]>
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 | 168 | { |
f1eca35a | 169 | unsigned long mapsize, section_nr, i; |
04753278 YG |
170 | struct mem_section *ms; |
171 | struct page *page, *memmap; | |
f1eca35a | 172 | struct mem_section_usage *usage; |
04753278 | 173 | |
04753278 YG |
174 | section_nr = pfn_to_section_nr(start_pfn); |
175 | ms = __nr_to_section(section_nr); | |
176 | ||
177 | /* Get section's memmap address */ | |
178 | memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr); | |
179 | ||
180 | /* | |
181 | * Get page for the memmap's phys address | |
182 | * XXX: need more consideration for sparse_vmemmap... | |
183 | */ | |
184 | page = virt_to_page(memmap); | |
185 | mapsize = sizeof(struct page) * PAGES_PER_SECTION; | |
186 | mapsize = PAGE_ALIGN(mapsize) >> PAGE_SHIFT; | |
187 | ||
188 | /* remember memmap's page */ | |
189 | for (i = 0; i < mapsize; i++, page++) | |
190 | get_page_bootmem(section_nr, page, SECTION_INFO); | |
191 | ||
f1eca35a DW |
192 | usage = ms->usage; |
193 | page = virt_to_page(usage); | |
04753278 | 194 | |
f1eca35a | 195 | mapsize = PAGE_ALIGN(mem_section_usage_size()) >> PAGE_SHIFT; |
04753278 YG |
196 | |
197 | for (i = 0; i < mapsize; i++, page++) | |
af370fb8 | 198 | get_page_bootmem(section_nr, page, MIX_SECTION_INFO); |
04753278 YG |
199 | |
200 | } | |
46723bfa YI |
201 | #else /* CONFIG_SPARSEMEM_VMEMMAP */ |
202 | static void register_page_bootmem_info_section(unsigned long start_pfn) | |
203 | { | |
f1eca35a | 204 | unsigned long mapsize, section_nr, i; |
46723bfa YI |
205 | struct mem_section *ms; |
206 | struct page *page, *memmap; | |
f1eca35a | 207 | struct mem_section_usage *usage; |
46723bfa | 208 | |
46723bfa YI |
209 | section_nr = pfn_to_section_nr(start_pfn); |
210 | ms = __nr_to_section(section_nr); | |
211 | ||
212 | memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr); | |
213 | ||
214 | register_page_bootmem_memmap(section_nr, memmap, PAGES_PER_SECTION); | |
215 | ||
f1eca35a DW |
216 | usage = ms->usage; |
217 | page = virt_to_page(usage); | |
46723bfa | 218 | |
f1eca35a | 219 | mapsize = PAGE_ALIGN(mem_section_usage_size()) >> PAGE_SHIFT; |
46723bfa YI |
220 | |
221 | for (i = 0; i < mapsize; i++, page++) | |
222 | get_page_bootmem(section_nr, page, MIX_SECTION_INFO); | |
223 | } | |
224 | #endif /* !CONFIG_SPARSEMEM_VMEMMAP */ | |
04753278 | 225 | |
7ded384a | 226 | void __init register_page_bootmem_info_node(struct pglist_data *pgdat) |
04753278 YG |
227 | { |
228 | unsigned long i, pfn, end_pfn, nr_pages; | |
229 | int node = pgdat->node_id; | |
230 | struct page *page; | |
04753278 YG |
231 | |
232 | nr_pages = PAGE_ALIGN(sizeof(struct pglist_data)) >> PAGE_SHIFT; | |
233 | page = virt_to_page(pgdat); | |
234 | ||
235 | for (i = 0; i < nr_pages; i++, page++) | |
236 | get_page_bootmem(node, page, NODE_INFO); | |
237 | ||
04753278 | 238 | pfn = pgdat->node_start_pfn; |
c1f19495 | 239 | end_pfn = pgdat_end_pfn(pgdat); |
04753278 | 240 | |
7e9f5eb0 | 241 | /* register section info */ |
f14851af | 242 | for (; pfn < end_pfn; pfn += PAGES_PER_SECTION) { |
243 | /* | |
244 | * Some platforms can assign the same pfn to multiple nodes - on | |
245 | * node0 as well as nodeN. To avoid registering a pfn against | |
246 | * multiple nodes we check that this pfn does not already | |
7e9f5eb0 | 247 | * reside in some other nodes. |
f14851af | 248 | */ |
f65e91df | 249 | if (pfn_valid(pfn) && (early_pfn_to_nid(pfn) == node)) |
f14851af | 250 | register_page_bootmem_info_section(pfn); |
251 | } | |
04753278 | 252 | } |
46723bfa | 253 | #endif /* CONFIG_HAVE_BOOTMEM_INFO_NODE */ |
04753278 | 254 | |
7ea62160 DW |
255 | static int check_pfn_span(unsigned long pfn, unsigned long nr_pages, |
256 | const char *reason) | |
257 | { | |
258 | /* | |
259 | * Disallow all operations smaller than a sub-section and only | |
260 | * allow operations smaller than a section for | |
261 | * SPARSEMEM_VMEMMAP. Note that check_hotplug_memory_range() | |
262 | * enforces a larger memory_block_size_bytes() granularity for | |
263 | * memory that will be marked online, so this check should only | |
264 | * fire for direct arch_{add,remove}_memory() users outside of | |
265 | * add_memory_resource(). | |
266 | */ | |
267 | unsigned long min_align; | |
268 | ||
269 | if (IS_ENABLED(CONFIG_SPARSEMEM_VMEMMAP)) | |
270 | min_align = PAGES_PER_SUBSECTION; | |
271 | else | |
272 | min_align = PAGES_PER_SECTION; | |
273 | if (!IS_ALIGNED(pfn, min_align) | |
274 | || !IS_ALIGNED(nr_pages, min_align)) { | |
275 | WARN(1, "Misaligned __%s_pages start: %#lx end: #%lx\n", | |
276 | reason, pfn, pfn + nr_pages - 1); | |
277 | return -EINVAL; | |
278 | } | |
279 | return 0; | |
280 | } | |
281 | ||
4edd7cef DR |
282 | /* |
283 | * Reasonably generic function for adding memory. It is | |
284 | * expected that archs that support memory hotplug will | |
285 | * call this function after deciding the zone to which to | |
286 | * add the new pages. | |
287 | */ | |
7ea62160 DW |
288 | int __ref __add_pages(int nid, unsigned long pfn, unsigned long nr_pages, |
289 | struct mhp_restrictions *restrictions) | |
4edd7cef DR |
290 | { |
291 | unsigned long i; | |
7ea62160 | 292 | int start_sec, end_sec, err; |
940519f0 | 293 | struct vmem_altmap *altmap = restrictions->altmap; |
4b94ffdc | 294 | |
4b94ffdc DW |
295 | if (altmap) { |
296 | /* | |
297 | * Validate altmap is within bounds of the total request | |
298 | */ | |
7ea62160 | 299 | if (altmap->base_pfn != pfn |
4b94ffdc DW |
300 | || vmem_altmap_offset(altmap) > nr_pages) { |
301 | pr_warn_once("memory add fail, invalid altmap\n"); | |
7ea62160 | 302 | return -EINVAL; |
4b94ffdc DW |
303 | } |
304 | altmap->alloc = 0; | |
305 | } | |
306 | ||
7ea62160 DW |
307 | err = check_pfn_span(pfn, nr_pages, "add"); |
308 | if (err) | |
309 | return err; | |
310 | ||
311 | start_sec = pfn_to_section_nr(pfn); | |
312 | end_sec = pfn_to_section_nr(pfn + nr_pages - 1); | |
4edd7cef | 313 | for (i = start_sec; i <= end_sec; i++) { |
7ea62160 DW |
314 | unsigned long pfns; |
315 | ||
316 | pfns = min(nr_pages, PAGES_PER_SECTION | |
317 | - (pfn & ~PAGE_SECTION_MASK)); | |
ba72b4c8 DW |
318 | err = sparse_add_section(nid, pfn, pfns, altmap); |
319 | if (err) | |
320 | break; | |
7ea62160 DW |
321 | pfn += pfns; |
322 | nr_pages -= pfns; | |
f64ac5e6 | 323 | cond_resched(); |
4edd7cef | 324 | } |
c435a390 | 325 | vmemmap_populate_print_last(); |
4edd7cef DR |
326 | return err; |
327 | } | |
4edd7cef | 328 | |
815121d2 | 329 | /* find the smallest valid pfn in the range [start_pfn, end_pfn) */ |
d09b0137 | 330 | static unsigned long find_smallest_section_pfn(int nid, struct zone *zone, |
815121d2 YI |
331 | unsigned long start_pfn, |
332 | unsigned long end_pfn) | |
333 | { | |
49ba3c6b DW |
334 | for (; start_pfn < end_pfn; start_pfn += PAGES_PER_SUBSECTION) { |
335 | if (unlikely(!pfn_valid(start_pfn))) | |
815121d2 YI |
336 | continue; |
337 | ||
338 | if (unlikely(pfn_to_nid(start_pfn) != nid)) | |
339 | continue; | |
340 | ||
341 | if (zone && zone != page_zone(pfn_to_page(start_pfn))) | |
342 | continue; | |
343 | ||
344 | return start_pfn; | |
345 | } | |
346 | ||
347 | return 0; | |
348 | } | |
349 | ||
350 | /* find the biggest valid pfn in the range [start_pfn, end_pfn). */ | |
d09b0137 | 351 | static unsigned long find_biggest_section_pfn(int nid, struct zone *zone, |
815121d2 YI |
352 | unsigned long start_pfn, |
353 | unsigned long end_pfn) | |
354 | { | |
815121d2 YI |
355 | unsigned long pfn; |
356 | ||
357 | /* pfn is the end pfn of a memory section. */ | |
358 | pfn = end_pfn - 1; | |
49ba3c6b DW |
359 | for (; pfn >= start_pfn; pfn -= PAGES_PER_SUBSECTION) { |
360 | if (unlikely(!pfn_valid(pfn))) | |
815121d2 YI |
361 | continue; |
362 | ||
363 | if (unlikely(pfn_to_nid(pfn) != nid)) | |
364 | continue; | |
365 | ||
366 | if (zone && zone != page_zone(pfn_to_page(pfn))) | |
367 | continue; | |
368 | ||
369 | return pfn; | |
370 | } | |
371 | ||
372 | return 0; | |
373 | } | |
374 | ||
375 | static void shrink_zone_span(struct zone *zone, unsigned long start_pfn, | |
376 | unsigned long end_pfn) | |
377 | { | |
c33bc315 XQ |
378 | unsigned long zone_start_pfn = zone->zone_start_pfn; |
379 | unsigned long z = zone_end_pfn(zone); /* zone_end_pfn namespace clash */ | |
380 | unsigned long zone_end_pfn = z; | |
815121d2 | 381 | unsigned long pfn; |
815121d2 YI |
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; | |
49ba3c6b DW |
418 | for (; pfn < zone_end_pfn; pfn += PAGES_PER_SUBSECTION) { |
419 | if (unlikely(!pfn_valid(pfn))) | |
815121d2 YI |
420 | continue; |
421 | ||
422 | if (page_zone(pfn_to_page(pfn)) != zone) | |
423 | continue; | |
424 | ||
49ba3c6b DW |
425 | /* Skip range to be removed */ |
426 | if (pfn >= start_pfn && pfn < end_pfn) | |
815121d2 YI |
427 | continue; |
428 | ||
429 | /* If we find valid section, we have nothing to do */ | |
430 | zone_span_writeunlock(zone); | |
431 | return; | |
432 | } | |
433 | ||
434 | /* The zone has no valid section */ | |
435 | zone->zone_start_pfn = 0; | |
436 | zone->spanned_pages = 0; | |
437 | zone_span_writeunlock(zone); | |
438 | } | |
439 | ||
440 | static void shrink_pgdat_span(struct pglist_data *pgdat, | |
441 | unsigned long start_pfn, unsigned long end_pfn) | |
442 | { | |
83285c72 XQ |
443 | unsigned long pgdat_start_pfn = pgdat->node_start_pfn; |
444 | unsigned long p = pgdat_end_pfn(pgdat); /* pgdat_end_pfn namespace clash */ | |
445 | unsigned long pgdat_end_pfn = p; | |
815121d2 | 446 | unsigned long pfn; |
815121d2 YI |
447 | int nid = pgdat->node_id; |
448 | ||
449 | if (pgdat_start_pfn == start_pfn) { | |
450 | /* | |
451 | * If the section is smallest section in the pgdat, it need | |
452 | * shrink pgdat->node_start_pfn and pgdat->node_spanned_pages. | |
453 | * In this case, we find second smallest valid mem_section | |
454 | * for shrinking zone. | |
455 | */ | |
456 | pfn = find_smallest_section_pfn(nid, NULL, end_pfn, | |
457 | pgdat_end_pfn); | |
458 | if (pfn) { | |
459 | pgdat->node_start_pfn = pfn; | |
460 | pgdat->node_spanned_pages = pgdat_end_pfn - pfn; | |
461 | } | |
462 | } else if (pgdat_end_pfn == end_pfn) { | |
463 | /* | |
464 | * If the section is biggest section in the pgdat, it need | |
465 | * shrink pgdat->node_spanned_pages. | |
466 | * In this case, we find second biggest valid mem_section for | |
467 | * shrinking zone. | |
468 | */ | |
469 | pfn = find_biggest_section_pfn(nid, NULL, pgdat_start_pfn, | |
470 | start_pfn); | |
471 | if (pfn) | |
472 | pgdat->node_spanned_pages = pfn - pgdat_start_pfn + 1; | |
473 | } | |
474 | ||
475 | /* | |
476 | * If the section is not biggest or smallest mem_section in the pgdat, | |
477 | * it only creates a hole in the pgdat. So in this case, we need not | |
478 | * change the pgdat. | |
479 | * But perhaps, the pgdat has only hole data. Thus it check the pgdat | |
480 | * has only hole or not. | |
481 | */ | |
482 | pfn = pgdat_start_pfn; | |
49ba3c6b DW |
483 | for (; pfn < pgdat_end_pfn; pfn += PAGES_PER_SUBSECTION) { |
484 | if (unlikely(!pfn_valid(pfn))) | |
815121d2 YI |
485 | continue; |
486 | ||
487 | if (pfn_to_nid(pfn) != nid) | |
488 | continue; | |
489 | ||
49ba3c6b DW |
490 | /* Skip range to be removed */ |
491 | if (pfn >= start_pfn && pfn < end_pfn) | |
815121d2 YI |
492 | continue; |
493 | ||
494 | /* If we find valid section, we have nothing to do */ | |
495 | return; | |
496 | } | |
497 | ||
498 | /* The pgdat has no valid section */ | |
499 | pgdat->node_start_pfn = 0; | |
500 | pgdat->node_spanned_pages = 0; | |
501 | } | |
502 | ||
7ea62160 DW |
503 | static void __remove_zone(struct zone *zone, unsigned long start_pfn, |
504 | unsigned long nr_pages) | |
815121d2 YI |
505 | { |
506 | struct pglist_data *pgdat = zone->zone_pgdat; | |
815121d2 YI |
507 | unsigned long flags; |
508 | ||
815121d2 YI |
509 | pgdat_resize_lock(zone->zone_pgdat, &flags); |
510 | shrink_zone_span(zone, start_pfn, start_pfn + nr_pages); | |
511 | shrink_pgdat_span(pgdat, start_pfn, start_pfn + nr_pages); | |
512 | pgdat_resize_unlock(zone->zone_pgdat, &flags); | |
513 | } | |
514 | ||
7ea62160 DW |
515 | static void __remove_section(struct zone *zone, unsigned long pfn, |
516 | unsigned long nr_pages, unsigned long map_offset, | |
517 | struct vmem_altmap *altmap) | |
ea01ea93 | 518 | { |
7ea62160 | 519 | struct mem_section *ms = __nr_to_section(pfn_to_section_nr(pfn)); |
ea01ea93 | 520 | |
9d1d887d DH |
521 | if (WARN_ON_ONCE(!valid_section(ms))) |
522 | return; | |
ea01ea93 | 523 | |
7ea62160 | 524 | __remove_zone(zone, pfn, nr_pages); |
ba72b4c8 | 525 | sparse_remove_section(ms, pfn, nr_pages, map_offset, altmap); |
ea01ea93 BP |
526 | } |
527 | ||
ea01ea93 BP |
528 | /** |
529 | * __remove_pages() - remove sections of pages from a zone | |
530 | * @zone: zone from which pages need to be removed | |
7ea62160 | 531 | * @pfn: starting pageframe (must be aligned to start of a section) |
ea01ea93 | 532 | * @nr_pages: number of pages to remove (must be multiple of section size) |
e8b098fc | 533 | * @altmap: alternative device page map or %NULL if default memmap is used |
ea01ea93 BP |
534 | * |
535 | * Generic helper function to remove section mappings and sysfs entries | |
536 | * for the section of the memory we are removing. Caller needs to make | |
537 | * sure that pages are marked reserved and zones are adjust properly by | |
538 | * calling offline_pages(). | |
539 | */ | |
7ea62160 | 540 | void __remove_pages(struct zone *zone, unsigned long pfn, |
ac5c9426 | 541 | unsigned long nr_pages, struct vmem_altmap *altmap) |
ea01ea93 | 542 | { |
4b94ffdc | 543 | unsigned long map_offset = 0; |
7ea62160 | 544 | int i, start_sec, end_sec; |
4b94ffdc | 545 | |
96da4350 | 546 | map_offset = vmem_altmap_offset(altmap); |
ea01ea93 | 547 | |
7cf91a98 JK |
548 | clear_zone_contiguous(zone); |
549 | ||
7ea62160 DW |
550 | if (check_pfn_span(pfn, nr_pages, "remove")) |
551 | return; | |
ea01ea93 | 552 | |
7ea62160 DW |
553 | start_sec = pfn_to_section_nr(pfn); |
554 | end_sec = pfn_to_section_nr(pfn + nr_pages - 1); | |
555 | for (i = start_sec; i <= end_sec; i++) { | |
556 | unsigned long pfns; | |
4b94ffdc | 557 | |
dd33ad7b | 558 | cond_resched(); |
7ea62160 DW |
559 | pfns = min(nr_pages, PAGES_PER_SECTION |
560 | - (pfn & ~PAGE_SECTION_MASK)); | |
561 | __remove_section(zone, pfn, pfns, map_offset, altmap); | |
562 | pfn += pfns; | |
563 | nr_pages -= pfns; | |
4b94ffdc | 564 | map_offset = 0; |
ea01ea93 | 565 | } |
7cf91a98 JK |
566 | |
567 | set_zone_contiguous(zone); | |
ea01ea93 | 568 | } |
ea01ea93 | 569 | |
9d0ad8ca DK |
570 | int set_online_page_callback(online_page_callback_t callback) |
571 | { | |
572 | int rc = -EINVAL; | |
573 | ||
bfc8c901 VD |
574 | get_online_mems(); |
575 | mutex_lock(&online_page_callback_lock); | |
9d0ad8ca DK |
576 | |
577 | if (online_page_callback == generic_online_page) { | |
578 | online_page_callback = callback; | |
579 | rc = 0; | |
580 | } | |
581 | ||
bfc8c901 VD |
582 | mutex_unlock(&online_page_callback_lock); |
583 | put_online_mems(); | |
9d0ad8ca DK |
584 | |
585 | return rc; | |
586 | } | |
587 | EXPORT_SYMBOL_GPL(set_online_page_callback); | |
588 | ||
589 | int restore_online_page_callback(online_page_callback_t callback) | |
590 | { | |
591 | int rc = -EINVAL; | |
592 | ||
bfc8c901 VD |
593 | get_online_mems(); |
594 | mutex_lock(&online_page_callback_lock); | |
9d0ad8ca DK |
595 | |
596 | if (online_page_callback == callback) { | |
597 | online_page_callback = generic_online_page; | |
598 | rc = 0; | |
599 | } | |
600 | ||
bfc8c901 VD |
601 | mutex_unlock(&online_page_callback_lock); |
602 | put_online_mems(); | |
9d0ad8ca DK |
603 | |
604 | return rc; | |
605 | } | |
606 | EXPORT_SYMBOL_GPL(restore_online_page_callback); | |
607 | ||
608 | void __online_page_set_limits(struct page *page) | |
180c06ef | 609 | { |
9d0ad8ca DK |
610 | } |
611 | EXPORT_SYMBOL_GPL(__online_page_set_limits); | |
612 | ||
613 | void __online_page_increment_counters(struct page *page) | |
614 | { | |
3dcc0571 | 615 | adjust_managed_page_count(page, 1); |
9d0ad8ca DK |
616 | } |
617 | EXPORT_SYMBOL_GPL(__online_page_increment_counters); | |
180c06ef | 618 | |
9d0ad8ca DK |
619 | void __online_page_free(struct page *page) |
620 | { | |
3dcc0571 | 621 | __free_reserved_page(page); |
180c06ef | 622 | } |
9d0ad8ca DK |
623 | EXPORT_SYMBOL_GPL(__online_page_free); |
624 | ||
a9cd410a | 625 | static void generic_online_page(struct page *page, unsigned int order) |
9d0ad8ca | 626 | { |
cd02cf1a | 627 | kernel_map_pages(page, 1 << order, 1); |
a9cd410a AK |
628 | __free_pages_core(page, order); |
629 | totalram_pages_add(1UL << order); | |
630 | #ifdef CONFIG_HIGHMEM | |
631 | if (PageHighMem(page)) | |
632 | totalhigh_pages_add(1UL << order); | |
633 | #endif | |
634 | } | |
635 | ||
636 | static int online_pages_blocks(unsigned long start, unsigned long nr_pages) | |
637 | { | |
638 | unsigned long end = start + nr_pages; | |
639 | int order, onlined_pages = 0; | |
640 | ||
641 | while (start < end) { | |
642 | order = min(MAX_ORDER - 1, | |
643 | get_order(PFN_PHYS(end) - PFN_PHYS(start))); | |
644 | (*online_page_callback)(pfn_to_page(start), order); | |
645 | ||
646 | onlined_pages += (1UL << order); | |
647 | start += (1UL << order); | |
648 | } | |
649 | return onlined_pages; | |
9d0ad8ca | 650 | } |
180c06ef | 651 | |
75884fb1 KH |
652 | static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages, |
653 | void *arg) | |
3947be19 | 654 | { |
75884fb1 | 655 | unsigned long onlined_pages = *(unsigned long *)arg; |
2d070eab | 656 | |
75884fb1 | 657 | if (PageReserved(pfn_to_page(start_pfn))) |
a9cd410a | 658 | onlined_pages += online_pages_blocks(start_pfn, nr_pages); |
2d070eab MH |
659 | |
660 | online_mem_sections(start_pfn, start_pfn + nr_pages); | |
661 | ||
75884fb1 KH |
662 | *(unsigned long *)arg = onlined_pages; |
663 | return 0; | |
664 | } | |
665 | ||
d9713679 LJ |
666 | /* check which state of node_states will be changed when online memory */ |
667 | static void node_states_check_changes_online(unsigned long nr_pages, | |
668 | struct zone *zone, struct memory_notify *arg) | |
669 | { | |
670 | int nid = zone_to_nid(zone); | |
d9713679 | 671 | |
98fa15f3 AK |
672 | arg->status_change_nid = NUMA_NO_NODE; |
673 | arg->status_change_nid_normal = NUMA_NO_NODE; | |
674 | arg->status_change_nid_high = NUMA_NO_NODE; | |
d9713679 | 675 | |
8efe33f4 OS |
676 | if (!node_state(nid, N_MEMORY)) |
677 | arg->status_change_nid = nid; | |
678 | if (zone_idx(zone) <= ZONE_NORMAL && !node_state(nid, N_NORMAL_MEMORY)) | |
d9713679 | 679 | arg->status_change_nid_normal = nid; |
6715ddf9 | 680 | #ifdef CONFIG_HIGHMEM |
d3ba3ae1 | 681 | if (zone_idx(zone) <= ZONE_HIGHMEM && !node_state(nid, N_HIGH_MEMORY)) |
6715ddf9 | 682 | arg->status_change_nid_high = nid; |
6715ddf9 | 683 | #endif |
d9713679 LJ |
684 | } |
685 | ||
686 | static void node_states_set_node(int node, struct memory_notify *arg) | |
687 | { | |
688 | if (arg->status_change_nid_normal >= 0) | |
689 | node_set_state(node, N_NORMAL_MEMORY); | |
690 | ||
6715ddf9 LJ |
691 | if (arg->status_change_nid_high >= 0) |
692 | node_set_state(node, N_HIGH_MEMORY); | |
693 | ||
83d83612 OS |
694 | if (arg->status_change_nid >= 0) |
695 | node_set_state(node, N_MEMORY); | |
d9713679 LJ |
696 | } |
697 | ||
f1dd2cd1 MH |
698 | static void __meminit resize_zone_range(struct zone *zone, unsigned long start_pfn, |
699 | unsigned long nr_pages) | |
700 | { | |
701 | unsigned long old_end_pfn = zone_end_pfn(zone); | |
702 | ||
703 | if (zone_is_empty(zone) || start_pfn < zone->zone_start_pfn) | |
704 | zone->zone_start_pfn = start_pfn; | |
705 | ||
706 | zone->spanned_pages = max(start_pfn + nr_pages, old_end_pfn) - zone->zone_start_pfn; | |
707 | } | |
708 | ||
709 | static void __meminit resize_pgdat_range(struct pglist_data *pgdat, unsigned long start_pfn, | |
710 | unsigned long nr_pages) | |
711 | { | |
712 | unsigned long old_end_pfn = pgdat_end_pfn(pgdat); | |
713 | ||
714 | if (!pgdat->node_spanned_pages || start_pfn < pgdat->node_start_pfn) | |
715 | pgdat->node_start_pfn = start_pfn; | |
716 | ||
717 | pgdat->node_spanned_pages = max(start_pfn + nr_pages, old_end_pfn) - pgdat->node_start_pfn; | |
718 | } | |
719 | ||
a99583e7 CH |
720 | void __ref move_pfn_range_to_zone(struct zone *zone, unsigned long start_pfn, |
721 | unsigned long nr_pages, struct vmem_altmap *altmap) | |
f1dd2cd1 MH |
722 | { |
723 | struct pglist_data *pgdat = zone->zone_pgdat; | |
724 | int nid = pgdat->node_id; | |
725 | unsigned long flags; | |
df429ac0 | 726 | |
f1dd2cd1 MH |
727 | clear_zone_contiguous(zone); |
728 | ||
729 | /* TODO Huh pgdat is irqsave while zone is not. It used to be like that before */ | |
730 | pgdat_resize_lock(pgdat, &flags); | |
731 | zone_span_writelock(zone); | |
fa004ab7 WY |
732 | if (zone_is_empty(zone)) |
733 | init_currently_empty_zone(zone, start_pfn, nr_pages); | |
f1dd2cd1 MH |
734 | resize_zone_range(zone, start_pfn, nr_pages); |
735 | zone_span_writeunlock(zone); | |
736 | resize_pgdat_range(pgdat, start_pfn, nr_pages); | |
737 | pgdat_resize_unlock(pgdat, &flags); | |
738 | ||
739 | /* | |
740 | * TODO now we have a visible range of pages which are not associated | |
741 | * with their zone properly. Not nice but set_pfnblock_flags_mask | |
742 | * expects the zone spans the pfn range. All the pages in the range | |
743 | * are reserved so nobody should be touching them so we should be safe | |
744 | */ | |
a99583e7 CH |
745 | memmap_init_zone(nr_pages, nid, zone_idx(zone), start_pfn, |
746 | MEMMAP_HOTPLUG, altmap); | |
f1dd2cd1 MH |
747 | |
748 | set_zone_contiguous(zone); | |
749 | } | |
750 | ||
c246a213 MH |
751 | /* |
752 | * Returns a default kernel memory zone for the given pfn range. | |
753 | * If no kernel zone covers this pfn range it will automatically go | |
754 | * to the ZONE_NORMAL. | |
755 | */ | |
c6f03e29 | 756 | static struct zone *default_kernel_zone_for_pfn(int nid, unsigned long start_pfn, |
c246a213 MH |
757 | unsigned long nr_pages) |
758 | { | |
759 | struct pglist_data *pgdat = NODE_DATA(nid); | |
760 | int zid; | |
761 | ||
762 | for (zid = 0; zid <= ZONE_NORMAL; zid++) { | |
763 | struct zone *zone = &pgdat->node_zones[zid]; | |
764 | ||
765 | if (zone_intersects(zone, start_pfn, nr_pages)) | |
766 | return zone; | |
767 | } | |
768 | ||
769 | return &pgdat->node_zones[ZONE_NORMAL]; | |
770 | } | |
771 | ||
c6f03e29 MH |
772 | static inline struct zone *default_zone_for_pfn(int nid, unsigned long start_pfn, |
773 | unsigned long nr_pages) | |
e5e68930 | 774 | { |
c6f03e29 MH |
775 | struct zone *kernel_zone = default_kernel_zone_for_pfn(nid, start_pfn, |
776 | nr_pages); | |
777 | struct zone *movable_zone = &NODE_DATA(nid)->node_zones[ZONE_MOVABLE]; | |
778 | bool in_kernel = zone_intersects(kernel_zone, start_pfn, nr_pages); | |
779 | bool in_movable = zone_intersects(movable_zone, start_pfn, nr_pages); | |
e5e68930 MH |
780 | |
781 | /* | |
c6f03e29 MH |
782 | * We inherit the existing zone in a simple case where zones do not |
783 | * overlap in the given range | |
e5e68930 | 784 | */ |
c6f03e29 MH |
785 | if (in_kernel ^ in_movable) |
786 | return (in_kernel) ? kernel_zone : movable_zone; | |
9f123ab5 | 787 | |
c6f03e29 MH |
788 | /* |
789 | * If the range doesn't belong to any zone or two zones overlap in the | |
790 | * given range then we use movable zone only if movable_node is | |
791 | * enabled because we always online to a kernel zone by default. | |
792 | */ | |
793 | return movable_node_enabled ? movable_zone : kernel_zone; | |
9f123ab5 MH |
794 | } |
795 | ||
e5e68930 MH |
796 | struct zone * zone_for_pfn_range(int online_type, int nid, unsigned start_pfn, |
797 | unsigned long nr_pages) | |
f1dd2cd1 | 798 | { |
c6f03e29 MH |
799 | if (online_type == MMOP_ONLINE_KERNEL) |
800 | return default_kernel_zone_for_pfn(nid, start_pfn, nr_pages); | |
f1dd2cd1 | 801 | |
c6f03e29 MH |
802 | if (online_type == MMOP_ONLINE_MOVABLE) |
803 | return &NODE_DATA(nid)->node_zones[ZONE_MOVABLE]; | |
df429ac0 | 804 | |
c6f03e29 | 805 | return default_zone_for_pfn(nid, start_pfn, nr_pages); |
e5e68930 MH |
806 | } |
807 | ||
808 | /* | |
809 | * Associates the given pfn range with the given node and the zone appropriate | |
810 | * for the given online type. | |
811 | */ | |
812 | static struct zone * __meminit move_pfn_range(int online_type, int nid, | |
813 | unsigned long start_pfn, unsigned long nr_pages) | |
814 | { | |
815 | struct zone *zone; | |
816 | ||
817 | zone = zone_for_pfn_range(online_type, nid, start_pfn, nr_pages); | |
a99583e7 | 818 | move_pfn_range_to_zone(zone, start_pfn, nr_pages, NULL); |
f1dd2cd1 | 819 | return zone; |
df429ac0 | 820 | } |
75884fb1 | 821 | |
511c2aba | 822 | int __ref online_pages(unsigned long pfn, unsigned long nr_pages, int online_type) |
75884fb1 | 823 | { |
aa47228a | 824 | unsigned long flags; |
3947be19 DH |
825 | unsigned long onlined_pages = 0; |
826 | struct zone *zone; | |
6811378e | 827 | int need_zonelists_rebuild = 0; |
7b78d335 YG |
828 | int nid; |
829 | int ret; | |
830 | struct memory_notify arg; | |
d0dc12e8 PT |
831 | struct memory_block *mem; |
832 | ||
381eab4a DH |
833 | mem_hotplug_begin(); |
834 | ||
d0dc12e8 PT |
835 | /* |
836 | * We can't use pfn_to_nid() because nid might be stored in struct page | |
837 | * which is not yet initialized. Instead, we find nid from memory block. | |
838 | */ | |
839 | mem = find_memory_block(__pfn_to_section(pfn)); | |
840 | nid = mem->nid; | |
89c02e69 | 841 | put_device(&mem->dev); |
7b78d335 | 842 | |
f1dd2cd1 MH |
843 | /* associate pfn range with the zone */ |
844 | zone = move_pfn_range(online_type, nid, pfn, nr_pages); | |
845 | ||
7b78d335 YG |
846 | arg.start_pfn = pfn; |
847 | arg.nr_pages = nr_pages; | |
d9713679 | 848 | node_states_check_changes_online(nr_pages, zone, &arg); |
7b78d335 | 849 | |
7b78d335 YG |
850 | ret = memory_notify(MEM_GOING_ONLINE, &arg); |
851 | ret = notifier_to_errno(ret); | |
e33e33b4 CY |
852 | if (ret) |
853 | goto failed_addition; | |
854 | ||
6811378e YG |
855 | /* |
856 | * If this zone is not populated, then it is not in zonelist. | |
857 | * This means the page allocator ignores this zone. | |
858 | * So, zonelist must be updated after online. | |
859 | */ | |
6dcd73d7 | 860 | if (!populated_zone(zone)) { |
6811378e | 861 | need_zonelists_rebuild = 1; |
72675e13 | 862 | setup_zone_pageset(zone); |
6dcd73d7 | 863 | } |
6811378e | 864 | |
908eedc6 | 865 | ret = walk_system_ram_range(pfn, nr_pages, &onlined_pages, |
75884fb1 | 866 | online_pages_range); |
fd8a4221 | 867 | if (ret) { |
6dcd73d7 WC |
868 | if (need_zonelists_rebuild) |
869 | zone_pcp_reset(zone); | |
e33e33b4 | 870 | goto failed_addition; |
fd8a4221 GL |
871 | } |
872 | ||
3947be19 | 873 | zone->present_pages += onlined_pages; |
aa47228a CS |
874 | |
875 | pgdat_resize_lock(zone->zone_pgdat, &flags); | |
f2937be5 | 876 | zone->zone_pgdat->node_present_pages += onlined_pages; |
aa47228a CS |
877 | pgdat_resize_unlock(zone->zone_pgdat, &flags); |
878 | ||
e900a918 DW |
879 | shuffle_zone(zone); |
880 | ||
08dff7b7 | 881 | if (onlined_pages) { |
e888ca35 | 882 | node_states_set_node(nid, &arg); |
08dff7b7 | 883 | if (need_zonelists_rebuild) |
72675e13 | 884 | build_all_zonelists(NULL); |
08dff7b7 JL |
885 | else |
886 | zone_pcp_update(zone); | |
887 | } | |
3947be19 | 888 | |
1b79acc9 KM |
889 | init_per_zone_wmark_min(); |
890 | ||
698b1b30 | 891 | if (onlined_pages) { |
e888ca35 | 892 | kswapd_run(nid); |
698b1b30 VB |
893 | kcompactd_run(nid); |
894 | } | |
61b13993 | 895 | |
1f522509 | 896 | vm_total_pages = nr_free_pagecache_pages(); |
2f7f24ec | 897 | |
2d1d43f6 | 898 | writeback_set_ratelimit(); |
7b78d335 YG |
899 | |
900 | if (onlined_pages) | |
901 | memory_notify(MEM_ONLINE, &arg); | |
381eab4a | 902 | mem_hotplug_done(); |
30467e0b | 903 | return 0; |
e33e33b4 CY |
904 | |
905 | failed_addition: | |
906 | pr_debug("online_pages [mem %#010llx-%#010llx] failed\n", | |
907 | (unsigned long long) pfn << PAGE_SHIFT, | |
908 | (((unsigned long long) pfn + nr_pages) << PAGE_SHIFT) - 1); | |
909 | memory_notify(MEM_CANCEL_ONLINE, &arg); | |
381eab4a | 910 | mem_hotplug_done(); |
e33e33b4 | 911 | return ret; |
3947be19 | 912 | } |
53947027 | 913 | #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */ |
bc02af93 | 914 | |
0bd85420 TC |
915 | static void reset_node_present_pages(pg_data_t *pgdat) |
916 | { | |
917 | struct zone *z; | |
918 | ||
919 | for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++) | |
920 | z->present_pages = 0; | |
921 | ||
922 | pgdat->node_present_pages = 0; | |
923 | } | |
924 | ||
e1319331 HS |
925 | /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */ |
926 | static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start) | |
9af3c2de YG |
927 | { |
928 | struct pglist_data *pgdat; | |
c8e861a5 | 929 | unsigned long start_pfn = PFN_DOWN(start); |
9af3c2de | 930 | |
a1e565aa TC |
931 | pgdat = NODE_DATA(nid); |
932 | if (!pgdat) { | |
933 | pgdat = arch_alloc_nodedata(nid); | |
934 | if (!pgdat) | |
935 | return NULL; | |
9af3c2de | 936 | |
a1e565aa | 937 | arch_refresh_nodedata(nid, pgdat); |
b0dc3a34 | 938 | } else { |
e716f2eb MG |
939 | /* |
940 | * Reset the nr_zones, order and classzone_idx before reuse. | |
941 | * Note that kswapd will init kswapd_classzone_idx properly | |
942 | * when it starts in the near future. | |
943 | */ | |
b0dc3a34 | 944 | pgdat->nr_zones = 0; |
38087d9b MG |
945 | pgdat->kswapd_order = 0; |
946 | pgdat->kswapd_classzone_idx = 0; | |
a1e565aa | 947 | } |
9af3c2de YG |
948 | |
949 | /* we can use NODE_DATA(nid) from here */ | |
950 | ||
03e85f9d OS |
951 | pgdat->node_id = nid; |
952 | pgdat->node_start_pfn = start_pfn; | |
953 | ||
9af3c2de | 954 | /* init node's zones as empty zones, we don't have any present pages.*/ |
03e85f9d | 955 | free_area_init_core_hotplug(nid); |
5830169f | 956 | pgdat->per_cpu_nodestats = alloc_percpu(struct per_cpu_nodestat); |
9af3c2de | 957 | |
959ecc48 KH |
958 | /* |
959 | * The node we allocated has no zone fallback lists. For avoiding | |
960 | * to access not-initialized zonelist, build here. | |
961 | */ | |
72675e13 | 962 | build_all_zonelists(pgdat); |
959ecc48 | 963 | |
0bd85420 TC |
964 | /* |
965 | * When memory is hot-added, all the memory is in offline state. So | |
966 | * clear all zones' present_pages because they will be updated in | |
967 | * online_pages() and offline_pages(). | |
968 | */ | |
03e85f9d | 969 | reset_node_managed_pages(pgdat); |
0bd85420 TC |
970 | reset_node_present_pages(pgdat); |
971 | ||
9af3c2de YG |
972 | return pgdat; |
973 | } | |
974 | ||
b9ff0360 | 975 | static void rollback_node_hotadd(int nid) |
9af3c2de | 976 | { |
b9ff0360 OS |
977 | pg_data_t *pgdat = NODE_DATA(nid); |
978 | ||
9af3c2de | 979 | arch_refresh_nodedata(nid, NULL); |
5830169f | 980 | free_percpu(pgdat->per_cpu_nodestats); |
9af3c2de YG |
981 | arch_free_nodedata(pgdat); |
982 | return; | |
983 | } | |
984 | ||
0a547039 | 985 | |
01b0f197 TK |
986 | /** |
987 | * try_online_node - online a node if offlined | |
e8b098fc | 988 | * @nid: the node ID |
b9ff0360 OS |
989 | * @start: start addr of the node |
990 | * @set_node_online: Whether we want to online the node | |
cf23422b | 991 | * called by cpu_up() to online a node without onlined memory. |
b9ff0360 OS |
992 | * |
993 | * Returns: | |
994 | * 1 -> a new node has been allocated | |
995 | * 0 -> the node is already online | |
996 | * -ENOMEM -> the node could not be allocated | |
cf23422b | 997 | */ |
b9ff0360 | 998 | static int __try_online_node(int nid, u64 start, bool set_node_online) |
cf23422b | 999 | { |
b9ff0360 OS |
1000 | pg_data_t *pgdat; |
1001 | int ret = 1; | |
cf23422b | 1002 | |
01b0f197 TK |
1003 | if (node_online(nid)) |
1004 | return 0; | |
1005 | ||
b9ff0360 | 1006 | pgdat = hotadd_new_pgdat(nid, start); |
7553e8f2 | 1007 | if (!pgdat) { |
01b0f197 | 1008 | pr_err("Cannot online node %d due to NULL pgdat\n", nid); |
cf23422b | 1009 | ret = -ENOMEM; |
1010 | goto out; | |
1011 | } | |
b9ff0360 OS |
1012 | |
1013 | if (set_node_online) { | |
1014 | node_set_online(nid); | |
1015 | ret = register_one_node(nid); | |
1016 | BUG_ON(ret); | |
1017 | } | |
cf23422b | 1018 | out: |
b9ff0360 OS |
1019 | return ret; |
1020 | } | |
1021 | ||
1022 | /* | |
1023 | * Users of this function always want to online/register the node | |
1024 | */ | |
1025 | int try_online_node(int nid) | |
1026 | { | |
1027 | int ret; | |
1028 | ||
1029 | mem_hotplug_begin(); | |
1030 | ret = __try_online_node(nid, 0, true); | |
bfc8c901 | 1031 | mem_hotplug_done(); |
cf23422b | 1032 | return ret; |
1033 | } | |
1034 | ||
27356f54 TK |
1035 | static int check_hotplug_memory_range(u64 start, u64 size) |
1036 | { | |
ba325585 | 1037 | /* memory range must be block size aligned */ |
cec3ebd0 DH |
1038 | if (!size || !IS_ALIGNED(start, memory_block_size_bytes()) || |
1039 | !IS_ALIGNED(size, memory_block_size_bytes())) { | |
ba325585 | 1040 | pr_err("Block size [%#lx] unaligned hotplug range: start %#llx, size %#llx", |
cec3ebd0 | 1041 | memory_block_size_bytes(), start, size); |
27356f54 TK |
1042 | return -EINVAL; |
1043 | } | |
1044 | ||
1045 | return 0; | |
1046 | } | |
1047 | ||
31bc3858 VK |
1048 | static int online_memory_block(struct memory_block *mem, void *arg) |
1049 | { | |
dc18d706 | 1050 | return device_online(&mem->dev); |
31bc3858 VK |
1051 | } |
1052 | ||
8df1d0e4 DH |
1053 | /* |
1054 | * NOTE: The caller must call lock_device_hotplug() to serialize hotplug | |
1055 | * and online/offline operations (triggered e.g. by sysfs). | |
1056 | * | |
1057 | * we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG | |
1058 | */ | |
f29d8e9c | 1059 | int __ref add_memory_resource(int nid, struct resource *res) |
bc02af93 | 1060 | { |
05f800a0 | 1061 | struct mhp_restrictions restrictions = {}; |
62cedb9f | 1062 | u64 start, size; |
b9ff0360 | 1063 | bool new_node = false; |
bc02af93 YG |
1064 | int ret; |
1065 | ||
62cedb9f DV |
1066 | start = res->start; |
1067 | size = resource_size(res); | |
1068 | ||
27356f54 TK |
1069 | ret = check_hotplug_memory_range(start, size); |
1070 | if (ret) | |
1071 | return ret; | |
1072 | ||
bfc8c901 | 1073 | mem_hotplug_begin(); |
ac13c462 | 1074 | |
7f36e3e5 TC |
1075 | /* |
1076 | * Add new range to memblock so that when hotadd_new_pgdat() is called | |
1077 | * to allocate new pgdat, get_pfn_range_for_nid() will be able to find | |
1078 | * this new range and calculate total pages correctly. The range will | |
1079 | * be removed at hot-remove time. | |
1080 | */ | |
1081 | memblock_add_node(start, size, nid); | |
1082 | ||
b9ff0360 OS |
1083 | ret = __try_online_node(nid, start, false); |
1084 | if (ret < 0) | |
1085 | goto error; | |
1086 | new_node = ret; | |
9af3c2de | 1087 | |
bc02af93 | 1088 | /* call arch's memory hotadd */ |
940519f0 | 1089 | ret = arch_add_memory(nid, start, size, &restrictions); |
9af3c2de YG |
1090 | if (ret < 0) |
1091 | goto error; | |
1092 | ||
db051a0d DH |
1093 | /* create memory block devices after memory was added */ |
1094 | ret = create_memory_block_devices(start, size); | |
1095 | if (ret) { | |
1096 | arch_remove_memory(nid, start, size, NULL); | |
1097 | goto error; | |
1098 | } | |
1099 | ||
a1e565aa | 1100 | if (new_node) { |
d5b6f6a3 | 1101 | /* If sysfs file of new node can't be created, cpu on the node |
0fc44159 YG |
1102 | * can't be hot-added. There is no rollback way now. |
1103 | * So, check by BUG_ON() to catch it reluctantly.. | |
d5b6f6a3 | 1104 | * We online node here. We can't roll back from here. |
0fc44159 | 1105 | */ |
d5b6f6a3 OS |
1106 | node_set_online(nid); |
1107 | ret = __register_one_node(nid); | |
0fc44159 YG |
1108 | BUG_ON(ret); |
1109 | } | |
1110 | ||
d5b6f6a3 | 1111 | /* link memory sections under this node.*/ |
4fbce633 | 1112 | ret = link_mem_sections(nid, PFN_DOWN(start), PFN_UP(start + size - 1)); |
d5b6f6a3 OS |
1113 | BUG_ON(ret); |
1114 | ||
d96ae530 AM |
1115 | /* create new memmap entry */ |
1116 | firmware_map_add_hotplug(start, start + size, "System RAM"); | |
1117 | ||
381eab4a DH |
1118 | /* device_online() will take the lock when calling online_pages() */ |
1119 | mem_hotplug_done(); | |
1120 | ||
31bc3858 | 1121 | /* online pages if requested */ |
f29d8e9c | 1122 | if (memhp_auto_online) |
fbcf73ce | 1123 | walk_memory_blocks(start, size, NULL, online_memory_block); |
31bc3858 | 1124 | |
381eab4a | 1125 | return ret; |
9af3c2de YG |
1126 | error: |
1127 | /* rollback pgdat allocation and others */ | |
b9ff0360 OS |
1128 | if (new_node) |
1129 | rollback_node_hotadd(nid); | |
7f36e3e5 | 1130 | memblock_remove(start, size); |
bfc8c901 | 1131 | mem_hotplug_done(); |
bc02af93 YG |
1132 | return ret; |
1133 | } | |
62cedb9f | 1134 | |
8df1d0e4 DH |
1135 | /* requires device_hotplug_lock, see add_memory_resource() */ |
1136 | int __ref __add_memory(int nid, u64 start, u64 size) | |
62cedb9f DV |
1137 | { |
1138 | struct resource *res; | |
1139 | int ret; | |
1140 | ||
1141 | res = register_memory_resource(start, size); | |
6f754ba4 VK |
1142 | if (IS_ERR(res)) |
1143 | return PTR_ERR(res); | |
62cedb9f | 1144 | |
f29d8e9c | 1145 | ret = add_memory_resource(nid, res); |
62cedb9f DV |
1146 | if (ret < 0) |
1147 | release_memory_resource(res); | |
1148 | return ret; | |
1149 | } | |
8df1d0e4 DH |
1150 | |
1151 | int add_memory(int nid, u64 start, u64 size) | |
1152 | { | |
1153 | int rc; | |
1154 | ||
1155 | lock_device_hotplug(); | |
1156 | rc = __add_memory(nid, start, size); | |
1157 | unlock_device_hotplug(); | |
1158 | ||
1159 | return rc; | |
1160 | } | |
bc02af93 | 1161 | EXPORT_SYMBOL_GPL(add_memory); |
0c0e6195 KH |
1162 | |
1163 | #ifdef CONFIG_MEMORY_HOTREMOVE | |
5c755e9f BP |
1164 | /* |
1165 | * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy | |
1166 | * set and the size of the free page is given by page_order(). Using this, | |
1167 | * the function determines if the pageblock contains only free pages. | |
1168 | * Due to buddy contraints, a free page at least the size of a pageblock will | |
1169 | * be located at the start of the pageblock | |
1170 | */ | |
1171 | static inline int pageblock_free(struct page *page) | |
1172 | { | |
1173 | return PageBuddy(page) && page_order(page) >= pageblock_order; | |
1174 | } | |
1175 | ||
891cb2a7 MH |
1176 | /* Return the pfn of the start of the next active pageblock after a given pfn */ |
1177 | static unsigned long next_active_pageblock(unsigned long pfn) | |
5c755e9f | 1178 | { |
891cb2a7 MH |
1179 | struct page *page = pfn_to_page(pfn); |
1180 | ||
5c755e9f | 1181 | /* Ensure the starting page is pageblock-aligned */ |
891cb2a7 | 1182 | BUG_ON(pfn & (pageblock_nr_pages - 1)); |
5c755e9f | 1183 | |
5c755e9f | 1184 | /* If the entire pageblock is free, move to the end of free page */ |
0dcc48c1 KH |
1185 | if (pageblock_free(page)) { |
1186 | int order; | |
1187 | /* be careful. we don't have locks, page_order can be changed.*/ | |
1188 | order = page_order(page); | |
1189 | if ((order < MAX_ORDER) && (order >= pageblock_order)) | |
891cb2a7 | 1190 | return pfn + (1 << order); |
0dcc48c1 | 1191 | } |
5c755e9f | 1192 | |
891cb2a7 | 1193 | return pfn + pageblock_nr_pages; |
5c755e9f BP |
1194 | } |
1195 | ||
891cb2a7 | 1196 | static bool is_pageblock_removable_nolock(unsigned long pfn) |
fb52bbae | 1197 | { |
891cb2a7 | 1198 | struct page *page = pfn_to_page(pfn); |
fb52bbae | 1199 | struct zone *zone; |
fb52bbae MM |
1200 | |
1201 | /* | |
1202 | * We have to be careful here because we are iterating over memory | |
1203 | * sections which are not zone aware so we might end up outside of | |
1204 | * the zone but still within the section. | |
1205 | * We have to take care about the node as well. If the node is offline | |
1206 | * its NODE_DATA will be NULL - see page_zone. | |
1207 | */ | |
1208 | if (!node_online(page_to_nid(page))) | |
1209 | return false; | |
1210 | ||
1211 | zone = page_zone(page); | |
1212 | pfn = page_to_pfn(page); | |
1213 | if (!zone_spans_pfn(zone, pfn)) | |
1214 | return false; | |
1215 | ||
d381c547 | 1216 | return !has_unmovable_pages(zone, page, 0, MIGRATE_MOVABLE, SKIP_HWPOISON); |
fb52bbae MM |
1217 | } |
1218 | ||
5c755e9f | 1219 | /* Checks if this range of memory is likely to be hot-removable. */ |
c98940f6 | 1220 | bool is_mem_section_removable(unsigned long start_pfn, unsigned long nr_pages) |
5c755e9f | 1221 | { |
891cb2a7 MH |
1222 | unsigned long end_pfn, pfn; |
1223 | ||
1224 | end_pfn = min(start_pfn + nr_pages, | |
1225 | zone_end_pfn(page_zone(pfn_to_page(start_pfn)))); | |
5c755e9f BP |
1226 | |
1227 | /* Check the starting page of each pageblock within the range */ | |
891cb2a7 MH |
1228 | for (pfn = start_pfn; pfn < end_pfn; pfn = next_active_pageblock(pfn)) { |
1229 | if (!is_pageblock_removable_nolock(pfn)) | |
c98940f6 | 1230 | return false; |
49ac8255 | 1231 | cond_resched(); |
5c755e9f BP |
1232 | } |
1233 | ||
1234 | /* All pageblocks in the memory block are likely to be hot-removable */ | |
c98940f6 | 1235 | return true; |
5c755e9f BP |
1236 | } |
1237 | ||
0c0e6195 | 1238 | /* |
deb88a2a | 1239 | * Confirm all pages in a range [start, end) belong to the same zone. |
a96dfddb | 1240 | * When true, return its valid [start, end). |
0c0e6195 | 1241 | */ |
a96dfddb TK |
1242 | int test_pages_in_a_zone(unsigned long start_pfn, unsigned long end_pfn, |
1243 | unsigned long *valid_start, unsigned long *valid_end) | |
0c0e6195 | 1244 | { |
5f0f2887 | 1245 | unsigned long pfn, sec_end_pfn; |
a96dfddb | 1246 | unsigned long start, end; |
0c0e6195 KH |
1247 | struct zone *zone = NULL; |
1248 | struct page *page; | |
1249 | int i; | |
deb88a2a | 1250 | for (pfn = start_pfn, sec_end_pfn = SECTION_ALIGN_UP(start_pfn + 1); |
0c0e6195 | 1251 | pfn < end_pfn; |
deb88a2a | 1252 | pfn = sec_end_pfn, sec_end_pfn += PAGES_PER_SECTION) { |
5f0f2887 AB |
1253 | /* Make sure the memory section is present first */ |
1254 | if (!present_section_nr(pfn_to_section_nr(pfn))) | |
0c0e6195 | 1255 | continue; |
5f0f2887 AB |
1256 | for (; pfn < sec_end_pfn && pfn < end_pfn; |
1257 | pfn += MAX_ORDER_NR_PAGES) { | |
1258 | i = 0; | |
1259 | /* This is just a CONFIG_HOLES_IN_ZONE check.*/ | |
1260 | while ((i < MAX_ORDER_NR_PAGES) && | |
1261 | !pfn_valid_within(pfn + i)) | |
1262 | i++; | |
d6d8c8a4 | 1263 | if (i == MAX_ORDER_NR_PAGES || pfn + i >= end_pfn) |
5f0f2887 | 1264 | continue; |
24feb47c MZ |
1265 | /* Check if we got outside of the zone */ |
1266 | if (zone && !zone_spans_pfn(zone, pfn + i)) | |
1267 | return 0; | |
5f0f2887 AB |
1268 | page = pfn_to_page(pfn + i); |
1269 | if (zone && page_zone(page) != zone) | |
1270 | return 0; | |
a96dfddb TK |
1271 | if (!zone) |
1272 | start = pfn + i; | |
5f0f2887 | 1273 | zone = page_zone(page); |
a96dfddb | 1274 | end = pfn + MAX_ORDER_NR_PAGES; |
5f0f2887 | 1275 | } |
0c0e6195 | 1276 | } |
deb88a2a | 1277 | |
a96dfddb TK |
1278 | if (zone) { |
1279 | *valid_start = start; | |
d6d8c8a4 | 1280 | *valid_end = min(end, end_pfn); |
deb88a2a | 1281 | return 1; |
a96dfddb | 1282 | } else { |
deb88a2a | 1283 | return 0; |
a96dfddb | 1284 | } |
0c0e6195 KH |
1285 | } |
1286 | ||
1287 | /* | |
0efadf48 YX |
1288 | * Scan pfn range [start,end) to find movable/migratable pages (LRU pages, |
1289 | * non-lru movable pages and hugepages). We scan pfn because it's much | |
1290 | * easier than scanning over linked list. This function returns the pfn | |
1291 | * of the first found movable page if it's found, otherwise 0. | |
0c0e6195 | 1292 | */ |
c8721bbb | 1293 | static unsigned long scan_movable_pages(unsigned long start, unsigned long end) |
0c0e6195 KH |
1294 | { |
1295 | unsigned long pfn; | |
eeb0efd0 | 1296 | |
0c0e6195 | 1297 | for (pfn = start; pfn < end; pfn++) { |
eeb0efd0 OS |
1298 | struct page *page, *head; |
1299 | unsigned long skip; | |
1300 | ||
1301 | if (!pfn_valid(pfn)) | |
1302 | continue; | |
1303 | page = pfn_to_page(pfn); | |
1304 | if (PageLRU(page)) | |
1305 | return pfn; | |
1306 | if (__PageMovable(page)) | |
1307 | return pfn; | |
1308 | ||
1309 | if (!PageHuge(page)) | |
1310 | continue; | |
1311 | head = compound_head(page); | |
39186cbe | 1312 | if (page_huge_active(head)) |
eeb0efd0 OS |
1313 | return pfn; |
1314 | skip = (1 << compound_order(head)) - (page - head); | |
1315 | pfn += skip - 1; | |
0c0e6195 KH |
1316 | } |
1317 | return 0; | |
1318 | } | |
1319 | ||
666feb21 | 1320 | static struct page *new_node_page(struct page *page, unsigned long private) |
394e31d2 | 1321 | { |
394e31d2 | 1322 | int nid = page_to_nid(page); |
231e97e2 | 1323 | nodemask_t nmask = node_states[N_MEMORY]; |
7f252f27 MH |
1324 | |
1325 | /* | |
1326 | * try to allocate from a different node but reuse this node if there | |
1327 | * are no other online nodes to be used (e.g. we are offlining a part | |
1328 | * of the only existing node) | |
1329 | */ | |
1330 | node_clear(nid, nmask); | |
1331 | if (nodes_empty(nmask)) | |
1332 | node_set(nid, nmask); | |
394e31d2 | 1333 | |
8b913238 | 1334 | return new_page_nodemask(page, nid, &nmask); |
394e31d2 XQ |
1335 | } |
1336 | ||
0c0e6195 KH |
1337 | static int |
1338 | do_migrate_range(unsigned long start_pfn, unsigned long end_pfn) | |
1339 | { | |
1340 | unsigned long pfn; | |
1341 | struct page *page; | |
0c0e6195 KH |
1342 | int ret = 0; |
1343 | LIST_HEAD(source); | |
1344 | ||
a85009c3 | 1345 | for (pfn = start_pfn; pfn < end_pfn; pfn++) { |
0c0e6195 KH |
1346 | if (!pfn_valid(pfn)) |
1347 | continue; | |
1348 | page = pfn_to_page(pfn); | |
c8721bbb NH |
1349 | |
1350 | if (PageHuge(page)) { | |
1351 | struct page *head = compound_head(page); | |
daf3538a OS |
1352 | pfn = page_to_pfn(head) + (1<<compound_order(head)) - 1; |
1353 | isolate_huge_page(head, &source); | |
c8721bbb | 1354 | continue; |
94723aaf | 1355 | } else if (PageTransHuge(page)) |
8135d892 NH |
1356 | pfn = page_to_pfn(compound_head(page)) |
1357 | + hpage_nr_pages(page) - 1; | |
c8721bbb | 1358 | |
b15c8726 MH |
1359 | /* |
1360 | * HWPoison pages have elevated reference counts so the migration would | |
1361 | * fail on them. It also doesn't make any sense to migrate them in the | |
1362 | * first place. Still try to unmap such a page in case it is still mapped | |
1363 | * (e.g. current hwpoison implementation doesn't unmap KSM pages but keep | |
1364 | * the unmap as the catch all safety net). | |
1365 | */ | |
1366 | if (PageHWPoison(page)) { | |
1367 | if (WARN_ON(PageLRU(page))) | |
1368 | isolate_lru_page(page); | |
1369 | if (page_mapped(page)) | |
1370 | try_to_unmap(page, TTU_IGNORE_MLOCK | TTU_IGNORE_ACCESS); | |
1371 | continue; | |
1372 | } | |
1373 | ||
700c2a46 | 1374 | if (!get_page_unless_zero(page)) |
0c0e6195 KH |
1375 | continue; |
1376 | /* | |
0efadf48 YX |
1377 | * We can skip free pages. And we can deal with pages on |
1378 | * LRU and non-lru movable pages. | |
0c0e6195 | 1379 | */ |
0efadf48 YX |
1380 | if (PageLRU(page)) |
1381 | ret = isolate_lru_page(page); | |
1382 | else | |
1383 | ret = isolate_movable_page(page, ISOLATE_UNEVICTABLE); | |
0c0e6195 | 1384 | if (!ret) { /* Success */ |
62695a84 | 1385 | list_add_tail(&page->lru, &source); |
0efadf48 YX |
1386 | if (!__PageMovable(page)) |
1387 | inc_node_page_state(page, NR_ISOLATED_ANON + | |
1388 | page_is_file_cache(page)); | |
6d9c285a | 1389 | |
0c0e6195 | 1390 | } else { |
2932c8b0 | 1391 | pr_warn("failed to isolate pfn %lx\n", pfn); |
0efadf48 | 1392 | dump_page(page, "isolation failed"); |
0c0e6195 | 1393 | } |
1723058e | 1394 | put_page(page); |
0c0e6195 | 1395 | } |
f3ab2636 | 1396 | if (!list_empty(&source)) { |
394e31d2 XQ |
1397 | /* Allocate a new page from the nearest neighbor node */ |
1398 | ret = migrate_pages(&source, new_node_page, NULL, 0, | |
9c620e2b | 1399 | MIGRATE_SYNC, MR_MEMORY_HOTPLUG); |
2932c8b0 MH |
1400 | if (ret) { |
1401 | list_for_each_entry(page, &source, lru) { | |
1402 | pr_warn("migrating pfn %lx failed ret:%d ", | |
1403 | page_to_pfn(page), ret); | |
1404 | dump_page(page, "migration failure"); | |
1405 | } | |
c8721bbb | 1406 | putback_movable_pages(&source); |
2932c8b0 | 1407 | } |
0c0e6195 | 1408 | } |
1723058e | 1409 | |
0c0e6195 KH |
1410 | return ret; |
1411 | } | |
1412 | ||
1413 | /* | |
1414 | * remove from free_area[] and mark all as Reserved. | |
1415 | */ | |
1416 | static int | |
1417 | offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages, | |
1418 | void *data) | |
1419 | { | |
5557c766 | 1420 | unsigned long *offlined_pages = (unsigned long *)data; |
0c0e6195 | 1421 | |
5557c766 MH |
1422 | *offlined_pages += __offline_isolated_pages(start, start + nr_pages); |
1423 | return 0; | |
0c0e6195 KH |
1424 | } |
1425 | ||
1426 | /* | |
1427 | * Check all pages in range, recoreded as memory resource, are isolated. | |
1428 | */ | |
1429 | static int | |
1430 | check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages, | |
1431 | void *data) | |
1432 | { | |
5557c766 | 1433 | return test_pages_isolated(start_pfn, start_pfn + nr_pages, true); |
0c0e6195 KH |
1434 | } |
1435 | ||
c5320926 TC |
1436 | static int __init cmdline_parse_movable_node(char *p) |
1437 | { | |
4932381e | 1438 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
55ac590c | 1439 | movable_node_enabled = true; |
4932381e MH |
1440 | #else |
1441 | pr_warn("movable_node parameter depends on CONFIG_HAVE_MEMBLOCK_NODE_MAP to work properly\n"); | |
1442 | #endif | |
c5320926 TC |
1443 | return 0; |
1444 | } | |
1445 | early_param("movable_node", cmdline_parse_movable_node); | |
1446 | ||
d9713679 LJ |
1447 | /* check which state of node_states will be changed when offline memory */ |
1448 | static void node_states_check_changes_offline(unsigned long nr_pages, | |
1449 | struct zone *zone, struct memory_notify *arg) | |
1450 | { | |
1451 | struct pglist_data *pgdat = zone->zone_pgdat; | |
1452 | unsigned long present_pages = 0; | |
86b27bea | 1453 | enum zone_type zt; |
d9713679 | 1454 | |
98fa15f3 AK |
1455 | arg->status_change_nid = NUMA_NO_NODE; |
1456 | arg->status_change_nid_normal = NUMA_NO_NODE; | |
1457 | arg->status_change_nid_high = NUMA_NO_NODE; | |
d9713679 LJ |
1458 | |
1459 | /* | |
86b27bea OS |
1460 | * Check whether node_states[N_NORMAL_MEMORY] will be changed. |
1461 | * If the memory to be offline is within the range | |
1462 | * [0..ZONE_NORMAL], and it is the last present memory there, | |
1463 | * the zones in that range will become empty after the offlining, | |
1464 | * thus we can determine that we need to clear the node from | |
1465 | * node_states[N_NORMAL_MEMORY]. | |
d9713679 | 1466 | */ |
86b27bea | 1467 | for (zt = 0; zt <= ZONE_NORMAL; zt++) |
d9713679 | 1468 | present_pages += pgdat->node_zones[zt].present_pages; |
86b27bea | 1469 | if (zone_idx(zone) <= ZONE_NORMAL && nr_pages >= present_pages) |
d9713679 | 1470 | arg->status_change_nid_normal = zone_to_nid(zone); |
d9713679 | 1471 | |
6715ddf9 LJ |
1472 | #ifdef CONFIG_HIGHMEM |
1473 | /* | |
86b27bea OS |
1474 | * node_states[N_HIGH_MEMORY] contains nodes which |
1475 | * have normal memory or high memory. | |
1476 | * Here we add the present_pages belonging to ZONE_HIGHMEM. | |
1477 | * If the zone is within the range of [0..ZONE_HIGHMEM), and | |
1478 | * we determine that the zones in that range become empty, | |
1479 | * we need to clear the node for N_HIGH_MEMORY. | |
6715ddf9 | 1480 | */ |
86b27bea OS |
1481 | present_pages += pgdat->node_zones[ZONE_HIGHMEM].present_pages; |
1482 | if (zone_idx(zone) <= ZONE_HIGHMEM && nr_pages >= present_pages) | |
6715ddf9 | 1483 | arg->status_change_nid_high = zone_to_nid(zone); |
6715ddf9 LJ |
1484 | #endif |
1485 | ||
d9713679 | 1486 | /* |
86b27bea OS |
1487 | * We have accounted the pages from [0..ZONE_NORMAL), and |
1488 | * in case of CONFIG_HIGHMEM the pages from ZONE_HIGHMEM | |
1489 | * as well. | |
1490 | * Here we count the possible pages from ZONE_MOVABLE. | |
1491 | * If after having accounted all the pages, we see that the nr_pages | |
1492 | * to be offlined is over or equal to the accounted pages, | |
1493 | * we know that the node will become empty, and so, we can clear | |
1494 | * it for N_MEMORY as well. | |
d9713679 | 1495 | */ |
86b27bea | 1496 | present_pages += pgdat->node_zones[ZONE_MOVABLE].present_pages; |
d9713679 | 1497 | |
d9713679 LJ |
1498 | if (nr_pages >= present_pages) |
1499 | arg->status_change_nid = zone_to_nid(zone); | |
d9713679 LJ |
1500 | } |
1501 | ||
1502 | static void node_states_clear_node(int node, struct memory_notify *arg) | |
1503 | { | |
1504 | if (arg->status_change_nid_normal >= 0) | |
1505 | node_clear_state(node, N_NORMAL_MEMORY); | |
1506 | ||
cf01f6f5 | 1507 | if (arg->status_change_nid_high >= 0) |
d9713679 | 1508 | node_clear_state(node, N_HIGH_MEMORY); |
6715ddf9 | 1509 | |
cf01f6f5 | 1510 | if (arg->status_change_nid >= 0) |
6715ddf9 | 1511 | node_clear_state(node, N_MEMORY); |
d9713679 LJ |
1512 | } |
1513 | ||
a16cee10 | 1514 | static int __ref __offline_pages(unsigned long start_pfn, |
ecde0f3e | 1515 | unsigned long end_pfn) |
0c0e6195 | 1516 | { |
ecde0f3e | 1517 | unsigned long pfn, nr_pages; |
5557c766 | 1518 | unsigned long offlined_pages = 0; |
9b7ea46a | 1519 | int ret, node, nr_isolate_pageblock; |
d702909f | 1520 | unsigned long flags; |
a96dfddb | 1521 | unsigned long valid_start, valid_end; |
0c0e6195 | 1522 | struct zone *zone; |
7b78d335 | 1523 | struct memory_notify arg; |
79605093 | 1524 | char *reason; |
0c0e6195 | 1525 | |
381eab4a DH |
1526 | mem_hotplug_begin(); |
1527 | ||
0c0e6195 KH |
1528 | /* This makes hotplug much easier...and readable. |
1529 | we assume this for now. .*/ | |
381eab4a DH |
1530 | if (!test_pages_in_a_zone(start_pfn, end_pfn, &valid_start, |
1531 | &valid_end)) { | |
79605093 MH |
1532 | ret = -EINVAL; |
1533 | reason = "multizone range"; | |
1534 | goto failed_removal; | |
381eab4a | 1535 | } |
7b78d335 | 1536 | |
a96dfddb | 1537 | zone = page_zone(pfn_to_page(valid_start)); |
7b78d335 YG |
1538 | node = zone_to_nid(zone); |
1539 | nr_pages = end_pfn - start_pfn; | |
1540 | ||
0c0e6195 | 1541 | /* set above range as isolated */ |
b023f468 | 1542 | ret = start_isolate_page_range(start_pfn, end_pfn, |
d381c547 MH |
1543 | MIGRATE_MOVABLE, |
1544 | SKIP_HWPOISON | REPORT_FAILURE); | |
9b7ea46a | 1545 | if (ret < 0) { |
79605093 MH |
1546 | reason = "failure to isolate range"; |
1547 | goto failed_removal; | |
381eab4a | 1548 | } |
9b7ea46a | 1549 | nr_isolate_pageblock = ret; |
7b78d335 YG |
1550 | |
1551 | arg.start_pfn = start_pfn; | |
1552 | arg.nr_pages = nr_pages; | |
d9713679 | 1553 | node_states_check_changes_offline(nr_pages, zone, &arg); |
7b78d335 YG |
1554 | |
1555 | ret = memory_notify(MEM_GOING_OFFLINE, &arg); | |
1556 | ret = notifier_to_errno(ret); | |
79605093 MH |
1557 | if (ret) { |
1558 | reason = "notifier failure"; | |
1559 | goto failed_removal_isolated; | |
1560 | } | |
7b78d335 | 1561 | |
bb8965bd MH |
1562 | do { |
1563 | for (pfn = start_pfn; pfn;) { | |
1564 | if (signal_pending(current)) { | |
1565 | ret = -EINTR; | |
1566 | reason = "signal backoff"; | |
1567 | goto failed_removal_isolated; | |
1568 | } | |
72b39cfc | 1569 | |
bb8965bd MH |
1570 | cond_resched(); |
1571 | lru_add_drain_all(); | |
bb8965bd MH |
1572 | |
1573 | pfn = scan_movable_pages(pfn, end_pfn); | |
1574 | if (pfn) { | |
1575 | /* | |
1576 | * TODO: fatal migration failures should bail | |
1577 | * out | |
1578 | */ | |
1579 | do_migrate_range(pfn, end_pfn); | |
1580 | } | |
1581 | } | |
0c0e6195 | 1582 | |
bb8965bd MH |
1583 | /* |
1584 | * Dissolve free hugepages in the memory block before doing | |
1585 | * offlining actually in order to make hugetlbfs's object | |
1586 | * counting consistent. | |
1587 | */ | |
1588 | ret = dissolve_free_huge_pages(start_pfn, end_pfn); | |
1589 | if (ret) { | |
1590 | reason = "failure to dissolve huge pages"; | |
1591 | goto failed_removal_isolated; | |
1592 | } | |
1593 | /* check again */ | |
5557c766 MH |
1594 | ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn, |
1595 | NULL, check_pages_isolated_cb); | |
1596 | } while (ret); | |
72b39cfc | 1597 | |
b3834be5 | 1598 | /* Ok, all of our target is isolated. |
0c0e6195 | 1599 | We cannot do rollback at this point. */ |
5557c766 MH |
1600 | walk_system_ram_range(start_pfn, end_pfn - start_pfn, |
1601 | &offlined_pages, offline_isolated_pages_cb); | |
1602 | pr_info("Offlined Pages %ld\n", offlined_pages); | |
9b7ea46a QC |
1603 | /* |
1604 | * Onlining will reset pagetype flags and makes migrate type | |
1605 | * MOVABLE, so just need to decrease the number of isolated | |
1606 | * pageblocks zone counter here. | |
1607 | */ | |
1608 | spin_lock_irqsave(&zone->lock, flags); | |
1609 | zone->nr_isolate_pageblock -= nr_isolate_pageblock; | |
1610 | spin_unlock_irqrestore(&zone->lock, flags); | |
1611 | ||
0c0e6195 | 1612 | /* removal success */ |
3dcc0571 | 1613 | adjust_managed_page_count(pfn_to_page(start_pfn), -offlined_pages); |
0c0e6195 | 1614 | zone->present_pages -= offlined_pages; |
d702909f CS |
1615 | |
1616 | pgdat_resize_lock(zone->zone_pgdat, &flags); | |
0c0e6195 | 1617 | zone->zone_pgdat->node_present_pages -= offlined_pages; |
d702909f | 1618 | pgdat_resize_unlock(zone->zone_pgdat, &flags); |
7b78d335 | 1619 | |
1b79acc9 KM |
1620 | init_per_zone_wmark_min(); |
1621 | ||
1e8537ba | 1622 | if (!populated_zone(zone)) { |
340175b7 | 1623 | zone_pcp_reset(zone); |
72675e13 | 1624 | build_all_zonelists(NULL); |
1e8537ba XQ |
1625 | } else |
1626 | zone_pcp_update(zone); | |
340175b7 | 1627 | |
d9713679 | 1628 | node_states_clear_node(node, &arg); |
698b1b30 | 1629 | if (arg.status_change_nid >= 0) { |
8fe23e05 | 1630 | kswapd_stop(node); |
698b1b30 VB |
1631 | kcompactd_stop(node); |
1632 | } | |
bce7394a | 1633 | |
0c0e6195 KH |
1634 | vm_total_pages = nr_free_pagecache_pages(); |
1635 | writeback_set_ratelimit(); | |
7b78d335 YG |
1636 | |
1637 | memory_notify(MEM_OFFLINE, &arg); | |
381eab4a | 1638 | mem_hotplug_done(); |
0c0e6195 KH |
1639 | return 0; |
1640 | ||
79605093 MH |
1641 | failed_removal_isolated: |
1642 | undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE); | |
c4efe484 | 1643 | memory_notify(MEM_CANCEL_OFFLINE, &arg); |
0c0e6195 | 1644 | failed_removal: |
79605093 | 1645 | pr_debug("memory offlining [mem %#010llx-%#010llx] failed due to %s\n", |
e33e33b4 | 1646 | (unsigned long long) start_pfn << PAGE_SHIFT, |
79605093 MH |
1647 | ((unsigned long long) end_pfn << PAGE_SHIFT) - 1, |
1648 | reason); | |
0c0e6195 | 1649 | /* pushback to free area */ |
381eab4a | 1650 | mem_hotplug_done(); |
0c0e6195 KH |
1651 | return ret; |
1652 | } | |
71088785 | 1653 | |
a16cee10 WC |
1654 | int offline_pages(unsigned long start_pfn, unsigned long nr_pages) |
1655 | { | |
ecde0f3e | 1656 | return __offline_pages(start_pfn, start_pfn + nr_pages); |
a16cee10 WC |
1657 | } |
1658 | ||
d6de9d53 | 1659 | static int check_memblock_offlined_cb(struct memory_block *mem, void *arg) |
bbc76be6 WC |
1660 | { |
1661 | int ret = !is_memblock_offlined(mem); | |
1662 | ||
349daa0f RD |
1663 | if (unlikely(ret)) { |
1664 | phys_addr_t beginpa, endpa; | |
1665 | ||
1666 | beginpa = PFN_PHYS(section_nr_to_pfn(mem->start_section_nr)); | |
1667 | endpa = PFN_PHYS(section_nr_to_pfn(mem->end_section_nr + 1))-1; | |
756a025f | 1668 | pr_warn("removing memory fails, because memory [%pa-%pa] is onlined\n", |
349daa0f | 1669 | &beginpa, &endpa); |
bbc76be6 | 1670 | |
eca499ab PT |
1671 | return -EBUSY; |
1672 | } | |
1673 | return 0; | |
bbc76be6 WC |
1674 | } |
1675 | ||
0f1cfe9d | 1676 | static int check_cpu_on_node(pg_data_t *pgdat) |
60a5a19e | 1677 | { |
60a5a19e TC |
1678 | int cpu; |
1679 | ||
1680 | for_each_present_cpu(cpu) { | |
1681 | if (cpu_to_node(cpu) == pgdat->node_id) | |
1682 | /* | |
1683 | * the cpu on this node isn't removed, and we can't | |
1684 | * offline this node. | |
1685 | */ | |
1686 | return -EBUSY; | |
1687 | } | |
1688 | ||
1689 | return 0; | |
1690 | } | |
1691 | ||
0f1cfe9d TK |
1692 | /** |
1693 | * try_offline_node | |
e8b098fc | 1694 | * @nid: the node ID |
0f1cfe9d TK |
1695 | * |
1696 | * Offline a node if all memory sections and cpus of the node are removed. | |
1697 | * | |
1698 | * NOTE: The caller must call lock_device_hotplug() to serialize hotplug | |
1699 | * and online/offline operations before this call. | |
1700 | */ | |
90b30cdc | 1701 | void try_offline_node(int nid) |
60a5a19e | 1702 | { |
d822b86a WC |
1703 | pg_data_t *pgdat = NODE_DATA(nid); |
1704 | unsigned long start_pfn = pgdat->node_start_pfn; | |
1705 | unsigned long end_pfn = start_pfn + pgdat->node_spanned_pages; | |
60a5a19e TC |
1706 | unsigned long pfn; |
1707 | ||
1708 | for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) { | |
1709 | unsigned long section_nr = pfn_to_section_nr(pfn); | |
1710 | ||
1711 | if (!present_section_nr(section_nr)) | |
1712 | continue; | |
1713 | ||
1714 | if (pfn_to_nid(pfn) != nid) | |
1715 | continue; | |
1716 | ||
1717 | /* | |
1718 | * some memory sections of this node are not removed, and we | |
1719 | * can't offline node now. | |
1720 | */ | |
1721 | return; | |
1722 | } | |
1723 | ||
46a3679b | 1724 | if (check_cpu_on_node(pgdat)) |
60a5a19e TC |
1725 | return; |
1726 | ||
1727 | /* | |
1728 | * all memory/cpu of this node are removed, we can offline this | |
1729 | * node now. | |
1730 | */ | |
1731 | node_set_offline(nid); | |
1732 | unregister_one_node(nid); | |
1733 | } | |
90b30cdc | 1734 | EXPORT_SYMBOL(try_offline_node); |
60a5a19e | 1735 | |
d9eb1417 DH |
1736 | static void __release_memory_resource(resource_size_t start, |
1737 | resource_size_t size) | |
1738 | { | |
1739 | int ret; | |
1740 | ||
1741 | /* | |
1742 | * When removing memory in the same granularity as it was added, | |
1743 | * this function never fails. It might only fail if resources | |
1744 | * have to be adjusted or split. We'll ignore the error, as | |
1745 | * removing of memory cannot fail. | |
1746 | */ | |
1747 | ret = release_mem_region_adjustable(&iomem_resource, start, size); | |
1748 | if (ret) { | |
1749 | resource_size_t endres = start + size - 1; | |
1750 | ||
1751 | pr_warn("Unable to release resource <%pa-%pa> (%d)\n", | |
1752 | &start, &endres, ret); | |
1753 | } | |
1754 | } | |
1755 | ||
eca499ab | 1756 | static int __ref try_remove_memory(int nid, u64 start, u64 size) |
bbc76be6 | 1757 | { |
eca499ab | 1758 | int rc = 0; |
993c1aad | 1759 | |
27356f54 TK |
1760 | BUG_ON(check_hotplug_memory_range(start, size)); |
1761 | ||
bfc8c901 | 1762 | mem_hotplug_begin(); |
6677e3ea YI |
1763 | |
1764 | /* | |
242831eb | 1765 | * All memory blocks must be offlined before removing memory. Check |
eca499ab | 1766 | * whether all memory blocks in question are offline and return error |
242831eb | 1767 | * if this is not the case. |
6677e3ea | 1768 | */ |
fbcf73ce | 1769 | rc = walk_memory_blocks(start, size, NULL, check_memblock_offlined_cb); |
eca499ab PT |
1770 | if (rc) |
1771 | goto done; | |
6677e3ea | 1772 | |
46c66c4b YI |
1773 | /* remove memmap entry */ |
1774 | firmware_map_remove(start, start + size, "System RAM"); | |
f9126ab9 XQ |
1775 | memblock_free(start, size); |
1776 | memblock_remove(start, size); | |
4c4b7f9b DH |
1777 | |
1778 | /* remove memory block devices before removing memory */ | |
1779 | remove_memory_block_devices(start, size); | |
46c66c4b | 1780 | |
2c2a5af6 | 1781 | arch_remove_memory(nid, start, size, NULL); |
d9eb1417 | 1782 | __release_memory_resource(start, size); |
24d335ca | 1783 | |
60a5a19e TC |
1784 | try_offline_node(nid); |
1785 | ||
eca499ab | 1786 | done: |
bfc8c901 | 1787 | mem_hotplug_done(); |
eca499ab | 1788 | return rc; |
71088785 | 1789 | } |
d15e5926 | 1790 | |
eca499ab PT |
1791 | /** |
1792 | * remove_memory | |
1793 | * @nid: the node ID | |
1794 | * @start: physical address of the region to remove | |
1795 | * @size: size of the region to remove | |
1796 | * | |
1797 | * NOTE: The caller must call lock_device_hotplug() to serialize hotplug | |
1798 | * and online/offline operations before this call, as required by | |
1799 | * try_offline_node(). | |
1800 | */ | |
1801 | void __remove_memory(int nid, u64 start, u64 size) | |
1802 | { | |
1803 | ||
1804 | /* | |
1805 | * trigger BUG() is some memory is not offlined prior to calling this | |
1806 | * function | |
1807 | */ | |
1808 | if (try_remove_memory(nid, start, size)) | |
1809 | BUG(); | |
1810 | } | |
1811 | ||
1812 | /* | |
1813 | * Remove memory if every memory block is offline, otherwise return -EBUSY is | |
1814 | * some memory is not offline | |
1815 | */ | |
1816 | int remove_memory(int nid, u64 start, u64 size) | |
d15e5926 | 1817 | { |
eca499ab PT |
1818 | int rc; |
1819 | ||
d15e5926 | 1820 | lock_device_hotplug(); |
eca499ab | 1821 | rc = try_remove_memory(nid, start, size); |
d15e5926 | 1822 | unlock_device_hotplug(); |
eca499ab PT |
1823 | |
1824 | return rc; | |
d15e5926 | 1825 | } |
71088785 | 1826 | EXPORT_SYMBOL_GPL(remove_memory); |
aba6efc4 | 1827 | #endif /* CONFIG_MEMORY_HOTREMOVE */ |