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); | |
45e0b78b KM |
135 | } |
136 | ||
53947027 | 137 | #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE |
46723bfa YI |
138 | void get_page_bootmem(unsigned long info, struct page *page, |
139 | unsigned long type) | |
04753278 | 140 | { |
ddffe98d | 141 | page->freelist = (void *)type; |
04753278 YG |
142 | SetPagePrivate(page); |
143 | set_page_private(page, info); | |
fe896d18 | 144 | page_ref_inc(page); |
04753278 YG |
145 | } |
146 | ||
170a5a7e | 147 | void put_page_bootmem(struct page *page) |
04753278 | 148 | { |
5f24ce5f | 149 | unsigned long type; |
04753278 | 150 | |
ddffe98d | 151 | type = (unsigned long) page->freelist; |
5f24ce5f AA |
152 | BUG_ON(type < MEMORY_HOTPLUG_MIN_BOOTMEM_TYPE || |
153 | type > MEMORY_HOTPLUG_MAX_BOOTMEM_TYPE); | |
04753278 | 154 | |
fe896d18 | 155 | if (page_ref_dec_return(page) == 1) { |
ddffe98d | 156 | page->freelist = NULL; |
04753278 YG |
157 | ClearPagePrivate(page); |
158 | set_page_private(page, 0); | |
5f24ce5f | 159 | INIT_LIST_HEAD(&page->lru); |
170a5a7e | 160 | free_reserved_page(page); |
04753278 | 161 | } |
04753278 YG |
162 | } |
163 | ||
46723bfa YI |
164 | #ifdef CONFIG_HAVE_BOOTMEM_INFO_NODE |
165 | #ifndef CONFIG_SPARSEMEM_VMEMMAP | |
d92bc318 | 166 | static void register_page_bootmem_info_section(unsigned long start_pfn) |
04753278 | 167 | { |
f1eca35a | 168 | unsigned long mapsize, section_nr, i; |
04753278 YG |
169 | struct mem_section *ms; |
170 | struct page *page, *memmap; | |
f1eca35a | 171 | struct mem_section_usage *usage; |
04753278 | 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 | ||
f1eca35a DW |
191 | usage = ms->usage; |
192 | page = virt_to_page(usage); | |
04753278 | 193 | |
f1eca35a | 194 | mapsize = PAGE_ALIGN(mem_section_usage_size()) >> PAGE_SHIFT; |
04753278 YG |
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 | { | |
f1eca35a | 203 | unsigned long mapsize, section_nr, i; |
46723bfa YI |
204 | struct mem_section *ms; |
205 | struct page *page, *memmap; | |
f1eca35a | 206 | struct mem_section_usage *usage; |
46723bfa | 207 | |
46723bfa YI |
208 | section_nr = pfn_to_section_nr(start_pfn); |
209 | ms = __nr_to_section(section_nr); | |
210 | ||
211 | memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr); | |
212 | ||
213 | register_page_bootmem_memmap(section_nr, memmap, PAGES_PER_SECTION); | |
214 | ||
f1eca35a DW |
215 | usage = ms->usage; |
216 | page = virt_to_page(usage); | |
46723bfa | 217 | |
f1eca35a | 218 | mapsize = PAGE_ALIGN(mem_section_usage_size()) >> PAGE_SHIFT; |
46723bfa YI |
219 | |
220 | for (i = 0; i < mapsize; i++, page++) | |
221 | get_page_bootmem(section_nr, page, MIX_SECTION_INFO); | |
222 | } | |
223 | #endif /* !CONFIG_SPARSEMEM_VMEMMAP */ | |
04753278 | 224 | |
7ded384a | 225 | void __init register_page_bootmem_info_node(struct pglist_data *pgdat) |
04753278 YG |
226 | { |
227 | unsigned long i, pfn, end_pfn, nr_pages; | |
228 | int node = pgdat->node_id; | |
229 | struct page *page; | |
04753278 YG |
230 | |
231 | nr_pages = PAGE_ALIGN(sizeof(struct pglist_data)) >> PAGE_SHIFT; | |
232 | page = virt_to_page(pgdat); | |
233 | ||
234 | for (i = 0; i < nr_pages; i++, page++) | |
235 | get_page_bootmem(node, page, NODE_INFO); | |
236 | ||
04753278 | 237 | pfn = pgdat->node_start_pfn; |
c1f19495 | 238 | end_pfn = pgdat_end_pfn(pgdat); |
04753278 | 239 | |
7e9f5eb0 | 240 | /* register section info */ |
f14851af | 241 | for (; pfn < end_pfn; pfn += PAGES_PER_SECTION) { |
242 | /* | |
243 | * Some platforms can assign the same pfn to multiple nodes - on | |
244 | * node0 as well as nodeN. To avoid registering a pfn against | |
245 | * multiple nodes we check that this pfn does not already | |
7e9f5eb0 | 246 | * reside in some other nodes. |
f14851af | 247 | */ |
f65e91df | 248 | if (pfn_valid(pfn) && (early_pfn_to_nid(pfn) == node)) |
f14851af | 249 | register_page_bootmem_info_section(pfn); |
250 | } | |
04753278 | 251 | } |
46723bfa | 252 | #endif /* CONFIG_HAVE_BOOTMEM_INFO_NODE */ |
04753278 | 253 | |
7ea62160 DW |
254 | static int check_pfn_span(unsigned long pfn, unsigned long nr_pages, |
255 | const char *reason) | |
256 | { | |
257 | /* | |
258 | * Disallow all operations smaller than a sub-section and only | |
259 | * allow operations smaller than a section for | |
260 | * SPARSEMEM_VMEMMAP. Note that check_hotplug_memory_range() | |
261 | * enforces a larger memory_block_size_bytes() granularity for | |
262 | * memory that will be marked online, so this check should only | |
263 | * fire for direct arch_{add,remove}_memory() users outside of | |
264 | * add_memory_resource(). | |
265 | */ | |
266 | unsigned long min_align; | |
267 | ||
268 | if (IS_ENABLED(CONFIG_SPARSEMEM_VMEMMAP)) | |
269 | min_align = PAGES_PER_SUBSECTION; | |
270 | else | |
271 | min_align = PAGES_PER_SECTION; | |
272 | if (!IS_ALIGNED(pfn, min_align) | |
273 | || !IS_ALIGNED(nr_pages, min_align)) { | |
274 | WARN(1, "Misaligned __%s_pages start: %#lx end: #%lx\n", | |
275 | reason, pfn, pfn + nr_pages - 1); | |
276 | return -EINVAL; | |
277 | } | |
278 | return 0; | |
279 | } | |
280 | ||
4edd7cef DR |
281 | /* |
282 | * Reasonably generic function for adding memory. It is | |
283 | * expected that archs that support memory hotplug will | |
284 | * call this function after deciding the zone to which to | |
285 | * add the new pages. | |
286 | */ | |
7ea62160 DW |
287 | int __ref __add_pages(int nid, unsigned long pfn, unsigned long nr_pages, |
288 | struct mhp_restrictions *restrictions) | |
4edd7cef | 289 | { |
9a845030 DW |
290 | int err; |
291 | unsigned long nr, start_sec, end_sec; | |
940519f0 | 292 | struct vmem_altmap *altmap = restrictions->altmap; |
4b94ffdc | 293 | |
4b94ffdc DW |
294 | if (altmap) { |
295 | /* | |
296 | * Validate altmap is within bounds of the total request | |
297 | */ | |
7ea62160 | 298 | if (altmap->base_pfn != pfn |
4b94ffdc DW |
299 | || vmem_altmap_offset(altmap) > nr_pages) { |
300 | pr_warn_once("memory add fail, invalid altmap\n"); | |
7ea62160 | 301 | return -EINVAL; |
4b94ffdc DW |
302 | } |
303 | altmap->alloc = 0; | |
304 | } | |
305 | ||
7ea62160 DW |
306 | err = check_pfn_span(pfn, nr_pages, "add"); |
307 | if (err) | |
308 | return err; | |
309 | ||
310 | start_sec = pfn_to_section_nr(pfn); | |
311 | end_sec = pfn_to_section_nr(pfn + nr_pages - 1); | |
9a845030 | 312 | for (nr = start_sec; nr <= end_sec; nr++) { |
7ea62160 DW |
313 | unsigned long pfns; |
314 | ||
315 | pfns = min(nr_pages, PAGES_PER_SECTION | |
316 | - (pfn & ~PAGE_SECTION_MASK)); | |
ba72b4c8 DW |
317 | err = sparse_add_section(nid, pfn, pfns, altmap); |
318 | if (err) | |
319 | break; | |
7ea62160 DW |
320 | pfn += pfns; |
321 | nr_pages -= pfns; | |
f64ac5e6 | 322 | cond_resched(); |
4edd7cef | 323 | } |
c435a390 | 324 | vmemmap_populate_print_last(); |
4edd7cef DR |
325 | return err; |
326 | } | |
4edd7cef | 327 | |
815121d2 | 328 | /* find the smallest valid pfn in the range [start_pfn, end_pfn) */ |
d09b0137 | 329 | static unsigned long find_smallest_section_pfn(int nid, struct zone *zone, |
815121d2 YI |
330 | unsigned long start_pfn, |
331 | unsigned long end_pfn) | |
332 | { | |
49ba3c6b DW |
333 | for (; start_pfn < end_pfn; start_pfn += PAGES_PER_SUBSECTION) { |
334 | if (unlikely(!pfn_valid(start_pfn))) | |
815121d2 YI |
335 | continue; |
336 | ||
337 | if (unlikely(pfn_to_nid(start_pfn) != nid)) | |
338 | continue; | |
339 | ||
340 | if (zone && zone != page_zone(pfn_to_page(start_pfn))) | |
341 | continue; | |
342 | ||
343 | return start_pfn; | |
344 | } | |
345 | ||
346 | return 0; | |
347 | } | |
348 | ||
349 | /* find the biggest valid pfn in the range [start_pfn, end_pfn). */ | |
d09b0137 | 350 | static unsigned long find_biggest_section_pfn(int nid, struct zone *zone, |
815121d2 YI |
351 | unsigned long start_pfn, |
352 | unsigned long end_pfn) | |
353 | { | |
815121d2 YI |
354 | unsigned long pfn; |
355 | ||
356 | /* pfn is the end pfn of a memory section. */ | |
357 | pfn = end_pfn - 1; | |
49ba3c6b DW |
358 | for (; pfn >= start_pfn; pfn -= PAGES_PER_SUBSECTION) { |
359 | if (unlikely(!pfn_valid(pfn))) | |
815121d2 YI |
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 | 380 | unsigned long pfn; |
815121d2 YI |
381 | int nid = zone_to_nid(zone); |
382 | ||
383 | zone_span_writelock(zone); | |
384 | if (zone_start_pfn == start_pfn) { | |
385 | /* | |
386 | * If the section is smallest section in the zone, it need | |
387 | * shrink zone->zone_start_pfn and zone->zone_spanned_pages. | |
388 | * In this case, we find second smallest valid mem_section | |
389 | * for shrinking zone. | |
390 | */ | |
391 | pfn = find_smallest_section_pfn(nid, zone, end_pfn, | |
392 | zone_end_pfn); | |
393 | if (pfn) { | |
394 | zone->zone_start_pfn = pfn; | |
395 | zone->spanned_pages = zone_end_pfn - pfn; | |
396 | } | |
397 | } else if (zone_end_pfn == end_pfn) { | |
398 | /* | |
399 | * If the section is biggest section in the zone, it need | |
400 | * shrink zone->spanned_pages. | |
401 | * In this case, we find second biggest valid mem_section for | |
402 | * shrinking zone. | |
403 | */ | |
404 | pfn = find_biggest_section_pfn(nid, zone, zone_start_pfn, | |
405 | start_pfn); | |
406 | if (pfn) | |
407 | zone->spanned_pages = pfn - zone_start_pfn + 1; | |
408 | } | |
409 | ||
410 | /* | |
411 | * The section is not biggest or smallest mem_section in the zone, it | |
412 | * only creates a hole in the zone. So in this case, we need not | |
413 | * change the zone. But perhaps, the zone has only hole data. Thus | |
414 | * it check the zone has only hole or not. | |
415 | */ | |
416 | pfn = zone_start_pfn; | |
49ba3c6b DW |
417 | for (; pfn < zone_end_pfn; pfn += PAGES_PER_SUBSECTION) { |
418 | if (unlikely(!pfn_valid(pfn))) | |
815121d2 YI |
419 | continue; |
420 | ||
421 | if (page_zone(pfn_to_page(pfn)) != zone) | |
422 | continue; | |
423 | ||
49ba3c6b DW |
424 | /* Skip range to be removed */ |
425 | if (pfn >= start_pfn && pfn < end_pfn) | |
815121d2 YI |
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 | 445 | unsigned long pfn; |
815121d2 YI |
446 | int nid = pgdat->node_id; |
447 | ||
448 | if (pgdat_start_pfn == start_pfn) { | |
449 | /* | |
450 | * If the section is smallest section in the pgdat, it need | |
451 | * shrink pgdat->node_start_pfn and pgdat->node_spanned_pages. | |
452 | * In this case, we find second smallest valid mem_section | |
453 | * for shrinking zone. | |
454 | */ | |
455 | pfn = find_smallest_section_pfn(nid, NULL, end_pfn, | |
456 | pgdat_end_pfn); | |
457 | if (pfn) { | |
458 | pgdat->node_start_pfn = pfn; | |
459 | pgdat->node_spanned_pages = pgdat_end_pfn - pfn; | |
460 | } | |
461 | } else if (pgdat_end_pfn == end_pfn) { | |
462 | /* | |
463 | * If the section is biggest section in the pgdat, it need | |
464 | * shrink pgdat->node_spanned_pages. | |
465 | * In this case, we find second biggest valid mem_section for | |
466 | * shrinking zone. | |
467 | */ | |
468 | pfn = find_biggest_section_pfn(nid, NULL, pgdat_start_pfn, | |
469 | start_pfn); | |
470 | if (pfn) | |
471 | pgdat->node_spanned_pages = pfn - pgdat_start_pfn + 1; | |
472 | } | |
473 | ||
474 | /* | |
475 | * If the section is not biggest or smallest mem_section in the pgdat, | |
476 | * it only creates a hole in the pgdat. So in this case, we need not | |
477 | * change the pgdat. | |
478 | * But perhaps, the pgdat has only hole data. Thus it check the pgdat | |
479 | * has only hole or not. | |
480 | */ | |
481 | pfn = pgdat_start_pfn; | |
49ba3c6b DW |
482 | for (; pfn < pgdat_end_pfn; pfn += PAGES_PER_SUBSECTION) { |
483 | if (unlikely(!pfn_valid(pfn))) | |
815121d2 YI |
484 | continue; |
485 | ||
486 | if (pfn_to_nid(pfn) != nid) | |
487 | continue; | |
488 | ||
49ba3c6b DW |
489 | /* Skip range to be removed */ |
490 | if (pfn >= start_pfn && pfn < end_pfn) | |
815121d2 YI |
491 | continue; |
492 | ||
493 | /* If we find valid section, we have nothing to do */ | |
494 | return; | |
495 | } | |
496 | ||
497 | /* The pgdat has no valid section */ | |
498 | pgdat->node_start_pfn = 0; | |
499 | pgdat->node_spanned_pages = 0; | |
500 | } | |
501 | ||
7ea62160 DW |
502 | static void __remove_zone(struct zone *zone, unsigned long start_pfn, |
503 | unsigned long nr_pages) | |
815121d2 YI |
504 | { |
505 | struct pglist_data *pgdat = zone->zone_pgdat; | |
815121d2 YI |
506 | unsigned long flags; |
507 | ||
815121d2 YI |
508 | pgdat_resize_lock(zone->zone_pgdat, &flags); |
509 | shrink_zone_span(zone, start_pfn, start_pfn + nr_pages); | |
510 | shrink_pgdat_span(pgdat, start_pfn, start_pfn + nr_pages); | |
511 | pgdat_resize_unlock(zone->zone_pgdat, &flags); | |
512 | } | |
513 | ||
7ea62160 DW |
514 | static void __remove_section(struct zone *zone, unsigned long pfn, |
515 | unsigned long nr_pages, unsigned long map_offset, | |
516 | struct vmem_altmap *altmap) | |
ea01ea93 | 517 | { |
7ea62160 | 518 | struct mem_section *ms = __nr_to_section(pfn_to_section_nr(pfn)); |
ea01ea93 | 519 | |
9d1d887d DH |
520 | if (WARN_ON_ONCE(!valid_section(ms))) |
521 | return; | |
ea01ea93 | 522 | |
7ea62160 | 523 | __remove_zone(zone, pfn, nr_pages); |
ba72b4c8 | 524 | sparse_remove_section(ms, pfn, nr_pages, map_offset, altmap); |
ea01ea93 BP |
525 | } |
526 | ||
ea01ea93 BP |
527 | /** |
528 | * __remove_pages() - remove sections of pages from a zone | |
529 | * @zone: zone from which pages need to be removed | |
7ea62160 | 530 | * @pfn: starting pageframe (must be aligned to start of a section) |
ea01ea93 | 531 | * @nr_pages: number of pages to remove (must be multiple of section size) |
e8b098fc | 532 | * @altmap: alternative device page map or %NULL if default memmap is used |
ea01ea93 BP |
533 | * |
534 | * Generic helper function to remove section mappings and sysfs entries | |
535 | * for the section of the memory we are removing. Caller needs to make | |
536 | * sure that pages are marked reserved and zones are adjust properly by | |
537 | * calling offline_pages(). | |
538 | */ | |
7ea62160 | 539 | void __remove_pages(struct zone *zone, unsigned long pfn, |
ac5c9426 | 540 | unsigned long nr_pages, struct vmem_altmap *altmap) |
ea01ea93 | 541 | { |
4b94ffdc | 542 | unsigned long map_offset = 0; |
9a845030 | 543 | unsigned long nr, start_sec, end_sec; |
4b94ffdc | 544 | |
96da4350 | 545 | map_offset = vmem_altmap_offset(altmap); |
ea01ea93 | 546 | |
7cf91a98 JK |
547 | clear_zone_contiguous(zone); |
548 | ||
7ea62160 DW |
549 | if (check_pfn_span(pfn, nr_pages, "remove")) |
550 | return; | |
ea01ea93 | 551 | |
7ea62160 DW |
552 | start_sec = pfn_to_section_nr(pfn); |
553 | end_sec = pfn_to_section_nr(pfn + nr_pages - 1); | |
9a845030 | 554 | for (nr = start_sec; nr <= end_sec; nr++) { |
7ea62160 | 555 | unsigned long pfns; |
4b94ffdc | 556 | |
dd33ad7b | 557 | cond_resched(); |
7ea62160 DW |
558 | pfns = min(nr_pages, PAGES_PER_SECTION |
559 | - (pfn & ~PAGE_SECTION_MASK)); | |
560 | __remove_section(zone, pfn, pfns, map_offset, altmap); | |
561 | pfn += pfns; | |
562 | nr_pages -= pfns; | |
4b94ffdc | 563 | map_offset = 0; |
ea01ea93 | 564 | } |
7cf91a98 JK |
565 | |
566 | set_zone_contiguous(zone); | |
ea01ea93 | 567 | } |
ea01ea93 | 568 | |
9d0ad8ca DK |
569 | int set_online_page_callback(online_page_callback_t callback) |
570 | { | |
571 | int rc = -EINVAL; | |
572 | ||
bfc8c901 VD |
573 | get_online_mems(); |
574 | mutex_lock(&online_page_callback_lock); | |
9d0ad8ca DK |
575 | |
576 | if (online_page_callback == generic_online_page) { | |
577 | online_page_callback = callback; | |
578 | rc = 0; | |
579 | } | |
580 | ||
bfc8c901 VD |
581 | mutex_unlock(&online_page_callback_lock); |
582 | put_online_mems(); | |
9d0ad8ca DK |
583 | |
584 | return rc; | |
585 | } | |
586 | EXPORT_SYMBOL_GPL(set_online_page_callback); | |
587 | ||
588 | int restore_online_page_callback(online_page_callback_t callback) | |
589 | { | |
590 | int rc = -EINVAL; | |
591 | ||
bfc8c901 VD |
592 | get_online_mems(); |
593 | mutex_lock(&online_page_callback_lock); | |
9d0ad8ca DK |
594 | |
595 | if (online_page_callback == callback) { | |
596 | online_page_callback = generic_online_page; | |
597 | rc = 0; | |
598 | } | |
599 | ||
bfc8c901 VD |
600 | mutex_unlock(&online_page_callback_lock); |
601 | put_online_mems(); | |
9d0ad8ca DK |
602 | |
603 | return rc; | |
604 | } | |
605 | EXPORT_SYMBOL_GPL(restore_online_page_callback); | |
606 | ||
607 | void __online_page_set_limits(struct page *page) | |
180c06ef | 608 | { |
9d0ad8ca DK |
609 | } |
610 | EXPORT_SYMBOL_GPL(__online_page_set_limits); | |
611 | ||
612 | void __online_page_increment_counters(struct page *page) | |
613 | { | |
3dcc0571 | 614 | adjust_managed_page_count(page, 1); |
9d0ad8ca DK |
615 | } |
616 | EXPORT_SYMBOL_GPL(__online_page_increment_counters); | |
180c06ef | 617 | |
9d0ad8ca DK |
618 | void __online_page_free(struct page *page) |
619 | { | |
3dcc0571 | 620 | __free_reserved_page(page); |
180c06ef | 621 | } |
9d0ad8ca DK |
622 | EXPORT_SYMBOL_GPL(__online_page_free); |
623 | ||
a9cd410a | 624 | static void generic_online_page(struct page *page, unsigned int order) |
9d0ad8ca | 625 | { |
cd02cf1a | 626 | kernel_map_pages(page, 1 << order, 1); |
a9cd410a AK |
627 | __free_pages_core(page, order); |
628 | totalram_pages_add(1UL << order); | |
629 | #ifdef CONFIG_HIGHMEM | |
630 | if (PageHighMem(page)) | |
631 | totalhigh_pages_add(1UL << order); | |
632 | #endif | |
633 | } | |
634 | ||
75884fb1 KH |
635 | static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages, |
636 | void *arg) | |
3947be19 | 637 | { |
b2c2ab20 DH |
638 | const unsigned long end_pfn = start_pfn + nr_pages; |
639 | unsigned long pfn; | |
640 | int order; | |
641 | ||
642 | /* | |
643 | * Online the pages. The callback might decide to keep some pages | |
644 | * PG_reserved (to add them to the buddy later), but we still account | |
645 | * them as being online/belonging to this zone ("present"). | |
646 | */ | |
647 | for (pfn = start_pfn; pfn < end_pfn; pfn += 1ul << order) { | |
648 | order = min(MAX_ORDER - 1, get_order(PFN_PHYS(end_pfn - pfn))); | |
bd02cc01 DH |
649 | /* __free_pages_core() wants pfns to be aligned to the order */ |
650 | if (WARN_ON_ONCE(!IS_ALIGNED(pfn, 1ul << order))) | |
651 | order = 0; | |
b2c2ab20 DH |
652 | (*online_page_callback)(pfn_to_page(pfn), order); |
653 | } | |
2d070eab | 654 | |
b2c2ab20 DH |
655 | /* mark all involved sections as online */ |
656 | online_mem_sections(start_pfn, end_pfn); | |
2d070eab | 657 | |
b2c2ab20 | 658 | *(unsigned long *)arg += nr_pages; |
75884fb1 KH |
659 | return 0; |
660 | } | |
661 | ||
d9713679 LJ |
662 | /* check which state of node_states will be changed when online memory */ |
663 | static void node_states_check_changes_online(unsigned long nr_pages, | |
664 | struct zone *zone, struct memory_notify *arg) | |
665 | { | |
666 | int nid = zone_to_nid(zone); | |
d9713679 | 667 | |
98fa15f3 AK |
668 | arg->status_change_nid = NUMA_NO_NODE; |
669 | arg->status_change_nid_normal = NUMA_NO_NODE; | |
670 | arg->status_change_nid_high = NUMA_NO_NODE; | |
d9713679 | 671 | |
8efe33f4 OS |
672 | if (!node_state(nid, N_MEMORY)) |
673 | arg->status_change_nid = nid; | |
674 | if (zone_idx(zone) <= ZONE_NORMAL && !node_state(nid, N_NORMAL_MEMORY)) | |
d9713679 | 675 | arg->status_change_nid_normal = nid; |
6715ddf9 | 676 | #ifdef CONFIG_HIGHMEM |
d3ba3ae1 | 677 | if (zone_idx(zone) <= ZONE_HIGHMEM && !node_state(nid, N_HIGH_MEMORY)) |
6715ddf9 | 678 | arg->status_change_nid_high = nid; |
6715ddf9 | 679 | #endif |
d9713679 LJ |
680 | } |
681 | ||
682 | static void node_states_set_node(int node, struct memory_notify *arg) | |
683 | { | |
684 | if (arg->status_change_nid_normal >= 0) | |
685 | node_set_state(node, N_NORMAL_MEMORY); | |
686 | ||
6715ddf9 LJ |
687 | if (arg->status_change_nid_high >= 0) |
688 | node_set_state(node, N_HIGH_MEMORY); | |
689 | ||
83d83612 OS |
690 | if (arg->status_change_nid >= 0) |
691 | node_set_state(node, N_MEMORY); | |
d9713679 LJ |
692 | } |
693 | ||
f1dd2cd1 MH |
694 | static void __meminit resize_zone_range(struct zone *zone, unsigned long start_pfn, |
695 | unsigned long nr_pages) | |
696 | { | |
697 | unsigned long old_end_pfn = zone_end_pfn(zone); | |
698 | ||
699 | if (zone_is_empty(zone) || start_pfn < zone->zone_start_pfn) | |
700 | zone->zone_start_pfn = start_pfn; | |
701 | ||
702 | zone->spanned_pages = max(start_pfn + nr_pages, old_end_pfn) - zone->zone_start_pfn; | |
703 | } | |
704 | ||
705 | static void __meminit resize_pgdat_range(struct pglist_data *pgdat, unsigned long start_pfn, | |
706 | unsigned long nr_pages) | |
707 | { | |
708 | unsigned long old_end_pfn = pgdat_end_pfn(pgdat); | |
709 | ||
710 | if (!pgdat->node_spanned_pages || start_pfn < pgdat->node_start_pfn) | |
711 | pgdat->node_start_pfn = start_pfn; | |
712 | ||
713 | pgdat->node_spanned_pages = max(start_pfn + nr_pages, old_end_pfn) - pgdat->node_start_pfn; | |
f1dd2cd1 | 714 | |
3fccb74c DH |
715 | } |
716 | /* | |
717 | * Associate the pfn range with the given zone, initializing the memmaps | |
718 | * and resizing the pgdat/zone data to span the added pages. After this | |
719 | * call, all affected pages are PG_reserved. | |
720 | */ | |
a99583e7 CH |
721 | void __ref move_pfn_range_to_zone(struct zone *zone, unsigned long start_pfn, |
722 | unsigned long nr_pages, struct vmem_altmap *altmap) | |
f1dd2cd1 MH |
723 | { |
724 | struct pglist_data *pgdat = zone->zone_pgdat; | |
725 | int nid = pgdat->node_id; | |
726 | unsigned long flags; | |
df429ac0 | 727 | |
f1dd2cd1 MH |
728 | clear_zone_contiguous(zone); |
729 | ||
730 | /* TODO Huh pgdat is irqsave while zone is not. It used to be like that before */ | |
731 | pgdat_resize_lock(pgdat, &flags); | |
732 | zone_span_writelock(zone); | |
fa004ab7 WY |
733 | if (zone_is_empty(zone)) |
734 | init_currently_empty_zone(zone, start_pfn, nr_pages); | |
f1dd2cd1 MH |
735 | resize_zone_range(zone, start_pfn, nr_pages); |
736 | zone_span_writeunlock(zone); | |
737 | resize_pgdat_range(pgdat, start_pfn, nr_pages); | |
738 | pgdat_resize_unlock(pgdat, &flags); | |
739 | ||
740 | /* | |
741 | * TODO now we have a visible range of pages which are not associated | |
742 | * with their zone properly. Not nice but set_pfnblock_flags_mask | |
743 | * expects the zone spans the pfn range. All the pages in the range | |
744 | * are reserved so nobody should be touching them so we should be safe | |
745 | */ | |
a99583e7 CH |
746 | memmap_init_zone(nr_pages, nid, zone_idx(zone), start_pfn, |
747 | MEMMAP_HOTPLUG, altmap); | |
f1dd2cd1 MH |
748 | |
749 | set_zone_contiguous(zone); | |
750 | } | |
751 | ||
c246a213 MH |
752 | /* |
753 | * Returns a default kernel memory zone for the given pfn range. | |
754 | * If no kernel zone covers this pfn range it will automatically go | |
755 | * to the ZONE_NORMAL. | |
756 | */ | |
c6f03e29 | 757 | static struct zone *default_kernel_zone_for_pfn(int nid, unsigned long start_pfn, |
c246a213 MH |
758 | unsigned long nr_pages) |
759 | { | |
760 | struct pglist_data *pgdat = NODE_DATA(nid); | |
761 | int zid; | |
762 | ||
763 | for (zid = 0; zid <= ZONE_NORMAL; zid++) { | |
764 | struct zone *zone = &pgdat->node_zones[zid]; | |
765 | ||
766 | if (zone_intersects(zone, start_pfn, nr_pages)) | |
767 | return zone; | |
768 | } | |
769 | ||
770 | return &pgdat->node_zones[ZONE_NORMAL]; | |
771 | } | |
772 | ||
c6f03e29 MH |
773 | static inline struct zone *default_zone_for_pfn(int nid, unsigned long start_pfn, |
774 | unsigned long nr_pages) | |
e5e68930 | 775 | { |
c6f03e29 MH |
776 | struct zone *kernel_zone = default_kernel_zone_for_pfn(nid, start_pfn, |
777 | nr_pages); | |
778 | struct zone *movable_zone = &NODE_DATA(nid)->node_zones[ZONE_MOVABLE]; | |
779 | bool in_kernel = zone_intersects(kernel_zone, start_pfn, nr_pages); | |
780 | bool in_movable = zone_intersects(movable_zone, start_pfn, nr_pages); | |
e5e68930 MH |
781 | |
782 | /* | |
c6f03e29 MH |
783 | * We inherit the existing zone in a simple case where zones do not |
784 | * overlap in the given range | |
e5e68930 | 785 | */ |
c6f03e29 MH |
786 | if (in_kernel ^ in_movable) |
787 | return (in_kernel) ? kernel_zone : movable_zone; | |
9f123ab5 | 788 | |
c6f03e29 MH |
789 | /* |
790 | * If the range doesn't belong to any zone or two zones overlap in the | |
791 | * given range then we use movable zone only if movable_node is | |
792 | * enabled because we always online to a kernel zone by default. | |
793 | */ | |
794 | return movable_node_enabled ? movable_zone : kernel_zone; | |
9f123ab5 MH |
795 | } |
796 | ||
e5e68930 MH |
797 | struct zone * zone_for_pfn_range(int online_type, int nid, unsigned start_pfn, |
798 | unsigned long nr_pages) | |
f1dd2cd1 | 799 | { |
c6f03e29 MH |
800 | if (online_type == MMOP_ONLINE_KERNEL) |
801 | return default_kernel_zone_for_pfn(nid, start_pfn, nr_pages); | |
f1dd2cd1 | 802 | |
c6f03e29 MH |
803 | if (online_type == MMOP_ONLINE_MOVABLE) |
804 | return &NODE_DATA(nid)->node_zones[ZONE_MOVABLE]; | |
df429ac0 | 805 | |
c6f03e29 | 806 | return default_zone_for_pfn(nid, start_pfn, nr_pages); |
e5e68930 MH |
807 | } |
808 | ||
511c2aba | 809 | int __ref online_pages(unsigned long pfn, unsigned long nr_pages, int online_type) |
75884fb1 | 810 | { |
aa47228a | 811 | unsigned long flags; |
3947be19 DH |
812 | unsigned long onlined_pages = 0; |
813 | struct zone *zone; | |
6811378e | 814 | int need_zonelists_rebuild = 0; |
7b78d335 YG |
815 | int nid; |
816 | int ret; | |
817 | struct memory_notify arg; | |
d0dc12e8 PT |
818 | struct memory_block *mem; |
819 | ||
381eab4a DH |
820 | mem_hotplug_begin(); |
821 | ||
d0dc12e8 PT |
822 | /* |
823 | * We can't use pfn_to_nid() because nid might be stored in struct page | |
824 | * which is not yet initialized. Instead, we find nid from memory block. | |
825 | */ | |
826 | mem = find_memory_block(__pfn_to_section(pfn)); | |
827 | nid = mem->nid; | |
89c02e69 | 828 | put_device(&mem->dev); |
7b78d335 | 829 | |
f1dd2cd1 | 830 | /* associate pfn range with the zone */ |
3fccb74c DH |
831 | zone = zone_for_pfn_range(online_type, nid, pfn, nr_pages); |
832 | move_pfn_range_to_zone(zone, pfn, nr_pages, NULL); | |
f1dd2cd1 | 833 | |
7b78d335 YG |
834 | arg.start_pfn = pfn; |
835 | arg.nr_pages = nr_pages; | |
d9713679 | 836 | node_states_check_changes_online(nr_pages, zone, &arg); |
7b78d335 | 837 | |
7b78d335 YG |
838 | ret = memory_notify(MEM_GOING_ONLINE, &arg); |
839 | ret = notifier_to_errno(ret); | |
e33e33b4 CY |
840 | if (ret) |
841 | goto failed_addition; | |
842 | ||
6811378e YG |
843 | /* |
844 | * If this zone is not populated, then it is not in zonelist. | |
845 | * This means the page allocator ignores this zone. | |
846 | * So, zonelist must be updated after online. | |
847 | */ | |
6dcd73d7 | 848 | if (!populated_zone(zone)) { |
6811378e | 849 | need_zonelists_rebuild = 1; |
72675e13 | 850 | setup_zone_pageset(zone); |
6dcd73d7 | 851 | } |
6811378e | 852 | |
908eedc6 | 853 | ret = walk_system_ram_range(pfn, nr_pages, &onlined_pages, |
75884fb1 | 854 | online_pages_range); |
fd8a4221 | 855 | if (ret) { |
ca9a46f8 | 856 | /* not a single memory resource was applicable */ |
6dcd73d7 WC |
857 | if (need_zonelists_rebuild) |
858 | zone_pcp_reset(zone); | |
e33e33b4 | 859 | goto failed_addition; |
fd8a4221 GL |
860 | } |
861 | ||
3947be19 | 862 | zone->present_pages += onlined_pages; |
aa47228a CS |
863 | |
864 | pgdat_resize_lock(zone->zone_pgdat, &flags); | |
f2937be5 | 865 | zone->zone_pgdat->node_present_pages += onlined_pages; |
aa47228a CS |
866 | pgdat_resize_unlock(zone->zone_pgdat, &flags); |
867 | ||
e900a918 DW |
868 | shuffle_zone(zone); |
869 | ||
ca9a46f8 DH |
870 | node_states_set_node(nid, &arg); |
871 | if (need_zonelists_rebuild) | |
872 | build_all_zonelists(NULL); | |
873 | else | |
874 | zone_pcp_update(zone); | |
3947be19 | 875 | |
1b79acc9 KM |
876 | init_per_zone_wmark_min(); |
877 | ||
ca9a46f8 DH |
878 | kswapd_run(nid); |
879 | kcompactd_run(nid); | |
61b13993 | 880 | |
1f522509 | 881 | vm_total_pages = nr_free_pagecache_pages(); |
2f7f24ec | 882 | |
2d1d43f6 | 883 | writeback_set_ratelimit(); |
7b78d335 | 884 | |
ca9a46f8 | 885 | memory_notify(MEM_ONLINE, &arg); |
381eab4a | 886 | mem_hotplug_done(); |
30467e0b | 887 | return 0; |
e33e33b4 CY |
888 | |
889 | failed_addition: | |
890 | pr_debug("online_pages [mem %#010llx-%#010llx] failed\n", | |
891 | (unsigned long long) pfn << PAGE_SHIFT, | |
892 | (((unsigned long long) pfn + nr_pages) << PAGE_SHIFT) - 1); | |
893 | memory_notify(MEM_CANCEL_ONLINE, &arg); | |
381eab4a | 894 | mem_hotplug_done(); |
e33e33b4 | 895 | return ret; |
3947be19 | 896 | } |
53947027 | 897 | #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */ |
bc02af93 | 898 | |
0bd85420 TC |
899 | static void reset_node_present_pages(pg_data_t *pgdat) |
900 | { | |
901 | struct zone *z; | |
902 | ||
903 | for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++) | |
904 | z->present_pages = 0; | |
905 | ||
906 | pgdat->node_present_pages = 0; | |
907 | } | |
908 | ||
e1319331 HS |
909 | /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */ |
910 | static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start) | |
9af3c2de YG |
911 | { |
912 | struct pglist_data *pgdat; | |
c8e861a5 | 913 | unsigned long start_pfn = PFN_DOWN(start); |
9af3c2de | 914 | |
a1e565aa TC |
915 | pgdat = NODE_DATA(nid); |
916 | if (!pgdat) { | |
917 | pgdat = arch_alloc_nodedata(nid); | |
918 | if (!pgdat) | |
919 | return NULL; | |
9af3c2de | 920 | |
33fce011 WY |
921 | pgdat->per_cpu_nodestats = |
922 | alloc_percpu(struct per_cpu_nodestat); | |
a1e565aa | 923 | arch_refresh_nodedata(nid, pgdat); |
b0dc3a34 | 924 | } else { |
33fce011 | 925 | int cpu; |
e716f2eb MG |
926 | /* |
927 | * Reset the nr_zones, order and classzone_idx before reuse. | |
928 | * Note that kswapd will init kswapd_classzone_idx properly | |
929 | * when it starts in the near future. | |
930 | */ | |
b0dc3a34 | 931 | pgdat->nr_zones = 0; |
38087d9b MG |
932 | pgdat->kswapd_order = 0; |
933 | pgdat->kswapd_classzone_idx = 0; | |
33fce011 WY |
934 | for_each_online_cpu(cpu) { |
935 | struct per_cpu_nodestat *p; | |
936 | ||
937 | p = per_cpu_ptr(pgdat->per_cpu_nodestats, cpu); | |
938 | memset(p, 0, sizeof(*p)); | |
939 | } | |
a1e565aa | 940 | } |
9af3c2de YG |
941 | |
942 | /* we can use NODE_DATA(nid) from here */ | |
943 | ||
03e85f9d OS |
944 | pgdat->node_id = nid; |
945 | pgdat->node_start_pfn = start_pfn; | |
946 | ||
9af3c2de | 947 | /* init node's zones as empty zones, we don't have any present pages.*/ |
03e85f9d | 948 | free_area_init_core_hotplug(nid); |
9af3c2de | 949 | |
959ecc48 KH |
950 | /* |
951 | * The node we allocated has no zone fallback lists. For avoiding | |
952 | * to access not-initialized zonelist, build here. | |
953 | */ | |
72675e13 | 954 | build_all_zonelists(pgdat); |
959ecc48 | 955 | |
0bd85420 TC |
956 | /* |
957 | * When memory is hot-added, all the memory is in offline state. So | |
958 | * clear all zones' present_pages because they will be updated in | |
959 | * online_pages() and offline_pages(). | |
960 | */ | |
03e85f9d | 961 | reset_node_managed_pages(pgdat); |
0bd85420 TC |
962 | reset_node_present_pages(pgdat); |
963 | ||
9af3c2de YG |
964 | return pgdat; |
965 | } | |
966 | ||
b9ff0360 | 967 | static void rollback_node_hotadd(int nid) |
9af3c2de | 968 | { |
b9ff0360 OS |
969 | pg_data_t *pgdat = NODE_DATA(nid); |
970 | ||
9af3c2de | 971 | arch_refresh_nodedata(nid, NULL); |
5830169f | 972 | free_percpu(pgdat->per_cpu_nodestats); |
9af3c2de | 973 | arch_free_nodedata(pgdat); |
9af3c2de YG |
974 | } |
975 | ||
0a547039 | 976 | |
01b0f197 TK |
977 | /** |
978 | * try_online_node - online a node if offlined | |
e8b098fc | 979 | * @nid: the node ID |
b9ff0360 OS |
980 | * @start: start addr of the node |
981 | * @set_node_online: Whether we want to online the node | |
cf23422b | 982 | * called by cpu_up() to online a node without onlined memory. |
b9ff0360 OS |
983 | * |
984 | * Returns: | |
985 | * 1 -> a new node has been allocated | |
986 | * 0 -> the node is already online | |
987 | * -ENOMEM -> the node could not be allocated | |
cf23422b | 988 | */ |
b9ff0360 | 989 | static int __try_online_node(int nid, u64 start, bool set_node_online) |
cf23422b | 990 | { |
b9ff0360 OS |
991 | pg_data_t *pgdat; |
992 | int ret = 1; | |
cf23422b | 993 | |
01b0f197 TK |
994 | if (node_online(nid)) |
995 | return 0; | |
996 | ||
b9ff0360 | 997 | pgdat = hotadd_new_pgdat(nid, start); |
7553e8f2 | 998 | if (!pgdat) { |
01b0f197 | 999 | pr_err("Cannot online node %d due to NULL pgdat\n", nid); |
cf23422b | 1000 | ret = -ENOMEM; |
1001 | goto out; | |
1002 | } | |
b9ff0360 OS |
1003 | |
1004 | if (set_node_online) { | |
1005 | node_set_online(nid); | |
1006 | ret = register_one_node(nid); | |
1007 | BUG_ON(ret); | |
1008 | } | |
cf23422b | 1009 | out: |
b9ff0360 OS |
1010 | return ret; |
1011 | } | |
1012 | ||
1013 | /* | |
1014 | * Users of this function always want to online/register the node | |
1015 | */ | |
1016 | int try_online_node(int nid) | |
1017 | { | |
1018 | int ret; | |
1019 | ||
1020 | mem_hotplug_begin(); | |
1021 | ret = __try_online_node(nid, 0, true); | |
bfc8c901 | 1022 | mem_hotplug_done(); |
cf23422b | 1023 | return ret; |
1024 | } | |
1025 | ||
27356f54 TK |
1026 | static int check_hotplug_memory_range(u64 start, u64 size) |
1027 | { | |
ba325585 | 1028 | /* memory range must be block size aligned */ |
cec3ebd0 DH |
1029 | if (!size || !IS_ALIGNED(start, memory_block_size_bytes()) || |
1030 | !IS_ALIGNED(size, memory_block_size_bytes())) { | |
ba325585 | 1031 | pr_err("Block size [%#lx] unaligned hotplug range: start %#llx, size %#llx", |
cec3ebd0 | 1032 | memory_block_size_bytes(), start, size); |
27356f54 TK |
1033 | return -EINVAL; |
1034 | } | |
1035 | ||
1036 | return 0; | |
1037 | } | |
1038 | ||
31bc3858 VK |
1039 | static int online_memory_block(struct memory_block *mem, void *arg) |
1040 | { | |
dc18d706 | 1041 | return device_online(&mem->dev); |
31bc3858 VK |
1042 | } |
1043 | ||
8df1d0e4 DH |
1044 | /* |
1045 | * NOTE: The caller must call lock_device_hotplug() to serialize hotplug | |
1046 | * and online/offline operations (triggered e.g. by sysfs). | |
1047 | * | |
1048 | * we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG | |
1049 | */ | |
f29d8e9c | 1050 | int __ref add_memory_resource(int nid, struct resource *res) |
bc02af93 | 1051 | { |
05f800a0 | 1052 | struct mhp_restrictions restrictions = {}; |
62cedb9f | 1053 | u64 start, size; |
b9ff0360 | 1054 | bool new_node = false; |
bc02af93 YG |
1055 | int ret; |
1056 | ||
62cedb9f DV |
1057 | start = res->start; |
1058 | size = resource_size(res); | |
1059 | ||
27356f54 TK |
1060 | ret = check_hotplug_memory_range(start, size); |
1061 | if (ret) | |
1062 | return ret; | |
1063 | ||
bfc8c901 | 1064 | mem_hotplug_begin(); |
ac13c462 | 1065 | |
7f36e3e5 TC |
1066 | /* |
1067 | * Add new range to memblock so that when hotadd_new_pgdat() is called | |
1068 | * to allocate new pgdat, get_pfn_range_for_nid() will be able to find | |
1069 | * this new range and calculate total pages correctly. The range will | |
1070 | * be removed at hot-remove time. | |
1071 | */ | |
1072 | memblock_add_node(start, size, nid); | |
1073 | ||
b9ff0360 OS |
1074 | ret = __try_online_node(nid, start, false); |
1075 | if (ret < 0) | |
1076 | goto error; | |
1077 | new_node = ret; | |
9af3c2de | 1078 | |
bc02af93 | 1079 | /* call arch's memory hotadd */ |
940519f0 | 1080 | ret = arch_add_memory(nid, start, size, &restrictions); |
9af3c2de YG |
1081 | if (ret < 0) |
1082 | goto error; | |
1083 | ||
db051a0d DH |
1084 | /* create memory block devices after memory was added */ |
1085 | ret = create_memory_block_devices(start, size); | |
1086 | if (ret) { | |
1087 | arch_remove_memory(nid, start, size, NULL); | |
1088 | goto error; | |
1089 | } | |
1090 | ||
a1e565aa | 1091 | if (new_node) { |
d5b6f6a3 | 1092 | /* If sysfs file of new node can't be created, cpu on the node |
0fc44159 YG |
1093 | * can't be hot-added. There is no rollback way now. |
1094 | * So, check by BUG_ON() to catch it reluctantly.. | |
d5b6f6a3 | 1095 | * We online node here. We can't roll back from here. |
0fc44159 | 1096 | */ |
d5b6f6a3 OS |
1097 | node_set_online(nid); |
1098 | ret = __register_one_node(nid); | |
0fc44159 YG |
1099 | BUG_ON(ret); |
1100 | } | |
1101 | ||
d5b6f6a3 | 1102 | /* link memory sections under this node.*/ |
4fbce633 | 1103 | ret = link_mem_sections(nid, PFN_DOWN(start), PFN_UP(start + size - 1)); |
d5b6f6a3 OS |
1104 | BUG_ON(ret); |
1105 | ||
d96ae530 | 1106 | /* create new memmap entry */ |
1107 | firmware_map_add_hotplug(start, start + size, "System RAM"); | |
1108 | ||
381eab4a DH |
1109 | /* device_online() will take the lock when calling online_pages() */ |
1110 | mem_hotplug_done(); | |
1111 | ||
31bc3858 | 1112 | /* online pages if requested */ |
f29d8e9c | 1113 | if (memhp_auto_online) |
fbcf73ce | 1114 | walk_memory_blocks(start, size, NULL, online_memory_block); |
31bc3858 | 1115 | |
381eab4a | 1116 | return ret; |
9af3c2de YG |
1117 | error: |
1118 | /* rollback pgdat allocation and others */ | |
b9ff0360 OS |
1119 | if (new_node) |
1120 | rollback_node_hotadd(nid); | |
7f36e3e5 | 1121 | memblock_remove(start, size); |
bfc8c901 | 1122 | mem_hotplug_done(); |
bc02af93 YG |
1123 | return ret; |
1124 | } | |
62cedb9f | 1125 | |
8df1d0e4 DH |
1126 | /* requires device_hotplug_lock, see add_memory_resource() */ |
1127 | int __ref __add_memory(int nid, u64 start, u64 size) | |
62cedb9f DV |
1128 | { |
1129 | struct resource *res; | |
1130 | int ret; | |
1131 | ||
1132 | res = register_memory_resource(start, size); | |
6f754ba4 VK |
1133 | if (IS_ERR(res)) |
1134 | return PTR_ERR(res); | |
62cedb9f | 1135 | |
f29d8e9c | 1136 | ret = add_memory_resource(nid, res); |
62cedb9f DV |
1137 | if (ret < 0) |
1138 | release_memory_resource(res); | |
1139 | return ret; | |
1140 | } | |
8df1d0e4 DH |
1141 | |
1142 | int add_memory(int nid, u64 start, u64 size) | |
1143 | { | |
1144 | int rc; | |
1145 | ||
1146 | lock_device_hotplug(); | |
1147 | rc = __add_memory(nid, start, size); | |
1148 | unlock_device_hotplug(); | |
1149 | ||
1150 | return rc; | |
1151 | } | |
bc02af93 | 1152 | EXPORT_SYMBOL_GPL(add_memory); |
0c0e6195 KH |
1153 | |
1154 | #ifdef CONFIG_MEMORY_HOTREMOVE | |
5c755e9f BP |
1155 | /* |
1156 | * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy | |
1157 | * set and the size of the free page is given by page_order(). Using this, | |
1158 | * the function determines if the pageblock contains only free pages. | |
1159 | * Due to buddy contraints, a free page at least the size of a pageblock will | |
1160 | * be located at the start of the pageblock | |
1161 | */ | |
1162 | static inline int pageblock_free(struct page *page) | |
1163 | { | |
1164 | return PageBuddy(page) && page_order(page) >= pageblock_order; | |
1165 | } | |
1166 | ||
891cb2a7 MH |
1167 | /* Return the pfn of the start of the next active pageblock after a given pfn */ |
1168 | static unsigned long next_active_pageblock(unsigned long pfn) | |
5c755e9f | 1169 | { |
891cb2a7 MH |
1170 | struct page *page = pfn_to_page(pfn); |
1171 | ||
5c755e9f | 1172 | /* Ensure the starting page is pageblock-aligned */ |
891cb2a7 | 1173 | BUG_ON(pfn & (pageblock_nr_pages - 1)); |
5c755e9f | 1174 | |
5c755e9f | 1175 | /* If the entire pageblock is free, move to the end of free page */ |
0dcc48c1 KH |
1176 | if (pageblock_free(page)) { |
1177 | int order; | |
1178 | /* be careful. we don't have locks, page_order can be changed.*/ | |
1179 | order = page_order(page); | |
1180 | if ((order < MAX_ORDER) && (order >= pageblock_order)) | |
891cb2a7 | 1181 | return pfn + (1 << order); |
0dcc48c1 | 1182 | } |
5c755e9f | 1183 | |
891cb2a7 | 1184 | return pfn + pageblock_nr_pages; |
5c755e9f BP |
1185 | } |
1186 | ||
891cb2a7 | 1187 | static bool is_pageblock_removable_nolock(unsigned long pfn) |
fb52bbae | 1188 | { |
891cb2a7 | 1189 | struct page *page = pfn_to_page(pfn); |
fb52bbae | 1190 | struct zone *zone; |
fb52bbae MM |
1191 | |
1192 | /* | |
1193 | * We have to be careful here because we are iterating over memory | |
1194 | * sections which are not zone aware so we might end up outside of | |
1195 | * the zone but still within the section. | |
1196 | * We have to take care about the node as well. If the node is offline | |
1197 | * its NODE_DATA will be NULL - see page_zone. | |
1198 | */ | |
1199 | if (!node_online(page_to_nid(page))) | |
1200 | return false; | |
1201 | ||
1202 | zone = page_zone(page); | |
1203 | pfn = page_to_pfn(page); | |
1204 | if (!zone_spans_pfn(zone, pfn)) | |
1205 | return false; | |
1206 | ||
d381c547 | 1207 | return !has_unmovable_pages(zone, page, 0, MIGRATE_MOVABLE, SKIP_HWPOISON); |
fb52bbae MM |
1208 | } |
1209 | ||
5c755e9f | 1210 | /* Checks if this range of memory is likely to be hot-removable. */ |
c98940f6 | 1211 | bool is_mem_section_removable(unsigned long start_pfn, unsigned long nr_pages) |
5c755e9f | 1212 | { |
891cb2a7 MH |
1213 | unsigned long end_pfn, pfn; |
1214 | ||
1215 | end_pfn = min(start_pfn + nr_pages, | |
1216 | zone_end_pfn(page_zone(pfn_to_page(start_pfn)))); | |
5c755e9f BP |
1217 | |
1218 | /* Check the starting page of each pageblock within the range */ | |
891cb2a7 MH |
1219 | for (pfn = start_pfn; pfn < end_pfn; pfn = next_active_pageblock(pfn)) { |
1220 | if (!is_pageblock_removable_nolock(pfn)) | |
c98940f6 | 1221 | return false; |
49ac8255 | 1222 | cond_resched(); |
5c755e9f BP |
1223 | } |
1224 | ||
1225 | /* All pageblocks in the memory block are likely to be hot-removable */ | |
c98940f6 | 1226 | return true; |
5c755e9f BP |
1227 | } |
1228 | ||
0c0e6195 | 1229 | /* |
deb88a2a | 1230 | * Confirm all pages in a range [start, end) belong to the same zone. |
a96dfddb | 1231 | * When true, return its valid [start, end). |
0c0e6195 | 1232 | */ |
a96dfddb TK |
1233 | int test_pages_in_a_zone(unsigned long start_pfn, unsigned long end_pfn, |
1234 | unsigned long *valid_start, unsigned long *valid_end) | |
0c0e6195 | 1235 | { |
5f0f2887 | 1236 | unsigned long pfn, sec_end_pfn; |
a96dfddb | 1237 | unsigned long start, end; |
0c0e6195 KH |
1238 | struct zone *zone = NULL; |
1239 | struct page *page; | |
1240 | int i; | |
deb88a2a | 1241 | for (pfn = start_pfn, sec_end_pfn = SECTION_ALIGN_UP(start_pfn + 1); |
0c0e6195 | 1242 | pfn < end_pfn; |
deb88a2a | 1243 | pfn = sec_end_pfn, sec_end_pfn += PAGES_PER_SECTION) { |
5f0f2887 AB |
1244 | /* Make sure the memory section is present first */ |
1245 | if (!present_section_nr(pfn_to_section_nr(pfn))) | |
0c0e6195 | 1246 | continue; |
5f0f2887 AB |
1247 | for (; pfn < sec_end_pfn && pfn < end_pfn; |
1248 | pfn += MAX_ORDER_NR_PAGES) { | |
1249 | i = 0; | |
1250 | /* This is just a CONFIG_HOLES_IN_ZONE check.*/ | |
1251 | while ((i < MAX_ORDER_NR_PAGES) && | |
1252 | !pfn_valid_within(pfn + i)) | |
1253 | i++; | |
d6d8c8a4 | 1254 | if (i == MAX_ORDER_NR_PAGES || pfn + i >= end_pfn) |
5f0f2887 | 1255 | continue; |
24feb47c MZ |
1256 | /* Check if we got outside of the zone */ |
1257 | if (zone && !zone_spans_pfn(zone, pfn + i)) | |
1258 | return 0; | |
5f0f2887 AB |
1259 | page = pfn_to_page(pfn + i); |
1260 | if (zone && page_zone(page) != zone) | |
1261 | return 0; | |
a96dfddb TK |
1262 | if (!zone) |
1263 | start = pfn + i; | |
5f0f2887 | 1264 | zone = page_zone(page); |
a96dfddb | 1265 | end = pfn + MAX_ORDER_NR_PAGES; |
5f0f2887 | 1266 | } |
0c0e6195 | 1267 | } |
deb88a2a | 1268 | |
a96dfddb TK |
1269 | if (zone) { |
1270 | *valid_start = start; | |
d6d8c8a4 | 1271 | *valid_end = min(end, end_pfn); |
deb88a2a | 1272 | return 1; |
a96dfddb | 1273 | } else { |
deb88a2a | 1274 | return 0; |
a96dfddb | 1275 | } |
0c0e6195 KH |
1276 | } |
1277 | ||
1278 | /* | |
0efadf48 YX |
1279 | * Scan pfn range [start,end) to find movable/migratable pages (LRU pages, |
1280 | * non-lru movable pages and hugepages). We scan pfn because it's much | |
1281 | * easier than scanning over linked list. This function returns the pfn | |
1282 | * of the first found movable page if it's found, otherwise 0. | |
0c0e6195 | 1283 | */ |
c8721bbb | 1284 | static unsigned long scan_movable_pages(unsigned long start, unsigned long end) |
0c0e6195 KH |
1285 | { |
1286 | unsigned long pfn; | |
eeb0efd0 | 1287 | |
0c0e6195 | 1288 | for (pfn = start; pfn < end; pfn++) { |
eeb0efd0 OS |
1289 | struct page *page, *head; |
1290 | unsigned long skip; | |
1291 | ||
1292 | if (!pfn_valid(pfn)) | |
1293 | continue; | |
1294 | page = pfn_to_page(pfn); | |
1295 | if (PageLRU(page)) | |
1296 | return pfn; | |
1297 | if (__PageMovable(page)) | |
1298 | return pfn; | |
1299 | ||
1300 | if (!PageHuge(page)) | |
1301 | continue; | |
1302 | head = compound_head(page); | |
39186cbe | 1303 | if (page_huge_active(head)) |
eeb0efd0 | 1304 | return pfn; |
d8c6546b | 1305 | skip = compound_nr(head) - (page - head); |
eeb0efd0 | 1306 | pfn += skip - 1; |
0c0e6195 KH |
1307 | } |
1308 | return 0; | |
1309 | } | |
1310 | ||
666feb21 | 1311 | static struct page *new_node_page(struct page *page, unsigned long private) |
394e31d2 | 1312 | { |
394e31d2 | 1313 | int nid = page_to_nid(page); |
231e97e2 | 1314 | nodemask_t nmask = node_states[N_MEMORY]; |
7f252f27 MH |
1315 | |
1316 | /* | |
1317 | * try to allocate from a different node but reuse this node if there | |
1318 | * are no other online nodes to be used (e.g. we are offlining a part | |
1319 | * of the only existing node) | |
1320 | */ | |
1321 | node_clear(nid, nmask); | |
1322 | if (nodes_empty(nmask)) | |
1323 | node_set(nid, nmask); | |
394e31d2 | 1324 | |
8b913238 | 1325 | return new_page_nodemask(page, nid, &nmask); |
394e31d2 XQ |
1326 | } |
1327 | ||
0c0e6195 KH |
1328 | static int |
1329 | do_migrate_range(unsigned long start_pfn, unsigned long end_pfn) | |
1330 | { | |
1331 | unsigned long pfn; | |
1332 | struct page *page; | |
0c0e6195 KH |
1333 | int ret = 0; |
1334 | LIST_HEAD(source); | |
1335 | ||
a85009c3 | 1336 | for (pfn = start_pfn; pfn < end_pfn; pfn++) { |
0c0e6195 KH |
1337 | if (!pfn_valid(pfn)) |
1338 | continue; | |
1339 | page = pfn_to_page(pfn); | |
c8721bbb NH |
1340 | |
1341 | if (PageHuge(page)) { | |
1342 | struct page *head = compound_head(page); | |
d8c6546b | 1343 | pfn = page_to_pfn(head) + compound_nr(head) - 1; |
daf3538a | 1344 | isolate_huge_page(head, &source); |
c8721bbb | 1345 | continue; |
94723aaf | 1346 | } else if (PageTransHuge(page)) |
8135d892 NH |
1347 | pfn = page_to_pfn(compound_head(page)) |
1348 | + hpage_nr_pages(page) - 1; | |
c8721bbb | 1349 | |
b15c8726 MH |
1350 | /* |
1351 | * HWPoison pages have elevated reference counts so the migration would | |
1352 | * fail on them. It also doesn't make any sense to migrate them in the | |
1353 | * first place. Still try to unmap such a page in case it is still mapped | |
1354 | * (e.g. current hwpoison implementation doesn't unmap KSM pages but keep | |
1355 | * the unmap as the catch all safety net). | |
1356 | */ | |
1357 | if (PageHWPoison(page)) { | |
1358 | if (WARN_ON(PageLRU(page))) | |
1359 | isolate_lru_page(page); | |
1360 | if (page_mapped(page)) | |
1361 | try_to_unmap(page, TTU_IGNORE_MLOCK | TTU_IGNORE_ACCESS); | |
1362 | continue; | |
1363 | } | |
1364 | ||
700c2a46 | 1365 | if (!get_page_unless_zero(page)) |
0c0e6195 KH |
1366 | continue; |
1367 | /* | |
0efadf48 YX |
1368 | * We can skip free pages. And we can deal with pages on |
1369 | * LRU and non-lru movable pages. | |
0c0e6195 | 1370 | */ |
0efadf48 YX |
1371 | if (PageLRU(page)) |
1372 | ret = isolate_lru_page(page); | |
1373 | else | |
1374 | ret = isolate_movable_page(page, ISOLATE_UNEVICTABLE); | |
0c0e6195 | 1375 | if (!ret) { /* Success */ |
62695a84 | 1376 | list_add_tail(&page->lru, &source); |
0efadf48 YX |
1377 | if (!__PageMovable(page)) |
1378 | inc_node_page_state(page, NR_ISOLATED_ANON + | |
1379 | page_is_file_cache(page)); | |
6d9c285a | 1380 | |
0c0e6195 | 1381 | } else { |
2932c8b0 | 1382 | pr_warn("failed to isolate pfn %lx\n", pfn); |
0efadf48 | 1383 | dump_page(page, "isolation failed"); |
0c0e6195 | 1384 | } |
1723058e | 1385 | put_page(page); |
0c0e6195 | 1386 | } |
f3ab2636 | 1387 | if (!list_empty(&source)) { |
394e31d2 XQ |
1388 | /* Allocate a new page from the nearest neighbor node */ |
1389 | ret = migrate_pages(&source, new_node_page, NULL, 0, | |
9c620e2b | 1390 | MIGRATE_SYNC, MR_MEMORY_HOTPLUG); |
2932c8b0 MH |
1391 | if (ret) { |
1392 | list_for_each_entry(page, &source, lru) { | |
1393 | pr_warn("migrating pfn %lx failed ret:%d ", | |
1394 | page_to_pfn(page), ret); | |
1395 | dump_page(page, "migration failure"); | |
1396 | } | |
c8721bbb | 1397 | putback_movable_pages(&source); |
2932c8b0 | 1398 | } |
0c0e6195 | 1399 | } |
1723058e | 1400 | |
0c0e6195 KH |
1401 | return ret; |
1402 | } | |
1403 | ||
1404 | /* | |
1405 | * remove from free_area[] and mark all as Reserved. | |
1406 | */ | |
1407 | static int | |
1408 | offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages, | |
1409 | void *data) | |
1410 | { | |
5557c766 | 1411 | unsigned long *offlined_pages = (unsigned long *)data; |
0c0e6195 | 1412 | |
5557c766 MH |
1413 | *offlined_pages += __offline_isolated_pages(start, start + nr_pages); |
1414 | return 0; | |
0c0e6195 KH |
1415 | } |
1416 | ||
1417 | /* | |
1418 | * Check all pages in range, recoreded as memory resource, are isolated. | |
1419 | */ | |
1420 | static int | |
1421 | check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages, | |
1422 | void *data) | |
1423 | { | |
5557c766 | 1424 | return test_pages_isolated(start_pfn, start_pfn + nr_pages, true); |
0c0e6195 KH |
1425 | } |
1426 | ||
c5320926 TC |
1427 | static int __init cmdline_parse_movable_node(char *p) |
1428 | { | |
4932381e | 1429 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
55ac590c | 1430 | movable_node_enabled = true; |
4932381e MH |
1431 | #else |
1432 | pr_warn("movable_node parameter depends on CONFIG_HAVE_MEMBLOCK_NODE_MAP to work properly\n"); | |
1433 | #endif | |
c5320926 TC |
1434 | return 0; |
1435 | } | |
1436 | early_param("movable_node", cmdline_parse_movable_node); | |
1437 | ||
d9713679 LJ |
1438 | /* check which state of node_states will be changed when offline memory */ |
1439 | static void node_states_check_changes_offline(unsigned long nr_pages, | |
1440 | struct zone *zone, struct memory_notify *arg) | |
1441 | { | |
1442 | struct pglist_data *pgdat = zone->zone_pgdat; | |
1443 | unsigned long present_pages = 0; | |
86b27bea | 1444 | enum zone_type zt; |
d9713679 | 1445 | |
98fa15f3 AK |
1446 | arg->status_change_nid = NUMA_NO_NODE; |
1447 | arg->status_change_nid_normal = NUMA_NO_NODE; | |
1448 | arg->status_change_nid_high = NUMA_NO_NODE; | |
d9713679 LJ |
1449 | |
1450 | /* | |
86b27bea OS |
1451 | * Check whether node_states[N_NORMAL_MEMORY] will be changed. |
1452 | * If the memory to be offline is within the range | |
1453 | * [0..ZONE_NORMAL], and it is the last present memory there, | |
1454 | * the zones in that range will become empty after the offlining, | |
1455 | * thus we can determine that we need to clear the node from | |
1456 | * node_states[N_NORMAL_MEMORY]. | |
d9713679 | 1457 | */ |
86b27bea | 1458 | for (zt = 0; zt <= ZONE_NORMAL; zt++) |
d9713679 | 1459 | present_pages += pgdat->node_zones[zt].present_pages; |
86b27bea | 1460 | if (zone_idx(zone) <= ZONE_NORMAL && nr_pages >= present_pages) |
d9713679 | 1461 | arg->status_change_nid_normal = zone_to_nid(zone); |
d9713679 | 1462 | |
6715ddf9 LJ |
1463 | #ifdef CONFIG_HIGHMEM |
1464 | /* | |
86b27bea OS |
1465 | * node_states[N_HIGH_MEMORY] contains nodes which |
1466 | * have normal memory or high memory. | |
1467 | * Here we add the present_pages belonging to ZONE_HIGHMEM. | |
1468 | * If the zone is within the range of [0..ZONE_HIGHMEM), and | |
1469 | * we determine that the zones in that range become empty, | |
1470 | * we need to clear the node for N_HIGH_MEMORY. | |
6715ddf9 | 1471 | */ |
86b27bea OS |
1472 | present_pages += pgdat->node_zones[ZONE_HIGHMEM].present_pages; |
1473 | if (zone_idx(zone) <= ZONE_HIGHMEM && nr_pages >= present_pages) | |
6715ddf9 | 1474 | arg->status_change_nid_high = zone_to_nid(zone); |
6715ddf9 LJ |
1475 | #endif |
1476 | ||
d9713679 | 1477 | /* |
86b27bea OS |
1478 | * We have accounted the pages from [0..ZONE_NORMAL), and |
1479 | * in case of CONFIG_HIGHMEM the pages from ZONE_HIGHMEM | |
1480 | * as well. | |
1481 | * Here we count the possible pages from ZONE_MOVABLE. | |
1482 | * If after having accounted all the pages, we see that the nr_pages | |
1483 | * to be offlined is over or equal to the accounted pages, | |
1484 | * we know that the node will become empty, and so, we can clear | |
1485 | * it for N_MEMORY as well. | |
d9713679 | 1486 | */ |
86b27bea | 1487 | present_pages += pgdat->node_zones[ZONE_MOVABLE].present_pages; |
d9713679 | 1488 | |
d9713679 LJ |
1489 | if (nr_pages >= present_pages) |
1490 | arg->status_change_nid = zone_to_nid(zone); | |
d9713679 LJ |
1491 | } |
1492 | ||
1493 | static void node_states_clear_node(int node, struct memory_notify *arg) | |
1494 | { | |
1495 | if (arg->status_change_nid_normal >= 0) | |
1496 | node_clear_state(node, N_NORMAL_MEMORY); | |
1497 | ||
cf01f6f5 | 1498 | if (arg->status_change_nid_high >= 0) |
d9713679 | 1499 | node_clear_state(node, N_HIGH_MEMORY); |
6715ddf9 | 1500 | |
cf01f6f5 | 1501 | if (arg->status_change_nid >= 0) |
6715ddf9 | 1502 | node_clear_state(node, N_MEMORY); |
d9713679 LJ |
1503 | } |
1504 | ||
a16cee10 | 1505 | static int __ref __offline_pages(unsigned long start_pfn, |
ecde0f3e | 1506 | unsigned long end_pfn) |
0c0e6195 | 1507 | { |
ecde0f3e | 1508 | unsigned long pfn, nr_pages; |
5557c766 | 1509 | unsigned long offlined_pages = 0; |
9b7ea46a | 1510 | int ret, node, nr_isolate_pageblock; |
d702909f | 1511 | unsigned long flags; |
a96dfddb | 1512 | unsigned long valid_start, valid_end; |
0c0e6195 | 1513 | struct zone *zone; |
7b78d335 | 1514 | struct memory_notify arg; |
79605093 | 1515 | char *reason; |
0c0e6195 | 1516 | |
381eab4a DH |
1517 | mem_hotplug_begin(); |
1518 | ||
0c0e6195 KH |
1519 | /* This makes hotplug much easier...and readable. |
1520 | we assume this for now. .*/ | |
381eab4a DH |
1521 | if (!test_pages_in_a_zone(start_pfn, end_pfn, &valid_start, |
1522 | &valid_end)) { | |
79605093 MH |
1523 | ret = -EINVAL; |
1524 | reason = "multizone range"; | |
1525 | goto failed_removal; | |
381eab4a | 1526 | } |
7b78d335 | 1527 | |
a96dfddb | 1528 | zone = page_zone(pfn_to_page(valid_start)); |
7b78d335 YG |
1529 | node = zone_to_nid(zone); |
1530 | nr_pages = end_pfn - start_pfn; | |
1531 | ||
0c0e6195 | 1532 | /* set above range as isolated */ |
b023f468 | 1533 | ret = start_isolate_page_range(start_pfn, end_pfn, |
d381c547 MH |
1534 | MIGRATE_MOVABLE, |
1535 | SKIP_HWPOISON | REPORT_FAILURE); | |
9b7ea46a | 1536 | if (ret < 0) { |
79605093 MH |
1537 | reason = "failure to isolate range"; |
1538 | goto failed_removal; | |
381eab4a | 1539 | } |
9b7ea46a | 1540 | nr_isolate_pageblock = ret; |
7b78d335 YG |
1541 | |
1542 | arg.start_pfn = start_pfn; | |
1543 | arg.nr_pages = nr_pages; | |
d9713679 | 1544 | node_states_check_changes_offline(nr_pages, zone, &arg); |
7b78d335 YG |
1545 | |
1546 | ret = memory_notify(MEM_GOING_OFFLINE, &arg); | |
1547 | ret = notifier_to_errno(ret); | |
79605093 MH |
1548 | if (ret) { |
1549 | reason = "notifier failure"; | |
1550 | goto failed_removal_isolated; | |
1551 | } | |
7b78d335 | 1552 | |
bb8965bd MH |
1553 | do { |
1554 | for (pfn = start_pfn; pfn;) { | |
1555 | if (signal_pending(current)) { | |
1556 | ret = -EINTR; | |
1557 | reason = "signal backoff"; | |
1558 | goto failed_removal_isolated; | |
1559 | } | |
72b39cfc | 1560 | |
bb8965bd MH |
1561 | cond_resched(); |
1562 | lru_add_drain_all(); | |
bb8965bd MH |
1563 | |
1564 | pfn = scan_movable_pages(pfn, end_pfn); | |
1565 | if (pfn) { | |
1566 | /* | |
1567 | * TODO: fatal migration failures should bail | |
1568 | * out | |
1569 | */ | |
1570 | do_migrate_range(pfn, end_pfn); | |
1571 | } | |
1572 | } | |
0c0e6195 | 1573 | |
bb8965bd MH |
1574 | /* |
1575 | * Dissolve free hugepages in the memory block before doing | |
1576 | * offlining actually in order to make hugetlbfs's object | |
1577 | * counting consistent. | |
1578 | */ | |
1579 | ret = dissolve_free_huge_pages(start_pfn, end_pfn); | |
1580 | if (ret) { | |
1581 | reason = "failure to dissolve huge pages"; | |
1582 | goto failed_removal_isolated; | |
1583 | } | |
1584 | /* check again */ | |
5557c766 MH |
1585 | ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn, |
1586 | NULL, check_pages_isolated_cb); | |
1587 | } while (ret); | |
72b39cfc | 1588 | |
b3834be5 | 1589 | /* Ok, all of our target is isolated. |
0c0e6195 | 1590 | We cannot do rollback at this point. */ |
5557c766 MH |
1591 | walk_system_ram_range(start_pfn, end_pfn - start_pfn, |
1592 | &offlined_pages, offline_isolated_pages_cb); | |
1593 | pr_info("Offlined Pages %ld\n", offlined_pages); | |
9b7ea46a QC |
1594 | /* |
1595 | * Onlining will reset pagetype flags and makes migrate type | |
1596 | * MOVABLE, so just need to decrease the number of isolated | |
1597 | * pageblocks zone counter here. | |
1598 | */ | |
1599 | spin_lock_irqsave(&zone->lock, flags); | |
1600 | zone->nr_isolate_pageblock -= nr_isolate_pageblock; | |
1601 | spin_unlock_irqrestore(&zone->lock, flags); | |
1602 | ||
0c0e6195 | 1603 | /* removal success */ |
3dcc0571 | 1604 | adjust_managed_page_count(pfn_to_page(start_pfn), -offlined_pages); |
0c0e6195 | 1605 | zone->present_pages -= offlined_pages; |
d702909f CS |
1606 | |
1607 | pgdat_resize_lock(zone->zone_pgdat, &flags); | |
0c0e6195 | 1608 | zone->zone_pgdat->node_present_pages -= offlined_pages; |
d702909f | 1609 | pgdat_resize_unlock(zone->zone_pgdat, &flags); |
7b78d335 | 1610 | |
1b79acc9 KM |
1611 | init_per_zone_wmark_min(); |
1612 | ||
1e8537ba | 1613 | if (!populated_zone(zone)) { |
340175b7 | 1614 | zone_pcp_reset(zone); |
72675e13 | 1615 | build_all_zonelists(NULL); |
1e8537ba XQ |
1616 | } else |
1617 | zone_pcp_update(zone); | |
340175b7 | 1618 | |
d9713679 | 1619 | node_states_clear_node(node, &arg); |
698b1b30 | 1620 | if (arg.status_change_nid >= 0) { |
8fe23e05 | 1621 | kswapd_stop(node); |
698b1b30 VB |
1622 | kcompactd_stop(node); |
1623 | } | |
bce7394a | 1624 | |
0c0e6195 KH |
1625 | vm_total_pages = nr_free_pagecache_pages(); |
1626 | writeback_set_ratelimit(); | |
7b78d335 YG |
1627 | |
1628 | memory_notify(MEM_OFFLINE, &arg); | |
381eab4a | 1629 | mem_hotplug_done(); |
0c0e6195 KH |
1630 | return 0; |
1631 | ||
79605093 MH |
1632 | failed_removal_isolated: |
1633 | undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE); | |
c4efe484 | 1634 | memory_notify(MEM_CANCEL_OFFLINE, &arg); |
0c0e6195 | 1635 | failed_removal: |
79605093 | 1636 | pr_debug("memory offlining [mem %#010llx-%#010llx] failed due to %s\n", |
e33e33b4 | 1637 | (unsigned long long) start_pfn << PAGE_SHIFT, |
79605093 MH |
1638 | ((unsigned long long) end_pfn << PAGE_SHIFT) - 1, |
1639 | reason); | |
0c0e6195 | 1640 | /* pushback to free area */ |
381eab4a | 1641 | mem_hotplug_done(); |
0c0e6195 KH |
1642 | return ret; |
1643 | } | |
71088785 | 1644 | |
a16cee10 WC |
1645 | int offline_pages(unsigned long start_pfn, unsigned long nr_pages) |
1646 | { | |
ecde0f3e | 1647 | return __offline_pages(start_pfn, start_pfn + nr_pages); |
a16cee10 WC |
1648 | } |
1649 | ||
d6de9d53 | 1650 | static int check_memblock_offlined_cb(struct memory_block *mem, void *arg) |
bbc76be6 WC |
1651 | { |
1652 | int ret = !is_memblock_offlined(mem); | |
1653 | ||
349daa0f RD |
1654 | if (unlikely(ret)) { |
1655 | phys_addr_t beginpa, endpa; | |
1656 | ||
1657 | beginpa = PFN_PHYS(section_nr_to_pfn(mem->start_section_nr)); | |
b6c88d3b | 1658 | endpa = beginpa + memory_block_size_bytes() - 1; |
756a025f | 1659 | pr_warn("removing memory fails, because memory [%pa-%pa] is onlined\n", |
349daa0f | 1660 | &beginpa, &endpa); |
bbc76be6 | 1661 | |
eca499ab PT |
1662 | return -EBUSY; |
1663 | } | |
1664 | return 0; | |
bbc76be6 WC |
1665 | } |
1666 | ||
0f1cfe9d | 1667 | static int check_cpu_on_node(pg_data_t *pgdat) |
60a5a19e | 1668 | { |
60a5a19e TC |
1669 | int cpu; |
1670 | ||
1671 | for_each_present_cpu(cpu) { | |
1672 | if (cpu_to_node(cpu) == pgdat->node_id) | |
1673 | /* | |
1674 | * the cpu on this node isn't removed, and we can't | |
1675 | * offline this node. | |
1676 | */ | |
1677 | return -EBUSY; | |
1678 | } | |
1679 | ||
1680 | return 0; | |
1681 | } | |
1682 | ||
0f1cfe9d TK |
1683 | /** |
1684 | * try_offline_node | |
e8b098fc | 1685 | * @nid: the node ID |
0f1cfe9d TK |
1686 | * |
1687 | * Offline a node if all memory sections and cpus of the node are removed. | |
1688 | * | |
1689 | * NOTE: The caller must call lock_device_hotplug() to serialize hotplug | |
1690 | * and online/offline operations before this call. | |
1691 | */ | |
90b30cdc | 1692 | void try_offline_node(int nid) |
60a5a19e | 1693 | { |
d822b86a WC |
1694 | pg_data_t *pgdat = NODE_DATA(nid); |
1695 | unsigned long start_pfn = pgdat->node_start_pfn; | |
1696 | unsigned long end_pfn = start_pfn + pgdat->node_spanned_pages; | |
60a5a19e TC |
1697 | unsigned long pfn; |
1698 | ||
1699 | for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) { | |
1700 | unsigned long section_nr = pfn_to_section_nr(pfn); | |
1701 | ||
1702 | if (!present_section_nr(section_nr)) | |
1703 | continue; | |
1704 | ||
1705 | if (pfn_to_nid(pfn) != nid) | |
1706 | continue; | |
1707 | ||
1708 | /* | |
1709 | * some memory sections of this node are not removed, and we | |
1710 | * can't offline node now. | |
1711 | */ | |
1712 | return; | |
1713 | } | |
1714 | ||
46a3679b | 1715 | if (check_cpu_on_node(pgdat)) |
60a5a19e TC |
1716 | return; |
1717 | ||
1718 | /* | |
1719 | * all memory/cpu of this node are removed, we can offline this | |
1720 | * node now. | |
1721 | */ | |
1722 | node_set_offline(nid); | |
1723 | unregister_one_node(nid); | |
1724 | } | |
90b30cdc | 1725 | EXPORT_SYMBOL(try_offline_node); |
60a5a19e | 1726 | |
d9eb1417 DH |
1727 | static void __release_memory_resource(resource_size_t start, |
1728 | resource_size_t size) | |
1729 | { | |
1730 | int ret; | |
1731 | ||
1732 | /* | |
1733 | * When removing memory in the same granularity as it was added, | |
1734 | * this function never fails. It might only fail if resources | |
1735 | * have to be adjusted or split. We'll ignore the error, as | |
1736 | * removing of memory cannot fail. | |
1737 | */ | |
1738 | ret = release_mem_region_adjustable(&iomem_resource, start, size); | |
1739 | if (ret) { | |
1740 | resource_size_t endres = start + size - 1; | |
1741 | ||
1742 | pr_warn("Unable to release resource <%pa-%pa> (%d)\n", | |
1743 | &start, &endres, ret); | |
1744 | } | |
1745 | } | |
1746 | ||
eca499ab | 1747 | static int __ref try_remove_memory(int nid, u64 start, u64 size) |
bbc76be6 | 1748 | { |
eca499ab | 1749 | int rc = 0; |
993c1aad | 1750 | |
27356f54 TK |
1751 | BUG_ON(check_hotplug_memory_range(start, size)); |
1752 | ||
bfc8c901 | 1753 | mem_hotplug_begin(); |
6677e3ea YI |
1754 | |
1755 | /* | |
242831eb | 1756 | * All memory blocks must be offlined before removing memory. Check |
eca499ab | 1757 | * whether all memory blocks in question are offline and return error |
242831eb | 1758 | * if this is not the case. |
6677e3ea | 1759 | */ |
fbcf73ce | 1760 | rc = walk_memory_blocks(start, size, NULL, check_memblock_offlined_cb); |
eca499ab PT |
1761 | if (rc) |
1762 | goto done; | |
6677e3ea | 1763 | |
46c66c4b YI |
1764 | /* remove memmap entry */ |
1765 | firmware_map_remove(start, start + size, "System RAM"); | |
f9126ab9 XQ |
1766 | memblock_free(start, size); |
1767 | memblock_remove(start, size); | |
4c4b7f9b DH |
1768 | |
1769 | /* remove memory block devices before removing memory */ | |
1770 | remove_memory_block_devices(start, size); | |
46c66c4b | 1771 | |
2c2a5af6 | 1772 | arch_remove_memory(nid, start, size, NULL); |
d9eb1417 | 1773 | __release_memory_resource(start, size); |
24d335ca | 1774 | |
60a5a19e TC |
1775 | try_offline_node(nid); |
1776 | ||
eca499ab | 1777 | done: |
bfc8c901 | 1778 | mem_hotplug_done(); |
eca499ab | 1779 | return rc; |
71088785 | 1780 | } |
d15e5926 | 1781 | |
eca499ab PT |
1782 | /** |
1783 | * remove_memory | |
1784 | * @nid: the node ID | |
1785 | * @start: physical address of the region to remove | |
1786 | * @size: size of the region to remove | |
1787 | * | |
1788 | * NOTE: The caller must call lock_device_hotplug() to serialize hotplug | |
1789 | * and online/offline operations before this call, as required by | |
1790 | * try_offline_node(). | |
1791 | */ | |
1792 | void __remove_memory(int nid, u64 start, u64 size) | |
1793 | { | |
1794 | ||
1795 | /* | |
29a90db9 | 1796 | * trigger BUG() if some memory is not offlined prior to calling this |
eca499ab PT |
1797 | * function |
1798 | */ | |
1799 | if (try_remove_memory(nid, start, size)) | |
1800 | BUG(); | |
1801 | } | |
1802 | ||
1803 | /* | |
1804 | * Remove memory if every memory block is offline, otherwise return -EBUSY is | |
1805 | * some memory is not offline | |
1806 | */ | |
1807 | int remove_memory(int nid, u64 start, u64 size) | |
d15e5926 | 1808 | { |
eca499ab PT |
1809 | int rc; |
1810 | ||
d15e5926 | 1811 | lock_device_hotplug(); |
eca499ab | 1812 | rc = try_remove_memory(nid, start, size); |
d15e5926 | 1813 | unlock_device_hotplug(); |
eca499ab PT |
1814 | |
1815 | return rc; | |
d15e5926 | 1816 | } |
71088785 | 1817 | EXPORT_SYMBOL_GPL(remove_memory); |
aba6efc4 | 1818 | #endif /* CONFIG_MEMORY_HOTREMOVE */ |