Commit | Line | Data |
---|---|---|
457c8996 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
3947be19 DH |
2 | /* |
3 | * linux/mm/memory_hotplug.c | |
4 | * | |
5 | * Copyright (C) | |
6 | */ | |
7 | ||
3947be19 DH |
8 | #include <linux/stddef.h> |
9 | #include <linux/mm.h> | |
174cd4b1 | 10 | #include <linux/sched/signal.h> |
3947be19 DH |
11 | #include <linux/swap.h> |
12 | #include <linux/interrupt.h> | |
13 | #include <linux/pagemap.h> | |
3947be19 | 14 | #include <linux/compiler.h> |
b95f1b31 | 15 | #include <linux/export.h> |
3947be19 | 16 | #include <linux/pagevec.h> |
2d1d43f6 | 17 | #include <linux/writeback.h> |
3947be19 DH |
18 | #include <linux/slab.h> |
19 | #include <linux/sysctl.h> | |
20 | #include <linux/cpu.h> | |
21 | #include <linux/memory.h> | |
4b94ffdc | 22 | #include <linux/memremap.h> |
3947be19 DH |
23 | #include <linux/memory_hotplug.h> |
24 | #include <linux/highmem.h> | |
25 | #include <linux/vmalloc.h> | |
0a547039 | 26 | #include <linux/ioport.h> |
0c0e6195 KH |
27 | #include <linux/delay.h> |
28 | #include <linux/migrate.h> | |
29 | #include <linux/page-isolation.h> | |
71088785 | 30 | #include <linux/pfn.h> |
6ad696d2 | 31 | #include <linux/suspend.h> |
6d9c285a | 32 | #include <linux/mm_inline.h> |
d96ae530 | 33 | #include <linux/firmware-map.h> |
60a5a19e | 34 | #include <linux/stop_machine.h> |
c8721bbb | 35 | #include <linux/hugetlb.h> |
c5320926 | 36 | #include <linux/memblock.h> |
698b1b30 | 37 | #include <linux/compaction.h> |
b15c8726 | 38 | #include <linux/rmap.h> |
3947be19 DH |
39 | |
40 | #include <asm/tlbflush.h> | |
41 | ||
1e5ad9a3 | 42 | #include "internal.h" |
e900a918 | 43 | #include "shuffle.h" |
1e5ad9a3 | 44 | |
9d0ad8ca DK |
45 | /* |
46 | * online_page_callback contains pointer to current page onlining function. | |
47 | * Initially it is generic_online_page(). If it is required it could be | |
48 | * changed by calling set_online_page_callback() for callback registration | |
49 | * and restore_online_page_callback() for generic callback restore. | |
50 | */ | |
51 | ||
a9cd410a | 52 | static void generic_online_page(struct page *page, unsigned int order); |
9d0ad8ca DK |
53 | |
54 | static online_page_callback_t online_page_callback = generic_online_page; | |
bfc8c901 | 55 | static DEFINE_MUTEX(online_page_callback_lock); |
9d0ad8ca | 56 | |
3f906ba2 | 57 | DEFINE_STATIC_PERCPU_RWSEM(mem_hotplug_lock); |
bfc8c901 | 58 | |
3f906ba2 TG |
59 | void get_online_mems(void) |
60 | { | |
61 | percpu_down_read(&mem_hotplug_lock); | |
62 | } | |
bfc8c901 | 63 | |
3f906ba2 TG |
64 | void put_online_mems(void) |
65 | { | |
66 | percpu_up_read(&mem_hotplug_lock); | |
67 | } | |
bfc8c901 | 68 | |
4932381e MH |
69 | bool movable_node_enabled = false; |
70 | ||
8604d9e5 | 71 | #ifndef CONFIG_MEMORY_HOTPLUG_DEFAULT_ONLINE |
31bc3858 | 72 | bool memhp_auto_online; |
8604d9e5 VK |
73 | #else |
74 | bool memhp_auto_online = true; | |
75 | #endif | |
31bc3858 VK |
76 | EXPORT_SYMBOL_GPL(memhp_auto_online); |
77 | ||
86dd995d VK |
78 | static int __init setup_memhp_default_state(char *str) |
79 | { | |
80 | if (!strcmp(str, "online")) | |
81 | memhp_auto_online = true; | |
82 | else if (!strcmp(str, "offline")) | |
83 | memhp_auto_online = false; | |
84 | ||
85 | return 1; | |
86 | } | |
87 | __setup("memhp_default_state=", setup_memhp_default_state); | |
88 | ||
30467e0b | 89 | void mem_hotplug_begin(void) |
20d6c96b | 90 | { |
3f906ba2 TG |
91 | cpus_read_lock(); |
92 | percpu_down_write(&mem_hotplug_lock); | |
20d6c96b KM |
93 | } |
94 | ||
30467e0b | 95 | void mem_hotplug_done(void) |
bfc8c901 | 96 | { |
3f906ba2 TG |
97 | percpu_up_write(&mem_hotplug_lock); |
98 | cpus_read_unlock(); | |
bfc8c901 | 99 | } |
20d6c96b | 100 | |
357b4da5 JG |
101 | u64 max_mem_size = U64_MAX; |
102 | ||
45e0b78b KM |
103 | /* add this memory to iomem resource */ |
104 | static struct resource *register_memory_resource(u64 start, u64 size) | |
105 | { | |
2794129e DH |
106 | struct resource *res; |
107 | unsigned long flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY; | |
108 | char *resource_name = "System RAM"; | |
357b4da5 JG |
109 | |
110 | if (start + size > max_mem_size) | |
111 | return ERR_PTR(-E2BIG); | |
112 | ||
2794129e DH |
113 | /* |
114 | * Request ownership of the new memory range. This might be | |
115 | * a child of an existing resource that was present but | |
116 | * not marked as busy. | |
117 | */ | |
118 | res = __request_region(&iomem_resource, start, size, | |
119 | resource_name, flags); | |
120 | ||
121 | if (!res) { | |
122 | pr_debug("Unable to reserve System RAM region: %016llx->%016llx\n", | |
123 | start, start + size); | |
6f754ba4 | 124 | return ERR_PTR(-EEXIST); |
45e0b78b KM |
125 | } |
126 | return res; | |
127 | } | |
128 | ||
129 | static void release_memory_resource(struct resource *res) | |
130 | { | |
131 | if (!res) | |
132 | return; | |
133 | release_resource(res); | |
134 | kfree(res); | |
135 | return; | |
136 | } | |
137 | ||
53947027 | 138 | #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE |
46723bfa YI |
139 | void get_page_bootmem(unsigned long info, struct page *page, |
140 | unsigned long type) | |
04753278 | 141 | { |
ddffe98d | 142 | page->freelist = (void *)type; |
04753278 YG |
143 | SetPagePrivate(page); |
144 | set_page_private(page, info); | |
fe896d18 | 145 | page_ref_inc(page); |
04753278 YG |
146 | } |
147 | ||
170a5a7e | 148 | void put_page_bootmem(struct page *page) |
04753278 | 149 | { |
5f24ce5f | 150 | unsigned long type; |
04753278 | 151 | |
ddffe98d | 152 | type = (unsigned long) page->freelist; |
5f24ce5f AA |
153 | BUG_ON(type < MEMORY_HOTPLUG_MIN_BOOTMEM_TYPE || |
154 | type > MEMORY_HOTPLUG_MAX_BOOTMEM_TYPE); | |
04753278 | 155 | |
fe896d18 | 156 | if (page_ref_dec_return(page) == 1) { |
ddffe98d | 157 | page->freelist = NULL; |
04753278 YG |
158 | ClearPagePrivate(page); |
159 | set_page_private(page, 0); | |
5f24ce5f | 160 | INIT_LIST_HEAD(&page->lru); |
170a5a7e | 161 | free_reserved_page(page); |
04753278 | 162 | } |
04753278 YG |
163 | } |
164 | ||
46723bfa YI |
165 | #ifdef CONFIG_HAVE_BOOTMEM_INFO_NODE |
166 | #ifndef CONFIG_SPARSEMEM_VMEMMAP | |
d92bc318 | 167 | static void register_page_bootmem_info_section(unsigned long start_pfn) |
04753278 YG |
168 | { |
169 | unsigned long *usemap, mapsize, section_nr, i; | |
170 | struct mem_section *ms; | |
171 | struct page *page, *memmap; | |
172 | ||
04753278 YG |
173 | section_nr = pfn_to_section_nr(start_pfn); |
174 | ms = __nr_to_section(section_nr); | |
175 | ||
176 | /* Get section's memmap address */ | |
177 | memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr); | |
178 | ||
179 | /* | |
180 | * Get page for the memmap's phys address | |
181 | * XXX: need more consideration for sparse_vmemmap... | |
182 | */ | |
183 | page = virt_to_page(memmap); | |
184 | mapsize = sizeof(struct page) * PAGES_PER_SECTION; | |
185 | mapsize = PAGE_ALIGN(mapsize) >> PAGE_SHIFT; | |
186 | ||
187 | /* remember memmap's page */ | |
188 | for (i = 0; i < mapsize; i++, page++) | |
189 | get_page_bootmem(section_nr, page, SECTION_INFO); | |
190 | ||
9ac9322d | 191 | usemap = ms->pageblock_flags; |
04753278 YG |
192 | page = virt_to_page(usemap); |
193 | ||
194 | mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT; | |
195 | ||
196 | for (i = 0; i < mapsize; i++, page++) | |
af370fb8 | 197 | get_page_bootmem(section_nr, page, MIX_SECTION_INFO); |
04753278 YG |
198 | |
199 | } | |
46723bfa YI |
200 | #else /* CONFIG_SPARSEMEM_VMEMMAP */ |
201 | static void register_page_bootmem_info_section(unsigned long start_pfn) | |
202 | { | |
203 | unsigned long *usemap, mapsize, section_nr, i; | |
204 | struct mem_section *ms; | |
205 | struct page *page, *memmap; | |
206 | ||
46723bfa YI |
207 | section_nr = pfn_to_section_nr(start_pfn); |
208 | ms = __nr_to_section(section_nr); | |
209 | ||
210 | memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr); | |
211 | ||
212 | register_page_bootmem_memmap(section_nr, memmap, PAGES_PER_SECTION); | |
213 | ||
9ac9322d | 214 | usemap = ms->pageblock_flags; |
46723bfa YI |
215 | page = virt_to_page(usemap); |
216 | ||
217 | mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT; | |
218 | ||
219 | for (i = 0; i < mapsize; i++, page++) | |
220 | get_page_bootmem(section_nr, page, MIX_SECTION_INFO); | |
221 | } | |
222 | #endif /* !CONFIG_SPARSEMEM_VMEMMAP */ | |
04753278 | 223 | |
7ded384a | 224 | void __init register_page_bootmem_info_node(struct pglist_data *pgdat) |
04753278 YG |
225 | { |
226 | unsigned long i, pfn, end_pfn, nr_pages; | |
227 | int node = pgdat->node_id; | |
228 | struct page *page; | |
04753278 YG |
229 | |
230 | nr_pages = PAGE_ALIGN(sizeof(struct pglist_data)) >> PAGE_SHIFT; | |
231 | page = virt_to_page(pgdat); | |
232 | ||
233 | for (i = 0; i < nr_pages; i++, page++) | |
234 | get_page_bootmem(node, page, NODE_INFO); | |
235 | ||
04753278 | 236 | pfn = pgdat->node_start_pfn; |
c1f19495 | 237 | end_pfn = pgdat_end_pfn(pgdat); |
04753278 | 238 | |
7e9f5eb0 | 239 | /* register section info */ |
f14851af | 240 | for (; pfn < end_pfn; pfn += PAGES_PER_SECTION) { |
241 | /* | |
242 | * Some platforms can assign the same pfn to multiple nodes - on | |
243 | * node0 as well as nodeN. To avoid registering a pfn against | |
244 | * multiple nodes we check that this pfn does not already | |
7e9f5eb0 | 245 | * reside in some other nodes. |
f14851af | 246 | */ |
f65e91df | 247 | if (pfn_valid(pfn) && (early_pfn_to_nid(pfn) == node)) |
f14851af | 248 | register_page_bootmem_info_section(pfn); |
249 | } | |
04753278 | 250 | } |
46723bfa | 251 | #endif /* CONFIG_HAVE_BOOTMEM_INFO_NODE */ |
04753278 | 252 | |
f1dd2cd1 | 253 | static int __meminit __add_section(int nid, unsigned long phys_start_pfn, |
7b73d978 | 254 | struct vmem_altmap *altmap, bool want_memblock) |
3947be19 | 255 | { |
3947be19 DH |
256 | int ret; |
257 | ||
ebd15302 KH |
258 | if (pfn_valid(phys_start_pfn)) |
259 | return -EEXIST; | |
260 | ||
4e0d2e7e | 261 | ret = sparse_add_one_section(nid, phys_start_pfn, altmap); |
3947be19 DH |
262 | if (ret < 0) |
263 | return ret; | |
264 | ||
1b862aec MH |
265 | if (!want_memblock) |
266 | return 0; | |
267 | ||
fc44f7f9 | 268 | return hotplug_memory_register(nid, __pfn_to_section(phys_start_pfn)); |
3947be19 DH |
269 | } |
270 | ||
4edd7cef DR |
271 | /* |
272 | * Reasonably generic function for adding memory. It is | |
273 | * expected that archs that support memory hotplug will | |
274 | * call this function after deciding the zone to which to | |
275 | * add the new pages. | |
276 | */ | |
f1dd2cd1 | 277 | int __ref __add_pages(int nid, unsigned long phys_start_pfn, |
940519f0 | 278 | unsigned long nr_pages, struct mhp_restrictions *restrictions) |
4edd7cef DR |
279 | { |
280 | unsigned long i; | |
281 | int err = 0; | |
282 | int start_sec, end_sec; | |
940519f0 | 283 | struct vmem_altmap *altmap = restrictions->altmap; |
4b94ffdc | 284 | |
4edd7cef DR |
285 | /* during initialize mem_map, align hot-added range to section */ |
286 | start_sec = pfn_to_section_nr(phys_start_pfn); | |
287 | end_sec = pfn_to_section_nr(phys_start_pfn + nr_pages - 1); | |
288 | ||
4b94ffdc DW |
289 | if (altmap) { |
290 | /* | |
291 | * Validate altmap is within bounds of the total request | |
292 | */ | |
293 | if (altmap->base_pfn != phys_start_pfn | |
294 | || vmem_altmap_offset(altmap) > nr_pages) { | |
295 | pr_warn_once("memory add fail, invalid altmap\n"); | |
7cf91a98 JK |
296 | err = -EINVAL; |
297 | goto out; | |
4b94ffdc DW |
298 | } |
299 | altmap->alloc = 0; | |
300 | } | |
301 | ||
4edd7cef | 302 | for (i = start_sec; i <= end_sec; i++) { |
7b73d978 | 303 | err = __add_section(nid, section_nr_to_pfn(i), altmap, |
940519f0 | 304 | restrictions->flags & MHP_MEMBLOCK_API); |
4edd7cef DR |
305 | |
306 | /* | |
307 | * EEXIST is finally dealt with by ioresource collision | |
308 | * check. see add_memory() => register_memory_resource() | |
309 | * Warning will be printed if there is collision. | |
310 | */ | |
311 | if (err && (err != -EEXIST)) | |
312 | break; | |
313 | err = 0; | |
f64ac5e6 | 314 | cond_resched(); |
4edd7cef | 315 | } |
c435a390 | 316 | vmemmap_populate_print_last(); |
7cf91a98 | 317 | out: |
4edd7cef DR |
318 | return err; |
319 | } | |
4edd7cef | 320 | |
815121d2 | 321 | /* find the smallest valid pfn in the range [start_pfn, end_pfn) */ |
d09b0137 | 322 | static unsigned long find_smallest_section_pfn(int nid, struct zone *zone, |
815121d2 YI |
323 | unsigned long start_pfn, |
324 | unsigned long end_pfn) | |
325 | { | |
326 | struct mem_section *ms; | |
327 | ||
328 | for (; start_pfn < end_pfn; start_pfn += PAGES_PER_SECTION) { | |
329 | ms = __pfn_to_section(start_pfn); | |
330 | ||
331 | if (unlikely(!valid_section(ms))) | |
332 | continue; | |
333 | ||
334 | if (unlikely(pfn_to_nid(start_pfn) != nid)) | |
335 | continue; | |
336 | ||
337 | if (zone && zone != page_zone(pfn_to_page(start_pfn))) | |
338 | continue; | |
339 | ||
340 | return start_pfn; | |
341 | } | |
342 | ||
343 | return 0; | |
344 | } | |
345 | ||
346 | /* find the biggest valid pfn in the range [start_pfn, end_pfn). */ | |
d09b0137 | 347 | static unsigned long find_biggest_section_pfn(int nid, struct zone *zone, |
815121d2 YI |
348 | unsigned long start_pfn, |
349 | unsigned long end_pfn) | |
350 | { | |
351 | struct mem_section *ms; | |
352 | unsigned long pfn; | |
353 | ||
354 | /* pfn is the end pfn of a memory section. */ | |
355 | pfn = end_pfn - 1; | |
356 | for (; pfn >= start_pfn; pfn -= PAGES_PER_SECTION) { | |
357 | ms = __pfn_to_section(pfn); | |
358 | ||
359 | if (unlikely(!valid_section(ms))) | |
360 | continue; | |
361 | ||
362 | if (unlikely(pfn_to_nid(pfn) != nid)) | |
363 | continue; | |
364 | ||
365 | if (zone && zone != page_zone(pfn_to_page(pfn))) | |
366 | continue; | |
367 | ||
368 | return pfn; | |
369 | } | |
370 | ||
371 | return 0; | |
372 | } | |
373 | ||
374 | static void shrink_zone_span(struct zone *zone, unsigned long start_pfn, | |
375 | unsigned long end_pfn) | |
376 | { | |
c33bc315 XQ |
377 | unsigned long zone_start_pfn = zone->zone_start_pfn; |
378 | unsigned long z = zone_end_pfn(zone); /* zone_end_pfn namespace clash */ | |
379 | unsigned long zone_end_pfn = z; | |
815121d2 YI |
380 | unsigned long pfn; |
381 | struct mem_section *ms; | |
382 | int nid = zone_to_nid(zone); | |
383 | ||
384 | zone_span_writelock(zone); | |
385 | if (zone_start_pfn == start_pfn) { | |
386 | /* | |
387 | * If the section is smallest section in the zone, it need | |
388 | * shrink zone->zone_start_pfn and zone->zone_spanned_pages. | |
389 | * In this case, we find second smallest valid mem_section | |
390 | * for shrinking zone. | |
391 | */ | |
392 | pfn = find_smallest_section_pfn(nid, zone, end_pfn, | |
393 | zone_end_pfn); | |
394 | if (pfn) { | |
395 | zone->zone_start_pfn = pfn; | |
396 | zone->spanned_pages = zone_end_pfn - pfn; | |
397 | } | |
398 | } else if (zone_end_pfn == end_pfn) { | |
399 | /* | |
400 | * If the section is biggest section in the zone, it need | |
401 | * shrink zone->spanned_pages. | |
402 | * In this case, we find second biggest valid mem_section for | |
403 | * shrinking zone. | |
404 | */ | |
405 | pfn = find_biggest_section_pfn(nid, zone, zone_start_pfn, | |
406 | start_pfn); | |
407 | if (pfn) | |
408 | zone->spanned_pages = pfn - zone_start_pfn + 1; | |
409 | } | |
410 | ||
411 | /* | |
412 | * The section is not biggest or smallest mem_section in the zone, it | |
413 | * only creates a hole in the zone. So in this case, we need not | |
414 | * change the zone. But perhaps, the zone has only hole data. Thus | |
415 | * it check the zone has only hole or not. | |
416 | */ | |
417 | pfn = zone_start_pfn; | |
418 | for (; pfn < zone_end_pfn; pfn += PAGES_PER_SECTION) { | |
419 | ms = __pfn_to_section(pfn); | |
420 | ||
421 | if (unlikely(!valid_section(ms))) | |
422 | continue; | |
423 | ||
424 | if (page_zone(pfn_to_page(pfn)) != zone) | |
425 | continue; | |
426 | ||
427 | /* If the section is current section, it continues the loop */ | |
428 | if (start_pfn == pfn) | |
429 | continue; | |
430 | ||
431 | /* If we find valid section, we have nothing to do */ | |
432 | zone_span_writeunlock(zone); | |
433 | return; | |
434 | } | |
435 | ||
436 | /* The zone has no valid section */ | |
437 | zone->zone_start_pfn = 0; | |
438 | zone->spanned_pages = 0; | |
439 | zone_span_writeunlock(zone); | |
440 | } | |
441 | ||
442 | static void shrink_pgdat_span(struct pglist_data *pgdat, | |
443 | unsigned long start_pfn, unsigned long end_pfn) | |
444 | { | |
83285c72 XQ |
445 | unsigned long pgdat_start_pfn = pgdat->node_start_pfn; |
446 | unsigned long p = pgdat_end_pfn(pgdat); /* pgdat_end_pfn namespace clash */ | |
447 | unsigned long pgdat_end_pfn = p; | |
815121d2 YI |
448 | unsigned long pfn; |
449 | struct mem_section *ms; | |
450 | int nid = pgdat->node_id; | |
451 | ||
452 | if (pgdat_start_pfn == start_pfn) { | |
453 | /* | |
454 | * If the section is smallest section in the pgdat, it need | |
455 | * shrink pgdat->node_start_pfn and pgdat->node_spanned_pages. | |
456 | * In this case, we find second smallest valid mem_section | |
457 | * for shrinking zone. | |
458 | */ | |
459 | pfn = find_smallest_section_pfn(nid, NULL, end_pfn, | |
460 | pgdat_end_pfn); | |
461 | if (pfn) { | |
462 | pgdat->node_start_pfn = pfn; | |
463 | pgdat->node_spanned_pages = pgdat_end_pfn - pfn; | |
464 | } | |
465 | } else if (pgdat_end_pfn == end_pfn) { | |
466 | /* | |
467 | * If the section is biggest section in the pgdat, it need | |
468 | * shrink pgdat->node_spanned_pages. | |
469 | * In this case, we find second biggest valid mem_section for | |
470 | * shrinking zone. | |
471 | */ | |
472 | pfn = find_biggest_section_pfn(nid, NULL, pgdat_start_pfn, | |
473 | start_pfn); | |
474 | if (pfn) | |
475 | pgdat->node_spanned_pages = pfn - pgdat_start_pfn + 1; | |
476 | } | |
477 | ||
478 | /* | |
479 | * If the section is not biggest or smallest mem_section in the pgdat, | |
480 | * it only creates a hole in the pgdat. So in this case, we need not | |
481 | * change the pgdat. | |
482 | * But perhaps, the pgdat has only hole data. Thus it check the pgdat | |
483 | * has only hole or not. | |
484 | */ | |
485 | pfn = pgdat_start_pfn; | |
486 | for (; pfn < pgdat_end_pfn; pfn += PAGES_PER_SECTION) { | |
487 | ms = __pfn_to_section(pfn); | |
488 | ||
489 | if (unlikely(!valid_section(ms))) | |
490 | continue; | |
491 | ||
492 | if (pfn_to_nid(pfn) != nid) | |
493 | continue; | |
494 | ||
495 | /* If the section is current section, it continues the loop */ | |
496 | if (start_pfn == pfn) | |
497 | continue; | |
498 | ||
499 | /* If we find valid section, we have nothing to do */ | |
500 | return; | |
501 | } | |
502 | ||
503 | /* The pgdat has no valid section */ | |
504 | pgdat->node_start_pfn = 0; | |
505 | pgdat->node_spanned_pages = 0; | |
506 | } | |
507 | ||
508 | static void __remove_zone(struct zone *zone, unsigned long start_pfn) | |
509 | { | |
510 | struct pglist_data *pgdat = zone->zone_pgdat; | |
511 | int nr_pages = PAGES_PER_SECTION; | |
815121d2 YI |
512 | unsigned long flags; |
513 | ||
815121d2 YI |
514 | pgdat_resize_lock(zone->zone_pgdat, &flags); |
515 | shrink_zone_span(zone, start_pfn, start_pfn + nr_pages); | |
516 | shrink_pgdat_span(pgdat, start_pfn, start_pfn + nr_pages); | |
517 | pgdat_resize_unlock(zone->zone_pgdat, &flags); | |
518 | } | |
519 | ||
9d1d887d DH |
520 | static void __remove_section(struct zone *zone, struct mem_section *ms, |
521 | unsigned long map_offset, | |
522 | struct vmem_altmap *altmap) | |
ea01ea93 | 523 | { |
815121d2 YI |
524 | unsigned long start_pfn; |
525 | int scn_nr; | |
ea01ea93 | 526 | |
9d1d887d DH |
527 | if (WARN_ON_ONCE(!valid_section(ms))) |
528 | return; | |
ea01ea93 | 529 | |
cb7b3a36 | 530 | unregister_memory_section(ms); |
ea01ea93 | 531 | |
815121d2 | 532 | scn_nr = __section_nr(ms); |
1dd2bfc8 | 533 | start_pfn = section_nr_to_pfn((unsigned long)scn_nr); |
815121d2 YI |
534 | __remove_zone(zone, start_pfn); |
535 | ||
24b6d416 | 536 | sparse_remove_one_section(zone, ms, map_offset, altmap); |
ea01ea93 BP |
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 | */ | |
ac5c9426 DH |
551 | void __remove_pages(struct zone *zone, unsigned long phys_start_pfn, |
552 | unsigned long nr_pages, struct vmem_altmap *altmap) | |
ea01ea93 | 553 | { |
fe74ebb1 | 554 | unsigned long i; |
4b94ffdc | 555 | unsigned long map_offset = 0; |
9d1d887d | 556 | int sections_to_remove; |
4b94ffdc DW |
557 | |
558 | /* In the ZONE_DEVICE case device driver owns the memory region */ | |
514caf23 CH |
559 | if (is_dev_zone(zone)) |
560 | map_offset = vmem_altmap_offset(altmap); | |
ea01ea93 | 561 | |
7cf91a98 JK |
562 | clear_zone_contiguous(zone); |
563 | ||
ea01ea93 BP |
564 | /* |
565 | * We can only remove entire sections | |
566 | */ | |
567 | BUG_ON(phys_start_pfn & ~PAGE_SECTION_MASK); | |
568 | BUG_ON(nr_pages % PAGES_PER_SECTION); | |
569 | ||
ea01ea93 BP |
570 | sections_to_remove = nr_pages / PAGES_PER_SECTION; |
571 | for (i = 0; i < sections_to_remove; i++) { | |
572 | unsigned long pfn = phys_start_pfn + i*PAGES_PER_SECTION; | |
4b94ffdc | 573 | |
dd33ad7b | 574 | cond_resched(); |
9d1d887d DH |
575 | __remove_section(zone, __pfn_to_section(pfn), map_offset, |
576 | altmap); | |
4b94ffdc | 577 | map_offset = 0; |
ea01ea93 | 578 | } |
7cf91a98 JK |
579 | |
580 | set_zone_contiguous(zone); | |
ea01ea93 | 581 | } |
ea01ea93 | 582 | |
9d0ad8ca DK |
583 | int set_online_page_callback(online_page_callback_t callback) |
584 | { | |
585 | int rc = -EINVAL; | |
586 | ||
bfc8c901 VD |
587 | get_online_mems(); |
588 | mutex_lock(&online_page_callback_lock); | |
9d0ad8ca DK |
589 | |
590 | if (online_page_callback == generic_online_page) { | |
591 | online_page_callback = callback; | |
592 | rc = 0; | |
593 | } | |
594 | ||
bfc8c901 VD |
595 | mutex_unlock(&online_page_callback_lock); |
596 | put_online_mems(); | |
9d0ad8ca DK |
597 | |
598 | return rc; | |
599 | } | |
600 | EXPORT_SYMBOL_GPL(set_online_page_callback); | |
601 | ||
602 | int restore_online_page_callback(online_page_callback_t callback) | |
603 | { | |
604 | int rc = -EINVAL; | |
605 | ||
bfc8c901 VD |
606 | get_online_mems(); |
607 | mutex_lock(&online_page_callback_lock); | |
9d0ad8ca DK |
608 | |
609 | if (online_page_callback == callback) { | |
610 | online_page_callback = generic_online_page; | |
611 | rc = 0; | |
612 | } | |
613 | ||
bfc8c901 VD |
614 | mutex_unlock(&online_page_callback_lock); |
615 | put_online_mems(); | |
9d0ad8ca DK |
616 | |
617 | return rc; | |
618 | } | |
619 | EXPORT_SYMBOL_GPL(restore_online_page_callback); | |
620 | ||
621 | void __online_page_set_limits(struct page *page) | |
180c06ef | 622 | { |
9d0ad8ca DK |
623 | } |
624 | EXPORT_SYMBOL_GPL(__online_page_set_limits); | |
625 | ||
626 | void __online_page_increment_counters(struct page *page) | |
627 | { | |
3dcc0571 | 628 | adjust_managed_page_count(page, 1); |
9d0ad8ca DK |
629 | } |
630 | EXPORT_SYMBOL_GPL(__online_page_increment_counters); | |
180c06ef | 631 | |
9d0ad8ca DK |
632 | void __online_page_free(struct page *page) |
633 | { | |
3dcc0571 | 634 | __free_reserved_page(page); |
180c06ef | 635 | } |
9d0ad8ca DK |
636 | EXPORT_SYMBOL_GPL(__online_page_free); |
637 | ||
a9cd410a | 638 | static void generic_online_page(struct page *page, unsigned int order) |
9d0ad8ca | 639 | { |
cd02cf1a | 640 | kernel_map_pages(page, 1 << order, 1); |
a9cd410a AK |
641 | __free_pages_core(page, order); |
642 | totalram_pages_add(1UL << order); | |
643 | #ifdef CONFIG_HIGHMEM | |
644 | if (PageHighMem(page)) | |
645 | totalhigh_pages_add(1UL << order); | |
646 | #endif | |
647 | } | |
648 | ||
649 | static int online_pages_blocks(unsigned long start, unsigned long nr_pages) | |
650 | { | |
651 | unsigned long end = start + nr_pages; | |
652 | int order, onlined_pages = 0; | |
653 | ||
654 | while (start < end) { | |
655 | order = min(MAX_ORDER - 1, | |
656 | get_order(PFN_PHYS(end) - PFN_PHYS(start))); | |
657 | (*online_page_callback)(pfn_to_page(start), order); | |
658 | ||
659 | onlined_pages += (1UL << order); | |
660 | start += (1UL << order); | |
661 | } | |
662 | return onlined_pages; | |
9d0ad8ca | 663 | } |
180c06ef | 664 | |
75884fb1 KH |
665 | static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages, |
666 | void *arg) | |
3947be19 | 667 | { |
75884fb1 | 668 | unsigned long onlined_pages = *(unsigned long *)arg; |
2d070eab | 669 | |
75884fb1 | 670 | if (PageReserved(pfn_to_page(start_pfn))) |
a9cd410a | 671 | onlined_pages += online_pages_blocks(start_pfn, nr_pages); |
2d070eab MH |
672 | |
673 | online_mem_sections(start_pfn, start_pfn + nr_pages); | |
674 | ||
75884fb1 KH |
675 | *(unsigned long *)arg = onlined_pages; |
676 | return 0; | |
677 | } | |
678 | ||
d9713679 LJ |
679 | /* check which state of node_states will be changed when online memory */ |
680 | static void node_states_check_changes_online(unsigned long nr_pages, | |
681 | struct zone *zone, struct memory_notify *arg) | |
682 | { | |
683 | int nid = zone_to_nid(zone); | |
d9713679 | 684 | |
98fa15f3 AK |
685 | arg->status_change_nid = NUMA_NO_NODE; |
686 | arg->status_change_nid_normal = NUMA_NO_NODE; | |
687 | arg->status_change_nid_high = NUMA_NO_NODE; | |
d9713679 | 688 | |
8efe33f4 OS |
689 | if (!node_state(nid, N_MEMORY)) |
690 | arg->status_change_nid = nid; | |
691 | if (zone_idx(zone) <= ZONE_NORMAL && !node_state(nid, N_NORMAL_MEMORY)) | |
d9713679 | 692 | arg->status_change_nid_normal = nid; |
6715ddf9 | 693 | #ifdef CONFIG_HIGHMEM |
d3ba3ae1 | 694 | if (zone_idx(zone) <= ZONE_HIGHMEM && !node_state(nid, N_HIGH_MEMORY)) |
6715ddf9 | 695 | arg->status_change_nid_high = nid; |
6715ddf9 | 696 | #endif |
d9713679 LJ |
697 | } |
698 | ||
699 | static void node_states_set_node(int node, struct memory_notify *arg) | |
700 | { | |
701 | if (arg->status_change_nid_normal >= 0) | |
702 | node_set_state(node, N_NORMAL_MEMORY); | |
703 | ||
6715ddf9 LJ |
704 | if (arg->status_change_nid_high >= 0) |
705 | node_set_state(node, N_HIGH_MEMORY); | |
706 | ||
83d83612 OS |
707 | if (arg->status_change_nid >= 0) |
708 | node_set_state(node, N_MEMORY); | |
d9713679 LJ |
709 | } |
710 | ||
f1dd2cd1 MH |
711 | static void __meminit resize_zone_range(struct zone *zone, unsigned long start_pfn, |
712 | unsigned long nr_pages) | |
713 | { | |
714 | unsigned long old_end_pfn = zone_end_pfn(zone); | |
715 | ||
716 | if (zone_is_empty(zone) || start_pfn < zone->zone_start_pfn) | |
717 | zone->zone_start_pfn = start_pfn; | |
718 | ||
719 | zone->spanned_pages = max(start_pfn + nr_pages, old_end_pfn) - zone->zone_start_pfn; | |
720 | } | |
721 | ||
722 | static void __meminit resize_pgdat_range(struct pglist_data *pgdat, unsigned long start_pfn, | |
723 | unsigned long nr_pages) | |
724 | { | |
725 | unsigned long old_end_pfn = pgdat_end_pfn(pgdat); | |
726 | ||
727 | if (!pgdat->node_spanned_pages || start_pfn < pgdat->node_start_pfn) | |
728 | pgdat->node_start_pfn = start_pfn; | |
729 | ||
730 | pgdat->node_spanned_pages = max(start_pfn + nr_pages, old_end_pfn) - pgdat->node_start_pfn; | |
731 | } | |
732 | ||
a99583e7 CH |
733 | void __ref move_pfn_range_to_zone(struct zone *zone, unsigned long start_pfn, |
734 | unsigned long nr_pages, struct vmem_altmap *altmap) | |
f1dd2cd1 MH |
735 | { |
736 | struct pglist_data *pgdat = zone->zone_pgdat; | |
737 | int nid = pgdat->node_id; | |
738 | unsigned long flags; | |
df429ac0 | 739 | |
f1dd2cd1 MH |
740 | clear_zone_contiguous(zone); |
741 | ||
742 | /* TODO Huh pgdat is irqsave while zone is not. It used to be like that before */ | |
743 | pgdat_resize_lock(pgdat, &flags); | |
744 | zone_span_writelock(zone); | |
fa004ab7 WY |
745 | if (zone_is_empty(zone)) |
746 | init_currently_empty_zone(zone, start_pfn, nr_pages); | |
f1dd2cd1 MH |
747 | resize_zone_range(zone, start_pfn, nr_pages); |
748 | zone_span_writeunlock(zone); | |
749 | resize_pgdat_range(pgdat, start_pfn, nr_pages); | |
750 | pgdat_resize_unlock(pgdat, &flags); | |
751 | ||
752 | /* | |
753 | * TODO now we have a visible range of pages which are not associated | |
754 | * with their zone properly. Not nice but set_pfnblock_flags_mask | |
755 | * expects the zone spans the pfn range. All the pages in the range | |
756 | * are reserved so nobody should be touching them so we should be safe | |
757 | */ | |
a99583e7 CH |
758 | memmap_init_zone(nr_pages, nid, zone_idx(zone), start_pfn, |
759 | MEMMAP_HOTPLUG, altmap); | |
f1dd2cd1 MH |
760 | |
761 | set_zone_contiguous(zone); | |
762 | } | |
763 | ||
c246a213 MH |
764 | /* |
765 | * Returns a default kernel memory zone for the given pfn range. | |
766 | * If no kernel zone covers this pfn range it will automatically go | |
767 | * to the ZONE_NORMAL. | |
768 | */ | |
c6f03e29 | 769 | static struct zone *default_kernel_zone_for_pfn(int nid, unsigned long start_pfn, |
c246a213 MH |
770 | unsigned long nr_pages) |
771 | { | |
772 | struct pglist_data *pgdat = NODE_DATA(nid); | |
773 | int zid; | |
774 | ||
775 | for (zid = 0; zid <= ZONE_NORMAL; zid++) { | |
776 | struct zone *zone = &pgdat->node_zones[zid]; | |
777 | ||
778 | if (zone_intersects(zone, start_pfn, nr_pages)) | |
779 | return zone; | |
780 | } | |
781 | ||
782 | return &pgdat->node_zones[ZONE_NORMAL]; | |
783 | } | |
784 | ||
c6f03e29 MH |
785 | static inline struct zone *default_zone_for_pfn(int nid, unsigned long start_pfn, |
786 | unsigned long nr_pages) | |
e5e68930 | 787 | { |
c6f03e29 MH |
788 | struct zone *kernel_zone = default_kernel_zone_for_pfn(nid, start_pfn, |
789 | nr_pages); | |
790 | struct zone *movable_zone = &NODE_DATA(nid)->node_zones[ZONE_MOVABLE]; | |
791 | bool in_kernel = zone_intersects(kernel_zone, start_pfn, nr_pages); | |
792 | bool in_movable = zone_intersects(movable_zone, start_pfn, nr_pages); | |
e5e68930 MH |
793 | |
794 | /* | |
c6f03e29 MH |
795 | * We inherit the existing zone in a simple case where zones do not |
796 | * overlap in the given range | |
e5e68930 | 797 | */ |
c6f03e29 MH |
798 | if (in_kernel ^ in_movable) |
799 | return (in_kernel) ? kernel_zone : movable_zone; | |
9f123ab5 | 800 | |
c6f03e29 MH |
801 | /* |
802 | * If the range doesn't belong to any zone or two zones overlap in the | |
803 | * given range then we use movable zone only if movable_node is | |
804 | * enabled because we always online to a kernel zone by default. | |
805 | */ | |
806 | return movable_node_enabled ? movable_zone : kernel_zone; | |
9f123ab5 MH |
807 | } |
808 | ||
e5e68930 MH |
809 | struct zone * zone_for_pfn_range(int online_type, int nid, unsigned start_pfn, |
810 | unsigned long nr_pages) | |
f1dd2cd1 | 811 | { |
c6f03e29 MH |
812 | if (online_type == MMOP_ONLINE_KERNEL) |
813 | return default_kernel_zone_for_pfn(nid, start_pfn, nr_pages); | |
f1dd2cd1 | 814 | |
c6f03e29 MH |
815 | if (online_type == MMOP_ONLINE_MOVABLE) |
816 | return &NODE_DATA(nid)->node_zones[ZONE_MOVABLE]; | |
df429ac0 | 817 | |
c6f03e29 | 818 | return default_zone_for_pfn(nid, start_pfn, nr_pages); |
e5e68930 MH |
819 | } |
820 | ||
821 | /* | |
822 | * Associates the given pfn range with the given node and the zone appropriate | |
823 | * for the given online type. | |
824 | */ | |
825 | static struct zone * __meminit move_pfn_range(int online_type, int nid, | |
826 | unsigned long start_pfn, unsigned long nr_pages) | |
827 | { | |
828 | struct zone *zone; | |
829 | ||
830 | zone = zone_for_pfn_range(online_type, nid, start_pfn, nr_pages); | |
a99583e7 | 831 | move_pfn_range_to_zone(zone, start_pfn, nr_pages, NULL); |
f1dd2cd1 | 832 | return zone; |
df429ac0 | 833 | } |
75884fb1 | 834 | |
511c2aba | 835 | int __ref online_pages(unsigned long pfn, unsigned long nr_pages, int online_type) |
75884fb1 | 836 | { |
aa47228a | 837 | unsigned long flags; |
3947be19 DH |
838 | unsigned long onlined_pages = 0; |
839 | struct zone *zone; | |
6811378e | 840 | int need_zonelists_rebuild = 0; |
7b78d335 YG |
841 | int nid; |
842 | int ret; | |
843 | struct memory_notify arg; | |
d0dc12e8 PT |
844 | struct memory_block *mem; |
845 | ||
381eab4a DH |
846 | mem_hotplug_begin(); |
847 | ||
d0dc12e8 PT |
848 | /* |
849 | * We can't use pfn_to_nid() because nid might be stored in struct page | |
850 | * which is not yet initialized. Instead, we find nid from memory block. | |
851 | */ | |
852 | mem = find_memory_block(__pfn_to_section(pfn)); | |
853 | nid = mem->nid; | |
89c02e69 | 854 | put_device(&mem->dev); |
7b78d335 | 855 | |
f1dd2cd1 MH |
856 | /* associate pfn range with the zone */ |
857 | zone = move_pfn_range(online_type, nid, pfn, nr_pages); | |
858 | ||
7b78d335 YG |
859 | arg.start_pfn = pfn; |
860 | arg.nr_pages = nr_pages; | |
d9713679 | 861 | node_states_check_changes_online(nr_pages, zone, &arg); |
7b78d335 | 862 | |
7b78d335 YG |
863 | ret = memory_notify(MEM_GOING_ONLINE, &arg); |
864 | ret = notifier_to_errno(ret); | |
e33e33b4 CY |
865 | if (ret) |
866 | goto failed_addition; | |
867 | ||
6811378e YG |
868 | /* |
869 | * If this zone is not populated, then it is not in zonelist. | |
870 | * This means the page allocator ignores this zone. | |
871 | * So, zonelist must be updated after online. | |
872 | */ | |
6dcd73d7 | 873 | if (!populated_zone(zone)) { |
6811378e | 874 | need_zonelists_rebuild = 1; |
72675e13 | 875 | setup_zone_pageset(zone); |
6dcd73d7 | 876 | } |
6811378e | 877 | |
908eedc6 | 878 | ret = walk_system_ram_range(pfn, nr_pages, &onlined_pages, |
75884fb1 | 879 | online_pages_range); |
fd8a4221 | 880 | if (ret) { |
6dcd73d7 WC |
881 | if (need_zonelists_rebuild) |
882 | zone_pcp_reset(zone); | |
e33e33b4 | 883 | goto failed_addition; |
fd8a4221 GL |
884 | } |
885 | ||
3947be19 | 886 | zone->present_pages += onlined_pages; |
aa47228a CS |
887 | |
888 | pgdat_resize_lock(zone->zone_pgdat, &flags); | |
f2937be5 | 889 | zone->zone_pgdat->node_present_pages += onlined_pages; |
aa47228a CS |
890 | pgdat_resize_unlock(zone->zone_pgdat, &flags); |
891 | ||
e900a918 DW |
892 | shuffle_zone(zone); |
893 | ||
08dff7b7 | 894 | if (onlined_pages) { |
e888ca35 | 895 | node_states_set_node(nid, &arg); |
08dff7b7 | 896 | if (need_zonelists_rebuild) |
72675e13 | 897 | build_all_zonelists(NULL); |
08dff7b7 JL |
898 | else |
899 | zone_pcp_update(zone); | |
900 | } | |
3947be19 | 901 | |
1b79acc9 KM |
902 | init_per_zone_wmark_min(); |
903 | ||
698b1b30 | 904 | if (onlined_pages) { |
e888ca35 | 905 | kswapd_run(nid); |
698b1b30 VB |
906 | kcompactd_run(nid); |
907 | } | |
61b13993 | 908 | |
1f522509 | 909 | vm_total_pages = nr_free_pagecache_pages(); |
2f7f24ec | 910 | |
2d1d43f6 | 911 | writeback_set_ratelimit(); |
7b78d335 YG |
912 | |
913 | if (onlined_pages) | |
914 | memory_notify(MEM_ONLINE, &arg); | |
381eab4a | 915 | mem_hotplug_done(); |
30467e0b | 916 | return 0; |
e33e33b4 CY |
917 | |
918 | failed_addition: | |
919 | pr_debug("online_pages [mem %#010llx-%#010llx] failed\n", | |
920 | (unsigned long long) pfn << PAGE_SHIFT, | |
921 | (((unsigned long long) pfn + nr_pages) << PAGE_SHIFT) - 1); | |
922 | memory_notify(MEM_CANCEL_ONLINE, &arg); | |
381eab4a | 923 | mem_hotplug_done(); |
e33e33b4 | 924 | return ret; |
3947be19 | 925 | } |
53947027 | 926 | #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */ |
bc02af93 | 927 | |
0bd85420 TC |
928 | static void reset_node_present_pages(pg_data_t *pgdat) |
929 | { | |
930 | struct zone *z; | |
931 | ||
932 | for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++) | |
933 | z->present_pages = 0; | |
934 | ||
935 | pgdat->node_present_pages = 0; | |
936 | } | |
937 | ||
e1319331 HS |
938 | /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */ |
939 | static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start) | |
9af3c2de YG |
940 | { |
941 | struct pglist_data *pgdat; | |
c8e861a5 | 942 | unsigned long start_pfn = PFN_DOWN(start); |
9af3c2de | 943 | |
a1e565aa TC |
944 | pgdat = NODE_DATA(nid); |
945 | if (!pgdat) { | |
946 | pgdat = arch_alloc_nodedata(nid); | |
947 | if (!pgdat) | |
948 | return NULL; | |
9af3c2de | 949 | |
a1e565aa | 950 | arch_refresh_nodedata(nid, pgdat); |
b0dc3a34 | 951 | } else { |
e716f2eb MG |
952 | /* |
953 | * Reset the nr_zones, order and classzone_idx before reuse. | |
954 | * Note that kswapd will init kswapd_classzone_idx properly | |
955 | * when it starts in the near future. | |
956 | */ | |
b0dc3a34 | 957 | pgdat->nr_zones = 0; |
38087d9b MG |
958 | pgdat->kswapd_order = 0; |
959 | pgdat->kswapd_classzone_idx = 0; | |
a1e565aa | 960 | } |
9af3c2de YG |
961 | |
962 | /* we can use NODE_DATA(nid) from here */ | |
963 | ||
03e85f9d OS |
964 | pgdat->node_id = nid; |
965 | pgdat->node_start_pfn = start_pfn; | |
966 | ||
9af3c2de | 967 | /* init node's zones as empty zones, we don't have any present pages.*/ |
03e85f9d | 968 | free_area_init_core_hotplug(nid); |
5830169f | 969 | pgdat->per_cpu_nodestats = alloc_percpu(struct per_cpu_nodestat); |
9af3c2de | 970 | |
959ecc48 KH |
971 | /* |
972 | * The node we allocated has no zone fallback lists. For avoiding | |
973 | * to access not-initialized zonelist, build here. | |
974 | */ | |
72675e13 | 975 | build_all_zonelists(pgdat); |
959ecc48 | 976 | |
0bd85420 TC |
977 | /* |
978 | * When memory is hot-added, all the memory is in offline state. So | |
979 | * clear all zones' present_pages because they will be updated in | |
980 | * online_pages() and offline_pages(). | |
981 | */ | |
03e85f9d | 982 | reset_node_managed_pages(pgdat); |
0bd85420 TC |
983 | reset_node_present_pages(pgdat); |
984 | ||
9af3c2de YG |
985 | return pgdat; |
986 | } | |
987 | ||
b9ff0360 | 988 | static void rollback_node_hotadd(int nid) |
9af3c2de | 989 | { |
b9ff0360 OS |
990 | pg_data_t *pgdat = NODE_DATA(nid); |
991 | ||
9af3c2de | 992 | arch_refresh_nodedata(nid, NULL); |
5830169f | 993 | free_percpu(pgdat->per_cpu_nodestats); |
9af3c2de YG |
994 | arch_free_nodedata(pgdat); |
995 | return; | |
996 | } | |
997 | ||
0a547039 | 998 | |
01b0f197 TK |
999 | /** |
1000 | * try_online_node - online a node if offlined | |
e8b098fc | 1001 | * @nid: the node ID |
b9ff0360 OS |
1002 | * @start: start addr of the node |
1003 | * @set_node_online: Whether we want to online the node | |
cf23422b | 1004 | * called by cpu_up() to online a node without onlined memory. |
b9ff0360 OS |
1005 | * |
1006 | * Returns: | |
1007 | * 1 -> a new node has been allocated | |
1008 | * 0 -> the node is already online | |
1009 | * -ENOMEM -> the node could not be allocated | |
cf23422b | 1010 | */ |
b9ff0360 | 1011 | static int __try_online_node(int nid, u64 start, bool set_node_online) |
cf23422b | 1012 | { |
b9ff0360 OS |
1013 | pg_data_t *pgdat; |
1014 | int ret = 1; | |
cf23422b | 1015 | |
01b0f197 TK |
1016 | if (node_online(nid)) |
1017 | return 0; | |
1018 | ||
b9ff0360 | 1019 | pgdat = hotadd_new_pgdat(nid, start); |
7553e8f2 | 1020 | if (!pgdat) { |
01b0f197 | 1021 | pr_err("Cannot online node %d due to NULL pgdat\n", nid); |
cf23422b | 1022 | ret = -ENOMEM; |
1023 | goto out; | |
1024 | } | |
b9ff0360 OS |
1025 | |
1026 | if (set_node_online) { | |
1027 | node_set_online(nid); | |
1028 | ret = register_one_node(nid); | |
1029 | BUG_ON(ret); | |
1030 | } | |
cf23422b | 1031 | out: |
b9ff0360 OS |
1032 | return ret; |
1033 | } | |
1034 | ||
1035 | /* | |
1036 | * Users of this function always want to online/register the node | |
1037 | */ | |
1038 | int try_online_node(int nid) | |
1039 | { | |
1040 | int ret; | |
1041 | ||
1042 | mem_hotplug_begin(); | |
1043 | ret = __try_online_node(nid, 0, true); | |
bfc8c901 | 1044 | mem_hotplug_done(); |
cf23422b | 1045 | return ret; |
1046 | } | |
1047 | ||
27356f54 TK |
1048 | static int check_hotplug_memory_range(u64 start, u64 size) |
1049 | { | |
ba325585 | 1050 | /* memory range must be block size aligned */ |
cec3ebd0 DH |
1051 | if (!size || !IS_ALIGNED(start, memory_block_size_bytes()) || |
1052 | !IS_ALIGNED(size, memory_block_size_bytes())) { | |
ba325585 | 1053 | pr_err("Block size [%#lx] unaligned hotplug range: start %#llx, size %#llx", |
cec3ebd0 | 1054 | memory_block_size_bytes(), start, size); |
27356f54 TK |
1055 | return -EINVAL; |
1056 | } | |
1057 | ||
1058 | return 0; | |
1059 | } | |
1060 | ||
31bc3858 VK |
1061 | static int online_memory_block(struct memory_block *mem, void *arg) |
1062 | { | |
dc18d706 | 1063 | return device_online(&mem->dev); |
31bc3858 VK |
1064 | } |
1065 | ||
8df1d0e4 DH |
1066 | /* |
1067 | * NOTE: The caller must call lock_device_hotplug() to serialize hotplug | |
1068 | * and online/offline operations (triggered e.g. by sysfs). | |
1069 | * | |
1070 | * we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG | |
1071 | */ | |
f29d8e9c | 1072 | int __ref add_memory_resource(int nid, struct resource *res) |
bc02af93 | 1073 | { |
940519f0 MH |
1074 | struct mhp_restrictions restrictions = { |
1075 | .flags = MHP_MEMBLOCK_API, | |
1076 | }; | |
62cedb9f | 1077 | u64 start, size; |
b9ff0360 | 1078 | bool new_node = false; |
bc02af93 YG |
1079 | int ret; |
1080 | ||
62cedb9f DV |
1081 | start = res->start; |
1082 | size = resource_size(res); | |
1083 | ||
27356f54 TK |
1084 | ret = check_hotplug_memory_range(start, size); |
1085 | if (ret) | |
1086 | return ret; | |
1087 | ||
bfc8c901 | 1088 | mem_hotplug_begin(); |
ac13c462 | 1089 | |
7f36e3e5 TC |
1090 | /* |
1091 | * Add new range to memblock so that when hotadd_new_pgdat() is called | |
1092 | * to allocate new pgdat, get_pfn_range_for_nid() will be able to find | |
1093 | * this new range and calculate total pages correctly. The range will | |
1094 | * be removed at hot-remove time. | |
1095 | */ | |
1096 | memblock_add_node(start, size, nid); | |
1097 | ||
b9ff0360 OS |
1098 | ret = __try_online_node(nid, start, false); |
1099 | if (ret < 0) | |
1100 | goto error; | |
1101 | new_node = ret; | |
9af3c2de | 1102 | |
bc02af93 | 1103 | /* call arch's memory hotadd */ |
940519f0 | 1104 | ret = arch_add_memory(nid, start, size, &restrictions); |
9af3c2de YG |
1105 | if (ret < 0) |
1106 | goto error; | |
1107 | ||
a1e565aa | 1108 | if (new_node) { |
d5b6f6a3 | 1109 | /* If sysfs file of new node can't be created, cpu on the node |
0fc44159 YG |
1110 | * can't be hot-added. There is no rollback way now. |
1111 | * So, check by BUG_ON() to catch it reluctantly.. | |
d5b6f6a3 | 1112 | * We online node here. We can't roll back from here. |
0fc44159 | 1113 | */ |
d5b6f6a3 OS |
1114 | node_set_online(nid); |
1115 | ret = __register_one_node(nid); | |
0fc44159 YG |
1116 | BUG_ON(ret); |
1117 | } | |
1118 | ||
d5b6f6a3 | 1119 | /* link memory sections under this node.*/ |
4fbce633 | 1120 | ret = link_mem_sections(nid, PFN_DOWN(start), PFN_UP(start + size - 1)); |
d5b6f6a3 OS |
1121 | BUG_ON(ret); |
1122 | ||
d96ae530 | 1123 | /* create new memmap entry */ |
1124 | firmware_map_add_hotplug(start, start + size, "System RAM"); | |
1125 | ||
381eab4a DH |
1126 | /* device_online() will take the lock when calling online_pages() */ |
1127 | mem_hotplug_done(); | |
1128 | ||
31bc3858 | 1129 | /* online pages if requested */ |
f29d8e9c | 1130 | if (memhp_auto_online) |
31bc3858 VK |
1131 | walk_memory_range(PFN_DOWN(start), PFN_UP(start + size - 1), |
1132 | NULL, online_memory_block); | |
1133 | ||
381eab4a | 1134 | return ret; |
9af3c2de YG |
1135 | error: |
1136 | /* rollback pgdat allocation and others */ | |
b9ff0360 OS |
1137 | if (new_node) |
1138 | rollback_node_hotadd(nid); | |
7f36e3e5 | 1139 | memblock_remove(start, size); |
bfc8c901 | 1140 | mem_hotplug_done(); |
bc02af93 YG |
1141 | return ret; |
1142 | } | |
62cedb9f | 1143 | |
8df1d0e4 DH |
1144 | /* requires device_hotplug_lock, see add_memory_resource() */ |
1145 | int __ref __add_memory(int nid, u64 start, u64 size) | |
62cedb9f DV |
1146 | { |
1147 | struct resource *res; | |
1148 | int ret; | |
1149 | ||
1150 | res = register_memory_resource(start, size); | |
6f754ba4 VK |
1151 | if (IS_ERR(res)) |
1152 | return PTR_ERR(res); | |
62cedb9f | 1153 | |
f29d8e9c | 1154 | ret = add_memory_resource(nid, res); |
62cedb9f DV |
1155 | if (ret < 0) |
1156 | release_memory_resource(res); | |
1157 | return ret; | |
1158 | } | |
8df1d0e4 DH |
1159 | |
1160 | int add_memory(int nid, u64 start, u64 size) | |
1161 | { | |
1162 | int rc; | |
1163 | ||
1164 | lock_device_hotplug(); | |
1165 | rc = __add_memory(nid, start, size); | |
1166 | unlock_device_hotplug(); | |
1167 | ||
1168 | return rc; | |
1169 | } | |
bc02af93 | 1170 | EXPORT_SYMBOL_GPL(add_memory); |
0c0e6195 KH |
1171 | |
1172 | #ifdef CONFIG_MEMORY_HOTREMOVE | |
5c755e9f BP |
1173 | /* |
1174 | * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy | |
1175 | * set and the size of the free page is given by page_order(). Using this, | |
1176 | * the function determines if the pageblock contains only free pages. | |
1177 | * Due to buddy contraints, a free page at least the size of a pageblock will | |
1178 | * be located at the start of the pageblock | |
1179 | */ | |
1180 | static inline int pageblock_free(struct page *page) | |
1181 | { | |
1182 | return PageBuddy(page) && page_order(page) >= pageblock_order; | |
1183 | } | |
1184 | ||
891cb2a7 MH |
1185 | /* Return the pfn of the start of the next active pageblock after a given pfn */ |
1186 | static unsigned long next_active_pageblock(unsigned long pfn) | |
5c755e9f | 1187 | { |
891cb2a7 MH |
1188 | struct page *page = pfn_to_page(pfn); |
1189 | ||
5c755e9f | 1190 | /* Ensure the starting page is pageblock-aligned */ |
891cb2a7 | 1191 | BUG_ON(pfn & (pageblock_nr_pages - 1)); |
5c755e9f | 1192 | |
5c755e9f | 1193 | /* If the entire pageblock is free, move to the end of free page */ |
0dcc48c1 KH |
1194 | if (pageblock_free(page)) { |
1195 | int order; | |
1196 | /* be careful. we don't have locks, page_order can be changed.*/ | |
1197 | order = page_order(page); | |
1198 | if ((order < MAX_ORDER) && (order >= pageblock_order)) | |
891cb2a7 | 1199 | return pfn + (1 << order); |
0dcc48c1 | 1200 | } |
5c755e9f | 1201 | |
891cb2a7 | 1202 | return pfn + pageblock_nr_pages; |
5c755e9f BP |
1203 | } |
1204 | ||
891cb2a7 | 1205 | static bool is_pageblock_removable_nolock(unsigned long pfn) |
fb52bbae | 1206 | { |
891cb2a7 | 1207 | struct page *page = pfn_to_page(pfn); |
fb52bbae | 1208 | struct zone *zone; |
fb52bbae MM |
1209 | |
1210 | /* | |
1211 | * We have to be careful here because we are iterating over memory | |
1212 | * sections which are not zone aware so we might end up outside of | |
1213 | * the zone but still within the section. | |
1214 | * We have to take care about the node as well. If the node is offline | |
1215 | * its NODE_DATA will be NULL - see page_zone. | |
1216 | */ | |
1217 | if (!node_online(page_to_nid(page))) | |
1218 | return false; | |
1219 | ||
1220 | zone = page_zone(page); | |
1221 | pfn = page_to_pfn(page); | |
1222 | if (!zone_spans_pfn(zone, pfn)) | |
1223 | return false; | |
1224 | ||
d381c547 | 1225 | return !has_unmovable_pages(zone, page, 0, MIGRATE_MOVABLE, SKIP_HWPOISON); |
fb52bbae MM |
1226 | } |
1227 | ||
5c755e9f | 1228 | /* Checks if this range of memory is likely to be hot-removable. */ |
c98940f6 | 1229 | bool is_mem_section_removable(unsigned long start_pfn, unsigned long nr_pages) |
5c755e9f | 1230 | { |
891cb2a7 MH |
1231 | unsigned long end_pfn, pfn; |
1232 | ||
1233 | end_pfn = min(start_pfn + nr_pages, | |
1234 | zone_end_pfn(page_zone(pfn_to_page(start_pfn)))); | |
5c755e9f BP |
1235 | |
1236 | /* Check the starting page of each pageblock within the range */ | |
891cb2a7 MH |
1237 | for (pfn = start_pfn; pfn < end_pfn; pfn = next_active_pageblock(pfn)) { |
1238 | if (!is_pageblock_removable_nolock(pfn)) | |
c98940f6 | 1239 | return false; |
49ac8255 | 1240 | cond_resched(); |
5c755e9f BP |
1241 | } |
1242 | ||
1243 | /* All pageblocks in the memory block are likely to be hot-removable */ | |
c98940f6 | 1244 | return true; |
5c755e9f BP |
1245 | } |
1246 | ||
0c0e6195 | 1247 | /* |
deb88a2a | 1248 | * Confirm all pages in a range [start, end) belong to the same zone. |
a96dfddb | 1249 | * When true, return its valid [start, end). |
0c0e6195 | 1250 | */ |
a96dfddb TK |
1251 | int test_pages_in_a_zone(unsigned long start_pfn, unsigned long end_pfn, |
1252 | unsigned long *valid_start, unsigned long *valid_end) | |
0c0e6195 | 1253 | { |
5f0f2887 | 1254 | unsigned long pfn, sec_end_pfn; |
a96dfddb | 1255 | unsigned long start, end; |
0c0e6195 KH |
1256 | struct zone *zone = NULL; |
1257 | struct page *page; | |
1258 | int i; | |
deb88a2a | 1259 | for (pfn = start_pfn, sec_end_pfn = SECTION_ALIGN_UP(start_pfn + 1); |
0c0e6195 | 1260 | pfn < end_pfn; |
deb88a2a | 1261 | pfn = sec_end_pfn, sec_end_pfn += PAGES_PER_SECTION) { |
5f0f2887 AB |
1262 | /* Make sure the memory section is present first */ |
1263 | if (!present_section_nr(pfn_to_section_nr(pfn))) | |
0c0e6195 | 1264 | continue; |
5f0f2887 AB |
1265 | for (; pfn < sec_end_pfn && pfn < end_pfn; |
1266 | pfn += MAX_ORDER_NR_PAGES) { | |
1267 | i = 0; | |
1268 | /* This is just a CONFIG_HOLES_IN_ZONE check.*/ | |
1269 | while ((i < MAX_ORDER_NR_PAGES) && | |
1270 | !pfn_valid_within(pfn + i)) | |
1271 | i++; | |
d6d8c8a4 | 1272 | if (i == MAX_ORDER_NR_PAGES || pfn + i >= end_pfn) |
5f0f2887 | 1273 | continue; |
24feb47c MZ |
1274 | /* Check if we got outside of the zone */ |
1275 | if (zone && !zone_spans_pfn(zone, pfn + i)) | |
1276 | return 0; | |
5f0f2887 AB |
1277 | page = pfn_to_page(pfn + i); |
1278 | if (zone && page_zone(page) != zone) | |
1279 | return 0; | |
a96dfddb TK |
1280 | if (!zone) |
1281 | start = pfn + i; | |
5f0f2887 | 1282 | zone = page_zone(page); |
a96dfddb | 1283 | end = pfn + MAX_ORDER_NR_PAGES; |
5f0f2887 | 1284 | } |
0c0e6195 | 1285 | } |
deb88a2a | 1286 | |
a96dfddb TK |
1287 | if (zone) { |
1288 | *valid_start = start; | |
d6d8c8a4 | 1289 | *valid_end = min(end, end_pfn); |
deb88a2a | 1290 | return 1; |
a96dfddb | 1291 | } else { |
deb88a2a | 1292 | return 0; |
a96dfddb | 1293 | } |
0c0e6195 KH |
1294 | } |
1295 | ||
1296 | /* | |
0efadf48 YX |
1297 | * Scan pfn range [start,end) to find movable/migratable pages (LRU pages, |
1298 | * non-lru movable pages and hugepages). We scan pfn because it's much | |
1299 | * easier than scanning over linked list. This function returns the pfn | |
1300 | * of the first found movable page if it's found, otherwise 0. | |
0c0e6195 | 1301 | */ |
c8721bbb | 1302 | static unsigned long scan_movable_pages(unsigned long start, unsigned long end) |
0c0e6195 KH |
1303 | { |
1304 | unsigned long pfn; | |
eeb0efd0 | 1305 | |
0c0e6195 | 1306 | for (pfn = start; pfn < end; pfn++) { |
eeb0efd0 OS |
1307 | struct page *page, *head; |
1308 | unsigned long skip; | |
1309 | ||
1310 | if (!pfn_valid(pfn)) | |
1311 | continue; | |
1312 | page = pfn_to_page(pfn); | |
1313 | if (PageLRU(page)) | |
1314 | return pfn; | |
1315 | if (__PageMovable(page)) | |
1316 | return pfn; | |
1317 | ||
1318 | if (!PageHuge(page)) | |
1319 | continue; | |
1320 | head = compound_head(page); | |
39186cbe | 1321 | if (page_huge_active(head)) |
eeb0efd0 OS |
1322 | return pfn; |
1323 | skip = (1 << compound_order(head)) - (page - head); | |
1324 | pfn += skip - 1; | |
0c0e6195 KH |
1325 | } |
1326 | return 0; | |
1327 | } | |
1328 | ||
666feb21 | 1329 | static struct page *new_node_page(struct page *page, unsigned long private) |
394e31d2 | 1330 | { |
394e31d2 | 1331 | int nid = page_to_nid(page); |
231e97e2 | 1332 | nodemask_t nmask = node_states[N_MEMORY]; |
7f252f27 MH |
1333 | |
1334 | /* | |
1335 | * try to allocate from a different node but reuse this node if there | |
1336 | * are no other online nodes to be used (e.g. we are offlining a part | |
1337 | * of the only existing node) | |
1338 | */ | |
1339 | node_clear(nid, nmask); | |
1340 | if (nodes_empty(nmask)) | |
1341 | node_set(nid, nmask); | |
394e31d2 | 1342 | |
8b913238 | 1343 | return new_page_nodemask(page, nid, &nmask); |
394e31d2 XQ |
1344 | } |
1345 | ||
0c0e6195 KH |
1346 | static int |
1347 | do_migrate_range(unsigned long start_pfn, unsigned long end_pfn) | |
1348 | { | |
1349 | unsigned long pfn; | |
1350 | struct page *page; | |
0c0e6195 KH |
1351 | int ret = 0; |
1352 | LIST_HEAD(source); | |
1353 | ||
a85009c3 | 1354 | for (pfn = start_pfn; pfn < end_pfn; pfn++) { |
0c0e6195 KH |
1355 | if (!pfn_valid(pfn)) |
1356 | continue; | |
1357 | page = pfn_to_page(pfn); | |
c8721bbb NH |
1358 | |
1359 | if (PageHuge(page)) { | |
1360 | struct page *head = compound_head(page); | |
daf3538a OS |
1361 | pfn = page_to_pfn(head) + (1<<compound_order(head)) - 1; |
1362 | isolate_huge_page(head, &source); | |
c8721bbb | 1363 | continue; |
94723aaf | 1364 | } else if (PageTransHuge(page)) |
8135d892 NH |
1365 | pfn = page_to_pfn(compound_head(page)) |
1366 | + hpage_nr_pages(page) - 1; | |
c8721bbb | 1367 | |
b15c8726 MH |
1368 | /* |
1369 | * HWPoison pages have elevated reference counts so the migration would | |
1370 | * fail on them. It also doesn't make any sense to migrate them in the | |
1371 | * first place. Still try to unmap such a page in case it is still mapped | |
1372 | * (e.g. current hwpoison implementation doesn't unmap KSM pages but keep | |
1373 | * the unmap as the catch all safety net). | |
1374 | */ | |
1375 | if (PageHWPoison(page)) { | |
1376 | if (WARN_ON(PageLRU(page))) | |
1377 | isolate_lru_page(page); | |
1378 | if (page_mapped(page)) | |
1379 | try_to_unmap(page, TTU_IGNORE_MLOCK | TTU_IGNORE_ACCESS); | |
1380 | continue; | |
1381 | } | |
1382 | ||
700c2a46 | 1383 | if (!get_page_unless_zero(page)) |
0c0e6195 KH |
1384 | continue; |
1385 | /* | |
0efadf48 YX |
1386 | * We can skip free pages. And we can deal with pages on |
1387 | * LRU and non-lru movable pages. | |
0c0e6195 | 1388 | */ |
0efadf48 YX |
1389 | if (PageLRU(page)) |
1390 | ret = isolate_lru_page(page); | |
1391 | else | |
1392 | ret = isolate_movable_page(page, ISOLATE_UNEVICTABLE); | |
0c0e6195 | 1393 | if (!ret) { /* Success */ |
62695a84 | 1394 | list_add_tail(&page->lru, &source); |
0efadf48 YX |
1395 | if (!__PageMovable(page)) |
1396 | inc_node_page_state(page, NR_ISOLATED_ANON + | |
1397 | page_is_file_cache(page)); | |
6d9c285a | 1398 | |
0c0e6195 | 1399 | } else { |
2932c8b0 | 1400 | pr_warn("failed to isolate pfn %lx\n", pfn); |
0efadf48 | 1401 | dump_page(page, "isolation failed"); |
0c0e6195 | 1402 | } |
1723058e | 1403 | put_page(page); |
0c0e6195 | 1404 | } |
f3ab2636 | 1405 | if (!list_empty(&source)) { |
394e31d2 XQ |
1406 | /* Allocate a new page from the nearest neighbor node */ |
1407 | ret = migrate_pages(&source, new_node_page, NULL, 0, | |
9c620e2b | 1408 | MIGRATE_SYNC, MR_MEMORY_HOTPLUG); |
2932c8b0 MH |
1409 | if (ret) { |
1410 | list_for_each_entry(page, &source, lru) { | |
1411 | pr_warn("migrating pfn %lx failed ret:%d ", | |
1412 | page_to_pfn(page), ret); | |
1413 | dump_page(page, "migration failure"); | |
1414 | } | |
c8721bbb | 1415 | putback_movable_pages(&source); |
2932c8b0 | 1416 | } |
0c0e6195 | 1417 | } |
1723058e | 1418 | |
0c0e6195 KH |
1419 | return ret; |
1420 | } | |
1421 | ||
1422 | /* | |
1423 | * remove from free_area[] and mark all as Reserved. | |
1424 | */ | |
1425 | static int | |
1426 | offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages, | |
1427 | void *data) | |
1428 | { | |
5557c766 | 1429 | unsigned long *offlined_pages = (unsigned long *)data; |
0c0e6195 | 1430 | |
5557c766 MH |
1431 | *offlined_pages += __offline_isolated_pages(start, start + nr_pages); |
1432 | return 0; | |
0c0e6195 KH |
1433 | } |
1434 | ||
1435 | /* | |
1436 | * Check all pages in range, recoreded as memory resource, are isolated. | |
1437 | */ | |
1438 | static int | |
1439 | check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages, | |
1440 | void *data) | |
1441 | { | |
5557c766 | 1442 | return test_pages_isolated(start_pfn, start_pfn + nr_pages, true); |
0c0e6195 KH |
1443 | } |
1444 | ||
c5320926 TC |
1445 | static int __init cmdline_parse_movable_node(char *p) |
1446 | { | |
4932381e | 1447 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
55ac590c | 1448 | movable_node_enabled = true; |
4932381e MH |
1449 | #else |
1450 | pr_warn("movable_node parameter depends on CONFIG_HAVE_MEMBLOCK_NODE_MAP to work properly\n"); | |
1451 | #endif | |
c5320926 TC |
1452 | return 0; |
1453 | } | |
1454 | early_param("movable_node", cmdline_parse_movable_node); | |
1455 | ||
d9713679 LJ |
1456 | /* check which state of node_states will be changed when offline memory */ |
1457 | static void node_states_check_changes_offline(unsigned long nr_pages, | |
1458 | struct zone *zone, struct memory_notify *arg) | |
1459 | { | |
1460 | struct pglist_data *pgdat = zone->zone_pgdat; | |
1461 | unsigned long present_pages = 0; | |
86b27bea | 1462 | enum zone_type zt; |
d9713679 | 1463 | |
98fa15f3 AK |
1464 | arg->status_change_nid = NUMA_NO_NODE; |
1465 | arg->status_change_nid_normal = NUMA_NO_NODE; | |
1466 | arg->status_change_nid_high = NUMA_NO_NODE; | |
d9713679 LJ |
1467 | |
1468 | /* | |
86b27bea OS |
1469 | * Check whether node_states[N_NORMAL_MEMORY] will be changed. |
1470 | * If the memory to be offline is within the range | |
1471 | * [0..ZONE_NORMAL], and it is the last present memory there, | |
1472 | * the zones in that range will become empty after the offlining, | |
1473 | * thus we can determine that we need to clear the node from | |
1474 | * node_states[N_NORMAL_MEMORY]. | |
d9713679 | 1475 | */ |
86b27bea | 1476 | for (zt = 0; zt <= ZONE_NORMAL; zt++) |
d9713679 | 1477 | present_pages += pgdat->node_zones[zt].present_pages; |
86b27bea | 1478 | if (zone_idx(zone) <= ZONE_NORMAL && nr_pages >= present_pages) |
d9713679 | 1479 | arg->status_change_nid_normal = zone_to_nid(zone); |
d9713679 | 1480 | |
6715ddf9 LJ |
1481 | #ifdef CONFIG_HIGHMEM |
1482 | /* | |
86b27bea OS |
1483 | * node_states[N_HIGH_MEMORY] contains nodes which |
1484 | * have normal memory or high memory. | |
1485 | * Here we add the present_pages belonging to ZONE_HIGHMEM. | |
1486 | * If the zone is within the range of [0..ZONE_HIGHMEM), and | |
1487 | * we determine that the zones in that range become empty, | |
1488 | * we need to clear the node for N_HIGH_MEMORY. | |
6715ddf9 | 1489 | */ |
86b27bea OS |
1490 | present_pages += pgdat->node_zones[ZONE_HIGHMEM].present_pages; |
1491 | if (zone_idx(zone) <= ZONE_HIGHMEM && nr_pages >= present_pages) | |
6715ddf9 | 1492 | arg->status_change_nid_high = zone_to_nid(zone); |
6715ddf9 LJ |
1493 | #endif |
1494 | ||
d9713679 | 1495 | /* |
86b27bea OS |
1496 | * We have accounted the pages from [0..ZONE_NORMAL), and |
1497 | * in case of CONFIG_HIGHMEM the pages from ZONE_HIGHMEM | |
1498 | * as well. | |
1499 | * Here we count the possible pages from ZONE_MOVABLE. | |
1500 | * If after having accounted all the pages, we see that the nr_pages | |
1501 | * to be offlined is over or equal to the accounted pages, | |
1502 | * we know that the node will become empty, and so, we can clear | |
1503 | * it for N_MEMORY as well. | |
d9713679 | 1504 | */ |
86b27bea | 1505 | present_pages += pgdat->node_zones[ZONE_MOVABLE].present_pages; |
d9713679 | 1506 | |
d9713679 LJ |
1507 | if (nr_pages >= present_pages) |
1508 | arg->status_change_nid = zone_to_nid(zone); | |
d9713679 LJ |
1509 | } |
1510 | ||
1511 | static void node_states_clear_node(int node, struct memory_notify *arg) | |
1512 | { | |
1513 | if (arg->status_change_nid_normal >= 0) | |
1514 | node_clear_state(node, N_NORMAL_MEMORY); | |
1515 | ||
cf01f6f5 | 1516 | if (arg->status_change_nid_high >= 0) |
d9713679 | 1517 | node_clear_state(node, N_HIGH_MEMORY); |
6715ddf9 | 1518 | |
cf01f6f5 | 1519 | if (arg->status_change_nid >= 0) |
6715ddf9 | 1520 | node_clear_state(node, N_MEMORY); |
d9713679 LJ |
1521 | } |
1522 | ||
a16cee10 | 1523 | static int __ref __offline_pages(unsigned long start_pfn, |
ecde0f3e | 1524 | unsigned long end_pfn) |
0c0e6195 | 1525 | { |
ecde0f3e | 1526 | unsigned long pfn, nr_pages; |
5557c766 | 1527 | unsigned long offlined_pages = 0; |
9b7ea46a | 1528 | int ret, node, nr_isolate_pageblock; |
d702909f | 1529 | unsigned long flags; |
a96dfddb | 1530 | unsigned long valid_start, valid_end; |
0c0e6195 | 1531 | struct zone *zone; |
7b78d335 | 1532 | struct memory_notify arg; |
79605093 | 1533 | char *reason; |
0c0e6195 | 1534 | |
381eab4a DH |
1535 | mem_hotplug_begin(); |
1536 | ||
0c0e6195 KH |
1537 | /* This makes hotplug much easier...and readable. |
1538 | we assume this for now. .*/ | |
381eab4a DH |
1539 | if (!test_pages_in_a_zone(start_pfn, end_pfn, &valid_start, |
1540 | &valid_end)) { | |
79605093 MH |
1541 | ret = -EINVAL; |
1542 | reason = "multizone range"; | |
1543 | goto failed_removal; | |
381eab4a | 1544 | } |
7b78d335 | 1545 | |
a96dfddb | 1546 | zone = page_zone(pfn_to_page(valid_start)); |
7b78d335 YG |
1547 | node = zone_to_nid(zone); |
1548 | nr_pages = end_pfn - start_pfn; | |
1549 | ||
0c0e6195 | 1550 | /* set above range as isolated */ |
b023f468 | 1551 | ret = start_isolate_page_range(start_pfn, end_pfn, |
d381c547 MH |
1552 | MIGRATE_MOVABLE, |
1553 | SKIP_HWPOISON | REPORT_FAILURE); | |
9b7ea46a | 1554 | if (ret < 0) { |
79605093 MH |
1555 | reason = "failure to isolate range"; |
1556 | goto failed_removal; | |
381eab4a | 1557 | } |
9b7ea46a | 1558 | nr_isolate_pageblock = ret; |
7b78d335 YG |
1559 | |
1560 | arg.start_pfn = start_pfn; | |
1561 | arg.nr_pages = nr_pages; | |
d9713679 | 1562 | node_states_check_changes_offline(nr_pages, zone, &arg); |
7b78d335 YG |
1563 | |
1564 | ret = memory_notify(MEM_GOING_OFFLINE, &arg); | |
1565 | ret = notifier_to_errno(ret); | |
79605093 MH |
1566 | if (ret) { |
1567 | reason = "notifier failure"; | |
1568 | goto failed_removal_isolated; | |
1569 | } | |
7b78d335 | 1570 | |
bb8965bd MH |
1571 | do { |
1572 | for (pfn = start_pfn; pfn;) { | |
1573 | if (signal_pending(current)) { | |
1574 | ret = -EINTR; | |
1575 | reason = "signal backoff"; | |
1576 | goto failed_removal_isolated; | |
1577 | } | |
72b39cfc | 1578 | |
bb8965bd MH |
1579 | cond_resched(); |
1580 | lru_add_drain_all(); | |
bb8965bd MH |
1581 | |
1582 | pfn = scan_movable_pages(pfn, end_pfn); | |
1583 | if (pfn) { | |
1584 | /* | |
1585 | * TODO: fatal migration failures should bail | |
1586 | * out | |
1587 | */ | |
1588 | do_migrate_range(pfn, end_pfn); | |
1589 | } | |
1590 | } | |
0c0e6195 | 1591 | |
bb8965bd MH |
1592 | /* |
1593 | * Dissolve free hugepages in the memory block before doing | |
1594 | * offlining actually in order to make hugetlbfs's object | |
1595 | * counting consistent. | |
1596 | */ | |
1597 | ret = dissolve_free_huge_pages(start_pfn, end_pfn); | |
1598 | if (ret) { | |
1599 | reason = "failure to dissolve huge pages"; | |
1600 | goto failed_removal_isolated; | |
1601 | } | |
1602 | /* check again */ | |
5557c766 MH |
1603 | ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn, |
1604 | NULL, check_pages_isolated_cb); | |
1605 | } while (ret); | |
72b39cfc | 1606 | |
b3834be5 | 1607 | /* Ok, all of our target is isolated. |
0c0e6195 | 1608 | We cannot do rollback at this point. */ |
5557c766 MH |
1609 | walk_system_ram_range(start_pfn, end_pfn - start_pfn, |
1610 | &offlined_pages, offline_isolated_pages_cb); | |
1611 | pr_info("Offlined Pages %ld\n", offlined_pages); | |
9b7ea46a QC |
1612 | /* |
1613 | * Onlining will reset pagetype flags and makes migrate type | |
1614 | * MOVABLE, so just need to decrease the number of isolated | |
1615 | * pageblocks zone counter here. | |
1616 | */ | |
1617 | spin_lock_irqsave(&zone->lock, flags); | |
1618 | zone->nr_isolate_pageblock -= nr_isolate_pageblock; | |
1619 | spin_unlock_irqrestore(&zone->lock, flags); | |
1620 | ||
0c0e6195 | 1621 | /* removal success */ |
3dcc0571 | 1622 | adjust_managed_page_count(pfn_to_page(start_pfn), -offlined_pages); |
0c0e6195 | 1623 | zone->present_pages -= offlined_pages; |
d702909f CS |
1624 | |
1625 | pgdat_resize_lock(zone->zone_pgdat, &flags); | |
0c0e6195 | 1626 | zone->zone_pgdat->node_present_pages -= offlined_pages; |
d702909f | 1627 | pgdat_resize_unlock(zone->zone_pgdat, &flags); |
7b78d335 | 1628 | |
1b79acc9 KM |
1629 | init_per_zone_wmark_min(); |
1630 | ||
1e8537ba | 1631 | if (!populated_zone(zone)) { |
340175b7 | 1632 | zone_pcp_reset(zone); |
72675e13 | 1633 | build_all_zonelists(NULL); |
1e8537ba XQ |
1634 | } else |
1635 | zone_pcp_update(zone); | |
340175b7 | 1636 | |
d9713679 | 1637 | node_states_clear_node(node, &arg); |
698b1b30 | 1638 | if (arg.status_change_nid >= 0) { |
8fe23e05 | 1639 | kswapd_stop(node); |
698b1b30 VB |
1640 | kcompactd_stop(node); |
1641 | } | |
bce7394a | 1642 | |
0c0e6195 KH |
1643 | vm_total_pages = nr_free_pagecache_pages(); |
1644 | writeback_set_ratelimit(); | |
7b78d335 YG |
1645 | |
1646 | memory_notify(MEM_OFFLINE, &arg); | |
381eab4a | 1647 | mem_hotplug_done(); |
0c0e6195 KH |
1648 | return 0; |
1649 | ||
79605093 MH |
1650 | failed_removal_isolated: |
1651 | undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE); | |
c4efe484 | 1652 | memory_notify(MEM_CANCEL_OFFLINE, &arg); |
0c0e6195 | 1653 | failed_removal: |
79605093 | 1654 | pr_debug("memory offlining [mem %#010llx-%#010llx] failed due to %s\n", |
e33e33b4 | 1655 | (unsigned long long) start_pfn << PAGE_SHIFT, |
79605093 MH |
1656 | ((unsigned long long) end_pfn << PAGE_SHIFT) - 1, |
1657 | reason); | |
0c0e6195 | 1658 | /* pushback to free area */ |
381eab4a | 1659 | mem_hotplug_done(); |
0c0e6195 KH |
1660 | return ret; |
1661 | } | |
71088785 | 1662 | |
a16cee10 WC |
1663 | int offline_pages(unsigned long start_pfn, unsigned long nr_pages) |
1664 | { | |
ecde0f3e | 1665 | return __offline_pages(start_pfn, start_pfn + nr_pages); |
a16cee10 | 1666 | } |
e2ff3940 | 1667 | #endif /* CONFIG_MEMORY_HOTREMOVE */ |
a16cee10 | 1668 | |
bbc76be6 WC |
1669 | /** |
1670 | * walk_memory_range - walks through all mem sections in [start_pfn, end_pfn) | |
1671 | * @start_pfn: start pfn of the memory range | |
e05c4bbf | 1672 | * @end_pfn: end pfn of the memory range |
bbc76be6 WC |
1673 | * @arg: argument passed to func |
1674 | * @func: callback for each memory section walked | |
1675 | * | |
1676 | * This function walks through all present mem sections in range | |
1677 | * [start_pfn, end_pfn) and call func on each mem section. | |
1678 | * | |
1679 | * Returns the return value of func. | |
1680 | */ | |
e2ff3940 | 1681 | int walk_memory_range(unsigned long start_pfn, unsigned long end_pfn, |
bbc76be6 | 1682 | void *arg, int (*func)(struct memory_block *, void *)) |
71088785 | 1683 | { |
e90bdb7f WC |
1684 | struct memory_block *mem = NULL; |
1685 | struct mem_section *section; | |
e90bdb7f WC |
1686 | unsigned long pfn, section_nr; |
1687 | int ret; | |
e90bdb7f WC |
1688 | |
1689 | for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) { | |
1690 | section_nr = pfn_to_section_nr(pfn); | |
1691 | if (!present_section_nr(section_nr)) | |
1692 | continue; | |
1693 | ||
1694 | section = __nr_to_section(section_nr); | |
1695 | /* same memblock? */ | |
1696 | if (mem) | |
1697 | if ((section_nr >= mem->start_section_nr) && | |
1698 | (section_nr <= mem->end_section_nr)) | |
1699 | continue; | |
1700 | ||
1701 | mem = find_memory_block_hinted(section, mem); | |
1702 | if (!mem) | |
1703 | continue; | |
1704 | ||
bbc76be6 | 1705 | ret = func(mem, arg); |
e90bdb7f | 1706 | if (ret) { |
bbc76be6 WC |
1707 | kobject_put(&mem->dev.kobj); |
1708 | return ret; | |
e90bdb7f WC |
1709 | } |
1710 | } | |
1711 | ||
1712 | if (mem) | |
1713 | kobject_put(&mem->dev.kobj); | |
1714 | ||
bbc76be6 WC |
1715 | return 0; |
1716 | } | |
1717 | ||
e2ff3940 | 1718 | #ifdef CONFIG_MEMORY_HOTREMOVE |
d6de9d53 | 1719 | static int check_memblock_offlined_cb(struct memory_block *mem, void *arg) |
bbc76be6 WC |
1720 | { |
1721 | int ret = !is_memblock_offlined(mem); | |
1722 | ||
349daa0f RD |
1723 | if (unlikely(ret)) { |
1724 | phys_addr_t beginpa, endpa; | |
1725 | ||
1726 | beginpa = PFN_PHYS(section_nr_to_pfn(mem->start_section_nr)); | |
1727 | endpa = PFN_PHYS(section_nr_to_pfn(mem->end_section_nr + 1))-1; | |
756a025f | 1728 | pr_warn("removing memory fails, because memory [%pa-%pa] is onlined\n", |
349daa0f | 1729 | &beginpa, &endpa); |
bbc76be6 | 1730 | |
eca499ab PT |
1731 | return -EBUSY; |
1732 | } | |
1733 | return 0; | |
bbc76be6 WC |
1734 | } |
1735 | ||
0f1cfe9d | 1736 | static int check_cpu_on_node(pg_data_t *pgdat) |
60a5a19e | 1737 | { |
60a5a19e TC |
1738 | int cpu; |
1739 | ||
1740 | for_each_present_cpu(cpu) { | |
1741 | if (cpu_to_node(cpu) == pgdat->node_id) | |
1742 | /* | |
1743 | * the cpu on this node isn't removed, and we can't | |
1744 | * offline this node. | |
1745 | */ | |
1746 | return -EBUSY; | |
1747 | } | |
1748 | ||
1749 | return 0; | |
1750 | } | |
1751 | ||
0f1cfe9d TK |
1752 | /** |
1753 | * try_offline_node | |
e8b098fc | 1754 | * @nid: the node ID |
0f1cfe9d TK |
1755 | * |
1756 | * Offline a node if all memory sections and cpus of the node are removed. | |
1757 | * | |
1758 | * NOTE: The caller must call lock_device_hotplug() to serialize hotplug | |
1759 | * and online/offline operations before this call. | |
1760 | */ | |
90b30cdc | 1761 | void try_offline_node(int nid) |
60a5a19e | 1762 | { |
d822b86a WC |
1763 | pg_data_t *pgdat = NODE_DATA(nid); |
1764 | unsigned long start_pfn = pgdat->node_start_pfn; | |
1765 | unsigned long end_pfn = start_pfn + pgdat->node_spanned_pages; | |
60a5a19e TC |
1766 | unsigned long pfn; |
1767 | ||
1768 | for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) { | |
1769 | unsigned long section_nr = pfn_to_section_nr(pfn); | |
1770 | ||
1771 | if (!present_section_nr(section_nr)) | |
1772 | continue; | |
1773 | ||
1774 | if (pfn_to_nid(pfn) != nid) | |
1775 | continue; | |
1776 | ||
1777 | /* | |
1778 | * some memory sections of this node are not removed, and we | |
1779 | * can't offline node now. | |
1780 | */ | |
1781 | return; | |
1782 | } | |
1783 | ||
46a3679b | 1784 | if (check_cpu_on_node(pgdat)) |
60a5a19e TC |
1785 | return; |
1786 | ||
1787 | /* | |
1788 | * all memory/cpu of this node are removed, we can offline this | |
1789 | * node now. | |
1790 | */ | |
1791 | node_set_offline(nid); | |
1792 | unregister_one_node(nid); | |
1793 | } | |
90b30cdc | 1794 | EXPORT_SYMBOL(try_offline_node); |
60a5a19e | 1795 | |
d9eb1417 DH |
1796 | static void __release_memory_resource(resource_size_t start, |
1797 | resource_size_t size) | |
1798 | { | |
1799 | int ret; | |
1800 | ||
1801 | /* | |
1802 | * When removing memory in the same granularity as it was added, | |
1803 | * this function never fails. It might only fail if resources | |
1804 | * have to be adjusted or split. We'll ignore the error, as | |
1805 | * removing of memory cannot fail. | |
1806 | */ | |
1807 | ret = release_mem_region_adjustable(&iomem_resource, start, size); | |
1808 | if (ret) { | |
1809 | resource_size_t endres = start + size - 1; | |
1810 | ||
1811 | pr_warn("Unable to release resource <%pa-%pa> (%d)\n", | |
1812 | &start, &endres, ret); | |
1813 | } | |
1814 | } | |
1815 | ||
eca499ab | 1816 | static int __ref try_remove_memory(int nid, u64 start, u64 size) |
bbc76be6 | 1817 | { |
eca499ab | 1818 | int rc = 0; |
993c1aad | 1819 | |
27356f54 TK |
1820 | BUG_ON(check_hotplug_memory_range(start, size)); |
1821 | ||
bfc8c901 | 1822 | mem_hotplug_begin(); |
6677e3ea YI |
1823 | |
1824 | /* | |
242831eb | 1825 | * All memory blocks must be offlined before removing memory. Check |
eca499ab | 1826 | * whether all memory blocks in question are offline and return error |
242831eb | 1827 | * if this is not the case. |
6677e3ea | 1828 | */ |
eca499ab PT |
1829 | rc = walk_memory_range(PFN_DOWN(start), PFN_UP(start + size - 1), NULL, |
1830 | check_memblock_offlined_cb); | |
1831 | if (rc) | |
1832 | goto done; | |
6677e3ea | 1833 | |
46c66c4b YI |
1834 | /* remove memmap entry */ |
1835 | firmware_map_remove(start, start + size, "System RAM"); | |
f9126ab9 XQ |
1836 | memblock_free(start, size); |
1837 | memblock_remove(start, size); | |
46c66c4b | 1838 | |
2c2a5af6 | 1839 | arch_remove_memory(nid, start, size, NULL); |
d9eb1417 | 1840 | __release_memory_resource(start, size); |
24d335ca | 1841 | |
60a5a19e TC |
1842 | try_offline_node(nid); |
1843 | ||
eca499ab | 1844 | done: |
bfc8c901 | 1845 | mem_hotplug_done(); |
eca499ab | 1846 | return rc; |
71088785 | 1847 | } |
d15e5926 | 1848 | |
eca499ab PT |
1849 | /** |
1850 | * remove_memory | |
1851 | * @nid: the node ID | |
1852 | * @start: physical address of the region to remove | |
1853 | * @size: size of the region to remove | |
1854 | * | |
1855 | * NOTE: The caller must call lock_device_hotplug() to serialize hotplug | |
1856 | * and online/offline operations before this call, as required by | |
1857 | * try_offline_node(). | |
1858 | */ | |
1859 | void __remove_memory(int nid, u64 start, u64 size) | |
1860 | { | |
1861 | ||
1862 | /* | |
1863 | * trigger BUG() is some memory is not offlined prior to calling this | |
1864 | * function | |
1865 | */ | |
1866 | if (try_remove_memory(nid, start, size)) | |
1867 | BUG(); | |
1868 | } | |
1869 | ||
1870 | /* | |
1871 | * Remove memory if every memory block is offline, otherwise return -EBUSY is | |
1872 | * some memory is not offline | |
1873 | */ | |
1874 | int remove_memory(int nid, u64 start, u64 size) | |
d15e5926 | 1875 | { |
eca499ab PT |
1876 | int rc; |
1877 | ||
d15e5926 | 1878 | lock_device_hotplug(); |
eca499ab | 1879 | rc = try_remove_memory(nid, start, size); |
d15e5926 | 1880 | unlock_device_hotplug(); |
eca499ab PT |
1881 | |
1882 | return rc; | |
d15e5926 | 1883 | } |
71088785 | 1884 | EXPORT_SYMBOL_GPL(remove_memory); |
aba6efc4 | 1885 | #endif /* CONFIG_MEMORY_HOTREMOVE */ |