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) { |
4996eed8 | 78 | pr_debug("System RAM resource %pR cannot be added\n", res); |
45e0b78b KM |
79 | kfree(res); |
80 | res = NULL; | |
81 | } | |
82 | return res; | |
83 | } | |
84 | ||
85 | static void release_memory_resource(struct resource *res) | |
86 | { | |
87 | if (!res) | |
88 | return; | |
89 | release_resource(res); | |
90 | kfree(res); | |
91 | return; | |
92 | } | |
93 | ||
53947027 | 94 | #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE |
46723bfa YI |
95 | void get_page_bootmem(unsigned long info, struct page *page, |
96 | unsigned long type) | |
04753278 | 97 | { |
5f24ce5f | 98 | page->lru.next = (struct list_head *) type; |
04753278 YG |
99 | SetPagePrivate(page); |
100 | set_page_private(page, info); | |
101 | atomic_inc(&page->_count); | |
102 | } | |
103 | ||
23ce932a RM |
104 | /* reference to __meminit __free_pages_bootmem is valid |
105 | * so use __ref to tell modpost not to generate a warning */ | |
106 | void __ref put_page_bootmem(struct page *page) | |
04753278 | 107 | { |
5f24ce5f | 108 | unsigned long type; |
9feedc9d | 109 | static DEFINE_MUTEX(ppb_lock); |
04753278 | 110 | |
5f24ce5f AA |
111 | type = (unsigned long) page->lru.next; |
112 | BUG_ON(type < MEMORY_HOTPLUG_MIN_BOOTMEM_TYPE || | |
113 | type > MEMORY_HOTPLUG_MAX_BOOTMEM_TYPE); | |
04753278 YG |
114 | |
115 | if (atomic_dec_return(&page->_count) == 1) { | |
116 | ClearPagePrivate(page); | |
117 | set_page_private(page, 0); | |
5f24ce5f | 118 | INIT_LIST_HEAD(&page->lru); |
9feedc9d JL |
119 | |
120 | /* | |
121 | * Please refer to comment for __free_pages_bootmem() | |
122 | * for why we serialize here. | |
123 | */ | |
124 | mutex_lock(&ppb_lock); | |
04753278 | 125 | __free_pages_bootmem(page, 0); |
9feedc9d | 126 | mutex_unlock(&ppb_lock); |
c60514b6 | 127 | totalram_pages++; |
04753278 YG |
128 | } |
129 | ||
130 | } | |
131 | ||
46723bfa YI |
132 | #ifdef CONFIG_HAVE_BOOTMEM_INFO_NODE |
133 | #ifndef CONFIG_SPARSEMEM_VMEMMAP | |
d92bc318 | 134 | static void register_page_bootmem_info_section(unsigned long start_pfn) |
04753278 YG |
135 | { |
136 | unsigned long *usemap, mapsize, section_nr, i; | |
137 | struct mem_section *ms; | |
138 | struct page *page, *memmap; | |
139 | ||
04753278 YG |
140 | section_nr = pfn_to_section_nr(start_pfn); |
141 | ms = __nr_to_section(section_nr); | |
142 | ||
143 | /* Get section's memmap address */ | |
144 | memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr); | |
145 | ||
146 | /* | |
147 | * Get page for the memmap's phys address | |
148 | * XXX: need more consideration for sparse_vmemmap... | |
149 | */ | |
150 | page = virt_to_page(memmap); | |
151 | mapsize = sizeof(struct page) * PAGES_PER_SECTION; | |
152 | mapsize = PAGE_ALIGN(mapsize) >> PAGE_SHIFT; | |
153 | ||
154 | /* remember memmap's page */ | |
155 | for (i = 0; i < mapsize; i++, page++) | |
156 | get_page_bootmem(section_nr, page, SECTION_INFO); | |
157 | ||
158 | usemap = __nr_to_section(section_nr)->pageblock_flags; | |
159 | page = virt_to_page(usemap); | |
160 | ||
161 | mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT; | |
162 | ||
163 | for (i = 0; i < mapsize; i++, page++) | |
af370fb8 | 164 | get_page_bootmem(section_nr, page, MIX_SECTION_INFO); |
04753278 YG |
165 | |
166 | } | |
46723bfa YI |
167 | #else /* CONFIG_SPARSEMEM_VMEMMAP */ |
168 | static void register_page_bootmem_info_section(unsigned long start_pfn) | |
169 | { | |
170 | unsigned long *usemap, mapsize, section_nr, i; | |
171 | struct mem_section *ms; | |
172 | struct page *page, *memmap; | |
173 | ||
174 | if (!pfn_valid(start_pfn)) | |
175 | return; | |
176 | ||
177 | section_nr = pfn_to_section_nr(start_pfn); | |
178 | ms = __nr_to_section(section_nr); | |
179 | ||
180 | memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr); | |
181 | ||
182 | register_page_bootmem_memmap(section_nr, memmap, PAGES_PER_SECTION); | |
183 | ||
184 | usemap = __nr_to_section(section_nr)->pageblock_flags; | |
185 | page = virt_to_page(usemap); | |
186 | ||
187 | mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT; | |
188 | ||
189 | for (i = 0; i < mapsize; i++, page++) | |
190 | get_page_bootmem(section_nr, page, MIX_SECTION_INFO); | |
191 | } | |
192 | #endif /* !CONFIG_SPARSEMEM_VMEMMAP */ | |
04753278 YG |
193 | |
194 | void register_page_bootmem_info_node(struct pglist_data *pgdat) | |
195 | { | |
196 | unsigned long i, pfn, end_pfn, nr_pages; | |
197 | int node = pgdat->node_id; | |
198 | struct page *page; | |
199 | struct zone *zone; | |
200 | ||
201 | nr_pages = PAGE_ALIGN(sizeof(struct pglist_data)) >> PAGE_SHIFT; | |
202 | page = virt_to_page(pgdat); | |
203 | ||
204 | for (i = 0; i < nr_pages; i++, page++) | |
205 | get_page_bootmem(node, page, NODE_INFO); | |
206 | ||
207 | zone = &pgdat->node_zones[0]; | |
208 | for (; zone < pgdat->node_zones + MAX_NR_ZONES - 1; zone++) { | |
209 | if (zone->wait_table) { | |
210 | nr_pages = zone->wait_table_hash_nr_entries | |
211 | * sizeof(wait_queue_head_t); | |
212 | nr_pages = PAGE_ALIGN(nr_pages) >> PAGE_SHIFT; | |
213 | page = virt_to_page(zone->wait_table); | |
214 | ||
215 | for (i = 0; i < nr_pages; i++, page++) | |
216 | get_page_bootmem(node, page, NODE_INFO); | |
217 | } | |
218 | } | |
219 | ||
220 | pfn = pgdat->node_start_pfn; | |
c1f19495 | 221 | end_pfn = pgdat_end_pfn(pgdat); |
04753278 YG |
222 | |
223 | /* register_section info */ | |
f14851af | 224 | for (; pfn < end_pfn; pfn += PAGES_PER_SECTION) { |
225 | /* | |
226 | * Some platforms can assign the same pfn to multiple nodes - on | |
227 | * node0 as well as nodeN. To avoid registering a pfn against | |
228 | * multiple nodes we check that this pfn does not already | |
229 | * reside in some other node. | |
230 | */ | |
231 | if (pfn_valid(pfn) && (pfn_to_nid(pfn) == node)) | |
232 | register_page_bootmem_info_section(pfn); | |
233 | } | |
04753278 | 234 | } |
46723bfa | 235 | #endif /* CONFIG_HAVE_BOOTMEM_INFO_NODE */ |
04753278 | 236 | |
76cdd58e HC |
237 | static void grow_zone_span(struct zone *zone, unsigned long start_pfn, |
238 | unsigned long end_pfn) | |
239 | { | |
240 | unsigned long old_zone_end_pfn; | |
241 | ||
242 | zone_span_writelock(zone); | |
243 | ||
244 | old_zone_end_pfn = zone->zone_start_pfn + zone->spanned_pages; | |
712cd386 | 245 | if (!zone->spanned_pages || start_pfn < zone->zone_start_pfn) |
76cdd58e HC |
246 | zone->zone_start_pfn = start_pfn; |
247 | ||
248 | zone->spanned_pages = max(old_zone_end_pfn, end_pfn) - | |
249 | zone->zone_start_pfn; | |
250 | ||
251 | zone_span_writeunlock(zone); | |
252 | } | |
253 | ||
511c2aba LJ |
254 | static void resize_zone(struct zone *zone, unsigned long start_pfn, |
255 | unsigned long end_pfn) | |
256 | { | |
257 | zone_span_writelock(zone); | |
258 | ||
e455a9b9 LJ |
259 | if (end_pfn - start_pfn) { |
260 | zone->zone_start_pfn = start_pfn; | |
261 | zone->spanned_pages = end_pfn - start_pfn; | |
262 | } else { | |
263 | /* | |
264 | * make it consist as free_area_init_core(), | |
265 | * if spanned_pages = 0, then keep start_pfn = 0 | |
266 | */ | |
267 | zone->zone_start_pfn = 0; | |
268 | zone->spanned_pages = 0; | |
269 | } | |
511c2aba LJ |
270 | |
271 | zone_span_writeunlock(zone); | |
272 | } | |
273 | ||
274 | static void fix_zone_id(struct zone *zone, unsigned long start_pfn, | |
275 | unsigned long end_pfn) | |
276 | { | |
277 | enum zone_type zid = zone_idx(zone); | |
278 | int nid = zone->zone_pgdat->node_id; | |
279 | unsigned long pfn; | |
280 | ||
281 | for (pfn = start_pfn; pfn < end_pfn; pfn++) | |
282 | set_page_links(pfn_to_page(pfn), zid, nid, pfn); | |
283 | } | |
284 | ||
f6bbb78e CS |
285 | /* Can fail with -ENOMEM from allocating a wait table with vmalloc() or |
286 | * alloc_bootmem_node_nopanic() */ | |
287 | static int __ref ensure_zone_is_initialized(struct zone *zone, | |
288 | unsigned long start_pfn, unsigned long num_pages) | |
289 | { | |
290 | if (!zone_is_initialized(zone)) | |
291 | return init_currently_empty_zone(zone, start_pfn, num_pages, | |
292 | MEMMAP_HOTPLUG); | |
293 | return 0; | |
294 | } | |
295 | ||
e455a9b9 | 296 | static int __meminit move_pfn_range_left(struct zone *z1, struct zone *z2, |
511c2aba LJ |
297 | unsigned long start_pfn, unsigned long end_pfn) |
298 | { | |
e455a9b9 | 299 | int ret; |
511c2aba | 300 | unsigned long flags; |
e455a9b9 LJ |
301 | unsigned long z1_start_pfn; |
302 | ||
64dd1b29 CS |
303 | ret = ensure_zone_is_initialized(z1, start_pfn, end_pfn - start_pfn); |
304 | if (ret) | |
305 | return ret; | |
511c2aba LJ |
306 | |
307 | pgdat_resize_lock(z1->zone_pgdat, &flags); | |
308 | ||
309 | /* can't move pfns which are higher than @z2 */ | |
108bcc96 | 310 | if (end_pfn > zone_end_pfn(z2)) |
511c2aba | 311 | goto out_fail; |
834405c3 | 312 | /* the move out part must be at the left most of @z2 */ |
511c2aba LJ |
313 | if (start_pfn > z2->zone_start_pfn) |
314 | goto out_fail; | |
315 | /* must included/overlap */ | |
316 | if (end_pfn <= z2->zone_start_pfn) | |
317 | goto out_fail; | |
318 | ||
e455a9b9 LJ |
319 | /* use start_pfn for z1's start_pfn if z1 is empty */ |
320 | if (z1->spanned_pages) | |
321 | z1_start_pfn = z1->zone_start_pfn; | |
322 | else | |
323 | z1_start_pfn = start_pfn; | |
324 | ||
325 | resize_zone(z1, z1_start_pfn, end_pfn); | |
108bcc96 | 326 | resize_zone(z2, end_pfn, zone_end_pfn(z2)); |
511c2aba LJ |
327 | |
328 | pgdat_resize_unlock(z1->zone_pgdat, &flags); | |
329 | ||
330 | fix_zone_id(z1, start_pfn, end_pfn); | |
331 | ||
332 | return 0; | |
333 | out_fail: | |
334 | pgdat_resize_unlock(z1->zone_pgdat, &flags); | |
335 | return -1; | |
336 | } | |
337 | ||
e455a9b9 | 338 | static int __meminit move_pfn_range_right(struct zone *z1, struct zone *z2, |
511c2aba LJ |
339 | unsigned long start_pfn, unsigned long end_pfn) |
340 | { | |
e455a9b9 | 341 | int ret; |
511c2aba | 342 | unsigned long flags; |
e455a9b9 LJ |
343 | unsigned long z2_end_pfn; |
344 | ||
64dd1b29 CS |
345 | ret = ensure_zone_is_initialized(z2, start_pfn, end_pfn - start_pfn); |
346 | if (ret) | |
347 | return ret; | |
511c2aba LJ |
348 | |
349 | pgdat_resize_lock(z1->zone_pgdat, &flags); | |
350 | ||
351 | /* can't move pfns which are lower than @z1 */ | |
352 | if (z1->zone_start_pfn > start_pfn) | |
353 | goto out_fail; | |
354 | /* the move out part mast at the right most of @z1 */ | |
108bcc96 | 355 | if (zone_end_pfn(z1) > end_pfn) |
511c2aba LJ |
356 | goto out_fail; |
357 | /* must included/overlap */ | |
108bcc96 | 358 | if (start_pfn >= zone_end_pfn(z1)) |
511c2aba LJ |
359 | goto out_fail; |
360 | ||
e455a9b9 LJ |
361 | /* use end_pfn for z2's end_pfn if z2 is empty */ |
362 | if (z2->spanned_pages) | |
108bcc96 | 363 | z2_end_pfn = zone_end_pfn(z2); |
e455a9b9 LJ |
364 | else |
365 | z2_end_pfn = end_pfn; | |
366 | ||
511c2aba | 367 | resize_zone(z1, z1->zone_start_pfn, start_pfn); |
e455a9b9 | 368 | resize_zone(z2, start_pfn, z2_end_pfn); |
511c2aba LJ |
369 | |
370 | pgdat_resize_unlock(z1->zone_pgdat, &flags); | |
371 | ||
372 | fix_zone_id(z2, start_pfn, end_pfn); | |
373 | ||
374 | return 0; | |
375 | out_fail: | |
376 | pgdat_resize_unlock(z1->zone_pgdat, &flags); | |
377 | return -1; | |
378 | } | |
379 | ||
76cdd58e HC |
380 | static void grow_pgdat_span(struct pglist_data *pgdat, unsigned long start_pfn, |
381 | unsigned long end_pfn) | |
382 | { | |
383 | unsigned long old_pgdat_end_pfn = | |
384 | pgdat->node_start_pfn + pgdat->node_spanned_pages; | |
385 | ||
712cd386 | 386 | if (!pgdat->node_spanned_pages || start_pfn < pgdat->node_start_pfn) |
76cdd58e HC |
387 | pgdat->node_start_pfn = start_pfn; |
388 | ||
389 | pgdat->node_spanned_pages = max(old_pgdat_end_pfn, end_pfn) - | |
390 | pgdat->node_start_pfn; | |
391 | } | |
392 | ||
31168481 | 393 | static int __meminit __add_zone(struct zone *zone, unsigned long phys_start_pfn) |
3947be19 DH |
394 | { |
395 | struct pglist_data *pgdat = zone->zone_pgdat; | |
396 | int nr_pages = PAGES_PER_SECTION; | |
397 | int nid = pgdat->node_id; | |
398 | int zone_type; | |
76cdd58e | 399 | unsigned long flags; |
64dd1b29 | 400 | int ret; |
3947be19 DH |
401 | |
402 | zone_type = zone - pgdat->node_zones; | |
64dd1b29 CS |
403 | ret = ensure_zone_is_initialized(zone, phys_start_pfn, nr_pages); |
404 | if (ret) | |
405 | return ret; | |
76cdd58e | 406 | |
76cdd58e HC |
407 | pgdat_resize_lock(zone->zone_pgdat, &flags); |
408 | grow_zone_span(zone, phys_start_pfn, phys_start_pfn + nr_pages); | |
409 | grow_pgdat_span(zone->zone_pgdat, phys_start_pfn, | |
410 | phys_start_pfn + nr_pages); | |
411 | pgdat_resize_unlock(zone->zone_pgdat, &flags); | |
a2f3aa02 DH |
412 | memmap_init_zone(nr_pages, nid, zone_type, |
413 | phys_start_pfn, MEMMAP_HOTPLUG); | |
718127cc | 414 | return 0; |
3947be19 DH |
415 | } |
416 | ||
c04fc586 GH |
417 | static int __meminit __add_section(int nid, struct zone *zone, |
418 | unsigned long phys_start_pfn) | |
3947be19 | 419 | { |
3947be19 | 420 | int nr_pages = PAGES_PER_SECTION; |
3947be19 DH |
421 | int ret; |
422 | ||
ebd15302 KH |
423 | if (pfn_valid(phys_start_pfn)) |
424 | return -EEXIST; | |
425 | ||
0b0acbec | 426 | ret = sparse_add_one_section(zone, phys_start_pfn, nr_pages); |
3947be19 DH |
427 | |
428 | if (ret < 0) | |
429 | return ret; | |
430 | ||
718127cc YG |
431 | ret = __add_zone(zone, phys_start_pfn); |
432 | ||
433 | if (ret < 0) | |
434 | return ret; | |
435 | ||
c04fc586 | 436 | return register_new_memory(nid, __pfn_to_section(phys_start_pfn)); |
3947be19 DH |
437 | } |
438 | ||
4edd7cef DR |
439 | /* |
440 | * Reasonably generic function for adding memory. It is | |
441 | * expected that archs that support memory hotplug will | |
442 | * call this function after deciding the zone to which to | |
443 | * add the new pages. | |
444 | */ | |
445 | int __ref __add_pages(int nid, struct zone *zone, unsigned long phys_start_pfn, | |
446 | unsigned long nr_pages) | |
447 | { | |
448 | unsigned long i; | |
449 | int err = 0; | |
450 | int start_sec, end_sec; | |
451 | /* during initialize mem_map, align hot-added range to section */ | |
452 | start_sec = pfn_to_section_nr(phys_start_pfn); | |
453 | end_sec = pfn_to_section_nr(phys_start_pfn + nr_pages - 1); | |
454 | ||
455 | for (i = start_sec; i <= end_sec; i++) { | |
456 | err = __add_section(nid, zone, i << PFN_SECTION_SHIFT); | |
457 | ||
458 | /* | |
459 | * EEXIST is finally dealt with by ioresource collision | |
460 | * check. see add_memory() => register_memory_resource() | |
461 | * Warning will be printed if there is collision. | |
462 | */ | |
463 | if (err && (err != -EEXIST)) | |
464 | break; | |
465 | err = 0; | |
466 | } | |
467 | ||
468 | return err; | |
469 | } | |
470 | EXPORT_SYMBOL_GPL(__add_pages); | |
471 | ||
472 | #ifdef CONFIG_MEMORY_HOTREMOVE | |
815121d2 YI |
473 | /* find the smallest valid pfn in the range [start_pfn, end_pfn) */ |
474 | static int find_smallest_section_pfn(int nid, struct zone *zone, | |
475 | unsigned long start_pfn, | |
476 | unsigned long end_pfn) | |
477 | { | |
478 | struct mem_section *ms; | |
479 | ||
480 | for (; start_pfn < end_pfn; start_pfn += PAGES_PER_SECTION) { | |
481 | ms = __pfn_to_section(start_pfn); | |
482 | ||
483 | if (unlikely(!valid_section(ms))) | |
484 | continue; | |
485 | ||
486 | if (unlikely(pfn_to_nid(start_pfn) != nid)) | |
487 | continue; | |
488 | ||
489 | if (zone && zone != page_zone(pfn_to_page(start_pfn))) | |
490 | continue; | |
491 | ||
492 | return start_pfn; | |
493 | } | |
494 | ||
495 | return 0; | |
496 | } | |
497 | ||
498 | /* find the biggest valid pfn in the range [start_pfn, end_pfn). */ | |
499 | static int find_biggest_section_pfn(int nid, struct zone *zone, | |
500 | unsigned long start_pfn, | |
501 | unsigned long end_pfn) | |
502 | { | |
503 | struct mem_section *ms; | |
504 | unsigned long pfn; | |
505 | ||
506 | /* pfn is the end pfn of a memory section. */ | |
507 | pfn = end_pfn - 1; | |
508 | for (; pfn >= start_pfn; pfn -= PAGES_PER_SECTION) { | |
509 | ms = __pfn_to_section(pfn); | |
510 | ||
511 | if (unlikely(!valid_section(ms))) | |
512 | continue; | |
513 | ||
514 | if (unlikely(pfn_to_nid(pfn) != nid)) | |
515 | continue; | |
516 | ||
517 | if (zone && zone != page_zone(pfn_to_page(pfn))) | |
518 | continue; | |
519 | ||
520 | return pfn; | |
521 | } | |
522 | ||
523 | return 0; | |
524 | } | |
525 | ||
526 | static void shrink_zone_span(struct zone *zone, unsigned long start_pfn, | |
527 | unsigned long end_pfn) | |
528 | { | |
529 | unsigned long zone_start_pfn = zone->zone_start_pfn; | |
530 | unsigned long zone_end_pfn = zone->zone_start_pfn + zone->spanned_pages; | |
531 | unsigned long pfn; | |
532 | struct mem_section *ms; | |
533 | int nid = zone_to_nid(zone); | |
534 | ||
535 | zone_span_writelock(zone); | |
536 | if (zone_start_pfn == start_pfn) { | |
537 | /* | |
538 | * If the section is smallest section in the zone, it need | |
539 | * shrink zone->zone_start_pfn and zone->zone_spanned_pages. | |
540 | * In this case, we find second smallest valid mem_section | |
541 | * for shrinking zone. | |
542 | */ | |
543 | pfn = find_smallest_section_pfn(nid, zone, end_pfn, | |
544 | zone_end_pfn); | |
545 | if (pfn) { | |
546 | zone->zone_start_pfn = pfn; | |
547 | zone->spanned_pages = zone_end_pfn - pfn; | |
548 | } | |
549 | } else if (zone_end_pfn == end_pfn) { | |
550 | /* | |
551 | * If the section is biggest section in the zone, it need | |
552 | * shrink zone->spanned_pages. | |
553 | * In this case, we find second biggest valid mem_section for | |
554 | * shrinking zone. | |
555 | */ | |
556 | pfn = find_biggest_section_pfn(nid, zone, zone_start_pfn, | |
557 | start_pfn); | |
558 | if (pfn) | |
559 | zone->spanned_pages = pfn - zone_start_pfn + 1; | |
560 | } | |
561 | ||
562 | /* | |
563 | * The section is not biggest or smallest mem_section in the zone, it | |
564 | * only creates a hole in the zone. So in this case, we need not | |
565 | * change the zone. But perhaps, the zone has only hole data. Thus | |
566 | * it check the zone has only hole or not. | |
567 | */ | |
568 | pfn = zone_start_pfn; | |
569 | for (; pfn < zone_end_pfn; pfn += PAGES_PER_SECTION) { | |
570 | ms = __pfn_to_section(pfn); | |
571 | ||
572 | if (unlikely(!valid_section(ms))) | |
573 | continue; | |
574 | ||
575 | if (page_zone(pfn_to_page(pfn)) != zone) | |
576 | continue; | |
577 | ||
578 | /* If the section is current section, it continues the loop */ | |
579 | if (start_pfn == pfn) | |
580 | continue; | |
581 | ||
582 | /* If we find valid section, we have nothing to do */ | |
583 | zone_span_writeunlock(zone); | |
584 | return; | |
585 | } | |
586 | ||
587 | /* The zone has no valid section */ | |
588 | zone->zone_start_pfn = 0; | |
589 | zone->spanned_pages = 0; | |
590 | zone_span_writeunlock(zone); | |
591 | } | |
592 | ||
593 | static void shrink_pgdat_span(struct pglist_data *pgdat, | |
594 | unsigned long start_pfn, unsigned long end_pfn) | |
595 | { | |
596 | unsigned long pgdat_start_pfn = pgdat->node_start_pfn; | |
597 | unsigned long pgdat_end_pfn = | |
598 | pgdat->node_start_pfn + pgdat->node_spanned_pages; | |
599 | unsigned long pfn; | |
600 | struct mem_section *ms; | |
601 | int nid = pgdat->node_id; | |
602 | ||
603 | if (pgdat_start_pfn == start_pfn) { | |
604 | /* | |
605 | * If the section is smallest section in the pgdat, it need | |
606 | * shrink pgdat->node_start_pfn and pgdat->node_spanned_pages. | |
607 | * In this case, we find second smallest valid mem_section | |
608 | * for shrinking zone. | |
609 | */ | |
610 | pfn = find_smallest_section_pfn(nid, NULL, end_pfn, | |
611 | pgdat_end_pfn); | |
612 | if (pfn) { | |
613 | pgdat->node_start_pfn = pfn; | |
614 | pgdat->node_spanned_pages = pgdat_end_pfn - pfn; | |
615 | } | |
616 | } else if (pgdat_end_pfn == end_pfn) { | |
617 | /* | |
618 | * If the section is biggest section in the pgdat, it need | |
619 | * shrink pgdat->node_spanned_pages. | |
620 | * In this case, we find second biggest valid mem_section for | |
621 | * shrinking zone. | |
622 | */ | |
623 | pfn = find_biggest_section_pfn(nid, NULL, pgdat_start_pfn, | |
624 | start_pfn); | |
625 | if (pfn) | |
626 | pgdat->node_spanned_pages = pfn - pgdat_start_pfn + 1; | |
627 | } | |
628 | ||
629 | /* | |
630 | * If the section is not biggest or smallest mem_section in the pgdat, | |
631 | * it only creates a hole in the pgdat. So in this case, we need not | |
632 | * change the pgdat. | |
633 | * But perhaps, the pgdat has only hole data. Thus it check the pgdat | |
634 | * has only hole or not. | |
635 | */ | |
636 | pfn = pgdat_start_pfn; | |
637 | for (; pfn < pgdat_end_pfn; pfn += PAGES_PER_SECTION) { | |
638 | ms = __pfn_to_section(pfn); | |
639 | ||
640 | if (unlikely(!valid_section(ms))) | |
641 | continue; | |
642 | ||
643 | if (pfn_to_nid(pfn) != nid) | |
644 | continue; | |
645 | ||
646 | /* If the section is current section, it continues the loop */ | |
647 | if (start_pfn == pfn) | |
648 | continue; | |
649 | ||
650 | /* If we find valid section, we have nothing to do */ | |
651 | return; | |
652 | } | |
653 | ||
654 | /* The pgdat has no valid section */ | |
655 | pgdat->node_start_pfn = 0; | |
656 | pgdat->node_spanned_pages = 0; | |
657 | } | |
658 | ||
659 | static void __remove_zone(struct zone *zone, unsigned long start_pfn) | |
660 | { | |
661 | struct pglist_data *pgdat = zone->zone_pgdat; | |
662 | int nr_pages = PAGES_PER_SECTION; | |
663 | int zone_type; | |
664 | unsigned long flags; | |
665 | ||
666 | zone_type = zone - pgdat->node_zones; | |
667 | ||
668 | pgdat_resize_lock(zone->zone_pgdat, &flags); | |
669 | shrink_zone_span(zone, start_pfn, start_pfn + nr_pages); | |
670 | shrink_pgdat_span(pgdat, start_pfn, start_pfn + nr_pages); | |
671 | pgdat_resize_unlock(zone->zone_pgdat, &flags); | |
672 | } | |
673 | ||
ea01ea93 BP |
674 | static int __remove_section(struct zone *zone, struct mem_section *ms) |
675 | { | |
815121d2 YI |
676 | unsigned long start_pfn; |
677 | int scn_nr; | |
ea01ea93 BP |
678 | int ret = -EINVAL; |
679 | ||
680 | if (!valid_section(ms)) | |
681 | return ret; | |
682 | ||
683 | ret = unregister_memory_section(ms); | |
684 | if (ret) | |
685 | return ret; | |
686 | ||
815121d2 YI |
687 | scn_nr = __section_nr(ms); |
688 | start_pfn = section_nr_to_pfn(scn_nr); | |
689 | __remove_zone(zone, start_pfn); | |
690 | ||
ea01ea93 | 691 | sparse_remove_one_section(zone, ms); |
ea01ea93 BP |
692 | return 0; |
693 | } | |
694 | ||
ea01ea93 BP |
695 | /** |
696 | * __remove_pages() - remove sections of pages from a zone | |
697 | * @zone: zone from which pages need to be removed | |
698 | * @phys_start_pfn: starting pageframe (must be aligned to start of a section) | |
699 | * @nr_pages: number of pages to remove (must be multiple of section size) | |
700 | * | |
701 | * Generic helper function to remove section mappings and sysfs entries | |
702 | * for the section of the memory we are removing. Caller needs to make | |
703 | * sure that pages are marked reserved and zones are adjust properly by | |
704 | * calling offline_pages(). | |
705 | */ | |
706 | int __remove_pages(struct zone *zone, unsigned long phys_start_pfn, | |
707 | unsigned long nr_pages) | |
708 | { | |
fe74ebb1 | 709 | unsigned long i; |
ea01ea93 | 710 | int sections_to_remove; |
fe74ebb1 TK |
711 | resource_size_t start, size; |
712 | int ret = 0; | |
ea01ea93 BP |
713 | |
714 | /* | |
715 | * We can only remove entire sections | |
716 | */ | |
717 | BUG_ON(phys_start_pfn & ~PAGE_SECTION_MASK); | |
718 | BUG_ON(nr_pages % PAGES_PER_SECTION); | |
719 | ||
fe74ebb1 TK |
720 | start = phys_start_pfn << PAGE_SHIFT; |
721 | size = nr_pages * PAGE_SIZE; | |
722 | ret = release_mem_region_adjustable(&iomem_resource, start, size); | |
348f9f05 RD |
723 | if (ret) { |
724 | resource_size_t endres = start + size - 1; | |
725 | ||
726 | pr_warn("Unable to release resource <%pa-%pa> (%d)\n", | |
727 | &start, &endres, ret); | |
728 | } | |
d760afd4 | 729 | |
ea01ea93 BP |
730 | sections_to_remove = nr_pages / PAGES_PER_SECTION; |
731 | for (i = 0; i < sections_to_remove; i++) { | |
732 | unsigned long pfn = phys_start_pfn + i*PAGES_PER_SECTION; | |
733 | ret = __remove_section(zone, __pfn_to_section(pfn)); | |
734 | if (ret) | |
735 | break; | |
736 | } | |
737 | return ret; | |
738 | } | |
739 | EXPORT_SYMBOL_GPL(__remove_pages); | |
4edd7cef | 740 | #endif /* CONFIG_MEMORY_HOTREMOVE */ |
ea01ea93 | 741 | |
9d0ad8ca DK |
742 | int set_online_page_callback(online_page_callback_t callback) |
743 | { | |
744 | int rc = -EINVAL; | |
745 | ||
746 | lock_memory_hotplug(); | |
747 | ||
748 | if (online_page_callback == generic_online_page) { | |
749 | online_page_callback = callback; | |
750 | rc = 0; | |
751 | } | |
752 | ||
753 | unlock_memory_hotplug(); | |
754 | ||
755 | return rc; | |
756 | } | |
757 | EXPORT_SYMBOL_GPL(set_online_page_callback); | |
758 | ||
759 | int restore_online_page_callback(online_page_callback_t callback) | |
760 | { | |
761 | int rc = -EINVAL; | |
762 | ||
763 | lock_memory_hotplug(); | |
764 | ||
765 | if (online_page_callback == callback) { | |
766 | online_page_callback = generic_online_page; | |
767 | rc = 0; | |
768 | } | |
769 | ||
770 | unlock_memory_hotplug(); | |
771 | ||
772 | return rc; | |
773 | } | |
774 | EXPORT_SYMBOL_GPL(restore_online_page_callback); | |
775 | ||
776 | void __online_page_set_limits(struct page *page) | |
180c06ef | 777 | { |
4738e1b9 JB |
778 | unsigned long pfn = page_to_pfn(page); |
779 | ||
4738e1b9 JB |
780 | if (pfn >= num_physpages) |
781 | num_physpages = pfn + 1; | |
9d0ad8ca DK |
782 | } |
783 | EXPORT_SYMBOL_GPL(__online_page_set_limits); | |
784 | ||
785 | void __online_page_increment_counters(struct page *page) | |
786 | { | |
787 | totalram_pages++; | |
180c06ef JF |
788 | |
789 | #ifdef CONFIG_HIGHMEM | |
790 | if (PageHighMem(page)) | |
791 | totalhigh_pages++; | |
792 | #endif | |
9d0ad8ca DK |
793 | } |
794 | EXPORT_SYMBOL_GPL(__online_page_increment_counters); | |
180c06ef | 795 | |
9d0ad8ca DK |
796 | void __online_page_free(struct page *page) |
797 | { | |
180c06ef JF |
798 | ClearPageReserved(page); |
799 | init_page_count(page); | |
800 | __free_page(page); | |
801 | } | |
9d0ad8ca DK |
802 | EXPORT_SYMBOL_GPL(__online_page_free); |
803 | ||
804 | static void generic_online_page(struct page *page) | |
805 | { | |
806 | __online_page_set_limits(page); | |
807 | __online_page_increment_counters(page); | |
808 | __online_page_free(page); | |
809 | } | |
180c06ef | 810 | |
75884fb1 KH |
811 | static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages, |
812 | void *arg) | |
3947be19 DH |
813 | { |
814 | unsigned long i; | |
75884fb1 KH |
815 | unsigned long onlined_pages = *(unsigned long *)arg; |
816 | struct page *page; | |
817 | if (PageReserved(pfn_to_page(start_pfn))) | |
818 | for (i = 0; i < nr_pages; i++) { | |
819 | page = pfn_to_page(start_pfn + i); | |
9d0ad8ca | 820 | (*online_page_callback)(page); |
75884fb1 KH |
821 | onlined_pages++; |
822 | } | |
823 | *(unsigned long *)arg = onlined_pages; | |
824 | return 0; | |
825 | } | |
826 | ||
09285af7 | 827 | #ifdef CONFIG_MOVABLE_NODE |
79a4dcef TC |
828 | /* |
829 | * When CONFIG_MOVABLE_NODE, we permit onlining of a node which doesn't have | |
830 | * normal memory. | |
831 | */ | |
09285af7 LJ |
832 | static bool can_online_high_movable(struct zone *zone) |
833 | { | |
834 | return true; | |
835 | } | |
79a4dcef | 836 | #else /* CONFIG_MOVABLE_NODE */ |
74d42d8f LJ |
837 | /* ensure every online node has NORMAL memory */ |
838 | static bool can_online_high_movable(struct zone *zone) | |
839 | { | |
840 | return node_state(zone_to_nid(zone), N_NORMAL_MEMORY); | |
841 | } | |
79a4dcef | 842 | #endif /* CONFIG_MOVABLE_NODE */ |
74d42d8f | 843 | |
d9713679 LJ |
844 | /* check which state of node_states will be changed when online memory */ |
845 | static void node_states_check_changes_online(unsigned long nr_pages, | |
846 | struct zone *zone, struct memory_notify *arg) | |
847 | { | |
848 | int nid = zone_to_nid(zone); | |
849 | enum zone_type zone_last = ZONE_NORMAL; | |
850 | ||
851 | /* | |
6715ddf9 LJ |
852 | * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY] |
853 | * contains nodes which have zones of 0...ZONE_NORMAL, | |
854 | * set zone_last to ZONE_NORMAL. | |
d9713679 | 855 | * |
6715ddf9 LJ |
856 | * If we don't have HIGHMEM nor movable node, |
857 | * node_states[N_NORMAL_MEMORY] contains nodes which have zones of | |
858 | * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE. | |
d9713679 | 859 | */ |
6715ddf9 | 860 | if (N_MEMORY == N_NORMAL_MEMORY) |
d9713679 LJ |
861 | zone_last = ZONE_MOVABLE; |
862 | ||
863 | /* | |
864 | * if the memory to be online is in a zone of 0...zone_last, and | |
865 | * the zones of 0...zone_last don't have memory before online, we will | |
866 | * need to set the node to node_states[N_NORMAL_MEMORY] after | |
867 | * the memory is online. | |
868 | */ | |
869 | if (zone_idx(zone) <= zone_last && !node_state(nid, N_NORMAL_MEMORY)) | |
870 | arg->status_change_nid_normal = nid; | |
871 | else | |
872 | arg->status_change_nid_normal = -1; | |
873 | ||
6715ddf9 LJ |
874 | #ifdef CONFIG_HIGHMEM |
875 | /* | |
876 | * If we have movable node, node_states[N_HIGH_MEMORY] | |
877 | * contains nodes which have zones of 0...ZONE_HIGHMEM, | |
878 | * set zone_last to ZONE_HIGHMEM. | |
879 | * | |
880 | * If we don't have movable node, node_states[N_NORMAL_MEMORY] | |
881 | * contains nodes which have zones of 0...ZONE_MOVABLE, | |
882 | * set zone_last to ZONE_MOVABLE. | |
883 | */ | |
884 | zone_last = ZONE_HIGHMEM; | |
885 | if (N_MEMORY == N_HIGH_MEMORY) | |
886 | zone_last = ZONE_MOVABLE; | |
887 | ||
888 | if (zone_idx(zone) <= zone_last && !node_state(nid, N_HIGH_MEMORY)) | |
889 | arg->status_change_nid_high = nid; | |
890 | else | |
891 | arg->status_change_nid_high = -1; | |
892 | #else | |
893 | arg->status_change_nid_high = arg->status_change_nid_normal; | |
894 | #endif | |
895 | ||
d9713679 LJ |
896 | /* |
897 | * if the node don't have memory befor online, we will need to | |
6715ddf9 | 898 | * set the node to node_states[N_MEMORY] after the memory |
d9713679 LJ |
899 | * is online. |
900 | */ | |
6715ddf9 | 901 | if (!node_state(nid, N_MEMORY)) |
d9713679 LJ |
902 | arg->status_change_nid = nid; |
903 | else | |
904 | arg->status_change_nid = -1; | |
905 | } | |
906 | ||
907 | static void node_states_set_node(int node, struct memory_notify *arg) | |
908 | { | |
909 | if (arg->status_change_nid_normal >= 0) | |
910 | node_set_state(node, N_NORMAL_MEMORY); | |
911 | ||
6715ddf9 LJ |
912 | if (arg->status_change_nid_high >= 0) |
913 | node_set_state(node, N_HIGH_MEMORY); | |
914 | ||
915 | node_set_state(node, N_MEMORY); | |
d9713679 LJ |
916 | } |
917 | ||
75884fb1 | 918 | |
511c2aba | 919 | int __ref online_pages(unsigned long pfn, unsigned long nr_pages, int online_type) |
75884fb1 | 920 | { |
aa47228a | 921 | unsigned long flags; |
3947be19 DH |
922 | unsigned long onlined_pages = 0; |
923 | struct zone *zone; | |
6811378e | 924 | int need_zonelists_rebuild = 0; |
7b78d335 YG |
925 | int nid; |
926 | int ret; | |
927 | struct memory_notify arg; | |
928 | ||
925268a0 | 929 | lock_memory_hotplug(); |
d9713679 LJ |
930 | /* |
931 | * This doesn't need a lock to do pfn_to_page(). | |
932 | * The section can't be removed here because of the | |
933 | * memory_block->state_mutex. | |
934 | */ | |
935 | zone = page_zone(pfn_to_page(pfn)); | |
936 | ||
74d42d8f LJ |
937 | if ((zone_idx(zone) > ZONE_NORMAL || online_type == ONLINE_MOVABLE) && |
938 | !can_online_high_movable(zone)) { | |
939 | unlock_memory_hotplug(); | |
940 | return -1; | |
941 | } | |
942 | ||
511c2aba LJ |
943 | if (online_type == ONLINE_KERNEL && zone_idx(zone) == ZONE_MOVABLE) { |
944 | if (move_pfn_range_left(zone - 1, zone, pfn, pfn + nr_pages)) { | |
945 | unlock_memory_hotplug(); | |
946 | return -1; | |
947 | } | |
948 | } | |
949 | if (online_type == ONLINE_MOVABLE && zone_idx(zone) == ZONE_MOVABLE - 1) { | |
950 | if (move_pfn_range_right(zone, zone + 1, pfn, pfn + nr_pages)) { | |
951 | unlock_memory_hotplug(); | |
952 | return -1; | |
953 | } | |
954 | } | |
955 | ||
956 | /* Previous code may changed the zone of the pfn range */ | |
957 | zone = page_zone(pfn_to_page(pfn)); | |
958 | ||
7b78d335 YG |
959 | arg.start_pfn = pfn; |
960 | arg.nr_pages = nr_pages; | |
d9713679 | 961 | node_states_check_changes_online(nr_pages, zone, &arg); |
7b78d335 YG |
962 | |
963 | nid = page_to_nid(pfn_to_page(pfn)); | |
3947be19 | 964 | |
7b78d335 YG |
965 | ret = memory_notify(MEM_GOING_ONLINE, &arg); |
966 | ret = notifier_to_errno(ret); | |
967 | if (ret) { | |
968 | memory_notify(MEM_CANCEL_ONLINE, &arg); | |
925268a0 | 969 | unlock_memory_hotplug(); |
7b78d335 YG |
970 | return ret; |
971 | } | |
6811378e YG |
972 | /* |
973 | * If this zone is not populated, then it is not in zonelist. | |
974 | * This means the page allocator ignores this zone. | |
975 | * So, zonelist must be updated after online. | |
976 | */ | |
4eaf3f64 | 977 | mutex_lock(&zonelists_mutex); |
6dcd73d7 | 978 | if (!populated_zone(zone)) { |
6811378e | 979 | need_zonelists_rebuild = 1; |
6dcd73d7 WC |
980 | build_all_zonelists(NULL, zone); |
981 | } | |
6811378e | 982 | |
908eedc6 | 983 | ret = walk_system_ram_range(pfn, nr_pages, &onlined_pages, |
75884fb1 | 984 | online_pages_range); |
fd8a4221 | 985 | if (ret) { |
6dcd73d7 WC |
986 | if (need_zonelists_rebuild) |
987 | zone_pcp_reset(zone); | |
4eaf3f64 | 988 | mutex_unlock(&zonelists_mutex); |
a62e2f4f BH |
989 | printk(KERN_DEBUG "online_pages [mem %#010llx-%#010llx] failed\n", |
990 | (unsigned long long) pfn << PAGE_SHIFT, | |
991 | (((unsigned long long) pfn + nr_pages) | |
992 | << PAGE_SHIFT) - 1); | |
fd8a4221 | 993 | memory_notify(MEM_CANCEL_ONLINE, &arg); |
925268a0 | 994 | unlock_memory_hotplug(); |
fd8a4221 GL |
995 | return ret; |
996 | } | |
997 | ||
9feedc9d | 998 | zone->managed_pages += onlined_pages; |
3947be19 | 999 | zone->present_pages += onlined_pages; |
aa47228a CS |
1000 | |
1001 | pgdat_resize_lock(zone->zone_pgdat, &flags); | |
f2937be5 | 1002 | zone->zone_pgdat->node_present_pages += onlined_pages; |
aa47228a CS |
1003 | pgdat_resize_unlock(zone->zone_pgdat, &flags); |
1004 | ||
08dff7b7 | 1005 | if (onlined_pages) { |
d9713679 | 1006 | node_states_set_node(zone_to_nid(zone), &arg); |
08dff7b7 | 1007 | if (need_zonelists_rebuild) |
6dcd73d7 | 1008 | build_all_zonelists(NULL, NULL); |
08dff7b7 JL |
1009 | else |
1010 | zone_pcp_update(zone); | |
1011 | } | |
3947be19 | 1012 | |
4eaf3f64 | 1013 | mutex_unlock(&zonelists_mutex); |
1b79acc9 KM |
1014 | |
1015 | init_per_zone_wmark_min(); | |
1016 | ||
08dff7b7 | 1017 | if (onlined_pages) |
7ea1530a | 1018 | kswapd_run(zone_to_nid(zone)); |
61b13993 | 1019 | |
1f522509 | 1020 | vm_total_pages = nr_free_pagecache_pages(); |
2f7f24ec | 1021 | |
2d1d43f6 | 1022 | writeback_set_ratelimit(); |
7b78d335 YG |
1023 | |
1024 | if (onlined_pages) | |
1025 | memory_notify(MEM_ONLINE, &arg); | |
925268a0 | 1026 | unlock_memory_hotplug(); |
7b78d335 | 1027 | |
3947be19 DH |
1028 | return 0; |
1029 | } | |
53947027 | 1030 | #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */ |
bc02af93 | 1031 | |
e1319331 HS |
1032 | /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */ |
1033 | static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start) | |
9af3c2de YG |
1034 | { |
1035 | struct pglist_data *pgdat; | |
1036 | unsigned long zones_size[MAX_NR_ZONES] = {0}; | |
1037 | unsigned long zholes_size[MAX_NR_ZONES] = {0}; | |
1038 | unsigned long start_pfn = start >> PAGE_SHIFT; | |
1039 | ||
a1e565aa TC |
1040 | pgdat = NODE_DATA(nid); |
1041 | if (!pgdat) { | |
1042 | pgdat = arch_alloc_nodedata(nid); | |
1043 | if (!pgdat) | |
1044 | return NULL; | |
9af3c2de | 1045 | |
a1e565aa TC |
1046 | arch_refresh_nodedata(nid, pgdat); |
1047 | } | |
9af3c2de YG |
1048 | |
1049 | /* we can use NODE_DATA(nid) from here */ | |
1050 | ||
1051 | /* init node's zones as empty zones, we don't have any present pages.*/ | |
9109fb7b | 1052 | free_area_init_node(nid, zones_size, start_pfn, zholes_size); |
9af3c2de | 1053 | |
959ecc48 KH |
1054 | /* |
1055 | * The node we allocated has no zone fallback lists. For avoiding | |
1056 | * to access not-initialized zonelist, build here. | |
1057 | */ | |
f957db4f | 1058 | mutex_lock(&zonelists_mutex); |
9adb62a5 | 1059 | build_all_zonelists(pgdat, NULL); |
f957db4f | 1060 | mutex_unlock(&zonelists_mutex); |
959ecc48 | 1061 | |
9af3c2de YG |
1062 | return pgdat; |
1063 | } | |
1064 | ||
1065 | static void rollback_node_hotadd(int nid, pg_data_t *pgdat) | |
1066 | { | |
1067 | arch_refresh_nodedata(nid, NULL); | |
1068 | arch_free_nodedata(pgdat); | |
1069 | return; | |
1070 | } | |
1071 | ||
0a547039 | 1072 | |
cf23422b | 1073 | /* |
1074 | * called by cpu_up() to online a node without onlined memory. | |
1075 | */ | |
1076 | int mem_online_node(int nid) | |
1077 | { | |
1078 | pg_data_t *pgdat; | |
1079 | int ret; | |
1080 | ||
20d6c96b | 1081 | lock_memory_hotplug(); |
cf23422b | 1082 | pgdat = hotadd_new_pgdat(nid, 0); |
7553e8f2 | 1083 | if (!pgdat) { |
cf23422b | 1084 | ret = -ENOMEM; |
1085 | goto out; | |
1086 | } | |
1087 | node_set_online(nid); | |
1088 | ret = register_one_node(nid); | |
1089 | BUG_ON(ret); | |
1090 | ||
1091 | out: | |
20d6c96b | 1092 | unlock_memory_hotplug(); |
cf23422b | 1093 | return ret; |
1094 | } | |
1095 | ||
31168481 AV |
1096 | /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */ |
1097 | int __ref add_memory(int nid, u64 start, u64 size) | |
bc02af93 | 1098 | { |
9af3c2de | 1099 | pg_data_t *pgdat = NULL; |
a1e565aa TC |
1100 | bool new_pgdat; |
1101 | bool new_node; | |
ebd15302 | 1102 | struct resource *res; |
bc02af93 YG |
1103 | int ret; |
1104 | ||
20d6c96b | 1105 | lock_memory_hotplug(); |
6ad696d2 | 1106 | |
ebd15302 | 1107 | res = register_memory_resource(start, size); |
6ad696d2 | 1108 | ret = -EEXIST; |
ebd15302 | 1109 | if (!res) |
6ad696d2 | 1110 | goto out; |
ebd15302 | 1111 | |
a1e565aa TC |
1112 | { /* Stupid hack to suppress address-never-null warning */ |
1113 | void *p = NODE_DATA(nid); | |
1114 | new_pgdat = !p; | |
1115 | } | |
1116 | new_node = !node_online(nid); | |
1117 | if (new_node) { | |
9af3c2de | 1118 | pgdat = hotadd_new_pgdat(nid, start); |
6ad696d2 | 1119 | ret = -ENOMEM; |
9af3c2de | 1120 | if (!pgdat) |
41b9e2d7 | 1121 | goto error; |
9af3c2de YG |
1122 | } |
1123 | ||
bc02af93 YG |
1124 | /* call arch's memory hotadd */ |
1125 | ret = arch_add_memory(nid, start, size); | |
1126 | ||
9af3c2de YG |
1127 | if (ret < 0) |
1128 | goto error; | |
1129 | ||
0fc44159 | 1130 | /* we online node here. we can't roll back from here. */ |
9af3c2de YG |
1131 | node_set_online(nid); |
1132 | ||
a1e565aa | 1133 | if (new_node) { |
0fc44159 YG |
1134 | ret = register_one_node(nid); |
1135 | /* | |
1136 | * If sysfs file of new node can't create, cpu on the node | |
1137 | * can't be hot-added. There is no rollback way now. | |
1138 | * So, check by BUG_ON() to catch it reluctantly.. | |
1139 | */ | |
1140 | BUG_ON(ret); | |
1141 | } | |
1142 | ||
d96ae530 | 1143 | /* create new memmap entry */ |
1144 | firmware_map_add_hotplug(start, start + size, "System RAM"); | |
1145 | ||
6ad696d2 AK |
1146 | goto out; |
1147 | ||
9af3c2de YG |
1148 | error: |
1149 | /* rollback pgdat allocation and others */ | |
1150 | if (new_pgdat) | |
1151 | rollback_node_hotadd(nid, pgdat); | |
a864b9d0 | 1152 | release_memory_resource(res); |
9af3c2de | 1153 | |
6ad696d2 | 1154 | out: |
20d6c96b | 1155 | unlock_memory_hotplug(); |
bc02af93 YG |
1156 | return ret; |
1157 | } | |
1158 | EXPORT_SYMBOL_GPL(add_memory); | |
0c0e6195 KH |
1159 | |
1160 | #ifdef CONFIG_MEMORY_HOTREMOVE | |
5c755e9f BP |
1161 | /* |
1162 | * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy | |
1163 | * set and the size of the free page is given by page_order(). Using this, | |
1164 | * the function determines if the pageblock contains only free pages. | |
1165 | * Due to buddy contraints, a free page at least the size of a pageblock will | |
1166 | * be located at the start of the pageblock | |
1167 | */ | |
1168 | static inline int pageblock_free(struct page *page) | |
1169 | { | |
1170 | return PageBuddy(page) && page_order(page) >= pageblock_order; | |
1171 | } | |
1172 | ||
1173 | /* Return the start of the next active pageblock after a given page */ | |
1174 | static struct page *next_active_pageblock(struct page *page) | |
1175 | { | |
5c755e9f BP |
1176 | /* Ensure the starting page is pageblock-aligned */ |
1177 | BUG_ON(page_to_pfn(page) & (pageblock_nr_pages - 1)); | |
1178 | ||
5c755e9f | 1179 | /* If the entire pageblock is free, move to the end of free page */ |
0dcc48c1 KH |
1180 | if (pageblock_free(page)) { |
1181 | int order; | |
1182 | /* be careful. we don't have locks, page_order can be changed.*/ | |
1183 | order = page_order(page); | |
1184 | if ((order < MAX_ORDER) && (order >= pageblock_order)) | |
1185 | return page + (1 << order); | |
1186 | } | |
5c755e9f | 1187 | |
0dcc48c1 | 1188 | return page + pageblock_nr_pages; |
5c755e9f BP |
1189 | } |
1190 | ||
1191 | /* Checks if this range of memory is likely to be hot-removable. */ | |
1192 | int is_mem_section_removable(unsigned long start_pfn, unsigned long nr_pages) | |
1193 | { | |
5c755e9f BP |
1194 | struct page *page = pfn_to_page(start_pfn); |
1195 | struct page *end_page = page + nr_pages; | |
1196 | ||
1197 | /* Check the starting page of each pageblock within the range */ | |
1198 | for (; page < end_page; page = next_active_pageblock(page)) { | |
49ac8255 | 1199 | if (!is_pageblock_removable_nolock(page)) |
5c755e9f | 1200 | return 0; |
49ac8255 | 1201 | cond_resched(); |
5c755e9f BP |
1202 | } |
1203 | ||
1204 | /* All pageblocks in the memory block are likely to be hot-removable */ | |
1205 | return 1; | |
1206 | } | |
1207 | ||
0c0e6195 KH |
1208 | /* |
1209 | * Confirm all pages in a range [start, end) is belongs to the same zone. | |
1210 | */ | |
1211 | static int test_pages_in_a_zone(unsigned long start_pfn, unsigned long end_pfn) | |
1212 | { | |
1213 | unsigned long pfn; | |
1214 | struct zone *zone = NULL; | |
1215 | struct page *page; | |
1216 | int i; | |
1217 | for (pfn = start_pfn; | |
1218 | pfn < end_pfn; | |
1219 | pfn += MAX_ORDER_NR_PAGES) { | |
1220 | i = 0; | |
1221 | /* This is just a CONFIG_HOLES_IN_ZONE check.*/ | |
1222 | while ((i < MAX_ORDER_NR_PAGES) && !pfn_valid_within(pfn + i)) | |
1223 | i++; | |
1224 | if (i == MAX_ORDER_NR_PAGES) | |
1225 | continue; | |
1226 | page = pfn_to_page(pfn + i); | |
1227 | if (zone && page_zone(page) != zone) | |
1228 | return 0; | |
1229 | zone = page_zone(page); | |
1230 | } | |
1231 | return 1; | |
1232 | } | |
1233 | ||
1234 | /* | |
1235 | * Scanning pfn is much easier than scanning lru list. | |
1236 | * Scan pfn from start to end and Find LRU page. | |
1237 | */ | |
7bbc0905 | 1238 | static unsigned long scan_lru_pages(unsigned long start, unsigned long end) |
0c0e6195 KH |
1239 | { |
1240 | unsigned long pfn; | |
1241 | struct page *page; | |
1242 | for (pfn = start; pfn < end; pfn++) { | |
1243 | if (pfn_valid(pfn)) { | |
1244 | page = pfn_to_page(pfn); | |
1245 | if (PageLRU(page)) | |
1246 | return pfn; | |
1247 | } | |
1248 | } | |
1249 | return 0; | |
1250 | } | |
1251 | ||
0c0e6195 KH |
1252 | #define NR_OFFLINE_AT_ONCE_PAGES (256) |
1253 | static int | |
1254 | do_migrate_range(unsigned long start_pfn, unsigned long end_pfn) | |
1255 | { | |
1256 | unsigned long pfn; | |
1257 | struct page *page; | |
1258 | int move_pages = NR_OFFLINE_AT_ONCE_PAGES; | |
1259 | int not_managed = 0; | |
1260 | int ret = 0; | |
1261 | LIST_HEAD(source); | |
1262 | ||
1263 | for (pfn = start_pfn; pfn < end_pfn && move_pages > 0; pfn++) { | |
1264 | if (!pfn_valid(pfn)) | |
1265 | continue; | |
1266 | page = pfn_to_page(pfn); | |
700c2a46 | 1267 | if (!get_page_unless_zero(page)) |
0c0e6195 KH |
1268 | continue; |
1269 | /* | |
1270 | * We can skip free pages. And we can only deal with pages on | |
1271 | * LRU. | |
1272 | */ | |
62695a84 | 1273 | ret = isolate_lru_page(page); |
0c0e6195 | 1274 | if (!ret) { /* Success */ |
700c2a46 | 1275 | put_page(page); |
62695a84 | 1276 | list_add_tail(&page->lru, &source); |
0c0e6195 | 1277 | move_pages--; |
6d9c285a KM |
1278 | inc_zone_page_state(page, NR_ISOLATED_ANON + |
1279 | page_is_file_cache(page)); | |
1280 | ||
0c0e6195 | 1281 | } else { |
0c0e6195 | 1282 | #ifdef CONFIG_DEBUG_VM |
718a3821 WF |
1283 | printk(KERN_ALERT "removing pfn %lx from LRU failed\n", |
1284 | pfn); | |
1285 | dump_page(page); | |
0c0e6195 | 1286 | #endif |
700c2a46 | 1287 | put_page(page); |
25985edc | 1288 | /* Because we don't have big zone->lock. we should |
809c4449 BL |
1289 | check this again here. */ |
1290 | if (page_count(page)) { | |
1291 | not_managed++; | |
f3ab2636 | 1292 | ret = -EBUSY; |
809c4449 BL |
1293 | break; |
1294 | } | |
0c0e6195 KH |
1295 | } |
1296 | } | |
f3ab2636 BL |
1297 | if (!list_empty(&source)) { |
1298 | if (not_managed) { | |
1299 | putback_lru_pages(&source); | |
1300 | goto out; | |
1301 | } | |
74c08f98 MK |
1302 | |
1303 | /* | |
1304 | * alloc_migrate_target should be improooooved!! | |
1305 | * migrate_pages returns # of failed pages. | |
1306 | */ | |
1307 | ret = migrate_pages(&source, alloc_migrate_target, 0, | |
9c620e2b | 1308 | MIGRATE_SYNC, MR_MEMORY_HOTPLUG); |
f3ab2636 | 1309 | if (ret) |
0c0e6195 | 1310 | putback_lru_pages(&source); |
0c0e6195 | 1311 | } |
0c0e6195 KH |
1312 | out: |
1313 | return ret; | |
1314 | } | |
1315 | ||
1316 | /* | |
1317 | * remove from free_area[] and mark all as Reserved. | |
1318 | */ | |
1319 | static int | |
1320 | offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages, | |
1321 | void *data) | |
1322 | { | |
1323 | __offline_isolated_pages(start, start + nr_pages); | |
1324 | return 0; | |
1325 | } | |
1326 | ||
1327 | static void | |
1328 | offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn) | |
1329 | { | |
908eedc6 | 1330 | walk_system_ram_range(start_pfn, end_pfn - start_pfn, NULL, |
0c0e6195 KH |
1331 | offline_isolated_pages_cb); |
1332 | } | |
1333 | ||
1334 | /* | |
1335 | * Check all pages in range, recoreded as memory resource, are isolated. | |
1336 | */ | |
1337 | static int | |
1338 | check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages, | |
1339 | void *data) | |
1340 | { | |
1341 | int ret; | |
1342 | long offlined = *(long *)data; | |
b023f468 | 1343 | ret = test_pages_isolated(start_pfn, start_pfn + nr_pages, true); |
0c0e6195 KH |
1344 | offlined = nr_pages; |
1345 | if (!ret) | |
1346 | *(long *)data += offlined; | |
1347 | return ret; | |
1348 | } | |
1349 | ||
1350 | static long | |
1351 | check_pages_isolated(unsigned long start_pfn, unsigned long end_pfn) | |
1352 | { | |
1353 | long offlined = 0; | |
1354 | int ret; | |
1355 | ||
908eedc6 | 1356 | ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn, &offlined, |
0c0e6195 KH |
1357 | check_pages_isolated_cb); |
1358 | if (ret < 0) | |
1359 | offlined = (long)ret; | |
1360 | return offlined; | |
1361 | } | |
1362 | ||
09285af7 | 1363 | #ifdef CONFIG_MOVABLE_NODE |
79a4dcef TC |
1364 | /* |
1365 | * When CONFIG_MOVABLE_NODE, we permit offlining of a node which doesn't have | |
1366 | * normal memory. | |
1367 | */ | |
09285af7 LJ |
1368 | static bool can_offline_normal(struct zone *zone, unsigned long nr_pages) |
1369 | { | |
1370 | return true; | |
1371 | } | |
79a4dcef | 1372 | #else /* CONFIG_MOVABLE_NODE */ |
74d42d8f LJ |
1373 | /* ensure the node has NORMAL memory if it is still online */ |
1374 | static bool can_offline_normal(struct zone *zone, unsigned long nr_pages) | |
1375 | { | |
1376 | struct pglist_data *pgdat = zone->zone_pgdat; | |
1377 | unsigned long present_pages = 0; | |
1378 | enum zone_type zt; | |
1379 | ||
1380 | for (zt = 0; zt <= ZONE_NORMAL; zt++) | |
1381 | present_pages += pgdat->node_zones[zt].present_pages; | |
1382 | ||
1383 | if (present_pages > nr_pages) | |
1384 | return true; | |
1385 | ||
1386 | present_pages = 0; | |
1387 | for (; zt <= ZONE_MOVABLE; zt++) | |
1388 | present_pages += pgdat->node_zones[zt].present_pages; | |
1389 | ||
1390 | /* | |
1391 | * we can't offline the last normal memory until all | |
1392 | * higher memory is offlined. | |
1393 | */ | |
1394 | return present_pages == 0; | |
1395 | } | |
79a4dcef | 1396 | #endif /* CONFIG_MOVABLE_NODE */ |
74d42d8f | 1397 | |
d9713679 LJ |
1398 | /* check which state of node_states will be changed when offline memory */ |
1399 | static void node_states_check_changes_offline(unsigned long nr_pages, | |
1400 | struct zone *zone, struct memory_notify *arg) | |
1401 | { | |
1402 | struct pglist_data *pgdat = zone->zone_pgdat; | |
1403 | unsigned long present_pages = 0; | |
1404 | enum zone_type zt, zone_last = ZONE_NORMAL; | |
1405 | ||
1406 | /* | |
6715ddf9 LJ |
1407 | * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY] |
1408 | * contains nodes which have zones of 0...ZONE_NORMAL, | |
1409 | * set zone_last to ZONE_NORMAL. | |
d9713679 | 1410 | * |
6715ddf9 LJ |
1411 | * If we don't have HIGHMEM nor movable node, |
1412 | * node_states[N_NORMAL_MEMORY] contains nodes which have zones of | |
1413 | * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE. | |
d9713679 | 1414 | */ |
6715ddf9 | 1415 | if (N_MEMORY == N_NORMAL_MEMORY) |
d9713679 LJ |
1416 | zone_last = ZONE_MOVABLE; |
1417 | ||
1418 | /* | |
1419 | * check whether node_states[N_NORMAL_MEMORY] will be changed. | |
1420 | * If the memory to be offline is in a zone of 0...zone_last, | |
1421 | * and it is the last present memory, 0...zone_last will | |
1422 | * become empty after offline , thus we can determind we will | |
1423 | * need to clear the node from node_states[N_NORMAL_MEMORY]. | |
1424 | */ | |
1425 | for (zt = 0; zt <= zone_last; zt++) | |
1426 | present_pages += pgdat->node_zones[zt].present_pages; | |
1427 | if (zone_idx(zone) <= zone_last && nr_pages >= present_pages) | |
1428 | arg->status_change_nid_normal = zone_to_nid(zone); | |
1429 | else | |
1430 | arg->status_change_nid_normal = -1; | |
1431 | ||
6715ddf9 LJ |
1432 | #ifdef CONFIG_HIGHMEM |
1433 | /* | |
1434 | * If we have movable node, node_states[N_HIGH_MEMORY] | |
1435 | * contains nodes which have zones of 0...ZONE_HIGHMEM, | |
1436 | * set zone_last to ZONE_HIGHMEM. | |
1437 | * | |
1438 | * If we don't have movable node, node_states[N_NORMAL_MEMORY] | |
1439 | * contains nodes which have zones of 0...ZONE_MOVABLE, | |
1440 | * set zone_last to ZONE_MOVABLE. | |
1441 | */ | |
1442 | zone_last = ZONE_HIGHMEM; | |
1443 | if (N_MEMORY == N_HIGH_MEMORY) | |
1444 | zone_last = ZONE_MOVABLE; | |
1445 | ||
1446 | for (; zt <= zone_last; zt++) | |
1447 | present_pages += pgdat->node_zones[zt].present_pages; | |
1448 | if (zone_idx(zone) <= zone_last && nr_pages >= present_pages) | |
1449 | arg->status_change_nid_high = zone_to_nid(zone); | |
1450 | else | |
1451 | arg->status_change_nid_high = -1; | |
1452 | #else | |
1453 | arg->status_change_nid_high = arg->status_change_nid_normal; | |
1454 | #endif | |
1455 | ||
d9713679 LJ |
1456 | /* |
1457 | * node_states[N_HIGH_MEMORY] contains nodes which have 0...ZONE_MOVABLE | |
1458 | */ | |
1459 | zone_last = ZONE_MOVABLE; | |
1460 | ||
1461 | /* | |
1462 | * check whether node_states[N_HIGH_MEMORY] will be changed | |
1463 | * If we try to offline the last present @nr_pages from the node, | |
1464 | * we can determind we will need to clear the node from | |
1465 | * node_states[N_HIGH_MEMORY]. | |
1466 | */ | |
1467 | for (; zt <= zone_last; zt++) | |
1468 | present_pages += pgdat->node_zones[zt].present_pages; | |
1469 | if (nr_pages >= present_pages) | |
1470 | arg->status_change_nid = zone_to_nid(zone); | |
1471 | else | |
1472 | arg->status_change_nid = -1; | |
1473 | } | |
1474 | ||
1475 | static void node_states_clear_node(int node, struct memory_notify *arg) | |
1476 | { | |
1477 | if (arg->status_change_nid_normal >= 0) | |
1478 | node_clear_state(node, N_NORMAL_MEMORY); | |
1479 | ||
6715ddf9 LJ |
1480 | if ((N_MEMORY != N_NORMAL_MEMORY) && |
1481 | (arg->status_change_nid_high >= 0)) | |
d9713679 | 1482 | node_clear_state(node, N_HIGH_MEMORY); |
6715ddf9 LJ |
1483 | |
1484 | if ((N_MEMORY != N_HIGH_MEMORY) && | |
1485 | (arg->status_change_nid >= 0)) | |
1486 | node_clear_state(node, N_MEMORY); | |
d9713679 LJ |
1487 | } |
1488 | ||
a16cee10 | 1489 | static int __ref __offline_pages(unsigned long start_pfn, |
0c0e6195 KH |
1490 | unsigned long end_pfn, unsigned long timeout) |
1491 | { | |
1492 | unsigned long pfn, nr_pages, expire; | |
1493 | long offlined_pages; | |
7b78d335 | 1494 | int ret, drain, retry_max, node; |
d702909f | 1495 | unsigned long flags; |
0c0e6195 | 1496 | struct zone *zone; |
7b78d335 | 1497 | struct memory_notify arg; |
0c0e6195 KH |
1498 | |
1499 | BUG_ON(start_pfn >= end_pfn); | |
1500 | /* at least, alignment against pageblock is necessary */ | |
1501 | if (!IS_ALIGNED(start_pfn, pageblock_nr_pages)) | |
1502 | return -EINVAL; | |
1503 | if (!IS_ALIGNED(end_pfn, pageblock_nr_pages)) | |
1504 | return -EINVAL; | |
1505 | /* This makes hotplug much easier...and readable. | |
1506 | we assume this for now. .*/ | |
1507 | if (!test_pages_in_a_zone(start_pfn, end_pfn)) | |
1508 | return -EINVAL; | |
7b78d335 | 1509 | |
20d6c96b | 1510 | lock_memory_hotplug(); |
6ad696d2 | 1511 | |
7b78d335 YG |
1512 | zone = page_zone(pfn_to_page(start_pfn)); |
1513 | node = zone_to_nid(zone); | |
1514 | nr_pages = end_pfn - start_pfn; | |
1515 | ||
74d42d8f LJ |
1516 | ret = -EINVAL; |
1517 | if (zone_idx(zone) <= ZONE_NORMAL && !can_offline_normal(zone, nr_pages)) | |
1518 | goto out; | |
1519 | ||
0c0e6195 | 1520 | /* set above range as isolated */ |
b023f468 WC |
1521 | ret = start_isolate_page_range(start_pfn, end_pfn, |
1522 | MIGRATE_MOVABLE, true); | |
0c0e6195 | 1523 | if (ret) |
6ad696d2 | 1524 | goto out; |
7b78d335 YG |
1525 | |
1526 | arg.start_pfn = start_pfn; | |
1527 | arg.nr_pages = nr_pages; | |
d9713679 | 1528 | node_states_check_changes_offline(nr_pages, zone, &arg); |
7b78d335 YG |
1529 | |
1530 | ret = memory_notify(MEM_GOING_OFFLINE, &arg); | |
1531 | ret = notifier_to_errno(ret); | |
1532 | if (ret) | |
1533 | goto failed_removal; | |
1534 | ||
0c0e6195 KH |
1535 | pfn = start_pfn; |
1536 | expire = jiffies + timeout; | |
1537 | drain = 0; | |
1538 | retry_max = 5; | |
1539 | repeat: | |
1540 | /* start memory hot removal */ | |
1541 | ret = -EAGAIN; | |
1542 | if (time_after(jiffies, expire)) | |
1543 | goto failed_removal; | |
1544 | ret = -EINTR; | |
1545 | if (signal_pending(current)) | |
1546 | goto failed_removal; | |
1547 | ret = 0; | |
1548 | if (drain) { | |
1549 | lru_add_drain_all(); | |
0c0e6195 | 1550 | cond_resched(); |
9f8f2172 | 1551 | drain_all_pages(); |
0c0e6195 KH |
1552 | } |
1553 | ||
1554 | pfn = scan_lru_pages(start_pfn, end_pfn); | |
1555 | if (pfn) { /* We have page on LRU */ | |
1556 | ret = do_migrate_range(pfn, end_pfn); | |
1557 | if (!ret) { | |
1558 | drain = 1; | |
1559 | goto repeat; | |
1560 | } else { | |
1561 | if (ret < 0) | |
1562 | if (--retry_max == 0) | |
1563 | goto failed_removal; | |
1564 | yield(); | |
1565 | drain = 1; | |
1566 | goto repeat; | |
1567 | } | |
1568 | } | |
b3834be5 | 1569 | /* drain all zone's lru pagevec, this is asynchronous... */ |
0c0e6195 | 1570 | lru_add_drain_all(); |
0c0e6195 | 1571 | yield(); |
b3834be5 | 1572 | /* drain pcp pages, this is synchronous. */ |
9f8f2172 | 1573 | drain_all_pages(); |
0c0e6195 KH |
1574 | /* check again */ |
1575 | offlined_pages = check_pages_isolated(start_pfn, end_pfn); | |
1576 | if (offlined_pages < 0) { | |
1577 | ret = -EBUSY; | |
1578 | goto failed_removal; | |
1579 | } | |
1580 | printk(KERN_INFO "Offlined Pages %ld\n", offlined_pages); | |
b3834be5 | 1581 | /* Ok, all of our target is isolated. |
0c0e6195 KH |
1582 | We cannot do rollback at this point. */ |
1583 | offline_isolated_pages(start_pfn, end_pfn); | |
dbc0e4ce | 1584 | /* reset pagetype flags and makes migrate type to be MOVABLE */ |
0815f3d8 | 1585 | undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE); |
0c0e6195 | 1586 | /* removal success */ |
9feedc9d | 1587 | zone->managed_pages -= offlined_pages; |
0c0e6195 | 1588 | zone->present_pages -= offlined_pages; |
d702909f CS |
1589 | |
1590 | pgdat_resize_lock(zone->zone_pgdat, &flags); | |
0c0e6195 | 1591 | zone->zone_pgdat->node_present_pages -= offlined_pages; |
d702909f CS |
1592 | pgdat_resize_unlock(zone->zone_pgdat, &flags); |
1593 | ||
0c0e6195 | 1594 | totalram_pages -= offlined_pages; |
7b78d335 | 1595 | |
1b79acc9 KM |
1596 | init_per_zone_wmark_min(); |
1597 | ||
1e8537ba | 1598 | if (!populated_zone(zone)) { |
340175b7 | 1599 | zone_pcp_reset(zone); |
1e8537ba XQ |
1600 | mutex_lock(&zonelists_mutex); |
1601 | build_all_zonelists(NULL, NULL); | |
1602 | mutex_unlock(&zonelists_mutex); | |
1603 | } else | |
1604 | zone_pcp_update(zone); | |
340175b7 | 1605 | |
d9713679 LJ |
1606 | node_states_clear_node(node, &arg); |
1607 | if (arg.status_change_nid >= 0) | |
8fe23e05 | 1608 | kswapd_stop(node); |
bce7394a | 1609 | |
0c0e6195 KH |
1610 | vm_total_pages = nr_free_pagecache_pages(); |
1611 | writeback_set_ratelimit(); | |
7b78d335 YG |
1612 | |
1613 | memory_notify(MEM_OFFLINE, &arg); | |
20d6c96b | 1614 | unlock_memory_hotplug(); |
0c0e6195 KH |
1615 | return 0; |
1616 | ||
1617 | failed_removal: | |
a62e2f4f BH |
1618 | printk(KERN_INFO "memory offlining [mem %#010llx-%#010llx] failed\n", |
1619 | (unsigned long long) start_pfn << PAGE_SHIFT, | |
1620 | ((unsigned long long) end_pfn << PAGE_SHIFT) - 1); | |
7b78d335 | 1621 | memory_notify(MEM_CANCEL_OFFLINE, &arg); |
0c0e6195 | 1622 | /* pushback to free area */ |
0815f3d8 | 1623 | undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE); |
7b78d335 | 1624 | |
6ad696d2 | 1625 | out: |
20d6c96b | 1626 | unlock_memory_hotplug(); |
0c0e6195 KH |
1627 | return ret; |
1628 | } | |
71088785 | 1629 | |
a16cee10 WC |
1630 | int offline_pages(unsigned long start_pfn, unsigned long nr_pages) |
1631 | { | |
1632 | return __offline_pages(start_pfn, start_pfn + nr_pages, 120 * HZ); | |
1633 | } | |
1634 | ||
bbc76be6 WC |
1635 | /** |
1636 | * walk_memory_range - walks through all mem sections in [start_pfn, end_pfn) | |
1637 | * @start_pfn: start pfn of the memory range | |
e05c4bbf | 1638 | * @end_pfn: end pfn of the memory range |
bbc76be6 WC |
1639 | * @arg: argument passed to func |
1640 | * @func: callback for each memory section walked | |
1641 | * | |
1642 | * This function walks through all present mem sections in range | |
1643 | * [start_pfn, end_pfn) and call func on each mem section. | |
1644 | * | |
1645 | * Returns the return value of func. | |
1646 | */ | |
1647 | static int walk_memory_range(unsigned long start_pfn, unsigned long end_pfn, | |
1648 | void *arg, int (*func)(struct memory_block *, void *)) | |
71088785 | 1649 | { |
e90bdb7f WC |
1650 | struct memory_block *mem = NULL; |
1651 | struct mem_section *section; | |
e90bdb7f WC |
1652 | unsigned long pfn, section_nr; |
1653 | int ret; | |
e90bdb7f WC |
1654 | |
1655 | for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) { | |
1656 | section_nr = pfn_to_section_nr(pfn); | |
1657 | if (!present_section_nr(section_nr)) | |
1658 | continue; | |
1659 | ||
1660 | section = __nr_to_section(section_nr); | |
1661 | /* same memblock? */ | |
1662 | if (mem) | |
1663 | if ((section_nr >= mem->start_section_nr) && | |
1664 | (section_nr <= mem->end_section_nr)) | |
1665 | continue; | |
1666 | ||
1667 | mem = find_memory_block_hinted(section, mem); | |
1668 | if (!mem) | |
1669 | continue; | |
1670 | ||
bbc76be6 | 1671 | ret = func(mem, arg); |
e90bdb7f | 1672 | if (ret) { |
bbc76be6 WC |
1673 | kobject_put(&mem->dev.kobj); |
1674 | return ret; | |
e90bdb7f WC |
1675 | } |
1676 | } | |
1677 | ||
1678 | if (mem) | |
1679 | kobject_put(&mem->dev.kobj); | |
1680 | ||
bbc76be6 WC |
1681 | return 0; |
1682 | } | |
1683 | ||
1684 | /** | |
1685 | * offline_memory_block_cb - callback function for offlining memory block | |
1686 | * @mem: the memory block to be offlined | |
1687 | * @arg: buffer to hold error msg | |
1688 | * | |
1689 | * Always return 0, and put the error msg in arg if any. | |
1690 | */ | |
1691 | static int offline_memory_block_cb(struct memory_block *mem, void *arg) | |
1692 | { | |
1693 | int *ret = arg; | |
1694 | int error = offline_memory_block(mem); | |
1695 | ||
1696 | if (error != 0 && *ret == 0) | |
1697 | *ret = error; | |
1698 | ||
1699 | return 0; | |
1700 | } | |
1701 | ||
1702 | static int is_memblock_offlined_cb(struct memory_block *mem, void *arg) | |
1703 | { | |
1704 | int ret = !is_memblock_offlined(mem); | |
1705 | ||
349daa0f RD |
1706 | if (unlikely(ret)) { |
1707 | phys_addr_t beginpa, endpa; | |
1708 | ||
1709 | beginpa = PFN_PHYS(section_nr_to_pfn(mem->start_section_nr)); | |
1710 | endpa = PFN_PHYS(section_nr_to_pfn(mem->end_section_nr + 1))-1; | |
bbc76be6 | 1711 | pr_warn("removing memory fails, because memory " |
349daa0f RD |
1712 | "[%pa-%pa] is onlined\n", |
1713 | &beginpa, &endpa); | |
1714 | } | |
bbc76be6 WC |
1715 | |
1716 | return ret; | |
1717 | } | |
1718 | ||
60a5a19e TC |
1719 | static int check_cpu_on_node(void *data) |
1720 | { | |
1721 | struct pglist_data *pgdat = data; | |
1722 | int cpu; | |
1723 | ||
1724 | for_each_present_cpu(cpu) { | |
1725 | if (cpu_to_node(cpu) == pgdat->node_id) | |
1726 | /* | |
1727 | * the cpu on this node isn't removed, and we can't | |
1728 | * offline this node. | |
1729 | */ | |
1730 | return -EBUSY; | |
1731 | } | |
1732 | ||
1733 | return 0; | |
1734 | } | |
1735 | ||
e13fe869 WC |
1736 | static void unmap_cpu_on_node(void *data) |
1737 | { | |
1738 | #ifdef CONFIG_ACPI_NUMA | |
1739 | struct pglist_data *pgdat = data; | |
1740 | int cpu; | |
1741 | ||
1742 | for_each_possible_cpu(cpu) | |
1743 | if (cpu_to_node(cpu) == pgdat->node_id) | |
1744 | numa_clear_node(cpu); | |
1745 | #endif | |
1746 | } | |
1747 | ||
1748 | static int check_and_unmap_cpu_on_node(void *data) | |
1749 | { | |
1750 | int ret = check_cpu_on_node(data); | |
1751 | ||
1752 | if (ret) | |
1753 | return ret; | |
1754 | ||
1755 | /* | |
1756 | * the node will be offlined when we come here, so we can clear | |
1757 | * the cpu_to_node() now. | |
1758 | */ | |
1759 | ||
1760 | unmap_cpu_on_node(data); | |
1761 | return 0; | |
1762 | } | |
1763 | ||
60a5a19e | 1764 | /* offline the node if all memory sections of this node are removed */ |
90b30cdc | 1765 | void try_offline_node(int nid) |
60a5a19e | 1766 | { |
d822b86a WC |
1767 | pg_data_t *pgdat = NODE_DATA(nid); |
1768 | unsigned long start_pfn = pgdat->node_start_pfn; | |
1769 | unsigned long end_pfn = start_pfn + pgdat->node_spanned_pages; | |
60a5a19e | 1770 | unsigned long pfn; |
d822b86a WC |
1771 | struct page *pgdat_page = virt_to_page(pgdat); |
1772 | int i; | |
60a5a19e TC |
1773 | |
1774 | for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) { | |
1775 | unsigned long section_nr = pfn_to_section_nr(pfn); | |
1776 | ||
1777 | if (!present_section_nr(section_nr)) | |
1778 | continue; | |
1779 | ||
1780 | if (pfn_to_nid(pfn) != nid) | |
1781 | continue; | |
1782 | ||
1783 | /* | |
1784 | * some memory sections of this node are not removed, and we | |
1785 | * can't offline node now. | |
1786 | */ | |
1787 | return; | |
1788 | } | |
1789 | ||
e13fe869 | 1790 | if (stop_machine(check_and_unmap_cpu_on_node, pgdat, NULL)) |
60a5a19e TC |
1791 | return; |
1792 | ||
1793 | /* | |
1794 | * all memory/cpu of this node are removed, we can offline this | |
1795 | * node now. | |
1796 | */ | |
1797 | node_set_offline(nid); | |
1798 | unregister_one_node(nid); | |
d822b86a WC |
1799 | |
1800 | if (!PageSlab(pgdat_page) && !PageCompound(pgdat_page)) | |
1801 | /* node data is allocated from boot memory */ | |
1802 | return; | |
1803 | ||
1804 | /* free waittable in each zone */ | |
1805 | for (i = 0; i < MAX_NR_ZONES; i++) { | |
1806 | struct zone *zone = pgdat->node_zones + i; | |
1807 | ||
ca4b3f30 JW |
1808 | /* |
1809 | * wait_table may be allocated from boot memory, | |
1810 | * here only free if it's allocated by vmalloc. | |
1811 | */ | |
1812 | if (is_vmalloc_addr(zone->wait_table)) | |
d822b86a WC |
1813 | vfree(zone->wait_table); |
1814 | } | |
1815 | ||
1816 | /* | |
1817 | * Since there is no way to guarentee the address of pgdat/zone is not | |
1818 | * on stack of any kernel threads or used by other kernel objects | |
1819 | * without reference counting or other symchronizing method, do not | |
1820 | * reset node_data and free pgdat here. Just reset it to 0 and reuse | |
1821 | * the memory when the node is online again. | |
1822 | */ | |
1823 | memset(pgdat, 0, sizeof(*pgdat)); | |
60a5a19e | 1824 | } |
90b30cdc | 1825 | EXPORT_SYMBOL(try_offline_node); |
60a5a19e TC |
1826 | |
1827 | int __ref remove_memory(int nid, u64 start, u64 size) | |
bbc76be6 WC |
1828 | { |
1829 | unsigned long start_pfn, end_pfn; | |
1830 | int ret = 0; | |
1831 | int retry = 1; | |
1832 | ||
1833 | start_pfn = PFN_DOWN(start); | |
f8749452 | 1834 | end_pfn = PFN_UP(start + size - 1); |
bbc76be6 WC |
1835 | |
1836 | /* | |
1837 | * When CONFIG_MEMCG is on, one memory block may be used by other | |
1838 | * blocks to store page cgroup when onlining pages. But we don't know | |
1839 | * in what order pages are onlined. So we iterate twice to offline | |
1840 | * memory: | |
1841 | * 1st iterate: offline every non primary memory block. | |
1842 | * 2nd iterate: offline primary (i.e. first added) memory block. | |
1843 | */ | |
1844 | repeat: | |
1845 | walk_memory_range(start_pfn, end_pfn, &ret, | |
1846 | offline_memory_block_cb); | |
1847 | if (ret) { | |
1848 | if (!retry) | |
1849 | return ret; | |
1850 | ||
1851 | retry = 0; | |
1852 | ret = 0; | |
993c1aad WC |
1853 | goto repeat; |
1854 | } | |
1855 | ||
6677e3ea YI |
1856 | lock_memory_hotplug(); |
1857 | ||
1858 | /* | |
1859 | * we have offlined all memory blocks like this: | |
1860 | * 1. lock memory hotplug | |
1861 | * 2. offline a memory block | |
1862 | * 3. unlock memory hotplug | |
1863 | * | |
1864 | * repeat step1-3 to offline the memory block. All memory blocks | |
1865 | * must be offlined before removing memory. But we don't hold the | |
1866 | * lock in the whole operation. So we should check whether all | |
1867 | * memory blocks are offlined. | |
1868 | */ | |
1869 | ||
bbc76be6 WC |
1870 | ret = walk_memory_range(start_pfn, end_pfn, NULL, |
1871 | is_memblock_offlined_cb); | |
1872 | if (ret) { | |
1873 | unlock_memory_hotplug(); | |
1874 | return ret; | |
6677e3ea YI |
1875 | } |
1876 | ||
46c66c4b YI |
1877 | /* remove memmap entry */ |
1878 | firmware_map_remove(start, start + size, "System RAM"); | |
1879 | ||
24d335ca WC |
1880 | arch_remove_memory(start, size); |
1881 | ||
60a5a19e TC |
1882 | try_offline_node(nid); |
1883 | ||
6677e3ea YI |
1884 | unlock_memory_hotplug(); |
1885 | ||
e90bdb7f | 1886 | return 0; |
71088785 | 1887 | } |
48e94196 | 1888 | #else |
a16cee10 WC |
1889 | int offline_pages(unsigned long start_pfn, unsigned long nr_pages) |
1890 | { | |
1891 | return -EINVAL; | |
1892 | } | |
60a5a19e | 1893 | int remove_memory(int nid, u64 start, u64 size) |
48e94196 KH |
1894 | { |
1895 | return -EINVAL; | |
1896 | } | |
0c0e6195 | 1897 | #endif /* CONFIG_MEMORY_HOTREMOVE */ |
71088785 | 1898 | EXPORT_SYMBOL_GPL(remove_memory); |