]>
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 | ||
9d0ad8ca | 52 | static online_page_callback_t online_page_callback = generic_online_page; |
bfc8c901 | 53 | static DEFINE_MUTEX(online_page_callback_lock); |
9d0ad8ca | 54 | |
3f906ba2 | 55 | DEFINE_STATIC_PERCPU_RWSEM(mem_hotplug_lock); |
bfc8c901 | 56 | |
accfb3e0 SF |
57 | static int default_kernel_zone = ZONE_NORMAL; |
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 |
862919e5 | 72 | int memhp_default_online_type = MMOP_OFFLINE; |
8604d9e5 | 73 | #else |
862919e5 | 74 | int memhp_default_online_type = MMOP_ONLINE; |
8604d9e5 | 75 | #endif |
31bc3858 | 76 | |
86dd995d VK |
77 | static int __init setup_memhp_default_state(char *str) |
78 | { | |
5f47adf7 DH |
79 | const int online_type = memhp_online_type_from_str(str); |
80 | ||
81 | if (online_type >= 0) | |
82 | memhp_default_online_type = online_type; | |
86dd995d VK |
83 | |
84 | return 1; | |
85 | } | |
86 | __setup("memhp_default_state=", setup_memhp_default_state); | |
87 | ||
30467e0b | 88 | void mem_hotplug_begin(void) |
20d6c96b | 89 | { |
3f906ba2 TG |
90 | cpus_read_lock(); |
91 | percpu_down_write(&mem_hotplug_lock); | |
20d6c96b KM |
92 | } |
93 | ||
30467e0b | 94 | void mem_hotplug_done(void) |
bfc8c901 | 95 | { |
3f906ba2 TG |
96 | percpu_up_write(&mem_hotplug_lock); |
97 | cpus_read_unlock(); | |
bfc8c901 | 98 | } |
20d6c96b | 99 | |
357b4da5 JG |
100 | u64 max_mem_size = U64_MAX; |
101 | ||
45e0b78b | 102 | /* add this memory to iomem resource */ |
7b7b2721 DH |
103 | static struct resource *register_memory_resource(u64 start, u64 size, |
104 | const char *resource_name) | |
45e0b78b | 105 | { |
2794129e DH |
106 | struct resource *res; |
107 | unsigned long flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY; | |
7b7b2721 DH |
108 | |
109 | if (strcmp(resource_name, "System RAM")) | |
7cf603d1 | 110 | flags |= IORESOURCE_SYSRAM_DRIVER_MANAGED; |
357b4da5 | 111 | |
f3cd4c86 BH |
112 | /* |
113 | * Make sure value parsed from 'mem=' only restricts memory adding | |
114 | * while booting, so that memory hotplug won't be impacted. Please | |
115 | * refer to document of 'mem=' in kernel-parameters.txt for more | |
116 | * details. | |
117 | */ | |
118 | if (start + size > max_mem_size && system_state < SYSTEM_RUNNING) | |
357b4da5 JG |
119 | return ERR_PTR(-E2BIG); |
120 | ||
2794129e DH |
121 | /* |
122 | * Request ownership of the new memory range. This might be | |
123 | * a child of an existing resource that was present but | |
124 | * not marked as busy. | |
125 | */ | |
126 | res = __request_region(&iomem_resource, start, size, | |
127 | resource_name, flags); | |
128 | ||
129 | if (!res) { | |
130 | pr_debug("Unable to reserve System RAM region: %016llx->%016llx\n", | |
131 | start, start + size); | |
6f754ba4 | 132 | return ERR_PTR(-EEXIST); |
45e0b78b KM |
133 | } |
134 | return res; | |
135 | } | |
136 | ||
137 | static void release_memory_resource(struct resource *res) | |
138 | { | |
139 | if (!res) | |
140 | return; | |
141 | release_resource(res); | |
142 | kfree(res); | |
45e0b78b KM |
143 | } |
144 | ||
53947027 | 145 | #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE |
46723bfa YI |
146 | void get_page_bootmem(unsigned long info, struct page *page, |
147 | unsigned long type) | |
04753278 | 148 | { |
ddffe98d | 149 | page->freelist = (void *)type; |
04753278 YG |
150 | SetPagePrivate(page); |
151 | set_page_private(page, info); | |
fe896d18 | 152 | page_ref_inc(page); |
04753278 YG |
153 | } |
154 | ||
170a5a7e | 155 | void put_page_bootmem(struct page *page) |
04753278 | 156 | { |
5f24ce5f | 157 | unsigned long type; |
04753278 | 158 | |
ddffe98d | 159 | type = (unsigned long) page->freelist; |
5f24ce5f AA |
160 | BUG_ON(type < MEMORY_HOTPLUG_MIN_BOOTMEM_TYPE || |
161 | type > MEMORY_HOTPLUG_MAX_BOOTMEM_TYPE); | |
04753278 | 162 | |
fe896d18 | 163 | if (page_ref_dec_return(page) == 1) { |
ddffe98d | 164 | page->freelist = NULL; |
04753278 YG |
165 | ClearPagePrivate(page); |
166 | set_page_private(page, 0); | |
5f24ce5f | 167 | INIT_LIST_HEAD(&page->lru); |
170a5a7e | 168 | free_reserved_page(page); |
04753278 | 169 | } |
04753278 YG |
170 | } |
171 | ||
46723bfa YI |
172 | #ifdef CONFIG_HAVE_BOOTMEM_INFO_NODE |
173 | #ifndef CONFIG_SPARSEMEM_VMEMMAP | |
d92bc318 | 174 | static void register_page_bootmem_info_section(unsigned long start_pfn) |
04753278 | 175 | { |
f1eca35a | 176 | unsigned long mapsize, section_nr, i; |
04753278 YG |
177 | struct mem_section *ms; |
178 | struct page *page, *memmap; | |
f1eca35a | 179 | struct mem_section_usage *usage; |
04753278 | 180 | |
04753278 YG |
181 | section_nr = pfn_to_section_nr(start_pfn); |
182 | ms = __nr_to_section(section_nr); | |
183 | ||
184 | /* Get section's memmap address */ | |
185 | memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr); | |
186 | ||
187 | /* | |
188 | * Get page for the memmap's phys address | |
189 | * XXX: need more consideration for sparse_vmemmap... | |
190 | */ | |
191 | page = virt_to_page(memmap); | |
192 | mapsize = sizeof(struct page) * PAGES_PER_SECTION; | |
193 | mapsize = PAGE_ALIGN(mapsize) >> PAGE_SHIFT; | |
194 | ||
195 | /* remember memmap's page */ | |
196 | for (i = 0; i < mapsize; i++, page++) | |
197 | get_page_bootmem(section_nr, page, SECTION_INFO); | |
198 | ||
f1eca35a DW |
199 | usage = ms->usage; |
200 | page = virt_to_page(usage); | |
04753278 | 201 | |
f1eca35a | 202 | mapsize = PAGE_ALIGN(mem_section_usage_size()) >> PAGE_SHIFT; |
04753278 YG |
203 | |
204 | for (i = 0; i < mapsize; i++, page++) | |
af370fb8 | 205 | get_page_bootmem(section_nr, page, MIX_SECTION_INFO); |
04753278 YG |
206 | |
207 | } | |
46723bfa YI |
208 | #else /* CONFIG_SPARSEMEM_VMEMMAP */ |
209 | static void register_page_bootmem_info_section(unsigned long start_pfn) | |
210 | { | |
f1eca35a | 211 | unsigned long mapsize, section_nr, i; |
46723bfa YI |
212 | struct mem_section *ms; |
213 | struct page *page, *memmap; | |
f1eca35a | 214 | struct mem_section_usage *usage; |
46723bfa | 215 | |
46723bfa YI |
216 | section_nr = pfn_to_section_nr(start_pfn); |
217 | ms = __nr_to_section(section_nr); | |
218 | ||
219 | memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr); | |
220 | ||
221 | register_page_bootmem_memmap(section_nr, memmap, PAGES_PER_SECTION); | |
222 | ||
f1eca35a DW |
223 | usage = ms->usage; |
224 | page = virt_to_page(usage); | |
46723bfa | 225 | |
f1eca35a | 226 | mapsize = PAGE_ALIGN(mem_section_usage_size()) >> PAGE_SHIFT; |
46723bfa YI |
227 | |
228 | for (i = 0; i < mapsize; i++, page++) | |
229 | get_page_bootmem(section_nr, page, MIX_SECTION_INFO); | |
230 | } | |
231 | #endif /* !CONFIG_SPARSEMEM_VMEMMAP */ | |
04753278 | 232 | |
7ded384a | 233 | void __init register_page_bootmem_info_node(struct pglist_data *pgdat) |
04753278 YG |
234 | { |
235 | unsigned long i, pfn, end_pfn, nr_pages; | |
236 | int node = pgdat->node_id; | |
237 | struct page *page; | |
04753278 YG |
238 | |
239 | nr_pages = PAGE_ALIGN(sizeof(struct pglist_data)) >> PAGE_SHIFT; | |
240 | page = virt_to_page(pgdat); | |
241 | ||
242 | for (i = 0; i < nr_pages; i++, page++) | |
243 | get_page_bootmem(node, page, NODE_INFO); | |
244 | ||
04753278 | 245 | pfn = pgdat->node_start_pfn; |
c1f19495 | 246 | end_pfn = pgdat_end_pfn(pgdat); |
04753278 | 247 | |
7e9f5eb0 | 248 | /* register section info */ |
f14851af | 249 | for (; pfn < end_pfn; pfn += PAGES_PER_SECTION) { |
250 | /* | |
251 | * Some platforms can assign the same pfn to multiple nodes - on | |
252 | * node0 as well as nodeN. To avoid registering a pfn against | |
253 | * multiple nodes we check that this pfn does not already | |
7e9f5eb0 | 254 | * reside in some other nodes. |
f14851af | 255 | */ |
f65e91df | 256 | if (pfn_valid(pfn) && (early_pfn_to_nid(pfn) == node)) |
f14851af | 257 | register_page_bootmem_info_section(pfn); |
258 | } | |
04753278 | 259 | } |
46723bfa | 260 | #endif /* CONFIG_HAVE_BOOTMEM_INFO_NODE */ |
04753278 | 261 | |
7ea62160 DW |
262 | static int check_pfn_span(unsigned long pfn, unsigned long nr_pages, |
263 | const char *reason) | |
264 | { | |
265 | /* | |
266 | * Disallow all operations smaller than a sub-section and only | |
267 | * allow operations smaller than a section for | |
268 | * SPARSEMEM_VMEMMAP. Note that check_hotplug_memory_range() | |
269 | * enforces a larger memory_block_size_bytes() granularity for | |
270 | * memory that will be marked online, so this check should only | |
271 | * fire for direct arch_{add,remove}_memory() users outside of | |
272 | * add_memory_resource(). | |
273 | */ | |
274 | unsigned long min_align; | |
275 | ||
276 | if (IS_ENABLED(CONFIG_SPARSEMEM_VMEMMAP)) | |
277 | min_align = PAGES_PER_SUBSECTION; | |
278 | else | |
279 | min_align = PAGES_PER_SECTION; | |
280 | if (!IS_ALIGNED(pfn, min_align) | |
281 | || !IS_ALIGNED(nr_pages, min_align)) { | |
282 | WARN(1, "Misaligned __%s_pages start: %#lx end: #%lx\n", | |
283 | reason, pfn, pfn + nr_pages - 1); | |
284 | return -EINVAL; | |
285 | } | |
286 | return 0; | |
287 | } | |
288 | ||
dca4436d AS |
289 | static int check_hotplug_memory_addressable(unsigned long pfn, |
290 | unsigned long nr_pages) | |
291 | { | |
292 | const u64 max_addr = PFN_PHYS(pfn + nr_pages) - 1; | |
293 | ||
294 | if (max_addr >> MAX_PHYSMEM_BITS) { | |
295 | const u64 max_allowed = (1ull << (MAX_PHYSMEM_BITS + 1)) - 1; | |
296 | WARN(1, | |
297 | "Hotplugged memory exceeds maximum addressable address, range=%#llx-%#llx, maximum=%#llx\n", | |
298 | (u64)PFN_PHYS(pfn), max_addr, max_allowed); | |
299 | return -E2BIG; | |
300 | } | |
301 | ||
302 | return 0; | |
303 | } | |
304 | ||
4edd7cef DR |
305 | /* |
306 | * Reasonably generic function for adding memory. It is | |
307 | * expected that archs that support memory hotplug will | |
308 | * call this function after deciding the zone to which to | |
309 | * add the new pages. | |
310 | */ | |
7ea62160 | 311 | int __ref __add_pages(int nid, unsigned long pfn, unsigned long nr_pages, |
f5637d3b | 312 | struct mhp_params *params) |
4edd7cef | 313 | { |
6cdd0b30 DH |
314 | const unsigned long end_pfn = pfn + nr_pages; |
315 | unsigned long cur_nr_pages; | |
9a845030 | 316 | int err; |
f5637d3b | 317 | struct vmem_altmap *altmap = params->altmap; |
4b94ffdc | 318 | |
bfeb022f LG |
319 | if (WARN_ON_ONCE(!params->pgprot.pgprot)) |
320 | return -EINVAL; | |
321 | ||
dca4436d AS |
322 | err = check_hotplug_memory_addressable(pfn, nr_pages); |
323 | if (err) | |
324 | return err; | |
325 | ||
4b94ffdc DW |
326 | if (altmap) { |
327 | /* | |
328 | * Validate altmap is within bounds of the total request | |
329 | */ | |
7ea62160 | 330 | if (altmap->base_pfn != pfn |
4b94ffdc DW |
331 | || vmem_altmap_offset(altmap) > nr_pages) { |
332 | pr_warn_once("memory add fail, invalid altmap\n"); | |
7ea62160 | 333 | return -EINVAL; |
4b94ffdc DW |
334 | } |
335 | altmap->alloc = 0; | |
336 | } | |
337 | ||
7ea62160 DW |
338 | err = check_pfn_span(pfn, nr_pages, "add"); |
339 | if (err) | |
340 | return err; | |
341 | ||
6cdd0b30 DH |
342 | for (; pfn < end_pfn; pfn += cur_nr_pages) { |
343 | /* Select all remaining pages up to the next section boundary */ | |
344 | cur_nr_pages = min(end_pfn - pfn, | |
345 | SECTION_ALIGN_UP(pfn + 1) - pfn); | |
346 | err = sparse_add_section(nid, pfn, cur_nr_pages, altmap); | |
ba72b4c8 DW |
347 | if (err) |
348 | break; | |
f64ac5e6 | 349 | cond_resched(); |
4edd7cef | 350 | } |
c435a390 | 351 | vmemmap_populate_print_last(); |
4edd7cef DR |
352 | return err; |
353 | } | |
4edd7cef | 354 | |
815121d2 | 355 | /* find the smallest valid pfn in the range [start_pfn, end_pfn) */ |
d09b0137 | 356 | static unsigned long find_smallest_section_pfn(int nid, struct zone *zone, |
815121d2 YI |
357 | unsigned long start_pfn, |
358 | unsigned long end_pfn) | |
359 | { | |
49ba3c6b | 360 | for (; start_pfn < end_pfn; start_pfn += PAGES_PER_SUBSECTION) { |
7ce700bf | 361 | if (unlikely(!pfn_to_online_page(start_pfn))) |
815121d2 YI |
362 | continue; |
363 | ||
364 | if (unlikely(pfn_to_nid(start_pfn) != nid)) | |
365 | continue; | |
366 | ||
9b05158f | 367 | if (zone != page_zone(pfn_to_page(start_pfn))) |
815121d2 YI |
368 | continue; |
369 | ||
370 | return start_pfn; | |
371 | } | |
372 | ||
373 | return 0; | |
374 | } | |
375 | ||
376 | /* find the biggest valid pfn in the range [start_pfn, end_pfn). */ | |
d09b0137 | 377 | static unsigned long find_biggest_section_pfn(int nid, struct zone *zone, |
815121d2 YI |
378 | unsigned long start_pfn, |
379 | unsigned long end_pfn) | |
380 | { | |
815121d2 YI |
381 | unsigned long pfn; |
382 | ||
383 | /* pfn is the end pfn of a memory section. */ | |
384 | pfn = end_pfn - 1; | |
49ba3c6b | 385 | for (; pfn >= start_pfn; pfn -= PAGES_PER_SUBSECTION) { |
7ce700bf | 386 | if (unlikely(!pfn_to_online_page(pfn))) |
815121d2 YI |
387 | continue; |
388 | ||
389 | if (unlikely(pfn_to_nid(pfn) != nid)) | |
390 | continue; | |
391 | ||
9b05158f | 392 | if (zone != page_zone(pfn_to_page(pfn))) |
815121d2 YI |
393 | continue; |
394 | ||
395 | return pfn; | |
396 | } | |
397 | ||
398 | return 0; | |
399 | } | |
400 | ||
401 | static void shrink_zone_span(struct zone *zone, unsigned long start_pfn, | |
402 | unsigned long end_pfn) | |
403 | { | |
815121d2 | 404 | unsigned long pfn; |
815121d2 YI |
405 | int nid = zone_to_nid(zone); |
406 | ||
407 | zone_span_writelock(zone); | |
5d12071c | 408 | if (zone->zone_start_pfn == start_pfn) { |
815121d2 YI |
409 | /* |
410 | * If the section is smallest section in the zone, it need | |
411 | * shrink zone->zone_start_pfn and zone->zone_spanned_pages. | |
412 | * In this case, we find second smallest valid mem_section | |
413 | * for shrinking zone. | |
414 | */ | |
415 | pfn = find_smallest_section_pfn(nid, zone, end_pfn, | |
5d12071c | 416 | zone_end_pfn(zone)); |
815121d2 | 417 | if (pfn) { |
5d12071c | 418 | zone->spanned_pages = zone_end_pfn(zone) - pfn; |
815121d2 | 419 | zone->zone_start_pfn = pfn; |
950b68d9 DH |
420 | } else { |
421 | zone->zone_start_pfn = 0; | |
422 | zone->spanned_pages = 0; | |
815121d2 | 423 | } |
5d12071c | 424 | } else if (zone_end_pfn(zone) == end_pfn) { |
815121d2 YI |
425 | /* |
426 | * If the section is biggest section in the zone, it need | |
427 | * shrink zone->spanned_pages. | |
428 | * In this case, we find second biggest valid mem_section for | |
429 | * shrinking zone. | |
430 | */ | |
5d12071c | 431 | pfn = find_biggest_section_pfn(nid, zone, zone->zone_start_pfn, |
815121d2 YI |
432 | start_pfn); |
433 | if (pfn) | |
5d12071c | 434 | zone->spanned_pages = pfn - zone->zone_start_pfn + 1; |
950b68d9 DH |
435 | else { |
436 | zone->zone_start_pfn = 0; | |
437 | zone->spanned_pages = 0; | |
438 | } | |
815121d2 | 439 | } |
815121d2 YI |
440 | zone_span_writeunlock(zone); |
441 | } | |
442 | ||
00d6c019 | 443 | static void update_pgdat_span(struct pglist_data *pgdat) |
815121d2 | 444 | { |
00d6c019 DH |
445 | unsigned long node_start_pfn = 0, node_end_pfn = 0; |
446 | struct zone *zone; | |
447 | ||
448 | for (zone = pgdat->node_zones; | |
449 | zone < pgdat->node_zones + MAX_NR_ZONES; zone++) { | |
450 | unsigned long zone_end_pfn = zone->zone_start_pfn + | |
451 | zone->spanned_pages; | |
452 | ||
453 | /* No need to lock the zones, they can't change. */ | |
656d5711 DH |
454 | if (!zone->spanned_pages) |
455 | continue; | |
456 | if (!node_end_pfn) { | |
457 | node_start_pfn = zone->zone_start_pfn; | |
458 | node_end_pfn = zone_end_pfn; | |
459 | continue; | |
460 | } | |
461 | ||
00d6c019 DH |
462 | if (zone_end_pfn > node_end_pfn) |
463 | node_end_pfn = zone_end_pfn; | |
464 | if (zone->zone_start_pfn < node_start_pfn) | |
465 | node_start_pfn = zone->zone_start_pfn; | |
815121d2 YI |
466 | } |
467 | ||
00d6c019 DH |
468 | pgdat->node_start_pfn = node_start_pfn; |
469 | pgdat->node_spanned_pages = node_end_pfn - node_start_pfn; | |
815121d2 YI |
470 | } |
471 | ||
feee6b29 DH |
472 | void __ref remove_pfn_range_from_zone(struct zone *zone, |
473 | unsigned long start_pfn, | |
474 | unsigned long nr_pages) | |
815121d2 | 475 | { |
b7e3debd | 476 | const unsigned long end_pfn = start_pfn + nr_pages; |
815121d2 | 477 | struct pglist_data *pgdat = zone->zone_pgdat; |
b7e3debd | 478 | unsigned long pfn, cur_nr_pages, flags; |
815121d2 | 479 | |
d33695b1 | 480 | /* Poison struct pages because they are now uninitialized again. */ |
b7e3debd BW |
481 | for (pfn = start_pfn; pfn < end_pfn; pfn += cur_nr_pages) { |
482 | cond_resched(); | |
483 | ||
484 | /* Select all remaining pages up to the next section boundary */ | |
485 | cur_nr_pages = | |
486 | min(end_pfn - pfn, SECTION_ALIGN_UP(pfn + 1) - pfn); | |
487 | page_init_poison(pfn_to_page(pfn), | |
488 | sizeof(struct page) * cur_nr_pages); | |
489 | } | |
d33695b1 | 490 | |
7ce700bf DH |
491 | #ifdef CONFIG_ZONE_DEVICE |
492 | /* | |
493 | * Zone shrinking code cannot properly deal with ZONE_DEVICE. So | |
494 | * we will not try to shrink the zones - which is okay as | |
495 | * set_zone_contiguous() cannot deal with ZONE_DEVICE either way. | |
496 | */ | |
497 | if (zone_idx(zone) == ZONE_DEVICE) | |
498 | return; | |
499 | #endif | |
500 | ||
feee6b29 DH |
501 | clear_zone_contiguous(zone); |
502 | ||
815121d2 YI |
503 | pgdat_resize_lock(zone->zone_pgdat, &flags); |
504 | shrink_zone_span(zone, start_pfn, start_pfn + nr_pages); | |
00d6c019 | 505 | update_pgdat_span(pgdat); |
815121d2 | 506 | pgdat_resize_unlock(zone->zone_pgdat, &flags); |
feee6b29 DH |
507 | |
508 | set_zone_contiguous(zone); | |
815121d2 YI |
509 | } |
510 | ||
feee6b29 DH |
511 | static void __remove_section(unsigned long pfn, unsigned long nr_pages, |
512 | unsigned long map_offset, | |
513 | struct vmem_altmap *altmap) | |
ea01ea93 | 514 | { |
10404901 | 515 | struct mem_section *ms = __pfn_to_section(pfn); |
ea01ea93 | 516 | |
9d1d887d DH |
517 | if (WARN_ON_ONCE(!valid_section(ms))) |
518 | return; | |
ea01ea93 | 519 | |
ba72b4c8 | 520 | sparse_remove_section(ms, pfn, nr_pages, map_offset, altmap); |
ea01ea93 BP |
521 | } |
522 | ||
ea01ea93 | 523 | /** |
feee6b29 | 524 | * __remove_pages() - remove sections of pages |
7ea62160 | 525 | * @pfn: starting pageframe (must be aligned to start of a section) |
ea01ea93 | 526 | * @nr_pages: number of pages to remove (must be multiple of section size) |
e8b098fc | 527 | * @altmap: alternative device page map or %NULL if default memmap is used |
ea01ea93 BP |
528 | * |
529 | * Generic helper function to remove section mappings and sysfs entries | |
530 | * for the section of the memory we are removing. Caller needs to make | |
531 | * sure that pages are marked reserved and zones are adjust properly by | |
532 | * calling offline_pages(). | |
533 | */ | |
feee6b29 DH |
534 | void __remove_pages(unsigned long pfn, unsigned long nr_pages, |
535 | struct vmem_altmap *altmap) | |
ea01ea93 | 536 | { |
52fb87c8 DH |
537 | const unsigned long end_pfn = pfn + nr_pages; |
538 | unsigned long cur_nr_pages; | |
4b94ffdc | 539 | unsigned long map_offset = 0; |
4b94ffdc | 540 | |
96da4350 | 541 | map_offset = vmem_altmap_offset(altmap); |
ea01ea93 | 542 | |
7ea62160 DW |
543 | if (check_pfn_span(pfn, nr_pages, "remove")) |
544 | return; | |
ea01ea93 | 545 | |
52fb87c8 | 546 | for (; pfn < end_pfn; pfn += cur_nr_pages) { |
dd33ad7b | 547 | cond_resched(); |
52fb87c8 | 548 | /* Select all remaining pages up to the next section boundary */ |
a11b9419 DH |
549 | cur_nr_pages = min(end_pfn - pfn, |
550 | SECTION_ALIGN_UP(pfn + 1) - pfn); | |
52fb87c8 | 551 | __remove_section(pfn, cur_nr_pages, map_offset, altmap); |
4b94ffdc | 552 | map_offset = 0; |
ea01ea93 | 553 | } |
ea01ea93 | 554 | } |
ea01ea93 | 555 | |
9d0ad8ca DK |
556 | int set_online_page_callback(online_page_callback_t callback) |
557 | { | |
558 | int rc = -EINVAL; | |
559 | ||
bfc8c901 VD |
560 | get_online_mems(); |
561 | mutex_lock(&online_page_callback_lock); | |
9d0ad8ca DK |
562 | |
563 | if (online_page_callback == generic_online_page) { | |
564 | online_page_callback = callback; | |
565 | rc = 0; | |
566 | } | |
567 | ||
bfc8c901 VD |
568 | mutex_unlock(&online_page_callback_lock); |
569 | put_online_mems(); | |
9d0ad8ca DK |
570 | |
571 | return rc; | |
572 | } | |
573 | EXPORT_SYMBOL_GPL(set_online_page_callback); | |
574 | ||
575 | int restore_online_page_callback(online_page_callback_t callback) | |
576 | { | |
577 | int rc = -EINVAL; | |
578 | ||
bfc8c901 VD |
579 | get_online_mems(); |
580 | mutex_lock(&online_page_callback_lock); | |
9d0ad8ca DK |
581 | |
582 | if (online_page_callback == callback) { | |
583 | online_page_callback = generic_online_page; | |
584 | rc = 0; | |
585 | } | |
586 | ||
bfc8c901 VD |
587 | mutex_unlock(&online_page_callback_lock); |
588 | put_online_mems(); | |
9d0ad8ca DK |
589 | |
590 | return rc; | |
591 | } | |
592 | EXPORT_SYMBOL_GPL(restore_online_page_callback); | |
593 | ||
18db1491 | 594 | void generic_online_page(struct page *page, unsigned int order) |
9d0ad8ca | 595 | { |
c87cbc1f VB |
596 | /* |
597 | * Freeing the page with debug_pagealloc enabled will try to unmap it, | |
598 | * so we should map it first. This is better than introducing a special | |
599 | * case in page freeing fast path. | |
600 | */ | |
77bc7fd6 | 601 | debug_pagealloc_map_pages(page, 1 << order); |
a9cd410a AK |
602 | __free_pages_core(page, order); |
603 | totalram_pages_add(1UL << order); | |
604 | #ifdef CONFIG_HIGHMEM | |
605 | if (PageHighMem(page)) | |
606 | totalhigh_pages_add(1UL << order); | |
607 | #endif | |
608 | } | |
18db1491 | 609 | EXPORT_SYMBOL_GPL(generic_online_page); |
a9cd410a | 610 | |
aac65321 | 611 | static void online_pages_range(unsigned long start_pfn, unsigned long nr_pages) |
3947be19 | 612 | { |
b2c2ab20 DH |
613 | const unsigned long end_pfn = start_pfn + nr_pages; |
614 | unsigned long pfn; | |
b2c2ab20 DH |
615 | |
616 | /* | |
aac65321 DH |
617 | * Online the pages in MAX_ORDER - 1 aligned chunks. The callback might |
618 | * decide to not expose all pages to the buddy (e.g., expose them | |
619 | * later). We account all pages as being online and belonging to this | |
620 | * zone ("present"). | |
b2c2ab20 | 621 | */ |
aac65321 DH |
622 | for (pfn = start_pfn; pfn < end_pfn; pfn += MAX_ORDER_NR_PAGES) |
623 | (*online_page_callback)(pfn_to_page(pfn), MAX_ORDER - 1); | |
2d070eab | 624 | |
b2c2ab20 DH |
625 | /* mark all involved sections as online */ |
626 | online_mem_sections(start_pfn, end_pfn); | |
75884fb1 KH |
627 | } |
628 | ||
d9713679 LJ |
629 | /* check which state of node_states will be changed when online memory */ |
630 | static void node_states_check_changes_online(unsigned long nr_pages, | |
631 | struct zone *zone, struct memory_notify *arg) | |
632 | { | |
633 | int nid = zone_to_nid(zone); | |
d9713679 | 634 | |
98fa15f3 AK |
635 | arg->status_change_nid = NUMA_NO_NODE; |
636 | arg->status_change_nid_normal = NUMA_NO_NODE; | |
637 | arg->status_change_nid_high = NUMA_NO_NODE; | |
d9713679 | 638 | |
8efe33f4 OS |
639 | if (!node_state(nid, N_MEMORY)) |
640 | arg->status_change_nid = nid; | |
641 | if (zone_idx(zone) <= ZONE_NORMAL && !node_state(nid, N_NORMAL_MEMORY)) | |
d9713679 | 642 | arg->status_change_nid_normal = nid; |
6715ddf9 | 643 | #ifdef CONFIG_HIGHMEM |
d3ba3ae1 | 644 | if (zone_idx(zone) <= ZONE_HIGHMEM && !node_state(nid, N_HIGH_MEMORY)) |
6715ddf9 | 645 | arg->status_change_nid_high = nid; |
6715ddf9 | 646 | #endif |
d9713679 LJ |
647 | } |
648 | ||
649 | static void node_states_set_node(int node, struct memory_notify *arg) | |
650 | { | |
651 | if (arg->status_change_nid_normal >= 0) | |
652 | node_set_state(node, N_NORMAL_MEMORY); | |
653 | ||
6715ddf9 LJ |
654 | if (arg->status_change_nid_high >= 0) |
655 | node_set_state(node, N_HIGH_MEMORY); | |
656 | ||
83d83612 OS |
657 | if (arg->status_change_nid >= 0) |
658 | node_set_state(node, N_MEMORY); | |
d9713679 LJ |
659 | } |
660 | ||
f1dd2cd1 MH |
661 | static void __meminit resize_zone_range(struct zone *zone, unsigned long start_pfn, |
662 | unsigned long nr_pages) | |
663 | { | |
664 | unsigned long old_end_pfn = zone_end_pfn(zone); | |
665 | ||
666 | if (zone_is_empty(zone) || start_pfn < zone->zone_start_pfn) | |
667 | zone->zone_start_pfn = start_pfn; | |
668 | ||
669 | zone->spanned_pages = max(start_pfn + nr_pages, old_end_pfn) - zone->zone_start_pfn; | |
670 | } | |
671 | ||
672 | static void __meminit resize_pgdat_range(struct pglist_data *pgdat, unsigned long start_pfn, | |
673 | unsigned long nr_pages) | |
674 | { | |
675 | unsigned long old_end_pfn = pgdat_end_pfn(pgdat); | |
676 | ||
677 | if (!pgdat->node_spanned_pages || start_pfn < pgdat->node_start_pfn) | |
678 | pgdat->node_start_pfn = start_pfn; | |
679 | ||
680 | pgdat->node_spanned_pages = max(start_pfn + nr_pages, old_end_pfn) - pgdat->node_start_pfn; | |
f1dd2cd1 | 681 | |
3fccb74c DH |
682 | } |
683 | /* | |
684 | * Associate the pfn range with the given zone, initializing the memmaps | |
685 | * and resizing the pgdat/zone data to span the added pages. After this | |
686 | * call, all affected pages are PG_reserved. | |
d882c006 DH |
687 | * |
688 | * All aligned pageblocks are initialized to the specified migratetype | |
689 | * (usually MIGRATE_MOVABLE). Besides setting the migratetype, no related | |
690 | * zone stats (e.g., nr_isolate_pageblock) are touched. | |
3fccb74c | 691 | */ |
a99583e7 | 692 | void __ref move_pfn_range_to_zone(struct zone *zone, unsigned long start_pfn, |
d882c006 DH |
693 | unsigned long nr_pages, |
694 | struct vmem_altmap *altmap, int migratetype) | |
f1dd2cd1 MH |
695 | { |
696 | struct pglist_data *pgdat = zone->zone_pgdat; | |
697 | int nid = pgdat->node_id; | |
698 | unsigned long flags; | |
df429ac0 | 699 | |
f1dd2cd1 MH |
700 | clear_zone_contiguous(zone); |
701 | ||
702 | /* TODO Huh pgdat is irqsave while zone is not. It used to be like that before */ | |
703 | pgdat_resize_lock(pgdat, &flags); | |
704 | zone_span_writelock(zone); | |
fa004ab7 WY |
705 | if (zone_is_empty(zone)) |
706 | init_currently_empty_zone(zone, start_pfn, nr_pages); | |
f1dd2cd1 MH |
707 | resize_zone_range(zone, start_pfn, nr_pages); |
708 | zone_span_writeunlock(zone); | |
709 | resize_pgdat_range(pgdat, start_pfn, nr_pages); | |
710 | pgdat_resize_unlock(pgdat, &flags); | |
711 | ||
712 | /* | |
713 | * TODO now we have a visible range of pages which are not associated | |
714 | * with their zone properly. Not nice but set_pfnblock_flags_mask | |
715 | * expects the zone spans the pfn range. All the pages in the range | |
716 | * are reserved so nobody should be touching them so we should be safe | |
717 | */ | |
dc2da7b4 | 718 | memmap_init_zone(nr_pages, nid, zone_idx(zone), start_pfn, 0, |
d882c006 | 719 | MEMINIT_HOTPLUG, altmap, migratetype); |
f1dd2cd1 MH |
720 | |
721 | set_zone_contiguous(zone); | |
722 | } | |
723 | ||
accfb3e0 SF |
724 | void set_default_mem_hotplug_zone(enum zone_type zone) |
725 | { | |
726 | default_kernel_zone = zone; | |
727 | } | |
728 | ||
729 | #ifdef CONFIG_HIGHMEM | |
730 | #define MAX_KERNEL_ZONE ZONE_HIGHMEM | |
731 | #else | |
732 | #define MAX_KERNEL_ZONE ZONE_NORMAL | |
733 | #endif | |
734 | ||
c246a213 MH |
735 | /* |
736 | * Returns a default kernel memory zone for the given pfn range. | |
737 | * If no kernel zone covers this pfn range it will automatically go | |
accfb3e0 | 738 | * to the MAX_KERNEL_ZONE. |
c246a213 | 739 | */ |
c6f03e29 | 740 | static struct zone *default_kernel_zone_for_pfn(int nid, unsigned long start_pfn, |
c246a213 MH |
741 | unsigned long nr_pages) |
742 | { | |
743 | struct pglist_data *pgdat = NODE_DATA(nid); | |
744 | int zid; | |
745 | ||
accfb3e0 | 746 | for (zid = 0; zid <= MAX_KERNEL_ZONE; zid++) { |
c246a213 MH |
747 | struct zone *zone = &pgdat->node_zones[zid]; |
748 | ||
749 | if (zone_intersects(zone, start_pfn, nr_pages)) | |
750 | return zone; | |
751 | } | |
752 | ||
accfb3e0 | 753 | return &pgdat->node_zones[default_kernel_zone]; |
c246a213 MH |
754 | } |
755 | ||
c6f03e29 MH |
756 | static inline struct zone *default_zone_for_pfn(int nid, unsigned long start_pfn, |
757 | unsigned long nr_pages) | |
e5e68930 | 758 | { |
c6f03e29 MH |
759 | struct zone *kernel_zone = default_kernel_zone_for_pfn(nid, start_pfn, |
760 | nr_pages); | |
761 | struct zone *movable_zone = &NODE_DATA(nid)->node_zones[ZONE_MOVABLE]; | |
762 | bool in_kernel = zone_intersects(kernel_zone, start_pfn, nr_pages); | |
763 | bool in_movable = zone_intersects(movable_zone, start_pfn, nr_pages); | |
e5e68930 MH |
764 | |
765 | /* | |
c6f03e29 MH |
766 | * We inherit the existing zone in a simple case where zones do not |
767 | * overlap in the given range | |
e5e68930 | 768 | */ |
c6f03e29 MH |
769 | if (in_kernel ^ in_movable) |
770 | return (in_kernel) ? kernel_zone : movable_zone; | |
9f123ab5 | 771 | |
c6f03e29 MH |
772 | /* |
773 | * If the range doesn't belong to any zone or two zones overlap in the | |
774 | * given range then we use movable zone only if movable_node is | |
775 | * enabled because we always online to a kernel zone by default. | |
776 | */ | |
777 | return movable_node_enabled ? movable_zone : kernel_zone; | |
9f123ab5 MH |
778 | } |
779 | ||
e5e68930 MH |
780 | struct zone * zone_for_pfn_range(int online_type, int nid, unsigned start_pfn, |
781 | unsigned long nr_pages) | |
f1dd2cd1 | 782 | { |
c6f03e29 MH |
783 | if (online_type == MMOP_ONLINE_KERNEL) |
784 | return default_kernel_zone_for_pfn(nid, start_pfn, nr_pages); | |
f1dd2cd1 | 785 | |
c6f03e29 MH |
786 | if (online_type == MMOP_ONLINE_MOVABLE) |
787 | return &NODE_DATA(nid)->node_zones[ZONE_MOVABLE]; | |
df429ac0 | 788 | |
c6f03e29 | 789 | return default_zone_for_pfn(nid, start_pfn, nr_pages); |
e5e68930 MH |
790 | } |
791 | ||
bd5c2344 DH |
792 | int __ref online_pages(unsigned long pfn, unsigned long nr_pages, |
793 | int online_type, int nid) | |
75884fb1 | 794 | { |
aa47228a | 795 | unsigned long flags; |
3947be19 | 796 | struct zone *zone; |
6811378e | 797 | int need_zonelists_rebuild = 0; |
7b78d335 YG |
798 | int ret; |
799 | struct memory_notify arg; | |
d0dc12e8 | 800 | |
4986fac1 DH |
801 | /* We can only online full sections (e.g., SECTION_IS_ONLINE) */ |
802 | if (WARN_ON_ONCE(!nr_pages || | |
803 | !IS_ALIGNED(pfn | nr_pages, PAGES_PER_SECTION))) | |
804 | return -EINVAL; | |
805 | ||
381eab4a DH |
806 | mem_hotplug_begin(); |
807 | ||
f1dd2cd1 | 808 | /* associate pfn range with the zone */ |
3fccb74c | 809 | zone = zone_for_pfn_range(online_type, nid, pfn, nr_pages); |
b30c5927 | 810 | move_pfn_range_to_zone(zone, pfn, nr_pages, NULL, MIGRATE_ISOLATE); |
f1dd2cd1 | 811 | |
7b78d335 YG |
812 | arg.start_pfn = pfn; |
813 | arg.nr_pages = nr_pages; | |
d9713679 | 814 | node_states_check_changes_online(nr_pages, zone, &arg); |
7b78d335 | 815 | |
7b78d335 YG |
816 | ret = memory_notify(MEM_GOING_ONLINE, &arg); |
817 | ret = notifier_to_errno(ret); | |
e33e33b4 CY |
818 | if (ret) |
819 | goto failed_addition; | |
820 | ||
b30c5927 DH |
821 | /* |
822 | * Fixup the number of isolated pageblocks before marking the sections | |
823 | * onlining, such that undo_isolate_page_range() works correctly. | |
824 | */ | |
825 | spin_lock_irqsave(&zone->lock, flags); | |
826 | zone->nr_isolate_pageblock += nr_pages / pageblock_nr_pages; | |
827 | spin_unlock_irqrestore(&zone->lock, flags); | |
828 | ||
6811378e YG |
829 | /* |
830 | * If this zone is not populated, then it is not in zonelist. | |
831 | * This means the page allocator ignores this zone. | |
832 | * So, zonelist must be updated after online. | |
833 | */ | |
6dcd73d7 | 834 | if (!populated_zone(zone)) { |
6811378e | 835 | need_zonelists_rebuild = 1; |
72675e13 | 836 | setup_zone_pageset(zone); |
6dcd73d7 | 837 | } |
6811378e | 838 | |
aac65321 DH |
839 | online_pages_range(pfn, nr_pages); |
840 | zone->present_pages += nr_pages; | |
aa47228a CS |
841 | |
842 | pgdat_resize_lock(zone->zone_pgdat, &flags); | |
aac65321 | 843 | zone->zone_pgdat->node_present_pages += nr_pages; |
aa47228a CS |
844 | pgdat_resize_unlock(zone->zone_pgdat, &flags); |
845 | ||
b30c5927 DH |
846 | node_states_set_node(nid, &arg); |
847 | if (need_zonelists_rebuild) | |
848 | build_all_zonelists(NULL); | |
849 | zone_pcp_update(zone); | |
850 | ||
851 | /* Basic onlining is complete, allow allocation of onlined pages. */ | |
852 | undo_isolate_page_range(pfn, pfn + nr_pages, MIGRATE_MOVABLE); | |
853 | ||
93146d98 | 854 | /* |
b86c5fc4 DH |
855 | * Freshly onlined pages aren't shuffled (e.g., all pages are placed to |
856 | * the tail of the freelist when undoing isolation). Shuffle the whole | |
857 | * zone to make sure the just onlined pages are properly distributed | |
858 | * across the whole freelist - to create an initial shuffle. | |
93146d98 | 859 | */ |
e900a918 DW |
860 | shuffle_zone(zone); |
861 | ||
1b79acc9 KM |
862 | init_per_zone_wmark_min(); |
863 | ||
ca9a46f8 DH |
864 | kswapd_run(nid); |
865 | kcompactd_run(nid); | |
61b13993 | 866 | |
2d1d43f6 | 867 | writeback_set_ratelimit(); |
7b78d335 | 868 | |
ca9a46f8 | 869 | memory_notify(MEM_ONLINE, &arg); |
381eab4a | 870 | mem_hotplug_done(); |
30467e0b | 871 | return 0; |
e33e33b4 CY |
872 | |
873 | failed_addition: | |
874 | pr_debug("online_pages [mem %#010llx-%#010llx] failed\n", | |
875 | (unsigned long long) pfn << PAGE_SHIFT, | |
876 | (((unsigned long long) pfn + nr_pages) << PAGE_SHIFT) - 1); | |
877 | memory_notify(MEM_CANCEL_ONLINE, &arg); | |
feee6b29 | 878 | remove_pfn_range_from_zone(zone, pfn, nr_pages); |
381eab4a | 879 | mem_hotplug_done(); |
e33e33b4 | 880 | return ret; |
3947be19 | 881 | } |
53947027 | 882 | #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */ |
bc02af93 | 883 | |
0bd85420 TC |
884 | static void reset_node_present_pages(pg_data_t *pgdat) |
885 | { | |
886 | struct zone *z; | |
887 | ||
888 | for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++) | |
889 | z->present_pages = 0; | |
890 | ||
891 | pgdat->node_present_pages = 0; | |
892 | } | |
893 | ||
e1319331 | 894 | /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */ |
c68ab18c | 895 | static pg_data_t __ref *hotadd_new_pgdat(int nid) |
9af3c2de YG |
896 | { |
897 | struct pglist_data *pgdat; | |
9af3c2de | 898 | |
a1e565aa TC |
899 | pgdat = NODE_DATA(nid); |
900 | if (!pgdat) { | |
901 | pgdat = arch_alloc_nodedata(nid); | |
902 | if (!pgdat) | |
903 | return NULL; | |
9af3c2de | 904 | |
33fce011 WY |
905 | pgdat->per_cpu_nodestats = |
906 | alloc_percpu(struct per_cpu_nodestat); | |
a1e565aa | 907 | arch_refresh_nodedata(nid, pgdat); |
b0dc3a34 | 908 | } else { |
33fce011 | 909 | int cpu; |
e716f2eb | 910 | /* |
97a225e6 JK |
911 | * Reset the nr_zones, order and highest_zoneidx before reuse. |
912 | * Note that kswapd will init kswapd_highest_zoneidx properly | |
e716f2eb MG |
913 | * when it starts in the near future. |
914 | */ | |
b0dc3a34 | 915 | pgdat->nr_zones = 0; |
38087d9b | 916 | pgdat->kswapd_order = 0; |
97a225e6 | 917 | pgdat->kswapd_highest_zoneidx = 0; |
33fce011 WY |
918 | for_each_online_cpu(cpu) { |
919 | struct per_cpu_nodestat *p; | |
920 | ||
921 | p = per_cpu_ptr(pgdat->per_cpu_nodestats, cpu); | |
922 | memset(p, 0, sizeof(*p)); | |
923 | } | |
a1e565aa | 924 | } |
9af3c2de YG |
925 | |
926 | /* we can use NODE_DATA(nid) from here */ | |
03e85f9d | 927 | pgdat->node_id = nid; |
c68ab18c | 928 | pgdat->node_start_pfn = 0; |
03e85f9d | 929 | |
9af3c2de | 930 | /* init node's zones as empty zones, we don't have any present pages.*/ |
03e85f9d | 931 | free_area_init_core_hotplug(nid); |
9af3c2de | 932 | |
959ecc48 KH |
933 | /* |
934 | * The node we allocated has no zone fallback lists. For avoiding | |
935 | * to access not-initialized zonelist, build here. | |
936 | */ | |
72675e13 | 937 | build_all_zonelists(pgdat); |
959ecc48 | 938 | |
0bd85420 TC |
939 | /* |
940 | * When memory is hot-added, all the memory is in offline state. So | |
941 | * clear all zones' present_pages because they will be updated in | |
942 | * online_pages() and offline_pages(). | |
943 | */ | |
03e85f9d | 944 | reset_node_managed_pages(pgdat); |
0bd85420 TC |
945 | reset_node_present_pages(pgdat); |
946 | ||
9af3c2de YG |
947 | return pgdat; |
948 | } | |
949 | ||
b9ff0360 | 950 | static void rollback_node_hotadd(int nid) |
9af3c2de | 951 | { |
b9ff0360 OS |
952 | pg_data_t *pgdat = NODE_DATA(nid); |
953 | ||
9af3c2de | 954 | arch_refresh_nodedata(nid, NULL); |
5830169f | 955 | free_percpu(pgdat->per_cpu_nodestats); |
9af3c2de | 956 | arch_free_nodedata(pgdat); |
9af3c2de YG |
957 | } |
958 | ||
0a547039 | 959 | |
01b0f197 TK |
960 | /** |
961 | * try_online_node - online a node if offlined | |
e8b098fc | 962 | * @nid: the node ID |
b9ff0360 | 963 | * @set_node_online: Whether we want to online the node |
cf23422b | 964 | * called by cpu_up() to online a node without onlined memory. |
b9ff0360 OS |
965 | * |
966 | * Returns: | |
967 | * 1 -> a new node has been allocated | |
968 | * 0 -> the node is already online | |
969 | * -ENOMEM -> the node could not be allocated | |
cf23422b | 970 | */ |
c68ab18c | 971 | static int __try_online_node(int nid, bool set_node_online) |
cf23422b | 972 | { |
b9ff0360 OS |
973 | pg_data_t *pgdat; |
974 | int ret = 1; | |
cf23422b | 975 | |
01b0f197 TK |
976 | if (node_online(nid)) |
977 | return 0; | |
978 | ||
c68ab18c | 979 | pgdat = hotadd_new_pgdat(nid); |
7553e8f2 | 980 | if (!pgdat) { |
01b0f197 | 981 | pr_err("Cannot online node %d due to NULL pgdat\n", nid); |
cf23422b | 982 | ret = -ENOMEM; |
983 | goto out; | |
984 | } | |
b9ff0360 OS |
985 | |
986 | if (set_node_online) { | |
987 | node_set_online(nid); | |
988 | ret = register_one_node(nid); | |
989 | BUG_ON(ret); | |
990 | } | |
cf23422b | 991 | out: |
b9ff0360 OS |
992 | return ret; |
993 | } | |
994 | ||
995 | /* | |
996 | * Users of this function always want to online/register the node | |
997 | */ | |
998 | int try_online_node(int nid) | |
999 | { | |
1000 | int ret; | |
1001 | ||
1002 | mem_hotplug_begin(); | |
c68ab18c | 1003 | ret = __try_online_node(nid, true); |
bfc8c901 | 1004 | mem_hotplug_done(); |
cf23422b | 1005 | return ret; |
1006 | } | |
1007 | ||
27356f54 TK |
1008 | static int check_hotplug_memory_range(u64 start, u64 size) |
1009 | { | |
ba325585 | 1010 | /* memory range must be block size aligned */ |
cec3ebd0 DH |
1011 | if (!size || !IS_ALIGNED(start, memory_block_size_bytes()) || |
1012 | !IS_ALIGNED(size, memory_block_size_bytes())) { | |
ba325585 | 1013 | pr_err("Block size [%#lx] unaligned hotplug range: start %#llx, size %#llx", |
cec3ebd0 | 1014 | memory_block_size_bytes(), start, size); |
27356f54 TK |
1015 | return -EINVAL; |
1016 | } | |
1017 | ||
1018 | return 0; | |
1019 | } | |
1020 | ||
31bc3858 VK |
1021 | static int online_memory_block(struct memory_block *mem, void *arg) |
1022 | { | |
862919e5 | 1023 | mem->online_type = memhp_default_online_type; |
dc18d706 | 1024 | return device_online(&mem->dev); |
31bc3858 VK |
1025 | } |
1026 | ||
8df1d0e4 DH |
1027 | /* |
1028 | * NOTE: The caller must call lock_device_hotplug() to serialize hotplug | |
1029 | * and online/offline operations (triggered e.g. by sysfs). | |
1030 | * | |
1031 | * we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG | |
1032 | */ | |
b6117199 | 1033 | int __ref add_memory_resource(int nid, struct resource *res, mhp_t mhp_flags) |
bc02af93 | 1034 | { |
302cba40 | 1035 | struct mhp_params params = { .pgprot = pgprot_mhp(PAGE_KERNEL) }; |
62cedb9f | 1036 | u64 start, size; |
b9ff0360 | 1037 | bool new_node = false; |
bc02af93 YG |
1038 | int ret; |
1039 | ||
62cedb9f DV |
1040 | start = res->start; |
1041 | size = resource_size(res); | |
1042 | ||
27356f54 TK |
1043 | ret = check_hotplug_memory_range(start, size); |
1044 | if (ret) | |
1045 | return ret; | |
1046 | ||
fa6d9ec7 VV |
1047 | if (!node_possible(nid)) { |
1048 | WARN(1, "node %d was absent from the node_possible_map\n", nid); | |
1049 | return -EINVAL; | |
1050 | } | |
1051 | ||
bfc8c901 | 1052 | mem_hotplug_begin(); |
ac13c462 | 1053 | |
52219aea DH |
1054 | if (IS_ENABLED(CONFIG_ARCH_KEEP_MEMBLOCK)) |
1055 | memblock_add_node(start, size, nid); | |
7f36e3e5 | 1056 | |
c68ab18c | 1057 | ret = __try_online_node(nid, false); |
b9ff0360 OS |
1058 | if (ret < 0) |
1059 | goto error; | |
1060 | new_node = ret; | |
9af3c2de | 1061 | |
bc02af93 | 1062 | /* call arch's memory hotadd */ |
f5637d3b | 1063 | ret = arch_add_memory(nid, start, size, ¶ms); |
9af3c2de YG |
1064 | if (ret < 0) |
1065 | goto error; | |
1066 | ||
db051a0d DH |
1067 | /* create memory block devices after memory was added */ |
1068 | ret = create_memory_block_devices(start, size); | |
1069 | if (ret) { | |
1070 | arch_remove_memory(nid, start, size, NULL); | |
1071 | goto error; | |
1072 | } | |
1073 | ||
a1e565aa | 1074 | if (new_node) { |
d5b6f6a3 | 1075 | /* If sysfs file of new node can't be created, cpu on the node |
0fc44159 YG |
1076 | * can't be hot-added. There is no rollback way now. |
1077 | * So, check by BUG_ON() to catch it reluctantly.. | |
d5b6f6a3 | 1078 | * We online node here. We can't roll back from here. |
0fc44159 | 1079 | */ |
d5b6f6a3 OS |
1080 | node_set_online(nid); |
1081 | ret = __register_one_node(nid); | |
0fc44159 YG |
1082 | BUG_ON(ret); |
1083 | } | |
1084 | ||
d5b6f6a3 | 1085 | /* link memory sections under this node.*/ |
90c7eaeb LD |
1086 | link_mem_sections(nid, PFN_DOWN(start), PFN_UP(start + size - 1), |
1087 | MEMINIT_HOTPLUG); | |
d5b6f6a3 | 1088 | |
d96ae530 | 1089 | /* create new memmap entry */ |
7b7b2721 DH |
1090 | if (!strcmp(res->name, "System RAM")) |
1091 | firmware_map_add_hotplug(start, start + size, "System RAM"); | |
d96ae530 | 1092 | |
381eab4a DH |
1093 | /* device_online() will take the lock when calling online_pages() */ |
1094 | mem_hotplug_done(); | |
1095 | ||
9ca6551e DH |
1096 | /* |
1097 | * In case we're allowed to merge the resource, flag it and trigger | |
1098 | * merging now that adding succeeded. | |
1099 | */ | |
1100 | if (mhp_flags & MEMHP_MERGE_RESOURCE) | |
1101 | merge_system_ram_resource(res); | |
1102 | ||
31bc3858 | 1103 | /* online pages if requested */ |
862919e5 | 1104 | if (memhp_default_online_type != MMOP_OFFLINE) |
fbcf73ce | 1105 | walk_memory_blocks(start, size, NULL, online_memory_block); |
31bc3858 | 1106 | |
381eab4a | 1107 | return ret; |
9af3c2de YG |
1108 | error: |
1109 | /* rollback pgdat allocation and others */ | |
b9ff0360 OS |
1110 | if (new_node) |
1111 | rollback_node_hotadd(nid); | |
52219aea DH |
1112 | if (IS_ENABLED(CONFIG_ARCH_KEEP_MEMBLOCK)) |
1113 | memblock_remove(start, size); | |
bfc8c901 | 1114 | mem_hotplug_done(); |
bc02af93 YG |
1115 | return ret; |
1116 | } | |
62cedb9f | 1117 | |
8df1d0e4 | 1118 | /* requires device_hotplug_lock, see add_memory_resource() */ |
b6117199 | 1119 | int __ref __add_memory(int nid, u64 start, u64 size, mhp_t mhp_flags) |
62cedb9f DV |
1120 | { |
1121 | struct resource *res; | |
1122 | int ret; | |
1123 | ||
7b7b2721 | 1124 | res = register_memory_resource(start, size, "System RAM"); |
6f754ba4 VK |
1125 | if (IS_ERR(res)) |
1126 | return PTR_ERR(res); | |
62cedb9f | 1127 | |
b6117199 | 1128 | ret = add_memory_resource(nid, res, mhp_flags); |
62cedb9f DV |
1129 | if (ret < 0) |
1130 | release_memory_resource(res); | |
1131 | return ret; | |
1132 | } | |
8df1d0e4 | 1133 | |
b6117199 | 1134 | int add_memory(int nid, u64 start, u64 size, mhp_t mhp_flags) |
8df1d0e4 DH |
1135 | { |
1136 | int rc; | |
1137 | ||
1138 | lock_device_hotplug(); | |
b6117199 | 1139 | rc = __add_memory(nid, start, size, mhp_flags); |
8df1d0e4 DH |
1140 | unlock_device_hotplug(); |
1141 | ||
1142 | return rc; | |
1143 | } | |
bc02af93 | 1144 | EXPORT_SYMBOL_GPL(add_memory); |
0c0e6195 | 1145 | |
7b7b2721 DH |
1146 | /* |
1147 | * Add special, driver-managed memory to the system as system RAM. Such | |
1148 | * memory is not exposed via the raw firmware-provided memmap as system | |
1149 | * RAM, instead, it is detected and added by a driver - during cold boot, | |
1150 | * after a reboot, and after kexec. | |
1151 | * | |
1152 | * Reasons why this memory should not be used for the initial memmap of a | |
1153 | * kexec kernel or for placing kexec images: | |
1154 | * - The booting kernel is in charge of determining how this memory will be | |
1155 | * used (e.g., use persistent memory as system RAM) | |
1156 | * - Coordination with a hypervisor is required before this memory | |
1157 | * can be used (e.g., inaccessible parts). | |
1158 | * | |
1159 | * For this memory, no entries in /sys/firmware/memmap ("raw firmware-provided | |
1160 | * memory map") are created. Also, the created memory resource is flagged | |
7cf603d1 | 1161 | * with IORESOURCE_SYSRAM_DRIVER_MANAGED, so in-kernel users can special-case |
7b7b2721 DH |
1162 | * this memory as well (esp., not place kexec images onto it). |
1163 | * | |
1164 | * The resource_name (visible via /proc/iomem) has to have the format | |
1165 | * "System RAM ($DRIVER)". | |
1166 | */ | |
1167 | int add_memory_driver_managed(int nid, u64 start, u64 size, | |
b6117199 | 1168 | const char *resource_name, mhp_t mhp_flags) |
7b7b2721 DH |
1169 | { |
1170 | struct resource *res; | |
1171 | int rc; | |
1172 | ||
1173 | if (!resource_name || | |
1174 | strstr(resource_name, "System RAM (") != resource_name || | |
1175 | resource_name[strlen(resource_name) - 1] != ')') | |
1176 | return -EINVAL; | |
1177 | ||
1178 | lock_device_hotplug(); | |
1179 | ||
1180 | res = register_memory_resource(start, size, resource_name); | |
1181 | if (IS_ERR(res)) { | |
1182 | rc = PTR_ERR(res); | |
1183 | goto out_unlock; | |
1184 | } | |
1185 | ||
b6117199 | 1186 | rc = add_memory_resource(nid, res, mhp_flags); |
7b7b2721 DH |
1187 | if (rc < 0) |
1188 | release_memory_resource(res); | |
1189 | ||
1190 | out_unlock: | |
1191 | unlock_device_hotplug(); | |
1192 | return rc; | |
1193 | } | |
1194 | EXPORT_SYMBOL_GPL(add_memory_driver_managed); | |
1195 | ||
0c0e6195 KH |
1196 | #ifdef CONFIG_MEMORY_HOTREMOVE |
1197 | /* | |
92917998 DH |
1198 | * Confirm all pages in a range [start, end) belong to the same zone (skipping |
1199 | * memory holes). When true, return the zone. | |
0c0e6195 | 1200 | */ |
92917998 DH |
1201 | struct zone *test_pages_in_a_zone(unsigned long start_pfn, |
1202 | unsigned long end_pfn) | |
0c0e6195 | 1203 | { |
5f0f2887 | 1204 | unsigned long pfn, sec_end_pfn; |
0c0e6195 KH |
1205 | struct zone *zone = NULL; |
1206 | struct page *page; | |
1207 | int i; | |
deb88a2a | 1208 | for (pfn = start_pfn, sec_end_pfn = SECTION_ALIGN_UP(start_pfn + 1); |
0c0e6195 | 1209 | pfn < end_pfn; |
deb88a2a | 1210 | pfn = sec_end_pfn, sec_end_pfn += PAGES_PER_SECTION) { |
5f0f2887 AB |
1211 | /* Make sure the memory section is present first */ |
1212 | if (!present_section_nr(pfn_to_section_nr(pfn))) | |
0c0e6195 | 1213 | continue; |
5f0f2887 AB |
1214 | for (; pfn < sec_end_pfn && pfn < end_pfn; |
1215 | pfn += MAX_ORDER_NR_PAGES) { | |
1216 | i = 0; | |
1217 | /* This is just a CONFIG_HOLES_IN_ZONE check.*/ | |
1218 | while ((i < MAX_ORDER_NR_PAGES) && | |
1219 | !pfn_valid_within(pfn + i)) | |
1220 | i++; | |
d6d8c8a4 | 1221 | if (i == MAX_ORDER_NR_PAGES || pfn + i >= end_pfn) |
5f0f2887 | 1222 | continue; |
24feb47c MZ |
1223 | /* Check if we got outside of the zone */ |
1224 | if (zone && !zone_spans_pfn(zone, pfn + i)) | |
92917998 | 1225 | return NULL; |
5f0f2887 AB |
1226 | page = pfn_to_page(pfn + i); |
1227 | if (zone && page_zone(page) != zone) | |
92917998 | 1228 | return NULL; |
5f0f2887 AB |
1229 | zone = page_zone(page); |
1230 | } | |
0c0e6195 | 1231 | } |
deb88a2a | 1232 | |
92917998 | 1233 | return zone; |
0c0e6195 KH |
1234 | } |
1235 | ||
1236 | /* | |
0efadf48 | 1237 | * Scan pfn range [start,end) to find movable/migratable pages (LRU pages, |
aa218795 DH |
1238 | * non-lru movable pages and hugepages). Will skip over most unmovable |
1239 | * pages (esp., pages that can be skipped when offlining), but bail out on | |
1240 | * definitely unmovable pages. | |
1241 | * | |
1242 | * Returns: | |
1243 | * 0 in case a movable page is found and movable_pfn was updated. | |
1244 | * -ENOENT in case no movable page was found. | |
1245 | * -EBUSY in case a definitely unmovable page was found. | |
0c0e6195 | 1246 | */ |
aa218795 DH |
1247 | static int scan_movable_pages(unsigned long start, unsigned long end, |
1248 | unsigned long *movable_pfn) | |
0c0e6195 KH |
1249 | { |
1250 | unsigned long pfn; | |
eeb0efd0 | 1251 | |
0c0e6195 | 1252 | for (pfn = start; pfn < end; pfn++) { |
eeb0efd0 OS |
1253 | struct page *page, *head; |
1254 | unsigned long skip; | |
1255 | ||
1256 | if (!pfn_valid(pfn)) | |
1257 | continue; | |
1258 | page = pfn_to_page(pfn); | |
1259 | if (PageLRU(page)) | |
aa218795 | 1260 | goto found; |
eeb0efd0 | 1261 | if (__PageMovable(page)) |
aa218795 DH |
1262 | goto found; |
1263 | ||
1264 | /* | |
1265 | * PageOffline() pages that are not marked __PageMovable() and | |
1266 | * have a reference count > 0 (after MEM_GOING_OFFLINE) are | |
1267 | * definitely unmovable. If their reference count would be 0, | |
1268 | * they could at least be skipped when offlining memory. | |
1269 | */ | |
1270 | if (PageOffline(page) && page_count(page)) | |
1271 | return -EBUSY; | |
eeb0efd0 OS |
1272 | |
1273 | if (!PageHuge(page)) | |
1274 | continue; | |
1275 | head = compound_head(page); | |
39186cbe | 1276 | if (page_huge_active(head)) |
aa218795 | 1277 | goto found; |
d8c6546b | 1278 | skip = compound_nr(head) - (page - head); |
eeb0efd0 | 1279 | pfn += skip - 1; |
0c0e6195 | 1280 | } |
aa218795 DH |
1281 | return -ENOENT; |
1282 | found: | |
1283 | *movable_pfn = pfn; | |
0c0e6195 KH |
1284 | return 0; |
1285 | } | |
1286 | ||
0c0e6195 KH |
1287 | static int |
1288 | do_migrate_range(unsigned long start_pfn, unsigned long end_pfn) | |
1289 | { | |
1290 | unsigned long pfn; | |
6c357848 | 1291 | struct page *page, *head; |
0c0e6195 KH |
1292 | int ret = 0; |
1293 | LIST_HEAD(source); | |
1294 | ||
a85009c3 | 1295 | for (pfn = start_pfn; pfn < end_pfn; pfn++) { |
0c0e6195 KH |
1296 | if (!pfn_valid(pfn)) |
1297 | continue; | |
1298 | page = pfn_to_page(pfn); | |
6c357848 | 1299 | head = compound_head(page); |
c8721bbb NH |
1300 | |
1301 | if (PageHuge(page)) { | |
d8c6546b | 1302 | pfn = page_to_pfn(head) + compound_nr(head) - 1; |
daf3538a | 1303 | isolate_huge_page(head, &source); |
c8721bbb | 1304 | continue; |
94723aaf | 1305 | } else if (PageTransHuge(page)) |
6c357848 | 1306 | pfn = page_to_pfn(head) + thp_nr_pages(page) - 1; |
c8721bbb | 1307 | |
b15c8726 MH |
1308 | /* |
1309 | * HWPoison pages have elevated reference counts so the migration would | |
1310 | * fail on them. It also doesn't make any sense to migrate them in the | |
1311 | * first place. Still try to unmap such a page in case it is still mapped | |
1312 | * (e.g. current hwpoison implementation doesn't unmap KSM pages but keep | |
1313 | * the unmap as the catch all safety net). | |
1314 | */ | |
1315 | if (PageHWPoison(page)) { | |
1316 | if (WARN_ON(PageLRU(page))) | |
1317 | isolate_lru_page(page); | |
1318 | if (page_mapped(page)) | |
013339df | 1319 | try_to_unmap(page, TTU_IGNORE_MLOCK); |
b15c8726 MH |
1320 | continue; |
1321 | } | |
1322 | ||
700c2a46 | 1323 | if (!get_page_unless_zero(page)) |
0c0e6195 KH |
1324 | continue; |
1325 | /* | |
0efadf48 YX |
1326 | * We can skip free pages. And we can deal with pages on |
1327 | * LRU and non-lru movable pages. | |
0c0e6195 | 1328 | */ |
0efadf48 YX |
1329 | if (PageLRU(page)) |
1330 | ret = isolate_lru_page(page); | |
1331 | else | |
1332 | ret = isolate_movable_page(page, ISOLATE_UNEVICTABLE); | |
0c0e6195 | 1333 | if (!ret) { /* Success */ |
62695a84 | 1334 | list_add_tail(&page->lru, &source); |
0efadf48 YX |
1335 | if (!__PageMovable(page)) |
1336 | inc_node_page_state(page, NR_ISOLATED_ANON + | |
9de4f22a | 1337 | page_is_file_lru(page)); |
6d9c285a | 1338 | |
0c0e6195 | 1339 | } else { |
2932c8b0 | 1340 | pr_warn("failed to isolate pfn %lx\n", pfn); |
0efadf48 | 1341 | dump_page(page, "isolation failed"); |
0c0e6195 | 1342 | } |
1723058e | 1343 | put_page(page); |
0c0e6195 | 1344 | } |
f3ab2636 | 1345 | if (!list_empty(&source)) { |
203e6e5c JK |
1346 | nodemask_t nmask = node_states[N_MEMORY]; |
1347 | struct migration_target_control mtc = { | |
1348 | .nmask = &nmask, | |
1349 | .gfp_mask = GFP_USER | __GFP_MOVABLE | __GFP_RETRY_MAYFAIL, | |
1350 | }; | |
1351 | ||
1352 | /* | |
1353 | * We have checked that migration range is on a single zone so | |
1354 | * we can use the nid of the first page to all the others. | |
1355 | */ | |
1356 | mtc.nid = page_to_nid(list_first_entry(&source, struct page, lru)); | |
1357 | ||
1358 | /* | |
1359 | * try to allocate from a different node but reuse this node | |
1360 | * if there are no other online nodes to be used (e.g. we are | |
1361 | * offlining a part of the only existing node) | |
1362 | */ | |
1363 | node_clear(mtc.nid, nmask); | |
1364 | if (nodes_empty(nmask)) | |
1365 | node_set(mtc.nid, nmask); | |
1366 | ret = migrate_pages(&source, alloc_migration_target, NULL, | |
1367 | (unsigned long)&mtc, MIGRATE_SYNC, MR_MEMORY_HOTPLUG); | |
2932c8b0 MH |
1368 | if (ret) { |
1369 | list_for_each_entry(page, &source, lru) { | |
1370 | pr_warn("migrating pfn %lx failed ret:%d ", | |
1371 | page_to_pfn(page), ret); | |
1372 | dump_page(page, "migration failure"); | |
1373 | } | |
c8721bbb | 1374 | putback_movable_pages(&source); |
2932c8b0 | 1375 | } |
0c0e6195 | 1376 | } |
1723058e | 1377 | |
0c0e6195 KH |
1378 | return ret; |
1379 | } | |
1380 | ||
c5320926 TC |
1381 | static int __init cmdline_parse_movable_node(char *p) |
1382 | { | |
55ac590c | 1383 | movable_node_enabled = true; |
c5320926 TC |
1384 | return 0; |
1385 | } | |
1386 | early_param("movable_node", cmdline_parse_movable_node); | |
1387 | ||
d9713679 LJ |
1388 | /* check which state of node_states will be changed when offline memory */ |
1389 | static void node_states_check_changes_offline(unsigned long nr_pages, | |
1390 | struct zone *zone, struct memory_notify *arg) | |
1391 | { | |
1392 | struct pglist_data *pgdat = zone->zone_pgdat; | |
1393 | unsigned long present_pages = 0; | |
86b27bea | 1394 | enum zone_type zt; |
d9713679 | 1395 | |
98fa15f3 AK |
1396 | arg->status_change_nid = NUMA_NO_NODE; |
1397 | arg->status_change_nid_normal = NUMA_NO_NODE; | |
1398 | arg->status_change_nid_high = NUMA_NO_NODE; | |
d9713679 LJ |
1399 | |
1400 | /* | |
86b27bea OS |
1401 | * Check whether node_states[N_NORMAL_MEMORY] will be changed. |
1402 | * If the memory to be offline is within the range | |
1403 | * [0..ZONE_NORMAL], and it is the last present memory there, | |
1404 | * the zones in that range will become empty after the offlining, | |
1405 | * thus we can determine that we need to clear the node from | |
1406 | * node_states[N_NORMAL_MEMORY]. | |
d9713679 | 1407 | */ |
86b27bea | 1408 | for (zt = 0; zt <= ZONE_NORMAL; zt++) |
d9713679 | 1409 | present_pages += pgdat->node_zones[zt].present_pages; |
86b27bea | 1410 | if (zone_idx(zone) <= ZONE_NORMAL && nr_pages >= present_pages) |
d9713679 | 1411 | arg->status_change_nid_normal = zone_to_nid(zone); |
d9713679 | 1412 | |
6715ddf9 LJ |
1413 | #ifdef CONFIG_HIGHMEM |
1414 | /* | |
86b27bea OS |
1415 | * node_states[N_HIGH_MEMORY] contains nodes which |
1416 | * have normal memory or high memory. | |
1417 | * Here we add the present_pages belonging to ZONE_HIGHMEM. | |
1418 | * If the zone is within the range of [0..ZONE_HIGHMEM), and | |
1419 | * we determine that the zones in that range become empty, | |
1420 | * we need to clear the node for N_HIGH_MEMORY. | |
6715ddf9 | 1421 | */ |
86b27bea OS |
1422 | present_pages += pgdat->node_zones[ZONE_HIGHMEM].present_pages; |
1423 | if (zone_idx(zone) <= ZONE_HIGHMEM && nr_pages >= present_pages) | |
6715ddf9 | 1424 | arg->status_change_nid_high = zone_to_nid(zone); |
6715ddf9 LJ |
1425 | #endif |
1426 | ||
d9713679 | 1427 | /* |
86b27bea OS |
1428 | * We have accounted the pages from [0..ZONE_NORMAL), and |
1429 | * in case of CONFIG_HIGHMEM the pages from ZONE_HIGHMEM | |
1430 | * as well. | |
1431 | * Here we count the possible pages from ZONE_MOVABLE. | |
1432 | * If after having accounted all the pages, we see that the nr_pages | |
1433 | * to be offlined is over or equal to the accounted pages, | |
1434 | * we know that the node will become empty, and so, we can clear | |
1435 | * it for N_MEMORY as well. | |
d9713679 | 1436 | */ |
86b27bea | 1437 | present_pages += pgdat->node_zones[ZONE_MOVABLE].present_pages; |
d9713679 | 1438 | |
d9713679 LJ |
1439 | if (nr_pages >= present_pages) |
1440 | arg->status_change_nid = zone_to_nid(zone); | |
d9713679 LJ |
1441 | } |
1442 | ||
1443 | static void node_states_clear_node(int node, struct memory_notify *arg) | |
1444 | { | |
1445 | if (arg->status_change_nid_normal >= 0) | |
1446 | node_clear_state(node, N_NORMAL_MEMORY); | |
1447 | ||
cf01f6f5 | 1448 | if (arg->status_change_nid_high >= 0) |
d9713679 | 1449 | node_clear_state(node, N_HIGH_MEMORY); |
6715ddf9 | 1450 | |
cf01f6f5 | 1451 | if (arg->status_change_nid >= 0) |
6715ddf9 | 1452 | node_clear_state(node, N_MEMORY); |
d9713679 LJ |
1453 | } |
1454 | ||
c5e79ef5 DH |
1455 | static int count_system_ram_pages_cb(unsigned long start_pfn, |
1456 | unsigned long nr_pages, void *data) | |
1457 | { | |
1458 | unsigned long *nr_system_ram_pages = data; | |
1459 | ||
1460 | *nr_system_ram_pages += nr_pages; | |
1461 | return 0; | |
1462 | } | |
1463 | ||
73a11c96 | 1464 | int __ref offline_pages(unsigned long start_pfn, unsigned long nr_pages) |
0c0e6195 | 1465 | { |
73a11c96 | 1466 | const unsigned long end_pfn = start_pfn + nr_pages; |
0a1a9a00 | 1467 | unsigned long pfn, system_ram_pages = 0; |
d702909f | 1468 | unsigned long flags; |
0c0e6195 | 1469 | struct zone *zone; |
7b78d335 | 1470 | struct memory_notify arg; |
ea15153c | 1471 | int ret, node; |
79605093 | 1472 | char *reason; |
0c0e6195 | 1473 | |
4986fac1 DH |
1474 | /* We can only offline full sections (e.g., SECTION_IS_ONLINE) */ |
1475 | if (WARN_ON_ONCE(!nr_pages || | |
1476 | !IS_ALIGNED(start_pfn | nr_pages, PAGES_PER_SECTION))) | |
1477 | return -EINVAL; | |
1478 | ||
381eab4a DH |
1479 | mem_hotplug_begin(); |
1480 | ||
c5e79ef5 DH |
1481 | /* |
1482 | * Don't allow to offline memory blocks that contain holes. | |
1483 | * Consequently, memory blocks with holes can never get onlined | |
1484 | * via the hotplug path - online_pages() - as hotplugged memory has | |
1485 | * no holes. This way, we e.g., don't have to worry about marking | |
1486 | * memory holes PG_reserved, don't need pfn_valid() checks, and can | |
1487 | * avoid using walk_system_ram_range() later. | |
1488 | */ | |
73a11c96 | 1489 | walk_system_ram_range(start_pfn, nr_pages, &system_ram_pages, |
c5e79ef5 | 1490 | count_system_ram_pages_cb); |
73a11c96 | 1491 | if (system_ram_pages != nr_pages) { |
c5e79ef5 DH |
1492 | ret = -EINVAL; |
1493 | reason = "memory holes"; | |
1494 | goto failed_removal; | |
1495 | } | |
1496 | ||
0c0e6195 KH |
1497 | /* This makes hotplug much easier...and readable. |
1498 | we assume this for now. .*/ | |
92917998 DH |
1499 | zone = test_pages_in_a_zone(start_pfn, end_pfn); |
1500 | if (!zone) { | |
79605093 MH |
1501 | ret = -EINVAL; |
1502 | reason = "multizone range"; | |
1503 | goto failed_removal; | |
381eab4a | 1504 | } |
7b78d335 | 1505 | node = zone_to_nid(zone); |
7b78d335 | 1506 | |
ec6e8c7e VB |
1507 | /* |
1508 | * Disable pcplists so that page isolation cannot race with freeing | |
1509 | * in a way that pages from isolated pageblock are left on pcplists. | |
1510 | */ | |
1511 | zone_pcp_disable(zone); | |
1512 | ||
0c0e6195 | 1513 | /* set above range as isolated */ |
b023f468 | 1514 | ret = start_isolate_page_range(start_pfn, end_pfn, |
d381c547 | 1515 | MIGRATE_MOVABLE, |
756d25be | 1516 | MEMORY_OFFLINE | REPORT_FAILURE); |
3fa0c7c7 | 1517 | if (ret) { |
79605093 | 1518 | reason = "failure to isolate range"; |
ec6e8c7e | 1519 | goto failed_removal_pcplists_disabled; |
381eab4a | 1520 | } |
7b78d335 YG |
1521 | |
1522 | arg.start_pfn = start_pfn; | |
1523 | arg.nr_pages = nr_pages; | |
d9713679 | 1524 | node_states_check_changes_offline(nr_pages, zone, &arg); |
7b78d335 YG |
1525 | |
1526 | ret = memory_notify(MEM_GOING_OFFLINE, &arg); | |
1527 | ret = notifier_to_errno(ret); | |
79605093 MH |
1528 | if (ret) { |
1529 | reason = "notifier failure"; | |
1530 | goto failed_removal_isolated; | |
1531 | } | |
7b78d335 | 1532 | |
bb8965bd | 1533 | do { |
aa218795 DH |
1534 | pfn = start_pfn; |
1535 | do { | |
bb8965bd MH |
1536 | if (signal_pending(current)) { |
1537 | ret = -EINTR; | |
1538 | reason = "signal backoff"; | |
1539 | goto failed_removal_isolated; | |
1540 | } | |
72b39cfc | 1541 | |
bb8965bd MH |
1542 | cond_resched(); |
1543 | lru_add_drain_all(); | |
bb8965bd | 1544 | |
aa218795 DH |
1545 | ret = scan_movable_pages(pfn, end_pfn, &pfn); |
1546 | if (!ret) { | |
bb8965bd MH |
1547 | /* |
1548 | * TODO: fatal migration failures should bail | |
1549 | * out | |
1550 | */ | |
1551 | do_migrate_range(pfn, end_pfn); | |
1552 | } | |
aa218795 DH |
1553 | } while (!ret); |
1554 | ||
1555 | if (ret != -ENOENT) { | |
1556 | reason = "unmovable page"; | |
1557 | goto failed_removal_isolated; | |
bb8965bd | 1558 | } |
0c0e6195 | 1559 | |
bb8965bd MH |
1560 | /* |
1561 | * Dissolve free hugepages in the memory block before doing | |
1562 | * offlining actually in order to make hugetlbfs's object | |
1563 | * counting consistent. | |
1564 | */ | |
1565 | ret = dissolve_free_huge_pages(start_pfn, end_pfn); | |
1566 | if (ret) { | |
1567 | reason = "failure to dissolve huge pages"; | |
1568 | goto failed_removal_isolated; | |
1569 | } | |
0a1a9a00 | 1570 | |
0a1a9a00 | 1571 | ret = test_pages_isolated(start_pfn, end_pfn, MEMORY_OFFLINE); |
ec6e8c7e | 1572 | |
5557c766 | 1573 | } while (ret); |
72b39cfc | 1574 | |
0a1a9a00 DH |
1575 | /* Mark all sections offline and remove free pages from the buddy. */ |
1576 | __offline_isolated_pages(start_pfn, end_pfn); | |
7c33023a | 1577 | pr_debug("Offlined Pages %ld\n", nr_pages); |
0a1a9a00 | 1578 | |
9b7ea46a | 1579 | /* |
b30c5927 DH |
1580 | * The memory sections are marked offline, and the pageblock flags |
1581 | * effectively stale; nobody should be touching them. Fixup the number | |
1582 | * of isolated pageblocks, memory onlining will properly revert this. | |
9b7ea46a QC |
1583 | */ |
1584 | spin_lock_irqsave(&zone->lock, flags); | |
ea15153c | 1585 | zone->nr_isolate_pageblock -= nr_pages / pageblock_nr_pages; |
9b7ea46a QC |
1586 | spin_unlock_irqrestore(&zone->lock, flags); |
1587 | ||
ec6e8c7e VB |
1588 | zone_pcp_enable(zone); |
1589 | ||
0c0e6195 | 1590 | /* removal success */ |
0a1a9a00 DH |
1591 | adjust_managed_page_count(pfn_to_page(start_pfn), -nr_pages); |
1592 | zone->present_pages -= nr_pages; | |
d702909f CS |
1593 | |
1594 | pgdat_resize_lock(zone->zone_pgdat, &flags); | |
0a1a9a00 | 1595 | zone->zone_pgdat->node_present_pages -= nr_pages; |
d702909f | 1596 | pgdat_resize_unlock(zone->zone_pgdat, &flags); |
7b78d335 | 1597 | |
1b79acc9 KM |
1598 | init_per_zone_wmark_min(); |
1599 | ||
1e8537ba | 1600 | if (!populated_zone(zone)) { |
340175b7 | 1601 | zone_pcp_reset(zone); |
72675e13 | 1602 | build_all_zonelists(NULL); |
1e8537ba XQ |
1603 | } else |
1604 | zone_pcp_update(zone); | |
340175b7 | 1605 | |
d9713679 | 1606 | node_states_clear_node(node, &arg); |
698b1b30 | 1607 | if (arg.status_change_nid >= 0) { |
8fe23e05 | 1608 | kswapd_stop(node); |
698b1b30 VB |
1609 | kcompactd_stop(node); |
1610 | } | |
bce7394a | 1611 | |
0c0e6195 | 1612 | writeback_set_ratelimit(); |
7b78d335 YG |
1613 | |
1614 | memory_notify(MEM_OFFLINE, &arg); | |
feee6b29 | 1615 | remove_pfn_range_from_zone(zone, start_pfn, nr_pages); |
381eab4a | 1616 | mem_hotplug_done(); |
0c0e6195 KH |
1617 | return 0; |
1618 | ||
79605093 MH |
1619 | failed_removal_isolated: |
1620 | undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE); | |
c4efe484 | 1621 | memory_notify(MEM_CANCEL_OFFLINE, &arg); |
ec6e8c7e VB |
1622 | failed_removal_pcplists_disabled: |
1623 | zone_pcp_enable(zone); | |
0c0e6195 | 1624 | failed_removal: |
79605093 | 1625 | pr_debug("memory offlining [mem %#010llx-%#010llx] failed due to %s\n", |
e33e33b4 | 1626 | (unsigned long long) start_pfn << PAGE_SHIFT, |
79605093 MH |
1627 | ((unsigned long long) end_pfn << PAGE_SHIFT) - 1, |
1628 | reason); | |
0c0e6195 | 1629 | /* pushback to free area */ |
381eab4a | 1630 | mem_hotplug_done(); |
0c0e6195 KH |
1631 | return ret; |
1632 | } | |
71088785 | 1633 | |
d6de9d53 | 1634 | static int check_memblock_offlined_cb(struct memory_block *mem, void *arg) |
bbc76be6 WC |
1635 | { |
1636 | int ret = !is_memblock_offlined(mem); | |
1637 | ||
349daa0f RD |
1638 | if (unlikely(ret)) { |
1639 | phys_addr_t beginpa, endpa; | |
1640 | ||
1641 | beginpa = PFN_PHYS(section_nr_to_pfn(mem->start_section_nr)); | |
b6c88d3b | 1642 | endpa = beginpa + memory_block_size_bytes() - 1; |
756a025f | 1643 | pr_warn("removing memory fails, because memory [%pa-%pa] is onlined\n", |
349daa0f | 1644 | &beginpa, &endpa); |
bbc76be6 | 1645 | |
eca499ab PT |
1646 | return -EBUSY; |
1647 | } | |
1648 | return 0; | |
bbc76be6 WC |
1649 | } |
1650 | ||
0f1cfe9d | 1651 | static int check_cpu_on_node(pg_data_t *pgdat) |
60a5a19e | 1652 | { |
60a5a19e TC |
1653 | int cpu; |
1654 | ||
1655 | for_each_present_cpu(cpu) { | |
1656 | if (cpu_to_node(cpu) == pgdat->node_id) | |
1657 | /* | |
1658 | * the cpu on this node isn't removed, and we can't | |
1659 | * offline this node. | |
1660 | */ | |
1661 | return -EBUSY; | |
1662 | } | |
1663 | ||
1664 | return 0; | |
1665 | } | |
1666 | ||
2c91f8fc DH |
1667 | static int check_no_memblock_for_node_cb(struct memory_block *mem, void *arg) |
1668 | { | |
1669 | int nid = *(int *)arg; | |
1670 | ||
1671 | /* | |
1672 | * If a memory block belongs to multiple nodes, the stored nid is not | |
1673 | * reliable. However, such blocks are always online (e.g., cannot get | |
1674 | * offlined) and, therefore, are still spanned by the node. | |
1675 | */ | |
1676 | return mem->nid == nid ? -EEXIST : 0; | |
1677 | } | |
1678 | ||
0f1cfe9d TK |
1679 | /** |
1680 | * try_offline_node | |
e8b098fc | 1681 | * @nid: the node ID |
0f1cfe9d TK |
1682 | * |
1683 | * Offline a node if all memory sections and cpus of the node are removed. | |
1684 | * | |
1685 | * NOTE: The caller must call lock_device_hotplug() to serialize hotplug | |
1686 | * and online/offline operations before this call. | |
1687 | */ | |
90b30cdc | 1688 | void try_offline_node(int nid) |
60a5a19e | 1689 | { |
d822b86a | 1690 | pg_data_t *pgdat = NODE_DATA(nid); |
2c91f8fc | 1691 | int rc; |
60a5a19e | 1692 | |
2c91f8fc DH |
1693 | /* |
1694 | * If the node still spans pages (especially ZONE_DEVICE), don't | |
1695 | * offline it. A node spans memory after move_pfn_range_to_zone(), | |
1696 | * e.g., after the memory block was onlined. | |
1697 | */ | |
1698 | if (pgdat->node_spanned_pages) | |
1699 | return; | |
60a5a19e | 1700 | |
2c91f8fc DH |
1701 | /* |
1702 | * Especially offline memory blocks might not be spanned by the | |
1703 | * node. They will get spanned by the node once they get onlined. | |
1704 | * However, they link to the node in sysfs and can get onlined later. | |
1705 | */ | |
1706 | rc = for_each_memory_block(&nid, check_no_memblock_for_node_cb); | |
1707 | if (rc) | |
60a5a19e | 1708 | return; |
60a5a19e | 1709 | |
46a3679b | 1710 | if (check_cpu_on_node(pgdat)) |
60a5a19e TC |
1711 | return; |
1712 | ||
1713 | /* | |
1714 | * all memory/cpu of this node are removed, we can offline this | |
1715 | * node now. | |
1716 | */ | |
1717 | node_set_offline(nid); | |
1718 | unregister_one_node(nid); | |
1719 | } | |
90b30cdc | 1720 | EXPORT_SYMBOL(try_offline_node); |
60a5a19e | 1721 | |
eca499ab | 1722 | static int __ref try_remove_memory(int nid, u64 start, u64 size) |
bbc76be6 | 1723 | { |
eca499ab | 1724 | int rc = 0; |
993c1aad | 1725 | |
27356f54 TK |
1726 | BUG_ON(check_hotplug_memory_range(start, size)); |
1727 | ||
6677e3ea | 1728 | /* |
242831eb | 1729 | * All memory blocks must be offlined before removing memory. Check |
eca499ab | 1730 | * whether all memory blocks in question are offline and return error |
242831eb | 1731 | * if this is not the case. |
6677e3ea | 1732 | */ |
fbcf73ce | 1733 | rc = walk_memory_blocks(start, size, NULL, check_memblock_offlined_cb); |
eca499ab | 1734 | if (rc) |
b4223a51 | 1735 | return rc; |
6677e3ea | 1736 | |
46c66c4b YI |
1737 | /* remove memmap entry */ |
1738 | firmware_map_remove(start, start + size, "System RAM"); | |
4c4b7f9b | 1739 | |
f1037ec0 DW |
1740 | /* |
1741 | * Memory block device removal under the device_hotplug_lock is | |
1742 | * a barrier against racing online attempts. | |
1743 | */ | |
4c4b7f9b | 1744 | remove_memory_block_devices(start, size); |
46c66c4b | 1745 | |
f1037ec0 DW |
1746 | mem_hotplug_begin(); |
1747 | ||
2c2a5af6 | 1748 | arch_remove_memory(nid, start, size, NULL); |
52219aea DH |
1749 | |
1750 | if (IS_ENABLED(CONFIG_ARCH_KEEP_MEMBLOCK)) { | |
1751 | memblock_free(start, size); | |
1752 | memblock_remove(start, size); | |
1753 | } | |
1754 | ||
cb8e3c8b | 1755 | release_mem_region_adjustable(start, size); |
24d335ca | 1756 | |
60a5a19e TC |
1757 | try_offline_node(nid); |
1758 | ||
bfc8c901 | 1759 | mem_hotplug_done(); |
b4223a51 | 1760 | return 0; |
71088785 | 1761 | } |
d15e5926 | 1762 | |
eca499ab PT |
1763 | /** |
1764 | * remove_memory | |
1765 | * @nid: the node ID | |
1766 | * @start: physical address of the region to remove | |
1767 | * @size: size of the region to remove | |
1768 | * | |
1769 | * NOTE: The caller must call lock_device_hotplug() to serialize hotplug | |
1770 | * and online/offline operations before this call, as required by | |
1771 | * try_offline_node(). | |
1772 | */ | |
1773 | void __remove_memory(int nid, u64 start, u64 size) | |
1774 | { | |
1775 | ||
1776 | /* | |
29a90db9 | 1777 | * trigger BUG() if some memory is not offlined prior to calling this |
eca499ab PT |
1778 | * function |
1779 | */ | |
1780 | if (try_remove_memory(nid, start, size)) | |
1781 | BUG(); | |
1782 | } | |
1783 | ||
1784 | /* | |
1785 | * Remove memory if every memory block is offline, otherwise return -EBUSY is | |
1786 | * some memory is not offline | |
1787 | */ | |
1788 | int remove_memory(int nid, u64 start, u64 size) | |
d15e5926 | 1789 | { |
eca499ab PT |
1790 | int rc; |
1791 | ||
d15e5926 | 1792 | lock_device_hotplug(); |
eca499ab | 1793 | rc = try_remove_memory(nid, start, size); |
d15e5926 | 1794 | unlock_device_hotplug(); |
eca499ab PT |
1795 | |
1796 | return rc; | |
d15e5926 | 1797 | } |
71088785 | 1798 | EXPORT_SYMBOL_GPL(remove_memory); |
08b3acd7 | 1799 | |
8dc4bb58 DH |
1800 | static int try_offline_memory_block(struct memory_block *mem, void *arg) |
1801 | { | |
1802 | uint8_t online_type = MMOP_ONLINE_KERNEL; | |
1803 | uint8_t **online_types = arg; | |
1804 | struct page *page; | |
1805 | int rc; | |
1806 | ||
1807 | /* | |
1808 | * Sense the online_type via the zone of the memory block. Offlining | |
1809 | * with multiple zones within one memory block will be rejected | |
1810 | * by offlining code ... so we don't care about that. | |
1811 | */ | |
1812 | page = pfn_to_online_page(section_nr_to_pfn(mem->start_section_nr)); | |
1813 | if (page && zone_idx(page_zone(page)) == ZONE_MOVABLE) | |
1814 | online_type = MMOP_ONLINE_MOVABLE; | |
1815 | ||
1816 | rc = device_offline(&mem->dev); | |
1817 | /* | |
1818 | * Default is MMOP_OFFLINE - change it only if offlining succeeded, | |
1819 | * so try_reonline_memory_block() can do the right thing. | |
1820 | */ | |
1821 | if (!rc) | |
1822 | **online_types = online_type; | |
1823 | ||
1824 | (*online_types)++; | |
1825 | /* Ignore if already offline. */ | |
1826 | return rc < 0 ? rc : 0; | |
1827 | } | |
1828 | ||
1829 | static int try_reonline_memory_block(struct memory_block *mem, void *arg) | |
1830 | { | |
1831 | uint8_t **online_types = arg; | |
1832 | int rc; | |
1833 | ||
1834 | if (**online_types != MMOP_OFFLINE) { | |
1835 | mem->online_type = **online_types; | |
1836 | rc = device_online(&mem->dev); | |
1837 | if (rc < 0) | |
1838 | pr_warn("%s: Failed to re-online memory: %d", | |
1839 | __func__, rc); | |
1840 | } | |
1841 | ||
1842 | /* Continue processing all remaining memory blocks. */ | |
1843 | (*online_types)++; | |
1844 | return 0; | |
1845 | } | |
1846 | ||
08b3acd7 | 1847 | /* |
8dc4bb58 DH |
1848 | * Try to offline and remove memory. Might take a long time to finish in case |
1849 | * memory is still in use. Primarily useful for memory devices that logically | |
1850 | * unplugged all memory (so it's no longer in use) and want to offline + remove | |
1851 | * that memory. | |
08b3acd7 DH |
1852 | */ |
1853 | int offline_and_remove_memory(int nid, u64 start, u64 size) | |
1854 | { | |
8dc4bb58 DH |
1855 | const unsigned long mb_count = size / memory_block_size_bytes(); |
1856 | uint8_t *online_types, *tmp; | |
1857 | int rc; | |
08b3acd7 DH |
1858 | |
1859 | if (!IS_ALIGNED(start, memory_block_size_bytes()) || | |
8dc4bb58 DH |
1860 | !IS_ALIGNED(size, memory_block_size_bytes()) || !size) |
1861 | return -EINVAL; | |
1862 | ||
1863 | /* | |
1864 | * We'll remember the old online type of each memory block, so we can | |
1865 | * try to revert whatever we did when offlining one memory block fails | |
1866 | * after offlining some others succeeded. | |
1867 | */ | |
1868 | online_types = kmalloc_array(mb_count, sizeof(*online_types), | |
1869 | GFP_KERNEL); | |
1870 | if (!online_types) | |
1871 | return -ENOMEM; | |
1872 | /* | |
1873 | * Initialize all states to MMOP_OFFLINE, so when we abort processing in | |
1874 | * try_offline_memory_block(), we'll skip all unprocessed blocks in | |
1875 | * try_reonline_memory_block(). | |
1876 | */ | |
1877 | memset(online_types, MMOP_OFFLINE, mb_count); | |
08b3acd7 DH |
1878 | |
1879 | lock_device_hotplug(); | |
8dc4bb58 DH |
1880 | |
1881 | tmp = online_types; | |
1882 | rc = walk_memory_blocks(start, size, &tmp, try_offline_memory_block); | |
08b3acd7 DH |
1883 | |
1884 | /* | |
8dc4bb58 | 1885 | * In case we succeeded to offline all memory, remove it. |
08b3acd7 DH |
1886 | * This cannot fail as it cannot get onlined in the meantime. |
1887 | */ | |
1888 | if (!rc) { | |
1889 | rc = try_remove_memory(nid, start, size); | |
8dc4bb58 DH |
1890 | if (rc) |
1891 | pr_err("%s: Failed to remove memory: %d", __func__, rc); | |
1892 | } | |
1893 | ||
1894 | /* | |
1895 | * Rollback what we did. While memory onlining might theoretically fail | |
1896 | * (nacked by a notifier), it barely ever happens. | |
1897 | */ | |
1898 | if (rc) { | |
1899 | tmp = online_types; | |
1900 | walk_memory_blocks(start, size, &tmp, | |
1901 | try_reonline_memory_block); | |
08b3acd7 DH |
1902 | } |
1903 | unlock_device_hotplug(); | |
1904 | ||
8dc4bb58 | 1905 | kfree(online_types); |
08b3acd7 DH |
1906 | return rc; |
1907 | } | |
1908 | EXPORT_SYMBOL_GPL(offline_and_remove_memory); | |
aba6efc4 | 1909 | #endif /* CONFIG_MEMORY_HOTREMOVE */ |