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 | |
3f906ba2 TG |
57 | void get_online_mems(void) |
58 | { | |
59 | percpu_down_read(&mem_hotplug_lock); | |
60 | } | |
bfc8c901 | 61 | |
3f906ba2 TG |
62 | void put_online_mems(void) |
63 | { | |
64 | percpu_up_read(&mem_hotplug_lock); | |
65 | } | |
bfc8c901 | 66 | |
4932381e MH |
67 | bool movable_node_enabled = false; |
68 | ||
8604d9e5 | 69 | #ifndef CONFIG_MEMORY_HOTPLUG_DEFAULT_ONLINE |
862919e5 | 70 | int memhp_default_online_type = MMOP_OFFLINE; |
8604d9e5 | 71 | #else |
862919e5 | 72 | int memhp_default_online_type = MMOP_ONLINE; |
8604d9e5 | 73 | #endif |
31bc3858 | 74 | |
86dd995d VK |
75 | static int __init setup_memhp_default_state(char *str) |
76 | { | |
5f47adf7 DH |
77 | const int online_type = memhp_online_type_from_str(str); |
78 | ||
79 | if (online_type >= 0) | |
80 | memhp_default_online_type = online_type; | |
86dd995d VK |
81 | |
82 | return 1; | |
83 | } | |
84 | __setup("memhp_default_state=", setup_memhp_default_state); | |
85 | ||
30467e0b | 86 | void mem_hotplug_begin(void) |
20d6c96b | 87 | { |
3f906ba2 TG |
88 | cpus_read_lock(); |
89 | percpu_down_write(&mem_hotplug_lock); | |
20d6c96b KM |
90 | } |
91 | ||
30467e0b | 92 | void mem_hotplug_done(void) |
bfc8c901 | 93 | { |
3f906ba2 TG |
94 | percpu_up_write(&mem_hotplug_lock); |
95 | cpus_read_unlock(); | |
bfc8c901 | 96 | } |
20d6c96b | 97 | |
357b4da5 JG |
98 | u64 max_mem_size = U64_MAX; |
99 | ||
45e0b78b | 100 | /* add this memory to iomem resource */ |
7b7b2721 DH |
101 | static struct resource *register_memory_resource(u64 start, u64 size, |
102 | const char *resource_name) | |
45e0b78b | 103 | { |
2794129e DH |
104 | struct resource *res; |
105 | unsigned long flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY; | |
7b7b2721 DH |
106 | |
107 | if (strcmp(resource_name, "System RAM")) | |
108 | flags |= IORESOURCE_MEM_DRIVER_MANAGED; | |
357b4da5 | 109 | |
f3cd4c86 BH |
110 | /* |
111 | * Make sure value parsed from 'mem=' only restricts memory adding | |
112 | * while booting, so that memory hotplug won't be impacted. Please | |
113 | * refer to document of 'mem=' in kernel-parameters.txt for more | |
114 | * details. | |
115 | */ | |
116 | if (start + size > max_mem_size && system_state < SYSTEM_RUNNING) | |
357b4da5 JG |
117 | return ERR_PTR(-E2BIG); |
118 | ||
2794129e DH |
119 | /* |
120 | * Request ownership of the new memory range. This might be | |
121 | * a child of an existing resource that was present but | |
122 | * not marked as busy. | |
123 | */ | |
124 | res = __request_region(&iomem_resource, start, size, | |
125 | resource_name, flags); | |
126 | ||
127 | if (!res) { | |
128 | pr_debug("Unable to reserve System RAM region: %016llx->%016llx\n", | |
129 | start, start + size); | |
6f754ba4 | 130 | return ERR_PTR(-EEXIST); |
45e0b78b KM |
131 | } |
132 | return res; | |
133 | } | |
134 | ||
135 | static void release_memory_resource(struct resource *res) | |
136 | { | |
137 | if (!res) | |
138 | return; | |
139 | release_resource(res); | |
140 | kfree(res); | |
45e0b78b KM |
141 | } |
142 | ||
53947027 | 143 | #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE |
46723bfa YI |
144 | void get_page_bootmem(unsigned long info, struct page *page, |
145 | unsigned long type) | |
04753278 | 146 | { |
ddffe98d | 147 | page->freelist = (void *)type; |
04753278 YG |
148 | SetPagePrivate(page); |
149 | set_page_private(page, info); | |
fe896d18 | 150 | page_ref_inc(page); |
04753278 YG |
151 | } |
152 | ||
170a5a7e | 153 | void put_page_bootmem(struct page *page) |
04753278 | 154 | { |
5f24ce5f | 155 | unsigned long type; |
04753278 | 156 | |
ddffe98d | 157 | type = (unsigned long) page->freelist; |
5f24ce5f AA |
158 | BUG_ON(type < MEMORY_HOTPLUG_MIN_BOOTMEM_TYPE || |
159 | type > MEMORY_HOTPLUG_MAX_BOOTMEM_TYPE); | |
04753278 | 160 | |
fe896d18 | 161 | if (page_ref_dec_return(page) == 1) { |
ddffe98d | 162 | page->freelist = NULL; |
04753278 YG |
163 | ClearPagePrivate(page); |
164 | set_page_private(page, 0); | |
5f24ce5f | 165 | INIT_LIST_HEAD(&page->lru); |
170a5a7e | 166 | free_reserved_page(page); |
04753278 | 167 | } |
04753278 YG |
168 | } |
169 | ||
46723bfa YI |
170 | #ifdef CONFIG_HAVE_BOOTMEM_INFO_NODE |
171 | #ifndef CONFIG_SPARSEMEM_VMEMMAP | |
d92bc318 | 172 | static void register_page_bootmem_info_section(unsigned long start_pfn) |
04753278 | 173 | { |
f1eca35a | 174 | unsigned long mapsize, section_nr, i; |
04753278 YG |
175 | struct mem_section *ms; |
176 | struct page *page, *memmap; | |
f1eca35a | 177 | struct mem_section_usage *usage; |
04753278 | 178 | |
04753278 YG |
179 | section_nr = pfn_to_section_nr(start_pfn); |
180 | ms = __nr_to_section(section_nr); | |
181 | ||
182 | /* Get section's memmap address */ | |
183 | memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr); | |
184 | ||
185 | /* | |
186 | * Get page for the memmap's phys address | |
187 | * XXX: need more consideration for sparse_vmemmap... | |
188 | */ | |
189 | page = virt_to_page(memmap); | |
190 | mapsize = sizeof(struct page) * PAGES_PER_SECTION; | |
191 | mapsize = PAGE_ALIGN(mapsize) >> PAGE_SHIFT; | |
192 | ||
193 | /* remember memmap's page */ | |
194 | for (i = 0; i < mapsize; i++, page++) | |
195 | get_page_bootmem(section_nr, page, SECTION_INFO); | |
196 | ||
f1eca35a DW |
197 | usage = ms->usage; |
198 | page = virt_to_page(usage); | |
04753278 | 199 | |
f1eca35a | 200 | mapsize = PAGE_ALIGN(mem_section_usage_size()) >> PAGE_SHIFT; |
04753278 YG |
201 | |
202 | for (i = 0; i < mapsize; i++, page++) | |
af370fb8 | 203 | get_page_bootmem(section_nr, page, MIX_SECTION_INFO); |
04753278 YG |
204 | |
205 | } | |
46723bfa YI |
206 | #else /* CONFIG_SPARSEMEM_VMEMMAP */ |
207 | static void register_page_bootmem_info_section(unsigned long start_pfn) | |
208 | { | |
f1eca35a | 209 | unsigned long mapsize, section_nr, i; |
46723bfa YI |
210 | struct mem_section *ms; |
211 | struct page *page, *memmap; | |
f1eca35a | 212 | struct mem_section_usage *usage; |
46723bfa | 213 | |
46723bfa YI |
214 | section_nr = pfn_to_section_nr(start_pfn); |
215 | ms = __nr_to_section(section_nr); | |
216 | ||
217 | memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr); | |
218 | ||
219 | register_page_bootmem_memmap(section_nr, memmap, PAGES_PER_SECTION); | |
220 | ||
f1eca35a DW |
221 | usage = ms->usage; |
222 | page = virt_to_page(usage); | |
46723bfa | 223 | |
f1eca35a | 224 | mapsize = PAGE_ALIGN(mem_section_usage_size()) >> PAGE_SHIFT; |
46723bfa YI |
225 | |
226 | for (i = 0; i < mapsize; i++, page++) | |
227 | get_page_bootmem(section_nr, page, MIX_SECTION_INFO); | |
228 | } | |
229 | #endif /* !CONFIG_SPARSEMEM_VMEMMAP */ | |
04753278 | 230 | |
7ded384a | 231 | void __init register_page_bootmem_info_node(struct pglist_data *pgdat) |
04753278 YG |
232 | { |
233 | unsigned long i, pfn, end_pfn, nr_pages; | |
234 | int node = pgdat->node_id; | |
235 | struct page *page; | |
04753278 YG |
236 | |
237 | nr_pages = PAGE_ALIGN(sizeof(struct pglist_data)) >> PAGE_SHIFT; | |
238 | page = virt_to_page(pgdat); | |
239 | ||
240 | for (i = 0; i < nr_pages; i++, page++) | |
241 | get_page_bootmem(node, page, NODE_INFO); | |
242 | ||
04753278 | 243 | pfn = pgdat->node_start_pfn; |
c1f19495 | 244 | end_pfn = pgdat_end_pfn(pgdat); |
04753278 | 245 | |
7e9f5eb0 | 246 | /* register section info */ |
f14851af | 247 | for (; pfn < end_pfn; pfn += PAGES_PER_SECTION) { |
248 | /* | |
249 | * Some platforms can assign the same pfn to multiple nodes - on | |
250 | * node0 as well as nodeN. To avoid registering a pfn against | |
251 | * multiple nodes we check that this pfn does not already | |
7e9f5eb0 | 252 | * reside in some other nodes. |
f14851af | 253 | */ |
f65e91df | 254 | if (pfn_valid(pfn) && (early_pfn_to_nid(pfn) == node)) |
f14851af | 255 | register_page_bootmem_info_section(pfn); |
256 | } | |
04753278 | 257 | } |
46723bfa | 258 | #endif /* CONFIG_HAVE_BOOTMEM_INFO_NODE */ |
04753278 | 259 | |
7ea62160 DW |
260 | static int check_pfn_span(unsigned long pfn, unsigned long nr_pages, |
261 | const char *reason) | |
262 | { | |
263 | /* | |
264 | * Disallow all operations smaller than a sub-section and only | |
265 | * allow operations smaller than a section for | |
266 | * SPARSEMEM_VMEMMAP. Note that check_hotplug_memory_range() | |
267 | * enforces a larger memory_block_size_bytes() granularity for | |
268 | * memory that will be marked online, so this check should only | |
269 | * fire for direct arch_{add,remove}_memory() users outside of | |
270 | * add_memory_resource(). | |
271 | */ | |
272 | unsigned long min_align; | |
273 | ||
274 | if (IS_ENABLED(CONFIG_SPARSEMEM_VMEMMAP)) | |
275 | min_align = PAGES_PER_SUBSECTION; | |
276 | else | |
277 | min_align = PAGES_PER_SECTION; | |
278 | if (!IS_ALIGNED(pfn, min_align) | |
279 | || !IS_ALIGNED(nr_pages, min_align)) { | |
280 | WARN(1, "Misaligned __%s_pages start: %#lx end: #%lx\n", | |
281 | reason, pfn, pfn + nr_pages - 1); | |
282 | return -EINVAL; | |
283 | } | |
284 | return 0; | |
285 | } | |
286 | ||
dca4436d AS |
287 | static int check_hotplug_memory_addressable(unsigned long pfn, |
288 | unsigned long nr_pages) | |
289 | { | |
290 | const u64 max_addr = PFN_PHYS(pfn + nr_pages) - 1; | |
291 | ||
292 | if (max_addr >> MAX_PHYSMEM_BITS) { | |
293 | const u64 max_allowed = (1ull << (MAX_PHYSMEM_BITS + 1)) - 1; | |
294 | WARN(1, | |
295 | "Hotplugged memory exceeds maximum addressable address, range=%#llx-%#llx, maximum=%#llx\n", | |
296 | (u64)PFN_PHYS(pfn), max_addr, max_allowed); | |
297 | return -E2BIG; | |
298 | } | |
299 | ||
300 | return 0; | |
301 | } | |
302 | ||
4edd7cef DR |
303 | /* |
304 | * Reasonably generic function for adding memory. It is | |
305 | * expected that archs that support memory hotplug will | |
306 | * call this function after deciding the zone to which to | |
307 | * add the new pages. | |
308 | */ | |
7ea62160 | 309 | int __ref __add_pages(int nid, unsigned long pfn, unsigned long nr_pages, |
f5637d3b | 310 | struct mhp_params *params) |
4edd7cef | 311 | { |
6cdd0b30 DH |
312 | const unsigned long end_pfn = pfn + nr_pages; |
313 | unsigned long cur_nr_pages; | |
9a845030 | 314 | int err; |
f5637d3b | 315 | struct vmem_altmap *altmap = params->altmap; |
4b94ffdc | 316 | |
bfeb022f LG |
317 | if (WARN_ON_ONCE(!params->pgprot.pgprot)) |
318 | return -EINVAL; | |
319 | ||
dca4436d AS |
320 | err = check_hotplug_memory_addressable(pfn, nr_pages); |
321 | if (err) | |
322 | return err; | |
323 | ||
4b94ffdc DW |
324 | if (altmap) { |
325 | /* | |
326 | * Validate altmap is within bounds of the total request | |
327 | */ | |
7ea62160 | 328 | if (altmap->base_pfn != pfn |
4b94ffdc DW |
329 | || vmem_altmap_offset(altmap) > nr_pages) { |
330 | pr_warn_once("memory add fail, invalid altmap\n"); | |
7ea62160 | 331 | return -EINVAL; |
4b94ffdc DW |
332 | } |
333 | altmap->alloc = 0; | |
334 | } | |
335 | ||
7ea62160 DW |
336 | err = check_pfn_span(pfn, nr_pages, "add"); |
337 | if (err) | |
338 | return err; | |
339 | ||
6cdd0b30 DH |
340 | for (; pfn < end_pfn; pfn += cur_nr_pages) { |
341 | /* Select all remaining pages up to the next section boundary */ | |
342 | cur_nr_pages = min(end_pfn - pfn, | |
343 | SECTION_ALIGN_UP(pfn + 1) - pfn); | |
344 | err = sparse_add_section(nid, pfn, cur_nr_pages, altmap); | |
ba72b4c8 DW |
345 | if (err) |
346 | break; | |
f64ac5e6 | 347 | cond_resched(); |
4edd7cef | 348 | } |
c435a390 | 349 | vmemmap_populate_print_last(); |
4edd7cef DR |
350 | return err; |
351 | } | |
4edd7cef | 352 | |
815121d2 | 353 | /* find the smallest valid pfn in the range [start_pfn, end_pfn) */ |
d09b0137 | 354 | static unsigned long find_smallest_section_pfn(int nid, struct zone *zone, |
815121d2 YI |
355 | unsigned long start_pfn, |
356 | unsigned long end_pfn) | |
357 | { | |
49ba3c6b | 358 | for (; start_pfn < end_pfn; start_pfn += PAGES_PER_SUBSECTION) { |
7ce700bf | 359 | if (unlikely(!pfn_to_online_page(start_pfn))) |
815121d2 YI |
360 | continue; |
361 | ||
362 | if (unlikely(pfn_to_nid(start_pfn) != nid)) | |
363 | continue; | |
364 | ||
9b05158f | 365 | if (zone != page_zone(pfn_to_page(start_pfn))) |
815121d2 YI |
366 | continue; |
367 | ||
368 | return start_pfn; | |
369 | } | |
370 | ||
371 | return 0; | |
372 | } | |
373 | ||
374 | /* find the biggest valid pfn in the range [start_pfn, end_pfn). */ | |
d09b0137 | 375 | static unsigned long find_biggest_section_pfn(int nid, struct zone *zone, |
815121d2 YI |
376 | unsigned long start_pfn, |
377 | unsigned long end_pfn) | |
378 | { | |
815121d2 YI |
379 | unsigned long pfn; |
380 | ||
381 | /* pfn is the end pfn of a memory section. */ | |
382 | pfn = end_pfn - 1; | |
49ba3c6b | 383 | for (; pfn >= start_pfn; pfn -= PAGES_PER_SUBSECTION) { |
7ce700bf | 384 | if (unlikely(!pfn_to_online_page(pfn))) |
815121d2 YI |
385 | continue; |
386 | ||
387 | if (unlikely(pfn_to_nid(pfn) != nid)) | |
388 | continue; | |
389 | ||
9b05158f | 390 | if (zone != page_zone(pfn_to_page(pfn))) |
815121d2 YI |
391 | continue; |
392 | ||
393 | return pfn; | |
394 | } | |
395 | ||
396 | return 0; | |
397 | } | |
398 | ||
399 | static void shrink_zone_span(struct zone *zone, unsigned long start_pfn, | |
400 | unsigned long end_pfn) | |
401 | { | |
815121d2 | 402 | unsigned long pfn; |
815121d2 YI |
403 | int nid = zone_to_nid(zone); |
404 | ||
405 | zone_span_writelock(zone); | |
5d12071c | 406 | if (zone->zone_start_pfn == start_pfn) { |
815121d2 YI |
407 | /* |
408 | * If the section is smallest section in the zone, it need | |
409 | * shrink zone->zone_start_pfn and zone->zone_spanned_pages. | |
410 | * In this case, we find second smallest valid mem_section | |
411 | * for shrinking zone. | |
412 | */ | |
413 | pfn = find_smallest_section_pfn(nid, zone, end_pfn, | |
5d12071c | 414 | zone_end_pfn(zone)); |
815121d2 | 415 | if (pfn) { |
5d12071c | 416 | zone->spanned_pages = zone_end_pfn(zone) - pfn; |
815121d2 | 417 | zone->zone_start_pfn = pfn; |
950b68d9 DH |
418 | } else { |
419 | zone->zone_start_pfn = 0; | |
420 | zone->spanned_pages = 0; | |
815121d2 | 421 | } |
5d12071c | 422 | } else if (zone_end_pfn(zone) == end_pfn) { |
815121d2 YI |
423 | /* |
424 | * If the section is biggest section in the zone, it need | |
425 | * shrink zone->spanned_pages. | |
426 | * In this case, we find second biggest valid mem_section for | |
427 | * shrinking zone. | |
428 | */ | |
5d12071c | 429 | pfn = find_biggest_section_pfn(nid, zone, zone->zone_start_pfn, |
815121d2 YI |
430 | start_pfn); |
431 | if (pfn) | |
5d12071c | 432 | zone->spanned_pages = pfn - zone->zone_start_pfn + 1; |
950b68d9 DH |
433 | else { |
434 | zone->zone_start_pfn = 0; | |
435 | zone->spanned_pages = 0; | |
436 | } | |
815121d2 | 437 | } |
815121d2 YI |
438 | zone_span_writeunlock(zone); |
439 | } | |
440 | ||
00d6c019 | 441 | static void update_pgdat_span(struct pglist_data *pgdat) |
815121d2 | 442 | { |
00d6c019 DH |
443 | unsigned long node_start_pfn = 0, node_end_pfn = 0; |
444 | struct zone *zone; | |
445 | ||
446 | for (zone = pgdat->node_zones; | |
447 | zone < pgdat->node_zones + MAX_NR_ZONES; zone++) { | |
448 | unsigned long zone_end_pfn = zone->zone_start_pfn + | |
449 | zone->spanned_pages; | |
450 | ||
451 | /* No need to lock the zones, they can't change. */ | |
656d5711 DH |
452 | if (!zone->spanned_pages) |
453 | continue; | |
454 | if (!node_end_pfn) { | |
455 | node_start_pfn = zone->zone_start_pfn; | |
456 | node_end_pfn = zone_end_pfn; | |
457 | continue; | |
458 | } | |
459 | ||
00d6c019 DH |
460 | if (zone_end_pfn > node_end_pfn) |
461 | node_end_pfn = zone_end_pfn; | |
462 | if (zone->zone_start_pfn < node_start_pfn) | |
463 | node_start_pfn = zone->zone_start_pfn; | |
815121d2 YI |
464 | } |
465 | ||
00d6c019 DH |
466 | pgdat->node_start_pfn = node_start_pfn; |
467 | pgdat->node_spanned_pages = node_end_pfn - node_start_pfn; | |
815121d2 YI |
468 | } |
469 | ||
feee6b29 DH |
470 | void __ref remove_pfn_range_from_zone(struct zone *zone, |
471 | unsigned long start_pfn, | |
472 | unsigned long nr_pages) | |
815121d2 | 473 | { |
b7e3debd | 474 | const unsigned long end_pfn = start_pfn + nr_pages; |
815121d2 | 475 | struct pglist_data *pgdat = zone->zone_pgdat; |
b7e3debd | 476 | unsigned long pfn, cur_nr_pages, flags; |
815121d2 | 477 | |
d33695b1 | 478 | /* Poison struct pages because they are now uninitialized again. */ |
b7e3debd BW |
479 | for (pfn = start_pfn; pfn < end_pfn; pfn += cur_nr_pages) { |
480 | cond_resched(); | |
481 | ||
482 | /* Select all remaining pages up to the next section boundary */ | |
483 | cur_nr_pages = | |
484 | min(end_pfn - pfn, SECTION_ALIGN_UP(pfn + 1) - pfn); | |
485 | page_init_poison(pfn_to_page(pfn), | |
486 | sizeof(struct page) * cur_nr_pages); | |
487 | } | |
d33695b1 | 488 | |
7ce700bf DH |
489 | #ifdef CONFIG_ZONE_DEVICE |
490 | /* | |
491 | * Zone shrinking code cannot properly deal with ZONE_DEVICE. So | |
492 | * we will not try to shrink the zones - which is okay as | |
493 | * set_zone_contiguous() cannot deal with ZONE_DEVICE either way. | |
494 | */ | |
495 | if (zone_idx(zone) == ZONE_DEVICE) | |
496 | return; | |
497 | #endif | |
498 | ||
feee6b29 DH |
499 | clear_zone_contiguous(zone); |
500 | ||
815121d2 YI |
501 | pgdat_resize_lock(zone->zone_pgdat, &flags); |
502 | shrink_zone_span(zone, start_pfn, start_pfn + nr_pages); | |
00d6c019 | 503 | update_pgdat_span(pgdat); |
815121d2 | 504 | pgdat_resize_unlock(zone->zone_pgdat, &flags); |
feee6b29 DH |
505 | |
506 | set_zone_contiguous(zone); | |
815121d2 YI |
507 | } |
508 | ||
feee6b29 DH |
509 | static void __remove_section(unsigned long pfn, unsigned long nr_pages, |
510 | unsigned long map_offset, | |
511 | struct vmem_altmap *altmap) | |
ea01ea93 | 512 | { |
10404901 | 513 | struct mem_section *ms = __pfn_to_section(pfn); |
ea01ea93 | 514 | |
9d1d887d DH |
515 | if (WARN_ON_ONCE(!valid_section(ms))) |
516 | return; | |
ea01ea93 | 517 | |
ba72b4c8 | 518 | sparse_remove_section(ms, pfn, nr_pages, map_offset, altmap); |
ea01ea93 BP |
519 | } |
520 | ||
ea01ea93 | 521 | /** |
feee6b29 | 522 | * __remove_pages() - remove sections of pages |
7ea62160 | 523 | * @pfn: starting pageframe (must be aligned to start of a section) |
ea01ea93 | 524 | * @nr_pages: number of pages to remove (must be multiple of section size) |
e8b098fc | 525 | * @altmap: alternative device page map or %NULL if default memmap is used |
ea01ea93 BP |
526 | * |
527 | * Generic helper function to remove section mappings and sysfs entries | |
528 | * for the section of the memory we are removing. Caller needs to make | |
529 | * sure that pages are marked reserved and zones are adjust properly by | |
530 | * calling offline_pages(). | |
531 | */ | |
feee6b29 DH |
532 | void __remove_pages(unsigned long pfn, unsigned long nr_pages, |
533 | struct vmem_altmap *altmap) | |
ea01ea93 | 534 | { |
52fb87c8 DH |
535 | const unsigned long end_pfn = pfn + nr_pages; |
536 | unsigned long cur_nr_pages; | |
4b94ffdc | 537 | unsigned long map_offset = 0; |
4b94ffdc | 538 | |
96da4350 | 539 | map_offset = vmem_altmap_offset(altmap); |
ea01ea93 | 540 | |
7ea62160 DW |
541 | if (check_pfn_span(pfn, nr_pages, "remove")) |
542 | return; | |
ea01ea93 | 543 | |
52fb87c8 | 544 | for (; pfn < end_pfn; pfn += cur_nr_pages) { |
dd33ad7b | 545 | cond_resched(); |
52fb87c8 | 546 | /* Select all remaining pages up to the next section boundary */ |
a11b9419 DH |
547 | cur_nr_pages = min(end_pfn - pfn, |
548 | SECTION_ALIGN_UP(pfn + 1) - pfn); | |
52fb87c8 | 549 | __remove_section(pfn, cur_nr_pages, map_offset, altmap); |
4b94ffdc | 550 | map_offset = 0; |
ea01ea93 | 551 | } |
ea01ea93 | 552 | } |
ea01ea93 | 553 | |
9d0ad8ca DK |
554 | int set_online_page_callback(online_page_callback_t callback) |
555 | { | |
556 | int rc = -EINVAL; | |
557 | ||
bfc8c901 VD |
558 | get_online_mems(); |
559 | mutex_lock(&online_page_callback_lock); | |
9d0ad8ca DK |
560 | |
561 | if (online_page_callback == generic_online_page) { | |
562 | online_page_callback = callback; | |
563 | rc = 0; | |
564 | } | |
565 | ||
bfc8c901 VD |
566 | mutex_unlock(&online_page_callback_lock); |
567 | put_online_mems(); | |
9d0ad8ca DK |
568 | |
569 | return rc; | |
570 | } | |
571 | EXPORT_SYMBOL_GPL(set_online_page_callback); | |
572 | ||
573 | int restore_online_page_callback(online_page_callback_t callback) | |
574 | { | |
575 | int rc = -EINVAL; | |
576 | ||
bfc8c901 VD |
577 | get_online_mems(); |
578 | mutex_lock(&online_page_callback_lock); | |
9d0ad8ca DK |
579 | |
580 | if (online_page_callback == callback) { | |
581 | online_page_callback = generic_online_page; | |
582 | rc = 0; | |
583 | } | |
584 | ||
bfc8c901 VD |
585 | mutex_unlock(&online_page_callback_lock); |
586 | put_online_mems(); | |
9d0ad8ca DK |
587 | |
588 | return rc; | |
589 | } | |
590 | EXPORT_SYMBOL_GPL(restore_online_page_callback); | |
591 | ||
18db1491 | 592 | void generic_online_page(struct page *page, unsigned int order) |
9d0ad8ca | 593 | { |
c87cbc1f VB |
594 | /* |
595 | * Freeing the page with debug_pagealloc enabled will try to unmap it, | |
596 | * so we should map it first. This is better than introducing a special | |
597 | * case in page freeing fast path. | |
598 | */ | |
599 | if (debug_pagealloc_enabled_static()) | |
600 | kernel_map_pages(page, 1 << order, 1); | |
a9cd410a AK |
601 | __free_pages_core(page, order); |
602 | totalram_pages_add(1UL << order); | |
603 | #ifdef CONFIG_HIGHMEM | |
604 | if (PageHighMem(page)) | |
605 | totalhigh_pages_add(1UL << order); | |
606 | #endif | |
607 | } | |
18db1491 | 608 | EXPORT_SYMBOL_GPL(generic_online_page); |
a9cd410a | 609 | |
75884fb1 KH |
610 | static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages, |
611 | void *arg) | |
3947be19 | 612 | { |
b2c2ab20 DH |
613 | const unsigned long end_pfn = start_pfn + nr_pages; |
614 | unsigned long pfn; | |
615 | int order; | |
616 | ||
617 | /* | |
618 | * Online the pages. The callback might decide to keep some pages | |
619 | * PG_reserved (to add them to the buddy later), but we still account | |
620 | * them as being online/belonging to this zone ("present"). | |
621 | */ | |
622 | for (pfn = start_pfn; pfn < end_pfn; pfn += 1ul << order) { | |
623 | order = min(MAX_ORDER - 1, get_order(PFN_PHYS(end_pfn - pfn))); | |
bd02cc01 DH |
624 | /* __free_pages_core() wants pfns to be aligned to the order */ |
625 | if (WARN_ON_ONCE(!IS_ALIGNED(pfn, 1ul << order))) | |
626 | order = 0; | |
b2c2ab20 DH |
627 | (*online_page_callback)(pfn_to_page(pfn), order); |
628 | } | |
2d070eab | 629 | |
b2c2ab20 DH |
630 | /* mark all involved sections as online */ |
631 | online_mem_sections(start_pfn, end_pfn); | |
2d070eab | 632 | |
b2c2ab20 | 633 | *(unsigned long *)arg += nr_pages; |
75884fb1 KH |
634 | return 0; |
635 | } | |
636 | ||
d9713679 LJ |
637 | /* check which state of node_states will be changed when online memory */ |
638 | static void node_states_check_changes_online(unsigned long nr_pages, | |
639 | struct zone *zone, struct memory_notify *arg) | |
640 | { | |
641 | int nid = zone_to_nid(zone); | |
d9713679 | 642 | |
98fa15f3 AK |
643 | arg->status_change_nid = NUMA_NO_NODE; |
644 | arg->status_change_nid_normal = NUMA_NO_NODE; | |
645 | arg->status_change_nid_high = NUMA_NO_NODE; | |
d9713679 | 646 | |
8efe33f4 OS |
647 | if (!node_state(nid, N_MEMORY)) |
648 | arg->status_change_nid = nid; | |
649 | if (zone_idx(zone) <= ZONE_NORMAL && !node_state(nid, N_NORMAL_MEMORY)) | |
d9713679 | 650 | arg->status_change_nid_normal = nid; |
6715ddf9 | 651 | #ifdef CONFIG_HIGHMEM |
d3ba3ae1 | 652 | if (zone_idx(zone) <= ZONE_HIGHMEM && !node_state(nid, N_HIGH_MEMORY)) |
6715ddf9 | 653 | arg->status_change_nid_high = nid; |
6715ddf9 | 654 | #endif |
d9713679 LJ |
655 | } |
656 | ||
657 | static void node_states_set_node(int node, struct memory_notify *arg) | |
658 | { | |
659 | if (arg->status_change_nid_normal >= 0) | |
660 | node_set_state(node, N_NORMAL_MEMORY); | |
661 | ||
6715ddf9 LJ |
662 | if (arg->status_change_nid_high >= 0) |
663 | node_set_state(node, N_HIGH_MEMORY); | |
664 | ||
83d83612 OS |
665 | if (arg->status_change_nid >= 0) |
666 | node_set_state(node, N_MEMORY); | |
d9713679 LJ |
667 | } |
668 | ||
f1dd2cd1 MH |
669 | static void __meminit resize_zone_range(struct zone *zone, unsigned long start_pfn, |
670 | unsigned long nr_pages) | |
671 | { | |
672 | unsigned long old_end_pfn = zone_end_pfn(zone); | |
673 | ||
674 | if (zone_is_empty(zone) || start_pfn < zone->zone_start_pfn) | |
675 | zone->zone_start_pfn = start_pfn; | |
676 | ||
677 | zone->spanned_pages = max(start_pfn + nr_pages, old_end_pfn) - zone->zone_start_pfn; | |
678 | } | |
679 | ||
680 | static void __meminit resize_pgdat_range(struct pglist_data *pgdat, unsigned long start_pfn, | |
681 | unsigned long nr_pages) | |
682 | { | |
683 | unsigned long old_end_pfn = pgdat_end_pfn(pgdat); | |
684 | ||
685 | if (!pgdat->node_spanned_pages || start_pfn < pgdat->node_start_pfn) | |
686 | pgdat->node_start_pfn = start_pfn; | |
687 | ||
688 | pgdat->node_spanned_pages = max(start_pfn + nr_pages, old_end_pfn) - pgdat->node_start_pfn; | |
f1dd2cd1 | 689 | |
3fccb74c DH |
690 | } |
691 | /* | |
692 | * Associate the pfn range with the given zone, initializing the memmaps | |
693 | * and resizing the pgdat/zone data to span the added pages. After this | |
694 | * call, all affected pages are PG_reserved. | |
695 | */ | |
a99583e7 CH |
696 | void __ref move_pfn_range_to_zone(struct zone *zone, unsigned long start_pfn, |
697 | unsigned long nr_pages, struct vmem_altmap *altmap) | |
f1dd2cd1 MH |
698 | { |
699 | struct pglist_data *pgdat = zone->zone_pgdat; | |
700 | int nid = pgdat->node_id; | |
701 | unsigned long flags; | |
df429ac0 | 702 | |
f1dd2cd1 MH |
703 | clear_zone_contiguous(zone); |
704 | ||
705 | /* TODO Huh pgdat is irqsave while zone is not. It used to be like that before */ | |
706 | pgdat_resize_lock(pgdat, &flags); | |
707 | zone_span_writelock(zone); | |
fa004ab7 WY |
708 | if (zone_is_empty(zone)) |
709 | init_currently_empty_zone(zone, start_pfn, nr_pages); | |
f1dd2cd1 MH |
710 | resize_zone_range(zone, start_pfn, nr_pages); |
711 | zone_span_writeunlock(zone); | |
712 | resize_pgdat_range(pgdat, start_pfn, nr_pages); | |
713 | pgdat_resize_unlock(pgdat, &flags); | |
714 | ||
715 | /* | |
716 | * TODO now we have a visible range of pages which are not associated | |
717 | * with their zone properly. Not nice but set_pfnblock_flags_mask | |
718 | * expects the zone spans the pfn range. All the pages in the range | |
719 | * are reserved so nobody should be touching them so we should be safe | |
720 | */ | |
a99583e7 CH |
721 | memmap_init_zone(nr_pages, nid, zone_idx(zone), start_pfn, |
722 | MEMMAP_HOTPLUG, altmap); | |
f1dd2cd1 MH |
723 | |
724 | set_zone_contiguous(zone); | |
725 | } | |
726 | ||
c246a213 MH |
727 | /* |
728 | * Returns a default kernel memory zone for the given pfn range. | |
729 | * If no kernel zone covers this pfn range it will automatically go | |
730 | * to the ZONE_NORMAL. | |
731 | */ | |
c6f03e29 | 732 | static struct zone *default_kernel_zone_for_pfn(int nid, unsigned long start_pfn, |
c246a213 MH |
733 | unsigned long nr_pages) |
734 | { | |
735 | struct pglist_data *pgdat = NODE_DATA(nid); | |
736 | int zid; | |
737 | ||
738 | for (zid = 0; zid <= ZONE_NORMAL; zid++) { | |
739 | struct zone *zone = &pgdat->node_zones[zid]; | |
740 | ||
741 | if (zone_intersects(zone, start_pfn, nr_pages)) | |
742 | return zone; | |
743 | } | |
744 | ||
745 | return &pgdat->node_zones[ZONE_NORMAL]; | |
746 | } | |
747 | ||
c6f03e29 MH |
748 | static inline struct zone *default_zone_for_pfn(int nid, unsigned long start_pfn, |
749 | unsigned long nr_pages) | |
e5e68930 | 750 | { |
c6f03e29 MH |
751 | struct zone *kernel_zone = default_kernel_zone_for_pfn(nid, start_pfn, |
752 | nr_pages); | |
753 | struct zone *movable_zone = &NODE_DATA(nid)->node_zones[ZONE_MOVABLE]; | |
754 | bool in_kernel = zone_intersects(kernel_zone, start_pfn, nr_pages); | |
755 | bool in_movable = zone_intersects(movable_zone, start_pfn, nr_pages); | |
e5e68930 MH |
756 | |
757 | /* | |
c6f03e29 MH |
758 | * We inherit the existing zone in a simple case where zones do not |
759 | * overlap in the given range | |
e5e68930 | 760 | */ |
c6f03e29 MH |
761 | if (in_kernel ^ in_movable) |
762 | return (in_kernel) ? kernel_zone : movable_zone; | |
9f123ab5 | 763 | |
c6f03e29 MH |
764 | /* |
765 | * If the range doesn't belong to any zone or two zones overlap in the | |
766 | * given range then we use movable zone only if movable_node is | |
767 | * enabled because we always online to a kernel zone by default. | |
768 | */ | |
769 | return movable_node_enabled ? movable_zone : kernel_zone; | |
9f123ab5 MH |
770 | } |
771 | ||
e5e68930 MH |
772 | struct zone * zone_for_pfn_range(int online_type, int nid, unsigned start_pfn, |
773 | unsigned long nr_pages) | |
f1dd2cd1 | 774 | { |
c6f03e29 MH |
775 | if (online_type == MMOP_ONLINE_KERNEL) |
776 | return default_kernel_zone_for_pfn(nid, start_pfn, nr_pages); | |
f1dd2cd1 | 777 | |
c6f03e29 MH |
778 | if (online_type == MMOP_ONLINE_MOVABLE) |
779 | return &NODE_DATA(nid)->node_zones[ZONE_MOVABLE]; | |
df429ac0 | 780 | |
c6f03e29 | 781 | return default_zone_for_pfn(nid, start_pfn, nr_pages); |
e5e68930 MH |
782 | } |
783 | ||
bd5c2344 DH |
784 | int __ref online_pages(unsigned long pfn, unsigned long nr_pages, |
785 | int online_type, int nid) | |
75884fb1 | 786 | { |
aa47228a | 787 | unsigned long flags; |
3947be19 DH |
788 | unsigned long onlined_pages = 0; |
789 | struct zone *zone; | |
6811378e | 790 | int need_zonelists_rebuild = 0; |
7b78d335 YG |
791 | int ret; |
792 | struct memory_notify arg; | |
d0dc12e8 | 793 | |
381eab4a DH |
794 | mem_hotplug_begin(); |
795 | ||
f1dd2cd1 | 796 | /* associate pfn range with the zone */ |
3fccb74c DH |
797 | zone = zone_for_pfn_range(online_type, nid, pfn, nr_pages); |
798 | move_pfn_range_to_zone(zone, pfn, nr_pages, NULL); | |
f1dd2cd1 | 799 | |
7b78d335 YG |
800 | arg.start_pfn = pfn; |
801 | arg.nr_pages = nr_pages; | |
d9713679 | 802 | node_states_check_changes_online(nr_pages, zone, &arg); |
7b78d335 | 803 | |
7b78d335 YG |
804 | ret = memory_notify(MEM_GOING_ONLINE, &arg); |
805 | ret = notifier_to_errno(ret); | |
e33e33b4 CY |
806 | if (ret) |
807 | goto failed_addition; | |
808 | ||
6811378e YG |
809 | /* |
810 | * If this zone is not populated, then it is not in zonelist. | |
811 | * This means the page allocator ignores this zone. | |
812 | * So, zonelist must be updated after online. | |
813 | */ | |
6dcd73d7 | 814 | if (!populated_zone(zone)) { |
6811378e | 815 | need_zonelists_rebuild = 1; |
72675e13 | 816 | setup_zone_pageset(zone); |
6dcd73d7 | 817 | } |
6811378e | 818 | |
908eedc6 | 819 | ret = walk_system_ram_range(pfn, nr_pages, &onlined_pages, |
75884fb1 | 820 | online_pages_range); |
fd8a4221 | 821 | if (ret) { |
ca9a46f8 | 822 | /* not a single memory resource was applicable */ |
6dcd73d7 WC |
823 | if (need_zonelists_rebuild) |
824 | zone_pcp_reset(zone); | |
e33e33b4 | 825 | goto failed_addition; |
fd8a4221 GL |
826 | } |
827 | ||
3947be19 | 828 | zone->present_pages += onlined_pages; |
aa47228a CS |
829 | |
830 | pgdat_resize_lock(zone->zone_pgdat, &flags); | |
f2937be5 | 831 | zone->zone_pgdat->node_present_pages += onlined_pages; |
aa47228a CS |
832 | pgdat_resize_unlock(zone->zone_pgdat, &flags); |
833 | ||
e900a918 DW |
834 | shuffle_zone(zone); |
835 | ||
ca9a46f8 DH |
836 | node_states_set_node(nid, &arg); |
837 | if (need_zonelists_rebuild) | |
838 | build_all_zonelists(NULL); | |
839 | else | |
840 | zone_pcp_update(zone); | |
3947be19 | 841 | |
1b79acc9 KM |
842 | init_per_zone_wmark_min(); |
843 | ||
ca9a46f8 DH |
844 | kswapd_run(nid); |
845 | kcompactd_run(nid); | |
61b13993 | 846 | |
1f522509 | 847 | vm_total_pages = nr_free_pagecache_pages(); |
2f7f24ec | 848 | |
2d1d43f6 | 849 | writeback_set_ratelimit(); |
7b78d335 | 850 | |
ca9a46f8 | 851 | memory_notify(MEM_ONLINE, &arg); |
381eab4a | 852 | mem_hotplug_done(); |
30467e0b | 853 | return 0; |
e33e33b4 CY |
854 | |
855 | failed_addition: | |
856 | pr_debug("online_pages [mem %#010llx-%#010llx] failed\n", | |
857 | (unsigned long long) pfn << PAGE_SHIFT, | |
858 | (((unsigned long long) pfn + nr_pages) << PAGE_SHIFT) - 1); | |
859 | memory_notify(MEM_CANCEL_ONLINE, &arg); | |
feee6b29 | 860 | remove_pfn_range_from_zone(zone, pfn, nr_pages); |
381eab4a | 861 | mem_hotplug_done(); |
e33e33b4 | 862 | return ret; |
3947be19 | 863 | } |
53947027 | 864 | #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */ |
bc02af93 | 865 | |
0bd85420 TC |
866 | static void reset_node_present_pages(pg_data_t *pgdat) |
867 | { | |
868 | struct zone *z; | |
869 | ||
870 | for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++) | |
871 | z->present_pages = 0; | |
872 | ||
873 | pgdat->node_present_pages = 0; | |
874 | } | |
875 | ||
e1319331 | 876 | /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */ |
c68ab18c | 877 | static pg_data_t __ref *hotadd_new_pgdat(int nid) |
9af3c2de YG |
878 | { |
879 | struct pglist_data *pgdat; | |
9af3c2de | 880 | |
a1e565aa TC |
881 | pgdat = NODE_DATA(nid); |
882 | if (!pgdat) { | |
883 | pgdat = arch_alloc_nodedata(nid); | |
884 | if (!pgdat) | |
885 | return NULL; | |
9af3c2de | 886 | |
33fce011 WY |
887 | pgdat->per_cpu_nodestats = |
888 | alloc_percpu(struct per_cpu_nodestat); | |
a1e565aa | 889 | arch_refresh_nodedata(nid, pgdat); |
b0dc3a34 | 890 | } else { |
33fce011 | 891 | int cpu; |
e716f2eb | 892 | /* |
97a225e6 JK |
893 | * Reset the nr_zones, order and highest_zoneidx before reuse. |
894 | * Note that kswapd will init kswapd_highest_zoneidx properly | |
e716f2eb MG |
895 | * when it starts in the near future. |
896 | */ | |
b0dc3a34 | 897 | pgdat->nr_zones = 0; |
38087d9b | 898 | pgdat->kswapd_order = 0; |
97a225e6 | 899 | pgdat->kswapd_highest_zoneidx = 0; |
33fce011 WY |
900 | for_each_online_cpu(cpu) { |
901 | struct per_cpu_nodestat *p; | |
902 | ||
903 | p = per_cpu_ptr(pgdat->per_cpu_nodestats, cpu); | |
904 | memset(p, 0, sizeof(*p)); | |
905 | } | |
a1e565aa | 906 | } |
9af3c2de YG |
907 | |
908 | /* we can use NODE_DATA(nid) from here */ | |
03e85f9d | 909 | pgdat->node_id = nid; |
c68ab18c | 910 | pgdat->node_start_pfn = 0; |
03e85f9d | 911 | |
9af3c2de | 912 | /* init node's zones as empty zones, we don't have any present pages.*/ |
03e85f9d | 913 | free_area_init_core_hotplug(nid); |
9af3c2de | 914 | |
959ecc48 KH |
915 | /* |
916 | * The node we allocated has no zone fallback lists. For avoiding | |
917 | * to access not-initialized zonelist, build here. | |
918 | */ | |
72675e13 | 919 | build_all_zonelists(pgdat); |
959ecc48 | 920 | |
0bd85420 TC |
921 | /* |
922 | * When memory is hot-added, all the memory is in offline state. So | |
923 | * clear all zones' present_pages because they will be updated in | |
924 | * online_pages() and offline_pages(). | |
925 | */ | |
03e85f9d | 926 | reset_node_managed_pages(pgdat); |
0bd85420 TC |
927 | reset_node_present_pages(pgdat); |
928 | ||
9af3c2de YG |
929 | return pgdat; |
930 | } | |
931 | ||
b9ff0360 | 932 | static void rollback_node_hotadd(int nid) |
9af3c2de | 933 | { |
b9ff0360 OS |
934 | pg_data_t *pgdat = NODE_DATA(nid); |
935 | ||
9af3c2de | 936 | arch_refresh_nodedata(nid, NULL); |
5830169f | 937 | free_percpu(pgdat->per_cpu_nodestats); |
9af3c2de | 938 | arch_free_nodedata(pgdat); |
9af3c2de YG |
939 | } |
940 | ||
0a547039 | 941 | |
01b0f197 TK |
942 | /** |
943 | * try_online_node - online a node if offlined | |
e8b098fc | 944 | * @nid: the node ID |
b9ff0360 | 945 | * @set_node_online: Whether we want to online the node |
cf23422b | 946 | * called by cpu_up() to online a node without onlined memory. |
b9ff0360 OS |
947 | * |
948 | * Returns: | |
949 | * 1 -> a new node has been allocated | |
950 | * 0 -> the node is already online | |
951 | * -ENOMEM -> the node could not be allocated | |
cf23422b | 952 | */ |
c68ab18c | 953 | static int __try_online_node(int nid, bool set_node_online) |
cf23422b | 954 | { |
b9ff0360 OS |
955 | pg_data_t *pgdat; |
956 | int ret = 1; | |
cf23422b | 957 | |
01b0f197 TK |
958 | if (node_online(nid)) |
959 | return 0; | |
960 | ||
c68ab18c | 961 | pgdat = hotadd_new_pgdat(nid); |
7553e8f2 | 962 | if (!pgdat) { |
01b0f197 | 963 | pr_err("Cannot online node %d due to NULL pgdat\n", nid); |
cf23422b | 964 | ret = -ENOMEM; |
965 | goto out; | |
966 | } | |
b9ff0360 OS |
967 | |
968 | if (set_node_online) { | |
969 | node_set_online(nid); | |
970 | ret = register_one_node(nid); | |
971 | BUG_ON(ret); | |
972 | } | |
cf23422b | 973 | out: |
b9ff0360 OS |
974 | return ret; |
975 | } | |
976 | ||
977 | /* | |
978 | * Users of this function always want to online/register the node | |
979 | */ | |
980 | int try_online_node(int nid) | |
981 | { | |
982 | int ret; | |
983 | ||
984 | mem_hotplug_begin(); | |
c68ab18c | 985 | ret = __try_online_node(nid, true); |
bfc8c901 | 986 | mem_hotplug_done(); |
cf23422b | 987 | return ret; |
988 | } | |
989 | ||
27356f54 TK |
990 | static int check_hotplug_memory_range(u64 start, u64 size) |
991 | { | |
ba325585 | 992 | /* memory range must be block size aligned */ |
cec3ebd0 DH |
993 | if (!size || !IS_ALIGNED(start, memory_block_size_bytes()) || |
994 | !IS_ALIGNED(size, memory_block_size_bytes())) { | |
ba325585 | 995 | pr_err("Block size [%#lx] unaligned hotplug range: start %#llx, size %#llx", |
cec3ebd0 | 996 | memory_block_size_bytes(), start, size); |
27356f54 TK |
997 | return -EINVAL; |
998 | } | |
999 | ||
1000 | return 0; | |
1001 | } | |
1002 | ||
31bc3858 VK |
1003 | static int online_memory_block(struct memory_block *mem, void *arg) |
1004 | { | |
862919e5 | 1005 | mem->online_type = memhp_default_online_type; |
dc18d706 | 1006 | return device_online(&mem->dev); |
31bc3858 VK |
1007 | } |
1008 | ||
8df1d0e4 DH |
1009 | /* |
1010 | * NOTE: The caller must call lock_device_hotplug() to serialize hotplug | |
1011 | * and online/offline operations (triggered e.g. by sysfs). | |
1012 | * | |
1013 | * we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG | |
1014 | */ | |
f29d8e9c | 1015 | int __ref add_memory_resource(int nid, struct resource *res) |
bc02af93 | 1016 | { |
bfeb022f | 1017 | struct mhp_params params = { .pgprot = PAGE_KERNEL }; |
62cedb9f | 1018 | u64 start, size; |
b9ff0360 | 1019 | bool new_node = false; |
bc02af93 YG |
1020 | int ret; |
1021 | ||
62cedb9f DV |
1022 | start = res->start; |
1023 | size = resource_size(res); | |
1024 | ||
27356f54 TK |
1025 | ret = check_hotplug_memory_range(start, size); |
1026 | if (ret) | |
1027 | return ret; | |
1028 | ||
fa6d9ec7 VV |
1029 | if (!node_possible(nid)) { |
1030 | WARN(1, "node %d was absent from the node_possible_map\n", nid); | |
1031 | return -EINVAL; | |
1032 | } | |
1033 | ||
bfc8c901 | 1034 | mem_hotplug_begin(); |
ac13c462 | 1035 | |
52219aea DH |
1036 | if (IS_ENABLED(CONFIG_ARCH_KEEP_MEMBLOCK)) |
1037 | memblock_add_node(start, size, nid); | |
7f36e3e5 | 1038 | |
c68ab18c | 1039 | ret = __try_online_node(nid, false); |
b9ff0360 OS |
1040 | if (ret < 0) |
1041 | goto error; | |
1042 | new_node = ret; | |
9af3c2de | 1043 | |
bc02af93 | 1044 | /* call arch's memory hotadd */ |
f5637d3b | 1045 | ret = arch_add_memory(nid, start, size, ¶ms); |
9af3c2de YG |
1046 | if (ret < 0) |
1047 | goto error; | |
1048 | ||
db051a0d DH |
1049 | /* create memory block devices after memory was added */ |
1050 | ret = create_memory_block_devices(start, size); | |
1051 | if (ret) { | |
1052 | arch_remove_memory(nid, start, size, NULL); | |
1053 | goto error; | |
1054 | } | |
1055 | ||
a1e565aa | 1056 | if (new_node) { |
d5b6f6a3 | 1057 | /* If sysfs file of new node can't be created, cpu on the node |
0fc44159 YG |
1058 | * can't be hot-added. There is no rollback way now. |
1059 | * So, check by BUG_ON() to catch it reluctantly.. | |
d5b6f6a3 | 1060 | * We online node here. We can't roll back from here. |
0fc44159 | 1061 | */ |
d5b6f6a3 OS |
1062 | node_set_online(nid); |
1063 | ret = __register_one_node(nid); | |
0fc44159 YG |
1064 | BUG_ON(ret); |
1065 | } | |
1066 | ||
d5b6f6a3 | 1067 | /* link memory sections under this node.*/ |
4fbce633 | 1068 | ret = link_mem_sections(nid, PFN_DOWN(start), PFN_UP(start + size - 1)); |
d5b6f6a3 OS |
1069 | BUG_ON(ret); |
1070 | ||
d96ae530 | 1071 | /* create new memmap entry */ |
7b7b2721 DH |
1072 | if (!strcmp(res->name, "System RAM")) |
1073 | firmware_map_add_hotplug(start, start + size, "System RAM"); | |
d96ae530 | 1074 | |
381eab4a DH |
1075 | /* device_online() will take the lock when calling online_pages() */ |
1076 | mem_hotplug_done(); | |
1077 | ||
31bc3858 | 1078 | /* online pages if requested */ |
862919e5 | 1079 | if (memhp_default_online_type != MMOP_OFFLINE) |
fbcf73ce | 1080 | walk_memory_blocks(start, size, NULL, online_memory_block); |
31bc3858 | 1081 | |
381eab4a | 1082 | return ret; |
9af3c2de YG |
1083 | error: |
1084 | /* rollback pgdat allocation and others */ | |
b9ff0360 OS |
1085 | if (new_node) |
1086 | rollback_node_hotadd(nid); | |
52219aea DH |
1087 | if (IS_ENABLED(CONFIG_ARCH_KEEP_MEMBLOCK)) |
1088 | memblock_remove(start, size); | |
bfc8c901 | 1089 | mem_hotplug_done(); |
bc02af93 YG |
1090 | return ret; |
1091 | } | |
62cedb9f | 1092 | |
8df1d0e4 DH |
1093 | /* requires device_hotplug_lock, see add_memory_resource() */ |
1094 | int __ref __add_memory(int nid, u64 start, u64 size) | |
62cedb9f DV |
1095 | { |
1096 | struct resource *res; | |
1097 | int ret; | |
1098 | ||
7b7b2721 | 1099 | res = register_memory_resource(start, size, "System RAM"); |
6f754ba4 VK |
1100 | if (IS_ERR(res)) |
1101 | return PTR_ERR(res); | |
62cedb9f | 1102 | |
f29d8e9c | 1103 | ret = add_memory_resource(nid, res); |
62cedb9f DV |
1104 | if (ret < 0) |
1105 | release_memory_resource(res); | |
1106 | return ret; | |
1107 | } | |
8df1d0e4 DH |
1108 | |
1109 | int add_memory(int nid, u64 start, u64 size) | |
1110 | { | |
1111 | int rc; | |
1112 | ||
1113 | lock_device_hotplug(); | |
1114 | rc = __add_memory(nid, start, size); | |
1115 | unlock_device_hotplug(); | |
1116 | ||
1117 | return rc; | |
1118 | } | |
bc02af93 | 1119 | EXPORT_SYMBOL_GPL(add_memory); |
0c0e6195 | 1120 | |
7b7b2721 DH |
1121 | /* |
1122 | * Add special, driver-managed memory to the system as system RAM. Such | |
1123 | * memory is not exposed via the raw firmware-provided memmap as system | |
1124 | * RAM, instead, it is detected and added by a driver - during cold boot, | |
1125 | * after a reboot, and after kexec. | |
1126 | * | |
1127 | * Reasons why this memory should not be used for the initial memmap of a | |
1128 | * kexec kernel or for placing kexec images: | |
1129 | * - The booting kernel is in charge of determining how this memory will be | |
1130 | * used (e.g., use persistent memory as system RAM) | |
1131 | * - Coordination with a hypervisor is required before this memory | |
1132 | * can be used (e.g., inaccessible parts). | |
1133 | * | |
1134 | * For this memory, no entries in /sys/firmware/memmap ("raw firmware-provided | |
1135 | * memory map") are created. Also, the created memory resource is flagged | |
1136 | * with IORESOURCE_MEM_DRIVER_MANAGED, so in-kernel users can special-case | |
1137 | * this memory as well (esp., not place kexec images onto it). | |
1138 | * | |
1139 | * The resource_name (visible via /proc/iomem) has to have the format | |
1140 | * "System RAM ($DRIVER)". | |
1141 | */ | |
1142 | int add_memory_driver_managed(int nid, u64 start, u64 size, | |
1143 | const char *resource_name) | |
1144 | { | |
1145 | struct resource *res; | |
1146 | int rc; | |
1147 | ||
1148 | if (!resource_name || | |
1149 | strstr(resource_name, "System RAM (") != resource_name || | |
1150 | resource_name[strlen(resource_name) - 1] != ')') | |
1151 | return -EINVAL; | |
1152 | ||
1153 | lock_device_hotplug(); | |
1154 | ||
1155 | res = register_memory_resource(start, size, resource_name); | |
1156 | if (IS_ERR(res)) { | |
1157 | rc = PTR_ERR(res); | |
1158 | goto out_unlock; | |
1159 | } | |
1160 | ||
1161 | rc = add_memory_resource(nid, res); | |
1162 | if (rc < 0) | |
1163 | release_memory_resource(res); | |
1164 | ||
1165 | out_unlock: | |
1166 | unlock_device_hotplug(); | |
1167 | return rc; | |
1168 | } | |
1169 | EXPORT_SYMBOL_GPL(add_memory_driver_managed); | |
1170 | ||
0c0e6195 KH |
1171 | #ifdef CONFIG_MEMORY_HOTREMOVE |
1172 | /* | |
92917998 DH |
1173 | * Confirm all pages in a range [start, end) belong to the same zone (skipping |
1174 | * memory holes). When true, return the zone. | |
0c0e6195 | 1175 | */ |
92917998 DH |
1176 | struct zone *test_pages_in_a_zone(unsigned long start_pfn, |
1177 | unsigned long end_pfn) | |
0c0e6195 | 1178 | { |
5f0f2887 | 1179 | unsigned long pfn, sec_end_pfn; |
0c0e6195 KH |
1180 | struct zone *zone = NULL; |
1181 | struct page *page; | |
1182 | int i; | |
deb88a2a | 1183 | for (pfn = start_pfn, sec_end_pfn = SECTION_ALIGN_UP(start_pfn + 1); |
0c0e6195 | 1184 | pfn < end_pfn; |
deb88a2a | 1185 | pfn = sec_end_pfn, sec_end_pfn += PAGES_PER_SECTION) { |
5f0f2887 AB |
1186 | /* Make sure the memory section is present first */ |
1187 | if (!present_section_nr(pfn_to_section_nr(pfn))) | |
0c0e6195 | 1188 | continue; |
5f0f2887 AB |
1189 | for (; pfn < sec_end_pfn && pfn < end_pfn; |
1190 | pfn += MAX_ORDER_NR_PAGES) { | |
1191 | i = 0; | |
1192 | /* This is just a CONFIG_HOLES_IN_ZONE check.*/ | |
1193 | while ((i < MAX_ORDER_NR_PAGES) && | |
1194 | !pfn_valid_within(pfn + i)) | |
1195 | i++; | |
d6d8c8a4 | 1196 | if (i == MAX_ORDER_NR_PAGES || pfn + i >= end_pfn) |
5f0f2887 | 1197 | continue; |
24feb47c MZ |
1198 | /* Check if we got outside of the zone */ |
1199 | if (zone && !zone_spans_pfn(zone, pfn + i)) | |
92917998 | 1200 | return NULL; |
5f0f2887 AB |
1201 | page = pfn_to_page(pfn + i); |
1202 | if (zone && page_zone(page) != zone) | |
92917998 | 1203 | return NULL; |
5f0f2887 AB |
1204 | zone = page_zone(page); |
1205 | } | |
0c0e6195 | 1206 | } |
deb88a2a | 1207 | |
92917998 | 1208 | return zone; |
0c0e6195 KH |
1209 | } |
1210 | ||
1211 | /* | |
0efadf48 | 1212 | * Scan pfn range [start,end) to find movable/migratable pages (LRU pages, |
aa218795 DH |
1213 | * non-lru movable pages and hugepages). Will skip over most unmovable |
1214 | * pages (esp., pages that can be skipped when offlining), but bail out on | |
1215 | * definitely unmovable pages. | |
1216 | * | |
1217 | * Returns: | |
1218 | * 0 in case a movable page is found and movable_pfn was updated. | |
1219 | * -ENOENT in case no movable page was found. | |
1220 | * -EBUSY in case a definitely unmovable page was found. | |
0c0e6195 | 1221 | */ |
aa218795 DH |
1222 | static int scan_movable_pages(unsigned long start, unsigned long end, |
1223 | unsigned long *movable_pfn) | |
0c0e6195 KH |
1224 | { |
1225 | unsigned long pfn; | |
eeb0efd0 | 1226 | |
0c0e6195 | 1227 | for (pfn = start; pfn < end; pfn++) { |
eeb0efd0 OS |
1228 | struct page *page, *head; |
1229 | unsigned long skip; | |
1230 | ||
1231 | if (!pfn_valid(pfn)) | |
1232 | continue; | |
1233 | page = pfn_to_page(pfn); | |
1234 | if (PageLRU(page)) | |
aa218795 | 1235 | goto found; |
eeb0efd0 | 1236 | if (__PageMovable(page)) |
aa218795 DH |
1237 | goto found; |
1238 | ||
1239 | /* | |
1240 | * PageOffline() pages that are not marked __PageMovable() and | |
1241 | * have a reference count > 0 (after MEM_GOING_OFFLINE) are | |
1242 | * definitely unmovable. If their reference count would be 0, | |
1243 | * they could at least be skipped when offlining memory. | |
1244 | */ | |
1245 | if (PageOffline(page) && page_count(page)) | |
1246 | return -EBUSY; | |
eeb0efd0 OS |
1247 | |
1248 | if (!PageHuge(page)) | |
1249 | continue; | |
1250 | head = compound_head(page); | |
39186cbe | 1251 | if (page_huge_active(head)) |
aa218795 | 1252 | goto found; |
d8c6546b | 1253 | skip = compound_nr(head) - (page - head); |
eeb0efd0 | 1254 | pfn += skip - 1; |
0c0e6195 | 1255 | } |
aa218795 DH |
1256 | return -ENOENT; |
1257 | found: | |
1258 | *movable_pfn = pfn; | |
0c0e6195 KH |
1259 | return 0; |
1260 | } | |
1261 | ||
666feb21 | 1262 | static struct page *new_node_page(struct page *page, unsigned long private) |
394e31d2 | 1263 | { |
394e31d2 | 1264 | int nid = page_to_nid(page); |
231e97e2 | 1265 | nodemask_t nmask = node_states[N_MEMORY]; |
7f252f27 MH |
1266 | |
1267 | /* | |
1268 | * try to allocate from a different node but reuse this node if there | |
1269 | * are no other online nodes to be used (e.g. we are offlining a part | |
1270 | * of the only existing node) | |
1271 | */ | |
1272 | node_clear(nid, nmask); | |
1273 | if (nodes_empty(nmask)) | |
1274 | node_set(nid, nmask); | |
394e31d2 | 1275 | |
8b913238 | 1276 | return new_page_nodemask(page, nid, &nmask); |
394e31d2 XQ |
1277 | } |
1278 | ||
0c0e6195 KH |
1279 | static int |
1280 | do_migrate_range(unsigned long start_pfn, unsigned long end_pfn) | |
1281 | { | |
1282 | unsigned long pfn; | |
1283 | struct page *page; | |
0c0e6195 KH |
1284 | int ret = 0; |
1285 | LIST_HEAD(source); | |
1286 | ||
a85009c3 | 1287 | for (pfn = start_pfn; pfn < end_pfn; pfn++) { |
0c0e6195 KH |
1288 | if (!pfn_valid(pfn)) |
1289 | continue; | |
1290 | page = pfn_to_page(pfn); | |
c8721bbb NH |
1291 | |
1292 | if (PageHuge(page)) { | |
1293 | struct page *head = compound_head(page); | |
d8c6546b | 1294 | pfn = page_to_pfn(head) + compound_nr(head) - 1; |
daf3538a | 1295 | isolate_huge_page(head, &source); |
c8721bbb | 1296 | continue; |
94723aaf | 1297 | } else if (PageTransHuge(page)) |
8135d892 NH |
1298 | pfn = page_to_pfn(compound_head(page)) |
1299 | + hpage_nr_pages(page) - 1; | |
c8721bbb | 1300 | |
b15c8726 MH |
1301 | /* |
1302 | * HWPoison pages have elevated reference counts so the migration would | |
1303 | * fail on them. It also doesn't make any sense to migrate them in the | |
1304 | * first place. Still try to unmap such a page in case it is still mapped | |
1305 | * (e.g. current hwpoison implementation doesn't unmap KSM pages but keep | |
1306 | * the unmap as the catch all safety net). | |
1307 | */ | |
1308 | if (PageHWPoison(page)) { | |
1309 | if (WARN_ON(PageLRU(page))) | |
1310 | isolate_lru_page(page); | |
1311 | if (page_mapped(page)) | |
1312 | try_to_unmap(page, TTU_IGNORE_MLOCK | TTU_IGNORE_ACCESS); | |
1313 | continue; | |
1314 | } | |
1315 | ||
700c2a46 | 1316 | if (!get_page_unless_zero(page)) |
0c0e6195 KH |
1317 | continue; |
1318 | /* | |
0efadf48 YX |
1319 | * We can skip free pages. And we can deal with pages on |
1320 | * LRU and non-lru movable pages. | |
0c0e6195 | 1321 | */ |
0efadf48 YX |
1322 | if (PageLRU(page)) |
1323 | ret = isolate_lru_page(page); | |
1324 | else | |
1325 | ret = isolate_movable_page(page, ISOLATE_UNEVICTABLE); | |
0c0e6195 | 1326 | if (!ret) { /* Success */ |
62695a84 | 1327 | list_add_tail(&page->lru, &source); |
0efadf48 YX |
1328 | if (!__PageMovable(page)) |
1329 | inc_node_page_state(page, NR_ISOLATED_ANON + | |
9de4f22a | 1330 | page_is_file_lru(page)); |
6d9c285a | 1331 | |
0c0e6195 | 1332 | } else { |
2932c8b0 | 1333 | pr_warn("failed to isolate pfn %lx\n", pfn); |
0efadf48 | 1334 | dump_page(page, "isolation failed"); |
0c0e6195 | 1335 | } |
1723058e | 1336 | put_page(page); |
0c0e6195 | 1337 | } |
f3ab2636 | 1338 | if (!list_empty(&source)) { |
394e31d2 XQ |
1339 | /* Allocate a new page from the nearest neighbor node */ |
1340 | ret = migrate_pages(&source, new_node_page, NULL, 0, | |
9c620e2b | 1341 | MIGRATE_SYNC, MR_MEMORY_HOTPLUG); |
2932c8b0 MH |
1342 | if (ret) { |
1343 | list_for_each_entry(page, &source, lru) { | |
1344 | pr_warn("migrating pfn %lx failed ret:%d ", | |
1345 | page_to_pfn(page), ret); | |
1346 | dump_page(page, "migration failure"); | |
1347 | } | |
c8721bbb | 1348 | putback_movable_pages(&source); |
2932c8b0 | 1349 | } |
0c0e6195 | 1350 | } |
1723058e | 1351 | |
0c0e6195 KH |
1352 | return ret; |
1353 | } | |
1354 | ||
0ee5f4f3 | 1355 | /* Mark all sections offline and remove all free pages from the buddy. */ |
0c0e6195 KH |
1356 | static int |
1357 | offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages, | |
1358 | void *data) | |
1359 | { | |
5557c766 | 1360 | unsigned long *offlined_pages = (unsigned long *)data; |
0c0e6195 | 1361 | |
5557c766 MH |
1362 | *offlined_pages += __offline_isolated_pages(start, start + nr_pages); |
1363 | return 0; | |
0c0e6195 KH |
1364 | } |
1365 | ||
1366 | /* | |
52cfc245 | 1367 | * Check all pages in range, recorded as memory resource, are isolated. |
0c0e6195 KH |
1368 | */ |
1369 | static int | |
1370 | check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages, | |
1371 | void *data) | |
1372 | { | |
756d25be DH |
1373 | return test_pages_isolated(start_pfn, start_pfn + nr_pages, |
1374 | MEMORY_OFFLINE); | |
0c0e6195 KH |
1375 | } |
1376 | ||
c5320926 TC |
1377 | static int __init cmdline_parse_movable_node(char *p) |
1378 | { | |
55ac590c | 1379 | movable_node_enabled = true; |
c5320926 TC |
1380 | return 0; |
1381 | } | |
1382 | early_param("movable_node", cmdline_parse_movable_node); | |
1383 | ||
d9713679 LJ |
1384 | /* check which state of node_states will be changed when offline memory */ |
1385 | static void node_states_check_changes_offline(unsigned long nr_pages, | |
1386 | struct zone *zone, struct memory_notify *arg) | |
1387 | { | |
1388 | struct pglist_data *pgdat = zone->zone_pgdat; | |
1389 | unsigned long present_pages = 0; | |
86b27bea | 1390 | enum zone_type zt; |
d9713679 | 1391 | |
98fa15f3 AK |
1392 | arg->status_change_nid = NUMA_NO_NODE; |
1393 | arg->status_change_nid_normal = NUMA_NO_NODE; | |
1394 | arg->status_change_nid_high = NUMA_NO_NODE; | |
d9713679 LJ |
1395 | |
1396 | /* | |
86b27bea OS |
1397 | * Check whether node_states[N_NORMAL_MEMORY] will be changed. |
1398 | * If the memory to be offline is within the range | |
1399 | * [0..ZONE_NORMAL], and it is the last present memory there, | |
1400 | * the zones in that range will become empty after the offlining, | |
1401 | * thus we can determine that we need to clear the node from | |
1402 | * node_states[N_NORMAL_MEMORY]. | |
d9713679 | 1403 | */ |
86b27bea | 1404 | for (zt = 0; zt <= ZONE_NORMAL; zt++) |
d9713679 | 1405 | present_pages += pgdat->node_zones[zt].present_pages; |
86b27bea | 1406 | if (zone_idx(zone) <= ZONE_NORMAL && nr_pages >= present_pages) |
d9713679 | 1407 | arg->status_change_nid_normal = zone_to_nid(zone); |
d9713679 | 1408 | |
6715ddf9 LJ |
1409 | #ifdef CONFIG_HIGHMEM |
1410 | /* | |
86b27bea OS |
1411 | * node_states[N_HIGH_MEMORY] contains nodes which |
1412 | * have normal memory or high memory. | |
1413 | * Here we add the present_pages belonging to ZONE_HIGHMEM. | |
1414 | * If the zone is within the range of [0..ZONE_HIGHMEM), and | |
1415 | * we determine that the zones in that range become empty, | |
1416 | * we need to clear the node for N_HIGH_MEMORY. | |
6715ddf9 | 1417 | */ |
86b27bea OS |
1418 | present_pages += pgdat->node_zones[ZONE_HIGHMEM].present_pages; |
1419 | if (zone_idx(zone) <= ZONE_HIGHMEM && nr_pages >= present_pages) | |
6715ddf9 | 1420 | arg->status_change_nid_high = zone_to_nid(zone); |
6715ddf9 LJ |
1421 | #endif |
1422 | ||
d9713679 | 1423 | /* |
86b27bea OS |
1424 | * We have accounted the pages from [0..ZONE_NORMAL), and |
1425 | * in case of CONFIG_HIGHMEM the pages from ZONE_HIGHMEM | |
1426 | * as well. | |
1427 | * Here we count the possible pages from ZONE_MOVABLE. | |
1428 | * If after having accounted all the pages, we see that the nr_pages | |
1429 | * to be offlined is over or equal to the accounted pages, | |
1430 | * we know that the node will become empty, and so, we can clear | |
1431 | * it for N_MEMORY as well. | |
d9713679 | 1432 | */ |
86b27bea | 1433 | present_pages += pgdat->node_zones[ZONE_MOVABLE].present_pages; |
d9713679 | 1434 | |
d9713679 LJ |
1435 | if (nr_pages >= present_pages) |
1436 | arg->status_change_nid = zone_to_nid(zone); | |
d9713679 LJ |
1437 | } |
1438 | ||
1439 | static void node_states_clear_node(int node, struct memory_notify *arg) | |
1440 | { | |
1441 | if (arg->status_change_nid_normal >= 0) | |
1442 | node_clear_state(node, N_NORMAL_MEMORY); | |
1443 | ||
cf01f6f5 | 1444 | if (arg->status_change_nid_high >= 0) |
d9713679 | 1445 | node_clear_state(node, N_HIGH_MEMORY); |
6715ddf9 | 1446 | |
cf01f6f5 | 1447 | if (arg->status_change_nid >= 0) |
6715ddf9 | 1448 | node_clear_state(node, N_MEMORY); |
d9713679 LJ |
1449 | } |
1450 | ||
c5e79ef5 DH |
1451 | static int count_system_ram_pages_cb(unsigned long start_pfn, |
1452 | unsigned long nr_pages, void *data) | |
1453 | { | |
1454 | unsigned long *nr_system_ram_pages = data; | |
1455 | ||
1456 | *nr_system_ram_pages += nr_pages; | |
1457 | return 0; | |
1458 | } | |
1459 | ||
a16cee10 | 1460 | static int __ref __offline_pages(unsigned long start_pfn, |
ecde0f3e | 1461 | unsigned long end_pfn) |
0c0e6195 | 1462 | { |
c5e79ef5 | 1463 | unsigned long pfn, nr_pages = 0; |
5557c766 | 1464 | unsigned long offlined_pages = 0; |
9b7ea46a | 1465 | int ret, node, nr_isolate_pageblock; |
d702909f | 1466 | unsigned long flags; |
0c0e6195 | 1467 | struct zone *zone; |
7b78d335 | 1468 | struct memory_notify arg; |
79605093 | 1469 | char *reason; |
0c0e6195 | 1470 | |
381eab4a DH |
1471 | mem_hotplug_begin(); |
1472 | ||
c5e79ef5 DH |
1473 | /* |
1474 | * Don't allow to offline memory blocks that contain holes. | |
1475 | * Consequently, memory blocks with holes can never get onlined | |
1476 | * via the hotplug path - online_pages() - as hotplugged memory has | |
1477 | * no holes. This way, we e.g., don't have to worry about marking | |
1478 | * memory holes PG_reserved, don't need pfn_valid() checks, and can | |
1479 | * avoid using walk_system_ram_range() later. | |
1480 | */ | |
1481 | walk_system_ram_range(start_pfn, end_pfn - start_pfn, &nr_pages, | |
1482 | count_system_ram_pages_cb); | |
1483 | if (nr_pages != end_pfn - start_pfn) { | |
1484 | ret = -EINVAL; | |
1485 | reason = "memory holes"; | |
1486 | goto failed_removal; | |
1487 | } | |
1488 | ||
0c0e6195 KH |
1489 | /* This makes hotplug much easier...and readable. |
1490 | we assume this for now. .*/ | |
92917998 DH |
1491 | zone = test_pages_in_a_zone(start_pfn, end_pfn); |
1492 | if (!zone) { | |
79605093 MH |
1493 | ret = -EINVAL; |
1494 | reason = "multizone range"; | |
1495 | goto failed_removal; | |
381eab4a | 1496 | } |
7b78d335 | 1497 | node = zone_to_nid(zone); |
7b78d335 | 1498 | |
0c0e6195 | 1499 | /* set above range as isolated */ |
b023f468 | 1500 | ret = start_isolate_page_range(start_pfn, end_pfn, |
d381c547 | 1501 | MIGRATE_MOVABLE, |
756d25be | 1502 | MEMORY_OFFLINE | REPORT_FAILURE); |
9b7ea46a | 1503 | if (ret < 0) { |
79605093 MH |
1504 | reason = "failure to isolate range"; |
1505 | goto failed_removal; | |
381eab4a | 1506 | } |
9b7ea46a | 1507 | nr_isolate_pageblock = ret; |
7b78d335 YG |
1508 | |
1509 | arg.start_pfn = start_pfn; | |
1510 | arg.nr_pages = nr_pages; | |
d9713679 | 1511 | node_states_check_changes_offline(nr_pages, zone, &arg); |
7b78d335 YG |
1512 | |
1513 | ret = memory_notify(MEM_GOING_OFFLINE, &arg); | |
1514 | ret = notifier_to_errno(ret); | |
79605093 MH |
1515 | if (ret) { |
1516 | reason = "notifier failure"; | |
1517 | goto failed_removal_isolated; | |
1518 | } | |
7b78d335 | 1519 | |
bb8965bd | 1520 | do { |
aa218795 DH |
1521 | pfn = start_pfn; |
1522 | do { | |
bb8965bd MH |
1523 | if (signal_pending(current)) { |
1524 | ret = -EINTR; | |
1525 | reason = "signal backoff"; | |
1526 | goto failed_removal_isolated; | |
1527 | } | |
72b39cfc | 1528 | |
bb8965bd MH |
1529 | cond_resched(); |
1530 | lru_add_drain_all(); | |
bb8965bd | 1531 | |
aa218795 DH |
1532 | ret = scan_movable_pages(pfn, end_pfn, &pfn); |
1533 | if (!ret) { | |
bb8965bd MH |
1534 | /* |
1535 | * TODO: fatal migration failures should bail | |
1536 | * out | |
1537 | */ | |
1538 | do_migrate_range(pfn, end_pfn); | |
1539 | } | |
aa218795 DH |
1540 | } while (!ret); |
1541 | ||
1542 | if (ret != -ENOENT) { | |
1543 | reason = "unmovable page"; | |
1544 | goto failed_removal_isolated; | |
bb8965bd | 1545 | } |
0c0e6195 | 1546 | |
bb8965bd MH |
1547 | /* |
1548 | * Dissolve free hugepages in the memory block before doing | |
1549 | * offlining actually in order to make hugetlbfs's object | |
1550 | * counting consistent. | |
1551 | */ | |
1552 | ret = dissolve_free_huge_pages(start_pfn, end_pfn); | |
1553 | if (ret) { | |
1554 | reason = "failure to dissolve huge pages"; | |
1555 | goto failed_removal_isolated; | |
1556 | } | |
1557 | /* check again */ | |
5557c766 MH |
1558 | ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn, |
1559 | NULL, check_pages_isolated_cb); | |
1560 | } while (ret); | |
72b39cfc | 1561 | |
b3834be5 | 1562 | /* Ok, all of our target is isolated. |
0c0e6195 | 1563 | We cannot do rollback at this point. */ |
5557c766 MH |
1564 | walk_system_ram_range(start_pfn, end_pfn - start_pfn, |
1565 | &offlined_pages, offline_isolated_pages_cb); | |
1566 | pr_info("Offlined Pages %ld\n", offlined_pages); | |
9b7ea46a QC |
1567 | /* |
1568 | * Onlining will reset pagetype flags and makes migrate type | |
1569 | * MOVABLE, so just need to decrease the number of isolated | |
1570 | * pageblocks zone counter here. | |
1571 | */ | |
1572 | spin_lock_irqsave(&zone->lock, flags); | |
1573 | zone->nr_isolate_pageblock -= nr_isolate_pageblock; | |
1574 | spin_unlock_irqrestore(&zone->lock, flags); | |
1575 | ||
0c0e6195 | 1576 | /* removal success */ |
3dcc0571 | 1577 | adjust_managed_page_count(pfn_to_page(start_pfn), -offlined_pages); |
0c0e6195 | 1578 | zone->present_pages -= offlined_pages; |
d702909f CS |
1579 | |
1580 | pgdat_resize_lock(zone->zone_pgdat, &flags); | |
0c0e6195 | 1581 | zone->zone_pgdat->node_present_pages -= offlined_pages; |
d702909f | 1582 | pgdat_resize_unlock(zone->zone_pgdat, &flags); |
7b78d335 | 1583 | |
1b79acc9 KM |
1584 | init_per_zone_wmark_min(); |
1585 | ||
1e8537ba | 1586 | if (!populated_zone(zone)) { |
340175b7 | 1587 | zone_pcp_reset(zone); |
72675e13 | 1588 | build_all_zonelists(NULL); |
1e8537ba XQ |
1589 | } else |
1590 | zone_pcp_update(zone); | |
340175b7 | 1591 | |
d9713679 | 1592 | node_states_clear_node(node, &arg); |
698b1b30 | 1593 | if (arg.status_change_nid >= 0) { |
8fe23e05 | 1594 | kswapd_stop(node); |
698b1b30 VB |
1595 | kcompactd_stop(node); |
1596 | } | |
bce7394a | 1597 | |
0c0e6195 KH |
1598 | vm_total_pages = nr_free_pagecache_pages(); |
1599 | writeback_set_ratelimit(); | |
7b78d335 YG |
1600 | |
1601 | memory_notify(MEM_OFFLINE, &arg); | |
feee6b29 | 1602 | remove_pfn_range_from_zone(zone, start_pfn, nr_pages); |
381eab4a | 1603 | mem_hotplug_done(); |
0c0e6195 KH |
1604 | return 0; |
1605 | ||
79605093 MH |
1606 | failed_removal_isolated: |
1607 | undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE); | |
c4efe484 | 1608 | memory_notify(MEM_CANCEL_OFFLINE, &arg); |
0c0e6195 | 1609 | failed_removal: |
79605093 | 1610 | pr_debug("memory offlining [mem %#010llx-%#010llx] failed due to %s\n", |
e33e33b4 | 1611 | (unsigned long long) start_pfn << PAGE_SHIFT, |
79605093 MH |
1612 | ((unsigned long long) end_pfn << PAGE_SHIFT) - 1, |
1613 | reason); | |
0c0e6195 | 1614 | /* pushback to free area */ |
381eab4a | 1615 | mem_hotplug_done(); |
0c0e6195 KH |
1616 | return ret; |
1617 | } | |
71088785 | 1618 | |
a16cee10 WC |
1619 | int offline_pages(unsigned long start_pfn, unsigned long nr_pages) |
1620 | { | |
ecde0f3e | 1621 | return __offline_pages(start_pfn, start_pfn + nr_pages); |
a16cee10 WC |
1622 | } |
1623 | ||
d6de9d53 | 1624 | static int check_memblock_offlined_cb(struct memory_block *mem, void *arg) |
bbc76be6 WC |
1625 | { |
1626 | int ret = !is_memblock_offlined(mem); | |
1627 | ||
349daa0f RD |
1628 | if (unlikely(ret)) { |
1629 | phys_addr_t beginpa, endpa; | |
1630 | ||
1631 | beginpa = PFN_PHYS(section_nr_to_pfn(mem->start_section_nr)); | |
b6c88d3b | 1632 | endpa = beginpa + memory_block_size_bytes() - 1; |
756a025f | 1633 | pr_warn("removing memory fails, because memory [%pa-%pa] is onlined\n", |
349daa0f | 1634 | &beginpa, &endpa); |
bbc76be6 | 1635 | |
eca499ab PT |
1636 | return -EBUSY; |
1637 | } | |
1638 | return 0; | |
bbc76be6 WC |
1639 | } |
1640 | ||
0f1cfe9d | 1641 | static int check_cpu_on_node(pg_data_t *pgdat) |
60a5a19e | 1642 | { |
60a5a19e TC |
1643 | int cpu; |
1644 | ||
1645 | for_each_present_cpu(cpu) { | |
1646 | if (cpu_to_node(cpu) == pgdat->node_id) | |
1647 | /* | |
1648 | * the cpu on this node isn't removed, and we can't | |
1649 | * offline this node. | |
1650 | */ | |
1651 | return -EBUSY; | |
1652 | } | |
1653 | ||
1654 | return 0; | |
1655 | } | |
1656 | ||
2c91f8fc DH |
1657 | static int check_no_memblock_for_node_cb(struct memory_block *mem, void *arg) |
1658 | { | |
1659 | int nid = *(int *)arg; | |
1660 | ||
1661 | /* | |
1662 | * If a memory block belongs to multiple nodes, the stored nid is not | |
1663 | * reliable. However, such blocks are always online (e.g., cannot get | |
1664 | * offlined) and, therefore, are still spanned by the node. | |
1665 | */ | |
1666 | return mem->nid == nid ? -EEXIST : 0; | |
1667 | } | |
1668 | ||
0f1cfe9d TK |
1669 | /** |
1670 | * try_offline_node | |
e8b098fc | 1671 | * @nid: the node ID |
0f1cfe9d TK |
1672 | * |
1673 | * Offline a node if all memory sections and cpus of the node are removed. | |
1674 | * | |
1675 | * NOTE: The caller must call lock_device_hotplug() to serialize hotplug | |
1676 | * and online/offline operations before this call. | |
1677 | */ | |
90b30cdc | 1678 | void try_offline_node(int nid) |
60a5a19e | 1679 | { |
d822b86a | 1680 | pg_data_t *pgdat = NODE_DATA(nid); |
2c91f8fc | 1681 | int rc; |
60a5a19e | 1682 | |
2c91f8fc DH |
1683 | /* |
1684 | * If the node still spans pages (especially ZONE_DEVICE), don't | |
1685 | * offline it. A node spans memory after move_pfn_range_to_zone(), | |
1686 | * e.g., after the memory block was onlined. | |
1687 | */ | |
1688 | if (pgdat->node_spanned_pages) | |
1689 | return; | |
60a5a19e | 1690 | |
2c91f8fc DH |
1691 | /* |
1692 | * Especially offline memory blocks might not be spanned by the | |
1693 | * node. They will get spanned by the node once they get onlined. | |
1694 | * However, they link to the node in sysfs and can get onlined later. | |
1695 | */ | |
1696 | rc = for_each_memory_block(&nid, check_no_memblock_for_node_cb); | |
1697 | if (rc) | |
60a5a19e | 1698 | return; |
60a5a19e | 1699 | |
46a3679b | 1700 | if (check_cpu_on_node(pgdat)) |
60a5a19e TC |
1701 | return; |
1702 | ||
1703 | /* | |
1704 | * all memory/cpu of this node are removed, we can offline this | |
1705 | * node now. | |
1706 | */ | |
1707 | node_set_offline(nid); | |
1708 | unregister_one_node(nid); | |
1709 | } | |
90b30cdc | 1710 | EXPORT_SYMBOL(try_offline_node); |
60a5a19e | 1711 | |
d9eb1417 DH |
1712 | static void __release_memory_resource(resource_size_t start, |
1713 | resource_size_t size) | |
1714 | { | |
1715 | int ret; | |
1716 | ||
1717 | /* | |
1718 | * When removing memory in the same granularity as it was added, | |
1719 | * this function never fails. It might only fail if resources | |
1720 | * have to be adjusted or split. We'll ignore the error, as | |
1721 | * removing of memory cannot fail. | |
1722 | */ | |
1723 | ret = release_mem_region_adjustable(&iomem_resource, start, size); | |
1724 | if (ret) { | |
1725 | resource_size_t endres = start + size - 1; | |
1726 | ||
1727 | pr_warn("Unable to release resource <%pa-%pa> (%d)\n", | |
1728 | &start, &endres, ret); | |
1729 | } | |
1730 | } | |
1731 | ||
eca499ab | 1732 | static int __ref try_remove_memory(int nid, u64 start, u64 size) |
bbc76be6 | 1733 | { |
eca499ab | 1734 | int rc = 0; |
993c1aad | 1735 | |
27356f54 TK |
1736 | BUG_ON(check_hotplug_memory_range(start, size)); |
1737 | ||
6677e3ea | 1738 | /* |
242831eb | 1739 | * All memory blocks must be offlined before removing memory. Check |
eca499ab | 1740 | * whether all memory blocks in question are offline and return error |
242831eb | 1741 | * if this is not the case. |
6677e3ea | 1742 | */ |
fbcf73ce | 1743 | rc = walk_memory_blocks(start, size, NULL, check_memblock_offlined_cb); |
eca499ab PT |
1744 | if (rc) |
1745 | goto done; | |
6677e3ea | 1746 | |
46c66c4b YI |
1747 | /* remove memmap entry */ |
1748 | firmware_map_remove(start, start + size, "System RAM"); | |
4c4b7f9b | 1749 | |
f1037ec0 DW |
1750 | /* |
1751 | * Memory block device removal under the device_hotplug_lock is | |
1752 | * a barrier against racing online attempts. | |
1753 | */ | |
4c4b7f9b | 1754 | remove_memory_block_devices(start, size); |
46c66c4b | 1755 | |
f1037ec0 DW |
1756 | mem_hotplug_begin(); |
1757 | ||
2c2a5af6 | 1758 | arch_remove_memory(nid, start, size, NULL); |
52219aea DH |
1759 | |
1760 | if (IS_ENABLED(CONFIG_ARCH_KEEP_MEMBLOCK)) { | |
1761 | memblock_free(start, size); | |
1762 | memblock_remove(start, size); | |
1763 | } | |
1764 | ||
d9eb1417 | 1765 | __release_memory_resource(start, size); |
24d335ca | 1766 | |
60a5a19e TC |
1767 | try_offline_node(nid); |
1768 | ||
eca499ab | 1769 | done: |
bfc8c901 | 1770 | mem_hotplug_done(); |
eca499ab | 1771 | return rc; |
71088785 | 1772 | } |
d15e5926 | 1773 | |
eca499ab PT |
1774 | /** |
1775 | * remove_memory | |
1776 | * @nid: the node ID | |
1777 | * @start: physical address of the region to remove | |
1778 | * @size: size of the region to remove | |
1779 | * | |
1780 | * NOTE: The caller must call lock_device_hotplug() to serialize hotplug | |
1781 | * and online/offline operations before this call, as required by | |
1782 | * try_offline_node(). | |
1783 | */ | |
1784 | void __remove_memory(int nid, u64 start, u64 size) | |
1785 | { | |
1786 | ||
1787 | /* | |
29a90db9 | 1788 | * trigger BUG() if some memory is not offlined prior to calling this |
eca499ab PT |
1789 | * function |
1790 | */ | |
1791 | if (try_remove_memory(nid, start, size)) | |
1792 | BUG(); | |
1793 | } | |
1794 | ||
1795 | /* | |
1796 | * Remove memory if every memory block is offline, otherwise return -EBUSY is | |
1797 | * some memory is not offline | |
1798 | */ | |
1799 | int remove_memory(int nid, u64 start, u64 size) | |
d15e5926 | 1800 | { |
eca499ab PT |
1801 | int rc; |
1802 | ||
d15e5926 | 1803 | lock_device_hotplug(); |
eca499ab | 1804 | rc = try_remove_memory(nid, start, size); |
d15e5926 | 1805 | unlock_device_hotplug(); |
eca499ab PT |
1806 | |
1807 | return rc; | |
d15e5926 | 1808 | } |
71088785 | 1809 | EXPORT_SYMBOL_GPL(remove_memory); |
08b3acd7 DH |
1810 | |
1811 | /* | |
1812 | * Try to offline and remove a memory block. Might take a long time to | |
1813 | * finish in case memory is still in use. Primarily useful for memory devices | |
1814 | * that logically unplugged all memory (so it's no longer in use) and want to | |
1815 | * offline + remove the memory block. | |
1816 | */ | |
1817 | int offline_and_remove_memory(int nid, u64 start, u64 size) | |
1818 | { | |
1819 | struct memory_block *mem; | |
1820 | int rc = -EINVAL; | |
1821 | ||
1822 | if (!IS_ALIGNED(start, memory_block_size_bytes()) || | |
1823 | size != memory_block_size_bytes()) | |
1824 | return rc; | |
1825 | ||
1826 | lock_device_hotplug(); | |
1827 | mem = find_memory_block(__pfn_to_section(PFN_DOWN(start))); | |
1828 | if (mem) | |
1829 | rc = device_offline(&mem->dev); | |
1830 | /* Ignore if the device is already offline. */ | |
1831 | if (rc > 0) | |
1832 | rc = 0; | |
1833 | ||
1834 | /* | |
1835 | * In case we succeeded to offline the memory block, remove it. | |
1836 | * This cannot fail as it cannot get onlined in the meantime. | |
1837 | */ | |
1838 | if (!rc) { | |
1839 | rc = try_remove_memory(nid, start, size); | |
1840 | WARN_ON_ONCE(rc); | |
1841 | } | |
1842 | unlock_device_hotplug(); | |
1843 | ||
1844 | return rc; | |
1845 | } | |
1846 | EXPORT_SYMBOL_GPL(offline_and_remove_memory); | |
aba6efc4 | 1847 | #endif /* CONFIG_MEMORY_HOTREMOVE */ |