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 KM |
100 | /* add this memory to iomem resource */ |
101 | static struct resource *register_memory_resource(u64 start, u64 size) | |
102 | { | |
2794129e DH |
103 | struct resource *res; |
104 | unsigned long flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY; | |
105 | char *resource_name = "System RAM"; | |
357b4da5 | 106 | |
f3cd4c86 BH |
107 | /* |
108 | * Make sure value parsed from 'mem=' only restricts memory adding | |
109 | * while booting, so that memory hotplug won't be impacted. Please | |
110 | * refer to document of 'mem=' in kernel-parameters.txt for more | |
111 | * details. | |
112 | */ | |
113 | if (start + size > max_mem_size && system_state < SYSTEM_RUNNING) | |
357b4da5 JG |
114 | return ERR_PTR(-E2BIG); |
115 | ||
2794129e DH |
116 | /* |
117 | * Request ownership of the new memory range. This might be | |
118 | * a child of an existing resource that was present but | |
119 | * not marked as busy. | |
120 | */ | |
121 | res = __request_region(&iomem_resource, start, size, | |
122 | resource_name, flags); | |
123 | ||
124 | if (!res) { | |
125 | pr_debug("Unable to reserve System RAM region: %016llx->%016llx\n", | |
126 | start, start + size); | |
6f754ba4 | 127 | return ERR_PTR(-EEXIST); |
45e0b78b KM |
128 | } |
129 | return res; | |
130 | } | |
131 | ||
132 | static void release_memory_resource(struct resource *res) | |
133 | { | |
134 | if (!res) | |
135 | return; | |
136 | release_resource(res); | |
137 | kfree(res); | |
45e0b78b KM |
138 | } |
139 | ||
53947027 | 140 | #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE |
46723bfa YI |
141 | void get_page_bootmem(unsigned long info, struct page *page, |
142 | unsigned long type) | |
04753278 | 143 | { |
ddffe98d | 144 | page->freelist = (void *)type; |
04753278 YG |
145 | SetPagePrivate(page); |
146 | set_page_private(page, info); | |
fe896d18 | 147 | page_ref_inc(page); |
04753278 YG |
148 | } |
149 | ||
170a5a7e | 150 | void put_page_bootmem(struct page *page) |
04753278 | 151 | { |
5f24ce5f | 152 | unsigned long type; |
04753278 | 153 | |
ddffe98d | 154 | type = (unsigned long) page->freelist; |
5f24ce5f AA |
155 | BUG_ON(type < MEMORY_HOTPLUG_MIN_BOOTMEM_TYPE || |
156 | type > MEMORY_HOTPLUG_MAX_BOOTMEM_TYPE); | |
04753278 | 157 | |
fe896d18 | 158 | if (page_ref_dec_return(page) == 1) { |
ddffe98d | 159 | page->freelist = NULL; |
04753278 YG |
160 | ClearPagePrivate(page); |
161 | set_page_private(page, 0); | |
5f24ce5f | 162 | INIT_LIST_HEAD(&page->lru); |
170a5a7e | 163 | free_reserved_page(page); |
04753278 | 164 | } |
04753278 YG |
165 | } |
166 | ||
46723bfa YI |
167 | #ifdef CONFIG_HAVE_BOOTMEM_INFO_NODE |
168 | #ifndef CONFIG_SPARSEMEM_VMEMMAP | |
d92bc318 | 169 | static void register_page_bootmem_info_section(unsigned long start_pfn) |
04753278 | 170 | { |
f1eca35a | 171 | unsigned long mapsize, section_nr, i; |
04753278 YG |
172 | struct mem_section *ms; |
173 | struct page *page, *memmap; | |
f1eca35a | 174 | struct mem_section_usage *usage; |
04753278 | 175 | |
04753278 YG |
176 | section_nr = pfn_to_section_nr(start_pfn); |
177 | ms = __nr_to_section(section_nr); | |
178 | ||
179 | /* Get section's memmap address */ | |
180 | memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr); | |
181 | ||
182 | /* | |
183 | * Get page for the memmap's phys address | |
184 | * XXX: need more consideration for sparse_vmemmap... | |
185 | */ | |
186 | page = virt_to_page(memmap); | |
187 | mapsize = sizeof(struct page) * PAGES_PER_SECTION; | |
188 | mapsize = PAGE_ALIGN(mapsize) >> PAGE_SHIFT; | |
189 | ||
190 | /* remember memmap's page */ | |
191 | for (i = 0; i < mapsize; i++, page++) | |
192 | get_page_bootmem(section_nr, page, SECTION_INFO); | |
193 | ||
f1eca35a DW |
194 | usage = ms->usage; |
195 | page = virt_to_page(usage); | |
04753278 | 196 | |
f1eca35a | 197 | mapsize = PAGE_ALIGN(mem_section_usage_size()) >> PAGE_SHIFT; |
04753278 YG |
198 | |
199 | for (i = 0; i < mapsize; i++, page++) | |
af370fb8 | 200 | get_page_bootmem(section_nr, page, MIX_SECTION_INFO); |
04753278 YG |
201 | |
202 | } | |
46723bfa YI |
203 | #else /* CONFIG_SPARSEMEM_VMEMMAP */ |
204 | static void register_page_bootmem_info_section(unsigned long start_pfn) | |
205 | { | |
f1eca35a | 206 | unsigned long mapsize, section_nr, i; |
46723bfa YI |
207 | struct mem_section *ms; |
208 | struct page *page, *memmap; | |
f1eca35a | 209 | struct mem_section_usage *usage; |
46723bfa | 210 | |
46723bfa YI |
211 | section_nr = pfn_to_section_nr(start_pfn); |
212 | ms = __nr_to_section(section_nr); | |
213 | ||
214 | memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr); | |
215 | ||
216 | register_page_bootmem_memmap(section_nr, memmap, PAGES_PER_SECTION); | |
217 | ||
f1eca35a DW |
218 | usage = ms->usage; |
219 | page = virt_to_page(usage); | |
46723bfa | 220 | |
f1eca35a | 221 | mapsize = PAGE_ALIGN(mem_section_usage_size()) >> PAGE_SHIFT; |
46723bfa YI |
222 | |
223 | for (i = 0; i < mapsize; i++, page++) | |
224 | get_page_bootmem(section_nr, page, MIX_SECTION_INFO); | |
225 | } | |
226 | #endif /* !CONFIG_SPARSEMEM_VMEMMAP */ | |
04753278 | 227 | |
7ded384a | 228 | void __init register_page_bootmem_info_node(struct pglist_data *pgdat) |
04753278 YG |
229 | { |
230 | unsigned long i, pfn, end_pfn, nr_pages; | |
231 | int node = pgdat->node_id; | |
232 | struct page *page; | |
04753278 YG |
233 | |
234 | nr_pages = PAGE_ALIGN(sizeof(struct pglist_data)) >> PAGE_SHIFT; | |
235 | page = virt_to_page(pgdat); | |
236 | ||
237 | for (i = 0; i < nr_pages; i++, page++) | |
238 | get_page_bootmem(node, page, NODE_INFO); | |
239 | ||
04753278 | 240 | pfn = pgdat->node_start_pfn; |
c1f19495 | 241 | end_pfn = pgdat_end_pfn(pgdat); |
04753278 | 242 | |
7e9f5eb0 | 243 | /* register section info */ |
f14851af | 244 | for (; pfn < end_pfn; pfn += PAGES_PER_SECTION) { |
245 | /* | |
246 | * Some platforms can assign the same pfn to multiple nodes - on | |
247 | * node0 as well as nodeN. To avoid registering a pfn against | |
248 | * multiple nodes we check that this pfn does not already | |
7e9f5eb0 | 249 | * reside in some other nodes. |
f14851af | 250 | */ |
f65e91df | 251 | if (pfn_valid(pfn) && (early_pfn_to_nid(pfn) == node)) |
f14851af | 252 | register_page_bootmem_info_section(pfn); |
253 | } | |
04753278 | 254 | } |
46723bfa | 255 | #endif /* CONFIG_HAVE_BOOTMEM_INFO_NODE */ |
04753278 | 256 | |
7ea62160 DW |
257 | static int check_pfn_span(unsigned long pfn, unsigned long nr_pages, |
258 | const char *reason) | |
259 | { | |
260 | /* | |
261 | * Disallow all operations smaller than a sub-section and only | |
262 | * allow operations smaller than a section for | |
263 | * SPARSEMEM_VMEMMAP. Note that check_hotplug_memory_range() | |
264 | * enforces a larger memory_block_size_bytes() granularity for | |
265 | * memory that will be marked online, so this check should only | |
266 | * fire for direct arch_{add,remove}_memory() users outside of | |
267 | * add_memory_resource(). | |
268 | */ | |
269 | unsigned long min_align; | |
270 | ||
271 | if (IS_ENABLED(CONFIG_SPARSEMEM_VMEMMAP)) | |
272 | min_align = PAGES_PER_SUBSECTION; | |
273 | else | |
274 | min_align = PAGES_PER_SECTION; | |
275 | if (!IS_ALIGNED(pfn, min_align) | |
276 | || !IS_ALIGNED(nr_pages, min_align)) { | |
277 | WARN(1, "Misaligned __%s_pages start: %#lx end: #%lx\n", | |
278 | reason, pfn, pfn + nr_pages - 1); | |
279 | return -EINVAL; | |
280 | } | |
281 | return 0; | |
282 | } | |
283 | ||
dca4436d AS |
284 | static int check_hotplug_memory_addressable(unsigned long pfn, |
285 | unsigned long nr_pages) | |
286 | { | |
287 | const u64 max_addr = PFN_PHYS(pfn + nr_pages) - 1; | |
288 | ||
289 | if (max_addr >> MAX_PHYSMEM_BITS) { | |
290 | const u64 max_allowed = (1ull << (MAX_PHYSMEM_BITS + 1)) - 1; | |
291 | WARN(1, | |
292 | "Hotplugged memory exceeds maximum addressable address, range=%#llx-%#llx, maximum=%#llx\n", | |
293 | (u64)PFN_PHYS(pfn), max_addr, max_allowed); | |
294 | return -E2BIG; | |
295 | } | |
296 | ||
297 | return 0; | |
298 | } | |
299 | ||
4edd7cef DR |
300 | /* |
301 | * Reasonably generic function for adding memory. It is | |
302 | * expected that archs that support memory hotplug will | |
303 | * call this function after deciding the zone to which to | |
304 | * add the new pages. | |
305 | */ | |
7ea62160 | 306 | int __ref __add_pages(int nid, unsigned long pfn, unsigned long nr_pages, |
f5637d3b | 307 | struct mhp_params *params) |
4edd7cef | 308 | { |
6cdd0b30 DH |
309 | const unsigned long end_pfn = pfn + nr_pages; |
310 | unsigned long cur_nr_pages; | |
9a845030 | 311 | int err; |
f5637d3b | 312 | struct vmem_altmap *altmap = params->altmap; |
4b94ffdc | 313 | |
bfeb022f LG |
314 | if (WARN_ON_ONCE(!params->pgprot.pgprot)) |
315 | return -EINVAL; | |
316 | ||
dca4436d AS |
317 | err = check_hotplug_memory_addressable(pfn, nr_pages); |
318 | if (err) | |
319 | return err; | |
320 | ||
4b94ffdc DW |
321 | if (altmap) { |
322 | /* | |
323 | * Validate altmap is within bounds of the total request | |
324 | */ | |
7ea62160 | 325 | if (altmap->base_pfn != pfn |
4b94ffdc DW |
326 | || vmem_altmap_offset(altmap) > nr_pages) { |
327 | pr_warn_once("memory add fail, invalid altmap\n"); | |
7ea62160 | 328 | return -EINVAL; |
4b94ffdc DW |
329 | } |
330 | altmap->alloc = 0; | |
331 | } | |
332 | ||
7ea62160 DW |
333 | err = check_pfn_span(pfn, nr_pages, "add"); |
334 | if (err) | |
335 | return err; | |
336 | ||
6cdd0b30 DH |
337 | for (; pfn < end_pfn; pfn += cur_nr_pages) { |
338 | /* Select all remaining pages up to the next section boundary */ | |
339 | cur_nr_pages = min(end_pfn - pfn, | |
340 | SECTION_ALIGN_UP(pfn + 1) - pfn); | |
341 | err = sparse_add_section(nid, pfn, cur_nr_pages, altmap); | |
ba72b4c8 DW |
342 | if (err) |
343 | break; | |
f64ac5e6 | 344 | cond_resched(); |
4edd7cef | 345 | } |
c435a390 | 346 | vmemmap_populate_print_last(); |
4edd7cef DR |
347 | return err; |
348 | } | |
4edd7cef | 349 | |
815121d2 | 350 | /* find the smallest valid pfn in the range [start_pfn, end_pfn) */ |
d09b0137 | 351 | static unsigned long find_smallest_section_pfn(int nid, struct zone *zone, |
815121d2 YI |
352 | unsigned long start_pfn, |
353 | unsigned long end_pfn) | |
354 | { | |
49ba3c6b | 355 | for (; start_pfn < end_pfn; start_pfn += PAGES_PER_SUBSECTION) { |
7ce700bf | 356 | if (unlikely(!pfn_to_online_page(start_pfn))) |
815121d2 YI |
357 | continue; |
358 | ||
359 | if (unlikely(pfn_to_nid(start_pfn) != nid)) | |
360 | continue; | |
361 | ||
9b05158f | 362 | if (zone != page_zone(pfn_to_page(start_pfn))) |
815121d2 YI |
363 | continue; |
364 | ||
365 | return start_pfn; | |
366 | } | |
367 | ||
368 | return 0; | |
369 | } | |
370 | ||
371 | /* find the biggest valid pfn in the range [start_pfn, end_pfn). */ | |
d09b0137 | 372 | static unsigned long find_biggest_section_pfn(int nid, struct zone *zone, |
815121d2 YI |
373 | unsigned long start_pfn, |
374 | unsigned long end_pfn) | |
375 | { | |
815121d2 YI |
376 | unsigned long pfn; |
377 | ||
378 | /* pfn is the end pfn of a memory section. */ | |
379 | pfn = end_pfn - 1; | |
49ba3c6b | 380 | for (; pfn >= start_pfn; pfn -= PAGES_PER_SUBSECTION) { |
7ce700bf | 381 | if (unlikely(!pfn_to_online_page(pfn))) |
815121d2 YI |
382 | continue; |
383 | ||
384 | if (unlikely(pfn_to_nid(pfn) != nid)) | |
385 | continue; | |
386 | ||
9b05158f | 387 | if (zone != page_zone(pfn_to_page(pfn))) |
815121d2 YI |
388 | continue; |
389 | ||
390 | return pfn; | |
391 | } | |
392 | ||
393 | return 0; | |
394 | } | |
395 | ||
396 | static void shrink_zone_span(struct zone *zone, unsigned long start_pfn, | |
397 | unsigned long end_pfn) | |
398 | { | |
815121d2 | 399 | unsigned long pfn; |
815121d2 YI |
400 | int nid = zone_to_nid(zone); |
401 | ||
402 | zone_span_writelock(zone); | |
5d12071c | 403 | if (zone->zone_start_pfn == start_pfn) { |
815121d2 YI |
404 | /* |
405 | * If the section is smallest section in the zone, it need | |
406 | * shrink zone->zone_start_pfn and zone->zone_spanned_pages. | |
407 | * In this case, we find second smallest valid mem_section | |
408 | * for shrinking zone. | |
409 | */ | |
410 | pfn = find_smallest_section_pfn(nid, zone, end_pfn, | |
5d12071c | 411 | zone_end_pfn(zone)); |
815121d2 | 412 | if (pfn) { |
5d12071c | 413 | zone->spanned_pages = zone_end_pfn(zone) - pfn; |
815121d2 | 414 | zone->zone_start_pfn = pfn; |
950b68d9 DH |
415 | } else { |
416 | zone->zone_start_pfn = 0; | |
417 | zone->spanned_pages = 0; | |
815121d2 | 418 | } |
5d12071c | 419 | } else if (zone_end_pfn(zone) == end_pfn) { |
815121d2 YI |
420 | /* |
421 | * If the section is biggest section in the zone, it need | |
422 | * shrink zone->spanned_pages. | |
423 | * In this case, we find second biggest valid mem_section for | |
424 | * shrinking zone. | |
425 | */ | |
5d12071c | 426 | pfn = find_biggest_section_pfn(nid, zone, zone->zone_start_pfn, |
815121d2 YI |
427 | start_pfn); |
428 | if (pfn) | |
5d12071c | 429 | zone->spanned_pages = pfn - zone->zone_start_pfn + 1; |
950b68d9 DH |
430 | else { |
431 | zone->zone_start_pfn = 0; | |
432 | zone->spanned_pages = 0; | |
433 | } | |
815121d2 | 434 | } |
815121d2 YI |
435 | zone_span_writeunlock(zone); |
436 | } | |
437 | ||
00d6c019 | 438 | static void update_pgdat_span(struct pglist_data *pgdat) |
815121d2 | 439 | { |
00d6c019 DH |
440 | unsigned long node_start_pfn = 0, node_end_pfn = 0; |
441 | struct zone *zone; | |
442 | ||
443 | for (zone = pgdat->node_zones; | |
444 | zone < pgdat->node_zones + MAX_NR_ZONES; zone++) { | |
445 | unsigned long zone_end_pfn = zone->zone_start_pfn + | |
446 | zone->spanned_pages; | |
447 | ||
448 | /* No need to lock the zones, they can't change. */ | |
656d5711 DH |
449 | if (!zone->spanned_pages) |
450 | continue; | |
451 | if (!node_end_pfn) { | |
452 | node_start_pfn = zone->zone_start_pfn; | |
453 | node_end_pfn = zone_end_pfn; | |
454 | continue; | |
455 | } | |
456 | ||
00d6c019 DH |
457 | if (zone_end_pfn > node_end_pfn) |
458 | node_end_pfn = zone_end_pfn; | |
459 | if (zone->zone_start_pfn < node_start_pfn) | |
460 | node_start_pfn = zone->zone_start_pfn; | |
815121d2 YI |
461 | } |
462 | ||
00d6c019 DH |
463 | pgdat->node_start_pfn = node_start_pfn; |
464 | pgdat->node_spanned_pages = node_end_pfn - node_start_pfn; | |
815121d2 YI |
465 | } |
466 | ||
feee6b29 DH |
467 | void __ref remove_pfn_range_from_zone(struct zone *zone, |
468 | unsigned long start_pfn, | |
469 | unsigned long nr_pages) | |
815121d2 YI |
470 | { |
471 | struct pglist_data *pgdat = zone->zone_pgdat; | |
815121d2 YI |
472 | unsigned long flags; |
473 | ||
d33695b1 DH |
474 | /* Poison struct pages because they are now uninitialized again. */ |
475 | page_init_poison(pfn_to_page(start_pfn), sizeof(struct page) * nr_pages); | |
476 | ||
7ce700bf DH |
477 | #ifdef CONFIG_ZONE_DEVICE |
478 | /* | |
479 | * Zone shrinking code cannot properly deal with ZONE_DEVICE. So | |
480 | * we will not try to shrink the zones - which is okay as | |
481 | * set_zone_contiguous() cannot deal with ZONE_DEVICE either way. | |
482 | */ | |
483 | if (zone_idx(zone) == ZONE_DEVICE) | |
484 | return; | |
485 | #endif | |
486 | ||
feee6b29 DH |
487 | clear_zone_contiguous(zone); |
488 | ||
815121d2 YI |
489 | pgdat_resize_lock(zone->zone_pgdat, &flags); |
490 | shrink_zone_span(zone, start_pfn, start_pfn + nr_pages); | |
00d6c019 | 491 | update_pgdat_span(pgdat); |
815121d2 | 492 | pgdat_resize_unlock(zone->zone_pgdat, &flags); |
feee6b29 DH |
493 | |
494 | set_zone_contiguous(zone); | |
815121d2 YI |
495 | } |
496 | ||
feee6b29 DH |
497 | static void __remove_section(unsigned long pfn, unsigned long nr_pages, |
498 | unsigned long map_offset, | |
499 | struct vmem_altmap *altmap) | |
ea01ea93 | 500 | { |
10404901 | 501 | struct mem_section *ms = __pfn_to_section(pfn); |
ea01ea93 | 502 | |
9d1d887d DH |
503 | if (WARN_ON_ONCE(!valid_section(ms))) |
504 | return; | |
ea01ea93 | 505 | |
ba72b4c8 | 506 | sparse_remove_section(ms, pfn, nr_pages, map_offset, altmap); |
ea01ea93 BP |
507 | } |
508 | ||
ea01ea93 | 509 | /** |
feee6b29 | 510 | * __remove_pages() - remove sections of pages |
7ea62160 | 511 | * @pfn: starting pageframe (must be aligned to start of a section) |
ea01ea93 | 512 | * @nr_pages: number of pages to remove (must be multiple of section size) |
e8b098fc | 513 | * @altmap: alternative device page map or %NULL if default memmap is used |
ea01ea93 BP |
514 | * |
515 | * Generic helper function to remove section mappings and sysfs entries | |
516 | * for the section of the memory we are removing. Caller needs to make | |
517 | * sure that pages are marked reserved and zones are adjust properly by | |
518 | * calling offline_pages(). | |
519 | */ | |
feee6b29 DH |
520 | void __remove_pages(unsigned long pfn, unsigned long nr_pages, |
521 | struct vmem_altmap *altmap) | |
ea01ea93 | 522 | { |
52fb87c8 DH |
523 | const unsigned long end_pfn = pfn + nr_pages; |
524 | unsigned long cur_nr_pages; | |
4b94ffdc | 525 | unsigned long map_offset = 0; |
4b94ffdc | 526 | |
96da4350 | 527 | map_offset = vmem_altmap_offset(altmap); |
ea01ea93 | 528 | |
7ea62160 DW |
529 | if (check_pfn_span(pfn, nr_pages, "remove")) |
530 | return; | |
ea01ea93 | 531 | |
52fb87c8 | 532 | for (; pfn < end_pfn; pfn += cur_nr_pages) { |
dd33ad7b | 533 | cond_resched(); |
52fb87c8 | 534 | /* Select all remaining pages up to the next section boundary */ |
a11b9419 DH |
535 | cur_nr_pages = min(end_pfn - pfn, |
536 | SECTION_ALIGN_UP(pfn + 1) - pfn); | |
52fb87c8 | 537 | __remove_section(pfn, cur_nr_pages, map_offset, altmap); |
4b94ffdc | 538 | map_offset = 0; |
ea01ea93 | 539 | } |
ea01ea93 | 540 | } |
ea01ea93 | 541 | |
9d0ad8ca DK |
542 | int set_online_page_callback(online_page_callback_t callback) |
543 | { | |
544 | int rc = -EINVAL; | |
545 | ||
bfc8c901 VD |
546 | get_online_mems(); |
547 | mutex_lock(&online_page_callback_lock); | |
9d0ad8ca DK |
548 | |
549 | if (online_page_callback == generic_online_page) { | |
550 | online_page_callback = callback; | |
551 | rc = 0; | |
552 | } | |
553 | ||
bfc8c901 VD |
554 | mutex_unlock(&online_page_callback_lock); |
555 | put_online_mems(); | |
9d0ad8ca DK |
556 | |
557 | return rc; | |
558 | } | |
559 | EXPORT_SYMBOL_GPL(set_online_page_callback); | |
560 | ||
561 | int restore_online_page_callback(online_page_callback_t callback) | |
562 | { | |
563 | int rc = -EINVAL; | |
564 | ||
bfc8c901 VD |
565 | get_online_mems(); |
566 | mutex_lock(&online_page_callback_lock); | |
9d0ad8ca DK |
567 | |
568 | if (online_page_callback == callback) { | |
569 | online_page_callback = generic_online_page; | |
570 | rc = 0; | |
571 | } | |
572 | ||
bfc8c901 VD |
573 | mutex_unlock(&online_page_callback_lock); |
574 | put_online_mems(); | |
9d0ad8ca DK |
575 | |
576 | return rc; | |
577 | } | |
578 | EXPORT_SYMBOL_GPL(restore_online_page_callback); | |
579 | ||
18db1491 | 580 | void generic_online_page(struct page *page, unsigned int order) |
9d0ad8ca | 581 | { |
c87cbc1f VB |
582 | /* |
583 | * Freeing the page with debug_pagealloc enabled will try to unmap it, | |
584 | * so we should map it first. This is better than introducing a special | |
585 | * case in page freeing fast path. | |
586 | */ | |
587 | if (debug_pagealloc_enabled_static()) | |
588 | kernel_map_pages(page, 1 << order, 1); | |
a9cd410a AK |
589 | __free_pages_core(page, order); |
590 | totalram_pages_add(1UL << order); | |
591 | #ifdef CONFIG_HIGHMEM | |
592 | if (PageHighMem(page)) | |
593 | totalhigh_pages_add(1UL << order); | |
594 | #endif | |
595 | } | |
18db1491 | 596 | EXPORT_SYMBOL_GPL(generic_online_page); |
a9cd410a | 597 | |
75884fb1 KH |
598 | static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages, |
599 | void *arg) | |
3947be19 | 600 | { |
b2c2ab20 DH |
601 | const unsigned long end_pfn = start_pfn + nr_pages; |
602 | unsigned long pfn; | |
603 | int order; | |
604 | ||
605 | /* | |
606 | * Online the pages. The callback might decide to keep some pages | |
607 | * PG_reserved (to add them to the buddy later), but we still account | |
608 | * them as being online/belonging to this zone ("present"). | |
609 | */ | |
610 | for (pfn = start_pfn; pfn < end_pfn; pfn += 1ul << order) { | |
611 | order = min(MAX_ORDER - 1, get_order(PFN_PHYS(end_pfn - pfn))); | |
bd02cc01 DH |
612 | /* __free_pages_core() wants pfns to be aligned to the order */ |
613 | if (WARN_ON_ONCE(!IS_ALIGNED(pfn, 1ul << order))) | |
614 | order = 0; | |
b2c2ab20 DH |
615 | (*online_page_callback)(pfn_to_page(pfn), order); |
616 | } | |
2d070eab | 617 | |
b2c2ab20 DH |
618 | /* mark all involved sections as online */ |
619 | online_mem_sections(start_pfn, end_pfn); | |
2d070eab | 620 | |
b2c2ab20 | 621 | *(unsigned long *)arg += nr_pages; |
75884fb1 KH |
622 | return 0; |
623 | } | |
624 | ||
d9713679 LJ |
625 | /* check which state of node_states will be changed when online memory */ |
626 | static void node_states_check_changes_online(unsigned long nr_pages, | |
627 | struct zone *zone, struct memory_notify *arg) | |
628 | { | |
629 | int nid = zone_to_nid(zone); | |
d9713679 | 630 | |
98fa15f3 AK |
631 | arg->status_change_nid = NUMA_NO_NODE; |
632 | arg->status_change_nid_normal = NUMA_NO_NODE; | |
633 | arg->status_change_nid_high = NUMA_NO_NODE; | |
d9713679 | 634 | |
8efe33f4 OS |
635 | if (!node_state(nid, N_MEMORY)) |
636 | arg->status_change_nid = nid; | |
637 | if (zone_idx(zone) <= ZONE_NORMAL && !node_state(nid, N_NORMAL_MEMORY)) | |
d9713679 | 638 | arg->status_change_nid_normal = nid; |
6715ddf9 | 639 | #ifdef CONFIG_HIGHMEM |
d3ba3ae1 | 640 | if (zone_idx(zone) <= ZONE_HIGHMEM && !node_state(nid, N_HIGH_MEMORY)) |
6715ddf9 | 641 | arg->status_change_nid_high = nid; |
6715ddf9 | 642 | #endif |
d9713679 LJ |
643 | } |
644 | ||
645 | static void node_states_set_node(int node, struct memory_notify *arg) | |
646 | { | |
647 | if (arg->status_change_nid_normal >= 0) | |
648 | node_set_state(node, N_NORMAL_MEMORY); | |
649 | ||
6715ddf9 LJ |
650 | if (arg->status_change_nid_high >= 0) |
651 | node_set_state(node, N_HIGH_MEMORY); | |
652 | ||
83d83612 OS |
653 | if (arg->status_change_nid >= 0) |
654 | node_set_state(node, N_MEMORY); | |
d9713679 LJ |
655 | } |
656 | ||
f1dd2cd1 MH |
657 | static void __meminit resize_zone_range(struct zone *zone, unsigned long start_pfn, |
658 | unsigned long nr_pages) | |
659 | { | |
660 | unsigned long old_end_pfn = zone_end_pfn(zone); | |
661 | ||
662 | if (zone_is_empty(zone) || start_pfn < zone->zone_start_pfn) | |
663 | zone->zone_start_pfn = start_pfn; | |
664 | ||
665 | zone->spanned_pages = max(start_pfn + nr_pages, old_end_pfn) - zone->zone_start_pfn; | |
666 | } | |
667 | ||
668 | static void __meminit resize_pgdat_range(struct pglist_data *pgdat, unsigned long start_pfn, | |
669 | unsigned long nr_pages) | |
670 | { | |
671 | unsigned long old_end_pfn = pgdat_end_pfn(pgdat); | |
672 | ||
673 | if (!pgdat->node_spanned_pages || start_pfn < pgdat->node_start_pfn) | |
674 | pgdat->node_start_pfn = start_pfn; | |
675 | ||
676 | pgdat->node_spanned_pages = max(start_pfn + nr_pages, old_end_pfn) - pgdat->node_start_pfn; | |
f1dd2cd1 | 677 | |
3fccb74c DH |
678 | } |
679 | /* | |
680 | * Associate the pfn range with the given zone, initializing the memmaps | |
681 | * and resizing the pgdat/zone data to span the added pages. After this | |
682 | * call, all affected pages are PG_reserved. | |
683 | */ | |
a99583e7 CH |
684 | void __ref move_pfn_range_to_zone(struct zone *zone, unsigned long start_pfn, |
685 | unsigned long nr_pages, struct vmem_altmap *altmap) | |
f1dd2cd1 MH |
686 | { |
687 | struct pglist_data *pgdat = zone->zone_pgdat; | |
688 | int nid = pgdat->node_id; | |
689 | unsigned long flags; | |
df429ac0 | 690 | |
f1dd2cd1 MH |
691 | clear_zone_contiguous(zone); |
692 | ||
693 | /* TODO Huh pgdat is irqsave while zone is not. It used to be like that before */ | |
694 | pgdat_resize_lock(pgdat, &flags); | |
695 | zone_span_writelock(zone); | |
fa004ab7 WY |
696 | if (zone_is_empty(zone)) |
697 | init_currently_empty_zone(zone, start_pfn, nr_pages); | |
f1dd2cd1 MH |
698 | resize_zone_range(zone, start_pfn, nr_pages); |
699 | zone_span_writeunlock(zone); | |
700 | resize_pgdat_range(pgdat, start_pfn, nr_pages); | |
701 | pgdat_resize_unlock(pgdat, &flags); | |
702 | ||
703 | /* | |
704 | * TODO now we have a visible range of pages which are not associated | |
705 | * with their zone properly. Not nice but set_pfnblock_flags_mask | |
706 | * expects the zone spans the pfn range. All the pages in the range | |
707 | * are reserved so nobody should be touching them so we should be safe | |
708 | */ | |
a99583e7 CH |
709 | memmap_init_zone(nr_pages, nid, zone_idx(zone), start_pfn, |
710 | MEMMAP_HOTPLUG, altmap); | |
f1dd2cd1 MH |
711 | |
712 | set_zone_contiguous(zone); | |
713 | } | |
714 | ||
c246a213 MH |
715 | /* |
716 | * Returns a default kernel memory zone for the given pfn range. | |
717 | * If no kernel zone covers this pfn range it will automatically go | |
718 | * to the ZONE_NORMAL. | |
719 | */ | |
c6f03e29 | 720 | static struct zone *default_kernel_zone_for_pfn(int nid, unsigned long start_pfn, |
c246a213 MH |
721 | unsigned long nr_pages) |
722 | { | |
723 | struct pglist_data *pgdat = NODE_DATA(nid); | |
724 | int zid; | |
725 | ||
726 | for (zid = 0; zid <= ZONE_NORMAL; zid++) { | |
727 | struct zone *zone = &pgdat->node_zones[zid]; | |
728 | ||
729 | if (zone_intersects(zone, start_pfn, nr_pages)) | |
730 | return zone; | |
731 | } | |
732 | ||
733 | return &pgdat->node_zones[ZONE_NORMAL]; | |
734 | } | |
735 | ||
c6f03e29 MH |
736 | static inline struct zone *default_zone_for_pfn(int nid, unsigned long start_pfn, |
737 | unsigned long nr_pages) | |
e5e68930 | 738 | { |
c6f03e29 MH |
739 | struct zone *kernel_zone = default_kernel_zone_for_pfn(nid, start_pfn, |
740 | nr_pages); | |
741 | struct zone *movable_zone = &NODE_DATA(nid)->node_zones[ZONE_MOVABLE]; | |
742 | bool in_kernel = zone_intersects(kernel_zone, start_pfn, nr_pages); | |
743 | bool in_movable = zone_intersects(movable_zone, start_pfn, nr_pages); | |
e5e68930 MH |
744 | |
745 | /* | |
c6f03e29 MH |
746 | * We inherit the existing zone in a simple case where zones do not |
747 | * overlap in the given range | |
e5e68930 | 748 | */ |
c6f03e29 MH |
749 | if (in_kernel ^ in_movable) |
750 | return (in_kernel) ? kernel_zone : movable_zone; | |
9f123ab5 | 751 | |
c6f03e29 MH |
752 | /* |
753 | * If the range doesn't belong to any zone or two zones overlap in the | |
754 | * given range then we use movable zone only if movable_node is | |
755 | * enabled because we always online to a kernel zone by default. | |
756 | */ | |
757 | return movable_node_enabled ? movable_zone : kernel_zone; | |
9f123ab5 MH |
758 | } |
759 | ||
e5e68930 MH |
760 | struct zone * zone_for_pfn_range(int online_type, int nid, unsigned start_pfn, |
761 | unsigned long nr_pages) | |
f1dd2cd1 | 762 | { |
c6f03e29 MH |
763 | if (online_type == MMOP_ONLINE_KERNEL) |
764 | return default_kernel_zone_for_pfn(nid, start_pfn, nr_pages); | |
f1dd2cd1 | 765 | |
c6f03e29 MH |
766 | if (online_type == MMOP_ONLINE_MOVABLE) |
767 | return &NODE_DATA(nid)->node_zones[ZONE_MOVABLE]; | |
df429ac0 | 768 | |
c6f03e29 | 769 | return default_zone_for_pfn(nid, start_pfn, nr_pages); |
e5e68930 MH |
770 | } |
771 | ||
bd5c2344 DH |
772 | int __ref online_pages(unsigned long pfn, unsigned long nr_pages, |
773 | int online_type, int nid) | |
75884fb1 | 774 | { |
aa47228a | 775 | unsigned long flags; |
3947be19 DH |
776 | unsigned long onlined_pages = 0; |
777 | struct zone *zone; | |
6811378e | 778 | int need_zonelists_rebuild = 0; |
7b78d335 YG |
779 | int ret; |
780 | struct memory_notify arg; | |
d0dc12e8 | 781 | |
381eab4a DH |
782 | mem_hotplug_begin(); |
783 | ||
f1dd2cd1 | 784 | /* associate pfn range with the zone */ |
3fccb74c DH |
785 | zone = zone_for_pfn_range(online_type, nid, pfn, nr_pages); |
786 | move_pfn_range_to_zone(zone, pfn, nr_pages, NULL); | |
f1dd2cd1 | 787 | |
7b78d335 YG |
788 | arg.start_pfn = pfn; |
789 | arg.nr_pages = nr_pages; | |
d9713679 | 790 | node_states_check_changes_online(nr_pages, zone, &arg); |
7b78d335 | 791 | |
7b78d335 YG |
792 | ret = memory_notify(MEM_GOING_ONLINE, &arg); |
793 | ret = notifier_to_errno(ret); | |
e33e33b4 CY |
794 | if (ret) |
795 | goto failed_addition; | |
796 | ||
6811378e YG |
797 | /* |
798 | * If this zone is not populated, then it is not in zonelist. | |
799 | * This means the page allocator ignores this zone. | |
800 | * So, zonelist must be updated after online. | |
801 | */ | |
6dcd73d7 | 802 | if (!populated_zone(zone)) { |
6811378e | 803 | need_zonelists_rebuild = 1; |
72675e13 | 804 | setup_zone_pageset(zone); |
6dcd73d7 | 805 | } |
6811378e | 806 | |
908eedc6 | 807 | ret = walk_system_ram_range(pfn, nr_pages, &onlined_pages, |
75884fb1 | 808 | online_pages_range); |
fd8a4221 | 809 | if (ret) { |
ca9a46f8 | 810 | /* not a single memory resource was applicable */ |
6dcd73d7 WC |
811 | if (need_zonelists_rebuild) |
812 | zone_pcp_reset(zone); | |
e33e33b4 | 813 | goto failed_addition; |
fd8a4221 GL |
814 | } |
815 | ||
3947be19 | 816 | zone->present_pages += onlined_pages; |
aa47228a CS |
817 | |
818 | pgdat_resize_lock(zone->zone_pgdat, &flags); | |
f2937be5 | 819 | zone->zone_pgdat->node_present_pages += onlined_pages; |
aa47228a CS |
820 | pgdat_resize_unlock(zone->zone_pgdat, &flags); |
821 | ||
e900a918 DW |
822 | shuffle_zone(zone); |
823 | ||
ca9a46f8 DH |
824 | node_states_set_node(nid, &arg); |
825 | if (need_zonelists_rebuild) | |
826 | build_all_zonelists(NULL); | |
827 | else | |
828 | zone_pcp_update(zone); | |
3947be19 | 829 | |
1b79acc9 KM |
830 | init_per_zone_wmark_min(); |
831 | ||
ca9a46f8 DH |
832 | kswapd_run(nid); |
833 | kcompactd_run(nid); | |
61b13993 | 834 | |
1f522509 | 835 | vm_total_pages = nr_free_pagecache_pages(); |
2f7f24ec | 836 | |
2d1d43f6 | 837 | writeback_set_ratelimit(); |
7b78d335 | 838 | |
ca9a46f8 | 839 | memory_notify(MEM_ONLINE, &arg); |
381eab4a | 840 | mem_hotplug_done(); |
30467e0b | 841 | return 0; |
e33e33b4 CY |
842 | |
843 | failed_addition: | |
844 | pr_debug("online_pages [mem %#010llx-%#010llx] failed\n", | |
845 | (unsigned long long) pfn << PAGE_SHIFT, | |
846 | (((unsigned long long) pfn + nr_pages) << PAGE_SHIFT) - 1); | |
847 | memory_notify(MEM_CANCEL_ONLINE, &arg); | |
feee6b29 | 848 | remove_pfn_range_from_zone(zone, pfn, nr_pages); |
381eab4a | 849 | mem_hotplug_done(); |
e33e33b4 | 850 | return ret; |
3947be19 | 851 | } |
53947027 | 852 | #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */ |
bc02af93 | 853 | |
0bd85420 TC |
854 | static void reset_node_present_pages(pg_data_t *pgdat) |
855 | { | |
856 | struct zone *z; | |
857 | ||
858 | for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++) | |
859 | z->present_pages = 0; | |
860 | ||
861 | pgdat->node_present_pages = 0; | |
862 | } | |
863 | ||
e1319331 HS |
864 | /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */ |
865 | static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start) | |
9af3c2de YG |
866 | { |
867 | struct pglist_data *pgdat; | |
c8e861a5 | 868 | unsigned long start_pfn = PFN_DOWN(start); |
9af3c2de | 869 | |
a1e565aa TC |
870 | pgdat = NODE_DATA(nid); |
871 | if (!pgdat) { | |
872 | pgdat = arch_alloc_nodedata(nid); | |
873 | if (!pgdat) | |
874 | return NULL; | |
9af3c2de | 875 | |
33fce011 WY |
876 | pgdat->per_cpu_nodestats = |
877 | alloc_percpu(struct per_cpu_nodestat); | |
a1e565aa | 878 | arch_refresh_nodedata(nid, pgdat); |
b0dc3a34 | 879 | } else { |
33fce011 | 880 | int cpu; |
e716f2eb MG |
881 | /* |
882 | * Reset the nr_zones, order and classzone_idx before reuse. | |
883 | * Note that kswapd will init kswapd_classzone_idx properly | |
884 | * when it starts in the near future. | |
885 | */ | |
b0dc3a34 | 886 | pgdat->nr_zones = 0; |
38087d9b MG |
887 | pgdat->kswapd_order = 0; |
888 | pgdat->kswapd_classzone_idx = 0; | |
33fce011 WY |
889 | for_each_online_cpu(cpu) { |
890 | struct per_cpu_nodestat *p; | |
891 | ||
892 | p = per_cpu_ptr(pgdat->per_cpu_nodestats, cpu); | |
893 | memset(p, 0, sizeof(*p)); | |
894 | } | |
a1e565aa | 895 | } |
9af3c2de YG |
896 | |
897 | /* we can use NODE_DATA(nid) from here */ | |
898 | ||
03e85f9d OS |
899 | pgdat->node_id = nid; |
900 | pgdat->node_start_pfn = start_pfn; | |
901 | ||
9af3c2de | 902 | /* init node's zones as empty zones, we don't have any present pages.*/ |
03e85f9d | 903 | free_area_init_core_hotplug(nid); |
9af3c2de | 904 | |
959ecc48 KH |
905 | /* |
906 | * The node we allocated has no zone fallback lists. For avoiding | |
907 | * to access not-initialized zonelist, build here. | |
908 | */ | |
72675e13 | 909 | build_all_zonelists(pgdat); |
959ecc48 | 910 | |
0bd85420 TC |
911 | /* |
912 | * When memory is hot-added, all the memory is in offline state. So | |
913 | * clear all zones' present_pages because they will be updated in | |
914 | * online_pages() and offline_pages(). | |
915 | */ | |
03e85f9d | 916 | reset_node_managed_pages(pgdat); |
0bd85420 TC |
917 | reset_node_present_pages(pgdat); |
918 | ||
9af3c2de YG |
919 | return pgdat; |
920 | } | |
921 | ||
b9ff0360 | 922 | static void rollback_node_hotadd(int nid) |
9af3c2de | 923 | { |
b9ff0360 OS |
924 | pg_data_t *pgdat = NODE_DATA(nid); |
925 | ||
9af3c2de | 926 | arch_refresh_nodedata(nid, NULL); |
5830169f | 927 | free_percpu(pgdat->per_cpu_nodestats); |
9af3c2de | 928 | arch_free_nodedata(pgdat); |
9af3c2de YG |
929 | } |
930 | ||
0a547039 | 931 | |
01b0f197 TK |
932 | /** |
933 | * try_online_node - online a node if offlined | |
e8b098fc | 934 | * @nid: the node ID |
b9ff0360 OS |
935 | * @start: start addr of the node |
936 | * @set_node_online: Whether we want to online the node | |
cf23422b | 937 | * called by cpu_up() to online a node without onlined memory. |
b9ff0360 OS |
938 | * |
939 | * Returns: | |
940 | * 1 -> a new node has been allocated | |
941 | * 0 -> the node is already online | |
942 | * -ENOMEM -> the node could not be allocated | |
cf23422b | 943 | */ |
b9ff0360 | 944 | static int __try_online_node(int nid, u64 start, bool set_node_online) |
cf23422b | 945 | { |
b9ff0360 OS |
946 | pg_data_t *pgdat; |
947 | int ret = 1; | |
cf23422b | 948 | |
01b0f197 TK |
949 | if (node_online(nid)) |
950 | return 0; | |
951 | ||
b9ff0360 | 952 | pgdat = hotadd_new_pgdat(nid, start); |
7553e8f2 | 953 | if (!pgdat) { |
01b0f197 | 954 | pr_err("Cannot online node %d due to NULL pgdat\n", nid); |
cf23422b | 955 | ret = -ENOMEM; |
956 | goto out; | |
957 | } | |
b9ff0360 OS |
958 | |
959 | if (set_node_online) { | |
960 | node_set_online(nid); | |
961 | ret = register_one_node(nid); | |
962 | BUG_ON(ret); | |
963 | } | |
cf23422b | 964 | out: |
b9ff0360 OS |
965 | return ret; |
966 | } | |
967 | ||
968 | /* | |
969 | * Users of this function always want to online/register the node | |
970 | */ | |
971 | int try_online_node(int nid) | |
972 | { | |
973 | int ret; | |
974 | ||
975 | mem_hotplug_begin(); | |
976 | ret = __try_online_node(nid, 0, true); | |
bfc8c901 | 977 | mem_hotplug_done(); |
cf23422b | 978 | return ret; |
979 | } | |
980 | ||
27356f54 TK |
981 | static int check_hotplug_memory_range(u64 start, u64 size) |
982 | { | |
ba325585 | 983 | /* memory range must be block size aligned */ |
cec3ebd0 DH |
984 | if (!size || !IS_ALIGNED(start, memory_block_size_bytes()) || |
985 | !IS_ALIGNED(size, memory_block_size_bytes())) { | |
ba325585 | 986 | pr_err("Block size [%#lx] unaligned hotplug range: start %#llx, size %#llx", |
cec3ebd0 | 987 | memory_block_size_bytes(), start, size); |
27356f54 TK |
988 | return -EINVAL; |
989 | } | |
990 | ||
991 | return 0; | |
992 | } | |
993 | ||
31bc3858 VK |
994 | static int online_memory_block(struct memory_block *mem, void *arg) |
995 | { | |
862919e5 | 996 | mem->online_type = memhp_default_online_type; |
dc18d706 | 997 | return device_online(&mem->dev); |
31bc3858 VK |
998 | } |
999 | ||
8df1d0e4 DH |
1000 | /* |
1001 | * NOTE: The caller must call lock_device_hotplug() to serialize hotplug | |
1002 | * and online/offline operations (triggered e.g. by sysfs). | |
1003 | * | |
1004 | * we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG | |
1005 | */ | |
f29d8e9c | 1006 | int __ref add_memory_resource(int nid, struct resource *res) |
bc02af93 | 1007 | { |
bfeb022f | 1008 | struct mhp_params params = { .pgprot = PAGE_KERNEL }; |
62cedb9f | 1009 | u64 start, size; |
b9ff0360 | 1010 | bool new_node = false; |
bc02af93 YG |
1011 | int ret; |
1012 | ||
62cedb9f DV |
1013 | start = res->start; |
1014 | size = resource_size(res); | |
1015 | ||
27356f54 TK |
1016 | ret = check_hotplug_memory_range(start, size); |
1017 | if (ret) | |
1018 | return ret; | |
1019 | ||
bfc8c901 | 1020 | mem_hotplug_begin(); |
ac13c462 | 1021 | |
7f36e3e5 TC |
1022 | /* |
1023 | * Add new range to memblock so that when hotadd_new_pgdat() is called | |
1024 | * to allocate new pgdat, get_pfn_range_for_nid() will be able to find | |
1025 | * this new range and calculate total pages correctly. The range will | |
1026 | * be removed at hot-remove time. | |
1027 | */ | |
1028 | memblock_add_node(start, size, nid); | |
1029 | ||
b9ff0360 OS |
1030 | ret = __try_online_node(nid, start, false); |
1031 | if (ret < 0) | |
1032 | goto error; | |
1033 | new_node = ret; | |
9af3c2de | 1034 | |
bc02af93 | 1035 | /* call arch's memory hotadd */ |
f5637d3b | 1036 | ret = arch_add_memory(nid, start, size, ¶ms); |
9af3c2de YG |
1037 | if (ret < 0) |
1038 | goto error; | |
1039 | ||
db051a0d DH |
1040 | /* create memory block devices after memory was added */ |
1041 | ret = create_memory_block_devices(start, size); | |
1042 | if (ret) { | |
1043 | arch_remove_memory(nid, start, size, NULL); | |
1044 | goto error; | |
1045 | } | |
1046 | ||
a1e565aa | 1047 | if (new_node) { |
d5b6f6a3 | 1048 | /* If sysfs file of new node can't be created, cpu on the node |
0fc44159 YG |
1049 | * can't be hot-added. There is no rollback way now. |
1050 | * So, check by BUG_ON() to catch it reluctantly.. | |
d5b6f6a3 | 1051 | * We online node here. We can't roll back from here. |
0fc44159 | 1052 | */ |
d5b6f6a3 OS |
1053 | node_set_online(nid); |
1054 | ret = __register_one_node(nid); | |
0fc44159 YG |
1055 | BUG_ON(ret); |
1056 | } | |
1057 | ||
d5b6f6a3 | 1058 | /* link memory sections under this node.*/ |
4fbce633 | 1059 | ret = link_mem_sections(nid, PFN_DOWN(start), PFN_UP(start + size - 1)); |
d5b6f6a3 OS |
1060 | BUG_ON(ret); |
1061 | ||
d96ae530 | 1062 | /* create new memmap entry */ |
1063 | firmware_map_add_hotplug(start, start + size, "System RAM"); | |
1064 | ||
381eab4a DH |
1065 | /* device_online() will take the lock when calling online_pages() */ |
1066 | mem_hotplug_done(); | |
1067 | ||
31bc3858 | 1068 | /* online pages if requested */ |
862919e5 | 1069 | if (memhp_default_online_type != MMOP_OFFLINE) |
fbcf73ce | 1070 | walk_memory_blocks(start, size, NULL, online_memory_block); |
31bc3858 | 1071 | |
381eab4a | 1072 | return ret; |
9af3c2de YG |
1073 | error: |
1074 | /* rollback pgdat allocation and others */ | |
b9ff0360 OS |
1075 | if (new_node) |
1076 | rollback_node_hotadd(nid); | |
7f36e3e5 | 1077 | memblock_remove(start, size); |
bfc8c901 | 1078 | mem_hotplug_done(); |
bc02af93 YG |
1079 | return ret; |
1080 | } | |
62cedb9f | 1081 | |
8df1d0e4 DH |
1082 | /* requires device_hotplug_lock, see add_memory_resource() */ |
1083 | int __ref __add_memory(int nid, u64 start, u64 size) | |
62cedb9f DV |
1084 | { |
1085 | struct resource *res; | |
1086 | int ret; | |
1087 | ||
1088 | res = register_memory_resource(start, size); | |
6f754ba4 VK |
1089 | if (IS_ERR(res)) |
1090 | return PTR_ERR(res); | |
62cedb9f | 1091 | |
f29d8e9c | 1092 | ret = add_memory_resource(nid, res); |
62cedb9f DV |
1093 | if (ret < 0) |
1094 | release_memory_resource(res); | |
1095 | return ret; | |
1096 | } | |
8df1d0e4 DH |
1097 | |
1098 | int add_memory(int nid, u64 start, u64 size) | |
1099 | { | |
1100 | int rc; | |
1101 | ||
1102 | lock_device_hotplug(); | |
1103 | rc = __add_memory(nid, start, size); | |
1104 | unlock_device_hotplug(); | |
1105 | ||
1106 | return rc; | |
1107 | } | |
bc02af93 | 1108 | EXPORT_SYMBOL_GPL(add_memory); |
0c0e6195 KH |
1109 | |
1110 | #ifdef CONFIG_MEMORY_HOTREMOVE | |
5c755e9f BP |
1111 | /* |
1112 | * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy | |
1113 | * set and the size of the free page is given by page_order(). Using this, | |
1114 | * the function determines if the pageblock contains only free pages. | |
1115 | * Due to buddy contraints, a free page at least the size of a pageblock will | |
1116 | * be located at the start of the pageblock | |
1117 | */ | |
1118 | static inline int pageblock_free(struct page *page) | |
1119 | { | |
1120 | return PageBuddy(page) && page_order(page) >= pageblock_order; | |
1121 | } | |
1122 | ||
891cb2a7 MH |
1123 | /* Return the pfn of the start of the next active pageblock after a given pfn */ |
1124 | static unsigned long next_active_pageblock(unsigned long pfn) | |
5c755e9f | 1125 | { |
891cb2a7 MH |
1126 | struct page *page = pfn_to_page(pfn); |
1127 | ||
5c755e9f | 1128 | /* Ensure the starting page is pageblock-aligned */ |
891cb2a7 | 1129 | BUG_ON(pfn & (pageblock_nr_pages - 1)); |
5c755e9f | 1130 | |
5c755e9f | 1131 | /* If the entire pageblock is free, move to the end of free page */ |
0dcc48c1 KH |
1132 | if (pageblock_free(page)) { |
1133 | int order; | |
1134 | /* be careful. we don't have locks, page_order can be changed.*/ | |
1135 | order = page_order(page); | |
1136 | if ((order < MAX_ORDER) && (order >= pageblock_order)) | |
891cb2a7 | 1137 | return pfn + (1 << order); |
0dcc48c1 | 1138 | } |
5c755e9f | 1139 | |
891cb2a7 | 1140 | return pfn + pageblock_nr_pages; |
5c755e9f BP |
1141 | } |
1142 | ||
891cb2a7 | 1143 | static bool is_pageblock_removable_nolock(unsigned long pfn) |
fb52bbae | 1144 | { |
891cb2a7 | 1145 | struct page *page = pfn_to_page(pfn); |
fb52bbae | 1146 | struct zone *zone; |
fb52bbae MM |
1147 | |
1148 | /* | |
1149 | * We have to be careful here because we are iterating over memory | |
1150 | * sections which are not zone aware so we might end up outside of | |
1151 | * the zone but still within the section. | |
1152 | * We have to take care about the node as well. If the node is offline | |
1153 | * its NODE_DATA will be NULL - see page_zone. | |
1154 | */ | |
1155 | if (!node_online(page_to_nid(page))) | |
1156 | return false; | |
1157 | ||
1158 | zone = page_zone(page); | |
1159 | pfn = page_to_pfn(page); | |
1160 | if (!zone_spans_pfn(zone, pfn)) | |
1161 | return false; | |
1162 | ||
fe4c86c9 | 1163 | return !has_unmovable_pages(zone, page, MIGRATE_MOVABLE, |
756d25be | 1164 | MEMORY_OFFLINE); |
fb52bbae MM |
1165 | } |
1166 | ||
5c755e9f | 1167 | /* Checks if this range of memory is likely to be hot-removable. */ |
c98940f6 | 1168 | bool is_mem_section_removable(unsigned long start_pfn, unsigned long nr_pages) |
5c755e9f | 1169 | { |
891cb2a7 MH |
1170 | unsigned long end_pfn, pfn; |
1171 | ||
1172 | end_pfn = min(start_pfn + nr_pages, | |
1173 | zone_end_pfn(page_zone(pfn_to_page(start_pfn)))); | |
5c755e9f BP |
1174 | |
1175 | /* Check the starting page of each pageblock within the range */ | |
891cb2a7 MH |
1176 | for (pfn = start_pfn; pfn < end_pfn; pfn = next_active_pageblock(pfn)) { |
1177 | if (!is_pageblock_removable_nolock(pfn)) | |
c98940f6 | 1178 | return false; |
49ac8255 | 1179 | cond_resched(); |
5c755e9f BP |
1180 | } |
1181 | ||
1182 | /* All pageblocks in the memory block are likely to be hot-removable */ | |
c98940f6 | 1183 | return true; |
5c755e9f BP |
1184 | } |
1185 | ||
0c0e6195 | 1186 | /* |
92917998 DH |
1187 | * Confirm all pages in a range [start, end) belong to the same zone (skipping |
1188 | * memory holes). When true, return the zone. | |
0c0e6195 | 1189 | */ |
92917998 DH |
1190 | struct zone *test_pages_in_a_zone(unsigned long start_pfn, |
1191 | unsigned long end_pfn) | |
0c0e6195 | 1192 | { |
5f0f2887 | 1193 | unsigned long pfn, sec_end_pfn; |
0c0e6195 KH |
1194 | struct zone *zone = NULL; |
1195 | struct page *page; | |
1196 | int i; | |
deb88a2a | 1197 | for (pfn = start_pfn, sec_end_pfn = SECTION_ALIGN_UP(start_pfn + 1); |
0c0e6195 | 1198 | pfn < end_pfn; |
deb88a2a | 1199 | pfn = sec_end_pfn, sec_end_pfn += PAGES_PER_SECTION) { |
5f0f2887 AB |
1200 | /* Make sure the memory section is present first */ |
1201 | if (!present_section_nr(pfn_to_section_nr(pfn))) | |
0c0e6195 | 1202 | continue; |
5f0f2887 AB |
1203 | for (; pfn < sec_end_pfn && pfn < end_pfn; |
1204 | pfn += MAX_ORDER_NR_PAGES) { | |
1205 | i = 0; | |
1206 | /* This is just a CONFIG_HOLES_IN_ZONE check.*/ | |
1207 | while ((i < MAX_ORDER_NR_PAGES) && | |
1208 | !pfn_valid_within(pfn + i)) | |
1209 | i++; | |
d6d8c8a4 | 1210 | if (i == MAX_ORDER_NR_PAGES || pfn + i >= end_pfn) |
5f0f2887 | 1211 | continue; |
24feb47c MZ |
1212 | /* Check if we got outside of the zone */ |
1213 | if (zone && !zone_spans_pfn(zone, pfn + i)) | |
92917998 | 1214 | return NULL; |
5f0f2887 AB |
1215 | page = pfn_to_page(pfn + i); |
1216 | if (zone && page_zone(page) != zone) | |
92917998 | 1217 | return NULL; |
5f0f2887 AB |
1218 | zone = page_zone(page); |
1219 | } | |
0c0e6195 | 1220 | } |
deb88a2a | 1221 | |
92917998 | 1222 | return zone; |
0c0e6195 KH |
1223 | } |
1224 | ||
1225 | /* | |
0efadf48 YX |
1226 | * Scan pfn range [start,end) to find movable/migratable pages (LRU pages, |
1227 | * non-lru movable pages and hugepages). We scan pfn because it's much | |
1228 | * easier than scanning over linked list. This function returns the pfn | |
1229 | * of the first found movable page if it's found, otherwise 0. | |
0c0e6195 | 1230 | */ |
c8721bbb | 1231 | static unsigned long scan_movable_pages(unsigned long start, unsigned long end) |
0c0e6195 KH |
1232 | { |
1233 | unsigned long pfn; | |
eeb0efd0 | 1234 | |
0c0e6195 | 1235 | for (pfn = start; pfn < end; pfn++) { |
eeb0efd0 OS |
1236 | struct page *page, *head; |
1237 | unsigned long skip; | |
1238 | ||
1239 | if (!pfn_valid(pfn)) | |
1240 | continue; | |
1241 | page = pfn_to_page(pfn); | |
1242 | if (PageLRU(page)) | |
1243 | return pfn; | |
1244 | if (__PageMovable(page)) | |
1245 | return pfn; | |
1246 | ||
1247 | if (!PageHuge(page)) | |
1248 | continue; | |
1249 | head = compound_head(page); | |
39186cbe | 1250 | if (page_huge_active(head)) |
eeb0efd0 | 1251 | return pfn; |
d8c6546b | 1252 | skip = compound_nr(head) - (page - head); |
eeb0efd0 | 1253 | pfn += skip - 1; |
0c0e6195 KH |
1254 | } |
1255 | return 0; | |
1256 | } | |
1257 | ||
666feb21 | 1258 | static struct page *new_node_page(struct page *page, unsigned long private) |
394e31d2 | 1259 | { |
394e31d2 | 1260 | int nid = page_to_nid(page); |
231e97e2 | 1261 | nodemask_t nmask = node_states[N_MEMORY]; |
7f252f27 MH |
1262 | |
1263 | /* | |
1264 | * try to allocate from a different node but reuse this node if there | |
1265 | * are no other online nodes to be used (e.g. we are offlining a part | |
1266 | * of the only existing node) | |
1267 | */ | |
1268 | node_clear(nid, nmask); | |
1269 | if (nodes_empty(nmask)) | |
1270 | node_set(nid, nmask); | |
394e31d2 | 1271 | |
8b913238 | 1272 | return new_page_nodemask(page, nid, &nmask); |
394e31d2 XQ |
1273 | } |
1274 | ||
0c0e6195 KH |
1275 | static int |
1276 | do_migrate_range(unsigned long start_pfn, unsigned long end_pfn) | |
1277 | { | |
1278 | unsigned long pfn; | |
1279 | struct page *page; | |
0c0e6195 KH |
1280 | int ret = 0; |
1281 | LIST_HEAD(source); | |
1282 | ||
a85009c3 | 1283 | for (pfn = start_pfn; pfn < end_pfn; pfn++) { |
0c0e6195 KH |
1284 | if (!pfn_valid(pfn)) |
1285 | continue; | |
1286 | page = pfn_to_page(pfn); | |
c8721bbb NH |
1287 | |
1288 | if (PageHuge(page)) { | |
1289 | struct page *head = compound_head(page); | |
d8c6546b | 1290 | pfn = page_to_pfn(head) + compound_nr(head) - 1; |
daf3538a | 1291 | isolate_huge_page(head, &source); |
c8721bbb | 1292 | continue; |
94723aaf | 1293 | } else if (PageTransHuge(page)) |
8135d892 NH |
1294 | pfn = page_to_pfn(compound_head(page)) |
1295 | + hpage_nr_pages(page) - 1; | |
c8721bbb | 1296 | |
b15c8726 MH |
1297 | /* |
1298 | * HWPoison pages have elevated reference counts so the migration would | |
1299 | * fail on them. It also doesn't make any sense to migrate them in the | |
1300 | * first place. Still try to unmap such a page in case it is still mapped | |
1301 | * (e.g. current hwpoison implementation doesn't unmap KSM pages but keep | |
1302 | * the unmap as the catch all safety net). | |
1303 | */ | |
1304 | if (PageHWPoison(page)) { | |
1305 | if (WARN_ON(PageLRU(page))) | |
1306 | isolate_lru_page(page); | |
1307 | if (page_mapped(page)) | |
1308 | try_to_unmap(page, TTU_IGNORE_MLOCK | TTU_IGNORE_ACCESS); | |
1309 | continue; | |
1310 | } | |
1311 | ||
700c2a46 | 1312 | if (!get_page_unless_zero(page)) |
0c0e6195 KH |
1313 | continue; |
1314 | /* | |
0efadf48 YX |
1315 | * We can skip free pages. And we can deal with pages on |
1316 | * LRU and non-lru movable pages. | |
0c0e6195 | 1317 | */ |
0efadf48 YX |
1318 | if (PageLRU(page)) |
1319 | ret = isolate_lru_page(page); | |
1320 | else | |
1321 | ret = isolate_movable_page(page, ISOLATE_UNEVICTABLE); | |
0c0e6195 | 1322 | if (!ret) { /* Success */ |
62695a84 | 1323 | list_add_tail(&page->lru, &source); |
0efadf48 YX |
1324 | if (!__PageMovable(page)) |
1325 | inc_node_page_state(page, NR_ISOLATED_ANON + | |
9de4f22a | 1326 | page_is_file_lru(page)); |
6d9c285a | 1327 | |
0c0e6195 | 1328 | } else { |
2932c8b0 | 1329 | pr_warn("failed to isolate pfn %lx\n", pfn); |
0efadf48 | 1330 | dump_page(page, "isolation failed"); |
0c0e6195 | 1331 | } |
1723058e | 1332 | put_page(page); |
0c0e6195 | 1333 | } |
f3ab2636 | 1334 | if (!list_empty(&source)) { |
394e31d2 XQ |
1335 | /* Allocate a new page from the nearest neighbor node */ |
1336 | ret = migrate_pages(&source, new_node_page, NULL, 0, | |
9c620e2b | 1337 | MIGRATE_SYNC, MR_MEMORY_HOTPLUG); |
2932c8b0 MH |
1338 | if (ret) { |
1339 | list_for_each_entry(page, &source, lru) { | |
1340 | pr_warn("migrating pfn %lx failed ret:%d ", | |
1341 | page_to_pfn(page), ret); | |
1342 | dump_page(page, "migration failure"); | |
1343 | } | |
c8721bbb | 1344 | putback_movable_pages(&source); |
2932c8b0 | 1345 | } |
0c0e6195 | 1346 | } |
1723058e | 1347 | |
0c0e6195 KH |
1348 | return ret; |
1349 | } | |
1350 | ||
0ee5f4f3 | 1351 | /* Mark all sections offline and remove all free pages from the buddy. */ |
0c0e6195 KH |
1352 | static int |
1353 | offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages, | |
1354 | void *data) | |
1355 | { | |
5557c766 | 1356 | unsigned long *offlined_pages = (unsigned long *)data; |
0c0e6195 | 1357 | |
5557c766 MH |
1358 | *offlined_pages += __offline_isolated_pages(start, start + nr_pages); |
1359 | return 0; | |
0c0e6195 KH |
1360 | } |
1361 | ||
1362 | /* | |
1363 | * Check all pages in range, recoreded as memory resource, are isolated. | |
1364 | */ | |
1365 | static int | |
1366 | check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages, | |
1367 | void *data) | |
1368 | { | |
756d25be DH |
1369 | return test_pages_isolated(start_pfn, start_pfn + nr_pages, |
1370 | MEMORY_OFFLINE); | |
0c0e6195 KH |
1371 | } |
1372 | ||
c5320926 TC |
1373 | static int __init cmdline_parse_movable_node(char *p) |
1374 | { | |
4932381e | 1375 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
55ac590c | 1376 | movable_node_enabled = true; |
4932381e MH |
1377 | #else |
1378 | pr_warn("movable_node parameter depends on CONFIG_HAVE_MEMBLOCK_NODE_MAP to work properly\n"); | |
1379 | #endif | |
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 MH |
1520 | do { |
1521 | for (pfn = start_pfn; pfn;) { | |
1522 | if (signal_pending(current)) { | |
1523 | ret = -EINTR; | |
1524 | reason = "signal backoff"; | |
1525 | goto failed_removal_isolated; | |
1526 | } | |
72b39cfc | 1527 | |
bb8965bd MH |
1528 | cond_resched(); |
1529 | lru_add_drain_all(); | |
bb8965bd MH |
1530 | |
1531 | pfn = scan_movable_pages(pfn, end_pfn); | |
1532 | if (pfn) { | |
1533 | /* | |
1534 | * TODO: fatal migration failures should bail | |
1535 | * out | |
1536 | */ | |
1537 | do_migrate_range(pfn, end_pfn); | |
1538 | } | |
1539 | } | |
0c0e6195 | 1540 | |
bb8965bd MH |
1541 | /* |
1542 | * Dissolve free hugepages in the memory block before doing | |
1543 | * offlining actually in order to make hugetlbfs's object | |
1544 | * counting consistent. | |
1545 | */ | |
1546 | ret = dissolve_free_huge_pages(start_pfn, end_pfn); | |
1547 | if (ret) { | |
1548 | reason = "failure to dissolve huge pages"; | |
1549 | goto failed_removal_isolated; | |
1550 | } | |
1551 | /* check again */ | |
5557c766 MH |
1552 | ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn, |
1553 | NULL, check_pages_isolated_cb); | |
1554 | } while (ret); | |
72b39cfc | 1555 | |
b3834be5 | 1556 | /* Ok, all of our target is isolated. |
0c0e6195 | 1557 | We cannot do rollback at this point. */ |
5557c766 MH |
1558 | walk_system_ram_range(start_pfn, end_pfn - start_pfn, |
1559 | &offlined_pages, offline_isolated_pages_cb); | |
1560 | pr_info("Offlined Pages %ld\n", offlined_pages); | |
9b7ea46a QC |
1561 | /* |
1562 | * Onlining will reset pagetype flags and makes migrate type | |
1563 | * MOVABLE, so just need to decrease the number of isolated | |
1564 | * pageblocks zone counter here. | |
1565 | */ | |
1566 | spin_lock_irqsave(&zone->lock, flags); | |
1567 | zone->nr_isolate_pageblock -= nr_isolate_pageblock; | |
1568 | spin_unlock_irqrestore(&zone->lock, flags); | |
1569 | ||
0c0e6195 | 1570 | /* removal success */ |
3dcc0571 | 1571 | adjust_managed_page_count(pfn_to_page(start_pfn), -offlined_pages); |
0c0e6195 | 1572 | zone->present_pages -= offlined_pages; |
d702909f CS |
1573 | |
1574 | pgdat_resize_lock(zone->zone_pgdat, &flags); | |
0c0e6195 | 1575 | zone->zone_pgdat->node_present_pages -= offlined_pages; |
d702909f | 1576 | pgdat_resize_unlock(zone->zone_pgdat, &flags); |
7b78d335 | 1577 | |
1b79acc9 KM |
1578 | init_per_zone_wmark_min(); |
1579 | ||
1e8537ba | 1580 | if (!populated_zone(zone)) { |
340175b7 | 1581 | zone_pcp_reset(zone); |
72675e13 | 1582 | build_all_zonelists(NULL); |
1e8537ba XQ |
1583 | } else |
1584 | zone_pcp_update(zone); | |
340175b7 | 1585 | |
d9713679 | 1586 | node_states_clear_node(node, &arg); |
698b1b30 | 1587 | if (arg.status_change_nid >= 0) { |
8fe23e05 | 1588 | kswapd_stop(node); |
698b1b30 VB |
1589 | kcompactd_stop(node); |
1590 | } | |
bce7394a | 1591 | |
0c0e6195 KH |
1592 | vm_total_pages = nr_free_pagecache_pages(); |
1593 | writeback_set_ratelimit(); | |
7b78d335 YG |
1594 | |
1595 | memory_notify(MEM_OFFLINE, &arg); | |
feee6b29 | 1596 | remove_pfn_range_from_zone(zone, start_pfn, nr_pages); |
381eab4a | 1597 | mem_hotplug_done(); |
0c0e6195 KH |
1598 | return 0; |
1599 | ||
79605093 MH |
1600 | failed_removal_isolated: |
1601 | undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE); | |
c4efe484 | 1602 | memory_notify(MEM_CANCEL_OFFLINE, &arg); |
0c0e6195 | 1603 | failed_removal: |
79605093 | 1604 | pr_debug("memory offlining [mem %#010llx-%#010llx] failed due to %s\n", |
e33e33b4 | 1605 | (unsigned long long) start_pfn << PAGE_SHIFT, |
79605093 MH |
1606 | ((unsigned long long) end_pfn << PAGE_SHIFT) - 1, |
1607 | reason); | |
0c0e6195 | 1608 | /* pushback to free area */ |
381eab4a | 1609 | mem_hotplug_done(); |
0c0e6195 KH |
1610 | return ret; |
1611 | } | |
71088785 | 1612 | |
a16cee10 WC |
1613 | int offline_pages(unsigned long start_pfn, unsigned long nr_pages) |
1614 | { | |
ecde0f3e | 1615 | return __offline_pages(start_pfn, start_pfn + nr_pages); |
a16cee10 WC |
1616 | } |
1617 | ||
d6de9d53 | 1618 | static int check_memblock_offlined_cb(struct memory_block *mem, void *arg) |
bbc76be6 WC |
1619 | { |
1620 | int ret = !is_memblock_offlined(mem); | |
1621 | ||
349daa0f RD |
1622 | if (unlikely(ret)) { |
1623 | phys_addr_t beginpa, endpa; | |
1624 | ||
1625 | beginpa = PFN_PHYS(section_nr_to_pfn(mem->start_section_nr)); | |
b6c88d3b | 1626 | endpa = beginpa + memory_block_size_bytes() - 1; |
756a025f | 1627 | pr_warn("removing memory fails, because memory [%pa-%pa] is onlined\n", |
349daa0f | 1628 | &beginpa, &endpa); |
bbc76be6 | 1629 | |
eca499ab PT |
1630 | return -EBUSY; |
1631 | } | |
1632 | return 0; | |
bbc76be6 WC |
1633 | } |
1634 | ||
0f1cfe9d | 1635 | static int check_cpu_on_node(pg_data_t *pgdat) |
60a5a19e | 1636 | { |
60a5a19e TC |
1637 | int cpu; |
1638 | ||
1639 | for_each_present_cpu(cpu) { | |
1640 | if (cpu_to_node(cpu) == pgdat->node_id) | |
1641 | /* | |
1642 | * the cpu on this node isn't removed, and we can't | |
1643 | * offline this node. | |
1644 | */ | |
1645 | return -EBUSY; | |
1646 | } | |
1647 | ||
1648 | return 0; | |
1649 | } | |
1650 | ||
2c91f8fc DH |
1651 | static int check_no_memblock_for_node_cb(struct memory_block *mem, void *arg) |
1652 | { | |
1653 | int nid = *(int *)arg; | |
1654 | ||
1655 | /* | |
1656 | * If a memory block belongs to multiple nodes, the stored nid is not | |
1657 | * reliable. However, such blocks are always online (e.g., cannot get | |
1658 | * offlined) and, therefore, are still spanned by the node. | |
1659 | */ | |
1660 | return mem->nid == nid ? -EEXIST : 0; | |
1661 | } | |
1662 | ||
0f1cfe9d TK |
1663 | /** |
1664 | * try_offline_node | |
e8b098fc | 1665 | * @nid: the node ID |
0f1cfe9d TK |
1666 | * |
1667 | * Offline a node if all memory sections and cpus of the node are removed. | |
1668 | * | |
1669 | * NOTE: The caller must call lock_device_hotplug() to serialize hotplug | |
1670 | * and online/offline operations before this call. | |
1671 | */ | |
90b30cdc | 1672 | void try_offline_node(int nid) |
60a5a19e | 1673 | { |
d822b86a | 1674 | pg_data_t *pgdat = NODE_DATA(nid); |
2c91f8fc | 1675 | int rc; |
60a5a19e | 1676 | |
2c91f8fc DH |
1677 | /* |
1678 | * If the node still spans pages (especially ZONE_DEVICE), don't | |
1679 | * offline it. A node spans memory after move_pfn_range_to_zone(), | |
1680 | * e.g., after the memory block was onlined. | |
1681 | */ | |
1682 | if (pgdat->node_spanned_pages) | |
1683 | return; | |
60a5a19e | 1684 | |
2c91f8fc DH |
1685 | /* |
1686 | * Especially offline memory blocks might not be spanned by the | |
1687 | * node. They will get spanned by the node once they get onlined. | |
1688 | * However, they link to the node in sysfs and can get onlined later. | |
1689 | */ | |
1690 | rc = for_each_memory_block(&nid, check_no_memblock_for_node_cb); | |
1691 | if (rc) | |
60a5a19e | 1692 | return; |
60a5a19e | 1693 | |
46a3679b | 1694 | if (check_cpu_on_node(pgdat)) |
60a5a19e TC |
1695 | return; |
1696 | ||
1697 | /* | |
1698 | * all memory/cpu of this node are removed, we can offline this | |
1699 | * node now. | |
1700 | */ | |
1701 | node_set_offline(nid); | |
1702 | unregister_one_node(nid); | |
1703 | } | |
90b30cdc | 1704 | EXPORT_SYMBOL(try_offline_node); |
60a5a19e | 1705 | |
d9eb1417 DH |
1706 | static void __release_memory_resource(resource_size_t start, |
1707 | resource_size_t size) | |
1708 | { | |
1709 | int ret; | |
1710 | ||
1711 | /* | |
1712 | * When removing memory in the same granularity as it was added, | |
1713 | * this function never fails. It might only fail if resources | |
1714 | * have to be adjusted or split. We'll ignore the error, as | |
1715 | * removing of memory cannot fail. | |
1716 | */ | |
1717 | ret = release_mem_region_adjustable(&iomem_resource, start, size); | |
1718 | if (ret) { | |
1719 | resource_size_t endres = start + size - 1; | |
1720 | ||
1721 | pr_warn("Unable to release resource <%pa-%pa> (%d)\n", | |
1722 | &start, &endres, ret); | |
1723 | } | |
1724 | } | |
1725 | ||
eca499ab | 1726 | static int __ref try_remove_memory(int nid, u64 start, u64 size) |
bbc76be6 | 1727 | { |
eca499ab | 1728 | int rc = 0; |
993c1aad | 1729 | |
27356f54 TK |
1730 | BUG_ON(check_hotplug_memory_range(start, size)); |
1731 | ||
6677e3ea | 1732 | /* |
242831eb | 1733 | * All memory blocks must be offlined before removing memory. Check |
eca499ab | 1734 | * whether all memory blocks in question are offline and return error |
242831eb | 1735 | * if this is not the case. |
6677e3ea | 1736 | */ |
fbcf73ce | 1737 | rc = walk_memory_blocks(start, size, NULL, check_memblock_offlined_cb); |
eca499ab PT |
1738 | if (rc) |
1739 | goto done; | |
6677e3ea | 1740 | |
46c66c4b YI |
1741 | /* remove memmap entry */ |
1742 | firmware_map_remove(start, start + size, "System RAM"); | |
4c4b7f9b | 1743 | |
f1037ec0 DW |
1744 | /* |
1745 | * Memory block device removal under the device_hotplug_lock is | |
1746 | * a barrier against racing online attempts. | |
1747 | */ | |
4c4b7f9b | 1748 | remove_memory_block_devices(start, size); |
46c66c4b | 1749 | |
f1037ec0 DW |
1750 | mem_hotplug_begin(); |
1751 | ||
2c2a5af6 | 1752 | arch_remove_memory(nid, start, size, NULL); |
32d1fe8f AK |
1753 | memblock_free(start, size); |
1754 | memblock_remove(start, size); | |
d9eb1417 | 1755 | __release_memory_resource(start, size); |
24d335ca | 1756 | |
60a5a19e TC |
1757 | try_offline_node(nid); |
1758 | ||
eca499ab | 1759 | done: |
bfc8c901 | 1760 | mem_hotplug_done(); |
eca499ab | 1761 | return rc; |
71088785 | 1762 | } |
d15e5926 | 1763 | |
eca499ab PT |
1764 | /** |
1765 | * remove_memory | |
1766 | * @nid: the node ID | |
1767 | * @start: physical address of the region to remove | |
1768 | * @size: size of the region to remove | |
1769 | * | |
1770 | * NOTE: The caller must call lock_device_hotplug() to serialize hotplug | |
1771 | * and online/offline operations before this call, as required by | |
1772 | * try_offline_node(). | |
1773 | */ | |
1774 | void __remove_memory(int nid, u64 start, u64 size) | |
1775 | { | |
1776 | ||
1777 | /* | |
29a90db9 | 1778 | * trigger BUG() if some memory is not offlined prior to calling this |
eca499ab PT |
1779 | * function |
1780 | */ | |
1781 | if (try_remove_memory(nid, start, size)) | |
1782 | BUG(); | |
1783 | } | |
1784 | ||
1785 | /* | |
1786 | * Remove memory if every memory block is offline, otherwise return -EBUSY is | |
1787 | * some memory is not offline | |
1788 | */ | |
1789 | int remove_memory(int nid, u64 start, u64 size) | |
d15e5926 | 1790 | { |
eca499ab PT |
1791 | int rc; |
1792 | ||
d15e5926 | 1793 | lock_device_hotplug(); |
eca499ab | 1794 | rc = try_remove_memory(nid, start, size); |
d15e5926 | 1795 | unlock_device_hotplug(); |
eca499ab PT |
1796 | |
1797 | return rc; | |
d15e5926 | 1798 | } |
71088785 | 1799 | EXPORT_SYMBOL_GPL(remove_memory); |
aba6efc4 | 1800 | #endif /* CONFIG_MEMORY_HOTREMOVE */ |