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); | |
20d6c96b KM |
55 | } |
56 | ||
57 | void unlock_memory_hotplug(void) | |
58 | { | |
20d6c96b KM |
59 | mutex_unlock(&mem_hotplug_mutex); |
60 | } | |
61 | ||
62 | ||
45e0b78b KM |
63 | /* add this memory to iomem resource */ |
64 | static struct resource *register_memory_resource(u64 start, u64 size) | |
65 | { | |
66 | struct resource *res; | |
67 | res = kzalloc(sizeof(struct resource), GFP_KERNEL); | |
68 | BUG_ON(!res); | |
69 | ||
70 | res->name = "System RAM"; | |
71 | res->start = start; | |
72 | res->end = start + size - 1; | |
887c3cb1 | 73 | res->flags = IORESOURCE_MEM | IORESOURCE_BUSY; |
45e0b78b | 74 | if (request_resource(&iomem_resource, res) < 0) { |
4996eed8 | 75 | pr_debug("System RAM resource %pR cannot be added\n", res); |
45e0b78b KM |
76 | kfree(res); |
77 | res = NULL; | |
78 | } | |
79 | return res; | |
80 | } | |
81 | ||
82 | static void release_memory_resource(struct resource *res) | |
83 | { | |
84 | if (!res) | |
85 | return; | |
86 | release_resource(res); | |
87 | kfree(res); | |
88 | return; | |
89 | } | |
90 | ||
53947027 | 91 | #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE |
46723bfa YI |
92 | void get_page_bootmem(unsigned long info, struct page *page, |
93 | unsigned long type) | |
04753278 | 94 | { |
5f24ce5f | 95 | page->lru.next = (struct list_head *) type; |
04753278 YG |
96 | SetPagePrivate(page); |
97 | set_page_private(page, info); | |
98 | atomic_inc(&page->_count); | |
99 | } | |
100 | ||
170a5a7e | 101 | void put_page_bootmem(struct page *page) |
04753278 | 102 | { |
5f24ce5f | 103 | unsigned long type; |
04753278 | 104 | |
5f24ce5f AA |
105 | type = (unsigned long) page->lru.next; |
106 | BUG_ON(type < MEMORY_HOTPLUG_MIN_BOOTMEM_TYPE || | |
107 | type > MEMORY_HOTPLUG_MAX_BOOTMEM_TYPE); | |
04753278 YG |
108 | |
109 | if (atomic_dec_return(&page->_count) == 1) { | |
110 | ClearPagePrivate(page); | |
111 | set_page_private(page, 0); | |
5f24ce5f | 112 | INIT_LIST_HEAD(&page->lru); |
170a5a7e | 113 | free_reserved_page(page); |
04753278 | 114 | } |
04753278 YG |
115 | } |
116 | ||
46723bfa YI |
117 | #ifdef CONFIG_HAVE_BOOTMEM_INFO_NODE |
118 | #ifndef CONFIG_SPARSEMEM_VMEMMAP | |
d92bc318 | 119 | static void register_page_bootmem_info_section(unsigned long start_pfn) |
04753278 YG |
120 | { |
121 | unsigned long *usemap, mapsize, section_nr, i; | |
122 | struct mem_section *ms; | |
123 | struct page *page, *memmap; | |
124 | ||
04753278 YG |
125 | section_nr = pfn_to_section_nr(start_pfn); |
126 | ms = __nr_to_section(section_nr); | |
127 | ||
128 | /* Get section's memmap address */ | |
129 | memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr); | |
130 | ||
131 | /* | |
132 | * Get page for the memmap's phys address | |
133 | * XXX: need more consideration for sparse_vmemmap... | |
134 | */ | |
135 | page = virt_to_page(memmap); | |
136 | mapsize = sizeof(struct page) * PAGES_PER_SECTION; | |
137 | mapsize = PAGE_ALIGN(mapsize) >> PAGE_SHIFT; | |
138 | ||
139 | /* remember memmap's page */ | |
140 | for (i = 0; i < mapsize; i++, page++) | |
141 | get_page_bootmem(section_nr, page, SECTION_INFO); | |
142 | ||
143 | usemap = __nr_to_section(section_nr)->pageblock_flags; | |
144 | page = virt_to_page(usemap); | |
145 | ||
146 | mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT; | |
147 | ||
148 | for (i = 0; i < mapsize; i++, page++) | |
af370fb8 | 149 | get_page_bootmem(section_nr, page, MIX_SECTION_INFO); |
04753278 YG |
150 | |
151 | } | |
46723bfa YI |
152 | #else /* CONFIG_SPARSEMEM_VMEMMAP */ |
153 | static void register_page_bootmem_info_section(unsigned long start_pfn) | |
154 | { | |
155 | unsigned long *usemap, mapsize, section_nr, i; | |
156 | struct mem_section *ms; | |
157 | struct page *page, *memmap; | |
158 | ||
159 | if (!pfn_valid(start_pfn)) | |
160 | return; | |
161 | ||
162 | section_nr = pfn_to_section_nr(start_pfn); | |
163 | ms = __nr_to_section(section_nr); | |
164 | ||
165 | memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr); | |
166 | ||
167 | register_page_bootmem_memmap(section_nr, memmap, PAGES_PER_SECTION); | |
168 | ||
169 | usemap = __nr_to_section(section_nr)->pageblock_flags; | |
170 | page = virt_to_page(usemap); | |
171 | ||
172 | mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT; | |
173 | ||
174 | for (i = 0; i < mapsize; i++, page++) | |
175 | get_page_bootmem(section_nr, page, MIX_SECTION_INFO); | |
176 | } | |
177 | #endif /* !CONFIG_SPARSEMEM_VMEMMAP */ | |
04753278 YG |
178 | |
179 | void register_page_bootmem_info_node(struct pglist_data *pgdat) | |
180 | { | |
181 | unsigned long i, pfn, end_pfn, nr_pages; | |
182 | int node = pgdat->node_id; | |
183 | struct page *page; | |
184 | struct zone *zone; | |
185 | ||
186 | nr_pages = PAGE_ALIGN(sizeof(struct pglist_data)) >> PAGE_SHIFT; | |
187 | page = virt_to_page(pgdat); | |
188 | ||
189 | for (i = 0; i < nr_pages; i++, page++) | |
190 | get_page_bootmem(node, page, NODE_INFO); | |
191 | ||
192 | zone = &pgdat->node_zones[0]; | |
193 | for (; zone < pgdat->node_zones + MAX_NR_ZONES - 1; zone++) { | |
139c2d75 | 194 | if (zone_is_initialized(zone)) { |
04753278 YG |
195 | nr_pages = zone->wait_table_hash_nr_entries |
196 | * sizeof(wait_queue_head_t); | |
197 | nr_pages = PAGE_ALIGN(nr_pages) >> PAGE_SHIFT; | |
198 | page = virt_to_page(zone->wait_table); | |
199 | ||
200 | for (i = 0; i < nr_pages; i++, page++) | |
201 | get_page_bootmem(node, page, NODE_INFO); | |
202 | } | |
203 | } | |
204 | ||
205 | pfn = pgdat->node_start_pfn; | |
c1f19495 | 206 | end_pfn = pgdat_end_pfn(pgdat); |
04753278 | 207 | |
7e9f5eb0 | 208 | /* register section info */ |
f14851af | 209 | for (; pfn < end_pfn; pfn += PAGES_PER_SECTION) { |
210 | /* | |
211 | * Some platforms can assign the same pfn to multiple nodes - on | |
212 | * node0 as well as nodeN. To avoid registering a pfn against | |
213 | * multiple nodes we check that this pfn does not already | |
7e9f5eb0 | 214 | * reside in some other nodes. |
f14851af | 215 | */ |
216 | if (pfn_valid(pfn) && (pfn_to_nid(pfn) == node)) | |
217 | register_page_bootmem_info_section(pfn); | |
218 | } | |
04753278 | 219 | } |
46723bfa | 220 | #endif /* CONFIG_HAVE_BOOTMEM_INFO_NODE */ |
04753278 | 221 | |
76cdd58e HC |
222 | static void grow_zone_span(struct zone *zone, unsigned long start_pfn, |
223 | unsigned long end_pfn) | |
224 | { | |
225 | unsigned long old_zone_end_pfn; | |
226 | ||
227 | zone_span_writelock(zone); | |
228 | ||
c33bc315 | 229 | old_zone_end_pfn = zone_end_pfn(zone); |
8080fc03 | 230 | if (zone_is_empty(zone) || start_pfn < zone->zone_start_pfn) |
76cdd58e HC |
231 | zone->zone_start_pfn = start_pfn; |
232 | ||
233 | zone->spanned_pages = max(old_zone_end_pfn, end_pfn) - | |
234 | zone->zone_start_pfn; | |
235 | ||
236 | zone_span_writeunlock(zone); | |
237 | } | |
238 | ||
511c2aba LJ |
239 | static void resize_zone(struct zone *zone, unsigned long start_pfn, |
240 | unsigned long end_pfn) | |
241 | { | |
242 | zone_span_writelock(zone); | |
243 | ||
e455a9b9 LJ |
244 | if (end_pfn - start_pfn) { |
245 | zone->zone_start_pfn = start_pfn; | |
246 | zone->spanned_pages = end_pfn - start_pfn; | |
247 | } else { | |
248 | /* | |
249 | * make it consist as free_area_init_core(), | |
250 | * if spanned_pages = 0, then keep start_pfn = 0 | |
251 | */ | |
252 | zone->zone_start_pfn = 0; | |
253 | zone->spanned_pages = 0; | |
254 | } | |
511c2aba LJ |
255 | |
256 | zone_span_writeunlock(zone); | |
257 | } | |
258 | ||
259 | static void fix_zone_id(struct zone *zone, unsigned long start_pfn, | |
260 | unsigned long end_pfn) | |
261 | { | |
262 | enum zone_type zid = zone_idx(zone); | |
263 | int nid = zone->zone_pgdat->node_id; | |
264 | unsigned long pfn; | |
265 | ||
266 | for (pfn = start_pfn; pfn < end_pfn; pfn++) | |
267 | set_page_links(pfn_to_page(pfn), zid, nid, pfn); | |
268 | } | |
269 | ||
f6bbb78e CS |
270 | /* Can fail with -ENOMEM from allocating a wait table with vmalloc() or |
271 | * alloc_bootmem_node_nopanic() */ | |
272 | static int __ref ensure_zone_is_initialized(struct zone *zone, | |
273 | unsigned long start_pfn, unsigned long num_pages) | |
274 | { | |
275 | if (!zone_is_initialized(zone)) | |
276 | return init_currently_empty_zone(zone, start_pfn, num_pages, | |
277 | MEMMAP_HOTPLUG); | |
278 | return 0; | |
279 | } | |
280 | ||
e455a9b9 | 281 | static int __meminit move_pfn_range_left(struct zone *z1, struct zone *z2, |
511c2aba LJ |
282 | unsigned long start_pfn, unsigned long end_pfn) |
283 | { | |
e455a9b9 | 284 | int ret; |
511c2aba | 285 | unsigned long flags; |
e455a9b9 LJ |
286 | unsigned long z1_start_pfn; |
287 | ||
64dd1b29 CS |
288 | ret = ensure_zone_is_initialized(z1, start_pfn, end_pfn - start_pfn); |
289 | if (ret) | |
290 | return ret; | |
511c2aba LJ |
291 | |
292 | pgdat_resize_lock(z1->zone_pgdat, &flags); | |
293 | ||
294 | /* can't move pfns which are higher than @z2 */ | |
108bcc96 | 295 | if (end_pfn > zone_end_pfn(z2)) |
511c2aba | 296 | goto out_fail; |
834405c3 | 297 | /* the move out part must be at the left most of @z2 */ |
511c2aba LJ |
298 | if (start_pfn > z2->zone_start_pfn) |
299 | goto out_fail; | |
300 | /* must included/overlap */ | |
301 | if (end_pfn <= z2->zone_start_pfn) | |
302 | goto out_fail; | |
303 | ||
e455a9b9 | 304 | /* use start_pfn for z1's start_pfn if z1 is empty */ |
8080fc03 | 305 | if (!zone_is_empty(z1)) |
e455a9b9 LJ |
306 | z1_start_pfn = z1->zone_start_pfn; |
307 | else | |
308 | z1_start_pfn = start_pfn; | |
309 | ||
310 | resize_zone(z1, z1_start_pfn, end_pfn); | |
108bcc96 | 311 | resize_zone(z2, end_pfn, zone_end_pfn(z2)); |
511c2aba LJ |
312 | |
313 | pgdat_resize_unlock(z1->zone_pgdat, &flags); | |
314 | ||
315 | fix_zone_id(z1, start_pfn, end_pfn); | |
316 | ||
317 | return 0; | |
318 | out_fail: | |
319 | pgdat_resize_unlock(z1->zone_pgdat, &flags); | |
320 | return -1; | |
321 | } | |
322 | ||
e455a9b9 | 323 | static int __meminit move_pfn_range_right(struct zone *z1, struct zone *z2, |
511c2aba LJ |
324 | unsigned long start_pfn, unsigned long end_pfn) |
325 | { | |
e455a9b9 | 326 | int ret; |
511c2aba | 327 | unsigned long flags; |
e455a9b9 LJ |
328 | unsigned long z2_end_pfn; |
329 | ||
64dd1b29 CS |
330 | ret = ensure_zone_is_initialized(z2, start_pfn, end_pfn - start_pfn); |
331 | if (ret) | |
332 | return ret; | |
511c2aba LJ |
333 | |
334 | pgdat_resize_lock(z1->zone_pgdat, &flags); | |
335 | ||
336 | /* can't move pfns which are lower than @z1 */ | |
337 | if (z1->zone_start_pfn > start_pfn) | |
338 | goto out_fail; | |
339 | /* the move out part mast at the right most of @z1 */ | |
108bcc96 | 340 | if (zone_end_pfn(z1) > end_pfn) |
511c2aba LJ |
341 | goto out_fail; |
342 | /* must included/overlap */ | |
108bcc96 | 343 | if (start_pfn >= zone_end_pfn(z1)) |
511c2aba LJ |
344 | goto out_fail; |
345 | ||
e455a9b9 | 346 | /* use end_pfn for z2's end_pfn if z2 is empty */ |
8080fc03 | 347 | if (!zone_is_empty(z2)) |
108bcc96 | 348 | z2_end_pfn = zone_end_pfn(z2); |
e455a9b9 LJ |
349 | else |
350 | z2_end_pfn = end_pfn; | |
351 | ||
511c2aba | 352 | resize_zone(z1, z1->zone_start_pfn, start_pfn); |
e455a9b9 | 353 | resize_zone(z2, start_pfn, z2_end_pfn); |
511c2aba LJ |
354 | |
355 | pgdat_resize_unlock(z1->zone_pgdat, &flags); | |
356 | ||
357 | fix_zone_id(z2, start_pfn, end_pfn); | |
358 | ||
359 | return 0; | |
360 | out_fail: | |
361 | pgdat_resize_unlock(z1->zone_pgdat, &flags); | |
362 | return -1; | |
363 | } | |
364 | ||
76cdd58e HC |
365 | static void grow_pgdat_span(struct pglist_data *pgdat, unsigned long start_pfn, |
366 | unsigned long end_pfn) | |
367 | { | |
83285c72 | 368 | unsigned long old_pgdat_end_pfn = pgdat_end_pfn(pgdat); |
76cdd58e | 369 | |
712cd386 | 370 | if (!pgdat->node_spanned_pages || start_pfn < pgdat->node_start_pfn) |
76cdd58e HC |
371 | pgdat->node_start_pfn = start_pfn; |
372 | ||
373 | pgdat->node_spanned_pages = max(old_pgdat_end_pfn, end_pfn) - | |
374 | pgdat->node_start_pfn; | |
375 | } | |
376 | ||
31168481 | 377 | static int __meminit __add_zone(struct zone *zone, unsigned long phys_start_pfn) |
3947be19 DH |
378 | { |
379 | struct pglist_data *pgdat = zone->zone_pgdat; | |
380 | int nr_pages = PAGES_PER_SECTION; | |
381 | int nid = pgdat->node_id; | |
382 | int zone_type; | |
76cdd58e | 383 | unsigned long flags; |
64dd1b29 | 384 | int ret; |
3947be19 DH |
385 | |
386 | zone_type = zone - pgdat->node_zones; | |
64dd1b29 CS |
387 | ret = ensure_zone_is_initialized(zone, phys_start_pfn, nr_pages); |
388 | if (ret) | |
389 | return ret; | |
76cdd58e | 390 | |
76cdd58e HC |
391 | pgdat_resize_lock(zone->zone_pgdat, &flags); |
392 | grow_zone_span(zone, phys_start_pfn, phys_start_pfn + nr_pages); | |
393 | grow_pgdat_span(zone->zone_pgdat, phys_start_pfn, | |
394 | phys_start_pfn + nr_pages); | |
395 | pgdat_resize_unlock(zone->zone_pgdat, &flags); | |
a2f3aa02 DH |
396 | memmap_init_zone(nr_pages, nid, zone_type, |
397 | phys_start_pfn, MEMMAP_HOTPLUG); | |
718127cc | 398 | return 0; |
3947be19 DH |
399 | } |
400 | ||
c04fc586 GH |
401 | static int __meminit __add_section(int nid, struct zone *zone, |
402 | unsigned long phys_start_pfn) | |
3947be19 | 403 | { |
3947be19 | 404 | int nr_pages = PAGES_PER_SECTION; |
3947be19 DH |
405 | int ret; |
406 | ||
ebd15302 KH |
407 | if (pfn_valid(phys_start_pfn)) |
408 | return -EEXIST; | |
409 | ||
0b0acbec | 410 | ret = sparse_add_one_section(zone, phys_start_pfn, nr_pages); |
3947be19 DH |
411 | |
412 | if (ret < 0) | |
413 | return ret; | |
414 | ||
718127cc YG |
415 | ret = __add_zone(zone, phys_start_pfn); |
416 | ||
417 | if (ret < 0) | |
418 | return ret; | |
419 | ||
c04fc586 | 420 | return register_new_memory(nid, __pfn_to_section(phys_start_pfn)); |
3947be19 DH |
421 | } |
422 | ||
4edd7cef DR |
423 | /* |
424 | * Reasonably generic function for adding memory. It is | |
425 | * expected that archs that support memory hotplug will | |
426 | * call this function after deciding the zone to which to | |
427 | * add the new pages. | |
428 | */ | |
429 | int __ref __add_pages(int nid, struct zone *zone, unsigned long phys_start_pfn, | |
430 | unsigned long nr_pages) | |
431 | { | |
432 | unsigned long i; | |
433 | int err = 0; | |
434 | int start_sec, end_sec; | |
435 | /* during initialize mem_map, align hot-added range to section */ | |
436 | start_sec = pfn_to_section_nr(phys_start_pfn); | |
437 | end_sec = pfn_to_section_nr(phys_start_pfn + nr_pages - 1); | |
438 | ||
439 | for (i = start_sec; i <= end_sec; i++) { | |
440 | err = __add_section(nid, zone, i << PFN_SECTION_SHIFT); | |
441 | ||
442 | /* | |
443 | * EEXIST is finally dealt with by ioresource collision | |
444 | * check. see add_memory() => register_memory_resource() | |
445 | * Warning will be printed if there is collision. | |
446 | */ | |
447 | if (err && (err != -EEXIST)) | |
448 | break; | |
449 | err = 0; | |
450 | } | |
451 | ||
452 | return err; | |
453 | } | |
454 | EXPORT_SYMBOL_GPL(__add_pages); | |
455 | ||
456 | #ifdef CONFIG_MEMORY_HOTREMOVE | |
815121d2 YI |
457 | /* find the smallest valid pfn in the range [start_pfn, end_pfn) */ |
458 | static int find_smallest_section_pfn(int nid, struct zone *zone, | |
459 | unsigned long start_pfn, | |
460 | unsigned long end_pfn) | |
461 | { | |
462 | struct mem_section *ms; | |
463 | ||
464 | for (; start_pfn < end_pfn; start_pfn += PAGES_PER_SECTION) { | |
465 | ms = __pfn_to_section(start_pfn); | |
466 | ||
467 | if (unlikely(!valid_section(ms))) | |
468 | continue; | |
469 | ||
470 | if (unlikely(pfn_to_nid(start_pfn) != nid)) | |
471 | continue; | |
472 | ||
473 | if (zone && zone != page_zone(pfn_to_page(start_pfn))) | |
474 | continue; | |
475 | ||
476 | return start_pfn; | |
477 | } | |
478 | ||
479 | return 0; | |
480 | } | |
481 | ||
482 | /* find the biggest valid pfn in the range [start_pfn, end_pfn). */ | |
483 | static int find_biggest_section_pfn(int nid, struct zone *zone, | |
484 | unsigned long start_pfn, | |
485 | unsigned long end_pfn) | |
486 | { | |
487 | struct mem_section *ms; | |
488 | unsigned long pfn; | |
489 | ||
490 | /* pfn is the end pfn of a memory section. */ | |
491 | pfn = end_pfn - 1; | |
492 | for (; pfn >= start_pfn; pfn -= PAGES_PER_SECTION) { | |
493 | ms = __pfn_to_section(pfn); | |
494 | ||
495 | if (unlikely(!valid_section(ms))) | |
496 | continue; | |
497 | ||
498 | if (unlikely(pfn_to_nid(pfn) != nid)) | |
499 | continue; | |
500 | ||
501 | if (zone && zone != page_zone(pfn_to_page(pfn))) | |
502 | continue; | |
503 | ||
504 | return pfn; | |
505 | } | |
506 | ||
507 | return 0; | |
508 | } | |
509 | ||
510 | static void shrink_zone_span(struct zone *zone, unsigned long start_pfn, | |
511 | unsigned long end_pfn) | |
512 | { | |
c33bc315 XQ |
513 | unsigned long zone_start_pfn = zone->zone_start_pfn; |
514 | unsigned long z = zone_end_pfn(zone); /* zone_end_pfn namespace clash */ | |
515 | unsigned long zone_end_pfn = z; | |
815121d2 YI |
516 | unsigned long pfn; |
517 | struct mem_section *ms; | |
518 | int nid = zone_to_nid(zone); | |
519 | ||
520 | zone_span_writelock(zone); | |
521 | if (zone_start_pfn == start_pfn) { | |
522 | /* | |
523 | * If the section is smallest section in the zone, it need | |
524 | * shrink zone->zone_start_pfn and zone->zone_spanned_pages. | |
525 | * In this case, we find second smallest valid mem_section | |
526 | * for shrinking zone. | |
527 | */ | |
528 | pfn = find_smallest_section_pfn(nid, zone, end_pfn, | |
529 | zone_end_pfn); | |
530 | if (pfn) { | |
531 | zone->zone_start_pfn = pfn; | |
532 | zone->spanned_pages = zone_end_pfn - pfn; | |
533 | } | |
534 | } else if (zone_end_pfn == end_pfn) { | |
535 | /* | |
536 | * If the section is biggest section in the zone, it need | |
537 | * shrink zone->spanned_pages. | |
538 | * In this case, we find second biggest valid mem_section for | |
539 | * shrinking zone. | |
540 | */ | |
541 | pfn = find_biggest_section_pfn(nid, zone, zone_start_pfn, | |
542 | start_pfn); | |
543 | if (pfn) | |
544 | zone->spanned_pages = pfn - zone_start_pfn + 1; | |
545 | } | |
546 | ||
547 | /* | |
548 | * The section is not biggest or smallest mem_section in the zone, it | |
549 | * only creates a hole in the zone. So in this case, we need not | |
550 | * change the zone. But perhaps, the zone has only hole data. Thus | |
551 | * it check the zone has only hole or not. | |
552 | */ | |
553 | pfn = zone_start_pfn; | |
554 | for (; pfn < zone_end_pfn; pfn += PAGES_PER_SECTION) { | |
555 | ms = __pfn_to_section(pfn); | |
556 | ||
557 | if (unlikely(!valid_section(ms))) | |
558 | continue; | |
559 | ||
560 | if (page_zone(pfn_to_page(pfn)) != zone) | |
561 | continue; | |
562 | ||
563 | /* If the section is current section, it continues the loop */ | |
564 | if (start_pfn == pfn) | |
565 | continue; | |
566 | ||
567 | /* If we find valid section, we have nothing to do */ | |
568 | zone_span_writeunlock(zone); | |
569 | return; | |
570 | } | |
571 | ||
572 | /* The zone has no valid section */ | |
573 | zone->zone_start_pfn = 0; | |
574 | zone->spanned_pages = 0; | |
575 | zone_span_writeunlock(zone); | |
576 | } | |
577 | ||
578 | static void shrink_pgdat_span(struct pglist_data *pgdat, | |
579 | unsigned long start_pfn, unsigned long end_pfn) | |
580 | { | |
83285c72 XQ |
581 | unsigned long pgdat_start_pfn = pgdat->node_start_pfn; |
582 | unsigned long p = pgdat_end_pfn(pgdat); /* pgdat_end_pfn namespace clash */ | |
583 | unsigned long pgdat_end_pfn = p; | |
815121d2 YI |
584 | unsigned long pfn; |
585 | struct mem_section *ms; | |
586 | int nid = pgdat->node_id; | |
587 | ||
588 | if (pgdat_start_pfn == start_pfn) { | |
589 | /* | |
590 | * If the section is smallest section in the pgdat, it need | |
591 | * shrink pgdat->node_start_pfn and pgdat->node_spanned_pages. | |
592 | * In this case, we find second smallest valid mem_section | |
593 | * for shrinking zone. | |
594 | */ | |
595 | pfn = find_smallest_section_pfn(nid, NULL, end_pfn, | |
596 | pgdat_end_pfn); | |
597 | if (pfn) { | |
598 | pgdat->node_start_pfn = pfn; | |
599 | pgdat->node_spanned_pages = pgdat_end_pfn - pfn; | |
600 | } | |
601 | } else if (pgdat_end_pfn == end_pfn) { | |
602 | /* | |
603 | * If the section is biggest section in the pgdat, it need | |
604 | * shrink pgdat->node_spanned_pages. | |
605 | * In this case, we find second biggest valid mem_section for | |
606 | * shrinking zone. | |
607 | */ | |
608 | pfn = find_biggest_section_pfn(nid, NULL, pgdat_start_pfn, | |
609 | start_pfn); | |
610 | if (pfn) | |
611 | pgdat->node_spanned_pages = pfn - pgdat_start_pfn + 1; | |
612 | } | |
613 | ||
614 | /* | |
615 | * If the section is not biggest or smallest mem_section in the pgdat, | |
616 | * it only creates a hole in the pgdat. So in this case, we need not | |
617 | * change the pgdat. | |
618 | * But perhaps, the pgdat has only hole data. Thus it check the pgdat | |
619 | * has only hole or not. | |
620 | */ | |
621 | pfn = pgdat_start_pfn; | |
622 | for (; pfn < pgdat_end_pfn; pfn += PAGES_PER_SECTION) { | |
623 | ms = __pfn_to_section(pfn); | |
624 | ||
625 | if (unlikely(!valid_section(ms))) | |
626 | continue; | |
627 | ||
628 | if (pfn_to_nid(pfn) != nid) | |
629 | continue; | |
630 | ||
631 | /* If the section is current section, it continues the loop */ | |
632 | if (start_pfn == pfn) | |
633 | continue; | |
634 | ||
635 | /* If we find valid section, we have nothing to do */ | |
636 | return; | |
637 | } | |
638 | ||
639 | /* The pgdat has no valid section */ | |
640 | pgdat->node_start_pfn = 0; | |
641 | pgdat->node_spanned_pages = 0; | |
642 | } | |
643 | ||
644 | static void __remove_zone(struct zone *zone, unsigned long start_pfn) | |
645 | { | |
646 | struct pglist_data *pgdat = zone->zone_pgdat; | |
647 | int nr_pages = PAGES_PER_SECTION; | |
648 | int zone_type; | |
649 | unsigned long flags; | |
650 | ||
651 | zone_type = zone - pgdat->node_zones; | |
652 | ||
653 | pgdat_resize_lock(zone->zone_pgdat, &flags); | |
654 | shrink_zone_span(zone, start_pfn, start_pfn + nr_pages); | |
655 | shrink_pgdat_span(pgdat, start_pfn, start_pfn + nr_pages); | |
656 | pgdat_resize_unlock(zone->zone_pgdat, &flags); | |
657 | } | |
658 | ||
ea01ea93 BP |
659 | static int __remove_section(struct zone *zone, struct mem_section *ms) |
660 | { | |
815121d2 YI |
661 | unsigned long start_pfn; |
662 | int scn_nr; | |
ea01ea93 BP |
663 | int ret = -EINVAL; |
664 | ||
665 | if (!valid_section(ms)) | |
666 | return ret; | |
667 | ||
668 | ret = unregister_memory_section(ms); | |
669 | if (ret) | |
670 | return ret; | |
671 | ||
815121d2 YI |
672 | scn_nr = __section_nr(ms); |
673 | start_pfn = section_nr_to_pfn(scn_nr); | |
674 | __remove_zone(zone, start_pfn); | |
675 | ||
ea01ea93 | 676 | sparse_remove_one_section(zone, ms); |
ea01ea93 BP |
677 | return 0; |
678 | } | |
679 | ||
ea01ea93 BP |
680 | /** |
681 | * __remove_pages() - remove sections of pages from a zone | |
682 | * @zone: zone from which pages need to be removed | |
683 | * @phys_start_pfn: starting pageframe (must be aligned to start of a section) | |
684 | * @nr_pages: number of pages to remove (must be multiple of section size) | |
685 | * | |
686 | * Generic helper function to remove section mappings and sysfs entries | |
687 | * for the section of the memory we are removing. Caller needs to make | |
688 | * sure that pages are marked reserved and zones are adjust properly by | |
689 | * calling offline_pages(). | |
690 | */ | |
691 | int __remove_pages(struct zone *zone, unsigned long phys_start_pfn, | |
692 | unsigned long nr_pages) | |
693 | { | |
fe74ebb1 | 694 | unsigned long i; |
ea01ea93 | 695 | int sections_to_remove; |
fe74ebb1 TK |
696 | resource_size_t start, size; |
697 | int ret = 0; | |
ea01ea93 BP |
698 | |
699 | /* | |
700 | * We can only remove entire sections | |
701 | */ | |
702 | BUG_ON(phys_start_pfn & ~PAGE_SECTION_MASK); | |
703 | BUG_ON(nr_pages % PAGES_PER_SECTION); | |
704 | ||
fe74ebb1 TK |
705 | start = phys_start_pfn << PAGE_SHIFT; |
706 | size = nr_pages * PAGE_SIZE; | |
707 | ret = release_mem_region_adjustable(&iomem_resource, start, size); | |
348f9f05 RD |
708 | if (ret) { |
709 | resource_size_t endres = start + size - 1; | |
710 | ||
711 | pr_warn("Unable to release resource <%pa-%pa> (%d)\n", | |
712 | &start, &endres, ret); | |
713 | } | |
d760afd4 | 714 | |
ea01ea93 BP |
715 | sections_to_remove = nr_pages / PAGES_PER_SECTION; |
716 | for (i = 0; i < sections_to_remove; i++) { | |
717 | unsigned long pfn = phys_start_pfn + i*PAGES_PER_SECTION; | |
718 | ret = __remove_section(zone, __pfn_to_section(pfn)); | |
719 | if (ret) | |
720 | break; | |
721 | } | |
722 | return ret; | |
723 | } | |
724 | EXPORT_SYMBOL_GPL(__remove_pages); | |
4edd7cef | 725 | #endif /* CONFIG_MEMORY_HOTREMOVE */ |
ea01ea93 | 726 | |
9d0ad8ca DK |
727 | int set_online_page_callback(online_page_callback_t callback) |
728 | { | |
729 | int rc = -EINVAL; | |
730 | ||
731 | lock_memory_hotplug(); | |
732 | ||
733 | if (online_page_callback == generic_online_page) { | |
734 | online_page_callback = callback; | |
735 | rc = 0; | |
736 | } | |
737 | ||
738 | unlock_memory_hotplug(); | |
739 | ||
740 | return rc; | |
741 | } | |
742 | EXPORT_SYMBOL_GPL(set_online_page_callback); | |
743 | ||
744 | int restore_online_page_callback(online_page_callback_t callback) | |
745 | { | |
746 | int rc = -EINVAL; | |
747 | ||
748 | lock_memory_hotplug(); | |
749 | ||
750 | if (online_page_callback == callback) { | |
751 | online_page_callback = generic_online_page; | |
752 | rc = 0; | |
753 | } | |
754 | ||
755 | unlock_memory_hotplug(); | |
756 | ||
757 | return rc; | |
758 | } | |
759 | EXPORT_SYMBOL_GPL(restore_online_page_callback); | |
760 | ||
761 | void __online_page_set_limits(struct page *page) | |
180c06ef | 762 | { |
9d0ad8ca DK |
763 | } |
764 | EXPORT_SYMBOL_GPL(__online_page_set_limits); | |
765 | ||
766 | void __online_page_increment_counters(struct page *page) | |
767 | { | |
3dcc0571 | 768 | adjust_managed_page_count(page, 1); |
9d0ad8ca DK |
769 | } |
770 | EXPORT_SYMBOL_GPL(__online_page_increment_counters); | |
180c06ef | 771 | |
9d0ad8ca DK |
772 | void __online_page_free(struct page *page) |
773 | { | |
3dcc0571 | 774 | __free_reserved_page(page); |
180c06ef | 775 | } |
9d0ad8ca DK |
776 | EXPORT_SYMBOL_GPL(__online_page_free); |
777 | ||
778 | static void generic_online_page(struct page *page) | |
779 | { | |
780 | __online_page_set_limits(page); | |
781 | __online_page_increment_counters(page); | |
782 | __online_page_free(page); | |
783 | } | |
180c06ef | 784 | |
75884fb1 KH |
785 | static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages, |
786 | void *arg) | |
3947be19 DH |
787 | { |
788 | unsigned long i; | |
75884fb1 KH |
789 | unsigned long onlined_pages = *(unsigned long *)arg; |
790 | struct page *page; | |
791 | if (PageReserved(pfn_to_page(start_pfn))) | |
792 | for (i = 0; i < nr_pages; i++) { | |
793 | page = pfn_to_page(start_pfn + i); | |
9d0ad8ca | 794 | (*online_page_callback)(page); |
75884fb1 KH |
795 | onlined_pages++; |
796 | } | |
797 | *(unsigned long *)arg = onlined_pages; | |
798 | return 0; | |
799 | } | |
800 | ||
09285af7 | 801 | #ifdef CONFIG_MOVABLE_NODE |
79a4dcef TC |
802 | /* |
803 | * When CONFIG_MOVABLE_NODE, we permit onlining of a node which doesn't have | |
804 | * normal memory. | |
805 | */ | |
09285af7 LJ |
806 | static bool can_online_high_movable(struct zone *zone) |
807 | { | |
808 | return true; | |
809 | } | |
79a4dcef | 810 | #else /* CONFIG_MOVABLE_NODE */ |
74d42d8f LJ |
811 | /* ensure every online node has NORMAL memory */ |
812 | static bool can_online_high_movable(struct zone *zone) | |
813 | { | |
814 | return node_state(zone_to_nid(zone), N_NORMAL_MEMORY); | |
815 | } | |
79a4dcef | 816 | #endif /* CONFIG_MOVABLE_NODE */ |
74d42d8f | 817 | |
d9713679 LJ |
818 | /* check which state of node_states will be changed when online memory */ |
819 | static void node_states_check_changes_online(unsigned long nr_pages, | |
820 | struct zone *zone, struct memory_notify *arg) | |
821 | { | |
822 | int nid = zone_to_nid(zone); | |
823 | enum zone_type zone_last = ZONE_NORMAL; | |
824 | ||
825 | /* | |
6715ddf9 LJ |
826 | * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY] |
827 | * contains nodes which have zones of 0...ZONE_NORMAL, | |
828 | * set zone_last to ZONE_NORMAL. | |
d9713679 | 829 | * |
6715ddf9 LJ |
830 | * If we don't have HIGHMEM nor movable node, |
831 | * node_states[N_NORMAL_MEMORY] contains nodes which have zones of | |
832 | * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE. | |
d9713679 | 833 | */ |
6715ddf9 | 834 | if (N_MEMORY == N_NORMAL_MEMORY) |
d9713679 LJ |
835 | zone_last = ZONE_MOVABLE; |
836 | ||
837 | /* | |
838 | * if the memory to be online is in a zone of 0...zone_last, and | |
839 | * the zones of 0...zone_last don't have memory before online, we will | |
840 | * need to set the node to node_states[N_NORMAL_MEMORY] after | |
841 | * the memory is online. | |
842 | */ | |
843 | if (zone_idx(zone) <= zone_last && !node_state(nid, N_NORMAL_MEMORY)) | |
844 | arg->status_change_nid_normal = nid; | |
845 | else | |
846 | arg->status_change_nid_normal = -1; | |
847 | ||
6715ddf9 LJ |
848 | #ifdef CONFIG_HIGHMEM |
849 | /* | |
850 | * If we have movable node, node_states[N_HIGH_MEMORY] | |
851 | * contains nodes which have zones of 0...ZONE_HIGHMEM, | |
852 | * set zone_last to ZONE_HIGHMEM. | |
853 | * | |
854 | * If we don't have movable node, node_states[N_NORMAL_MEMORY] | |
855 | * contains nodes which have zones of 0...ZONE_MOVABLE, | |
856 | * set zone_last to ZONE_MOVABLE. | |
857 | */ | |
858 | zone_last = ZONE_HIGHMEM; | |
859 | if (N_MEMORY == N_HIGH_MEMORY) | |
860 | zone_last = ZONE_MOVABLE; | |
861 | ||
862 | if (zone_idx(zone) <= zone_last && !node_state(nid, N_HIGH_MEMORY)) | |
863 | arg->status_change_nid_high = nid; | |
864 | else | |
865 | arg->status_change_nid_high = -1; | |
866 | #else | |
867 | arg->status_change_nid_high = arg->status_change_nid_normal; | |
868 | #endif | |
869 | ||
d9713679 LJ |
870 | /* |
871 | * if the node don't have memory befor online, we will need to | |
6715ddf9 | 872 | * set the node to node_states[N_MEMORY] after the memory |
d9713679 LJ |
873 | * is online. |
874 | */ | |
6715ddf9 | 875 | if (!node_state(nid, N_MEMORY)) |
d9713679 LJ |
876 | arg->status_change_nid = nid; |
877 | else | |
878 | arg->status_change_nid = -1; | |
879 | } | |
880 | ||
881 | static void node_states_set_node(int node, struct memory_notify *arg) | |
882 | { | |
883 | if (arg->status_change_nid_normal >= 0) | |
884 | node_set_state(node, N_NORMAL_MEMORY); | |
885 | ||
6715ddf9 LJ |
886 | if (arg->status_change_nid_high >= 0) |
887 | node_set_state(node, N_HIGH_MEMORY); | |
888 | ||
889 | node_set_state(node, N_MEMORY); | |
d9713679 LJ |
890 | } |
891 | ||
75884fb1 | 892 | |
511c2aba | 893 | int __ref online_pages(unsigned long pfn, unsigned long nr_pages, int online_type) |
75884fb1 | 894 | { |
aa47228a | 895 | unsigned long flags; |
3947be19 DH |
896 | unsigned long onlined_pages = 0; |
897 | struct zone *zone; | |
6811378e | 898 | int need_zonelists_rebuild = 0; |
7b78d335 YG |
899 | int nid; |
900 | int ret; | |
901 | struct memory_notify arg; | |
902 | ||
925268a0 | 903 | lock_memory_hotplug(); |
d9713679 LJ |
904 | /* |
905 | * This doesn't need a lock to do pfn_to_page(). | |
906 | * The section can't be removed here because of the | |
907 | * memory_block->state_mutex. | |
908 | */ | |
909 | zone = page_zone(pfn_to_page(pfn)); | |
910 | ||
74d42d8f LJ |
911 | if ((zone_idx(zone) > ZONE_NORMAL || online_type == ONLINE_MOVABLE) && |
912 | !can_online_high_movable(zone)) { | |
913 | unlock_memory_hotplug(); | |
0a1be150 | 914 | return -EINVAL; |
74d42d8f LJ |
915 | } |
916 | ||
511c2aba LJ |
917 | if (online_type == ONLINE_KERNEL && zone_idx(zone) == ZONE_MOVABLE) { |
918 | if (move_pfn_range_left(zone - 1, zone, pfn, pfn + nr_pages)) { | |
919 | unlock_memory_hotplug(); | |
0a1be150 | 920 | return -EINVAL; |
511c2aba LJ |
921 | } |
922 | } | |
923 | if (online_type == ONLINE_MOVABLE && zone_idx(zone) == ZONE_MOVABLE - 1) { | |
924 | if (move_pfn_range_right(zone, zone + 1, pfn, pfn + nr_pages)) { | |
925 | unlock_memory_hotplug(); | |
0a1be150 | 926 | return -EINVAL; |
511c2aba LJ |
927 | } |
928 | } | |
929 | ||
930 | /* Previous code may changed the zone of the pfn range */ | |
931 | zone = page_zone(pfn_to_page(pfn)); | |
932 | ||
7b78d335 YG |
933 | arg.start_pfn = pfn; |
934 | arg.nr_pages = nr_pages; | |
d9713679 | 935 | node_states_check_changes_online(nr_pages, zone, &arg); |
7b78d335 YG |
936 | |
937 | nid = page_to_nid(pfn_to_page(pfn)); | |
3947be19 | 938 | |
7b78d335 YG |
939 | ret = memory_notify(MEM_GOING_ONLINE, &arg); |
940 | ret = notifier_to_errno(ret); | |
941 | if (ret) { | |
942 | memory_notify(MEM_CANCEL_ONLINE, &arg); | |
925268a0 | 943 | unlock_memory_hotplug(); |
7b78d335 YG |
944 | return ret; |
945 | } | |
6811378e YG |
946 | /* |
947 | * If this zone is not populated, then it is not in zonelist. | |
948 | * This means the page allocator ignores this zone. | |
949 | * So, zonelist must be updated after online. | |
950 | */ | |
4eaf3f64 | 951 | mutex_lock(&zonelists_mutex); |
6dcd73d7 | 952 | if (!populated_zone(zone)) { |
6811378e | 953 | need_zonelists_rebuild = 1; |
6dcd73d7 WC |
954 | build_all_zonelists(NULL, zone); |
955 | } | |
6811378e | 956 | |
908eedc6 | 957 | ret = walk_system_ram_range(pfn, nr_pages, &onlined_pages, |
75884fb1 | 958 | online_pages_range); |
fd8a4221 | 959 | if (ret) { |
6dcd73d7 WC |
960 | if (need_zonelists_rebuild) |
961 | zone_pcp_reset(zone); | |
4eaf3f64 | 962 | mutex_unlock(&zonelists_mutex); |
a62e2f4f BH |
963 | printk(KERN_DEBUG "online_pages [mem %#010llx-%#010llx] failed\n", |
964 | (unsigned long long) pfn << PAGE_SHIFT, | |
965 | (((unsigned long long) pfn + nr_pages) | |
966 | << PAGE_SHIFT) - 1); | |
fd8a4221 | 967 | memory_notify(MEM_CANCEL_ONLINE, &arg); |
925268a0 | 968 | unlock_memory_hotplug(); |
fd8a4221 GL |
969 | return ret; |
970 | } | |
971 | ||
3947be19 | 972 | zone->present_pages += onlined_pages; |
aa47228a CS |
973 | |
974 | pgdat_resize_lock(zone->zone_pgdat, &flags); | |
f2937be5 | 975 | zone->zone_pgdat->node_present_pages += onlined_pages; |
aa47228a CS |
976 | pgdat_resize_unlock(zone->zone_pgdat, &flags); |
977 | ||
08dff7b7 | 978 | if (onlined_pages) { |
d9713679 | 979 | node_states_set_node(zone_to_nid(zone), &arg); |
08dff7b7 | 980 | if (need_zonelists_rebuild) |
6dcd73d7 | 981 | build_all_zonelists(NULL, NULL); |
08dff7b7 JL |
982 | else |
983 | zone_pcp_update(zone); | |
984 | } | |
3947be19 | 985 | |
4eaf3f64 | 986 | mutex_unlock(&zonelists_mutex); |
1b79acc9 KM |
987 | |
988 | init_per_zone_wmark_min(); | |
989 | ||
08dff7b7 | 990 | if (onlined_pages) |
7ea1530a | 991 | kswapd_run(zone_to_nid(zone)); |
61b13993 | 992 | |
1f522509 | 993 | vm_total_pages = nr_free_pagecache_pages(); |
2f7f24ec | 994 | |
2d1d43f6 | 995 | writeback_set_ratelimit(); |
7b78d335 YG |
996 | |
997 | if (onlined_pages) | |
998 | memory_notify(MEM_ONLINE, &arg); | |
925268a0 | 999 | unlock_memory_hotplug(); |
7b78d335 | 1000 | |
3947be19 DH |
1001 | return 0; |
1002 | } | |
53947027 | 1003 | #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */ |
bc02af93 | 1004 | |
e1319331 HS |
1005 | /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */ |
1006 | static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start) | |
9af3c2de YG |
1007 | { |
1008 | struct pglist_data *pgdat; | |
1009 | unsigned long zones_size[MAX_NR_ZONES] = {0}; | |
1010 | unsigned long zholes_size[MAX_NR_ZONES] = {0}; | |
1011 | unsigned long start_pfn = start >> PAGE_SHIFT; | |
1012 | ||
a1e565aa TC |
1013 | pgdat = NODE_DATA(nid); |
1014 | if (!pgdat) { | |
1015 | pgdat = arch_alloc_nodedata(nid); | |
1016 | if (!pgdat) | |
1017 | return NULL; | |
9af3c2de | 1018 | |
a1e565aa TC |
1019 | arch_refresh_nodedata(nid, pgdat); |
1020 | } | |
9af3c2de YG |
1021 | |
1022 | /* we can use NODE_DATA(nid) from here */ | |
1023 | ||
1024 | /* init node's zones as empty zones, we don't have any present pages.*/ | |
9109fb7b | 1025 | free_area_init_node(nid, zones_size, start_pfn, zholes_size); |
9af3c2de | 1026 | |
959ecc48 KH |
1027 | /* |
1028 | * The node we allocated has no zone fallback lists. For avoiding | |
1029 | * to access not-initialized zonelist, build here. | |
1030 | */ | |
f957db4f | 1031 | mutex_lock(&zonelists_mutex); |
9adb62a5 | 1032 | build_all_zonelists(pgdat, NULL); |
f957db4f | 1033 | mutex_unlock(&zonelists_mutex); |
959ecc48 | 1034 | |
9af3c2de YG |
1035 | return pgdat; |
1036 | } | |
1037 | ||
1038 | static void rollback_node_hotadd(int nid, pg_data_t *pgdat) | |
1039 | { | |
1040 | arch_refresh_nodedata(nid, NULL); | |
1041 | arch_free_nodedata(pgdat); | |
1042 | return; | |
1043 | } | |
1044 | ||
0a547039 | 1045 | |
cf23422b | 1046 | /* |
1047 | * called by cpu_up() to online a node without onlined memory. | |
1048 | */ | |
1049 | int mem_online_node(int nid) | |
1050 | { | |
1051 | pg_data_t *pgdat; | |
1052 | int ret; | |
1053 | ||
20d6c96b | 1054 | lock_memory_hotplug(); |
cf23422b | 1055 | pgdat = hotadd_new_pgdat(nid, 0); |
7553e8f2 | 1056 | if (!pgdat) { |
cf23422b | 1057 | ret = -ENOMEM; |
1058 | goto out; | |
1059 | } | |
1060 | node_set_online(nid); | |
1061 | ret = register_one_node(nid); | |
1062 | BUG_ON(ret); | |
1063 | ||
1064 | out: | |
20d6c96b | 1065 | unlock_memory_hotplug(); |
cf23422b | 1066 | return ret; |
1067 | } | |
1068 | ||
27356f54 TK |
1069 | static int check_hotplug_memory_range(u64 start, u64 size) |
1070 | { | |
1071 | u64 start_pfn = start >> PAGE_SHIFT; | |
1072 | u64 nr_pages = size >> PAGE_SHIFT; | |
1073 | ||
1074 | /* Memory range must be aligned with section */ | |
1075 | if ((start_pfn & ~PAGE_SECTION_MASK) || | |
1076 | (nr_pages % PAGES_PER_SECTION) || (!nr_pages)) { | |
1077 | pr_err("Section-unaligned hotplug range: start 0x%llx, size 0x%llx\n", | |
1078 | (unsigned long long)start, | |
1079 | (unsigned long long)size); | |
1080 | return -EINVAL; | |
1081 | } | |
1082 | ||
1083 | return 0; | |
1084 | } | |
1085 | ||
31168481 AV |
1086 | /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */ |
1087 | int __ref add_memory(int nid, u64 start, u64 size) | |
bc02af93 | 1088 | { |
9af3c2de | 1089 | pg_data_t *pgdat = NULL; |
a1e565aa TC |
1090 | bool new_pgdat; |
1091 | bool new_node; | |
ebd15302 | 1092 | struct resource *res; |
bc02af93 YG |
1093 | int ret; |
1094 | ||
27356f54 TK |
1095 | ret = check_hotplug_memory_range(start, size); |
1096 | if (ret) | |
1097 | return ret; | |
1098 | ||
20d6c96b | 1099 | lock_memory_hotplug(); |
6ad696d2 | 1100 | |
ebd15302 | 1101 | res = register_memory_resource(start, size); |
6ad696d2 | 1102 | ret = -EEXIST; |
ebd15302 | 1103 | if (!res) |
6ad696d2 | 1104 | goto out; |
ebd15302 | 1105 | |
a1e565aa TC |
1106 | { /* Stupid hack to suppress address-never-null warning */ |
1107 | void *p = NODE_DATA(nid); | |
1108 | new_pgdat = !p; | |
1109 | } | |
1110 | new_node = !node_online(nid); | |
1111 | if (new_node) { | |
9af3c2de | 1112 | pgdat = hotadd_new_pgdat(nid, start); |
6ad696d2 | 1113 | ret = -ENOMEM; |
9af3c2de | 1114 | if (!pgdat) |
41b9e2d7 | 1115 | goto error; |
9af3c2de YG |
1116 | } |
1117 | ||
bc02af93 YG |
1118 | /* call arch's memory hotadd */ |
1119 | ret = arch_add_memory(nid, start, size); | |
1120 | ||
9af3c2de YG |
1121 | if (ret < 0) |
1122 | goto error; | |
1123 | ||
0fc44159 | 1124 | /* we online node here. we can't roll back from here. */ |
9af3c2de YG |
1125 | node_set_online(nid); |
1126 | ||
a1e565aa | 1127 | if (new_node) { |
0fc44159 YG |
1128 | ret = register_one_node(nid); |
1129 | /* | |
1130 | * If sysfs file of new node can't create, cpu on the node | |
1131 | * can't be hot-added. There is no rollback way now. | |
1132 | * So, check by BUG_ON() to catch it reluctantly.. | |
1133 | */ | |
1134 | BUG_ON(ret); | |
1135 | } | |
1136 | ||
d96ae530 | 1137 | /* create new memmap entry */ |
1138 | firmware_map_add_hotplug(start, start + size, "System RAM"); | |
1139 | ||
6ad696d2 AK |
1140 | goto out; |
1141 | ||
9af3c2de YG |
1142 | error: |
1143 | /* rollback pgdat allocation and others */ | |
1144 | if (new_pgdat) | |
1145 | rollback_node_hotadd(nid, pgdat); | |
a864b9d0 | 1146 | release_memory_resource(res); |
9af3c2de | 1147 | |
6ad696d2 | 1148 | out: |
20d6c96b | 1149 | unlock_memory_hotplug(); |
bc02af93 YG |
1150 | return ret; |
1151 | } | |
1152 | EXPORT_SYMBOL_GPL(add_memory); | |
0c0e6195 KH |
1153 | |
1154 | #ifdef CONFIG_MEMORY_HOTREMOVE | |
5c755e9f BP |
1155 | /* |
1156 | * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy | |
1157 | * set and the size of the free page is given by page_order(). Using this, | |
1158 | * the function determines if the pageblock contains only free pages. | |
1159 | * Due to buddy contraints, a free page at least the size of a pageblock will | |
1160 | * be located at the start of the pageblock | |
1161 | */ | |
1162 | static inline int pageblock_free(struct page *page) | |
1163 | { | |
1164 | return PageBuddy(page) && page_order(page) >= pageblock_order; | |
1165 | } | |
1166 | ||
1167 | /* Return the start of the next active pageblock after a given page */ | |
1168 | static struct page *next_active_pageblock(struct page *page) | |
1169 | { | |
5c755e9f BP |
1170 | /* Ensure the starting page is pageblock-aligned */ |
1171 | BUG_ON(page_to_pfn(page) & (pageblock_nr_pages - 1)); | |
1172 | ||
5c755e9f | 1173 | /* If the entire pageblock is free, move to the end of free page */ |
0dcc48c1 KH |
1174 | if (pageblock_free(page)) { |
1175 | int order; | |
1176 | /* be careful. we don't have locks, page_order can be changed.*/ | |
1177 | order = page_order(page); | |
1178 | if ((order < MAX_ORDER) && (order >= pageblock_order)) | |
1179 | return page + (1 << order); | |
1180 | } | |
5c755e9f | 1181 | |
0dcc48c1 | 1182 | return page + pageblock_nr_pages; |
5c755e9f BP |
1183 | } |
1184 | ||
1185 | /* Checks if this range of memory is likely to be hot-removable. */ | |
1186 | int is_mem_section_removable(unsigned long start_pfn, unsigned long nr_pages) | |
1187 | { | |
5c755e9f BP |
1188 | struct page *page = pfn_to_page(start_pfn); |
1189 | struct page *end_page = page + nr_pages; | |
1190 | ||
1191 | /* Check the starting page of each pageblock within the range */ | |
1192 | for (; page < end_page; page = next_active_pageblock(page)) { | |
49ac8255 | 1193 | if (!is_pageblock_removable_nolock(page)) |
5c755e9f | 1194 | return 0; |
49ac8255 | 1195 | cond_resched(); |
5c755e9f BP |
1196 | } |
1197 | ||
1198 | /* All pageblocks in the memory block are likely to be hot-removable */ | |
1199 | return 1; | |
1200 | } | |
1201 | ||
0c0e6195 KH |
1202 | /* |
1203 | * Confirm all pages in a range [start, end) is belongs to the same zone. | |
1204 | */ | |
1205 | static int test_pages_in_a_zone(unsigned long start_pfn, unsigned long end_pfn) | |
1206 | { | |
1207 | unsigned long pfn; | |
1208 | struct zone *zone = NULL; | |
1209 | struct page *page; | |
1210 | int i; | |
1211 | for (pfn = start_pfn; | |
1212 | pfn < end_pfn; | |
1213 | pfn += MAX_ORDER_NR_PAGES) { | |
1214 | i = 0; | |
1215 | /* This is just a CONFIG_HOLES_IN_ZONE check.*/ | |
1216 | while ((i < MAX_ORDER_NR_PAGES) && !pfn_valid_within(pfn + i)) | |
1217 | i++; | |
1218 | if (i == MAX_ORDER_NR_PAGES) | |
1219 | continue; | |
1220 | page = pfn_to_page(pfn + i); | |
1221 | if (zone && page_zone(page) != zone) | |
1222 | return 0; | |
1223 | zone = page_zone(page); | |
1224 | } | |
1225 | return 1; | |
1226 | } | |
1227 | ||
1228 | /* | |
c8721bbb NH |
1229 | * Scan pfn range [start,end) to find movable/migratable pages (LRU pages |
1230 | * and hugepages). We scan pfn because it's much easier than scanning over | |
1231 | * linked list. This function returns the pfn of the first found movable | |
1232 | * page if it's found, otherwise 0. | |
0c0e6195 | 1233 | */ |
c8721bbb | 1234 | static unsigned long scan_movable_pages(unsigned long start, unsigned long end) |
0c0e6195 KH |
1235 | { |
1236 | unsigned long pfn; | |
1237 | struct page *page; | |
1238 | for (pfn = start; pfn < end; pfn++) { | |
1239 | if (pfn_valid(pfn)) { | |
1240 | page = pfn_to_page(pfn); | |
1241 | if (PageLRU(page)) | |
1242 | return pfn; | |
c8721bbb NH |
1243 | if (PageHuge(page)) { |
1244 | if (is_hugepage_active(page)) | |
1245 | return pfn; | |
1246 | else | |
1247 | pfn = round_up(pfn + 1, | |
1248 | 1 << compound_order(page)) - 1; | |
1249 | } | |
0c0e6195 KH |
1250 | } |
1251 | } | |
1252 | return 0; | |
1253 | } | |
1254 | ||
0c0e6195 KH |
1255 | #define NR_OFFLINE_AT_ONCE_PAGES (256) |
1256 | static int | |
1257 | do_migrate_range(unsigned long start_pfn, unsigned long end_pfn) | |
1258 | { | |
1259 | unsigned long pfn; | |
1260 | struct page *page; | |
1261 | int move_pages = NR_OFFLINE_AT_ONCE_PAGES; | |
1262 | int not_managed = 0; | |
1263 | int ret = 0; | |
1264 | LIST_HEAD(source); | |
1265 | ||
1266 | for (pfn = start_pfn; pfn < end_pfn && move_pages > 0; pfn++) { | |
1267 | if (!pfn_valid(pfn)) | |
1268 | continue; | |
1269 | page = pfn_to_page(pfn); | |
c8721bbb NH |
1270 | |
1271 | if (PageHuge(page)) { | |
1272 | struct page *head = compound_head(page); | |
1273 | pfn = page_to_pfn(head) + (1<<compound_order(head)) - 1; | |
1274 | if (compound_order(head) > PFN_SECTION_SHIFT) { | |
1275 | ret = -EBUSY; | |
1276 | break; | |
1277 | } | |
1278 | if (isolate_huge_page(page, &source)) | |
1279 | move_pages -= 1 << compound_order(head); | |
1280 | continue; | |
1281 | } | |
1282 | ||
700c2a46 | 1283 | if (!get_page_unless_zero(page)) |
0c0e6195 KH |
1284 | continue; |
1285 | /* | |
1286 | * We can skip free pages. And we can only deal with pages on | |
1287 | * LRU. | |
1288 | */ | |
62695a84 | 1289 | ret = isolate_lru_page(page); |
0c0e6195 | 1290 | if (!ret) { /* Success */ |
700c2a46 | 1291 | put_page(page); |
62695a84 | 1292 | list_add_tail(&page->lru, &source); |
0c0e6195 | 1293 | move_pages--; |
6d9c285a KM |
1294 | inc_zone_page_state(page, NR_ISOLATED_ANON + |
1295 | page_is_file_cache(page)); | |
1296 | ||
0c0e6195 | 1297 | } else { |
0c0e6195 | 1298 | #ifdef CONFIG_DEBUG_VM |
718a3821 WF |
1299 | printk(KERN_ALERT "removing pfn %lx from LRU failed\n", |
1300 | pfn); | |
1301 | dump_page(page); | |
0c0e6195 | 1302 | #endif |
700c2a46 | 1303 | put_page(page); |
25985edc | 1304 | /* Because we don't have big zone->lock. we should |
809c4449 BL |
1305 | check this again here. */ |
1306 | if (page_count(page)) { | |
1307 | not_managed++; | |
f3ab2636 | 1308 | ret = -EBUSY; |
809c4449 BL |
1309 | break; |
1310 | } | |
0c0e6195 KH |
1311 | } |
1312 | } | |
f3ab2636 BL |
1313 | if (!list_empty(&source)) { |
1314 | if (not_managed) { | |
c8721bbb | 1315 | putback_movable_pages(&source); |
f3ab2636 BL |
1316 | goto out; |
1317 | } | |
74c08f98 MK |
1318 | |
1319 | /* | |
1320 | * alloc_migrate_target should be improooooved!! | |
1321 | * migrate_pages returns # of failed pages. | |
1322 | */ | |
1323 | ret = migrate_pages(&source, alloc_migrate_target, 0, | |
9c620e2b | 1324 | MIGRATE_SYNC, MR_MEMORY_HOTPLUG); |
f3ab2636 | 1325 | if (ret) |
c8721bbb | 1326 | putback_movable_pages(&source); |
0c0e6195 | 1327 | } |
0c0e6195 KH |
1328 | out: |
1329 | return ret; | |
1330 | } | |
1331 | ||
1332 | /* | |
1333 | * remove from free_area[] and mark all as Reserved. | |
1334 | */ | |
1335 | static int | |
1336 | offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages, | |
1337 | void *data) | |
1338 | { | |
1339 | __offline_isolated_pages(start, start + nr_pages); | |
1340 | return 0; | |
1341 | } | |
1342 | ||
1343 | static void | |
1344 | offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn) | |
1345 | { | |
908eedc6 | 1346 | walk_system_ram_range(start_pfn, end_pfn - start_pfn, NULL, |
0c0e6195 KH |
1347 | offline_isolated_pages_cb); |
1348 | } | |
1349 | ||
1350 | /* | |
1351 | * Check all pages in range, recoreded as memory resource, are isolated. | |
1352 | */ | |
1353 | static int | |
1354 | check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages, | |
1355 | void *data) | |
1356 | { | |
1357 | int ret; | |
1358 | long offlined = *(long *)data; | |
b023f468 | 1359 | ret = test_pages_isolated(start_pfn, start_pfn + nr_pages, true); |
0c0e6195 KH |
1360 | offlined = nr_pages; |
1361 | if (!ret) | |
1362 | *(long *)data += offlined; | |
1363 | return ret; | |
1364 | } | |
1365 | ||
1366 | static long | |
1367 | check_pages_isolated(unsigned long start_pfn, unsigned long end_pfn) | |
1368 | { | |
1369 | long offlined = 0; | |
1370 | int ret; | |
1371 | ||
908eedc6 | 1372 | ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn, &offlined, |
0c0e6195 KH |
1373 | check_pages_isolated_cb); |
1374 | if (ret < 0) | |
1375 | offlined = (long)ret; | |
1376 | return offlined; | |
1377 | } | |
1378 | ||
09285af7 | 1379 | #ifdef CONFIG_MOVABLE_NODE |
79a4dcef TC |
1380 | /* |
1381 | * When CONFIG_MOVABLE_NODE, we permit offlining of a node which doesn't have | |
1382 | * normal memory. | |
1383 | */ | |
09285af7 LJ |
1384 | static bool can_offline_normal(struct zone *zone, unsigned long nr_pages) |
1385 | { | |
1386 | return true; | |
1387 | } | |
79a4dcef | 1388 | #else /* CONFIG_MOVABLE_NODE */ |
74d42d8f LJ |
1389 | /* ensure the node has NORMAL memory if it is still online */ |
1390 | static bool can_offline_normal(struct zone *zone, unsigned long nr_pages) | |
1391 | { | |
1392 | struct pglist_data *pgdat = zone->zone_pgdat; | |
1393 | unsigned long present_pages = 0; | |
1394 | enum zone_type zt; | |
1395 | ||
1396 | for (zt = 0; zt <= ZONE_NORMAL; zt++) | |
1397 | present_pages += pgdat->node_zones[zt].present_pages; | |
1398 | ||
1399 | if (present_pages > nr_pages) | |
1400 | return true; | |
1401 | ||
1402 | present_pages = 0; | |
1403 | for (; zt <= ZONE_MOVABLE; zt++) | |
1404 | present_pages += pgdat->node_zones[zt].present_pages; | |
1405 | ||
1406 | /* | |
1407 | * we can't offline the last normal memory until all | |
1408 | * higher memory is offlined. | |
1409 | */ | |
1410 | return present_pages == 0; | |
1411 | } | |
79a4dcef | 1412 | #endif /* CONFIG_MOVABLE_NODE */ |
74d42d8f | 1413 | |
d9713679 LJ |
1414 | /* check which state of node_states will be changed when offline memory */ |
1415 | static void node_states_check_changes_offline(unsigned long nr_pages, | |
1416 | struct zone *zone, struct memory_notify *arg) | |
1417 | { | |
1418 | struct pglist_data *pgdat = zone->zone_pgdat; | |
1419 | unsigned long present_pages = 0; | |
1420 | enum zone_type zt, zone_last = ZONE_NORMAL; | |
1421 | ||
1422 | /* | |
6715ddf9 LJ |
1423 | * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY] |
1424 | * contains nodes which have zones of 0...ZONE_NORMAL, | |
1425 | * set zone_last to ZONE_NORMAL. | |
d9713679 | 1426 | * |
6715ddf9 LJ |
1427 | * If we don't have HIGHMEM nor movable node, |
1428 | * node_states[N_NORMAL_MEMORY] contains nodes which have zones of | |
1429 | * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE. | |
d9713679 | 1430 | */ |
6715ddf9 | 1431 | if (N_MEMORY == N_NORMAL_MEMORY) |
d9713679 LJ |
1432 | zone_last = ZONE_MOVABLE; |
1433 | ||
1434 | /* | |
1435 | * check whether node_states[N_NORMAL_MEMORY] will be changed. | |
1436 | * If the memory to be offline is in a zone of 0...zone_last, | |
1437 | * and it is the last present memory, 0...zone_last will | |
1438 | * become empty after offline , thus we can determind we will | |
1439 | * need to clear the node from node_states[N_NORMAL_MEMORY]. | |
1440 | */ | |
1441 | for (zt = 0; zt <= zone_last; zt++) | |
1442 | present_pages += pgdat->node_zones[zt].present_pages; | |
1443 | if (zone_idx(zone) <= zone_last && nr_pages >= present_pages) | |
1444 | arg->status_change_nid_normal = zone_to_nid(zone); | |
1445 | else | |
1446 | arg->status_change_nid_normal = -1; | |
1447 | ||
6715ddf9 LJ |
1448 | #ifdef CONFIG_HIGHMEM |
1449 | /* | |
1450 | * If we have movable node, node_states[N_HIGH_MEMORY] | |
1451 | * contains nodes which have zones of 0...ZONE_HIGHMEM, | |
1452 | * set zone_last to ZONE_HIGHMEM. | |
1453 | * | |
1454 | * If we don't have movable node, node_states[N_NORMAL_MEMORY] | |
1455 | * contains nodes which have zones of 0...ZONE_MOVABLE, | |
1456 | * set zone_last to ZONE_MOVABLE. | |
1457 | */ | |
1458 | zone_last = ZONE_HIGHMEM; | |
1459 | if (N_MEMORY == N_HIGH_MEMORY) | |
1460 | zone_last = ZONE_MOVABLE; | |
1461 | ||
1462 | for (; zt <= zone_last; zt++) | |
1463 | present_pages += pgdat->node_zones[zt].present_pages; | |
1464 | if (zone_idx(zone) <= zone_last && nr_pages >= present_pages) | |
1465 | arg->status_change_nid_high = zone_to_nid(zone); | |
1466 | else | |
1467 | arg->status_change_nid_high = -1; | |
1468 | #else | |
1469 | arg->status_change_nid_high = arg->status_change_nid_normal; | |
1470 | #endif | |
1471 | ||
d9713679 LJ |
1472 | /* |
1473 | * node_states[N_HIGH_MEMORY] contains nodes which have 0...ZONE_MOVABLE | |
1474 | */ | |
1475 | zone_last = ZONE_MOVABLE; | |
1476 | ||
1477 | /* | |
1478 | * check whether node_states[N_HIGH_MEMORY] will be changed | |
1479 | * If we try to offline the last present @nr_pages from the node, | |
1480 | * we can determind we will need to clear the node from | |
1481 | * node_states[N_HIGH_MEMORY]. | |
1482 | */ | |
1483 | for (; zt <= zone_last; zt++) | |
1484 | present_pages += pgdat->node_zones[zt].present_pages; | |
1485 | if (nr_pages >= present_pages) | |
1486 | arg->status_change_nid = zone_to_nid(zone); | |
1487 | else | |
1488 | arg->status_change_nid = -1; | |
1489 | } | |
1490 | ||
1491 | static void node_states_clear_node(int node, struct memory_notify *arg) | |
1492 | { | |
1493 | if (arg->status_change_nid_normal >= 0) | |
1494 | node_clear_state(node, N_NORMAL_MEMORY); | |
1495 | ||
6715ddf9 LJ |
1496 | if ((N_MEMORY != N_NORMAL_MEMORY) && |
1497 | (arg->status_change_nid_high >= 0)) | |
d9713679 | 1498 | node_clear_state(node, N_HIGH_MEMORY); |
6715ddf9 LJ |
1499 | |
1500 | if ((N_MEMORY != N_HIGH_MEMORY) && | |
1501 | (arg->status_change_nid >= 0)) | |
1502 | node_clear_state(node, N_MEMORY); | |
d9713679 LJ |
1503 | } |
1504 | ||
a16cee10 | 1505 | static int __ref __offline_pages(unsigned long start_pfn, |
0c0e6195 KH |
1506 | unsigned long end_pfn, unsigned long timeout) |
1507 | { | |
1508 | unsigned long pfn, nr_pages, expire; | |
1509 | long offlined_pages; | |
7b78d335 | 1510 | int ret, drain, retry_max, node; |
d702909f | 1511 | unsigned long flags; |
0c0e6195 | 1512 | struct zone *zone; |
7b78d335 | 1513 | struct memory_notify arg; |
0c0e6195 | 1514 | |
0c0e6195 KH |
1515 | /* at least, alignment against pageblock is necessary */ |
1516 | if (!IS_ALIGNED(start_pfn, pageblock_nr_pages)) | |
1517 | return -EINVAL; | |
1518 | if (!IS_ALIGNED(end_pfn, pageblock_nr_pages)) | |
1519 | return -EINVAL; | |
1520 | /* This makes hotplug much easier...and readable. | |
1521 | we assume this for now. .*/ | |
1522 | if (!test_pages_in_a_zone(start_pfn, end_pfn)) | |
1523 | return -EINVAL; | |
7b78d335 | 1524 | |
20d6c96b | 1525 | lock_memory_hotplug(); |
6ad696d2 | 1526 | |
7b78d335 YG |
1527 | zone = page_zone(pfn_to_page(start_pfn)); |
1528 | node = zone_to_nid(zone); | |
1529 | nr_pages = end_pfn - start_pfn; | |
1530 | ||
74d42d8f LJ |
1531 | ret = -EINVAL; |
1532 | if (zone_idx(zone) <= ZONE_NORMAL && !can_offline_normal(zone, nr_pages)) | |
1533 | goto out; | |
1534 | ||
0c0e6195 | 1535 | /* set above range as isolated */ |
b023f468 WC |
1536 | ret = start_isolate_page_range(start_pfn, end_pfn, |
1537 | MIGRATE_MOVABLE, true); | |
0c0e6195 | 1538 | if (ret) |
6ad696d2 | 1539 | goto out; |
7b78d335 YG |
1540 | |
1541 | arg.start_pfn = start_pfn; | |
1542 | arg.nr_pages = nr_pages; | |
d9713679 | 1543 | node_states_check_changes_offline(nr_pages, zone, &arg); |
7b78d335 YG |
1544 | |
1545 | ret = memory_notify(MEM_GOING_OFFLINE, &arg); | |
1546 | ret = notifier_to_errno(ret); | |
1547 | if (ret) | |
1548 | goto failed_removal; | |
1549 | ||
0c0e6195 KH |
1550 | pfn = start_pfn; |
1551 | expire = jiffies + timeout; | |
1552 | drain = 0; | |
1553 | retry_max = 5; | |
1554 | repeat: | |
1555 | /* start memory hot removal */ | |
1556 | ret = -EAGAIN; | |
1557 | if (time_after(jiffies, expire)) | |
1558 | goto failed_removal; | |
1559 | ret = -EINTR; | |
1560 | if (signal_pending(current)) | |
1561 | goto failed_removal; | |
1562 | ret = 0; | |
1563 | if (drain) { | |
1564 | lru_add_drain_all(); | |
0c0e6195 | 1565 | cond_resched(); |
9f8f2172 | 1566 | drain_all_pages(); |
0c0e6195 KH |
1567 | } |
1568 | ||
c8721bbb NH |
1569 | pfn = scan_movable_pages(start_pfn, end_pfn); |
1570 | if (pfn) { /* We have movable pages */ | |
0c0e6195 KH |
1571 | ret = do_migrate_range(pfn, end_pfn); |
1572 | if (!ret) { | |
1573 | drain = 1; | |
1574 | goto repeat; | |
1575 | } else { | |
1576 | if (ret < 0) | |
1577 | if (--retry_max == 0) | |
1578 | goto failed_removal; | |
1579 | yield(); | |
1580 | drain = 1; | |
1581 | goto repeat; | |
1582 | } | |
1583 | } | |
b3834be5 | 1584 | /* drain all zone's lru pagevec, this is asynchronous... */ |
0c0e6195 | 1585 | lru_add_drain_all(); |
0c0e6195 | 1586 | yield(); |
b3834be5 | 1587 | /* drain pcp pages, this is synchronous. */ |
9f8f2172 | 1588 | drain_all_pages(); |
c8721bbb NH |
1589 | /* |
1590 | * dissolve free hugepages in the memory block before doing offlining | |
1591 | * actually in order to make hugetlbfs's object counting consistent. | |
1592 | */ | |
1593 | dissolve_free_huge_pages(start_pfn, end_pfn); | |
0c0e6195 KH |
1594 | /* check again */ |
1595 | offlined_pages = check_pages_isolated(start_pfn, end_pfn); | |
1596 | if (offlined_pages < 0) { | |
1597 | ret = -EBUSY; | |
1598 | goto failed_removal; | |
1599 | } | |
1600 | printk(KERN_INFO "Offlined Pages %ld\n", offlined_pages); | |
b3834be5 | 1601 | /* Ok, all of our target is isolated. |
0c0e6195 KH |
1602 | We cannot do rollback at this point. */ |
1603 | offline_isolated_pages(start_pfn, end_pfn); | |
dbc0e4ce | 1604 | /* reset pagetype flags and makes migrate type to be MOVABLE */ |
0815f3d8 | 1605 | undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE); |
0c0e6195 | 1606 | /* removal success */ |
3dcc0571 | 1607 | adjust_managed_page_count(pfn_to_page(start_pfn), -offlined_pages); |
0c0e6195 | 1608 | zone->present_pages -= offlined_pages; |
d702909f CS |
1609 | |
1610 | pgdat_resize_lock(zone->zone_pgdat, &flags); | |
0c0e6195 | 1611 | zone->zone_pgdat->node_present_pages -= offlined_pages; |
d702909f | 1612 | pgdat_resize_unlock(zone->zone_pgdat, &flags); |
7b78d335 | 1613 | |
1b79acc9 KM |
1614 | init_per_zone_wmark_min(); |
1615 | ||
1e8537ba | 1616 | if (!populated_zone(zone)) { |
340175b7 | 1617 | zone_pcp_reset(zone); |
1e8537ba XQ |
1618 | mutex_lock(&zonelists_mutex); |
1619 | build_all_zonelists(NULL, NULL); | |
1620 | mutex_unlock(&zonelists_mutex); | |
1621 | } else | |
1622 | zone_pcp_update(zone); | |
340175b7 | 1623 | |
d9713679 LJ |
1624 | node_states_clear_node(node, &arg); |
1625 | if (arg.status_change_nid >= 0) | |
8fe23e05 | 1626 | kswapd_stop(node); |
bce7394a | 1627 | |
0c0e6195 KH |
1628 | vm_total_pages = nr_free_pagecache_pages(); |
1629 | writeback_set_ratelimit(); | |
7b78d335 YG |
1630 | |
1631 | memory_notify(MEM_OFFLINE, &arg); | |
20d6c96b | 1632 | unlock_memory_hotplug(); |
0c0e6195 KH |
1633 | return 0; |
1634 | ||
1635 | failed_removal: | |
a62e2f4f BH |
1636 | printk(KERN_INFO "memory offlining [mem %#010llx-%#010llx] failed\n", |
1637 | (unsigned long long) start_pfn << PAGE_SHIFT, | |
1638 | ((unsigned long long) end_pfn << PAGE_SHIFT) - 1); | |
7b78d335 | 1639 | memory_notify(MEM_CANCEL_OFFLINE, &arg); |
0c0e6195 | 1640 | /* pushback to free area */ |
0815f3d8 | 1641 | undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE); |
7b78d335 | 1642 | |
6ad696d2 | 1643 | out: |
20d6c96b | 1644 | unlock_memory_hotplug(); |
0c0e6195 KH |
1645 | return ret; |
1646 | } | |
71088785 | 1647 | |
a16cee10 WC |
1648 | int offline_pages(unsigned long start_pfn, unsigned long nr_pages) |
1649 | { | |
1650 | return __offline_pages(start_pfn, start_pfn + nr_pages, 120 * HZ); | |
1651 | } | |
e2ff3940 | 1652 | #endif /* CONFIG_MEMORY_HOTREMOVE */ |
a16cee10 | 1653 | |
bbc76be6 WC |
1654 | /** |
1655 | * walk_memory_range - walks through all mem sections in [start_pfn, end_pfn) | |
1656 | * @start_pfn: start pfn of the memory range | |
e05c4bbf | 1657 | * @end_pfn: end pfn of the memory range |
bbc76be6 WC |
1658 | * @arg: argument passed to func |
1659 | * @func: callback for each memory section walked | |
1660 | * | |
1661 | * This function walks through all present mem sections in range | |
1662 | * [start_pfn, end_pfn) and call func on each mem section. | |
1663 | * | |
1664 | * Returns the return value of func. | |
1665 | */ | |
e2ff3940 | 1666 | int walk_memory_range(unsigned long start_pfn, unsigned long end_pfn, |
bbc76be6 | 1667 | void *arg, int (*func)(struct memory_block *, void *)) |
71088785 | 1668 | { |
e90bdb7f WC |
1669 | struct memory_block *mem = NULL; |
1670 | struct mem_section *section; | |
e90bdb7f WC |
1671 | unsigned long pfn, section_nr; |
1672 | int ret; | |
e90bdb7f WC |
1673 | |
1674 | for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) { | |
1675 | section_nr = pfn_to_section_nr(pfn); | |
1676 | if (!present_section_nr(section_nr)) | |
1677 | continue; | |
1678 | ||
1679 | section = __nr_to_section(section_nr); | |
1680 | /* same memblock? */ | |
1681 | if (mem) | |
1682 | if ((section_nr >= mem->start_section_nr) && | |
1683 | (section_nr <= mem->end_section_nr)) | |
1684 | continue; | |
1685 | ||
1686 | mem = find_memory_block_hinted(section, mem); | |
1687 | if (!mem) | |
1688 | continue; | |
1689 | ||
bbc76be6 | 1690 | ret = func(mem, arg); |
e90bdb7f | 1691 | if (ret) { |
bbc76be6 WC |
1692 | kobject_put(&mem->dev.kobj); |
1693 | return ret; | |
e90bdb7f WC |
1694 | } |
1695 | } | |
1696 | ||
1697 | if (mem) | |
1698 | kobject_put(&mem->dev.kobj); | |
1699 | ||
bbc76be6 WC |
1700 | return 0; |
1701 | } | |
1702 | ||
e2ff3940 | 1703 | #ifdef CONFIG_MEMORY_HOTREMOVE |
d6de9d53 | 1704 | static int check_memblock_offlined_cb(struct memory_block *mem, void *arg) |
bbc76be6 WC |
1705 | { |
1706 | int ret = !is_memblock_offlined(mem); | |
1707 | ||
349daa0f RD |
1708 | if (unlikely(ret)) { |
1709 | phys_addr_t beginpa, endpa; | |
1710 | ||
1711 | beginpa = PFN_PHYS(section_nr_to_pfn(mem->start_section_nr)); | |
1712 | endpa = PFN_PHYS(section_nr_to_pfn(mem->end_section_nr + 1))-1; | |
bbc76be6 | 1713 | pr_warn("removing memory fails, because memory " |
349daa0f RD |
1714 | "[%pa-%pa] is onlined\n", |
1715 | &beginpa, &endpa); | |
1716 | } | |
bbc76be6 WC |
1717 | |
1718 | return ret; | |
1719 | } | |
1720 | ||
0f1cfe9d | 1721 | static int check_cpu_on_node(pg_data_t *pgdat) |
60a5a19e | 1722 | { |
60a5a19e TC |
1723 | int cpu; |
1724 | ||
1725 | for_each_present_cpu(cpu) { | |
1726 | if (cpu_to_node(cpu) == pgdat->node_id) | |
1727 | /* | |
1728 | * the cpu on this node isn't removed, and we can't | |
1729 | * offline this node. | |
1730 | */ | |
1731 | return -EBUSY; | |
1732 | } | |
1733 | ||
1734 | return 0; | |
1735 | } | |
1736 | ||
0f1cfe9d | 1737 | static void unmap_cpu_on_node(pg_data_t *pgdat) |
e13fe869 WC |
1738 | { |
1739 | #ifdef CONFIG_ACPI_NUMA | |
e13fe869 WC |
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 | ||
0f1cfe9d | 1748 | static int check_and_unmap_cpu_on_node(pg_data_t *pgdat) |
e13fe869 | 1749 | { |
0f1cfe9d | 1750 | int ret; |
e13fe869 | 1751 | |
0f1cfe9d | 1752 | ret = check_cpu_on_node(pgdat); |
e13fe869 WC |
1753 | if (ret) |
1754 | return ret; | |
1755 | ||
1756 | /* | |
1757 | * the node will be offlined when we come here, so we can clear | |
1758 | * the cpu_to_node() now. | |
1759 | */ | |
1760 | ||
0f1cfe9d | 1761 | unmap_cpu_on_node(pgdat); |
e13fe869 WC |
1762 | return 0; |
1763 | } | |
1764 | ||
0f1cfe9d TK |
1765 | /** |
1766 | * try_offline_node | |
1767 | * | |
1768 | * Offline a node if all memory sections and cpus of the node are removed. | |
1769 | * | |
1770 | * NOTE: The caller must call lock_device_hotplug() to serialize hotplug | |
1771 | * and online/offline operations before this call. | |
1772 | */ | |
90b30cdc | 1773 | void try_offline_node(int nid) |
60a5a19e | 1774 | { |
d822b86a WC |
1775 | pg_data_t *pgdat = NODE_DATA(nid); |
1776 | unsigned long start_pfn = pgdat->node_start_pfn; | |
1777 | unsigned long end_pfn = start_pfn + pgdat->node_spanned_pages; | |
60a5a19e | 1778 | unsigned long pfn; |
d822b86a WC |
1779 | struct page *pgdat_page = virt_to_page(pgdat); |
1780 | int i; | |
60a5a19e TC |
1781 | |
1782 | for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) { | |
1783 | unsigned long section_nr = pfn_to_section_nr(pfn); | |
1784 | ||
1785 | if (!present_section_nr(section_nr)) | |
1786 | continue; | |
1787 | ||
1788 | if (pfn_to_nid(pfn) != nid) | |
1789 | continue; | |
1790 | ||
1791 | /* | |
1792 | * some memory sections of this node are not removed, and we | |
1793 | * can't offline node now. | |
1794 | */ | |
1795 | return; | |
1796 | } | |
1797 | ||
0f1cfe9d | 1798 | if (check_and_unmap_cpu_on_node(pgdat)) |
60a5a19e TC |
1799 | return; |
1800 | ||
1801 | /* | |
1802 | * all memory/cpu of this node are removed, we can offline this | |
1803 | * node now. | |
1804 | */ | |
1805 | node_set_offline(nid); | |
1806 | unregister_one_node(nid); | |
d822b86a WC |
1807 | |
1808 | if (!PageSlab(pgdat_page) && !PageCompound(pgdat_page)) | |
1809 | /* node data is allocated from boot memory */ | |
1810 | return; | |
1811 | ||
1812 | /* free waittable in each zone */ | |
1813 | for (i = 0; i < MAX_NR_ZONES; i++) { | |
1814 | struct zone *zone = pgdat->node_zones + i; | |
1815 | ||
ca4b3f30 JW |
1816 | /* |
1817 | * wait_table may be allocated from boot memory, | |
1818 | * here only free if it's allocated by vmalloc. | |
1819 | */ | |
1820 | if (is_vmalloc_addr(zone->wait_table)) | |
d822b86a WC |
1821 | vfree(zone->wait_table); |
1822 | } | |
1823 | ||
1824 | /* | |
1825 | * Since there is no way to guarentee the address of pgdat/zone is not | |
1826 | * on stack of any kernel threads or used by other kernel objects | |
1827 | * without reference counting or other symchronizing method, do not | |
1828 | * reset node_data and free pgdat here. Just reset it to 0 and reuse | |
1829 | * the memory when the node is online again. | |
1830 | */ | |
1831 | memset(pgdat, 0, sizeof(*pgdat)); | |
60a5a19e | 1832 | } |
90b30cdc | 1833 | EXPORT_SYMBOL(try_offline_node); |
60a5a19e | 1834 | |
0f1cfe9d TK |
1835 | /** |
1836 | * remove_memory | |
1837 | * | |
1838 | * NOTE: The caller must call lock_device_hotplug() to serialize hotplug | |
1839 | * and online/offline operations before this call, as required by | |
1840 | * try_offline_node(). | |
1841 | */ | |
242831eb | 1842 | void __ref remove_memory(int nid, u64 start, u64 size) |
bbc76be6 | 1843 | { |
242831eb | 1844 | int ret; |
993c1aad | 1845 | |
27356f54 TK |
1846 | BUG_ON(check_hotplug_memory_range(start, size)); |
1847 | ||
6677e3ea YI |
1848 | lock_memory_hotplug(); |
1849 | ||
1850 | /* | |
242831eb RW |
1851 | * All memory blocks must be offlined before removing memory. Check |
1852 | * whether all memory blocks in question are offline and trigger a BUG() | |
1853 | * if this is not the case. | |
6677e3ea | 1854 | */ |
242831eb | 1855 | ret = walk_memory_range(PFN_DOWN(start), PFN_UP(start + size - 1), NULL, |
d6de9d53 | 1856 | check_memblock_offlined_cb); |
bbc76be6 WC |
1857 | if (ret) { |
1858 | unlock_memory_hotplug(); | |
242831eb | 1859 | BUG(); |
6677e3ea YI |
1860 | } |
1861 | ||
46c66c4b YI |
1862 | /* remove memmap entry */ |
1863 | firmware_map_remove(start, start + size, "System RAM"); | |
1864 | ||
24d335ca WC |
1865 | arch_remove_memory(start, size); |
1866 | ||
60a5a19e TC |
1867 | try_offline_node(nid); |
1868 | ||
6677e3ea | 1869 | unlock_memory_hotplug(); |
71088785 | 1870 | } |
71088785 | 1871 | EXPORT_SYMBOL_GPL(remove_memory); |
aba6efc4 | 1872 | #endif /* CONFIG_MEMORY_HOTREMOVE */ |