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 | |
3f906ba2 | 55 | DEFINE_STATIC_PERCPU_RWSEM(mem_hotplug_lock); |
bfc8c901 | 56 | |
3f906ba2 TG |
57 | void get_online_mems(void) |
58 | { | |
59 | percpu_down_read(&mem_hotplug_lock); | |
60 | } | |
bfc8c901 | 61 | |
3f906ba2 TG |
62 | void put_online_mems(void) |
63 | { | |
64 | percpu_up_read(&mem_hotplug_lock); | |
65 | } | |
bfc8c901 | 66 | |
4932381e MH |
67 | bool movable_node_enabled = false; |
68 | ||
8604d9e5 | 69 | #ifndef CONFIG_MEMORY_HOTPLUG_DEFAULT_ONLINE |
31bc3858 | 70 | bool memhp_auto_online; |
8604d9e5 VK |
71 | #else |
72 | bool memhp_auto_online = true; | |
73 | #endif | |
31bc3858 VK |
74 | EXPORT_SYMBOL_GPL(memhp_auto_online); |
75 | ||
86dd995d VK |
76 | static int __init setup_memhp_default_state(char *str) |
77 | { | |
78 | if (!strcmp(str, "online")) | |
79 | memhp_auto_online = true; | |
80 | else if (!strcmp(str, "offline")) | |
81 | memhp_auto_online = false; | |
82 | ||
83 | return 1; | |
84 | } | |
85 | __setup("memhp_default_state=", setup_memhp_default_state); | |
86 | ||
30467e0b | 87 | void mem_hotplug_begin(void) |
20d6c96b | 88 | { |
3f906ba2 TG |
89 | cpus_read_lock(); |
90 | percpu_down_write(&mem_hotplug_lock); | |
20d6c96b KM |
91 | } |
92 | ||
30467e0b | 93 | void mem_hotplug_done(void) |
bfc8c901 | 94 | { |
3f906ba2 TG |
95 | percpu_up_write(&mem_hotplug_lock); |
96 | cpus_read_unlock(); | |
bfc8c901 | 97 | } |
20d6c96b | 98 | |
45e0b78b KM |
99 | /* add this memory to iomem resource */ |
100 | static struct resource *register_memory_resource(u64 start, u64 size) | |
101 | { | |
5042db43 | 102 | struct resource *res, *conflict; |
45e0b78b | 103 | res = kzalloc(sizeof(struct resource), GFP_KERNEL); |
6f754ba4 VK |
104 | if (!res) |
105 | return ERR_PTR(-ENOMEM); | |
45e0b78b KM |
106 | |
107 | res->name = "System RAM"; | |
108 | res->start = start; | |
109 | res->end = start + size - 1; | |
782b8664 | 110 | res->flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY; |
5042db43 JG |
111 | conflict = request_resource_conflict(&iomem_resource, res); |
112 | if (conflict) { | |
113 | if (conflict->desc == IORES_DESC_DEVICE_PRIVATE_MEMORY) { | |
114 | pr_debug("Device unaddressable memory block " | |
115 | "memory hotplug at %#010llx !\n", | |
116 | (unsigned long long)start); | |
117 | } | |
4996eed8 | 118 | pr_debug("System RAM resource %pR cannot be added\n", res); |
45e0b78b | 119 | kfree(res); |
6f754ba4 | 120 | return ERR_PTR(-EEXIST); |
45e0b78b KM |
121 | } |
122 | return res; | |
123 | } | |
124 | ||
125 | static void release_memory_resource(struct resource *res) | |
126 | { | |
127 | if (!res) | |
128 | return; | |
129 | release_resource(res); | |
130 | kfree(res); | |
131 | return; | |
132 | } | |
133 | ||
53947027 | 134 | #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE |
46723bfa YI |
135 | void get_page_bootmem(unsigned long info, struct page *page, |
136 | unsigned long type) | |
04753278 | 137 | { |
ddffe98d | 138 | page->freelist = (void *)type; |
04753278 YG |
139 | SetPagePrivate(page); |
140 | set_page_private(page, info); | |
fe896d18 | 141 | page_ref_inc(page); |
04753278 YG |
142 | } |
143 | ||
170a5a7e | 144 | void put_page_bootmem(struct page *page) |
04753278 | 145 | { |
5f24ce5f | 146 | unsigned long type; |
04753278 | 147 | |
ddffe98d | 148 | type = (unsigned long) page->freelist; |
5f24ce5f AA |
149 | BUG_ON(type < MEMORY_HOTPLUG_MIN_BOOTMEM_TYPE || |
150 | type > MEMORY_HOTPLUG_MAX_BOOTMEM_TYPE); | |
04753278 | 151 | |
fe896d18 | 152 | if (page_ref_dec_return(page) == 1) { |
ddffe98d | 153 | page->freelist = NULL; |
04753278 YG |
154 | ClearPagePrivate(page); |
155 | set_page_private(page, 0); | |
5f24ce5f | 156 | INIT_LIST_HEAD(&page->lru); |
170a5a7e | 157 | free_reserved_page(page); |
04753278 | 158 | } |
04753278 YG |
159 | } |
160 | ||
46723bfa YI |
161 | #ifdef CONFIG_HAVE_BOOTMEM_INFO_NODE |
162 | #ifndef CONFIG_SPARSEMEM_VMEMMAP | |
d92bc318 | 163 | static void register_page_bootmem_info_section(unsigned long start_pfn) |
04753278 YG |
164 | { |
165 | unsigned long *usemap, mapsize, section_nr, i; | |
166 | struct mem_section *ms; | |
167 | struct page *page, *memmap; | |
168 | ||
04753278 YG |
169 | section_nr = pfn_to_section_nr(start_pfn); |
170 | ms = __nr_to_section(section_nr); | |
171 | ||
172 | /* Get section's memmap address */ | |
173 | memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr); | |
174 | ||
175 | /* | |
176 | * Get page for the memmap's phys address | |
177 | * XXX: need more consideration for sparse_vmemmap... | |
178 | */ | |
179 | page = virt_to_page(memmap); | |
180 | mapsize = sizeof(struct page) * PAGES_PER_SECTION; | |
181 | mapsize = PAGE_ALIGN(mapsize) >> PAGE_SHIFT; | |
182 | ||
183 | /* remember memmap's page */ | |
184 | for (i = 0; i < mapsize; i++, page++) | |
185 | get_page_bootmem(section_nr, page, SECTION_INFO); | |
186 | ||
9ac9322d | 187 | usemap = ms->pageblock_flags; |
04753278 YG |
188 | page = virt_to_page(usemap); |
189 | ||
190 | mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT; | |
191 | ||
192 | for (i = 0; i < mapsize; i++, page++) | |
af370fb8 | 193 | get_page_bootmem(section_nr, page, MIX_SECTION_INFO); |
04753278 YG |
194 | |
195 | } | |
46723bfa YI |
196 | #else /* CONFIG_SPARSEMEM_VMEMMAP */ |
197 | static void register_page_bootmem_info_section(unsigned long start_pfn) | |
198 | { | |
199 | unsigned long *usemap, mapsize, section_nr, i; | |
200 | struct mem_section *ms; | |
201 | struct page *page, *memmap; | |
202 | ||
46723bfa YI |
203 | section_nr = pfn_to_section_nr(start_pfn); |
204 | ms = __nr_to_section(section_nr); | |
205 | ||
206 | memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr); | |
207 | ||
208 | register_page_bootmem_memmap(section_nr, memmap, PAGES_PER_SECTION); | |
209 | ||
9ac9322d | 210 | usemap = ms->pageblock_flags; |
46723bfa YI |
211 | page = virt_to_page(usemap); |
212 | ||
213 | mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT; | |
214 | ||
215 | for (i = 0; i < mapsize; i++, page++) | |
216 | get_page_bootmem(section_nr, page, MIX_SECTION_INFO); | |
217 | } | |
218 | #endif /* !CONFIG_SPARSEMEM_VMEMMAP */ | |
04753278 | 219 | |
7ded384a | 220 | void __init register_page_bootmem_info_node(struct pglist_data *pgdat) |
04753278 YG |
221 | { |
222 | unsigned long i, pfn, end_pfn, nr_pages; | |
223 | int node = pgdat->node_id; | |
224 | struct page *page; | |
04753278 YG |
225 | |
226 | nr_pages = PAGE_ALIGN(sizeof(struct pglist_data)) >> PAGE_SHIFT; | |
227 | page = virt_to_page(pgdat); | |
228 | ||
229 | for (i = 0; i < nr_pages; i++, page++) | |
230 | get_page_bootmem(node, page, NODE_INFO); | |
231 | ||
04753278 | 232 | pfn = pgdat->node_start_pfn; |
c1f19495 | 233 | end_pfn = pgdat_end_pfn(pgdat); |
04753278 | 234 | |
7e9f5eb0 | 235 | /* register section info */ |
f14851af | 236 | for (; pfn < end_pfn; pfn += PAGES_PER_SECTION) { |
237 | /* | |
238 | * Some platforms can assign the same pfn to multiple nodes - on | |
239 | * node0 as well as nodeN. To avoid registering a pfn against | |
240 | * multiple nodes we check that this pfn does not already | |
7e9f5eb0 | 241 | * reside in some other nodes. |
f14851af | 242 | */ |
f65e91df | 243 | if (pfn_valid(pfn) && (early_pfn_to_nid(pfn) == node)) |
f14851af | 244 | register_page_bootmem_info_section(pfn); |
245 | } | |
04753278 | 246 | } |
46723bfa | 247 | #endif /* CONFIG_HAVE_BOOTMEM_INFO_NODE */ |
04753278 | 248 | |
f1dd2cd1 | 249 | static int __meminit __add_section(int nid, unsigned long phys_start_pfn, |
7b73d978 | 250 | struct vmem_altmap *altmap, bool want_memblock) |
3947be19 | 251 | { |
3947be19 DH |
252 | int ret; |
253 | ||
ebd15302 KH |
254 | if (pfn_valid(phys_start_pfn)) |
255 | return -EEXIST; | |
256 | ||
7b73d978 | 257 | ret = sparse_add_one_section(NODE_DATA(nid), phys_start_pfn, altmap); |
3947be19 DH |
258 | if (ret < 0) |
259 | return ret; | |
260 | ||
1b862aec MH |
261 | if (!want_memblock) |
262 | return 0; | |
263 | ||
fc44f7f9 | 264 | return hotplug_memory_register(nid, __pfn_to_section(phys_start_pfn)); |
3947be19 DH |
265 | } |
266 | ||
4edd7cef DR |
267 | /* |
268 | * Reasonably generic function for adding memory. It is | |
269 | * expected that archs that support memory hotplug will | |
270 | * call this function after deciding the zone to which to | |
271 | * add the new pages. | |
272 | */ | |
f1dd2cd1 | 273 | int __ref __add_pages(int nid, unsigned long phys_start_pfn, |
24e6d5a5 CH |
274 | unsigned long nr_pages, struct vmem_altmap *altmap, |
275 | bool want_memblock) | |
4edd7cef DR |
276 | { |
277 | unsigned long i; | |
278 | int err = 0; | |
279 | int start_sec, end_sec; | |
4b94ffdc | 280 | |
4edd7cef DR |
281 | /* during initialize mem_map, align hot-added range to section */ |
282 | start_sec = pfn_to_section_nr(phys_start_pfn); | |
283 | end_sec = pfn_to_section_nr(phys_start_pfn + nr_pages - 1); | |
284 | ||
4b94ffdc DW |
285 | if (altmap) { |
286 | /* | |
287 | * Validate altmap is within bounds of the total request | |
288 | */ | |
289 | if (altmap->base_pfn != phys_start_pfn | |
290 | || vmem_altmap_offset(altmap) > nr_pages) { | |
291 | pr_warn_once("memory add fail, invalid altmap\n"); | |
7cf91a98 JK |
292 | err = -EINVAL; |
293 | goto out; | |
4b94ffdc DW |
294 | } |
295 | altmap->alloc = 0; | |
296 | } | |
297 | ||
4edd7cef | 298 | for (i = start_sec; i <= end_sec; i++) { |
7b73d978 CH |
299 | err = __add_section(nid, section_nr_to_pfn(i), altmap, |
300 | want_memblock); | |
4edd7cef DR |
301 | |
302 | /* | |
303 | * EEXIST is finally dealt with by ioresource collision | |
304 | * check. see add_memory() => register_memory_resource() | |
305 | * Warning will be printed if there is collision. | |
306 | */ | |
307 | if (err && (err != -EEXIST)) | |
308 | break; | |
309 | err = 0; | |
f64ac5e6 | 310 | cond_resched(); |
4edd7cef | 311 | } |
c435a390 | 312 | vmemmap_populate_print_last(); |
7cf91a98 | 313 | out: |
4edd7cef DR |
314 | return err; |
315 | } | |
4edd7cef DR |
316 | |
317 | #ifdef CONFIG_MEMORY_HOTREMOVE | |
815121d2 | 318 | /* find the smallest valid pfn in the range [start_pfn, end_pfn) */ |
d09b0137 | 319 | static unsigned long find_smallest_section_pfn(int nid, struct zone *zone, |
815121d2 YI |
320 | unsigned long start_pfn, |
321 | unsigned long end_pfn) | |
322 | { | |
323 | struct mem_section *ms; | |
324 | ||
325 | for (; start_pfn < end_pfn; start_pfn += PAGES_PER_SECTION) { | |
326 | ms = __pfn_to_section(start_pfn); | |
327 | ||
328 | if (unlikely(!valid_section(ms))) | |
329 | continue; | |
330 | ||
331 | if (unlikely(pfn_to_nid(start_pfn) != nid)) | |
332 | continue; | |
333 | ||
334 | if (zone && zone != page_zone(pfn_to_page(start_pfn))) | |
335 | continue; | |
336 | ||
337 | return start_pfn; | |
338 | } | |
339 | ||
340 | return 0; | |
341 | } | |
342 | ||
343 | /* find the biggest valid pfn in the range [start_pfn, end_pfn). */ | |
d09b0137 | 344 | static unsigned long find_biggest_section_pfn(int nid, struct zone *zone, |
815121d2 YI |
345 | unsigned long start_pfn, |
346 | unsigned long end_pfn) | |
347 | { | |
348 | struct mem_section *ms; | |
349 | unsigned long pfn; | |
350 | ||
351 | /* pfn is the end pfn of a memory section. */ | |
352 | pfn = end_pfn - 1; | |
353 | for (; pfn >= start_pfn; pfn -= PAGES_PER_SECTION) { | |
354 | ms = __pfn_to_section(pfn); | |
355 | ||
356 | if (unlikely(!valid_section(ms))) | |
357 | continue; | |
358 | ||
359 | if (unlikely(pfn_to_nid(pfn) != nid)) | |
360 | continue; | |
361 | ||
362 | if (zone && zone != page_zone(pfn_to_page(pfn))) | |
363 | continue; | |
364 | ||
365 | return pfn; | |
366 | } | |
367 | ||
368 | return 0; | |
369 | } | |
370 | ||
371 | static void shrink_zone_span(struct zone *zone, unsigned long start_pfn, | |
372 | unsigned long end_pfn) | |
373 | { | |
c33bc315 XQ |
374 | unsigned long zone_start_pfn = zone->zone_start_pfn; |
375 | unsigned long z = zone_end_pfn(zone); /* zone_end_pfn namespace clash */ | |
376 | unsigned long zone_end_pfn = z; | |
815121d2 YI |
377 | unsigned long pfn; |
378 | struct mem_section *ms; | |
379 | int nid = zone_to_nid(zone); | |
380 | ||
381 | zone_span_writelock(zone); | |
382 | if (zone_start_pfn == start_pfn) { | |
383 | /* | |
384 | * If the section is smallest section in the zone, it need | |
385 | * shrink zone->zone_start_pfn and zone->zone_spanned_pages. | |
386 | * In this case, we find second smallest valid mem_section | |
387 | * for shrinking zone. | |
388 | */ | |
389 | pfn = find_smallest_section_pfn(nid, zone, end_pfn, | |
390 | zone_end_pfn); | |
391 | if (pfn) { | |
392 | zone->zone_start_pfn = pfn; | |
393 | zone->spanned_pages = zone_end_pfn - pfn; | |
394 | } | |
395 | } else if (zone_end_pfn == end_pfn) { | |
396 | /* | |
397 | * If the section is biggest section in the zone, it need | |
398 | * shrink zone->spanned_pages. | |
399 | * In this case, we find second biggest valid mem_section for | |
400 | * shrinking zone. | |
401 | */ | |
402 | pfn = find_biggest_section_pfn(nid, zone, zone_start_pfn, | |
403 | start_pfn); | |
404 | if (pfn) | |
405 | zone->spanned_pages = pfn - zone_start_pfn + 1; | |
406 | } | |
407 | ||
408 | /* | |
409 | * The section is not biggest or smallest mem_section in the zone, it | |
410 | * only creates a hole in the zone. So in this case, we need not | |
411 | * change the zone. But perhaps, the zone has only hole data. Thus | |
412 | * it check the zone has only hole or not. | |
413 | */ | |
414 | pfn = zone_start_pfn; | |
415 | for (; pfn < zone_end_pfn; pfn += PAGES_PER_SECTION) { | |
416 | ms = __pfn_to_section(pfn); | |
417 | ||
418 | if (unlikely(!valid_section(ms))) | |
419 | continue; | |
420 | ||
421 | if (page_zone(pfn_to_page(pfn)) != zone) | |
422 | continue; | |
423 | ||
424 | /* If the section is current section, it continues the loop */ | |
425 | if (start_pfn == pfn) | |
426 | continue; | |
427 | ||
428 | /* If we find valid section, we have nothing to do */ | |
429 | zone_span_writeunlock(zone); | |
430 | return; | |
431 | } | |
432 | ||
433 | /* The zone has no valid section */ | |
434 | zone->zone_start_pfn = 0; | |
435 | zone->spanned_pages = 0; | |
436 | zone_span_writeunlock(zone); | |
437 | } | |
438 | ||
439 | static void shrink_pgdat_span(struct pglist_data *pgdat, | |
440 | unsigned long start_pfn, unsigned long end_pfn) | |
441 | { | |
83285c72 XQ |
442 | unsigned long pgdat_start_pfn = pgdat->node_start_pfn; |
443 | unsigned long p = pgdat_end_pfn(pgdat); /* pgdat_end_pfn namespace clash */ | |
444 | unsigned long pgdat_end_pfn = p; | |
815121d2 YI |
445 | unsigned long pfn; |
446 | struct mem_section *ms; | |
447 | int nid = pgdat->node_id; | |
448 | ||
449 | if (pgdat_start_pfn == start_pfn) { | |
450 | /* | |
451 | * If the section is smallest section in the pgdat, it need | |
452 | * shrink pgdat->node_start_pfn and pgdat->node_spanned_pages. | |
453 | * In this case, we find second smallest valid mem_section | |
454 | * for shrinking zone. | |
455 | */ | |
456 | pfn = find_smallest_section_pfn(nid, NULL, end_pfn, | |
457 | pgdat_end_pfn); | |
458 | if (pfn) { | |
459 | pgdat->node_start_pfn = pfn; | |
460 | pgdat->node_spanned_pages = pgdat_end_pfn - pfn; | |
461 | } | |
462 | } else if (pgdat_end_pfn == end_pfn) { | |
463 | /* | |
464 | * If the section is biggest section in the pgdat, it need | |
465 | * shrink pgdat->node_spanned_pages. | |
466 | * In this case, we find second biggest valid mem_section for | |
467 | * shrinking zone. | |
468 | */ | |
469 | pfn = find_biggest_section_pfn(nid, NULL, pgdat_start_pfn, | |
470 | start_pfn); | |
471 | if (pfn) | |
472 | pgdat->node_spanned_pages = pfn - pgdat_start_pfn + 1; | |
473 | } | |
474 | ||
475 | /* | |
476 | * If the section is not biggest or smallest mem_section in the pgdat, | |
477 | * it only creates a hole in the pgdat. So in this case, we need not | |
478 | * change the pgdat. | |
479 | * But perhaps, the pgdat has only hole data. Thus it check the pgdat | |
480 | * has only hole or not. | |
481 | */ | |
482 | pfn = pgdat_start_pfn; | |
483 | for (; pfn < pgdat_end_pfn; pfn += PAGES_PER_SECTION) { | |
484 | ms = __pfn_to_section(pfn); | |
485 | ||
486 | if (unlikely(!valid_section(ms))) | |
487 | continue; | |
488 | ||
489 | if (pfn_to_nid(pfn) != nid) | |
490 | continue; | |
491 | ||
492 | /* If the section is current section, it continues the loop */ | |
493 | if (start_pfn == pfn) | |
494 | continue; | |
495 | ||
496 | /* If we find valid section, we have nothing to do */ | |
497 | return; | |
498 | } | |
499 | ||
500 | /* The pgdat has no valid section */ | |
501 | pgdat->node_start_pfn = 0; | |
502 | pgdat->node_spanned_pages = 0; | |
503 | } | |
504 | ||
505 | static void __remove_zone(struct zone *zone, unsigned long start_pfn) | |
506 | { | |
507 | struct pglist_data *pgdat = zone->zone_pgdat; | |
508 | int nr_pages = PAGES_PER_SECTION; | |
815121d2 YI |
509 | unsigned long flags; |
510 | ||
815121d2 YI |
511 | pgdat_resize_lock(zone->zone_pgdat, &flags); |
512 | shrink_zone_span(zone, start_pfn, start_pfn + nr_pages); | |
513 | shrink_pgdat_span(pgdat, start_pfn, start_pfn + nr_pages); | |
514 | pgdat_resize_unlock(zone->zone_pgdat, &flags); | |
515 | } | |
516 | ||
4b94ffdc | 517 | static int __remove_section(struct zone *zone, struct mem_section *ms, |
24b6d416 | 518 | unsigned long map_offset, struct vmem_altmap *altmap) |
ea01ea93 | 519 | { |
815121d2 YI |
520 | unsigned long start_pfn; |
521 | int scn_nr; | |
ea01ea93 BP |
522 | int ret = -EINVAL; |
523 | ||
524 | if (!valid_section(ms)) | |
525 | return ret; | |
526 | ||
527 | ret = unregister_memory_section(ms); | |
528 | if (ret) | |
529 | return ret; | |
530 | ||
815121d2 | 531 | scn_nr = __section_nr(ms); |
1dd2bfc8 | 532 | start_pfn = section_nr_to_pfn((unsigned long)scn_nr); |
815121d2 YI |
533 | __remove_zone(zone, start_pfn); |
534 | ||
24b6d416 | 535 | sparse_remove_one_section(zone, ms, map_offset, altmap); |
ea01ea93 BP |
536 | return 0; |
537 | } | |
538 | ||
ea01ea93 BP |
539 | /** |
540 | * __remove_pages() - remove sections of pages from a zone | |
541 | * @zone: zone from which pages need to be removed | |
542 | * @phys_start_pfn: starting pageframe (must be aligned to start of a section) | |
543 | * @nr_pages: number of pages to remove (must be multiple of section size) | |
e8b098fc | 544 | * @altmap: alternative device page map or %NULL if default memmap is used |
ea01ea93 BP |
545 | * |
546 | * Generic helper function to remove section mappings and sysfs entries | |
547 | * for the section of the memory we are removing. Caller needs to make | |
548 | * sure that pages are marked reserved and zones are adjust properly by | |
549 | * calling offline_pages(). | |
550 | */ | |
551 | int __remove_pages(struct zone *zone, unsigned long phys_start_pfn, | |
da024512 | 552 | unsigned long nr_pages, struct vmem_altmap *altmap) |
ea01ea93 | 553 | { |
fe74ebb1 | 554 | unsigned long i; |
4b94ffdc DW |
555 | unsigned long map_offset = 0; |
556 | int sections_to_remove, ret = 0; | |
557 | ||
558 | /* In the ZONE_DEVICE case device driver owns the memory region */ | |
559 | if (is_dev_zone(zone)) { | |
4b94ffdc DW |
560 | if (altmap) |
561 | map_offset = vmem_altmap_offset(altmap); | |
562 | } else { | |
563 | resource_size_t start, size; | |
564 | ||
565 | start = phys_start_pfn << PAGE_SHIFT; | |
566 | size = nr_pages * PAGE_SIZE; | |
567 | ||
568 | ret = release_mem_region_adjustable(&iomem_resource, start, | |
569 | size); | |
570 | if (ret) { | |
571 | resource_size_t endres = start + size - 1; | |
572 | ||
573 | pr_warn("Unable to release resource <%pa-%pa> (%d)\n", | |
574 | &start, &endres, ret); | |
575 | } | |
576 | } | |
ea01ea93 | 577 | |
7cf91a98 JK |
578 | clear_zone_contiguous(zone); |
579 | ||
ea01ea93 BP |
580 | /* |
581 | * We can only remove entire sections | |
582 | */ | |
583 | BUG_ON(phys_start_pfn & ~PAGE_SECTION_MASK); | |
584 | BUG_ON(nr_pages % PAGES_PER_SECTION); | |
585 | ||
ea01ea93 BP |
586 | sections_to_remove = nr_pages / PAGES_PER_SECTION; |
587 | for (i = 0; i < sections_to_remove; i++) { | |
588 | unsigned long pfn = phys_start_pfn + i*PAGES_PER_SECTION; | |
4b94ffdc | 589 | |
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 | |
b93e0f32 | 842 | /* Must be protected by mem_hotplug_begin() or a device_lock */ |
511c2aba | 843 | int __ref online_pages(unsigned long pfn, unsigned long nr_pages, int online_type) |
75884fb1 | 844 | { |
aa47228a | 845 | unsigned long flags; |
3947be19 DH |
846 | unsigned long onlined_pages = 0; |
847 | struct zone *zone; | |
6811378e | 848 | int need_zonelists_rebuild = 0; |
7b78d335 YG |
849 | int nid; |
850 | int ret; | |
851 | struct memory_notify arg; | |
d0dc12e8 PT |
852 | struct memory_block *mem; |
853 | ||
854 | /* | |
855 | * We can't use pfn_to_nid() because nid might be stored in struct page | |
856 | * which is not yet initialized. Instead, we find nid from memory block. | |
857 | */ | |
858 | mem = find_memory_block(__pfn_to_section(pfn)); | |
859 | nid = mem->nid; | |
7b78d335 | 860 | |
f1dd2cd1 MH |
861 | /* associate pfn range with the zone */ |
862 | zone = move_pfn_range(online_type, nid, pfn, nr_pages); | |
863 | ||
7b78d335 YG |
864 | arg.start_pfn = pfn; |
865 | arg.nr_pages = nr_pages; | |
d9713679 | 866 | node_states_check_changes_online(nr_pages, zone, &arg); |
7b78d335 | 867 | |
7b78d335 YG |
868 | ret = memory_notify(MEM_GOING_ONLINE, &arg); |
869 | ret = notifier_to_errno(ret); | |
e33e33b4 CY |
870 | if (ret) |
871 | goto failed_addition; | |
872 | ||
6811378e YG |
873 | /* |
874 | * If this zone is not populated, then it is not in zonelist. | |
875 | * This means the page allocator ignores this zone. | |
876 | * So, zonelist must be updated after online. | |
877 | */ | |
6dcd73d7 | 878 | if (!populated_zone(zone)) { |
6811378e | 879 | need_zonelists_rebuild = 1; |
72675e13 | 880 | setup_zone_pageset(zone); |
6dcd73d7 | 881 | } |
6811378e | 882 | |
908eedc6 | 883 | ret = walk_system_ram_range(pfn, nr_pages, &onlined_pages, |
75884fb1 | 884 | online_pages_range); |
fd8a4221 | 885 | if (ret) { |
6dcd73d7 WC |
886 | if (need_zonelists_rebuild) |
887 | zone_pcp_reset(zone); | |
e33e33b4 | 888 | goto failed_addition; |
fd8a4221 GL |
889 | } |
890 | ||
3947be19 | 891 | zone->present_pages += onlined_pages; |
aa47228a CS |
892 | |
893 | pgdat_resize_lock(zone->zone_pgdat, &flags); | |
f2937be5 | 894 | zone->zone_pgdat->node_present_pages += onlined_pages; |
aa47228a CS |
895 | pgdat_resize_unlock(zone->zone_pgdat, &flags); |
896 | ||
08dff7b7 | 897 | if (onlined_pages) { |
e888ca35 | 898 | node_states_set_node(nid, &arg); |
08dff7b7 | 899 | if (need_zonelists_rebuild) |
72675e13 | 900 | build_all_zonelists(NULL); |
08dff7b7 JL |
901 | else |
902 | zone_pcp_update(zone); | |
903 | } | |
3947be19 | 904 | |
1b79acc9 KM |
905 | init_per_zone_wmark_min(); |
906 | ||
698b1b30 | 907 | if (onlined_pages) { |
e888ca35 | 908 | kswapd_run(nid); |
698b1b30 VB |
909 | kcompactd_run(nid); |
910 | } | |
61b13993 | 911 | |
1f522509 | 912 | vm_total_pages = nr_free_pagecache_pages(); |
2f7f24ec | 913 | |
2d1d43f6 | 914 | writeback_set_ratelimit(); |
7b78d335 YG |
915 | |
916 | if (onlined_pages) | |
917 | memory_notify(MEM_ONLINE, &arg); | |
30467e0b | 918 | return 0; |
e33e33b4 CY |
919 | |
920 | failed_addition: | |
921 | pr_debug("online_pages [mem %#010llx-%#010llx] failed\n", | |
922 | (unsigned long long) pfn << PAGE_SHIFT, | |
923 | (((unsigned long long) pfn + nr_pages) << PAGE_SHIFT) - 1); | |
924 | memory_notify(MEM_CANCEL_ONLINE, &arg); | |
925 | return ret; | |
3947be19 | 926 | } |
53947027 | 927 | #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */ |
bc02af93 | 928 | |
0bd85420 TC |
929 | static void reset_node_present_pages(pg_data_t *pgdat) |
930 | { | |
931 | struct zone *z; | |
932 | ||
933 | for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++) | |
934 | z->present_pages = 0; | |
935 | ||
936 | pgdat->node_present_pages = 0; | |
937 | } | |
938 | ||
e1319331 HS |
939 | /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */ |
940 | static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start) | |
9af3c2de YG |
941 | { |
942 | struct pglist_data *pgdat; | |
c8e861a5 | 943 | unsigned long start_pfn = PFN_DOWN(start); |
9af3c2de | 944 | |
a1e565aa TC |
945 | pgdat = NODE_DATA(nid); |
946 | if (!pgdat) { | |
947 | pgdat = arch_alloc_nodedata(nid); | |
948 | if (!pgdat) | |
949 | return NULL; | |
9af3c2de | 950 | |
a1e565aa | 951 | arch_refresh_nodedata(nid, pgdat); |
b0dc3a34 | 952 | } else { |
e716f2eb MG |
953 | /* |
954 | * Reset the nr_zones, order and classzone_idx before reuse. | |
955 | * Note that kswapd will init kswapd_classzone_idx properly | |
956 | * when it starts in the near future. | |
957 | */ | |
b0dc3a34 | 958 | pgdat->nr_zones = 0; |
38087d9b MG |
959 | pgdat->kswapd_order = 0; |
960 | pgdat->kswapd_classzone_idx = 0; | |
a1e565aa | 961 | } |
9af3c2de YG |
962 | |
963 | /* we can use NODE_DATA(nid) from here */ | |
964 | ||
03e85f9d OS |
965 | pgdat->node_id = nid; |
966 | pgdat->node_start_pfn = start_pfn; | |
967 | ||
9af3c2de | 968 | /* init node's zones as empty zones, we don't have any present pages.*/ |
03e85f9d | 969 | free_area_init_core_hotplug(nid); |
5830169f | 970 | pgdat->per_cpu_nodestats = alloc_percpu(struct per_cpu_nodestat); |
9af3c2de | 971 | |
959ecc48 KH |
972 | /* |
973 | * The node we allocated has no zone fallback lists. For avoiding | |
974 | * to access not-initialized zonelist, build here. | |
975 | */ | |
72675e13 | 976 | build_all_zonelists(pgdat); |
959ecc48 | 977 | |
0bd85420 TC |
978 | /* |
979 | * When memory is hot-added, all the memory is in offline state. So | |
980 | * clear all zones' present_pages because they will be updated in | |
981 | * online_pages() and offline_pages(). | |
982 | */ | |
03e85f9d | 983 | reset_node_managed_pages(pgdat); |
0bd85420 TC |
984 | reset_node_present_pages(pgdat); |
985 | ||
9af3c2de YG |
986 | return pgdat; |
987 | } | |
988 | ||
b9ff0360 | 989 | static void rollback_node_hotadd(int nid) |
9af3c2de | 990 | { |
b9ff0360 OS |
991 | pg_data_t *pgdat = NODE_DATA(nid); |
992 | ||
9af3c2de | 993 | arch_refresh_nodedata(nid, NULL); |
5830169f | 994 | free_percpu(pgdat->per_cpu_nodestats); |
9af3c2de YG |
995 | arch_free_nodedata(pgdat); |
996 | return; | |
997 | } | |
998 | ||
0a547039 | 999 | |
01b0f197 TK |
1000 | /** |
1001 | * try_online_node - online a node if offlined | |
e8b098fc | 1002 | * @nid: the node ID |
b9ff0360 OS |
1003 | * @start: start addr of the node |
1004 | * @set_node_online: Whether we want to online the node | |
cf23422b | 1005 | * called by cpu_up() to online a node without onlined memory. |
b9ff0360 OS |
1006 | * |
1007 | * Returns: | |
1008 | * 1 -> a new node has been allocated | |
1009 | * 0 -> the node is already online | |
1010 | * -ENOMEM -> the node could not be allocated | |
cf23422b | 1011 | */ |
b9ff0360 | 1012 | static int __try_online_node(int nid, u64 start, bool set_node_online) |
cf23422b | 1013 | { |
b9ff0360 OS |
1014 | pg_data_t *pgdat; |
1015 | int ret = 1; | |
cf23422b | 1016 | |
01b0f197 TK |
1017 | if (node_online(nid)) |
1018 | return 0; | |
1019 | ||
b9ff0360 | 1020 | pgdat = hotadd_new_pgdat(nid, start); |
7553e8f2 | 1021 | if (!pgdat) { |
01b0f197 | 1022 | pr_err("Cannot online node %d due to NULL pgdat\n", nid); |
cf23422b | 1023 | ret = -ENOMEM; |
1024 | goto out; | |
1025 | } | |
b9ff0360 OS |
1026 | |
1027 | if (set_node_online) { | |
1028 | node_set_online(nid); | |
1029 | ret = register_one_node(nid); | |
1030 | BUG_ON(ret); | |
1031 | } | |
cf23422b | 1032 | out: |
b9ff0360 OS |
1033 | return ret; |
1034 | } | |
1035 | ||
1036 | /* | |
1037 | * Users of this function always want to online/register the node | |
1038 | */ | |
1039 | int try_online_node(int nid) | |
1040 | { | |
1041 | int ret; | |
1042 | ||
1043 | mem_hotplug_begin(); | |
1044 | ret = __try_online_node(nid, 0, true); | |
bfc8c901 | 1045 | mem_hotplug_done(); |
cf23422b | 1046 | return ret; |
1047 | } | |
1048 | ||
27356f54 TK |
1049 | static int check_hotplug_memory_range(u64 start, u64 size) |
1050 | { | |
ba325585 PT |
1051 | unsigned long block_sz = memory_block_size_bytes(); |
1052 | u64 block_nr_pages = block_sz >> PAGE_SHIFT; | |
27356f54 | 1053 | u64 nr_pages = size >> PAGE_SHIFT; |
ba325585 | 1054 | u64 start_pfn = PFN_DOWN(start); |
27356f54 | 1055 | |
ba325585 PT |
1056 | /* memory range must be block size aligned */ |
1057 | if (!nr_pages || !IS_ALIGNED(start_pfn, block_nr_pages) || | |
1058 | !IS_ALIGNED(nr_pages, block_nr_pages)) { | |
1059 | pr_err("Block size [%#lx] unaligned hotplug range: start %#llx, size %#llx", | |
1060 | block_sz, start, size); | |
27356f54 TK |
1061 | return -EINVAL; |
1062 | } | |
1063 | ||
1064 | return 0; | |
1065 | } | |
1066 | ||
31bc3858 VK |
1067 | static int online_memory_block(struct memory_block *mem, void *arg) |
1068 | { | |
dc18d706 | 1069 | return device_online(&mem->dev); |
31bc3858 VK |
1070 | } |
1071 | ||
31168481 | 1072 | /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */ |
31bc3858 | 1073 | int __ref add_memory_resource(int nid, struct resource *res, bool online) |
bc02af93 | 1074 | { |
62cedb9f | 1075 | u64 start, size; |
b9ff0360 | 1076 | bool new_node = false; |
bc02af93 YG |
1077 | int ret; |
1078 | ||
62cedb9f DV |
1079 | start = res->start; |
1080 | size = resource_size(res); | |
1081 | ||
27356f54 TK |
1082 | ret = check_hotplug_memory_range(start, size); |
1083 | if (ret) | |
1084 | return ret; | |
1085 | ||
bfc8c901 | 1086 | mem_hotplug_begin(); |
ac13c462 | 1087 | |
7f36e3e5 TC |
1088 | /* |
1089 | * Add new range to memblock so that when hotadd_new_pgdat() is called | |
1090 | * to allocate new pgdat, get_pfn_range_for_nid() will be able to find | |
1091 | * this new range and calculate total pages correctly. The range will | |
1092 | * be removed at hot-remove time. | |
1093 | */ | |
1094 | memblock_add_node(start, size, nid); | |
1095 | ||
b9ff0360 OS |
1096 | ret = __try_online_node(nid, start, false); |
1097 | if (ret < 0) | |
1098 | goto error; | |
1099 | new_node = ret; | |
9af3c2de | 1100 | |
bc02af93 | 1101 | /* call arch's memory hotadd */ |
24e6d5a5 | 1102 | ret = arch_add_memory(nid, start, size, NULL, true); |
9af3c2de YG |
1103 | if (ret < 0) |
1104 | goto error; | |
1105 | ||
a1e565aa | 1106 | if (new_node) { |
d5b6f6a3 | 1107 | /* If sysfs file of new node can't be created, cpu on the node |
0fc44159 YG |
1108 | * can't be hot-added. There is no rollback way now. |
1109 | * So, check by BUG_ON() to catch it reluctantly.. | |
d5b6f6a3 | 1110 | * We online node here. We can't roll back from here. |
0fc44159 | 1111 | */ |
d5b6f6a3 OS |
1112 | node_set_online(nid); |
1113 | ret = __register_one_node(nid); | |
0fc44159 YG |
1114 | BUG_ON(ret); |
1115 | } | |
1116 | ||
d5b6f6a3 | 1117 | /* link memory sections under this node.*/ |
4fbce633 | 1118 | ret = link_mem_sections(nid, PFN_DOWN(start), PFN_UP(start + size - 1)); |
d5b6f6a3 OS |
1119 | BUG_ON(ret); |
1120 | ||
d96ae530 | 1121 | /* create new memmap entry */ |
1122 | firmware_map_add_hotplug(start, start + size, "System RAM"); | |
1123 | ||
31bc3858 VK |
1124 | /* online pages if requested */ |
1125 | if (online) | |
1126 | walk_memory_range(PFN_DOWN(start), PFN_UP(start + size - 1), | |
1127 | NULL, online_memory_block); | |
1128 | ||
6ad696d2 AK |
1129 | goto out; |
1130 | ||
9af3c2de YG |
1131 | error: |
1132 | /* rollback pgdat allocation and others */ | |
b9ff0360 OS |
1133 | if (new_node) |
1134 | rollback_node_hotadd(nid); | |
7f36e3e5 | 1135 | memblock_remove(start, size); |
9af3c2de | 1136 | |
6ad696d2 | 1137 | out: |
bfc8c901 | 1138 | mem_hotplug_done(); |
bc02af93 YG |
1139 | return ret; |
1140 | } | |
62cedb9f DV |
1141 | EXPORT_SYMBOL_GPL(add_memory_resource); |
1142 | ||
1143 | int __ref add_memory(int nid, u64 start, u64 size) | |
1144 | { | |
1145 | struct resource *res; | |
1146 | int ret; | |
1147 | ||
1148 | res = register_memory_resource(start, size); | |
6f754ba4 VK |
1149 | if (IS_ERR(res)) |
1150 | return PTR_ERR(res); | |
62cedb9f | 1151 | |
31bc3858 | 1152 | ret = add_memory_resource(nid, res, memhp_auto_online); |
62cedb9f DV |
1153 | if (ret < 0) |
1154 | release_memory_resource(res); | |
1155 | return ret; | |
1156 | } | |
bc02af93 | 1157 | EXPORT_SYMBOL_GPL(add_memory); |
0c0e6195 KH |
1158 | |
1159 | #ifdef CONFIG_MEMORY_HOTREMOVE | |
5c755e9f BP |
1160 | /* |
1161 | * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy | |
1162 | * set and the size of the free page is given by page_order(). Using this, | |
1163 | * the function determines if the pageblock contains only free pages. | |
1164 | * Due to buddy contraints, a free page at least the size of a pageblock will | |
1165 | * be located at the start of the pageblock | |
1166 | */ | |
1167 | static inline int pageblock_free(struct page *page) | |
1168 | { | |
1169 | return PageBuddy(page) && page_order(page) >= pageblock_order; | |
1170 | } | |
1171 | ||
1172 | /* Return the start of the next active pageblock after a given page */ | |
1173 | static struct page *next_active_pageblock(struct page *page) | |
1174 | { | |
5c755e9f BP |
1175 | /* Ensure the starting page is pageblock-aligned */ |
1176 | BUG_ON(page_to_pfn(page) & (pageblock_nr_pages - 1)); | |
1177 | ||
5c755e9f | 1178 | /* If the entire pageblock is free, move to the end of free page */ |
0dcc48c1 KH |
1179 | if (pageblock_free(page)) { |
1180 | int order; | |
1181 | /* be careful. we don't have locks, page_order can be changed.*/ | |
1182 | order = page_order(page); | |
1183 | if ((order < MAX_ORDER) && (order >= pageblock_order)) | |
1184 | return page + (1 << order); | |
1185 | } | |
5c755e9f | 1186 | |
0dcc48c1 | 1187 | return page + pageblock_nr_pages; |
5c755e9f BP |
1188 | } |
1189 | ||
fb52bbae MM |
1190 | static bool is_pageblock_removable_nolock(struct page *page) |
1191 | { | |
1192 | struct zone *zone; | |
1193 | unsigned long pfn; | |
1194 | ||
1195 | /* | |
1196 | * We have to be careful here because we are iterating over memory | |
1197 | * sections which are not zone aware so we might end up outside of | |
1198 | * the zone but still within the section. | |
1199 | * We have to take care about the node as well. If the node is offline | |
1200 | * its NODE_DATA will be NULL - see page_zone. | |
1201 | */ | |
1202 | if (!node_online(page_to_nid(page))) | |
1203 | return false; | |
1204 | ||
1205 | zone = page_zone(page); | |
1206 | pfn = page_to_pfn(page); | |
1207 | if (!zone_spans_pfn(zone, pfn)) | |
1208 | return false; | |
1209 | ||
1210 | return !has_unmovable_pages(zone, page, 0, MIGRATE_MOVABLE, true); | |
1211 | } | |
1212 | ||
5c755e9f | 1213 | /* Checks if this range of memory is likely to be hot-removable. */ |
c98940f6 | 1214 | bool is_mem_section_removable(unsigned long start_pfn, unsigned long nr_pages) |
5c755e9f | 1215 | { |
5c755e9f BP |
1216 | struct page *page = pfn_to_page(start_pfn); |
1217 | struct page *end_page = page + nr_pages; | |
1218 | ||
1219 | /* Check the starting page of each pageblock within the range */ | |
1220 | for (; page < end_page; page = next_active_pageblock(page)) { | |
49ac8255 | 1221 | if (!is_pageblock_removable_nolock(page)) |
c98940f6 | 1222 | return false; |
49ac8255 | 1223 | cond_resched(); |
5c755e9f BP |
1224 | } |
1225 | ||
1226 | /* All pageblocks in the memory block are likely to be hot-removable */ | |
c98940f6 | 1227 | return true; |
5c755e9f BP |
1228 | } |
1229 | ||
0c0e6195 | 1230 | /* |
deb88a2a | 1231 | * Confirm all pages in a range [start, end) belong to the same zone. |
a96dfddb | 1232 | * When true, return its valid [start, end). |
0c0e6195 | 1233 | */ |
a96dfddb TK |
1234 | int test_pages_in_a_zone(unsigned long start_pfn, unsigned long end_pfn, |
1235 | unsigned long *valid_start, unsigned long *valid_end) | |
0c0e6195 | 1236 | { |
5f0f2887 | 1237 | unsigned long pfn, sec_end_pfn; |
a96dfddb | 1238 | unsigned long start, end; |
0c0e6195 KH |
1239 | struct zone *zone = NULL; |
1240 | struct page *page; | |
1241 | int i; | |
deb88a2a | 1242 | for (pfn = start_pfn, sec_end_pfn = SECTION_ALIGN_UP(start_pfn + 1); |
0c0e6195 | 1243 | pfn < end_pfn; |
deb88a2a | 1244 | pfn = sec_end_pfn, sec_end_pfn += PAGES_PER_SECTION) { |
5f0f2887 AB |
1245 | /* Make sure the memory section is present first */ |
1246 | if (!present_section_nr(pfn_to_section_nr(pfn))) | |
0c0e6195 | 1247 | continue; |
5f0f2887 AB |
1248 | for (; pfn < sec_end_pfn && pfn < end_pfn; |
1249 | pfn += MAX_ORDER_NR_PAGES) { | |
1250 | i = 0; | |
1251 | /* This is just a CONFIG_HOLES_IN_ZONE check.*/ | |
1252 | while ((i < MAX_ORDER_NR_PAGES) && | |
1253 | !pfn_valid_within(pfn + i)) | |
1254 | i++; | |
d6d8c8a4 | 1255 | if (i == MAX_ORDER_NR_PAGES || pfn + i >= end_pfn) |
5f0f2887 AB |
1256 | continue; |
1257 | page = pfn_to_page(pfn + i); | |
1258 | if (zone && page_zone(page) != zone) | |
1259 | return 0; | |
a96dfddb TK |
1260 | if (!zone) |
1261 | start = pfn + i; | |
5f0f2887 | 1262 | zone = page_zone(page); |
a96dfddb | 1263 | end = pfn + MAX_ORDER_NR_PAGES; |
5f0f2887 | 1264 | } |
0c0e6195 | 1265 | } |
deb88a2a | 1266 | |
a96dfddb TK |
1267 | if (zone) { |
1268 | *valid_start = start; | |
d6d8c8a4 | 1269 | *valid_end = min(end, end_pfn); |
deb88a2a | 1270 | return 1; |
a96dfddb | 1271 | } else { |
deb88a2a | 1272 | return 0; |
a96dfddb | 1273 | } |
0c0e6195 KH |
1274 | } |
1275 | ||
1276 | /* | |
0efadf48 YX |
1277 | * Scan pfn range [start,end) to find movable/migratable pages (LRU pages, |
1278 | * non-lru movable pages and hugepages). We scan pfn because it's much | |
1279 | * easier than scanning over linked list. This function returns the pfn | |
1280 | * of the first found movable page if it's found, otherwise 0. | |
0c0e6195 | 1281 | */ |
c8721bbb | 1282 | static unsigned long scan_movable_pages(unsigned long start, unsigned long end) |
0c0e6195 KH |
1283 | { |
1284 | unsigned long pfn; | |
1285 | struct page *page; | |
1286 | for (pfn = start; pfn < end; pfn++) { | |
1287 | if (pfn_valid(pfn)) { | |
1288 | page = pfn_to_page(pfn); | |
1289 | if (PageLRU(page)) | |
1290 | return pfn; | |
0efadf48 YX |
1291 | if (__PageMovable(page)) |
1292 | return pfn; | |
c8721bbb | 1293 | if (PageHuge(page)) { |
464c7ffb AK |
1294 | if (hugepage_migration_supported(page_hstate(page)) && |
1295 | page_huge_active(page)) | |
c8721bbb NH |
1296 | return pfn; |
1297 | else | |
1298 | pfn = round_up(pfn + 1, | |
1299 | 1 << compound_order(page)) - 1; | |
1300 | } | |
0c0e6195 KH |
1301 | } |
1302 | } | |
1303 | return 0; | |
1304 | } | |
1305 | ||
666feb21 | 1306 | static struct page *new_node_page(struct page *page, unsigned long private) |
394e31d2 | 1307 | { |
394e31d2 | 1308 | int nid = page_to_nid(page); |
231e97e2 | 1309 | nodemask_t nmask = node_states[N_MEMORY]; |
7f252f27 MH |
1310 | |
1311 | /* | |
1312 | * try to allocate from a different node but reuse this node if there | |
1313 | * are no other online nodes to be used (e.g. we are offlining a part | |
1314 | * of the only existing node) | |
1315 | */ | |
1316 | node_clear(nid, nmask); | |
1317 | if (nodes_empty(nmask)) | |
1318 | node_set(nid, nmask); | |
394e31d2 | 1319 | |
8b913238 | 1320 | return new_page_nodemask(page, nid, &nmask); |
394e31d2 XQ |
1321 | } |
1322 | ||
0c0e6195 KH |
1323 | #define NR_OFFLINE_AT_ONCE_PAGES (256) |
1324 | static int | |
1325 | do_migrate_range(unsigned long start_pfn, unsigned long end_pfn) | |
1326 | { | |
1327 | unsigned long pfn; | |
1328 | struct page *page; | |
1329 | int move_pages = NR_OFFLINE_AT_ONCE_PAGES; | |
1330 | int not_managed = 0; | |
1331 | int ret = 0; | |
1332 | LIST_HEAD(source); | |
1333 | ||
1334 | for (pfn = start_pfn; pfn < end_pfn && move_pages > 0; pfn++) { | |
1335 | if (!pfn_valid(pfn)) | |
1336 | continue; | |
1337 | page = pfn_to_page(pfn); | |
c8721bbb NH |
1338 | |
1339 | if (PageHuge(page)) { | |
1340 | struct page *head = compound_head(page); | |
1341 | pfn = page_to_pfn(head) + (1<<compound_order(head)) - 1; | |
1342 | if (compound_order(head) > PFN_SECTION_SHIFT) { | |
1343 | ret = -EBUSY; | |
1344 | break; | |
1345 | } | |
1346 | if (isolate_huge_page(page, &source)) | |
1347 | move_pages -= 1 << compound_order(head); | |
1348 | continue; | |
94723aaf | 1349 | } else if (PageTransHuge(page)) |
8135d892 NH |
1350 | pfn = page_to_pfn(compound_head(page)) |
1351 | + hpage_nr_pages(page) - 1; | |
c8721bbb | 1352 | |
700c2a46 | 1353 | if (!get_page_unless_zero(page)) |
0c0e6195 KH |
1354 | continue; |
1355 | /* | |
0efadf48 YX |
1356 | * We can skip free pages. And we can deal with pages on |
1357 | * LRU and non-lru movable pages. | |
0c0e6195 | 1358 | */ |
0efadf48 YX |
1359 | if (PageLRU(page)) |
1360 | ret = isolate_lru_page(page); | |
1361 | else | |
1362 | ret = isolate_movable_page(page, ISOLATE_UNEVICTABLE); | |
0c0e6195 | 1363 | if (!ret) { /* Success */ |
700c2a46 | 1364 | put_page(page); |
62695a84 | 1365 | list_add_tail(&page->lru, &source); |
0c0e6195 | 1366 | move_pages--; |
0efadf48 YX |
1367 | if (!__PageMovable(page)) |
1368 | inc_node_page_state(page, NR_ISOLATED_ANON + | |
1369 | page_is_file_cache(page)); | |
6d9c285a | 1370 | |
0c0e6195 | 1371 | } else { |
0c0e6195 | 1372 | #ifdef CONFIG_DEBUG_VM |
0efadf48 YX |
1373 | pr_alert("failed to isolate pfn %lx\n", pfn); |
1374 | dump_page(page, "isolation failed"); | |
0c0e6195 | 1375 | #endif |
700c2a46 | 1376 | put_page(page); |
25985edc | 1377 | /* Because we don't have big zone->lock. we should |
809c4449 BL |
1378 | check this again here. */ |
1379 | if (page_count(page)) { | |
1380 | not_managed++; | |
f3ab2636 | 1381 | ret = -EBUSY; |
809c4449 BL |
1382 | break; |
1383 | } | |
0c0e6195 KH |
1384 | } |
1385 | } | |
f3ab2636 BL |
1386 | if (!list_empty(&source)) { |
1387 | if (not_managed) { | |
c8721bbb | 1388 | putback_movable_pages(&source); |
f3ab2636 BL |
1389 | goto out; |
1390 | } | |
74c08f98 | 1391 | |
394e31d2 XQ |
1392 | /* Allocate a new page from the nearest neighbor node */ |
1393 | ret = migrate_pages(&source, new_node_page, NULL, 0, | |
9c620e2b | 1394 | MIGRATE_SYNC, MR_MEMORY_HOTPLUG); |
f3ab2636 | 1395 | if (ret) |
c8721bbb | 1396 | putback_movable_pages(&source); |
0c0e6195 | 1397 | } |
0c0e6195 KH |
1398 | out: |
1399 | return ret; | |
1400 | } | |
1401 | ||
1402 | /* | |
1403 | * remove from free_area[] and mark all as Reserved. | |
1404 | */ | |
1405 | static int | |
1406 | offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages, | |
1407 | void *data) | |
1408 | { | |
1409 | __offline_isolated_pages(start, start + nr_pages); | |
1410 | return 0; | |
1411 | } | |
1412 | ||
1413 | static void | |
1414 | offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn) | |
1415 | { | |
908eedc6 | 1416 | walk_system_ram_range(start_pfn, end_pfn - start_pfn, NULL, |
0c0e6195 KH |
1417 | offline_isolated_pages_cb); |
1418 | } | |
1419 | ||
1420 | /* | |
1421 | * Check all pages in range, recoreded as memory resource, are isolated. | |
1422 | */ | |
1423 | static int | |
1424 | check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages, | |
1425 | void *data) | |
1426 | { | |
1427 | int ret; | |
1428 | long offlined = *(long *)data; | |
b023f468 | 1429 | ret = test_pages_isolated(start_pfn, start_pfn + nr_pages, true); |
0c0e6195 KH |
1430 | offlined = nr_pages; |
1431 | if (!ret) | |
1432 | *(long *)data += offlined; | |
1433 | return ret; | |
1434 | } | |
1435 | ||
1436 | static long | |
1437 | check_pages_isolated(unsigned long start_pfn, unsigned long end_pfn) | |
1438 | { | |
1439 | long offlined = 0; | |
1440 | int ret; | |
1441 | ||
908eedc6 | 1442 | ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn, &offlined, |
0c0e6195 KH |
1443 | check_pages_isolated_cb); |
1444 | if (ret < 0) | |
1445 | offlined = (long)ret; | |
1446 | return offlined; | |
1447 | } | |
1448 | ||
c5320926 TC |
1449 | static int __init cmdline_parse_movable_node(char *p) |
1450 | { | |
4932381e | 1451 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
55ac590c | 1452 | movable_node_enabled = true; |
4932381e MH |
1453 | #else |
1454 | pr_warn("movable_node parameter depends on CONFIG_HAVE_MEMBLOCK_NODE_MAP to work properly\n"); | |
1455 | #endif | |
c5320926 TC |
1456 | return 0; |
1457 | } | |
1458 | early_param("movable_node", cmdline_parse_movable_node); | |
1459 | ||
d9713679 LJ |
1460 | /* check which state of node_states will be changed when offline memory */ |
1461 | static void node_states_check_changes_offline(unsigned long nr_pages, | |
1462 | struct zone *zone, struct memory_notify *arg) | |
1463 | { | |
1464 | struct pglist_data *pgdat = zone->zone_pgdat; | |
1465 | unsigned long present_pages = 0; | |
86b27bea | 1466 | enum zone_type zt; |
d9713679 | 1467 | |
86b27bea OS |
1468 | arg->status_change_nid = -1; |
1469 | arg->status_change_nid_normal = -1; | |
1470 | arg->status_change_nid_high = -1; | |
d9713679 LJ |
1471 | |
1472 | /* | |
86b27bea OS |
1473 | * Check whether node_states[N_NORMAL_MEMORY] will be changed. |
1474 | * If the memory to be offline is within the range | |
1475 | * [0..ZONE_NORMAL], and it is the last present memory there, | |
1476 | * the zones in that range will become empty after the offlining, | |
1477 | * thus we can determine that we need to clear the node from | |
1478 | * node_states[N_NORMAL_MEMORY]. | |
d9713679 | 1479 | */ |
86b27bea | 1480 | for (zt = 0; zt <= ZONE_NORMAL; zt++) |
d9713679 | 1481 | present_pages += pgdat->node_zones[zt].present_pages; |
86b27bea | 1482 | if (zone_idx(zone) <= ZONE_NORMAL && nr_pages >= present_pages) |
d9713679 | 1483 | arg->status_change_nid_normal = zone_to_nid(zone); |
d9713679 | 1484 | |
6715ddf9 LJ |
1485 | #ifdef CONFIG_HIGHMEM |
1486 | /* | |
86b27bea OS |
1487 | * node_states[N_HIGH_MEMORY] contains nodes which |
1488 | * have normal memory or high memory. | |
1489 | * Here we add the present_pages belonging to ZONE_HIGHMEM. | |
1490 | * If the zone is within the range of [0..ZONE_HIGHMEM), and | |
1491 | * we determine that the zones in that range become empty, | |
1492 | * we need to clear the node for N_HIGH_MEMORY. | |
6715ddf9 | 1493 | */ |
86b27bea OS |
1494 | present_pages += pgdat->node_zones[ZONE_HIGHMEM].present_pages; |
1495 | if (zone_idx(zone) <= ZONE_HIGHMEM && nr_pages >= present_pages) | |
6715ddf9 | 1496 | arg->status_change_nid_high = zone_to_nid(zone); |
6715ddf9 LJ |
1497 | #endif |
1498 | ||
d9713679 | 1499 | /* |
86b27bea OS |
1500 | * We have accounted the pages from [0..ZONE_NORMAL), and |
1501 | * in case of CONFIG_HIGHMEM the pages from ZONE_HIGHMEM | |
1502 | * as well. | |
1503 | * Here we count the possible pages from ZONE_MOVABLE. | |
1504 | * If after having accounted all the pages, we see that the nr_pages | |
1505 | * to be offlined is over or equal to the accounted pages, | |
1506 | * we know that the node will become empty, and so, we can clear | |
1507 | * it for N_MEMORY as well. | |
d9713679 | 1508 | */ |
86b27bea | 1509 | present_pages += pgdat->node_zones[ZONE_MOVABLE].present_pages; |
d9713679 | 1510 | |
d9713679 LJ |
1511 | if (nr_pages >= present_pages) |
1512 | arg->status_change_nid = zone_to_nid(zone); | |
d9713679 LJ |
1513 | } |
1514 | ||
1515 | static void node_states_clear_node(int node, struct memory_notify *arg) | |
1516 | { | |
1517 | if (arg->status_change_nid_normal >= 0) | |
1518 | node_clear_state(node, N_NORMAL_MEMORY); | |
1519 | ||
cf01f6f5 | 1520 | if (arg->status_change_nid_high >= 0) |
d9713679 | 1521 | node_clear_state(node, N_HIGH_MEMORY); |
6715ddf9 | 1522 | |
cf01f6f5 | 1523 | if (arg->status_change_nid >= 0) |
6715ddf9 | 1524 | node_clear_state(node, N_MEMORY); |
d9713679 LJ |
1525 | } |
1526 | ||
a16cee10 | 1527 | static int __ref __offline_pages(unsigned long start_pfn, |
ecde0f3e | 1528 | unsigned long end_pfn) |
0c0e6195 | 1529 | { |
ecde0f3e | 1530 | unsigned long pfn, nr_pages; |
0c0e6195 | 1531 | long offlined_pages; |
72b39cfc | 1532 | int ret, node; |
d702909f | 1533 | unsigned long flags; |
a96dfddb | 1534 | unsigned long valid_start, valid_end; |
0c0e6195 | 1535 | struct zone *zone; |
7b78d335 | 1536 | struct memory_notify arg; |
0c0e6195 | 1537 | |
0c0e6195 KH |
1538 | /* at least, alignment against pageblock is necessary */ |
1539 | if (!IS_ALIGNED(start_pfn, pageblock_nr_pages)) | |
1540 | return -EINVAL; | |
1541 | if (!IS_ALIGNED(end_pfn, pageblock_nr_pages)) | |
1542 | return -EINVAL; | |
1543 | /* This makes hotplug much easier...and readable. | |
1544 | we assume this for now. .*/ | |
a96dfddb | 1545 | if (!test_pages_in_a_zone(start_pfn, end_pfn, &valid_start, &valid_end)) |
0c0e6195 | 1546 | return -EINVAL; |
7b78d335 | 1547 | |
a96dfddb | 1548 | zone = page_zone(pfn_to_page(valid_start)); |
7b78d335 YG |
1549 | node = zone_to_nid(zone); |
1550 | nr_pages = end_pfn - start_pfn; | |
1551 | ||
0c0e6195 | 1552 | /* set above range as isolated */ |
b023f468 WC |
1553 | ret = start_isolate_page_range(start_pfn, end_pfn, |
1554 | MIGRATE_MOVABLE, true); | |
0c0e6195 | 1555 | if (ret) |
30467e0b | 1556 | return ret; |
7b78d335 YG |
1557 | |
1558 | arg.start_pfn = start_pfn; | |
1559 | arg.nr_pages = nr_pages; | |
d9713679 | 1560 | node_states_check_changes_offline(nr_pages, zone, &arg); |
7b78d335 YG |
1561 | |
1562 | ret = memory_notify(MEM_GOING_OFFLINE, &arg); | |
1563 | ret = notifier_to_errno(ret); | |
1564 | if (ret) | |
1565 | goto failed_removal; | |
1566 | ||
0c0e6195 | 1567 | pfn = start_pfn; |
0c0e6195 KH |
1568 | repeat: |
1569 | /* start memory hot removal */ | |
0c0e6195 KH |
1570 | ret = -EINTR; |
1571 | if (signal_pending(current)) | |
1572 | goto failed_removal; | |
72b39cfc MH |
1573 | |
1574 | cond_resched(); | |
9852a721 | 1575 | lru_add_drain_all(); |
72b39cfc | 1576 | drain_all_pages(zone); |
0c0e6195 | 1577 | |
c8721bbb NH |
1578 | pfn = scan_movable_pages(start_pfn, end_pfn); |
1579 | if (pfn) { /* We have movable pages */ | |
0c0e6195 | 1580 | ret = do_migrate_range(pfn, end_pfn); |
72b39cfc | 1581 | goto repeat; |
0c0e6195 | 1582 | } |
72b39cfc | 1583 | |
c8721bbb NH |
1584 | /* |
1585 | * dissolve free hugepages in the memory block before doing offlining | |
1586 | * actually in order to make hugetlbfs's object counting consistent. | |
1587 | */ | |
082d5b6b GS |
1588 | ret = dissolve_free_huge_pages(start_pfn, end_pfn); |
1589 | if (ret) | |
1590 | goto failed_removal; | |
0c0e6195 KH |
1591 | /* check again */ |
1592 | offlined_pages = check_pages_isolated(start_pfn, end_pfn); | |
72b39cfc MH |
1593 | if (offlined_pages < 0) |
1594 | goto repeat; | |
e33e33b4 | 1595 | pr_info("Offlined Pages %ld\n", offlined_pages); |
b3834be5 | 1596 | /* Ok, all of our target is isolated. |
0c0e6195 KH |
1597 | We cannot do rollback at this point. */ |
1598 | offline_isolated_pages(start_pfn, end_pfn); | |
dbc0e4ce | 1599 | /* reset pagetype flags and makes migrate type to be MOVABLE */ |
0815f3d8 | 1600 | undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE); |
0c0e6195 | 1601 | /* removal success */ |
3dcc0571 | 1602 | adjust_managed_page_count(pfn_to_page(start_pfn), -offlined_pages); |
0c0e6195 | 1603 | zone->present_pages -= offlined_pages; |
d702909f CS |
1604 | |
1605 | pgdat_resize_lock(zone->zone_pgdat, &flags); | |
0c0e6195 | 1606 | zone->zone_pgdat->node_present_pages -= offlined_pages; |
d702909f | 1607 | pgdat_resize_unlock(zone->zone_pgdat, &flags); |
7b78d335 | 1608 | |
1b79acc9 KM |
1609 | init_per_zone_wmark_min(); |
1610 | ||
1e8537ba | 1611 | if (!populated_zone(zone)) { |
340175b7 | 1612 | zone_pcp_reset(zone); |
72675e13 | 1613 | build_all_zonelists(NULL); |
1e8537ba XQ |
1614 | } else |
1615 | zone_pcp_update(zone); | |
340175b7 | 1616 | |
d9713679 | 1617 | node_states_clear_node(node, &arg); |
698b1b30 | 1618 | if (arg.status_change_nid >= 0) { |
8fe23e05 | 1619 | kswapd_stop(node); |
698b1b30 VB |
1620 | kcompactd_stop(node); |
1621 | } | |
bce7394a | 1622 | |
0c0e6195 KH |
1623 | vm_total_pages = nr_free_pagecache_pages(); |
1624 | writeback_set_ratelimit(); | |
7b78d335 YG |
1625 | |
1626 | memory_notify(MEM_OFFLINE, &arg); | |
0c0e6195 KH |
1627 | return 0; |
1628 | ||
1629 | failed_removal: | |
e33e33b4 CY |
1630 | pr_debug("memory offlining [mem %#010llx-%#010llx] failed\n", |
1631 | (unsigned long long) start_pfn << PAGE_SHIFT, | |
1632 | ((unsigned long long) end_pfn << PAGE_SHIFT) - 1); | |
7b78d335 | 1633 | memory_notify(MEM_CANCEL_OFFLINE, &arg); |
0c0e6195 | 1634 | /* pushback to free area */ |
0815f3d8 | 1635 | undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE); |
0c0e6195 KH |
1636 | return ret; |
1637 | } | |
71088785 | 1638 | |
b93e0f32 | 1639 | /* Must be protected by mem_hotplug_begin() or a device_lock */ |
a16cee10 WC |
1640 | int offline_pages(unsigned long start_pfn, unsigned long nr_pages) |
1641 | { | |
ecde0f3e | 1642 | return __offline_pages(start_pfn, start_pfn + nr_pages); |
a16cee10 | 1643 | } |
e2ff3940 | 1644 | #endif /* CONFIG_MEMORY_HOTREMOVE */ |
a16cee10 | 1645 | |
bbc76be6 WC |
1646 | /** |
1647 | * walk_memory_range - walks through all mem sections in [start_pfn, end_pfn) | |
1648 | * @start_pfn: start pfn of the memory range | |
e05c4bbf | 1649 | * @end_pfn: end pfn of the memory range |
bbc76be6 WC |
1650 | * @arg: argument passed to func |
1651 | * @func: callback for each memory section walked | |
1652 | * | |
1653 | * This function walks through all present mem sections in range | |
1654 | * [start_pfn, end_pfn) and call func on each mem section. | |
1655 | * | |
1656 | * Returns the return value of func. | |
1657 | */ | |
e2ff3940 | 1658 | int walk_memory_range(unsigned long start_pfn, unsigned long end_pfn, |
bbc76be6 | 1659 | void *arg, int (*func)(struct memory_block *, void *)) |
71088785 | 1660 | { |
e90bdb7f WC |
1661 | struct memory_block *mem = NULL; |
1662 | struct mem_section *section; | |
e90bdb7f WC |
1663 | unsigned long pfn, section_nr; |
1664 | int ret; | |
e90bdb7f WC |
1665 | |
1666 | for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) { | |
1667 | section_nr = pfn_to_section_nr(pfn); | |
1668 | if (!present_section_nr(section_nr)) | |
1669 | continue; | |
1670 | ||
1671 | section = __nr_to_section(section_nr); | |
1672 | /* same memblock? */ | |
1673 | if (mem) | |
1674 | if ((section_nr >= mem->start_section_nr) && | |
1675 | (section_nr <= mem->end_section_nr)) | |
1676 | continue; | |
1677 | ||
1678 | mem = find_memory_block_hinted(section, mem); | |
1679 | if (!mem) | |
1680 | continue; | |
1681 | ||
bbc76be6 | 1682 | ret = func(mem, arg); |
e90bdb7f | 1683 | if (ret) { |
bbc76be6 WC |
1684 | kobject_put(&mem->dev.kobj); |
1685 | return ret; | |
e90bdb7f WC |
1686 | } |
1687 | } | |
1688 | ||
1689 | if (mem) | |
1690 | kobject_put(&mem->dev.kobj); | |
1691 | ||
bbc76be6 WC |
1692 | return 0; |
1693 | } | |
1694 | ||
e2ff3940 | 1695 | #ifdef CONFIG_MEMORY_HOTREMOVE |
d6de9d53 | 1696 | static int check_memblock_offlined_cb(struct memory_block *mem, void *arg) |
bbc76be6 WC |
1697 | { |
1698 | int ret = !is_memblock_offlined(mem); | |
1699 | ||
349daa0f RD |
1700 | if (unlikely(ret)) { |
1701 | phys_addr_t beginpa, endpa; | |
1702 | ||
1703 | beginpa = PFN_PHYS(section_nr_to_pfn(mem->start_section_nr)); | |
1704 | endpa = PFN_PHYS(section_nr_to_pfn(mem->end_section_nr + 1))-1; | |
756a025f | 1705 | pr_warn("removing memory fails, because memory [%pa-%pa] is onlined\n", |
349daa0f RD |
1706 | &beginpa, &endpa); |
1707 | } | |
bbc76be6 WC |
1708 | |
1709 | return ret; | |
1710 | } | |
1711 | ||
0f1cfe9d | 1712 | static int check_cpu_on_node(pg_data_t *pgdat) |
60a5a19e | 1713 | { |
60a5a19e TC |
1714 | int cpu; |
1715 | ||
1716 | for_each_present_cpu(cpu) { | |
1717 | if (cpu_to_node(cpu) == pgdat->node_id) | |
1718 | /* | |
1719 | * the cpu on this node isn't removed, and we can't | |
1720 | * offline this node. | |
1721 | */ | |
1722 | return -EBUSY; | |
1723 | } | |
1724 | ||
1725 | return 0; | |
1726 | } | |
1727 | ||
0f1cfe9d | 1728 | static void unmap_cpu_on_node(pg_data_t *pgdat) |
e13fe869 WC |
1729 | { |
1730 | #ifdef CONFIG_ACPI_NUMA | |
e13fe869 WC |
1731 | int cpu; |
1732 | ||
1733 | for_each_possible_cpu(cpu) | |
1734 | if (cpu_to_node(cpu) == pgdat->node_id) | |
1735 | numa_clear_node(cpu); | |
1736 | #endif | |
1737 | } | |
1738 | ||
0f1cfe9d | 1739 | static int check_and_unmap_cpu_on_node(pg_data_t *pgdat) |
e13fe869 | 1740 | { |
0f1cfe9d | 1741 | int ret; |
e13fe869 | 1742 | |
0f1cfe9d | 1743 | ret = check_cpu_on_node(pgdat); |
e13fe869 WC |
1744 | if (ret) |
1745 | return ret; | |
1746 | ||
1747 | /* | |
1748 | * the node will be offlined when we come here, so we can clear | |
1749 | * the cpu_to_node() now. | |
1750 | */ | |
1751 | ||
0f1cfe9d | 1752 | unmap_cpu_on_node(pgdat); |
e13fe869 WC |
1753 | return 0; |
1754 | } | |
1755 | ||
0f1cfe9d TK |
1756 | /** |
1757 | * try_offline_node | |
e8b098fc | 1758 | * @nid: the node ID |
0f1cfe9d TK |
1759 | * |
1760 | * Offline a node if all memory sections and cpus of the node are removed. | |
1761 | * | |
1762 | * NOTE: The caller must call lock_device_hotplug() to serialize hotplug | |
1763 | * and online/offline operations before this call. | |
1764 | */ | |
90b30cdc | 1765 | void try_offline_node(int nid) |
60a5a19e | 1766 | { |
d822b86a WC |
1767 | pg_data_t *pgdat = NODE_DATA(nid); |
1768 | unsigned long start_pfn = pgdat->node_start_pfn; | |
1769 | unsigned long end_pfn = start_pfn + pgdat->node_spanned_pages; | |
60a5a19e TC |
1770 | unsigned long pfn; |
1771 | ||
1772 | for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) { | |
1773 | unsigned long section_nr = pfn_to_section_nr(pfn); | |
1774 | ||
1775 | if (!present_section_nr(section_nr)) | |
1776 | continue; | |
1777 | ||
1778 | if (pfn_to_nid(pfn) != nid) | |
1779 | continue; | |
1780 | ||
1781 | /* | |
1782 | * some memory sections of this node are not removed, and we | |
1783 | * can't offline node now. | |
1784 | */ | |
1785 | return; | |
1786 | } | |
1787 | ||
0f1cfe9d | 1788 | if (check_and_unmap_cpu_on_node(pgdat)) |
60a5a19e TC |
1789 | return; |
1790 | ||
1791 | /* | |
1792 | * all memory/cpu of this node are removed, we can offline this | |
1793 | * node now. | |
1794 | */ | |
1795 | node_set_offline(nid); | |
1796 | unregister_one_node(nid); | |
1797 | } | |
90b30cdc | 1798 | EXPORT_SYMBOL(try_offline_node); |
60a5a19e | 1799 | |
0f1cfe9d TK |
1800 | /** |
1801 | * remove_memory | |
e8b098fc MR |
1802 | * @nid: the node ID |
1803 | * @start: physical address of the region to remove | |
1804 | * @size: size of the region to remove | |
0f1cfe9d TK |
1805 | * |
1806 | * NOTE: The caller must call lock_device_hotplug() to serialize hotplug | |
1807 | * and online/offline operations before this call, as required by | |
1808 | * try_offline_node(). | |
1809 | */ | |
242831eb | 1810 | void __ref remove_memory(int nid, u64 start, u64 size) |
bbc76be6 | 1811 | { |
242831eb | 1812 | int ret; |
993c1aad | 1813 | |
27356f54 TK |
1814 | BUG_ON(check_hotplug_memory_range(start, size)); |
1815 | ||
bfc8c901 | 1816 | mem_hotplug_begin(); |
6677e3ea YI |
1817 | |
1818 | /* | |
242831eb RW |
1819 | * All memory blocks must be offlined before removing memory. Check |
1820 | * whether all memory blocks in question are offline and trigger a BUG() | |
1821 | * if this is not the case. | |
6677e3ea | 1822 | */ |
242831eb | 1823 | ret = walk_memory_range(PFN_DOWN(start), PFN_UP(start + size - 1), NULL, |
d6de9d53 | 1824 | check_memblock_offlined_cb); |
bfc8c901 | 1825 | if (ret) |
242831eb | 1826 | BUG(); |
6677e3ea | 1827 | |
46c66c4b YI |
1828 | /* remove memmap entry */ |
1829 | firmware_map_remove(start, start + size, "System RAM"); | |
f9126ab9 XQ |
1830 | memblock_free(start, size); |
1831 | memblock_remove(start, size); | |
46c66c4b | 1832 | |
da024512 | 1833 | arch_remove_memory(start, size, NULL); |
24d335ca | 1834 | |
60a5a19e TC |
1835 | try_offline_node(nid); |
1836 | ||
bfc8c901 | 1837 | mem_hotplug_done(); |
71088785 | 1838 | } |
71088785 | 1839 | EXPORT_SYMBOL_GPL(remove_memory); |
aba6efc4 | 1840 | #endif /* CONFIG_MEMORY_HOTREMOVE */ |