]>
Commit | Line | Data |
---|---|---|
3947be19 DH |
1 | /* |
2 | * linux/mm/memory_hotplug.c | |
3 | * | |
4 | * Copyright (C) | |
5 | */ | |
6 | ||
3947be19 DH |
7 | #include <linux/stddef.h> |
8 | #include <linux/mm.h> | |
9 | #include <linux/swap.h> | |
10 | #include <linux/interrupt.h> | |
11 | #include <linux/pagemap.h> | |
12 | #include <linux/bootmem.h> | |
13 | #include <linux/compiler.h> | |
b95f1b31 | 14 | #include <linux/export.h> |
3947be19 | 15 | #include <linux/pagevec.h> |
2d1d43f6 | 16 | #include <linux/writeback.h> |
3947be19 DH |
17 | #include <linux/slab.h> |
18 | #include <linux/sysctl.h> | |
19 | #include <linux/cpu.h> | |
20 | #include <linux/memory.h> | |
21 | #include <linux/memory_hotplug.h> | |
22 | #include <linux/highmem.h> | |
23 | #include <linux/vmalloc.h> | |
0a547039 | 24 | #include <linux/ioport.h> |
0c0e6195 KH |
25 | #include <linux/delay.h> |
26 | #include <linux/migrate.h> | |
27 | #include <linux/page-isolation.h> | |
71088785 | 28 | #include <linux/pfn.h> |
6ad696d2 | 29 | #include <linux/suspend.h> |
6d9c285a | 30 | #include <linux/mm_inline.h> |
d96ae530 | 31 | #include <linux/firmware-map.h> |
3947be19 DH |
32 | |
33 | #include <asm/tlbflush.h> | |
34 | ||
1e5ad9a3 AB |
35 | #include "internal.h" |
36 | ||
9d0ad8ca DK |
37 | /* |
38 | * online_page_callback contains pointer to current page onlining function. | |
39 | * Initially it is generic_online_page(). If it is required it could be | |
40 | * changed by calling set_online_page_callback() for callback registration | |
41 | * and restore_online_page_callback() for generic callback restore. | |
42 | */ | |
43 | ||
44 | static void generic_online_page(struct page *page); | |
45 | ||
46 | static online_page_callback_t online_page_callback = generic_online_page; | |
47 | ||
20d6c96b KM |
48 | DEFINE_MUTEX(mem_hotplug_mutex); |
49 | ||
50 | void lock_memory_hotplug(void) | |
51 | { | |
52 | mutex_lock(&mem_hotplug_mutex); | |
53 | ||
54 | /* for exclusive hibernation if CONFIG_HIBERNATION=y */ | |
55 | lock_system_sleep(); | |
56 | } | |
57 | ||
58 | void unlock_memory_hotplug(void) | |
59 | { | |
60 | unlock_system_sleep(); | |
61 | mutex_unlock(&mem_hotplug_mutex); | |
62 | } | |
63 | ||
64 | ||
45e0b78b KM |
65 | /* add this memory to iomem resource */ |
66 | static struct resource *register_memory_resource(u64 start, u64 size) | |
67 | { | |
68 | struct resource *res; | |
69 | res = kzalloc(sizeof(struct resource), GFP_KERNEL); | |
70 | BUG_ON(!res); | |
71 | ||
72 | res->name = "System RAM"; | |
73 | res->start = start; | |
74 | res->end = start + size - 1; | |
887c3cb1 | 75 | res->flags = IORESOURCE_MEM | IORESOURCE_BUSY; |
45e0b78b | 76 | if (request_resource(&iomem_resource, res) < 0) { |
a62e2f4f | 77 | printk("System RAM resource %pR cannot be added\n", res); |
45e0b78b KM |
78 | kfree(res); |
79 | res = NULL; | |
80 | } | |
81 | return res; | |
82 | } | |
83 | ||
84 | static void release_memory_resource(struct resource *res) | |
85 | { | |
86 | if (!res) | |
87 | return; | |
88 | release_resource(res); | |
89 | kfree(res); | |
90 | return; | |
91 | } | |
92 | ||
53947027 | 93 | #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE |
46723bfa YI |
94 | void get_page_bootmem(unsigned long info, struct page *page, |
95 | unsigned long type) | |
04753278 | 96 | { |
5f24ce5f | 97 | page->lru.next = (struct list_head *) type; |
04753278 YG |
98 | SetPagePrivate(page); |
99 | set_page_private(page, info); | |
100 | atomic_inc(&page->_count); | |
101 | } | |
102 | ||
23ce932a RM |
103 | /* reference to __meminit __free_pages_bootmem is valid |
104 | * so use __ref to tell modpost not to generate a warning */ | |
105 | void __ref put_page_bootmem(struct page *page) | |
04753278 | 106 | { |
5f24ce5f | 107 | unsigned long type; |
9feedc9d | 108 | static DEFINE_MUTEX(ppb_lock); |
04753278 | 109 | |
5f24ce5f AA |
110 | type = (unsigned long) page->lru.next; |
111 | BUG_ON(type < MEMORY_HOTPLUG_MIN_BOOTMEM_TYPE || | |
112 | type > MEMORY_HOTPLUG_MAX_BOOTMEM_TYPE); | |
04753278 YG |
113 | |
114 | if (atomic_dec_return(&page->_count) == 1) { | |
115 | ClearPagePrivate(page); | |
116 | set_page_private(page, 0); | |
5f24ce5f | 117 | INIT_LIST_HEAD(&page->lru); |
9feedc9d JL |
118 | |
119 | /* | |
120 | * Please refer to comment for __free_pages_bootmem() | |
121 | * for why we serialize here. | |
122 | */ | |
123 | mutex_lock(&ppb_lock); | |
04753278 | 124 | __free_pages_bootmem(page, 0); |
9feedc9d | 125 | mutex_unlock(&ppb_lock); |
04753278 YG |
126 | } |
127 | ||
128 | } | |
129 | ||
46723bfa YI |
130 | #ifdef CONFIG_HAVE_BOOTMEM_INFO_NODE |
131 | #ifndef CONFIG_SPARSEMEM_VMEMMAP | |
d92bc318 | 132 | static void register_page_bootmem_info_section(unsigned long start_pfn) |
04753278 YG |
133 | { |
134 | unsigned long *usemap, mapsize, section_nr, i; | |
135 | struct mem_section *ms; | |
136 | struct page *page, *memmap; | |
137 | ||
04753278 YG |
138 | section_nr = pfn_to_section_nr(start_pfn); |
139 | ms = __nr_to_section(section_nr); | |
140 | ||
141 | /* Get section's memmap address */ | |
142 | memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr); | |
143 | ||
144 | /* | |
145 | * Get page for the memmap's phys address | |
146 | * XXX: need more consideration for sparse_vmemmap... | |
147 | */ | |
148 | page = virt_to_page(memmap); | |
149 | mapsize = sizeof(struct page) * PAGES_PER_SECTION; | |
150 | mapsize = PAGE_ALIGN(mapsize) >> PAGE_SHIFT; | |
151 | ||
152 | /* remember memmap's page */ | |
153 | for (i = 0; i < mapsize; i++, page++) | |
154 | get_page_bootmem(section_nr, page, SECTION_INFO); | |
155 | ||
156 | usemap = __nr_to_section(section_nr)->pageblock_flags; | |
157 | page = virt_to_page(usemap); | |
158 | ||
159 | mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT; | |
160 | ||
161 | for (i = 0; i < mapsize; i++, page++) | |
af370fb8 | 162 | get_page_bootmem(section_nr, page, MIX_SECTION_INFO); |
04753278 YG |
163 | |
164 | } | |
46723bfa YI |
165 | #else /* CONFIG_SPARSEMEM_VMEMMAP */ |
166 | static void register_page_bootmem_info_section(unsigned long start_pfn) | |
167 | { | |
168 | unsigned long *usemap, mapsize, section_nr, i; | |
169 | struct mem_section *ms; | |
170 | struct page *page, *memmap; | |
171 | ||
172 | if (!pfn_valid(start_pfn)) | |
173 | return; | |
174 | ||
175 | section_nr = pfn_to_section_nr(start_pfn); | |
176 | ms = __nr_to_section(section_nr); | |
177 | ||
178 | memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr); | |
179 | ||
180 | register_page_bootmem_memmap(section_nr, memmap, PAGES_PER_SECTION); | |
181 | ||
182 | usemap = __nr_to_section(section_nr)->pageblock_flags; | |
183 | page = virt_to_page(usemap); | |
184 | ||
185 | mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT; | |
186 | ||
187 | for (i = 0; i < mapsize; i++, page++) | |
188 | get_page_bootmem(section_nr, page, MIX_SECTION_INFO); | |
189 | } | |
190 | #endif /* !CONFIG_SPARSEMEM_VMEMMAP */ | |
04753278 YG |
191 | |
192 | void register_page_bootmem_info_node(struct pglist_data *pgdat) | |
193 | { | |
194 | unsigned long i, pfn, end_pfn, nr_pages; | |
195 | int node = pgdat->node_id; | |
196 | struct page *page; | |
197 | struct zone *zone; | |
198 | ||
199 | nr_pages = PAGE_ALIGN(sizeof(struct pglist_data)) >> PAGE_SHIFT; | |
200 | page = virt_to_page(pgdat); | |
201 | ||
202 | for (i = 0; i < nr_pages; i++, page++) | |
203 | get_page_bootmem(node, page, NODE_INFO); | |
204 | ||
205 | zone = &pgdat->node_zones[0]; | |
206 | for (; zone < pgdat->node_zones + MAX_NR_ZONES - 1; zone++) { | |
207 | if (zone->wait_table) { | |
208 | nr_pages = zone->wait_table_hash_nr_entries | |
209 | * sizeof(wait_queue_head_t); | |
210 | nr_pages = PAGE_ALIGN(nr_pages) >> PAGE_SHIFT; | |
211 | page = virt_to_page(zone->wait_table); | |
212 | ||
213 | for (i = 0; i < nr_pages; i++, page++) | |
214 | get_page_bootmem(node, page, NODE_INFO); | |
215 | } | |
216 | } | |
217 | ||
218 | pfn = pgdat->node_start_pfn; | |
219 | end_pfn = pfn + pgdat->node_spanned_pages; | |
220 | ||
221 | /* register_section info */ | |
f14851af | 222 | for (; pfn < end_pfn; pfn += PAGES_PER_SECTION) { |
223 | /* | |
224 | * Some platforms can assign the same pfn to multiple nodes - on | |
225 | * node0 as well as nodeN. To avoid registering a pfn against | |
226 | * multiple nodes we check that this pfn does not already | |
227 | * reside in some other node. | |
228 | */ | |
229 | if (pfn_valid(pfn) && (pfn_to_nid(pfn) == node)) | |
230 | register_page_bootmem_info_section(pfn); | |
231 | } | |
04753278 | 232 | } |
46723bfa | 233 | #endif /* CONFIG_HAVE_BOOTMEM_INFO_NODE */ |
04753278 | 234 | |
76cdd58e HC |
235 | static void grow_zone_span(struct zone *zone, unsigned long start_pfn, |
236 | unsigned long end_pfn) | |
237 | { | |
238 | unsigned long old_zone_end_pfn; | |
239 | ||
240 | zone_span_writelock(zone); | |
241 | ||
242 | old_zone_end_pfn = zone->zone_start_pfn + zone->spanned_pages; | |
712cd386 | 243 | if (!zone->spanned_pages || start_pfn < zone->zone_start_pfn) |
76cdd58e HC |
244 | zone->zone_start_pfn = start_pfn; |
245 | ||
246 | zone->spanned_pages = max(old_zone_end_pfn, end_pfn) - | |
247 | zone->zone_start_pfn; | |
248 | ||
249 | zone_span_writeunlock(zone); | |
250 | } | |
251 | ||
511c2aba LJ |
252 | static void resize_zone(struct zone *zone, unsigned long start_pfn, |
253 | unsigned long end_pfn) | |
254 | { | |
255 | zone_span_writelock(zone); | |
256 | ||
e455a9b9 LJ |
257 | if (end_pfn - start_pfn) { |
258 | zone->zone_start_pfn = start_pfn; | |
259 | zone->spanned_pages = end_pfn - start_pfn; | |
260 | } else { | |
261 | /* | |
262 | * make it consist as free_area_init_core(), | |
263 | * if spanned_pages = 0, then keep start_pfn = 0 | |
264 | */ | |
265 | zone->zone_start_pfn = 0; | |
266 | zone->spanned_pages = 0; | |
267 | } | |
511c2aba LJ |
268 | |
269 | zone_span_writeunlock(zone); | |
270 | } | |
271 | ||
272 | static void fix_zone_id(struct zone *zone, unsigned long start_pfn, | |
273 | unsigned long end_pfn) | |
274 | { | |
275 | enum zone_type zid = zone_idx(zone); | |
276 | int nid = zone->zone_pgdat->node_id; | |
277 | unsigned long pfn; | |
278 | ||
279 | for (pfn = start_pfn; pfn < end_pfn; pfn++) | |
280 | set_page_links(pfn_to_page(pfn), zid, nid, pfn); | |
281 | } | |
282 | ||
e455a9b9 | 283 | static int __meminit move_pfn_range_left(struct zone *z1, struct zone *z2, |
511c2aba LJ |
284 | unsigned long start_pfn, unsigned long end_pfn) |
285 | { | |
e455a9b9 | 286 | int ret; |
511c2aba | 287 | unsigned long flags; |
e455a9b9 LJ |
288 | unsigned long z1_start_pfn; |
289 | ||
290 | if (!z1->wait_table) { | |
291 | ret = init_currently_empty_zone(z1, start_pfn, | |
292 | end_pfn - start_pfn, MEMMAP_HOTPLUG); | |
293 | if (ret) | |
294 | return ret; | |
295 | } | |
511c2aba LJ |
296 | |
297 | pgdat_resize_lock(z1->zone_pgdat, &flags); | |
298 | ||
299 | /* can't move pfns which are higher than @z2 */ | |
300 | if (end_pfn > z2->zone_start_pfn + z2->spanned_pages) | |
301 | goto out_fail; | |
302 | /* the move out part mast at the left most of @z2 */ | |
303 | if (start_pfn > z2->zone_start_pfn) | |
304 | goto out_fail; | |
305 | /* must included/overlap */ | |
306 | if (end_pfn <= z2->zone_start_pfn) | |
307 | goto out_fail; | |
308 | ||
e455a9b9 LJ |
309 | /* use start_pfn for z1's start_pfn if z1 is empty */ |
310 | if (z1->spanned_pages) | |
311 | z1_start_pfn = z1->zone_start_pfn; | |
312 | else | |
313 | z1_start_pfn = start_pfn; | |
314 | ||
315 | resize_zone(z1, z1_start_pfn, end_pfn); | |
511c2aba LJ |
316 | resize_zone(z2, end_pfn, z2->zone_start_pfn + z2->spanned_pages); |
317 | ||
318 | pgdat_resize_unlock(z1->zone_pgdat, &flags); | |
319 | ||
320 | fix_zone_id(z1, start_pfn, end_pfn); | |
321 | ||
322 | return 0; | |
323 | out_fail: | |
324 | pgdat_resize_unlock(z1->zone_pgdat, &flags); | |
325 | return -1; | |
326 | } | |
327 | ||
e455a9b9 | 328 | static int __meminit move_pfn_range_right(struct zone *z1, struct zone *z2, |
511c2aba LJ |
329 | unsigned long start_pfn, unsigned long end_pfn) |
330 | { | |
e455a9b9 | 331 | int ret; |
511c2aba | 332 | unsigned long flags; |
e455a9b9 LJ |
333 | unsigned long z2_end_pfn; |
334 | ||
335 | if (!z2->wait_table) { | |
336 | ret = init_currently_empty_zone(z2, start_pfn, | |
337 | end_pfn - start_pfn, MEMMAP_HOTPLUG); | |
338 | if (ret) | |
339 | return ret; | |
340 | } | |
511c2aba LJ |
341 | |
342 | pgdat_resize_lock(z1->zone_pgdat, &flags); | |
343 | ||
344 | /* can't move pfns which are lower than @z1 */ | |
345 | if (z1->zone_start_pfn > start_pfn) | |
346 | goto out_fail; | |
347 | /* the move out part mast at the right most of @z1 */ | |
348 | if (z1->zone_start_pfn + z1->spanned_pages > end_pfn) | |
349 | goto out_fail; | |
350 | /* must included/overlap */ | |
351 | if (start_pfn >= z1->zone_start_pfn + z1->spanned_pages) | |
352 | goto out_fail; | |
353 | ||
e455a9b9 LJ |
354 | /* use end_pfn for z2's end_pfn if z2 is empty */ |
355 | if (z2->spanned_pages) | |
356 | z2_end_pfn = z2->zone_start_pfn + z2->spanned_pages; | |
357 | else | |
358 | z2_end_pfn = end_pfn; | |
359 | ||
511c2aba | 360 | resize_zone(z1, z1->zone_start_pfn, start_pfn); |
e455a9b9 | 361 | resize_zone(z2, start_pfn, z2_end_pfn); |
511c2aba LJ |
362 | |
363 | pgdat_resize_unlock(z1->zone_pgdat, &flags); | |
364 | ||
365 | fix_zone_id(z2, start_pfn, end_pfn); | |
366 | ||
367 | return 0; | |
368 | out_fail: | |
369 | pgdat_resize_unlock(z1->zone_pgdat, &flags); | |
370 | return -1; | |
371 | } | |
372 | ||
76cdd58e HC |
373 | static void grow_pgdat_span(struct pglist_data *pgdat, unsigned long start_pfn, |
374 | unsigned long end_pfn) | |
375 | { | |
376 | unsigned long old_pgdat_end_pfn = | |
377 | pgdat->node_start_pfn + pgdat->node_spanned_pages; | |
378 | ||
712cd386 | 379 | if (!pgdat->node_spanned_pages || start_pfn < pgdat->node_start_pfn) |
76cdd58e HC |
380 | pgdat->node_start_pfn = start_pfn; |
381 | ||
382 | pgdat->node_spanned_pages = max(old_pgdat_end_pfn, end_pfn) - | |
383 | pgdat->node_start_pfn; | |
384 | } | |
385 | ||
31168481 | 386 | static int __meminit __add_zone(struct zone *zone, unsigned long phys_start_pfn) |
3947be19 DH |
387 | { |
388 | struct pglist_data *pgdat = zone->zone_pgdat; | |
389 | int nr_pages = PAGES_PER_SECTION; | |
390 | int nid = pgdat->node_id; | |
391 | int zone_type; | |
76cdd58e | 392 | unsigned long flags; |
3947be19 DH |
393 | |
394 | zone_type = zone - pgdat->node_zones; | |
76cdd58e HC |
395 | if (!zone->wait_table) { |
396 | int ret; | |
397 | ||
398 | ret = init_currently_empty_zone(zone, phys_start_pfn, | |
399 | nr_pages, MEMMAP_HOTPLUG); | |
400 | if (ret) | |
401 | return ret; | |
402 | } | |
403 | pgdat_resize_lock(zone->zone_pgdat, &flags); | |
404 | grow_zone_span(zone, phys_start_pfn, phys_start_pfn + nr_pages); | |
405 | grow_pgdat_span(zone->zone_pgdat, phys_start_pfn, | |
406 | phys_start_pfn + nr_pages); | |
407 | pgdat_resize_unlock(zone->zone_pgdat, &flags); | |
a2f3aa02 DH |
408 | memmap_init_zone(nr_pages, nid, zone_type, |
409 | phys_start_pfn, MEMMAP_HOTPLUG); | |
718127cc | 410 | return 0; |
3947be19 DH |
411 | } |
412 | ||
c04fc586 GH |
413 | static int __meminit __add_section(int nid, struct zone *zone, |
414 | unsigned long phys_start_pfn) | |
3947be19 | 415 | { |
3947be19 | 416 | int nr_pages = PAGES_PER_SECTION; |
3947be19 DH |
417 | int ret; |
418 | ||
ebd15302 KH |
419 | if (pfn_valid(phys_start_pfn)) |
420 | return -EEXIST; | |
421 | ||
0b0acbec | 422 | ret = sparse_add_one_section(zone, phys_start_pfn, nr_pages); |
3947be19 DH |
423 | |
424 | if (ret < 0) | |
425 | return ret; | |
426 | ||
718127cc YG |
427 | ret = __add_zone(zone, phys_start_pfn); |
428 | ||
429 | if (ret < 0) | |
430 | return ret; | |
431 | ||
c04fc586 | 432 | return register_new_memory(nid, __pfn_to_section(phys_start_pfn)); |
3947be19 DH |
433 | } |
434 | ||
815121d2 YI |
435 | /* find the smallest valid pfn in the range [start_pfn, end_pfn) */ |
436 | static int find_smallest_section_pfn(int nid, struct zone *zone, | |
437 | unsigned long start_pfn, | |
438 | unsigned long end_pfn) | |
439 | { | |
440 | struct mem_section *ms; | |
441 | ||
442 | for (; start_pfn < end_pfn; start_pfn += PAGES_PER_SECTION) { | |
443 | ms = __pfn_to_section(start_pfn); | |
444 | ||
445 | if (unlikely(!valid_section(ms))) | |
446 | continue; | |
447 | ||
448 | if (unlikely(pfn_to_nid(start_pfn) != nid)) | |
449 | continue; | |
450 | ||
451 | if (zone && zone != page_zone(pfn_to_page(start_pfn))) | |
452 | continue; | |
453 | ||
454 | return start_pfn; | |
455 | } | |
456 | ||
457 | return 0; | |
458 | } | |
459 | ||
460 | /* find the biggest valid pfn in the range [start_pfn, end_pfn). */ | |
461 | static int find_biggest_section_pfn(int nid, struct zone *zone, | |
462 | unsigned long start_pfn, | |
463 | unsigned long end_pfn) | |
464 | { | |
465 | struct mem_section *ms; | |
466 | unsigned long pfn; | |
467 | ||
468 | /* pfn is the end pfn of a memory section. */ | |
469 | pfn = end_pfn - 1; | |
470 | for (; pfn >= start_pfn; pfn -= PAGES_PER_SECTION) { | |
471 | ms = __pfn_to_section(pfn); | |
472 | ||
473 | if (unlikely(!valid_section(ms))) | |
474 | continue; | |
475 | ||
476 | if (unlikely(pfn_to_nid(pfn) != nid)) | |
477 | continue; | |
478 | ||
479 | if (zone && zone != page_zone(pfn_to_page(pfn))) | |
480 | continue; | |
481 | ||
482 | return pfn; | |
483 | } | |
484 | ||
485 | return 0; | |
486 | } | |
487 | ||
488 | static void shrink_zone_span(struct zone *zone, unsigned long start_pfn, | |
489 | unsigned long end_pfn) | |
490 | { | |
491 | unsigned long zone_start_pfn = zone->zone_start_pfn; | |
492 | unsigned long zone_end_pfn = zone->zone_start_pfn + zone->spanned_pages; | |
493 | unsigned long pfn; | |
494 | struct mem_section *ms; | |
495 | int nid = zone_to_nid(zone); | |
496 | ||
497 | zone_span_writelock(zone); | |
498 | if (zone_start_pfn == start_pfn) { | |
499 | /* | |
500 | * If the section is smallest section in the zone, it need | |
501 | * shrink zone->zone_start_pfn and zone->zone_spanned_pages. | |
502 | * In this case, we find second smallest valid mem_section | |
503 | * for shrinking zone. | |
504 | */ | |
505 | pfn = find_smallest_section_pfn(nid, zone, end_pfn, | |
506 | zone_end_pfn); | |
507 | if (pfn) { | |
508 | zone->zone_start_pfn = pfn; | |
509 | zone->spanned_pages = zone_end_pfn - pfn; | |
510 | } | |
511 | } else if (zone_end_pfn == end_pfn) { | |
512 | /* | |
513 | * If the section is biggest section in the zone, it need | |
514 | * shrink zone->spanned_pages. | |
515 | * In this case, we find second biggest valid mem_section for | |
516 | * shrinking zone. | |
517 | */ | |
518 | pfn = find_biggest_section_pfn(nid, zone, zone_start_pfn, | |
519 | start_pfn); | |
520 | if (pfn) | |
521 | zone->spanned_pages = pfn - zone_start_pfn + 1; | |
522 | } | |
523 | ||
524 | /* | |
525 | * The section is not biggest or smallest mem_section in the zone, it | |
526 | * only creates a hole in the zone. So in this case, we need not | |
527 | * change the zone. But perhaps, the zone has only hole data. Thus | |
528 | * it check the zone has only hole or not. | |
529 | */ | |
530 | pfn = zone_start_pfn; | |
531 | for (; pfn < zone_end_pfn; pfn += PAGES_PER_SECTION) { | |
532 | ms = __pfn_to_section(pfn); | |
533 | ||
534 | if (unlikely(!valid_section(ms))) | |
535 | continue; | |
536 | ||
537 | if (page_zone(pfn_to_page(pfn)) != zone) | |
538 | continue; | |
539 | ||
540 | /* If the section is current section, it continues the loop */ | |
541 | if (start_pfn == pfn) | |
542 | continue; | |
543 | ||
544 | /* If we find valid section, we have nothing to do */ | |
545 | zone_span_writeunlock(zone); | |
546 | return; | |
547 | } | |
548 | ||
549 | /* The zone has no valid section */ | |
550 | zone->zone_start_pfn = 0; | |
551 | zone->spanned_pages = 0; | |
552 | zone_span_writeunlock(zone); | |
553 | } | |
554 | ||
555 | static void shrink_pgdat_span(struct pglist_data *pgdat, | |
556 | unsigned long start_pfn, unsigned long end_pfn) | |
557 | { | |
558 | unsigned long pgdat_start_pfn = pgdat->node_start_pfn; | |
559 | unsigned long pgdat_end_pfn = | |
560 | pgdat->node_start_pfn + pgdat->node_spanned_pages; | |
561 | unsigned long pfn; | |
562 | struct mem_section *ms; | |
563 | int nid = pgdat->node_id; | |
564 | ||
565 | if (pgdat_start_pfn == start_pfn) { | |
566 | /* | |
567 | * If the section is smallest section in the pgdat, it need | |
568 | * shrink pgdat->node_start_pfn and pgdat->node_spanned_pages. | |
569 | * In this case, we find second smallest valid mem_section | |
570 | * for shrinking zone. | |
571 | */ | |
572 | pfn = find_smallest_section_pfn(nid, NULL, end_pfn, | |
573 | pgdat_end_pfn); | |
574 | if (pfn) { | |
575 | pgdat->node_start_pfn = pfn; | |
576 | pgdat->node_spanned_pages = pgdat_end_pfn - pfn; | |
577 | } | |
578 | } else if (pgdat_end_pfn == end_pfn) { | |
579 | /* | |
580 | * If the section is biggest section in the pgdat, it need | |
581 | * shrink pgdat->node_spanned_pages. | |
582 | * In this case, we find second biggest valid mem_section for | |
583 | * shrinking zone. | |
584 | */ | |
585 | pfn = find_biggest_section_pfn(nid, NULL, pgdat_start_pfn, | |
586 | start_pfn); | |
587 | if (pfn) | |
588 | pgdat->node_spanned_pages = pfn - pgdat_start_pfn + 1; | |
589 | } | |
590 | ||
591 | /* | |
592 | * If the section is not biggest or smallest mem_section in the pgdat, | |
593 | * it only creates a hole in the pgdat. So in this case, we need not | |
594 | * change the pgdat. | |
595 | * But perhaps, the pgdat has only hole data. Thus it check the pgdat | |
596 | * has only hole or not. | |
597 | */ | |
598 | pfn = pgdat_start_pfn; | |
599 | for (; pfn < pgdat_end_pfn; pfn += PAGES_PER_SECTION) { | |
600 | ms = __pfn_to_section(pfn); | |
601 | ||
602 | if (unlikely(!valid_section(ms))) | |
603 | continue; | |
604 | ||
605 | if (pfn_to_nid(pfn) != nid) | |
606 | continue; | |
607 | ||
608 | /* If the section is current section, it continues the loop */ | |
609 | if (start_pfn == pfn) | |
610 | continue; | |
611 | ||
612 | /* If we find valid section, we have nothing to do */ | |
613 | return; | |
614 | } | |
615 | ||
616 | /* The pgdat has no valid section */ | |
617 | pgdat->node_start_pfn = 0; | |
618 | pgdat->node_spanned_pages = 0; | |
619 | } | |
620 | ||
621 | static void __remove_zone(struct zone *zone, unsigned long start_pfn) | |
622 | { | |
623 | struct pglist_data *pgdat = zone->zone_pgdat; | |
624 | int nr_pages = PAGES_PER_SECTION; | |
625 | int zone_type; | |
626 | unsigned long flags; | |
627 | ||
628 | zone_type = zone - pgdat->node_zones; | |
629 | ||
630 | pgdat_resize_lock(zone->zone_pgdat, &flags); | |
631 | shrink_zone_span(zone, start_pfn, start_pfn + nr_pages); | |
632 | shrink_pgdat_span(pgdat, start_pfn, start_pfn + nr_pages); | |
633 | pgdat_resize_unlock(zone->zone_pgdat, &flags); | |
634 | } | |
635 | ||
ea01ea93 BP |
636 | static int __remove_section(struct zone *zone, struct mem_section *ms) |
637 | { | |
815121d2 YI |
638 | unsigned long start_pfn; |
639 | int scn_nr; | |
ea01ea93 BP |
640 | int ret = -EINVAL; |
641 | ||
642 | if (!valid_section(ms)) | |
643 | return ret; | |
644 | ||
645 | ret = unregister_memory_section(ms); | |
646 | if (ret) | |
647 | return ret; | |
648 | ||
815121d2 YI |
649 | scn_nr = __section_nr(ms); |
650 | start_pfn = section_nr_to_pfn(scn_nr); | |
651 | __remove_zone(zone, start_pfn); | |
652 | ||
ea01ea93 | 653 | sparse_remove_one_section(zone, ms); |
ea01ea93 BP |
654 | return 0; |
655 | } | |
656 | ||
3947be19 DH |
657 | /* |
658 | * Reasonably generic function for adding memory. It is | |
659 | * expected that archs that support memory hotplug will | |
660 | * call this function after deciding the zone to which to | |
661 | * add the new pages. | |
662 | */ | |
c04fc586 GH |
663 | int __ref __add_pages(int nid, struct zone *zone, unsigned long phys_start_pfn, |
664 | unsigned long nr_pages) | |
3947be19 DH |
665 | { |
666 | unsigned long i; | |
667 | int err = 0; | |
6f712711 KH |
668 | int start_sec, end_sec; |
669 | /* during initialize mem_map, align hot-added range to section */ | |
670 | start_sec = pfn_to_section_nr(phys_start_pfn); | |
671 | end_sec = pfn_to_section_nr(phys_start_pfn + nr_pages - 1); | |
3947be19 | 672 | |
6f712711 | 673 | for (i = start_sec; i <= end_sec; i++) { |
c04fc586 | 674 | err = __add_section(nid, zone, i << PFN_SECTION_SHIFT); |
3947be19 | 675 | |
6f712711 | 676 | /* |
183ff22b | 677 | * EEXIST is finally dealt with by ioresource collision |
6f712711 KH |
678 | * check. see add_memory() => register_memory_resource() |
679 | * Warning will be printed if there is collision. | |
bed120c6 JS |
680 | */ |
681 | if (err && (err != -EEXIST)) | |
3947be19 | 682 | break; |
6f712711 | 683 | err = 0; |
3947be19 DH |
684 | } |
685 | ||
686 | return err; | |
687 | } | |
bed120c6 | 688 | EXPORT_SYMBOL_GPL(__add_pages); |
3947be19 | 689 | |
ea01ea93 BP |
690 | /** |
691 | * __remove_pages() - remove sections of pages from a zone | |
692 | * @zone: zone from which pages need to be removed | |
693 | * @phys_start_pfn: starting pageframe (must be aligned to start of a section) | |
694 | * @nr_pages: number of pages to remove (must be multiple of section size) | |
695 | * | |
696 | * Generic helper function to remove section mappings and sysfs entries | |
697 | * for the section of the memory we are removing. Caller needs to make | |
698 | * sure that pages are marked reserved and zones are adjust properly by | |
699 | * calling offline_pages(). | |
700 | */ | |
701 | int __remove_pages(struct zone *zone, unsigned long phys_start_pfn, | |
702 | unsigned long nr_pages) | |
703 | { | |
704 | unsigned long i, ret = 0; | |
705 | int sections_to_remove; | |
706 | ||
707 | /* | |
708 | * We can only remove entire sections | |
709 | */ | |
710 | BUG_ON(phys_start_pfn & ~PAGE_SECTION_MASK); | |
711 | BUG_ON(nr_pages % PAGES_PER_SECTION); | |
712 | ||
d760afd4 YI |
713 | release_mem_region(phys_start_pfn << PAGE_SHIFT, nr_pages * PAGE_SIZE); |
714 | ||
ea01ea93 BP |
715 | sections_to_remove = nr_pages / PAGES_PER_SECTION; |
716 | for (i = 0; i < sections_to_remove; i++) { | |
717 | unsigned long pfn = phys_start_pfn + i*PAGES_PER_SECTION; | |
718 | ret = __remove_section(zone, __pfn_to_section(pfn)); | |
719 | if (ret) | |
720 | break; | |
721 | } | |
722 | return ret; | |
723 | } | |
724 | EXPORT_SYMBOL_GPL(__remove_pages); | |
725 | ||
9d0ad8ca DK |
726 | int set_online_page_callback(online_page_callback_t callback) |
727 | { | |
728 | int rc = -EINVAL; | |
729 | ||
730 | lock_memory_hotplug(); | |
731 | ||
732 | if (online_page_callback == generic_online_page) { | |
733 | online_page_callback = callback; | |
734 | rc = 0; | |
735 | } | |
736 | ||
737 | unlock_memory_hotplug(); | |
738 | ||
739 | return rc; | |
740 | } | |
741 | EXPORT_SYMBOL_GPL(set_online_page_callback); | |
742 | ||
743 | int restore_online_page_callback(online_page_callback_t callback) | |
744 | { | |
745 | int rc = -EINVAL; | |
746 | ||
747 | lock_memory_hotplug(); | |
748 | ||
749 | if (online_page_callback == callback) { | |
750 | online_page_callback = generic_online_page; | |
751 | rc = 0; | |
752 | } | |
753 | ||
754 | unlock_memory_hotplug(); | |
755 | ||
756 | return rc; | |
757 | } | |
758 | EXPORT_SYMBOL_GPL(restore_online_page_callback); | |
759 | ||
760 | void __online_page_set_limits(struct page *page) | |
180c06ef | 761 | { |
4738e1b9 JB |
762 | unsigned long pfn = page_to_pfn(page); |
763 | ||
4738e1b9 JB |
764 | if (pfn >= num_physpages) |
765 | num_physpages = pfn + 1; | |
9d0ad8ca DK |
766 | } |
767 | EXPORT_SYMBOL_GPL(__online_page_set_limits); | |
768 | ||
769 | void __online_page_increment_counters(struct page *page) | |
770 | { | |
771 | totalram_pages++; | |
180c06ef JF |
772 | |
773 | #ifdef CONFIG_HIGHMEM | |
774 | if (PageHighMem(page)) | |
775 | totalhigh_pages++; | |
776 | #endif | |
9d0ad8ca DK |
777 | } |
778 | EXPORT_SYMBOL_GPL(__online_page_increment_counters); | |
180c06ef | 779 | |
9d0ad8ca DK |
780 | void __online_page_free(struct page *page) |
781 | { | |
180c06ef JF |
782 | ClearPageReserved(page); |
783 | init_page_count(page); | |
784 | __free_page(page); | |
785 | } | |
9d0ad8ca DK |
786 | EXPORT_SYMBOL_GPL(__online_page_free); |
787 | ||
788 | static void generic_online_page(struct page *page) | |
789 | { | |
790 | __online_page_set_limits(page); | |
791 | __online_page_increment_counters(page); | |
792 | __online_page_free(page); | |
793 | } | |
180c06ef | 794 | |
75884fb1 KH |
795 | static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages, |
796 | void *arg) | |
3947be19 DH |
797 | { |
798 | unsigned long i; | |
75884fb1 KH |
799 | unsigned long onlined_pages = *(unsigned long *)arg; |
800 | struct page *page; | |
801 | if (PageReserved(pfn_to_page(start_pfn))) | |
802 | for (i = 0; i < nr_pages; i++) { | |
803 | page = pfn_to_page(start_pfn + i); | |
9d0ad8ca | 804 | (*online_page_callback)(page); |
75884fb1 KH |
805 | onlined_pages++; |
806 | } | |
807 | *(unsigned long *)arg = onlined_pages; | |
808 | return 0; | |
809 | } | |
810 | ||
09285af7 | 811 | #ifdef CONFIG_MOVABLE_NODE |
79a4dcef TC |
812 | /* |
813 | * When CONFIG_MOVABLE_NODE, we permit onlining of a node which doesn't have | |
814 | * normal memory. | |
815 | */ | |
09285af7 LJ |
816 | static bool can_online_high_movable(struct zone *zone) |
817 | { | |
818 | return true; | |
819 | } | |
79a4dcef | 820 | #else /* CONFIG_MOVABLE_NODE */ |
74d42d8f LJ |
821 | /* ensure every online node has NORMAL memory */ |
822 | static bool can_online_high_movable(struct zone *zone) | |
823 | { | |
824 | return node_state(zone_to_nid(zone), N_NORMAL_MEMORY); | |
825 | } | |
79a4dcef | 826 | #endif /* CONFIG_MOVABLE_NODE */ |
74d42d8f | 827 | |
d9713679 LJ |
828 | /* check which state of node_states will be changed when online memory */ |
829 | static void node_states_check_changes_online(unsigned long nr_pages, | |
830 | struct zone *zone, struct memory_notify *arg) | |
831 | { | |
832 | int nid = zone_to_nid(zone); | |
833 | enum zone_type zone_last = ZONE_NORMAL; | |
834 | ||
835 | /* | |
6715ddf9 LJ |
836 | * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY] |
837 | * contains nodes which have zones of 0...ZONE_NORMAL, | |
838 | * set zone_last to ZONE_NORMAL. | |
d9713679 | 839 | * |
6715ddf9 LJ |
840 | * If we don't have HIGHMEM nor movable node, |
841 | * node_states[N_NORMAL_MEMORY] contains nodes which have zones of | |
842 | * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE. | |
d9713679 | 843 | */ |
6715ddf9 | 844 | if (N_MEMORY == N_NORMAL_MEMORY) |
d9713679 LJ |
845 | zone_last = ZONE_MOVABLE; |
846 | ||
847 | /* | |
848 | * if the memory to be online is in a zone of 0...zone_last, and | |
849 | * the zones of 0...zone_last don't have memory before online, we will | |
850 | * need to set the node to node_states[N_NORMAL_MEMORY] after | |
851 | * the memory is online. | |
852 | */ | |
853 | if (zone_idx(zone) <= zone_last && !node_state(nid, N_NORMAL_MEMORY)) | |
854 | arg->status_change_nid_normal = nid; | |
855 | else | |
856 | arg->status_change_nid_normal = -1; | |
857 | ||
6715ddf9 LJ |
858 | #ifdef CONFIG_HIGHMEM |
859 | /* | |
860 | * If we have movable node, node_states[N_HIGH_MEMORY] | |
861 | * contains nodes which have zones of 0...ZONE_HIGHMEM, | |
862 | * set zone_last to ZONE_HIGHMEM. | |
863 | * | |
864 | * If we don't have movable node, node_states[N_NORMAL_MEMORY] | |
865 | * contains nodes which have zones of 0...ZONE_MOVABLE, | |
866 | * set zone_last to ZONE_MOVABLE. | |
867 | */ | |
868 | zone_last = ZONE_HIGHMEM; | |
869 | if (N_MEMORY == N_HIGH_MEMORY) | |
870 | zone_last = ZONE_MOVABLE; | |
871 | ||
872 | if (zone_idx(zone) <= zone_last && !node_state(nid, N_HIGH_MEMORY)) | |
873 | arg->status_change_nid_high = nid; | |
874 | else | |
875 | arg->status_change_nid_high = -1; | |
876 | #else | |
877 | arg->status_change_nid_high = arg->status_change_nid_normal; | |
878 | #endif | |
879 | ||
d9713679 LJ |
880 | /* |
881 | * if the node don't have memory befor online, we will need to | |
6715ddf9 | 882 | * set the node to node_states[N_MEMORY] after the memory |
d9713679 LJ |
883 | * is online. |
884 | */ | |
6715ddf9 | 885 | if (!node_state(nid, N_MEMORY)) |
d9713679 LJ |
886 | arg->status_change_nid = nid; |
887 | else | |
888 | arg->status_change_nid = -1; | |
889 | } | |
890 | ||
891 | static void node_states_set_node(int node, struct memory_notify *arg) | |
892 | { | |
893 | if (arg->status_change_nid_normal >= 0) | |
894 | node_set_state(node, N_NORMAL_MEMORY); | |
895 | ||
6715ddf9 LJ |
896 | if (arg->status_change_nid_high >= 0) |
897 | node_set_state(node, N_HIGH_MEMORY); | |
898 | ||
899 | node_set_state(node, N_MEMORY); | |
d9713679 LJ |
900 | } |
901 | ||
75884fb1 | 902 | |
511c2aba | 903 | int __ref online_pages(unsigned long pfn, unsigned long nr_pages, int online_type) |
75884fb1 | 904 | { |
3947be19 DH |
905 | unsigned long onlined_pages = 0; |
906 | struct zone *zone; | |
6811378e | 907 | int need_zonelists_rebuild = 0; |
7b78d335 YG |
908 | int nid; |
909 | int ret; | |
910 | struct memory_notify arg; | |
911 | ||
925268a0 | 912 | lock_memory_hotplug(); |
d9713679 LJ |
913 | /* |
914 | * This doesn't need a lock to do pfn_to_page(). | |
915 | * The section can't be removed here because of the | |
916 | * memory_block->state_mutex. | |
917 | */ | |
918 | zone = page_zone(pfn_to_page(pfn)); | |
919 | ||
74d42d8f LJ |
920 | if ((zone_idx(zone) > ZONE_NORMAL || online_type == ONLINE_MOVABLE) && |
921 | !can_online_high_movable(zone)) { | |
922 | unlock_memory_hotplug(); | |
923 | return -1; | |
924 | } | |
925 | ||
511c2aba LJ |
926 | if (online_type == ONLINE_KERNEL && zone_idx(zone) == ZONE_MOVABLE) { |
927 | if (move_pfn_range_left(zone - 1, zone, pfn, pfn + nr_pages)) { | |
928 | unlock_memory_hotplug(); | |
929 | return -1; | |
930 | } | |
931 | } | |
932 | if (online_type == ONLINE_MOVABLE && zone_idx(zone) == ZONE_MOVABLE - 1) { | |
933 | if (move_pfn_range_right(zone, zone + 1, pfn, pfn + nr_pages)) { | |
934 | unlock_memory_hotplug(); | |
935 | return -1; | |
936 | } | |
937 | } | |
938 | ||
939 | /* Previous code may changed the zone of the pfn range */ | |
940 | zone = page_zone(pfn_to_page(pfn)); | |
941 | ||
7b78d335 YG |
942 | arg.start_pfn = pfn; |
943 | arg.nr_pages = nr_pages; | |
d9713679 | 944 | node_states_check_changes_online(nr_pages, zone, &arg); |
7b78d335 YG |
945 | |
946 | nid = page_to_nid(pfn_to_page(pfn)); | |
3947be19 | 947 | |
7b78d335 YG |
948 | ret = memory_notify(MEM_GOING_ONLINE, &arg); |
949 | ret = notifier_to_errno(ret); | |
950 | if (ret) { | |
951 | memory_notify(MEM_CANCEL_ONLINE, &arg); | |
925268a0 | 952 | unlock_memory_hotplug(); |
7b78d335 YG |
953 | return ret; |
954 | } | |
6811378e YG |
955 | /* |
956 | * If this zone is not populated, then it is not in zonelist. | |
957 | * This means the page allocator ignores this zone. | |
958 | * So, zonelist must be updated after online. | |
959 | */ | |
4eaf3f64 | 960 | mutex_lock(&zonelists_mutex); |
6dcd73d7 | 961 | if (!populated_zone(zone)) { |
6811378e | 962 | need_zonelists_rebuild = 1; |
6dcd73d7 WC |
963 | build_all_zonelists(NULL, zone); |
964 | } | |
6811378e | 965 | |
908eedc6 | 966 | ret = walk_system_ram_range(pfn, nr_pages, &onlined_pages, |
75884fb1 | 967 | online_pages_range); |
fd8a4221 | 968 | if (ret) { |
6dcd73d7 WC |
969 | if (need_zonelists_rebuild) |
970 | zone_pcp_reset(zone); | |
4eaf3f64 | 971 | mutex_unlock(&zonelists_mutex); |
a62e2f4f BH |
972 | printk(KERN_DEBUG "online_pages [mem %#010llx-%#010llx] failed\n", |
973 | (unsigned long long) pfn << PAGE_SHIFT, | |
974 | (((unsigned long long) pfn + nr_pages) | |
975 | << PAGE_SHIFT) - 1); | |
fd8a4221 | 976 | memory_notify(MEM_CANCEL_ONLINE, &arg); |
925268a0 | 977 | unlock_memory_hotplug(); |
fd8a4221 GL |
978 | return ret; |
979 | } | |
980 | ||
9feedc9d | 981 | zone->managed_pages += onlined_pages; |
3947be19 | 982 | zone->present_pages += onlined_pages; |
f2937be5 | 983 | zone->zone_pgdat->node_present_pages += onlined_pages; |
08dff7b7 | 984 | if (onlined_pages) { |
d9713679 | 985 | node_states_set_node(zone_to_nid(zone), &arg); |
08dff7b7 | 986 | if (need_zonelists_rebuild) |
6dcd73d7 | 987 | build_all_zonelists(NULL, NULL); |
08dff7b7 JL |
988 | else |
989 | zone_pcp_update(zone); | |
990 | } | |
3947be19 | 991 | |
4eaf3f64 | 992 | mutex_unlock(&zonelists_mutex); |
1b79acc9 KM |
993 | |
994 | init_per_zone_wmark_min(); | |
995 | ||
08dff7b7 | 996 | if (onlined_pages) |
7ea1530a | 997 | kswapd_run(zone_to_nid(zone)); |
61b13993 | 998 | |
1f522509 | 999 | vm_total_pages = nr_free_pagecache_pages(); |
2f7f24ec | 1000 | |
2d1d43f6 | 1001 | writeback_set_ratelimit(); |
7b78d335 YG |
1002 | |
1003 | if (onlined_pages) | |
1004 | memory_notify(MEM_ONLINE, &arg); | |
925268a0 | 1005 | unlock_memory_hotplug(); |
7b78d335 | 1006 | |
3947be19 DH |
1007 | return 0; |
1008 | } | |
53947027 | 1009 | #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */ |
bc02af93 | 1010 | |
e1319331 HS |
1011 | /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */ |
1012 | static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start) | |
9af3c2de YG |
1013 | { |
1014 | struct pglist_data *pgdat; | |
1015 | unsigned long zones_size[MAX_NR_ZONES] = {0}; | |
1016 | unsigned long zholes_size[MAX_NR_ZONES] = {0}; | |
1017 | unsigned long start_pfn = start >> PAGE_SHIFT; | |
1018 | ||
1019 | pgdat = arch_alloc_nodedata(nid); | |
1020 | if (!pgdat) | |
1021 | return NULL; | |
1022 | ||
1023 | arch_refresh_nodedata(nid, pgdat); | |
1024 | ||
1025 | /* we can use NODE_DATA(nid) from here */ | |
1026 | ||
1027 | /* init node's zones as empty zones, we don't have any present pages.*/ | |
9109fb7b | 1028 | free_area_init_node(nid, zones_size, start_pfn, zholes_size); |
9af3c2de | 1029 | |
959ecc48 KH |
1030 | /* |
1031 | * The node we allocated has no zone fallback lists. For avoiding | |
1032 | * to access not-initialized zonelist, build here. | |
1033 | */ | |
f957db4f | 1034 | mutex_lock(&zonelists_mutex); |
9adb62a5 | 1035 | build_all_zonelists(pgdat, NULL); |
f957db4f | 1036 | mutex_unlock(&zonelists_mutex); |
959ecc48 | 1037 | |
9af3c2de YG |
1038 | return pgdat; |
1039 | } | |
1040 | ||
1041 | static void rollback_node_hotadd(int nid, pg_data_t *pgdat) | |
1042 | { | |
1043 | arch_refresh_nodedata(nid, NULL); | |
1044 | arch_free_nodedata(pgdat); | |
1045 | return; | |
1046 | } | |
1047 | ||
0a547039 | 1048 | |
cf23422b | 1049 | /* |
1050 | * called by cpu_up() to online a node without onlined memory. | |
1051 | */ | |
1052 | int mem_online_node(int nid) | |
1053 | { | |
1054 | pg_data_t *pgdat; | |
1055 | int ret; | |
1056 | ||
20d6c96b | 1057 | lock_memory_hotplug(); |
cf23422b | 1058 | pgdat = hotadd_new_pgdat(nid, 0); |
7553e8f2 | 1059 | if (!pgdat) { |
cf23422b | 1060 | ret = -ENOMEM; |
1061 | goto out; | |
1062 | } | |
1063 | node_set_online(nid); | |
1064 | ret = register_one_node(nid); | |
1065 | BUG_ON(ret); | |
1066 | ||
1067 | out: | |
20d6c96b | 1068 | unlock_memory_hotplug(); |
cf23422b | 1069 | return ret; |
1070 | } | |
1071 | ||
31168481 AV |
1072 | /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */ |
1073 | int __ref add_memory(int nid, u64 start, u64 size) | |
bc02af93 | 1074 | { |
9af3c2de YG |
1075 | pg_data_t *pgdat = NULL; |
1076 | int new_pgdat = 0; | |
ebd15302 | 1077 | struct resource *res; |
bc02af93 YG |
1078 | int ret; |
1079 | ||
20d6c96b | 1080 | lock_memory_hotplug(); |
6ad696d2 | 1081 | |
ebd15302 | 1082 | res = register_memory_resource(start, size); |
6ad696d2 | 1083 | ret = -EEXIST; |
ebd15302 | 1084 | if (!res) |
6ad696d2 | 1085 | goto out; |
ebd15302 | 1086 | |
9af3c2de YG |
1087 | if (!node_online(nid)) { |
1088 | pgdat = hotadd_new_pgdat(nid, start); | |
6ad696d2 | 1089 | ret = -ENOMEM; |
9af3c2de | 1090 | if (!pgdat) |
41b9e2d7 | 1091 | goto error; |
9af3c2de | 1092 | new_pgdat = 1; |
9af3c2de YG |
1093 | } |
1094 | ||
bc02af93 YG |
1095 | /* call arch's memory hotadd */ |
1096 | ret = arch_add_memory(nid, start, size); | |
1097 | ||
9af3c2de YG |
1098 | if (ret < 0) |
1099 | goto error; | |
1100 | ||
0fc44159 | 1101 | /* we online node here. we can't roll back from here. */ |
9af3c2de YG |
1102 | node_set_online(nid); |
1103 | ||
0fc44159 YG |
1104 | if (new_pgdat) { |
1105 | ret = register_one_node(nid); | |
1106 | /* | |
1107 | * If sysfs file of new node can't create, cpu on the node | |
1108 | * can't be hot-added. There is no rollback way now. | |
1109 | * So, check by BUG_ON() to catch it reluctantly.. | |
1110 | */ | |
1111 | BUG_ON(ret); | |
1112 | } | |
1113 | ||
d96ae530 AM |
1114 | /* create new memmap entry */ |
1115 | firmware_map_add_hotplug(start, start + size, "System RAM"); | |
1116 | ||
6ad696d2 AK |
1117 | goto out; |
1118 | ||
9af3c2de YG |
1119 | error: |
1120 | /* rollback pgdat allocation and others */ | |
1121 | if (new_pgdat) | |
1122 | rollback_node_hotadd(nid, pgdat); | |
a864b9d0 | 1123 | release_memory_resource(res); |
9af3c2de | 1124 | |
6ad696d2 | 1125 | out: |
20d6c96b | 1126 | unlock_memory_hotplug(); |
bc02af93 YG |
1127 | return ret; |
1128 | } | |
1129 | EXPORT_SYMBOL_GPL(add_memory); | |
0c0e6195 KH |
1130 | |
1131 | #ifdef CONFIG_MEMORY_HOTREMOVE | |
5c755e9f BP |
1132 | /* |
1133 | * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy | |
1134 | * set and the size of the free page is given by page_order(). Using this, | |
1135 | * the function determines if the pageblock contains only free pages. | |
1136 | * Due to buddy contraints, a free page at least the size of a pageblock will | |
1137 | * be located at the start of the pageblock | |
1138 | */ | |
1139 | static inline int pageblock_free(struct page *page) | |
1140 | { | |
1141 | return PageBuddy(page) && page_order(page) >= pageblock_order; | |
1142 | } | |
1143 | ||
1144 | /* Return the start of the next active pageblock after a given page */ | |
1145 | static struct page *next_active_pageblock(struct page *page) | |
1146 | { | |
5c755e9f BP |
1147 | /* Ensure the starting page is pageblock-aligned */ |
1148 | BUG_ON(page_to_pfn(page) & (pageblock_nr_pages - 1)); | |
1149 | ||
5c755e9f | 1150 | /* If the entire pageblock is free, move to the end of free page */ |
0dcc48c1 KH |
1151 | if (pageblock_free(page)) { |
1152 | int order; | |
1153 | /* be careful. we don't have locks, page_order can be changed.*/ | |
1154 | order = page_order(page); | |
1155 | if ((order < MAX_ORDER) && (order >= pageblock_order)) | |
1156 | return page + (1 << order); | |
1157 | } | |
5c755e9f | 1158 | |
0dcc48c1 | 1159 | return page + pageblock_nr_pages; |
5c755e9f BP |
1160 | } |
1161 | ||
1162 | /* Checks if this range of memory is likely to be hot-removable. */ | |
1163 | int is_mem_section_removable(unsigned long start_pfn, unsigned long nr_pages) | |
1164 | { | |
5c755e9f BP |
1165 | struct page *page = pfn_to_page(start_pfn); |
1166 | struct page *end_page = page + nr_pages; | |
1167 | ||
1168 | /* Check the starting page of each pageblock within the range */ | |
1169 | for (; page < end_page; page = next_active_pageblock(page)) { | |
49ac8255 | 1170 | if (!is_pageblock_removable_nolock(page)) |
5c755e9f | 1171 | return 0; |
49ac8255 | 1172 | cond_resched(); |
5c755e9f BP |
1173 | } |
1174 | ||
1175 | /* All pageblocks in the memory block are likely to be hot-removable */ | |
1176 | return 1; | |
1177 | } | |
1178 | ||
0c0e6195 KH |
1179 | /* |
1180 | * Confirm all pages in a range [start, end) is belongs to the same zone. | |
1181 | */ | |
1182 | static int test_pages_in_a_zone(unsigned long start_pfn, unsigned long end_pfn) | |
1183 | { | |
1184 | unsigned long pfn; | |
1185 | struct zone *zone = NULL; | |
1186 | struct page *page; | |
1187 | int i; | |
1188 | for (pfn = start_pfn; | |
1189 | pfn < end_pfn; | |
1190 | pfn += MAX_ORDER_NR_PAGES) { | |
1191 | i = 0; | |
1192 | /* This is just a CONFIG_HOLES_IN_ZONE check.*/ | |
1193 | while ((i < MAX_ORDER_NR_PAGES) && !pfn_valid_within(pfn + i)) | |
1194 | i++; | |
1195 | if (i == MAX_ORDER_NR_PAGES) | |
1196 | continue; | |
1197 | page = pfn_to_page(pfn + i); | |
1198 | if (zone && page_zone(page) != zone) | |
1199 | return 0; | |
1200 | zone = page_zone(page); | |
1201 | } | |
1202 | return 1; | |
1203 | } | |
1204 | ||
1205 | /* | |
1206 | * Scanning pfn is much easier than scanning lru list. | |
1207 | * Scan pfn from start to end and Find LRU page. | |
1208 | */ | |
7bbc0905 | 1209 | static unsigned long scan_lru_pages(unsigned long start, unsigned long end) |
0c0e6195 KH |
1210 | { |
1211 | unsigned long pfn; | |
1212 | struct page *page; | |
1213 | for (pfn = start; pfn < end; pfn++) { | |
1214 | if (pfn_valid(pfn)) { | |
1215 | page = pfn_to_page(pfn); | |
1216 | if (PageLRU(page)) | |
1217 | return pfn; | |
1218 | } | |
1219 | } | |
1220 | return 0; | |
1221 | } | |
1222 | ||
0c0e6195 KH |
1223 | #define NR_OFFLINE_AT_ONCE_PAGES (256) |
1224 | static int | |
1225 | do_migrate_range(unsigned long start_pfn, unsigned long end_pfn) | |
1226 | { | |
1227 | unsigned long pfn; | |
1228 | struct page *page; | |
1229 | int move_pages = NR_OFFLINE_AT_ONCE_PAGES; | |
1230 | int not_managed = 0; | |
1231 | int ret = 0; | |
1232 | LIST_HEAD(source); | |
1233 | ||
1234 | for (pfn = start_pfn; pfn < end_pfn && move_pages > 0; pfn++) { | |
1235 | if (!pfn_valid(pfn)) | |
1236 | continue; | |
1237 | page = pfn_to_page(pfn); | |
700c2a46 | 1238 | if (!get_page_unless_zero(page)) |
0c0e6195 KH |
1239 | continue; |
1240 | /* | |
1241 | * We can skip free pages. And we can only deal with pages on | |
1242 | * LRU. | |
1243 | */ | |
62695a84 | 1244 | ret = isolate_lru_page(page); |
0c0e6195 | 1245 | if (!ret) { /* Success */ |
700c2a46 | 1246 | put_page(page); |
62695a84 | 1247 | list_add_tail(&page->lru, &source); |
0c0e6195 | 1248 | move_pages--; |
6d9c285a KM |
1249 | inc_zone_page_state(page, NR_ISOLATED_ANON + |
1250 | page_is_file_cache(page)); | |
1251 | ||
0c0e6195 | 1252 | } else { |
0c0e6195 | 1253 | #ifdef CONFIG_DEBUG_VM |
718a3821 WF |
1254 | printk(KERN_ALERT "removing pfn %lx from LRU failed\n", |
1255 | pfn); | |
1256 | dump_page(page); | |
0c0e6195 | 1257 | #endif |
700c2a46 | 1258 | put_page(page); |
25985edc | 1259 | /* Because we don't have big zone->lock. we should |
809c4449 BL |
1260 | check this again here. */ |
1261 | if (page_count(page)) { | |
1262 | not_managed++; | |
f3ab2636 | 1263 | ret = -EBUSY; |
809c4449 BL |
1264 | break; |
1265 | } | |
0c0e6195 KH |
1266 | } |
1267 | } | |
f3ab2636 BL |
1268 | if (!list_empty(&source)) { |
1269 | if (not_managed) { | |
1270 | putback_lru_pages(&source); | |
1271 | goto out; | |
1272 | } | |
74c08f98 MK |
1273 | |
1274 | /* | |
1275 | * alloc_migrate_target should be improooooved!! | |
1276 | * migrate_pages returns # of failed pages. | |
1277 | */ | |
1278 | ret = migrate_pages(&source, alloc_migrate_target, 0, | |
7b2a2d4a MG |
1279 | true, MIGRATE_SYNC, |
1280 | MR_MEMORY_HOTPLUG); | |
f3ab2636 | 1281 | if (ret) |
0c0e6195 | 1282 | putback_lru_pages(&source); |
0c0e6195 | 1283 | } |
0c0e6195 KH |
1284 | out: |
1285 | return ret; | |
1286 | } | |
1287 | ||
1288 | /* | |
1289 | * remove from free_area[] and mark all as Reserved. | |
1290 | */ | |
1291 | static int | |
1292 | offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages, | |
1293 | void *data) | |
1294 | { | |
1295 | __offline_isolated_pages(start, start + nr_pages); | |
1296 | return 0; | |
1297 | } | |
1298 | ||
1299 | static void | |
1300 | offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn) | |
1301 | { | |
908eedc6 | 1302 | walk_system_ram_range(start_pfn, end_pfn - start_pfn, NULL, |
0c0e6195 KH |
1303 | offline_isolated_pages_cb); |
1304 | } | |
1305 | ||
1306 | /* | |
1307 | * Check all pages in range, recoreded as memory resource, are isolated. | |
1308 | */ | |
1309 | static int | |
1310 | check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages, | |
1311 | void *data) | |
1312 | { | |
1313 | int ret; | |
1314 | long offlined = *(long *)data; | |
b023f468 | 1315 | ret = test_pages_isolated(start_pfn, start_pfn + nr_pages, true); |
0c0e6195 KH |
1316 | offlined = nr_pages; |
1317 | if (!ret) | |
1318 | *(long *)data += offlined; | |
1319 | return ret; | |
1320 | } | |
1321 | ||
1322 | static long | |
1323 | check_pages_isolated(unsigned long start_pfn, unsigned long end_pfn) | |
1324 | { | |
1325 | long offlined = 0; | |
1326 | int ret; | |
1327 | ||
908eedc6 | 1328 | ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn, &offlined, |
0c0e6195 KH |
1329 | check_pages_isolated_cb); |
1330 | if (ret < 0) | |
1331 | offlined = (long)ret; | |
1332 | return offlined; | |
1333 | } | |
1334 | ||
09285af7 | 1335 | #ifdef CONFIG_MOVABLE_NODE |
79a4dcef TC |
1336 | /* |
1337 | * When CONFIG_MOVABLE_NODE, we permit offlining of a node which doesn't have | |
1338 | * normal memory. | |
1339 | */ | |
09285af7 LJ |
1340 | static bool can_offline_normal(struct zone *zone, unsigned long nr_pages) |
1341 | { | |
1342 | return true; | |
1343 | } | |
79a4dcef | 1344 | #else /* CONFIG_MOVABLE_NODE */ |
74d42d8f LJ |
1345 | /* ensure the node has NORMAL memory if it is still online */ |
1346 | static bool can_offline_normal(struct zone *zone, unsigned long nr_pages) | |
1347 | { | |
1348 | struct pglist_data *pgdat = zone->zone_pgdat; | |
1349 | unsigned long present_pages = 0; | |
1350 | enum zone_type zt; | |
1351 | ||
1352 | for (zt = 0; zt <= ZONE_NORMAL; zt++) | |
1353 | present_pages += pgdat->node_zones[zt].present_pages; | |
1354 | ||
1355 | if (present_pages > nr_pages) | |
1356 | return true; | |
1357 | ||
1358 | present_pages = 0; | |
1359 | for (; zt <= ZONE_MOVABLE; zt++) | |
1360 | present_pages += pgdat->node_zones[zt].present_pages; | |
1361 | ||
1362 | /* | |
1363 | * we can't offline the last normal memory until all | |
1364 | * higher memory is offlined. | |
1365 | */ | |
1366 | return present_pages == 0; | |
1367 | } | |
79a4dcef | 1368 | #endif /* CONFIG_MOVABLE_NODE */ |
74d42d8f | 1369 | |
d9713679 LJ |
1370 | /* check which state of node_states will be changed when offline memory */ |
1371 | static void node_states_check_changes_offline(unsigned long nr_pages, | |
1372 | struct zone *zone, struct memory_notify *arg) | |
1373 | { | |
1374 | struct pglist_data *pgdat = zone->zone_pgdat; | |
1375 | unsigned long present_pages = 0; | |
1376 | enum zone_type zt, zone_last = ZONE_NORMAL; | |
1377 | ||
1378 | /* | |
6715ddf9 LJ |
1379 | * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY] |
1380 | * contains nodes which have zones of 0...ZONE_NORMAL, | |
1381 | * set zone_last to ZONE_NORMAL. | |
d9713679 | 1382 | * |
6715ddf9 LJ |
1383 | * If we don't have HIGHMEM nor movable node, |
1384 | * node_states[N_NORMAL_MEMORY] contains nodes which have zones of | |
1385 | * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE. | |
d9713679 | 1386 | */ |
6715ddf9 | 1387 | if (N_MEMORY == N_NORMAL_MEMORY) |
d9713679 LJ |
1388 | zone_last = ZONE_MOVABLE; |
1389 | ||
1390 | /* | |
1391 | * check whether node_states[N_NORMAL_MEMORY] will be changed. | |
1392 | * If the memory to be offline is in a zone of 0...zone_last, | |
1393 | * and it is the last present memory, 0...zone_last will | |
1394 | * become empty after offline , thus we can determind we will | |
1395 | * need to clear the node from node_states[N_NORMAL_MEMORY]. | |
1396 | */ | |
1397 | for (zt = 0; zt <= zone_last; zt++) | |
1398 | present_pages += pgdat->node_zones[zt].present_pages; | |
1399 | if (zone_idx(zone) <= zone_last && nr_pages >= present_pages) | |
1400 | arg->status_change_nid_normal = zone_to_nid(zone); | |
1401 | else | |
1402 | arg->status_change_nid_normal = -1; | |
1403 | ||
6715ddf9 LJ |
1404 | #ifdef CONFIG_HIGHMEM |
1405 | /* | |
1406 | * If we have movable node, node_states[N_HIGH_MEMORY] | |
1407 | * contains nodes which have zones of 0...ZONE_HIGHMEM, | |
1408 | * set zone_last to ZONE_HIGHMEM. | |
1409 | * | |
1410 | * If we don't have movable node, node_states[N_NORMAL_MEMORY] | |
1411 | * contains nodes which have zones of 0...ZONE_MOVABLE, | |
1412 | * set zone_last to ZONE_MOVABLE. | |
1413 | */ | |
1414 | zone_last = ZONE_HIGHMEM; | |
1415 | if (N_MEMORY == N_HIGH_MEMORY) | |
1416 | zone_last = ZONE_MOVABLE; | |
1417 | ||
1418 | for (; zt <= zone_last; zt++) | |
1419 | present_pages += pgdat->node_zones[zt].present_pages; | |
1420 | if (zone_idx(zone) <= zone_last && nr_pages >= present_pages) | |
1421 | arg->status_change_nid_high = zone_to_nid(zone); | |
1422 | else | |
1423 | arg->status_change_nid_high = -1; | |
1424 | #else | |
1425 | arg->status_change_nid_high = arg->status_change_nid_normal; | |
1426 | #endif | |
1427 | ||
d9713679 LJ |
1428 | /* |
1429 | * node_states[N_HIGH_MEMORY] contains nodes which have 0...ZONE_MOVABLE | |
1430 | */ | |
1431 | zone_last = ZONE_MOVABLE; | |
1432 | ||
1433 | /* | |
1434 | * check whether node_states[N_HIGH_MEMORY] will be changed | |
1435 | * If we try to offline the last present @nr_pages from the node, | |
1436 | * we can determind we will need to clear the node from | |
1437 | * node_states[N_HIGH_MEMORY]. | |
1438 | */ | |
1439 | for (; zt <= zone_last; zt++) | |
1440 | present_pages += pgdat->node_zones[zt].present_pages; | |
1441 | if (nr_pages >= present_pages) | |
1442 | arg->status_change_nid = zone_to_nid(zone); | |
1443 | else | |
1444 | arg->status_change_nid = -1; | |
1445 | } | |
1446 | ||
1447 | static void node_states_clear_node(int node, struct memory_notify *arg) | |
1448 | { | |
1449 | if (arg->status_change_nid_normal >= 0) | |
1450 | node_clear_state(node, N_NORMAL_MEMORY); | |
1451 | ||
6715ddf9 LJ |
1452 | if ((N_MEMORY != N_NORMAL_MEMORY) && |
1453 | (arg->status_change_nid_high >= 0)) | |
d9713679 | 1454 | node_clear_state(node, N_HIGH_MEMORY); |
6715ddf9 LJ |
1455 | |
1456 | if ((N_MEMORY != N_HIGH_MEMORY) && | |
1457 | (arg->status_change_nid >= 0)) | |
1458 | node_clear_state(node, N_MEMORY); | |
d9713679 LJ |
1459 | } |
1460 | ||
a16cee10 | 1461 | static int __ref __offline_pages(unsigned long start_pfn, |
0c0e6195 KH |
1462 | unsigned long end_pfn, unsigned long timeout) |
1463 | { | |
1464 | unsigned long pfn, nr_pages, expire; | |
1465 | long offlined_pages; | |
7b78d335 | 1466 | int ret, drain, retry_max, node; |
0c0e6195 | 1467 | struct zone *zone; |
7b78d335 | 1468 | struct memory_notify arg; |
0c0e6195 KH |
1469 | |
1470 | BUG_ON(start_pfn >= end_pfn); | |
1471 | /* at least, alignment against pageblock is necessary */ | |
1472 | if (!IS_ALIGNED(start_pfn, pageblock_nr_pages)) | |
1473 | return -EINVAL; | |
1474 | if (!IS_ALIGNED(end_pfn, pageblock_nr_pages)) | |
1475 | return -EINVAL; | |
1476 | /* This makes hotplug much easier...and readable. | |
1477 | we assume this for now. .*/ | |
1478 | if (!test_pages_in_a_zone(start_pfn, end_pfn)) | |
1479 | return -EINVAL; | |
7b78d335 | 1480 | |
20d6c96b | 1481 | lock_memory_hotplug(); |
6ad696d2 | 1482 | |
7b78d335 YG |
1483 | zone = page_zone(pfn_to_page(start_pfn)); |
1484 | node = zone_to_nid(zone); | |
1485 | nr_pages = end_pfn - start_pfn; | |
1486 | ||
74d42d8f LJ |
1487 | ret = -EINVAL; |
1488 | if (zone_idx(zone) <= ZONE_NORMAL && !can_offline_normal(zone, nr_pages)) | |
1489 | goto out; | |
1490 | ||
0c0e6195 | 1491 | /* set above range as isolated */ |
b023f468 WC |
1492 | ret = start_isolate_page_range(start_pfn, end_pfn, |
1493 | MIGRATE_MOVABLE, true); | |
0c0e6195 | 1494 | if (ret) |
6ad696d2 | 1495 | goto out; |
7b78d335 YG |
1496 | |
1497 | arg.start_pfn = start_pfn; | |
1498 | arg.nr_pages = nr_pages; | |
d9713679 | 1499 | node_states_check_changes_offline(nr_pages, zone, &arg); |
7b78d335 YG |
1500 | |
1501 | ret = memory_notify(MEM_GOING_OFFLINE, &arg); | |
1502 | ret = notifier_to_errno(ret); | |
1503 | if (ret) | |
1504 | goto failed_removal; | |
1505 | ||
0c0e6195 KH |
1506 | pfn = start_pfn; |
1507 | expire = jiffies + timeout; | |
1508 | drain = 0; | |
1509 | retry_max = 5; | |
1510 | repeat: | |
1511 | /* start memory hot removal */ | |
1512 | ret = -EAGAIN; | |
1513 | if (time_after(jiffies, expire)) | |
1514 | goto failed_removal; | |
1515 | ret = -EINTR; | |
1516 | if (signal_pending(current)) | |
1517 | goto failed_removal; | |
1518 | ret = 0; | |
1519 | if (drain) { | |
1520 | lru_add_drain_all(); | |
0c0e6195 | 1521 | cond_resched(); |
9f8f2172 | 1522 | drain_all_pages(); |
0c0e6195 KH |
1523 | } |
1524 | ||
1525 | pfn = scan_lru_pages(start_pfn, end_pfn); | |
1526 | if (pfn) { /* We have page on LRU */ | |
1527 | ret = do_migrate_range(pfn, end_pfn); | |
1528 | if (!ret) { | |
1529 | drain = 1; | |
1530 | goto repeat; | |
1531 | } else { | |
1532 | if (ret < 0) | |
1533 | if (--retry_max == 0) | |
1534 | goto failed_removal; | |
1535 | yield(); | |
1536 | drain = 1; | |
1537 | goto repeat; | |
1538 | } | |
1539 | } | |
b3834be5 | 1540 | /* drain all zone's lru pagevec, this is asynchronous... */ |
0c0e6195 | 1541 | lru_add_drain_all(); |
0c0e6195 | 1542 | yield(); |
b3834be5 | 1543 | /* drain pcp pages, this is synchronous. */ |
9f8f2172 | 1544 | drain_all_pages(); |
0c0e6195 KH |
1545 | /* check again */ |
1546 | offlined_pages = check_pages_isolated(start_pfn, end_pfn); | |
1547 | if (offlined_pages < 0) { | |
1548 | ret = -EBUSY; | |
1549 | goto failed_removal; | |
1550 | } | |
1551 | printk(KERN_INFO "Offlined Pages %ld\n", offlined_pages); | |
b3834be5 | 1552 | /* Ok, all of our target is isolated. |
0c0e6195 KH |
1553 | We cannot do rollback at this point. */ |
1554 | offline_isolated_pages(start_pfn, end_pfn); | |
dbc0e4ce | 1555 | /* reset pagetype flags and makes migrate type to be MOVABLE */ |
0815f3d8 | 1556 | undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE); |
0c0e6195 | 1557 | /* removal success */ |
9feedc9d | 1558 | zone->managed_pages -= offlined_pages; |
0c0e6195 KH |
1559 | zone->present_pages -= offlined_pages; |
1560 | zone->zone_pgdat->node_present_pages -= offlined_pages; | |
1561 | totalram_pages -= offlined_pages; | |
7b78d335 | 1562 | |
1b79acc9 KM |
1563 | init_per_zone_wmark_min(); |
1564 | ||
1e8537ba | 1565 | if (!populated_zone(zone)) { |
340175b7 | 1566 | zone_pcp_reset(zone); |
1e8537ba XQ |
1567 | mutex_lock(&zonelists_mutex); |
1568 | build_all_zonelists(NULL, NULL); | |
1569 | mutex_unlock(&zonelists_mutex); | |
1570 | } else | |
1571 | zone_pcp_update(zone); | |
340175b7 | 1572 | |
d9713679 LJ |
1573 | node_states_clear_node(node, &arg); |
1574 | if (arg.status_change_nid >= 0) | |
8fe23e05 | 1575 | kswapd_stop(node); |
bce7394a | 1576 | |
0c0e6195 KH |
1577 | vm_total_pages = nr_free_pagecache_pages(); |
1578 | writeback_set_ratelimit(); | |
7b78d335 YG |
1579 | |
1580 | memory_notify(MEM_OFFLINE, &arg); | |
20d6c96b | 1581 | unlock_memory_hotplug(); |
0c0e6195 KH |
1582 | return 0; |
1583 | ||
1584 | failed_removal: | |
a62e2f4f BH |
1585 | printk(KERN_INFO "memory offlining [mem %#010llx-%#010llx] failed\n", |
1586 | (unsigned long long) start_pfn << PAGE_SHIFT, | |
1587 | ((unsigned long long) end_pfn << PAGE_SHIFT) - 1); | |
7b78d335 | 1588 | memory_notify(MEM_CANCEL_OFFLINE, &arg); |
0c0e6195 | 1589 | /* pushback to free area */ |
0815f3d8 | 1590 | undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE); |
7b78d335 | 1591 | |
6ad696d2 | 1592 | out: |
20d6c96b | 1593 | unlock_memory_hotplug(); |
0c0e6195 KH |
1594 | return ret; |
1595 | } | |
71088785 | 1596 | |
a16cee10 WC |
1597 | int offline_pages(unsigned long start_pfn, unsigned long nr_pages) |
1598 | { | |
1599 | return __offline_pages(start_pfn, start_pfn + nr_pages, 120 * HZ); | |
1600 | } | |
1601 | ||
bbc76be6 WC |
1602 | /** |
1603 | * walk_memory_range - walks through all mem sections in [start_pfn, end_pfn) | |
1604 | * @start_pfn: start pfn of the memory range | |
1605 | * @end_pfn: end pft of the memory range | |
1606 | * @arg: argument passed to func | |
1607 | * @func: callback for each memory section walked | |
1608 | * | |
1609 | * This function walks through all present mem sections in range | |
1610 | * [start_pfn, end_pfn) and call func on each mem section. | |
1611 | * | |
1612 | * Returns the return value of func. | |
1613 | */ | |
1614 | static int walk_memory_range(unsigned long start_pfn, unsigned long end_pfn, | |
1615 | void *arg, int (*func)(struct memory_block *, void *)) | |
71088785 | 1616 | { |
e90bdb7f WC |
1617 | struct memory_block *mem = NULL; |
1618 | struct mem_section *section; | |
e90bdb7f WC |
1619 | unsigned long pfn, section_nr; |
1620 | int ret; | |
e90bdb7f WC |
1621 | |
1622 | for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) { | |
1623 | section_nr = pfn_to_section_nr(pfn); | |
1624 | if (!present_section_nr(section_nr)) | |
1625 | continue; | |
1626 | ||
1627 | section = __nr_to_section(section_nr); | |
1628 | /* same memblock? */ | |
1629 | if (mem) | |
1630 | if ((section_nr >= mem->start_section_nr) && | |
1631 | (section_nr <= mem->end_section_nr)) | |
1632 | continue; | |
1633 | ||
1634 | mem = find_memory_block_hinted(section, mem); | |
1635 | if (!mem) | |
1636 | continue; | |
1637 | ||
bbc76be6 | 1638 | ret = func(mem, arg); |
e90bdb7f | 1639 | if (ret) { |
bbc76be6 WC |
1640 | kobject_put(&mem->dev.kobj); |
1641 | return ret; | |
e90bdb7f WC |
1642 | } |
1643 | } | |
1644 | ||
1645 | if (mem) | |
1646 | kobject_put(&mem->dev.kobj); | |
1647 | ||
bbc76be6 WC |
1648 | return 0; |
1649 | } | |
1650 | ||
1651 | /** | |
1652 | * offline_memory_block_cb - callback function for offlining memory block | |
1653 | * @mem: the memory block to be offlined | |
1654 | * @arg: buffer to hold error msg | |
1655 | * | |
1656 | * Always return 0, and put the error msg in arg if any. | |
1657 | */ | |
1658 | static int offline_memory_block_cb(struct memory_block *mem, void *arg) | |
1659 | { | |
1660 | int *ret = arg; | |
1661 | int error = offline_memory_block(mem); | |
1662 | ||
1663 | if (error != 0 && *ret == 0) | |
1664 | *ret = error; | |
1665 | ||
1666 | return 0; | |
1667 | } | |
1668 | ||
1669 | static int is_memblock_offlined_cb(struct memory_block *mem, void *arg) | |
1670 | { | |
1671 | int ret = !is_memblock_offlined(mem); | |
1672 | ||
1673 | if (unlikely(ret)) | |
1674 | pr_warn("removing memory fails, because memory " | |
1675 | "[%#010llx-%#010llx] is onlined\n", | |
1676 | PFN_PHYS(section_nr_to_pfn(mem->start_section_nr)), | |
1677 | PFN_PHYS(section_nr_to_pfn(mem->end_section_nr + 1))-1); | |
1678 | ||
1679 | return ret; | |
1680 | } | |
1681 | ||
46c66c4b | 1682 | int __ref remove_memory(u64 start, u64 size) |
bbc76be6 WC |
1683 | { |
1684 | unsigned long start_pfn, end_pfn; | |
1685 | int ret = 0; | |
1686 | int retry = 1; | |
1687 | ||
1688 | start_pfn = PFN_DOWN(start); | |
1689 | end_pfn = start_pfn + PFN_DOWN(size); | |
1690 | ||
1691 | /* | |
1692 | * When CONFIG_MEMCG is on, one memory block may be used by other | |
1693 | * blocks to store page cgroup when onlining pages. But we don't know | |
1694 | * in what order pages are onlined. So we iterate twice to offline | |
1695 | * memory: | |
1696 | * 1st iterate: offline every non primary memory block. | |
1697 | * 2nd iterate: offline primary (i.e. first added) memory block. | |
1698 | */ | |
1699 | repeat: | |
1700 | walk_memory_range(start_pfn, end_pfn, &ret, | |
1701 | offline_memory_block_cb); | |
1702 | if (ret) { | |
1703 | if (!retry) | |
1704 | return ret; | |
1705 | ||
1706 | retry = 0; | |
1707 | ret = 0; | |
993c1aad WC |
1708 | goto repeat; |
1709 | } | |
1710 | ||
6677e3ea YI |
1711 | lock_memory_hotplug(); |
1712 | ||
1713 | /* | |
1714 | * we have offlined all memory blocks like this: | |
1715 | * 1. lock memory hotplug | |
1716 | * 2. offline a memory block | |
1717 | * 3. unlock memory hotplug | |
1718 | * | |
1719 | * repeat step1-3 to offline the memory block. All memory blocks | |
1720 | * must be offlined before removing memory. But we don't hold the | |
1721 | * lock in the whole operation. So we should check whether all | |
1722 | * memory blocks are offlined. | |
1723 | */ | |
1724 | ||
bbc76be6 WC |
1725 | ret = walk_memory_range(start_pfn, end_pfn, NULL, |
1726 | is_memblock_offlined_cb); | |
1727 | if (ret) { | |
1728 | unlock_memory_hotplug(); | |
1729 | return ret; | |
6677e3ea YI |
1730 | } |
1731 | ||
46c66c4b YI |
1732 | /* remove memmap entry */ |
1733 | firmware_map_remove(start, start + size, "System RAM"); | |
1734 | ||
24d335ca WC |
1735 | arch_remove_memory(start, size); |
1736 | ||
6677e3ea YI |
1737 | unlock_memory_hotplug(); |
1738 | ||
e90bdb7f | 1739 | return 0; |
71088785 | 1740 | } |
48e94196 | 1741 | #else |
a16cee10 WC |
1742 | int offline_pages(unsigned long start_pfn, unsigned long nr_pages) |
1743 | { | |
1744 | return -EINVAL; | |
1745 | } | |
48e94196 KH |
1746 | int remove_memory(u64 start, u64 size) |
1747 | { | |
1748 | return -EINVAL; | |
1749 | } | |
0c0e6195 | 1750 | #endif /* CONFIG_MEMORY_HOTREMOVE */ |
71088785 | 1751 | EXPORT_SYMBOL_GPL(remove_memory); |