Commit | Line | Data |
---|---|---|
3947be19 DH |
1 | /* |
2 | * linux/mm/memory_hotplug.c | |
3 | * | |
4 | * Copyright (C) | |
5 | */ | |
6 | ||
3947be19 DH |
7 | #include <linux/stddef.h> |
8 | #include <linux/mm.h> | |
174cd4b1 | 9 | #include <linux/sched/signal.h> |
3947be19 DH |
10 | #include <linux/swap.h> |
11 | #include <linux/interrupt.h> | |
12 | #include <linux/pagemap.h> | |
3947be19 | 13 | #include <linux/compiler.h> |
b95f1b31 | 14 | #include <linux/export.h> |
3947be19 | 15 | #include <linux/pagevec.h> |
2d1d43f6 | 16 | #include <linux/writeback.h> |
3947be19 DH |
17 | #include <linux/slab.h> |
18 | #include <linux/sysctl.h> | |
19 | #include <linux/cpu.h> | |
20 | #include <linux/memory.h> | |
4b94ffdc | 21 | #include <linux/memremap.h> |
3947be19 DH |
22 | #include <linux/memory_hotplug.h> |
23 | #include <linux/highmem.h> | |
24 | #include <linux/vmalloc.h> | |
0a547039 | 25 | #include <linux/ioport.h> |
0c0e6195 KH |
26 | #include <linux/delay.h> |
27 | #include <linux/migrate.h> | |
28 | #include <linux/page-isolation.h> | |
71088785 | 29 | #include <linux/pfn.h> |
6ad696d2 | 30 | #include <linux/suspend.h> |
6d9c285a | 31 | #include <linux/mm_inline.h> |
d96ae530 | 32 | #include <linux/firmware-map.h> |
60a5a19e | 33 | #include <linux/stop_machine.h> |
c8721bbb | 34 | #include <linux/hugetlb.h> |
c5320926 | 35 | #include <linux/memblock.h> |
f784a3f1 | 36 | #include <linux/bootmem.h> |
698b1b30 | 37 | #include <linux/compaction.h> |
3947be19 DH |
38 | |
39 | #include <asm/tlbflush.h> | |
40 | ||
1e5ad9a3 AB |
41 | #include "internal.h" |
42 | ||
9d0ad8ca DK |
43 | /* |
44 | * online_page_callback contains pointer to current page onlining function. | |
45 | * Initially it is generic_online_page(). If it is required it could be | |
46 | * changed by calling set_online_page_callback() for callback registration | |
47 | * and restore_online_page_callback() for generic callback restore. | |
48 | */ | |
49 | ||
50 | static void generic_online_page(struct page *page); | |
51 | ||
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 | |
45e0b78b KM |
99 | /* add this memory to iomem resource */ |
100 | static struct resource *register_memory_resource(u64 start, u64 size) | |
101 | { | |
5042db43 | 102 | struct resource *res, *conflict; |
45e0b78b | 103 | res = kzalloc(sizeof(struct resource), GFP_KERNEL); |
6f754ba4 VK |
104 | if (!res) |
105 | return ERR_PTR(-ENOMEM); | |
45e0b78b KM |
106 | |
107 | res->name = "System RAM"; | |
108 | res->start = start; | |
109 | res->end = start + size - 1; | |
782b8664 | 110 | res->flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY; |
5042db43 JG |
111 | conflict = request_resource_conflict(&iomem_resource, res); |
112 | if (conflict) { | |
113 | if (conflict->desc == IORES_DESC_DEVICE_PRIVATE_MEMORY) { | |
114 | pr_debug("Device unaddressable memory block " | |
115 | "memory hotplug at %#010llx !\n", | |
116 | (unsigned long long)start); | |
117 | } | |
4996eed8 | 118 | pr_debug("System RAM resource %pR cannot be added\n", res); |
45e0b78b | 119 | kfree(res); |
6f754ba4 | 120 | return ERR_PTR(-EEXIST); |
45e0b78b KM |
121 | } |
122 | return res; | |
123 | } | |
124 | ||
125 | static void release_memory_resource(struct resource *res) | |
126 | { | |
127 | if (!res) | |
128 | return; | |
129 | release_resource(res); | |
130 | kfree(res); | |
131 | return; | |
132 | } | |
133 | ||
53947027 | 134 | #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE |
46723bfa YI |
135 | void get_page_bootmem(unsigned long info, struct page *page, |
136 | unsigned long type) | |
04753278 | 137 | { |
ddffe98d | 138 | page->freelist = (void *)type; |
04753278 YG |
139 | SetPagePrivate(page); |
140 | set_page_private(page, info); | |
fe896d18 | 141 | page_ref_inc(page); |
04753278 YG |
142 | } |
143 | ||
170a5a7e | 144 | void put_page_bootmem(struct page *page) |
04753278 | 145 | { |
5f24ce5f | 146 | unsigned long type; |
04753278 | 147 | |
ddffe98d | 148 | type = (unsigned long) page->freelist; |
5f24ce5f AA |
149 | BUG_ON(type < MEMORY_HOTPLUG_MIN_BOOTMEM_TYPE || |
150 | type > MEMORY_HOTPLUG_MAX_BOOTMEM_TYPE); | |
04753278 | 151 | |
fe896d18 | 152 | if (page_ref_dec_return(page) == 1) { |
ddffe98d | 153 | page->freelist = NULL; |
04753278 YG |
154 | ClearPagePrivate(page); |
155 | set_page_private(page, 0); | |
5f24ce5f | 156 | INIT_LIST_HEAD(&page->lru); |
170a5a7e | 157 | free_reserved_page(page); |
04753278 | 158 | } |
04753278 YG |
159 | } |
160 | ||
46723bfa YI |
161 | #ifdef CONFIG_HAVE_BOOTMEM_INFO_NODE |
162 | #ifndef CONFIG_SPARSEMEM_VMEMMAP | |
d92bc318 | 163 | static void register_page_bootmem_info_section(unsigned long start_pfn) |
04753278 YG |
164 | { |
165 | unsigned long *usemap, mapsize, section_nr, i; | |
166 | struct mem_section *ms; | |
167 | struct page *page, *memmap; | |
168 | ||
04753278 YG |
169 | section_nr = pfn_to_section_nr(start_pfn); |
170 | ms = __nr_to_section(section_nr); | |
171 | ||
172 | /* Get section's memmap address */ | |
173 | memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr); | |
174 | ||
175 | /* | |
176 | * Get page for the memmap's phys address | |
177 | * XXX: need more consideration for sparse_vmemmap... | |
178 | */ | |
179 | page = virt_to_page(memmap); | |
180 | mapsize = sizeof(struct page) * PAGES_PER_SECTION; | |
181 | mapsize = PAGE_ALIGN(mapsize) >> PAGE_SHIFT; | |
182 | ||
183 | /* remember memmap's page */ | |
184 | for (i = 0; i < mapsize; i++, page++) | |
185 | get_page_bootmem(section_nr, page, SECTION_INFO); | |
186 | ||
187 | usemap = __nr_to_section(section_nr)->pageblock_flags; | |
188 | page = virt_to_page(usemap); | |
189 | ||
190 | mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT; | |
191 | ||
192 | for (i = 0; i < mapsize; i++, page++) | |
af370fb8 | 193 | get_page_bootmem(section_nr, page, MIX_SECTION_INFO); |
04753278 YG |
194 | |
195 | } | |
46723bfa YI |
196 | #else /* CONFIG_SPARSEMEM_VMEMMAP */ |
197 | static void register_page_bootmem_info_section(unsigned long start_pfn) | |
198 | { | |
199 | unsigned long *usemap, mapsize, section_nr, i; | |
200 | struct mem_section *ms; | |
201 | struct page *page, *memmap; | |
202 | ||
203 | if (!pfn_valid(start_pfn)) | |
204 | return; | |
205 | ||
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 | ||
213 | usemap = __nr_to_section(section_nr)->pageblock_flags; | |
214 | page = virt_to_page(usemap); | |
215 | ||
216 | mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT; | |
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 | |
f1dd2cd1 | 252 | static int __meminit __add_section(int nid, unsigned long phys_start_pfn, |
7b73d978 | 253 | struct vmem_altmap *altmap, bool want_memblock) |
3947be19 | 254 | { |
3947be19 | 255 | int ret; |
f1dd2cd1 | 256 | int i; |
3947be19 | 257 | |
ebd15302 KH |
258 | if (pfn_valid(phys_start_pfn)) |
259 | return -EEXIST; | |
260 | ||
7b73d978 | 261 | ret = sparse_add_one_section(NODE_DATA(nid), phys_start_pfn, altmap); |
3947be19 DH |
262 | if (ret < 0) |
263 | return ret; | |
264 | ||
f1dd2cd1 MH |
265 | /* |
266 | * Make all the pages reserved so that nobody will stumble over half | |
267 | * initialized state. | |
1b7176ae | 268 | * FIXME: We also have to associate it with a node because page_to_nid |
f1dd2cd1 MH |
269 | * relies on having page with the proper node. |
270 | */ | |
271 | for (i = 0; i < PAGES_PER_SECTION; i++) { | |
272 | unsigned long pfn = phys_start_pfn + i; | |
273 | struct page *page; | |
274 | if (!pfn_valid(pfn)) | |
275 | continue; | |
718127cc | 276 | |
f1dd2cd1 MH |
277 | page = pfn_to_page(pfn); |
278 | set_page_node(page, nid); | |
279 | SetPageReserved(page); | |
280 | } | |
718127cc | 281 | |
1b862aec MH |
282 | if (!want_memblock) |
283 | return 0; | |
284 | ||
c04fc586 | 285 | return register_new_memory(nid, __pfn_to_section(phys_start_pfn)); |
3947be19 DH |
286 | } |
287 | ||
4edd7cef DR |
288 | /* |
289 | * Reasonably generic function for adding memory. It is | |
290 | * expected that archs that support memory hotplug will | |
291 | * call this function after deciding the zone to which to | |
292 | * add the new pages. | |
293 | */ | |
f1dd2cd1 | 294 | int __ref __add_pages(int nid, unsigned long phys_start_pfn, |
24e6d5a5 CH |
295 | unsigned long nr_pages, struct vmem_altmap *altmap, |
296 | bool want_memblock) | |
4edd7cef DR |
297 | { |
298 | unsigned long i; | |
299 | int err = 0; | |
300 | int start_sec, end_sec; | |
4b94ffdc | 301 | |
4edd7cef DR |
302 | /* during initialize mem_map, align hot-added range to section */ |
303 | start_sec = pfn_to_section_nr(phys_start_pfn); | |
304 | end_sec = pfn_to_section_nr(phys_start_pfn + nr_pages - 1); | |
305 | ||
4b94ffdc DW |
306 | if (altmap) { |
307 | /* | |
308 | * Validate altmap is within bounds of the total request | |
309 | */ | |
310 | if (altmap->base_pfn != phys_start_pfn | |
311 | || vmem_altmap_offset(altmap) > nr_pages) { | |
312 | pr_warn_once("memory add fail, invalid altmap\n"); | |
7cf91a98 JK |
313 | err = -EINVAL; |
314 | goto out; | |
4b94ffdc DW |
315 | } |
316 | altmap->alloc = 0; | |
317 | } | |
318 | ||
4edd7cef | 319 | for (i = start_sec; i <= end_sec; i++) { |
7b73d978 CH |
320 | err = __add_section(nid, section_nr_to_pfn(i), altmap, |
321 | want_memblock); | |
4edd7cef DR |
322 | |
323 | /* | |
324 | * EEXIST is finally dealt with by ioresource collision | |
325 | * check. see add_memory() => register_memory_resource() | |
326 | * Warning will be printed if there is collision. | |
327 | */ | |
328 | if (err && (err != -EEXIST)) | |
329 | break; | |
330 | err = 0; | |
f64ac5e6 | 331 | cond_resched(); |
4edd7cef | 332 | } |
c435a390 | 333 | vmemmap_populate_print_last(); |
7cf91a98 | 334 | out: |
4edd7cef DR |
335 | return err; |
336 | } | |
4edd7cef DR |
337 | |
338 | #ifdef CONFIG_MEMORY_HOTREMOVE | |
815121d2 | 339 | /* find the smallest valid pfn in the range [start_pfn, end_pfn) */ |
d09b0137 | 340 | static unsigned long find_smallest_section_pfn(int nid, struct zone *zone, |
815121d2 YI |
341 | unsigned long start_pfn, |
342 | unsigned long end_pfn) | |
343 | { | |
344 | struct mem_section *ms; | |
345 | ||
346 | for (; start_pfn < end_pfn; start_pfn += PAGES_PER_SECTION) { | |
347 | ms = __pfn_to_section(start_pfn); | |
348 | ||
349 | if (unlikely(!valid_section(ms))) | |
350 | continue; | |
351 | ||
352 | if (unlikely(pfn_to_nid(start_pfn) != nid)) | |
353 | continue; | |
354 | ||
355 | if (zone && zone != page_zone(pfn_to_page(start_pfn))) | |
356 | continue; | |
357 | ||
358 | return start_pfn; | |
359 | } | |
360 | ||
361 | return 0; | |
362 | } | |
363 | ||
364 | /* find the biggest valid pfn in the range [start_pfn, end_pfn). */ | |
d09b0137 | 365 | static unsigned long find_biggest_section_pfn(int nid, struct zone *zone, |
815121d2 YI |
366 | unsigned long start_pfn, |
367 | unsigned long end_pfn) | |
368 | { | |
369 | struct mem_section *ms; | |
370 | unsigned long pfn; | |
371 | ||
372 | /* pfn is the end pfn of a memory section. */ | |
373 | pfn = end_pfn - 1; | |
374 | for (; pfn >= start_pfn; pfn -= PAGES_PER_SECTION) { | |
375 | ms = __pfn_to_section(pfn); | |
376 | ||
377 | if (unlikely(!valid_section(ms))) | |
378 | continue; | |
379 | ||
380 | if (unlikely(pfn_to_nid(pfn) != nid)) | |
381 | continue; | |
382 | ||
383 | if (zone && zone != page_zone(pfn_to_page(pfn))) | |
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 YI |
398 | unsigned long pfn; |
399 | struct mem_section *ms; | |
400 | int nid = zone_to_nid(zone); | |
401 | ||
402 | zone_span_writelock(zone); | |
403 | if (zone_start_pfn == start_pfn) { | |
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, | |
411 | zone_end_pfn); | |
412 | if (pfn) { | |
413 | zone->zone_start_pfn = pfn; | |
414 | zone->spanned_pages = zone_end_pfn - pfn; | |
415 | } | |
416 | } else if (zone_end_pfn == end_pfn) { | |
417 | /* | |
418 | * If the section is biggest section in the zone, it need | |
419 | * shrink zone->spanned_pages. | |
420 | * In this case, we find second biggest valid mem_section for | |
421 | * shrinking zone. | |
422 | */ | |
423 | pfn = find_biggest_section_pfn(nid, zone, zone_start_pfn, | |
424 | start_pfn); | |
425 | if (pfn) | |
426 | zone->spanned_pages = pfn - zone_start_pfn + 1; | |
427 | } | |
428 | ||
429 | /* | |
430 | * The section is not biggest or smallest mem_section in the zone, it | |
431 | * only creates a hole in the zone. So in this case, we need not | |
432 | * change the zone. But perhaps, the zone has only hole data. Thus | |
433 | * it check the zone has only hole or not. | |
434 | */ | |
435 | pfn = zone_start_pfn; | |
436 | for (; pfn < zone_end_pfn; pfn += PAGES_PER_SECTION) { | |
437 | ms = __pfn_to_section(pfn); | |
438 | ||
439 | if (unlikely(!valid_section(ms))) | |
440 | continue; | |
441 | ||
442 | if (page_zone(pfn_to_page(pfn)) != zone) | |
443 | continue; | |
444 | ||
445 | /* If the section is current section, it continues the loop */ | |
446 | if (start_pfn == pfn) | |
447 | continue; | |
448 | ||
449 | /* If we find valid section, we have nothing to do */ | |
450 | zone_span_writeunlock(zone); | |
451 | return; | |
452 | } | |
453 | ||
454 | /* The zone has no valid section */ | |
455 | zone->zone_start_pfn = 0; | |
456 | zone->spanned_pages = 0; | |
457 | zone_span_writeunlock(zone); | |
458 | } | |
459 | ||
460 | static void shrink_pgdat_span(struct pglist_data *pgdat, | |
461 | unsigned long start_pfn, unsigned long end_pfn) | |
462 | { | |
83285c72 XQ |
463 | unsigned long pgdat_start_pfn = pgdat->node_start_pfn; |
464 | unsigned long p = pgdat_end_pfn(pgdat); /* pgdat_end_pfn namespace clash */ | |
465 | unsigned long pgdat_end_pfn = p; | |
815121d2 YI |
466 | unsigned long pfn; |
467 | struct mem_section *ms; | |
468 | int nid = pgdat->node_id; | |
469 | ||
470 | if (pgdat_start_pfn == start_pfn) { | |
471 | /* | |
472 | * If the section is smallest section in the pgdat, it need | |
473 | * shrink pgdat->node_start_pfn and pgdat->node_spanned_pages. | |
474 | * In this case, we find second smallest valid mem_section | |
475 | * for shrinking zone. | |
476 | */ | |
477 | pfn = find_smallest_section_pfn(nid, NULL, end_pfn, | |
478 | pgdat_end_pfn); | |
479 | if (pfn) { | |
480 | pgdat->node_start_pfn = pfn; | |
481 | pgdat->node_spanned_pages = pgdat_end_pfn - pfn; | |
482 | } | |
483 | } else if (pgdat_end_pfn == end_pfn) { | |
484 | /* | |
485 | * If the section is biggest section in the pgdat, it need | |
486 | * shrink pgdat->node_spanned_pages. | |
487 | * In this case, we find second biggest valid mem_section for | |
488 | * shrinking zone. | |
489 | */ | |
490 | pfn = find_biggest_section_pfn(nid, NULL, pgdat_start_pfn, | |
491 | start_pfn); | |
492 | if (pfn) | |
493 | pgdat->node_spanned_pages = pfn - pgdat_start_pfn + 1; | |
494 | } | |
495 | ||
496 | /* | |
497 | * If the section is not biggest or smallest mem_section in the pgdat, | |
498 | * it only creates a hole in the pgdat. So in this case, we need not | |
499 | * change the pgdat. | |
500 | * But perhaps, the pgdat has only hole data. Thus it check the pgdat | |
501 | * has only hole or not. | |
502 | */ | |
503 | pfn = pgdat_start_pfn; | |
504 | for (; pfn < pgdat_end_pfn; pfn += PAGES_PER_SECTION) { | |
505 | ms = __pfn_to_section(pfn); | |
506 | ||
507 | if (unlikely(!valid_section(ms))) | |
508 | continue; | |
509 | ||
510 | if (pfn_to_nid(pfn) != nid) | |
511 | continue; | |
512 | ||
513 | /* If the section is current section, it continues the loop */ | |
514 | if (start_pfn == pfn) | |
515 | continue; | |
516 | ||
517 | /* If we find valid section, we have nothing to do */ | |
518 | return; | |
519 | } | |
520 | ||
521 | /* The pgdat has no valid section */ | |
522 | pgdat->node_start_pfn = 0; | |
523 | pgdat->node_spanned_pages = 0; | |
524 | } | |
525 | ||
526 | static void __remove_zone(struct zone *zone, unsigned long start_pfn) | |
527 | { | |
528 | struct pglist_data *pgdat = zone->zone_pgdat; | |
529 | int nr_pages = PAGES_PER_SECTION; | |
815121d2 YI |
530 | unsigned long flags; |
531 | ||
815121d2 YI |
532 | pgdat_resize_lock(zone->zone_pgdat, &flags); |
533 | shrink_zone_span(zone, start_pfn, start_pfn + nr_pages); | |
534 | shrink_pgdat_span(pgdat, start_pfn, start_pfn + nr_pages); | |
535 | pgdat_resize_unlock(zone->zone_pgdat, &flags); | |
536 | } | |
537 | ||
4b94ffdc DW |
538 | static int __remove_section(struct zone *zone, struct mem_section *ms, |
539 | unsigned long map_offset) | |
ea01ea93 | 540 | { |
815121d2 YI |
541 | unsigned long start_pfn; |
542 | int scn_nr; | |
ea01ea93 BP |
543 | int ret = -EINVAL; |
544 | ||
545 | if (!valid_section(ms)) | |
546 | return ret; | |
547 | ||
548 | ret = unregister_memory_section(ms); | |
549 | if (ret) | |
550 | return ret; | |
551 | ||
815121d2 | 552 | scn_nr = __section_nr(ms); |
1dd2bfc8 | 553 | start_pfn = section_nr_to_pfn((unsigned long)scn_nr); |
815121d2 YI |
554 | __remove_zone(zone, start_pfn); |
555 | ||
4b94ffdc | 556 | sparse_remove_one_section(zone, ms, map_offset); |
ea01ea93 BP |
557 | return 0; |
558 | } | |
559 | ||
ea01ea93 BP |
560 | /** |
561 | * __remove_pages() - remove sections of pages from a zone | |
562 | * @zone: zone from which pages need to be removed | |
563 | * @phys_start_pfn: starting pageframe (must be aligned to start of a section) | |
564 | * @nr_pages: number of pages to remove (must be multiple of section size) | |
565 | * | |
566 | * Generic helper function to remove section mappings and sysfs entries | |
567 | * for the section of the memory we are removing. Caller needs to make | |
568 | * sure that pages are marked reserved and zones are adjust properly by | |
569 | * calling offline_pages(). | |
570 | */ | |
571 | int __remove_pages(struct zone *zone, unsigned long phys_start_pfn, | |
da024512 | 572 | unsigned long nr_pages, struct vmem_altmap *altmap) |
ea01ea93 | 573 | { |
fe74ebb1 | 574 | unsigned long i; |
4b94ffdc DW |
575 | unsigned long map_offset = 0; |
576 | int sections_to_remove, ret = 0; | |
577 | ||
578 | /* In the ZONE_DEVICE case device driver owns the memory region */ | |
579 | if (is_dev_zone(zone)) { | |
4b94ffdc DW |
580 | if (altmap) |
581 | map_offset = vmem_altmap_offset(altmap); | |
582 | } else { | |
583 | resource_size_t start, size; | |
584 | ||
585 | start = phys_start_pfn << PAGE_SHIFT; | |
586 | size = nr_pages * PAGE_SIZE; | |
587 | ||
588 | ret = release_mem_region_adjustable(&iomem_resource, start, | |
589 | size); | |
590 | if (ret) { | |
591 | resource_size_t endres = start + size - 1; | |
592 | ||
593 | pr_warn("Unable to release resource <%pa-%pa> (%d)\n", | |
594 | &start, &endres, ret); | |
595 | } | |
596 | } | |
ea01ea93 | 597 | |
7cf91a98 JK |
598 | clear_zone_contiguous(zone); |
599 | ||
ea01ea93 BP |
600 | /* |
601 | * We can only remove entire sections | |
602 | */ | |
603 | BUG_ON(phys_start_pfn & ~PAGE_SECTION_MASK); | |
604 | BUG_ON(nr_pages % PAGES_PER_SECTION); | |
605 | ||
ea01ea93 BP |
606 | sections_to_remove = nr_pages / PAGES_PER_SECTION; |
607 | for (i = 0; i < sections_to_remove; i++) { | |
608 | unsigned long pfn = phys_start_pfn + i*PAGES_PER_SECTION; | |
4b94ffdc DW |
609 | |
610 | ret = __remove_section(zone, __pfn_to_section(pfn), map_offset); | |
611 | map_offset = 0; | |
ea01ea93 BP |
612 | if (ret) |
613 | break; | |
614 | } | |
7cf91a98 JK |
615 | |
616 | set_zone_contiguous(zone); | |
617 | ||
ea01ea93 BP |
618 | return ret; |
619 | } | |
4edd7cef | 620 | #endif /* CONFIG_MEMORY_HOTREMOVE */ |
ea01ea93 | 621 | |
9d0ad8ca DK |
622 | int set_online_page_callback(online_page_callback_t callback) |
623 | { | |
624 | int rc = -EINVAL; | |
625 | ||
bfc8c901 VD |
626 | get_online_mems(); |
627 | mutex_lock(&online_page_callback_lock); | |
9d0ad8ca DK |
628 | |
629 | if (online_page_callback == generic_online_page) { | |
630 | online_page_callback = callback; | |
631 | rc = 0; | |
632 | } | |
633 | ||
bfc8c901 VD |
634 | mutex_unlock(&online_page_callback_lock); |
635 | put_online_mems(); | |
9d0ad8ca DK |
636 | |
637 | return rc; | |
638 | } | |
639 | EXPORT_SYMBOL_GPL(set_online_page_callback); | |
640 | ||
641 | int restore_online_page_callback(online_page_callback_t callback) | |
642 | { | |
643 | int rc = -EINVAL; | |
644 | ||
bfc8c901 VD |
645 | get_online_mems(); |
646 | mutex_lock(&online_page_callback_lock); | |
9d0ad8ca DK |
647 | |
648 | if (online_page_callback == callback) { | |
649 | online_page_callback = generic_online_page; | |
650 | rc = 0; | |
651 | } | |
652 | ||
bfc8c901 VD |
653 | mutex_unlock(&online_page_callback_lock); |
654 | put_online_mems(); | |
9d0ad8ca DK |
655 | |
656 | return rc; | |
657 | } | |
658 | EXPORT_SYMBOL_GPL(restore_online_page_callback); | |
659 | ||
660 | void __online_page_set_limits(struct page *page) | |
180c06ef | 661 | { |
9d0ad8ca DK |
662 | } |
663 | EXPORT_SYMBOL_GPL(__online_page_set_limits); | |
664 | ||
665 | void __online_page_increment_counters(struct page *page) | |
666 | { | |
3dcc0571 | 667 | adjust_managed_page_count(page, 1); |
9d0ad8ca DK |
668 | } |
669 | EXPORT_SYMBOL_GPL(__online_page_increment_counters); | |
180c06ef | 670 | |
9d0ad8ca DK |
671 | void __online_page_free(struct page *page) |
672 | { | |
3dcc0571 | 673 | __free_reserved_page(page); |
180c06ef | 674 | } |
9d0ad8ca DK |
675 | EXPORT_SYMBOL_GPL(__online_page_free); |
676 | ||
677 | static void generic_online_page(struct page *page) | |
678 | { | |
679 | __online_page_set_limits(page); | |
680 | __online_page_increment_counters(page); | |
681 | __online_page_free(page); | |
682 | } | |
180c06ef | 683 | |
75884fb1 KH |
684 | static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages, |
685 | void *arg) | |
3947be19 DH |
686 | { |
687 | unsigned long i; | |
75884fb1 KH |
688 | unsigned long onlined_pages = *(unsigned long *)arg; |
689 | struct page *page; | |
2d070eab | 690 | |
75884fb1 KH |
691 | if (PageReserved(pfn_to_page(start_pfn))) |
692 | for (i = 0; i < nr_pages; i++) { | |
693 | page = pfn_to_page(start_pfn + i); | |
9d0ad8ca | 694 | (*online_page_callback)(page); |
75884fb1 KH |
695 | onlined_pages++; |
696 | } | |
2d070eab MH |
697 | |
698 | online_mem_sections(start_pfn, start_pfn + nr_pages); | |
699 | ||
75884fb1 KH |
700 | *(unsigned long *)arg = onlined_pages; |
701 | return 0; | |
702 | } | |
703 | ||
d9713679 LJ |
704 | /* check which state of node_states will be changed when online memory */ |
705 | static void node_states_check_changes_online(unsigned long nr_pages, | |
706 | struct zone *zone, struct memory_notify *arg) | |
707 | { | |
708 | int nid = zone_to_nid(zone); | |
709 | enum zone_type zone_last = ZONE_NORMAL; | |
710 | ||
711 | /* | |
6715ddf9 LJ |
712 | * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY] |
713 | * contains nodes which have zones of 0...ZONE_NORMAL, | |
714 | * set zone_last to ZONE_NORMAL. | |
d9713679 | 715 | * |
6715ddf9 LJ |
716 | * If we don't have HIGHMEM nor movable node, |
717 | * node_states[N_NORMAL_MEMORY] contains nodes which have zones of | |
718 | * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE. | |
d9713679 | 719 | */ |
6715ddf9 | 720 | if (N_MEMORY == N_NORMAL_MEMORY) |
d9713679 LJ |
721 | zone_last = ZONE_MOVABLE; |
722 | ||
723 | /* | |
724 | * if the memory to be online is in a zone of 0...zone_last, and | |
725 | * the zones of 0...zone_last don't have memory before online, we will | |
726 | * need to set the node to node_states[N_NORMAL_MEMORY] after | |
727 | * the memory is online. | |
728 | */ | |
729 | if (zone_idx(zone) <= zone_last && !node_state(nid, N_NORMAL_MEMORY)) | |
730 | arg->status_change_nid_normal = nid; | |
731 | else | |
732 | arg->status_change_nid_normal = -1; | |
733 | ||
6715ddf9 LJ |
734 | #ifdef CONFIG_HIGHMEM |
735 | /* | |
736 | * If we have movable node, node_states[N_HIGH_MEMORY] | |
737 | * contains nodes which have zones of 0...ZONE_HIGHMEM, | |
738 | * set zone_last to ZONE_HIGHMEM. | |
739 | * | |
740 | * If we don't have movable node, node_states[N_NORMAL_MEMORY] | |
741 | * contains nodes which have zones of 0...ZONE_MOVABLE, | |
742 | * set zone_last to ZONE_MOVABLE. | |
743 | */ | |
744 | zone_last = ZONE_HIGHMEM; | |
745 | if (N_MEMORY == N_HIGH_MEMORY) | |
746 | zone_last = ZONE_MOVABLE; | |
747 | ||
748 | if (zone_idx(zone) <= zone_last && !node_state(nid, N_HIGH_MEMORY)) | |
749 | arg->status_change_nid_high = nid; | |
750 | else | |
751 | arg->status_change_nid_high = -1; | |
752 | #else | |
753 | arg->status_change_nid_high = arg->status_change_nid_normal; | |
754 | #endif | |
755 | ||
d9713679 LJ |
756 | /* |
757 | * if the node don't have memory befor online, we will need to | |
6715ddf9 | 758 | * set the node to node_states[N_MEMORY] after the memory |
d9713679 LJ |
759 | * is online. |
760 | */ | |
6715ddf9 | 761 | if (!node_state(nid, N_MEMORY)) |
d9713679 LJ |
762 | arg->status_change_nid = nid; |
763 | else | |
764 | arg->status_change_nid = -1; | |
765 | } | |
766 | ||
767 | static void node_states_set_node(int node, struct memory_notify *arg) | |
768 | { | |
769 | if (arg->status_change_nid_normal >= 0) | |
770 | node_set_state(node, N_NORMAL_MEMORY); | |
771 | ||
6715ddf9 LJ |
772 | if (arg->status_change_nid_high >= 0) |
773 | node_set_state(node, N_HIGH_MEMORY); | |
774 | ||
775 | node_set_state(node, N_MEMORY); | |
d9713679 LJ |
776 | } |
777 | ||
f1dd2cd1 MH |
778 | static void __meminit resize_zone_range(struct zone *zone, unsigned long start_pfn, |
779 | unsigned long nr_pages) | |
780 | { | |
781 | unsigned long old_end_pfn = zone_end_pfn(zone); | |
782 | ||
783 | if (zone_is_empty(zone) || start_pfn < zone->zone_start_pfn) | |
784 | zone->zone_start_pfn = start_pfn; | |
785 | ||
786 | zone->spanned_pages = max(start_pfn + nr_pages, old_end_pfn) - zone->zone_start_pfn; | |
787 | } | |
788 | ||
789 | static void __meminit resize_pgdat_range(struct pglist_data *pgdat, unsigned long start_pfn, | |
790 | unsigned long nr_pages) | |
791 | { | |
792 | unsigned long old_end_pfn = pgdat_end_pfn(pgdat); | |
793 | ||
794 | if (!pgdat->node_spanned_pages || start_pfn < pgdat->node_start_pfn) | |
795 | pgdat->node_start_pfn = start_pfn; | |
796 | ||
797 | pgdat->node_spanned_pages = max(start_pfn + nr_pages, old_end_pfn) - pgdat->node_start_pfn; | |
798 | } | |
799 | ||
cdf72f25 | 800 | void __ref move_pfn_range_to_zone(struct zone *zone, |
f1dd2cd1 MH |
801 | unsigned long start_pfn, unsigned long nr_pages) |
802 | { | |
803 | struct pglist_data *pgdat = zone->zone_pgdat; | |
804 | int nid = pgdat->node_id; | |
805 | unsigned long flags; | |
df429ac0 | 806 | |
f1dd2cd1 MH |
807 | if (zone_is_empty(zone)) |
808 | init_currently_empty_zone(zone, start_pfn, nr_pages); | |
df429ac0 | 809 | |
f1dd2cd1 MH |
810 | clear_zone_contiguous(zone); |
811 | ||
812 | /* TODO Huh pgdat is irqsave while zone is not. It used to be like that before */ | |
813 | pgdat_resize_lock(pgdat, &flags); | |
814 | zone_span_writelock(zone); | |
815 | resize_zone_range(zone, start_pfn, nr_pages); | |
816 | zone_span_writeunlock(zone); | |
817 | resize_pgdat_range(pgdat, start_pfn, nr_pages); | |
818 | pgdat_resize_unlock(pgdat, &flags); | |
819 | ||
820 | /* | |
821 | * TODO now we have a visible range of pages which are not associated | |
822 | * with their zone properly. Not nice but set_pfnblock_flags_mask | |
823 | * expects the zone spans the pfn range. All the pages in the range | |
824 | * are reserved so nobody should be touching them so we should be safe | |
825 | */ | |
826 | memmap_init_zone(nr_pages, nid, zone_idx(zone), start_pfn, MEMMAP_HOTPLUG); | |
827 | ||
828 | set_zone_contiguous(zone); | |
829 | } | |
830 | ||
c246a213 MH |
831 | /* |
832 | * Returns a default kernel memory zone for the given pfn range. | |
833 | * If no kernel zone covers this pfn range it will automatically go | |
834 | * to the ZONE_NORMAL. | |
835 | */ | |
c6f03e29 | 836 | static struct zone *default_kernel_zone_for_pfn(int nid, unsigned long start_pfn, |
c246a213 MH |
837 | unsigned long nr_pages) |
838 | { | |
839 | struct pglist_data *pgdat = NODE_DATA(nid); | |
840 | int zid; | |
841 | ||
842 | for (zid = 0; zid <= ZONE_NORMAL; zid++) { | |
843 | struct zone *zone = &pgdat->node_zones[zid]; | |
844 | ||
845 | if (zone_intersects(zone, start_pfn, nr_pages)) | |
846 | return zone; | |
847 | } | |
848 | ||
849 | return &pgdat->node_zones[ZONE_NORMAL]; | |
850 | } | |
851 | ||
c6f03e29 MH |
852 | static inline struct zone *default_zone_for_pfn(int nid, unsigned long start_pfn, |
853 | unsigned long nr_pages) | |
e5e68930 | 854 | { |
c6f03e29 MH |
855 | struct zone *kernel_zone = default_kernel_zone_for_pfn(nid, start_pfn, |
856 | nr_pages); | |
857 | struct zone *movable_zone = &NODE_DATA(nid)->node_zones[ZONE_MOVABLE]; | |
858 | bool in_kernel = zone_intersects(kernel_zone, start_pfn, nr_pages); | |
859 | bool in_movable = zone_intersects(movable_zone, start_pfn, nr_pages); | |
e5e68930 MH |
860 | |
861 | /* | |
c6f03e29 MH |
862 | * We inherit the existing zone in a simple case where zones do not |
863 | * overlap in the given range | |
e5e68930 | 864 | */ |
c6f03e29 MH |
865 | if (in_kernel ^ in_movable) |
866 | return (in_kernel) ? kernel_zone : movable_zone; | |
9f123ab5 | 867 | |
c6f03e29 MH |
868 | /* |
869 | * If the range doesn't belong to any zone or two zones overlap in the | |
870 | * given range then we use movable zone only if movable_node is | |
871 | * enabled because we always online to a kernel zone by default. | |
872 | */ | |
873 | return movable_node_enabled ? movable_zone : kernel_zone; | |
9f123ab5 MH |
874 | } |
875 | ||
e5e68930 MH |
876 | struct zone * zone_for_pfn_range(int online_type, int nid, unsigned start_pfn, |
877 | unsigned long nr_pages) | |
f1dd2cd1 | 878 | { |
c6f03e29 MH |
879 | if (online_type == MMOP_ONLINE_KERNEL) |
880 | return default_kernel_zone_for_pfn(nid, start_pfn, nr_pages); | |
f1dd2cd1 | 881 | |
c6f03e29 MH |
882 | if (online_type == MMOP_ONLINE_MOVABLE) |
883 | return &NODE_DATA(nid)->node_zones[ZONE_MOVABLE]; | |
df429ac0 | 884 | |
c6f03e29 | 885 | return default_zone_for_pfn(nid, start_pfn, nr_pages); |
e5e68930 MH |
886 | } |
887 | ||
888 | /* | |
889 | * Associates the given pfn range with the given node and the zone appropriate | |
890 | * for the given online type. | |
891 | */ | |
892 | static struct zone * __meminit move_pfn_range(int online_type, int nid, | |
893 | unsigned long start_pfn, unsigned long nr_pages) | |
894 | { | |
895 | struct zone *zone; | |
896 | ||
897 | zone = zone_for_pfn_range(online_type, nid, start_pfn, nr_pages); | |
f1dd2cd1 MH |
898 | move_pfn_range_to_zone(zone, start_pfn, nr_pages); |
899 | return zone; | |
df429ac0 | 900 | } |
75884fb1 | 901 | |
b93e0f32 | 902 | /* Must be protected by mem_hotplug_begin() or a device_lock */ |
511c2aba | 903 | int __ref online_pages(unsigned long pfn, unsigned long nr_pages, int online_type) |
75884fb1 | 904 | { |
aa47228a | 905 | unsigned long flags; |
3947be19 DH |
906 | unsigned long onlined_pages = 0; |
907 | struct zone *zone; | |
6811378e | 908 | int need_zonelists_rebuild = 0; |
7b78d335 YG |
909 | int nid; |
910 | int ret; | |
911 | struct memory_notify arg; | |
912 | ||
f1dd2cd1 | 913 | nid = pfn_to_nid(pfn); |
f1dd2cd1 MH |
914 | /* associate pfn range with the zone */ |
915 | zone = move_pfn_range(online_type, nid, pfn, nr_pages); | |
916 | ||
7b78d335 YG |
917 | arg.start_pfn = pfn; |
918 | arg.nr_pages = nr_pages; | |
d9713679 | 919 | node_states_check_changes_online(nr_pages, zone, &arg); |
7b78d335 | 920 | |
7b78d335 YG |
921 | ret = memory_notify(MEM_GOING_ONLINE, &arg); |
922 | ret = notifier_to_errno(ret); | |
e33e33b4 CY |
923 | if (ret) |
924 | goto failed_addition; | |
925 | ||
6811378e YG |
926 | /* |
927 | * If this zone is not populated, then it is not in zonelist. | |
928 | * This means the page allocator ignores this zone. | |
929 | * So, zonelist must be updated after online. | |
930 | */ | |
6dcd73d7 | 931 | if (!populated_zone(zone)) { |
6811378e | 932 | need_zonelists_rebuild = 1; |
72675e13 | 933 | setup_zone_pageset(zone); |
6dcd73d7 | 934 | } |
6811378e | 935 | |
908eedc6 | 936 | ret = walk_system_ram_range(pfn, nr_pages, &onlined_pages, |
75884fb1 | 937 | online_pages_range); |
fd8a4221 | 938 | if (ret) { |
6dcd73d7 WC |
939 | if (need_zonelists_rebuild) |
940 | zone_pcp_reset(zone); | |
e33e33b4 | 941 | goto failed_addition; |
fd8a4221 GL |
942 | } |
943 | ||
3947be19 | 944 | zone->present_pages += onlined_pages; |
aa47228a CS |
945 | |
946 | pgdat_resize_lock(zone->zone_pgdat, &flags); | |
f2937be5 | 947 | zone->zone_pgdat->node_present_pages += onlined_pages; |
aa47228a CS |
948 | pgdat_resize_unlock(zone->zone_pgdat, &flags); |
949 | ||
08dff7b7 | 950 | if (onlined_pages) { |
e888ca35 | 951 | node_states_set_node(nid, &arg); |
08dff7b7 | 952 | if (need_zonelists_rebuild) |
72675e13 | 953 | build_all_zonelists(NULL); |
08dff7b7 JL |
954 | else |
955 | zone_pcp_update(zone); | |
956 | } | |
3947be19 | 957 | |
1b79acc9 KM |
958 | init_per_zone_wmark_min(); |
959 | ||
698b1b30 | 960 | if (onlined_pages) { |
e888ca35 | 961 | kswapd_run(nid); |
698b1b30 VB |
962 | kcompactd_run(nid); |
963 | } | |
61b13993 | 964 | |
1f522509 | 965 | vm_total_pages = nr_free_pagecache_pages(); |
2f7f24ec | 966 | |
2d1d43f6 | 967 | writeback_set_ratelimit(); |
7b78d335 YG |
968 | |
969 | if (onlined_pages) | |
970 | memory_notify(MEM_ONLINE, &arg); | |
30467e0b | 971 | return 0; |
e33e33b4 CY |
972 | |
973 | failed_addition: | |
974 | pr_debug("online_pages [mem %#010llx-%#010llx] failed\n", | |
975 | (unsigned long long) pfn << PAGE_SHIFT, | |
976 | (((unsigned long long) pfn + nr_pages) << PAGE_SHIFT) - 1); | |
977 | memory_notify(MEM_CANCEL_ONLINE, &arg); | |
978 | return ret; | |
3947be19 | 979 | } |
53947027 | 980 | #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */ |
bc02af93 | 981 | |
0bd85420 TC |
982 | static void reset_node_present_pages(pg_data_t *pgdat) |
983 | { | |
984 | struct zone *z; | |
985 | ||
986 | for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++) | |
987 | z->present_pages = 0; | |
988 | ||
989 | pgdat->node_present_pages = 0; | |
990 | } | |
991 | ||
e1319331 HS |
992 | /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */ |
993 | static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start) | |
9af3c2de YG |
994 | { |
995 | struct pglist_data *pgdat; | |
996 | unsigned long zones_size[MAX_NR_ZONES] = {0}; | |
997 | unsigned long zholes_size[MAX_NR_ZONES] = {0}; | |
c8e861a5 | 998 | unsigned long start_pfn = PFN_DOWN(start); |
9af3c2de | 999 | |
a1e565aa TC |
1000 | pgdat = NODE_DATA(nid); |
1001 | if (!pgdat) { | |
1002 | pgdat = arch_alloc_nodedata(nid); | |
1003 | if (!pgdat) | |
1004 | return NULL; | |
9af3c2de | 1005 | |
a1e565aa | 1006 | arch_refresh_nodedata(nid, pgdat); |
b0dc3a34 | 1007 | } else { |
e716f2eb MG |
1008 | /* |
1009 | * Reset the nr_zones, order and classzone_idx before reuse. | |
1010 | * Note that kswapd will init kswapd_classzone_idx properly | |
1011 | * when it starts in the near future. | |
1012 | */ | |
b0dc3a34 | 1013 | pgdat->nr_zones = 0; |
38087d9b MG |
1014 | pgdat->kswapd_order = 0; |
1015 | pgdat->kswapd_classzone_idx = 0; | |
a1e565aa | 1016 | } |
9af3c2de YG |
1017 | |
1018 | /* we can use NODE_DATA(nid) from here */ | |
1019 | ||
1020 | /* init node's zones as empty zones, we don't have any present pages.*/ | |
9109fb7b | 1021 | free_area_init_node(nid, zones_size, start_pfn, zholes_size); |
5830169f | 1022 | pgdat->per_cpu_nodestats = alloc_percpu(struct per_cpu_nodestat); |
9af3c2de | 1023 | |
959ecc48 KH |
1024 | /* |
1025 | * The node we allocated has no zone fallback lists. For avoiding | |
1026 | * to access not-initialized zonelist, build here. | |
1027 | */ | |
72675e13 | 1028 | build_all_zonelists(pgdat); |
959ecc48 | 1029 | |
f784a3f1 TC |
1030 | /* |
1031 | * zone->managed_pages is set to an approximate value in | |
1032 | * free_area_init_core(), which will cause | |
1033 | * /sys/device/system/node/nodeX/meminfo has wrong data. | |
1034 | * So reset it to 0 before any memory is onlined. | |
1035 | */ | |
1036 | reset_node_managed_pages(pgdat); | |
1037 | ||
0bd85420 TC |
1038 | /* |
1039 | * When memory is hot-added, all the memory is in offline state. So | |
1040 | * clear all zones' present_pages because they will be updated in | |
1041 | * online_pages() and offline_pages(). | |
1042 | */ | |
1043 | reset_node_present_pages(pgdat); | |
1044 | ||
9af3c2de YG |
1045 | return pgdat; |
1046 | } | |
1047 | ||
1048 | static void rollback_node_hotadd(int nid, pg_data_t *pgdat) | |
1049 | { | |
1050 | arch_refresh_nodedata(nid, NULL); | |
5830169f | 1051 | free_percpu(pgdat->per_cpu_nodestats); |
9af3c2de YG |
1052 | arch_free_nodedata(pgdat); |
1053 | return; | |
1054 | } | |
1055 | ||
0a547039 | 1056 | |
01b0f197 TK |
1057 | /** |
1058 | * try_online_node - online a node if offlined | |
1059 | * | |
cf23422b | 1060 | * called by cpu_up() to online a node without onlined memory. |
1061 | */ | |
01b0f197 | 1062 | int try_online_node(int nid) |
cf23422b | 1063 | { |
1064 | pg_data_t *pgdat; | |
1065 | int ret; | |
1066 | ||
01b0f197 TK |
1067 | if (node_online(nid)) |
1068 | return 0; | |
1069 | ||
bfc8c901 | 1070 | mem_hotplug_begin(); |
cf23422b | 1071 | pgdat = hotadd_new_pgdat(nid, 0); |
7553e8f2 | 1072 | if (!pgdat) { |
01b0f197 | 1073 | pr_err("Cannot online node %d due to NULL pgdat\n", nid); |
cf23422b | 1074 | ret = -ENOMEM; |
1075 | goto out; | |
1076 | } | |
1077 | node_set_online(nid); | |
1078 | ret = register_one_node(nid); | |
1079 | BUG_ON(ret); | |
cf23422b | 1080 | out: |
bfc8c901 | 1081 | mem_hotplug_done(); |
cf23422b | 1082 | return ret; |
1083 | } | |
1084 | ||
27356f54 TK |
1085 | static int check_hotplug_memory_range(u64 start, u64 size) |
1086 | { | |
c8e861a5 | 1087 | u64 start_pfn = PFN_DOWN(start); |
27356f54 TK |
1088 | u64 nr_pages = size >> PAGE_SHIFT; |
1089 | ||
1090 | /* Memory range must be aligned with section */ | |
1091 | if ((start_pfn & ~PAGE_SECTION_MASK) || | |
1092 | (nr_pages % PAGES_PER_SECTION) || (!nr_pages)) { | |
1093 | pr_err("Section-unaligned hotplug range: start 0x%llx, size 0x%llx\n", | |
1094 | (unsigned long long)start, | |
1095 | (unsigned long long)size); | |
1096 | return -EINVAL; | |
1097 | } | |
1098 | ||
1099 | return 0; | |
1100 | } | |
1101 | ||
31bc3858 VK |
1102 | static int online_memory_block(struct memory_block *mem, void *arg) |
1103 | { | |
dc18d706 | 1104 | return device_online(&mem->dev); |
31bc3858 VK |
1105 | } |
1106 | ||
31168481 | 1107 | /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */ |
31bc3858 | 1108 | int __ref add_memory_resource(int nid, struct resource *res, bool online) |
bc02af93 | 1109 | { |
62cedb9f | 1110 | u64 start, size; |
9af3c2de | 1111 | pg_data_t *pgdat = NULL; |
a1e565aa TC |
1112 | bool new_pgdat; |
1113 | bool new_node; | |
bc02af93 YG |
1114 | int ret; |
1115 | ||
62cedb9f DV |
1116 | start = res->start; |
1117 | size = resource_size(res); | |
1118 | ||
27356f54 TK |
1119 | ret = check_hotplug_memory_range(start, size); |
1120 | if (ret) | |
1121 | return ret; | |
1122 | ||
a1e565aa TC |
1123 | { /* Stupid hack to suppress address-never-null warning */ |
1124 | void *p = NODE_DATA(nid); | |
1125 | new_pgdat = !p; | |
1126 | } | |
ac13c462 | 1127 | |
bfc8c901 | 1128 | mem_hotplug_begin(); |
ac13c462 | 1129 | |
7f36e3e5 TC |
1130 | /* |
1131 | * Add new range to memblock so that when hotadd_new_pgdat() is called | |
1132 | * to allocate new pgdat, get_pfn_range_for_nid() will be able to find | |
1133 | * this new range and calculate total pages correctly. The range will | |
1134 | * be removed at hot-remove time. | |
1135 | */ | |
1136 | memblock_add_node(start, size, nid); | |
1137 | ||
a1e565aa TC |
1138 | new_node = !node_online(nid); |
1139 | if (new_node) { | |
9af3c2de | 1140 | pgdat = hotadd_new_pgdat(nid, start); |
6ad696d2 | 1141 | ret = -ENOMEM; |
9af3c2de | 1142 | if (!pgdat) |
41b9e2d7 | 1143 | goto error; |
9af3c2de YG |
1144 | } |
1145 | ||
bc02af93 | 1146 | /* call arch's memory hotadd */ |
24e6d5a5 | 1147 | ret = arch_add_memory(nid, start, size, NULL, true); |
bc02af93 | 1148 | |
9af3c2de YG |
1149 | if (ret < 0) |
1150 | goto error; | |
1151 | ||
0fc44159 | 1152 | /* we online node here. we can't roll back from here. */ |
9af3c2de YG |
1153 | node_set_online(nid); |
1154 | ||
a1e565aa | 1155 | if (new_node) { |
9037a993 MH |
1156 | unsigned long start_pfn = start >> PAGE_SHIFT; |
1157 | unsigned long nr_pages = size >> PAGE_SHIFT; | |
1158 | ||
1159 | ret = __register_one_node(nid); | |
1160 | if (ret) | |
1161 | goto register_fail; | |
1162 | ||
1163 | /* | |
1164 | * link memory sections under this node. This is already | |
1165 | * done when creatig memory section in register_new_memory | |
1166 | * but that depends to have the node registered so offline | |
1167 | * nodes have to go through register_node. | |
1168 | * TODO clean up this mess. | |
1169 | */ | |
1170 | ret = link_mem_sections(nid, start_pfn, nr_pages); | |
1171 | register_fail: | |
0fc44159 YG |
1172 | /* |
1173 | * If sysfs file of new node can't create, cpu on the node | |
1174 | * can't be hot-added. There is no rollback way now. | |
1175 | * So, check by BUG_ON() to catch it reluctantly.. | |
1176 | */ | |
1177 | BUG_ON(ret); | |
1178 | } | |
1179 | ||
d96ae530 | 1180 | /* create new memmap entry */ |
1181 | firmware_map_add_hotplug(start, start + size, "System RAM"); | |
1182 | ||
31bc3858 VK |
1183 | /* online pages if requested */ |
1184 | if (online) | |
1185 | walk_memory_range(PFN_DOWN(start), PFN_UP(start + size - 1), | |
1186 | NULL, online_memory_block); | |
1187 | ||
6ad696d2 AK |
1188 | goto out; |
1189 | ||
9af3c2de YG |
1190 | error: |
1191 | /* rollback pgdat allocation and others */ | |
dbac61a3 | 1192 | if (new_pgdat && pgdat) |
9af3c2de | 1193 | rollback_node_hotadd(nid, pgdat); |
7f36e3e5 | 1194 | memblock_remove(start, size); |
9af3c2de | 1195 | |
6ad696d2 | 1196 | out: |
bfc8c901 | 1197 | mem_hotplug_done(); |
bc02af93 YG |
1198 | return ret; |
1199 | } | |
62cedb9f DV |
1200 | EXPORT_SYMBOL_GPL(add_memory_resource); |
1201 | ||
1202 | int __ref add_memory(int nid, u64 start, u64 size) | |
1203 | { | |
1204 | struct resource *res; | |
1205 | int ret; | |
1206 | ||
1207 | res = register_memory_resource(start, size); | |
6f754ba4 VK |
1208 | if (IS_ERR(res)) |
1209 | return PTR_ERR(res); | |
62cedb9f | 1210 | |
31bc3858 | 1211 | ret = add_memory_resource(nid, res, memhp_auto_online); |
62cedb9f DV |
1212 | if (ret < 0) |
1213 | release_memory_resource(res); | |
1214 | return ret; | |
1215 | } | |
bc02af93 | 1216 | EXPORT_SYMBOL_GPL(add_memory); |
0c0e6195 KH |
1217 | |
1218 | #ifdef CONFIG_MEMORY_HOTREMOVE | |
5c755e9f BP |
1219 | /* |
1220 | * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy | |
1221 | * set and the size of the free page is given by page_order(). Using this, | |
1222 | * the function determines if the pageblock contains only free pages. | |
1223 | * Due to buddy contraints, a free page at least the size of a pageblock will | |
1224 | * be located at the start of the pageblock | |
1225 | */ | |
1226 | static inline int pageblock_free(struct page *page) | |
1227 | { | |
1228 | return PageBuddy(page) && page_order(page) >= pageblock_order; | |
1229 | } | |
1230 | ||
1231 | /* Return the start of the next active pageblock after a given page */ | |
1232 | static struct page *next_active_pageblock(struct page *page) | |
1233 | { | |
5c755e9f BP |
1234 | /* Ensure the starting page is pageblock-aligned */ |
1235 | BUG_ON(page_to_pfn(page) & (pageblock_nr_pages - 1)); | |
1236 | ||
5c755e9f | 1237 | /* If the entire pageblock is free, move to the end of free page */ |
0dcc48c1 KH |
1238 | if (pageblock_free(page)) { |
1239 | int order; | |
1240 | /* be careful. we don't have locks, page_order can be changed.*/ | |
1241 | order = page_order(page); | |
1242 | if ((order < MAX_ORDER) && (order >= pageblock_order)) | |
1243 | return page + (1 << order); | |
1244 | } | |
5c755e9f | 1245 | |
0dcc48c1 | 1246 | return page + pageblock_nr_pages; |
5c755e9f BP |
1247 | } |
1248 | ||
1249 | /* Checks if this range of memory is likely to be hot-removable. */ | |
c98940f6 | 1250 | bool is_mem_section_removable(unsigned long start_pfn, unsigned long nr_pages) |
5c755e9f | 1251 | { |
5c755e9f BP |
1252 | struct page *page = pfn_to_page(start_pfn); |
1253 | struct page *end_page = page + nr_pages; | |
1254 | ||
1255 | /* Check the starting page of each pageblock within the range */ | |
1256 | for (; page < end_page; page = next_active_pageblock(page)) { | |
49ac8255 | 1257 | if (!is_pageblock_removable_nolock(page)) |
c98940f6 | 1258 | return false; |
49ac8255 | 1259 | cond_resched(); |
5c755e9f BP |
1260 | } |
1261 | ||
1262 | /* All pageblocks in the memory block are likely to be hot-removable */ | |
c98940f6 | 1263 | return true; |
5c755e9f BP |
1264 | } |
1265 | ||
0c0e6195 | 1266 | /* |
deb88a2a | 1267 | * Confirm all pages in a range [start, end) belong to the same zone. |
a96dfddb | 1268 | * When true, return its valid [start, end). |
0c0e6195 | 1269 | */ |
a96dfddb TK |
1270 | int test_pages_in_a_zone(unsigned long start_pfn, unsigned long end_pfn, |
1271 | unsigned long *valid_start, unsigned long *valid_end) | |
0c0e6195 | 1272 | { |
5f0f2887 | 1273 | unsigned long pfn, sec_end_pfn; |
a96dfddb | 1274 | unsigned long start, end; |
0c0e6195 KH |
1275 | struct zone *zone = NULL; |
1276 | struct page *page; | |
1277 | int i; | |
deb88a2a | 1278 | for (pfn = start_pfn, sec_end_pfn = SECTION_ALIGN_UP(start_pfn + 1); |
0c0e6195 | 1279 | pfn < end_pfn; |
deb88a2a | 1280 | pfn = sec_end_pfn, sec_end_pfn += PAGES_PER_SECTION) { |
5f0f2887 AB |
1281 | /* Make sure the memory section is present first */ |
1282 | if (!present_section_nr(pfn_to_section_nr(pfn))) | |
0c0e6195 | 1283 | continue; |
5f0f2887 AB |
1284 | for (; pfn < sec_end_pfn && pfn < end_pfn; |
1285 | pfn += MAX_ORDER_NR_PAGES) { | |
1286 | i = 0; | |
1287 | /* This is just a CONFIG_HOLES_IN_ZONE check.*/ | |
1288 | while ((i < MAX_ORDER_NR_PAGES) && | |
1289 | !pfn_valid_within(pfn + i)) | |
1290 | i++; | |
d6d8c8a4 | 1291 | if (i == MAX_ORDER_NR_PAGES || pfn + i >= end_pfn) |
5f0f2887 AB |
1292 | continue; |
1293 | page = pfn_to_page(pfn + i); | |
1294 | if (zone && page_zone(page) != zone) | |
1295 | return 0; | |
a96dfddb TK |
1296 | if (!zone) |
1297 | start = pfn + i; | |
5f0f2887 | 1298 | zone = page_zone(page); |
a96dfddb | 1299 | end = pfn + MAX_ORDER_NR_PAGES; |
5f0f2887 | 1300 | } |
0c0e6195 | 1301 | } |
deb88a2a | 1302 | |
a96dfddb TK |
1303 | if (zone) { |
1304 | *valid_start = start; | |
d6d8c8a4 | 1305 | *valid_end = min(end, end_pfn); |
deb88a2a | 1306 | return 1; |
a96dfddb | 1307 | } else { |
deb88a2a | 1308 | return 0; |
a96dfddb | 1309 | } |
0c0e6195 KH |
1310 | } |
1311 | ||
1312 | /* | |
0efadf48 YX |
1313 | * Scan pfn range [start,end) to find movable/migratable pages (LRU pages, |
1314 | * non-lru movable pages and hugepages). We scan pfn because it's much | |
1315 | * easier than scanning over linked list. This function returns the pfn | |
1316 | * of the first found movable page if it's found, otherwise 0. | |
0c0e6195 | 1317 | */ |
c8721bbb | 1318 | static unsigned long scan_movable_pages(unsigned long start, unsigned long end) |
0c0e6195 KH |
1319 | { |
1320 | unsigned long pfn; | |
1321 | struct page *page; | |
1322 | for (pfn = start; pfn < end; pfn++) { | |
1323 | if (pfn_valid(pfn)) { | |
1324 | page = pfn_to_page(pfn); | |
1325 | if (PageLRU(page)) | |
1326 | return pfn; | |
0efadf48 YX |
1327 | if (__PageMovable(page)) |
1328 | return pfn; | |
c8721bbb | 1329 | if (PageHuge(page)) { |
7e1f049e | 1330 | if (page_huge_active(page)) |
c8721bbb NH |
1331 | return pfn; |
1332 | else | |
1333 | pfn = round_up(pfn + 1, | |
1334 | 1 << compound_order(page)) - 1; | |
1335 | } | |
0c0e6195 KH |
1336 | } |
1337 | } | |
1338 | return 0; | |
1339 | } | |
1340 | ||
394e31d2 XQ |
1341 | static struct page *new_node_page(struct page *page, unsigned long private, |
1342 | int **result) | |
1343 | { | |
394e31d2 | 1344 | int nid = page_to_nid(page); |
231e97e2 | 1345 | nodemask_t nmask = node_states[N_MEMORY]; |
7f252f27 MH |
1346 | |
1347 | /* | |
1348 | * try to allocate from a different node but reuse this node if there | |
1349 | * are no other online nodes to be used (e.g. we are offlining a part | |
1350 | * of the only existing node) | |
1351 | */ | |
1352 | node_clear(nid, nmask); | |
1353 | if (nodes_empty(nmask)) | |
1354 | node_set(nid, nmask); | |
394e31d2 | 1355 | |
8b913238 | 1356 | return new_page_nodemask(page, nid, &nmask); |
394e31d2 XQ |
1357 | } |
1358 | ||
0c0e6195 KH |
1359 | #define NR_OFFLINE_AT_ONCE_PAGES (256) |
1360 | static int | |
1361 | do_migrate_range(unsigned long start_pfn, unsigned long end_pfn) | |
1362 | { | |
1363 | unsigned long pfn; | |
1364 | struct page *page; | |
1365 | int move_pages = NR_OFFLINE_AT_ONCE_PAGES; | |
1366 | int not_managed = 0; | |
1367 | int ret = 0; | |
1368 | LIST_HEAD(source); | |
1369 | ||
1370 | for (pfn = start_pfn; pfn < end_pfn && move_pages > 0; pfn++) { | |
1371 | if (!pfn_valid(pfn)) | |
1372 | continue; | |
1373 | page = pfn_to_page(pfn); | |
c8721bbb NH |
1374 | |
1375 | if (PageHuge(page)) { | |
1376 | struct page *head = compound_head(page); | |
1377 | pfn = page_to_pfn(head) + (1<<compound_order(head)) - 1; | |
1378 | if (compound_order(head) > PFN_SECTION_SHIFT) { | |
1379 | ret = -EBUSY; | |
1380 | break; | |
1381 | } | |
1382 | if (isolate_huge_page(page, &source)) | |
1383 | move_pages -= 1 << compound_order(head); | |
1384 | continue; | |
8135d892 NH |
1385 | } else if (thp_migration_supported() && PageTransHuge(page)) |
1386 | pfn = page_to_pfn(compound_head(page)) | |
1387 | + hpage_nr_pages(page) - 1; | |
c8721bbb | 1388 | |
700c2a46 | 1389 | if (!get_page_unless_zero(page)) |
0c0e6195 KH |
1390 | continue; |
1391 | /* | |
0efadf48 YX |
1392 | * We can skip free pages. And we can deal with pages on |
1393 | * LRU and non-lru movable pages. | |
0c0e6195 | 1394 | */ |
0efadf48 YX |
1395 | if (PageLRU(page)) |
1396 | ret = isolate_lru_page(page); | |
1397 | else | |
1398 | ret = isolate_movable_page(page, ISOLATE_UNEVICTABLE); | |
0c0e6195 | 1399 | if (!ret) { /* Success */ |
700c2a46 | 1400 | put_page(page); |
62695a84 | 1401 | list_add_tail(&page->lru, &source); |
0c0e6195 | 1402 | move_pages--; |
0efadf48 YX |
1403 | if (!__PageMovable(page)) |
1404 | inc_node_page_state(page, NR_ISOLATED_ANON + | |
1405 | page_is_file_cache(page)); | |
6d9c285a | 1406 | |
0c0e6195 | 1407 | } else { |
0c0e6195 | 1408 | #ifdef CONFIG_DEBUG_VM |
0efadf48 YX |
1409 | pr_alert("failed to isolate pfn %lx\n", pfn); |
1410 | dump_page(page, "isolation failed"); | |
0c0e6195 | 1411 | #endif |
700c2a46 | 1412 | put_page(page); |
25985edc | 1413 | /* Because we don't have big zone->lock. we should |
809c4449 BL |
1414 | check this again here. */ |
1415 | if (page_count(page)) { | |
1416 | not_managed++; | |
f3ab2636 | 1417 | ret = -EBUSY; |
809c4449 BL |
1418 | break; |
1419 | } | |
0c0e6195 KH |
1420 | } |
1421 | } | |
f3ab2636 BL |
1422 | if (!list_empty(&source)) { |
1423 | if (not_managed) { | |
c8721bbb | 1424 | putback_movable_pages(&source); |
f3ab2636 BL |
1425 | goto out; |
1426 | } | |
74c08f98 | 1427 | |
394e31d2 XQ |
1428 | /* Allocate a new page from the nearest neighbor node */ |
1429 | ret = migrate_pages(&source, new_node_page, NULL, 0, | |
9c620e2b | 1430 | MIGRATE_SYNC, MR_MEMORY_HOTPLUG); |
f3ab2636 | 1431 | if (ret) |
c8721bbb | 1432 | putback_movable_pages(&source); |
0c0e6195 | 1433 | } |
0c0e6195 KH |
1434 | out: |
1435 | return ret; | |
1436 | } | |
1437 | ||
1438 | /* | |
1439 | * remove from free_area[] and mark all as Reserved. | |
1440 | */ | |
1441 | static int | |
1442 | offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages, | |
1443 | void *data) | |
1444 | { | |
1445 | __offline_isolated_pages(start, start + nr_pages); | |
1446 | return 0; | |
1447 | } | |
1448 | ||
1449 | static void | |
1450 | offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn) | |
1451 | { | |
908eedc6 | 1452 | walk_system_ram_range(start_pfn, end_pfn - start_pfn, NULL, |
0c0e6195 KH |
1453 | offline_isolated_pages_cb); |
1454 | } | |
1455 | ||
1456 | /* | |
1457 | * Check all pages in range, recoreded as memory resource, are isolated. | |
1458 | */ | |
1459 | static int | |
1460 | check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages, | |
1461 | void *data) | |
1462 | { | |
1463 | int ret; | |
1464 | long offlined = *(long *)data; | |
b023f468 | 1465 | ret = test_pages_isolated(start_pfn, start_pfn + nr_pages, true); |
0c0e6195 KH |
1466 | offlined = nr_pages; |
1467 | if (!ret) | |
1468 | *(long *)data += offlined; | |
1469 | return ret; | |
1470 | } | |
1471 | ||
1472 | static long | |
1473 | check_pages_isolated(unsigned long start_pfn, unsigned long end_pfn) | |
1474 | { | |
1475 | long offlined = 0; | |
1476 | int ret; | |
1477 | ||
908eedc6 | 1478 | ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn, &offlined, |
0c0e6195 KH |
1479 | check_pages_isolated_cb); |
1480 | if (ret < 0) | |
1481 | offlined = (long)ret; | |
1482 | return offlined; | |
1483 | } | |
1484 | ||
c5320926 TC |
1485 | static int __init cmdline_parse_movable_node(char *p) |
1486 | { | |
4932381e | 1487 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
55ac590c | 1488 | movable_node_enabled = true; |
4932381e MH |
1489 | #else |
1490 | pr_warn("movable_node parameter depends on CONFIG_HAVE_MEMBLOCK_NODE_MAP to work properly\n"); | |
1491 | #endif | |
c5320926 TC |
1492 | return 0; |
1493 | } | |
1494 | early_param("movable_node", cmdline_parse_movable_node); | |
1495 | ||
d9713679 LJ |
1496 | /* check which state of node_states will be changed when offline memory */ |
1497 | static void node_states_check_changes_offline(unsigned long nr_pages, | |
1498 | struct zone *zone, struct memory_notify *arg) | |
1499 | { | |
1500 | struct pglist_data *pgdat = zone->zone_pgdat; | |
1501 | unsigned long present_pages = 0; | |
1502 | enum zone_type zt, zone_last = ZONE_NORMAL; | |
1503 | ||
1504 | /* | |
6715ddf9 LJ |
1505 | * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY] |
1506 | * contains nodes which have zones of 0...ZONE_NORMAL, | |
1507 | * set zone_last to ZONE_NORMAL. | |
d9713679 | 1508 | * |
6715ddf9 LJ |
1509 | * If we don't have HIGHMEM nor movable node, |
1510 | * node_states[N_NORMAL_MEMORY] contains nodes which have zones of | |
1511 | * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE. | |
d9713679 | 1512 | */ |
6715ddf9 | 1513 | if (N_MEMORY == N_NORMAL_MEMORY) |
d9713679 LJ |
1514 | zone_last = ZONE_MOVABLE; |
1515 | ||
1516 | /* | |
1517 | * check whether node_states[N_NORMAL_MEMORY] will be changed. | |
1518 | * If the memory to be offline is in a zone of 0...zone_last, | |
1519 | * and it is the last present memory, 0...zone_last will | |
1520 | * become empty after offline , thus we can determind we will | |
1521 | * need to clear the node from node_states[N_NORMAL_MEMORY]. | |
1522 | */ | |
1523 | for (zt = 0; zt <= zone_last; zt++) | |
1524 | present_pages += pgdat->node_zones[zt].present_pages; | |
1525 | if (zone_idx(zone) <= zone_last && nr_pages >= present_pages) | |
1526 | arg->status_change_nid_normal = zone_to_nid(zone); | |
1527 | else | |
1528 | arg->status_change_nid_normal = -1; | |
1529 | ||
6715ddf9 LJ |
1530 | #ifdef CONFIG_HIGHMEM |
1531 | /* | |
1532 | * If we have movable node, node_states[N_HIGH_MEMORY] | |
1533 | * contains nodes which have zones of 0...ZONE_HIGHMEM, | |
1534 | * set zone_last to ZONE_HIGHMEM. | |
1535 | * | |
1536 | * If we don't have movable node, node_states[N_NORMAL_MEMORY] | |
1537 | * contains nodes which have zones of 0...ZONE_MOVABLE, | |
1538 | * set zone_last to ZONE_MOVABLE. | |
1539 | */ | |
1540 | zone_last = ZONE_HIGHMEM; | |
1541 | if (N_MEMORY == N_HIGH_MEMORY) | |
1542 | zone_last = ZONE_MOVABLE; | |
1543 | ||
1544 | for (; zt <= zone_last; zt++) | |
1545 | present_pages += pgdat->node_zones[zt].present_pages; | |
1546 | if (zone_idx(zone) <= zone_last && nr_pages >= present_pages) | |
1547 | arg->status_change_nid_high = zone_to_nid(zone); | |
1548 | else | |
1549 | arg->status_change_nid_high = -1; | |
1550 | #else | |
1551 | arg->status_change_nid_high = arg->status_change_nid_normal; | |
1552 | #endif | |
1553 | ||
d9713679 LJ |
1554 | /* |
1555 | * node_states[N_HIGH_MEMORY] contains nodes which have 0...ZONE_MOVABLE | |
1556 | */ | |
1557 | zone_last = ZONE_MOVABLE; | |
1558 | ||
1559 | /* | |
1560 | * check whether node_states[N_HIGH_MEMORY] will be changed | |
1561 | * If we try to offline the last present @nr_pages from the node, | |
1562 | * we can determind we will need to clear the node from | |
1563 | * node_states[N_HIGH_MEMORY]. | |
1564 | */ | |
1565 | for (; zt <= zone_last; zt++) | |
1566 | present_pages += pgdat->node_zones[zt].present_pages; | |
1567 | if (nr_pages >= present_pages) | |
1568 | arg->status_change_nid = zone_to_nid(zone); | |
1569 | else | |
1570 | arg->status_change_nid = -1; | |
1571 | } | |
1572 | ||
1573 | static void node_states_clear_node(int node, struct memory_notify *arg) | |
1574 | { | |
1575 | if (arg->status_change_nid_normal >= 0) | |
1576 | node_clear_state(node, N_NORMAL_MEMORY); | |
1577 | ||
6715ddf9 LJ |
1578 | if ((N_MEMORY != N_NORMAL_MEMORY) && |
1579 | (arg->status_change_nid_high >= 0)) | |
d9713679 | 1580 | node_clear_state(node, N_HIGH_MEMORY); |
6715ddf9 LJ |
1581 | |
1582 | if ((N_MEMORY != N_HIGH_MEMORY) && | |
1583 | (arg->status_change_nid >= 0)) | |
1584 | node_clear_state(node, N_MEMORY); | |
d9713679 LJ |
1585 | } |
1586 | ||
a16cee10 | 1587 | static int __ref __offline_pages(unsigned long start_pfn, |
ecde0f3e | 1588 | unsigned long end_pfn) |
0c0e6195 | 1589 | { |
ecde0f3e | 1590 | unsigned long pfn, nr_pages; |
0c0e6195 | 1591 | long offlined_pages; |
72b39cfc | 1592 | int ret, node; |
d702909f | 1593 | unsigned long flags; |
a96dfddb | 1594 | unsigned long valid_start, valid_end; |
0c0e6195 | 1595 | struct zone *zone; |
7b78d335 | 1596 | struct memory_notify arg; |
0c0e6195 | 1597 | |
0c0e6195 KH |
1598 | /* at least, alignment against pageblock is necessary */ |
1599 | if (!IS_ALIGNED(start_pfn, pageblock_nr_pages)) | |
1600 | return -EINVAL; | |
1601 | if (!IS_ALIGNED(end_pfn, pageblock_nr_pages)) | |
1602 | return -EINVAL; | |
1603 | /* This makes hotplug much easier...and readable. | |
1604 | we assume this for now. .*/ | |
a96dfddb | 1605 | if (!test_pages_in_a_zone(start_pfn, end_pfn, &valid_start, &valid_end)) |
0c0e6195 | 1606 | return -EINVAL; |
7b78d335 | 1607 | |
a96dfddb | 1608 | zone = page_zone(pfn_to_page(valid_start)); |
7b78d335 YG |
1609 | node = zone_to_nid(zone); |
1610 | nr_pages = end_pfn - start_pfn; | |
1611 | ||
0c0e6195 | 1612 | /* set above range as isolated */ |
b023f468 WC |
1613 | ret = start_isolate_page_range(start_pfn, end_pfn, |
1614 | MIGRATE_MOVABLE, true); | |
0c0e6195 | 1615 | if (ret) |
30467e0b | 1616 | return ret; |
7b78d335 YG |
1617 | |
1618 | arg.start_pfn = start_pfn; | |
1619 | arg.nr_pages = nr_pages; | |
d9713679 | 1620 | node_states_check_changes_offline(nr_pages, zone, &arg); |
7b78d335 YG |
1621 | |
1622 | ret = memory_notify(MEM_GOING_OFFLINE, &arg); | |
1623 | ret = notifier_to_errno(ret); | |
1624 | if (ret) | |
1625 | goto failed_removal; | |
1626 | ||
0c0e6195 | 1627 | pfn = start_pfn; |
0c0e6195 KH |
1628 | repeat: |
1629 | /* start memory hot removal */ | |
0c0e6195 KH |
1630 | ret = -EINTR; |
1631 | if (signal_pending(current)) | |
1632 | goto failed_removal; | |
72b39cfc MH |
1633 | |
1634 | cond_resched(); | |
1635 | lru_add_drain_all_cpuslocked(); | |
1636 | drain_all_pages(zone); | |
0c0e6195 | 1637 | |
c8721bbb NH |
1638 | pfn = scan_movable_pages(start_pfn, end_pfn); |
1639 | if (pfn) { /* We have movable pages */ | |
0c0e6195 | 1640 | ret = do_migrate_range(pfn, end_pfn); |
72b39cfc | 1641 | goto repeat; |
0c0e6195 | 1642 | } |
72b39cfc | 1643 | |
c8721bbb NH |
1644 | /* |
1645 | * dissolve free hugepages in the memory block before doing offlining | |
1646 | * actually in order to make hugetlbfs's object counting consistent. | |
1647 | */ | |
082d5b6b GS |
1648 | ret = dissolve_free_huge_pages(start_pfn, end_pfn); |
1649 | if (ret) | |
1650 | goto failed_removal; | |
0c0e6195 KH |
1651 | /* check again */ |
1652 | offlined_pages = check_pages_isolated(start_pfn, end_pfn); | |
72b39cfc MH |
1653 | if (offlined_pages < 0) |
1654 | goto repeat; | |
e33e33b4 | 1655 | pr_info("Offlined Pages %ld\n", offlined_pages); |
b3834be5 | 1656 | /* Ok, all of our target is isolated. |
0c0e6195 KH |
1657 | We cannot do rollback at this point. */ |
1658 | offline_isolated_pages(start_pfn, end_pfn); | |
dbc0e4ce | 1659 | /* reset pagetype flags and makes migrate type to be MOVABLE */ |
0815f3d8 | 1660 | undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE); |
0c0e6195 | 1661 | /* removal success */ |
3dcc0571 | 1662 | adjust_managed_page_count(pfn_to_page(start_pfn), -offlined_pages); |
0c0e6195 | 1663 | zone->present_pages -= offlined_pages; |
d702909f CS |
1664 | |
1665 | pgdat_resize_lock(zone->zone_pgdat, &flags); | |
0c0e6195 | 1666 | zone->zone_pgdat->node_present_pages -= offlined_pages; |
d702909f | 1667 | pgdat_resize_unlock(zone->zone_pgdat, &flags); |
7b78d335 | 1668 | |
1b79acc9 KM |
1669 | init_per_zone_wmark_min(); |
1670 | ||
1e8537ba | 1671 | if (!populated_zone(zone)) { |
340175b7 | 1672 | zone_pcp_reset(zone); |
72675e13 | 1673 | build_all_zonelists(NULL); |
1e8537ba XQ |
1674 | } else |
1675 | zone_pcp_update(zone); | |
340175b7 | 1676 | |
d9713679 | 1677 | node_states_clear_node(node, &arg); |
698b1b30 | 1678 | if (arg.status_change_nid >= 0) { |
8fe23e05 | 1679 | kswapd_stop(node); |
698b1b30 VB |
1680 | kcompactd_stop(node); |
1681 | } | |
bce7394a | 1682 | |
0c0e6195 KH |
1683 | vm_total_pages = nr_free_pagecache_pages(); |
1684 | writeback_set_ratelimit(); | |
7b78d335 YG |
1685 | |
1686 | memory_notify(MEM_OFFLINE, &arg); | |
0c0e6195 KH |
1687 | return 0; |
1688 | ||
1689 | failed_removal: | |
e33e33b4 CY |
1690 | pr_debug("memory offlining [mem %#010llx-%#010llx] failed\n", |
1691 | (unsigned long long) start_pfn << PAGE_SHIFT, | |
1692 | ((unsigned long long) end_pfn << PAGE_SHIFT) - 1); | |
7b78d335 | 1693 | memory_notify(MEM_CANCEL_OFFLINE, &arg); |
0c0e6195 | 1694 | /* pushback to free area */ |
0815f3d8 | 1695 | undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE); |
0c0e6195 KH |
1696 | return ret; |
1697 | } | |
71088785 | 1698 | |
b93e0f32 | 1699 | /* Must be protected by mem_hotplug_begin() or a device_lock */ |
a16cee10 WC |
1700 | int offline_pages(unsigned long start_pfn, unsigned long nr_pages) |
1701 | { | |
ecde0f3e | 1702 | return __offline_pages(start_pfn, start_pfn + nr_pages); |
a16cee10 | 1703 | } |
e2ff3940 | 1704 | #endif /* CONFIG_MEMORY_HOTREMOVE */ |
a16cee10 | 1705 | |
bbc76be6 WC |
1706 | /** |
1707 | * walk_memory_range - walks through all mem sections in [start_pfn, end_pfn) | |
1708 | * @start_pfn: start pfn of the memory range | |
e05c4bbf | 1709 | * @end_pfn: end pfn of the memory range |
bbc76be6 WC |
1710 | * @arg: argument passed to func |
1711 | * @func: callback for each memory section walked | |
1712 | * | |
1713 | * This function walks through all present mem sections in range | |
1714 | * [start_pfn, end_pfn) and call func on each mem section. | |
1715 | * | |
1716 | * Returns the return value of func. | |
1717 | */ | |
e2ff3940 | 1718 | int walk_memory_range(unsigned long start_pfn, unsigned long end_pfn, |
bbc76be6 | 1719 | void *arg, int (*func)(struct memory_block *, void *)) |
71088785 | 1720 | { |
e90bdb7f WC |
1721 | struct memory_block *mem = NULL; |
1722 | struct mem_section *section; | |
e90bdb7f WC |
1723 | unsigned long pfn, section_nr; |
1724 | int ret; | |
e90bdb7f WC |
1725 | |
1726 | for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) { | |
1727 | section_nr = pfn_to_section_nr(pfn); | |
1728 | if (!present_section_nr(section_nr)) | |
1729 | continue; | |
1730 | ||
1731 | section = __nr_to_section(section_nr); | |
1732 | /* same memblock? */ | |
1733 | if (mem) | |
1734 | if ((section_nr >= mem->start_section_nr) && | |
1735 | (section_nr <= mem->end_section_nr)) | |
1736 | continue; | |
1737 | ||
1738 | mem = find_memory_block_hinted(section, mem); | |
1739 | if (!mem) | |
1740 | continue; | |
1741 | ||
bbc76be6 | 1742 | ret = func(mem, arg); |
e90bdb7f | 1743 | if (ret) { |
bbc76be6 WC |
1744 | kobject_put(&mem->dev.kobj); |
1745 | return ret; | |
e90bdb7f WC |
1746 | } |
1747 | } | |
1748 | ||
1749 | if (mem) | |
1750 | kobject_put(&mem->dev.kobj); | |
1751 | ||
bbc76be6 WC |
1752 | return 0; |
1753 | } | |
1754 | ||
e2ff3940 | 1755 | #ifdef CONFIG_MEMORY_HOTREMOVE |
d6de9d53 | 1756 | static int check_memblock_offlined_cb(struct memory_block *mem, void *arg) |
bbc76be6 WC |
1757 | { |
1758 | int ret = !is_memblock_offlined(mem); | |
1759 | ||
349daa0f RD |
1760 | if (unlikely(ret)) { |
1761 | phys_addr_t beginpa, endpa; | |
1762 | ||
1763 | beginpa = PFN_PHYS(section_nr_to_pfn(mem->start_section_nr)); | |
1764 | endpa = PFN_PHYS(section_nr_to_pfn(mem->end_section_nr + 1))-1; | |
756a025f | 1765 | pr_warn("removing memory fails, because memory [%pa-%pa] is onlined\n", |
349daa0f RD |
1766 | &beginpa, &endpa); |
1767 | } | |
bbc76be6 WC |
1768 | |
1769 | return ret; | |
1770 | } | |
1771 | ||
0f1cfe9d | 1772 | static int check_cpu_on_node(pg_data_t *pgdat) |
60a5a19e | 1773 | { |
60a5a19e TC |
1774 | int cpu; |
1775 | ||
1776 | for_each_present_cpu(cpu) { | |
1777 | if (cpu_to_node(cpu) == pgdat->node_id) | |
1778 | /* | |
1779 | * the cpu on this node isn't removed, and we can't | |
1780 | * offline this node. | |
1781 | */ | |
1782 | return -EBUSY; | |
1783 | } | |
1784 | ||
1785 | return 0; | |
1786 | } | |
1787 | ||
0f1cfe9d | 1788 | static void unmap_cpu_on_node(pg_data_t *pgdat) |
e13fe869 WC |
1789 | { |
1790 | #ifdef CONFIG_ACPI_NUMA | |
e13fe869 WC |
1791 | int cpu; |
1792 | ||
1793 | for_each_possible_cpu(cpu) | |
1794 | if (cpu_to_node(cpu) == pgdat->node_id) | |
1795 | numa_clear_node(cpu); | |
1796 | #endif | |
1797 | } | |
1798 | ||
0f1cfe9d | 1799 | static int check_and_unmap_cpu_on_node(pg_data_t *pgdat) |
e13fe869 | 1800 | { |
0f1cfe9d | 1801 | int ret; |
e13fe869 | 1802 | |
0f1cfe9d | 1803 | ret = check_cpu_on_node(pgdat); |
e13fe869 WC |
1804 | if (ret) |
1805 | return ret; | |
1806 | ||
1807 | /* | |
1808 | * the node will be offlined when we come here, so we can clear | |
1809 | * the cpu_to_node() now. | |
1810 | */ | |
1811 | ||
0f1cfe9d | 1812 | unmap_cpu_on_node(pgdat); |
e13fe869 WC |
1813 | return 0; |
1814 | } | |
1815 | ||
0f1cfe9d TK |
1816 | /** |
1817 | * try_offline_node | |
1818 | * | |
1819 | * Offline a node if all memory sections and cpus of the node are removed. | |
1820 | * | |
1821 | * NOTE: The caller must call lock_device_hotplug() to serialize hotplug | |
1822 | * and online/offline operations before this call. | |
1823 | */ | |
90b30cdc | 1824 | void try_offline_node(int nid) |
60a5a19e | 1825 | { |
d822b86a WC |
1826 | pg_data_t *pgdat = NODE_DATA(nid); |
1827 | unsigned long start_pfn = pgdat->node_start_pfn; | |
1828 | unsigned long end_pfn = start_pfn + pgdat->node_spanned_pages; | |
60a5a19e TC |
1829 | unsigned long pfn; |
1830 | ||
1831 | for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) { | |
1832 | unsigned long section_nr = pfn_to_section_nr(pfn); | |
1833 | ||
1834 | if (!present_section_nr(section_nr)) | |
1835 | continue; | |
1836 | ||
1837 | if (pfn_to_nid(pfn) != nid) | |
1838 | continue; | |
1839 | ||
1840 | /* | |
1841 | * some memory sections of this node are not removed, and we | |
1842 | * can't offline node now. | |
1843 | */ | |
1844 | return; | |
1845 | } | |
1846 | ||
0f1cfe9d | 1847 | if (check_and_unmap_cpu_on_node(pgdat)) |
60a5a19e TC |
1848 | return; |
1849 | ||
1850 | /* | |
1851 | * all memory/cpu of this node are removed, we can offline this | |
1852 | * node now. | |
1853 | */ | |
1854 | node_set_offline(nid); | |
1855 | unregister_one_node(nid); | |
1856 | } | |
90b30cdc | 1857 | EXPORT_SYMBOL(try_offline_node); |
60a5a19e | 1858 | |
0f1cfe9d TK |
1859 | /** |
1860 | * remove_memory | |
1861 | * | |
1862 | * NOTE: The caller must call lock_device_hotplug() to serialize hotplug | |
1863 | * and online/offline operations before this call, as required by | |
1864 | * try_offline_node(). | |
1865 | */ | |
242831eb | 1866 | void __ref remove_memory(int nid, u64 start, u64 size) |
bbc76be6 | 1867 | { |
242831eb | 1868 | int ret; |
993c1aad | 1869 | |
27356f54 TK |
1870 | BUG_ON(check_hotplug_memory_range(start, size)); |
1871 | ||
bfc8c901 | 1872 | mem_hotplug_begin(); |
6677e3ea YI |
1873 | |
1874 | /* | |
242831eb RW |
1875 | * All memory blocks must be offlined before removing memory. Check |
1876 | * whether all memory blocks in question are offline and trigger a BUG() | |
1877 | * if this is not the case. | |
6677e3ea | 1878 | */ |
242831eb | 1879 | ret = walk_memory_range(PFN_DOWN(start), PFN_UP(start + size - 1), NULL, |
d6de9d53 | 1880 | check_memblock_offlined_cb); |
bfc8c901 | 1881 | if (ret) |
242831eb | 1882 | BUG(); |
6677e3ea | 1883 | |
46c66c4b YI |
1884 | /* remove memmap entry */ |
1885 | firmware_map_remove(start, start + size, "System RAM"); | |
f9126ab9 XQ |
1886 | memblock_free(start, size); |
1887 | memblock_remove(start, size); | |
46c66c4b | 1888 | |
da024512 | 1889 | arch_remove_memory(start, size, NULL); |
24d335ca | 1890 | |
60a5a19e TC |
1891 | try_offline_node(nid); |
1892 | ||
bfc8c901 | 1893 | mem_hotplug_done(); |
71088785 | 1894 | } |
71088785 | 1895 | EXPORT_SYMBOL_GPL(remove_memory); |
aba6efc4 | 1896 | #endif /* CONFIG_MEMORY_HOTREMOVE */ |