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 |
31bc3858 | 70 | bool memhp_auto_online; |
8604d9e5 VK |
71 | #else |
72 | bool memhp_auto_online = true; | |
73 | #endif | |
31bc3858 VK |
74 | EXPORT_SYMBOL_GPL(memhp_auto_online); |
75 | ||
86dd995d VK |
76 | static int __init setup_memhp_default_state(char *str) |
77 | { | |
78 | if (!strcmp(str, "online")) | |
79 | memhp_auto_online = true; | |
80 | else if (!strcmp(str, "offline")) | |
81 | memhp_auto_online = false; | |
82 | ||
83 | return 1; | |
84 | } | |
85 | __setup("memhp_default_state=", setup_memhp_default_state); | |
86 | ||
30467e0b | 87 | void mem_hotplug_begin(void) |
20d6c96b | 88 | { |
3f906ba2 TG |
89 | cpus_read_lock(); |
90 | percpu_down_write(&mem_hotplug_lock); | |
20d6c96b KM |
91 | } |
92 | ||
30467e0b | 93 | void mem_hotplug_done(void) |
bfc8c901 | 94 | { |
3f906ba2 TG |
95 | percpu_up_write(&mem_hotplug_lock); |
96 | cpus_read_unlock(); | |
bfc8c901 | 97 | } |
20d6c96b | 98 | |
357b4da5 JG |
99 | u64 max_mem_size = U64_MAX; |
100 | ||
45e0b78b KM |
101 | /* add this memory to iomem resource */ |
102 | static struct resource *register_memory_resource(u64 start, u64 size) | |
103 | { | |
2794129e DH |
104 | struct resource *res; |
105 | unsigned long flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY; | |
106 | char *resource_name = "System RAM"; | |
357b4da5 JG |
107 | |
108 | if (start + size > max_mem_size) | |
109 | return ERR_PTR(-E2BIG); | |
110 | ||
2794129e DH |
111 | /* |
112 | * Request ownership of the new memory range. This might be | |
113 | * a child of an existing resource that was present but | |
114 | * not marked as busy. | |
115 | */ | |
116 | res = __request_region(&iomem_resource, start, size, | |
117 | resource_name, flags); | |
118 | ||
119 | if (!res) { | |
120 | pr_debug("Unable to reserve System RAM region: %016llx->%016llx\n", | |
121 | start, start + size); | |
6f754ba4 | 122 | return ERR_PTR(-EEXIST); |
45e0b78b KM |
123 | } |
124 | return res; | |
125 | } | |
126 | ||
127 | static void release_memory_resource(struct resource *res) | |
128 | { | |
129 | if (!res) | |
130 | return; | |
131 | release_resource(res); | |
132 | kfree(res); | |
45e0b78b KM |
133 | } |
134 | ||
53947027 | 135 | #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE |
46723bfa YI |
136 | void get_page_bootmem(unsigned long info, struct page *page, |
137 | unsigned long type) | |
04753278 | 138 | { |
ddffe98d | 139 | page->freelist = (void *)type; |
04753278 YG |
140 | SetPagePrivate(page); |
141 | set_page_private(page, info); | |
fe896d18 | 142 | page_ref_inc(page); |
04753278 YG |
143 | } |
144 | ||
170a5a7e | 145 | void put_page_bootmem(struct page *page) |
04753278 | 146 | { |
5f24ce5f | 147 | unsigned long type; |
04753278 | 148 | |
ddffe98d | 149 | type = (unsigned long) page->freelist; |
5f24ce5f AA |
150 | BUG_ON(type < MEMORY_HOTPLUG_MIN_BOOTMEM_TYPE || |
151 | type > MEMORY_HOTPLUG_MAX_BOOTMEM_TYPE); | |
04753278 | 152 | |
fe896d18 | 153 | if (page_ref_dec_return(page) == 1) { |
ddffe98d | 154 | page->freelist = NULL; |
04753278 YG |
155 | ClearPagePrivate(page); |
156 | set_page_private(page, 0); | |
5f24ce5f | 157 | INIT_LIST_HEAD(&page->lru); |
170a5a7e | 158 | free_reserved_page(page); |
04753278 | 159 | } |
04753278 YG |
160 | } |
161 | ||
46723bfa YI |
162 | #ifdef CONFIG_HAVE_BOOTMEM_INFO_NODE |
163 | #ifndef CONFIG_SPARSEMEM_VMEMMAP | |
d92bc318 | 164 | static void register_page_bootmem_info_section(unsigned long start_pfn) |
04753278 | 165 | { |
f1eca35a | 166 | unsigned long mapsize, section_nr, i; |
04753278 YG |
167 | struct mem_section *ms; |
168 | struct page *page, *memmap; | |
f1eca35a | 169 | struct mem_section_usage *usage; |
04753278 | 170 | |
04753278 YG |
171 | section_nr = pfn_to_section_nr(start_pfn); |
172 | ms = __nr_to_section(section_nr); | |
173 | ||
174 | /* Get section's memmap address */ | |
175 | memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr); | |
176 | ||
177 | /* | |
178 | * Get page for the memmap's phys address | |
179 | * XXX: need more consideration for sparse_vmemmap... | |
180 | */ | |
181 | page = virt_to_page(memmap); | |
182 | mapsize = sizeof(struct page) * PAGES_PER_SECTION; | |
183 | mapsize = PAGE_ALIGN(mapsize) >> PAGE_SHIFT; | |
184 | ||
185 | /* remember memmap's page */ | |
186 | for (i = 0; i < mapsize; i++, page++) | |
187 | get_page_bootmem(section_nr, page, SECTION_INFO); | |
188 | ||
f1eca35a DW |
189 | usage = ms->usage; |
190 | page = virt_to_page(usage); | |
04753278 | 191 | |
f1eca35a | 192 | mapsize = PAGE_ALIGN(mem_section_usage_size()) >> PAGE_SHIFT; |
04753278 YG |
193 | |
194 | for (i = 0; i < mapsize; i++, page++) | |
af370fb8 | 195 | get_page_bootmem(section_nr, page, MIX_SECTION_INFO); |
04753278 YG |
196 | |
197 | } | |
46723bfa YI |
198 | #else /* CONFIG_SPARSEMEM_VMEMMAP */ |
199 | static void register_page_bootmem_info_section(unsigned long start_pfn) | |
200 | { | |
f1eca35a | 201 | unsigned long mapsize, section_nr, i; |
46723bfa YI |
202 | struct mem_section *ms; |
203 | struct page *page, *memmap; | |
f1eca35a | 204 | struct mem_section_usage *usage; |
46723bfa | 205 | |
46723bfa YI |
206 | section_nr = pfn_to_section_nr(start_pfn); |
207 | ms = __nr_to_section(section_nr); | |
208 | ||
209 | memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr); | |
210 | ||
211 | register_page_bootmem_memmap(section_nr, memmap, PAGES_PER_SECTION); | |
212 | ||
f1eca35a DW |
213 | usage = ms->usage; |
214 | page = virt_to_page(usage); | |
46723bfa | 215 | |
f1eca35a | 216 | mapsize = PAGE_ALIGN(mem_section_usage_size()) >> PAGE_SHIFT; |
46723bfa YI |
217 | |
218 | for (i = 0; i < mapsize; i++, page++) | |
219 | get_page_bootmem(section_nr, page, MIX_SECTION_INFO); | |
220 | } | |
221 | #endif /* !CONFIG_SPARSEMEM_VMEMMAP */ | |
04753278 | 222 | |
7ded384a | 223 | void __init register_page_bootmem_info_node(struct pglist_data *pgdat) |
04753278 YG |
224 | { |
225 | unsigned long i, pfn, end_pfn, nr_pages; | |
226 | int node = pgdat->node_id; | |
227 | struct page *page; | |
04753278 YG |
228 | |
229 | nr_pages = PAGE_ALIGN(sizeof(struct pglist_data)) >> PAGE_SHIFT; | |
230 | page = virt_to_page(pgdat); | |
231 | ||
232 | for (i = 0; i < nr_pages; i++, page++) | |
233 | get_page_bootmem(node, page, NODE_INFO); | |
234 | ||
04753278 | 235 | pfn = pgdat->node_start_pfn; |
c1f19495 | 236 | end_pfn = pgdat_end_pfn(pgdat); |
04753278 | 237 | |
7e9f5eb0 | 238 | /* register section info */ |
f14851af | 239 | for (; pfn < end_pfn; pfn += PAGES_PER_SECTION) { |
240 | /* | |
241 | * Some platforms can assign the same pfn to multiple nodes - on | |
242 | * node0 as well as nodeN. To avoid registering a pfn against | |
243 | * multiple nodes we check that this pfn does not already | |
7e9f5eb0 | 244 | * reside in some other nodes. |
f14851af | 245 | */ |
f65e91df | 246 | if (pfn_valid(pfn) && (early_pfn_to_nid(pfn) == node)) |
f14851af | 247 | register_page_bootmem_info_section(pfn); |
248 | } | |
04753278 | 249 | } |
46723bfa | 250 | #endif /* CONFIG_HAVE_BOOTMEM_INFO_NODE */ |
04753278 | 251 | |
7ea62160 DW |
252 | static int check_pfn_span(unsigned long pfn, unsigned long nr_pages, |
253 | const char *reason) | |
254 | { | |
255 | /* | |
256 | * Disallow all operations smaller than a sub-section and only | |
257 | * allow operations smaller than a section for | |
258 | * SPARSEMEM_VMEMMAP. Note that check_hotplug_memory_range() | |
259 | * enforces a larger memory_block_size_bytes() granularity for | |
260 | * memory that will be marked online, so this check should only | |
261 | * fire for direct arch_{add,remove}_memory() users outside of | |
262 | * add_memory_resource(). | |
263 | */ | |
264 | unsigned long min_align; | |
265 | ||
266 | if (IS_ENABLED(CONFIG_SPARSEMEM_VMEMMAP)) | |
267 | min_align = PAGES_PER_SUBSECTION; | |
268 | else | |
269 | min_align = PAGES_PER_SECTION; | |
270 | if (!IS_ALIGNED(pfn, min_align) | |
271 | || !IS_ALIGNED(nr_pages, min_align)) { | |
272 | WARN(1, "Misaligned __%s_pages start: %#lx end: #%lx\n", | |
273 | reason, pfn, pfn + nr_pages - 1); | |
274 | return -EINVAL; | |
275 | } | |
276 | return 0; | |
277 | } | |
278 | ||
dca4436d AS |
279 | static int check_hotplug_memory_addressable(unsigned long pfn, |
280 | unsigned long nr_pages) | |
281 | { | |
282 | const u64 max_addr = PFN_PHYS(pfn + nr_pages) - 1; | |
283 | ||
284 | if (max_addr >> MAX_PHYSMEM_BITS) { | |
285 | const u64 max_allowed = (1ull << (MAX_PHYSMEM_BITS + 1)) - 1; | |
286 | WARN(1, | |
287 | "Hotplugged memory exceeds maximum addressable address, range=%#llx-%#llx, maximum=%#llx\n", | |
288 | (u64)PFN_PHYS(pfn), max_addr, max_allowed); | |
289 | return -E2BIG; | |
290 | } | |
291 | ||
292 | return 0; | |
293 | } | |
294 | ||
4edd7cef DR |
295 | /* |
296 | * Reasonably generic function for adding memory. It is | |
297 | * expected that archs that support memory hotplug will | |
298 | * call this function after deciding the zone to which to | |
299 | * add the new pages. | |
300 | */ | |
7ea62160 DW |
301 | int __ref __add_pages(int nid, unsigned long pfn, unsigned long nr_pages, |
302 | struct mhp_restrictions *restrictions) | |
4edd7cef | 303 | { |
9a845030 DW |
304 | int err; |
305 | unsigned long nr, start_sec, end_sec; | |
940519f0 | 306 | struct vmem_altmap *altmap = restrictions->altmap; |
4b94ffdc | 307 | |
dca4436d AS |
308 | err = check_hotplug_memory_addressable(pfn, nr_pages); |
309 | if (err) | |
310 | return err; | |
311 | ||
4b94ffdc DW |
312 | if (altmap) { |
313 | /* | |
314 | * Validate altmap is within bounds of the total request | |
315 | */ | |
7ea62160 | 316 | if (altmap->base_pfn != pfn |
4b94ffdc DW |
317 | || vmem_altmap_offset(altmap) > nr_pages) { |
318 | pr_warn_once("memory add fail, invalid altmap\n"); | |
7ea62160 | 319 | return -EINVAL; |
4b94ffdc DW |
320 | } |
321 | altmap->alloc = 0; | |
322 | } | |
323 | ||
7ea62160 DW |
324 | err = check_pfn_span(pfn, nr_pages, "add"); |
325 | if (err) | |
326 | return err; | |
327 | ||
328 | start_sec = pfn_to_section_nr(pfn); | |
329 | end_sec = pfn_to_section_nr(pfn + nr_pages - 1); | |
9a845030 | 330 | for (nr = start_sec; nr <= end_sec; nr++) { |
7ea62160 DW |
331 | unsigned long pfns; |
332 | ||
333 | pfns = min(nr_pages, PAGES_PER_SECTION | |
334 | - (pfn & ~PAGE_SECTION_MASK)); | |
ba72b4c8 DW |
335 | err = sparse_add_section(nid, pfn, pfns, altmap); |
336 | if (err) | |
337 | break; | |
7ea62160 DW |
338 | pfn += pfns; |
339 | nr_pages -= pfns; | |
f64ac5e6 | 340 | cond_resched(); |
4edd7cef | 341 | } |
c435a390 | 342 | vmemmap_populate_print_last(); |
4edd7cef DR |
343 | return err; |
344 | } | |
4edd7cef | 345 | |
815121d2 | 346 | /* find the smallest valid pfn in the range [start_pfn, end_pfn) */ |
d09b0137 | 347 | static unsigned long find_smallest_section_pfn(int nid, struct zone *zone, |
815121d2 YI |
348 | unsigned long start_pfn, |
349 | unsigned long end_pfn) | |
350 | { | |
49ba3c6b | 351 | for (; start_pfn < end_pfn; start_pfn += PAGES_PER_SUBSECTION) { |
7ce700bf | 352 | if (unlikely(!pfn_to_online_page(start_pfn))) |
815121d2 YI |
353 | continue; |
354 | ||
355 | if (unlikely(pfn_to_nid(start_pfn) != nid)) | |
356 | continue; | |
357 | ||
9b05158f | 358 | if (zone != page_zone(pfn_to_page(start_pfn))) |
815121d2 YI |
359 | continue; |
360 | ||
361 | return start_pfn; | |
362 | } | |
363 | ||
364 | return 0; | |
365 | } | |
366 | ||
367 | /* find the biggest valid pfn in the range [start_pfn, end_pfn). */ | |
d09b0137 | 368 | static unsigned long find_biggest_section_pfn(int nid, struct zone *zone, |
815121d2 YI |
369 | unsigned long start_pfn, |
370 | unsigned long end_pfn) | |
371 | { | |
815121d2 YI |
372 | unsigned long pfn; |
373 | ||
374 | /* pfn is the end pfn of a memory section. */ | |
375 | pfn = end_pfn - 1; | |
49ba3c6b | 376 | for (; pfn >= start_pfn; pfn -= PAGES_PER_SUBSECTION) { |
7ce700bf | 377 | if (unlikely(!pfn_to_online_page(pfn))) |
815121d2 YI |
378 | continue; |
379 | ||
380 | if (unlikely(pfn_to_nid(pfn) != nid)) | |
381 | continue; | |
382 | ||
9b05158f | 383 | if (zone != page_zone(pfn_to_page(pfn))) |
815121d2 YI |
384 | continue; |
385 | ||
386 | return pfn; | |
387 | } | |
388 | ||
389 | return 0; | |
390 | } | |
391 | ||
392 | static void shrink_zone_span(struct zone *zone, unsigned long start_pfn, | |
393 | unsigned long end_pfn) | |
394 | { | |
c33bc315 XQ |
395 | unsigned long zone_start_pfn = zone->zone_start_pfn; |
396 | unsigned long z = zone_end_pfn(zone); /* zone_end_pfn namespace clash */ | |
397 | unsigned long zone_end_pfn = z; | |
815121d2 | 398 | unsigned long pfn; |
815121d2 YI |
399 | int nid = zone_to_nid(zone); |
400 | ||
401 | zone_span_writelock(zone); | |
402 | if (zone_start_pfn == start_pfn) { | |
403 | /* | |
404 | * If the section is smallest section in the zone, it need | |
405 | * shrink zone->zone_start_pfn and zone->zone_spanned_pages. | |
406 | * In this case, we find second smallest valid mem_section | |
407 | * for shrinking zone. | |
408 | */ | |
409 | pfn = find_smallest_section_pfn(nid, zone, end_pfn, | |
410 | zone_end_pfn); | |
411 | if (pfn) { | |
412 | zone->zone_start_pfn = pfn; | |
413 | zone->spanned_pages = zone_end_pfn - pfn; | |
950b68d9 DH |
414 | } else { |
415 | zone->zone_start_pfn = 0; | |
416 | zone->spanned_pages = 0; | |
815121d2 YI |
417 | } |
418 | } else if (zone_end_pfn == end_pfn) { | |
419 | /* | |
420 | * If the section is biggest section in the zone, it need | |
421 | * shrink zone->spanned_pages. | |
422 | * In this case, we find second biggest valid mem_section for | |
423 | * shrinking zone. | |
424 | */ | |
425 | pfn = find_biggest_section_pfn(nid, zone, zone_start_pfn, | |
426 | start_pfn); | |
427 | if (pfn) | |
428 | zone->spanned_pages = pfn - zone_start_pfn + 1; | |
950b68d9 DH |
429 | else { |
430 | zone->zone_start_pfn = 0; | |
431 | zone->spanned_pages = 0; | |
432 | } | |
815121d2 | 433 | } |
815121d2 YI |
434 | zone_span_writeunlock(zone); |
435 | } | |
436 | ||
00d6c019 | 437 | static void update_pgdat_span(struct pglist_data *pgdat) |
815121d2 | 438 | { |
00d6c019 DH |
439 | unsigned long node_start_pfn = 0, node_end_pfn = 0; |
440 | struct zone *zone; | |
441 | ||
442 | for (zone = pgdat->node_zones; | |
443 | zone < pgdat->node_zones + MAX_NR_ZONES; zone++) { | |
444 | unsigned long zone_end_pfn = zone->zone_start_pfn + | |
445 | zone->spanned_pages; | |
446 | ||
447 | /* No need to lock the zones, they can't change. */ | |
656d5711 DH |
448 | if (!zone->spanned_pages) |
449 | continue; | |
450 | if (!node_end_pfn) { | |
451 | node_start_pfn = zone->zone_start_pfn; | |
452 | node_end_pfn = zone_end_pfn; | |
453 | continue; | |
454 | } | |
455 | ||
00d6c019 DH |
456 | if (zone_end_pfn > node_end_pfn) |
457 | node_end_pfn = zone_end_pfn; | |
458 | if (zone->zone_start_pfn < node_start_pfn) | |
459 | node_start_pfn = zone->zone_start_pfn; | |
815121d2 YI |
460 | } |
461 | ||
00d6c019 DH |
462 | pgdat->node_start_pfn = node_start_pfn; |
463 | pgdat->node_spanned_pages = node_end_pfn - node_start_pfn; | |
815121d2 YI |
464 | } |
465 | ||
feee6b29 DH |
466 | void __ref remove_pfn_range_from_zone(struct zone *zone, |
467 | unsigned long start_pfn, | |
468 | unsigned long nr_pages) | |
815121d2 YI |
469 | { |
470 | struct pglist_data *pgdat = zone->zone_pgdat; | |
815121d2 YI |
471 | unsigned long flags; |
472 | ||
d33695b1 DH |
473 | /* Poison struct pages because they are now uninitialized again. */ |
474 | page_init_poison(pfn_to_page(start_pfn), sizeof(struct page) * nr_pages); | |
475 | ||
7ce700bf DH |
476 | #ifdef CONFIG_ZONE_DEVICE |
477 | /* | |
478 | * Zone shrinking code cannot properly deal with ZONE_DEVICE. So | |
479 | * we will not try to shrink the zones - which is okay as | |
480 | * set_zone_contiguous() cannot deal with ZONE_DEVICE either way. | |
481 | */ | |
482 | if (zone_idx(zone) == ZONE_DEVICE) | |
483 | return; | |
484 | #endif | |
485 | ||
feee6b29 DH |
486 | clear_zone_contiguous(zone); |
487 | ||
815121d2 YI |
488 | pgdat_resize_lock(zone->zone_pgdat, &flags); |
489 | shrink_zone_span(zone, start_pfn, start_pfn + nr_pages); | |
00d6c019 | 490 | update_pgdat_span(pgdat); |
815121d2 | 491 | pgdat_resize_unlock(zone->zone_pgdat, &flags); |
feee6b29 DH |
492 | |
493 | set_zone_contiguous(zone); | |
815121d2 YI |
494 | } |
495 | ||
feee6b29 DH |
496 | static void __remove_section(unsigned long pfn, unsigned long nr_pages, |
497 | unsigned long map_offset, | |
498 | struct vmem_altmap *altmap) | |
ea01ea93 | 499 | { |
7ea62160 | 500 | struct mem_section *ms = __nr_to_section(pfn_to_section_nr(pfn)); |
ea01ea93 | 501 | |
9d1d887d DH |
502 | if (WARN_ON_ONCE(!valid_section(ms))) |
503 | return; | |
ea01ea93 | 504 | |
ba72b4c8 | 505 | sparse_remove_section(ms, pfn, nr_pages, map_offset, altmap); |
ea01ea93 BP |
506 | } |
507 | ||
ea01ea93 | 508 | /** |
feee6b29 | 509 | * __remove_pages() - remove sections of pages |
7ea62160 | 510 | * @pfn: starting pageframe (must be aligned to start of a section) |
ea01ea93 | 511 | * @nr_pages: number of pages to remove (must be multiple of section size) |
e8b098fc | 512 | * @altmap: alternative device page map or %NULL if default memmap is used |
ea01ea93 BP |
513 | * |
514 | * Generic helper function to remove section mappings and sysfs entries | |
515 | * for the section of the memory we are removing. Caller needs to make | |
516 | * sure that pages are marked reserved and zones are adjust properly by | |
517 | * calling offline_pages(). | |
518 | */ | |
feee6b29 DH |
519 | void __remove_pages(unsigned long pfn, unsigned long nr_pages, |
520 | struct vmem_altmap *altmap) | |
ea01ea93 | 521 | { |
4b94ffdc | 522 | unsigned long map_offset = 0; |
9a845030 | 523 | unsigned long nr, start_sec, end_sec; |
4b94ffdc | 524 | |
96da4350 | 525 | map_offset = vmem_altmap_offset(altmap); |
ea01ea93 | 526 | |
7ea62160 DW |
527 | if (check_pfn_span(pfn, nr_pages, "remove")) |
528 | return; | |
ea01ea93 | 529 | |
7ea62160 DW |
530 | start_sec = pfn_to_section_nr(pfn); |
531 | end_sec = pfn_to_section_nr(pfn + nr_pages - 1); | |
9a845030 | 532 | for (nr = start_sec; nr <= end_sec; nr++) { |
7ea62160 | 533 | unsigned long pfns; |
4b94ffdc | 534 | |
dd33ad7b | 535 | cond_resched(); |
7ea62160 DW |
536 | pfns = min(nr_pages, PAGES_PER_SECTION |
537 | - (pfn & ~PAGE_SECTION_MASK)); | |
feee6b29 | 538 | __remove_section(pfn, pfns, map_offset, altmap); |
7ea62160 DW |
539 | pfn += pfns; |
540 | nr_pages -= pfns; | |
4b94ffdc | 541 | map_offset = 0; |
ea01ea93 | 542 | } |
ea01ea93 | 543 | } |
ea01ea93 | 544 | |
9d0ad8ca DK |
545 | int set_online_page_callback(online_page_callback_t callback) |
546 | { | |
547 | int rc = -EINVAL; | |
548 | ||
bfc8c901 VD |
549 | get_online_mems(); |
550 | mutex_lock(&online_page_callback_lock); | |
9d0ad8ca DK |
551 | |
552 | if (online_page_callback == generic_online_page) { | |
553 | online_page_callback = callback; | |
554 | rc = 0; | |
555 | } | |
556 | ||
bfc8c901 VD |
557 | mutex_unlock(&online_page_callback_lock); |
558 | put_online_mems(); | |
9d0ad8ca DK |
559 | |
560 | return rc; | |
561 | } | |
562 | EXPORT_SYMBOL_GPL(set_online_page_callback); | |
563 | ||
564 | int restore_online_page_callback(online_page_callback_t callback) | |
565 | { | |
566 | int rc = -EINVAL; | |
567 | ||
bfc8c901 VD |
568 | get_online_mems(); |
569 | mutex_lock(&online_page_callback_lock); | |
9d0ad8ca DK |
570 | |
571 | if (online_page_callback == callback) { | |
572 | online_page_callback = generic_online_page; | |
573 | rc = 0; | |
574 | } | |
575 | ||
bfc8c901 VD |
576 | mutex_unlock(&online_page_callback_lock); |
577 | put_online_mems(); | |
9d0ad8ca DK |
578 | |
579 | return rc; | |
580 | } | |
581 | EXPORT_SYMBOL_GPL(restore_online_page_callback); | |
582 | ||
18db1491 | 583 | void generic_online_page(struct page *page, unsigned int order) |
9d0ad8ca | 584 | { |
cd02cf1a | 585 | kernel_map_pages(page, 1 << order, 1); |
a9cd410a AK |
586 | __free_pages_core(page, order); |
587 | totalram_pages_add(1UL << order); | |
588 | #ifdef CONFIG_HIGHMEM | |
589 | if (PageHighMem(page)) | |
590 | totalhigh_pages_add(1UL << order); | |
591 | #endif | |
592 | } | |
18db1491 | 593 | EXPORT_SYMBOL_GPL(generic_online_page); |
a9cd410a | 594 | |
75884fb1 KH |
595 | static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages, |
596 | void *arg) | |
3947be19 | 597 | { |
b2c2ab20 DH |
598 | const unsigned long end_pfn = start_pfn + nr_pages; |
599 | unsigned long pfn; | |
600 | int order; | |
601 | ||
602 | /* | |
603 | * Online the pages. The callback might decide to keep some pages | |
604 | * PG_reserved (to add them to the buddy later), but we still account | |
605 | * them as being online/belonging to this zone ("present"). | |
606 | */ | |
607 | for (pfn = start_pfn; pfn < end_pfn; pfn += 1ul << order) { | |
608 | order = min(MAX_ORDER - 1, get_order(PFN_PHYS(end_pfn - pfn))); | |
bd02cc01 DH |
609 | /* __free_pages_core() wants pfns to be aligned to the order */ |
610 | if (WARN_ON_ONCE(!IS_ALIGNED(pfn, 1ul << order))) | |
611 | order = 0; | |
b2c2ab20 DH |
612 | (*online_page_callback)(pfn_to_page(pfn), order); |
613 | } | |
2d070eab | 614 | |
b2c2ab20 DH |
615 | /* mark all involved sections as online */ |
616 | online_mem_sections(start_pfn, end_pfn); | |
2d070eab | 617 | |
b2c2ab20 | 618 | *(unsigned long *)arg += nr_pages; |
75884fb1 KH |
619 | return 0; |
620 | } | |
621 | ||
d9713679 LJ |
622 | /* check which state of node_states will be changed when online memory */ |
623 | static void node_states_check_changes_online(unsigned long nr_pages, | |
624 | struct zone *zone, struct memory_notify *arg) | |
625 | { | |
626 | int nid = zone_to_nid(zone); | |
d9713679 | 627 | |
98fa15f3 AK |
628 | arg->status_change_nid = NUMA_NO_NODE; |
629 | arg->status_change_nid_normal = NUMA_NO_NODE; | |
630 | arg->status_change_nid_high = NUMA_NO_NODE; | |
d9713679 | 631 | |
8efe33f4 OS |
632 | if (!node_state(nid, N_MEMORY)) |
633 | arg->status_change_nid = nid; | |
634 | if (zone_idx(zone) <= ZONE_NORMAL && !node_state(nid, N_NORMAL_MEMORY)) | |
d9713679 | 635 | arg->status_change_nid_normal = nid; |
6715ddf9 | 636 | #ifdef CONFIG_HIGHMEM |
d3ba3ae1 | 637 | if (zone_idx(zone) <= ZONE_HIGHMEM && !node_state(nid, N_HIGH_MEMORY)) |
6715ddf9 | 638 | arg->status_change_nid_high = nid; |
6715ddf9 | 639 | #endif |
d9713679 LJ |
640 | } |
641 | ||
642 | static void node_states_set_node(int node, struct memory_notify *arg) | |
643 | { | |
644 | if (arg->status_change_nid_normal >= 0) | |
645 | node_set_state(node, N_NORMAL_MEMORY); | |
646 | ||
6715ddf9 LJ |
647 | if (arg->status_change_nid_high >= 0) |
648 | node_set_state(node, N_HIGH_MEMORY); | |
649 | ||
83d83612 OS |
650 | if (arg->status_change_nid >= 0) |
651 | node_set_state(node, N_MEMORY); | |
d9713679 LJ |
652 | } |
653 | ||
f1dd2cd1 MH |
654 | static void __meminit resize_zone_range(struct zone *zone, unsigned long start_pfn, |
655 | unsigned long nr_pages) | |
656 | { | |
657 | unsigned long old_end_pfn = zone_end_pfn(zone); | |
658 | ||
659 | if (zone_is_empty(zone) || start_pfn < zone->zone_start_pfn) | |
660 | zone->zone_start_pfn = start_pfn; | |
661 | ||
662 | zone->spanned_pages = max(start_pfn + nr_pages, old_end_pfn) - zone->zone_start_pfn; | |
663 | } | |
664 | ||
665 | static void __meminit resize_pgdat_range(struct pglist_data *pgdat, unsigned long start_pfn, | |
666 | unsigned long nr_pages) | |
667 | { | |
668 | unsigned long old_end_pfn = pgdat_end_pfn(pgdat); | |
669 | ||
670 | if (!pgdat->node_spanned_pages || start_pfn < pgdat->node_start_pfn) | |
671 | pgdat->node_start_pfn = start_pfn; | |
672 | ||
673 | pgdat->node_spanned_pages = max(start_pfn + nr_pages, old_end_pfn) - pgdat->node_start_pfn; | |
f1dd2cd1 | 674 | |
3fccb74c DH |
675 | } |
676 | /* | |
677 | * Associate the pfn range with the given zone, initializing the memmaps | |
678 | * and resizing the pgdat/zone data to span the added pages. After this | |
679 | * call, all affected pages are PG_reserved. | |
680 | */ | |
a99583e7 CH |
681 | void __ref move_pfn_range_to_zone(struct zone *zone, unsigned long start_pfn, |
682 | unsigned long nr_pages, struct vmem_altmap *altmap) | |
f1dd2cd1 MH |
683 | { |
684 | struct pglist_data *pgdat = zone->zone_pgdat; | |
685 | int nid = pgdat->node_id; | |
686 | unsigned long flags; | |
df429ac0 | 687 | |
f1dd2cd1 MH |
688 | clear_zone_contiguous(zone); |
689 | ||
690 | /* TODO Huh pgdat is irqsave while zone is not. It used to be like that before */ | |
691 | pgdat_resize_lock(pgdat, &flags); | |
692 | zone_span_writelock(zone); | |
fa004ab7 WY |
693 | if (zone_is_empty(zone)) |
694 | init_currently_empty_zone(zone, start_pfn, nr_pages); | |
f1dd2cd1 MH |
695 | resize_zone_range(zone, start_pfn, nr_pages); |
696 | zone_span_writeunlock(zone); | |
697 | resize_pgdat_range(pgdat, start_pfn, nr_pages); | |
698 | pgdat_resize_unlock(pgdat, &flags); | |
699 | ||
700 | /* | |
701 | * TODO now we have a visible range of pages which are not associated | |
702 | * with their zone properly. Not nice but set_pfnblock_flags_mask | |
703 | * expects the zone spans the pfn range. All the pages in the range | |
704 | * are reserved so nobody should be touching them so we should be safe | |
705 | */ | |
a99583e7 CH |
706 | memmap_init_zone(nr_pages, nid, zone_idx(zone), start_pfn, |
707 | MEMMAP_HOTPLUG, altmap); | |
f1dd2cd1 MH |
708 | |
709 | set_zone_contiguous(zone); | |
710 | } | |
711 | ||
c246a213 MH |
712 | /* |
713 | * Returns a default kernel memory zone for the given pfn range. | |
714 | * If no kernel zone covers this pfn range it will automatically go | |
715 | * to the ZONE_NORMAL. | |
716 | */ | |
c6f03e29 | 717 | static struct zone *default_kernel_zone_for_pfn(int nid, unsigned long start_pfn, |
c246a213 MH |
718 | unsigned long nr_pages) |
719 | { | |
720 | struct pglist_data *pgdat = NODE_DATA(nid); | |
721 | int zid; | |
722 | ||
723 | for (zid = 0; zid <= ZONE_NORMAL; zid++) { | |
724 | struct zone *zone = &pgdat->node_zones[zid]; | |
725 | ||
726 | if (zone_intersects(zone, start_pfn, nr_pages)) | |
727 | return zone; | |
728 | } | |
729 | ||
730 | return &pgdat->node_zones[ZONE_NORMAL]; | |
731 | } | |
732 | ||
c6f03e29 MH |
733 | static inline struct zone *default_zone_for_pfn(int nid, unsigned long start_pfn, |
734 | unsigned long nr_pages) | |
e5e68930 | 735 | { |
c6f03e29 MH |
736 | struct zone *kernel_zone = default_kernel_zone_for_pfn(nid, start_pfn, |
737 | nr_pages); | |
738 | struct zone *movable_zone = &NODE_DATA(nid)->node_zones[ZONE_MOVABLE]; | |
739 | bool in_kernel = zone_intersects(kernel_zone, start_pfn, nr_pages); | |
740 | bool in_movable = zone_intersects(movable_zone, start_pfn, nr_pages); | |
e5e68930 MH |
741 | |
742 | /* | |
c6f03e29 MH |
743 | * We inherit the existing zone in a simple case where zones do not |
744 | * overlap in the given range | |
e5e68930 | 745 | */ |
c6f03e29 MH |
746 | if (in_kernel ^ in_movable) |
747 | return (in_kernel) ? kernel_zone : movable_zone; | |
9f123ab5 | 748 | |
c6f03e29 MH |
749 | /* |
750 | * If the range doesn't belong to any zone or two zones overlap in the | |
751 | * given range then we use movable zone only if movable_node is | |
752 | * enabled because we always online to a kernel zone by default. | |
753 | */ | |
754 | return movable_node_enabled ? movable_zone : kernel_zone; | |
9f123ab5 MH |
755 | } |
756 | ||
e5e68930 MH |
757 | struct zone * zone_for_pfn_range(int online_type, int nid, unsigned start_pfn, |
758 | unsigned long nr_pages) | |
f1dd2cd1 | 759 | { |
c6f03e29 MH |
760 | if (online_type == MMOP_ONLINE_KERNEL) |
761 | return default_kernel_zone_for_pfn(nid, start_pfn, nr_pages); | |
f1dd2cd1 | 762 | |
c6f03e29 MH |
763 | if (online_type == MMOP_ONLINE_MOVABLE) |
764 | return &NODE_DATA(nid)->node_zones[ZONE_MOVABLE]; | |
df429ac0 | 765 | |
c6f03e29 | 766 | return default_zone_for_pfn(nid, start_pfn, nr_pages); |
e5e68930 MH |
767 | } |
768 | ||
bd5c2344 DH |
769 | int __ref online_pages(unsigned long pfn, unsigned long nr_pages, |
770 | int online_type, int nid) | |
75884fb1 | 771 | { |
aa47228a | 772 | unsigned long flags; |
3947be19 DH |
773 | unsigned long onlined_pages = 0; |
774 | struct zone *zone; | |
6811378e | 775 | int need_zonelists_rebuild = 0; |
7b78d335 YG |
776 | int ret; |
777 | struct memory_notify arg; | |
d0dc12e8 | 778 | |
381eab4a DH |
779 | mem_hotplug_begin(); |
780 | ||
f1dd2cd1 | 781 | /* associate pfn range with the zone */ |
3fccb74c DH |
782 | zone = zone_for_pfn_range(online_type, nid, pfn, nr_pages); |
783 | move_pfn_range_to_zone(zone, pfn, nr_pages, NULL); | |
f1dd2cd1 | 784 | |
7b78d335 YG |
785 | arg.start_pfn = pfn; |
786 | arg.nr_pages = nr_pages; | |
d9713679 | 787 | node_states_check_changes_online(nr_pages, zone, &arg); |
7b78d335 | 788 | |
7b78d335 YG |
789 | ret = memory_notify(MEM_GOING_ONLINE, &arg); |
790 | ret = notifier_to_errno(ret); | |
e33e33b4 CY |
791 | if (ret) |
792 | goto failed_addition; | |
793 | ||
6811378e YG |
794 | /* |
795 | * If this zone is not populated, then it is not in zonelist. | |
796 | * This means the page allocator ignores this zone. | |
797 | * So, zonelist must be updated after online. | |
798 | */ | |
6dcd73d7 | 799 | if (!populated_zone(zone)) { |
6811378e | 800 | need_zonelists_rebuild = 1; |
72675e13 | 801 | setup_zone_pageset(zone); |
6dcd73d7 | 802 | } |
6811378e | 803 | |
908eedc6 | 804 | ret = walk_system_ram_range(pfn, nr_pages, &onlined_pages, |
75884fb1 | 805 | online_pages_range); |
fd8a4221 | 806 | if (ret) { |
ca9a46f8 | 807 | /* not a single memory resource was applicable */ |
6dcd73d7 WC |
808 | if (need_zonelists_rebuild) |
809 | zone_pcp_reset(zone); | |
e33e33b4 | 810 | goto failed_addition; |
fd8a4221 GL |
811 | } |
812 | ||
3947be19 | 813 | zone->present_pages += onlined_pages; |
aa47228a CS |
814 | |
815 | pgdat_resize_lock(zone->zone_pgdat, &flags); | |
f2937be5 | 816 | zone->zone_pgdat->node_present_pages += onlined_pages; |
aa47228a CS |
817 | pgdat_resize_unlock(zone->zone_pgdat, &flags); |
818 | ||
e900a918 DW |
819 | shuffle_zone(zone); |
820 | ||
ca9a46f8 DH |
821 | node_states_set_node(nid, &arg); |
822 | if (need_zonelists_rebuild) | |
823 | build_all_zonelists(NULL); | |
824 | else | |
825 | zone_pcp_update(zone); | |
3947be19 | 826 | |
1b79acc9 KM |
827 | init_per_zone_wmark_min(); |
828 | ||
ca9a46f8 DH |
829 | kswapd_run(nid); |
830 | kcompactd_run(nid); | |
61b13993 | 831 | |
1f522509 | 832 | vm_total_pages = nr_free_pagecache_pages(); |
2f7f24ec | 833 | |
2d1d43f6 | 834 | writeback_set_ratelimit(); |
7b78d335 | 835 | |
ca9a46f8 | 836 | memory_notify(MEM_ONLINE, &arg); |
381eab4a | 837 | mem_hotplug_done(); |
30467e0b | 838 | return 0; |
e33e33b4 CY |
839 | |
840 | failed_addition: | |
841 | pr_debug("online_pages [mem %#010llx-%#010llx] failed\n", | |
842 | (unsigned long long) pfn << PAGE_SHIFT, | |
843 | (((unsigned long long) pfn + nr_pages) << PAGE_SHIFT) - 1); | |
844 | memory_notify(MEM_CANCEL_ONLINE, &arg); | |
feee6b29 | 845 | remove_pfn_range_from_zone(zone, pfn, nr_pages); |
381eab4a | 846 | mem_hotplug_done(); |
e33e33b4 | 847 | return ret; |
3947be19 | 848 | } |
53947027 | 849 | #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */ |
bc02af93 | 850 | |
0bd85420 TC |
851 | static void reset_node_present_pages(pg_data_t *pgdat) |
852 | { | |
853 | struct zone *z; | |
854 | ||
855 | for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++) | |
856 | z->present_pages = 0; | |
857 | ||
858 | pgdat->node_present_pages = 0; | |
859 | } | |
860 | ||
e1319331 HS |
861 | /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */ |
862 | static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start) | |
9af3c2de YG |
863 | { |
864 | struct pglist_data *pgdat; | |
c8e861a5 | 865 | unsigned long start_pfn = PFN_DOWN(start); |
9af3c2de | 866 | |
a1e565aa TC |
867 | pgdat = NODE_DATA(nid); |
868 | if (!pgdat) { | |
869 | pgdat = arch_alloc_nodedata(nid); | |
870 | if (!pgdat) | |
871 | return NULL; | |
9af3c2de | 872 | |
33fce011 WY |
873 | pgdat->per_cpu_nodestats = |
874 | alloc_percpu(struct per_cpu_nodestat); | |
a1e565aa | 875 | arch_refresh_nodedata(nid, pgdat); |
b0dc3a34 | 876 | } else { |
33fce011 | 877 | int cpu; |
e716f2eb MG |
878 | /* |
879 | * Reset the nr_zones, order and classzone_idx before reuse. | |
880 | * Note that kswapd will init kswapd_classzone_idx properly | |
881 | * when it starts in the near future. | |
882 | */ | |
b0dc3a34 | 883 | pgdat->nr_zones = 0; |
38087d9b MG |
884 | pgdat->kswapd_order = 0; |
885 | pgdat->kswapd_classzone_idx = 0; | |
33fce011 WY |
886 | for_each_online_cpu(cpu) { |
887 | struct per_cpu_nodestat *p; | |
888 | ||
889 | p = per_cpu_ptr(pgdat->per_cpu_nodestats, cpu); | |
890 | memset(p, 0, sizeof(*p)); | |
891 | } | |
a1e565aa | 892 | } |
9af3c2de YG |
893 | |
894 | /* we can use NODE_DATA(nid) from here */ | |
895 | ||
03e85f9d OS |
896 | pgdat->node_id = nid; |
897 | pgdat->node_start_pfn = start_pfn; | |
898 | ||
9af3c2de | 899 | /* init node's zones as empty zones, we don't have any present pages.*/ |
03e85f9d | 900 | free_area_init_core_hotplug(nid); |
9af3c2de | 901 | |
959ecc48 KH |
902 | /* |
903 | * The node we allocated has no zone fallback lists. For avoiding | |
904 | * to access not-initialized zonelist, build here. | |
905 | */ | |
72675e13 | 906 | build_all_zonelists(pgdat); |
959ecc48 | 907 | |
0bd85420 TC |
908 | /* |
909 | * When memory is hot-added, all the memory is in offline state. So | |
910 | * clear all zones' present_pages because they will be updated in | |
911 | * online_pages() and offline_pages(). | |
912 | */ | |
03e85f9d | 913 | reset_node_managed_pages(pgdat); |
0bd85420 TC |
914 | reset_node_present_pages(pgdat); |
915 | ||
9af3c2de YG |
916 | return pgdat; |
917 | } | |
918 | ||
b9ff0360 | 919 | static void rollback_node_hotadd(int nid) |
9af3c2de | 920 | { |
b9ff0360 OS |
921 | pg_data_t *pgdat = NODE_DATA(nid); |
922 | ||
9af3c2de | 923 | arch_refresh_nodedata(nid, NULL); |
5830169f | 924 | free_percpu(pgdat->per_cpu_nodestats); |
9af3c2de | 925 | arch_free_nodedata(pgdat); |
9af3c2de YG |
926 | } |
927 | ||
0a547039 | 928 | |
01b0f197 TK |
929 | /** |
930 | * try_online_node - online a node if offlined | |
e8b098fc | 931 | * @nid: the node ID |
b9ff0360 OS |
932 | * @start: start addr of the node |
933 | * @set_node_online: Whether we want to online the node | |
cf23422b | 934 | * called by cpu_up() to online a node without onlined memory. |
b9ff0360 OS |
935 | * |
936 | * Returns: | |
937 | * 1 -> a new node has been allocated | |
938 | * 0 -> the node is already online | |
939 | * -ENOMEM -> the node could not be allocated | |
cf23422b | 940 | */ |
b9ff0360 | 941 | static int __try_online_node(int nid, u64 start, bool set_node_online) |
cf23422b | 942 | { |
b9ff0360 OS |
943 | pg_data_t *pgdat; |
944 | int ret = 1; | |
cf23422b | 945 | |
01b0f197 TK |
946 | if (node_online(nid)) |
947 | return 0; | |
948 | ||
b9ff0360 | 949 | pgdat = hotadd_new_pgdat(nid, start); |
7553e8f2 | 950 | if (!pgdat) { |
01b0f197 | 951 | pr_err("Cannot online node %d due to NULL pgdat\n", nid); |
cf23422b | 952 | ret = -ENOMEM; |
953 | goto out; | |
954 | } | |
b9ff0360 OS |
955 | |
956 | if (set_node_online) { | |
957 | node_set_online(nid); | |
958 | ret = register_one_node(nid); | |
959 | BUG_ON(ret); | |
960 | } | |
cf23422b | 961 | out: |
b9ff0360 OS |
962 | return ret; |
963 | } | |
964 | ||
965 | /* | |
966 | * Users of this function always want to online/register the node | |
967 | */ | |
968 | int try_online_node(int nid) | |
969 | { | |
970 | int ret; | |
971 | ||
972 | mem_hotplug_begin(); | |
973 | ret = __try_online_node(nid, 0, true); | |
bfc8c901 | 974 | mem_hotplug_done(); |
cf23422b | 975 | return ret; |
976 | } | |
977 | ||
27356f54 TK |
978 | static int check_hotplug_memory_range(u64 start, u64 size) |
979 | { | |
ba325585 | 980 | /* memory range must be block size aligned */ |
cec3ebd0 DH |
981 | if (!size || !IS_ALIGNED(start, memory_block_size_bytes()) || |
982 | !IS_ALIGNED(size, memory_block_size_bytes())) { | |
ba325585 | 983 | pr_err("Block size [%#lx] unaligned hotplug range: start %#llx, size %#llx", |
cec3ebd0 | 984 | memory_block_size_bytes(), start, size); |
27356f54 TK |
985 | return -EINVAL; |
986 | } | |
987 | ||
988 | return 0; | |
989 | } | |
990 | ||
31bc3858 VK |
991 | static int online_memory_block(struct memory_block *mem, void *arg) |
992 | { | |
dc18d706 | 993 | return device_online(&mem->dev); |
31bc3858 VK |
994 | } |
995 | ||
8df1d0e4 DH |
996 | /* |
997 | * NOTE: The caller must call lock_device_hotplug() to serialize hotplug | |
998 | * and online/offline operations (triggered e.g. by sysfs). | |
999 | * | |
1000 | * we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG | |
1001 | */ | |
f29d8e9c | 1002 | int __ref add_memory_resource(int nid, struct resource *res) |
bc02af93 | 1003 | { |
05f800a0 | 1004 | struct mhp_restrictions restrictions = {}; |
62cedb9f | 1005 | u64 start, size; |
b9ff0360 | 1006 | bool new_node = false; |
bc02af93 YG |
1007 | int ret; |
1008 | ||
62cedb9f DV |
1009 | start = res->start; |
1010 | size = resource_size(res); | |
1011 | ||
27356f54 TK |
1012 | ret = check_hotplug_memory_range(start, size); |
1013 | if (ret) | |
1014 | return ret; | |
1015 | ||
bfc8c901 | 1016 | mem_hotplug_begin(); |
ac13c462 | 1017 | |
7f36e3e5 TC |
1018 | /* |
1019 | * Add new range to memblock so that when hotadd_new_pgdat() is called | |
1020 | * to allocate new pgdat, get_pfn_range_for_nid() will be able to find | |
1021 | * this new range and calculate total pages correctly. The range will | |
1022 | * be removed at hot-remove time. | |
1023 | */ | |
1024 | memblock_add_node(start, size, nid); | |
1025 | ||
b9ff0360 OS |
1026 | ret = __try_online_node(nid, start, false); |
1027 | if (ret < 0) | |
1028 | goto error; | |
1029 | new_node = ret; | |
9af3c2de | 1030 | |
bc02af93 | 1031 | /* call arch's memory hotadd */ |
940519f0 | 1032 | ret = arch_add_memory(nid, start, size, &restrictions); |
9af3c2de YG |
1033 | if (ret < 0) |
1034 | goto error; | |
1035 | ||
db051a0d DH |
1036 | /* create memory block devices after memory was added */ |
1037 | ret = create_memory_block_devices(start, size); | |
1038 | if (ret) { | |
1039 | arch_remove_memory(nid, start, size, NULL); | |
1040 | goto error; | |
1041 | } | |
1042 | ||
a1e565aa | 1043 | if (new_node) { |
d5b6f6a3 | 1044 | /* If sysfs file of new node can't be created, cpu on the node |
0fc44159 YG |
1045 | * can't be hot-added. There is no rollback way now. |
1046 | * So, check by BUG_ON() to catch it reluctantly.. | |
d5b6f6a3 | 1047 | * We online node here. We can't roll back from here. |
0fc44159 | 1048 | */ |
d5b6f6a3 OS |
1049 | node_set_online(nid); |
1050 | ret = __register_one_node(nid); | |
0fc44159 YG |
1051 | BUG_ON(ret); |
1052 | } | |
1053 | ||
d5b6f6a3 | 1054 | /* link memory sections under this node.*/ |
4fbce633 | 1055 | ret = link_mem_sections(nid, PFN_DOWN(start), PFN_UP(start + size - 1)); |
d5b6f6a3 OS |
1056 | BUG_ON(ret); |
1057 | ||
d96ae530 | 1058 | /* create new memmap entry */ |
1059 | firmware_map_add_hotplug(start, start + size, "System RAM"); | |
1060 | ||
381eab4a DH |
1061 | /* device_online() will take the lock when calling online_pages() */ |
1062 | mem_hotplug_done(); | |
1063 | ||
31bc3858 | 1064 | /* online pages if requested */ |
f29d8e9c | 1065 | if (memhp_auto_online) |
fbcf73ce | 1066 | walk_memory_blocks(start, size, NULL, online_memory_block); |
31bc3858 | 1067 | |
381eab4a | 1068 | return ret; |
9af3c2de YG |
1069 | error: |
1070 | /* rollback pgdat allocation and others */ | |
b9ff0360 OS |
1071 | if (new_node) |
1072 | rollback_node_hotadd(nid); | |
7f36e3e5 | 1073 | memblock_remove(start, size); |
bfc8c901 | 1074 | mem_hotplug_done(); |
bc02af93 YG |
1075 | return ret; |
1076 | } | |
62cedb9f | 1077 | |
8df1d0e4 DH |
1078 | /* requires device_hotplug_lock, see add_memory_resource() */ |
1079 | int __ref __add_memory(int nid, u64 start, u64 size) | |
62cedb9f DV |
1080 | { |
1081 | struct resource *res; | |
1082 | int ret; | |
1083 | ||
1084 | res = register_memory_resource(start, size); | |
6f754ba4 VK |
1085 | if (IS_ERR(res)) |
1086 | return PTR_ERR(res); | |
62cedb9f | 1087 | |
f29d8e9c | 1088 | ret = add_memory_resource(nid, res); |
62cedb9f DV |
1089 | if (ret < 0) |
1090 | release_memory_resource(res); | |
1091 | return ret; | |
1092 | } | |
8df1d0e4 DH |
1093 | |
1094 | int add_memory(int nid, u64 start, u64 size) | |
1095 | { | |
1096 | int rc; | |
1097 | ||
1098 | lock_device_hotplug(); | |
1099 | rc = __add_memory(nid, start, size); | |
1100 | unlock_device_hotplug(); | |
1101 | ||
1102 | return rc; | |
1103 | } | |
bc02af93 | 1104 | EXPORT_SYMBOL_GPL(add_memory); |
0c0e6195 KH |
1105 | |
1106 | #ifdef CONFIG_MEMORY_HOTREMOVE | |
5c755e9f BP |
1107 | /* |
1108 | * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy | |
1109 | * set and the size of the free page is given by page_order(). Using this, | |
1110 | * the function determines if the pageblock contains only free pages. | |
1111 | * Due to buddy contraints, a free page at least the size of a pageblock will | |
1112 | * be located at the start of the pageblock | |
1113 | */ | |
1114 | static inline int pageblock_free(struct page *page) | |
1115 | { | |
1116 | return PageBuddy(page) && page_order(page) >= pageblock_order; | |
1117 | } | |
1118 | ||
891cb2a7 MH |
1119 | /* Return the pfn of the start of the next active pageblock after a given pfn */ |
1120 | static unsigned long next_active_pageblock(unsigned long pfn) | |
5c755e9f | 1121 | { |
891cb2a7 MH |
1122 | struct page *page = pfn_to_page(pfn); |
1123 | ||
5c755e9f | 1124 | /* Ensure the starting page is pageblock-aligned */ |
891cb2a7 | 1125 | BUG_ON(pfn & (pageblock_nr_pages - 1)); |
5c755e9f | 1126 | |
5c755e9f | 1127 | /* If the entire pageblock is free, move to the end of free page */ |
0dcc48c1 KH |
1128 | if (pageblock_free(page)) { |
1129 | int order; | |
1130 | /* be careful. we don't have locks, page_order can be changed.*/ | |
1131 | order = page_order(page); | |
1132 | if ((order < MAX_ORDER) && (order >= pageblock_order)) | |
891cb2a7 | 1133 | return pfn + (1 << order); |
0dcc48c1 | 1134 | } |
5c755e9f | 1135 | |
891cb2a7 | 1136 | return pfn + pageblock_nr_pages; |
5c755e9f BP |
1137 | } |
1138 | ||
891cb2a7 | 1139 | static bool is_pageblock_removable_nolock(unsigned long pfn) |
fb52bbae | 1140 | { |
891cb2a7 | 1141 | struct page *page = pfn_to_page(pfn); |
fb52bbae | 1142 | struct zone *zone; |
fb52bbae MM |
1143 | |
1144 | /* | |
1145 | * We have to be careful here because we are iterating over memory | |
1146 | * sections which are not zone aware so we might end up outside of | |
1147 | * the zone but still within the section. | |
1148 | * We have to take care about the node as well. If the node is offline | |
1149 | * its NODE_DATA will be NULL - see page_zone. | |
1150 | */ | |
1151 | if (!node_online(page_to_nid(page))) | |
1152 | return false; | |
1153 | ||
1154 | zone = page_zone(page); | |
1155 | pfn = page_to_pfn(page); | |
1156 | if (!zone_spans_pfn(zone, pfn)) | |
1157 | return false; | |
1158 | ||
fe4c86c9 | 1159 | return !has_unmovable_pages(zone, page, MIGRATE_MOVABLE, |
756d25be | 1160 | MEMORY_OFFLINE); |
fb52bbae MM |
1161 | } |
1162 | ||
5c755e9f | 1163 | /* Checks if this range of memory is likely to be hot-removable. */ |
c98940f6 | 1164 | bool is_mem_section_removable(unsigned long start_pfn, unsigned long nr_pages) |
5c755e9f | 1165 | { |
891cb2a7 MH |
1166 | unsigned long end_pfn, pfn; |
1167 | ||
1168 | end_pfn = min(start_pfn + nr_pages, | |
1169 | zone_end_pfn(page_zone(pfn_to_page(start_pfn)))); | |
5c755e9f BP |
1170 | |
1171 | /* Check the starting page of each pageblock within the range */ | |
891cb2a7 MH |
1172 | for (pfn = start_pfn; pfn < end_pfn; pfn = next_active_pageblock(pfn)) { |
1173 | if (!is_pageblock_removable_nolock(pfn)) | |
c98940f6 | 1174 | return false; |
49ac8255 | 1175 | cond_resched(); |
5c755e9f BP |
1176 | } |
1177 | ||
1178 | /* All pageblocks in the memory block are likely to be hot-removable */ | |
c98940f6 | 1179 | return true; |
5c755e9f BP |
1180 | } |
1181 | ||
0c0e6195 | 1182 | /* |
deb88a2a | 1183 | * Confirm all pages in a range [start, end) belong to the same zone. |
a96dfddb | 1184 | * When true, return its valid [start, end). |
0c0e6195 | 1185 | */ |
a96dfddb TK |
1186 | int test_pages_in_a_zone(unsigned long start_pfn, unsigned long end_pfn, |
1187 | unsigned long *valid_start, unsigned long *valid_end) | |
0c0e6195 | 1188 | { |
5f0f2887 | 1189 | unsigned long pfn, sec_end_pfn; |
a96dfddb | 1190 | unsigned long start, end; |
0c0e6195 KH |
1191 | struct zone *zone = NULL; |
1192 | struct page *page; | |
1193 | int i; | |
deb88a2a | 1194 | for (pfn = start_pfn, sec_end_pfn = SECTION_ALIGN_UP(start_pfn + 1); |
0c0e6195 | 1195 | pfn < end_pfn; |
deb88a2a | 1196 | pfn = sec_end_pfn, sec_end_pfn += PAGES_PER_SECTION) { |
5f0f2887 AB |
1197 | /* Make sure the memory section is present first */ |
1198 | if (!present_section_nr(pfn_to_section_nr(pfn))) | |
0c0e6195 | 1199 | continue; |
5f0f2887 AB |
1200 | for (; pfn < sec_end_pfn && pfn < end_pfn; |
1201 | pfn += MAX_ORDER_NR_PAGES) { | |
1202 | i = 0; | |
1203 | /* This is just a CONFIG_HOLES_IN_ZONE check.*/ | |
1204 | while ((i < MAX_ORDER_NR_PAGES) && | |
1205 | !pfn_valid_within(pfn + i)) | |
1206 | i++; | |
d6d8c8a4 | 1207 | if (i == MAX_ORDER_NR_PAGES || pfn + i >= end_pfn) |
5f0f2887 | 1208 | continue; |
24feb47c MZ |
1209 | /* Check if we got outside of the zone */ |
1210 | if (zone && !zone_spans_pfn(zone, pfn + i)) | |
1211 | return 0; | |
5f0f2887 AB |
1212 | page = pfn_to_page(pfn + i); |
1213 | if (zone && page_zone(page) != zone) | |
1214 | return 0; | |
a96dfddb TK |
1215 | if (!zone) |
1216 | start = pfn + i; | |
5f0f2887 | 1217 | zone = page_zone(page); |
a96dfddb | 1218 | end = pfn + MAX_ORDER_NR_PAGES; |
5f0f2887 | 1219 | } |
0c0e6195 | 1220 | } |
deb88a2a | 1221 | |
a96dfddb TK |
1222 | if (zone) { |
1223 | *valid_start = start; | |
d6d8c8a4 | 1224 | *valid_end = min(end, end_pfn); |
deb88a2a | 1225 | return 1; |
a96dfddb | 1226 | } else { |
deb88a2a | 1227 | return 0; |
a96dfddb | 1228 | } |
0c0e6195 KH |
1229 | } |
1230 | ||
1231 | /* | |
0efadf48 YX |
1232 | * Scan pfn range [start,end) to find movable/migratable pages (LRU pages, |
1233 | * non-lru movable pages and hugepages). We scan pfn because it's much | |
1234 | * easier than scanning over linked list. This function returns the pfn | |
1235 | * of the first found movable page if it's found, otherwise 0. | |
0c0e6195 | 1236 | */ |
c8721bbb | 1237 | static unsigned long scan_movable_pages(unsigned long start, unsigned long end) |
0c0e6195 KH |
1238 | { |
1239 | unsigned long pfn; | |
eeb0efd0 | 1240 | |
0c0e6195 | 1241 | for (pfn = start; pfn < end; pfn++) { |
eeb0efd0 OS |
1242 | struct page *page, *head; |
1243 | unsigned long skip; | |
1244 | ||
1245 | if (!pfn_valid(pfn)) | |
1246 | continue; | |
1247 | page = pfn_to_page(pfn); | |
1248 | if (PageLRU(page)) | |
1249 | return pfn; | |
1250 | if (__PageMovable(page)) | |
1251 | return pfn; | |
1252 | ||
1253 | if (!PageHuge(page)) | |
1254 | continue; | |
1255 | head = compound_head(page); | |
39186cbe | 1256 | if (page_huge_active(head)) |
eeb0efd0 | 1257 | return pfn; |
d8c6546b | 1258 | skip = compound_nr(head) - (page - head); |
eeb0efd0 | 1259 | pfn += skip - 1; |
0c0e6195 KH |
1260 | } |
1261 | return 0; | |
1262 | } | |
1263 | ||
666feb21 | 1264 | static struct page *new_node_page(struct page *page, unsigned long private) |
394e31d2 | 1265 | { |
394e31d2 | 1266 | int nid = page_to_nid(page); |
231e97e2 | 1267 | nodemask_t nmask = node_states[N_MEMORY]; |
7f252f27 MH |
1268 | |
1269 | /* | |
1270 | * try to allocate from a different node but reuse this node if there | |
1271 | * are no other online nodes to be used (e.g. we are offlining a part | |
1272 | * of the only existing node) | |
1273 | */ | |
1274 | node_clear(nid, nmask); | |
1275 | if (nodes_empty(nmask)) | |
1276 | node_set(nid, nmask); | |
394e31d2 | 1277 | |
8b913238 | 1278 | return new_page_nodemask(page, nid, &nmask); |
394e31d2 XQ |
1279 | } |
1280 | ||
0c0e6195 KH |
1281 | static int |
1282 | do_migrate_range(unsigned long start_pfn, unsigned long end_pfn) | |
1283 | { | |
1284 | unsigned long pfn; | |
1285 | struct page *page; | |
0c0e6195 KH |
1286 | int ret = 0; |
1287 | LIST_HEAD(source); | |
1288 | ||
a85009c3 | 1289 | for (pfn = start_pfn; pfn < end_pfn; pfn++) { |
0c0e6195 KH |
1290 | if (!pfn_valid(pfn)) |
1291 | continue; | |
1292 | page = pfn_to_page(pfn); | |
c8721bbb NH |
1293 | |
1294 | if (PageHuge(page)) { | |
1295 | struct page *head = compound_head(page); | |
d8c6546b | 1296 | pfn = page_to_pfn(head) + compound_nr(head) - 1; |
daf3538a | 1297 | isolate_huge_page(head, &source); |
c8721bbb | 1298 | continue; |
94723aaf | 1299 | } else if (PageTransHuge(page)) |
8135d892 NH |
1300 | pfn = page_to_pfn(compound_head(page)) |
1301 | + hpage_nr_pages(page) - 1; | |
c8721bbb | 1302 | |
b15c8726 MH |
1303 | /* |
1304 | * HWPoison pages have elevated reference counts so the migration would | |
1305 | * fail on them. It also doesn't make any sense to migrate them in the | |
1306 | * first place. Still try to unmap such a page in case it is still mapped | |
1307 | * (e.g. current hwpoison implementation doesn't unmap KSM pages but keep | |
1308 | * the unmap as the catch all safety net). | |
1309 | */ | |
1310 | if (PageHWPoison(page)) { | |
1311 | if (WARN_ON(PageLRU(page))) | |
1312 | isolate_lru_page(page); | |
1313 | if (page_mapped(page)) | |
1314 | try_to_unmap(page, TTU_IGNORE_MLOCK | TTU_IGNORE_ACCESS); | |
1315 | continue; | |
1316 | } | |
1317 | ||
700c2a46 | 1318 | if (!get_page_unless_zero(page)) |
0c0e6195 KH |
1319 | continue; |
1320 | /* | |
0efadf48 YX |
1321 | * We can skip free pages. And we can deal with pages on |
1322 | * LRU and non-lru movable pages. | |
0c0e6195 | 1323 | */ |
0efadf48 YX |
1324 | if (PageLRU(page)) |
1325 | ret = isolate_lru_page(page); | |
1326 | else | |
1327 | ret = isolate_movable_page(page, ISOLATE_UNEVICTABLE); | |
0c0e6195 | 1328 | if (!ret) { /* Success */ |
62695a84 | 1329 | list_add_tail(&page->lru, &source); |
0efadf48 YX |
1330 | if (!__PageMovable(page)) |
1331 | inc_node_page_state(page, NR_ISOLATED_ANON + | |
1332 | page_is_file_cache(page)); | |
6d9c285a | 1333 | |
0c0e6195 | 1334 | } else { |
2932c8b0 | 1335 | pr_warn("failed to isolate pfn %lx\n", pfn); |
0efadf48 | 1336 | dump_page(page, "isolation failed"); |
0c0e6195 | 1337 | } |
1723058e | 1338 | put_page(page); |
0c0e6195 | 1339 | } |
f3ab2636 | 1340 | if (!list_empty(&source)) { |
394e31d2 XQ |
1341 | /* Allocate a new page from the nearest neighbor node */ |
1342 | ret = migrate_pages(&source, new_node_page, NULL, 0, | |
9c620e2b | 1343 | MIGRATE_SYNC, MR_MEMORY_HOTPLUG); |
2932c8b0 MH |
1344 | if (ret) { |
1345 | list_for_each_entry(page, &source, lru) { | |
1346 | pr_warn("migrating pfn %lx failed ret:%d ", | |
1347 | page_to_pfn(page), ret); | |
1348 | dump_page(page, "migration failure"); | |
1349 | } | |
c8721bbb | 1350 | putback_movable_pages(&source); |
2932c8b0 | 1351 | } |
0c0e6195 | 1352 | } |
1723058e | 1353 | |
0c0e6195 KH |
1354 | return ret; |
1355 | } | |
1356 | ||
0ee5f4f3 | 1357 | /* Mark all sections offline and remove all free pages from the buddy. */ |
0c0e6195 KH |
1358 | static int |
1359 | offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages, | |
1360 | void *data) | |
1361 | { | |
5557c766 | 1362 | unsigned long *offlined_pages = (unsigned long *)data; |
0c0e6195 | 1363 | |
5557c766 MH |
1364 | *offlined_pages += __offline_isolated_pages(start, start + nr_pages); |
1365 | return 0; | |
0c0e6195 KH |
1366 | } |
1367 | ||
1368 | /* | |
1369 | * Check all pages in range, recoreded as memory resource, are isolated. | |
1370 | */ | |
1371 | static int | |
1372 | check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages, | |
1373 | void *data) | |
1374 | { | |
756d25be DH |
1375 | return test_pages_isolated(start_pfn, start_pfn + nr_pages, |
1376 | MEMORY_OFFLINE); | |
0c0e6195 KH |
1377 | } |
1378 | ||
c5320926 TC |
1379 | static int __init cmdline_parse_movable_node(char *p) |
1380 | { | |
4932381e | 1381 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
55ac590c | 1382 | movable_node_enabled = true; |
4932381e MH |
1383 | #else |
1384 | pr_warn("movable_node parameter depends on CONFIG_HAVE_MEMBLOCK_NODE_MAP to work properly\n"); | |
1385 | #endif | |
c5320926 TC |
1386 | return 0; |
1387 | } | |
1388 | early_param("movable_node", cmdline_parse_movable_node); | |
1389 | ||
d9713679 LJ |
1390 | /* check which state of node_states will be changed when offline memory */ |
1391 | static void node_states_check_changes_offline(unsigned long nr_pages, | |
1392 | struct zone *zone, struct memory_notify *arg) | |
1393 | { | |
1394 | struct pglist_data *pgdat = zone->zone_pgdat; | |
1395 | unsigned long present_pages = 0; | |
86b27bea | 1396 | enum zone_type zt; |
d9713679 | 1397 | |
98fa15f3 AK |
1398 | arg->status_change_nid = NUMA_NO_NODE; |
1399 | arg->status_change_nid_normal = NUMA_NO_NODE; | |
1400 | arg->status_change_nid_high = NUMA_NO_NODE; | |
d9713679 LJ |
1401 | |
1402 | /* | |
86b27bea OS |
1403 | * Check whether node_states[N_NORMAL_MEMORY] will be changed. |
1404 | * If the memory to be offline is within the range | |
1405 | * [0..ZONE_NORMAL], and it is the last present memory there, | |
1406 | * the zones in that range will become empty after the offlining, | |
1407 | * thus we can determine that we need to clear the node from | |
1408 | * node_states[N_NORMAL_MEMORY]. | |
d9713679 | 1409 | */ |
86b27bea | 1410 | for (zt = 0; zt <= ZONE_NORMAL; zt++) |
d9713679 | 1411 | present_pages += pgdat->node_zones[zt].present_pages; |
86b27bea | 1412 | if (zone_idx(zone) <= ZONE_NORMAL && nr_pages >= present_pages) |
d9713679 | 1413 | arg->status_change_nid_normal = zone_to_nid(zone); |
d9713679 | 1414 | |
6715ddf9 LJ |
1415 | #ifdef CONFIG_HIGHMEM |
1416 | /* | |
86b27bea OS |
1417 | * node_states[N_HIGH_MEMORY] contains nodes which |
1418 | * have normal memory or high memory. | |
1419 | * Here we add the present_pages belonging to ZONE_HIGHMEM. | |
1420 | * If the zone is within the range of [0..ZONE_HIGHMEM), and | |
1421 | * we determine that the zones in that range become empty, | |
1422 | * we need to clear the node for N_HIGH_MEMORY. | |
6715ddf9 | 1423 | */ |
86b27bea OS |
1424 | present_pages += pgdat->node_zones[ZONE_HIGHMEM].present_pages; |
1425 | if (zone_idx(zone) <= ZONE_HIGHMEM && nr_pages >= present_pages) | |
6715ddf9 | 1426 | arg->status_change_nid_high = zone_to_nid(zone); |
6715ddf9 LJ |
1427 | #endif |
1428 | ||
d9713679 | 1429 | /* |
86b27bea OS |
1430 | * We have accounted the pages from [0..ZONE_NORMAL), and |
1431 | * in case of CONFIG_HIGHMEM the pages from ZONE_HIGHMEM | |
1432 | * as well. | |
1433 | * Here we count the possible pages from ZONE_MOVABLE. | |
1434 | * If after having accounted all the pages, we see that the nr_pages | |
1435 | * to be offlined is over or equal to the accounted pages, | |
1436 | * we know that the node will become empty, and so, we can clear | |
1437 | * it for N_MEMORY as well. | |
d9713679 | 1438 | */ |
86b27bea | 1439 | present_pages += pgdat->node_zones[ZONE_MOVABLE].present_pages; |
d9713679 | 1440 | |
d9713679 LJ |
1441 | if (nr_pages >= present_pages) |
1442 | arg->status_change_nid = zone_to_nid(zone); | |
d9713679 LJ |
1443 | } |
1444 | ||
1445 | static void node_states_clear_node(int node, struct memory_notify *arg) | |
1446 | { | |
1447 | if (arg->status_change_nid_normal >= 0) | |
1448 | node_clear_state(node, N_NORMAL_MEMORY); | |
1449 | ||
cf01f6f5 | 1450 | if (arg->status_change_nid_high >= 0) |
d9713679 | 1451 | node_clear_state(node, N_HIGH_MEMORY); |
6715ddf9 | 1452 | |
cf01f6f5 | 1453 | if (arg->status_change_nid >= 0) |
6715ddf9 | 1454 | node_clear_state(node, N_MEMORY); |
d9713679 LJ |
1455 | } |
1456 | ||
c5e79ef5 DH |
1457 | static int count_system_ram_pages_cb(unsigned long start_pfn, |
1458 | unsigned long nr_pages, void *data) | |
1459 | { | |
1460 | unsigned long *nr_system_ram_pages = data; | |
1461 | ||
1462 | *nr_system_ram_pages += nr_pages; | |
1463 | return 0; | |
1464 | } | |
1465 | ||
a16cee10 | 1466 | static int __ref __offline_pages(unsigned long start_pfn, |
ecde0f3e | 1467 | unsigned long end_pfn) |
0c0e6195 | 1468 | { |
c5e79ef5 | 1469 | unsigned long pfn, nr_pages = 0; |
5557c766 | 1470 | unsigned long offlined_pages = 0; |
9b7ea46a | 1471 | int ret, node, nr_isolate_pageblock; |
d702909f | 1472 | unsigned long flags; |
a96dfddb | 1473 | unsigned long valid_start, valid_end; |
0c0e6195 | 1474 | struct zone *zone; |
7b78d335 | 1475 | struct memory_notify arg; |
79605093 | 1476 | char *reason; |
0c0e6195 | 1477 | |
381eab4a DH |
1478 | mem_hotplug_begin(); |
1479 | ||
c5e79ef5 DH |
1480 | /* |
1481 | * Don't allow to offline memory blocks that contain holes. | |
1482 | * Consequently, memory blocks with holes can never get onlined | |
1483 | * via the hotplug path - online_pages() - as hotplugged memory has | |
1484 | * no holes. This way, we e.g., don't have to worry about marking | |
1485 | * memory holes PG_reserved, don't need pfn_valid() checks, and can | |
1486 | * avoid using walk_system_ram_range() later. | |
1487 | */ | |
1488 | walk_system_ram_range(start_pfn, end_pfn - start_pfn, &nr_pages, | |
1489 | count_system_ram_pages_cb); | |
1490 | if (nr_pages != end_pfn - start_pfn) { | |
1491 | ret = -EINVAL; | |
1492 | reason = "memory holes"; | |
1493 | goto failed_removal; | |
1494 | } | |
1495 | ||
0c0e6195 KH |
1496 | /* This makes hotplug much easier...and readable. |
1497 | we assume this for now. .*/ | |
381eab4a DH |
1498 | if (!test_pages_in_a_zone(start_pfn, end_pfn, &valid_start, |
1499 | &valid_end)) { | |
79605093 MH |
1500 | ret = -EINVAL; |
1501 | reason = "multizone range"; | |
1502 | goto failed_removal; | |
381eab4a | 1503 | } |
7b78d335 | 1504 | |
a96dfddb | 1505 | zone = page_zone(pfn_to_page(valid_start)); |
7b78d335 | 1506 | node = zone_to_nid(zone); |
7b78d335 | 1507 | |
0c0e6195 | 1508 | /* set above range as isolated */ |
b023f468 | 1509 | ret = start_isolate_page_range(start_pfn, end_pfn, |
d381c547 | 1510 | MIGRATE_MOVABLE, |
756d25be | 1511 | MEMORY_OFFLINE | REPORT_FAILURE); |
9b7ea46a | 1512 | if (ret < 0) { |
79605093 MH |
1513 | reason = "failure to isolate range"; |
1514 | goto failed_removal; | |
381eab4a | 1515 | } |
9b7ea46a | 1516 | nr_isolate_pageblock = ret; |
7b78d335 YG |
1517 | |
1518 | arg.start_pfn = start_pfn; | |
1519 | arg.nr_pages = nr_pages; | |
d9713679 | 1520 | node_states_check_changes_offline(nr_pages, zone, &arg); |
7b78d335 YG |
1521 | |
1522 | ret = memory_notify(MEM_GOING_OFFLINE, &arg); | |
1523 | ret = notifier_to_errno(ret); | |
79605093 MH |
1524 | if (ret) { |
1525 | reason = "notifier failure"; | |
1526 | goto failed_removal_isolated; | |
1527 | } | |
7b78d335 | 1528 | |
bb8965bd MH |
1529 | do { |
1530 | for (pfn = start_pfn; pfn;) { | |
1531 | if (signal_pending(current)) { | |
1532 | ret = -EINTR; | |
1533 | reason = "signal backoff"; | |
1534 | goto failed_removal_isolated; | |
1535 | } | |
72b39cfc | 1536 | |
bb8965bd MH |
1537 | cond_resched(); |
1538 | lru_add_drain_all(); | |
bb8965bd MH |
1539 | |
1540 | pfn = scan_movable_pages(pfn, end_pfn); | |
1541 | if (pfn) { | |
1542 | /* | |
1543 | * TODO: fatal migration failures should bail | |
1544 | * out | |
1545 | */ | |
1546 | do_migrate_range(pfn, end_pfn); | |
1547 | } | |
1548 | } | |
0c0e6195 | 1549 | |
bb8965bd MH |
1550 | /* |
1551 | * Dissolve free hugepages in the memory block before doing | |
1552 | * offlining actually in order to make hugetlbfs's object | |
1553 | * counting consistent. | |
1554 | */ | |
1555 | ret = dissolve_free_huge_pages(start_pfn, end_pfn); | |
1556 | if (ret) { | |
1557 | reason = "failure to dissolve huge pages"; | |
1558 | goto failed_removal_isolated; | |
1559 | } | |
1560 | /* check again */ | |
5557c766 MH |
1561 | ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn, |
1562 | NULL, check_pages_isolated_cb); | |
1563 | } while (ret); | |
72b39cfc | 1564 | |
b3834be5 | 1565 | /* Ok, all of our target is isolated. |
0c0e6195 | 1566 | We cannot do rollback at this point. */ |
5557c766 MH |
1567 | walk_system_ram_range(start_pfn, end_pfn - start_pfn, |
1568 | &offlined_pages, offline_isolated_pages_cb); | |
1569 | pr_info("Offlined Pages %ld\n", offlined_pages); | |
9b7ea46a QC |
1570 | /* |
1571 | * Onlining will reset pagetype flags and makes migrate type | |
1572 | * MOVABLE, so just need to decrease the number of isolated | |
1573 | * pageblocks zone counter here. | |
1574 | */ | |
1575 | spin_lock_irqsave(&zone->lock, flags); | |
1576 | zone->nr_isolate_pageblock -= nr_isolate_pageblock; | |
1577 | spin_unlock_irqrestore(&zone->lock, flags); | |
1578 | ||
0c0e6195 | 1579 | /* removal success */ |
3dcc0571 | 1580 | adjust_managed_page_count(pfn_to_page(start_pfn), -offlined_pages); |
0c0e6195 | 1581 | zone->present_pages -= offlined_pages; |
d702909f CS |
1582 | |
1583 | pgdat_resize_lock(zone->zone_pgdat, &flags); | |
0c0e6195 | 1584 | zone->zone_pgdat->node_present_pages -= offlined_pages; |
d702909f | 1585 | pgdat_resize_unlock(zone->zone_pgdat, &flags); |
7b78d335 | 1586 | |
1b79acc9 KM |
1587 | init_per_zone_wmark_min(); |
1588 | ||
1e8537ba | 1589 | if (!populated_zone(zone)) { |
340175b7 | 1590 | zone_pcp_reset(zone); |
72675e13 | 1591 | build_all_zonelists(NULL); |
1e8537ba XQ |
1592 | } else |
1593 | zone_pcp_update(zone); | |
340175b7 | 1594 | |
d9713679 | 1595 | node_states_clear_node(node, &arg); |
698b1b30 | 1596 | if (arg.status_change_nid >= 0) { |
8fe23e05 | 1597 | kswapd_stop(node); |
698b1b30 VB |
1598 | kcompactd_stop(node); |
1599 | } | |
bce7394a | 1600 | |
0c0e6195 KH |
1601 | vm_total_pages = nr_free_pagecache_pages(); |
1602 | writeback_set_ratelimit(); | |
7b78d335 YG |
1603 | |
1604 | memory_notify(MEM_OFFLINE, &arg); | |
feee6b29 | 1605 | remove_pfn_range_from_zone(zone, start_pfn, nr_pages); |
381eab4a | 1606 | mem_hotplug_done(); |
0c0e6195 KH |
1607 | return 0; |
1608 | ||
79605093 MH |
1609 | failed_removal_isolated: |
1610 | undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE); | |
c4efe484 | 1611 | memory_notify(MEM_CANCEL_OFFLINE, &arg); |
0c0e6195 | 1612 | failed_removal: |
79605093 | 1613 | pr_debug("memory offlining [mem %#010llx-%#010llx] failed due to %s\n", |
e33e33b4 | 1614 | (unsigned long long) start_pfn << PAGE_SHIFT, |
79605093 MH |
1615 | ((unsigned long long) end_pfn << PAGE_SHIFT) - 1, |
1616 | reason); | |
0c0e6195 | 1617 | /* pushback to free area */ |
381eab4a | 1618 | mem_hotplug_done(); |
0c0e6195 KH |
1619 | return ret; |
1620 | } | |
71088785 | 1621 | |
a16cee10 WC |
1622 | int offline_pages(unsigned long start_pfn, unsigned long nr_pages) |
1623 | { | |
ecde0f3e | 1624 | return __offline_pages(start_pfn, start_pfn + nr_pages); |
a16cee10 WC |
1625 | } |
1626 | ||
d6de9d53 | 1627 | static int check_memblock_offlined_cb(struct memory_block *mem, void *arg) |
bbc76be6 WC |
1628 | { |
1629 | int ret = !is_memblock_offlined(mem); | |
1630 | ||
349daa0f RD |
1631 | if (unlikely(ret)) { |
1632 | phys_addr_t beginpa, endpa; | |
1633 | ||
1634 | beginpa = PFN_PHYS(section_nr_to_pfn(mem->start_section_nr)); | |
b6c88d3b | 1635 | endpa = beginpa + memory_block_size_bytes() - 1; |
756a025f | 1636 | pr_warn("removing memory fails, because memory [%pa-%pa] is onlined\n", |
349daa0f | 1637 | &beginpa, &endpa); |
bbc76be6 | 1638 | |
eca499ab PT |
1639 | return -EBUSY; |
1640 | } | |
1641 | return 0; | |
bbc76be6 WC |
1642 | } |
1643 | ||
0f1cfe9d | 1644 | static int check_cpu_on_node(pg_data_t *pgdat) |
60a5a19e | 1645 | { |
60a5a19e TC |
1646 | int cpu; |
1647 | ||
1648 | for_each_present_cpu(cpu) { | |
1649 | if (cpu_to_node(cpu) == pgdat->node_id) | |
1650 | /* | |
1651 | * the cpu on this node isn't removed, and we can't | |
1652 | * offline this node. | |
1653 | */ | |
1654 | return -EBUSY; | |
1655 | } | |
1656 | ||
1657 | return 0; | |
1658 | } | |
1659 | ||
2c91f8fc DH |
1660 | static int check_no_memblock_for_node_cb(struct memory_block *mem, void *arg) |
1661 | { | |
1662 | int nid = *(int *)arg; | |
1663 | ||
1664 | /* | |
1665 | * If a memory block belongs to multiple nodes, the stored nid is not | |
1666 | * reliable. However, such blocks are always online (e.g., cannot get | |
1667 | * offlined) and, therefore, are still spanned by the node. | |
1668 | */ | |
1669 | return mem->nid == nid ? -EEXIST : 0; | |
1670 | } | |
1671 | ||
0f1cfe9d TK |
1672 | /** |
1673 | * try_offline_node | |
e8b098fc | 1674 | * @nid: the node ID |
0f1cfe9d TK |
1675 | * |
1676 | * Offline a node if all memory sections and cpus of the node are removed. | |
1677 | * | |
1678 | * NOTE: The caller must call lock_device_hotplug() to serialize hotplug | |
1679 | * and online/offline operations before this call. | |
1680 | */ | |
90b30cdc | 1681 | void try_offline_node(int nid) |
60a5a19e | 1682 | { |
d822b86a | 1683 | pg_data_t *pgdat = NODE_DATA(nid); |
2c91f8fc | 1684 | int rc; |
60a5a19e | 1685 | |
2c91f8fc DH |
1686 | /* |
1687 | * If the node still spans pages (especially ZONE_DEVICE), don't | |
1688 | * offline it. A node spans memory after move_pfn_range_to_zone(), | |
1689 | * e.g., after the memory block was onlined. | |
1690 | */ | |
1691 | if (pgdat->node_spanned_pages) | |
1692 | return; | |
60a5a19e | 1693 | |
2c91f8fc DH |
1694 | /* |
1695 | * Especially offline memory blocks might not be spanned by the | |
1696 | * node. They will get spanned by the node once they get onlined. | |
1697 | * However, they link to the node in sysfs and can get onlined later. | |
1698 | */ | |
1699 | rc = for_each_memory_block(&nid, check_no_memblock_for_node_cb); | |
1700 | if (rc) | |
60a5a19e | 1701 | return; |
60a5a19e | 1702 | |
46a3679b | 1703 | if (check_cpu_on_node(pgdat)) |
60a5a19e TC |
1704 | return; |
1705 | ||
1706 | /* | |
1707 | * all memory/cpu of this node are removed, we can offline this | |
1708 | * node now. | |
1709 | */ | |
1710 | node_set_offline(nid); | |
1711 | unregister_one_node(nid); | |
1712 | } | |
90b30cdc | 1713 | EXPORT_SYMBOL(try_offline_node); |
60a5a19e | 1714 | |
d9eb1417 DH |
1715 | static void __release_memory_resource(resource_size_t start, |
1716 | resource_size_t size) | |
1717 | { | |
1718 | int ret; | |
1719 | ||
1720 | /* | |
1721 | * When removing memory in the same granularity as it was added, | |
1722 | * this function never fails. It might only fail if resources | |
1723 | * have to be adjusted or split. We'll ignore the error, as | |
1724 | * removing of memory cannot fail. | |
1725 | */ | |
1726 | ret = release_mem_region_adjustable(&iomem_resource, start, size); | |
1727 | if (ret) { | |
1728 | resource_size_t endres = start + size - 1; | |
1729 | ||
1730 | pr_warn("Unable to release resource <%pa-%pa> (%d)\n", | |
1731 | &start, &endres, ret); | |
1732 | } | |
1733 | } | |
1734 | ||
eca499ab | 1735 | static int __ref try_remove_memory(int nid, u64 start, u64 size) |
bbc76be6 | 1736 | { |
eca499ab | 1737 | int rc = 0; |
993c1aad | 1738 | |
27356f54 TK |
1739 | BUG_ON(check_hotplug_memory_range(start, size)); |
1740 | ||
6677e3ea | 1741 | /* |
242831eb | 1742 | * All memory blocks must be offlined before removing memory. Check |
eca499ab | 1743 | * whether all memory blocks in question are offline and return error |
242831eb | 1744 | * if this is not the case. |
6677e3ea | 1745 | */ |
fbcf73ce | 1746 | rc = walk_memory_blocks(start, size, NULL, check_memblock_offlined_cb); |
eca499ab PT |
1747 | if (rc) |
1748 | goto done; | |
6677e3ea | 1749 | |
46c66c4b YI |
1750 | /* remove memmap entry */ |
1751 | firmware_map_remove(start, start + size, "System RAM"); | |
4c4b7f9b | 1752 | |
f1037ec0 DW |
1753 | /* |
1754 | * Memory block device removal under the device_hotplug_lock is | |
1755 | * a barrier against racing online attempts. | |
1756 | */ | |
4c4b7f9b | 1757 | remove_memory_block_devices(start, size); |
46c66c4b | 1758 | |
f1037ec0 DW |
1759 | mem_hotplug_begin(); |
1760 | ||
2c2a5af6 | 1761 | arch_remove_memory(nid, start, size, NULL); |
32d1fe8f AK |
1762 | memblock_free(start, size); |
1763 | memblock_remove(start, size); | |
d9eb1417 | 1764 | __release_memory_resource(start, size); |
24d335ca | 1765 | |
60a5a19e TC |
1766 | try_offline_node(nid); |
1767 | ||
eca499ab | 1768 | done: |
bfc8c901 | 1769 | mem_hotplug_done(); |
eca499ab | 1770 | return rc; |
71088785 | 1771 | } |
d15e5926 | 1772 | |
eca499ab PT |
1773 | /** |
1774 | * remove_memory | |
1775 | * @nid: the node ID | |
1776 | * @start: physical address of the region to remove | |
1777 | * @size: size of the region to remove | |
1778 | * | |
1779 | * NOTE: The caller must call lock_device_hotplug() to serialize hotplug | |
1780 | * and online/offline operations before this call, as required by | |
1781 | * try_offline_node(). | |
1782 | */ | |
1783 | void __remove_memory(int nid, u64 start, u64 size) | |
1784 | { | |
1785 | ||
1786 | /* | |
29a90db9 | 1787 | * trigger BUG() if some memory is not offlined prior to calling this |
eca499ab PT |
1788 | * function |
1789 | */ | |
1790 | if (try_remove_memory(nid, start, size)) | |
1791 | BUG(); | |
1792 | } | |
1793 | ||
1794 | /* | |
1795 | * Remove memory if every memory block is offline, otherwise return -EBUSY is | |
1796 | * some memory is not offline | |
1797 | */ | |
1798 | int remove_memory(int nid, u64 start, u64 size) | |
d15e5926 | 1799 | { |
eca499ab PT |
1800 | int rc; |
1801 | ||
d15e5926 | 1802 | lock_device_hotplug(); |
eca499ab | 1803 | rc = try_remove_memory(nid, start, size); |
d15e5926 | 1804 | unlock_device_hotplug(); |
eca499ab PT |
1805 | |
1806 | return rc; | |
d15e5926 | 1807 | } |
71088785 | 1808 | EXPORT_SYMBOL_GPL(remove_memory); |
aba6efc4 | 1809 | #endif /* CONFIG_MEMORY_HOTREMOVE */ |