]>
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> | |
174cd4b1 | 9 | #include <linux/sched/signal.h> |
3947be19 DH |
10 | #include <linux/swap.h> |
11 | #include <linux/interrupt.h> | |
12 | #include <linux/pagemap.h> | |
3947be19 | 13 | #include <linux/compiler.h> |
b95f1b31 | 14 | #include <linux/export.h> |
3947be19 | 15 | #include <linux/pagevec.h> |
2d1d43f6 | 16 | #include <linux/writeback.h> |
3947be19 DH |
17 | #include <linux/slab.h> |
18 | #include <linux/sysctl.h> | |
19 | #include <linux/cpu.h> | |
20 | #include <linux/memory.h> | |
4b94ffdc | 21 | #include <linux/memremap.h> |
3947be19 DH |
22 | #include <linux/memory_hotplug.h> |
23 | #include <linux/highmem.h> | |
24 | #include <linux/vmalloc.h> | |
0a547039 | 25 | #include <linux/ioport.h> |
0c0e6195 KH |
26 | #include <linux/delay.h> |
27 | #include <linux/migrate.h> | |
28 | #include <linux/page-isolation.h> | |
71088785 | 29 | #include <linux/pfn.h> |
6ad696d2 | 30 | #include <linux/suspend.h> |
6d9c285a | 31 | #include <linux/mm_inline.h> |
d96ae530 | 32 | #include <linux/firmware-map.h> |
60a5a19e | 33 | #include <linux/stop_machine.h> |
c8721bbb | 34 | #include <linux/hugetlb.h> |
c5320926 | 35 | #include <linux/memblock.h> |
f784a3f1 | 36 | #include <linux/bootmem.h> |
698b1b30 | 37 | #include <linux/compaction.h> |
3947be19 DH |
38 | |
39 | #include <asm/tlbflush.h> | |
40 | ||
1e5ad9a3 AB |
41 | #include "internal.h" |
42 | ||
9d0ad8ca DK |
43 | /* |
44 | * online_page_callback contains pointer to current page onlining function. | |
45 | * Initially it is generic_online_page(). If it is required it could be | |
46 | * changed by calling set_online_page_callback() for callback registration | |
47 | * and restore_online_page_callback() for generic callback restore. | |
48 | */ | |
49 | ||
50 | static void generic_online_page(struct page *page); | |
51 | ||
52 | static online_page_callback_t online_page_callback = generic_online_page; | |
bfc8c901 | 53 | static DEFINE_MUTEX(online_page_callback_lock); |
9d0ad8ca | 54 | |
bfc8c901 VD |
55 | /* The same as the cpu_hotplug lock, but for memory hotplug. */ |
56 | static struct { | |
57 | struct task_struct *active_writer; | |
58 | struct mutex lock; /* Synchronizes accesses to refcount, */ | |
59 | /* | |
60 | * Also blocks the new readers during | |
61 | * an ongoing mem hotplug operation. | |
62 | */ | |
63 | int refcount; | |
64 | ||
65 | #ifdef CONFIG_DEBUG_LOCK_ALLOC | |
66 | struct lockdep_map dep_map; | |
67 | #endif | |
68 | } mem_hotplug = { | |
69 | .active_writer = NULL, | |
70 | .lock = __MUTEX_INITIALIZER(mem_hotplug.lock), | |
71 | .refcount = 0, | |
72 | #ifdef CONFIG_DEBUG_LOCK_ALLOC | |
73 | .dep_map = {.name = "mem_hotplug.lock" }, | |
74 | #endif | |
75 | }; | |
76 | ||
77 | /* Lockdep annotations for get/put_online_mems() and mem_hotplug_begin/end() */ | |
78 | #define memhp_lock_acquire_read() lock_map_acquire_read(&mem_hotplug.dep_map) | |
79 | #define memhp_lock_acquire() lock_map_acquire(&mem_hotplug.dep_map) | |
80 | #define memhp_lock_release() lock_map_release(&mem_hotplug.dep_map) | |
81 | ||
4932381e MH |
82 | bool movable_node_enabled = false; |
83 | ||
8604d9e5 | 84 | #ifndef CONFIG_MEMORY_HOTPLUG_DEFAULT_ONLINE |
31bc3858 | 85 | bool memhp_auto_online; |
8604d9e5 VK |
86 | #else |
87 | bool memhp_auto_online = true; | |
88 | #endif | |
31bc3858 VK |
89 | EXPORT_SYMBOL_GPL(memhp_auto_online); |
90 | ||
86dd995d VK |
91 | static int __init setup_memhp_default_state(char *str) |
92 | { | |
93 | if (!strcmp(str, "online")) | |
94 | memhp_auto_online = true; | |
95 | else if (!strcmp(str, "offline")) | |
96 | memhp_auto_online = false; | |
97 | ||
98 | return 1; | |
99 | } | |
100 | __setup("memhp_default_state=", setup_memhp_default_state); | |
101 | ||
bfc8c901 VD |
102 | void get_online_mems(void) |
103 | { | |
104 | might_sleep(); | |
105 | if (mem_hotplug.active_writer == current) | |
106 | return; | |
107 | memhp_lock_acquire_read(); | |
108 | mutex_lock(&mem_hotplug.lock); | |
109 | mem_hotplug.refcount++; | |
110 | mutex_unlock(&mem_hotplug.lock); | |
111 | ||
112 | } | |
20d6c96b | 113 | |
bfc8c901 | 114 | void put_online_mems(void) |
20d6c96b | 115 | { |
bfc8c901 VD |
116 | if (mem_hotplug.active_writer == current) |
117 | return; | |
118 | mutex_lock(&mem_hotplug.lock); | |
119 | ||
120 | if (WARN_ON(!mem_hotplug.refcount)) | |
121 | mem_hotplug.refcount++; /* try to fix things up */ | |
122 | ||
123 | if (!--mem_hotplug.refcount && unlikely(mem_hotplug.active_writer)) | |
124 | wake_up_process(mem_hotplug.active_writer); | |
125 | mutex_unlock(&mem_hotplug.lock); | |
126 | memhp_lock_release(); | |
127 | ||
20d6c96b KM |
128 | } |
129 | ||
55adc1d0 HC |
130 | /* Serializes write accesses to mem_hotplug.active_writer. */ |
131 | static DEFINE_MUTEX(memory_add_remove_lock); | |
132 | ||
30467e0b | 133 | void mem_hotplug_begin(void) |
20d6c96b | 134 | { |
55adc1d0 | 135 | mutex_lock(&memory_add_remove_lock); |
3fc21924 | 136 | |
bfc8c901 VD |
137 | mem_hotplug.active_writer = current; |
138 | ||
139 | memhp_lock_acquire(); | |
140 | for (;;) { | |
141 | mutex_lock(&mem_hotplug.lock); | |
142 | if (likely(!mem_hotplug.refcount)) | |
143 | break; | |
144 | __set_current_state(TASK_UNINTERRUPTIBLE); | |
145 | mutex_unlock(&mem_hotplug.lock); | |
146 | schedule(); | |
147 | } | |
20d6c96b KM |
148 | } |
149 | ||
30467e0b | 150 | void mem_hotplug_done(void) |
bfc8c901 VD |
151 | { |
152 | mem_hotplug.active_writer = NULL; | |
153 | mutex_unlock(&mem_hotplug.lock); | |
154 | memhp_lock_release(); | |
55adc1d0 | 155 | mutex_unlock(&memory_add_remove_lock); |
bfc8c901 | 156 | } |
20d6c96b | 157 | |
45e0b78b KM |
158 | /* add this memory to iomem resource */ |
159 | static struct resource *register_memory_resource(u64 start, u64 size) | |
160 | { | |
161 | struct resource *res; | |
162 | res = kzalloc(sizeof(struct resource), GFP_KERNEL); | |
6f754ba4 VK |
163 | if (!res) |
164 | return ERR_PTR(-ENOMEM); | |
45e0b78b KM |
165 | |
166 | res->name = "System RAM"; | |
167 | res->start = start; | |
168 | res->end = start + size - 1; | |
782b8664 | 169 | res->flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY; |
45e0b78b | 170 | if (request_resource(&iomem_resource, res) < 0) { |
4996eed8 | 171 | pr_debug("System RAM resource %pR cannot be added\n", res); |
45e0b78b | 172 | kfree(res); |
6f754ba4 | 173 | return ERR_PTR(-EEXIST); |
45e0b78b KM |
174 | } |
175 | return res; | |
176 | } | |
177 | ||
178 | static void release_memory_resource(struct resource *res) | |
179 | { | |
180 | if (!res) | |
181 | return; | |
182 | release_resource(res); | |
183 | kfree(res); | |
184 | return; | |
185 | } | |
186 | ||
53947027 | 187 | #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE |
46723bfa YI |
188 | void get_page_bootmem(unsigned long info, struct page *page, |
189 | unsigned long type) | |
04753278 | 190 | { |
ddffe98d | 191 | page->freelist = (void *)type; |
04753278 YG |
192 | SetPagePrivate(page); |
193 | set_page_private(page, info); | |
fe896d18 | 194 | page_ref_inc(page); |
04753278 YG |
195 | } |
196 | ||
170a5a7e | 197 | void put_page_bootmem(struct page *page) |
04753278 | 198 | { |
5f24ce5f | 199 | unsigned long type; |
04753278 | 200 | |
ddffe98d | 201 | type = (unsigned long) page->freelist; |
5f24ce5f AA |
202 | BUG_ON(type < MEMORY_HOTPLUG_MIN_BOOTMEM_TYPE || |
203 | type > MEMORY_HOTPLUG_MAX_BOOTMEM_TYPE); | |
04753278 | 204 | |
fe896d18 | 205 | if (page_ref_dec_return(page) == 1) { |
ddffe98d | 206 | page->freelist = NULL; |
04753278 YG |
207 | ClearPagePrivate(page); |
208 | set_page_private(page, 0); | |
5f24ce5f | 209 | INIT_LIST_HEAD(&page->lru); |
170a5a7e | 210 | free_reserved_page(page); |
04753278 | 211 | } |
04753278 YG |
212 | } |
213 | ||
46723bfa YI |
214 | #ifdef CONFIG_HAVE_BOOTMEM_INFO_NODE |
215 | #ifndef CONFIG_SPARSEMEM_VMEMMAP | |
d92bc318 | 216 | static void register_page_bootmem_info_section(unsigned long start_pfn) |
04753278 YG |
217 | { |
218 | unsigned long *usemap, mapsize, section_nr, i; | |
219 | struct mem_section *ms; | |
220 | struct page *page, *memmap; | |
221 | ||
04753278 YG |
222 | section_nr = pfn_to_section_nr(start_pfn); |
223 | ms = __nr_to_section(section_nr); | |
224 | ||
225 | /* Get section's memmap address */ | |
226 | memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr); | |
227 | ||
228 | /* | |
229 | * Get page for the memmap's phys address | |
230 | * XXX: need more consideration for sparse_vmemmap... | |
231 | */ | |
232 | page = virt_to_page(memmap); | |
233 | mapsize = sizeof(struct page) * PAGES_PER_SECTION; | |
234 | mapsize = PAGE_ALIGN(mapsize) >> PAGE_SHIFT; | |
235 | ||
236 | /* remember memmap's page */ | |
237 | for (i = 0; i < mapsize; i++, page++) | |
238 | get_page_bootmem(section_nr, page, SECTION_INFO); | |
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++) | |
af370fb8 | 246 | get_page_bootmem(section_nr, page, MIX_SECTION_INFO); |
04753278 YG |
247 | |
248 | } | |
46723bfa YI |
249 | #else /* CONFIG_SPARSEMEM_VMEMMAP */ |
250 | static void register_page_bootmem_info_section(unsigned long start_pfn) | |
251 | { | |
252 | unsigned long *usemap, mapsize, section_nr, i; | |
253 | struct mem_section *ms; | |
254 | struct page *page, *memmap; | |
255 | ||
256 | if (!pfn_valid(start_pfn)) | |
257 | return; | |
258 | ||
259 | section_nr = pfn_to_section_nr(start_pfn); | |
260 | ms = __nr_to_section(section_nr); | |
261 | ||
262 | memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr); | |
263 | ||
264 | register_page_bootmem_memmap(section_nr, memmap, PAGES_PER_SECTION); | |
265 | ||
266 | usemap = __nr_to_section(section_nr)->pageblock_flags; | |
267 | page = virt_to_page(usemap); | |
268 | ||
269 | mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT; | |
270 | ||
271 | for (i = 0; i < mapsize; i++, page++) | |
272 | get_page_bootmem(section_nr, page, MIX_SECTION_INFO); | |
273 | } | |
274 | #endif /* !CONFIG_SPARSEMEM_VMEMMAP */ | |
04753278 | 275 | |
7ded384a | 276 | void __init register_page_bootmem_info_node(struct pglist_data *pgdat) |
04753278 YG |
277 | { |
278 | unsigned long i, pfn, end_pfn, nr_pages; | |
279 | int node = pgdat->node_id; | |
280 | struct page *page; | |
04753278 YG |
281 | |
282 | nr_pages = PAGE_ALIGN(sizeof(struct pglist_data)) >> PAGE_SHIFT; | |
283 | page = virt_to_page(pgdat); | |
284 | ||
285 | for (i = 0; i < nr_pages; i++, page++) | |
286 | get_page_bootmem(node, page, NODE_INFO); | |
287 | ||
04753278 | 288 | pfn = pgdat->node_start_pfn; |
c1f19495 | 289 | end_pfn = pgdat_end_pfn(pgdat); |
04753278 | 290 | |
7e9f5eb0 | 291 | /* register section info */ |
f14851af | 292 | for (; pfn < end_pfn; pfn += PAGES_PER_SECTION) { |
293 | /* | |
294 | * Some platforms can assign the same pfn to multiple nodes - on | |
295 | * node0 as well as nodeN. To avoid registering a pfn against | |
296 | * multiple nodes we check that this pfn does not already | |
7e9f5eb0 | 297 | * reside in some other nodes. |
f14851af | 298 | */ |
f65e91df | 299 | if (pfn_valid(pfn) && (early_pfn_to_nid(pfn) == node)) |
f14851af | 300 | register_page_bootmem_info_section(pfn); |
301 | } | |
04753278 | 302 | } |
46723bfa | 303 | #endif /* CONFIG_HAVE_BOOTMEM_INFO_NODE */ |
04753278 | 304 | |
f1dd2cd1 MH |
305 | static int __meminit __add_section(int nid, unsigned long phys_start_pfn, |
306 | bool want_memblock) | |
3947be19 | 307 | { |
3947be19 | 308 | int ret; |
f1dd2cd1 | 309 | int i; |
3947be19 | 310 | |
ebd15302 KH |
311 | if (pfn_valid(phys_start_pfn)) |
312 | return -EEXIST; | |
313 | ||
f1dd2cd1 | 314 | ret = sparse_add_one_section(NODE_DATA(nid), phys_start_pfn); |
3947be19 DH |
315 | if (ret < 0) |
316 | return ret; | |
317 | ||
f1dd2cd1 MH |
318 | /* |
319 | * Make all the pages reserved so that nobody will stumble over half | |
320 | * initialized state. | |
321 | * FIXME: We also have to associate it with a node because pfn_to_node | |
322 | * relies on having page with the proper node. | |
323 | */ | |
324 | for (i = 0; i < PAGES_PER_SECTION; i++) { | |
325 | unsigned long pfn = phys_start_pfn + i; | |
326 | struct page *page; | |
327 | if (!pfn_valid(pfn)) | |
328 | continue; | |
718127cc | 329 | |
f1dd2cd1 MH |
330 | page = pfn_to_page(pfn); |
331 | set_page_node(page, nid); | |
332 | SetPageReserved(page); | |
333 | } | |
718127cc | 334 | |
1b862aec MH |
335 | if (!want_memblock) |
336 | return 0; | |
337 | ||
c04fc586 | 338 | return register_new_memory(nid, __pfn_to_section(phys_start_pfn)); |
3947be19 DH |
339 | } |
340 | ||
4edd7cef DR |
341 | /* |
342 | * Reasonably generic function for adding memory. It is | |
343 | * expected that archs that support memory hotplug will | |
344 | * call this function after deciding the zone to which to | |
345 | * add the new pages. | |
346 | */ | |
f1dd2cd1 | 347 | int __ref __add_pages(int nid, unsigned long phys_start_pfn, |
1b862aec | 348 | unsigned long nr_pages, bool want_memblock) |
4edd7cef DR |
349 | { |
350 | unsigned long i; | |
351 | int err = 0; | |
352 | int start_sec, end_sec; | |
4b94ffdc DW |
353 | struct vmem_altmap *altmap; |
354 | ||
4edd7cef DR |
355 | /* during initialize mem_map, align hot-added range to section */ |
356 | start_sec = pfn_to_section_nr(phys_start_pfn); | |
357 | end_sec = pfn_to_section_nr(phys_start_pfn + nr_pages - 1); | |
358 | ||
4b94ffdc DW |
359 | altmap = to_vmem_altmap((unsigned long) pfn_to_page(phys_start_pfn)); |
360 | if (altmap) { | |
361 | /* | |
362 | * Validate altmap is within bounds of the total request | |
363 | */ | |
364 | if (altmap->base_pfn != phys_start_pfn | |
365 | || vmem_altmap_offset(altmap) > nr_pages) { | |
366 | pr_warn_once("memory add fail, invalid altmap\n"); | |
7cf91a98 JK |
367 | err = -EINVAL; |
368 | goto out; | |
4b94ffdc DW |
369 | } |
370 | altmap->alloc = 0; | |
371 | } | |
372 | ||
4edd7cef | 373 | for (i = start_sec; i <= end_sec; i++) { |
f1dd2cd1 | 374 | err = __add_section(nid, section_nr_to_pfn(i), want_memblock); |
4edd7cef DR |
375 | |
376 | /* | |
377 | * EEXIST is finally dealt with by ioresource collision | |
378 | * check. see add_memory() => register_memory_resource() | |
379 | * Warning will be printed if there is collision. | |
380 | */ | |
381 | if (err && (err != -EEXIST)) | |
382 | break; | |
383 | err = 0; | |
384 | } | |
c435a390 | 385 | vmemmap_populate_print_last(); |
7cf91a98 | 386 | out: |
4edd7cef DR |
387 | return err; |
388 | } | |
389 | EXPORT_SYMBOL_GPL(__add_pages); | |
390 | ||
391 | #ifdef CONFIG_MEMORY_HOTREMOVE | |
815121d2 YI |
392 | /* find the smallest valid pfn in the range [start_pfn, end_pfn) */ |
393 | static int find_smallest_section_pfn(int nid, struct zone *zone, | |
394 | unsigned long start_pfn, | |
395 | unsigned long end_pfn) | |
396 | { | |
397 | struct mem_section *ms; | |
398 | ||
399 | for (; start_pfn < end_pfn; start_pfn += PAGES_PER_SECTION) { | |
400 | ms = __pfn_to_section(start_pfn); | |
401 | ||
402 | if (unlikely(!valid_section(ms))) | |
403 | continue; | |
404 | ||
405 | if (unlikely(pfn_to_nid(start_pfn) != nid)) | |
406 | continue; | |
407 | ||
408 | if (zone && zone != page_zone(pfn_to_page(start_pfn))) | |
409 | continue; | |
410 | ||
411 | return start_pfn; | |
412 | } | |
413 | ||
414 | return 0; | |
415 | } | |
416 | ||
417 | /* find the biggest valid pfn in the range [start_pfn, end_pfn). */ | |
418 | static int find_biggest_section_pfn(int nid, struct zone *zone, | |
419 | unsigned long start_pfn, | |
420 | unsigned long end_pfn) | |
421 | { | |
422 | struct mem_section *ms; | |
423 | unsigned long pfn; | |
424 | ||
425 | /* pfn is the end pfn of a memory section. */ | |
426 | pfn = end_pfn - 1; | |
427 | for (; pfn >= start_pfn; pfn -= PAGES_PER_SECTION) { | |
428 | ms = __pfn_to_section(pfn); | |
429 | ||
430 | if (unlikely(!valid_section(ms))) | |
431 | continue; | |
432 | ||
433 | if (unlikely(pfn_to_nid(pfn) != nid)) | |
434 | continue; | |
435 | ||
436 | if (zone && zone != page_zone(pfn_to_page(pfn))) | |
437 | continue; | |
438 | ||
439 | return pfn; | |
440 | } | |
441 | ||
442 | return 0; | |
443 | } | |
444 | ||
445 | static void shrink_zone_span(struct zone *zone, unsigned long start_pfn, | |
446 | unsigned long end_pfn) | |
447 | { | |
c33bc315 XQ |
448 | unsigned long zone_start_pfn = zone->zone_start_pfn; |
449 | unsigned long z = zone_end_pfn(zone); /* zone_end_pfn namespace clash */ | |
450 | unsigned long zone_end_pfn = z; | |
815121d2 YI |
451 | unsigned long pfn; |
452 | struct mem_section *ms; | |
453 | int nid = zone_to_nid(zone); | |
454 | ||
455 | zone_span_writelock(zone); | |
456 | if (zone_start_pfn == start_pfn) { | |
457 | /* | |
458 | * If the section is smallest section in the zone, it need | |
459 | * shrink zone->zone_start_pfn and zone->zone_spanned_pages. | |
460 | * In this case, we find second smallest valid mem_section | |
461 | * for shrinking zone. | |
462 | */ | |
463 | pfn = find_smallest_section_pfn(nid, zone, end_pfn, | |
464 | zone_end_pfn); | |
465 | if (pfn) { | |
466 | zone->zone_start_pfn = pfn; | |
467 | zone->spanned_pages = zone_end_pfn - pfn; | |
468 | } | |
469 | } else if (zone_end_pfn == end_pfn) { | |
470 | /* | |
471 | * If the section is biggest section in the zone, it need | |
472 | * shrink zone->spanned_pages. | |
473 | * In this case, we find second biggest valid mem_section for | |
474 | * shrinking zone. | |
475 | */ | |
476 | pfn = find_biggest_section_pfn(nid, zone, zone_start_pfn, | |
477 | start_pfn); | |
478 | if (pfn) | |
479 | zone->spanned_pages = pfn - zone_start_pfn + 1; | |
480 | } | |
481 | ||
482 | /* | |
483 | * The section is not biggest or smallest mem_section in the zone, it | |
484 | * only creates a hole in the zone. So in this case, we need not | |
485 | * change the zone. But perhaps, the zone has only hole data. Thus | |
486 | * it check the zone has only hole or not. | |
487 | */ | |
488 | pfn = zone_start_pfn; | |
489 | for (; pfn < zone_end_pfn; pfn += PAGES_PER_SECTION) { | |
490 | ms = __pfn_to_section(pfn); | |
491 | ||
492 | if (unlikely(!valid_section(ms))) | |
493 | continue; | |
494 | ||
495 | if (page_zone(pfn_to_page(pfn)) != zone) | |
496 | continue; | |
497 | ||
498 | /* If the section is current section, it continues the loop */ | |
499 | if (start_pfn == pfn) | |
500 | continue; | |
501 | ||
502 | /* If we find valid section, we have nothing to do */ | |
503 | zone_span_writeunlock(zone); | |
504 | return; | |
505 | } | |
506 | ||
507 | /* The zone has no valid section */ | |
508 | zone->zone_start_pfn = 0; | |
509 | zone->spanned_pages = 0; | |
510 | zone_span_writeunlock(zone); | |
511 | } | |
512 | ||
513 | static void shrink_pgdat_span(struct pglist_data *pgdat, | |
514 | unsigned long start_pfn, unsigned long end_pfn) | |
515 | { | |
83285c72 XQ |
516 | unsigned long pgdat_start_pfn = pgdat->node_start_pfn; |
517 | unsigned long p = pgdat_end_pfn(pgdat); /* pgdat_end_pfn namespace clash */ | |
518 | unsigned long pgdat_end_pfn = p; | |
815121d2 YI |
519 | unsigned long pfn; |
520 | struct mem_section *ms; | |
521 | int nid = pgdat->node_id; | |
522 | ||
523 | if (pgdat_start_pfn == start_pfn) { | |
524 | /* | |
525 | * If the section is smallest section in the pgdat, it need | |
526 | * shrink pgdat->node_start_pfn and pgdat->node_spanned_pages. | |
527 | * In this case, we find second smallest valid mem_section | |
528 | * for shrinking zone. | |
529 | */ | |
530 | pfn = find_smallest_section_pfn(nid, NULL, end_pfn, | |
531 | pgdat_end_pfn); | |
532 | if (pfn) { | |
533 | pgdat->node_start_pfn = pfn; | |
534 | pgdat->node_spanned_pages = pgdat_end_pfn - pfn; | |
535 | } | |
536 | } else if (pgdat_end_pfn == end_pfn) { | |
537 | /* | |
538 | * If the section is biggest section in the pgdat, it need | |
539 | * shrink pgdat->node_spanned_pages. | |
540 | * In this case, we find second biggest valid mem_section for | |
541 | * shrinking zone. | |
542 | */ | |
543 | pfn = find_biggest_section_pfn(nid, NULL, pgdat_start_pfn, | |
544 | start_pfn); | |
545 | if (pfn) | |
546 | pgdat->node_spanned_pages = pfn - pgdat_start_pfn + 1; | |
547 | } | |
548 | ||
549 | /* | |
550 | * If the section is not biggest or smallest mem_section in the pgdat, | |
551 | * it only creates a hole in the pgdat. So in this case, we need not | |
552 | * change the pgdat. | |
553 | * But perhaps, the pgdat has only hole data. Thus it check the pgdat | |
554 | * has only hole or not. | |
555 | */ | |
556 | pfn = pgdat_start_pfn; | |
557 | for (; pfn < pgdat_end_pfn; pfn += PAGES_PER_SECTION) { | |
558 | ms = __pfn_to_section(pfn); | |
559 | ||
560 | if (unlikely(!valid_section(ms))) | |
561 | continue; | |
562 | ||
563 | if (pfn_to_nid(pfn) != nid) | |
564 | continue; | |
565 | ||
566 | /* If the section is current section, it continues the loop */ | |
567 | if (start_pfn == pfn) | |
568 | continue; | |
569 | ||
570 | /* If we find valid section, we have nothing to do */ | |
571 | return; | |
572 | } | |
573 | ||
574 | /* The pgdat has no valid section */ | |
575 | pgdat->node_start_pfn = 0; | |
576 | pgdat->node_spanned_pages = 0; | |
577 | } | |
578 | ||
579 | static void __remove_zone(struct zone *zone, unsigned long start_pfn) | |
580 | { | |
581 | struct pglist_data *pgdat = zone->zone_pgdat; | |
582 | int nr_pages = PAGES_PER_SECTION; | |
583 | int zone_type; | |
584 | unsigned long flags; | |
585 | ||
586 | zone_type = zone - pgdat->node_zones; | |
587 | ||
588 | pgdat_resize_lock(zone->zone_pgdat, &flags); | |
589 | shrink_zone_span(zone, start_pfn, start_pfn + nr_pages); | |
590 | shrink_pgdat_span(pgdat, start_pfn, start_pfn + nr_pages); | |
591 | pgdat_resize_unlock(zone->zone_pgdat, &flags); | |
592 | } | |
593 | ||
4b94ffdc DW |
594 | static int __remove_section(struct zone *zone, struct mem_section *ms, |
595 | unsigned long map_offset) | |
ea01ea93 | 596 | { |
815121d2 YI |
597 | unsigned long start_pfn; |
598 | int scn_nr; | |
ea01ea93 BP |
599 | int ret = -EINVAL; |
600 | ||
601 | if (!valid_section(ms)) | |
602 | return ret; | |
603 | ||
604 | ret = unregister_memory_section(ms); | |
605 | if (ret) | |
606 | return ret; | |
607 | ||
815121d2 YI |
608 | scn_nr = __section_nr(ms); |
609 | start_pfn = section_nr_to_pfn(scn_nr); | |
610 | __remove_zone(zone, start_pfn); | |
611 | ||
4b94ffdc | 612 | sparse_remove_one_section(zone, ms, map_offset); |
ea01ea93 BP |
613 | return 0; |
614 | } | |
615 | ||
ea01ea93 BP |
616 | /** |
617 | * __remove_pages() - remove sections of pages from a zone | |
618 | * @zone: zone from which pages need to be removed | |
619 | * @phys_start_pfn: starting pageframe (must be aligned to start of a section) | |
620 | * @nr_pages: number of pages to remove (must be multiple of section size) | |
621 | * | |
622 | * Generic helper function to remove section mappings and sysfs entries | |
623 | * for the section of the memory we are removing. Caller needs to make | |
624 | * sure that pages are marked reserved and zones are adjust properly by | |
625 | * calling offline_pages(). | |
626 | */ | |
627 | int __remove_pages(struct zone *zone, unsigned long phys_start_pfn, | |
628 | unsigned long nr_pages) | |
629 | { | |
fe74ebb1 | 630 | unsigned long i; |
4b94ffdc DW |
631 | unsigned long map_offset = 0; |
632 | int sections_to_remove, ret = 0; | |
633 | ||
634 | /* In the ZONE_DEVICE case device driver owns the memory region */ | |
635 | if (is_dev_zone(zone)) { | |
636 | struct page *page = pfn_to_page(phys_start_pfn); | |
637 | struct vmem_altmap *altmap; | |
638 | ||
639 | altmap = to_vmem_altmap((unsigned long) page); | |
640 | if (altmap) | |
641 | map_offset = vmem_altmap_offset(altmap); | |
642 | } else { | |
643 | resource_size_t start, size; | |
644 | ||
645 | start = phys_start_pfn << PAGE_SHIFT; | |
646 | size = nr_pages * PAGE_SIZE; | |
647 | ||
648 | ret = release_mem_region_adjustable(&iomem_resource, start, | |
649 | size); | |
650 | if (ret) { | |
651 | resource_size_t endres = start + size - 1; | |
652 | ||
653 | pr_warn("Unable to release resource <%pa-%pa> (%d)\n", | |
654 | &start, &endres, ret); | |
655 | } | |
656 | } | |
ea01ea93 | 657 | |
7cf91a98 JK |
658 | clear_zone_contiguous(zone); |
659 | ||
ea01ea93 BP |
660 | /* |
661 | * We can only remove entire sections | |
662 | */ | |
663 | BUG_ON(phys_start_pfn & ~PAGE_SECTION_MASK); | |
664 | BUG_ON(nr_pages % PAGES_PER_SECTION); | |
665 | ||
ea01ea93 BP |
666 | sections_to_remove = nr_pages / PAGES_PER_SECTION; |
667 | for (i = 0; i < sections_to_remove; i++) { | |
668 | unsigned long pfn = phys_start_pfn + i*PAGES_PER_SECTION; | |
4b94ffdc DW |
669 | |
670 | ret = __remove_section(zone, __pfn_to_section(pfn), map_offset); | |
671 | map_offset = 0; | |
ea01ea93 BP |
672 | if (ret) |
673 | break; | |
674 | } | |
7cf91a98 JK |
675 | |
676 | set_zone_contiguous(zone); | |
677 | ||
ea01ea93 BP |
678 | return ret; |
679 | } | |
4edd7cef | 680 | #endif /* CONFIG_MEMORY_HOTREMOVE */ |
ea01ea93 | 681 | |
9d0ad8ca DK |
682 | int set_online_page_callback(online_page_callback_t callback) |
683 | { | |
684 | int rc = -EINVAL; | |
685 | ||
bfc8c901 VD |
686 | get_online_mems(); |
687 | mutex_lock(&online_page_callback_lock); | |
9d0ad8ca DK |
688 | |
689 | if (online_page_callback == generic_online_page) { | |
690 | online_page_callback = callback; | |
691 | rc = 0; | |
692 | } | |
693 | ||
bfc8c901 VD |
694 | mutex_unlock(&online_page_callback_lock); |
695 | put_online_mems(); | |
9d0ad8ca DK |
696 | |
697 | return rc; | |
698 | } | |
699 | EXPORT_SYMBOL_GPL(set_online_page_callback); | |
700 | ||
701 | int restore_online_page_callback(online_page_callback_t callback) | |
702 | { | |
703 | int rc = -EINVAL; | |
704 | ||
bfc8c901 VD |
705 | get_online_mems(); |
706 | mutex_lock(&online_page_callback_lock); | |
9d0ad8ca DK |
707 | |
708 | if (online_page_callback == callback) { | |
709 | online_page_callback = generic_online_page; | |
710 | rc = 0; | |
711 | } | |
712 | ||
bfc8c901 VD |
713 | mutex_unlock(&online_page_callback_lock); |
714 | put_online_mems(); | |
9d0ad8ca DK |
715 | |
716 | return rc; | |
717 | } | |
718 | EXPORT_SYMBOL_GPL(restore_online_page_callback); | |
719 | ||
720 | void __online_page_set_limits(struct page *page) | |
180c06ef | 721 | { |
9d0ad8ca DK |
722 | } |
723 | EXPORT_SYMBOL_GPL(__online_page_set_limits); | |
724 | ||
725 | void __online_page_increment_counters(struct page *page) | |
726 | { | |
3dcc0571 | 727 | adjust_managed_page_count(page, 1); |
9d0ad8ca DK |
728 | } |
729 | EXPORT_SYMBOL_GPL(__online_page_increment_counters); | |
180c06ef | 730 | |
9d0ad8ca DK |
731 | void __online_page_free(struct page *page) |
732 | { | |
3dcc0571 | 733 | __free_reserved_page(page); |
180c06ef | 734 | } |
9d0ad8ca DK |
735 | EXPORT_SYMBOL_GPL(__online_page_free); |
736 | ||
737 | static void generic_online_page(struct page *page) | |
738 | { | |
739 | __online_page_set_limits(page); | |
740 | __online_page_increment_counters(page); | |
741 | __online_page_free(page); | |
742 | } | |
180c06ef | 743 | |
75884fb1 KH |
744 | static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages, |
745 | void *arg) | |
3947be19 DH |
746 | { |
747 | unsigned long i; | |
75884fb1 KH |
748 | unsigned long onlined_pages = *(unsigned long *)arg; |
749 | struct page *page; | |
2d070eab | 750 | |
75884fb1 KH |
751 | if (PageReserved(pfn_to_page(start_pfn))) |
752 | for (i = 0; i < nr_pages; i++) { | |
753 | page = pfn_to_page(start_pfn + i); | |
9d0ad8ca | 754 | (*online_page_callback)(page); |
75884fb1 KH |
755 | onlined_pages++; |
756 | } | |
2d070eab MH |
757 | |
758 | online_mem_sections(start_pfn, start_pfn + nr_pages); | |
759 | ||
75884fb1 KH |
760 | *(unsigned long *)arg = onlined_pages; |
761 | return 0; | |
762 | } | |
763 | ||
d9713679 LJ |
764 | /* check which state of node_states will be changed when online memory */ |
765 | static void node_states_check_changes_online(unsigned long nr_pages, | |
766 | struct zone *zone, struct memory_notify *arg) | |
767 | { | |
768 | int nid = zone_to_nid(zone); | |
769 | enum zone_type zone_last = ZONE_NORMAL; | |
770 | ||
771 | /* | |
6715ddf9 LJ |
772 | * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY] |
773 | * contains nodes which have zones of 0...ZONE_NORMAL, | |
774 | * set zone_last to ZONE_NORMAL. | |
d9713679 | 775 | * |
6715ddf9 LJ |
776 | * If we don't have HIGHMEM nor movable node, |
777 | * node_states[N_NORMAL_MEMORY] contains nodes which have zones of | |
778 | * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE. | |
d9713679 | 779 | */ |
6715ddf9 | 780 | if (N_MEMORY == N_NORMAL_MEMORY) |
d9713679 LJ |
781 | zone_last = ZONE_MOVABLE; |
782 | ||
783 | /* | |
784 | * if the memory to be online is in a zone of 0...zone_last, and | |
785 | * the zones of 0...zone_last don't have memory before online, we will | |
786 | * need to set the node to node_states[N_NORMAL_MEMORY] after | |
787 | * the memory is online. | |
788 | */ | |
789 | if (zone_idx(zone) <= zone_last && !node_state(nid, N_NORMAL_MEMORY)) | |
790 | arg->status_change_nid_normal = nid; | |
791 | else | |
792 | arg->status_change_nid_normal = -1; | |
793 | ||
6715ddf9 LJ |
794 | #ifdef CONFIG_HIGHMEM |
795 | /* | |
796 | * If we have movable node, node_states[N_HIGH_MEMORY] | |
797 | * contains nodes which have zones of 0...ZONE_HIGHMEM, | |
798 | * set zone_last to ZONE_HIGHMEM. | |
799 | * | |
800 | * If we don't have movable node, node_states[N_NORMAL_MEMORY] | |
801 | * contains nodes which have zones of 0...ZONE_MOVABLE, | |
802 | * set zone_last to ZONE_MOVABLE. | |
803 | */ | |
804 | zone_last = ZONE_HIGHMEM; | |
805 | if (N_MEMORY == N_HIGH_MEMORY) | |
806 | zone_last = ZONE_MOVABLE; | |
807 | ||
808 | if (zone_idx(zone) <= zone_last && !node_state(nid, N_HIGH_MEMORY)) | |
809 | arg->status_change_nid_high = nid; | |
810 | else | |
811 | arg->status_change_nid_high = -1; | |
812 | #else | |
813 | arg->status_change_nid_high = arg->status_change_nid_normal; | |
814 | #endif | |
815 | ||
d9713679 LJ |
816 | /* |
817 | * if the node don't have memory befor online, we will need to | |
6715ddf9 | 818 | * set the node to node_states[N_MEMORY] after the memory |
d9713679 LJ |
819 | * is online. |
820 | */ | |
6715ddf9 | 821 | if (!node_state(nid, N_MEMORY)) |
d9713679 LJ |
822 | arg->status_change_nid = nid; |
823 | else | |
824 | arg->status_change_nid = -1; | |
825 | } | |
826 | ||
827 | static void node_states_set_node(int node, struct memory_notify *arg) | |
828 | { | |
829 | if (arg->status_change_nid_normal >= 0) | |
830 | node_set_state(node, N_NORMAL_MEMORY); | |
831 | ||
6715ddf9 LJ |
832 | if (arg->status_change_nid_high >= 0) |
833 | node_set_state(node, N_HIGH_MEMORY); | |
834 | ||
835 | node_set_state(node, N_MEMORY); | |
d9713679 LJ |
836 | } |
837 | ||
f1dd2cd1 | 838 | bool allow_online_pfn_range(int nid, unsigned long pfn, unsigned long nr_pages, int online_type) |
df429ac0 | 839 | { |
f1dd2cd1 MH |
840 | struct pglist_data *pgdat = NODE_DATA(nid); |
841 | struct zone *movable_zone = &pgdat->node_zones[ZONE_MOVABLE]; | |
c246a213 | 842 | struct zone *default_zone = default_zone_for_pfn(nid, pfn, nr_pages); |
df429ac0 | 843 | |
f1dd2cd1 MH |
844 | /* |
845 | * TODO there shouldn't be any inherent reason to have ZONE_NORMAL | |
846 | * physically before ZONE_MOVABLE. All we need is they do not | |
847 | * overlap. Historically we didn't allow ZONE_NORMAL after ZONE_MOVABLE | |
848 | * though so let's stick with it for simplicity for now. | |
849 | * TODO make sure we do not overlap with ZONE_DEVICE | |
850 | */ | |
851 | if (online_type == MMOP_ONLINE_KERNEL) { | |
852 | if (zone_is_empty(movable_zone)) | |
853 | return true; | |
854 | return movable_zone->zone_start_pfn >= pfn + nr_pages; | |
855 | } else if (online_type == MMOP_ONLINE_MOVABLE) { | |
c246a213 | 856 | return zone_end_pfn(default_zone) <= pfn; |
f1dd2cd1 | 857 | } |
8a1f780e | 858 | |
f1dd2cd1 MH |
859 | /* MMOP_ONLINE_KEEP will always succeed and inherits the current zone */ |
860 | return online_type == MMOP_ONLINE_KEEP; | |
861 | } | |
df429ac0 | 862 | |
f1dd2cd1 MH |
863 | static void __meminit resize_zone_range(struct zone *zone, unsigned long start_pfn, |
864 | unsigned long nr_pages) | |
865 | { | |
866 | unsigned long old_end_pfn = zone_end_pfn(zone); | |
867 | ||
868 | if (zone_is_empty(zone) || start_pfn < zone->zone_start_pfn) | |
869 | zone->zone_start_pfn = start_pfn; | |
870 | ||
871 | zone->spanned_pages = max(start_pfn + nr_pages, old_end_pfn) - zone->zone_start_pfn; | |
872 | } | |
873 | ||
874 | static void __meminit resize_pgdat_range(struct pglist_data *pgdat, unsigned long start_pfn, | |
875 | unsigned long nr_pages) | |
876 | { | |
877 | unsigned long old_end_pfn = pgdat_end_pfn(pgdat); | |
878 | ||
879 | if (!pgdat->node_spanned_pages || start_pfn < pgdat->node_start_pfn) | |
880 | pgdat->node_start_pfn = start_pfn; | |
881 | ||
882 | pgdat->node_spanned_pages = max(start_pfn + nr_pages, old_end_pfn) - pgdat->node_start_pfn; | |
883 | } | |
884 | ||
cdf72f25 | 885 | void __ref move_pfn_range_to_zone(struct zone *zone, |
f1dd2cd1 MH |
886 | unsigned long start_pfn, unsigned long nr_pages) |
887 | { | |
888 | struct pglist_data *pgdat = zone->zone_pgdat; | |
889 | int nid = pgdat->node_id; | |
890 | unsigned long flags; | |
df429ac0 | 891 | |
f1dd2cd1 MH |
892 | if (zone_is_empty(zone)) |
893 | init_currently_empty_zone(zone, start_pfn, nr_pages); | |
df429ac0 | 894 | |
f1dd2cd1 MH |
895 | clear_zone_contiguous(zone); |
896 | ||
897 | /* TODO Huh pgdat is irqsave while zone is not. It used to be like that before */ | |
898 | pgdat_resize_lock(pgdat, &flags); | |
899 | zone_span_writelock(zone); | |
900 | resize_zone_range(zone, start_pfn, nr_pages); | |
901 | zone_span_writeunlock(zone); | |
902 | resize_pgdat_range(pgdat, start_pfn, nr_pages); | |
903 | pgdat_resize_unlock(pgdat, &flags); | |
904 | ||
905 | /* | |
906 | * TODO now we have a visible range of pages which are not associated | |
907 | * with their zone properly. Not nice but set_pfnblock_flags_mask | |
908 | * expects the zone spans the pfn range. All the pages in the range | |
909 | * are reserved so nobody should be touching them so we should be safe | |
910 | */ | |
911 | memmap_init_zone(nr_pages, nid, zone_idx(zone), start_pfn, MEMMAP_HOTPLUG); | |
912 | ||
913 | set_zone_contiguous(zone); | |
914 | } | |
915 | ||
c246a213 MH |
916 | /* |
917 | * Returns a default kernel memory zone for the given pfn range. | |
918 | * If no kernel zone covers this pfn range it will automatically go | |
919 | * to the ZONE_NORMAL. | |
920 | */ | |
921 | struct zone *default_zone_for_pfn(int nid, unsigned long start_pfn, | |
922 | unsigned long nr_pages) | |
923 | { | |
924 | struct pglist_data *pgdat = NODE_DATA(nid); | |
925 | int zid; | |
926 | ||
927 | for (zid = 0; zid <= ZONE_NORMAL; zid++) { | |
928 | struct zone *zone = &pgdat->node_zones[zid]; | |
929 | ||
930 | if (zone_intersects(zone, start_pfn, nr_pages)) | |
931 | return zone; | |
932 | } | |
933 | ||
934 | return &pgdat->node_zones[ZONE_NORMAL]; | |
935 | } | |
936 | ||
f1dd2cd1 MH |
937 | /* |
938 | * Associates the given pfn range with the given node and the zone appropriate | |
939 | * for the given online type. | |
940 | */ | |
941 | static struct zone * __meminit move_pfn_range(int online_type, int nid, | |
942 | unsigned long start_pfn, unsigned long nr_pages) | |
943 | { | |
944 | struct pglist_data *pgdat = NODE_DATA(nid); | |
c246a213 | 945 | struct zone *zone = default_zone_for_pfn(nid, start_pfn, nr_pages); |
f1dd2cd1 MH |
946 | |
947 | if (online_type == MMOP_ONLINE_KEEP) { | |
948 | struct zone *movable_zone = &pgdat->node_zones[ZONE_MOVABLE]; | |
949 | /* | |
a69578a1 MH |
950 | * MMOP_ONLINE_KEEP defaults to MMOP_ONLINE_KERNEL but use |
951 | * movable zone if that is not possible (e.g. we are within | |
952 | * or past the existing movable zone) | |
f1dd2cd1 | 953 | */ |
a69578a1 MH |
954 | if (!allow_online_pfn_range(nid, start_pfn, nr_pages, |
955 | MMOP_ONLINE_KERNEL)) | |
f1dd2cd1 MH |
956 | zone = movable_zone; |
957 | } else if (online_type == MMOP_ONLINE_MOVABLE) { | |
958 | zone = &pgdat->node_zones[ZONE_MOVABLE]; | |
df429ac0 RA |
959 | } |
960 | ||
f1dd2cd1 MH |
961 | move_pfn_range_to_zone(zone, start_pfn, nr_pages); |
962 | return zone; | |
df429ac0 | 963 | } |
75884fb1 | 964 | |
30467e0b | 965 | /* Must be protected by mem_hotplug_begin() */ |
511c2aba | 966 | int __ref online_pages(unsigned long pfn, unsigned long nr_pages, int online_type) |
75884fb1 | 967 | { |
aa47228a | 968 | unsigned long flags; |
3947be19 DH |
969 | unsigned long onlined_pages = 0; |
970 | struct zone *zone; | |
6811378e | 971 | int need_zonelists_rebuild = 0; |
7b78d335 YG |
972 | int nid; |
973 | int ret; | |
974 | struct memory_notify arg; | |
975 | ||
f1dd2cd1 MH |
976 | nid = pfn_to_nid(pfn); |
977 | if (!allow_online_pfn_range(nid, pfn, nr_pages, online_type)) | |
30467e0b | 978 | return -EINVAL; |
74d42d8f | 979 | |
f1dd2cd1 MH |
980 | /* associate pfn range with the zone */ |
981 | zone = move_pfn_range(online_type, nid, pfn, nr_pages); | |
982 | ||
7b78d335 YG |
983 | arg.start_pfn = pfn; |
984 | arg.nr_pages = nr_pages; | |
d9713679 | 985 | node_states_check_changes_online(nr_pages, zone, &arg); |
7b78d335 | 986 | |
7b78d335 YG |
987 | ret = memory_notify(MEM_GOING_ONLINE, &arg); |
988 | ret = notifier_to_errno(ret); | |
e33e33b4 CY |
989 | if (ret) |
990 | goto failed_addition; | |
991 | ||
6811378e YG |
992 | /* |
993 | * If this zone is not populated, then it is not in zonelist. | |
994 | * This means the page allocator ignores this zone. | |
995 | * So, zonelist must be updated after online. | |
996 | */ | |
4eaf3f64 | 997 | mutex_lock(&zonelists_mutex); |
6dcd73d7 | 998 | if (!populated_zone(zone)) { |
6811378e | 999 | need_zonelists_rebuild = 1; |
6dcd73d7 WC |
1000 | build_all_zonelists(NULL, zone); |
1001 | } | |
6811378e | 1002 | |
908eedc6 | 1003 | ret = walk_system_ram_range(pfn, nr_pages, &onlined_pages, |
75884fb1 | 1004 | online_pages_range); |
fd8a4221 | 1005 | if (ret) { |
6dcd73d7 WC |
1006 | if (need_zonelists_rebuild) |
1007 | zone_pcp_reset(zone); | |
4eaf3f64 | 1008 | mutex_unlock(&zonelists_mutex); |
e33e33b4 | 1009 | goto failed_addition; |
fd8a4221 GL |
1010 | } |
1011 | ||
3947be19 | 1012 | zone->present_pages += onlined_pages; |
aa47228a CS |
1013 | |
1014 | pgdat_resize_lock(zone->zone_pgdat, &flags); | |
f2937be5 | 1015 | zone->zone_pgdat->node_present_pages += onlined_pages; |
aa47228a CS |
1016 | pgdat_resize_unlock(zone->zone_pgdat, &flags); |
1017 | ||
08dff7b7 | 1018 | if (onlined_pages) { |
e888ca35 | 1019 | node_states_set_node(nid, &arg); |
08dff7b7 | 1020 | if (need_zonelists_rebuild) |
6dcd73d7 | 1021 | build_all_zonelists(NULL, NULL); |
08dff7b7 JL |
1022 | else |
1023 | zone_pcp_update(zone); | |
1024 | } | |
3947be19 | 1025 | |
4eaf3f64 | 1026 | mutex_unlock(&zonelists_mutex); |
1b79acc9 KM |
1027 | |
1028 | init_per_zone_wmark_min(); | |
1029 | ||
698b1b30 | 1030 | if (onlined_pages) { |
e888ca35 | 1031 | kswapd_run(nid); |
698b1b30 VB |
1032 | kcompactd_run(nid); |
1033 | } | |
61b13993 | 1034 | |
1f522509 | 1035 | vm_total_pages = nr_free_pagecache_pages(); |
2f7f24ec | 1036 | |
2d1d43f6 | 1037 | writeback_set_ratelimit(); |
7b78d335 YG |
1038 | |
1039 | if (onlined_pages) | |
1040 | memory_notify(MEM_ONLINE, &arg); | |
30467e0b | 1041 | return 0; |
e33e33b4 CY |
1042 | |
1043 | failed_addition: | |
1044 | pr_debug("online_pages [mem %#010llx-%#010llx] failed\n", | |
1045 | (unsigned long long) pfn << PAGE_SHIFT, | |
1046 | (((unsigned long long) pfn + nr_pages) << PAGE_SHIFT) - 1); | |
1047 | memory_notify(MEM_CANCEL_ONLINE, &arg); | |
1048 | return ret; | |
3947be19 | 1049 | } |
53947027 | 1050 | #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */ |
bc02af93 | 1051 | |
0bd85420 TC |
1052 | static void reset_node_present_pages(pg_data_t *pgdat) |
1053 | { | |
1054 | struct zone *z; | |
1055 | ||
1056 | for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++) | |
1057 | z->present_pages = 0; | |
1058 | ||
1059 | pgdat->node_present_pages = 0; | |
1060 | } | |
1061 | ||
e1319331 HS |
1062 | /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */ |
1063 | static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start) | |
9af3c2de YG |
1064 | { |
1065 | struct pglist_data *pgdat; | |
1066 | unsigned long zones_size[MAX_NR_ZONES] = {0}; | |
1067 | unsigned long zholes_size[MAX_NR_ZONES] = {0}; | |
c8e861a5 | 1068 | unsigned long start_pfn = PFN_DOWN(start); |
9af3c2de | 1069 | |
a1e565aa TC |
1070 | pgdat = NODE_DATA(nid); |
1071 | if (!pgdat) { | |
1072 | pgdat = arch_alloc_nodedata(nid); | |
1073 | if (!pgdat) | |
1074 | return NULL; | |
9af3c2de | 1075 | |
a1e565aa | 1076 | arch_refresh_nodedata(nid, pgdat); |
b0dc3a34 | 1077 | } else { |
e716f2eb MG |
1078 | /* |
1079 | * Reset the nr_zones, order and classzone_idx before reuse. | |
1080 | * Note that kswapd will init kswapd_classzone_idx properly | |
1081 | * when it starts in the near future. | |
1082 | */ | |
b0dc3a34 | 1083 | pgdat->nr_zones = 0; |
38087d9b MG |
1084 | pgdat->kswapd_order = 0; |
1085 | pgdat->kswapd_classzone_idx = 0; | |
a1e565aa | 1086 | } |
9af3c2de YG |
1087 | |
1088 | /* we can use NODE_DATA(nid) from here */ | |
1089 | ||
1090 | /* init node's zones as empty zones, we don't have any present pages.*/ | |
9109fb7b | 1091 | free_area_init_node(nid, zones_size, start_pfn, zholes_size); |
5830169f | 1092 | pgdat->per_cpu_nodestats = alloc_percpu(struct per_cpu_nodestat); |
9af3c2de | 1093 | |
959ecc48 KH |
1094 | /* |
1095 | * The node we allocated has no zone fallback lists. For avoiding | |
1096 | * to access not-initialized zonelist, build here. | |
1097 | */ | |
f957db4f | 1098 | mutex_lock(&zonelists_mutex); |
9adb62a5 | 1099 | build_all_zonelists(pgdat, NULL); |
f957db4f | 1100 | mutex_unlock(&zonelists_mutex); |
959ecc48 | 1101 | |
f784a3f1 TC |
1102 | /* |
1103 | * zone->managed_pages is set to an approximate value in | |
1104 | * free_area_init_core(), which will cause | |
1105 | * /sys/device/system/node/nodeX/meminfo has wrong data. | |
1106 | * So reset it to 0 before any memory is onlined. | |
1107 | */ | |
1108 | reset_node_managed_pages(pgdat); | |
1109 | ||
0bd85420 TC |
1110 | /* |
1111 | * When memory is hot-added, all the memory is in offline state. So | |
1112 | * clear all zones' present_pages because they will be updated in | |
1113 | * online_pages() and offline_pages(). | |
1114 | */ | |
1115 | reset_node_present_pages(pgdat); | |
1116 | ||
9af3c2de YG |
1117 | return pgdat; |
1118 | } | |
1119 | ||
1120 | static void rollback_node_hotadd(int nid, pg_data_t *pgdat) | |
1121 | { | |
1122 | arch_refresh_nodedata(nid, NULL); | |
5830169f | 1123 | free_percpu(pgdat->per_cpu_nodestats); |
9af3c2de YG |
1124 | arch_free_nodedata(pgdat); |
1125 | return; | |
1126 | } | |
1127 | ||
0a547039 | 1128 | |
01b0f197 TK |
1129 | /** |
1130 | * try_online_node - online a node if offlined | |
1131 | * | |
cf23422b | 1132 | * called by cpu_up() to online a node without onlined memory. |
1133 | */ | |
01b0f197 | 1134 | int try_online_node(int nid) |
cf23422b | 1135 | { |
1136 | pg_data_t *pgdat; | |
1137 | int ret; | |
1138 | ||
01b0f197 TK |
1139 | if (node_online(nid)) |
1140 | return 0; | |
1141 | ||
bfc8c901 | 1142 | mem_hotplug_begin(); |
cf23422b | 1143 | pgdat = hotadd_new_pgdat(nid, 0); |
7553e8f2 | 1144 | if (!pgdat) { |
01b0f197 | 1145 | pr_err("Cannot online node %d due to NULL pgdat\n", nid); |
cf23422b | 1146 | ret = -ENOMEM; |
1147 | goto out; | |
1148 | } | |
1149 | node_set_online(nid); | |
1150 | ret = register_one_node(nid); | |
1151 | BUG_ON(ret); | |
1152 | ||
01b0f197 TK |
1153 | if (pgdat->node_zonelists->_zonerefs->zone == NULL) { |
1154 | mutex_lock(&zonelists_mutex); | |
1155 | build_all_zonelists(NULL, NULL); | |
1156 | mutex_unlock(&zonelists_mutex); | |
1157 | } | |
1158 | ||
cf23422b | 1159 | out: |
bfc8c901 | 1160 | mem_hotplug_done(); |
cf23422b | 1161 | return ret; |
1162 | } | |
1163 | ||
27356f54 TK |
1164 | static int check_hotplug_memory_range(u64 start, u64 size) |
1165 | { | |
c8e861a5 | 1166 | u64 start_pfn = PFN_DOWN(start); |
27356f54 TK |
1167 | u64 nr_pages = size >> PAGE_SHIFT; |
1168 | ||
1169 | /* Memory range must be aligned with section */ | |
1170 | if ((start_pfn & ~PAGE_SECTION_MASK) || | |
1171 | (nr_pages % PAGES_PER_SECTION) || (!nr_pages)) { | |
1172 | pr_err("Section-unaligned hotplug range: start 0x%llx, size 0x%llx\n", | |
1173 | (unsigned long long)start, | |
1174 | (unsigned long long)size); | |
1175 | return -EINVAL; | |
1176 | } | |
1177 | ||
1178 | return 0; | |
1179 | } | |
1180 | ||
31bc3858 VK |
1181 | static int online_memory_block(struct memory_block *mem, void *arg) |
1182 | { | |
dc18d706 | 1183 | return device_online(&mem->dev); |
31bc3858 VK |
1184 | } |
1185 | ||
31168481 | 1186 | /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */ |
31bc3858 | 1187 | int __ref add_memory_resource(int nid, struct resource *res, bool online) |
bc02af93 | 1188 | { |
62cedb9f | 1189 | u64 start, size; |
9af3c2de | 1190 | pg_data_t *pgdat = NULL; |
a1e565aa TC |
1191 | bool new_pgdat; |
1192 | bool new_node; | |
bc02af93 YG |
1193 | int ret; |
1194 | ||
62cedb9f DV |
1195 | start = res->start; |
1196 | size = resource_size(res); | |
1197 | ||
27356f54 TK |
1198 | ret = check_hotplug_memory_range(start, size); |
1199 | if (ret) | |
1200 | return ret; | |
1201 | ||
a1e565aa TC |
1202 | { /* Stupid hack to suppress address-never-null warning */ |
1203 | void *p = NODE_DATA(nid); | |
1204 | new_pgdat = !p; | |
1205 | } | |
ac13c462 | 1206 | |
bfc8c901 | 1207 | mem_hotplug_begin(); |
ac13c462 | 1208 | |
7f36e3e5 TC |
1209 | /* |
1210 | * Add new range to memblock so that when hotadd_new_pgdat() is called | |
1211 | * to allocate new pgdat, get_pfn_range_for_nid() will be able to find | |
1212 | * this new range and calculate total pages correctly. The range will | |
1213 | * be removed at hot-remove time. | |
1214 | */ | |
1215 | memblock_add_node(start, size, nid); | |
1216 | ||
a1e565aa TC |
1217 | new_node = !node_online(nid); |
1218 | if (new_node) { | |
9af3c2de | 1219 | pgdat = hotadd_new_pgdat(nid, start); |
6ad696d2 | 1220 | ret = -ENOMEM; |
9af3c2de | 1221 | if (!pgdat) |
41b9e2d7 | 1222 | goto error; |
9af3c2de YG |
1223 | } |
1224 | ||
bc02af93 | 1225 | /* call arch's memory hotadd */ |
3d79a728 | 1226 | ret = arch_add_memory(nid, start, size, true); |
bc02af93 | 1227 | |
9af3c2de YG |
1228 | if (ret < 0) |
1229 | goto error; | |
1230 | ||
0fc44159 | 1231 | /* we online node here. we can't roll back from here. */ |
9af3c2de YG |
1232 | node_set_online(nid); |
1233 | ||
a1e565aa | 1234 | if (new_node) { |
9037a993 MH |
1235 | unsigned long start_pfn = start >> PAGE_SHIFT; |
1236 | unsigned long nr_pages = size >> PAGE_SHIFT; | |
1237 | ||
1238 | ret = __register_one_node(nid); | |
1239 | if (ret) | |
1240 | goto register_fail; | |
1241 | ||
1242 | /* | |
1243 | * link memory sections under this node. This is already | |
1244 | * done when creatig memory section in register_new_memory | |
1245 | * but that depends to have the node registered so offline | |
1246 | * nodes have to go through register_node. | |
1247 | * TODO clean up this mess. | |
1248 | */ | |
1249 | ret = link_mem_sections(nid, start_pfn, nr_pages); | |
1250 | register_fail: | |
0fc44159 YG |
1251 | /* |
1252 | * If sysfs file of new node can't create, cpu on the node | |
1253 | * can't be hot-added. There is no rollback way now. | |
1254 | * So, check by BUG_ON() to catch it reluctantly.. | |
1255 | */ | |
1256 | BUG_ON(ret); | |
1257 | } | |
1258 | ||
d96ae530 AM |
1259 | /* create new memmap entry */ |
1260 | firmware_map_add_hotplug(start, start + size, "System RAM"); | |
1261 | ||
31bc3858 VK |
1262 | /* online pages if requested */ |
1263 | if (online) | |
1264 | walk_memory_range(PFN_DOWN(start), PFN_UP(start + size - 1), | |
1265 | NULL, online_memory_block); | |
1266 | ||
6ad696d2 AK |
1267 | goto out; |
1268 | ||
9af3c2de YG |
1269 | error: |
1270 | /* rollback pgdat allocation and others */ | |
1271 | if (new_pgdat) | |
1272 | rollback_node_hotadd(nid, pgdat); | |
7f36e3e5 | 1273 | memblock_remove(start, size); |
9af3c2de | 1274 | |
6ad696d2 | 1275 | out: |
bfc8c901 | 1276 | mem_hotplug_done(); |
bc02af93 YG |
1277 | return ret; |
1278 | } | |
62cedb9f DV |
1279 | EXPORT_SYMBOL_GPL(add_memory_resource); |
1280 | ||
1281 | int __ref add_memory(int nid, u64 start, u64 size) | |
1282 | { | |
1283 | struct resource *res; | |
1284 | int ret; | |
1285 | ||
1286 | res = register_memory_resource(start, size); | |
6f754ba4 VK |
1287 | if (IS_ERR(res)) |
1288 | return PTR_ERR(res); | |
62cedb9f | 1289 | |
31bc3858 | 1290 | ret = add_memory_resource(nid, res, memhp_auto_online); |
62cedb9f DV |
1291 | if (ret < 0) |
1292 | release_memory_resource(res); | |
1293 | return ret; | |
1294 | } | |
bc02af93 | 1295 | EXPORT_SYMBOL_GPL(add_memory); |
0c0e6195 KH |
1296 | |
1297 | #ifdef CONFIG_MEMORY_HOTREMOVE | |
5c755e9f BP |
1298 | /* |
1299 | * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy | |
1300 | * set and the size of the free page is given by page_order(). Using this, | |
1301 | * the function determines if the pageblock contains only free pages. | |
1302 | * Due to buddy contraints, a free page at least the size of a pageblock will | |
1303 | * be located at the start of the pageblock | |
1304 | */ | |
1305 | static inline int pageblock_free(struct page *page) | |
1306 | { | |
1307 | return PageBuddy(page) && page_order(page) >= pageblock_order; | |
1308 | } | |
1309 | ||
1310 | /* Return the start of the next active pageblock after a given page */ | |
1311 | static struct page *next_active_pageblock(struct page *page) | |
1312 | { | |
5c755e9f BP |
1313 | /* Ensure the starting page is pageblock-aligned */ |
1314 | BUG_ON(page_to_pfn(page) & (pageblock_nr_pages - 1)); | |
1315 | ||
5c755e9f | 1316 | /* If the entire pageblock is free, move to the end of free page */ |
0dcc48c1 KH |
1317 | if (pageblock_free(page)) { |
1318 | int order; | |
1319 | /* be careful. we don't have locks, page_order can be changed.*/ | |
1320 | order = page_order(page); | |
1321 | if ((order < MAX_ORDER) && (order >= pageblock_order)) | |
1322 | return page + (1 << order); | |
1323 | } | |
5c755e9f | 1324 | |
0dcc48c1 | 1325 | return page + pageblock_nr_pages; |
5c755e9f BP |
1326 | } |
1327 | ||
1328 | /* Checks if this range of memory is likely to be hot-removable. */ | |
c98940f6 | 1329 | bool is_mem_section_removable(unsigned long start_pfn, unsigned long nr_pages) |
5c755e9f | 1330 | { |
5c755e9f BP |
1331 | struct page *page = pfn_to_page(start_pfn); |
1332 | struct page *end_page = page + nr_pages; | |
1333 | ||
1334 | /* Check the starting page of each pageblock within the range */ | |
1335 | for (; page < end_page; page = next_active_pageblock(page)) { | |
49ac8255 | 1336 | if (!is_pageblock_removable_nolock(page)) |
c98940f6 | 1337 | return false; |
49ac8255 | 1338 | cond_resched(); |
5c755e9f BP |
1339 | } |
1340 | ||
1341 | /* All pageblocks in the memory block are likely to be hot-removable */ | |
c98940f6 | 1342 | return true; |
5c755e9f BP |
1343 | } |
1344 | ||
0c0e6195 | 1345 | /* |
deb88a2a | 1346 | * Confirm all pages in a range [start, end) belong to the same zone. |
a96dfddb | 1347 | * When true, return its valid [start, end). |
0c0e6195 | 1348 | */ |
a96dfddb TK |
1349 | int test_pages_in_a_zone(unsigned long start_pfn, unsigned long end_pfn, |
1350 | unsigned long *valid_start, unsigned long *valid_end) | |
0c0e6195 | 1351 | { |
5f0f2887 | 1352 | unsigned long pfn, sec_end_pfn; |
a96dfddb | 1353 | unsigned long start, end; |
0c0e6195 KH |
1354 | struct zone *zone = NULL; |
1355 | struct page *page; | |
1356 | int i; | |
deb88a2a | 1357 | for (pfn = start_pfn, sec_end_pfn = SECTION_ALIGN_UP(start_pfn + 1); |
0c0e6195 | 1358 | pfn < end_pfn; |
deb88a2a | 1359 | pfn = sec_end_pfn, sec_end_pfn += PAGES_PER_SECTION) { |
5f0f2887 AB |
1360 | /* Make sure the memory section is present first */ |
1361 | if (!present_section_nr(pfn_to_section_nr(pfn))) | |
0c0e6195 | 1362 | continue; |
5f0f2887 AB |
1363 | for (; pfn < sec_end_pfn && pfn < end_pfn; |
1364 | pfn += MAX_ORDER_NR_PAGES) { | |
1365 | i = 0; | |
1366 | /* This is just a CONFIG_HOLES_IN_ZONE check.*/ | |
1367 | while ((i < MAX_ORDER_NR_PAGES) && | |
1368 | !pfn_valid_within(pfn + i)) | |
1369 | i++; | |
d6d8c8a4 | 1370 | if (i == MAX_ORDER_NR_PAGES || pfn + i >= end_pfn) |
5f0f2887 AB |
1371 | continue; |
1372 | page = pfn_to_page(pfn + i); | |
1373 | if (zone && page_zone(page) != zone) | |
1374 | return 0; | |
a96dfddb TK |
1375 | if (!zone) |
1376 | start = pfn + i; | |
5f0f2887 | 1377 | zone = page_zone(page); |
a96dfddb | 1378 | end = pfn + MAX_ORDER_NR_PAGES; |
5f0f2887 | 1379 | } |
0c0e6195 | 1380 | } |
deb88a2a | 1381 | |
a96dfddb TK |
1382 | if (zone) { |
1383 | *valid_start = start; | |
d6d8c8a4 | 1384 | *valid_end = min(end, end_pfn); |
deb88a2a | 1385 | return 1; |
a96dfddb | 1386 | } else { |
deb88a2a | 1387 | return 0; |
a96dfddb | 1388 | } |
0c0e6195 KH |
1389 | } |
1390 | ||
1391 | /* | |
0efadf48 YX |
1392 | * Scan pfn range [start,end) to find movable/migratable pages (LRU pages, |
1393 | * non-lru movable pages and hugepages). We scan pfn because it's much | |
1394 | * easier than scanning over linked list. This function returns the pfn | |
1395 | * of the first found movable page if it's found, otherwise 0. | |
0c0e6195 | 1396 | */ |
c8721bbb | 1397 | static unsigned long scan_movable_pages(unsigned long start, unsigned long end) |
0c0e6195 KH |
1398 | { |
1399 | unsigned long pfn; | |
1400 | struct page *page; | |
1401 | for (pfn = start; pfn < end; pfn++) { | |
1402 | if (pfn_valid(pfn)) { | |
1403 | page = pfn_to_page(pfn); | |
1404 | if (PageLRU(page)) | |
1405 | return pfn; | |
0efadf48 YX |
1406 | if (__PageMovable(page)) |
1407 | return pfn; | |
c8721bbb | 1408 | if (PageHuge(page)) { |
7e1f049e | 1409 | if (page_huge_active(page)) |
c8721bbb NH |
1410 | return pfn; |
1411 | else | |
1412 | pfn = round_up(pfn + 1, | |
1413 | 1 << compound_order(page)) - 1; | |
1414 | } | |
0c0e6195 KH |
1415 | } |
1416 | } | |
1417 | return 0; | |
1418 | } | |
1419 | ||
394e31d2 XQ |
1420 | static struct page *new_node_page(struct page *page, unsigned long private, |
1421 | int **result) | |
1422 | { | |
1423 | gfp_t gfp_mask = GFP_USER | __GFP_MOVABLE; | |
1424 | int nid = page_to_nid(page); | |
231e97e2 LZ |
1425 | nodemask_t nmask = node_states[N_MEMORY]; |
1426 | struct page *new_page = NULL; | |
394e31d2 XQ |
1427 | |
1428 | /* | |
1429 | * TODO: allocate a destination hugepage from a nearest neighbor node, | |
1430 | * accordance with memory policy of the user process if possible. For | |
1431 | * now as a simple work-around, we use the next node for destination. | |
1432 | */ | |
1433 | if (PageHuge(page)) | |
1434 | return alloc_huge_page_node(page_hstate(compound_head(page)), | |
1435 | next_node_in(nid, nmask)); | |
1436 | ||
231e97e2 | 1437 | node_clear(nid, nmask); |
9bb627be | 1438 | |
394e31d2 XQ |
1439 | if (PageHighMem(page) |
1440 | || (zone_idx(page_zone(page)) == ZONE_MOVABLE)) | |
1441 | gfp_mask |= __GFP_HIGHMEM; | |
1442 | ||
231e97e2 | 1443 | if (!nodes_empty(nmask)) |
04ec6264 | 1444 | new_page = __alloc_pages_nodemask(gfp_mask, 0, nid, &nmask); |
394e31d2 | 1445 | if (!new_page) |
04ec6264 | 1446 | new_page = __alloc_pages(gfp_mask, 0, nid); |
394e31d2 XQ |
1447 | |
1448 | return new_page; | |
1449 | } | |
1450 | ||
0c0e6195 KH |
1451 | #define NR_OFFLINE_AT_ONCE_PAGES (256) |
1452 | static int | |
1453 | do_migrate_range(unsigned long start_pfn, unsigned long end_pfn) | |
1454 | { | |
1455 | unsigned long pfn; | |
1456 | struct page *page; | |
1457 | int move_pages = NR_OFFLINE_AT_ONCE_PAGES; | |
1458 | int not_managed = 0; | |
1459 | int ret = 0; | |
1460 | LIST_HEAD(source); | |
1461 | ||
1462 | for (pfn = start_pfn; pfn < end_pfn && move_pages > 0; pfn++) { | |
1463 | if (!pfn_valid(pfn)) | |
1464 | continue; | |
1465 | page = pfn_to_page(pfn); | |
c8721bbb NH |
1466 | |
1467 | if (PageHuge(page)) { | |
1468 | struct page *head = compound_head(page); | |
1469 | pfn = page_to_pfn(head) + (1<<compound_order(head)) - 1; | |
1470 | if (compound_order(head) > PFN_SECTION_SHIFT) { | |
1471 | ret = -EBUSY; | |
1472 | break; | |
1473 | } | |
1474 | if (isolate_huge_page(page, &source)) | |
1475 | move_pages -= 1 << compound_order(head); | |
1476 | continue; | |
1477 | } | |
1478 | ||
700c2a46 | 1479 | if (!get_page_unless_zero(page)) |
0c0e6195 KH |
1480 | continue; |
1481 | /* | |
0efadf48 YX |
1482 | * We can skip free pages. And we can deal with pages on |
1483 | * LRU and non-lru movable pages. | |
0c0e6195 | 1484 | */ |
0efadf48 YX |
1485 | if (PageLRU(page)) |
1486 | ret = isolate_lru_page(page); | |
1487 | else | |
1488 | ret = isolate_movable_page(page, ISOLATE_UNEVICTABLE); | |
0c0e6195 | 1489 | if (!ret) { /* Success */ |
700c2a46 | 1490 | put_page(page); |
62695a84 | 1491 | list_add_tail(&page->lru, &source); |
0c0e6195 | 1492 | move_pages--; |
0efadf48 YX |
1493 | if (!__PageMovable(page)) |
1494 | inc_node_page_state(page, NR_ISOLATED_ANON + | |
1495 | page_is_file_cache(page)); | |
6d9c285a | 1496 | |
0c0e6195 | 1497 | } else { |
0c0e6195 | 1498 | #ifdef CONFIG_DEBUG_VM |
0efadf48 YX |
1499 | pr_alert("failed to isolate pfn %lx\n", pfn); |
1500 | dump_page(page, "isolation failed"); | |
0c0e6195 | 1501 | #endif |
700c2a46 | 1502 | put_page(page); |
25985edc | 1503 | /* Because we don't have big zone->lock. we should |
809c4449 BL |
1504 | check this again here. */ |
1505 | if (page_count(page)) { | |
1506 | not_managed++; | |
f3ab2636 | 1507 | ret = -EBUSY; |
809c4449 BL |
1508 | break; |
1509 | } | |
0c0e6195 KH |
1510 | } |
1511 | } | |
f3ab2636 BL |
1512 | if (!list_empty(&source)) { |
1513 | if (not_managed) { | |
c8721bbb | 1514 | putback_movable_pages(&source); |
f3ab2636 BL |
1515 | goto out; |
1516 | } | |
74c08f98 | 1517 | |
394e31d2 XQ |
1518 | /* Allocate a new page from the nearest neighbor node */ |
1519 | ret = migrate_pages(&source, new_node_page, NULL, 0, | |
9c620e2b | 1520 | MIGRATE_SYNC, MR_MEMORY_HOTPLUG); |
f3ab2636 | 1521 | if (ret) |
c8721bbb | 1522 | putback_movable_pages(&source); |
0c0e6195 | 1523 | } |
0c0e6195 KH |
1524 | out: |
1525 | return ret; | |
1526 | } | |
1527 | ||
1528 | /* | |
1529 | * remove from free_area[] and mark all as Reserved. | |
1530 | */ | |
1531 | static int | |
1532 | offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages, | |
1533 | void *data) | |
1534 | { | |
1535 | __offline_isolated_pages(start, start + nr_pages); | |
1536 | return 0; | |
1537 | } | |
1538 | ||
1539 | static void | |
1540 | offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn) | |
1541 | { | |
908eedc6 | 1542 | walk_system_ram_range(start_pfn, end_pfn - start_pfn, NULL, |
0c0e6195 KH |
1543 | offline_isolated_pages_cb); |
1544 | } | |
1545 | ||
1546 | /* | |
1547 | * Check all pages in range, recoreded as memory resource, are isolated. | |
1548 | */ | |
1549 | static int | |
1550 | check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages, | |
1551 | void *data) | |
1552 | { | |
1553 | int ret; | |
1554 | long offlined = *(long *)data; | |
b023f468 | 1555 | ret = test_pages_isolated(start_pfn, start_pfn + nr_pages, true); |
0c0e6195 KH |
1556 | offlined = nr_pages; |
1557 | if (!ret) | |
1558 | *(long *)data += offlined; | |
1559 | return ret; | |
1560 | } | |
1561 | ||
1562 | static long | |
1563 | check_pages_isolated(unsigned long start_pfn, unsigned long end_pfn) | |
1564 | { | |
1565 | long offlined = 0; | |
1566 | int ret; | |
1567 | ||
908eedc6 | 1568 | ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn, &offlined, |
0c0e6195 KH |
1569 | check_pages_isolated_cb); |
1570 | if (ret < 0) | |
1571 | offlined = (long)ret; | |
1572 | return offlined; | |
1573 | } | |
1574 | ||
c5320926 TC |
1575 | static int __init cmdline_parse_movable_node(char *p) |
1576 | { | |
4932381e | 1577 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
55ac590c | 1578 | movable_node_enabled = true; |
4932381e MH |
1579 | #else |
1580 | pr_warn("movable_node parameter depends on CONFIG_HAVE_MEMBLOCK_NODE_MAP to work properly\n"); | |
1581 | #endif | |
c5320926 TC |
1582 | return 0; |
1583 | } | |
1584 | early_param("movable_node", cmdline_parse_movable_node); | |
1585 | ||
d9713679 LJ |
1586 | /* check which state of node_states will be changed when offline memory */ |
1587 | static void node_states_check_changes_offline(unsigned long nr_pages, | |
1588 | struct zone *zone, struct memory_notify *arg) | |
1589 | { | |
1590 | struct pglist_data *pgdat = zone->zone_pgdat; | |
1591 | unsigned long present_pages = 0; | |
1592 | enum zone_type zt, zone_last = ZONE_NORMAL; | |
1593 | ||
1594 | /* | |
6715ddf9 LJ |
1595 | * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY] |
1596 | * contains nodes which have zones of 0...ZONE_NORMAL, | |
1597 | * set zone_last to ZONE_NORMAL. | |
d9713679 | 1598 | * |
6715ddf9 LJ |
1599 | * If we don't have HIGHMEM nor movable node, |
1600 | * node_states[N_NORMAL_MEMORY] contains nodes which have zones of | |
1601 | * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE. | |
d9713679 | 1602 | */ |
6715ddf9 | 1603 | if (N_MEMORY == N_NORMAL_MEMORY) |
d9713679 LJ |
1604 | zone_last = ZONE_MOVABLE; |
1605 | ||
1606 | /* | |
1607 | * check whether node_states[N_NORMAL_MEMORY] will be changed. | |
1608 | * If the memory to be offline is in a zone of 0...zone_last, | |
1609 | * and it is the last present memory, 0...zone_last will | |
1610 | * become empty after offline , thus we can determind we will | |
1611 | * need to clear the node from node_states[N_NORMAL_MEMORY]. | |
1612 | */ | |
1613 | for (zt = 0; zt <= zone_last; zt++) | |
1614 | present_pages += pgdat->node_zones[zt].present_pages; | |
1615 | if (zone_idx(zone) <= zone_last && nr_pages >= present_pages) | |
1616 | arg->status_change_nid_normal = zone_to_nid(zone); | |
1617 | else | |
1618 | arg->status_change_nid_normal = -1; | |
1619 | ||
6715ddf9 LJ |
1620 | #ifdef CONFIG_HIGHMEM |
1621 | /* | |
1622 | * If we have movable node, node_states[N_HIGH_MEMORY] | |
1623 | * contains nodes which have zones of 0...ZONE_HIGHMEM, | |
1624 | * set zone_last to ZONE_HIGHMEM. | |
1625 | * | |
1626 | * If we don't have movable node, node_states[N_NORMAL_MEMORY] | |
1627 | * contains nodes which have zones of 0...ZONE_MOVABLE, | |
1628 | * set zone_last to ZONE_MOVABLE. | |
1629 | */ | |
1630 | zone_last = ZONE_HIGHMEM; | |
1631 | if (N_MEMORY == N_HIGH_MEMORY) | |
1632 | zone_last = ZONE_MOVABLE; | |
1633 | ||
1634 | for (; zt <= zone_last; zt++) | |
1635 | present_pages += pgdat->node_zones[zt].present_pages; | |
1636 | if (zone_idx(zone) <= zone_last && nr_pages >= present_pages) | |
1637 | arg->status_change_nid_high = zone_to_nid(zone); | |
1638 | else | |
1639 | arg->status_change_nid_high = -1; | |
1640 | #else | |
1641 | arg->status_change_nid_high = arg->status_change_nid_normal; | |
1642 | #endif | |
1643 | ||
d9713679 LJ |
1644 | /* |
1645 | * node_states[N_HIGH_MEMORY] contains nodes which have 0...ZONE_MOVABLE | |
1646 | */ | |
1647 | zone_last = ZONE_MOVABLE; | |
1648 | ||
1649 | /* | |
1650 | * check whether node_states[N_HIGH_MEMORY] will be changed | |
1651 | * If we try to offline the last present @nr_pages from the node, | |
1652 | * we can determind we will need to clear the node from | |
1653 | * node_states[N_HIGH_MEMORY]. | |
1654 | */ | |
1655 | for (; zt <= zone_last; zt++) | |
1656 | present_pages += pgdat->node_zones[zt].present_pages; | |
1657 | if (nr_pages >= present_pages) | |
1658 | arg->status_change_nid = zone_to_nid(zone); | |
1659 | else | |
1660 | arg->status_change_nid = -1; | |
1661 | } | |
1662 | ||
1663 | static void node_states_clear_node(int node, struct memory_notify *arg) | |
1664 | { | |
1665 | if (arg->status_change_nid_normal >= 0) | |
1666 | node_clear_state(node, N_NORMAL_MEMORY); | |
1667 | ||
6715ddf9 LJ |
1668 | if ((N_MEMORY != N_NORMAL_MEMORY) && |
1669 | (arg->status_change_nid_high >= 0)) | |
d9713679 | 1670 | node_clear_state(node, N_HIGH_MEMORY); |
6715ddf9 LJ |
1671 | |
1672 | if ((N_MEMORY != N_HIGH_MEMORY) && | |
1673 | (arg->status_change_nid >= 0)) | |
1674 | node_clear_state(node, N_MEMORY); | |
d9713679 LJ |
1675 | } |
1676 | ||
a16cee10 | 1677 | static int __ref __offline_pages(unsigned long start_pfn, |
0c0e6195 KH |
1678 | unsigned long end_pfn, unsigned long timeout) |
1679 | { | |
1680 | unsigned long pfn, nr_pages, expire; | |
1681 | long offlined_pages; | |
7b78d335 | 1682 | int ret, drain, retry_max, node; |
d702909f | 1683 | unsigned long flags; |
a96dfddb | 1684 | unsigned long valid_start, valid_end; |
0c0e6195 | 1685 | struct zone *zone; |
7b78d335 | 1686 | struct memory_notify arg; |
0c0e6195 | 1687 | |
0c0e6195 KH |
1688 | /* at least, alignment against pageblock is necessary */ |
1689 | if (!IS_ALIGNED(start_pfn, pageblock_nr_pages)) | |
1690 | return -EINVAL; | |
1691 | if (!IS_ALIGNED(end_pfn, pageblock_nr_pages)) | |
1692 | return -EINVAL; | |
1693 | /* This makes hotplug much easier...and readable. | |
1694 | we assume this for now. .*/ | |
a96dfddb | 1695 | if (!test_pages_in_a_zone(start_pfn, end_pfn, &valid_start, &valid_end)) |
0c0e6195 | 1696 | return -EINVAL; |
7b78d335 | 1697 | |
a96dfddb | 1698 | zone = page_zone(pfn_to_page(valid_start)); |
7b78d335 YG |
1699 | node = zone_to_nid(zone); |
1700 | nr_pages = end_pfn - start_pfn; | |
1701 | ||
0c0e6195 | 1702 | /* set above range as isolated */ |
b023f468 WC |
1703 | ret = start_isolate_page_range(start_pfn, end_pfn, |
1704 | MIGRATE_MOVABLE, true); | |
0c0e6195 | 1705 | if (ret) |
30467e0b | 1706 | return ret; |
7b78d335 YG |
1707 | |
1708 | arg.start_pfn = start_pfn; | |
1709 | arg.nr_pages = nr_pages; | |
d9713679 | 1710 | node_states_check_changes_offline(nr_pages, zone, &arg); |
7b78d335 YG |
1711 | |
1712 | ret = memory_notify(MEM_GOING_OFFLINE, &arg); | |
1713 | ret = notifier_to_errno(ret); | |
1714 | if (ret) | |
1715 | goto failed_removal; | |
1716 | ||
0c0e6195 KH |
1717 | pfn = start_pfn; |
1718 | expire = jiffies + timeout; | |
1719 | drain = 0; | |
1720 | retry_max = 5; | |
1721 | repeat: | |
1722 | /* start memory hot removal */ | |
1723 | ret = -EAGAIN; | |
1724 | if (time_after(jiffies, expire)) | |
1725 | goto failed_removal; | |
1726 | ret = -EINTR; | |
1727 | if (signal_pending(current)) | |
1728 | goto failed_removal; | |
1729 | ret = 0; | |
1730 | if (drain) { | |
1731 | lru_add_drain_all(); | |
0c0e6195 | 1732 | cond_resched(); |
c0554329 | 1733 | drain_all_pages(zone); |
0c0e6195 KH |
1734 | } |
1735 | ||
c8721bbb NH |
1736 | pfn = scan_movable_pages(start_pfn, end_pfn); |
1737 | if (pfn) { /* We have movable pages */ | |
0c0e6195 KH |
1738 | ret = do_migrate_range(pfn, end_pfn); |
1739 | if (!ret) { | |
1740 | drain = 1; | |
1741 | goto repeat; | |
1742 | } else { | |
1743 | if (ret < 0) | |
1744 | if (--retry_max == 0) | |
1745 | goto failed_removal; | |
1746 | yield(); | |
1747 | drain = 1; | |
1748 | goto repeat; | |
1749 | } | |
1750 | } | |
b3834be5 | 1751 | /* drain all zone's lru pagevec, this is asynchronous... */ |
0c0e6195 | 1752 | lru_add_drain_all(); |
0c0e6195 | 1753 | yield(); |
b3834be5 | 1754 | /* drain pcp pages, this is synchronous. */ |
c0554329 | 1755 | drain_all_pages(zone); |
c8721bbb NH |
1756 | /* |
1757 | * dissolve free hugepages in the memory block before doing offlining | |
1758 | * actually in order to make hugetlbfs's object counting consistent. | |
1759 | */ | |
082d5b6b GS |
1760 | ret = dissolve_free_huge_pages(start_pfn, end_pfn); |
1761 | if (ret) | |
1762 | goto failed_removal; | |
0c0e6195 KH |
1763 | /* check again */ |
1764 | offlined_pages = check_pages_isolated(start_pfn, end_pfn); | |
1765 | if (offlined_pages < 0) { | |
1766 | ret = -EBUSY; | |
1767 | goto failed_removal; | |
1768 | } | |
e33e33b4 | 1769 | pr_info("Offlined Pages %ld\n", offlined_pages); |
b3834be5 | 1770 | /* Ok, all of our target is isolated. |
0c0e6195 KH |
1771 | We cannot do rollback at this point. */ |
1772 | offline_isolated_pages(start_pfn, end_pfn); | |
dbc0e4ce | 1773 | /* reset pagetype flags and makes migrate type to be MOVABLE */ |
0815f3d8 | 1774 | undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE); |
0c0e6195 | 1775 | /* removal success */ |
3dcc0571 | 1776 | adjust_managed_page_count(pfn_to_page(start_pfn), -offlined_pages); |
0c0e6195 | 1777 | zone->present_pages -= offlined_pages; |
d702909f CS |
1778 | |
1779 | pgdat_resize_lock(zone->zone_pgdat, &flags); | |
0c0e6195 | 1780 | zone->zone_pgdat->node_present_pages -= offlined_pages; |
d702909f | 1781 | pgdat_resize_unlock(zone->zone_pgdat, &flags); |
7b78d335 | 1782 | |
1b79acc9 KM |
1783 | init_per_zone_wmark_min(); |
1784 | ||
1e8537ba | 1785 | if (!populated_zone(zone)) { |
340175b7 | 1786 | zone_pcp_reset(zone); |
1e8537ba XQ |
1787 | mutex_lock(&zonelists_mutex); |
1788 | build_all_zonelists(NULL, NULL); | |
1789 | mutex_unlock(&zonelists_mutex); | |
1790 | } else | |
1791 | zone_pcp_update(zone); | |
340175b7 | 1792 | |
d9713679 | 1793 | node_states_clear_node(node, &arg); |
698b1b30 | 1794 | if (arg.status_change_nid >= 0) { |
8fe23e05 | 1795 | kswapd_stop(node); |
698b1b30 VB |
1796 | kcompactd_stop(node); |
1797 | } | |
bce7394a | 1798 | |
0c0e6195 KH |
1799 | vm_total_pages = nr_free_pagecache_pages(); |
1800 | writeback_set_ratelimit(); | |
7b78d335 YG |
1801 | |
1802 | memory_notify(MEM_OFFLINE, &arg); | |
0c0e6195 KH |
1803 | return 0; |
1804 | ||
1805 | failed_removal: | |
e33e33b4 CY |
1806 | pr_debug("memory offlining [mem %#010llx-%#010llx] failed\n", |
1807 | (unsigned long long) start_pfn << PAGE_SHIFT, | |
1808 | ((unsigned long long) end_pfn << PAGE_SHIFT) - 1); | |
7b78d335 | 1809 | memory_notify(MEM_CANCEL_OFFLINE, &arg); |
0c0e6195 | 1810 | /* pushback to free area */ |
0815f3d8 | 1811 | undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE); |
0c0e6195 KH |
1812 | return ret; |
1813 | } | |
71088785 | 1814 | |
30467e0b | 1815 | /* Must be protected by mem_hotplug_begin() */ |
a16cee10 WC |
1816 | int offline_pages(unsigned long start_pfn, unsigned long nr_pages) |
1817 | { | |
1818 | return __offline_pages(start_pfn, start_pfn + nr_pages, 120 * HZ); | |
1819 | } | |
e2ff3940 | 1820 | #endif /* CONFIG_MEMORY_HOTREMOVE */ |
a16cee10 | 1821 | |
bbc76be6 WC |
1822 | /** |
1823 | * walk_memory_range - walks through all mem sections in [start_pfn, end_pfn) | |
1824 | * @start_pfn: start pfn of the memory range | |
e05c4bbf | 1825 | * @end_pfn: end pfn of the memory range |
bbc76be6 WC |
1826 | * @arg: argument passed to func |
1827 | * @func: callback for each memory section walked | |
1828 | * | |
1829 | * This function walks through all present mem sections in range | |
1830 | * [start_pfn, end_pfn) and call func on each mem section. | |
1831 | * | |
1832 | * Returns the return value of func. | |
1833 | */ | |
e2ff3940 | 1834 | int walk_memory_range(unsigned long start_pfn, unsigned long end_pfn, |
bbc76be6 | 1835 | void *arg, int (*func)(struct memory_block *, void *)) |
71088785 | 1836 | { |
e90bdb7f WC |
1837 | struct memory_block *mem = NULL; |
1838 | struct mem_section *section; | |
e90bdb7f WC |
1839 | unsigned long pfn, section_nr; |
1840 | int ret; | |
e90bdb7f WC |
1841 | |
1842 | for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) { | |
1843 | section_nr = pfn_to_section_nr(pfn); | |
1844 | if (!present_section_nr(section_nr)) | |
1845 | continue; | |
1846 | ||
1847 | section = __nr_to_section(section_nr); | |
1848 | /* same memblock? */ | |
1849 | if (mem) | |
1850 | if ((section_nr >= mem->start_section_nr) && | |
1851 | (section_nr <= mem->end_section_nr)) | |
1852 | continue; | |
1853 | ||
1854 | mem = find_memory_block_hinted(section, mem); | |
1855 | if (!mem) | |
1856 | continue; | |
1857 | ||
bbc76be6 | 1858 | ret = func(mem, arg); |
e90bdb7f | 1859 | if (ret) { |
bbc76be6 WC |
1860 | kobject_put(&mem->dev.kobj); |
1861 | return ret; | |
e90bdb7f WC |
1862 | } |
1863 | } | |
1864 | ||
1865 | if (mem) | |
1866 | kobject_put(&mem->dev.kobj); | |
1867 | ||
bbc76be6 WC |
1868 | return 0; |
1869 | } | |
1870 | ||
e2ff3940 | 1871 | #ifdef CONFIG_MEMORY_HOTREMOVE |
d6de9d53 | 1872 | static int check_memblock_offlined_cb(struct memory_block *mem, void *arg) |
bbc76be6 WC |
1873 | { |
1874 | int ret = !is_memblock_offlined(mem); | |
1875 | ||
349daa0f RD |
1876 | if (unlikely(ret)) { |
1877 | phys_addr_t beginpa, endpa; | |
1878 | ||
1879 | beginpa = PFN_PHYS(section_nr_to_pfn(mem->start_section_nr)); | |
1880 | endpa = PFN_PHYS(section_nr_to_pfn(mem->end_section_nr + 1))-1; | |
756a025f | 1881 | pr_warn("removing memory fails, because memory [%pa-%pa] is onlined\n", |
349daa0f RD |
1882 | &beginpa, &endpa); |
1883 | } | |
bbc76be6 WC |
1884 | |
1885 | return ret; | |
1886 | } | |
1887 | ||
0f1cfe9d | 1888 | static int check_cpu_on_node(pg_data_t *pgdat) |
60a5a19e | 1889 | { |
60a5a19e TC |
1890 | int cpu; |
1891 | ||
1892 | for_each_present_cpu(cpu) { | |
1893 | if (cpu_to_node(cpu) == pgdat->node_id) | |
1894 | /* | |
1895 | * the cpu on this node isn't removed, and we can't | |
1896 | * offline this node. | |
1897 | */ | |
1898 | return -EBUSY; | |
1899 | } | |
1900 | ||
1901 | return 0; | |
1902 | } | |
1903 | ||
0f1cfe9d | 1904 | static void unmap_cpu_on_node(pg_data_t *pgdat) |
e13fe869 WC |
1905 | { |
1906 | #ifdef CONFIG_ACPI_NUMA | |
e13fe869 WC |
1907 | int cpu; |
1908 | ||
1909 | for_each_possible_cpu(cpu) | |
1910 | if (cpu_to_node(cpu) == pgdat->node_id) | |
1911 | numa_clear_node(cpu); | |
1912 | #endif | |
1913 | } | |
1914 | ||
0f1cfe9d | 1915 | static int check_and_unmap_cpu_on_node(pg_data_t *pgdat) |
e13fe869 | 1916 | { |
0f1cfe9d | 1917 | int ret; |
e13fe869 | 1918 | |
0f1cfe9d | 1919 | ret = check_cpu_on_node(pgdat); |
e13fe869 WC |
1920 | if (ret) |
1921 | return ret; | |
1922 | ||
1923 | /* | |
1924 | * the node will be offlined when we come here, so we can clear | |
1925 | * the cpu_to_node() now. | |
1926 | */ | |
1927 | ||
0f1cfe9d | 1928 | unmap_cpu_on_node(pgdat); |
e13fe869 WC |
1929 | return 0; |
1930 | } | |
1931 | ||
0f1cfe9d TK |
1932 | /** |
1933 | * try_offline_node | |
1934 | * | |
1935 | * Offline a node if all memory sections and cpus of the node are removed. | |
1936 | * | |
1937 | * NOTE: The caller must call lock_device_hotplug() to serialize hotplug | |
1938 | * and online/offline operations before this call. | |
1939 | */ | |
90b30cdc | 1940 | void try_offline_node(int nid) |
60a5a19e | 1941 | { |
d822b86a WC |
1942 | pg_data_t *pgdat = NODE_DATA(nid); |
1943 | unsigned long start_pfn = pgdat->node_start_pfn; | |
1944 | unsigned long end_pfn = start_pfn + pgdat->node_spanned_pages; | |
60a5a19e TC |
1945 | unsigned long pfn; |
1946 | ||
1947 | for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) { | |
1948 | unsigned long section_nr = pfn_to_section_nr(pfn); | |
1949 | ||
1950 | if (!present_section_nr(section_nr)) | |
1951 | continue; | |
1952 | ||
1953 | if (pfn_to_nid(pfn) != nid) | |
1954 | continue; | |
1955 | ||
1956 | /* | |
1957 | * some memory sections of this node are not removed, and we | |
1958 | * can't offline node now. | |
1959 | */ | |
1960 | return; | |
1961 | } | |
1962 | ||
0f1cfe9d | 1963 | if (check_and_unmap_cpu_on_node(pgdat)) |
60a5a19e TC |
1964 | return; |
1965 | ||
1966 | /* | |
1967 | * all memory/cpu of this node are removed, we can offline this | |
1968 | * node now. | |
1969 | */ | |
1970 | node_set_offline(nid); | |
1971 | unregister_one_node(nid); | |
1972 | } | |
90b30cdc | 1973 | EXPORT_SYMBOL(try_offline_node); |
60a5a19e | 1974 | |
0f1cfe9d TK |
1975 | /** |
1976 | * remove_memory | |
1977 | * | |
1978 | * NOTE: The caller must call lock_device_hotplug() to serialize hotplug | |
1979 | * and online/offline operations before this call, as required by | |
1980 | * try_offline_node(). | |
1981 | */ | |
242831eb | 1982 | void __ref remove_memory(int nid, u64 start, u64 size) |
bbc76be6 | 1983 | { |
242831eb | 1984 | int ret; |
993c1aad | 1985 | |
27356f54 TK |
1986 | BUG_ON(check_hotplug_memory_range(start, size)); |
1987 | ||
bfc8c901 | 1988 | mem_hotplug_begin(); |
6677e3ea YI |
1989 | |
1990 | /* | |
242831eb RW |
1991 | * All memory blocks must be offlined before removing memory. Check |
1992 | * whether all memory blocks in question are offline and trigger a BUG() | |
1993 | * if this is not the case. | |
6677e3ea | 1994 | */ |
242831eb | 1995 | ret = walk_memory_range(PFN_DOWN(start), PFN_UP(start + size - 1), NULL, |
d6de9d53 | 1996 | check_memblock_offlined_cb); |
bfc8c901 | 1997 | if (ret) |
242831eb | 1998 | BUG(); |
6677e3ea | 1999 | |
46c66c4b YI |
2000 | /* remove memmap entry */ |
2001 | firmware_map_remove(start, start + size, "System RAM"); | |
f9126ab9 XQ |
2002 | memblock_free(start, size); |
2003 | memblock_remove(start, size); | |
46c66c4b | 2004 | |
24d335ca WC |
2005 | arch_remove_memory(start, size); |
2006 | ||
60a5a19e TC |
2007 | try_offline_node(nid); |
2008 | ||
bfc8c901 | 2009 | mem_hotplug_done(); |
71088785 | 2010 | } |
71088785 | 2011 | EXPORT_SYMBOL_GPL(remove_memory); |
aba6efc4 | 2012 | #endif /* CONFIG_MEMORY_HOTREMOVE */ |