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> | |
3947be19 | 12 | #include <linux/compiler.h> |
b95f1b31 | 13 | #include <linux/export.h> |
3947be19 | 14 | #include <linux/pagevec.h> |
2d1d43f6 | 15 | #include <linux/writeback.h> |
3947be19 DH |
16 | #include <linux/slab.h> |
17 | #include <linux/sysctl.h> | |
18 | #include <linux/cpu.h> | |
19 | #include <linux/memory.h> | |
4b94ffdc | 20 | #include <linux/memremap.h> |
3947be19 DH |
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> |
c5320926 | 34 | #include <linux/memblock.h> |
f784a3f1 | 35 | #include <linux/bootmem.h> |
3947be19 DH |
36 | |
37 | #include <asm/tlbflush.h> | |
38 | ||
1e5ad9a3 AB |
39 | #include "internal.h" |
40 | ||
9d0ad8ca DK |
41 | /* |
42 | * online_page_callback contains pointer to current page onlining function. | |
43 | * Initially it is generic_online_page(). If it is required it could be | |
44 | * changed by calling set_online_page_callback() for callback registration | |
45 | * and restore_online_page_callback() for generic callback restore. | |
46 | */ | |
47 | ||
48 | static void generic_online_page(struct page *page); | |
49 | ||
50 | static online_page_callback_t online_page_callback = generic_online_page; | |
bfc8c901 | 51 | static DEFINE_MUTEX(online_page_callback_lock); |
9d0ad8ca | 52 | |
bfc8c901 VD |
53 | /* The same as the cpu_hotplug lock, but for memory hotplug. */ |
54 | static struct { | |
55 | struct task_struct *active_writer; | |
56 | struct mutex lock; /* Synchronizes accesses to refcount, */ | |
57 | /* | |
58 | * Also blocks the new readers during | |
59 | * an ongoing mem hotplug operation. | |
60 | */ | |
61 | int refcount; | |
62 | ||
63 | #ifdef CONFIG_DEBUG_LOCK_ALLOC | |
64 | struct lockdep_map dep_map; | |
65 | #endif | |
66 | } mem_hotplug = { | |
67 | .active_writer = NULL, | |
68 | .lock = __MUTEX_INITIALIZER(mem_hotplug.lock), | |
69 | .refcount = 0, | |
70 | #ifdef CONFIG_DEBUG_LOCK_ALLOC | |
71 | .dep_map = {.name = "mem_hotplug.lock" }, | |
72 | #endif | |
73 | }; | |
74 | ||
75 | /* Lockdep annotations for get/put_online_mems() and mem_hotplug_begin/end() */ | |
76 | #define memhp_lock_acquire_read() lock_map_acquire_read(&mem_hotplug.dep_map) | |
77 | #define memhp_lock_acquire() lock_map_acquire(&mem_hotplug.dep_map) | |
78 | #define memhp_lock_release() lock_map_release(&mem_hotplug.dep_map) | |
79 | ||
31bc3858 VK |
80 | bool memhp_auto_online; |
81 | EXPORT_SYMBOL_GPL(memhp_auto_online); | |
82 | ||
bfc8c901 VD |
83 | void get_online_mems(void) |
84 | { | |
85 | might_sleep(); | |
86 | if (mem_hotplug.active_writer == current) | |
87 | return; | |
88 | memhp_lock_acquire_read(); | |
89 | mutex_lock(&mem_hotplug.lock); | |
90 | mem_hotplug.refcount++; | |
91 | mutex_unlock(&mem_hotplug.lock); | |
92 | ||
93 | } | |
20d6c96b | 94 | |
bfc8c901 | 95 | void put_online_mems(void) |
20d6c96b | 96 | { |
bfc8c901 VD |
97 | if (mem_hotplug.active_writer == current) |
98 | return; | |
99 | mutex_lock(&mem_hotplug.lock); | |
100 | ||
101 | if (WARN_ON(!mem_hotplug.refcount)) | |
102 | mem_hotplug.refcount++; /* try to fix things up */ | |
103 | ||
104 | if (!--mem_hotplug.refcount && unlikely(mem_hotplug.active_writer)) | |
105 | wake_up_process(mem_hotplug.active_writer); | |
106 | mutex_unlock(&mem_hotplug.lock); | |
107 | memhp_lock_release(); | |
108 | ||
20d6c96b KM |
109 | } |
110 | ||
30467e0b | 111 | void mem_hotplug_begin(void) |
20d6c96b | 112 | { |
bfc8c901 VD |
113 | mem_hotplug.active_writer = current; |
114 | ||
115 | memhp_lock_acquire(); | |
116 | for (;;) { | |
117 | mutex_lock(&mem_hotplug.lock); | |
118 | if (likely(!mem_hotplug.refcount)) | |
119 | break; | |
120 | __set_current_state(TASK_UNINTERRUPTIBLE); | |
121 | mutex_unlock(&mem_hotplug.lock); | |
122 | schedule(); | |
123 | } | |
20d6c96b KM |
124 | } |
125 | ||
30467e0b | 126 | void mem_hotplug_done(void) |
bfc8c901 VD |
127 | { |
128 | mem_hotplug.active_writer = NULL; | |
129 | mutex_unlock(&mem_hotplug.lock); | |
130 | memhp_lock_release(); | |
131 | } | |
20d6c96b | 132 | |
45e0b78b KM |
133 | /* add this memory to iomem resource */ |
134 | static struct resource *register_memory_resource(u64 start, u64 size) | |
135 | { | |
136 | struct resource *res; | |
137 | res = kzalloc(sizeof(struct resource), GFP_KERNEL); | |
6f754ba4 VK |
138 | if (!res) |
139 | return ERR_PTR(-ENOMEM); | |
45e0b78b KM |
140 | |
141 | res->name = "System RAM"; | |
142 | res->start = start; | |
143 | res->end = start + size - 1; | |
782b8664 | 144 | res->flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY; |
45e0b78b | 145 | if (request_resource(&iomem_resource, res) < 0) { |
4996eed8 | 146 | pr_debug("System RAM resource %pR cannot be added\n", res); |
45e0b78b | 147 | kfree(res); |
6f754ba4 | 148 | return ERR_PTR(-EEXIST); |
45e0b78b KM |
149 | } |
150 | return res; | |
151 | } | |
152 | ||
153 | static void release_memory_resource(struct resource *res) | |
154 | { | |
155 | if (!res) | |
156 | return; | |
157 | release_resource(res); | |
158 | kfree(res); | |
159 | return; | |
160 | } | |
161 | ||
53947027 | 162 | #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE |
46723bfa YI |
163 | void get_page_bootmem(unsigned long info, struct page *page, |
164 | unsigned long type) | |
04753278 | 165 | { |
5f24ce5f | 166 | page->lru.next = (struct list_head *) type; |
04753278 YG |
167 | SetPagePrivate(page); |
168 | set_page_private(page, info); | |
169 | atomic_inc(&page->_count); | |
170 | } | |
171 | ||
170a5a7e | 172 | void put_page_bootmem(struct page *page) |
04753278 | 173 | { |
5f24ce5f | 174 | unsigned long type; |
04753278 | 175 | |
5f24ce5f AA |
176 | type = (unsigned long) page->lru.next; |
177 | BUG_ON(type < MEMORY_HOTPLUG_MIN_BOOTMEM_TYPE || | |
178 | type > MEMORY_HOTPLUG_MAX_BOOTMEM_TYPE); | |
04753278 YG |
179 | |
180 | if (atomic_dec_return(&page->_count) == 1) { | |
181 | ClearPagePrivate(page); | |
182 | set_page_private(page, 0); | |
5f24ce5f | 183 | INIT_LIST_HEAD(&page->lru); |
170a5a7e | 184 | free_reserved_page(page); |
04753278 | 185 | } |
04753278 YG |
186 | } |
187 | ||
46723bfa YI |
188 | #ifdef CONFIG_HAVE_BOOTMEM_INFO_NODE |
189 | #ifndef CONFIG_SPARSEMEM_VMEMMAP | |
d92bc318 | 190 | static void register_page_bootmem_info_section(unsigned long start_pfn) |
04753278 YG |
191 | { |
192 | unsigned long *usemap, mapsize, section_nr, i; | |
193 | struct mem_section *ms; | |
194 | struct page *page, *memmap; | |
195 | ||
04753278 YG |
196 | section_nr = pfn_to_section_nr(start_pfn); |
197 | ms = __nr_to_section(section_nr); | |
198 | ||
199 | /* Get section's memmap address */ | |
200 | memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr); | |
201 | ||
202 | /* | |
203 | * Get page for the memmap's phys address | |
204 | * XXX: need more consideration for sparse_vmemmap... | |
205 | */ | |
206 | page = virt_to_page(memmap); | |
207 | mapsize = sizeof(struct page) * PAGES_PER_SECTION; | |
208 | mapsize = PAGE_ALIGN(mapsize) >> PAGE_SHIFT; | |
209 | ||
210 | /* remember memmap's page */ | |
211 | for (i = 0; i < mapsize; i++, page++) | |
212 | get_page_bootmem(section_nr, page, SECTION_INFO); | |
213 | ||
214 | usemap = __nr_to_section(section_nr)->pageblock_flags; | |
215 | page = virt_to_page(usemap); | |
216 | ||
217 | mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT; | |
218 | ||
219 | for (i = 0; i < mapsize; i++, page++) | |
af370fb8 | 220 | get_page_bootmem(section_nr, page, MIX_SECTION_INFO); |
04753278 YG |
221 | |
222 | } | |
46723bfa YI |
223 | #else /* CONFIG_SPARSEMEM_VMEMMAP */ |
224 | static void register_page_bootmem_info_section(unsigned long start_pfn) | |
225 | { | |
226 | unsigned long *usemap, mapsize, section_nr, i; | |
227 | struct mem_section *ms; | |
228 | struct page *page, *memmap; | |
229 | ||
230 | if (!pfn_valid(start_pfn)) | |
231 | return; | |
232 | ||
233 | section_nr = pfn_to_section_nr(start_pfn); | |
234 | ms = __nr_to_section(section_nr); | |
235 | ||
236 | memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr); | |
237 | ||
238 | register_page_bootmem_memmap(section_nr, memmap, PAGES_PER_SECTION); | |
239 | ||
240 | usemap = __nr_to_section(section_nr)->pageblock_flags; | |
241 | page = virt_to_page(usemap); | |
242 | ||
243 | mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT; | |
244 | ||
245 | for (i = 0; i < mapsize; i++, page++) | |
246 | get_page_bootmem(section_nr, page, MIX_SECTION_INFO); | |
247 | } | |
248 | #endif /* !CONFIG_SPARSEMEM_VMEMMAP */ | |
04753278 YG |
249 | |
250 | void register_page_bootmem_info_node(struct pglist_data *pgdat) | |
251 | { | |
252 | unsigned long i, pfn, end_pfn, nr_pages; | |
253 | int node = pgdat->node_id; | |
254 | struct page *page; | |
255 | struct zone *zone; | |
256 | ||
257 | nr_pages = PAGE_ALIGN(sizeof(struct pglist_data)) >> PAGE_SHIFT; | |
258 | page = virt_to_page(pgdat); | |
259 | ||
260 | for (i = 0; i < nr_pages; i++, page++) | |
261 | get_page_bootmem(node, page, NODE_INFO); | |
262 | ||
263 | zone = &pgdat->node_zones[0]; | |
264 | for (; zone < pgdat->node_zones + MAX_NR_ZONES - 1; zone++) { | |
139c2d75 | 265 | if (zone_is_initialized(zone)) { |
04753278 YG |
266 | nr_pages = zone->wait_table_hash_nr_entries |
267 | * sizeof(wait_queue_head_t); | |
268 | nr_pages = PAGE_ALIGN(nr_pages) >> PAGE_SHIFT; | |
269 | page = virt_to_page(zone->wait_table); | |
270 | ||
271 | for (i = 0; i < nr_pages; i++, page++) | |
272 | get_page_bootmem(node, page, NODE_INFO); | |
273 | } | |
274 | } | |
275 | ||
276 | pfn = pgdat->node_start_pfn; | |
c1f19495 | 277 | end_pfn = pgdat_end_pfn(pgdat); |
04753278 | 278 | |
7e9f5eb0 | 279 | /* register section info */ |
f14851af | 280 | for (; pfn < end_pfn; pfn += PAGES_PER_SECTION) { |
281 | /* | |
282 | * Some platforms can assign the same pfn to multiple nodes - on | |
283 | * node0 as well as nodeN. To avoid registering a pfn against | |
284 | * multiple nodes we check that this pfn does not already | |
7e9f5eb0 | 285 | * reside in some other nodes. |
f14851af | 286 | */ |
287 | if (pfn_valid(pfn) && (pfn_to_nid(pfn) == node)) | |
288 | register_page_bootmem_info_section(pfn); | |
289 | } | |
04753278 | 290 | } |
46723bfa | 291 | #endif /* CONFIG_HAVE_BOOTMEM_INFO_NODE */ |
04753278 | 292 | |
f2765404 FF |
293 | static void __meminit grow_zone_span(struct zone *zone, unsigned long start_pfn, |
294 | unsigned long end_pfn) | |
76cdd58e HC |
295 | { |
296 | unsigned long old_zone_end_pfn; | |
297 | ||
298 | zone_span_writelock(zone); | |
299 | ||
c33bc315 | 300 | old_zone_end_pfn = zone_end_pfn(zone); |
8080fc03 | 301 | if (zone_is_empty(zone) || start_pfn < zone->zone_start_pfn) |
76cdd58e HC |
302 | zone->zone_start_pfn = start_pfn; |
303 | ||
304 | zone->spanned_pages = max(old_zone_end_pfn, end_pfn) - | |
305 | zone->zone_start_pfn; | |
306 | ||
307 | zone_span_writeunlock(zone); | |
308 | } | |
309 | ||
511c2aba LJ |
310 | static void resize_zone(struct zone *zone, unsigned long start_pfn, |
311 | unsigned long end_pfn) | |
312 | { | |
313 | zone_span_writelock(zone); | |
314 | ||
e455a9b9 LJ |
315 | if (end_pfn - start_pfn) { |
316 | zone->zone_start_pfn = start_pfn; | |
317 | zone->spanned_pages = end_pfn - start_pfn; | |
318 | } else { | |
319 | /* | |
320 | * make it consist as free_area_init_core(), | |
321 | * if spanned_pages = 0, then keep start_pfn = 0 | |
322 | */ | |
323 | zone->zone_start_pfn = 0; | |
324 | zone->spanned_pages = 0; | |
325 | } | |
511c2aba LJ |
326 | |
327 | zone_span_writeunlock(zone); | |
328 | } | |
329 | ||
330 | static void fix_zone_id(struct zone *zone, unsigned long start_pfn, | |
331 | unsigned long end_pfn) | |
332 | { | |
333 | enum zone_type zid = zone_idx(zone); | |
334 | int nid = zone->zone_pgdat->node_id; | |
335 | unsigned long pfn; | |
336 | ||
337 | for (pfn = start_pfn; pfn < end_pfn; pfn++) | |
338 | set_page_links(pfn_to_page(pfn), zid, nid, pfn); | |
339 | } | |
340 | ||
f6bbb78e | 341 | /* Can fail with -ENOMEM from allocating a wait table with vmalloc() or |
9e43aa2b | 342 | * alloc_bootmem_node_nopanic()/memblock_virt_alloc_node_nopanic() */ |
f6bbb78e CS |
343 | static int __ref ensure_zone_is_initialized(struct zone *zone, |
344 | unsigned long start_pfn, unsigned long num_pages) | |
345 | { | |
346 | if (!zone_is_initialized(zone)) | |
b171e409 YB |
347 | return init_currently_empty_zone(zone, start_pfn, num_pages); |
348 | ||
f6bbb78e CS |
349 | return 0; |
350 | } | |
351 | ||
e455a9b9 | 352 | static int __meminit move_pfn_range_left(struct zone *z1, struct zone *z2, |
511c2aba LJ |
353 | unsigned long start_pfn, unsigned long end_pfn) |
354 | { | |
e455a9b9 | 355 | int ret; |
511c2aba | 356 | unsigned long flags; |
e455a9b9 LJ |
357 | unsigned long z1_start_pfn; |
358 | ||
64dd1b29 CS |
359 | ret = ensure_zone_is_initialized(z1, start_pfn, end_pfn - start_pfn); |
360 | if (ret) | |
361 | return ret; | |
511c2aba LJ |
362 | |
363 | pgdat_resize_lock(z1->zone_pgdat, &flags); | |
364 | ||
365 | /* can't move pfns which are higher than @z2 */ | |
108bcc96 | 366 | if (end_pfn > zone_end_pfn(z2)) |
511c2aba | 367 | goto out_fail; |
834405c3 | 368 | /* the move out part must be at the left most of @z2 */ |
511c2aba LJ |
369 | if (start_pfn > z2->zone_start_pfn) |
370 | goto out_fail; | |
371 | /* must included/overlap */ | |
372 | if (end_pfn <= z2->zone_start_pfn) | |
373 | goto out_fail; | |
374 | ||
e455a9b9 | 375 | /* use start_pfn for z1's start_pfn if z1 is empty */ |
8080fc03 | 376 | if (!zone_is_empty(z1)) |
e455a9b9 LJ |
377 | z1_start_pfn = z1->zone_start_pfn; |
378 | else | |
379 | z1_start_pfn = start_pfn; | |
380 | ||
381 | resize_zone(z1, z1_start_pfn, end_pfn); | |
108bcc96 | 382 | resize_zone(z2, end_pfn, zone_end_pfn(z2)); |
511c2aba LJ |
383 | |
384 | pgdat_resize_unlock(z1->zone_pgdat, &flags); | |
385 | ||
386 | fix_zone_id(z1, start_pfn, end_pfn); | |
387 | ||
388 | return 0; | |
389 | out_fail: | |
390 | pgdat_resize_unlock(z1->zone_pgdat, &flags); | |
391 | return -1; | |
392 | } | |
393 | ||
e455a9b9 | 394 | static int __meminit move_pfn_range_right(struct zone *z1, struct zone *z2, |
511c2aba LJ |
395 | unsigned long start_pfn, unsigned long end_pfn) |
396 | { | |
e455a9b9 | 397 | int ret; |
511c2aba | 398 | unsigned long flags; |
e455a9b9 LJ |
399 | unsigned long z2_end_pfn; |
400 | ||
64dd1b29 CS |
401 | ret = ensure_zone_is_initialized(z2, start_pfn, end_pfn - start_pfn); |
402 | if (ret) | |
403 | return ret; | |
511c2aba LJ |
404 | |
405 | pgdat_resize_lock(z1->zone_pgdat, &flags); | |
406 | ||
407 | /* can't move pfns which are lower than @z1 */ | |
408 | if (z1->zone_start_pfn > start_pfn) | |
409 | goto out_fail; | |
410 | /* the move out part mast at the right most of @z1 */ | |
108bcc96 | 411 | if (zone_end_pfn(z1) > end_pfn) |
511c2aba LJ |
412 | goto out_fail; |
413 | /* must included/overlap */ | |
108bcc96 | 414 | if (start_pfn >= zone_end_pfn(z1)) |
511c2aba LJ |
415 | goto out_fail; |
416 | ||
e455a9b9 | 417 | /* use end_pfn for z2's end_pfn if z2 is empty */ |
8080fc03 | 418 | if (!zone_is_empty(z2)) |
108bcc96 | 419 | z2_end_pfn = zone_end_pfn(z2); |
e455a9b9 LJ |
420 | else |
421 | z2_end_pfn = end_pfn; | |
422 | ||
511c2aba | 423 | resize_zone(z1, z1->zone_start_pfn, start_pfn); |
e455a9b9 | 424 | resize_zone(z2, start_pfn, z2_end_pfn); |
511c2aba LJ |
425 | |
426 | pgdat_resize_unlock(z1->zone_pgdat, &flags); | |
427 | ||
428 | fix_zone_id(z2, start_pfn, end_pfn); | |
429 | ||
430 | return 0; | |
431 | out_fail: | |
432 | pgdat_resize_unlock(z1->zone_pgdat, &flags); | |
433 | return -1; | |
434 | } | |
435 | ||
f2765404 FF |
436 | static void __meminit grow_pgdat_span(struct pglist_data *pgdat, unsigned long start_pfn, |
437 | unsigned long end_pfn) | |
76cdd58e | 438 | { |
83285c72 | 439 | unsigned long old_pgdat_end_pfn = pgdat_end_pfn(pgdat); |
76cdd58e | 440 | |
712cd386 | 441 | if (!pgdat->node_spanned_pages || start_pfn < pgdat->node_start_pfn) |
76cdd58e HC |
442 | pgdat->node_start_pfn = start_pfn; |
443 | ||
444 | pgdat->node_spanned_pages = max(old_pgdat_end_pfn, end_pfn) - | |
445 | pgdat->node_start_pfn; | |
446 | } | |
447 | ||
31168481 | 448 | static int __meminit __add_zone(struct zone *zone, unsigned long phys_start_pfn) |
3947be19 DH |
449 | { |
450 | struct pglist_data *pgdat = zone->zone_pgdat; | |
451 | int nr_pages = PAGES_PER_SECTION; | |
452 | int nid = pgdat->node_id; | |
453 | int zone_type; | |
e298ff75 | 454 | unsigned long flags, pfn; |
64dd1b29 | 455 | int ret; |
3947be19 DH |
456 | |
457 | zone_type = zone - pgdat->node_zones; | |
64dd1b29 CS |
458 | ret = ensure_zone_is_initialized(zone, phys_start_pfn, nr_pages); |
459 | if (ret) | |
460 | return ret; | |
76cdd58e | 461 | |
76cdd58e HC |
462 | pgdat_resize_lock(zone->zone_pgdat, &flags); |
463 | grow_zone_span(zone, phys_start_pfn, phys_start_pfn + nr_pages); | |
464 | grow_pgdat_span(zone->zone_pgdat, phys_start_pfn, | |
465 | phys_start_pfn + nr_pages); | |
466 | pgdat_resize_unlock(zone->zone_pgdat, &flags); | |
a2f3aa02 DH |
467 | memmap_init_zone(nr_pages, nid, zone_type, |
468 | phys_start_pfn, MEMMAP_HOTPLUG); | |
e298ff75 MG |
469 | |
470 | /* online_page_range is called later and expects pages reserved */ | |
471 | for (pfn = phys_start_pfn; pfn < phys_start_pfn + nr_pages; pfn++) { | |
472 | if (!pfn_valid(pfn)) | |
473 | continue; | |
474 | ||
475 | SetPageReserved(pfn_to_page(pfn)); | |
476 | } | |
718127cc | 477 | return 0; |
3947be19 DH |
478 | } |
479 | ||
c04fc586 GH |
480 | static int __meminit __add_section(int nid, struct zone *zone, |
481 | unsigned long phys_start_pfn) | |
3947be19 | 482 | { |
3947be19 DH |
483 | int ret; |
484 | ||
ebd15302 KH |
485 | if (pfn_valid(phys_start_pfn)) |
486 | return -EEXIST; | |
487 | ||
85b35fea | 488 | ret = sparse_add_one_section(zone, phys_start_pfn); |
3947be19 DH |
489 | |
490 | if (ret < 0) | |
491 | return ret; | |
492 | ||
718127cc YG |
493 | ret = __add_zone(zone, phys_start_pfn); |
494 | ||
495 | if (ret < 0) | |
496 | return ret; | |
497 | ||
c04fc586 | 498 | return register_new_memory(nid, __pfn_to_section(phys_start_pfn)); |
3947be19 DH |
499 | } |
500 | ||
4edd7cef DR |
501 | /* |
502 | * Reasonably generic function for adding memory. It is | |
503 | * expected that archs that support memory hotplug will | |
504 | * call this function after deciding the zone to which to | |
505 | * add the new pages. | |
506 | */ | |
507 | int __ref __add_pages(int nid, struct zone *zone, unsigned long phys_start_pfn, | |
508 | unsigned long nr_pages) | |
509 | { | |
510 | unsigned long i; | |
511 | int err = 0; | |
512 | int start_sec, end_sec; | |
4b94ffdc DW |
513 | struct vmem_altmap *altmap; |
514 | ||
4edd7cef DR |
515 | /* during initialize mem_map, align hot-added range to section */ |
516 | start_sec = pfn_to_section_nr(phys_start_pfn); | |
517 | end_sec = pfn_to_section_nr(phys_start_pfn + nr_pages - 1); | |
518 | ||
4b94ffdc DW |
519 | altmap = to_vmem_altmap((unsigned long) pfn_to_page(phys_start_pfn)); |
520 | if (altmap) { | |
521 | /* | |
522 | * Validate altmap is within bounds of the total request | |
523 | */ | |
524 | if (altmap->base_pfn != phys_start_pfn | |
525 | || vmem_altmap_offset(altmap) > nr_pages) { | |
526 | pr_warn_once("memory add fail, invalid altmap\n"); | |
527 | return -EINVAL; | |
528 | } | |
529 | altmap->alloc = 0; | |
530 | } | |
531 | ||
4edd7cef | 532 | for (i = start_sec; i <= end_sec; i++) { |
19c07d5e | 533 | err = __add_section(nid, zone, section_nr_to_pfn(i)); |
4edd7cef DR |
534 | |
535 | /* | |
536 | * EEXIST is finally dealt with by ioresource collision | |
537 | * check. see add_memory() => register_memory_resource() | |
538 | * Warning will be printed if there is collision. | |
539 | */ | |
540 | if (err && (err != -EEXIST)) | |
541 | break; | |
542 | err = 0; | |
543 | } | |
c435a390 | 544 | vmemmap_populate_print_last(); |
4edd7cef DR |
545 | |
546 | return err; | |
547 | } | |
548 | EXPORT_SYMBOL_GPL(__add_pages); | |
549 | ||
550 | #ifdef CONFIG_MEMORY_HOTREMOVE | |
815121d2 YI |
551 | /* find the smallest valid pfn in the range [start_pfn, end_pfn) */ |
552 | static int find_smallest_section_pfn(int nid, struct zone *zone, | |
553 | unsigned long start_pfn, | |
554 | unsigned long end_pfn) | |
555 | { | |
556 | struct mem_section *ms; | |
557 | ||
558 | for (; start_pfn < end_pfn; start_pfn += PAGES_PER_SECTION) { | |
559 | ms = __pfn_to_section(start_pfn); | |
560 | ||
561 | if (unlikely(!valid_section(ms))) | |
562 | continue; | |
563 | ||
564 | if (unlikely(pfn_to_nid(start_pfn) != nid)) | |
565 | continue; | |
566 | ||
567 | if (zone && zone != page_zone(pfn_to_page(start_pfn))) | |
568 | continue; | |
569 | ||
570 | return start_pfn; | |
571 | } | |
572 | ||
573 | return 0; | |
574 | } | |
575 | ||
576 | /* find the biggest valid pfn in the range [start_pfn, end_pfn). */ | |
577 | static int find_biggest_section_pfn(int nid, struct zone *zone, | |
578 | unsigned long start_pfn, | |
579 | unsigned long end_pfn) | |
580 | { | |
581 | struct mem_section *ms; | |
582 | unsigned long pfn; | |
583 | ||
584 | /* pfn is the end pfn of a memory section. */ | |
585 | pfn = end_pfn - 1; | |
586 | for (; pfn >= start_pfn; pfn -= PAGES_PER_SECTION) { | |
587 | ms = __pfn_to_section(pfn); | |
588 | ||
589 | if (unlikely(!valid_section(ms))) | |
590 | continue; | |
591 | ||
592 | if (unlikely(pfn_to_nid(pfn) != nid)) | |
593 | continue; | |
594 | ||
595 | if (zone && zone != page_zone(pfn_to_page(pfn))) | |
596 | continue; | |
597 | ||
598 | return pfn; | |
599 | } | |
600 | ||
601 | return 0; | |
602 | } | |
603 | ||
604 | static void shrink_zone_span(struct zone *zone, unsigned long start_pfn, | |
605 | unsigned long end_pfn) | |
606 | { | |
c33bc315 XQ |
607 | unsigned long zone_start_pfn = zone->zone_start_pfn; |
608 | unsigned long z = zone_end_pfn(zone); /* zone_end_pfn namespace clash */ | |
609 | unsigned long zone_end_pfn = z; | |
815121d2 YI |
610 | unsigned long pfn; |
611 | struct mem_section *ms; | |
612 | int nid = zone_to_nid(zone); | |
613 | ||
614 | zone_span_writelock(zone); | |
615 | if (zone_start_pfn == start_pfn) { | |
616 | /* | |
617 | * If the section is smallest section in the zone, it need | |
618 | * shrink zone->zone_start_pfn and zone->zone_spanned_pages. | |
619 | * In this case, we find second smallest valid mem_section | |
620 | * for shrinking zone. | |
621 | */ | |
622 | pfn = find_smallest_section_pfn(nid, zone, end_pfn, | |
623 | zone_end_pfn); | |
624 | if (pfn) { | |
625 | zone->zone_start_pfn = pfn; | |
626 | zone->spanned_pages = zone_end_pfn - pfn; | |
627 | } | |
628 | } else if (zone_end_pfn == end_pfn) { | |
629 | /* | |
630 | * If the section is biggest section in the zone, it need | |
631 | * shrink zone->spanned_pages. | |
632 | * In this case, we find second biggest valid mem_section for | |
633 | * shrinking zone. | |
634 | */ | |
635 | pfn = find_biggest_section_pfn(nid, zone, zone_start_pfn, | |
636 | start_pfn); | |
637 | if (pfn) | |
638 | zone->spanned_pages = pfn - zone_start_pfn + 1; | |
639 | } | |
640 | ||
641 | /* | |
642 | * The section is not biggest or smallest mem_section in the zone, it | |
643 | * only creates a hole in the zone. So in this case, we need not | |
644 | * change the zone. But perhaps, the zone has only hole data. Thus | |
645 | * it check the zone has only hole or not. | |
646 | */ | |
647 | pfn = zone_start_pfn; | |
648 | for (; pfn < zone_end_pfn; pfn += PAGES_PER_SECTION) { | |
649 | ms = __pfn_to_section(pfn); | |
650 | ||
651 | if (unlikely(!valid_section(ms))) | |
652 | continue; | |
653 | ||
654 | if (page_zone(pfn_to_page(pfn)) != zone) | |
655 | continue; | |
656 | ||
657 | /* If the section is current section, it continues the loop */ | |
658 | if (start_pfn == pfn) | |
659 | continue; | |
660 | ||
661 | /* If we find valid section, we have nothing to do */ | |
662 | zone_span_writeunlock(zone); | |
663 | return; | |
664 | } | |
665 | ||
666 | /* The zone has no valid section */ | |
667 | zone->zone_start_pfn = 0; | |
668 | zone->spanned_pages = 0; | |
669 | zone_span_writeunlock(zone); | |
670 | } | |
671 | ||
672 | static void shrink_pgdat_span(struct pglist_data *pgdat, | |
673 | unsigned long start_pfn, unsigned long end_pfn) | |
674 | { | |
83285c72 XQ |
675 | unsigned long pgdat_start_pfn = pgdat->node_start_pfn; |
676 | unsigned long p = pgdat_end_pfn(pgdat); /* pgdat_end_pfn namespace clash */ | |
677 | unsigned long pgdat_end_pfn = p; | |
815121d2 YI |
678 | unsigned long pfn; |
679 | struct mem_section *ms; | |
680 | int nid = pgdat->node_id; | |
681 | ||
682 | if (pgdat_start_pfn == start_pfn) { | |
683 | /* | |
684 | * If the section is smallest section in the pgdat, it need | |
685 | * shrink pgdat->node_start_pfn and pgdat->node_spanned_pages. | |
686 | * In this case, we find second smallest valid mem_section | |
687 | * for shrinking zone. | |
688 | */ | |
689 | pfn = find_smallest_section_pfn(nid, NULL, end_pfn, | |
690 | pgdat_end_pfn); | |
691 | if (pfn) { | |
692 | pgdat->node_start_pfn = pfn; | |
693 | pgdat->node_spanned_pages = pgdat_end_pfn - pfn; | |
694 | } | |
695 | } else if (pgdat_end_pfn == end_pfn) { | |
696 | /* | |
697 | * If the section is biggest section in the pgdat, it need | |
698 | * shrink pgdat->node_spanned_pages. | |
699 | * In this case, we find second biggest valid mem_section for | |
700 | * shrinking zone. | |
701 | */ | |
702 | pfn = find_biggest_section_pfn(nid, NULL, pgdat_start_pfn, | |
703 | start_pfn); | |
704 | if (pfn) | |
705 | pgdat->node_spanned_pages = pfn - pgdat_start_pfn + 1; | |
706 | } | |
707 | ||
708 | /* | |
709 | * If the section is not biggest or smallest mem_section in the pgdat, | |
710 | * it only creates a hole in the pgdat. So in this case, we need not | |
711 | * change the pgdat. | |
712 | * But perhaps, the pgdat has only hole data. Thus it check the pgdat | |
713 | * has only hole or not. | |
714 | */ | |
715 | pfn = pgdat_start_pfn; | |
716 | for (; pfn < pgdat_end_pfn; pfn += PAGES_PER_SECTION) { | |
717 | ms = __pfn_to_section(pfn); | |
718 | ||
719 | if (unlikely(!valid_section(ms))) | |
720 | continue; | |
721 | ||
722 | if (pfn_to_nid(pfn) != nid) | |
723 | continue; | |
724 | ||
725 | /* If the section is current section, it continues the loop */ | |
726 | if (start_pfn == pfn) | |
727 | continue; | |
728 | ||
729 | /* If we find valid section, we have nothing to do */ | |
730 | return; | |
731 | } | |
732 | ||
733 | /* The pgdat has no valid section */ | |
734 | pgdat->node_start_pfn = 0; | |
735 | pgdat->node_spanned_pages = 0; | |
736 | } | |
737 | ||
738 | static void __remove_zone(struct zone *zone, unsigned long start_pfn) | |
739 | { | |
740 | struct pglist_data *pgdat = zone->zone_pgdat; | |
741 | int nr_pages = PAGES_PER_SECTION; | |
742 | int zone_type; | |
743 | unsigned long flags; | |
744 | ||
745 | zone_type = zone - pgdat->node_zones; | |
746 | ||
747 | pgdat_resize_lock(zone->zone_pgdat, &flags); | |
748 | shrink_zone_span(zone, start_pfn, start_pfn + nr_pages); | |
749 | shrink_pgdat_span(pgdat, start_pfn, start_pfn + nr_pages); | |
750 | pgdat_resize_unlock(zone->zone_pgdat, &flags); | |
751 | } | |
752 | ||
4b94ffdc DW |
753 | static int __remove_section(struct zone *zone, struct mem_section *ms, |
754 | unsigned long map_offset) | |
ea01ea93 | 755 | { |
815121d2 YI |
756 | unsigned long start_pfn; |
757 | int scn_nr; | |
ea01ea93 BP |
758 | int ret = -EINVAL; |
759 | ||
760 | if (!valid_section(ms)) | |
761 | return ret; | |
762 | ||
763 | ret = unregister_memory_section(ms); | |
764 | if (ret) | |
765 | return ret; | |
766 | ||
815121d2 YI |
767 | scn_nr = __section_nr(ms); |
768 | start_pfn = section_nr_to_pfn(scn_nr); | |
769 | __remove_zone(zone, start_pfn); | |
770 | ||
4b94ffdc | 771 | sparse_remove_one_section(zone, ms, map_offset); |
ea01ea93 BP |
772 | return 0; |
773 | } | |
774 | ||
ea01ea93 BP |
775 | /** |
776 | * __remove_pages() - remove sections of pages from a zone | |
777 | * @zone: zone from which pages need to be removed | |
778 | * @phys_start_pfn: starting pageframe (must be aligned to start of a section) | |
779 | * @nr_pages: number of pages to remove (must be multiple of section size) | |
780 | * | |
781 | * Generic helper function to remove section mappings and sysfs entries | |
782 | * for the section of the memory we are removing. Caller needs to make | |
783 | * sure that pages are marked reserved and zones are adjust properly by | |
784 | * calling offline_pages(). | |
785 | */ | |
786 | int __remove_pages(struct zone *zone, unsigned long phys_start_pfn, | |
787 | unsigned long nr_pages) | |
788 | { | |
fe74ebb1 | 789 | unsigned long i; |
4b94ffdc DW |
790 | unsigned long map_offset = 0; |
791 | int sections_to_remove, ret = 0; | |
792 | ||
793 | /* In the ZONE_DEVICE case device driver owns the memory region */ | |
794 | if (is_dev_zone(zone)) { | |
795 | struct page *page = pfn_to_page(phys_start_pfn); | |
796 | struct vmem_altmap *altmap; | |
797 | ||
798 | altmap = to_vmem_altmap((unsigned long) page); | |
799 | if (altmap) | |
800 | map_offset = vmem_altmap_offset(altmap); | |
801 | } else { | |
802 | resource_size_t start, size; | |
803 | ||
804 | start = phys_start_pfn << PAGE_SHIFT; | |
805 | size = nr_pages * PAGE_SIZE; | |
806 | ||
807 | ret = release_mem_region_adjustable(&iomem_resource, start, | |
808 | size); | |
809 | if (ret) { | |
810 | resource_size_t endres = start + size - 1; | |
811 | ||
812 | pr_warn("Unable to release resource <%pa-%pa> (%d)\n", | |
813 | &start, &endres, ret); | |
814 | } | |
815 | } | |
ea01ea93 BP |
816 | |
817 | /* | |
818 | * We can only remove entire sections | |
819 | */ | |
820 | BUG_ON(phys_start_pfn & ~PAGE_SECTION_MASK); | |
821 | BUG_ON(nr_pages % PAGES_PER_SECTION); | |
822 | ||
ea01ea93 BP |
823 | sections_to_remove = nr_pages / PAGES_PER_SECTION; |
824 | for (i = 0; i < sections_to_remove; i++) { | |
825 | unsigned long pfn = phys_start_pfn + i*PAGES_PER_SECTION; | |
4b94ffdc DW |
826 | |
827 | ret = __remove_section(zone, __pfn_to_section(pfn), map_offset); | |
828 | map_offset = 0; | |
ea01ea93 BP |
829 | if (ret) |
830 | break; | |
831 | } | |
832 | return ret; | |
833 | } | |
834 | EXPORT_SYMBOL_GPL(__remove_pages); | |
4edd7cef | 835 | #endif /* CONFIG_MEMORY_HOTREMOVE */ |
ea01ea93 | 836 | |
9d0ad8ca DK |
837 | int set_online_page_callback(online_page_callback_t callback) |
838 | { | |
839 | int rc = -EINVAL; | |
840 | ||
bfc8c901 VD |
841 | get_online_mems(); |
842 | mutex_lock(&online_page_callback_lock); | |
9d0ad8ca DK |
843 | |
844 | if (online_page_callback == generic_online_page) { | |
845 | online_page_callback = callback; | |
846 | rc = 0; | |
847 | } | |
848 | ||
bfc8c901 VD |
849 | mutex_unlock(&online_page_callback_lock); |
850 | put_online_mems(); | |
9d0ad8ca DK |
851 | |
852 | return rc; | |
853 | } | |
854 | EXPORT_SYMBOL_GPL(set_online_page_callback); | |
855 | ||
856 | int restore_online_page_callback(online_page_callback_t callback) | |
857 | { | |
858 | int rc = -EINVAL; | |
859 | ||
bfc8c901 VD |
860 | get_online_mems(); |
861 | mutex_lock(&online_page_callback_lock); | |
9d0ad8ca DK |
862 | |
863 | if (online_page_callback == callback) { | |
864 | online_page_callback = generic_online_page; | |
865 | rc = 0; | |
866 | } | |
867 | ||
bfc8c901 VD |
868 | mutex_unlock(&online_page_callback_lock); |
869 | put_online_mems(); | |
9d0ad8ca DK |
870 | |
871 | return rc; | |
872 | } | |
873 | EXPORT_SYMBOL_GPL(restore_online_page_callback); | |
874 | ||
875 | void __online_page_set_limits(struct page *page) | |
180c06ef | 876 | { |
9d0ad8ca DK |
877 | } |
878 | EXPORT_SYMBOL_GPL(__online_page_set_limits); | |
879 | ||
880 | void __online_page_increment_counters(struct page *page) | |
881 | { | |
3dcc0571 | 882 | adjust_managed_page_count(page, 1); |
9d0ad8ca DK |
883 | } |
884 | EXPORT_SYMBOL_GPL(__online_page_increment_counters); | |
180c06ef | 885 | |
9d0ad8ca DK |
886 | void __online_page_free(struct page *page) |
887 | { | |
3dcc0571 | 888 | __free_reserved_page(page); |
180c06ef | 889 | } |
9d0ad8ca DK |
890 | EXPORT_SYMBOL_GPL(__online_page_free); |
891 | ||
892 | static void generic_online_page(struct page *page) | |
893 | { | |
894 | __online_page_set_limits(page); | |
895 | __online_page_increment_counters(page); | |
896 | __online_page_free(page); | |
897 | } | |
180c06ef | 898 | |
75884fb1 KH |
899 | static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages, |
900 | void *arg) | |
3947be19 DH |
901 | { |
902 | unsigned long i; | |
75884fb1 KH |
903 | unsigned long onlined_pages = *(unsigned long *)arg; |
904 | struct page *page; | |
905 | if (PageReserved(pfn_to_page(start_pfn))) | |
906 | for (i = 0; i < nr_pages; i++) { | |
907 | page = pfn_to_page(start_pfn + i); | |
9d0ad8ca | 908 | (*online_page_callback)(page); |
75884fb1 KH |
909 | onlined_pages++; |
910 | } | |
911 | *(unsigned long *)arg = onlined_pages; | |
912 | return 0; | |
913 | } | |
914 | ||
09285af7 | 915 | #ifdef CONFIG_MOVABLE_NODE |
79a4dcef TC |
916 | /* |
917 | * When CONFIG_MOVABLE_NODE, we permit onlining of a node which doesn't have | |
918 | * normal memory. | |
919 | */ | |
09285af7 LJ |
920 | static bool can_online_high_movable(struct zone *zone) |
921 | { | |
922 | return true; | |
923 | } | |
79a4dcef | 924 | #else /* CONFIG_MOVABLE_NODE */ |
74d42d8f LJ |
925 | /* ensure every online node has NORMAL memory */ |
926 | static bool can_online_high_movable(struct zone *zone) | |
927 | { | |
928 | return node_state(zone_to_nid(zone), N_NORMAL_MEMORY); | |
929 | } | |
79a4dcef | 930 | #endif /* CONFIG_MOVABLE_NODE */ |
74d42d8f | 931 | |
d9713679 LJ |
932 | /* check which state of node_states will be changed when online memory */ |
933 | static void node_states_check_changes_online(unsigned long nr_pages, | |
934 | struct zone *zone, struct memory_notify *arg) | |
935 | { | |
936 | int nid = zone_to_nid(zone); | |
937 | enum zone_type zone_last = ZONE_NORMAL; | |
938 | ||
939 | /* | |
6715ddf9 LJ |
940 | * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY] |
941 | * contains nodes which have zones of 0...ZONE_NORMAL, | |
942 | * set zone_last to ZONE_NORMAL. | |
d9713679 | 943 | * |
6715ddf9 LJ |
944 | * If we don't have HIGHMEM nor movable node, |
945 | * node_states[N_NORMAL_MEMORY] contains nodes which have zones of | |
946 | * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE. | |
d9713679 | 947 | */ |
6715ddf9 | 948 | if (N_MEMORY == N_NORMAL_MEMORY) |
d9713679 LJ |
949 | zone_last = ZONE_MOVABLE; |
950 | ||
951 | /* | |
952 | * if the memory to be online is in a zone of 0...zone_last, and | |
953 | * the zones of 0...zone_last don't have memory before online, we will | |
954 | * need to set the node to node_states[N_NORMAL_MEMORY] after | |
955 | * the memory is online. | |
956 | */ | |
957 | if (zone_idx(zone) <= zone_last && !node_state(nid, N_NORMAL_MEMORY)) | |
958 | arg->status_change_nid_normal = nid; | |
959 | else | |
960 | arg->status_change_nid_normal = -1; | |
961 | ||
6715ddf9 LJ |
962 | #ifdef CONFIG_HIGHMEM |
963 | /* | |
964 | * If we have movable node, node_states[N_HIGH_MEMORY] | |
965 | * contains nodes which have zones of 0...ZONE_HIGHMEM, | |
966 | * set zone_last to ZONE_HIGHMEM. | |
967 | * | |
968 | * If we don't have movable node, node_states[N_NORMAL_MEMORY] | |
969 | * contains nodes which have zones of 0...ZONE_MOVABLE, | |
970 | * set zone_last to ZONE_MOVABLE. | |
971 | */ | |
972 | zone_last = ZONE_HIGHMEM; | |
973 | if (N_MEMORY == N_HIGH_MEMORY) | |
974 | zone_last = ZONE_MOVABLE; | |
975 | ||
976 | if (zone_idx(zone) <= zone_last && !node_state(nid, N_HIGH_MEMORY)) | |
977 | arg->status_change_nid_high = nid; | |
978 | else | |
979 | arg->status_change_nid_high = -1; | |
980 | #else | |
981 | arg->status_change_nid_high = arg->status_change_nid_normal; | |
982 | #endif | |
983 | ||
d9713679 LJ |
984 | /* |
985 | * if the node don't have memory befor online, we will need to | |
6715ddf9 | 986 | * set the node to node_states[N_MEMORY] after the memory |
d9713679 LJ |
987 | * is online. |
988 | */ | |
6715ddf9 | 989 | if (!node_state(nid, N_MEMORY)) |
d9713679 LJ |
990 | arg->status_change_nid = nid; |
991 | else | |
992 | arg->status_change_nid = -1; | |
993 | } | |
994 | ||
995 | static void node_states_set_node(int node, struct memory_notify *arg) | |
996 | { | |
997 | if (arg->status_change_nid_normal >= 0) | |
998 | node_set_state(node, N_NORMAL_MEMORY); | |
999 | ||
6715ddf9 LJ |
1000 | if (arg->status_change_nid_high >= 0) |
1001 | node_set_state(node, N_HIGH_MEMORY); | |
1002 | ||
1003 | node_set_state(node, N_MEMORY); | |
d9713679 LJ |
1004 | } |
1005 | ||
75884fb1 | 1006 | |
30467e0b | 1007 | /* Must be protected by mem_hotplug_begin() */ |
511c2aba | 1008 | int __ref online_pages(unsigned long pfn, unsigned long nr_pages, int online_type) |
75884fb1 | 1009 | { |
aa47228a | 1010 | unsigned long flags; |
3947be19 DH |
1011 | unsigned long onlined_pages = 0; |
1012 | struct zone *zone; | |
6811378e | 1013 | int need_zonelists_rebuild = 0; |
7b78d335 YG |
1014 | int nid; |
1015 | int ret; | |
1016 | struct memory_notify arg; | |
1017 | ||
d9713679 LJ |
1018 | /* |
1019 | * This doesn't need a lock to do pfn_to_page(). | |
1020 | * The section can't be removed here because of the | |
1021 | * memory_block->state_mutex. | |
1022 | */ | |
1023 | zone = page_zone(pfn_to_page(pfn)); | |
1024 | ||
4f7c6b49 TC |
1025 | if ((zone_idx(zone) > ZONE_NORMAL || |
1026 | online_type == MMOP_ONLINE_MOVABLE) && | |
bfc8c901 | 1027 | !can_online_high_movable(zone)) |
30467e0b | 1028 | return -EINVAL; |
74d42d8f | 1029 | |
4f7c6b49 TC |
1030 | if (online_type == MMOP_ONLINE_KERNEL && |
1031 | zone_idx(zone) == ZONE_MOVABLE) { | |
bfc8c901 | 1032 | if (move_pfn_range_left(zone - 1, zone, pfn, pfn + nr_pages)) |
30467e0b | 1033 | return -EINVAL; |
511c2aba | 1034 | } |
4f7c6b49 TC |
1035 | if (online_type == MMOP_ONLINE_MOVABLE && |
1036 | zone_idx(zone) == ZONE_MOVABLE - 1) { | |
bfc8c901 | 1037 | if (move_pfn_range_right(zone, zone + 1, pfn, pfn + nr_pages)) |
30467e0b | 1038 | return -EINVAL; |
511c2aba LJ |
1039 | } |
1040 | ||
1041 | /* Previous code may changed the zone of the pfn range */ | |
1042 | zone = page_zone(pfn_to_page(pfn)); | |
1043 | ||
7b78d335 YG |
1044 | arg.start_pfn = pfn; |
1045 | arg.nr_pages = nr_pages; | |
d9713679 | 1046 | node_states_check_changes_online(nr_pages, zone, &arg); |
7b78d335 | 1047 | |
9c2606b7 | 1048 | nid = pfn_to_nid(pfn); |
3947be19 | 1049 | |
7b78d335 YG |
1050 | ret = memory_notify(MEM_GOING_ONLINE, &arg); |
1051 | ret = notifier_to_errno(ret); | |
1052 | if (ret) { | |
1053 | memory_notify(MEM_CANCEL_ONLINE, &arg); | |
30467e0b | 1054 | return ret; |
7b78d335 | 1055 | } |
6811378e YG |
1056 | /* |
1057 | * If this zone is not populated, then it is not in zonelist. | |
1058 | * This means the page allocator ignores this zone. | |
1059 | * So, zonelist must be updated after online. | |
1060 | */ | |
4eaf3f64 | 1061 | mutex_lock(&zonelists_mutex); |
6dcd73d7 | 1062 | if (!populated_zone(zone)) { |
6811378e | 1063 | need_zonelists_rebuild = 1; |
6dcd73d7 WC |
1064 | build_all_zonelists(NULL, zone); |
1065 | } | |
6811378e | 1066 | |
908eedc6 | 1067 | ret = walk_system_ram_range(pfn, nr_pages, &onlined_pages, |
75884fb1 | 1068 | online_pages_range); |
fd8a4221 | 1069 | if (ret) { |
6dcd73d7 WC |
1070 | if (need_zonelists_rebuild) |
1071 | zone_pcp_reset(zone); | |
4eaf3f64 | 1072 | mutex_unlock(&zonelists_mutex); |
a62e2f4f BH |
1073 | printk(KERN_DEBUG "online_pages [mem %#010llx-%#010llx] failed\n", |
1074 | (unsigned long long) pfn << PAGE_SHIFT, | |
1075 | (((unsigned long long) pfn + nr_pages) | |
1076 | << PAGE_SHIFT) - 1); | |
fd8a4221 | 1077 | memory_notify(MEM_CANCEL_ONLINE, &arg); |
30467e0b | 1078 | return ret; |
fd8a4221 GL |
1079 | } |
1080 | ||
3947be19 | 1081 | zone->present_pages += onlined_pages; |
aa47228a CS |
1082 | |
1083 | pgdat_resize_lock(zone->zone_pgdat, &flags); | |
f2937be5 | 1084 | zone->zone_pgdat->node_present_pages += onlined_pages; |
aa47228a CS |
1085 | pgdat_resize_unlock(zone->zone_pgdat, &flags); |
1086 | ||
08dff7b7 | 1087 | if (onlined_pages) { |
d9713679 | 1088 | node_states_set_node(zone_to_nid(zone), &arg); |
08dff7b7 | 1089 | if (need_zonelists_rebuild) |
6dcd73d7 | 1090 | build_all_zonelists(NULL, NULL); |
08dff7b7 JL |
1091 | else |
1092 | zone_pcp_update(zone); | |
1093 | } | |
3947be19 | 1094 | |
4eaf3f64 | 1095 | mutex_unlock(&zonelists_mutex); |
1b79acc9 KM |
1096 | |
1097 | init_per_zone_wmark_min(); | |
1098 | ||
08dff7b7 | 1099 | if (onlined_pages) |
7ea1530a | 1100 | kswapd_run(zone_to_nid(zone)); |
61b13993 | 1101 | |
1f522509 | 1102 | vm_total_pages = nr_free_pagecache_pages(); |
2f7f24ec | 1103 | |
2d1d43f6 | 1104 | writeback_set_ratelimit(); |
7b78d335 YG |
1105 | |
1106 | if (onlined_pages) | |
1107 | memory_notify(MEM_ONLINE, &arg); | |
30467e0b | 1108 | return 0; |
3947be19 | 1109 | } |
53947027 | 1110 | #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */ |
bc02af93 | 1111 | |
0bd85420 TC |
1112 | static void reset_node_present_pages(pg_data_t *pgdat) |
1113 | { | |
1114 | struct zone *z; | |
1115 | ||
1116 | for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++) | |
1117 | z->present_pages = 0; | |
1118 | ||
1119 | pgdat->node_present_pages = 0; | |
1120 | } | |
1121 | ||
e1319331 HS |
1122 | /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */ |
1123 | static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start) | |
9af3c2de YG |
1124 | { |
1125 | struct pglist_data *pgdat; | |
1126 | unsigned long zones_size[MAX_NR_ZONES] = {0}; | |
1127 | unsigned long zholes_size[MAX_NR_ZONES] = {0}; | |
c8e861a5 | 1128 | unsigned long start_pfn = PFN_DOWN(start); |
9af3c2de | 1129 | |
a1e565aa TC |
1130 | pgdat = NODE_DATA(nid); |
1131 | if (!pgdat) { | |
1132 | pgdat = arch_alloc_nodedata(nid); | |
1133 | if (!pgdat) | |
1134 | return NULL; | |
9af3c2de | 1135 | |
a1e565aa | 1136 | arch_refresh_nodedata(nid, pgdat); |
b0dc3a34 GZ |
1137 | } else { |
1138 | /* Reset the nr_zones and classzone_idx to 0 before reuse */ | |
1139 | pgdat->nr_zones = 0; | |
1140 | pgdat->classzone_idx = 0; | |
a1e565aa | 1141 | } |
9af3c2de YG |
1142 | |
1143 | /* we can use NODE_DATA(nid) from here */ | |
1144 | ||
1145 | /* init node's zones as empty zones, we don't have any present pages.*/ | |
9109fb7b | 1146 | free_area_init_node(nid, zones_size, start_pfn, zholes_size); |
9af3c2de | 1147 | |
959ecc48 KH |
1148 | /* |
1149 | * The node we allocated has no zone fallback lists. For avoiding | |
1150 | * to access not-initialized zonelist, build here. | |
1151 | */ | |
f957db4f | 1152 | mutex_lock(&zonelists_mutex); |
9adb62a5 | 1153 | build_all_zonelists(pgdat, NULL); |
f957db4f | 1154 | mutex_unlock(&zonelists_mutex); |
959ecc48 | 1155 | |
f784a3f1 TC |
1156 | /* |
1157 | * zone->managed_pages is set to an approximate value in | |
1158 | * free_area_init_core(), which will cause | |
1159 | * /sys/device/system/node/nodeX/meminfo has wrong data. | |
1160 | * So reset it to 0 before any memory is onlined. | |
1161 | */ | |
1162 | reset_node_managed_pages(pgdat); | |
1163 | ||
0bd85420 TC |
1164 | /* |
1165 | * When memory is hot-added, all the memory is in offline state. So | |
1166 | * clear all zones' present_pages because they will be updated in | |
1167 | * online_pages() and offline_pages(). | |
1168 | */ | |
1169 | reset_node_present_pages(pgdat); | |
1170 | ||
9af3c2de YG |
1171 | return pgdat; |
1172 | } | |
1173 | ||
1174 | static void rollback_node_hotadd(int nid, pg_data_t *pgdat) | |
1175 | { | |
1176 | arch_refresh_nodedata(nid, NULL); | |
1177 | arch_free_nodedata(pgdat); | |
1178 | return; | |
1179 | } | |
1180 | ||
0a547039 | 1181 | |
01b0f197 TK |
1182 | /** |
1183 | * try_online_node - online a node if offlined | |
1184 | * | |
cf23422b | 1185 | * called by cpu_up() to online a node without onlined memory. |
1186 | */ | |
01b0f197 | 1187 | int try_online_node(int nid) |
cf23422b | 1188 | { |
1189 | pg_data_t *pgdat; | |
1190 | int ret; | |
1191 | ||
01b0f197 TK |
1192 | if (node_online(nid)) |
1193 | return 0; | |
1194 | ||
bfc8c901 | 1195 | mem_hotplug_begin(); |
cf23422b | 1196 | pgdat = hotadd_new_pgdat(nid, 0); |
7553e8f2 | 1197 | if (!pgdat) { |
01b0f197 | 1198 | pr_err("Cannot online node %d due to NULL pgdat\n", nid); |
cf23422b | 1199 | ret = -ENOMEM; |
1200 | goto out; | |
1201 | } | |
1202 | node_set_online(nid); | |
1203 | ret = register_one_node(nid); | |
1204 | BUG_ON(ret); | |
1205 | ||
01b0f197 TK |
1206 | if (pgdat->node_zonelists->_zonerefs->zone == NULL) { |
1207 | mutex_lock(&zonelists_mutex); | |
1208 | build_all_zonelists(NULL, NULL); | |
1209 | mutex_unlock(&zonelists_mutex); | |
1210 | } | |
1211 | ||
cf23422b | 1212 | out: |
bfc8c901 | 1213 | mem_hotplug_done(); |
cf23422b | 1214 | return ret; |
1215 | } | |
1216 | ||
27356f54 TK |
1217 | static int check_hotplug_memory_range(u64 start, u64 size) |
1218 | { | |
c8e861a5 | 1219 | u64 start_pfn = PFN_DOWN(start); |
27356f54 TK |
1220 | u64 nr_pages = size >> PAGE_SHIFT; |
1221 | ||
1222 | /* Memory range must be aligned with section */ | |
1223 | if ((start_pfn & ~PAGE_SECTION_MASK) || | |
1224 | (nr_pages % PAGES_PER_SECTION) || (!nr_pages)) { | |
1225 | pr_err("Section-unaligned hotplug range: start 0x%llx, size 0x%llx\n", | |
1226 | (unsigned long long)start, | |
1227 | (unsigned long long)size); | |
1228 | return -EINVAL; | |
1229 | } | |
1230 | ||
1231 | return 0; | |
1232 | } | |
1233 | ||
63264400 WN |
1234 | /* |
1235 | * If movable zone has already been setup, newly added memory should be check. | |
1236 | * If its address is higher than movable zone, it should be added as movable. | |
1237 | * Without this check, movable zone may overlap with other zone. | |
1238 | */ | |
1239 | static int should_add_memory_movable(int nid, u64 start, u64 size) | |
1240 | { | |
1241 | unsigned long start_pfn = start >> PAGE_SHIFT; | |
1242 | pg_data_t *pgdat = NODE_DATA(nid); | |
1243 | struct zone *movable_zone = pgdat->node_zones + ZONE_MOVABLE; | |
1244 | ||
1245 | if (zone_is_empty(movable_zone)) | |
1246 | return 0; | |
1247 | ||
1248 | if (movable_zone->zone_start_pfn <= start_pfn) | |
1249 | return 1; | |
1250 | ||
1251 | return 0; | |
1252 | } | |
1253 | ||
033fbae9 DW |
1254 | int zone_for_memory(int nid, u64 start, u64 size, int zone_default, |
1255 | bool for_device) | |
63264400 | 1256 | { |
033fbae9 DW |
1257 | #ifdef CONFIG_ZONE_DEVICE |
1258 | if (for_device) | |
1259 | return ZONE_DEVICE; | |
1260 | #endif | |
63264400 WN |
1261 | if (should_add_memory_movable(nid, start, size)) |
1262 | return ZONE_MOVABLE; | |
1263 | ||
1264 | return zone_default; | |
1265 | } | |
1266 | ||
31bc3858 VK |
1267 | static int online_memory_block(struct memory_block *mem, void *arg) |
1268 | { | |
1269 | return memory_block_change_state(mem, MEM_ONLINE, MEM_OFFLINE); | |
1270 | } | |
1271 | ||
31168481 | 1272 | /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */ |
31bc3858 | 1273 | int __ref add_memory_resource(int nid, struct resource *res, bool online) |
bc02af93 | 1274 | { |
62cedb9f | 1275 | u64 start, size; |
9af3c2de | 1276 | pg_data_t *pgdat = NULL; |
a1e565aa TC |
1277 | bool new_pgdat; |
1278 | bool new_node; | |
bc02af93 YG |
1279 | int ret; |
1280 | ||
62cedb9f DV |
1281 | start = res->start; |
1282 | size = resource_size(res); | |
1283 | ||
27356f54 TK |
1284 | ret = check_hotplug_memory_range(start, size); |
1285 | if (ret) | |
1286 | return ret; | |
1287 | ||
a1e565aa TC |
1288 | { /* Stupid hack to suppress address-never-null warning */ |
1289 | void *p = NODE_DATA(nid); | |
1290 | new_pgdat = !p; | |
1291 | } | |
ac13c462 | 1292 | |
bfc8c901 | 1293 | mem_hotplug_begin(); |
ac13c462 | 1294 | |
7f36e3e5 TC |
1295 | /* |
1296 | * Add new range to memblock so that when hotadd_new_pgdat() is called | |
1297 | * to allocate new pgdat, get_pfn_range_for_nid() will be able to find | |
1298 | * this new range and calculate total pages correctly. The range will | |
1299 | * be removed at hot-remove time. | |
1300 | */ | |
1301 | memblock_add_node(start, size, nid); | |
1302 | ||
a1e565aa TC |
1303 | new_node = !node_online(nid); |
1304 | if (new_node) { | |
9af3c2de | 1305 | pgdat = hotadd_new_pgdat(nid, start); |
6ad696d2 | 1306 | ret = -ENOMEM; |
9af3c2de | 1307 | if (!pgdat) |
41b9e2d7 | 1308 | goto error; |
9af3c2de YG |
1309 | } |
1310 | ||
bc02af93 | 1311 | /* call arch's memory hotadd */ |
033fbae9 | 1312 | ret = arch_add_memory(nid, start, size, false); |
bc02af93 | 1313 | |
9af3c2de YG |
1314 | if (ret < 0) |
1315 | goto error; | |
1316 | ||
0fc44159 | 1317 | /* we online node here. we can't roll back from here. */ |
9af3c2de YG |
1318 | node_set_online(nid); |
1319 | ||
a1e565aa | 1320 | if (new_node) { |
0fc44159 YG |
1321 | ret = register_one_node(nid); |
1322 | /* | |
1323 | * If sysfs file of new node can't create, cpu on the node | |
1324 | * can't be hot-added. There is no rollback way now. | |
1325 | * So, check by BUG_ON() to catch it reluctantly.. | |
1326 | */ | |
1327 | BUG_ON(ret); | |
1328 | } | |
1329 | ||
d96ae530 | 1330 | /* create new memmap entry */ |
1331 | firmware_map_add_hotplug(start, start + size, "System RAM"); | |
1332 | ||
31bc3858 VK |
1333 | /* online pages if requested */ |
1334 | if (online) | |
1335 | walk_memory_range(PFN_DOWN(start), PFN_UP(start + size - 1), | |
1336 | NULL, online_memory_block); | |
1337 | ||
6ad696d2 AK |
1338 | goto out; |
1339 | ||
9af3c2de YG |
1340 | error: |
1341 | /* rollback pgdat allocation and others */ | |
1342 | if (new_pgdat) | |
1343 | rollback_node_hotadd(nid, pgdat); | |
7f36e3e5 | 1344 | memblock_remove(start, size); |
9af3c2de | 1345 | |
6ad696d2 | 1346 | out: |
bfc8c901 | 1347 | mem_hotplug_done(); |
bc02af93 YG |
1348 | return ret; |
1349 | } | |
62cedb9f DV |
1350 | EXPORT_SYMBOL_GPL(add_memory_resource); |
1351 | ||
1352 | int __ref add_memory(int nid, u64 start, u64 size) | |
1353 | { | |
1354 | struct resource *res; | |
1355 | int ret; | |
1356 | ||
1357 | res = register_memory_resource(start, size); | |
6f754ba4 VK |
1358 | if (IS_ERR(res)) |
1359 | return PTR_ERR(res); | |
62cedb9f | 1360 | |
31bc3858 | 1361 | ret = add_memory_resource(nid, res, memhp_auto_online); |
62cedb9f DV |
1362 | if (ret < 0) |
1363 | release_memory_resource(res); | |
1364 | return ret; | |
1365 | } | |
bc02af93 | 1366 | EXPORT_SYMBOL_GPL(add_memory); |
0c0e6195 KH |
1367 | |
1368 | #ifdef CONFIG_MEMORY_HOTREMOVE | |
5c755e9f BP |
1369 | /* |
1370 | * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy | |
1371 | * set and the size of the free page is given by page_order(). Using this, | |
1372 | * the function determines if the pageblock contains only free pages. | |
1373 | * Due to buddy contraints, a free page at least the size of a pageblock will | |
1374 | * be located at the start of the pageblock | |
1375 | */ | |
1376 | static inline int pageblock_free(struct page *page) | |
1377 | { | |
1378 | return PageBuddy(page) && page_order(page) >= pageblock_order; | |
1379 | } | |
1380 | ||
1381 | /* Return the start of the next active pageblock after a given page */ | |
1382 | static struct page *next_active_pageblock(struct page *page) | |
1383 | { | |
5c755e9f BP |
1384 | /* Ensure the starting page is pageblock-aligned */ |
1385 | BUG_ON(page_to_pfn(page) & (pageblock_nr_pages - 1)); | |
1386 | ||
5c755e9f | 1387 | /* If the entire pageblock is free, move to the end of free page */ |
0dcc48c1 KH |
1388 | if (pageblock_free(page)) { |
1389 | int order; | |
1390 | /* be careful. we don't have locks, page_order can be changed.*/ | |
1391 | order = page_order(page); | |
1392 | if ((order < MAX_ORDER) && (order >= pageblock_order)) | |
1393 | return page + (1 << order); | |
1394 | } | |
5c755e9f | 1395 | |
0dcc48c1 | 1396 | return page + pageblock_nr_pages; |
5c755e9f BP |
1397 | } |
1398 | ||
1399 | /* Checks if this range of memory is likely to be hot-removable. */ | |
1400 | int is_mem_section_removable(unsigned long start_pfn, unsigned long nr_pages) | |
1401 | { | |
5c755e9f BP |
1402 | struct page *page = pfn_to_page(start_pfn); |
1403 | struct page *end_page = page + nr_pages; | |
1404 | ||
1405 | /* Check the starting page of each pageblock within the range */ | |
1406 | for (; page < end_page; page = next_active_pageblock(page)) { | |
49ac8255 | 1407 | if (!is_pageblock_removable_nolock(page)) |
5c755e9f | 1408 | return 0; |
49ac8255 | 1409 | cond_resched(); |
5c755e9f BP |
1410 | } |
1411 | ||
1412 | /* All pageblocks in the memory block are likely to be hot-removable */ | |
1413 | return 1; | |
1414 | } | |
1415 | ||
0c0e6195 KH |
1416 | /* |
1417 | * Confirm all pages in a range [start, end) is belongs to the same zone. | |
1418 | */ | |
ed2f2400 | 1419 | int test_pages_in_a_zone(unsigned long start_pfn, unsigned long end_pfn) |
0c0e6195 | 1420 | { |
5f0f2887 | 1421 | unsigned long pfn, sec_end_pfn; |
0c0e6195 KH |
1422 | struct zone *zone = NULL; |
1423 | struct page *page; | |
1424 | int i; | |
5f0f2887 | 1425 | for (pfn = start_pfn, sec_end_pfn = SECTION_ALIGN_UP(start_pfn); |
0c0e6195 | 1426 | pfn < end_pfn; |
5f0f2887 AB |
1427 | pfn = sec_end_pfn + 1, sec_end_pfn += PAGES_PER_SECTION) { |
1428 | /* Make sure the memory section is present first */ | |
1429 | if (!present_section_nr(pfn_to_section_nr(pfn))) | |
0c0e6195 | 1430 | continue; |
5f0f2887 AB |
1431 | for (; pfn < sec_end_pfn && pfn < end_pfn; |
1432 | pfn += MAX_ORDER_NR_PAGES) { | |
1433 | i = 0; | |
1434 | /* This is just a CONFIG_HOLES_IN_ZONE check.*/ | |
1435 | while ((i < MAX_ORDER_NR_PAGES) && | |
1436 | !pfn_valid_within(pfn + i)) | |
1437 | i++; | |
1438 | if (i == MAX_ORDER_NR_PAGES) | |
1439 | continue; | |
1440 | page = pfn_to_page(pfn + i); | |
1441 | if (zone && page_zone(page) != zone) | |
1442 | return 0; | |
1443 | zone = page_zone(page); | |
1444 | } | |
0c0e6195 KH |
1445 | } |
1446 | return 1; | |
1447 | } | |
1448 | ||
1449 | /* | |
c8721bbb NH |
1450 | * Scan pfn range [start,end) to find movable/migratable pages (LRU pages |
1451 | * and hugepages). We scan pfn because it's much easier than scanning over | |
1452 | * linked list. This function returns the pfn of the first found movable | |
1453 | * page if it's found, otherwise 0. | |
0c0e6195 | 1454 | */ |
c8721bbb | 1455 | static unsigned long scan_movable_pages(unsigned long start, unsigned long end) |
0c0e6195 KH |
1456 | { |
1457 | unsigned long pfn; | |
1458 | struct page *page; | |
1459 | for (pfn = start; pfn < end; pfn++) { | |
1460 | if (pfn_valid(pfn)) { | |
1461 | page = pfn_to_page(pfn); | |
1462 | if (PageLRU(page)) | |
1463 | return pfn; | |
c8721bbb | 1464 | if (PageHuge(page)) { |
7e1f049e | 1465 | if (page_huge_active(page)) |
c8721bbb NH |
1466 | return pfn; |
1467 | else | |
1468 | pfn = round_up(pfn + 1, | |
1469 | 1 << compound_order(page)) - 1; | |
1470 | } | |
0c0e6195 KH |
1471 | } |
1472 | } | |
1473 | return 0; | |
1474 | } | |
1475 | ||
0c0e6195 KH |
1476 | #define NR_OFFLINE_AT_ONCE_PAGES (256) |
1477 | static int | |
1478 | do_migrate_range(unsigned long start_pfn, unsigned long end_pfn) | |
1479 | { | |
1480 | unsigned long pfn; | |
1481 | struct page *page; | |
1482 | int move_pages = NR_OFFLINE_AT_ONCE_PAGES; | |
1483 | int not_managed = 0; | |
1484 | int ret = 0; | |
1485 | LIST_HEAD(source); | |
1486 | ||
1487 | for (pfn = start_pfn; pfn < end_pfn && move_pages > 0; pfn++) { | |
1488 | if (!pfn_valid(pfn)) | |
1489 | continue; | |
1490 | page = pfn_to_page(pfn); | |
c8721bbb NH |
1491 | |
1492 | if (PageHuge(page)) { | |
1493 | struct page *head = compound_head(page); | |
1494 | pfn = page_to_pfn(head) + (1<<compound_order(head)) - 1; | |
1495 | if (compound_order(head) > PFN_SECTION_SHIFT) { | |
1496 | ret = -EBUSY; | |
1497 | break; | |
1498 | } | |
1499 | if (isolate_huge_page(page, &source)) | |
1500 | move_pages -= 1 << compound_order(head); | |
1501 | continue; | |
1502 | } | |
1503 | ||
700c2a46 | 1504 | if (!get_page_unless_zero(page)) |
0c0e6195 KH |
1505 | continue; |
1506 | /* | |
1507 | * We can skip free pages. And we can only deal with pages on | |
1508 | * LRU. | |
1509 | */ | |
62695a84 | 1510 | ret = isolate_lru_page(page); |
0c0e6195 | 1511 | if (!ret) { /* Success */ |
700c2a46 | 1512 | put_page(page); |
62695a84 | 1513 | list_add_tail(&page->lru, &source); |
0c0e6195 | 1514 | move_pages--; |
6d9c285a KM |
1515 | inc_zone_page_state(page, NR_ISOLATED_ANON + |
1516 | page_is_file_cache(page)); | |
1517 | ||
0c0e6195 | 1518 | } else { |
0c0e6195 | 1519 | #ifdef CONFIG_DEBUG_VM |
718a3821 WF |
1520 | printk(KERN_ALERT "removing pfn %lx from LRU failed\n", |
1521 | pfn); | |
f0b791a3 | 1522 | dump_page(page, "failed to remove from LRU"); |
0c0e6195 | 1523 | #endif |
700c2a46 | 1524 | put_page(page); |
25985edc | 1525 | /* Because we don't have big zone->lock. we should |
809c4449 BL |
1526 | check this again here. */ |
1527 | if (page_count(page)) { | |
1528 | not_managed++; | |
f3ab2636 | 1529 | ret = -EBUSY; |
809c4449 BL |
1530 | break; |
1531 | } | |
0c0e6195 KH |
1532 | } |
1533 | } | |
f3ab2636 BL |
1534 | if (!list_empty(&source)) { |
1535 | if (not_managed) { | |
c8721bbb | 1536 | putback_movable_pages(&source); |
f3ab2636 BL |
1537 | goto out; |
1538 | } | |
74c08f98 MK |
1539 | |
1540 | /* | |
1541 | * alloc_migrate_target should be improooooved!! | |
1542 | * migrate_pages returns # of failed pages. | |
1543 | */ | |
68711a74 | 1544 | ret = migrate_pages(&source, alloc_migrate_target, NULL, 0, |
9c620e2b | 1545 | MIGRATE_SYNC, MR_MEMORY_HOTPLUG); |
f3ab2636 | 1546 | if (ret) |
c8721bbb | 1547 | putback_movable_pages(&source); |
0c0e6195 | 1548 | } |
0c0e6195 KH |
1549 | out: |
1550 | return ret; | |
1551 | } | |
1552 | ||
1553 | /* | |
1554 | * remove from free_area[] and mark all as Reserved. | |
1555 | */ | |
1556 | static int | |
1557 | offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages, | |
1558 | void *data) | |
1559 | { | |
1560 | __offline_isolated_pages(start, start + nr_pages); | |
1561 | return 0; | |
1562 | } | |
1563 | ||
1564 | static void | |
1565 | offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn) | |
1566 | { | |
908eedc6 | 1567 | walk_system_ram_range(start_pfn, end_pfn - start_pfn, NULL, |
0c0e6195 KH |
1568 | offline_isolated_pages_cb); |
1569 | } | |
1570 | ||
1571 | /* | |
1572 | * Check all pages in range, recoreded as memory resource, are isolated. | |
1573 | */ | |
1574 | static int | |
1575 | check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages, | |
1576 | void *data) | |
1577 | { | |
1578 | int ret; | |
1579 | long offlined = *(long *)data; | |
b023f468 | 1580 | ret = test_pages_isolated(start_pfn, start_pfn + nr_pages, true); |
0c0e6195 KH |
1581 | offlined = nr_pages; |
1582 | if (!ret) | |
1583 | *(long *)data += offlined; | |
1584 | return ret; | |
1585 | } | |
1586 | ||
1587 | static long | |
1588 | check_pages_isolated(unsigned long start_pfn, unsigned long end_pfn) | |
1589 | { | |
1590 | long offlined = 0; | |
1591 | int ret; | |
1592 | ||
908eedc6 | 1593 | ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn, &offlined, |
0c0e6195 KH |
1594 | check_pages_isolated_cb); |
1595 | if (ret < 0) | |
1596 | offlined = (long)ret; | |
1597 | return offlined; | |
1598 | } | |
1599 | ||
09285af7 | 1600 | #ifdef CONFIG_MOVABLE_NODE |
79a4dcef TC |
1601 | /* |
1602 | * When CONFIG_MOVABLE_NODE, we permit offlining of a node which doesn't have | |
1603 | * normal memory. | |
1604 | */ | |
09285af7 LJ |
1605 | static bool can_offline_normal(struct zone *zone, unsigned long nr_pages) |
1606 | { | |
1607 | return true; | |
1608 | } | |
79a4dcef | 1609 | #else /* CONFIG_MOVABLE_NODE */ |
74d42d8f LJ |
1610 | /* ensure the node has NORMAL memory if it is still online */ |
1611 | static bool can_offline_normal(struct zone *zone, unsigned long nr_pages) | |
1612 | { | |
1613 | struct pglist_data *pgdat = zone->zone_pgdat; | |
1614 | unsigned long present_pages = 0; | |
1615 | enum zone_type zt; | |
1616 | ||
1617 | for (zt = 0; zt <= ZONE_NORMAL; zt++) | |
1618 | present_pages += pgdat->node_zones[zt].present_pages; | |
1619 | ||
1620 | if (present_pages > nr_pages) | |
1621 | return true; | |
1622 | ||
1623 | present_pages = 0; | |
1624 | for (; zt <= ZONE_MOVABLE; zt++) | |
1625 | present_pages += pgdat->node_zones[zt].present_pages; | |
1626 | ||
1627 | /* | |
1628 | * we can't offline the last normal memory until all | |
1629 | * higher memory is offlined. | |
1630 | */ | |
1631 | return present_pages == 0; | |
1632 | } | |
79a4dcef | 1633 | #endif /* CONFIG_MOVABLE_NODE */ |
74d42d8f | 1634 | |
c5320926 TC |
1635 | static int __init cmdline_parse_movable_node(char *p) |
1636 | { | |
1637 | #ifdef CONFIG_MOVABLE_NODE | |
1638 | /* | |
1639 | * Memory used by the kernel cannot be hot-removed because Linux | |
1640 | * cannot migrate the kernel pages. When memory hotplug is | |
1641 | * enabled, we should prevent memblock from allocating memory | |
1642 | * for the kernel. | |
1643 | * | |
1644 | * ACPI SRAT records all hotpluggable memory ranges. But before | |
1645 | * SRAT is parsed, we don't know about it. | |
1646 | * | |
1647 | * The kernel image is loaded into memory at very early time. We | |
1648 | * cannot prevent this anyway. So on NUMA system, we set any | |
1649 | * node the kernel resides in as un-hotpluggable. | |
1650 | * | |
1651 | * Since on modern servers, one node could have double-digit | |
1652 | * gigabytes memory, we can assume the memory around the kernel | |
1653 | * image is also un-hotpluggable. So before SRAT is parsed, just | |
1654 | * allocate memory near the kernel image to try the best to keep | |
1655 | * the kernel away from hotpluggable memory. | |
1656 | */ | |
1657 | memblock_set_bottom_up(true); | |
55ac590c | 1658 | movable_node_enabled = true; |
c5320926 TC |
1659 | #else |
1660 | pr_warn("movable_node option not supported\n"); | |
1661 | #endif | |
1662 | return 0; | |
1663 | } | |
1664 | early_param("movable_node", cmdline_parse_movable_node); | |
1665 | ||
d9713679 LJ |
1666 | /* check which state of node_states will be changed when offline memory */ |
1667 | static void node_states_check_changes_offline(unsigned long nr_pages, | |
1668 | struct zone *zone, struct memory_notify *arg) | |
1669 | { | |
1670 | struct pglist_data *pgdat = zone->zone_pgdat; | |
1671 | unsigned long present_pages = 0; | |
1672 | enum zone_type zt, zone_last = ZONE_NORMAL; | |
1673 | ||
1674 | /* | |
6715ddf9 LJ |
1675 | * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY] |
1676 | * contains nodes which have zones of 0...ZONE_NORMAL, | |
1677 | * set zone_last to ZONE_NORMAL. | |
d9713679 | 1678 | * |
6715ddf9 LJ |
1679 | * If we don't have HIGHMEM nor movable node, |
1680 | * node_states[N_NORMAL_MEMORY] contains nodes which have zones of | |
1681 | * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE. | |
d9713679 | 1682 | */ |
6715ddf9 | 1683 | if (N_MEMORY == N_NORMAL_MEMORY) |
d9713679 LJ |
1684 | zone_last = ZONE_MOVABLE; |
1685 | ||
1686 | /* | |
1687 | * check whether node_states[N_NORMAL_MEMORY] will be changed. | |
1688 | * If the memory to be offline is in a zone of 0...zone_last, | |
1689 | * and it is the last present memory, 0...zone_last will | |
1690 | * become empty after offline , thus we can determind we will | |
1691 | * need to clear the node from node_states[N_NORMAL_MEMORY]. | |
1692 | */ | |
1693 | for (zt = 0; zt <= zone_last; zt++) | |
1694 | present_pages += pgdat->node_zones[zt].present_pages; | |
1695 | if (zone_idx(zone) <= zone_last && nr_pages >= present_pages) | |
1696 | arg->status_change_nid_normal = zone_to_nid(zone); | |
1697 | else | |
1698 | arg->status_change_nid_normal = -1; | |
1699 | ||
6715ddf9 LJ |
1700 | #ifdef CONFIG_HIGHMEM |
1701 | /* | |
1702 | * If we have movable node, node_states[N_HIGH_MEMORY] | |
1703 | * contains nodes which have zones of 0...ZONE_HIGHMEM, | |
1704 | * set zone_last to ZONE_HIGHMEM. | |
1705 | * | |
1706 | * If we don't have movable node, node_states[N_NORMAL_MEMORY] | |
1707 | * contains nodes which have zones of 0...ZONE_MOVABLE, | |
1708 | * set zone_last to ZONE_MOVABLE. | |
1709 | */ | |
1710 | zone_last = ZONE_HIGHMEM; | |
1711 | if (N_MEMORY == N_HIGH_MEMORY) | |
1712 | zone_last = ZONE_MOVABLE; | |
1713 | ||
1714 | for (; zt <= zone_last; zt++) | |
1715 | present_pages += pgdat->node_zones[zt].present_pages; | |
1716 | if (zone_idx(zone) <= zone_last && nr_pages >= present_pages) | |
1717 | arg->status_change_nid_high = zone_to_nid(zone); | |
1718 | else | |
1719 | arg->status_change_nid_high = -1; | |
1720 | #else | |
1721 | arg->status_change_nid_high = arg->status_change_nid_normal; | |
1722 | #endif | |
1723 | ||
d9713679 LJ |
1724 | /* |
1725 | * node_states[N_HIGH_MEMORY] contains nodes which have 0...ZONE_MOVABLE | |
1726 | */ | |
1727 | zone_last = ZONE_MOVABLE; | |
1728 | ||
1729 | /* | |
1730 | * check whether node_states[N_HIGH_MEMORY] will be changed | |
1731 | * If we try to offline the last present @nr_pages from the node, | |
1732 | * we can determind we will need to clear the node from | |
1733 | * node_states[N_HIGH_MEMORY]. | |
1734 | */ | |
1735 | for (; zt <= zone_last; zt++) | |
1736 | present_pages += pgdat->node_zones[zt].present_pages; | |
1737 | if (nr_pages >= present_pages) | |
1738 | arg->status_change_nid = zone_to_nid(zone); | |
1739 | else | |
1740 | arg->status_change_nid = -1; | |
1741 | } | |
1742 | ||
1743 | static void node_states_clear_node(int node, struct memory_notify *arg) | |
1744 | { | |
1745 | if (arg->status_change_nid_normal >= 0) | |
1746 | node_clear_state(node, N_NORMAL_MEMORY); | |
1747 | ||
6715ddf9 LJ |
1748 | if ((N_MEMORY != N_NORMAL_MEMORY) && |
1749 | (arg->status_change_nid_high >= 0)) | |
d9713679 | 1750 | node_clear_state(node, N_HIGH_MEMORY); |
6715ddf9 LJ |
1751 | |
1752 | if ((N_MEMORY != N_HIGH_MEMORY) && | |
1753 | (arg->status_change_nid >= 0)) | |
1754 | node_clear_state(node, N_MEMORY); | |
d9713679 LJ |
1755 | } |
1756 | ||
a16cee10 | 1757 | static int __ref __offline_pages(unsigned long start_pfn, |
0c0e6195 KH |
1758 | unsigned long end_pfn, unsigned long timeout) |
1759 | { | |
1760 | unsigned long pfn, nr_pages, expire; | |
1761 | long offlined_pages; | |
7b78d335 | 1762 | int ret, drain, retry_max, node; |
d702909f | 1763 | unsigned long flags; |
0c0e6195 | 1764 | struct zone *zone; |
7b78d335 | 1765 | struct memory_notify arg; |
0c0e6195 | 1766 | |
0c0e6195 KH |
1767 | /* at least, alignment against pageblock is necessary */ |
1768 | if (!IS_ALIGNED(start_pfn, pageblock_nr_pages)) | |
1769 | return -EINVAL; | |
1770 | if (!IS_ALIGNED(end_pfn, pageblock_nr_pages)) | |
1771 | return -EINVAL; | |
1772 | /* This makes hotplug much easier...and readable. | |
1773 | we assume this for now. .*/ | |
1774 | if (!test_pages_in_a_zone(start_pfn, end_pfn)) | |
1775 | return -EINVAL; | |
7b78d335 YG |
1776 | |
1777 | zone = page_zone(pfn_to_page(start_pfn)); | |
1778 | node = zone_to_nid(zone); | |
1779 | nr_pages = end_pfn - start_pfn; | |
1780 | ||
74d42d8f | 1781 | if (zone_idx(zone) <= ZONE_NORMAL && !can_offline_normal(zone, nr_pages)) |
30467e0b | 1782 | return -EINVAL; |
74d42d8f | 1783 | |
0c0e6195 | 1784 | /* set above range as isolated */ |
b023f468 WC |
1785 | ret = start_isolate_page_range(start_pfn, end_pfn, |
1786 | MIGRATE_MOVABLE, true); | |
0c0e6195 | 1787 | if (ret) |
30467e0b | 1788 | return ret; |
7b78d335 YG |
1789 | |
1790 | arg.start_pfn = start_pfn; | |
1791 | arg.nr_pages = nr_pages; | |
d9713679 | 1792 | node_states_check_changes_offline(nr_pages, zone, &arg); |
7b78d335 YG |
1793 | |
1794 | ret = memory_notify(MEM_GOING_OFFLINE, &arg); | |
1795 | ret = notifier_to_errno(ret); | |
1796 | if (ret) | |
1797 | goto failed_removal; | |
1798 | ||
0c0e6195 KH |
1799 | pfn = start_pfn; |
1800 | expire = jiffies + timeout; | |
1801 | drain = 0; | |
1802 | retry_max = 5; | |
1803 | repeat: | |
1804 | /* start memory hot removal */ | |
1805 | ret = -EAGAIN; | |
1806 | if (time_after(jiffies, expire)) | |
1807 | goto failed_removal; | |
1808 | ret = -EINTR; | |
1809 | if (signal_pending(current)) | |
1810 | goto failed_removal; | |
1811 | ret = 0; | |
1812 | if (drain) { | |
1813 | lru_add_drain_all(); | |
0c0e6195 | 1814 | cond_resched(); |
c0554329 | 1815 | drain_all_pages(zone); |
0c0e6195 KH |
1816 | } |
1817 | ||
c8721bbb NH |
1818 | pfn = scan_movable_pages(start_pfn, end_pfn); |
1819 | if (pfn) { /* We have movable pages */ | |
0c0e6195 KH |
1820 | ret = do_migrate_range(pfn, end_pfn); |
1821 | if (!ret) { | |
1822 | drain = 1; | |
1823 | goto repeat; | |
1824 | } else { | |
1825 | if (ret < 0) | |
1826 | if (--retry_max == 0) | |
1827 | goto failed_removal; | |
1828 | yield(); | |
1829 | drain = 1; | |
1830 | goto repeat; | |
1831 | } | |
1832 | } | |
b3834be5 | 1833 | /* drain all zone's lru pagevec, this is asynchronous... */ |
0c0e6195 | 1834 | lru_add_drain_all(); |
0c0e6195 | 1835 | yield(); |
b3834be5 | 1836 | /* drain pcp pages, this is synchronous. */ |
c0554329 | 1837 | drain_all_pages(zone); |
c8721bbb NH |
1838 | /* |
1839 | * dissolve free hugepages in the memory block before doing offlining | |
1840 | * actually in order to make hugetlbfs's object counting consistent. | |
1841 | */ | |
1842 | dissolve_free_huge_pages(start_pfn, end_pfn); | |
0c0e6195 KH |
1843 | /* check again */ |
1844 | offlined_pages = check_pages_isolated(start_pfn, end_pfn); | |
1845 | if (offlined_pages < 0) { | |
1846 | ret = -EBUSY; | |
1847 | goto failed_removal; | |
1848 | } | |
1849 | printk(KERN_INFO "Offlined Pages %ld\n", offlined_pages); | |
b3834be5 | 1850 | /* Ok, all of our target is isolated. |
0c0e6195 KH |
1851 | We cannot do rollback at this point. */ |
1852 | offline_isolated_pages(start_pfn, end_pfn); | |
dbc0e4ce | 1853 | /* reset pagetype flags and makes migrate type to be MOVABLE */ |
0815f3d8 | 1854 | undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE); |
0c0e6195 | 1855 | /* removal success */ |
3dcc0571 | 1856 | adjust_managed_page_count(pfn_to_page(start_pfn), -offlined_pages); |
0c0e6195 | 1857 | zone->present_pages -= offlined_pages; |
d702909f CS |
1858 | |
1859 | pgdat_resize_lock(zone->zone_pgdat, &flags); | |
0c0e6195 | 1860 | zone->zone_pgdat->node_present_pages -= offlined_pages; |
d702909f | 1861 | pgdat_resize_unlock(zone->zone_pgdat, &flags); |
7b78d335 | 1862 | |
1b79acc9 KM |
1863 | init_per_zone_wmark_min(); |
1864 | ||
1e8537ba | 1865 | if (!populated_zone(zone)) { |
340175b7 | 1866 | zone_pcp_reset(zone); |
1e8537ba XQ |
1867 | mutex_lock(&zonelists_mutex); |
1868 | build_all_zonelists(NULL, NULL); | |
1869 | mutex_unlock(&zonelists_mutex); | |
1870 | } else | |
1871 | zone_pcp_update(zone); | |
340175b7 | 1872 | |
d9713679 LJ |
1873 | node_states_clear_node(node, &arg); |
1874 | if (arg.status_change_nid >= 0) | |
8fe23e05 | 1875 | kswapd_stop(node); |
bce7394a | 1876 | |
0c0e6195 KH |
1877 | vm_total_pages = nr_free_pagecache_pages(); |
1878 | writeback_set_ratelimit(); | |
7b78d335 YG |
1879 | |
1880 | memory_notify(MEM_OFFLINE, &arg); | |
0c0e6195 KH |
1881 | return 0; |
1882 | ||
1883 | failed_removal: | |
a62e2f4f BH |
1884 | printk(KERN_INFO "memory offlining [mem %#010llx-%#010llx] failed\n", |
1885 | (unsigned long long) start_pfn << PAGE_SHIFT, | |
1886 | ((unsigned long long) end_pfn << PAGE_SHIFT) - 1); | |
7b78d335 | 1887 | memory_notify(MEM_CANCEL_OFFLINE, &arg); |
0c0e6195 | 1888 | /* pushback to free area */ |
0815f3d8 | 1889 | undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE); |
0c0e6195 KH |
1890 | return ret; |
1891 | } | |
71088785 | 1892 | |
30467e0b | 1893 | /* Must be protected by mem_hotplug_begin() */ |
a16cee10 WC |
1894 | int offline_pages(unsigned long start_pfn, unsigned long nr_pages) |
1895 | { | |
1896 | return __offline_pages(start_pfn, start_pfn + nr_pages, 120 * HZ); | |
1897 | } | |
e2ff3940 | 1898 | #endif /* CONFIG_MEMORY_HOTREMOVE */ |
a16cee10 | 1899 | |
bbc76be6 WC |
1900 | /** |
1901 | * walk_memory_range - walks through all mem sections in [start_pfn, end_pfn) | |
1902 | * @start_pfn: start pfn of the memory range | |
e05c4bbf | 1903 | * @end_pfn: end pfn of the memory range |
bbc76be6 WC |
1904 | * @arg: argument passed to func |
1905 | * @func: callback for each memory section walked | |
1906 | * | |
1907 | * This function walks through all present mem sections in range | |
1908 | * [start_pfn, end_pfn) and call func on each mem section. | |
1909 | * | |
1910 | * Returns the return value of func. | |
1911 | */ | |
e2ff3940 | 1912 | int walk_memory_range(unsigned long start_pfn, unsigned long end_pfn, |
bbc76be6 | 1913 | void *arg, int (*func)(struct memory_block *, void *)) |
71088785 | 1914 | { |
e90bdb7f WC |
1915 | struct memory_block *mem = NULL; |
1916 | struct mem_section *section; | |
e90bdb7f WC |
1917 | unsigned long pfn, section_nr; |
1918 | int ret; | |
e90bdb7f WC |
1919 | |
1920 | for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) { | |
1921 | section_nr = pfn_to_section_nr(pfn); | |
1922 | if (!present_section_nr(section_nr)) | |
1923 | continue; | |
1924 | ||
1925 | section = __nr_to_section(section_nr); | |
1926 | /* same memblock? */ | |
1927 | if (mem) | |
1928 | if ((section_nr >= mem->start_section_nr) && | |
1929 | (section_nr <= mem->end_section_nr)) | |
1930 | continue; | |
1931 | ||
1932 | mem = find_memory_block_hinted(section, mem); | |
1933 | if (!mem) | |
1934 | continue; | |
1935 | ||
bbc76be6 | 1936 | ret = func(mem, arg); |
e90bdb7f | 1937 | if (ret) { |
bbc76be6 WC |
1938 | kobject_put(&mem->dev.kobj); |
1939 | return ret; | |
e90bdb7f WC |
1940 | } |
1941 | } | |
1942 | ||
1943 | if (mem) | |
1944 | kobject_put(&mem->dev.kobj); | |
1945 | ||
bbc76be6 WC |
1946 | return 0; |
1947 | } | |
1948 | ||
e2ff3940 | 1949 | #ifdef CONFIG_MEMORY_HOTREMOVE |
d6de9d53 | 1950 | static int check_memblock_offlined_cb(struct memory_block *mem, void *arg) |
bbc76be6 WC |
1951 | { |
1952 | int ret = !is_memblock_offlined(mem); | |
1953 | ||
349daa0f RD |
1954 | if (unlikely(ret)) { |
1955 | phys_addr_t beginpa, endpa; | |
1956 | ||
1957 | beginpa = PFN_PHYS(section_nr_to_pfn(mem->start_section_nr)); | |
1958 | endpa = PFN_PHYS(section_nr_to_pfn(mem->end_section_nr + 1))-1; | |
bbc76be6 | 1959 | pr_warn("removing memory fails, because memory " |
349daa0f RD |
1960 | "[%pa-%pa] is onlined\n", |
1961 | &beginpa, &endpa); | |
1962 | } | |
bbc76be6 WC |
1963 | |
1964 | return ret; | |
1965 | } | |
1966 | ||
0f1cfe9d | 1967 | static int check_cpu_on_node(pg_data_t *pgdat) |
60a5a19e | 1968 | { |
60a5a19e TC |
1969 | int cpu; |
1970 | ||
1971 | for_each_present_cpu(cpu) { | |
1972 | if (cpu_to_node(cpu) == pgdat->node_id) | |
1973 | /* | |
1974 | * the cpu on this node isn't removed, and we can't | |
1975 | * offline this node. | |
1976 | */ | |
1977 | return -EBUSY; | |
1978 | } | |
1979 | ||
1980 | return 0; | |
1981 | } | |
1982 | ||
0f1cfe9d | 1983 | static void unmap_cpu_on_node(pg_data_t *pgdat) |
e13fe869 WC |
1984 | { |
1985 | #ifdef CONFIG_ACPI_NUMA | |
e13fe869 WC |
1986 | int cpu; |
1987 | ||
1988 | for_each_possible_cpu(cpu) | |
1989 | if (cpu_to_node(cpu) == pgdat->node_id) | |
1990 | numa_clear_node(cpu); | |
1991 | #endif | |
1992 | } | |
1993 | ||
0f1cfe9d | 1994 | static int check_and_unmap_cpu_on_node(pg_data_t *pgdat) |
e13fe869 | 1995 | { |
0f1cfe9d | 1996 | int ret; |
e13fe869 | 1997 | |
0f1cfe9d | 1998 | ret = check_cpu_on_node(pgdat); |
e13fe869 WC |
1999 | if (ret) |
2000 | return ret; | |
2001 | ||
2002 | /* | |
2003 | * the node will be offlined when we come here, so we can clear | |
2004 | * the cpu_to_node() now. | |
2005 | */ | |
2006 | ||
0f1cfe9d | 2007 | unmap_cpu_on_node(pgdat); |
e13fe869 WC |
2008 | return 0; |
2009 | } | |
2010 | ||
0f1cfe9d TK |
2011 | /** |
2012 | * try_offline_node | |
2013 | * | |
2014 | * Offline a node if all memory sections and cpus of the node are removed. | |
2015 | * | |
2016 | * NOTE: The caller must call lock_device_hotplug() to serialize hotplug | |
2017 | * and online/offline operations before this call. | |
2018 | */ | |
90b30cdc | 2019 | void try_offline_node(int nid) |
60a5a19e | 2020 | { |
d822b86a WC |
2021 | pg_data_t *pgdat = NODE_DATA(nid); |
2022 | unsigned long start_pfn = pgdat->node_start_pfn; | |
2023 | unsigned long end_pfn = start_pfn + pgdat->node_spanned_pages; | |
60a5a19e | 2024 | unsigned long pfn; |
d822b86a | 2025 | int i; |
60a5a19e TC |
2026 | |
2027 | for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) { | |
2028 | unsigned long section_nr = pfn_to_section_nr(pfn); | |
2029 | ||
2030 | if (!present_section_nr(section_nr)) | |
2031 | continue; | |
2032 | ||
2033 | if (pfn_to_nid(pfn) != nid) | |
2034 | continue; | |
2035 | ||
2036 | /* | |
2037 | * some memory sections of this node are not removed, and we | |
2038 | * can't offline node now. | |
2039 | */ | |
2040 | return; | |
2041 | } | |
2042 | ||
0f1cfe9d | 2043 | if (check_and_unmap_cpu_on_node(pgdat)) |
60a5a19e TC |
2044 | return; |
2045 | ||
2046 | /* | |
2047 | * all memory/cpu of this node are removed, we can offline this | |
2048 | * node now. | |
2049 | */ | |
2050 | node_set_offline(nid); | |
2051 | unregister_one_node(nid); | |
d822b86a | 2052 | |
d822b86a WC |
2053 | /* free waittable in each zone */ |
2054 | for (i = 0; i < MAX_NR_ZONES; i++) { | |
2055 | struct zone *zone = pgdat->node_zones + i; | |
2056 | ||
ca4b3f30 JW |
2057 | /* |
2058 | * wait_table may be allocated from boot memory, | |
2059 | * here only free if it's allocated by vmalloc. | |
2060 | */ | |
85bd8399 | 2061 | if (is_vmalloc_addr(zone->wait_table)) { |
d822b86a | 2062 | vfree(zone->wait_table); |
85bd8399 GZ |
2063 | zone->wait_table = NULL; |
2064 | } | |
d822b86a | 2065 | } |
60a5a19e | 2066 | } |
90b30cdc | 2067 | EXPORT_SYMBOL(try_offline_node); |
60a5a19e | 2068 | |
0f1cfe9d TK |
2069 | /** |
2070 | * remove_memory | |
2071 | * | |
2072 | * NOTE: The caller must call lock_device_hotplug() to serialize hotplug | |
2073 | * and online/offline operations before this call, as required by | |
2074 | * try_offline_node(). | |
2075 | */ | |
242831eb | 2076 | void __ref remove_memory(int nid, u64 start, u64 size) |
bbc76be6 | 2077 | { |
242831eb | 2078 | int ret; |
993c1aad | 2079 | |
27356f54 TK |
2080 | BUG_ON(check_hotplug_memory_range(start, size)); |
2081 | ||
bfc8c901 | 2082 | mem_hotplug_begin(); |
6677e3ea YI |
2083 | |
2084 | /* | |
242831eb RW |
2085 | * All memory blocks must be offlined before removing memory. Check |
2086 | * whether all memory blocks in question are offline and trigger a BUG() | |
2087 | * if this is not the case. | |
6677e3ea | 2088 | */ |
242831eb | 2089 | ret = walk_memory_range(PFN_DOWN(start), PFN_UP(start + size - 1), NULL, |
d6de9d53 | 2090 | check_memblock_offlined_cb); |
bfc8c901 | 2091 | if (ret) |
242831eb | 2092 | BUG(); |
6677e3ea | 2093 | |
46c66c4b YI |
2094 | /* remove memmap entry */ |
2095 | firmware_map_remove(start, start + size, "System RAM"); | |
f9126ab9 XQ |
2096 | memblock_free(start, size); |
2097 | memblock_remove(start, size); | |
46c66c4b | 2098 | |
24d335ca WC |
2099 | arch_remove_memory(start, size); |
2100 | ||
60a5a19e TC |
2101 | try_offline_node(nid); |
2102 | ||
bfc8c901 | 2103 | mem_hotplug_done(); |
71088785 | 2104 | } |
71088785 | 2105 | EXPORT_SYMBOL_GPL(remove_memory); |
aba6efc4 | 2106 | #endif /* CONFIG_MEMORY_HOTREMOVE */ |