]>
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> |
698b1b30 | 36 | #include <linux/compaction.h> |
3947be19 DH |
37 | |
38 | #include <asm/tlbflush.h> | |
39 | ||
1e5ad9a3 AB |
40 | #include "internal.h" |
41 | ||
9d0ad8ca DK |
42 | /* |
43 | * online_page_callback contains pointer to current page onlining function. | |
44 | * Initially it is generic_online_page(). If it is required it could be | |
45 | * changed by calling set_online_page_callback() for callback registration | |
46 | * and restore_online_page_callback() for generic callback restore. | |
47 | */ | |
48 | ||
49 | static void generic_online_page(struct page *page); | |
50 | ||
51 | static online_page_callback_t online_page_callback = generic_online_page; | |
bfc8c901 | 52 | static DEFINE_MUTEX(online_page_callback_lock); |
9d0ad8ca | 53 | |
bfc8c901 VD |
54 | /* The same as the cpu_hotplug lock, but for memory hotplug. */ |
55 | static struct { | |
56 | struct task_struct *active_writer; | |
57 | struct mutex lock; /* Synchronizes accesses to refcount, */ | |
58 | /* | |
59 | * Also blocks the new readers during | |
60 | * an ongoing mem hotplug operation. | |
61 | */ | |
62 | int refcount; | |
63 | ||
64 | #ifdef CONFIG_DEBUG_LOCK_ALLOC | |
65 | struct lockdep_map dep_map; | |
66 | #endif | |
67 | } mem_hotplug = { | |
68 | .active_writer = NULL, | |
69 | .lock = __MUTEX_INITIALIZER(mem_hotplug.lock), | |
70 | .refcount = 0, | |
71 | #ifdef CONFIG_DEBUG_LOCK_ALLOC | |
72 | .dep_map = {.name = "mem_hotplug.lock" }, | |
73 | #endif | |
74 | }; | |
75 | ||
76 | /* Lockdep annotations for get/put_online_mems() and mem_hotplug_begin/end() */ | |
77 | #define memhp_lock_acquire_read() lock_map_acquire_read(&mem_hotplug.dep_map) | |
78 | #define memhp_lock_acquire() lock_map_acquire(&mem_hotplug.dep_map) | |
79 | #define memhp_lock_release() lock_map_release(&mem_hotplug.dep_map) | |
80 | ||
8604d9e5 | 81 | #ifndef CONFIG_MEMORY_HOTPLUG_DEFAULT_ONLINE |
31bc3858 | 82 | bool memhp_auto_online; |
8604d9e5 VK |
83 | #else |
84 | bool memhp_auto_online = true; | |
85 | #endif | |
31bc3858 VK |
86 | EXPORT_SYMBOL_GPL(memhp_auto_online); |
87 | ||
86dd995d VK |
88 | static int __init setup_memhp_default_state(char *str) |
89 | { | |
90 | if (!strcmp(str, "online")) | |
91 | memhp_auto_online = true; | |
92 | else if (!strcmp(str, "offline")) | |
93 | memhp_auto_online = false; | |
94 | ||
95 | return 1; | |
96 | } | |
97 | __setup("memhp_default_state=", setup_memhp_default_state); | |
98 | ||
bfc8c901 VD |
99 | void get_online_mems(void) |
100 | { | |
101 | might_sleep(); | |
102 | if (mem_hotplug.active_writer == current) | |
103 | return; | |
104 | memhp_lock_acquire_read(); | |
105 | mutex_lock(&mem_hotplug.lock); | |
106 | mem_hotplug.refcount++; | |
107 | mutex_unlock(&mem_hotplug.lock); | |
108 | ||
109 | } | |
20d6c96b | 110 | |
bfc8c901 | 111 | void put_online_mems(void) |
20d6c96b | 112 | { |
bfc8c901 VD |
113 | if (mem_hotplug.active_writer == current) |
114 | return; | |
115 | mutex_lock(&mem_hotplug.lock); | |
116 | ||
117 | if (WARN_ON(!mem_hotplug.refcount)) | |
118 | mem_hotplug.refcount++; /* try to fix things up */ | |
119 | ||
120 | if (!--mem_hotplug.refcount && unlikely(mem_hotplug.active_writer)) | |
121 | wake_up_process(mem_hotplug.active_writer); | |
122 | mutex_unlock(&mem_hotplug.lock); | |
123 | memhp_lock_release(); | |
124 | ||
20d6c96b KM |
125 | } |
126 | ||
30467e0b | 127 | void mem_hotplug_begin(void) |
20d6c96b | 128 | { |
bfc8c901 VD |
129 | mem_hotplug.active_writer = current; |
130 | ||
131 | memhp_lock_acquire(); | |
132 | for (;;) { | |
133 | mutex_lock(&mem_hotplug.lock); | |
134 | if (likely(!mem_hotplug.refcount)) | |
135 | break; | |
136 | __set_current_state(TASK_UNINTERRUPTIBLE); | |
137 | mutex_unlock(&mem_hotplug.lock); | |
138 | schedule(); | |
139 | } | |
20d6c96b KM |
140 | } |
141 | ||
30467e0b | 142 | void mem_hotplug_done(void) |
bfc8c901 VD |
143 | { |
144 | mem_hotplug.active_writer = NULL; | |
145 | mutex_unlock(&mem_hotplug.lock); | |
146 | memhp_lock_release(); | |
147 | } | |
20d6c96b | 148 | |
45e0b78b KM |
149 | /* add this memory to iomem resource */ |
150 | static struct resource *register_memory_resource(u64 start, u64 size) | |
151 | { | |
152 | struct resource *res; | |
153 | res = kzalloc(sizeof(struct resource), GFP_KERNEL); | |
6f754ba4 VK |
154 | if (!res) |
155 | return ERR_PTR(-ENOMEM); | |
45e0b78b KM |
156 | |
157 | res->name = "System RAM"; | |
158 | res->start = start; | |
159 | res->end = start + size - 1; | |
782b8664 | 160 | res->flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY; |
45e0b78b | 161 | if (request_resource(&iomem_resource, res) < 0) { |
4996eed8 | 162 | pr_debug("System RAM resource %pR cannot be added\n", res); |
45e0b78b | 163 | kfree(res); |
6f754ba4 | 164 | return ERR_PTR(-EEXIST); |
45e0b78b KM |
165 | } |
166 | return res; | |
167 | } | |
168 | ||
169 | static void release_memory_resource(struct resource *res) | |
170 | { | |
171 | if (!res) | |
172 | return; | |
173 | release_resource(res); | |
174 | kfree(res); | |
175 | return; | |
176 | } | |
177 | ||
53947027 | 178 | #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE |
46723bfa YI |
179 | void get_page_bootmem(unsigned long info, struct page *page, |
180 | unsigned long type) | |
04753278 | 181 | { |
5f24ce5f | 182 | page->lru.next = (struct list_head *) type; |
04753278 YG |
183 | SetPagePrivate(page); |
184 | set_page_private(page, info); | |
fe896d18 | 185 | page_ref_inc(page); |
04753278 YG |
186 | } |
187 | ||
170a5a7e | 188 | void put_page_bootmem(struct page *page) |
04753278 | 189 | { |
5f24ce5f | 190 | unsigned long type; |
04753278 | 191 | |
5f24ce5f AA |
192 | type = (unsigned long) page->lru.next; |
193 | BUG_ON(type < MEMORY_HOTPLUG_MIN_BOOTMEM_TYPE || | |
194 | type > MEMORY_HOTPLUG_MAX_BOOTMEM_TYPE); | |
04753278 | 195 | |
fe896d18 | 196 | if (page_ref_dec_return(page) == 1) { |
04753278 YG |
197 | ClearPagePrivate(page); |
198 | set_page_private(page, 0); | |
5f24ce5f | 199 | INIT_LIST_HEAD(&page->lru); |
170a5a7e | 200 | free_reserved_page(page); |
04753278 | 201 | } |
04753278 YG |
202 | } |
203 | ||
46723bfa YI |
204 | #ifdef CONFIG_HAVE_BOOTMEM_INFO_NODE |
205 | #ifndef CONFIG_SPARSEMEM_VMEMMAP | |
d92bc318 | 206 | static void register_page_bootmem_info_section(unsigned long start_pfn) |
04753278 YG |
207 | { |
208 | unsigned long *usemap, mapsize, section_nr, i; | |
209 | struct mem_section *ms; | |
210 | struct page *page, *memmap; | |
211 | ||
04753278 YG |
212 | section_nr = pfn_to_section_nr(start_pfn); |
213 | ms = __nr_to_section(section_nr); | |
214 | ||
215 | /* Get section's memmap address */ | |
216 | memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr); | |
217 | ||
218 | /* | |
219 | * Get page for the memmap's phys address | |
220 | * XXX: need more consideration for sparse_vmemmap... | |
221 | */ | |
222 | page = virt_to_page(memmap); | |
223 | mapsize = sizeof(struct page) * PAGES_PER_SECTION; | |
224 | mapsize = PAGE_ALIGN(mapsize) >> PAGE_SHIFT; | |
225 | ||
226 | /* remember memmap's page */ | |
227 | for (i = 0; i < mapsize; i++, page++) | |
228 | get_page_bootmem(section_nr, page, SECTION_INFO); | |
229 | ||
230 | usemap = __nr_to_section(section_nr)->pageblock_flags; | |
231 | page = virt_to_page(usemap); | |
232 | ||
233 | mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT; | |
234 | ||
235 | for (i = 0; i < mapsize; i++, page++) | |
af370fb8 | 236 | get_page_bootmem(section_nr, page, MIX_SECTION_INFO); |
04753278 YG |
237 | |
238 | } | |
46723bfa YI |
239 | #else /* CONFIG_SPARSEMEM_VMEMMAP */ |
240 | static void register_page_bootmem_info_section(unsigned long start_pfn) | |
241 | { | |
242 | unsigned long *usemap, mapsize, section_nr, i; | |
243 | struct mem_section *ms; | |
244 | struct page *page, *memmap; | |
245 | ||
246 | if (!pfn_valid(start_pfn)) | |
247 | return; | |
248 | ||
249 | section_nr = pfn_to_section_nr(start_pfn); | |
250 | ms = __nr_to_section(section_nr); | |
251 | ||
252 | memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr); | |
253 | ||
254 | register_page_bootmem_memmap(section_nr, memmap, PAGES_PER_SECTION); | |
255 | ||
256 | usemap = __nr_to_section(section_nr)->pageblock_flags; | |
257 | page = virt_to_page(usemap); | |
258 | ||
259 | mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT; | |
260 | ||
261 | for (i = 0; i < mapsize; i++, page++) | |
262 | get_page_bootmem(section_nr, page, MIX_SECTION_INFO); | |
263 | } | |
264 | #endif /* !CONFIG_SPARSEMEM_VMEMMAP */ | |
04753278 | 265 | |
7ded384a | 266 | void __init register_page_bootmem_info_node(struct pglist_data *pgdat) |
04753278 YG |
267 | { |
268 | unsigned long i, pfn, end_pfn, nr_pages; | |
269 | int node = pgdat->node_id; | |
270 | struct page *page; | |
04753278 YG |
271 | |
272 | nr_pages = PAGE_ALIGN(sizeof(struct pglist_data)) >> PAGE_SHIFT; | |
273 | page = virt_to_page(pgdat); | |
274 | ||
275 | for (i = 0; i < nr_pages; i++, page++) | |
276 | get_page_bootmem(node, page, NODE_INFO); | |
277 | ||
04753278 | 278 | pfn = pgdat->node_start_pfn; |
c1f19495 | 279 | end_pfn = pgdat_end_pfn(pgdat); |
04753278 | 280 | |
7e9f5eb0 | 281 | /* register section info */ |
f14851af | 282 | for (; pfn < end_pfn; pfn += PAGES_PER_SECTION) { |
283 | /* | |
284 | * Some platforms can assign the same pfn to multiple nodes - on | |
285 | * node0 as well as nodeN. To avoid registering a pfn against | |
286 | * multiple nodes we check that this pfn does not already | |
7e9f5eb0 | 287 | * reside in some other nodes. |
f14851af | 288 | */ |
f65e91df | 289 | if (pfn_valid(pfn) && (early_pfn_to_nid(pfn) == node)) |
f14851af | 290 | register_page_bootmem_info_section(pfn); |
291 | } | |
04753278 | 292 | } |
46723bfa | 293 | #endif /* CONFIG_HAVE_BOOTMEM_INFO_NODE */ |
04753278 | 294 | |
f2765404 FF |
295 | static void __meminit grow_zone_span(struct zone *zone, unsigned long start_pfn, |
296 | unsigned long end_pfn) | |
76cdd58e HC |
297 | { |
298 | unsigned long old_zone_end_pfn; | |
299 | ||
300 | zone_span_writelock(zone); | |
301 | ||
c33bc315 | 302 | old_zone_end_pfn = zone_end_pfn(zone); |
8080fc03 | 303 | if (zone_is_empty(zone) || start_pfn < zone->zone_start_pfn) |
76cdd58e HC |
304 | zone->zone_start_pfn = start_pfn; |
305 | ||
306 | zone->spanned_pages = max(old_zone_end_pfn, end_pfn) - | |
307 | zone->zone_start_pfn; | |
308 | ||
309 | zone_span_writeunlock(zone); | |
310 | } | |
311 | ||
511c2aba LJ |
312 | static void resize_zone(struct zone *zone, unsigned long start_pfn, |
313 | unsigned long end_pfn) | |
314 | { | |
315 | zone_span_writelock(zone); | |
316 | ||
e455a9b9 LJ |
317 | if (end_pfn - start_pfn) { |
318 | zone->zone_start_pfn = start_pfn; | |
319 | zone->spanned_pages = end_pfn - start_pfn; | |
320 | } else { | |
321 | /* | |
322 | * make it consist as free_area_init_core(), | |
323 | * if spanned_pages = 0, then keep start_pfn = 0 | |
324 | */ | |
325 | zone->zone_start_pfn = 0; | |
326 | zone->spanned_pages = 0; | |
327 | } | |
511c2aba LJ |
328 | |
329 | zone_span_writeunlock(zone); | |
330 | } | |
331 | ||
332 | static void fix_zone_id(struct zone *zone, unsigned long start_pfn, | |
333 | unsigned long end_pfn) | |
334 | { | |
335 | enum zone_type zid = zone_idx(zone); | |
336 | int nid = zone->zone_pgdat->node_id; | |
337 | unsigned long pfn; | |
338 | ||
339 | for (pfn = start_pfn; pfn < end_pfn; pfn++) | |
340 | set_page_links(pfn_to_page(pfn), zid, nid, pfn); | |
341 | } | |
342 | ||
f6bbb78e | 343 | /* Can fail with -ENOMEM from allocating a wait table with vmalloc() or |
9e43aa2b | 344 | * alloc_bootmem_node_nopanic()/memblock_virt_alloc_node_nopanic() */ |
f6bbb78e CS |
345 | static int __ref ensure_zone_is_initialized(struct zone *zone, |
346 | unsigned long start_pfn, unsigned long num_pages) | |
347 | { | |
348 | if (!zone_is_initialized(zone)) | |
b171e409 YB |
349 | return init_currently_empty_zone(zone, start_pfn, num_pages); |
350 | ||
f6bbb78e CS |
351 | return 0; |
352 | } | |
353 | ||
e455a9b9 | 354 | static int __meminit move_pfn_range_left(struct zone *z1, struct zone *z2, |
511c2aba LJ |
355 | unsigned long start_pfn, unsigned long end_pfn) |
356 | { | |
e455a9b9 | 357 | int ret; |
511c2aba | 358 | unsigned long flags; |
e455a9b9 LJ |
359 | unsigned long z1_start_pfn; |
360 | ||
64dd1b29 CS |
361 | ret = ensure_zone_is_initialized(z1, start_pfn, end_pfn - start_pfn); |
362 | if (ret) | |
363 | return ret; | |
511c2aba LJ |
364 | |
365 | pgdat_resize_lock(z1->zone_pgdat, &flags); | |
366 | ||
367 | /* can't move pfns which are higher than @z2 */ | |
108bcc96 | 368 | if (end_pfn > zone_end_pfn(z2)) |
511c2aba | 369 | goto out_fail; |
834405c3 | 370 | /* the move out part must be at the left most of @z2 */ |
511c2aba LJ |
371 | if (start_pfn > z2->zone_start_pfn) |
372 | goto out_fail; | |
373 | /* must included/overlap */ | |
374 | if (end_pfn <= z2->zone_start_pfn) | |
375 | goto out_fail; | |
376 | ||
e455a9b9 | 377 | /* use start_pfn for z1's start_pfn if z1 is empty */ |
8080fc03 | 378 | if (!zone_is_empty(z1)) |
e455a9b9 LJ |
379 | z1_start_pfn = z1->zone_start_pfn; |
380 | else | |
381 | z1_start_pfn = start_pfn; | |
382 | ||
383 | resize_zone(z1, z1_start_pfn, end_pfn); | |
108bcc96 | 384 | resize_zone(z2, end_pfn, zone_end_pfn(z2)); |
511c2aba LJ |
385 | |
386 | pgdat_resize_unlock(z1->zone_pgdat, &flags); | |
387 | ||
388 | fix_zone_id(z1, start_pfn, end_pfn); | |
389 | ||
390 | return 0; | |
391 | out_fail: | |
392 | pgdat_resize_unlock(z1->zone_pgdat, &flags); | |
393 | return -1; | |
394 | } | |
395 | ||
e455a9b9 | 396 | static int __meminit move_pfn_range_right(struct zone *z1, struct zone *z2, |
511c2aba LJ |
397 | unsigned long start_pfn, unsigned long end_pfn) |
398 | { | |
e455a9b9 | 399 | int ret; |
511c2aba | 400 | unsigned long flags; |
e455a9b9 LJ |
401 | unsigned long z2_end_pfn; |
402 | ||
64dd1b29 CS |
403 | ret = ensure_zone_is_initialized(z2, start_pfn, end_pfn - start_pfn); |
404 | if (ret) | |
405 | return ret; | |
511c2aba LJ |
406 | |
407 | pgdat_resize_lock(z1->zone_pgdat, &flags); | |
408 | ||
409 | /* can't move pfns which are lower than @z1 */ | |
410 | if (z1->zone_start_pfn > start_pfn) | |
411 | goto out_fail; | |
412 | /* the move out part mast at the right most of @z1 */ | |
108bcc96 | 413 | if (zone_end_pfn(z1) > end_pfn) |
511c2aba LJ |
414 | goto out_fail; |
415 | /* must included/overlap */ | |
108bcc96 | 416 | if (start_pfn >= zone_end_pfn(z1)) |
511c2aba LJ |
417 | goto out_fail; |
418 | ||
e455a9b9 | 419 | /* use end_pfn for z2's end_pfn if z2 is empty */ |
8080fc03 | 420 | if (!zone_is_empty(z2)) |
108bcc96 | 421 | z2_end_pfn = zone_end_pfn(z2); |
e455a9b9 LJ |
422 | else |
423 | z2_end_pfn = end_pfn; | |
424 | ||
511c2aba | 425 | resize_zone(z1, z1->zone_start_pfn, start_pfn); |
e455a9b9 | 426 | resize_zone(z2, start_pfn, z2_end_pfn); |
511c2aba LJ |
427 | |
428 | pgdat_resize_unlock(z1->zone_pgdat, &flags); | |
429 | ||
430 | fix_zone_id(z2, start_pfn, end_pfn); | |
431 | ||
432 | return 0; | |
433 | out_fail: | |
434 | pgdat_resize_unlock(z1->zone_pgdat, &flags); | |
435 | return -1; | |
436 | } | |
437 | ||
e51e6c8f RA |
438 | static struct zone * __meminit move_pfn_range(int zone_shift, |
439 | unsigned long start_pfn, unsigned long end_pfn) | |
440 | { | |
441 | struct zone *zone = page_zone(pfn_to_page(start_pfn)); | |
442 | int ret = 0; | |
443 | ||
444 | if (zone_shift < 0) | |
445 | ret = move_pfn_range_left(zone + zone_shift, zone, | |
446 | start_pfn, end_pfn); | |
447 | else if (zone_shift) | |
448 | ret = move_pfn_range_right(zone, zone + zone_shift, | |
449 | start_pfn, end_pfn); | |
450 | ||
451 | if (ret) | |
452 | return NULL; | |
453 | ||
454 | return zone + zone_shift; | |
455 | } | |
456 | ||
f2765404 FF |
457 | static void __meminit grow_pgdat_span(struct pglist_data *pgdat, unsigned long start_pfn, |
458 | unsigned long end_pfn) | |
76cdd58e | 459 | { |
83285c72 | 460 | unsigned long old_pgdat_end_pfn = pgdat_end_pfn(pgdat); |
76cdd58e | 461 | |
712cd386 | 462 | if (!pgdat->node_spanned_pages || start_pfn < pgdat->node_start_pfn) |
76cdd58e HC |
463 | pgdat->node_start_pfn = start_pfn; |
464 | ||
465 | pgdat->node_spanned_pages = max(old_pgdat_end_pfn, end_pfn) - | |
466 | pgdat->node_start_pfn; | |
467 | } | |
468 | ||
31168481 | 469 | static int __meminit __add_zone(struct zone *zone, unsigned long phys_start_pfn) |
3947be19 DH |
470 | { |
471 | struct pglist_data *pgdat = zone->zone_pgdat; | |
472 | int nr_pages = PAGES_PER_SECTION; | |
473 | int nid = pgdat->node_id; | |
474 | int zone_type; | |
e298ff75 | 475 | unsigned long flags, pfn; |
64dd1b29 | 476 | int ret; |
3947be19 DH |
477 | |
478 | zone_type = zone - pgdat->node_zones; | |
64dd1b29 CS |
479 | ret = ensure_zone_is_initialized(zone, phys_start_pfn, nr_pages); |
480 | if (ret) | |
481 | return ret; | |
76cdd58e | 482 | |
76cdd58e HC |
483 | pgdat_resize_lock(zone->zone_pgdat, &flags); |
484 | grow_zone_span(zone, phys_start_pfn, phys_start_pfn + nr_pages); | |
485 | grow_pgdat_span(zone->zone_pgdat, phys_start_pfn, | |
486 | phys_start_pfn + nr_pages); | |
487 | pgdat_resize_unlock(zone->zone_pgdat, &flags); | |
a2f3aa02 DH |
488 | memmap_init_zone(nr_pages, nid, zone_type, |
489 | phys_start_pfn, MEMMAP_HOTPLUG); | |
e298ff75 MG |
490 | |
491 | /* online_page_range is called later and expects pages reserved */ | |
492 | for (pfn = phys_start_pfn; pfn < phys_start_pfn + nr_pages; pfn++) { | |
493 | if (!pfn_valid(pfn)) | |
494 | continue; | |
495 | ||
496 | SetPageReserved(pfn_to_page(pfn)); | |
497 | } | |
718127cc | 498 | return 0; |
3947be19 DH |
499 | } |
500 | ||
c04fc586 GH |
501 | static int __meminit __add_section(int nid, struct zone *zone, |
502 | unsigned long phys_start_pfn) | |
3947be19 | 503 | { |
3947be19 DH |
504 | int ret; |
505 | ||
ebd15302 KH |
506 | if (pfn_valid(phys_start_pfn)) |
507 | return -EEXIST; | |
508 | ||
85b35fea | 509 | ret = sparse_add_one_section(zone, phys_start_pfn); |
3947be19 DH |
510 | |
511 | if (ret < 0) | |
512 | return ret; | |
513 | ||
718127cc YG |
514 | ret = __add_zone(zone, phys_start_pfn); |
515 | ||
516 | if (ret < 0) | |
517 | return ret; | |
518 | ||
c04fc586 | 519 | return register_new_memory(nid, __pfn_to_section(phys_start_pfn)); |
3947be19 DH |
520 | } |
521 | ||
4edd7cef DR |
522 | /* |
523 | * Reasonably generic function for adding memory. It is | |
524 | * expected that archs that support memory hotplug will | |
525 | * call this function after deciding the zone to which to | |
526 | * add the new pages. | |
527 | */ | |
528 | int __ref __add_pages(int nid, struct zone *zone, unsigned long phys_start_pfn, | |
529 | unsigned long nr_pages) | |
530 | { | |
531 | unsigned long i; | |
532 | int err = 0; | |
533 | int start_sec, end_sec; | |
4b94ffdc DW |
534 | struct vmem_altmap *altmap; |
535 | ||
7cf91a98 JK |
536 | clear_zone_contiguous(zone); |
537 | ||
4edd7cef DR |
538 | /* during initialize mem_map, align hot-added range to section */ |
539 | start_sec = pfn_to_section_nr(phys_start_pfn); | |
540 | end_sec = pfn_to_section_nr(phys_start_pfn + nr_pages - 1); | |
541 | ||
4b94ffdc DW |
542 | altmap = to_vmem_altmap((unsigned long) pfn_to_page(phys_start_pfn)); |
543 | if (altmap) { | |
544 | /* | |
545 | * Validate altmap is within bounds of the total request | |
546 | */ | |
547 | if (altmap->base_pfn != phys_start_pfn | |
548 | || vmem_altmap_offset(altmap) > nr_pages) { | |
549 | pr_warn_once("memory add fail, invalid altmap\n"); | |
7cf91a98 JK |
550 | err = -EINVAL; |
551 | goto out; | |
4b94ffdc DW |
552 | } |
553 | altmap->alloc = 0; | |
554 | } | |
555 | ||
4edd7cef | 556 | for (i = start_sec; i <= end_sec; i++) { |
19c07d5e | 557 | err = __add_section(nid, zone, section_nr_to_pfn(i)); |
4edd7cef DR |
558 | |
559 | /* | |
560 | * EEXIST is finally dealt with by ioresource collision | |
561 | * check. see add_memory() => register_memory_resource() | |
562 | * Warning will be printed if there is collision. | |
563 | */ | |
564 | if (err && (err != -EEXIST)) | |
565 | break; | |
566 | err = 0; | |
567 | } | |
c435a390 | 568 | vmemmap_populate_print_last(); |
7cf91a98 JK |
569 | out: |
570 | set_zone_contiguous(zone); | |
4edd7cef DR |
571 | return err; |
572 | } | |
573 | EXPORT_SYMBOL_GPL(__add_pages); | |
574 | ||
575 | #ifdef CONFIG_MEMORY_HOTREMOVE | |
815121d2 YI |
576 | /* find the smallest valid pfn in the range [start_pfn, end_pfn) */ |
577 | static int find_smallest_section_pfn(int nid, struct zone *zone, | |
578 | unsigned long start_pfn, | |
579 | unsigned long end_pfn) | |
580 | { | |
581 | struct mem_section *ms; | |
582 | ||
583 | for (; start_pfn < end_pfn; start_pfn += PAGES_PER_SECTION) { | |
584 | ms = __pfn_to_section(start_pfn); | |
585 | ||
586 | if (unlikely(!valid_section(ms))) | |
587 | continue; | |
588 | ||
589 | if (unlikely(pfn_to_nid(start_pfn) != nid)) | |
590 | continue; | |
591 | ||
592 | if (zone && zone != page_zone(pfn_to_page(start_pfn))) | |
593 | continue; | |
594 | ||
595 | return start_pfn; | |
596 | } | |
597 | ||
598 | return 0; | |
599 | } | |
600 | ||
601 | /* find the biggest valid pfn in the range [start_pfn, end_pfn). */ | |
602 | static int find_biggest_section_pfn(int nid, struct zone *zone, | |
603 | unsigned long start_pfn, | |
604 | unsigned long end_pfn) | |
605 | { | |
606 | struct mem_section *ms; | |
607 | unsigned long pfn; | |
608 | ||
609 | /* pfn is the end pfn of a memory section. */ | |
610 | pfn = end_pfn - 1; | |
611 | for (; pfn >= start_pfn; pfn -= PAGES_PER_SECTION) { | |
612 | ms = __pfn_to_section(pfn); | |
613 | ||
614 | if (unlikely(!valid_section(ms))) | |
615 | continue; | |
616 | ||
617 | if (unlikely(pfn_to_nid(pfn) != nid)) | |
618 | continue; | |
619 | ||
620 | if (zone && zone != page_zone(pfn_to_page(pfn))) | |
621 | continue; | |
622 | ||
623 | return pfn; | |
624 | } | |
625 | ||
626 | return 0; | |
627 | } | |
628 | ||
629 | static void shrink_zone_span(struct zone *zone, unsigned long start_pfn, | |
630 | unsigned long end_pfn) | |
631 | { | |
c33bc315 XQ |
632 | unsigned long zone_start_pfn = zone->zone_start_pfn; |
633 | unsigned long z = zone_end_pfn(zone); /* zone_end_pfn namespace clash */ | |
634 | unsigned long zone_end_pfn = z; | |
815121d2 YI |
635 | unsigned long pfn; |
636 | struct mem_section *ms; | |
637 | int nid = zone_to_nid(zone); | |
638 | ||
639 | zone_span_writelock(zone); | |
640 | if (zone_start_pfn == start_pfn) { | |
641 | /* | |
642 | * If the section is smallest section in the zone, it need | |
643 | * shrink zone->zone_start_pfn and zone->zone_spanned_pages. | |
644 | * In this case, we find second smallest valid mem_section | |
645 | * for shrinking zone. | |
646 | */ | |
647 | pfn = find_smallest_section_pfn(nid, zone, end_pfn, | |
648 | zone_end_pfn); | |
649 | if (pfn) { | |
650 | zone->zone_start_pfn = pfn; | |
651 | zone->spanned_pages = zone_end_pfn - pfn; | |
652 | } | |
653 | } else if (zone_end_pfn == end_pfn) { | |
654 | /* | |
655 | * If the section is biggest section in the zone, it need | |
656 | * shrink zone->spanned_pages. | |
657 | * In this case, we find second biggest valid mem_section for | |
658 | * shrinking zone. | |
659 | */ | |
660 | pfn = find_biggest_section_pfn(nid, zone, zone_start_pfn, | |
661 | start_pfn); | |
662 | if (pfn) | |
663 | zone->spanned_pages = pfn - zone_start_pfn + 1; | |
664 | } | |
665 | ||
666 | /* | |
667 | * The section is not biggest or smallest mem_section in the zone, it | |
668 | * only creates a hole in the zone. So in this case, we need not | |
669 | * change the zone. But perhaps, the zone has only hole data. Thus | |
670 | * it check the zone has only hole or not. | |
671 | */ | |
672 | pfn = zone_start_pfn; | |
673 | for (; pfn < zone_end_pfn; pfn += PAGES_PER_SECTION) { | |
674 | ms = __pfn_to_section(pfn); | |
675 | ||
676 | if (unlikely(!valid_section(ms))) | |
677 | continue; | |
678 | ||
679 | if (page_zone(pfn_to_page(pfn)) != zone) | |
680 | continue; | |
681 | ||
682 | /* If the section is current section, it continues the loop */ | |
683 | if (start_pfn == pfn) | |
684 | continue; | |
685 | ||
686 | /* If we find valid section, we have nothing to do */ | |
687 | zone_span_writeunlock(zone); | |
688 | return; | |
689 | } | |
690 | ||
691 | /* The zone has no valid section */ | |
692 | zone->zone_start_pfn = 0; | |
693 | zone->spanned_pages = 0; | |
694 | zone_span_writeunlock(zone); | |
695 | } | |
696 | ||
697 | static void shrink_pgdat_span(struct pglist_data *pgdat, | |
698 | unsigned long start_pfn, unsigned long end_pfn) | |
699 | { | |
83285c72 XQ |
700 | unsigned long pgdat_start_pfn = pgdat->node_start_pfn; |
701 | unsigned long p = pgdat_end_pfn(pgdat); /* pgdat_end_pfn namespace clash */ | |
702 | unsigned long pgdat_end_pfn = p; | |
815121d2 YI |
703 | unsigned long pfn; |
704 | struct mem_section *ms; | |
705 | int nid = pgdat->node_id; | |
706 | ||
707 | if (pgdat_start_pfn == start_pfn) { | |
708 | /* | |
709 | * If the section is smallest section in the pgdat, it need | |
710 | * shrink pgdat->node_start_pfn and pgdat->node_spanned_pages. | |
711 | * In this case, we find second smallest valid mem_section | |
712 | * for shrinking zone. | |
713 | */ | |
714 | pfn = find_smallest_section_pfn(nid, NULL, end_pfn, | |
715 | pgdat_end_pfn); | |
716 | if (pfn) { | |
717 | pgdat->node_start_pfn = pfn; | |
718 | pgdat->node_spanned_pages = pgdat_end_pfn - pfn; | |
719 | } | |
720 | } else if (pgdat_end_pfn == end_pfn) { | |
721 | /* | |
722 | * If the section is biggest section in the pgdat, it need | |
723 | * shrink pgdat->node_spanned_pages. | |
724 | * In this case, we find second biggest valid mem_section for | |
725 | * shrinking zone. | |
726 | */ | |
727 | pfn = find_biggest_section_pfn(nid, NULL, pgdat_start_pfn, | |
728 | start_pfn); | |
729 | if (pfn) | |
730 | pgdat->node_spanned_pages = pfn - pgdat_start_pfn + 1; | |
731 | } | |
732 | ||
733 | /* | |
734 | * If the section is not biggest or smallest mem_section in the pgdat, | |
735 | * it only creates a hole in the pgdat. So in this case, we need not | |
736 | * change the pgdat. | |
737 | * But perhaps, the pgdat has only hole data. Thus it check the pgdat | |
738 | * has only hole or not. | |
739 | */ | |
740 | pfn = pgdat_start_pfn; | |
741 | for (; pfn < pgdat_end_pfn; pfn += PAGES_PER_SECTION) { | |
742 | ms = __pfn_to_section(pfn); | |
743 | ||
744 | if (unlikely(!valid_section(ms))) | |
745 | continue; | |
746 | ||
747 | if (pfn_to_nid(pfn) != nid) | |
748 | continue; | |
749 | ||
750 | /* If the section is current section, it continues the loop */ | |
751 | if (start_pfn == pfn) | |
752 | continue; | |
753 | ||
754 | /* If we find valid section, we have nothing to do */ | |
755 | return; | |
756 | } | |
757 | ||
758 | /* The pgdat has no valid section */ | |
759 | pgdat->node_start_pfn = 0; | |
760 | pgdat->node_spanned_pages = 0; | |
761 | } | |
762 | ||
763 | static void __remove_zone(struct zone *zone, unsigned long start_pfn) | |
764 | { | |
765 | struct pglist_data *pgdat = zone->zone_pgdat; | |
766 | int nr_pages = PAGES_PER_SECTION; | |
767 | int zone_type; | |
768 | unsigned long flags; | |
769 | ||
770 | zone_type = zone - pgdat->node_zones; | |
771 | ||
772 | pgdat_resize_lock(zone->zone_pgdat, &flags); | |
773 | shrink_zone_span(zone, start_pfn, start_pfn + nr_pages); | |
774 | shrink_pgdat_span(pgdat, start_pfn, start_pfn + nr_pages); | |
775 | pgdat_resize_unlock(zone->zone_pgdat, &flags); | |
776 | } | |
777 | ||
4b94ffdc DW |
778 | static int __remove_section(struct zone *zone, struct mem_section *ms, |
779 | unsigned long map_offset) | |
ea01ea93 | 780 | { |
815121d2 YI |
781 | unsigned long start_pfn; |
782 | int scn_nr; | |
ea01ea93 BP |
783 | int ret = -EINVAL; |
784 | ||
785 | if (!valid_section(ms)) | |
786 | return ret; | |
787 | ||
788 | ret = unregister_memory_section(ms); | |
789 | if (ret) | |
790 | return ret; | |
791 | ||
815121d2 YI |
792 | scn_nr = __section_nr(ms); |
793 | start_pfn = section_nr_to_pfn(scn_nr); | |
794 | __remove_zone(zone, start_pfn); | |
795 | ||
4b94ffdc | 796 | sparse_remove_one_section(zone, ms, map_offset); |
ea01ea93 BP |
797 | return 0; |
798 | } | |
799 | ||
ea01ea93 BP |
800 | /** |
801 | * __remove_pages() - remove sections of pages from a zone | |
802 | * @zone: zone from which pages need to be removed | |
803 | * @phys_start_pfn: starting pageframe (must be aligned to start of a section) | |
804 | * @nr_pages: number of pages to remove (must be multiple of section size) | |
805 | * | |
806 | * Generic helper function to remove section mappings and sysfs entries | |
807 | * for the section of the memory we are removing. Caller needs to make | |
808 | * sure that pages are marked reserved and zones are adjust properly by | |
809 | * calling offline_pages(). | |
810 | */ | |
811 | int __remove_pages(struct zone *zone, unsigned long phys_start_pfn, | |
812 | unsigned long nr_pages) | |
813 | { | |
fe74ebb1 | 814 | unsigned long i; |
4b94ffdc DW |
815 | unsigned long map_offset = 0; |
816 | int sections_to_remove, ret = 0; | |
817 | ||
818 | /* In the ZONE_DEVICE case device driver owns the memory region */ | |
819 | if (is_dev_zone(zone)) { | |
820 | struct page *page = pfn_to_page(phys_start_pfn); | |
821 | struct vmem_altmap *altmap; | |
822 | ||
823 | altmap = to_vmem_altmap((unsigned long) page); | |
824 | if (altmap) | |
825 | map_offset = vmem_altmap_offset(altmap); | |
826 | } else { | |
827 | resource_size_t start, size; | |
828 | ||
829 | start = phys_start_pfn << PAGE_SHIFT; | |
830 | size = nr_pages * PAGE_SIZE; | |
831 | ||
832 | ret = release_mem_region_adjustable(&iomem_resource, start, | |
833 | size); | |
834 | if (ret) { | |
835 | resource_size_t endres = start + size - 1; | |
836 | ||
837 | pr_warn("Unable to release resource <%pa-%pa> (%d)\n", | |
838 | &start, &endres, ret); | |
839 | } | |
840 | } | |
ea01ea93 | 841 | |
7cf91a98 JK |
842 | clear_zone_contiguous(zone); |
843 | ||
ea01ea93 BP |
844 | /* |
845 | * We can only remove entire sections | |
846 | */ | |
847 | BUG_ON(phys_start_pfn & ~PAGE_SECTION_MASK); | |
848 | BUG_ON(nr_pages % PAGES_PER_SECTION); | |
849 | ||
ea01ea93 BP |
850 | sections_to_remove = nr_pages / PAGES_PER_SECTION; |
851 | for (i = 0; i < sections_to_remove; i++) { | |
852 | unsigned long pfn = phys_start_pfn + i*PAGES_PER_SECTION; | |
4b94ffdc DW |
853 | |
854 | ret = __remove_section(zone, __pfn_to_section(pfn), map_offset); | |
855 | map_offset = 0; | |
ea01ea93 BP |
856 | if (ret) |
857 | break; | |
858 | } | |
7cf91a98 JK |
859 | |
860 | set_zone_contiguous(zone); | |
861 | ||
ea01ea93 BP |
862 | return ret; |
863 | } | |
864 | EXPORT_SYMBOL_GPL(__remove_pages); | |
4edd7cef | 865 | #endif /* CONFIG_MEMORY_HOTREMOVE */ |
ea01ea93 | 866 | |
9d0ad8ca DK |
867 | int set_online_page_callback(online_page_callback_t callback) |
868 | { | |
869 | int rc = -EINVAL; | |
870 | ||
bfc8c901 VD |
871 | get_online_mems(); |
872 | mutex_lock(&online_page_callback_lock); | |
9d0ad8ca DK |
873 | |
874 | if (online_page_callback == generic_online_page) { | |
875 | online_page_callback = callback; | |
876 | rc = 0; | |
877 | } | |
878 | ||
bfc8c901 VD |
879 | mutex_unlock(&online_page_callback_lock); |
880 | put_online_mems(); | |
9d0ad8ca DK |
881 | |
882 | return rc; | |
883 | } | |
884 | EXPORT_SYMBOL_GPL(set_online_page_callback); | |
885 | ||
886 | int restore_online_page_callback(online_page_callback_t callback) | |
887 | { | |
888 | int rc = -EINVAL; | |
889 | ||
bfc8c901 VD |
890 | get_online_mems(); |
891 | mutex_lock(&online_page_callback_lock); | |
9d0ad8ca DK |
892 | |
893 | if (online_page_callback == callback) { | |
894 | online_page_callback = generic_online_page; | |
895 | rc = 0; | |
896 | } | |
897 | ||
bfc8c901 VD |
898 | mutex_unlock(&online_page_callback_lock); |
899 | put_online_mems(); | |
9d0ad8ca DK |
900 | |
901 | return rc; | |
902 | } | |
903 | EXPORT_SYMBOL_GPL(restore_online_page_callback); | |
904 | ||
905 | void __online_page_set_limits(struct page *page) | |
180c06ef | 906 | { |
9d0ad8ca DK |
907 | } |
908 | EXPORT_SYMBOL_GPL(__online_page_set_limits); | |
909 | ||
910 | void __online_page_increment_counters(struct page *page) | |
911 | { | |
3dcc0571 | 912 | adjust_managed_page_count(page, 1); |
9d0ad8ca DK |
913 | } |
914 | EXPORT_SYMBOL_GPL(__online_page_increment_counters); | |
180c06ef | 915 | |
9d0ad8ca DK |
916 | void __online_page_free(struct page *page) |
917 | { | |
3dcc0571 | 918 | __free_reserved_page(page); |
180c06ef | 919 | } |
9d0ad8ca DK |
920 | EXPORT_SYMBOL_GPL(__online_page_free); |
921 | ||
922 | static void generic_online_page(struct page *page) | |
923 | { | |
924 | __online_page_set_limits(page); | |
925 | __online_page_increment_counters(page); | |
926 | __online_page_free(page); | |
927 | } | |
180c06ef | 928 | |
75884fb1 KH |
929 | static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages, |
930 | void *arg) | |
3947be19 DH |
931 | { |
932 | unsigned long i; | |
75884fb1 KH |
933 | unsigned long onlined_pages = *(unsigned long *)arg; |
934 | struct page *page; | |
935 | if (PageReserved(pfn_to_page(start_pfn))) | |
936 | for (i = 0; i < nr_pages; i++) { | |
937 | page = pfn_to_page(start_pfn + i); | |
9d0ad8ca | 938 | (*online_page_callback)(page); |
75884fb1 KH |
939 | onlined_pages++; |
940 | } | |
941 | *(unsigned long *)arg = onlined_pages; | |
942 | return 0; | |
943 | } | |
944 | ||
09285af7 | 945 | #ifdef CONFIG_MOVABLE_NODE |
79a4dcef TC |
946 | /* |
947 | * When CONFIG_MOVABLE_NODE, we permit onlining of a node which doesn't have | |
948 | * normal memory. | |
949 | */ | |
09285af7 LJ |
950 | static bool can_online_high_movable(struct zone *zone) |
951 | { | |
952 | return true; | |
953 | } | |
79a4dcef | 954 | #else /* CONFIG_MOVABLE_NODE */ |
74d42d8f LJ |
955 | /* ensure every online node has NORMAL memory */ |
956 | static bool can_online_high_movable(struct zone *zone) | |
957 | { | |
958 | return node_state(zone_to_nid(zone), N_NORMAL_MEMORY); | |
959 | } | |
79a4dcef | 960 | #endif /* CONFIG_MOVABLE_NODE */ |
74d42d8f | 961 | |
d9713679 LJ |
962 | /* check which state of node_states will be changed when online memory */ |
963 | static void node_states_check_changes_online(unsigned long nr_pages, | |
964 | struct zone *zone, struct memory_notify *arg) | |
965 | { | |
966 | int nid = zone_to_nid(zone); | |
967 | enum zone_type zone_last = ZONE_NORMAL; | |
968 | ||
969 | /* | |
6715ddf9 LJ |
970 | * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY] |
971 | * contains nodes which have zones of 0...ZONE_NORMAL, | |
972 | * set zone_last to ZONE_NORMAL. | |
d9713679 | 973 | * |
6715ddf9 LJ |
974 | * If we don't have HIGHMEM nor movable node, |
975 | * node_states[N_NORMAL_MEMORY] contains nodes which have zones of | |
976 | * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE. | |
d9713679 | 977 | */ |
6715ddf9 | 978 | if (N_MEMORY == N_NORMAL_MEMORY) |
d9713679 LJ |
979 | zone_last = ZONE_MOVABLE; |
980 | ||
981 | /* | |
982 | * if the memory to be online is in a zone of 0...zone_last, and | |
983 | * the zones of 0...zone_last don't have memory before online, we will | |
984 | * need to set the node to node_states[N_NORMAL_MEMORY] after | |
985 | * the memory is online. | |
986 | */ | |
987 | if (zone_idx(zone) <= zone_last && !node_state(nid, N_NORMAL_MEMORY)) | |
988 | arg->status_change_nid_normal = nid; | |
989 | else | |
990 | arg->status_change_nid_normal = -1; | |
991 | ||
6715ddf9 LJ |
992 | #ifdef CONFIG_HIGHMEM |
993 | /* | |
994 | * If we have movable node, node_states[N_HIGH_MEMORY] | |
995 | * contains nodes which have zones of 0...ZONE_HIGHMEM, | |
996 | * set zone_last to ZONE_HIGHMEM. | |
997 | * | |
998 | * If we don't have movable node, node_states[N_NORMAL_MEMORY] | |
999 | * contains nodes which have zones of 0...ZONE_MOVABLE, | |
1000 | * set zone_last to ZONE_MOVABLE. | |
1001 | */ | |
1002 | zone_last = ZONE_HIGHMEM; | |
1003 | if (N_MEMORY == N_HIGH_MEMORY) | |
1004 | zone_last = ZONE_MOVABLE; | |
1005 | ||
1006 | if (zone_idx(zone) <= zone_last && !node_state(nid, N_HIGH_MEMORY)) | |
1007 | arg->status_change_nid_high = nid; | |
1008 | else | |
1009 | arg->status_change_nid_high = -1; | |
1010 | #else | |
1011 | arg->status_change_nid_high = arg->status_change_nid_normal; | |
1012 | #endif | |
1013 | ||
d9713679 LJ |
1014 | /* |
1015 | * if the node don't have memory befor online, we will need to | |
6715ddf9 | 1016 | * set the node to node_states[N_MEMORY] after the memory |
d9713679 LJ |
1017 | * is online. |
1018 | */ | |
6715ddf9 | 1019 | if (!node_state(nid, N_MEMORY)) |
d9713679 LJ |
1020 | arg->status_change_nid = nid; |
1021 | else | |
1022 | arg->status_change_nid = -1; | |
1023 | } | |
1024 | ||
1025 | static void node_states_set_node(int node, struct memory_notify *arg) | |
1026 | { | |
1027 | if (arg->status_change_nid_normal >= 0) | |
1028 | node_set_state(node, N_NORMAL_MEMORY); | |
1029 | ||
6715ddf9 LJ |
1030 | if (arg->status_change_nid_high >= 0) |
1031 | node_set_state(node, N_HIGH_MEMORY); | |
1032 | ||
1033 | node_set_state(node, N_MEMORY); | |
d9713679 LJ |
1034 | } |
1035 | ||
df429ac0 RA |
1036 | int zone_can_shift(unsigned long pfn, unsigned long nr_pages, |
1037 | enum zone_type target) | |
1038 | { | |
1039 | struct zone *zone = page_zone(pfn_to_page(pfn)); | |
1040 | enum zone_type idx = zone_idx(zone); | |
1041 | int i; | |
1042 | ||
1043 | if (idx < target) { | |
1044 | /* pages must be at end of current zone */ | |
1045 | if (pfn + nr_pages != zone_end_pfn(zone)) | |
1046 | return 0; | |
1047 | ||
1048 | /* no zones in use between current zone and target */ | |
1049 | for (i = idx + 1; i < target; i++) | |
1050 | if (zone_is_initialized(zone - idx + i)) | |
1051 | return 0; | |
1052 | } | |
1053 | ||
1054 | if (target < idx) { | |
1055 | /* pages must be at beginning of current zone */ | |
1056 | if (pfn != zone->zone_start_pfn) | |
1057 | return 0; | |
1058 | ||
1059 | /* no zones in use between current zone and target */ | |
1060 | for (i = target + 1; i < idx; i++) | |
1061 | if (zone_is_initialized(zone - idx + i)) | |
1062 | return 0; | |
1063 | } | |
1064 | ||
1065 | return target - idx; | |
1066 | } | |
75884fb1 | 1067 | |
30467e0b | 1068 | /* Must be protected by mem_hotplug_begin() */ |
511c2aba | 1069 | int __ref online_pages(unsigned long pfn, unsigned long nr_pages, int online_type) |
75884fb1 | 1070 | { |
aa47228a | 1071 | unsigned long flags; |
3947be19 DH |
1072 | unsigned long onlined_pages = 0; |
1073 | struct zone *zone; | |
6811378e | 1074 | int need_zonelists_rebuild = 0; |
7b78d335 YG |
1075 | int nid; |
1076 | int ret; | |
1077 | struct memory_notify arg; | |
e51e6c8f | 1078 | int zone_shift = 0; |
7b78d335 | 1079 | |
d9713679 LJ |
1080 | /* |
1081 | * This doesn't need a lock to do pfn_to_page(). | |
1082 | * The section can't be removed here because of the | |
1083 | * memory_block->state_mutex. | |
1084 | */ | |
1085 | zone = page_zone(pfn_to_page(pfn)); | |
1086 | ||
4f7c6b49 TC |
1087 | if ((zone_idx(zone) > ZONE_NORMAL || |
1088 | online_type == MMOP_ONLINE_MOVABLE) && | |
bfc8c901 | 1089 | !can_online_high_movable(zone)) |
30467e0b | 1090 | return -EINVAL; |
74d42d8f | 1091 | |
df429ac0 RA |
1092 | if (online_type == MMOP_ONLINE_KERNEL) |
1093 | zone_shift = zone_can_shift(pfn, nr_pages, ZONE_NORMAL); | |
1094 | else if (online_type == MMOP_ONLINE_MOVABLE) | |
1095 | zone_shift = zone_can_shift(pfn, nr_pages, ZONE_MOVABLE); | |
511c2aba | 1096 | |
e51e6c8f RA |
1097 | zone = move_pfn_range(zone_shift, pfn, pfn + nr_pages); |
1098 | if (!zone) | |
1099 | return -EINVAL; | |
511c2aba | 1100 | |
7b78d335 YG |
1101 | arg.start_pfn = pfn; |
1102 | arg.nr_pages = nr_pages; | |
d9713679 | 1103 | node_states_check_changes_online(nr_pages, zone, &arg); |
7b78d335 | 1104 | |
e888ca35 | 1105 | nid = zone_to_nid(zone); |
3947be19 | 1106 | |
7b78d335 YG |
1107 | ret = memory_notify(MEM_GOING_ONLINE, &arg); |
1108 | ret = notifier_to_errno(ret); | |
e33e33b4 CY |
1109 | if (ret) |
1110 | goto failed_addition; | |
1111 | ||
6811378e YG |
1112 | /* |
1113 | * If this zone is not populated, then it is not in zonelist. | |
1114 | * This means the page allocator ignores this zone. | |
1115 | * So, zonelist must be updated after online. | |
1116 | */ | |
4eaf3f64 | 1117 | mutex_lock(&zonelists_mutex); |
6dcd73d7 | 1118 | if (!populated_zone(zone)) { |
6811378e | 1119 | need_zonelists_rebuild = 1; |
6dcd73d7 WC |
1120 | build_all_zonelists(NULL, zone); |
1121 | } | |
6811378e | 1122 | |
908eedc6 | 1123 | ret = walk_system_ram_range(pfn, nr_pages, &onlined_pages, |
75884fb1 | 1124 | online_pages_range); |
fd8a4221 | 1125 | if (ret) { |
6dcd73d7 WC |
1126 | if (need_zonelists_rebuild) |
1127 | zone_pcp_reset(zone); | |
4eaf3f64 | 1128 | mutex_unlock(&zonelists_mutex); |
e33e33b4 | 1129 | goto failed_addition; |
fd8a4221 GL |
1130 | } |
1131 | ||
3947be19 | 1132 | zone->present_pages += onlined_pages; |
aa47228a CS |
1133 | |
1134 | pgdat_resize_lock(zone->zone_pgdat, &flags); | |
f2937be5 | 1135 | zone->zone_pgdat->node_present_pages += onlined_pages; |
aa47228a CS |
1136 | pgdat_resize_unlock(zone->zone_pgdat, &flags); |
1137 | ||
08dff7b7 | 1138 | if (onlined_pages) { |
e888ca35 | 1139 | node_states_set_node(nid, &arg); |
08dff7b7 | 1140 | if (need_zonelists_rebuild) |
6dcd73d7 | 1141 | build_all_zonelists(NULL, NULL); |
08dff7b7 JL |
1142 | else |
1143 | zone_pcp_update(zone); | |
1144 | } | |
3947be19 | 1145 | |
4eaf3f64 | 1146 | mutex_unlock(&zonelists_mutex); |
1b79acc9 KM |
1147 | |
1148 | init_per_zone_wmark_min(); | |
1149 | ||
698b1b30 | 1150 | if (onlined_pages) { |
e888ca35 | 1151 | kswapd_run(nid); |
698b1b30 VB |
1152 | kcompactd_run(nid); |
1153 | } | |
61b13993 | 1154 | |
1f522509 | 1155 | vm_total_pages = nr_free_pagecache_pages(); |
2f7f24ec | 1156 | |
2d1d43f6 | 1157 | writeback_set_ratelimit(); |
7b78d335 YG |
1158 | |
1159 | if (onlined_pages) | |
1160 | memory_notify(MEM_ONLINE, &arg); | |
30467e0b | 1161 | return 0; |
e33e33b4 CY |
1162 | |
1163 | failed_addition: | |
1164 | pr_debug("online_pages [mem %#010llx-%#010llx] failed\n", | |
1165 | (unsigned long long) pfn << PAGE_SHIFT, | |
1166 | (((unsigned long long) pfn + nr_pages) << PAGE_SHIFT) - 1); | |
1167 | memory_notify(MEM_CANCEL_ONLINE, &arg); | |
1168 | return ret; | |
3947be19 | 1169 | } |
53947027 | 1170 | #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */ |
bc02af93 | 1171 | |
0bd85420 TC |
1172 | static void reset_node_present_pages(pg_data_t *pgdat) |
1173 | { | |
1174 | struct zone *z; | |
1175 | ||
1176 | for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++) | |
1177 | z->present_pages = 0; | |
1178 | ||
1179 | pgdat->node_present_pages = 0; | |
1180 | } | |
1181 | ||
e1319331 HS |
1182 | /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */ |
1183 | static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start) | |
9af3c2de YG |
1184 | { |
1185 | struct pglist_data *pgdat; | |
1186 | unsigned long zones_size[MAX_NR_ZONES] = {0}; | |
1187 | unsigned long zholes_size[MAX_NR_ZONES] = {0}; | |
c8e861a5 | 1188 | unsigned long start_pfn = PFN_DOWN(start); |
9af3c2de | 1189 | |
a1e565aa TC |
1190 | pgdat = NODE_DATA(nid); |
1191 | if (!pgdat) { | |
1192 | pgdat = arch_alloc_nodedata(nid); | |
1193 | if (!pgdat) | |
1194 | return NULL; | |
9af3c2de | 1195 | |
a1e565aa | 1196 | arch_refresh_nodedata(nid, pgdat); |
b0dc3a34 | 1197 | } else { |
38087d9b | 1198 | /* Reset the nr_zones, order and classzone_idx before reuse */ |
b0dc3a34 | 1199 | pgdat->nr_zones = 0; |
38087d9b MG |
1200 | pgdat->kswapd_order = 0; |
1201 | pgdat->kswapd_classzone_idx = 0; | |
a1e565aa | 1202 | } |
9af3c2de YG |
1203 | |
1204 | /* we can use NODE_DATA(nid) from here */ | |
1205 | ||
1206 | /* init node's zones as empty zones, we don't have any present pages.*/ | |
9109fb7b | 1207 | free_area_init_node(nid, zones_size, start_pfn, zholes_size); |
5830169f | 1208 | pgdat->per_cpu_nodestats = alloc_percpu(struct per_cpu_nodestat); |
9af3c2de | 1209 | |
959ecc48 KH |
1210 | /* |
1211 | * The node we allocated has no zone fallback lists. For avoiding | |
1212 | * to access not-initialized zonelist, build here. | |
1213 | */ | |
f957db4f | 1214 | mutex_lock(&zonelists_mutex); |
9adb62a5 | 1215 | build_all_zonelists(pgdat, NULL); |
f957db4f | 1216 | mutex_unlock(&zonelists_mutex); |
959ecc48 | 1217 | |
f784a3f1 TC |
1218 | /* |
1219 | * zone->managed_pages is set to an approximate value in | |
1220 | * free_area_init_core(), which will cause | |
1221 | * /sys/device/system/node/nodeX/meminfo has wrong data. | |
1222 | * So reset it to 0 before any memory is onlined. | |
1223 | */ | |
1224 | reset_node_managed_pages(pgdat); | |
1225 | ||
0bd85420 TC |
1226 | /* |
1227 | * When memory is hot-added, all the memory is in offline state. So | |
1228 | * clear all zones' present_pages because they will be updated in | |
1229 | * online_pages() and offline_pages(). | |
1230 | */ | |
1231 | reset_node_present_pages(pgdat); | |
1232 | ||
9af3c2de YG |
1233 | return pgdat; |
1234 | } | |
1235 | ||
1236 | static void rollback_node_hotadd(int nid, pg_data_t *pgdat) | |
1237 | { | |
1238 | arch_refresh_nodedata(nid, NULL); | |
5830169f | 1239 | free_percpu(pgdat->per_cpu_nodestats); |
9af3c2de YG |
1240 | arch_free_nodedata(pgdat); |
1241 | return; | |
1242 | } | |
1243 | ||
0a547039 | 1244 | |
01b0f197 TK |
1245 | /** |
1246 | * try_online_node - online a node if offlined | |
1247 | * | |
cf23422b | 1248 | * called by cpu_up() to online a node without onlined memory. |
1249 | */ | |
01b0f197 | 1250 | int try_online_node(int nid) |
cf23422b | 1251 | { |
1252 | pg_data_t *pgdat; | |
1253 | int ret; | |
1254 | ||
01b0f197 TK |
1255 | if (node_online(nid)) |
1256 | return 0; | |
1257 | ||
bfc8c901 | 1258 | mem_hotplug_begin(); |
cf23422b | 1259 | pgdat = hotadd_new_pgdat(nid, 0); |
7553e8f2 | 1260 | if (!pgdat) { |
01b0f197 | 1261 | pr_err("Cannot online node %d due to NULL pgdat\n", nid); |
cf23422b | 1262 | ret = -ENOMEM; |
1263 | goto out; | |
1264 | } | |
1265 | node_set_online(nid); | |
1266 | ret = register_one_node(nid); | |
1267 | BUG_ON(ret); | |
1268 | ||
01b0f197 TK |
1269 | if (pgdat->node_zonelists->_zonerefs->zone == NULL) { |
1270 | mutex_lock(&zonelists_mutex); | |
1271 | build_all_zonelists(NULL, NULL); | |
1272 | mutex_unlock(&zonelists_mutex); | |
1273 | } | |
1274 | ||
cf23422b | 1275 | out: |
bfc8c901 | 1276 | mem_hotplug_done(); |
cf23422b | 1277 | return ret; |
1278 | } | |
1279 | ||
27356f54 TK |
1280 | static int check_hotplug_memory_range(u64 start, u64 size) |
1281 | { | |
c8e861a5 | 1282 | u64 start_pfn = PFN_DOWN(start); |
27356f54 TK |
1283 | u64 nr_pages = size >> PAGE_SHIFT; |
1284 | ||
1285 | /* Memory range must be aligned with section */ | |
1286 | if ((start_pfn & ~PAGE_SECTION_MASK) || | |
1287 | (nr_pages % PAGES_PER_SECTION) || (!nr_pages)) { | |
1288 | pr_err("Section-unaligned hotplug range: start 0x%llx, size 0x%llx\n", | |
1289 | (unsigned long long)start, | |
1290 | (unsigned long long)size); | |
1291 | return -EINVAL; | |
1292 | } | |
1293 | ||
1294 | return 0; | |
1295 | } | |
1296 | ||
63264400 WN |
1297 | /* |
1298 | * If movable zone has already been setup, newly added memory should be check. | |
1299 | * If its address is higher than movable zone, it should be added as movable. | |
1300 | * Without this check, movable zone may overlap with other zone. | |
1301 | */ | |
1302 | static int should_add_memory_movable(int nid, u64 start, u64 size) | |
1303 | { | |
1304 | unsigned long start_pfn = start >> PAGE_SHIFT; | |
1305 | pg_data_t *pgdat = NODE_DATA(nid); | |
1306 | struct zone *movable_zone = pgdat->node_zones + ZONE_MOVABLE; | |
1307 | ||
1308 | if (zone_is_empty(movable_zone)) | |
1309 | return 0; | |
1310 | ||
1311 | if (movable_zone->zone_start_pfn <= start_pfn) | |
1312 | return 1; | |
1313 | ||
1314 | return 0; | |
1315 | } | |
1316 | ||
033fbae9 DW |
1317 | int zone_for_memory(int nid, u64 start, u64 size, int zone_default, |
1318 | bool for_device) | |
63264400 | 1319 | { |
033fbae9 DW |
1320 | #ifdef CONFIG_ZONE_DEVICE |
1321 | if (for_device) | |
1322 | return ZONE_DEVICE; | |
1323 | #endif | |
63264400 WN |
1324 | if (should_add_memory_movable(nid, start, size)) |
1325 | return ZONE_MOVABLE; | |
1326 | ||
1327 | return zone_default; | |
1328 | } | |
1329 | ||
31bc3858 VK |
1330 | static int online_memory_block(struct memory_block *mem, void *arg) |
1331 | { | |
1332 | return memory_block_change_state(mem, MEM_ONLINE, MEM_OFFLINE); | |
1333 | } | |
1334 | ||
31168481 | 1335 | /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */ |
31bc3858 | 1336 | int __ref add_memory_resource(int nid, struct resource *res, bool online) |
bc02af93 | 1337 | { |
62cedb9f | 1338 | u64 start, size; |
9af3c2de | 1339 | pg_data_t *pgdat = NULL; |
a1e565aa TC |
1340 | bool new_pgdat; |
1341 | bool new_node; | |
bc02af93 YG |
1342 | int ret; |
1343 | ||
62cedb9f DV |
1344 | start = res->start; |
1345 | size = resource_size(res); | |
1346 | ||
27356f54 TK |
1347 | ret = check_hotplug_memory_range(start, size); |
1348 | if (ret) | |
1349 | return ret; | |
1350 | ||
a1e565aa TC |
1351 | { /* Stupid hack to suppress address-never-null warning */ |
1352 | void *p = NODE_DATA(nid); | |
1353 | new_pgdat = !p; | |
1354 | } | |
ac13c462 | 1355 | |
bfc8c901 | 1356 | mem_hotplug_begin(); |
ac13c462 | 1357 | |
7f36e3e5 TC |
1358 | /* |
1359 | * Add new range to memblock so that when hotadd_new_pgdat() is called | |
1360 | * to allocate new pgdat, get_pfn_range_for_nid() will be able to find | |
1361 | * this new range and calculate total pages correctly. The range will | |
1362 | * be removed at hot-remove time. | |
1363 | */ | |
1364 | memblock_add_node(start, size, nid); | |
1365 | ||
a1e565aa TC |
1366 | new_node = !node_online(nid); |
1367 | if (new_node) { | |
9af3c2de | 1368 | pgdat = hotadd_new_pgdat(nid, start); |
6ad696d2 | 1369 | ret = -ENOMEM; |
9af3c2de | 1370 | if (!pgdat) |
41b9e2d7 | 1371 | goto error; |
9af3c2de YG |
1372 | } |
1373 | ||
bc02af93 | 1374 | /* call arch's memory hotadd */ |
033fbae9 | 1375 | ret = arch_add_memory(nid, start, size, false); |
bc02af93 | 1376 | |
9af3c2de YG |
1377 | if (ret < 0) |
1378 | goto error; | |
1379 | ||
0fc44159 | 1380 | /* we online node here. we can't roll back from here. */ |
9af3c2de YG |
1381 | node_set_online(nid); |
1382 | ||
a1e565aa | 1383 | if (new_node) { |
0fc44159 YG |
1384 | ret = register_one_node(nid); |
1385 | /* | |
1386 | * If sysfs file of new node can't create, cpu on the node | |
1387 | * can't be hot-added. There is no rollback way now. | |
1388 | * So, check by BUG_ON() to catch it reluctantly.. | |
1389 | */ | |
1390 | BUG_ON(ret); | |
1391 | } | |
1392 | ||
d96ae530 AM |
1393 | /* create new memmap entry */ |
1394 | firmware_map_add_hotplug(start, start + size, "System RAM"); | |
1395 | ||
31bc3858 VK |
1396 | /* online pages if requested */ |
1397 | if (online) | |
1398 | walk_memory_range(PFN_DOWN(start), PFN_UP(start + size - 1), | |
1399 | NULL, online_memory_block); | |
1400 | ||
6ad696d2 AK |
1401 | goto out; |
1402 | ||
9af3c2de YG |
1403 | error: |
1404 | /* rollback pgdat allocation and others */ | |
1405 | if (new_pgdat) | |
1406 | rollback_node_hotadd(nid, pgdat); | |
7f36e3e5 | 1407 | memblock_remove(start, size); |
9af3c2de | 1408 | |
6ad696d2 | 1409 | out: |
bfc8c901 | 1410 | mem_hotplug_done(); |
bc02af93 YG |
1411 | return ret; |
1412 | } | |
62cedb9f DV |
1413 | EXPORT_SYMBOL_GPL(add_memory_resource); |
1414 | ||
1415 | int __ref add_memory(int nid, u64 start, u64 size) | |
1416 | { | |
1417 | struct resource *res; | |
1418 | int ret; | |
1419 | ||
1420 | res = register_memory_resource(start, size); | |
6f754ba4 VK |
1421 | if (IS_ERR(res)) |
1422 | return PTR_ERR(res); | |
62cedb9f | 1423 | |
31bc3858 | 1424 | ret = add_memory_resource(nid, res, memhp_auto_online); |
62cedb9f DV |
1425 | if (ret < 0) |
1426 | release_memory_resource(res); | |
1427 | return ret; | |
1428 | } | |
bc02af93 | 1429 | EXPORT_SYMBOL_GPL(add_memory); |
0c0e6195 KH |
1430 | |
1431 | #ifdef CONFIG_MEMORY_HOTREMOVE | |
5c755e9f BP |
1432 | /* |
1433 | * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy | |
1434 | * set and the size of the free page is given by page_order(). Using this, | |
1435 | * the function determines if the pageblock contains only free pages. | |
1436 | * Due to buddy contraints, a free page at least the size of a pageblock will | |
1437 | * be located at the start of the pageblock | |
1438 | */ | |
1439 | static inline int pageblock_free(struct page *page) | |
1440 | { | |
1441 | return PageBuddy(page) && page_order(page) >= pageblock_order; | |
1442 | } | |
1443 | ||
1444 | /* Return the start of the next active pageblock after a given page */ | |
1445 | static struct page *next_active_pageblock(struct page *page) | |
1446 | { | |
5c755e9f BP |
1447 | /* Ensure the starting page is pageblock-aligned */ |
1448 | BUG_ON(page_to_pfn(page) & (pageblock_nr_pages - 1)); | |
1449 | ||
5c755e9f | 1450 | /* If the entire pageblock is free, move to the end of free page */ |
0dcc48c1 KH |
1451 | if (pageblock_free(page)) { |
1452 | int order; | |
1453 | /* be careful. we don't have locks, page_order can be changed.*/ | |
1454 | order = page_order(page); | |
1455 | if ((order < MAX_ORDER) && (order >= pageblock_order)) | |
1456 | return page + (1 << order); | |
1457 | } | |
5c755e9f | 1458 | |
0dcc48c1 | 1459 | return page + pageblock_nr_pages; |
5c755e9f BP |
1460 | } |
1461 | ||
1462 | /* Checks if this range of memory is likely to be hot-removable. */ | |
c98940f6 | 1463 | bool is_mem_section_removable(unsigned long start_pfn, unsigned long nr_pages) |
5c755e9f | 1464 | { |
5c755e9f BP |
1465 | struct page *page = pfn_to_page(start_pfn); |
1466 | struct page *end_page = page + nr_pages; | |
1467 | ||
1468 | /* Check the starting page of each pageblock within the range */ | |
1469 | for (; page < end_page; page = next_active_pageblock(page)) { | |
49ac8255 | 1470 | if (!is_pageblock_removable_nolock(page)) |
c98940f6 | 1471 | return false; |
49ac8255 | 1472 | cond_resched(); |
5c755e9f BP |
1473 | } |
1474 | ||
1475 | /* All pageblocks in the memory block are likely to be hot-removable */ | |
c98940f6 | 1476 | return true; |
5c755e9f BP |
1477 | } |
1478 | ||
0c0e6195 KH |
1479 | /* |
1480 | * Confirm all pages in a range [start, end) is belongs to the same zone. | |
1481 | */ | |
ed2f2400 | 1482 | int test_pages_in_a_zone(unsigned long start_pfn, unsigned long end_pfn) |
0c0e6195 | 1483 | { |
5f0f2887 | 1484 | unsigned long pfn, sec_end_pfn; |
0c0e6195 KH |
1485 | struct zone *zone = NULL; |
1486 | struct page *page; | |
1487 | int i; | |
5f0f2887 | 1488 | for (pfn = start_pfn, sec_end_pfn = SECTION_ALIGN_UP(start_pfn); |
0c0e6195 | 1489 | pfn < end_pfn; |
5f0f2887 AB |
1490 | pfn = sec_end_pfn + 1, sec_end_pfn += PAGES_PER_SECTION) { |
1491 | /* Make sure the memory section is present first */ | |
1492 | if (!present_section_nr(pfn_to_section_nr(pfn))) | |
0c0e6195 | 1493 | continue; |
5f0f2887 AB |
1494 | for (; pfn < sec_end_pfn && pfn < end_pfn; |
1495 | pfn += MAX_ORDER_NR_PAGES) { | |
1496 | i = 0; | |
1497 | /* This is just a CONFIG_HOLES_IN_ZONE check.*/ | |
1498 | while ((i < MAX_ORDER_NR_PAGES) && | |
1499 | !pfn_valid_within(pfn + i)) | |
1500 | i++; | |
1501 | if (i == MAX_ORDER_NR_PAGES) | |
1502 | continue; | |
1503 | page = pfn_to_page(pfn + i); | |
1504 | if (zone && page_zone(page) != zone) | |
1505 | return 0; | |
1506 | zone = page_zone(page); | |
1507 | } | |
0c0e6195 KH |
1508 | } |
1509 | return 1; | |
1510 | } | |
1511 | ||
1512 | /* | |
c8721bbb NH |
1513 | * Scan pfn range [start,end) to find movable/migratable pages (LRU pages |
1514 | * and hugepages). We scan pfn because it's much easier than scanning over | |
1515 | * linked list. This function returns the pfn of the first found movable | |
1516 | * page if it's found, otherwise 0. | |
0c0e6195 | 1517 | */ |
c8721bbb | 1518 | static unsigned long scan_movable_pages(unsigned long start, unsigned long end) |
0c0e6195 KH |
1519 | { |
1520 | unsigned long pfn; | |
1521 | struct page *page; | |
1522 | for (pfn = start; pfn < end; pfn++) { | |
1523 | if (pfn_valid(pfn)) { | |
1524 | page = pfn_to_page(pfn); | |
1525 | if (PageLRU(page)) | |
1526 | return pfn; | |
c8721bbb | 1527 | if (PageHuge(page)) { |
7e1f049e | 1528 | if (page_huge_active(page)) |
c8721bbb NH |
1529 | return pfn; |
1530 | else | |
1531 | pfn = round_up(pfn + 1, | |
1532 | 1 << compound_order(page)) - 1; | |
1533 | } | |
0c0e6195 KH |
1534 | } |
1535 | } | |
1536 | return 0; | |
1537 | } | |
1538 | ||
394e31d2 XQ |
1539 | static struct page *new_node_page(struct page *page, unsigned long private, |
1540 | int **result) | |
1541 | { | |
1542 | gfp_t gfp_mask = GFP_USER | __GFP_MOVABLE; | |
1543 | int nid = page_to_nid(page); | |
231e97e2 LZ |
1544 | nodemask_t nmask = node_states[N_MEMORY]; |
1545 | struct page *new_page = NULL; | |
394e31d2 XQ |
1546 | |
1547 | /* | |
1548 | * TODO: allocate a destination hugepage from a nearest neighbor node, | |
1549 | * accordance with memory policy of the user process if possible. For | |
1550 | * now as a simple work-around, we use the next node for destination. | |
1551 | */ | |
1552 | if (PageHuge(page)) | |
1553 | return alloc_huge_page_node(page_hstate(compound_head(page)), | |
1554 | next_node_in(nid, nmask)); | |
1555 | ||
231e97e2 | 1556 | node_clear(nid, nmask); |
9bb627be | 1557 | |
394e31d2 XQ |
1558 | if (PageHighMem(page) |
1559 | || (zone_idx(page_zone(page)) == ZONE_MOVABLE)) | |
1560 | gfp_mask |= __GFP_HIGHMEM; | |
1561 | ||
231e97e2 LZ |
1562 | if (!nodes_empty(nmask)) |
1563 | new_page = __alloc_pages_nodemask(gfp_mask, 0, | |
394e31d2 XQ |
1564 | node_zonelist(nid, gfp_mask), &nmask); |
1565 | if (!new_page) | |
1566 | new_page = __alloc_pages(gfp_mask, 0, | |
1567 | node_zonelist(nid, gfp_mask)); | |
1568 | ||
1569 | return new_page; | |
1570 | } | |
1571 | ||
0c0e6195 KH |
1572 | #define NR_OFFLINE_AT_ONCE_PAGES (256) |
1573 | static int | |
1574 | do_migrate_range(unsigned long start_pfn, unsigned long end_pfn) | |
1575 | { | |
1576 | unsigned long pfn; | |
1577 | struct page *page; | |
1578 | int move_pages = NR_OFFLINE_AT_ONCE_PAGES; | |
1579 | int not_managed = 0; | |
1580 | int ret = 0; | |
1581 | LIST_HEAD(source); | |
1582 | ||
1583 | for (pfn = start_pfn; pfn < end_pfn && move_pages > 0; pfn++) { | |
1584 | if (!pfn_valid(pfn)) | |
1585 | continue; | |
1586 | page = pfn_to_page(pfn); | |
c8721bbb NH |
1587 | |
1588 | if (PageHuge(page)) { | |
1589 | struct page *head = compound_head(page); | |
1590 | pfn = page_to_pfn(head) + (1<<compound_order(head)) - 1; | |
1591 | if (compound_order(head) > PFN_SECTION_SHIFT) { | |
1592 | ret = -EBUSY; | |
1593 | break; | |
1594 | } | |
1595 | if (isolate_huge_page(page, &source)) | |
1596 | move_pages -= 1 << compound_order(head); | |
1597 | continue; | |
1598 | } | |
1599 | ||
700c2a46 | 1600 | if (!get_page_unless_zero(page)) |
0c0e6195 KH |
1601 | continue; |
1602 | /* | |
1603 | * We can skip free pages. And we can only deal with pages on | |
1604 | * LRU. | |
1605 | */ | |
62695a84 | 1606 | ret = isolate_lru_page(page); |
0c0e6195 | 1607 | if (!ret) { /* Success */ |
700c2a46 | 1608 | put_page(page); |
62695a84 | 1609 | list_add_tail(&page->lru, &source); |
0c0e6195 | 1610 | move_pages--; |
599d0c95 | 1611 | inc_node_page_state(page, NR_ISOLATED_ANON + |
6d9c285a KM |
1612 | page_is_file_cache(page)); |
1613 | ||
0c0e6195 | 1614 | } else { |
0c0e6195 | 1615 | #ifdef CONFIG_DEBUG_VM |
e33e33b4 | 1616 | pr_alert("removing pfn %lx from LRU failed\n", pfn); |
f0b791a3 | 1617 | dump_page(page, "failed to remove from LRU"); |
0c0e6195 | 1618 | #endif |
700c2a46 | 1619 | put_page(page); |
25985edc | 1620 | /* Because we don't have big zone->lock. we should |
809c4449 BL |
1621 | check this again here. */ |
1622 | if (page_count(page)) { | |
1623 | not_managed++; | |
f3ab2636 | 1624 | ret = -EBUSY; |
809c4449 BL |
1625 | break; |
1626 | } | |
0c0e6195 KH |
1627 | } |
1628 | } | |
f3ab2636 BL |
1629 | if (!list_empty(&source)) { |
1630 | if (not_managed) { | |
c8721bbb | 1631 | putback_movable_pages(&source); |
f3ab2636 BL |
1632 | goto out; |
1633 | } | |
74c08f98 | 1634 | |
394e31d2 XQ |
1635 | /* Allocate a new page from the nearest neighbor node */ |
1636 | ret = migrate_pages(&source, new_node_page, NULL, 0, | |
9c620e2b | 1637 | MIGRATE_SYNC, MR_MEMORY_HOTPLUG); |
f3ab2636 | 1638 | if (ret) |
c8721bbb | 1639 | putback_movable_pages(&source); |
0c0e6195 | 1640 | } |
0c0e6195 KH |
1641 | out: |
1642 | return ret; | |
1643 | } | |
1644 | ||
1645 | /* | |
1646 | * remove from free_area[] and mark all as Reserved. | |
1647 | */ | |
1648 | static int | |
1649 | offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages, | |
1650 | void *data) | |
1651 | { | |
1652 | __offline_isolated_pages(start, start + nr_pages); | |
1653 | return 0; | |
1654 | } | |
1655 | ||
1656 | static void | |
1657 | offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn) | |
1658 | { | |
908eedc6 | 1659 | walk_system_ram_range(start_pfn, end_pfn - start_pfn, NULL, |
0c0e6195 KH |
1660 | offline_isolated_pages_cb); |
1661 | } | |
1662 | ||
1663 | /* | |
1664 | * Check all pages in range, recoreded as memory resource, are isolated. | |
1665 | */ | |
1666 | static int | |
1667 | check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages, | |
1668 | void *data) | |
1669 | { | |
1670 | int ret; | |
1671 | long offlined = *(long *)data; | |
b023f468 | 1672 | ret = test_pages_isolated(start_pfn, start_pfn + nr_pages, true); |
0c0e6195 KH |
1673 | offlined = nr_pages; |
1674 | if (!ret) | |
1675 | *(long *)data += offlined; | |
1676 | return ret; | |
1677 | } | |
1678 | ||
1679 | static long | |
1680 | check_pages_isolated(unsigned long start_pfn, unsigned long end_pfn) | |
1681 | { | |
1682 | long offlined = 0; | |
1683 | int ret; | |
1684 | ||
908eedc6 | 1685 | ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn, &offlined, |
0c0e6195 KH |
1686 | check_pages_isolated_cb); |
1687 | if (ret < 0) | |
1688 | offlined = (long)ret; | |
1689 | return offlined; | |
1690 | } | |
1691 | ||
09285af7 | 1692 | #ifdef CONFIG_MOVABLE_NODE |
79a4dcef TC |
1693 | /* |
1694 | * When CONFIG_MOVABLE_NODE, we permit offlining of a node which doesn't have | |
1695 | * normal memory. | |
1696 | */ | |
09285af7 LJ |
1697 | static bool can_offline_normal(struct zone *zone, unsigned long nr_pages) |
1698 | { | |
1699 | return true; | |
1700 | } | |
79a4dcef | 1701 | #else /* CONFIG_MOVABLE_NODE */ |
74d42d8f LJ |
1702 | /* ensure the node has NORMAL memory if it is still online */ |
1703 | static bool can_offline_normal(struct zone *zone, unsigned long nr_pages) | |
1704 | { | |
1705 | struct pglist_data *pgdat = zone->zone_pgdat; | |
1706 | unsigned long present_pages = 0; | |
1707 | enum zone_type zt; | |
1708 | ||
1709 | for (zt = 0; zt <= ZONE_NORMAL; zt++) | |
1710 | present_pages += pgdat->node_zones[zt].present_pages; | |
1711 | ||
1712 | if (present_pages > nr_pages) | |
1713 | return true; | |
1714 | ||
1715 | present_pages = 0; | |
1716 | for (; zt <= ZONE_MOVABLE; zt++) | |
1717 | present_pages += pgdat->node_zones[zt].present_pages; | |
1718 | ||
1719 | /* | |
1720 | * we can't offline the last normal memory until all | |
1721 | * higher memory is offlined. | |
1722 | */ | |
1723 | return present_pages == 0; | |
1724 | } | |
79a4dcef | 1725 | #endif /* CONFIG_MOVABLE_NODE */ |
74d42d8f | 1726 | |
c5320926 TC |
1727 | static int __init cmdline_parse_movable_node(char *p) |
1728 | { | |
1729 | #ifdef CONFIG_MOVABLE_NODE | |
55ac590c | 1730 | movable_node_enabled = true; |
c5320926 TC |
1731 | #else |
1732 | pr_warn("movable_node option not supported\n"); | |
1733 | #endif | |
1734 | return 0; | |
1735 | } | |
1736 | early_param("movable_node", cmdline_parse_movable_node); | |
1737 | ||
d9713679 LJ |
1738 | /* check which state of node_states will be changed when offline memory */ |
1739 | static void node_states_check_changes_offline(unsigned long nr_pages, | |
1740 | struct zone *zone, struct memory_notify *arg) | |
1741 | { | |
1742 | struct pglist_data *pgdat = zone->zone_pgdat; | |
1743 | unsigned long present_pages = 0; | |
1744 | enum zone_type zt, zone_last = ZONE_NORMAL; | |
1745 | ||
1746 | /* | |
6715ddf9 LJ |
1747 | * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY] |
1748 | * contains nodes which have zones of 0...ZONE_NORMAL, | |
1749 | * set zone_last to ZONE_NORMAL. | |
d9713679 | 1750 | * |
6715ddf9 LJ |
1751 | * If we don't have HIGHMEM nor movable node, |
1752 | * node_states[N_NORMAL_MEMORY] contains nodes which have zones of | |
1753 | * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE. | |
d9713679 | 1754 | */ |
6715ddf9 | 1755 | if (N_MEMORY == N_NORMAL_MEMORY) |
d9713679 LJ |
1756 | zone_last = ZONE_MOVABLE; |
1757 | ||
1758 | /* | |
1759 | * check whether node_states[N_NORMAL_MEMORY] will be changed. | |
1760 | * If the memory to be offline is in a zone of 0...zone_last, | |
1761 | * and it is the last present memory, 0...zone_last will | |
1762 | * become empty after offline , thus we can determind we will | |
1763 | * need to clear the node from node_states[N_NORMAL_MEMORY]. | |
1764 | */ | |
1765 | for (zt = 0; zt <= zone_last; zt++) | |
1766 | present_pages += pgdat->node_zones[zt].present_pages; | |
1767 | if (zone_idx(zone) <= zone_last && nr_pages >= present_pages) | |
1768 | arg->status_change_nid_normal = zone_to_nid(zone); | |
1769 | else | |
1770 | arg->status_change_nid_normal = -1; | |
1771 | ||
6715ddf9 LJ |
1772 | #ifdef CONFIG_HIGHMEM |
1773 | /* | |
1774 | * If we have movable node, node_states[N_HIGH_MEMORY] | |
1775 | * contains nodes which have zones of 0...ZONE_HIGHMEM, | |
1776 | * set zone_last to ZONE_HIGHMEM. | |
1777 | * | |
1778 | * If we don't have movable node, node_states[N_NORMAL_MEMORY] | |
1779 | * contains nodes which have zones of 0...ZONE_MOVABLE, | |
1780 | * set zone_last to ZONE_MOVABLE. | |
1781 | */ | |
1782 | zone_last = ZONE_HIGHMEM; | |
1783 | if (N_MEMORY == N_HIGH_MEMORY) | |
1784 | zone_last = ZONE_MOVABLE; | |
1785 | ||
1786 | for (; zt <= zone_last; zt++) | |
1787 | present_pages += pgdat->node_zones[zt].present_pages; | |
1788 | if (zone_idx(zone) <= zone_last && nr_pages >= present_pages) | |
1789 | arg->status_change_nid_high = zone_to_nid(zone); | |
1790 | else | |
1791 | arg->status_change_nid_high = -1; | |
1792 | #else | |
1793 | arg->status_change_nid_high = arg->status_change_nid_normal; | |
1794 | #endif | |
1795 | ||
d9713679 LJ |
1796 | /* |
1797 | * node_states[N_HIGH_MEMORY] contains nodes which have 0...ZONE_MOVABLE | |
1798 | */ | |
1799 | zone_last = ZONE_MOVABLE; | |
1800 | ||
1801 | /* | |
1802 | * check whether node_states[N_HIGH_MEMORY] will be changed | |
1803 | * If we try to offline the last present @nr_pages from the node, | |
1804 | * we can determind we will need to clear the node from | |
1805 | * node_states[N_HIGH_MEMORY]. | |
1806 | */ | |
1807 | for (; zt <= zone_last; zt++) | |
1808 | present_pages += pgdat->node_zones[zt].present_pages; | |
1809 | if (nr_pages >= present_pages) | |
1810 | arg->status_change_nid = zone_to_nid(zone); | |
1811 | else | |
1812 | arg->status_change_nid = -1; | |
1813 | } | |
1814 | ||
1815 | static void node_states_clear_node(int node, struct memory_notify *arg) | |
1816 | { | |
1817 | if (arg->status_change_nid_normal >= 0) | |
1818 | node_clear_state(node, N_NORMAL_MEMORY); | |
1819 | ||
6715ddf9 LJ |
1820 | if ((N_MEMORY != N_NORMAL_MEMORY) && |
1821 | (arg->status_change_nid_high >= 0)) | |
d9713679 | 1822 | node_clear_state(node, N_HIGH_MEMORY); |
6715ddf9 LJ |
1823 | |
1824 | if ((N_MEMORY != N_HIGH_MEMORY) && | |
1825 | (arg->status_change_nid >= 0)) | |
1826 | node_clear_state(node, N_MEMORY); | |
d9713679 LJ |
1827 | } |
1828 | ||
a16cee10 | 1829 | static int __ref __offline_pages(unsigned long start_pfn, |
0c0e6195 KH |
1830 | unsigned long end_pfn, unsigned long timeout) |
1831 | { | |
1832 | unsigned long pfn, nr_pages, expire; | |
1833 | long offlined_pages; | |
7b78d335 | 1834 | int ret, drain, retry_max, node; |
d702909f | 1835 | unsigned long flags; |
0c0e6195 | 1836 | struct zone *zone; |
7b78d335 | 1837 | struct memory_notify arg; |
0c0e6195 | 1838 | |
0c0e6195 KH |
1839 | /* at least, alignment against pageblock is necessary */ |
1840 | if (!IS_ALIGNED(start_pfn, pageblock_nr_pages)) | |
1841 | return -EINVAL; | |
1842 | if (!IS_ALIGNED(end_pfn, pageblock_nr_pages)) | |
1843 | return -EINVAL; | |
1844 | /* This makes hotplug much easier...and readable. | |
1845 | we assume this for now. .*/ | |
1846 | if (!test_pages_in_a_zone(start_pfn, end_pfn)) | |
1847 | return -EINVAL; | |
7b78d335 YG |
1848 | |
1849 | zone = page_zone(pfn_to_page(start_pfn)); | |
1850 | node = zone_to_nid(zone); | |
1851 | nr_pages = end_pfn - start_pfn; | |
1852 | ||
74d42d8f | 1853 | if (zone_idx(zone) <= ZONE_NORMAL && !can_offline_normal(zone, nr_pages)) |
30467e0b | 1854 | return -EINVAL; |
74d42d8f | 1855 | |
0c0e6195 | 1856 | /* set above range as isolated */ |
b023f468 WC |
1857 | ret = start_isolate_page_range(start_pfn, end_pfn, |
1858 | MIGRATE_MOVABLE, true); | |
0c0e6195 | 1859 | if (ret) |
30467e0b | 1860 | return ret; |
7b78d335 YG |
1861 | |
1862 | arg.start_pfn = start_pfn; | |
1863 | arg.nr_pages = nr_pages; | |
d9713679 | 1864 | node_states_check_changes_offline(nr_pages, zone, &arg); |
7b78d335 YG |
1865 | |
1866 | ret = memory_notify(MEM_GOING_OFFLINE, &arg); | |
1867 | ret = notifier_to_errno(ret); | |
1868 | if (ret) | |
1869 | goto failed_removal; | |
1870 | ||
0c0e6195 KH |
1871 | pfn = start_pfn; |
1872 | expire = jiffies + timeout; | |
1873 | drain = 0; | |
1874 | retry_max = 5; | |
1875 | repeat: | |
1876 | /* start memory hot removal */ | |
1877 | ret = -EAGAIN; | |
1878 | if (time_after(jiffies, expire)) | |
1879 | goto failed_removal; | |
1880 | ret = -EINTR; | |
1881 | if (signal_pending(current)) | |
1882 | goto failed_removal; | |
1883 | ret = 0; | |
1884 | if (drain) { | |
1885 | lru_add_drain_all(); | |
0c0e6195 | 1886 | cond_resched(); |
c0554329 | 1887 | drain_all_pages(zone); |
0c0e6195 KH |
1888 | } |
1889 | ||
c8721bbb NH |
1890 | pfn = scan_movable_pages(start_pfn, end_pfn); |
1891 | if (pfn) { /* We have movable pages */ | |
0c0e6195 KH |
1892 | ret = do_migrate_range(pfn, end_pfn); |
1893 | if (!ret) { | |
1894 | drain = 1; | |
1895 | goto repeat; | |
1896 | } else { | |
1897 | if (ret < 0) | |
1898 | if (--retry_max == 0) | |
1899 | goto failed_removal; | |
1900 | yield(); | |
1901 | drain = 1; | |
1902 | goto repeat; | |
1903 | } | |
1904 | } | |
b3834be5 | 1905 | /* drain all zone's lru pagevec, this is asynchronous... */ |
0c0e6195 | 1906 | lru_add_drain_all(); |
0c0e6195 | 1907 | yield(); |
b3834be5 | 1908 | /* drain pcp pages, this is synchronous. */ |
c0554329 | 1909 | drain_all_pages(zone); |
c8721bbb NH |
1910 | /* |
1911 | * dissolve free hugepages in the memory block before doing offlining | |
1912 | * actually in order to make hugetlbfs's object counting consistent. | |
1913 | */ | |
082d5b6b GS |
1914 | ret = dissolve_free_huge_pages(start_pfn, end_pfn); |
1915 | if (ret) | |
1916 | goto failed_removal; | |
0c0e6195 KH |
1917 | /* check again */ |
1918 | offlined_pages = check_pages_isolated(start_pfn, end_pfn); | |
1919 | if (offlined_pages < 0) { | |
1920 | ret = -EBUSY; | |
1921 | goto failed_removal; | |
1922 | } | |
e33e33b4 | 1923 | pr_info("Offlined Pages %ld\n", offlined_pages); |
b3834be5 | 1924 | /* Ok, all of our target is isolated. |
0c0e6195 KH |
1925 | We cannot do rollback at this point. */ |
1926 | offline_isolated_pages(start_pfn, end_pfn); | |
dbc0e4ce | 1927 | /* reset pagetype flags and makes migrate type to be MOVABLE */ |
0815f3d8 | 1928 | undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE); |
0c0e6195 | 1929 | /* removal success */ |
3dcc0571 | 1930 | adjust_managed_page_count(pfn_to_page(start_pfn), -offlined_pages); |
0c0e6195 | 1931 | zone->present_pages -= offlined_pages; |
d702909f CS |
1932 | |
1933 | pgdat_resize_lock(zone->zone_pgdat, &flags); | |
0c0e6195 | 1934 | zone->zone_pgdat->node_present_pages -= offlined_pages; |
d702909f | 1935 | pgdat_resize_unlock(zone->zone_pgdat, &flags); |
7b78d335 | 1936 | |
1b79acc9 KM |
1937 | init_per_zone_wmark_min(); |
1938 | ||
1e8537ba | 1939 | if (!populated_zone(zone)) { |
340175b7 | 1940 | zone_pcp_reset(zone); |
1e8537ba XQ |
1941 | mutex_lock(&zonelists_mutex); |
1942 | build_all_zonelists(NULL, NULL); | |
1943 | mutex_unlock(&zonelists_mutex); | |
1944 | } else | |
1945 | zone_pcp_update(zone); | |
340175b7 | 1946 | |
d9713679 | 1947 | node_states_clear_node(node, &arg); |
698b1b30 | 1948 | if (arg.status_change_nid >= 0) { |
8fe23e05 | 1949 | kswapd_stop(node); |
698b1b30 VB |
1950 | kcompactd_stop(node); |
1951 | } | |
bce7394a | 1952 | |
0c0e6195 KH |
1953 | vm_total_pages = nr_free_pagecache_pages(); |
1954 | writeback_set_ratelimit(); | |
7b78d335 YG |
1955 | |
1956 | memory_notify(MEM_OFFLINE, &arg); | |
0c0e6195 KH |
1957 | return 0; |
1958 | ||
1959 | failed_removal: | |
e33e33b4 CY |
1960 | pr_debug("memory offlining [mem %#010llx-%#010llx] failed\n", |
1961 | (unsigned long long) start_pfn << PAGE_SHIFT, | |
1962 | ((unsigned long long) end_pfn << PAGE_SHIFT) - 1); | |
7b78d335 | 1963 | memory_notify(MEM_CANCEL_OFFLINE, &arg); |
0c0e6195 | 1964 | /* pushback to free area */ |
0815f3d8 | 1965 | undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE); |
0c0e6195 KH |
1966 | return ret; |
1967 | } | |
71088785 | 1968 | |
30467e0b | 1969 | /* Must be protected by mem_hotplug_begin() */ |
a16cee10 WC |
1970 | int offline_pages(unsigned long start_pfn, unsigned long nr_pages) |
1971 | { | |
1972 | return __offline_pages(start_pfn, start_pfn + nr_pages, 120 * HZ); | |
1973 | } | |
e2ff3940 | 1974 | #endif /* CONFIG_MEMORY_HOTREMOVE */ |
a16cee10 | 1975 | |
bbc76be6 WC |
1976 | /** |
1977 | * walk_memory_range - walks through all mem sections in [start_pfn, end_pfn) | |
1978 | * @start_pfn: start pfn of the memory range | |
e05c4bbf | 1979 | * @end_pfn: end pfn of the memory range |
bbc76be6 WC |
1980 | * @arg: argument passed to func |
1981 | * @func: callback for each memory section walked | |
1982 | * | |
1983 | * This function walks through all present mem sections in range | |
1984 | * [start_pfn, end_pfn) and call func on each mem section. | |
1985 | * | |
1986 | * Returns the return value of func. | |
1987 | */ | |
e2ff3940 | 1988 | int walk_memory_range(unsigned long start_pfn, unsigned long end_pfn, |
bbc76be6 | 1989 | void *arg, int (*func)(struct memory_block *, void *)) |
71088785 | 1990 | { |
e90bdb7f WC |
1991 | struct memory_block *mem = NULL; |
1992 | struct mem_section *section; | |
e90bdb7f WC |
1993 | unsigned long pfn, section_nr; |
1994 | int ret; | |
e90bdb7f WC |
1995 | |
1996 | for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) { | |
1997 | section_nr = pfn_to_section_nr(pfn); | |
1998 | if (!present_section_nr(section_nr)) | |
1999 | continue; | |
2000 | ||
2001 | section = __nr_to_section(section_nr); | |
2002 | /* same memblock? */ | |
2003 | if (mem) | |
2004 | if ((section_nr >= mem->start_section_nr) && | |
2005 | (section_nr <= mem->end_section_nr)) | |
2006 | continue; | |
2007 | ||
2008 | mem = find_memory_block_hinted(section, mem); | |
2009 | if (!mem) | |
2010 | continue; | |
2011 | ||
bbc76be6 | 2012 | ret = func(mem, arg); |
e90bdb7f | 2013 | if (ret) { |
bbc76be6 WC |
2014 | kobject_put(&mem->dev.kobj); |
2015 | return ret; | |
e90bdb7f WC |
2016 | } |
2017 | } | |
2018 | ||
2019 | if (mem) | |
2020 | kobject_put(&mem->dev.kobj); | |
2021 | ||
bbc76be6 WC |
2022 | return 0; |
2023 | } | |
2024 | ||
e2ff3940 | 2025 | #ifdef CONFIG_MEMORY_HOTREMOVE |
d6de9d53 | 2026 | static int check_memblock_offlined_cb(struct memory_block *mem, void *arg) |
bbc76be6 WC |
2027 | { |
2028 | int ret = !is_memblock_offlined(mem); | |
2029 | ||
349daa0f RD |
2030 | if (unlikely(ret)) { |
2031 | phys_addr_t beginpa, endpa; | |
2032 | ||
2033 | beginpa = PFN_PHYS(section_nr_to_pfn(mem->start_section_nr)); | |
2034 | endpa = PFN_PHYS(section_nr_to_pfn(mem->end_section_nr + 1))-1; | |
756a025f | 2035 | pr_warn("removing memory fails, because memory [%pa-%pa] is onlined\n", |
349daa0f RD |
2036 | &beginpa, &endpa); |
2037 | } | |
bbc76be6 WC |
2038 | |
2039 | return ret; | |
2040 | } | |
2041 | ||
0f1cfe9d | 2042 | static int check_cpu_on_node(pg_data_t *pgdat) |
60a5a19e | 2043 | { |
60a5a19e TC |
2044 | int cpu; |
2045 | ||
2046 | for_each_present_cpu(cpu) { | |
2047 | if (cpu_to_node(cpu) == pgdat->node_id) | |
2048 | /* | |
2049 | * the cpu on this node isn't removed, and we can't | |
2050 | * offline this node. | |
2051 | */ | |
2052 | return -EBUSY; | |
2053 | } | |
2054 | ||
2055 | return 0; | |
2056 | } | |
2057 | ||
0f1cfe9d | 2058 | static void unmap_cpu_on_node(pg_data_t *pgdat) |
e13fe869 WC |
2059 | { |
2060 | #ifdef CONFIG_ACPI_NUMA | |
e13fe869 WC |
2061 | int cpu; |
2062 | ||
2063 | for_each_possible_cpu(cpu) | |
2064 | if (cpu_to_node(cpu) == pgdat->node_id) | |
2065 | numa_clear_node(cpu); | |
2066 | #endif | |
2067 | } | |
2068 | ||
0f1cfe9d | 2069 | static int check_and_unmap_cpu_on_node(pg_data_t *pgdat) |
e13fe869 | 2070 | { |
0f1cfe9d | 2071 | int ret; |
e13fe869 | 2072 | |
0f1cfe9d | 2073 | ret = check_cpu_on_node(pgdat); |
e13fe869 WC |
2074 | if (ret) |
2075 | return ret; | |
2076 | ||
2077 | /* | |
2078 | * the node will be offlined when we come here, so we can clear | |
2079 | * the cpu_to_node() now. | |
2080 | */ | |
2081 | ||
0f1cfe9d | 2082 | unmap_cpu_on_node(pgdat); |
e13fe869 WC |
2083 | return 0; |
2084 | } | |
2085 | ||
0f1cfe9d TK |
2086 | /** |
2087 | * try_offline_node | |
2088 | * | |
2089 | * Offline a node if all memory sections and cpus of the node are removed. | |
2090 | * | |
2091 | * NOTE: The caller must call lock_device_hotplug() to serialize hotplug | |
2092 | * and online/offline operations before this call. | |
2093 | */ | |
90b30cdc | 2094 | void try_offline_node(int nid) |
60a5a19e | 2095 | { |
d822b86a WC |
2096 | pg_data_t *pgdat = NODE_DATA(nid); |
2097 | unsigned long start_pfn = pgdat->node_start_pfn; | |
2098 | unsigned long end_pfn = start_pfn + pgdat->node_spanned_pages; | |
60a5a19e TC |
2099 | unsigned long pfn; |
2100 | ||
2101 | for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) { | |
2102 | unsigned long section_nr = pfn_to_section_nr(pfn); | |
2103 | ||
2104 | if (!present_section_nr(section_nr)) | |
2105 | continue; | |
2106 | ||
2107 | if (pfn_to_nid(pfn) != nid) | |
2108 | continue; | |
2109 | ||
2110 | /* | |
2111 | * some memory sections of this node are not removed, and we | |
2112 | * can't offline node now. | |
2113 | */ | |
2114 | return; | |
2115 | } | |
2116 | ||
0f1cfe9d | 2117 | if (check_and_unmap_cpu_on_node(pgdat)) |
60a5a19e TC |
2118 | return; |
2119 | ||
2120 | /* | |
2121 | * all memory/cpu of this node are removed, we can offline this | |
2122 | * node now. | |
2123 | */ | |
2124 | node_set_offline(nid); | |
2125 | unregister_one_node(nid); | |
2126 | } | |
90b30cdc | 2127 | EXPORT_SYMBOL(try_offline_node); |
60a5a19e | 2128 | |
0f1cfe9d TK |
2129 | /** |
2130 | * remove_memory | |
2131 | * | |
2132 | * NOTE: The caller must call lock_device_hotplug() to serialize hotplug | |
2133 | * and online/offline operations before this call, as required by | |
2134 | * try_offline_node(). | |
2135 | */ | |
242831eb | 2136 | void __ref remove_memory(int nid, u64 start, u64 size) |
bbc76be6 | 2137 | { |
242831eb | 2138 | int ret; |
993c1aad | 2139 | |
27356f54 TK |
2140 | BUG_ON(check_hotplug_memory_range(start, size)); |
2141 | ||
bfc8c901 | 2142 | mem_hotplug_begin(); |
6677e3ea YI |
2143 | |
2144 | /* | |
242831eb RW |
2145 | * All memory blocks must be offlined before removing memory. Check |
2146 | * whether all memory blocks in question are offline and trigger a BUG() | |
2147 | * if this is not the case. | |
6677e3ea | 2148 | */ |
242831eb | 2149 | ret = walk_memory_range(PFN_DOWN(start), PFN_UP(start + size - 1), NULL, |
d6de9d53 | 2150 | check_memblock_offlined_cb); |
bfc8c901 | 2151 | if (ret) |
242831eb | 2152 | BUG(); |
6677e3ea | 2153 | |
46c66c4b YI |
2154 | /* remove memmap entry */ |
2155 | firmware_map_remove(start, start + size, "System RAM"); | |
f9126ab9 XQ |
2156 | memblock_free(start, size); |
2157 | memblock_remove(start, size); | |
46c66c4b | 2158 | |
24d335ca WC |
2159 | arch_remove_memory(start, size); |
2160 | ||
60a5a19e TC |
2161 | try_offline_node(nid); |
2162 | ||
bfc8c901 | 2163 | mem_hotplug_done(); |
71088785 | 2164 | } |
71088785 | 2165 | EXPORT_SYMBOL_GPL(remove_memory); |
aba6efc4 | 2166 | #endif /* CONFIG_MEMORY_HOTREMOVE */ |