]>
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> |
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 | |
3f906ba2 | 54 | DEFINE_STATIC_PERCPU_RWSEM(mem_hotplug_lock); |
bfc8c901 | 55 | |
3f906ba2 TG |
56 | void get_online_mems(void) |
57 | { | |
58 | percpu_down_read(&mem_hotplug_lock); | |
59 | } | |
bfc8c901 | 60 | |
3f906ba2 TG |
61 | void put_online_mems(void) |
62 | { | |
63 | percpu_up_read(&mem_hotplug_lock); | |
64 | } | |
bfc8c901 | 65 | |
4932381e MH |
66 | bool movable_node_enabled = false; |
67 | ||
8604d9e5 | 68 | #ifndef CONFIG_MEMORY_HOTPLUG_DEFAULT_ONLINE |
31bc3858 | 69 | bool memhp_auto_online; |
8604d9e5 VK |
70 | #else |
71 | bool memhp_auto_online = true; | |
72 | #endif | |
31bc3858 VK |
73 | EXPORT_SYMBOL_GPL(memhp_auto_online); |
74 | ||
86dd995d VK |
75 | static int __init setup_memhp_default_state(char *str) |
76 | { | |
77 | if (!strcmp(str, "online")) | |
78 | memhp_auto_online = true; | |
79 | else if (!strcmp(str, "offline")) | |
80 | memhp_auto_online = false; | |
81 | ||
82 | return 1; | |
83 | } | |
84 | __setup("memhp_default_state=", setup_memhp_default_state); | |
85 | ||
30467e0b | 86 | void mem_hotplug_begin(void) |
20d6c96b | 87 | { |
3f906ba2 TG |
88 | cpus_read_lock(); |
89 | percpu_down_write(&mem_hotplug_lock); | |
20d6c96b KM |
90 | } |
91 | ||
30467e0b | 92 | void mem_hotplug_done(void) |
bfc8c901 | 93 | { |
3f906ba2 TG |
94 | percpu_up_write(&mem_hotplug_lock); |
95 | cpus_read_unlock(); | |
bfc8c901 | 96 | } |
20d6c96b | 97 | |
45e0b78b KM |
98 | /* add this memory to iomem resource */ |
99 | static struct resource *register_memory_resource(u64 start, u64 size) | |
100 | { | |
5042db43 | 101 | struct resource *res, *conflict; |
45e0b78b | 102 | res = kzalloc(sizeof(struct resource), GFP_KERNEL); |
6f754ba4 VK |
103 | if (!res) |
104 | return ERR_PTR(-ENOMEM); | |
45e0b78b KM |
105 | |
106 | res->name = "System RAM"; | |
107 | res->start = start; | |
108 | res->end = start + size - 1; | |
782b8664 | 109 | res->flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY; |
5042db43 JG |
110 | conflict = request_resource_conflict(&iomem_resource, res); |
111 | if (conflict) { | |
112 | if (conflict->desc == IORES_DESC_DEVICE_PRIVATE_MEMORY) { | |
113 | pr_debug("Device unaddressable memory block " | |
114 | "memory hotplug at %#010llx !\n", | |
115 | (unsigned long long)start); | |
116 | } | |
4996eed8 | 117 | pr_debug("System RAM resource %pR cannot be added\n", res); |
45e0b78b | 118 | kfree(res); |
6f754ba4 | 119 | return ERR_PTR(-EEXIST); |
45e0b78b KM |
120 | } |
121 | return res; | |
122 | } | |
123 | ||
124 | static void release_memory_resource(struct resource *res) | |
125 | { | |
126 | if (!res) | |
127 | return; | |
128 | release_resource(res); | |
129 | kfree(res); | |
130 | return; | |
131 | } | |
132 | ||
53947027 | 133 | #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE |
46723bfa YI |
134 | void get_page_bootmem(unsigned long info, struct page *page, |
135 | unsigned long type) | |
04753278 | 136 | { |
ddffe98d | 137 | page->freelist = (void *)type; |
04753278 YG |
138 | SetPagePrivate(page); |
139 | set_page_private(page, info); | |
fe896d18 | 140 | page_ref_inc(page); |
04753278 YG |
141 | } |
142 | ||
170a5a7e | 143 | void put_page_bootmem(struct page *page) |
04753278 | 144 | { |
5f24ce5f | 145 | unsigned long type; |
04753278 | 146 | |
ddffe98d | 147 | type = (unsigned long) page->freelist; |
5f24ce5f AA |
148 | BUG_ON(type < MEMORY_HOTPLUG_MIN_BOOTMEM_TYPE || |
149 | type > MEMORY_HOTPLUG_MAX_BOOTMEM_TYPE); | |
04753278 | 150 | |
fe896d18 | 151 | if (page_ref_dec_return(page) == 1) { |
ddffe98d | 152 | page->freelist = NULL; |
04753278 YG |
153 | ClearPagePrivate(page); |
154 | set_page_private(page, 0); | |
5f24ce5f | 155 | INIT_LIST_HEAD(&page->lru); |
170a5a7e | 156 | free_reserved_page(page); |
04753278 | 157 | } |
04753278 YG |
158 | } |
159 | ||
46723bfa YI |
160 | #ifdef CONFIG_HAVE_BOOTMEM_INFO_NODE |
161 | #ifndef CONFIG_SPARSEMEM_VMEMMAP | |
d92bc318 | 162 | static void register_page_bootmem_info_section(unsigned long start_pfn) |
04753278 YG |
163 | { |
164 | unsigned long *usemap, mapsize, section_nr, i; | |
165 | struct mem_section *ms; | |
166 | struct page *page, *memmap; | |
167 | ||
04753278 YG |
168 | section_nr = pfn_to_section_nr(start_pfn); |
169 | ms = __nr_to_section(section_nr); | |
170 | ||
171 | /* Get section's memmap address */ | |
172 | memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr); | |
173 | ||
174 | /* | |
175 | * Get page for the memmap's phys address | |
176 | * XXX: need more consideration for sparse_vmemmap... | |
177 | */ | |
178 | page = virt_to_page(memmap); | |
179 | mapsize = sizeof(struct page) * PAGES_PER_SECTION; | |
180 | mapsize = PAGE_ALIGN(mapsize) >> PAGE_SHIFT; | |
181 | ||
182 | /* remember memmap's page */ | |
183 | for (i = 0; i < mapsize; i++, page++) | |
184 | get_page_bootmem(section_nr, page, SECTION_INFO); | |
185 | ||
9ac9322d | 186 | usemap = ms->pageblock_flags; |
04753278 YG |
187 | page = virt_to_page(usemap); |
188 | ||
189 | mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT; | |
190 | ||
191 | for (i = 0; i < mapsize; i++, page++) | |
af370fb8 | 192 | get_page_bootmem(section_nr, page, MIX_SECTION_INFO); |
04753278 YG |
193 | |
194 | } | |
46723bfa YI |
195 | #else /* CONFIG_SPARSEMEM_VMEMMAP */ |
196 | static void register_page_bootmem_info_section(unsigned long start_pfn) | |
197 | { | |
198 | unsigned long *usemap, mapsize, section_nr, i; | |
199 | struct mem_section *ms; | |
200 | struct page *page, *memmap; | |
201 | ||
46723bfa YI |
202 | section_nr = pfn_to_section_nr(start_pfn); |
203 | ms = __nr_to_section(section_nr); | |
204 | ||
205 | memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr); | |
206 | ||
207 | register_page_bootmem_memmap(section_nr, memmap, PAGES_PER_SECTION); | |
208 | ||
9ac9322d | 209 | usemap = ms->pageblock_flags; |
46723bfa YI |
210 | page = virt_to_page(usemap); |
211 | ||
212 | mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT; | |
213 | ||
214 | for (i = 0; i < mapsize; i++, page++) | |
215 | get_page_bootmem(section_nr, page, MIX_SECTION_INFO); | |
216 | } | |
217 | #endif /* !CONFIG_SPARSEMEM_VMEMMAP */ | |
04753278 | 218 | |
7ded384a | 219 | void __init register_page_bootmem_info_node(struct pglist_data *pgdat) |
04753278 YG |
220 | { |
221 | unsigned long i, pfn, end_pfn, nr_pages; | |
222 | int node = pgdat->node_id; | |
223 | struct page *page; | |
04753278 YG |
224 | |
225 | nr_pages = PAGE_ALIGN(sizeof(struct pglist_data)) >> PAGE_SHIFT; | |
226 | page = virt_to_page(pgdat); | |
227 | ||
228 | for (i = 0; i < nr_pages; i++, page++) | |
229 | get_page_bootmem(node, page, NODE_INFO); | |
230 | ||
04753278 | 231 | pfn = pgdat->node_start_pfn; |
c1f19495 | 232 | end_pfn = pgdat_end_pfn(pgdat); |
04753278 | 233 | |
7e9f5eb0 | 234 | /* register section info */ |
f14851af | 235 | for (; pfn < end_pfn; pfn += PAGES_PER_SECTION) { |
236 | /* | |
237 | * Some platforms can assign the same pfn to multiple nodes - on | |
238 | * node0 as well as nodeN. To avoid registering a pfn against | |
239 | * multiple nodes we check that this pfn does not already | |
7e9f5eb0 | 240 | * reside in some other nodes. |
f14851af | 241 | */ |
f65e91df | 242 | if (pfn_valid(pfn) && (early_pfn_to_nid(pfn) == node)) |
f14851af | 243 | register_page_bootmem_info_section(pfn); |
244 | } | |
04753278 | 245 | } |
46723bfa | 246 | #endif /* CONFIG_HAVE_BOOTMEM_INFO_NODE */ |
04753278 | 247 | |
f1dd2cd1 | 248 | static int __meminit __add_section(int nid, unsigned long phys_start_pfn, |
7b73d978 | 249 | struct vmem_altmap *altmap, bool want_memblock) |
3947be19 | 250 | { |
3947be19 DH |
251 | int ret; |
252 | ||
ebd15302 KH |
253 | if (pfn_valid(phys_start_pfn)) |
254 | return -EEXIST; | |
255 | ||
7b73d978 | 256 | ret = sparse_add_one_section(NODE_DATA(nid), phys_start_pfn, altmap); |
3947be19 DH |
257 | if (ret < 0) |
258 | return ret; | |
259 | ||
1b862aec MH |
260 | if (!want_memblock) |
261 | return 0; | |
262 | ||
fc44f7f9 | 263 | return hotplug_memory_register(nid, __pfn_to_section(phys_start_pfn)); |
3947be19 DH |
264 | } |
265 | ||
4edd7cef DR |
266 | /* |
267 | * Reasonably generic function for adding memory. It is | |
268 | * expected that archs that support memory hotplug will | |
269 | * call this function after deciding the zone to which to | |
270 | * add the new pages. | |
271 | */ | |
f1dd2cd1 | 272 | int __ref __add_pages(int nid, unsigned long phys_start_pfn, |
24e6d5a5 CH |
273 | unsigned long nr_pages, struct vmem_altmap *altmap, |
274 | bool want_memblock) | |
4edd7cef DR |
275 | { |
276 | unsigned long i; | |
277 | int err = 0; | |
278 | int start_sec, end_sec; | |
4b94ffdc | 279 | |
4edd7cef DR |
280 | /* during initialize mem_map, align hot-added range to section */ |
281 | start_sec = pfn_to_section_nr(phys_start_pfn); | |
282 | end_sec = pfn_to_section_nr(phys_start_pfn + nr_pages - 1); | |
283 | ||
4b94ffdc DW |
284 | if (altmap) { |
285 | /* | |
286 | * Validate altmap is within bounds of the total request | |
287 | */ | |
288 | if (altmap->base_pfn != phys_start_pfn | |
289 | || vmem_altmap_offset(altmap) > nr_pages) { | |
290 | pr_warn_once("memory add fail, invalid altmap\n"); | |
7cf91a98 JK |
291 | err = -EINVAL; |
292 | goto out; | |
4b94ffdc DW |
293 | } |
294 | altmap->alloc = 0; | |
295 | } | |
296 | ||
4edd7cef | 297 | for (i = start_sec; i <= end_sec; i++) { |
7b73d978 CH |
298 | err = __add_section(nid, section_nr_to_pfn(i), altmap, |
299 | want_memblock); | |
4edd7cef DR |
300 | |
301 | /* | |
302 | * EEXIST is finally dealt with by ioresource collision | |
303 | * check. see add_memory() => register_memory_resource() | |
304 | * Warning will be printed if there is collision. | |
305 | */ | |
306 | if (err && (err != -EEXIST)) | |
307 | break; | |
308 | err = 0; | |
f64ac5e6 | 309 | cond_resched(); |
4edd7cef | 310 | } |
c435a390 | 311 | vmemmap_populate_print_last(); |
7cf91a98 | 312 | out: |
4edd7cef DR |
313 | return err; |
314 | } | |
4edd7cef DR |
315 | |
316 | #ifdef CONFIG_MEMORY_HOTREMOVE | |
815121d2 | 317 | /* find the smallest valid pfn in the range [start_pfn, end_pfn) */ |
d09b0137 | 318 | static unsigned long find_smallest_section_pfn(int nid, struct zone *zone, |
815121d2 YI |
319 | unsigned long start_pfn, |
320 | unsigned long end_pfn) | |
321 | { | |
322 | struct mem_section *ms; | |
323 | ||
324 | for (; start_pfn < end_pfn; start_pfn += PAGES_PER_SECTION) { | |
325 | ms = __pfn_to_section(start_pfn); | |
326 | ||
327 | if (unlikely(!valid_section(ms))) | |
328 | continue; | |
329 | ||
330 | if (unlikely(pfn_to_nid(start_pfn) != nid)) | |
331 | continue; | |
332 | ||
333 | if (zone && zone != page_zone(pfn_to_page(start_pfn))) | |
334 | continue; | |
335 | ||
336 | return start_pfn; | |
337 | } | |
338 | ||
339 | return 0; | |
340 | } | |
341 | ||
342 | /* find the biggest valid pfn in the range [start_pfn, end_pfn). */ | |
d09b0137 | 343 | static unsigned long find_biggest_section_pfn(int nid, struct zone *zone, |
815121d2 YI |
344 | unsigned long start_pfn, |
345 | unsigned long end_pfn) | |
346 | { | |
347 | struct mem_section *ms; | |
348 | unsigned long pfn; | |
349 | ||
350 | /* pfn is the end pfn of a memory section. */ | |
351 | pfn = end_pfn - 1; | |
352 | for (; pfn >= start_pfn; pfn -= PAGES_PER_SECTION) { | |
353 | ms = __pfn_to_section(pfn); | |
354 | ||
355 | if (unlikely(!valid_section(ms))) | |
356 | continue; | |
357 | ||
358 | if (unlikely(pfn_to_nid(pfn) != nid)) | |
359 | continue; | |
360 | ||
361 | if (zone && zone != page_zone(pfn_to_page(pfn))) | |
362 | continue; | |
363 | ||
364 | return pfn; | |
365 | } | |
366 | ||
367 | return 0; | |
368 | } | |
369 | ||
370 | static void shrink_zone_span(struct zone *zone, unsigned long start_pfn, | |
371 | unsigned long end_pfn) | |
372 | { | |
c33bc315 XQ |
373 | unsigned long zone_start_pfn = zone->zone_start_pfn; |
374 | unsigned long z = zone_end_pfn(zone); /* zone_end_pfn namespace clash */ | |
375 | unsigned long zone_end_pfn = z; | |
815121d2 YI |
376 | unsigned long pfn; |
377 | struct mem_section *ms; | |
378 | int nid = zone_to_nid(zone); | |
379 | ||
380 | zone_span_writelock(zone); | |
381 | if (zone_start_pfn == start_pfn) { | |
382 | /* | |
383 | * If the section is smallest section in the zone, it need | |
384 | * shrink zone->zone_start_pfn and zone->zone_spanned_pages. | |
385 | * In this case, we find second smallest valid mem_section | |
386 | * for shrinking zone. | |
387 | */ | |
388 | pfn = find_smallest_section_pfn(nid, zone, end_pfn, | |
389 | zone_end_pfn); | |
390 | if (pfn) { | |
391 | zone->zone_start_pfn = pfn; | |
392 | zone->spanned_pages = zone_end_pfn - pfn; | |
393 | } | |
394 | } else if (zone_end_pfn == end_pfn) { | |
395 | /* | |
396 | * If the section is biggest section in the zone, it need | |
397 | * shrink zone->spanned_pages. | |
398 | * In this case, we find second biggest valid mem_section for | |
399 | * shrinking zone. | |
400 | */ | |
401 | pfn = find_biggest_section_pfn(nid, zone, zone_start_pfn, | |
402 | start_pfn); | |
403 | if (pfn) | |
404 | zone->spanned_pages = pfn - zone_start_pfn + 1; | |
405 | } | |
406 | ||
407 | /* | |
408 | * The section is not biggest or smallest mem_section in the zone, it | |
409 | * only creates a hole in the zone. So in this case, we need not | |
410 | * change the zone. But perhaps, the zone has only hole data. Thus | |
411 | * it check the zone has only hole or not. | |
412 | */ | |
413 | pfn = zone_start_pfn; | |
414 | for (; pfn < zone_end_pfn; pfn += PAGES_PER_SECTION) { | |
415 | ms = __pfn_to_section(pfn); | |
416 | ||
417 | if (unlikely(!valid_section(ms))) | |
418 | continue; | |
419 | ||
420 | if (page_zone(pfn_to_page(pfn)) != zone) | |
421 | continue; | |
422 | ||
423 | /* If the section is current section, it continues the loop */ | |
424 | if (start_pfn == pfn) | |
425 | continue; | |
426 | ||
427 | /* If we find valid section, we have nothing to do */ | |
428 | zone_span_writeunlock(zone); | |
429 | return; | |
430 | } | |
431 | ||
432 | /* The zone has no valid section */ | |
433 | zone->zone_start_pfn = 0; | |
434 | zone->spanned_pages = 0; | |
435 | zone_span_writeunlock(zone); | |
436 | } | |
437 | ||
438 | static void shrink_pgdat_span(struct pglist_data *pgdat, | |
439 | unsigned long start_pfn, unsigned long end_pfn) | |
440 | { | |
83285c72 XQ |
441 | unsigned long pgdat_start_pfn = pgdat->node_start_pfn; |
442 | unsigned long p = pgdat_end_pfn(pgdat); /* pgdat_end_pfn namespace clash */ | |
443 | unsigned long pgdat_end_pfn = p; | |
815121d2 YI |
444 | unsigned long pfn; |
445 | struct mem_section *ms; | |
446 | int nid = pgdat->node_id; | |
447 | ||
448 | if (pgdat_start_pfn == start_pfn) { | |
449 | /* | |
450 | * If the section is smallest section in the pgdat, it need | |
451 | * shrink pgdat->node_start_pfn and pgdat->node_spanned_pages. | |
452 | * In this case, we find second smallest valid mem_section | |
453 | * for shrinking zone. | |
454 | */ | |
455 | pfn = find_smallest_section_pfn(nid, NULL, end_pfn, | |
456 | pgdat_end_pfn); | |
457 | if (pfn) { | |
458 | pgdat->node_start_pfn = pfn; | |
459 | pgdat->node_spanned_pages = pgdat_end_pfn - pfn; | |
460 | } | |
461 | } else if (pgdat_end_pfn == end_pfn) { | |
462 | /* | |
463 | * If the section is biggest section in the pgdat, it need | |
464 | * shrink pgdat->node_spanned_pages. | |
465 | * In this case, we find second biggest valid mem_section for | |
466 | * shrinking zone. | |
467 | */ | |
468 | pfn = find_biggest_section_pfn(nid, NULL, pgdat_start_pfn, | |
469 | start_pfn); | |
470 | if (pfn) | |
471 | pgdat->node_spanned_pages = pfn - pgdat_start_pfn + 1; | |
472 | } | |
473 | ||
474 | /* | |
475 | * If the section is not biggest or smallest mem_section in the pgdat, | |
476 | * it only creates a hole in the pgdat. So in this case, we need not | |
477 | * change the pgdat. | |
478 | * But perhaps, the pgdat has only hole data. Thus it check the pgdat | |
479 | * has only hole or not. | |
480 | */ | |
481 | pfn = pgdat_start_pfn; | |
482 | for (; pfn < pgdat_end_pfn; pfn += PAGES_PER_SECTION) { | |
483 | ms = __pfn_to_section(pfn); | |
484 | ||
485 | if (unlikely(!valid_section(ms))) | |
486 | continue; | |
487 | ||
488 | if (pfn_to_nid(pfn) != nid) | |
489 | continue; | |
490 | ||
491 | /* If the section is current section, it continues the loop */ | |
492 | if (start_pfn == pfn) | |
493 | continue; | |
494 | ||
495 | /* If we find valid section, we have nothing to do */ | |
496 | return; | |
497 | } | |
498 | ||
499 | /* The pgdat has no valid section */ | |
500 | pgdat->node_start_pfn = 0; | |
501 | pgdat->node_spanned_pages = 0; | |
502 | } | |
503 | ||
504 | static void __remove_zone(struct zone *zone, unsigned long start_pfn) | |
505 | { | |
506 | struct pglist_data *pgdat = zone->zone_pgdat; | |
507 | int nr_pages = PAGES_PER_SECTION; | |
815121d2 YI |
508 | unsigned long flags; |
509 | ||
815121d2 YI |
510 | pgdat_resize_lock(zone->zone_pgdat, &flags); |
511 | shrink_zone_span(zone, start_pfn, start_pfn + nr_pages); | |
512 | shrink_pgdat_span(pgdat, start_pfn, start_pfn + nr_pages); | |
513 | pgdat_resize_unlock(zone->zone_pgdat, &flags); | |
514 | } | |
515 | ||
4b94ffdc | 516 | static int __remove_section(struct zone *zone, struct mem_section *ms, |
24b6d416 | 517 | unsigned long map_offset, struct vmem_altmap *altmap) |
ea01ea93 | 518 | { |
815121d2 YI |
519 | unsigned long start_pfn; |
520 | int scn_nr; | |
ea01ea93 BP |
521 | int ret = -EINVAL; |
522 | ||
523 | if (!valid_section(ms)) | |
524 | return ret; | |
525 | ||
526 | ret = unregister_memory_section(ms); | |
527 | if (ret) | |
528 | return ret; | |
529 | ||
815121d2 | 530 | scn_nr = __section_nr(ms); |
1dd2bfc8 | 531 | start_pfn = section_nr_to_pfn((unsigned long)scn_nr); |
815121d2 YI |
532 | __remove_zone(zone, start_pfn); |
533 | ||
24b6d416 | 534 | sparse_remove_one_section(zone, ms, map_offset, altmap); |
ea01ea93 BP |
535 | return 0; |
536 | } | |
537 | ||
ea01ea93 BP |
538 | /** |
539 | * __remove_pages() - remove sections of pages from a zone | |
540 | * @zone: zone from which pages need to be removed | |
541 | * @phys_start_pfn: starting pageframe (must be aligned to start of a section) | |
542 | * @nr_pages: number of pages to remove (must be multiple of section size) | |
e8b098fc | 543 | * @altmap: alternative device page map or %NULL if default memmap is used |
ea01ea93 BP |
544 | * |
545 | * Generic helper function to remove section mappings and sysfs entries | |
546 | * for the section of the memory we are removing. Caller needs to make | |
547 | * sure that pages are marked reserved and zones are adjust properly by | |
548 | * calling offline_pages(). | |
549 | */ | |
550 | int __remove_pages(struct zone *zone, unsigned long phys_start_pfn, | |
da024512 | 551 | unsigned long nr_pages, struct vmem_altmap *altmap) |
ea01ea93 | 552 | { |
fe74ebb1 | 553 | unsigned long i; |
4b94ffdc DW |
554 | unsigned long map_offset = 0; |
555 | int sections_to_remove, ret = 0; | |
556 | ||
557 | /* In the ZONE_DEVICE case device driver owns the memory region */ | |
558 | if (is_dev_zone(zone)) { | |
4b94ffdc DW |
559 | if (altmap) |
560 | map_offset = vmem_altmap_offset(altmap); | |
561 | } else { | |
562 | resource_size_t start, size; | |
563 | ||
564 | start = phys_start_pfn << PAGE_SHIFT; | |
565 | size = nr_pages * PAGE_SIZE; | |
566 | ||
567 | ret = release_mem_region_adjustable(&iomem_resource, start, | |
568 | size); | |
569 | if (ret) { | |
570 | resource_size_t endres = start + size - 1; | |
571 | ||
572 | pr_warn("Unable to release resource <%pa-%pa> (%d)\n", | |
573 | &start, &endres, ret); | |
574 | } | |
575 | } | |
ea01ea93 | 576 | |
7cf91a98 JK |
577 | clear_zone_contiguous(zone); |
578 | ||
ea01ea93 BP |
579 | /* |
580 | * We can only remove entire sections | |
581 | */ | |
582 | BUG_ON(phys_start_pfn & ~PAGE_SECTION_MASK); | |
583 | BUG_ON(nr_pages % PAGES_PER_SECTION); | |
584 | ||
ea01ea93 BP |
585 | sections_to_remove = nr_pages / PAGES_PER_SECTION; |
586 | for (i = 0; i < sections_to_remove; i++) { | |
587 | unsigned long pfn = phys_start_pfn + i*PAGES_PER_SECTION; | |
4b94ffdc | 588 | |
dd33ad7b | 589 | cond_resched(); |
24b6d416 CH |
590 | ret = __remove_section(zone, __pfn_to_section(pfn), map_offset, |
591 | altmap); | |
4b94ffdc | 592 | map_offset = 0; |
ea01ea93 BP |
593 | if (ret) |
594 | break; | |
595 | } | |
7cf91a98 JK |
596 | |
597 | set_zone_contiguous(zone); | |
598 | ||
ea01ea93 BP |
599 | return ret; |
600 | } | |
4edd7cef | 601 | #endif /* CONFIG_MEMORY_HOTREMOVE */ |
ea01ea93 | 602 | |
9d0ad8ca DK |
603 | int set_online_page_callback(online_page_callback_t callback) |
604 | { | |
605 | int rc = -EINVAL; | |
606 | ||
bfc8c901 VD |
607 | get_online_mems(); |
608 | mutex_lock(&online_page_callback_lock); | |
9d0ad8ca DK |
609 | |
610 | if (online_page_callback == generic_online_page) { | |
611 | online_page_callback = callback; | |
612 | rc = 0; | |
613 | } | |
614 | ||
bfc8c901 VD |
615 | mutex_unlock(&online_page_callback_lock); |
616 | put_online_mems(); | |
9d0ad8ca DK |
617 | |
618 | return rc; | |
619 | } | |
620 | EXPORT_SYMBOL_GPL(set_online_page_callback); | |
621 | ||
622 | int restore_online_page_callback(online_page_callback_t callback) | |
623 | { | |
624 | int rc = -EINVAL; | |
625 | ||
bfc8c901 VD |
626 | get_online_mems(); |
627 | mutex_lock(&online_page_callback_lock); | |
9d0ad8ca DK |
628 | |
629 | if (online_page_callback == callback) { | |
630 | online_page_callback = generic_online_page; | |
631 | rc = 0; | |
632 | } | |
633 | ||
bfc8c901 VD |
634 | mutex_unlock(&online_page_callback_lock); |
635 | put_online_mems(); | |
9d0ad8ca DK |
636 | |
637 | return rc; | |
638 | } | |
639 | EXPORT_SYMBOL_GPL(restore_online_page_callback); | |
640 | ||
641 | void __online_page_set_limits(struct page *page) | |
180c06ef | 642 | { |
9d0ad8ca DK |
643 | } |
644 | EXPORT_SYMBOL_GPL(__online_page_set_limits); | |
645 | ||
646 | void __online_page_increment_counters(struct page *page) | |
647 | { | |
3dcc0571 | 648 | adjust_managed_page_count(page, 1); |
9d0ad8ca DK |
649 | } |
650 | EXPORT_SYMBOL_GPL(__online_page_increment_counters); | |
180c06ef | 651 | |
9d0ad8ca DK |
652 | void __online_page_free(struct page *page) |
653 | { | |
3dcc0571 | 654 | __free_reserved_page(page); |
180c06ef | 655 | } |
9d0ad8ca DK |
656 | EXPORT_SYMBOL_GPL(__online_page_free); |
657 | ||
658 | static void generic_online_page(struct page *page) | |
659 | { | |
660 | __online_page_set_limits(page); | |
661 | __online_page_increment_counters(page); | |
662 | __online_page_free(page); | |
663 | } | |
180c06ef | 664 | |
75884fb1 KH |
665 | static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages, |
666 | void *arg) | |
3947be19 DH |
667 | { |
668 | unsigned long i; | |
75884fb1 KH |
669 | unsigned long onlined_pages = *(unsigned long *)arg; |
670 | struct page *page; | |
2d070eab | 671 | |
75884fb1 KH |
672 | if (PageReserved(pfn_to_page(start_pfn))) |
673 | for (i = 0; i < nr_pages; i++) { | |
674 | page = pfn_to_page(start_pfn + i); | |
9d0ad8ca | 675 | (*online_page_callback)(page); |
75884fb1 KH |
676 | onlined_pages++; |
677 | } | |
2d070eab MH |
678 | |
679 | online_mem_sections(start_pfn, start_pfn + nr_pages); | |
680 | ||
75884fb1 KH |
681 | *(unsigned long *)arg = onlined_pages; |
682 | return 0; | |
683 | } | |
684 | ||
d9713679 LJ |
685 | /* check which state of node_states will be changed when online memory */ |
686 | static void node_states_check_changes_online(unsigned long nr_pages, | |
687 | struct zone *zone, struct memory_notify *arg) | |
688 | { | |
689 | int nid = zone_to_nid(zone); | |
d9713679 | 690 | |
8efe33f4 OS |
691 | arg->status_change_nid = -1; |
692 | arg->status_change_nid_normal = -1; | |
693 | arg->status_change_nid_high = -1; | |
d9713679 | 694 | |
8efe33f4 OS |
695 | if (!node_state(nid, N_MEMORY)) |
696 | arg->status_change_nid = nid; | |
697 | if (zone_idx(zone) <= ZONE_NORMAL && !node_state(nid, N_NORMAL_MEMORY)) | |
d9713679 | 698 | arg->status_change_nid_normal = nid; |
6715ddf9 | 699 | #ifdef CONFIG_HIGHMEM |
8efe33f4 | 700 | if (zone_idx(zone) <= N_HIGH_MEMORY && !node_state(nid, N_HIGH_MEMORY)) |
6715ddf9 | 701 | arg->status_change_nid_high = nid; |
6715ddf9 | 702 | #endif |
d9713679 LJ |
703 | } |
704 | ||
705 | static void node_states_set_node(int node, struct memory_notify *arg) | |
706 | { | |
707 | if (arg->status_change_nid_normal >= 0) | |
708 | node_set_state(node, N_NORMAL_MEMORY); | |
709 | ||
6715ddf9 LJ |
710 | if (arg->status_change_nid_high >= 0) |
711 | node_set_state(node, N_HIGH_MEMORY); | |
712 | ||
83d83612 OS |
713 | if (arg->status_change_nid >= 0) |
714 | node_set_state(node, N_MEMORY); | |
d9713679 LJ |
715 | } |
716 | ||
f1dd2cd1 MH |
717 | static void __meminit resize_zone_range(struct zone *zone, unsigned long start_pfn, |
718 | unsigned long nr_pages) | |
719 | { | |
720 | unsigned long old_end_pfn = zone_end_pfn(zone); | |
721 | ||
722 | if (zone_is_empty(zone) || start_pfn < zone->zone_start_pfn) | |
723 | zone->zone_start_pfn = start_pfn; | |
724 | ||
725 | zone->spanned_pages = max(start_pfn + nr_pages, old_end_pfn) - zone->zone_start_pfn; | |
726 | } | |
727 | ||
728 | static void __meminit resize_pgdat_range(struct pglist_data *pgdat, unsigned long start_pfn, | |
729 | unsigned long nr_pages) | |
730 | { | |
731 | unsigned long old_end_pfn = pgdat_end_pfn(pgdat); | |
732 | ||
733 | if (!pgdat->node_spanned_pages || start_pfn < pgdat->node_start_pfn) | |
734 | pgdat->node_start_pfn = start_pfn; | |
735 | ||
736 | pgdat->node_spanned_pages = max(start_pfn + nr_pages, old_end_pfn) - pgdat->node_start_pfn; | |
737 | } | |
738 | ||
a99583e7 CH |
739 | void __ref move_pfn_range_to_zone(struct zone *zone, unsigned long start_pfn, |
740 | unsigned long nr_pages, struct vmem_altmap *altmap) | |
f1dd2cd1 MH |
741 | { |
742 | struct pglist_data *pgdat = zone->zone_pgdat; | |
743 | int nid = pgdat->node_id; | |
744 | unsigned long flags; | |
df429ac0 | 745 | |
f1dd2cd1 MH |
746 | if (zone_is_empty(zone)) |
747 | init_currently_empty_zone(zone, start_pfn, nr_pages); | |
df429ac0 | 748 | |
f1dd2cd1 MH |
749 | clear_zone_contiguous(zone); |
750 | ||
751 | /* TODO Huh pgdat is irqsave while zone is not. It used to be like that before */ | |
752 | pgdat_resize_lock(pgdat, &flags); | |
753 | zone_span_writelock(zone); | |
754 | resize_zone_range(zone, start_pfn, nr_pages); | |
755 | zone_span_writeunlock(zone); | |
756 | resize_pgdat_range(pgdat, start_pfn, nr_pages); | |
757 | pgdat_resize_unlock(pgdat, &flags); | |
758 | ||
759 | /* | |
760 | * TODO now we have a visible range of pages which are not associated | |
761 | * with their zone properly. Not nice but set_pfnblock_flags_mask | |
762 | * expects the zone spans the pfn range. All the pages in the range | |
763 | * are reserved so nobody should be touching them so we should be safe | |
764 | */ | |
a99583e7 CH |
765 | memmap_init_zone(nr_pages, nid, zone_idx(zone), start_pfn, |
766 | MEMMAP_HOTPLUG, altmap); | |
f1dd2cd1 MH |
767 | |
768 | set_zone_contiguous(zone); | |
769 | } | |
770 | ||
c246a213 MH |
771 | /* |
772 | * Returns a default kernel memory zone for the given pfn range. | |
773 | * If no kernel zone covers this pfn range it will automatically go | |
774 | * to the ZONE_NORMAL. | |
775 | */ | |
c6f03e29 | 776 | static struct zone *default_kernel_zone_for_pfn(int nid, unsigned long start_pfn, |
c246a213 MH |
777 | unsigned long nr_pages) |
778 | { | |
779 | struct pglist_data *pgdat = NODE_DATA(nid); | |
780 | int zid; | |
781 | ||
782 | for (zid = 0; zid <= ZONE_NORMAL; zid++) { | |
783 | struct zone *zone = &pgdat->node_zones[zid]; | |
784 | ||
785 | if (zone_intersects(zone, start_pfn, nr_pages)) | |
786 | return zone; | |
787 | } | |
788 | ||
789 | return &pgdat->node_zones[ZONE_NORMAL]; | |
790 | } | |
791 | ||
c6f03e29 MH |
792 | static inline struct zone *default_zone_for_pfn(int nid, unsigned long start_pfn, |
793 | unsigned long nr_pages) | |
e5e68930 | 794 | { |
c6f03e29 MH |
795 | struct zone *kernel_zone = default_kernel_zone_for_pfn(nid, start_pfn, |
796 | nr_pages); | |
797 | struct zone *movable_zone = &NODE_DATA(nid)->node_zones[ZONE_MOVABLE]; | |
798 | bool in_kernel = zone_intersects(kernel_zone, start_pfn, nr_pages); | |
799 | bool in_movable = zone_intersects(movable_zone, start_pfn, nr_pages); | |
e5e68930 MH |
800 | |
801 | /* | |
c6f03e29 MH |
802 | * We inherit the existing zone in a simple case where zones do not |
803 | * overlap in the given range | |
e5e68930 | 804 | */ |
c6f03e29 MH |
805 | if (in_kernel ^ in_movable) |
806 | return (in_kernel) ? kernel_zone : movable_zone; | |
9f123ab5 | 807 | |
c6f03e29 MH |
808 | /* |
809 | * If the range doesn't belong to any zone or two zones overlap in the | |
810 | * given range then we use movable zone only if movable_node is | |
811 | * enabled because we always online to a kernel zone by default. | |
812 | */ | |
813 | return movable_node_enabled ? movable_zone : kernel_zone; | |
9f123ab5 MH |
814 | } |
815 | ||
e5e68930 MH |
816 | struct zone * zone_for_pfn_range(int online_type, int nid, unsigned start_pfn, |
817 | unsigned long nr_pages) | |
f1dd2cd1 | 818 | { |
c6f03e29 MH |
819 | if (online_type == MMOP_ONLINE_KERNEL) |
820 | return default_kernel_zone_for_pfn(nid, start_pfn, nr_pages); | |
f1dd2cd1 | 821 | |
c6f03e29 MH |
822 | if (online_type == MMOP_ONLINE_MOVABLE) |
823 | return &NODE_DATA(nid)->node_zones[ZONE_MOVABLE]; | |
df429ac0 | 824 | |
c6f03e29 | 825 | return default_zone_for_pfn(nid, start_pfn, nr_pages); |
e5e68930 MH |
826 | } |
827 | ||
828 | /* | |
829 | * Associates the given pfn range with the given node and the zone appropriate | |
830 | * for the given online type. | |
831 | */ | |
832 | static struct zone * __meminit move_pfn_range(int online_type, int nid, | |
833 | unsigned long start_pfn, unsigned long nr_pages) | |
834 | { | |
835 | struct zone *zone; | |
836 | ||
837 | zone = zone_for_pfn_range(online_type, nid, start_pfn, nr_pages); | |
a99583e7 | 838 | move_pfn_range_to_zone(zone, start_pfn, nr_pages, NULL); |
f1dd2cd1 | 839 | return zone; |
df429ac0 | 840 | } |
75884fb1 | 841 | |
511c2aba | 842 | int __ref online_pages(unsigned long pfn, unsigned long nr_pages, int online_type) |
75884fb1 | 843 | { |
aa47228a | 844 | unsigned long flags; |
3947be19 DH |
845 | unsigned long onlined_pages = 0; |
846 | struct zone *zone; | |
6811378e | 847 | int need_zonelists_rebuild = 0; |
7b78d335 YG |
848 | int nid; |
849 | int ret; | |
850 | struct memory_notify arg; | |
d0dc12e8 PT |
851 | struct memory_block *mem; |
852 | ||
381eab4a DH |
853 | mem_hotplug_begin(); |
854 | ||
d0dc12e8 PT |
855 | /* |
856 | * We can't use pfn_to_nid() because nid might be stored in struct page | |
857 | * which is not yet initialized. Instead, we find nid from memory block. | |
858 | */ | |
859 | mem = find_memory_block(__pfn_to_section(pfn)); | |
860 | nid = mem->nid; | |
7b78d335 | 861 | |
f1dd2cd1 MH |
862 | /* associate pfn range with the zone */ |
863 | zone = move_pfn_range(online_type, nid, pfn, nr_pages); | |
864 | ||
7b78d335 YG |
865 | arg.start_pfn = pfn; |
866 | arg.nr_pages = nr_pages; | |
d9713679 | 867 | node_states_check_changes_online(nr_pages, zone, &arg); |
7b78d335 | 868 | |
7b78d335 YG |
869 | ret = memory_notify(MEM_GOING_ONLINE, &arg); |
870 | ret = notifier_to_errno(ret); | |
e33e33b4 CY |
871 | if (ret) |
872 | goto failed_addition; | |
873 | ||
6811378e YG |
874 | /* |
875 | * If this zone is not populated, then it is not in zonelist. | |
876 | * This means the page allocator ignores this zone. | |
877 | * So, zonelist must be updated after online. | |
878 | */ | |
6dcd73d7 | 879 | if (!populated_zone(zone)) { |
6811378e | 880 | need_zonelists_rebuild = 1; |
72675e13 | 881 | setup_zone_pageset(zone); |
6dcd73d7 | 882 | } |
6811378e | 883 | |
908eedc6 | 884 | ret = walk_system_ram_range(pfn, nr_pages, &onlined_pages, |
75884fb1 | 885 | online_pages_range); |
fd8a4221 | 886 | if (ret) { |
6dcd73d7 WC |
887 | if (need_zonelists_rebuild) |
888 | zone_pcp_reset(zone); | |
e33e33b4 | 889 | goto failed_addition; |
fd8a4221 GL |
890 | } |
891 | ||
3947be19 | 892 | zone->present_pages += onlined_pages; |
aa47228a CS |
893 | |
894 | pgdat_resize_lock(zone->zone_pgdat, &flags); | |
f2937be5 | 895 | zone->zone_pgdat->node_present_pages += onlined_pages; |
aa47228a CS |
896 | pgdat_resize_unlock(zone->zone_pgdat, &flags); |
897 | ||
08dff7b7 | 898 | if (onlined_pages) { |
e888ca35 | 899 | node_states_set_node(nid, &arg); |
08dff7b7 | 900 | if (need_zonelists_rebuild) |
72675e13 | 901 | build_all_zonelists(NULL); |
08dff7b7 JL |
902 | else |
903 | zone_pcp_update(zone); | |
904 | } | |
3947be19 | 905 | |
1b79acc9 KM |
906 | init_per_zone_wmark_min(); |
907 | ||
698b1b30 | 908 | if (onlined_pages) { |
e888ca35 | 909 | kswapd_run(nid); |
698b1b30 VB |
910 | kcompactd_run(nid); |
911 | } | |
61b13993 | 912 | |
1f522509 | 913 | vm_total_pages = nr_free_pagecache_pages(); |
2f7f24ec | 914 | |
2d1d43f6 | 915 | writeback_set_ratelimit(); |
7b78d335 YG |
916 | |
917 | if (onlined_pages) | |
918 | memory_notify(MEM_ONLINE, &arg); | |
381eab4a | 919 | mem_hotplug_done(); |
30467e0b | 920 | return 0; |
e33e33b4 CY |
921 | |
922 | failed_addition: | |
923 | pr_debug("online_pages [mem %#010llx-%#010llx] failed\n", | |
924 | (unsigned long long) pfn << PAGE_SHIFT, | |
925 | (((unsigned long long) pfn + nr_pages) << PAGE_SHIFT) - 1); | |
926 | memory_notify(MEM_CANCEL_ONLINE, &arg); | |
381eab4a | 927 | mem_hotplug_done(); |
e33e33b4 | 928 | return ret; |
3947be19 | 929 | } |
53947027 | 930 | #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */ |
bc02af93 | 931 | |
0bd85420 TC |
932 | static void reset_node_present_pages(pg_data_t *pgdat) |
933 | { | |
934 | struct zone *z; | |
935 | ||
936 | for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++) | |
937 | z->present_pages = 0; | |
938 | ||
939 | pgdat->node_present_pages = 0; | |
940 | } | |
941 | ||
e1319331 HS |
942 | /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */ |
943 | static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start) | |
9af3c2de YG |
944 | { |
945 | struct pglist_data *pgdat; | |
c8e861a5 | 946 | unsigned long start_pfn = PFN_DOWN(start); |
9af3c2de | 947 | |
a1e565aa TC |
948 | pgdat = NODE_DATA(nid); |
949 | if (!pgdat) { | |
950 | pgdat = arch_alloc_nodedata(nid); | |
951 | if (!pgdat) | |
952 | return NULL; | |
9af3c2de | 953 | |
a1e565aa | 954 | arch_refresh_nodedata(nid, pgdat); |
b0dc3a34 | 955 | } else { |
e716f2eb MG |
956 | /* |
957 | * Reset the nr_zones, order and classzone_idx before reuse. | |
958 | * Note that kswapd will init kswapd_classzone_idx properly | |
959 | * when it starts in the near future. | |
960 | */ | |
b0dc3a34 | 961 | pgdat->nr_zones = 0; |
38087d9b MG |
962 | pgdat->kswapd_order = 0; |
963 | pgdat->kswapd_classzone_idx = 0; | |
a1e565aa | 964 | } |
9af3c2de YG |
965 | |
966 | /* we can use NODE_DATA(nid) from here */ | |
967 | ||
03e85f9d OS |
968 | pgdat->node_id = nid; |
969 | pgdat->node_start_pfn = start_pfn; | |
970 | ||
9af3c2de | 971 | /* init node's zones as empty zones, we don't have any present pages.*/ |
03e85f9d | 972 | free_area_init_core_hotplug(nid); |
5830169f | 973 | pgdat->per_cpu_nodestats = alloc_percpu(struct per_cpu_nodestat); |
9af3c2de | 974 | |
959ecc48 KH |
975 | /* |
976 | * The node we allocated has no zone fallback lists. For avoiding | |
977 | * to access not-initialized zonelist, build here. | |
978 | */ | |
72675e13 | 979 | build_all_zonelists(pgdat); |
959ecc48 | 980 | |
0bd85420 TC |
981 | /* |
982 | * When memory is hot-added, all the memory is in offline state. So | |
983 | * clear all zones' present_pages because they will be updated in | |
984 | * online_pages() and offline_pages(). | |
985 | */ | |
03e85f9d | 986 | reset_node_managed_pages(pgdat); |
0bd85420 TC |
987 | reset_node_present_pages(pgdat); |
988 | ||
9af3c2de YG |
989 | return pgdat; |
990 | } | |
991 | ||
b9ff0360 | 992 | static void rollback_node_hotadd(int nid) |
9af3c2de | 993 | { |
b9ff0360 OS |
994 | pg_data_t *pgdat = NODE_DATA(nid); |
995 | ||
9af3c2de | 996 | arch_refresh_nodedata(nid, NULL); |
5830169f | 997 | free_percpu(pgdat->per_cpu_nodestats); |
9af3c2de YG |
998 | arch_free_nodedata(pgdat); |
999 | return; | |
1000 | } | |
1001 | ||
0a547039 | 1002 | |
01b0f197 TK |
1003 | /** |
1004 | * try_online_node - online a node if offlined | |
e8b098fc | 1005 | * @nid: the node ID |
b9ff0360 OS |
1006 | * @start: start addr of the node |
1007 | * @set_node_online: Whether we want to online the node | |
cf23422b | 1008 | * called by cpu_up() to online a node without onlined memory. |
b9ff0360 OS |
1009 | * |
1010 | * Returns: | |
1011 | * 1 -> a new node has been allocated | |
1012 | * 0 -> the node is already online | |
1013 | * -ENOMEM -> the node could not be allocated | |
cf23422b | 1014 | */ |
b9ff0360 | 1015 | static int __try_online_node(int nid, u64 start, bool set_node_online) |
cf23422b | 1016 | { |
b9ff0360 OS |
1017 | pg_data_t *pgdat; |
1018 | int ret = 1; | |
cf23422b | 1019 | |
01b0f197 TK |
1020 | if (node_online(nid)) |
1021 | return 0; | |
1022 | ||
b9ff0360 | 1023 | pgdat = hotadd_new_pgdat(nid, start); |
7553e8f2 | 1024 | if (!pgdat) { |
01b0f197 | 1025 | pr_err("Cannot online node %d due to NULL pgdat\n", nid); |
cf23422b | 1026 | ret = -ENOMEM; |
1027 | goto out; | |
1028 | } | |
b9ff0360 OS |
1029 | |
1030 | if (set_node_online) { | |
1031 | node_set_online(nid); | |
1032 | ret = register_one_node(nid); | |
1033 | BUG_ON(ret); | |
1034 | } | |
cf23422b | 1035 | out: |
b9ff0360 OS |
1036 | return ret; |
1037 | } | |
1038 | ||
1039 | /* | |
1040 | * Users of this function always want to online/register the node | |
1041 | */ | |
1042 | int try_online_node(int nid) | |
1043 | { | |
1044 | int ret; | |
1045 | ||
1046 | mem_hotplug_begin(); | |
1047 | ret = __try_online_node(nid, 0, true); | |
bfc8c901 | 1048 | mem_hotplug_done(); |
cf23422b | 1049 | return ret; |
1050 | } | |
1051 | ||
27356f54 TK |
1052 | static int check_hotplug_memory_range(u64 start, u64 size) |
1053 | { | |
ba325585 PT |
1054 | unsigned long block_sz = memory_block_size_bytes(); |
1055 | u64 block_nr_pages = block_sz >> PAGE_SHIFT; | |
27356f54 | 1056 | u64 nr_pages = size >> PAGE_SHIFT; |
ba325585 | 1057 | u64 start_pfn = PFN_DOWN(start); |
27356f54 | 1058 | |
ba325585 PT |
1059 | /* memory range must be block size aligned */ |
1060 | if (!nr_pages || !IS_ALIGNED(start_pfn, block_nr_pages) || | |
1061 | !IS_ALIGNED(nr_pages, block_nr_pages)) { | |
1062 | pr_err("Block size [%#lx] unaligned hotplug range: start %#llx, size %#llx", | |
1063 | block_sz, start, size); | |
27356f54 TK |
1064 | return -EINVAL; |
1065 | } | |
1066 | ||
1067 | return 0; | |
1068 | } | |
1069 | ||
31bc3858 VK |
1070 | static int online_memory_block(struct memory_block *mem, void *arg) |
1071 | { | |
dc18d706 | 1072 | return device_online(&mem->dev); |
31bc3858 VK |
1073 | } |
1074 | ||
8df1d0e4 DH |
1075 | /* |
1076 | * NOTE: The caller must call lock_device_hotplug() to serialize hotplug | |
1077 | * and online/offline operations (triggered e.g. by sysfs). | |
1078 | * | |
1079 | * we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG | |
1080 | */ | |
31bc3858 | 1081 | int __ref add_memory_resource(int nid, struct resource *res, bool online) |
bc02af93 | 1082 | { |
62cedb9f | 1083 | u64 start, size; |
b9ff0360 | 1084 | bool new_node = false; |
bc02af93 YG |
1085 | int ret; |
1086 | ||
62cedb9f DV |
1087 | start = res->start; |
1088 | size = resource_size(res); | |
1089 | ||
27356f54 TK |
1090 | ret = check_hotplug_memory_range(start, size); |
1091 | if (ret) | |
1092 | return ret; | |
1093 | ||
bfc8c901 | 1094 | mem_hotplug_begin(); |
ac13c462 | 1095 | |
7f36e3e5 TC |
1096 | /* |
1097 | * Add new range to memblock so that when hotadd_new_pgdat() is called | |
1098 | * to allocate new pgdat, get_pfn_range_for_nid() will be able to find | |
1099 | * this new range and calculate total pages correctly. The range will | |
1100 | * be removed at hot-remove time. | |
1101 | */ | |
1102 | memblock_add_node(start, size, nid); | |
1103 | ||
b9ff0360 OS |
1104 | ret = __try_online_node(nid, start, false); |
1105 | if (ret < 0) | |
1106 | goto error; | |
1107 | new_node = ret; | |
9af3c2de | 1108 | |
bc02af93 | 1109 | /* call arch's memory hotadd */ |
24e6d5a5 | 1110 | ret = arch_add_memory(nid, start, size, NULL, true); |
9af3c2de YG |
1111 | if (ret < 0) |
1112 | goto error; | |
1113 | ||
a1e565aa | 1114 | if (new_node) { |
d5b6f6a3 | 1115 | /* If sysfs file of new node can't be created, cpu on the node |
0fc44159 YG |
1116 | * can't be hot-added. There is no rollback way now. |
1117 | * So, check by BUG_ON() to catch it reluctantly.. | |
d5b6f6a3 | 1118 | * We online node here. We can't roll back from here. |
0fc44159 | 1119 | */ |
d5b6f6a3 OS |
1120 | node_set_online(nid); |
1121 | ret = __register_one_node(nid); | |
0fc44159 YG |
1122 | BUG_ON(ret); |
1123 | } | |
1124 | ||
d5b6f6a3 | 1125 | /* link memory sections under this node.*/ |
4fbce633 | 1126 | ret = link_mem_sections(nid, PFN_DOWN(start), PFN_UP(start + size - 1)); |
d5b6f6a3 OS |
1127 | BUG_ON(ret); |
1128 | ||
d96ae530 AM |
1129 | /* create new memmap entry */ |
1130 | firmware_map_add_hotplug(start, start + size, "System RAM"); | |
1131 | ||
381eab4a DH |
1132 | /* device_online() will take the lock when calling online_pages() */ |
1133 | mem_hotplug_done(); | |
1134 | ||
31bc3858 VK |
1135 | /* online pages if requested */ |
1136 | if (online) | |
1137 | walk_memory_range(PFN_DOWN(start), PFN_UP(start + size - 1), | |
1138 | NULL, online_memory_block); | |
1139 | ||
381eab4a | 1140 | return ret; |
9af3c2de YG |
1141 | error: |
1142 | /* rollback pgdat allocation and others */ | |
b9ff0360 OS |
1143 | if (new_node) |
1144 | rollback_node_hotadd(nid); | |
7f36e3e5 | 1145 | memblock_remove(start, size); |
bfc8c901 | 1146 | mem_hotplug_done(); |
bc02af93 YG |
1147 | return ret; |
1148 | } | |
62cedb9f | 1149 | |
8df1d0e4 DH |
1150 | /* requires device_hotplug_lock, see add_memory_resource() */ |
1151 | int __ref __add_memory(int nid, u64 start, u64 size) | |
62cedb9f DV |
1152 | { |
1153 | struct resource *res; | |
1154 | int ret; | |
1155 | ||
1156 | res = register_memory_resource(start, size); | |
6f754ba4 VK |
1157 | if (IS_ERR(res)) |
1158 | return PTR_ERR(res); | |
62cedb9f | 1159 | |
31bc3858 | 1160 | ret = add_memory_resource(nid, res, memhp_auto_online); |
62cedb9f DV |
1161 | if (ret < 0) |
1162 | release_memory_resource(res); | |
1163 | return ret; | |
1164 | } | |
8df1d0e4 DH |
1165 | |
1166 | int add_memory(int nid, u64 start, u64 size) | |
1167 | { | |
1168 | int rc; | |
1169 | ||
1170 | lock_device_hotplug(); | |
1171 | rc = __add_memory(nid, start, size); | |
1172 | unlock_device_hotplug(); | |
1173 | ||
1174 | return rc; | |
1175 | } | |
bc02af93 | 1176 | EXPORT_SYMBOL_GPL(add_memory); |
0c0e6195 KH |
1177 | |
1178 | #ifdef CONFIG_MEMORY_HOTREMOVE | |
5c755e9f BP |
1179 | /* |
1180 | * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy | |
1181 | * set and the size of the free page is given by page_order(). Using this, | |
1182 | * the function determines if the pageblock contains only free pages. | |
1183 | * Due to buddy contraints, a free page at least the size of a pageblock will | |
1184 | * be located at the start of the pageblock | |
1185 | */ | |
1186 | static inline int pageblock_free(struct page *page) | |
1187 | { | |
1188 | return PageBuddy(page) && page_order(page) >= pageblock_order; | |
1189 | } | |
1190 | ||
1191 | /* Return the start of the next active pageblock after a given page */ | |
1192 | static struct page *next_active_pageblock(struct page *page) | |
1193 | { | |
5c755e9f BP |
1194 | /* Ensure the starting page is pageblock-aligned */ |
1195 | BUG_ON(page_to_pfn(page) & (pageblock_nr_pages - 1)); | |
1196 | ||
5c755e9f | 1197 | /* If the entire pageblock is free, move to the end of free page */ |
0dcc48c1 KH |
1198 | if (pageblock_free(page)) { |
1199 | int order; | |
1200 | /* be careful. we don't have locks, page_order can be changed.*/ | |
1201 | order = page_order(page); | |
1202 | if ((order < MAX_ORDER) && (order >= pageblock_order)) | |
1203 | return page + (1 << order); | |
1204 | } | |
5c755e9f | 1205 | |
0dcc48c1 | 1206 | return page + pageblock_nr_pages; |
5c755e9f BP |
1207 | } |
1208 | ||
fb52bbae MM |
1209 | static bool is_pageblock_removable_nolock(struct page *page) |
1210 | { | |
1211 | struct zone *zone; | |
1212 | unsigned long pfn; | |
1213 | ||
1214 | /* | |
1215 | * We have to be careful here because we are iterating over memory | |
1216 | * sections which are not zone aware so we might end up outside of | |
1217 | * the zone but still within the section. | |
1218 | * We have to take care about the node as well. If the node is offline | |
1219 | * its NODE_DATA will be NULL - see page_zone. | |
1220 | */ | |
1221 | if (!node_online(page_to_nid(page))) | |
1222 | return false; | |
1223 | ||
1224 | zone = page_zone(page); | |
1225 | pfn = page_to_pfn(page); | |
1226 | if (!zone_spans_pfn(zone, pfn)) | |
1227 | return false; | |
1228 | ||
1229 | return !has_unmovable_pages(zone, page, 0, MIGRATE_MOVABLE, true); | |
1230 | } | |
1231 | ||
5c755e9f | 1232 | /* Checks if this range of memory is likely to be hot-removable. */ |
c98940f6 | 1233 | bool is_mem_section_removable(unsigned long start_pfn, unsigned long nr_pages) |
5c755e9f | 1234 | { |
5c755e9f BP |
1235 | struct page *page = pfn_to_page(start_pfn); |
1236 | struct page *end_page = page + nr_pages; | |
1237 | ||
1238 | /* Check the starting page of each pageblock within the range */ | |
1239 | for (; page < end_page; page = next_active_pageblock(page)) { | |
49ac8255 | 1240 | if (!is_pageblock_removable_nolock(page)) |
c98940f6 | 1241 | return false; |
49ac8255 | 1242 | cond_resched(); |
5c755e9f BP |
1243 | } |
1244 | ||
1245 | /* All pageblocks in the memory block are likely to be hot-removable */ | |
c98940f6 | 1246 | return true; |
5c755e9f BP |
1247 | } |
1248 | ||
0c0e6195 | 1249 | /* |
deb88a2a | 1250 | * Confirm all pages in a range [start, end) belong to the same zone. |
a96dfddb | 1251 | * When true, return its valid [start, end). |
0c0e6195 | 1252 | */ |
a96dfddb TK |
1253 | int test_pages_in_a_zone(unsigned long start_pfn, unsigned long end_pfn, |
1254 | unsigned long *valid_start, unsigned long *valid_end) | |
0c0e6195 | 1255 | { |
5f0f2887 | 1256 | unsigned long pfn, sec_end_pfn; |
a96dfddb | 1257 | unsigned long start, end; |
0c0e6195 KH |
1258 | struct zone *zone = NULL; |
1259 | struct page *page; | |
1260 | int i; | |
deb88a2a | 1261 | for (pfn = start_pfn, sec_end_pfn = SECTION_ALIGN_UP(start_pfn + 1); |
0c0e6195 | 1262 | pfn < end_pfn; |
deb88a2a | 1263 | pfn = sec_end_pfn, sec_end_pfn += PAGES_PER_SECTION) { |
5f0f2887 AB |
1264 | /* Make sure the memory section is present first */ |
1265 | if (!present_section_nr(pfn_to_section_nr(pfn))) | |
0c0e6195 | 1266 | continue; |
5f0f2887 AB |
1267 | for (; pfn < sec_end_pfn && pfn < end_pfn; |
1268 | pfn += MAX_ORDER_NR_PAGES) { | |
1269 | i = 0; | |
1270 | /* This is just a CONFIG_HOLES_IN_ZONE check.*/ | |
1271 | while ((i < MAX_ORDER_NR_PAGES) && | |
1272 | !pfn_valid_within(pfn + i)) | |
1273 | i++; | |
d6d8c8a4 | 1274 | if (i == MAX_ORDER_NR_PAGES || pfn + i >= end_pfn) |
5f0f2887 AB |
1275 | continue; |
1276 | page = pfn_to_page(pfn + i); | |
1277 | if (zone && page_zone(page) != zone) | |
1278 | return 0; | |
a96dfddb TK |
1279 | if (!zone) |
1280 | start = pfn + i; | |
5f0f2887 | 1281 | zone = page_zone(page); |
a96dfddb | 1282 | end = pfn + MAX_ORDER_NR_PAGES; |
5f0f2887 | 1283 | } |
0c0e6195 | 1284 | } |
deb88a2a | 1285 | |
a96dfddb TK |
1286 | if (zone) { |
1287 | *valid_start = start; | |
d6d8c8a4 | 1288 | *valid_end = min(end, end_pfn); |
deb88a2a | 1289 | return 1; |
a96dfddb | 1290 | } else { |
deb88a2a | 1291 | return 0; |
a96dfddb | 1292 | } |
0c0e6195 KH |
1293 | } |
1294 | ||
1295 | /* | |
0efadf48 YX |
1296 | * Scan pfn range [start,end) to find movable/migratable pages (LRU pages, |
1297 | * non-lru movable pages and hugepages). We scan pfn because it's much | |
1298 | * easier than scanning over linked list. This function returns the pfn | |
1299 | * of the first found movable page if it's found, otherwise 0. | |
0c0e6195 | 1300 | */ |
c8721bbb | 1301 | static unsigned long scan_movable_pages(unsigned long start, unsigned long end) |
0c0e6195 KH |
1302 | { |
1303 | unsigned long pfn; | |
1304 | struct page *page; | |
1305 | for (pfn = start; pfn < end; pfn++) { | |
1306 | if (pfn_valid(pfn)) { | |
1307 | page = pfn_to_page(pfn); | |
1308 | if (PageLRU(page)) | |
1309 | return pfn; | |
0efadf48 YX |
1310 | if (__PageMovable(page)) |
1311 | return pfn; | |
c8721bbb | 1312 | if (PageHuge(page)) { |
464c7ffb AK |
1313 | if (hugepage_migration_supported(page_hstate(page)) && |
1314 | page_huge_active(page)) | |
c8721bbb NH |
1315 | return pfn; |
1316 | else | |
1317 | pfn = round_up(pfn + 1, | |
1318 | 1 << compound_order(page)) - 1; | |
1319 | } | |
0c0e6195 KH |
1320 | } |
1321 | } | |
1322 | return 0; | |
1323 | } | |
1324 | ||
666feb21 | 1325 | static struct page *new_node_page(struct page *page, unsigned long private) |
394e31d2 | 1326 | { |
394e31d2 | 1327 | int nid = page_to_nid(page); |
231e97e2 | 1328 | nodemask_t nmask = node_states[N_MEMORY]; |
7f252f27 MH |
1329 | |
1330 | /* | |
1331 | * try to allocate from a different node but reuse this node if there | |
1332 | * are no other online nodes to be used (e.g. we are offlining a part | |
1333 | * of the only existing node) | |
1334 | */ | |
1335 | node_clear(nid, nmask); | |
1336 | if (nodes_empty(nmask)) | |
1337 | node_set(nid, nmask); | |
394e31d2 | 1338 | |
8b913238 | 1339 | return new_page_nodemask(page, nid, &nmask); |
394e31d2 XQ |
1340 | } |
1341 | ||
0c0e6195 KH |
1342 | #define NR_OFFLINE_AT_ONCE_PAGES (256) |
1343 | static int | |
1344 | do_migrate_range(unsigned long start_pfn, unsigned long end_pfn) | |
1345 | { | |
1346 | unsigned long pfn; | |
1347 | struct page *page; | |
1348 | int move_pages = NR_OFFLINE_AT_ONCE_PAGES; | |
1349 | int not_managed = 0; | |
1350 | int ret = 0; | |
1351 | LIST_HEAD(source); | |
1352 | ||
1353 | for (pfn = start_pfn; pfn < end_pfn && move_pages > 0; pfn++) { | |
1354 | if (!pfn_valid(pfn)) | |
1355 | continue; | |
1356 | page = pfn_to_page(pfn); | |
c8721bbb NH |
1357 | |
1358 | if (PageHuge(page)) { | |
1359 | struct page *head = compound_head(page); | |
1360 | pfn = page_to_pfn(head) + (1<<compound_order(head)) - 1; | |
1361 | if (compound_order(head) > PFN_SECTION_SHIFT) { | |
1362 | ret = -EBUSY; | |
1363 | break; | |
1364 | } | |
1365 | if (isolate_huge_page(page, &source)) | |
1366 | move_pages -= 1 << compound_order(head); | |
1367 | continue; | |
94723aaf | 1368 | } else if (PageTransHuge(page)) |
8135d892 NH |
1369 | pfn = page_to_pfn(compound_head(page)) |
1370 | + hpage_nr_pages(page) - 1; | |
c8721bbb | 1371 | |
700c2a46 | 1372 | if (!get_page_unless_zero(page)) |
0c0e6195 KH |
1373 | continue; |
1374 | /* | |
0efadf48 YX |
1375 | * We can skip free pages. And we can deal with pages on |
1376 | * LRU and non-lru movable pages. | |
0c0e6195 | 1377 | */ |
0efadf48 YX |
1378 | if (PageLRU(page)) |
1379 | ret = isolate_lru_page(page); | |
1380 | else | |
1381 | ret = isolate_movable_page(page, ISOLATE_UNEVICTABLE); | |
0c0e6195 | 1382 | if (!ret) { /* Success */ |
700c2a46 | 1383 | put_page(page); |
62695a84 | 1384 | list_add_tail(&page->lru, &source); |
0c0e6195 | 1385 | move_pages--; |
0efadf48 YX |
1386 | if (!__PageMovable(page)) |
1387 | inc_node_page_state(page, NR_ISOLATED_ANON + | |
1388 | page_is_file_cache(page)); | |
6d9c285a | 1389 | |
0c0e6195 | 1390 | } else { |
0c0e6195 | 1391 | #ifdef CONFIG_DEBUG_VM |
0efadf48 YX |
1392 | pr_alert("failed to isolate pfn %lx\n", pfn); |
1393 | dump_page(page, "isolation failed"); | |
0c0e6195 | 1394 | #endif |
700c2a46 | 1395 | put_page(page); |
25985edc | 1396 | /* Because we don't have big zone->lock. we should |
809c4449 BL |
1397 | check this again here. */ |
1398 | if (page_count(page)) { | |
1399 | not_managed++; | |
f3ab2636 | 1400 | ret = -EBUSY; |
809c4449 BL |
1401 | break; |
1402 | } | |
0c0e6195 KH |
1403 | } |
1404 | } | |
f3ab2636 BL |
1405 | if (!list_empty(&source)) { |
1406 | if (not_managed) { | |
c8721bbb | 1407 | putback_movable_pages(&source); |
f3ab2636 BL |
1408 | goto out; |
1409 | } | |
74c08f98 | 1410 | |
394e31d2 XQ |
1411 | /* Allocate a new page from the nearest neighbor node */ |
1412 | ret = migrate_pages(&source, new_node_page, NULL, 0, | |
9c620e2b | 1413 | MIGRATE_SYNC, MR_MEMORY_HOTPLUG); |
f3ab2636 | 1414 | if (ret) |
c8721bbb | 1415 | putback_movable_pages(&source); |
0c0e6195 | 1416 | } |
0c0e6195 KH |
1417 | out: |
1418 | return ret; | |
1419 | } | |
1420 | ||
1421 | /* | |
1422 | * remove from free_area[] and mark all as Reserved. | |
1423 | */ | |
1424 | static int | |
1425 | offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages, | |
1426 | void *data) | |
1427 | { | |
1428 | __offline_isolated_pages(start, start + nr_pages); | |
1429 | return 0; | |
1430 | } | |
1431 | ||
1432 | static void | |
1433 | offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn) | |
1434 | { | |
908eedc6 | 1435 | walk_system_ram_range(start_pfn, end_pfn - start_pfn, NULL, |
0c0e6195 KH |
1436 | offline_isolated_pages_cb); |
1437 | } | |
1438 | ||
1439 | /* | |
1440 | * Check all pages in range, recoreded as memory resource, are isolated. | |
1441 | */ | |
1442 | static int | |
1443 | check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages, | |
1444 | void *data) | |
1445 | { | |
1446 | int ret; | |
1447 | long offlined = *(long *)data; | |
b023f468 | 1448 | ret = test_pages_isolated(start_pfn, start_pfn + nr_pages, true); |
0c0e6195 KH |
1449 | offlined = nr_pages; |
1450 | if (!ret) | |
1451 | *(long *)data += offlined; | |
1452 | return ret; | |
1453 | } | |
1454 | ||
1455 | static long | |
1456 | check_pages_isolated(unsigned long start_pfn, unsigned long end_pfn) | |
1457 | { | |
1458 | long offlined = 0; | |
1459 | int ret; | |
1460 | ||
908eedc6 | 1461 | ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn, &offlined, |
0c0e6195 KH |
1462 | check_pages_isolated_cb); |
1463 | if (ret < 0) | |
1464 | offlined = (long)ret; | |
1465 | return offlined; | |
1466 | } | |
1467 | ||
c5320926 TC |
1468 | static int __init cmdline_parse_movable_node(char *p) |
1469 | { | |
4932381e | 1470 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
55ac590c | 1471 | movable_node_enabled = true; |
4932381e MH |
1472 | #else |
1473 | pr_warn("movable_node parameter depends on CONFIG_HAVE_MEMBLOCK_NODE_MAP to work properly\n"); | |
1474 | #endif | |
c5320926 TC |
1475 | return 0; |
1476 | } | |
1477 | early_param("movable_node", cmdline_parse_movable_node); | |
1478 | ||
d9713679 LJ |
1479 | /* check which state of node_states will be changed when offline memory */ |
1480 | static void node_states_check_changes_offline(unsigned long nr_pages, | |
1481 | struct zone *zone, struct memory_notify *arg) | |
1482 | { | |
1483 | struct pglist_data *pgdat = zone->zone_pgdat; | |
1484 | unsigned long present_pages = 0; | |
86b27bea | 1485 | enum zone_type zt; |
d9713679 | 1486 | |
86b27bea OS |
1487 | arg->status_change_nid = -1; |
1488 | arg->status_change_nid_normal = -1; | |
1489 | arg->status_change_nid_high = -1; | |
d9713679 LJ |
1490 | |
1491 | /* | |
86b27bea OS |
1492 | * Check whether node_states[N_NORMAL_MEMORY] will be changed. |
1493 | * If the memory to be offline is within the range | |
1494 | * [0..ZONE_NORMAL], and it is the last present memory there, | |
1495 | * the zones in that range will become empty after the offlining, | |
1496 | * thus we can determine that we need to clear the node from | |
1497 | * node_states[N_NORMAL_MEMORY]. | |
d9713679 | 1498 | */ |
86b27bea | 1499 | for (zt = 0; zt <= ZONE_NORMAL; zt++) |
d9713679 | 1500 | present_pages += pgdat->node_zones[zt].present_pages; |
86b27bea | 1501 | if (zone_idx(zone) <= ZONE_NORMAL && nr_pages >= present_pages) |
d9713679 | 1502 | arg->status_change_nid_normal = zone_to_nid(zone); |
d9713679 | 1503 | |
6715ddf9 LJ |
1504 | #ifdef CONFIG_HIGHMEM |
1505 | /* | |
86b27bea OS |
1506 | * node_states[N_HIGH_MEMORY] contains nodes which |
1507 | * have normal memory or high memory. | |
1508 | * Here we add the present_pages belonging to ZONE_HIGHMEM. | |
1509 | * If the zone is within the range of [0..ZONE_HIGHMEM), and | |
1510 | * we determine that the zones in that range become empty, | |
1511 | * we need to clear the node for N_HIGH_MEMORY. | |
6715ddf9 | 1512 | */ |
86b27bea OS |
1513 | present_pages += pgdat->node_zones[ZONE_HIGHMEM].present_pages; |
1514 | if (zone_idx(zone) <= ZONE_HIGHMEM && nr_pages >= present_pages) | |
6715ddf9 | 1515 | arg->status_change_nid_high = zone_to_nid(zone); |
6715ddf9 LJ |
1516 | #endif |
1517 | ||
d9713679 | 1518 | /* |
86b27bea OS |
1519 | * We have accounted the pages from [0..ZONE_NORMAL), and |
1520 | * in case of CONFIG_HIGHMEM the pages from ZONE_HIGHMEM | |
1521 | * as well. | |
1522 | * Here we count the possible pages from ZONE_MOVABLE. | |
1523 | * If after having accounted all the pages, we see that the nr_pages | |
1524 | * to be offlined is over or equal to the accounted pages, | |
1525 | * we know that the node will become empty, and so, we can clear | |
1526 | * it for N_MEMORY as well. | |
d9713679 | 1527 | */ |
86b27bea | 1528 | present_pages += pgdat->node_zones[ZONE_MOVABLE].present_pages; |
d9713679 | 1529 | |
d9713679 LJ |
1530 | if (nr_pages >= present_pages) |
1531 | arg->status_change_nid = zone_to_nid(zone); | |
d9713679 LJ |
1532 | } |
1533 | ||
1534 | static void node_states_clear_node(int node, struct memory_notify *arg) | |
1535 | { | |
1536 | if (arg->status_change_nid_normal >= 0) | |
1537 | node_clear_state(node, N_NORMAL_MEMORY); | |
1538 | ||
cf01f6f5 | 1539 | if (arg->status_change_nid_high >= 0) |
d9713679 | 1540 | node_clear_state(node, N_HIGH_MEMORY); |
6715ddf9 | 1541 | |
cf01f6f5 | 1542 | if (arg->status_change_nid >= 0) |
6715ddf9 | 1543 | node_clear_state(node, N_MEMORY); |
d9713679 LJ |
1544 | } |
1545 | ||
a16cee10 | 1546 | static int __ref __offline_pages(unsigned long start_pfn, |
ecde0f3e | 1547 | unsigned long end_pfn) |
0c0e6195 | 1548 | { |
ecde0f3e | 1549 | unsigned long pfn, nr_pages; |
0c0e6195 | 1550 | long offlined_pages; |
72b39cfc | 1551 | int ret, node; |
d702909f | 1552 | unsigned long flags; |
a96dfddb | 1553 | unsigned long valid_start, valid_end; |
0c0e6195 | 1554 | struct zone *zone; |
7b78d335 | 1555 | struct memory_notify arg; |
0c0e6195 | 1556 | |
0c0e6195 KH |
1557 | /* at least, alignment against pageblock is necessary */ |
1558 | if (!IS_ALIGNED(start_pfn, pageblock_nr_pages)) | |
1559 | return -EINVAL; | |
1560 | if (!IS_ALIGNED(end_pfn, pageblock_nr_pages)) | |
1561 | return -EINVAL; | |
381eab4a DH |
1562 | |
1563 | mem_hotplug_begin(); | |
1564 | ||
0c0e6195 KH |
1565 | /* This makes hotplug much easier...and readable. |
1566 | we assume this for now. .*/ | |
381eab4a DH |
1567 | if (!test_pages_in_a_zone(start_pfn, end_pfn, &valid_start, |
1568 | &valid_end)) { | |
1569 | mem_hotplug_done(); | |
0c0e6195 | 1570 | return -EINVAL; |
381eab4a | 1571 | } |
7b78d335 | 1572 | |
a96dfddb | 1573 | zone = page_zone(pfn_to_page(valid_start)); |
7b78d335 YG |
1574 | node = zone_to_nid(zone); |
1575 | nr_pages = end_pfn - start_pfn; | |
1576 | ||
0c0e6195 | 1577 | /* set above range as isolated */ |
b023f468 WC |
1578 | ret = start_isolate_page_range(start_pfn, end_pfn, |
1579 | MIGRATE_MOVABLE, true); | |
381eab4a DH |
1580 | if (ret) { |
1581 | mem_hotplug_done(); | |
30467e0b | 1582 | return ret; |
381eab4a | 1583 | } |
7b78d335 YG |
1584 | |
1585 | arg.start_pfn = start_pfn; | |
1586 | arg.nr_pages = nr_pages; | |
d9713679 | 1587 | node_states_check_changes_offline(nr_pages, zone, &arg); |
7b78d335 YG |
1588 | |
1589 | ret = memory_notify(MEM_GOING_OFFLINE, &arg); | |
1590 | ret = notifier_to_errno(ret); | |
1591 | if (ret) | |
1592 | goto failed_removal; | |
1593 | ||
0c0e6195 | 1594 | pfn = start_pfn; |
0c0e6195 KH |
1595 | repeat: |
1596 | /* start memory hot removal */ | |
0c0e6195 KH |
1597 | ret = -EINTR; |
1598 | if (signal_pending(current)) | |
1599 | goto failed_removal; | |
72b39cfc MH |
1600 | |
1601 | cond_resched(); | |
9852a721 | 1602 | lru_add_drain_all(); |
72b39cfc | 1603 | drain_all_pages(zone); |
0c0e6195 | 1604 | |
c8721bbb NH |
1605 | pfn = scan_movable_pages(start_pfn, end_pfn); |
1606 | if (pfn) { /* We have movable pages */ | |
0c0e6195 | 1607 | ret = do_migrate_range(pfn, end_pfn); |
72b39cfc | 1608 | goto repeat; |
0c0e6195 | 1609 | } |
72b39cfc | 1610 | |
c8721bbb NH |
1611 | /* |
1612 | * dissolve free hugepages in the memory block before doing offlining | |
1613 | * actually in order to make hugetlbfs's object counting consistent. | |
1614 | */ | |
082d5b6b GS |
1615 | ret = dissolve_free_huge_pages(start_pfn, end_pfn); |
1616 | if (ret) | |
1617 | goto failed_removal; | |
0c0e6195 KH |
1618 | /* check again */ |
1619 | offlined_pages = check_pages_isolated(start_pfn, end_pfn); | |
72b39cfc MH |
1620 | if (offlined_pages < 0) |
1621 | goto repeat; | |
e33e33b4 | 1622 | pr_info("Offlined Pages %ld\n", offlined_pages); |
b3834be5 | 1623 | /* Ok, all of our target is isolated. |
0c0e6195 KH |
1624 | We cannot do rollback at this point. */ |
1625 | offline_isolated_pages(start_pfn, end_pfn); | |
dbc0e4ce | 1626 | /* reset pagetype flags and makes migrate type to be MOVABLE */ |
0815f3d8 | 1627 | undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE); |
0c0e6195 | 1628 | /* removal success */ |
3dcc0571 | 1629 | adjust_managed_page_count(pfn_to_page(start_pfn), -offlined_pages); |
0c0e6195 | 1630 | zone->present_pages -= offlined_pages; |
d702909f CS |
1631 | |
1632 | pgdat_resize_lock(zone->zone_pgdat, &flags); | |
0c0e6195 | 1633 | zone->zone_pgdat->node_present_pages -= offlined_pages; |
d702909f | 1634 | pgdat_resize_unlock(zone->zone_pgdat, &flags); |
7b78d335 | 1635 | |
1b79acc9 KM |
1636 | init_per_zone_wmark_min(); |
1637 | ||
1e8537ba | 1638 | if (!populated_zone(zone)) { |
340175b7 | 1639 | zone_pcp_reset(zone); |
72675e13 | 1640 | build_all_zonelists(NULL); |
1e8537ba XQ |
1641 | } else |
1642 | zone_pcp_update(zone); | |
340175b7 | 1643 | |
d9713679 | 1644 | node_states_clear_node(node, &arg); |
698b1b30 | 1645 | if (arg.status_change_nid >= 0) { |
8fe23e05 | 1646 | kswapd_stop(node); |
698b1b30 VB |
1647 | kcompactd_stop(node); |
1648 | } | |
bce7394a | 1649 | |
0c0e6195 KH |
1650 | vm_total_pages = nr_free_pagecache_pages(); |
1651 | writeback_set_ratelimit(); | |
7b78d335 YG |
1652 | |
1653 | memory_notify(MEM_OFFLINE, &arg); | |
381eab4a | 1654 | mem_hotplug_done(); |
0c0e6195 KH |
1655 | return 0; |
1656 | ||
1657 | failed_removal: | |
e33e33b4 CY |
1658 | pr_debug("memory offlining [mem %#010llx-%#010llx] failed\n", |
1659 | (unsigned long long) start_pfn << PAGE_SHIFT, | |
1660 | ((unsigned long long) end_pfn << PAGE_SHIFT) - 1); | |
7b78d335 | 1661 | memory_notify(MEM_CANCEL_OFFLINE, &arg); |
0c0e6195 | 1662 | /* pushback to free area */ |
0815f3d8 | 1663 | undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE); |
381eab4a | 1664 | mem_hotplug_done(); |
0c0e6195 KH |
1665 | return ret; |
1666 | } | |
71088785 | 1667 | |
a16cee10 WC |
1668 | int offline_pages(unsigned long start_pfn, unsigned long nr_pages) |
1669 | { | |
ecde0f3e | 1670 | return __offline_pages(start_pfn, start_pfn + nr_pages); |
a16cee10 | 1671 | } |
e2ff3940 | 1672 | #endif /* CONFIG_MEMORY_HOTREMOVE */ |
a16cee10 | 1673 | |
bbc76be6 WC |
1674 | /** |
1675 | * walk_memory_range - walks through all mem sections in [start_pfn, end_pfn) | |
1676 | * @start_pfn: start pfn of the memory range | |
e05c4bbf | 1677 | * @end_pfn: end pfn of the memory range |
bbc76be6 WC |
1678 | * @arg: argument passed to func |
1679 | * @func: callback for each memory section walked | |
1680 | * | |
1681 | * This function walks through all present mem sections in range | |
1682 | * [start_pfn, end_pfn) and call func on each mem section. | |
1683 | * | |
1684 | * Returns the return value of func. | |
1685 | */ | |
e2ff3940 | 1686 | int walk_memory_range(unsigned long start_pfn, unsigned long end_pfn, |
bbc76be6 | 1687 | void *arg, int (*func)(struct memory_block *, void *)) |
71088785 | 1688 | { |
e90bdb7f WC |
1689 | struct memory_block *mem = NULL; |
1690 | struct mem_section *section; | |
e90bdb7f WC |
1691 | unsigned long pfn, section_nr; |
1692 | int ret; | |
e90bdb7f WC |
1693 | |
1694 | for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) { | |
1695 | section_nr = pfn_to_section_nr(pfn); | |
1696 | if (!present_section_nr(section_nr)) | |
1697 | continue; | |
1698 | ||
1699 | section = __nr_to_section(section_nr); | |
1700 | /* same memblock? */ | |
1701 | if (mem) | |
1702 | if ((section_nr >= mem->start_section_nr) && | |
1703 | (section_nr <= mem->end_section_nr)) | |
1704 | continue; | |
1705 | ||
1706 | mem = find_memory_block_hinted(section, mem); | |
1707 | if (!mem) | |
1708 | continue; | |
1709 | ||
bbc76be6 | 1710 | ret = func(mem, arg); |
e90bdb7f | 1711 | if (ret) { |
bbc76be6 WC |
1712 | kobject_put(&mem->dev.kobj); |
1713 | return ret; | |
e90bdb7f WC |
1714 | } |
1715 | } | |
1716 | ||
1717 | if (mem) | |
1718 | kobject_put(&mem->dev.kobj); | |
1719 | ||
bbc76be6 WC |
1720 | return 0; |
1721 | } | |
1722 | ||
e2ff3940 | 1723 | #ifdef CONFIG_MEMORY_HOTREMOVE |
d6de9d53 | 1724 | static int check_memblock_offlined_cb(struct memory_block *mem, void *arg) |
bbc76be6 WC |
1725 | { |
1726 | int ret = !is_memblock_offlined(mem); | |
1727 | ||
349daa0f RD |
1728 | if (unlikely(ret)) { |
1729 | phys_addr_t beginpa, endpa; | |
1730 | ||
1731 | beginpa = PFN_PHYS(section_nr_to_pfn(mem->start_section_nr)); | |
1732 | endpa = PFN_PHYS(section_nr_to_pfn(mem->end_section_nr + 1))-1; | |
756a025f | 1733 | pr_warn("removing memory fails, because memory [%pa-%pa] is onlined\n", |
349daa0f RD |
1734 | &beginpa, &endpa); |
1735 | } | |
bbc76be6 WC |
1736 | |
1737 | return ret; | |
1738 | } | |
1739 | ||
0f1cfe9d | 1740 | static int check_cpu_on_node(pg_data_t *pgdat) |
60a5a19e | 1741 | { |
60a5a19e TC |
1742 | int cpu; |
1743 | ||
1744 | for_each_present_cpu(cpu) { | |
1745 | if (cpu_to_node(cpu) == pgdat->node_id) | |
1746 | /* | |
1747 | * the cpu on this node isn't removed, and we can't | |
1748 | * offline this node. | |
1749 | */ | |
1750 | return -EBUSY; | |
1751 | } | |
1752 | ||
1753 | return 0; | |
1754 | } | |
1755 | ||
0f1cfe9d | 1756 | static void unmap_cpu_on_node(pg_data_t *pgdat) |
e13fe869 WC |
1757 | { |
1758 | #ifdef CONFIG_ACPI_NUMA | |
e13fe869 WC |
1759 | int cpu; |
1760 | ||
1761 | for_each_possible_cpu(cpu) | |
1762 | if (cpu_to_node(cpu) == pgdat->node_id) | |
1763 | numa_clear_node(cpu); | |
1764 | #endif | |
1765 | } | |
1766 | ||
0f1cfe9d | 1767 | static int check_and_unmap_cpu_on_node(pg_data_t *pgdat) |
e13fe869 | 1768 | { |
0f1cfe9d | 1769 | int ret; |
e13fe869 | 1770 | |
0f1cfe9d | 1771 | ret = check_cpu_on_node(pgdat); |
e13fe869 WC |
1772 | if (ret) |
1773 | return ret; | |
1774 | ||
1775 | /* | |
1776 | * the node will be offlined when we come here, so we can clear | |
1777 | * the cpu_to_node() now. | |
1778 | */ | |
1779 | ||
0f1cfe9d | 1780 | unmap_cpu_on_node(pgdat); |
e13fe869 WC |
1781 | return 0; |
1782 | } | |
1783 | ||
0f1cfe9d TK |
1784 | /** |
1785 | * try_offline_node | |
e8b098fc | 1786 | * @nid: the node ID |
0f1cfe9d TK |
1787 | * |
1788 | * Offline a node if all memory sections and cpus of the node are removed. | |
1789 | * | |
1790 | * NOTE: The caller must call lock_device_hotplug() to serialize hotplug | |
1791 | * and online/offline operations before this call. | |
1792 | */ | |
90b30cdc | 1793 | void try_offline_node(int nid) |
60a5a19e | 1794 | { |
d822b86a WC |
1795 | pg_data_t *pgdat = NODE_DATA(nid); |
1796 | unsigned long start_pfn = pgdat->node_start_pfn; | |
1797 | unsigned long end_pfn = start_pfn + pgdat->node_spanned_pages; | |
60a5a19e TC |
1798 | unsigned long pfn; |
1799 | ||
1800 | for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) { | |
1801 | unsigned long section_nr = pfn_to_section_nr(pfn); | |
1802 | ||
1803 | if (!present_section_nr(section_nr)) | |
1804 | continue; | |
1805 | ||
1806 | if (pfn_to_nid(pfn) != nid) | |
1807 | continue; | |
1808 | ||
1809 | /* | |
1810 | * some memory sections of this node are not removed, and we | |
1811 | * can't offline node now. | |
1812 | */ | |
1813 | return; | |
1814 | } | |
1815 | ||
0f1cfe9d | 1816 | if (check_and_unmap_cpu_on_node(pgdat)) |
60a5a19e TC |
1817 | return; |
1818 | ||
1819 | /* | |
1820 | * all memory/cpu of this node are removed, we can offline this | |
1821 | * node now. | |
1822 | */ | |
1823 | node_set_offline(nid); | |
1824 | unregister_one_node(nid); | |
1825 | } | |
90b30cdc | 1826 | EXPORT_SYMBOL(try_offline_node); |
60a5a19e | 1827 | |
0f1cfe9d TK |
1828 | /** |
1829 | * remove_memory | |
e8b098fc MR |
1830 | * @nid: the node ID |
1831 | * @start: physical address of the region to remove | |
1832 | * @size: size of the region to remove | |
0f1cfe9d TK |
1833 | * |
1834 | * NOTE: The caller must call lock_device_hotplug() to serialize hotplug | |
1835 | * and online/offline operations before this call, as required by | |
1836 | * try_offline_node(). | |
1837 | */ | |
d15e5926 | 1838 | void __ref __remove_memory(int nid, u64 start, u64 size) |
bbc76be6 | 1839 | { |
242831eb | 1840 | int ret; |
993c1aad | 1841 | |
27356f54 TK |
1842 | BUG_ON(check_hotplug_memory_range(start, size)); |
1843 | ||
bfc8c901 | 1844 | mem_hotplug_begin(); |
6677e3ea YI |
1845 | |
1846 | /* | |
242831eb RW |
1847 | * All memory blocks must be offlined before removing memory. Check |
1848 | * whether all memory blocks in question are offline and trigger a BUG() | |
1849 | * if this is not the case. | |
6677e3ea | 1850 | */ |
242831eb | 1851 | ret = walk_memory_range(PFN_DOWN(start), PFN_UP(start + size - 1), NULL, |
d6de9d53 | 1852 | check_memblock_offlined_cb); |
bfc8c901 | 1853 | if (ret) |
242831eb | 1854 | BUG(); |
6677e3ea | 1855 | |
46c66c4b YI |
1856 | /* remove memmap entry */ |
1857 | firmware_map_remove(start, start + size, "System RAM"); | |
f9126ab9 XQ |
1858 | memblock_free(start, size); |
1859 | memblock_remove(start, size); | |
46c66c4b | 1860 | |
da024512 | 1861 | arch_remove_memory(start, size, NULL); |
24d335ca | 1862 | |
60a5a19e TC |
1863 | try_offline_node(nid); |
1864 | ||
bfc8c901 | 1865 | mem_hotplug_done(); |
71088785 | 1866 | } |
d15e5926 DH |
1867 | |
1868 | void remove_memory(int nid, u64 start, u64 size) | |
1869 | { | |
1870 | lock_device_hotplug(); | |
1871 | __remove_memory(nid, start, size); | |
1872 | unlock_device_hotplug(); | |
1873 | } | |
71088785 | 1874 | EXPORT_SYMBOL_GPL(remove_memory); |
aba6efc4 | 1875 | #endif /* CONFIG_MEMORY_HOTREMOVE */ |