_pfn == pfn) | |
611 | continue; | |
612 | ||
613 | /* If we find valid section, we have nothing to do */ | |
614 | return; | |
615 | } | |
616 | ||
617 | /* The pgdat has no valid section */ | |
618 | pgdat->node_start_pfn = 0; | |
619 | pgdat->node_spanned_pages = 0; | |
620 | } | |
621 | ||
622 | static void __remove_zone(struct zone *zone, unsigned long start_pfn) | |
623 | { | |
624 | struct pglist_data *pgdat = zone->zone_pgdat; | |
625 | int nr_pages = PAGES_PER_SECTION; | |
626 | int zone_type; | |
627 | unsigned long flags; | |
628 | ||
629 | zone_type = zone - pgdat->node_zones; | |
630 | ||
631 | pgdat_resize_lock(zone->zone_pgdat, &flags); | |
632 | shrink_zone_span(zone, start_pfn, start_pfn + nr_pages); | |
633 | shrink_pgdat_span(pgdat, start_pfn, start_pfn + nr_pages); | |
634 | pgdat_resize_unlock(zone->zone_pgdat, &flags); | |
635 | } | |
636 | ||
ea01ea93 BP |
637 | static int __remove_section(struct zone *zone, struct mem_section *ms) |
638 | { | |
815121d2 YI |
639 | unsigned long start_pfn; |
640 | int scn_nr; | |
ea01ea93 BP |
641 | int ret = -EINVAL; |
642 | ||
643 | if (!valid_section(ms)) | |
644 | return ret; | |
645 | ||
646 | ret = unregister_memory_section(ms); | |
647 | if (ret) | |
648 | return ret; | |
649 | ||
815121d2 YI |
650 | scn_nr = __section_nr(ms); |
651 | start_pfn = section_nr_to_pfn(scn_nr); | |
652 | __remove_zone(zone, start_pfn); | |
653 | ||
ea01ea93 | 654 | sparse_remove_one_section(zone, ms); |
ea01ea93 BP |
655 | return 0; |
656 | } | |
657 | ||
3947be19 DH |
658 | /* |
659 | * Reasonably generic function for adding memory. It is | |
660 | * expected that archs that support memory hotplug will | |
661 | * call this function after deciding the zone to which to | |
662 | * add the new pages. | |
663 | */ | |
c04fc586 GH |
664 | int __ref __add_pages(int nid, struct zone *zone, unsigned long phys_start_pfn, |
665 | unsigned long nr_pages) | |
3947be19 DH |
666 | { |
667 | unsigned long i; | |
668 | int err = 0; | |
6f712711 KH |
669 | int start_sec, end_sec; |
670 | /* during initialize mem_map, align hot-added range to section */ | |
671 | start_sec = pfn_to_section_nr(phys_start_pfn); | |
672 | end_sec = pfn_to_section_nr(phys_start_pfn + nr_pages - 1); | |
3947be19 | 673 | |
6f712711 | 674 | for (i = start_sec; i <= end_sec; i++) { |
c04fc586 | 675 | err = __add_section(nid, zone, i << PFN_SECTION_SHIFT); |
3947be19 | 676 | |
6f712711 | 677 | /* |
183ff22b | 678 | * EEXIST is finally dealt with by ioresource collision |
6f712711 KH |
679 | * check. see add_memory() => register_memory_resource() |
680 | * Warning will be printed if there is collision. | |
bed120c6 JS |
681 | */ |
682 | if (err && (err != -EEXIST)) | |
3947be19 | 683 | break; |
6f712711 | 684 | err = 0; |
3947be19 DH |
685 | } |
686 | ||
687 | return err; | |
688 | } | |
bed120c6 | 689 | EXPORT_SYMBOL_GPL(__add_pages); |
3947be19 | 690 | |
ea01ea93 BP |
691 | /** |
692 | * __remove_pages() - remove sections of pages from a zone | |
693 | * @zone: zone from which pages need to be removed | |
694 | * @phys_start_pfn: starting pageframe (must be aligned to start of a section) | |
695 | * @nr_pages: number of pages to remove (must be multiple of section size) | |
696 | * | |
697 | * Generic helper function to remove section mappings and sysfs entries | |
698 | * for the section of the memory we are removing. Caller needs to make | |
699 | * sure that pages are marked reserved and zones are adjust properly by | |
700 | * calling offline_pages(). | |
701 | */ | |
702 | int __remove_pages(struct zone *zone, unsigned long phys_start_pfn, | |
703 | unsigned long nr_pages) | |
704 | { | |
705 | unsigned long i, ret = 0; | |
706 | int sections_to_remove; | |
707 | ||
708 | /* | |
709 | * We can only remove entire sections | |
710 | */ | |
711 | BUG_ON(phys_start_pfn & ~PAGE_SECTION_MASK); | |
712 | BUG_ON(nr_pages % PAGES_PER_SECTION); | |
713 | ||
d760afd4 YI |
714 | release_mem_region(phys_start_pfn << PAGE_SHIFT, nr_pages * PAGE_SIZE); |
715 | ||
ea01ea93 BP |
716 | sections_to_remove = nr_pages / PAGES_PER_SECTION; |
717 | for (i = 0; i < sections_to_remove; i++) { | |
718 | unsigned long pfn = phys_start_pfn + i*PAGES_PER_SECTION; | |
719 | ret = __remove_section(zone, __pfn_to_section(pfn)); | |
720 | if (ret) | |
721 | break; | |
722 | } | |
723 | return ret; | |
724 | } | |
725 | EXPORT_SYMBOL_GPL(__remove_pages); | |
726 | ||
9d0ad8ca DK |
727 | int set_online_page_callback(online_page_callback_t callback) |
728 | { | |
729 | int rc = -EINVAL; | |
730 | ||
731 | lock_memory_hotplug(); | |
732 | ||
733 | if (online_page_callback == generic_online_page) { | |
734 | online_page_callback = callback; | |
735 | rc = 0; | |
736 | } | |
737 | ||
738 | unlock_memory_hotplug(); | |
739 | ||
740 | return rc; | |
741 | } | |
742 | EXPORT_SYMBOL_GPL(set_online_page_callback); | |
743 | ||
744 | int restore_online_page_callback(online_page_callback_t callback) | |
745 | { | |
746 | int rc = -EINVAL; | |
747 | ||
748 | lock_memory_hotplug(); | |
749 | ||
750 | if (online_page_callback == callback) { | |
751 | online_page_callback = generic_online_page; | |
752 | rc = 0; | |
753 | } | |
754 | ||
755 | unlock_memory_hotplug(); | |
756 | ||
757 | return rc; | |
758 | } | |
759 | EXPORT_SYMBOL_GPL(restore_online_page_callback); | |
760 | ||
761 | void __online_page_set_limits(struct page *page) | |
180c06ef | 762 | { |
4738e1b9 JB |
763 | unsigned long pfn = page_to_pfn(page); |
764 | ||
4738e1b9 JB |
765 | if (pfn >= num_physpages) |
766 | num_physpages = pfn + 1; | |
9d0ad8ca DK |
767 | } |
768 | EXPORT_SYMBOL_GPL(__online_page_set_limits); | |
769 | ||
770 | void __online_page_increment_counters(struct page *page) | |
771 | { | |
772 | totalram_pages++; | |
180c06ef JF |
773 | |
774 | #ifdef CONFIG_HIGHMEM | |
775 | if (PageHighMem(page)) | |
776 | totalhigh_pages++; | |
777 | #endif | |
9d0ad8ca DK |
778 | } |
779 | EXPORT_SYMBOL_GPL(__online_page_increment_counters); | |
180c06ef | 780 | |
9d0ad8ca DK |
781 | void __online_page_free(struct page *page) |
782 | { | |
180c06ef JF |
783 | ClearPageReserved(page); |
784 | init_page_count(page); | |
785 | __free_page(page); | |
786 | } | |
9d0ad8ca DK |
787 | EXPORT_SYMBOL_GPL(__online_page_free); |
788 | ||
789 | static void generic_online_page(struct page *page) | |
790 | { | |
791 | __online_page_set_limits(page); | |
792 | __online_page_increment_counters(page); | |
793 | __online_page_free(page); | |
794 | } | |
180c06ef | 795 | |
75884fb1 KH |
796 | static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages, |
797 | void *arg) | |
3947be19 DH |
798 | { |
799 | unsigned long i; | |
75884fb1 KH |
800 | unsigned long onlined_pages = *(unsigned long *)arg; |
801 | struct page *page; | |
802 | if (PageReserved(pfn_to_page(start_pfn))) | |
803 | for (i = 0; i < nr_pages; i++) { | |
804 | page = pfn_to_page(start_pfn + i); | |
9d0ad8ca | 805 | (*online_page_callback)(page); |
75884fb1 KH |
806 | onlined_pages++; |
807 | } | |
808 | *(unsigned long *)arg = onlined_pages; | |
809 | return 0; | |
810 | } | |
811 | ||
09285af7 | 812 | #ifdef CONFIG_MOVABLE_NODE |
79a4dcef TC |
813 | /* |
814 | * When CONFIG_MOVABLE_NODE, we permit onlining of a node which doesn't have | |
815 | * normal memory. | |
816 | */ | |
09285af7 LJ |
817 | static bool can_online_high_movable(struct zone *zone) |
818 | { | |
819 | return true; | |
820 | } | |
79a4dcef | 821 | #else /* CONFIG_MOVABLE_NODE */ |
74d42d8f LJ |
822 | /* ensure every online node has NORMAL memory */ |
823 | static bool can_online_high_movable(struct zone *zone) | |
824 | { | |
825 | return node_state(zone_to_nid(zone), N_NORMAL_MEMORY); | |
826 | } | |
79a4dcef | 827 | #endif /* CONFIG_MOVABLE_NODE */ |
74d42d8f | 828 | |
d9713679 LJ |
829 | /* check which state of node_states will be changed when online memory */ |
830 | static void node_states_check_changes_online(unsigned long nr_pages, | |
831 | struct zone *zone, struct memory_notify *arg) | |
832 | { | |
833 | int nid = zone_to_nid(zone); | |
834 | enum zone_type zone_last = ZONE_NORMAL; | |
835 | ||
836 | /* | |
6715ddf9 LJ |
837 | * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY] |
838 | * contains nodes which have zones of 0...ZONE_NORMAL, | |
839 | * set zone_last to ZONE_NORMAL. | |
d9713679 | 840 | * |
6715ddf9 LJ |
841 | * If we don't have HIGHMEM nor movable node, |
842 | * node_states[N_NORMAL_MEMORY] contains nodes which have zones of | |
843 | * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE. | |
d9713679 | 844 | */ |
6715ddf9 | 845 | if (N_MEMORY == N_NORMAL_MEMORY) |
d9713679 LJ |
846 | zone_last = ZONE_MOVABLE; |
847 | ||
848 | /* | |
849 | * if the memory to be online is in a zone of 0...zone_last, and | |
850 | * the zones of 0...zone_last don't have memory before online, we will | |
851 | * need to set the node to node_states[N_NORMAL_MEMORY] after | |
852 | * the memory is online. | |
853 | */ | |
854 | if (zone_idx(zone) <= zone_last && !node_state(nid, N_NORMAL_MEMORY)) | |
855 | arg->status_change_nid_normal = nid; | |
856 | else | |
857 | arg->status_change_nid_normal = -1; | |
858 | ||
6715ddf9 LJ |
859 | #ifdef CONFIG_HIGHMEM |
860 | /* | |
861 | * If we have movable node, node_states[N_HIGH_MEMORY] | |
862 | * contains nodes which have zones of 0...ZONE_HIGHMEM, | |
863 | * set zone_last to ZONE_HIGHMEM. | |
864 | * | |
865 | * If we don't have movable node, node_states[N_NORMAL_MEMORY] | |
866 | * contains nodes which have zones of 0...ZONE_MOVABLE, | |
867 | * set zone_last to ZONE_MOVABLE. | |
868 | */ | |
869 | zone_last = ZONE_HIGHMEM; | |
870 | if (N_MEMORY == N_HIGH_MEMORY) | |
871 | zone_last = ZONE_MOVABLE; | |
872 | ||
873 | if (zone_idx(zone) <= zone_last && !node_state(nid, N_HIGH_MEMORY)) | |
874 | arg->status_change_nid_high = nid; | |
875 | else | |
876 | arg->status_change_nid_high = -1; | |
877 | #else | |
878 | arg->status_change_nid_high = arg->status_change_nid_normal; | |
879 | #endif | |
880 | ||
d9713679 LJ |
881 | /* |
882 | * if the node don't have memory befor online, we will need to | |
6715ddf9 | 883 | * set the node to node_states[N_MEMORY] after the memory |
d9713679 LJ |
884 | * is online. |
885 | */ | |
6715ddf9 | 886 | if (!node_state(nid, N_MEMORY)) |
d9713679 LJ |
887 | arg->status_change_nid = nid; |
888 | else | |
889 | arg->status_change_nid = -1; | |
890 | } | |
891 | ||
892 | static void node_states_set_node(int node, struct memory_notify *arg) | |
893 | { | |
894 | if (arg->status_change_nid_normal >= 0) | |
895 | node_set_state(node, N_NORMAL_MEMORY); | |
896 | ||
6715ddf9 LJ |
897 | if (arg->status_change_nid_high >= 0) |
898 | node_set_state(node, N_HIGH_MEMORY); | |
899 | ||
900 | node_set_state(node, N_MEMORY); | |
d9713679 LJ |
901 | } |
902 | ||
75884fb1 | 903 | |
511c2aba | 904 | int __ref online_pages(unsigned long pfn, unsigned long nr_pages, int online_type) |
75884fb1 | 905 | { |
3947be19 DH |
906 | unsigned long onlined_pages = 0; |
907 | struct zone *zone; | |
6811378e | 908 | int need_zonelists_rebuild = 0; |
7b78d335 YG |
909 | int nid; |
910 | int ret; | |
911 | struct memory_notify arg; | |
912 | ||
925268a0 | 913 | lock_memory_hotplug(); |
d9713679 LJ |
914 | /* |
915 | * This doesn't need a lock to do pfn_to_page(). | |
916 | * The section can't be removed here because of the | |
917 | * memory_block->state_mutex. | |
918 | */ | |
919 | zone = page_zone(pfn_to_page(pfn)); | |
920 | ||
74d42d8f LJ |
921 | if ((zone_idx(zone) > ZONE_NORMAL || online_type == ONLINE_MOVABLE) && |
922 | !can_online_high_movable(zone)) { | |
923 | unlock_memory_hotplug(); | |
924 | return -1; | |
925 | } | |
926 | ||
511c2aba LJ |
927 | if (online_type == ONLINE_KERNEL && zone_idx(zone) == ZONE_MOVABLE) { |
928 | if (move_pfn_range_left(zone - 1, zone, pfn, pfn + nr_pages)) { | |
929 | unlock_memory_hotplug(); | |
930 | return -1; | |
931 | } | |
932 | } | |
933 | if (online_type == ONLINE_MOVABLE && zone_idx(zone) == ZONE_MOVABLE - 1) { | |
934 | if (move_pfn_range_right(zone, zone + 1, pfn, pfn + nr_pages)) { | |
935 | unlock_memory_hotplug(); | |
936 | return -1; | |
937 | } | |
938 | } | |
939 | ||
940 | /* Previous code may changed the zone of the pfn range */ | |
941 | zone = page_zone(pfn_to_page(pfn)); | |
942 | ||
7b78d335 YG |
943 | arg.start_pfn = pfn; |
944 | arg.nr_pages = nr_pages; | |
d9713679 | 945 | node_states_check_changes_online(nr_pages, zone, &arg); |
7b78d335 YG |
946 | |
947 | nid = page_to_nid(pfn_to_page(pfn)); | |
3947be19 | 948 | |
7b78d335 YG |
949 | ret = memory_notify(MEM_GOING_ONLINE, &arg); |
950 | ret = notifier_to_errno(ret); | |
951 | if (ret) { | |
952 | memory_notify(MEM_CANCEL_ONLINE, &arg); | |
925268a0 | 953 | unlock_memory_hotplug(); |
7b78d335 YG |
954 | return ret; |
955 | } | |
6811378e YG |
956 | /* |
957 | * If this zone is not populated, then it is not in zonelist. | |
958 | * This means the page allocator ignores this zone. | |
959 | * So, zonelist must be updated after online. | |
960 | */ | |
4eaf3f64 | 961 | mutex_lock(&zonelists_mutex); |
6dcd73d7 | 962 | if (!populated_zone(zone)) { |
6811378e | 963 | need_zonelists_rebuild = 1; |
6dcd73d7 WC |
964 | build_all_zonelists(NULL, zone); |
965 | } | |
6811378e | 966 | |
908eedc6 | 967 | ret = walk_system_ram_range(pfn, nr_pages, &onlined_pages, |
75884fb1 | 968 | online_pages_range); |
fd8a4221 | 969 | if (ret) { |
6dcd73d7 WC |
970 | if (need_zonelists_rebuild) |
971 | zone_pcp_reset(zone); | |
4eaf3f64 | 972 | mutex_unlock(&zonelists_mutex); |
a62e2f4f BH |
973 | printk(KERN_DEBUG "online_pages [mem %#010llx-%#010llx] failed\n", |
974 | (unsigned long long) pfn << PAGE_SHIFT, | |
975 | (((unsigned long long) pfn + nr_pages) | |
976 | << PAGE_SHIFT) - 1); | |
fd8a4221 | 977 | memory_notify(MEM_CANCEL_ONLINE, &arg); |
925268a0 | 978 | unlock_memory_hotplug(); |
fd8a4221 GL |
979 | return ret; |
980 | } | |
981 | ||
9feedc9d | 982 | zone->managed_pages += onlined_pages; |
3947be19 | 983 | zone->present_pages += onlined_pages; |
f2937be5 | 984 | zone->zone_pgdat->node_present_pages += onlined_pages; |
08dff7b7 | 985 | if (onlined_pages) { |
d9713679 | 986 | node_states_set_node(zone_to_nid(zone), &arg); |
08dff7b7 | 987 | if (need_zonelists_rebuild) |
6dcd73d7 | 988 | build_all_zonelists(NULL, NULL); |
08dff7b7 JL |
989 | else |
990 | zone_pcp_update(zone); | |
991 | } | |
3947be19 | 992 | |
4eaf3f64 | 993 | mutex_unlock(&zonelists_mutex); |
1b79acc9 KM |
994 | |
995 | init_per_zone_wmark_min(); | |
996 | ||
08dff7b7 | 997 | if (onlined_pages) |
7ea1530a | 998 | kswapd_run(zone_to_nid(zone)); |
61b13993 | 999 | |
1f522509 | 1000 | vm_total_pages = nr_free_pagecache_pages(); |
2f7f24ec | 1001 | |
2d1d43f6 | 1002 | writeback_set_ratelimit(); |
7b78d335 YG |
1003 | |
1004 | if (onlined_pages) | |
1005 | memory_notify(MEM_ONLINE, &arg); | |
925268a0 | 1006 | unlock_memory_hotplug(); |
7b78d335 | 1007 | |
3947be19 DH |
1008 | return 0; |
1009 | } | |
53947027 | 1010 | #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */ |
bc02af93 | 1011 | |
e1319331 HS |
1012 | /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */ |
1013 | static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start) | |
9af3c2de YG |
1014 | { |
1015 | struct pglist_data *pgdat; | |
1016 | unsigned long zones_size[MAX_NR_ZONES] = {0}; | |
1017 | unsigned long zholes_size[MAX_NR_ZONES] = {0}; | |
1018 | unsigned long start_pfn = start >> PAGE_SHIFT; | |
1019 | ||
1020 | pgdat = arch_alloc_nodedata(nid); | |
1021 | if (!pgdat) | |
1022 | return NULL; | |
1023 | ||
1024 | arch_refresh_nodedata(nid, pgdat); | |
1025 | ||
1026 | /* we can use NODE_DATA(nid) from here */ | |
1027 | ||
1028 | /* init node's zones as empty zones, we don't have any present pages.*/ | |
9109fb7b | 1029 | free_area_init_node(nid, zones_size, start_pfn, zholes_size); |
9af3c2de | 1030 | |
959ecc48 KH |
1031 | /* |
1032 | * The node we allocated has no zone fallback lists. For avoiding | |
1033 | * to access not-initialized zonelist, build here. | |
1034 | */ | |
f957db4f | 1035 | mutex_lock(&zonelists_mutex); |
9adb62a5 | 1036 | build_all_zonelists(pgdat, NULL); |
f957db4f | 1037 | mutex_unlock(&zonelists_mutex); |
959ecc48 | 1038 | |
9af3c2de YG |
1039 | return pgdat; |
1040 | } | |
1041 | ||
1042 | static void rollback_node_hotadd(int nid, pg_data_t *pgdat) | |
1043 | { | |
1044 | arch_refresh_nodedata(nid, NULL); | |
1045 | arch_free_nodedata(pgdat); | |
1046 | return; | |
1047 | } | |
1048 | ||
0a547039 | 1049 | |
cf23422b | 1050 | /* |
1051 | * called by cpu_up() to online a node without onlined memory. | |
1052 | */ | |
1053 | int mem_online_node(int nid) | |
1054 | { | |
1055 | pg_data_t *pgdat; | |
1056 | int ret; | |
1057 | ||
20d6c96b | 1058 | lock_memory_hotplug(); |
cf23422b | 1059 | pgdat = hotadd_new_pgdat(nid, 0); |
7553e8f2 | 1060 | if (!pgdat) { |
cf23422b | 1061 | ret = -ENOMEM; |
1062 | goto out; | |
1063 | } | |
1064 | node_set_online(nid); | |
1065 | ret = register_one_node(nid); | |
1066 | BUG_ON(ret); | |
1067 | ||
1068 | out: | |
20d6c96b | 1069 | unlock_memory_hotplug(); |
cf23422b | 1070 | return ret; |
1071 | } | |
1072 | ||
31168481 AV |
1073 | /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */ |
1074 | int __ref add_memory(int nid, u64 start, u64 size) | |
bc02af93 | 1075 | { |
9af3c2de YG |
1076 | pg_data_t *pgdat = NULL; |
1077 | int new_pgdat = 0; | |
ebd15302 | 1078 | struct resource *res; |
bc02af93 YG |
1079 | int ret; |
1080 | ||
20d6c96b | 1081 | lock_memory_hotplug(); |
6ad696d2 | 1082 | |
ebd15302 | 1083 | res = register_memory_resource(start, size); |
6ad696d2 | 1084 | ret = -EEXIST; |
ebd15302 | 1085 | if (!res) |
6ad696d2 | 1086 | goto out; |
ebd15302 | 1087 | |
9af3c2de YG |
1088 | if (!node_online(nid)) { |
1089 | pgdat = hotadd_new_pgdat(nid, start); | |
6ad696d2 | 1090 | ret = -ENOMEM; |
9af3c2de | 1091 | if (!pgdat) |
41b9e2d7 | 1092 | goto error; |
9af3c2de | 1093 | new_pgdat = 1; |
9af3c2de YG |
1094 | } |
1095 | ||
bc02af93 YG |
1096 | /* call arch's memory hotadd */ |
1097 | ret = arch_add_memory(nid, start, size); | |
1098 | ||
9af3c2de YG |
1099 | if (ret < 0) |
1100 | goto error; | |
1101 | ||
0fc44159 | 1102 | /* we online node here. we can't roll back from here. */ |
9af3c2de YG |
1103 | node_set_online(nid); |
1104 | ||
0fc44159 YG |
1105 | if (new_pgdat) { |
1106 | ret = register_one_node(nid); | |
1107 | /* | |
1108 | * If sysfs file of new node can't create, cpu on the node | |
1109 | * can't be hot-added. There is no rollback way now. | |
1110 | * So, check by BUG_ON() to catch it reluctantly.. | |
1111 | */ | |
1112 | BUG_ON(ret); | |
1113 | } | |
1114 | ||
d96ae530 AM |
1115 | /* create new memmap entry */ |
1116 | firmware_map_add_hotplug(start, start + size, "System RAM"); | |
1117 | ||
6ad696d2 AK |
1118 | goto out; |
1119 | ||
9af3c2de YG |
1120 | error: |
1121 | /* rollback pgdat allocation and others */ | |
1122 | if (new_pgdat) | |
1123 | rollback_node_hotadd(nid, pgdat); | |
a864b9d0 | 1124 | release_memory_resource(res); |
9af3c2de | 1125 | |
6ad696d2 | 1126 | out: |
20d6c96b | 1127 | unlock_memory_hotplug(); |
bc02af93 YG |
1128 | return ret; |
1129 | } | |
1130 | EXPORT_SYMBOL_GPL(add_memory); | |
0c0e6195 KH |
1131 | |
1132 | #ifdef CONFIG_MEMORY_HOTREMOVE | |
5c755e9f BP |
1133 | /* |
1134 | * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy | |
1135 | * set and the size of the free page is given by page_order(). Using this, | |
1136 | * the function determines if the pageblock contains only free pages. | |
1137 | * Due to buddy contraints, a free page at least the size of a pageblock will | |
1138 | * be located at the start of the pageblock | |
1139 | */ | |
1140 | static inline int pageblock_free(struct page *page) | |
1141 | { | |
1142 | return PageBuddy(page) && page_order(page) >= pageblock_order; | |
1143 | } | |
1144 | ||
1145 | /* Return the start of the next active pageblock after a given page */ | |
1146 | static struct page *next_active_pageblock(struct page *page) | |
1147 | { | |
5c755e9f BP |
1148 | /* Ensure the starting page is pageblock-aligned */ |
1149 | BUG_ON(page_to_pfn(page) & (pageblock_nr_pages - 1)); | |
1150 | ||
5c755e9f | 1151 | /* If the entire pageblock is free, move to the end of free page */ |
0dcc48c1 KH |
1152 | if (pageblock_free(page)) { |
1153 | int order; | |
1154 | /* be careful. we don't have locks, page_order can be changed.*/ | |
1155 | order = page_order(page); | |
1156 | if ((order < MAX_ORDER) && (order >= pageblock_order)) | |
1157 | return page + (1 << order); | |
1158 | } | |
5c755e9f | 1159 | |
0dcc48c1 | 1160 | return page + pageblock_nr_pages; |
5c755e9f BP |
1161 | } |
1162 | ||
1163 | /* Checks if this range of memory is likely to be hot-removable. */ | |
1164 | int is_mem_section_removable(unsigned long start_pfn, unsigned long nr_pages) | |
1165 | { | |
5c755e9f BP |
1166 | struct page *page = pfn_to_page(start_pfn); |
1167 | struct page *end_page = page + nr_pages; | |
1168 | ||
1169 | /* Check the starting page of each pageblock within the range */ | |
1170 | for (; page < end_page; page = next_active_pageblock(page)) { | |
49ac8255 | 1171 | if (!is_pageblock_removable_nolock(page)) |
5c755e9f | 1172 | return 0; |
49ac8255 | 1173 | cond_resched(); |
5c755e9f BP |
1174 | } |
1175 | ||
1176 | /* All pageblocks in the memory block are likely to be hot-removable */ | |
1177 | return 1; | |
1178 | } | |
1179 | ||
0c0e6195 KH |
1180 | /* |
1181 | * Confirm all pages in a range [start, end) is belongs to the same zone. | |
1182 | */ | |
1183 | static int test_pages_in_a_zone(unsigned long start_pfn, unsigned long end_pfn) | |
1184 | { | |
1185 | unsigned long pfn; | |
1186 | struct zone *zone = NULL; | |
1187 | struct page *page; | |
1188 | int i; | |
1189 | for (pfn = start_pfn; | |
1190 | pfn < end_pfn; | |
1191 | pfn += MAX_ORDER_NR_PAGES) { | |
1192 | i = 0; | |
1193 | /* This is just a CONFIG_HOLES_IN_ZONE check.*/ | |
1194 | while ((i < MAX_ORDER_NR_PAGES) && !pfn_valid_within(pfn + i)) | |
1195 | i++; | |
1196 | if (i == MAX_ORDER_NR_PAGES) | |
1197 | continue; | |
1198 | page = pfn_to_page(pfn + i); | |
1199 | if (zone && page_zone(page) != zone) | |
1200 | return 0; | |
1201 | zone = page_zone(page); | |
1202 | } | |
1203 | return 1; | |
1204 | } | |
1205 | ||
1206 | /* | |
1207 | * Scanning pfn is much easier than scanning lru list. | |
1208 | * Scan pfn from start to end and Find LRU page. | |
1209 | */ | |
7bbc0905 | 1210 | static unsigned long scan_lru_pages(unsigned long start, unsigned long end) |
0c0e6195 KH |
1211 | { |
1212 | unsigned long pfn; | |
1213 | struct page *page; | |
1214 | for (pfn = start; pfn < end; pfn++) { | |
1215 | if (pfn_valid(pfn)) { | |
1216 | page = pfn_to_page(pfn); | |
1217 | if (PageLRU(page)) | |
1218 | return pfn; | |
1219 | } | |
1220 | } | |
1221 | return 0; | |
1222 | } | |
1223 | ||
0c0e6195 KH |
1224 | #define NR_OFFLINE_AT_ONCE_PAGES (256) |
1225 | static int | |
1226 | do_migrate_range(unsigned long start_pfn, unsigned long end_pfn) | |
1227 | { | |
1228 | unsigned long pfn; | |
1229 | struct page *page; | |
1230 | int move_pages = NR_OFFLINE_AT_ONCE_PAGES; | |
1231 | int not_managed = 0; | |
1232 | int ret = 0; | |
1233 | LIST_HEAD(source); | |
1234 | ||
1235 | for (pfn = start_pfn; pfn < end_pfn && move_pages > 0; pfn++) { | |
1236 | if (!pfn_valid(pfn)) | |
1237 | continue; | |
1238 | page = pfn_to_page(pfn); | |
700c2a46 | 1239 | if (!get_page_unless_zero(page)) |
0c0e6195 KH |
1240 | continue; |
1241 | /* | |
1242 | * We can skip free pages. And we can only deal with pages on | |
1243 | * LRU. | |
1244 | */ | |
62695a84 | 1245 | ret = isolate_lru_page(page); |
0c0e6195 | 1246 | if (!ret) { /* Success */ |
700c2a46 | 1247 | put_page(page); |
62695a84 | 1248 | list_add_tail(&page->lru, &source); |
0c0e6195 | 1249 | move_pages--; |
6d9c285a KM |
1250 | inc_zone_page_state(page, NR_ISOLATED_ANON + |
1251 | page_is_file_cache(page)); | |
1252 | ||
0c0e6195 | 1253 | } else { |
0c0e6195 | 1254 | #ifdef CONFIG_DEBUG_VM |
718a3821 WF |
1255 | printk(KERN_ALERT "removing pfn %lx from LRU failed\n", |
1256 | pfn); | |
1257 | dump_page(page); | |
0c0e6195 | 1258 | #endif |
700c2a46 | 1259 | put_page(page); |
25985edc | 1260 | /* Because we don't have big zone->lock. we should |
809c4449 BL |
1261 | check this again here. */ |
1262 | if (page_count(page)) { | |
1263 | not_managed++; | |
f3ab2636 | 1264 | ret = -EBUSY; |
809c4449 BL |
1265 | break; |
1266 | } | |
0c0e6195 KH |
1267 | } |
1268 | } | |
f3ab2636 BL |
1269 | if (!list_empty(&source)) { |
1270 | if (not_managed) { | |
1271 | putback_lru_pages(&source); | |
1272 | goto out; | |
1273 | } | |
74c08f98 MK |
1274 | |
1275 | /* | |
1276 | * alloc_migrate_target should be improooooved!! | |
1277 | * migrate_pages returns # of failed pages. | |
1278 | */ | |
1279 | ret = migrate_pages(&source, alloc_migrate_target, 0, | |
7b2a2d4a MG |
1280 | true, MIGRATE_SYNC, |
1281 | MR_MEMORY_HOTPLUG); | |
f3ab2636 | 1282 | if (ret) |
0c0e6195 | 1283 | putback_lru_pages(&source); |
0c0e6195 | 1284 | } |
0c0e6195 KH |
1285 | out: |
1286 | return ret; | |
1287 | } | |
1288 | ||
1289 | /* | |
1290 | * remove from free_area[] and mark all as Reserved. | |
1291 | */ | |
1292 | static int | |
1293 | offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages, | |
1294 | void *data) | |
1295 | { | |
1296 | __offline_isolated_pages(start, start + nr_pages); | |
1297 | return 0; | |
1298 | } | |
1299 | ||
1300 | static void | |
1301 | offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn) | |
1302 | { | |
908eedc6 | 1303 | walk_system_ram_range(start_pfn, end_pfn - start_pfn, NULL, |
0c0e6195 KH |
1304 | offline_isolated_pages_cb); |
1305 | } | |
1306 | ||
1307 | /* | |
1308 | * Check all pages in range, recoreded as memory resource, are isolated. | |
1309 | */ | |
1310 | static int | |
1311 | check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages, | |
1312 | void *data) | |
1313 | { | |
1314 | int ret; | |
1315 | long offlined = *(long *)data; | |
b023f468 | 1316 | ret = test_pages_isolated(start_pfn, start_pfn + nr_pages, true); |
0c0e6195 KH |
1317 | offlined = nr_pages; |
1318 | if (!ret) | |
1319 | *(long *)data += offlined; | |
1320 | return ret; | |
1321 | } | |
1322 | ||
1323 | static long | |
1324 | check_pages_isolated(unsigned long start_pfn, unsigned long end_pfn) | |
1325 | { | |
1326 | long offlined = 0; | |
1327 | int ret; | |
1328 | ||
908eedc6 | 1329 | ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn, &offlined, |
0c0e6195 KH |
1330 | check_pages_isolated_cb); |
1331 | if (ret < 0) | |
1332 | offlined = (long)ret; | |
1333 | return offlined; | |
1334 | } | |
1335 | ||
09285af7 | 1336 | #ifdef CONFIG_MOVABLE_NODE |
79a4dcef TC |
1337 | /* |
1338 | * When CONFIG_MOVABLE_NODE, we permit offlining of a node which doesn't have | |
1339 | * normal memory. | |
1340 | */ | |
09285af7 LJ |
1341 | static bool can_offline_normal(struct zone *zone, unsigned long nr_pages) |
1342 | { | |
1343 | return true; | |
1344 | } | |
79a4dcef | 1345 | #else /* CONFIG_MOVABLE_NODE */ |
74d42d8f LJ |
1346 | /* ensure the node has NORMAL memory if it is still online */ |
1347 | static bool can_offline_normal(struct zone *zone, unsigned long nr_pages) | |
1348 | { | |
1349 | struct pglist_data *pgdat = zone->zone_pgdat; | |
1350 | unsigned long present_pages = 0; | |
1351 | enum zone_type zt; | |
1352 | ||
1353 | for (zt = 0; zt <= ZONE_NORMAL; zt++) | |
1354 | present_pages += pgdat->node_zones[zt].present_pages; | |
1355 | ||
1356 | if (present_pages > nr_pages) | |
1357 | return true; | |
1358 | ||
1359 | present_pages = 0; | |
1360 | for (; zt <= ZONE_MOVABLE; zt++) | |
1361 | present_pages += pgdat->node_zones[zt].present_pages; | |
1362 | ||
1363 | /* | |
1364 | * we can't offline the last normal memory until all | |
1365 | * higher memory is offlined. | |
1366 | */ | |
1367 | return present_pages == 0; | |
1368 | } | |
79a4dcef | 1369 | #endif /* CONFIG_MOVABLE_NODE */ |
74d42d8f | 1370 | |
d9713679 LJ |
1371 | /* check which state of node_states will be changed when offline memory */ |
1372 | static void node_states_check_changes_offline(unsigned long nr_pages, | |
1373 | struct zone *zone, struct memory_notify *arg) | |
1374 | { | |
1375 | struct pglist_data *pgdat = zone->zone_pgdat; | |
1376 | unsigned long present_pages = 0; | |
1377 | enum zone_type zt, zone_last = ZONE_NORMAL; | |
1378 | ||
1379 | /* | |
6715ddf9 LJ |
1380 | * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY] |
1381 | * contains nodes which have zones of 0...ZONE_NORMAL, | |
1382 | * set zone_last to ZONE_NORMAL. | |
d9713679 | 1383 | * |
6715ddf9 LJ |
1384 | * If we don't have HIGHMEM nor movable node, |
1385 | * node_states[N_NORMAL_MEMORY] contains nodes which have zones of | |
1386 | * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE. | |
d9713679 | 1387 | */ |
6715ddf9 | 1388 | if (N_MEMORY == N_NORMAL_MEMORY) |
d9713679 LJ |
1389 | zone_last = ZONE_MOVABLE; |
1390 | ||
1391 | /* | |
1392 | * check whether node_states[N_NORMAL_MEMORY] will be changed. | |
1393 | * If the memory to be offline is in a zone of 0...zone_last, | |
1394 | * and it is the last present memory, 0...zone_last will | |
1395 | * become empty after offline , thus we can determind we will | |
1396 | * need to clear the node from node_states[N_NORMAL_MEMORY]. | |
1397 | */ | |
1398 | for (zt = 0; zt <= zone_last; zt++) | |
1399 | present_pages += pgdat->node_zones[zt].present_pages; | |
1400 | if (zone_idx(zone) <= zone_last && nr_pages >= present_pages) | |
1401 | arg->status_change_nid_normal = zone_to_nid(zone); | |
1402 | else | |
1403 | arg->status_change_nid_normal = -1; | |
1404 | ||
6715ddf9 LJ |
1405 | #ifdef CONFIG_HIGHMEM |
1406 | /* | |
1407 | * If we have movable node, node_states[N_HIGH_MEMORY] | |
1408 | * contains nodes which have zones of 0...ZONE_HIGHMEM, | |
1409 | * set zone_last to ZONE_HIGHMEM. | |
1410 | * | |
1411 | * If we don't have movable node, node_states[N_NORMAL_MEMORY] | |
1412 | * contains nodes which have zones of 0...ZONE_MOVABLE, | |
1413 | * set zone_last to ZONE_MOVABLE. | |
1414 | */ | |
1415 | zone_last = ZONE_HIGHMEM; | |
1416 | if (N_MEMORY == N_HIGH_MEMORY) | |
1417 | zone_last = ZONE_MOVABLE; | |
1418 | ||
1419 | for (; zt <= zone_last; zt++) | |
1420 | present_pages += pgdat->node_zones[zt].present_pages; | |
1421 | if (zone_idx(zone) <= zone_last && nr_pages >= present_pages) | |
1422 | arg->status_change_nid_high = zone_to_nid(zone); | |
1423 | else | |
1424 | arg->status_change_nid_high = -1; | |
1425 | #else | |
1426 | arg->status_change_nid_high = arg->status_change_nid_normal; | |
1427 | #endif | |
1428 | ||
d9713679 LJ |
1429 | /* |
1430 | * node_states[N_HIGH_MEMORY] contains nodes which have 0...ZONE_MOVABLE | |
1431 | */ | |
1432 | zone_last = ZONE_MOVABLE; | |
1433 | ||
1434 | /* | |
1435 | * check whether node_states[N_HIGH_MEMORY] will be changed | |
1436 | * If we try to offline the last present @nr_pages from the node, | |
1437 | * we can determind we will need to clear the node from | |
1438 | * node_states[N_HIGH_MEMORY]. | |
1439 | */ | |
1440 | for (; zt <= zone_last; zt++) | |
1441 | present_pages += pgdat->node_zones[zt].present_pages; | |
1442 | if (nr_pages >= present_pages) | |
1443 | arg->status_change_nid = zone_to_nid(zone); | |
1444 | else | |
1445 | arg->status_change_nid = -1; | |
1446 | } | |
1447 | ||
1448 | static void node_states_clear_node(int node, struct memory_notify *arg) | |
1449 | { | |
1450 | if (arg->status_change_nid_normal >= 0) | |
1451 | node_clear_state(node, N_NORMAL_MEMORY); | |
1452 | ||
6715ddf9 LJ |
1453 | if ((N_MEMORY != N_NORMAL_MEMORY) && |
1454 | (arg->status_change_nid_high >= 0)) | |
d9713679 | 1455 | node_clear_state(node, N_HIGH_MEMORY); |
6715ddf9 LJ |
1456 | |
1457 | if ((N_MEMORY != N_HIGH_MEMORY) && | |
1458 | (arg->status_change_nid >= 0)) | |
1459 | node_clear_state(node, N_MEMORY); | |
d9713679 LJ |
1460 | } |
1461 | ||
a16cee10 | 1462 | static int __ref __offline_pages(unsigned long start_pfn, |
0c0e6195 KH |
1463 | unsigned long end_pfn, unsigned long timeout) |
1464 | { | |
1465 | unsigned long pfn, nr_pages, expire; | |
1466 | long offlined_pages; | |
7b78d335 | 1467 | int ret, drain, retry_max, node; |
0c0e6195 | 1468 | struct zone *zone; |
7b78d335 | 1469 | struct memory_notify arg; |
0c0e6195 KH |
1470 | |
1471 | BUG_ON(start_pfn >= end_pfn); | |
1472 | /* at least, alignment against pageblock is necessary */ | |
1473 | if (!IS_ALIGNED(start_pfn, pageblock_nr_pages)) | |
1474 | return -EINVAL; | |
1475 | if (!IS_ALIGNED(end_pfn, pageblock_nr_pages)) | |
1476 | return -EINVAL; | |
1477 | /* This makes hotplug much easier...and readable. | |
1478 | we assume this for now. .*/ | |
1479 | if (!test_pages_in_a_zone(start_pfn, end_pfn)) | |
1480 | return -EINVAL; | |
7b78d335 | 1481 | |
20d6c96b | 1482 | lock_memory_hotplug(); |
6ad696d2 | 1483 | |
7b78d335 YG |
1484 | zone = page_zone(pfn_to_page(start_pfn)); |
1485 | node = zone_to_nid(zone); | |
1486 | nr_pages = end_pfn - start_pfn; | |
1487 | ||
74d42d8f LJ |
1488 | ret = -EINVAL; |
1489 | if (zone_idx(zone) <= ZONE_NORMAL && !can_offline_normal(zone, nr_pages)) | |
1490 | goto out; | |
1491 | ||
0c0e6195 | 1492 | /* set above range as isolated */ |
b023f468 WC |
1493 | ret = start_isolate_page_range(start_pfn, end_pfn, |
1494 | MIGRATE_MOVABLE, true); | |
0c0e6195 | 1495 | if (ret) |
6ad696d2 | 1496 | goto out; |
7b78d335 YG |
1497 | |
1498 | arg.start_pfn = start_pfn; | |
1499 | arg.nr_pages = nr_pages; | |
d9713679 | 1500 | node_states_check_changes_offline(nr_pages, zone, &arg); |
7b78d335 YG |
1501 | |
1502 | ret = memory_notify(MEM_GOING_OFFLINE, &arg); | |
1503 | ret = notifier_to_errno(ret); | |
1504 | if (ret) | |
1505 | goto failed_removal; | |
1506 | ||
0c0e6195 KH |
1507 | pfn = start_pfn; |
1508 | expire = jiffies + timeout; | |
1509 | drain = 0; | |
1510 | retry_max = 5; | |
1511 | repeat: | |
1512 | /* start memory hot removal */ | |
1513 | ret = -EAGAIN; | |
1514 | if (time_after(jiffies, expire)) | |
1515 | goto failed_removal; | |
1516 | ret = -EINTR; | |
1517 | if (signal_pending(current)) | |
1518 | goto failed_removal; | |
1519 | ret = 0; | |
1520 | if (drain) { | |
1521 | lru_add_drain_all(); | |
0c0e6195 | 1522 | cond_resched(); |
9f8f2172 | 1523 | drain_all_pages(); |
0c0e6195 KH |
1524 | } |
1525 | ||
1526 | pfn = scan_lru_pages(start_pfn, end_pfn); | |
1527 | if (pfn) { /* We have page on LRU */ | |
1528 | ret = do_migrate_range(pfn, end_pfn); | |
1529 | if (!ret) { | |
1530 | drain = 1; | |
1531 | goto repeat; | |
1532 | } else { | |
1533 | if (ret < 0) | |
1534 | if (--retry_max == 0) | |
1535 | goto failed_removal; | |
1536 | yield(); | |
1537 | drain = 1; | |
1538 | goto repeat; | |
1539 | } | |
1540 | } | |
b3834be5 | 1541 | /* drain all zone's lru pagevec, this is asynchronous... */ |
0c0e6195 | 1542 | lru_add_drain_all(); |
0c0e6195 | 1543 | yield(); |
b3834be5 | 1544 | /* drain pcp pages, this is synchronous. */ |
9f8f2172 | 1545 | drain_all_pages(); |
0c0e6195 KH |
1546 | /* check again */ |
1547 | offlined_pages = check_pages_isolated(start_pfn, end_pfn); | |
1548 | if (offlined_pages < 0) { | |
1549 | ret = -EBUSY; | |
1550 | goto failed_removal; | |
1551 | } | |
1552 | printk(KERN_INFO "Offlined Pages %ld\n", offlined_pages); | |
b3834be5 | 1553 | /* Ok, all of our target is isolated. |
0c0e6195 KH |
1554 | We cannot do rollback at this point. */ |
1555 | offline_isolated_pages(start_pfn, end_pfn); | |
dbc0e4ce | 1556 | /* reset pagetype flags and makes migrate type to be MOVABLE */ |
0815f3d8 | 1557 | undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE); |
0c0e6195 | 1558 | /* removal success */ |
9feedc9d | 1559 | zone->managed_pages -= offlined_pages; |
0c0e6195 KH |
1560 | zone->present_pages -= offlined_pages; |
1561 | zone->zone_pgdat->node_present_pages -= offlined_pages; | |
1562 | totalram_pages -= offlined_pages; | |
7b78d335 | 1563 | |
1b79acc9 KM |
1564 | init_per_zone_wmark_min(); |
1565 | ||
1e8537ba | 1566 | if (!populated_zone(zone)) { |
340175b7 | 1567 | zone_pcp_reset(zone); |
1e8537ba XQ |
1568 | mutex_lock(&zonelists_mutex); |
1569 | build_all_zonelists(NULL, NULL); | |
1570 | mutex_unlock(&zonelists_mutex); | |
1571 | } else | |
1572 | zone_pcp_update(zone); | |
340175b7 | 1573 | |
d9713679 LJ |
1574 | node_states_clear_node(node, &arg); |
1575 | if (arg.status_change_nid >= 0) | |
8fe23e05 | 1576 | kswapd_stop(node); |
bce7394a | 1577 | |
0c0e6195 KH |
1578 | vm_total_pages = nr_free_pagecache_pages(); |
1579 | writeback_set_ratelimit(); | |
7b78d335 YG |
1580 | |
1581 | memory_notify(MEM_OFFLINE, &arg); | |
20d6c96b | 1582 | unlock_memory_hotplug(); |
0c0e6195 KH |
1583 | return 0; |
1584 | ||
1585 | failed_removal: | |
a62e2f4f BH |
1586 | printk(KERN_INFO "memory offlining [mem %#010llx-%#010llx] failed\n", |
1587 | (unsigned long long) start_pfn << PAGE_SHIFT, | |
1588 | ((unsigned long long) end_pfn << PAGE_SHIFT) - 1); | |
7b78d335 | 1589 | memory_notify(MEM_CANCEL_OFFLINE, &arg); |
0c0e6195 | 1590 | /* pushback to free area */ |
0815f3d8 | 1591 | undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE); |
7b78d335 | 1592 | |
6ad696d2 | 1593 | out: |
20d6c96b | 1594 | unlock_memory_hotplug(); |
0c0e6195 KH |
1595 | return ret; |
1596 | } | |
71088785 | 1597 | |
a16cee10 WC |
1598 | int offline_pages(unsigned long start_pfn, unsigned long nr_pages) |
1599 | { | |
1600 | return __offline_pages(start_pfn, start_pfn + nr_pages, 120 * HZ); | |
1601 | } | |
1602 | ||
bbc76be6 WC |
1603 | /** |
1604 | * walk_memory_range - walks through all mem sections in [start_pfn, end_pfn) | |
1605 | * @start_pfn: start pfn of the memory range | |
1606 | * @end_pfn: end pft of the memory range | |
1607 | * @arg: argument passed to func | |
1608 | * @func: callback for each memory section walked | |
1609 | * | |
1610 | * This function walks through all present mem sections in range | |
1611 | * [start_pfn, end_pfn) and call func on each mem section. | |
1612 | * | |
1613 | * Returns the return value of func. | |
1614 | */ | |
1615 | static int walk_memory_range(unsigned long start_pfn, unsigned long end_pfn, | |
1616 | void *arg, int (*func)(struct memory_block *, void *)) | |
71088785 | 1617 | { |
e90bdb7f WC |
1618 | struct memory_block *mem = NULL; |
1619 | struct mem_section *section; | |
e90bdb7f WC |
1620 | unsigned long pfn, section_nr; |
1621 | int ret; | |
e90bdb7f WC |
1622 | |
1623 | for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) { | |
1624 | section_nr = pfn_to_section_nr(pfn); | |
1625 | if (!present_section_nr(section_nr)) | |
1626 | continue; | |
1627 | ||
1628 | section = __nr_to_section(section_nr); | |
1629 | /* same memblock? */ | |
1630 | if (mem) | |
1631 | if ((section_nr >= mem->start_section_nr) && | |
1632 | (section_nr <= mem->end_section_nr)) | |
1633 | continue; | |
1634 | ||
1635 | mem = find_memory_block_hinted(section, mem); | |
1636 | if (!mem) | |
1637 | continue; | |
1638 | ||
bbc76be6 | 1639 | ret = func(mem, arg); |
e90bdb7f | 1640 | if (ret) { |
bbc76be6 WC |
1641 | kobject_put(&mem->dev.kobj); |
1642 | return ret; | |
e90bdb7f WC |
1643 | } |
1644 | } | |
1645 | ||
1646 | if (mem) | |
1647 | kobject_put(&mem->dev.kobj); | |
1648 | ||
bbc76be6 WC |
1649 | return 0; |
1650 | } | |
1651 | ||
1652 | /** | |
1653 | * offline_memory_block_cb - callback function for offlining memory block | |
1654 | * @mem: the memory block to be offlined | |
1655 | * @arg: buffer to hold error msg | |
1656 | * | |
1657 | * Always return 0, and put the error msg in arg if any. | |
1658 | */ | |
1659 | static int offline_memory_block_cb(struct memory_block *mem, void *arg) | |
1660 | { | |
1661 | int *ret = arg; | |
1662 | int error = offline_memory_block(mem); | |
1663 | ||
1664 | if (error != 0 && *ret == 0) | |
1665 | *ret = error; | |
1666 | ||
1667 | return 0; | |
1668 | } | |
1669 | ||
1670 | static int is_memblock_offlined_cb(struct memory_block *mem, void *arg) | |
1671 | { | |
1672 | int ret = !is_memblock_offlined(mem); | |
1673 | ||
1674 | if (unlikely(ret)) | |
1675 | pr_warn("removing memory fails, because memory " | |
1676 | "[%#010llx-%#010llx] is onlined\n", | |
1677 | PFN_PHYS(section_nr_to_pfn(mem->start_section_nr)), | |
1678 | PFN_PHYS(section_nr_to_pfn(mem->end_section_nr + 1))-1); | |
1679 | ||
1680 | return ret; | |
1681 | } | |
1682 | ||
60a5a19e TC |
1683 | static int check_cpu_on_node(void *data) |
1684 | { | |
1685 | struct pglist_data *pgdat = data; | |
1686 | int cpu; | |
1687 | ||
1688 | for_each_present_cpu(cpu) { | |
1689 | if (cpu_to_node(cpu) == pgdat->node_id) | |
1690 | /* | |
1691 | * the cpu on this node isn't removed, and we can't | |
1692 | * offline this node. | |
1693 | */ | |
1694 | return -EBUSY; | |
1695 | } | |
1696 | ||
1697 | return 0; | |
1698 | } | |
1699 | ||
1700 | /* offline the node if all memory sections of this node are removed */ | |
1701 | static void try_offline_node(int nid) | |
1702 | { | |
d822b86a WC |
1703 | pg_data_t *pgdat = NODE_DATA(nid); |
1704 | unsigned long start_pfn = pgdat->node_start_pfn; | |
1705 | unsigned long end_pfn = start_pfn + pgdat->node_spanned_pages; | |
60a5a19e | 1706 | unsigned long pfn; |
d822b86a WC |
1707 | struct page *pgdat_page = virt_to_page(pgdat); |
1708 | int i; | |
60a5a19e TC |
1709 | |
1710 | for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) { | |
1711 | unsigned long section_nr = pfn_to_section_nr(pfn); | |
1712 | ||
1713 | if (!present_section_nr(section_nr)) | |
1714 | continue; | |
1715 | ||
1716 | if (pfn_to_nid(pfn) != nid) | |
1717 | continue; | |
1718 | ||
1719 | /* | |
1720 | * some memory sections of this node are not removed, and we | |
1721 | * can't offline node now. | |
1722 | */ | |
1723 | return; | |
1724 | } | |
1725 | ||
d822b86a | 1726 | if (stop_machine(check_cpu_on_node, pgdat, NULL)) |
60a5a19e TC |
1727 | return; |
1728 | ||
1729 | /* | |
1730 | * all memory/cpu of this node are removed, we can offline this | |
1731 | * node now. | |
1732 | */ | |
1733 | node_set_offline(nid); | |
1734 | unregister_one_node(nid); | |
d822b86a WC |
1735 | |
1736 | if (!PageSlab(pgdat_page) && !PageCompound(pgdat_page)) | |
1737 | /* node data is allocated from boot memory */ | |
1738 | return; | |
1739 | ||
1740 | /* free waittable in each zone */ | |
1741 | for (i = 0; i < MAX_NR_ZONES; i++) { | |
1742 | struct zone *zone = pgdat->node_zones + i; | |
1743 | ||
1744 | if (zone->wait_table) | |
1745 | vfree(zone->wait_table); | |
1746 | } | |
1747 | ||
1748 | /* | |
1749 | * Since there is no way to guarentee the address of pgdat/zone is not | |
1750 | * on stack of any kernel threads or used by other kernel objects | |
1751 | * without reference counting or other symchronizing method, do not | |
1752 | * reset node_data and free pgdat here. Just reset it to 0 and reuse | |
1753 | * the memory when the node is online again. | |
1754 | */ | |
1755 | memset(pgdat, 0, sizeof(*pgdat)); | |
60a5a19e TC |
1756 | } |
1757 | ||
1758 | int __ref remove_memory(int nid, u64 start, u64 size) | |
bbc76be6 WC |
1759 | { |
1760 | unsigned long start_pfn, end_pfn; | |
1761 | int ret = 0; | |
1762 | int retry = 1; | |
1763 | ||
1764 | start_pfn = PFN_DOWN(start); | |
1765 | end_pfn = start_pfn + PFN_DOWN(size); | |
1766 | ||
1767 | /* | |
1768 | * When CONFIG_MEMCG is on, one memory block may be used by other | |
1769 | * blocks to store page cgroup when onlining pages. But we don't know | |
1770 | * in what order pages are onlined. So we iterate twice to offline | |
1771 | * memory: | |
1772 | * 1st iterate: offline every non primary memory block. | |
1773 | * 2nd iterate: offline primary (i.e. first added) memory block. | |
1774 | */ | |
1775 | repeat: | |
1776 | walk_memory_range(start_pfn, end_pfn, &ret, | |
1777 | offline_memory_block_cb); | |
1778 | if (ret) { | |
1779 | if (!retry) | |
1780 | return ret; | |
1781 | ||
1782 | retry = 0; | |
1783 | ret = 0; | |
993c1aad WC |
1784 | goto repeat; |
1785 | } | |
1786 | ||
6677e3ea YI |
1787 | lock_memory_hotplug(); |
1788 | ||
1789 | /* | |
1790 | * we have offlined all memory blocks like this: | |
1791 | * 1. lock memory hotplug | |
1792 | * 2. offline a memory block | |
1793 | * 3. unlock memory hotplug | |
1794 | * | |
1795 | * repeat step1-3 to offline the memory block. All memory blocks | |
1796 | * must be offlined before removing memory. But we don't hold the | |
1797 | * lock in the whole operation. So we should check whether all | |
1798 | * memory blocks are offlined. | |
1799 | */ | |
1800 | ||
bbc76be6 WC |
1801 | ret = walk_memory_range(start_pfn, end_pfn, NULL, |
1802 | is_memblock_offlined_cb); | |
1803 | if (ret) { | |
1804 | unlock_memory_hotplug(); | |
1805 | return ret; | |
6677e3ea YI |
1806 | } |
1807 | ||
46c66c4b YI |
1808 | /* remove memmap entry */ |
1809 | firmware_map_remove(start, start + size, "System RAM"); | |
1810 | ||
24d335ca WC |
1811 | arch_remove_memory(start, size); |
1812 | ||
60a5a19e TC |
1813 | try_offline_node(nid); |
1814 | ||
6677e3ea YI |
1815 | unlock_memory_hotplug(); |
1816 | ||
e90bdb7f | 1817 | return 0; |
71088785 | 1818 | } |
48e94196 | 1819 | #else |
a16cee10 WC |
1820 | int offline_pages(unsigned long start_pfn, unsigned long nr_pages) |
1821 | { | |
1822 | return -EINVAL; | |
1823 | } | |
60a5a19e | 1824 | int remove_memory(int nid, u64 start, u64 size) |
48e94196 KH |
1825 | { |
1826 | return -EINVAL; | |
1827 | } | |
0c0e6195 | 1828 | #endif /* CONFIG_MEMORY_HOTREMOVE */ |
71088785 | 1829 | EXPORT_SYMBOL_GPL(remove_memory); |