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> |
c8721bbb | 33 | #include <linux/hugetlb.h> |
3947be19 DH |
34 | |
35 | #include <asm/tlbflush.h> | |
36 | ||
1e5ad9a3 AB |
37 | #include "internal.h" |
38 | ||
9d0ad8ca DK |
39 | /* |
40 | * online_page_callback contains pointer to current page onlining function. | |
41 | * Initially it is generic_online_page(). If it is required it could be | |
42 | * changed by calling set_online_page_callback() for callback registration | |
43 | * and restore_online_page_callback() for generic callback restore. | |
44 | */ | |
45 | ||
46 | static void generic_online_page(struct page *page); | |
47 | ||
48 | static online_page_callback_t online_page_callback = generic_online_page; | |
49 | ||
20d6c96b KM |
50 | DEFINE_MUTEX(mem_hotplug_mutex); |
51 | ||
52 | void lock_memory_hotplug(void) | |
53 | { | |
54 | mutex_lock(&mem_hotplug_mutex); | |
55 | ||
56 | /* for exclusive hibernation if CONFIG_HIBERNATION=y */ | |
57 | lock_system_sleep(); | |
58 | } | |
59 | ||
60 | void unlock_memory_hotplug(void) | |
61 | { | |
62 | unlock_system_sleep(); | |
63 | mutex_unlock(&mem_hotplug_mutex); | |
64 | } | |
65 | ||
66 | ||
45e0b78b KM |
67 | /* add this memory to iomem resource */ |
68 | static struct resource *register_memory_resource(u64 start, u64 size) | |
69 | { | |
70 | struct resource *res; | |
71 | res = kzalloc(sizeof(struct resource), GFP_KERNEL); | |
72 | BUG_ON(!res); | |
73 | ||
74 | res->name = "System RAM"; | |
75 | res->start = start; | |
76 | res->end = start + size - 1; | |
887c3cb1 | 77 | res->flags = IORESOURCE_MEM | IORESOURCE_BUSY; |
45e0b78b | 78 | if (request_resource(&iomem_resource, res) < 0) { |
4996eed8 | 79 | pr_debug("System RAM resource %pR cannot be added\n", res); |
45e0b78b KM |
80 | kfree(res); |
81 | res = NULL; | |
82 | } | |
83 | return res; | |
84 | } | |
85 | ||
86 | static void release_memory_resource(struct resource *res) | |
87 | { | |
88 | if (!res) | |
89 | return; | |
90 | release_resource(res); | |
91 | kfree(res); | |
92 | return; | |
93 | } | |
94 | ||
53947027 | 95 | #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE |
46723bfa YI |
96 | void get_page_bootmem(unsigned long info, struct page *page, |
97 | unsigned long type) | |
04753278 | 98 | { |
5f24ce5f | 99 | page->lru.next = (struct list_head *) type; |
04753278 YG |
100 | SetPagePrivate(page); |
101 | set_page_private(page, info); | |
102 | atomic_inc(&page->_count); | |
103 | } | |
104 | ||
170a5a7e | 105 | void put_page_bootmem(struct page *page) |
04753278 | 106 | { |
5f24ce5f | 107 | unsigned long type; |
04753278 | 108 | |
5f24ce5f AA |
109 | type = (unsigned long) page->lru.next; |
110 | BUG_ON(type < MEMORY_HOTPLUG_MIN_BOOTMEM_TYPE || | |
111 | type > MEMORY_HOTPLUG_MAX_BOOTMEM_TYPE); | |
04753278 YG |
112 | |
113 | if (atomic_dec_return(&page->_count) == 1) { | |
114 | ClearPagePrivate(page); | |
115 | set_page_private(page, 0); | |
5f24ce5f | 116 | INIT_LIST_HEAD(&page->lru); |
170a5a7e | 117 | free_reserved_page(page); |
04753278 | 118 | } |
04753278 YG |
119 | } |
120 | ||
46723bfa YI |
121 | #ifdef CONFIG_HAVE_BOOTMEM_INFO_NODE |
122 | #ifndef CONFIG_SPARSEMEM_VMEMMAP | |
d92bc318 | 123 | static void register_page_bootmem_info_section(unsigned long start_pfn) |
04753278 YG |
124 | { |
125 | unsigned long *usemap, mapsize, section_nr, i; | |
126 | struct mem_section *ms; | |
127 | struct page *page, *memmap; | |
128 | ||
04753278 YG |
129 | section_nr = pfn_to_section_nr(start_pfn); |
130 | ms = __nr_to_section(section_nr); | |
131 | ||
132 | /* Get section's memmap address */ | |
133 | memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr); | |
134 | ||
135 | /* | |
136 | * Get page for the memmap's phys address | |
137 | * XXX: need more consideration for sparse_vmemmap... | |
138 | */ | |
139 | page = virt_to_page(memmap); | |
140 | mapsize = sizeof(struct page) * PAGES_PER_SECTION; | |
141 | mapsize = PAGE_ALIGN(mapsize) >> PAGE_SHIFT; | |
142 | ||
143 | /* remember memmap's page */ | |
144 | for (i = 0; i < mapsize; i++, page++) | |
145 | get_page_bootmem(section_nr, page, SECTION_INFO); | |
146 | ||
147 | usemap = __nr_to_section(section_nr)->pageblock_flags; | |
148 | page = virt_to_page(usemap); | |
149 | ||
150 | mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT; | |
151 | ||
152 | for (i = 0; i < mapsize; i++, page++) | |
af370fb8 | 153 | get_page_bootmem(section_nr, page, MIX_SECTION_INFO); |
04753278 YG |
154 | |
155 | } | |
46723bfa YI |
156 | #else /* CONFIG_SPARSEMEM_VMEMMAP */ |
157 | static void register_page_bootmem_info_section(unsigned long start_pfn) | |
158 | { | |
159 | unsigned long *usemap, mapsize, section_nr, i; | |
160 | struct mem_section *ms; | |
161 | struct page *page, *memmap; | |
162 | ||
163 | if (!pfn_valid(start_pfn)) | |
164 | return; | |
165 | ||
166 | section_nr = pfn_to_section_nr(start_pfn); | |
167 | ms = __nr_to_section(section_nr); | |
168 | ||
169 | memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr); | |
170 | ||
171 | register_page_bootmem_memmap(section_nr, memmap, PAGES_PER_SECTION); | |
172 | ||
173 | usemap = __nr_to_section(section_nr)->pageblock_flags; | |
174 | page = virt_to_page(usemap); | |
175 | ||
176 | mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT; | |
177 | ||
178 | for (i = 0; i < mapsize; i++, page++) | |
179 | get_page_bootmem(section_nr, page, MIX_SECTION_INFO); | |
180 | } | |
181 | #endif /* !CONFIG_SPARSEMEM_VMEMMAP */ | |
04753278 YG |
182 | |
183 | void register_page_bootmem_info_node(struct pglist_data *pgdat) | |
184 | { | |
185 | unsigned long i, pfn, end_pfn, nr_pages; | |
186 | int node = pgdat->node_id; | |
187 | struct page *page; | |
188 | struct zone *zone; | |
189 | ||
190 | nr_pages = PAGE_ALIGN(sizeof(struct pglist_data)) >> PAGE_SHIFT; | |
191 | page = virt_to_page(pgdat); | |
192 | ||
193 | for (i = 0; i < nr_pages; i++, page++) | |
194 | get_page_bootmem(node, page, NODE_INFO); | |
195 | ||
196 | zone = &pgdat->node_zones[0]; | |
197 | for (; zone < pgdat->node_zones + MAX_NR_ZONES - 1; zone++) { | |
139c2d75 | 198 | if (zone_is_initialized(zone)) { |
04753278 YG |
199 | nr_pages = zone->wait_table_hash_nr_entries |
200 | * sizeof(wait_queue_head_t); | |
201 | nr_pages = PAGE_ALIGN(nr_pages) >> PAGE_SHIFT; | |
202 | page = virt_to_page(zone->wait_table); | |
203 | ||
204 | for (i = 0; i < nr_pages; i++, page++) | |
205 | get_page_bootmem(node, page, NODE_INFO); | |
206 | } | |
207 | } | |
208 | ||
209 | pfn = pgdat->node_start_pfn; | |
c1f19495 | 210 | end_pfn = pgdat_end_pfn(pgdat); |
04753278 | 211 | |
7e9f5eb0 | 212 | /* register section info */ |
f14851af | 213 | for (; pfn < end_pfn; pfn += PAGES_PER_SECTION) { |
214 | /* | |
215 | * Some platforms can assign the same pfn to multiple nodes - on | |
216 | * node0 as well as nodeN. To avoid registering a pfn against | |
217 | * multiple nodes we check that this pfn does not already | |
7e9f5eb0 | 218 | * reside in some other nodes. |
f14851af | 219 | */ |
220 | if (pfn_valid(pfn) && (pfn_to_nid(pfn) == node)) | |
221 | register_page_bootmem_info_section(pfn); | |
222 | } | |
04753278 | 223 | } |
46723bfa | 224 | #endif /* CONFIG_HAVE_BOOTMEM_INFO_NODE */ |
04753278 | 225 | |
76cdd58e HC |
226 | static void grow_zone_span(struct zone *zone, unsigned long start_pfn, |
227 | unsigned long end_pfn) | |
228 | { | |
229 | unsigned long old_zone_end_pfn; | |
230 | ||
231 | zone_span_writelock(zone); | |
232 | ||
c33bc315 | 233 | old_zone_end_pfn = zone_end_pfn(zone); |
8080fc03 | 234 | if (zone_is_empty(zone) || start_pfn < zone->zone_start_pfn) |
76cdd58e HC |
235 | zone->zone_start_pfn = start_pfn; |
236 | ||
237 | zone->spanned_pages = max(old_zone_end_pfn, end_pfn) - | |
238 | zone->zone_start_pfn; | |
239 | ||
240 | zone_span_writeunlock(zone); | |
241 | } | |
242 | ||
511c2aba LJ |
243 | static void resize_zone(struct zone *zone, unsigned long start_pfn, |
244 | unsigned long end_pfn) | |
245 | { | |
246 | zone_span_writelock(zone); | |
247 | ||
e455a9b9 LJ |
248 | if (end_pfn - start_pfn) { |
249 | zone->zone_start_pfn = start_pfn; | |
250 | zone->spanned_pages = end_pfn - start_pfn; | |
251 | } else { | |
252 | /* | |
253 | * make it consist as free_area_init_core(), | |
254 | * if spanned_pages = 0, then keep start_pfn = 0 | |
255 | */ | |
256 | zone->zone_start_pfn = 0; | |
257 | zone->spanned_pages = 0; | |
258 | } | |
511c2aba LJ |
259 | |
260 | zone_span_writeunlock(zone); | |
261 | } | |
262 | ||
263 | static void fix_zone_id(struct zone *zone, unsigned long start_pfn, | |
264 | unsigned long end_pfn) | |
265 | { | |
266 | enum zone_type zid = zone_idx(zone); | |
267 | int nid = zone->zone_pgdat->node_id; | |
268 | unsigned long pfn; | |
269 | ||
270 | for (pfn = start_pfn; pfn < end_pfn; pfn++) | |
271 | set_page_links(pfn_to_page(pfn), zid, nid, pfn); | |
272 | } | |
273 | ||
f6bbb78e CS |
274 | /* Can fail with -ENOMEM from allocating a wait table with vmalloc() or |
275 | * alloc_bootmem_node_nopanic() */ | |
276 | static int __ref ensure_zone_is_initialized(struct zone *zone, | |
277 | unsigned long start_pfn, unsigned long num_pages) | |
278 | { | |
279 | if (!zone_is_initialized(zone)) | |
280 | return init_currently_empty_zone(zone, start_pfn, num_pages, | |
281 | MEMMAP_HOTPLUG); | |
282 | return 0; | |
283 | } | |
284 | ||
e455a9b9 | 285 | static int __meminit move_pfn_range_left(struct zone *z1, struct zone *z2, |
511c2aba LJ |
286 | unsigned long start_pfn, unsigned long end_pfn) |
287 | { | |
e455a9b9 | 288 | int ret; |
511c2aba | 289 | unsigned long flags; |
e455a9b9 LJ |
290 | unsigned long z1_start_pfn; |
291 | ||
64dd1b29 CS |
292 | ret = ensure_zone_is_initialized(z1, start_pfn, end_pfn - start_pfn); |
293 | if (ret) | |
294 | return ret; | |
511c2aba LJ |
295 | |
296 | pgdat_resize_lock(z1->zone_pgdat, &flags); | |
297 | ||
298 | /* can't move pfns which are higher than @z2 */ | |
108bcc96 | 299 | if (end_pfn > zone_end_pfn(z2)) |
511c2aba | 300 | goto out_fail; |
834405c3 | 301 | /* the move out part must be at the left most of @z2 */ |
511c2aba LJ |
302 | if (start_pfn > z2->zone_start_pfn) |
303 | goto out_fail; | |
304 | /* must included/overlap */ | |
305 | if (end_pfn <= z2->zone_start_pfn) | |
306 | goto out_fail; | |
307 | ||
e455a9b9 | 308 | /* use start_pfn for z1's start_pfn if z1 is empty */ |
8080fc03 | 309 | if (!zone_is_empty(z1)) |
e455a9b9 LJ |
310 | z1_start_pfn = z1->zone_start_pfn; |
311 | else | |
312 | z1_start_pfn = start_pfn; | |
313 | ||
314 | resize_zone(z1, z1_start_pfn, end_pfn); | |
108bcc96 | 315 | resize_zone(z2, end_pfn, zone_end_pfn(z2)); |
511c2aba LJ |
316 | |
317 | pgdat_resize_unlock(z1->zone_pgdat, &flags); | |
318 | ||
319 | fix_zone_id(z1, start_pfn, end_pfn); | |
320 | ||
321 | return 0; | |
322 | out_fail: | |
323 | pgdat_resize_unlock(z1->zone_pgdat, &flags); | |
324 | return -1; | |
325 | } | |
326 | ||
e455a9b9 | 327 | static int __meminit move_pfn_range_right(struct zone *z1, struct zone *z2, |
511c2aba LJ |
328 | unsigned long start_pfn, unsigned long end_pfn) |
329 | { | |
e455a9b9 | 330 | int ret; |
511c2aba | 331 | unsigned long flags; |
e455a9b9 LJ |
332 | unsigned long z2_end_pfn; |
333 | ||
64dd1b29 CS |
334 | ret = ensure_zone_is_initialized(z2, start_pfn, end_pfn - start_pfn); |
335 | if (ret) | |
336 | return ret; | |
511c2aba LJ |
337 | |
338 | pgdat_resize_lock(z1->zone_pgdat, &flags); | |
339 | ||
340 | /* can't move pfns which are lower than @z1 */ | |
341 | if (z1->zone_start_pfn > start_pfn) | |
342 | goto out_fail; | |
343 | /* the move out part mast at the right most of @z1 */ | |
108bcc96 | 344 | if (zone_end_pfn(z1) > end_pfn) |
511c2aba LJ |
345 | goto out_fail; |
346 | /* must included/overlap */ | |
108bcc96 | 347 | if (start_pfn >= zone_end_pfn(z1)) |
511c2aba LJ |
348 | goto out_fail; |
349 | ||
e455a9b9 | 350 | /* use end_pfn for z2's end_pfn if z2 is empty */ |
8080fc03 | 351 | if (!zone_is_empty(z2)) |
108bcc96 | 352 | z2_end_pfn = zone_end_pfn(z2); |
e455a9b9 LJ |
353 | else |
354 | z2_end_pfn = end_pfn; | |
355 | ||
511c2aba | 356 | resize_zone(z1, z1->zone_start_pfn, start_pfn); |
e455a9b9 | 357 | resize_zone(z2, start_pfn, z2_end_pfn); |
511c2aba LJ |
358 | |
359 | pgdat_resize_unlock(z1->zone_pgdat, &flags); | |
360 | ||
361 | fix_zone_id(z2, start_pfn, end_pfn); | |
362 | ||
363 | return 0; | |
364 | out_fail: | |
365 | pgdat_resize_unlock(z1->zone_pgdat, &flags); | |
366 | return -1; | |
367 | } | |
368 | ||
76cdd58e HC |
369 | static void grow_pgdat_span(struct pglist_data *pgdat, unsigned long start_pfn, |
370 | unsigned long end_pfn) | |
371 | { | |
372 | unsigned long old_pgdat_end_pfn = | |
373 | pgdat->node_start_pfn + pgdat->node_spanned_pages; | |
374 | ||
712cd386 | 375 | if (!pgdat->node_spanned_pages || start_pfn < pgdat->node_start_pfn) |
76cdd58e HC |
376 | pgdat->node_start_pfn = start_pfn; |
377 | ||
378 | pgdat->node_spanned_pages = max(old_pgdat_end_pfn, end_pfn) - | |
379 | pgdat->node_start_pfn; | |
380 | } | |
381 | ||
31168481 | 382 | static int __meminit __add_zone(struct zone *zone, unsigned long phys_start_pfn) |
3947be19 DH |
383 | { |
384 | struct pglist_data *pgdat = zone->zone_pgdat; | |
385 | int nr_pages = PAGES_PER_SECTION; | |
386 | int nid = pgdat->node_id; | |
387 | int zone_type; | |
76cdd58e | 388 | unsigned long flags; |
64dd1b29 | 389 | int ret; |
3947be19 DH |
390 | |
391 | zone_type = zone - pgdat->node_zones; | |
64dd1b29 CS |
392 | ret = ensure_zone_is_initialized(zone, phys_start_pfn, nr_pages); |
393 | if (ret) | |
394 | return ret; | |
76cdd58e | 395 | |
76cdd58e HC |
396 | pgdat_resize_lock(zone->zone_pgdat, &flags); |
397 | grow_zone_span(zone, phys_start_pfn, phys_start_pfn + nr_pages); | |
398 | grow_pgdat_span(zone->zone_pgdat, phys_start_pfn, | |
399 | phys_start_pfn + nr_pages); | |
400 | pgdat_resize_unlock(zone->zone_pgdat, &flags); | |
a2f3aa02 DH |
401 | memmap_init_zone(nr_pages, nid, zone_type, |
402 | phys_start_pfn, MEMMAP_HOTPLUG); | |
718127cc | 403 | return 0; |
3947be19 DH |
404 | } |
405 | ||
c04fc586 GH |
406 | static int __meminit __add_section(int nid, struct zone *zone, |
407 | unsigned long phys_start_pfn) | |
3947be19 | 408 | { |
3947be19 | 409 | int nr_pages = PAGES_PER_SECTION; |
3947be19 DH |
410 | int ret; |
411 | ||
ebd15302 KH |
412 | if (pfn_valid(phys_start_pfn)) |
413 | return -EEXIST; | |
414 | ||
0b0acbec | 415 | ret = sparse_add_one_section(zone, phys_start_pfn, nr_pages); |
3947be19 DH |
416 | |
417 | if (ret < 0) | |
418 | return ret; | |
419 | ||
718127cc YG |
420 | ret = __add_zone(zone, phys_start_pfn); |
421 | ||
422 | if (ret < 0) | |
423 | return ret; | |
424 | ||
c04fc586 | 425 | return register_new_memory(nid, __pfn_to_section(phys_start_pfn)); |
3947be19 DH |
426 | } |
427 | ||
4edd7cef DR |
428 | /* |
429 | * Reasonably generic function for adding memory. It is | |
430 | * expected that archs that support memory hotplug will | |
431 | * call this function after deciding the zone to which to | |
432 | * add the new pages. | |
433 | */ | |
434 | int __ref __add_pages(int nid, struct zone *zone, unsigned long phys_start_pfn, | |
435 | unsigned long nr_pages) | |
436 | { | |
437 | unsigned long i; | |
438 | int err = 0; | |
439 | int start_sec, end_sec; | |
440 | /* during initialize mem_map, align hot-added range to section */ | |
441 | start_sec = pfn_to_section_nr(phys_start_pfn); | |
442 | end_sec = pfn_to_section_nr(phys_start_pfn + nr_pages - 1); | |
443 | ||
444 | for (i = start_sec; i <= end_sec; i++) { | |
445 | err = __add_section(nid, zone, i << PFN_SECTION_SHIFT); | |
446 | ||
447 | /* | |
448 | * EEXIST is finally dealt with by ioresource collision | |
449 | * check. see add_memory() => register_memory_resource() | |
450 | * Warning will be printed if there is collision. | |
451 | */ | |
452 | if (err && (err != -EEXIST)) | |
453 | break; | |
454 | err = 0; | |
455 | } | |
456 | ||
457 | return err; | |
458 | } | |
459 | EXPORT_SYMBOL_GPL(__add_pages); | |
460 | ||
461 | #ifdef CONFIG_MEMORY_HOTREMOVE | |
815121d2 YI |
462 | /* find the smallest valid pfn in the range [start_pfn, end_pfn) */ |
463 | static int find_smallest_section_pfn(int nid, struct zone *zone, | |
464 | unsigned long start_pfn, | |
465 | unsigned long end_pfn) | |
466 | { | |
467 | struct mem_section *ms; | |
468 | ||
469 | for (; start_pfn < end_pfn; start_pfn += PAGES_PER_SECTION) { | |
470 | ms = __pfn_to_section(start_pfn); | |
471 | ||
472 | if (unlikely(!valid_section(ms))) | |
473 | continue; | |
474 | ||
475 | if (unlikely(pfn_to_nid(start_pfn) != nid)) | |
476 | continue; | |
477 | ||
478 | if (zone && zone != page_zone(pfn_to_page(start_pfn))) | |
479 | continue; | |
480 | ||
481 | return start_pfn; | |
482 | } | |
483 | ||
484 | return 0; | |
485 | } | |
486 | ||
487 | /* find the biggest valid pfn in the range [start_pfn, end_pfn). */ | |
488 | static int find_biggest_section_pfn(int nid, struct zone *zone, | |
489 | unsigned long start_pfn, | |
490 | unsigned long end_pfn) | |
491 | { | |
492 | struct mem_section *ms; | |
493 | unsigned long pfn; | |
494 | ||
495 | /* pfn is the end pfn of a memory section. */ | |
496 | pfn = end_pfn - 1; | |
497 | for (; pfn >= start_pfn; pfn -= PAGES_PER_SECTION) { | |
498 | ms = __pfn_to_section(pfn); | |
499 | ||
500 | if (unlikely(!valid_section(ms))) | |
501 | continue; | |
502 | ||
503 | if (unlikely(pfn_to_nid(pfn) != nid)) | |
504 | continue; | |
505 | ||
506 | if (zone && zone != page_zone(pfn_to_page(pfn))) | |
507 | continue; | |
508 | ||
509 | return pfn; | |
510 | } | |
511 | ||
512 | return 0; | |
513 | } | |
514 | ||
515 | static void shrink_zone_span(struct zone *zone, unsigned long start_pfn, | |
516 | unsigned long end_pfn) | |
517 | { | |
c33bc315 XQ |
518 | unsigned long zone_start_pfn = zone->zone_start_pfn; |
519 | unsigned long z = zone_end_pfn(zone); /* zone_end_pfn namespace clash */ | |
520 | unsigned long zone_end_pfn = z; | |
815121d2 YI |
521 | unsigned long pfn; |
522 | struct mem_section *ms; | |
523 | int nid = zone_to_nid(zone); | |
524 | ||
525 | zone_span_writelock(zone); | |
526 | if (zone_start_pfn == start_pfn) { | |
527 | /* | |
528 | * If the section is smallest section in the zone, it need | |
529 | * shrink zone->zone_start_pfn and zone->zone_spanned_pages. | |
530 | * In this case, we find second smallest valid mem_section | |
531 | * for shrinking zone. | |
532 | */ | |
533 | pfn = find_smallest_section_pfn(nid, zone, end_pfn, | |
534 | zone_end_pfn); | |
535 | if (pfn) { | |
536 | zone->zone_start_pfn = pfn; | |
537 | zone->spanned_pages = zone_end_pfn - pfn; | |
538 | } | |
539 | } else if (zone_end_pfn == end_pfn) { | |
540 | /* | |
541 | * If the section is biggest section in the zone, it need | |
542 | * shrink zone->spanned_pages. | |
543 | * In this case, we find second biggest valid mem_section for | |
544 | * shrinking zone. | |
545 | */ | |
546 | pfn = find_biggest_section_pfn(nid, zone, zone_start_pfn, | |
547 | start_pfn); | |
548 | if (pfn) | |
549 | zone->spanned_pages = pfn - zone_start_pfn + 1; | |
550 | } | |
551 | ||
552 | /* | |
553 | * The section is not biggest or smallest mem_section in the zone, it | |
554 | * only creates a hole in the zone. So in this case, we need not | |
555 | * change the zone. But perhaps, the zone has only hole data. Thus | |
556 | * it check the zone has only hole or not. | |
557 | */ | |
558 | pfn = zone_start_pfn; | |
559 | for (; pfn < zone_end_pfn; pfn += PAGES_PER_SECTION) { | |
560 | ms = __pfn_to_section(pfn); | |
561 | ||
562 | if (unlikely(!valid_section(ms))) | |
563 | continue; | |
564 | ||
565 | if (page_zone(pfn_to_page(pfn)) != zone) | |
566 | continue; | |
567 | ||
568 | /* If the section is current section, it continues the loop */ | |
569 | if (start_pfn == pfn) | |
570 | continue; | |
571 | ||
572 | /* If we find valid section, we have nothing to do */ | |
573 | zone_span_writeunlock(zone); | |
574 | return; | |
575 | } | |
576 | ||
577 | /* The zone has no valid section */ | |
578 | zone->zone_start_pfn = 0; | |
579 | zone->spanned_pages = 0; | |
580 | zone_span_writeunlock(zone); | |
581 | } | |
582 | ||
583 | static void shrink_pgdat_span(struct pglist_data *pgdat, | |
584 | unsigned long start_pfn, unsigned long end_pfn) | |
585 | { | |
586 | unsigned long pgdat_start_pfn = pgdat->node_start_pfn; | |
587 | unsigned long pgdat_end_pfn = | |
588 | pgdat->node_start_pfn + pgdat->node_spanned_pages; | |
589 | unsigned long pfn; | |
590 | struct mem_section *ms; | |
591 | int nid = pgdat->node_id; | |
592 | ||
593 | if (pgdat_start_pfn == start_pfn) { | |
594 | /* | |
595 | * If the section is smallest section in the pgdat, it need | |
596 | * shrink pgdat->node_start_pfn and pgdat->node_spanned_pages. | |
597 | * In this case, we find second smallest valid mem_section | |
598 | * for shrinking zone. | |
599 | */ | |
600 | pfn = find_smallest_section_pfn(nid, NULL, end_pfn, | |
601 | pgdat_end_pfn); | |
602 | if (pfn) { | |
603 | pgdat->node_start_pfn = pfn; | |
604 | pgdat->node_spanned_pages = pgdat_end_pfn - pfn; | |
605 | } | |
606 | } else if (pgdat_end_pfn == end_pfn) { | |
607 | /* | |
608 | * If the section is biggest section in the pgdat, it need | |
609 | * shrink pgdat->node_spanned_pages. | |
610 | * In this case, we find second biggest valid mem_section for | |
611 | * shrinking zone. | |
612 | */ | |
613 | pfn = find_biggest_section_pfn(nid, NULL, pgdat_start_pfn, | |
614 | start_pfn); | |
615 | if (pfn) | |
616 | pgdat->node_spanned_pages = pfn - pgdat_start_pfn + 1; | |
617 | } | |
618 | ||
619 | /* | |
620 | * If the section is not biggest or smallest mem_section in the pgdat, | |
621 | * it only creates a hole in the pgdat. So in this case, we need not | |
622 | * change the pgdat. | |
623 | * But perhaps, the pgdat has only hole data. Thus it check the pgdat | |
624 | * has only hole or not. | |
625 | */ | |
626 | pfn = pgdat_start_pfn; | |
627 | for (; pfn < pgdat_end_pfn; pfn += PAGES_PER_SECTION) { | |
628 | ms = __pfn_to_section(pfn); | |
629 | ||
630 | if (unlikely(!valid_section(ms))) | |
631 | continue; | |
632 | ||
633 | if (pfn_to_nid(pfn) != nid) | |
634 | continue; | |
635 | ||
636 | /* If the section is current section, it continues the loop */ | |
637 | if (start_pfn == pfn) | |
638 | continue; | |
639 | ||
640 | /* If we find valid section, we have nothing to do */ | |
641 | return; | |
642 | } | |
643 | ||
644 | /* The pgdat has no valid section */ | |
645 | pgdat->node_start_pfn = 0; | |
646 | pgdat->node_spanned_pages = 0; | |
647 | } | |
648 | ||
649 | static void __remove_zone(struct zone *zone, unsigned long start_pfn) | |
650 | { | |
651 | struct pglist_data *pgdat = zone->zone_pgdat; | |
652 | int nr_pages = PAGES_PER_SECTION; | |
653 | int zone_type; | |
654 | unsigned long flags; | |
655 | ||
656 | zone_type = zone - pgdat->node_zones; | |
657 | ||
658 | pgdat_resize_lock(zone->zone_pgdat, &flags); | |
659 | shrink_zone_span(zone, start_pfn, start_pfn + nr_pages); | |
660 | shrink_pgdat_span(pgdat, start_pfn, start_pfn + nr_pages); | |
661 | pgdat_resize_unlock(zone->zone_pgdat, &flags); | |
662 | } | |
663 | ||
ea01ea93 BP |
664 | static int __remove_section(struct zone *zone, struct mem_section *ms) |
665 | { | |
815121d2 YI |
666 | unsigned long start_pfn; |
667 | int scn_nr; | |
ea01ea93 BP |
668 | int ret = -EINVAL; |
669 | ||
670 | if (!valid_section(ms)) | |
671 | return ret; | |
672 | ||
673 | ret = unregister_memory_section(ms); | |
674 | if (ret) | |
675 | return ret; | |
676 | ||
815121d2 YI |
677 | scn_nr = __section_nr(ms); |
678 | start_pfn = section_nr_to_pfn(scn_nr); | |
679 | __remove_zone(zone, start_pfn); | |
680 | ||
ea01ea93 | 681 | sparse_remove_one_section(zone, ms); |
ea01ea93 BP |
682 | return 0; |
683 | } | |
684 | ||
ea01ea93 BP |
685 | /** |
686 | * __remove_pages() - remove sections of pages from a zone | |
687 | * @zone: zone from which pages need to be removed | |
688 | * @phys_start_pfn: starting pageframe (must be aligned to start of a section) | |
689 | * @nr_pages: number of pages to remove (must be multiple of section size) | |
690 | * | |
691 | * Generic helper function to remove section mappings and sysfs entries | |
692 | * for the section of the memory we are removing. Caller needs to make | |
693 | * sure that pages are marked reserved and zones are adjust properly by | |
694 | * calling offline_pages(). | |
695 | */ | |
696 | int __remove_pages(struct zone *zone, unsigned long phys_start_pfn, | |
697 | unsigned long nr_pages) | |
698 | { | |
fe74ebb1 | 699 | unsigned long i; |
ea01ea93 | 700 | int sections_to_remove; |
fe74ebb1 TK |
701 | resource_size_t start, size; |
702 | int ret = 0; | |
ea01ea93 BP |
703 | |
704 | /* | |
705 | * We can only remove entire sections | |
706 | */ | |
707 | BUG_ON(phys_start_pfn & ~PAGE_SECTION_MASK); | |
708 | BUG_ON(nr_pages % PAGES_PER_SECTION); | |
709 | ||
fe74ebb1 TK |
710 | start = phys_start_pfn << PAGE_SHIFT; |
711 | size = nr_pages * PAGE_SIZE; | |
712 | ret = release_mem_region_adjustable(&iomem_resource, start, size); | |
348f9f05 RD |
713 | if (ret) { |
714 | resource_size_t endres = start + size - 1; | |
715 | ||
716 | pr_warn("Unable to release resource <%pa-%pa> (%d)\n", | |
717 | &start, &endres, ret); | |
718 | } | |
d760afd4 | 719 | |
ea01ea93 BP |
720 | sections_to_remove = nr_pages / PAGES_PER_SECTION; |
721 | for (i = 0; i < sections_to_remove; i++) { | |
722 | unsigned long pfn = phys_start_pfn + i*PAGES_PER_SECTION; | |
723 | ret = __remove_section(zone, __pfn_to_section(pfn)); | |
724 | if (ret) | |
725 | break; | |
726 | } | |
727 | return ret; | |
728 | } | |
729 | EXPORT_SYMBOL_GPL(__remove_pages); | |
4edd7cef | 730 | #endif /* CONFIG_MEMORY_HOTREMOVE */ |
ea01ea93 | 731 | |
9d0ad8ca DK |
732 | int set_online_page_callback(online_page_callback_t callback) |
733 | { | |
734 | int rc = -EINVAL; | |
735 | ||
736 | lock_memory_hotplug(); | |
737 | ||
738 | if (online_page_callback == generic_online_page) { | |
739 | online_page_callback = callback; | |
740 | rc = 0; | |
741 | } | |
742 | ||
743 | unlock_memory_hotplug(); | |
744 | ||
745 | return rc; | |
746 | } | |
747 | EXPORT_SYMBOL_GPL(set_online_page_callback); | |
748 | ||
749 | int restore_online_page_callback(online_page_callback_t callback) | |
750 | { | |
751 | int rc = -EINVAL; | |
752 | ||
753 | lock_memory_hotplug(); | |
754 | ||
755 | if (online_page_callback == callback) { | |
756 | online_page_callback = generic_online_page; | |
757 | rc = 0; | |
758 | } | |
759 | ||
760 | unlock_memory_hotplug(); | |
761 | ||
762 | return rc; | |
763 | } | |
764 | EXPORT_SYMBOL_GPL(restore_online_page_callback); | |
765 | ||
766 | void __online_page_set_limits(struct page *page) | |
180c06ef | 767 | { |
9d0ad8ca DK |
768 | } |
769 | EXPORT_SYMBOL_GPL(__online_page_set_limits); | |
770 | ||
771 | void __online_page_increment_counters(struct page *page) | |
772 | { | |
3dcc0571 | 773 | adjust_managed_page_count(page, 1); |
9d0ad8ca DK |
774 | } |
775 | EXPORT_SYMBOL_GPL(__online_page_increment_counters); | |
180c06ef | 776 | |
9d0ad8ca DK |
777 | void __online_page_free(struct page *page) |
778 | { | |
3dcc0571 | 779 | __free_reserved_page(page); |
180c06ef | 780 | } |
9d0ad8ca DK |
781 | EXPORT_SYMBOL_GPL(__online_page_free); |
782 | ||
783 | static void generic_online_page(struct page *page) | |
784 | { | |
785 | __online_page_set_limits(page); | |
786 | __online_page_increment_counters(page); | |
787 | __online_page_free(page); | |
788 | } | |
180c06ef | 789 | |
75884fb1 KH |
790 | static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages, |
791 | void *arg) | |
3947be19 DH |
792 | { |
793 | unsigned long i; | |
75884fb1 KH |
794 | unsigned long onlined_pages = *(unsigned long *)arg; |
795 | struct page *page; | |
796 | if (PageReserved(pfn_to_page(start_pfn))) | |
797 | for (i = 0; i < nr_pages; i++) { | |
798 | page = pfn_to_page(start_pfn + i); | |
9d0ad8ca | 799 | (*online_page_callback)(page); |
75884fb1 KH |
800 | onlined_pages++; |
801 | } | |
802 | *(unsigned long *)arg = onlined_pages; | |
803 | return 0; | |
804 | } | |
805 | ||
09285af7 | 806 | #ifdef CONFIG_MOVABLE_NODE |
79a4dcef TC |
807 | /* |
808 | * When CONFIG_MOVABLE_NODE, we permit onlining of a node which doesn't have | |
809 | * normal memory. | |
810 | */ | |
09285af7 LJ |
811 | static bool can_online_high_movable(struct zone *zone) |
812 | { | |
813 | return true; | |
814 | } | |
79a4dcef | 815 | #else /* CONFIG_MOVABLE_NODE */ |
74d42d8f LJ |
816 | /* ensure every online node has NORMAL memory */ |
817 | static bool can_online_high_movable(struct zone *zone) | |
818 | { | |
819 | return node_state(zone_to_nid(zone), N_NORMAL_MEMORY); | |
820 | } | |
79a4dcef | 821 | #endif /* CONFIG_MOVABLE_NODE */ |
74d42d8f | 822 | |
d9713679 LJ |
823 | /* check which state of node_states will be changed when online memory */ |
824 | static void node_states_check_changes_online(unsigned long nr_pages, | |
825 | struct zone *zone, struct memory_notify *arg) | |
826 | { | |
827 | int nid = zone_to_nid(zone); | |
828 | enum zone_type zone_last = ZONE_NORMAL; | |
829 | ||
830 | /* | |
6715ddf9 LJ |
831 | * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY] |
832 | * contains nodes which have zones of 0...ZONE_NORMAL, | |
833 | * set zone_last to ZONE_NORMAL. | |
d9713679 | 834 | * |
6715ddf9 LJ |
835 | * If we don't have HIGHMEM nor movable node, |
836 | * node_states[N_NORMAL_MEMORY] contains nodes which have zones of | |
837 | * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE. | |
d9713679 | 838 | */ |
6715ddf9 | 839 | if (N_MEMORY == N_NORMAL_MEMORY) |
d9713679 LJ |
840 | zone_last = ZONE_MOVABLE; |
841 | ||
842 | /* | |
843 | * if the memory to be online is in a zone of 0...zone_last, and | |
844 | * the zones of 0...zone_last don't have memory before online, we will | |
845 | * need to set the node to node_states[N_NORMAL_MEMORY] after | |
846 | * the memory is online. | |
847 | */ | |
848 | if (zone_idx(zone) <= zone_last && !node_state(nid, N_NORMAL_MEMORY)) | |
849 | arg->status_change_nid_normal = nid; | |
850 | else | |
851 | arg->status_change_nid_normal = -1; | |
852 | ||
6715ddf9 LJ |
853 | #ifdef CONFIG_HIGHMEM |
854 | /* | |
855 | * If we have movable node, node_states[N_HIGH_MEMORY] | |
856 | * contains nodes which have zones of 0...ZONE_HIGHMEM, | |
857 | * set zone_last to ZONE_HIGHMEM. | |
858 | * | |
859 | * If we don't have movable node, node_states[N_NORMAL_MEMORY] | |
860 | * contains nodes which have zones of 0...ZONE_MOVABLE, | |
861 | * set zone_last to ZONE_MOVABLE. | |
862 | */ | |
863 | zone_last = ZONE_HIGHMEM; | |
864 | if (N_MEMORY == N_HIGH_MEMORY) | |
865 | zone_last = ZONE_MOVABLE; | |
866 | ||
867 | if (zone_idx(zone) <= zone_last && !node_state(nid, N_HIGH_MEMORY)) | |
868 | arg->status_change_nid_high = nid; | |
869 | else | |
870 | arg->status_change_nid_high = -1; | |
871 | #else | |
872 | arg->status_change_nid_high = arg->status_change_nid_normal; | |
873 | #endif | |
874 | ||
d9713679 LJ |
875 | /* |
876 | * if the node don't have memory befor online, we will need to | |
6715ddf9 | 877 | * set the node to node_states[N_MEMORY] after the memory |
d9713679 LJ |
878 | * is online. |
879 | */ | |
6715ddf9 | 880 | if (!node_state(nid, N_MEMORY)) |
d9713679 LJ |
881 | arg->status_change_nid = nid; |
882 | else | |
883 | arg->status_change_nid = -1; | |
884 | } | |
885 | ||
886 | static void node_states_set_node(int node, struct memory_notify *arg) | |
887 | { | |
888 | if (arg->status_change_nid_normal >= 0) | |
889 | node_set_state(node, N_NORMAL_MEMORY); | |
890 | ||
6715ddf9 LJ |
891 | if (arg->status_change_nid_high >= 0) |
892 | node_set_state(node, N_HIGH_MEMORY); | |
893 | ||
894 | node_set_state(node, N_MEMORY); | |
d9713679 LJ |
895 | } |
896 | ||
75884fb1 | 897 | |
511c2aba | 898 | int __ref online_pages(unsigned long pfn, unsigned long nr_pages, int online_type) |
75884fb1 | 899 | { |
aa47228a | 900 | unsigned long flags; |
3947be19 DH |
901 | unsigned long onlined_pages = 0; |
902 | struct zone *zone; | |
6811378e | 903 | int need_zonelists_rebuild = 0; |
7b78d335 YG |
904 | int nid; |
905 | int ret; | |
906 | struct memory_notify arg; | |
907 | ||
925268a0 | 908 | lock_memory_hotplug(); |
d9713679 LJ |
909 | /* |
910 | * This doesn't need a lock to do pfn_to_page(). | |
911 | * The section can't be removed here because of the | |
912 | * memory_block->state_mutex. | |
913 | */ | |
914 | zone = page_zone(pfn_to_page(pfn)); | |
915 | ||
74d42d8f LJ |
916 | if ((zone_idx(zone) > ZONE_NORMAL || online_type == ONLINE_MOVABLE) && |
917 | !can_online_high_movable(zone)) { | |
918 | unlock_memory_hotplug(); | |
0a1be150 | 919 | return -EINVAL; |
74d42d8f LJ |
920 | } |
921 | ||
511c2aba LJ |
922 | if (online_type == ONLINE_KERNEL && zone_idx(zone) == ZONE_MOVABLE) { |
923 | if (move_pfn_range_left(zone - 1, zone, pfn, pfn + nr_pages)) { | |
924 | unlock_memory_hotplug(); | |
0a1be150 | 925 | return -EINVAL; |
511c2aba LJ |
926 | } |
927 | } | |
928 | if (online_type == ONLINE_MOVABLE && zone_idx(zone) == ZONE_MOVABLE - 1) { | |
929 | if (move_pfn_range_right(zone, zone + 1, pfn, pfn + nr_pages)) { | |
930 | unlock_memory_hotplug(); | |
0a1be150 | 931 | return -EINVAL; |
511c2aba LJ |
932 | } |
933 | } | |
934 | ||
935 | /* Previous code may changed the zone of the pfn range */ | |
936 | zone = page_zone(pfn_to_page(pfn)); | |
937 | ||
7b78d335 YG |
938 | arg.start_pfn = pfn; |
939 | arg.nr_pages = nr_pages; | |
d9713679 | 940 | node_states_check_changes_online(nr_pages, zone, &arg); |
7b78d335 YG |
941 | |
942 | nid = page_to_nid(pfn_to_page(pfn)); | |
3947be19 | 943 | |
7b78d335 YG |
944 | ret = memory_notify(MEM_GOING_ONLINE, &arg); |
945 | ret = notifier_to_errno(ret); | |
946 | if (ret) { | |
947 | memory_notify(MEM_CANCEL_ONLINE, &arg); | |
925268a0 | 948 | unlock_memory_hotplug(); |
7b78d335 YG |
949 | return ret; |
950 | } | |
6811378e YG |
951 | /* |
952 | * If this zone is not populated, then it is not in zonelist. | |
953 | * This means the page allocator ignores this zone. | |
954 | * So, zonelist must be updated after online. | |
955 | */ | |
4eaf3f64 | 956 | mutex_lock(&zonelists_mutex); |
6dcd73d7 | 957 | if (!populated_zone(zone)) { |
6811378e | 958 | need_zonelists_rebuild = 1; |
6dcd73d7 WC |
959 | build_all_zonelists(NULL, zone); |
960 | } | |
6811378e | 961 | |
908eedc6 | 962 | ret = walk_system_ram_range(pfn, nr_pages, &onlined_pages, |
75884fb1 | 963 | online_pages_range); |
fd8a4221 | 964 | if (ret) { |
6dcd73d7 WC |
965 | if (need_zonelists_rebuild) |
966 | zone_pcp_reset(zone); | |
4eaf3f64 | 967 | mutex_unlock(&zonelists_mutex); |
a62e2f4f BH |
968 | printk(KERN_DEBUG "online_pages [mem %#010llx-%#010llx] failed\n", |
969 | (unsigned long long) pfn << PAGE_SHIFT, | |
970 | (((unsigned long long) pfn + nr_pages) | |
971 | << PAGE_SHIFT) - 1); | |
fd8a4221 | 972 | memory_notify(MEM_CANCEL_ONLINE, &arg); |
925268a0 | 973 | unlock_memory_hotplug(); |
fd8a4221 GL |
974 | return ret; |
975 | } | |
976 | ||
3947be19 | 977 | zone->present_pages += onlined_pages; |
aa47228a CS |
978 | |
979 | pgdat_resize_lock(zone->zone_pgdat, &flags); | |
f2937be5 | 980 | zone->zone_pgdat->node_present_pages += onlined_pages; |
aa47228a CS |
981 | pgdat_resize_unlock(zone->zone_pgdat, &flags); |
982 | ||
08dff7b7 | 983 | if (onlined_pages) { |
d9713679 | 984 | node_states_set_node(zone_to_nid(zone), &arg); |
08dff7b7 | 985 | if (need_zonelists_rebuild) |
6dcd73d7 | 986 | build_all_zonelists(NULL, NULL); |
08dff7b7 JL |
987 | else |
988 | zone_pcp_update(zone); | |
989 | } | |
3947be19 | 990 | |
4eaf3f64 | 991 | mutex_unlock(&zonelists_mutex); |
1b79acc9 KM |
992 | |
993 | init_per_zone_wmark_min(); | |
994 | ||
08dff7b7 | 995 | if (onlined_pages) |
7ea1530a | 996 | kswapd_run(zone_to_nid(zone)); |
61b13993 | 997 | |
1f522509 | 998 | vm_total_pages = nr_free_pagecache_pages(); |
2f7f24ec | 999 | |
2d1d43f6 | 1000 | writeback_set_ratelimit(); |
7b78d335 YG |
1001 | |
1002 | if (onlined_pages) | |
1003 | memory_notify(MEM_ONLINE, &arg); | |
925268a0 | 1004 | unlock_memory_hotplug(); |
7b78d335 | 1005 | |
3947be19 DH |
1006 | return 0; |
1007 | } | |
53947027 | 1008 | #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */ |
bc02af93 | 1009 | |
e1319331 HS |
1010 | /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */ |
1011 | static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start) | |
9af3c2de YG |
1012 | { |
1013 | struct pglist_data *pgdat; | |
1014 | unsigned long zones_size[MAX_NR_ZONES] = {0}; | |
1015 | unsigned long zholes_size[MAX_NR_ZONES] = {0}; | |
1016 | unsigned long start_pfn = start >> PAGE_SHIFT; | |
1017 | ||
a1e565aa TC |
1018 | pgdat = NODE_DATA(nid); |
1019 | if (!pgdat) { | |
1020 | pgdat = arch_alloc_nodedata(nid); | |
1021 | if (!pgdat) | |
1022 | return NULL; | |
9af3c2de | 1023 | |
a1e565aa TC |
1024 | arch_refresh_nodedata(nid, pgdat); |
1025 | } | |
9af3c2de YG |
1026 | |
1027 | /* we can use NODE_DATA(nid) from here */ | |
1028 | ||
1029 | /* init node's zones as empty zones, we don't have any present pages.*/ | |
9109fb7b | 1030 | free_area_init_node(nid, zones_size, start_pfn, zholes_size); |
9af3c2de | 1031 | |
959ecc48 KH |
1032 | /* |
1033 | * The node we allocated has no zone fallback lists. For avoiding | |
1034 | * to access not-initialized zonelist, build here. | |
1035 | */ | |
f957db4f | 1036 | mutex_lock(&zonelists_mutex); |
9adb62a5 | 1037 | build_all_zonelists(pgdat, NULL); |
f957db4f | 1038 | mutex_unlock(&zonelists_mutex); |
959ecc48 | 1039 | |
9af3c2de YG |
1040 | return pgdat; |
1041 | } | |
1042 | ||
1043 | static void rollback_node_hotadd(int nid, pg_data_t *pgdat) | |
1044 | { | |
1045 | arch_refresh_nodedata(nid, NULL); | |
1046 | arch_free_nodedata(pgdat); | |
1047 | return; | |
1048 | } | |
1049 | ||
0a547039 | 1050 | |
cf23422b | 1051 | /* |
1052 | * called by cpu_up() to online a node without onlined memory. | |
1053 | */ | |
1054 | int mem_online_node(int nid) | |
1055 | { | |
1056 | pg_data_t *pgdat; | |
1057 | int ret; | |
1058 | ||
20d6c96b | 1059 | lock_memory_hotplug(); |
cf23422b | 1060 | pgdat = hotadd_new_pgdat(nid, 0); |
7553e8f2 | 1061 | if (!pgdat) { |
cf23422b | 1062 | ret = -ENOMEM; |
1063 | goto out; | |
1064 | } | |
1065 | node_set_online(nid); | |
1066 | ret = register_one_node(nid); | |
1067 | BUG_ON(ret); | |
1068 | ||
1069 | out: | |
20d6c96b | 1070 | unlock_memory_hotplug(); |
cf23422b | 1071 | return ret; |
1072 | } | |
1073 | ||
27356f54 TK |
1074 | static int check_hotplug_memory_range(u64 start, u64 size) |
1075 | { | |
1076 | u64 start_pfn = start >> PAGE_SHIFT; | |
1077 | u64 nr_pages = size >> PAGE_SHIFT; | |
1078 | ||
1079 | /* Memory range must be aligned with section */ | |
1080 | if ((start_pfn & ~PAGE_SECTION_MASK) || | |
1081 | (nr_pages % PAGES_PER_SECTION) || (!nr_pages)) { | |
1082 | pr_err("Section-unaligned hotplug range: start 0x%llx, size 0x%llx\n", | |
1083 | (unsigned long long)start, | |
1084 | (unsigned long long)size); | |
1085 | return -EINVAL; | |
1086 | } | |
1087 | ||
1088 | return 0; | |
1089 | } | |
1090 | ||
31168481 AV |
1091 | /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */ |
1092 | int __ref add_memory(int nid, u64 start, u64 size) | |
bc02af93 | 1093 | { |
9af3c2de | 1094 | pg_data_t *pgdat = NULL; |
a1e565aa TC |
1095 | bool new_pgdat; |
1096 | bool new_node; | |
ebd15302 | 1097 | struct resource *res; |
bc02af93 YG |
1098 | int ret; |
1099 | ||
27356f54 TK |
1100 | ret = check_hotplug_memory_range(start, size); |
1101 | if (ret) | |
1102 | return ret; | |
1103 | ||
20d6c96b | 1104 | lock_memory_hotplug(); |
6ad696d2 | 1105 | |
ebd15302 | 1106 | res = register_memory_resource(start, size); |
6ad696d2 | 1107 | ret = -EEXIST; |
ebd15302 | 1108 | if (!res) |
6ad696d2 | 1109 | goto out; |
ebd15302 | 1110 | |
a1e565aa TC |
1111 | { /* Stupid hack to suppress address-never-null warning */ |
1112 | void *p = NODE_DATA(nid); | |
1113 | new_pgdat = !p; | |
1114 | } | |
1115 | new_node = !node_online(nid); | |
1116 | if (new_node) { | |
9af3c2de | 1117 | pgdat = hotadd_new_pgdat(nid, start); |
6ad696d2 | 1118 | ret = -ENOMEM; |
9af3c2de | 1119 | if (!pgdat) |
41b9e2d7 | 1120 | goto error; |
9af3c2de YG |
1121 | } |
1122 | ||
bc02af93 YG |
1123 | /* call arch's memory hotadd */ |
1124 | ret = arch_add_memory(nid, start, size); | |
1125 | ||
9af3c2de YG |
1126 | if (ret < 0) |
1127 | goto error; | |
1128 | ||
0fc44159 | 1129 | /* we online node here. we can't roll back from here. */ |
9af3c2de YG |
1130 | node_set_online(nid); |
1131 | ||
a1e565aa | 1132 | if (new_node) { |
0fc44159 YG |
1133 | ret = register_one_node(nid); |
1134 | /* | |
1135 | * If sysfs file of new node can't create, cpu on the node | |
1136 | * can't be hot-added. There is no rollback way now. | |
1137 | * So, check by BUG_ON() to catch it reluctantly.. | |
1138 | */ | |
1139 | BUG_ON(ret); | |
1140 | } | |
1141 | ||
d96ae530 | 1142 | /* create new memmap entry */ |
1143 | firmware_map_add_hotplug(start, start + size, "System RAM"); | |
1144 | ||
6ad696d2 AK |
1145 | goto out; |
1146 | ||
9af3c2de YG |
1147 | error: |
1148 | /* rollback pgdat allocation and others */ | |
1149 | if (new_pgdat) | |
1150 | rollback_node_hotadd(nid, pgdat); | |
a864b9d0 | 1151 | release_memory_resource(res); |
9af3c2de | 1152 | |
6ad696d2 | 1153 | out: |
20d6c96b | 1154 | unlock_memory_hotplug(); |
bc02af93 YG |
1155 | return ret; |
1156 | } | |
1157 | EXPORT_SYMBOL_GPL(add_memory); | |
0c0e6195 KH |
1158 | |
1159 | #ifdef CONFIG_MEMORY_HOTREMOVE | |
5c755e9f BP |
1160 | /* |
1161 | * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy | |
1162 | * set and the size of the free page is given by page_order(). Using this, | |
1163 | * the function determines if the pageblock contains only free pages. | |
1164 | * Due to buddy contraints, a free page at least the size of a pageblock will | |
1165 | * be located at the start of the pageblock | |
1166 | */ | |
1167 | static inline int pageblock_free(struct page *page) | |
1168 | { | |
1169 | return PageBuddy(page) && page_order(page) >= pageblock_order; | |
1170 | } | |
1171 | ||
1172 | /* Return the start of the next active pageblock after a given page */ | |
1173 | static struct page *next_active_pageblock(struct page *page) | |
1174 | { | |
5c755e9f BP |
1175 | /* Ensure the starting page is pageblock-aligned */ |
1176 | BUG_ON(page_to_pfn(page) & (pageblock_nr_pages - 1)); | |
1177 | ||
5c755e9f | 1178 | /* If the entire pageblock is free, move to the end of free page */ |
0dcc48c1 KH |
1179 | if (pageblock_free(page)) { |
1180 | int order; | |
1181 | /* be careful. we don't have locks, page_order can be changed.*/ | |
1182 | order = page_order(page); | |
1183 | if ((order < MAX_ORDER) && (order >= pageblock_order)) | |
1184 | return page + (1 << order); | |
1185 | } | |
5c755e9f | 1186 | |
0dcc48c1 | 1187 | return page + pageblock_nr_pages; |
5c755e9f BP |
1188 | } |
1189 | ||
1190 | /* Checks if this range of memory is likely to be hot-removable. */ | |
1191 | int is_mem_section_removable(unsigned long start_pfn, unsigned long nr_pages) | |
1192 | { | |
5c755e9f BP |
1193 | struct page *page = pfn_to_page(start_pfn); |
1194 | struct page *end_page = page + nr_pages; | |
1195 | ||
1196 | /* Check the starting page of each pageblock within the range */ | |
1197 | for (; page < end_page; page = next_active_pageblock(page)) { | |
49ac8255 | 1198 | if (!is_pageblock_removable_nolock(page)) |
5c755e9f | 1199 | return 0; |
49ac8255 | 1200 | cond_resched(); |
5c755e9f BP |
1201 | } |
1202 | ||
1203 | /* All pageblocks in the memory block are likely to be hot-removable */ | |
1204 | return 1; | |
1205 | } | |
1206 | ||
0c0e6195 KH |
1207 | /* |
1208 | * Confirm all pages in a range [start, end) is belongs to the same zone. | |
1209 | */ | |
1210 | static int test_pages_in_a_zone(unsigned long start_pfn, unsigned long end_pfn) | |
1211 | { | |
1212 | unsigned long pfn; | |
1213 | struct zone *zone = NULL; | |
1214 | struct page *page; | |
1215 | int i; | |
1216 | for (pfn = start_pfn; | |
1217 | pfn < end_pfn; | |
1218 | pfn += MAX_ORDER_NR_PAGES) { | |
1219 | i = 0; | |
1220 | /* This is just a CONFIG_HOLES_IN_ZONE check.*/ | |
1221 | while ((i < MAX_ORDER_NR_PAGES) && !pfn_valid_within(pfn + i)) | |
1222 | i++; | |
1223 | if (i == MAX_ORDER_NR_PAGES) | |
1224 | continue; | |
1225 | page = pfn_to_page(pfn + i); | |
1226 | if (zone && page_zone(page) != zone) | |
1227 | return 0; | |
1228 | zone = page_zone(page); | |
1229 | } | |
1230 | return 1; | |
1231 | } | |
1232 | ||
1233 | /* | |
c8721bbb NH |
1234 | * Scan pfn range [start,end) to find movable/migratable pages (LRU pages |
1235 | * and hugepages). We scan pfn because it's much easier than scanning over | |
1236 | * linked list. This function returns the pfn of the first found movable | |
1237 | * page if it's found, otherwise 0. | |
0c0e6195 | 1238 | */ |
c8721bbb | 1239 | static unsigned long scan_movable_pages(unsigned long start, unsigned long end) |
0c0e6195 KH |
1240 | { |
1241 | unsigned long pfn; | |
1242 | struct page *page; | |
1243 | for (pfn = start; pfn < end; pfn++) { | |
1244 | if (pfn_valid(pfn)) { | |
1245 | page = pfn_to_page(pfn); | |
1246 | if (PageLRU(page)) | |
1247 | return pfn; | |
c8721bbb NH |
1248 | if (PageHuge(page)) { |
1249 | if (is_hugepage_active(page)) | |
1250 | return pfn; | |
1251 | else | |
1252 | pfn = round_up(pfn + 1, | |
1253 | 1 << compound_order(page)) - 1; | |
1254 | } | |
0c0e6195 KH |
1255 | } |
1256 | } | |
1257 | return 0; | |
1258 | } | |
1259 | ||
0c0e6195 KH |
1260 | #define NR_OFFLINE_AT_ONCE_PAGES (256) |
1261 | static int | |
1262 | do_migrate_range(unsigned long start_pfn, unsigned long end_pfn) | |
1263 | { | |
1264 | unsigned long pfn; | |
1265 | struct page *page; | |
1266 | int move_pages = NR_OFFLINE_AT_ONCE_PAGES; | |
1267 | int not_managed = 0; | |
1268 | int ret = 0; | |
1269 | LIST_HEAD(source); | |
1270 | ||
1271 | for (pfn = start_pfn; pfn < end_pfn && move_pages > 0; pfn++) { | |
1272 | if (!pfn_valid(pfn)) | |
1273 | continue; | |
1274 | page = pfn_to_page(pfn); | |
c8721bbb NH |
1275 | |
1276 | if (PageHuge(page)) { | |
1277 | struct page *head = compound_head(page); | |
1278 | pfn = page_to_pfn(head) + (1<<compound_order(head)) - 1; | |
1279 | if (compound_order(head) > PFN_SECTION_SHIFT) { | |
1280 | ret = -EBUSY; | |
1281 | break; | |
1282 | } | |
1283 | if (isolate_huge_page(page, &source)) | |
1284 | move_pages -= 1 << compound_order(head); | |
1285 | continue; | |
1286 | } | |
1287 | ||
700c2a46 | 1288 | if (!get_page_unless_zero(page)) |
0c0e6195 KH |
1289 | continue; |
1290 | /* | |
1291 | * We can skip free pages. And we can only deal with pages on | |
1292 | * LRU. | |
1293 | */ | |
62695a84 | 1294 | ret = isolate_lru_page(page); |
0c0e6195 | 1295 | if (!ret) { /* Success */ |
700c2a46 | 1296 | put_page(page); |
62695a84 | 1297 | list_add_tail(&page->lru, &source); |
0c0e6195 | 1298 | move_pages--; |
6d9c285a KM |
1299 | inc_zone_page_state(page, NR_ISOLATED_ANON + |
1300 | page_is_file_cache(page)); | |
1301 | ||
0c0e6195 | 1302 | } else { |
0c0e6195 | 1303 | #ifdef CONFIG_DEBUG_VM |
718a3821 WF |
1304 | printk(KERN_ALERT "removing pfn %lx from LRU failed\n", |
1305 | pfn); | |
1306 | dump_page(page); | |
0c0e6195 | 1307 | #endif |
700c2a46 | 1308 | put_page(page); |
25985edc | 1309 | /* Because we don't have big zone->lock. we should |
809c4449 BL |
1310 | check this again here. */ |
1311 | if (page_count(page)) { | |
1312 | not_managed++; | |
f3ab2636 | 1313 | ret = -EBUSY; |
809c4449 BL |
1314 | break; |
1315 | } | |
0c0e6195 KH |
1316 | } |
1317 | } | |
f3ab2636 BL |
1318 | if (!list_empty(&source)) { |
1319 | if (not_managed) { | |
c8721bbb | 1320 | putback_movable_pages(&source); |
f3ab2636 BL |
1321 | goto out; |
1322 | } | |
74c08f98 MK |
1323 | |
1324 | /* | |
1325 | * alloc_migrate_target should be improooooved!! | |
1326 | * migrate_pages returns # of failed pages. | |
1327 | */ | |
1328 | ret = migrate_pages(&source, alloc_migrate_target, 0, | |
9c620e2b | 1329 | MIGRATE_SYNC, MR_MEMORY_HOTPLUG); |
f3ab2636 | 1330 | if (ret) |
c8721bbb | 1331 | putback_movable_pages(&source); |
0c0e6195 | 1332 | } |
0c0e6195 KH |
1333 | out: |
1334 | return ret; | |
1335 | } | |
1336 | ||
1337 | /* | |
1338 | * remove from free_area[] and mark all as Reserved. | |
1339 | */ | |
1340 | static int | |
1341 | offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages, | |
1342 | void *data) | |
1343 | { | |
1344 | __offline_isolated_pages(start, start + nr_pages); | |
1345 | return 0; | |
1346 | } | |
1347 | ||
1348 | static void | |
1349 | offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn) | |
1350 | { | |
908eedc6 | 1351 | walk_system_ram_range(start_pfn, end_pfn - start_pfn, NULL, |
0c0e6195 KH |
1352 | offline_isolated_pages_cb); |
1353 | } | |
1354 | ||
1355 | /* | |
1356 | * Check all pages in range, recoreded as memory resource, are isolated. | |
1357 | */ | |
1358 | static int | |
1359 | check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages, | |
1360 | void *data) | |
1361 | { | |
1362 | int ret; | |
1363 | long offlined = *(long *)data; | |
b023f468 | 1364 | ret = test_pages_isolated(start_pfn, start_pfn + nr_pages, true); |
0c0e6195 KH |
1365 | offlined = nr_pages; |
1366 | if (!ret) | |
1367 | *(long *)data += offlined; | |
1368 | return ret; | |
1369 | } | |
1370 | ||
1371 | static long | |
1372 | check_pages_isolated(unsigned long start_pfn, unsigned long end_pfn) | |
1373 | { | |
1374 | long offlined = 0; | |
1375 | int ret; | |
1376 | ||
908eedc6 | 1377 | ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn, &offlined, |
0c0e6195 KH |
1378 | check_pages_isolated_cb); |
1379 | if (ret < 0) | |
1380 | offlined = (long)ret; | |
1381 | return offlined; | |
1382 | } | |
1383 | ||
09285af7 | 1384 | #ifdef CONFIG_MOVABLE_NODE |
79a4dcef TC |
1385 | /* |
1386 | * When CONFIG_MOVABLE_NODE, we permit offlining of a node which doesn't have | |
1387 | * normal memory. | |
1388 | */ | |
09285af7 LJ |
1389 | static bool can_offline_normal(struct zone *zone, unsigned long nr_pages) |
1390 | { | |
1391 | return true; | |
1392 | } | |
79a4dcef | 1393 | #else /* CONFIG_MOVABLE_NODE */ |
74d42d8f LJ |
1394 | /* ensure the node has NORMAL memory if it is still online */ |
1395 | static bool can_offline_normal(struct zone *zone, unsigned long nr_pages) | |
1396 | { | |
1397 | struct pglist_data *pgdat = zone->zone_pgdat; | |
1398 | unsigned long present_pages = 0; | |
1399 | enum zone_type zt; | |
1400 | ||
1401 | for (zt = 0; zt <= ZONE_NORMAL; zt++) | |
1402 | present_pages += pgdat->node_zones[zt].present_pages; | |
1403 | ||
1404 | if (present_pages > nr_pages) | |
1405 | return true; | |
1406 | ||
1407 | present_pages = 0; | |
1408 | for (; zt <= ZONE_MOVABLE; zt++) | |
1409 | present_pages += pgdat->node_zones[zt].present_pages; | |
1410 | ||
1411 | /* | |
1412 | * we can't offline the last normal memory until all | |
1413 | * higher memory is offlined. | |
1414 | */ | |
1415 | return present_pages == 0; | |
1416 | } | |
79a4dcef | 1417 | #endif /* CONFIG_MOVABLE_NODE */ |
74d42d8f | 1418 | |
d9713679 LJ |
1419 | /* check which state of node_states will be changed when offline memory */ |
1420 | static void node_states_check_changes_offline(unsigned long nr_pages, | |
1421 | struct zone *zone, struct memory_notify *arg) | |
1422 | { | |
1423 | struct pglist_data *pgdat = zone->zone_pgdat; | |
1424 | unsigned long present_pages = 0; | |
1425 | enum zone_type zt, zone_last = ZONE_NORMAL; | |
1426 | ||
1427 | /* | |
6715ddf9 LJ |
1428 | * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY] |
1429 | * contains nodes which have zones of 0...ZONE_NORMAL, | |
1430 | * set zone_last to ZONE_NORMAL. | |
d9713679 | 1431 | * |
6715ddf9 LJ |
1432 | * If we don't have HIGHMEM nor movable node, |
1433 | * node_states[N_NORMAL_MEMORY] contains nodes which have zones of | |
1434 | * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE. | |
d9713679 | 1435 | */ |
6715ddf9 | 1436 | if (N_MEMORY == N_NORMAL_MEMORY) |
d9713679 LJ |
1437 | zone_last = ZONE_MOVABLE; |
1438 | ||
1439 | /* | |
1440 | * check whether node_states[N_NORMAL_MEMORY] will be changed. | |
1441 | * If the memory to be offline is in a zone of 0...zone_last, | |
1442 | * and it is the last present memory, 0...zone_last will | |
1443 | * become empty after offline , thus we can determind we will | |
1444 | * need to clear the node from node_states[N_NORMAL_MEMORY]. | |
1445 | */ | |
1446 | for (zt = 0; 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_normal = zone_to_nid(zone); | |
1450 | else | |
1451 | arg->status_change_nid_normal = -1; | |
1452 | ||
6715ddf9 LJ |
1453 | #ifdef CONFIG_HIGHMEM |
1454 | /* | |
1455 | * If we have movable node, node_states[N_HIGH_MEMORY] | |
1456 | * contains nodes which have zones of 0...ZONE_HIGHMEM, | |
1457 | * set zone_last to ZONE_HIGHMEM. | |
1458 | * | |
1459 | * If we don't have movable node, node_states[N_NORMAL_MEMORY] | |
1460 | * contains nodes which have zones of 0...ZONE_MOVABLE, | |
1461 | * set zone_last to ZONE_MOVABLE. | |
1462 | */ | |
1463 | zone_last = ZONE_HIGHMEM; | |
1464 | if (N_MEMORY == N_HIGH_MEMORY) | |
1465 | zone_last = ZONE_MOVABLE; | |
1466 | ||
1467 | for (; zt <= zone_last; zt++) | |
1468 | present_pages += pgdat->node_zones[zt].present_pages; | |
1469 | if (zone_idx(zone) <= zone_last && nr_pages >= present_pages) | |
1470 | arg->status_change_nid_high = zone_to_nid(zone); | |
1471 | else | |
1472 | arg->status_change_nid_high = -1; | |
1473 | #else | |
1474 | arg->status_change_nid_high = arg->status_change_nid_normal; | |
1475 | #endif | |
1476 | ||
d9713679 LJ |
1477 | /* |
1478 | * node_states[N_HIGH_MEMORY] contains nodes which have 0...ZONE_MOVABLE | |
1479 | */ | |
1480 | zone_last = ZONE_MOVABLE; | |
1481 | ||
1482 | /* | |
1483 | * check whether node_states[N_HIGH_MEMORY] will be changed | |
1484 | * If we try to offline the last present @nr_pages from the node, | |
1485 | * we can determind we will need to clear the node from | |
1486 | * node_states[N_HIGH_MEMORY]. | |
1487 | */ | |
1488 | for (; zt <= zone_last; zt++) | |
1489 | present_pages += pgdat->node_zones[zt].present_pages; | |
1490 | if (nr_pages >= present_pages) | |
1491 | arg->status_change_nid = zone_to_nid(zone); | |
1492 | else | |
1493 | arg->status_change_nid = -1; | |
1494 | } | |
1495 | ||
1496 | static void node_states_clear_node(int node, struct memory_notify *arg) | |
1497 | { | |
1498 | if (arg->status_change_nid_normal >= 0) | |
1499 | node_clear_state(node, N_NORMAL_MEMORY); | |
1500 | ||
6715ddf9 LJ |
1501 | if ((N_MEMORY != N_NORMAL_MEMORY) && |
1502 | (arg->status_change_nid_high >= 0)) | |
d9713679 | 1503 | node_clear_state(node, N_HIGH_MEMORY); |
6715ddf9 LJ |
1504 | |
1505 | if ((N_MEMORY != N_HIGH_MEMORY) && | |
1506 | (arg->status_change_nid >= 0)) | |
1507 | node_clear_state(node, N_MEMORY); | |
d9713679 LJ |
1508 | } |
1509 | ||
a16cee10 | 1510 | static int __ref __offline_pages(unsigned long start_pfn, |
0c0e6195 KH |
1511 | unsigned long end_pfn, unsigned long timeout) |
1512 | { | |
1513 | unsigned long pfn, nr_pages, expire; | |
1514 | long offlined_pages; | |
7b78d335 | 1515 | int ret, drain, retry_max, node; |
d702909f | 1516 | unsigned long flags; |
0c0e6195 | 1517 | struct zone *zone; |
7b78d335 | 1518 | struct memory_notify arg; |
0c0e6195 | 1519 | |
0c0e6195 KH |
1520 | /* at least, alignment against pageblock is necessary */ |
1521 | if (!IS_ALIGNED(start_pfn, pageblock_nr_pages)) | |
1522 | return -EINVAL; | |
1523 | if (!IS_ALIGNED(end_pfn, pageblock_nr_pages)) | |
1524 | return -EINVAL; | |
1525 | /* This makes hotplug much easier...and readable. | |
1526 | we assume this for now. .*/ | |
1527 | if (!test_pages_in_a_zone(start_pfn, end_pfn)) | |
1528 | return -EINVAL; | |
7b78d335 | 1529 | |
20d6c96b | 1530 | lock_memory_hotplug(); |
6ad696d2 | 1531 | |
7b78d335 YG |
1532 | zone = page_zone(pfn_to_page(start_pfn)); |
1533 | node = zone_to_nid(zone); | |
1534 | nr_pages = end_pfn - start_pfn; | |
1535 | ||
74d42d8f LJ |
1536 | ret = -EINVAL; |
1537 | if (zone_idx(zone) <= ZONE_NORMAL && !can_offline_normal(zone, nr_pages)) | |
1538 | goto out; | |
1539 | ||
0c0e6195 | 1540 | /* set above range as isolated */ |
b023f468 WC |
1541 | ret = start_isolate_page_range(start_pfn, end_pfn, |
1542 | MIGRATE_MOVABLE, true); | |
0c0e6195 | 1543 | if (ret) |
6ad696d2 | 1544 | goto out; |
7b78d335 YG |
1545 | |
1546 | arg.start_pfn = start_pfn; | |
1547 | arg.nr_pages = nr_pages; | |
d9713679 | 1548 | node_states_check_changes_offline(nr_pages, zone, &arg); |
7b78d335 YG |
1549 | |
1550 | ret = memory_notify(MEM_GOING_OFFLINE, &arg); | |
1551 | ret = notifier_to_errno(ret); | |
1552 | if (ret) | |
1553 | goto failed_removal; | |
1554 | ||
0c0e6195 KH |
1555 | pfn = start_pfn; |
1556 | expire = jiffies + timeout; | |
1557 | drain = 0; | |
1558 | retry_max = 5; | |
1559 | repeat: | |
1560 | /* start memory hot removal */ | |
1561 | ret = -EAGAIN; | |
1562 | if (time_after(jiffies, expire)) | |
1563 | goto failed_removal; | |
1564 | ret = -EINTR; | |
1565 | if (signal_pending(current)) | |
1566 | goto failed_removal; | |
1567 | ret = 0; | |
1568 | if (drain) { | |
1569 | lru_add_drain_all(); | |
0c0e6195 | 1570 | cond_resched(); |
9f8f2172 | 1571 | drain_all_pages(); |
0c0e6195 KH |
1572 | } |
1573 | ||
c8721bbb NH |
1574 | pfn = scan_movable_pages(start_pfn, end_pfn); |
1575 | if (pfn) { /* We have movable pages */ | |
0c0e6195 KH |
1576 | ret = do_migrate_range(pfn, end_pfn); |
1577 | if (!ret) { | |
1578 | drain = 1; | |
1579 | goto repeat; | |
1580 | } else { | |
1581 | if (ret < 0) | |
1582 | if (--retry_max == 0) | |
1583 | goto failed_removal; | |
1584 | yield(); | |
1585 | drain = 1; | |
1586 | goto repeat; | |
1587 | } | |
1588 | } | |
b3834be5 | 1589 | /* drain all zone's lru pagevec, this is asynchronous... */ |
0c0e6195 | 1590 | lru_add_drain_all(); |
0c0e6195 | 1591 | yield(); |
b3834be5 | 1592 | /* drain pcp pages, this is synchronous. */ |
9f8f2172 | 1593 | drain_all_pages(); |
c8721bbb NH |
1594 | /* |
1595 | * dissolve free hugepages in the memory block before doing offlining | |
1596 | * actually in order to make hugetlbfs's object counting consistent. | |
1597 | */ | |
1598 | dissolve_free_huge_pages(start_pfn, end_pfn); | |
0c0e6195 KH |
1599 | /* check again */ |
1600 | offlined_pages = check_pages_isolated(start_pfn, end_pfn); | |
1601 | if (offlined_pages < 0) { | |
1602 | ret = -EBUSY; | |
1603 | goto failed_removal; | |
1604 | } | |
1605 | printk(KERN_INFO "Offlined Pages %ld\n", offlined_pages); | |
b3834be5 | 1606 | /* Ok, all of our target is isolated. |
0c0e6195 KH |
1607 | We cannot do rollback at this point. */ |
1608 | offline_isolated_pages(start_pfn, end_pfn); | |
dbc0e4ce | 1609 | /* reset pagetype flags and makes migrate type to be MOVABLE */ |
0815f3d8 | 1610 | undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE); |
0c0e6195 | 1611 | /* removal success */ |
3dcc0571 | 1612 | adjust_managed_page_count(pfn_to_page(start_pfn), -offlined_pages); |
0c0e6195 | 1613 | zone->present_pages -= offlined_pages; |
d702909f CS |
1614 | |
1615 | pgdat_resize_lock(zone->zone_pgdat, &flags); | |
0c0e6195 | 1616 | zone->zone_pgdat->node_present_pages -= offlined_pages; |
d702909f | 1617 | pgdat_resize_unlock(zone->zone_pgdat, &flags); |
7b78d335 | 1618 | |
1b79acc9 KM |
1619 | init_per_zone_wmark_min(); |
1620 | ||
1e8537ba | 1621 | if (!populated_zone(zone)) { |
340175b7 | 1622 | zone_pcp_reset(zone); |
1e8537ba XQ |
1623 | mutex_lock(&zonelists_mutex); |
1624 | build_all_zonelists(NULL, NULL); | |
1625 | mutex_unlock(&zonelists_mutex); | |
1626 | } else | |
1627 | zone_pcp_update(zone); | |
340175b7 | 1628 | |
d9713679 LJ |
1629 | node_states_clear_node(node, &arg); |
1630 | if (arg.status_change_nid >= 0) | |
8fe23e05 | 1631 | kswapd_stop(node); |
bce7394a | 1632 | |
0c0e6195 KH |
1633 | vm_total_pages = nr_free_pagecache_pages(); |
1634 | writeback_set_ratelimit(); | |
7b78d335 YG |
1635 | |
1636 | memory_notify(MEM_OFFLINE, &arg); | |
20d6c96b | 1637 | unlock_memory_hotplug(); |
0c0e6195 KH |
1638 | return 0; |
1639 | ||
1640 | failed_removal: | |
a62e2f4f BH |
1641 | printk(KERN_INFO "memory offlining [mem %#010llx-%#010llx] failed\n", |
1642 | (unsigned long long) start_pfn << PAGE_SHIFT, | |
1643 | ((unsigned long long) end_pfn << PAGE_SHIFT) - 1); | |
7b78d335 | 1644 | memory_notify(MEM_CANCEL_OFFLINE, &arg); |
0c0e6195 | 1645 | /* pushback to free area */ |
0815f3d8 | 1646 | undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE); |
7b78d335 | 1647 | |
6ad696d2 | 1648 | out: |
20d6c96b | 1649 | unlock_memory_hotplug(); |
0c0e6195 KH |
1650 | return ret; |
1651 | } | |
71088785 | 1652 | |
a16cee10 WC |
1653 | int offline_pages(unsigned long start_pfn, unsigned long nr_pages) |
1654 | { | |
1655 | return __offline_pages(start_pfn, start_pfn + nr_pages, 120 * HZ); | |
1656 | } | |
e2ff3940 | 1657 | #endif /* CONFIG_MEMORY_HOTREMOVE */ |
a16cee10 | 1658 | |
bbc76be6 WC |
1659 | /** |
1660 | * walk_memory_range - walks through all mem sections in [start_pfn, end_pfn) | |
1661 | * @start_pfn: start pfn of the memory range | |
e05c4bbf | 1662 | * @end_pfn: end pfn of the memory range |
bbc76be6 WC |
1663 | * @arg: argument passed to func |
1664 | * @func: callback for each memory section walked | |
1665 | * | |
1666 | * This function walks through all present mem sections in range | |
1667 | * [start_pfn, end_pfn) and call func on each mem section. | |
1668 | * | |
1669 | * Returns the return value of func. | |
1670 | */ | |
e2ff3940 | 1671 | int walk_memory_range(unsigned long start_pfn, unsigned long end_pfn, |
bbc76be6 | 1672 | void *arg, int (*func)(struct memory_block *, void *)) |
71088785 | 1673 | { |
e90bdb7f WC |
1674 | struct memory_block *mem = NULL; |
1675 | struct mem_section *section; | |
e90bdb7f WC |
1676 | unsigned long pfn, section_nr; |
1677 | int ret; | |
e90bdb7f WC |
1678 | |
1679 | for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) { | |
1680 | section_nr = pfn_to_section_nr(pfn); | |
1681 | if (!present_section_nr(section_nr)) | |
1682 | continue; | |
1683 | ||
1684 | section = __nr_to_section(section_nr); | |
1685 | /* same memblock? */ | |
1686 | if (mem) | |
1687 | if ((section_nr >= mem->start_section_nr) && | |
1688 | (section_nr <= mem->end_section_nr)) | |
1689 | continue; | |
1690 | ||
1691 | mem = find_memory_block_hinted(section, mem); | |
1692 | if (!mem) | |
1693 | continue; | |
1694 | ||
bbc76be6 | 1695 | ret = func(mem, arg); |
e90bdb7f | 1696 | if (ret) { |
bbc76be6 WC |
1697 | kobject_put(&mem->dev.kobj); |
1698 | return ret; | |
e90bdb7f WC |
1699 | } |
1700 | } | |
1701 | ||
1702 | if (mem) | |
1703 | kobject_put(&mem->dev.kobj); | |
1704 | ||
bbc76be6 WC |
1705 | return 0; |
1706 | } | |
1707 | ||
e2ff3940 | 1708 | #ifdef CONFIG_MEMORY_HOTREMOVE |
bbc76be6 WC |
1709 | static int is_memblock_offlined_cb(struct memory_block *mem, void *arg) |
1710 | { | |
1711 | int ret = !is_memblock_offlined(mem); | |
1712 | ||
349daa0f RD |
1713 | if (unlikely(ret)) { |
1714 | phys_addr_t beginpa, endpa; | |
1715 | ||
1716 | beginpa = PFN_PHYS(section_nr_to_pfn(mem->start_section_nr)); | |
1717 | endpa = PFN_PHYS(section_nr_to_pfn(mem->end_section_nr + 1))-1; | |
bbc76be6 | 1718 | pr_warn("removing memory fails, because memory " |
349daa0f RD |
1719 | "[%pa-%pa] is onlined\n", |
1720 | &beginpa, &endpa); | |
1721 | } | |
bbc76be6 WC |
1722 | |
1723 | return ret; | |
1724 | } | |
1725 | ||
0f1cfe9d | 1726 | static int check_cpu_on_node(pg_data_t *pgdat) |
60a5a19e | 1727 | { |
60a5a19e TC |
1728 | int cpu; |
1729 | ||
1730 | for_each_present_cpu(cpu) { | |
1731 | if (cpu_to_node(cpu) == pgdat->node_id) | |
1732 | /* | |
1733 | * the cpu on this node isn't removed, and we can't | |
1734 | * offline this node. | |
1735 | */ | |
1736 | return -EBUSY; | |
1737 | } | |
1738 | ||
1739 | return 0; | |
1740 | } | |
1741 | ||
0f1cfe9d | 1742 | static void unmap_cpu_on_node(pg_data_t *pgdat) |
e13fe869 WC |
1743 | { |
1744 | #ifdef CONFIG_ACPI_NUMA | |
e13fe869 WC |
1745 | int cpu; |
1746 | ||
1747 | for_each_possible_cpu(cpu) | |
1748 | if (cpu_to_node(cpu) == pgdat->node_id) | |
1749 | numa_clear_node(cpu); | |
1750 | #endif | |
1751 | } | |
1752 | ||
0f1cfe9d | 1753 | static int check_and_unmap_cpu_on_node(pg_data_t *pgdat) |
e13fe869 | 1754 | { |
0f1cfe9d | 1755 | int ret; |
e13fe869 | 1756 | |
0f1cfe9d | 1757 | ret = check_cpu_on_node(pgdat); |
e13fe869 WC |
1758 | if (ret) |
1759 | return ret; | |
1760 | ||
1761 | /* | |
1762 | * the node will be offlined when we come here, so we can clear | |
1763 | * the cpu_to_node() now. | |
1764 | */ | |
1765 | ||
0f1cfe9d | 1766 | unmap_cpu_on_node(pgdat); |
e13fe869 WC |
1767 | return 0; |
1768 | } | |
1769 | ||
0f1cfe9d TK |
1770 | /** |
1771 | * try_offline_node | |
1772 | * | |
1773 | * Offline a node if all memory sections and cpus of the node are removed. | |
1774 | * | |
1775 | * NOTE: The caller must call lock_device_hotplug() to serialize hotplug | |
1776 | * and online/offline operations before this call. | |
1777 | */ | |
90b30cdc | 1778 | void try_offline_node(int nid) |
60a5a19e | 1779 | { |
d822b86a WC |
1780 | pg_data_t *pgdat = NODE_DATA(nid); |
1781 | unsigned long start_pfn = pgdat->node_start_pfn; | |
1782 | unsigned long end_pfn = start_pfn + pgdat->node_spanned_pages; | |
60a5a19e | 1783 | unsigned long pfn; |
d822b86a WC |
1784 | struct page *pgdat_page = virt_to_page(pgdat); |
1785 | int i; | |
60a5a19e TC |
1786 | |
1787 | for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) { | |
1788 | unsigned long section_nr = pfn_to_section_nr(pfn); | |
1789 | ||
1790 | if (!present_section_nr(section_nr)) | |
1791 | continue; | |
1792 | ||
1793 | if (pfn_to_nid(pfn) != nid) | |
1794 | continue; | |
1795 | ||
1796 | /* | |
1797 | * some memory sections of this node are not removed, and we | |
1798 | * can't offline node now. | |
1799 | */ | |
1800 | return; | |
1801 | } | |
1802 | ||
0f1cfe9d | 1803 | if (check_and_unmap_cpu_on_node(pgdat)) |
60a5a19e TC |
1804 | return; |
1805 | ||
1806 | /* | |
1807 | * all memory/cpu of this node are removed, we can offline this | |
1808 | * node now. | |
1809 | */ | |
1810 | node_set_offline(nid); | |
1811 | unregister_one_node(nid); | |
d822b86a WC |
1812 | |
1813 | if (!PageSlab(pgdat_page) && !PageCompound(pgdat_page)) | |
1814 | /* node data is allocated from boot memory */ | |
1815 | return; | |
1816 | ||
1817 | /* free waittable in each zone */ | |
1818 | for (i = 0; i < MAX_NR_ZONES; i++) { | |
1819 | struct zone *zone = pgdat->node_zones + i; | |
1820 | ||
ca4b3f30 JW |
1821 | /* |
1822 | * wait_table may be allocated from boot memory, | |
1823 | * here only free if it's allocated by vmalloc. | |
1824 | */ | |
1825 | if (is_vmalloc_addr(zone->wait_table)) | |
d822b86a WC |
1826 | vfree(zone->wait_table); |
1827 | } | |
1828 | ||
1829 | /* | |
1830 | * Since there is no way to guarentee the address of pgdat/zone is not | |
1831 | * on stack of any kernel threads or used by other kernel objects | |
1832 | * without reference counting or other symchronizing method, do not | |
1833 | * reset node_data and free pgdat here. Just reset it to 0 and reuse | |
1834 | * the memory when the node is online again. | |
1835 | */ | |
1836 | memset(pgdat, 0, sizeof(*pgdat)); | |
60a5a19e | 1837 | } |
90b30cdc | 1838 | EXPORT_SYMBOL(try_offline_node); |
60a5a19e | 1839 | |
0f1cfe9d TK |
1840 | /** |
1841 | * remove_memory | |
1842 | * | |
1843 | * NOTE: The caller must call lock_device_hotplug() to serialize hotplug | |
1844 | * and online/offline operations before this call, as required by | |
1845 | * try_offline_node(). | |
1846 | */ | |
242831eb | 1847 | void __ref remove_memory(int nid, u64 start, u64 size) |
bbc76be6 | 1848 | { |
242831eb | 1849 | int ret; |
993c1aad | 1850 | |
27356f54 TK |
1851 | BUG_ON(check_hotplug_memory_range(start, size)); |
1852 | ||
6677e3ea YI |
1853 | lock_memory_hotplug(); |
1854 | ||
1855 | /* | |
242831eb RW |
1856 | * All memory blocks must be offlined before removing memory. Check |
1857 | * whether all memory blocks in question are offline and trigger a BUG() | |
1858 | * if this is not the case. | |
6677e3ea | 1859 | */ |
242831eb | 1860 | ret = walk_memory_range(PFN_DOWN(start), PFN_UP(start + size - 1), NULL, |
bbc76be6 WC |
1861 | is_memblock_offlined_cb); |
1862 | if (ret) { | |
1863 | unlock_memory_hotplug(); | |
242831eb | 1864 | BUG(); |
6677e3ea YI |
1865 | } |
1866 | ||
46c66c4b YI |
1867 | /* remove memmap entry */ |
1868 | firmware_map_remove(start, start + size, "System RAM"); | |
1869 | ||
24d335ca WC |
1870 | arch_remove_memory(start, size); |
1871 | ||
60a5a19e TC |
1872 | try_offline_node(nid); |
1873 | ||
6677e3ea | 1874 | unlock_memory_hotplug(); |
71088785 | 1875 | } |
71088785 | 1876 | EXPORT_SYMBOL_GPL(remove_memory); |
aba6efc4 | 1877 | #endif /* CONFIG_MEMORY_HOTREMOVE */ |