]>
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> |
85953e73 | 38 | #include <linux/rmap.h> |
3947be19 DH |
39 | |
40 | #include <asm/tlbflush.h> | |
41 | ||
1e5ad9a3 AB |
42 | #include "internal.h" |
43 | ||
9d0ad8ca DK |
44 | /* |
45 | * online_page_callback contains pointer to current page onlining function. | |
46 | * Initially it is generic_online_page(). If it is required it could be | |
47 | * changed by calling set_online_page_callback() for callback registration | |
48 | * and restore_online_page_callback() for generic callback restore. | |
49 | */ | |
50 | ||
51 | static void generic_online_page(struct page *page); | |
52 | ||
53 | static online_page_callback_t online_page_callback = generic_online_page; | |
bfc8c901 | 54 | static DEFINE_MUTEX(online_page_callback_lock); |
9d0ad8ca | 55 | |
3f906ba2 | 56 | DEFINE_STATIC_PERCPU_RWSEM(mem_hotplug_lock); |
bfc8c901 | 57 | |
46e81a18 SF |
58 | static int default_kernel_zone = ZONE_NORMAL; |
59 | ||
3f906ba2 TG |
60 | void get_online_mems(void) |
61 | { | |
62 | percpu_down_read(&mem_hotplug_lock); | |
63 | } | |
bfc8c901 | 64 | |
3f906ba2 TG |
65 | void put_online_mems(void) |
66 | { | |
67 | percpu_up_read(&mem_hotplug_lock); | |
68 | } | |
bfc8c901 | 69 | |
4932381e MH |
70 | bool movable_node_enabled = false; |
71 | ||
8604d9e5 | 72 | #ifndef CONFIG_MEMORY_HOTPLUG_DEFAULT_ONLINE |
31bc3858 | 73 | bool memhp_auto_online; |
8604d9e5 VK |
74 | #else |
75 | bool memhp_auto_online = true; | |
76 | #endif | |
31bc3858 VK |
77 | EXPORT_SYMBOL_GPL(memhp_auto_online); |
78 | ||
86dd995d VK |
79 | static int __init setup_memhp_default_state(char *str) |
80 | { | |
81 | if (!strcmp(str, "online")) | |
82 | memhp_auto_online = true; | |
83 | else if (!strcmp(str, "offline")) | |
84 | memhp_auto_online = false; | |
85 | ||
86 | return 1; | |
87 | } | |
88 | __setup("memhp_default_state=", setup_memhp_default_state); | |
89 | ||
30467e0b | 90 | void mem_hotplug_begin(void) |
20d6c96b | 91 | { |
3f906ba2 TG |
92 | cpus_read_lock(); |
93 | percpu_down_write(&mem_hotplug_lock); | |
20d6c96b KM |
94 | } |
95 | ||
30467e0b | 96 | void mem_hotplug_done(void) |
bfc8c901 | 97 | { |
3f906ba2 TG |
98 | percpu_up_write(&mem_hotplug_lock); |
99 | cpus_read_unlock(); | |
bfc8c901 | 100 | } |
20d6c96b | 101 | |
45e0b78b KM |
102 | /* add this memory to iomem resource */ |
103 | static struct resource *register_memory_resource(u64 start, u64 size) | |
104 | { | |
5042db43 | 105 | struct resource *res, *conflict; |
45e0b78b | 106 | res = kzalloc(sizeof(struct resource), GFP_KERNEL); |
6f754ba4 VK |
107 | if (!res) |
108 | return ERR_PTR(-ENOMEM); | |
45e0b78b KM |
109 | |
110 | res->name = "System RAM"; | |
111 | res->start = start; | |
112 | res->end = start + size - 1; | |
782b8664 | 113 | res->flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY; |
5042db43 JG |
114 | conflict = request_resource_conflict(&iomem_resource, res); |
115 | if (conflict) { | |
116 | if (conflict->desc == IORES_DESC_DEVICE_PRIVATE_MEMORY) { | |
117 | pr_debug("Device unaddressable memory block " | |
118 | "memory hotplug at %#010llx !\n", | |
119 | (unsigned long long)start); | |
120 | } | |
4996eed8 | 121 | pr_debug("System RAM resource %pR cannot be added\n", res); |
45e0b78b | 122 | kfree(res); |
6f754ba4 | 123 | return ERR_PTR(-EEXIST); |
45e0b78b KM |
124 | } |
125 | return res; | |
126 | } | |
127 | ||
128 | static void release_memory_resource(struct resource *res) | |
129 | { | |
130 | if (!res) | |
131 | return; | |
132 | release_resource(res); | |
133 | kfree(res); | |
134 | return; | |
135 | } | |
136 | ||
53947027 | 137 | #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE |
46723bfa YI |
138 | void get_page_bootmem(unsigned long info, struct page *page, |
139 | unsigned long type) | |
04753278 | 140 | { |
ddffe98d | 141 | page->freelist = (void *)type; |
04753278 YG |
142 | SetPagePrivate(page); |
143 | set_page_private(page, info); | |
fe896d18 | 144 | page_ref_inc(page); |
04753278 YG |
145 | } |
146 | ||
170a5a7e | 147 | void put_page_bootmem(struct page *page) |
04753278 | 148 | { |
5f24ce5f | 149 | unsigned long type; |
04753278 | 150 | |
ddffe98d | 151 | type = (unsigned long) page->freelist; |
5f24ce5f AA |
152 | BUG_ON(type < MEMORY_HOTPLUG_MIN_BOOTMEM_TYPE || |
153 | type > MEMORY_HOTPLUG_MAX_BOOTMEM_TYPE); | |
04753278 | 154 | |
fe896d18 | 155 | if (page_ref_dec_return(page) == 1) { |
ddffe98d | 156 | page->freelist = NULL; |
04753278 YG |
157 | ClearPagePrivate(page); |
158 | set_page_private(page, 0); | |
5f24ce5f | 159 | INIT_LIST_HEAD(&page->lru); |
170a5a7e | 160 | free_reserved_page(page); |
04753278 | 161 | } |
04753278 YG |
162 | } |
163 | ||
46723bfa YI |
164 | #ifdef CONFIG_HAVE_BOOTMEM_INFO_NODE |
165 | #ifndef CONFIG_SPARSEMEM_VMEMMAP | |
d92bc318 | 166 | static void register_page_bootmem_info_section(unsigned long start_pfn) |
04753278 YG |
167 | { |
168 | unsigned long *usemap, mapsize, section_nr, i; | |
169 | struct mem_section *ms; | |
170 | struct page *page, *memmap; | |
171 | ||
04753278 YG |
172 | section_nr = pfn_to_section_nr(start_pfn); |
173 | ms = __nr_to_section(section_nr); | |
174 | ||
175 | /* Get section's memmap address */ | |
176 | memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr); | |
177 | ||
178 | /* | |
179 | * Get page for the memmap's phys address | |
180 | * XXX: need more consideration for sparse_vmemmap... | |
181 | */ | |
182 | page = virt_to_page(memmap); | |
183 | mapsize = sizeof(struct page) * PAGES_PER_SECTION; | |
184 | mapsize = PAGE_ALIGN(mapsize) >> PAGE_SHIFT; | |
185 | ||
186 | /* remember memmap's page */ | |
187 | for (i = 0; i < mapsize; i++, page++) | |
188 | get_page_bootmem(section_nr, page, SECTION_INFO); | |
189 | ||
190 | usemap = __nr_to_section(section_nr)->pageblock_flags; | |
191 | page = virt_to_page(usemap); | |
192 | ||
193 | mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT; | |
194 | ||
195 | for (i = 0; i < mapsize; i++, page++) | |
af370fb8 | 196 | get_page_bootmem(section_nr, page, MIX_SECTION_INFO); |
04753278 YG |
197 | |
198 | } | |
46723bfa YI |
199 | #else /* CONFIG_SPARSEMEM_VMEMMAP */ |
200 | static void register_page_bootmem_info_section(unsigned long start_pfn) | |
201 | { | |
202 | unsigned long *usemap, mapsize, section_nr, i; | |
203 | struct mem_section *ms; | |
204 | struct page *page, *memmap; | |
205 | ||
206 | if (!pfn_valid(start_pfn)) | |
207 | return; | |
208 | ||
209 | section_nr = pfn_to_section_nr(start_pfn); | |
210 | ms = __nr_to_section(section_nr); | |
211 | ||
212 | memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr); | |
213 | ||
214 | register_page_bootmem_memmap(section_nr, memmap, PAGES_PER_SECTION); | |
215 | ||
216 | usemap = __nr_to_section(section_nr)->pageblock_flags; | |
217 | page = virt_to_page(usemap); | |
218 | ||
219 | mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT; | |
220 | ||
221 | for (i = 0; i < mapsize; i++, page++) | |
222 | get_page_bootmem(section_nr, page, MIX_SECTION_INFO); | |
223 | } | |
224 | #endif /* !CONFIG_SPARSEMEM_VMEMMAP */ | |
04753278 | 225 | |
7ded384a | 226 | void __init register_page_bootmem_info_node(struct pglist_data *pgdat) |
04753278 YG |
227 | { |
228 | unsigned long i, pfn, end_pfn, nr_pages; | |
229 | int node = pgdat->node_id; | |
230 | struct page *page; | |
04753278 YG |
231 | |
232 | nr_pages = PAGE_ALIGN(sizeof(struct pglist_data)) >> PAGE_SHIFT; | |
233 | page = virt_to_page(pgdat); | |
234 | ||
235 | for (i = 0; i < nr_pages; i++, page++) | |
236 | get_page_bootmem(node, page, NODE_INFO); | |
237 | ||
04753278 | 238 | pfn = pgdat->node_start_pfn; |
c1f19495 | 239 | end_pfn = pgdat_end_pfn(pgdat); |
04753278 | 240 | |
7e9f5eb0 | 241 | /* register section info */ |
f14851af | 242 | for (; pfn < end_pfn; pfn += PAGES_PER_SECTION) { |
243 | /* | |
244 | * Some platforms can assign the same pfn to multiple nodes - on | |
245 | * node0 as well as nodeN. To avoid registering a pfn against | |
246 | * multiple nodes we check that this pfn does not already | |
7e9f5eb0 | 247 | * reside in some other nodes. |
f14851af | 248 | */ |
f65e91df | 249 | if (pfn_valid(pfn) && (early_pfn_to_nid(pfn) == node)) |
f14851af | 250 | register_page_bootmem_info_section(pfn); |
251 | } | |
04753278 | 252 | } |
46723bfa | 253 | #endif /* CONFIG_HAVE_BOOTMEM_INFO_NODE */ |
04753278 | 254 | |
f1dd2cd1 MH |
255 | static int __meminit __add_section(int nid, unsigned long phys_start_pfn, |
256 | bool want_memblock) | |
3947be19 | 257 | { |
3947be19 | 258 | int ret; |
f1dd2cd1 | 259 | int i; |
3947be19 | 260 | |
ebd15302 KH |
261 | if (pfn_valid(phys_start_pfn)) |
262 | return -EEXIST; | |
263 | ||
f1dd2cd1 | 264 | ret = sparse_add_one_section(NODE_DATA(nid), phys_start_pfn); |
3947be19 DH |
265 | if (ret < 0) |
266 | return ret; | |
267 | ||
f1dd2cd1 MH |
268 | /* |
269 | * Make all the pages reserved so that nobody will stumble over half | |
270 | * initialized state. | |
1b7176ae | 271 | * FIXME: We also have to associate it with a node because page_to_nid |
f1dd2cd1 MH |
272 | * relies on having page with the proper node. |
273 | */ | |
274 | for (i = 0; i < PAGES_PER_SECTION; i++) { | |
275 | unsigned long pfn = phys_start_pfn + i; | |
276 | struct page *page; | |
277 | if (!pfn_valid(pfn)) | |
278 | continue; | |
718127cc | 279 | |
f1dd2cd1 MH |
280 | page = pfn_to_page(pfn); |
281 | set_page_node(page, nid); | |
282 | SetPageReserved(page); | |
283 | } | |
718127cc | 284 | |
1b862aec MH |
285 | if (!want_memblock) |
286 | return 0; | |
287 | ||
c04fc586 | 288 | return register_new_memory(nid, __pfn_to_section(phys_start_pfn)); |
3947be19 DH |
289 | } |
290 | ||
4edd7cef DR |
291 | /* |
292 | * Reasonably generic function for adding memory. It is | |
293 | * expected that archs that support memory hotplug will | |
294 | * call this function after deciding the zone to which to | |
295 | * add the new pages. | |
296 | */ | |
f1dd2cd1 | 297 | int __ref __add_pages(int nid, unsigned long phys_start_pfn, |
1b862aec | 298 | unsigned long nr_pages, bool want_memblock) |
4edd7cef DR |
299 | { |
300 | unsigned long i; | |
301 | int err = 0; | |
302 | int start_sec, end_sec; | |
4b94ffdc DW |
303 | struct vmem_altmap *altmap; |
304 | ||
4edd7cef DR |
305 | /* during initialize mem_map, align hot-added range to section */ |
306 | start_sec = pfn_to_section_nr(phys_start_pfn); | |
307 | end_sec = pfn_to_section_nr(phys_start_pfn + nr_pages - 1); | |
308 | ||
4b94ffdc DW |
309 | altmap = to_vmem_altmap((unsigned long) pfn_to_page(phys_start_pfn)); |
310 | if (altmap) { | |
311 | /* | |
312 | * Validate altmap is within bounds of the total request | |
313 | */ | |
314 | if (altmap->base_pfn != phys_start_pfn | |
315 | || vmem_altmap_offset(altmap) > nr_pages) { | |
316 | pr_warn_once("memory add fail, invalid altmap\n"); | |
7cf91a98 JK |
317 | err = -EINVAL; |
318 | goto out; | |
4b94ffdc DW |
319 | } |
320 | altmap->alloc = 0; | |
321 | } | |
322 | ||
4edd7cef | 323 | for (i = start_sec; i <= end_sec; i++) { |
f1dd2cd1 | 324 | err = __add_section(nid, section_nr_to_pfn(i), want_memblock); |
4edd7cef DR |
325 | |
326 | /* | |
327 | * EEXIST is finally dealt with by ioresource collision | |
328 | * check. see add_memory() => register_memory_resource() | |
329 | * Warning will be printed if there is collision. | |
330 | */ | |
331 | if (err && (err != -EEXIST)) | |
332 | break; | |
333 | err = 0; | |
f64ac5e6 | 334 | cond_resched(); |
4edd7cef | 335 | } |
c435a390 | 336 | vmemmap_populate_print_last(); |
7cf91a98 | 337 | out: |
4edd7cef DR |
338 | return err; |
339 | } | |
340 | EXPORT_SYMBOL_GPL(__add_pages); | |
341 | ||
342 | #ifdef CONFIG_MEMORY_HOTREMOVE | |
815121d2 | 343 | /* find the smallest valid pfn in the range [start_pfn, end_pfn) */ |
d09b0137 | 344 | static unsigned long find_smallest_section_pfn(int nid, struct zone *zone, |
815121d2 YI |
345 | unsigned long start_pfn, |
346 | unsigned long end_pfn) | |
347 | { | |
815121d2 | 348 | for (; start_pfn < end_pfn; start_pfn += PAGES_PER_SECTION) { |
6c0bfcdc | 349 | if (unlikely(!pfn_to_online_page(start_pfn))) |
815121d2 YI |
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 | { | |
815121d2 YI |
369 | unsigned long pfn; |
370 | ||
371 | /* pfn is the end pfn of a memory section. */ | |
372 | pfn = end_pfn - 1; | |
373 | for (; pfn >= start_pfn; pfn -= PAGES_PER_SECTION) { | |
6c0bfcdc | 374 | if (unlikely(!pfn_to_online_page(pfn))) |
815121d2 YI |
375 | continue; |
376 | ||
377 | if (unlikely(pfn_to_nid(pfn) != nid)) | |
378 | continue; | |
379 | ||
380 | if (zone && zone != page_zone(pfn_to_page(pfn))) | |
381 | continue; | |
382 | ||
383 | return pfn; | |
384 | } | |
385 | ||
386 | return 0; | |
387 | } | |
388 | ||
389 | static void shrink_zone_span(struct zone *zone, unsigned long start_pfn, | |
390 | unsigned long end_pfn) | |
391 | { | |
c33bc315 XQ |
392 | unsigned long zone_start_pfn = zone->zone_start_pfn; |
393 | unsigned long z = zone_end_pfn(zone); /* zone_end_pfn namespace clash */ | |
394 | unsigned long zone_end_pfn = z; | |
815121d2 | 395 | unsigned long pfn; |
815121d2 YI |
396 | int nid = zone_to_nid(zone); |
397 | ||
398 | zone_span_writelock(zone); | |
399 | if (zone_start_pfn == start_pfn) { | |
400 | /* | |
401 | * If the section is smallest section in the zone, it need | |
402 | * shrink zone->zone_start_pfn and zone->zone_spanned_pages. | |
403 | * In this case, we find second smallest valid mem_section | |
404 | * for shrinking zone. | |
405 | */ | |
406 | pfn = find_smallest_section_pfn(nid, zone, end_pfn, | |
407 | zone_end_pfn); | |
408 | if (pfn) { | |
409 | zone->zone_start_pfn = pfn; | |
410 | zone->spanned_pages = zone_end_pfn - pfn; | |
411 | } | |
412 | } else if (zone_end_pfn == end_pfn) { | |
413 | /* | |
414 | * If the section is biggest section in the zone, it need | |
415 | * shrink zone->spanned_pages. | |
416 | * In this case, we find second biggest valid mem_section for | |
417 | * shrinking zone. | |
418 | */ | |
419 | pfn = find_biggest_section_pfn(nid, zone, zone_start_pfn, | |
420 | start_pfn); | |
421 | if (pfn) | |
422 | zone->spanned_pages = pfn - zone_start_pfn + 1; | |
423 | } | |
424 | ||
425 | /* | |
426 | * The section is not biggest or smallest mem_section in the zone, it | |
427 | * only creates a hole in the zone. So in this case, we need not | |
428 | * change the zone. But perhaps, the zone has only hole data. Thus | |
429 | * it check the zone has only hole or not. | |
430 | */ | |
431 | pfn = zone_start_pfn; | |
432 | for (; pfn < zone_end_pfn; pfn += PAGES_PER_SECTION) { | |
6c0bfcdc | 433 | if (unlikely(!pfn_to_online_page(pfn))) |
815121d2 YI |
434 | continue; |
435 | ||
436 | if (page_zone(pfn_to_page(pfn)) != zone) | |
437 | continue; | |
438 | ||
439 | /* If the section is current section, it continues the loop */ | |
440 | if (start_pfn == pfn) | |
441 | continue; | |
442 | ||
443 | /* If we find valid section, we have nothing to do */ | |
444 | zone_span_writeunlock(zone); | |
445 | return; | |
446 | } | |
447 | ||
448 | /* The zone has no valid section */ | |
449 | zone->zone_start_pfn = 0; | |
450 | zone->spanned_pages = 0; | |
451 | zone_span_writeunlock(zone); | |
452 | } | |
453 | ||
bada68c4 | 454 | static void update_pgdat_span(struct pglist_data *pgdat) |
815121d2 | 455 | { |
bada68c4 DH |
456 | unsigned long node_start_pfn = 0, node_end_pfn = 0; |
457 | struct zone *zone; | |
458 | ||
459 | for (zone = pgdat->node_zones; | |
460 | zone < pgdat->node_zones + MAX_NR_ZONES; zone++) { | |
461 | unsigned long zone_end_pfn = zone->zone_start_pfn + | |
462 | zone->spanned_pages; | |
463 | ||
464 | /* No need to lock the zones, they can't change. */ | |
530010f7 DH |
465 | if (!zone->spanned_pages) |
466 | continue; | |
467 | if (!node_end_pfn) { | |
468 | node_start_pfn = zone->zone_start_pfn; | |
469 | node_end_pfn = zone_end_pfn; | |
470 | continue; | |
471 | } | |
472 | ||
bada68c4 DH |
473 | if (zone_end_pfn > node_end_pfn) |
474 | node_end_pfn = zone_end_pfn; | |
475 | if (zone->zone_start_pfn < node_start_pfn) | |
476 | node_start_pfn = zone->zone_start_pfn; | |
815121d2 YI |
477 | } |
478 | ||
bada68c4 DH |
479 | pgdat->node_start_pfn = node_start_pfn; |
480 | pgdat->node_spanned_pages = node_end_pfn - node_start_pfn; | |
815121d2 YI |
481 | } |
482 | ||
483 | static void __remove_zone(struct zone *zone, unsigned long start_pfn) | |
484 | { | |
485 | struct pglist_data *pgdat = zone->zone_pgdat; | |
486 | int nr_pages = PAGES_PER_SECTION; | |
815121d2 YI |
487 | unsigned long flags; |
488 | ||
6c0bfcdc DH |
489 | #ifdef CONFIG_ZONE_DEVICE |
490 | /* | |
491 | * Zone shrinking code cannot properly deal with ZONE_DEVICE. So | |
492 | * we will not try to shrink the zones - which is okay as | |
493 | * set_zone_contiguous() cannot deal with ZONE_DEVICE either way. | |
494 | */ | |
495 | if (zone_idx(zone) == ZONE_DEVICE) | |
496 | return; | |
497 | #endif | |
498 | ||
815121d2 YI |
499 | pgdat_resize_lock(zone->zone_pgdat, &flags); |
500 | shrink_zone_span(zone, start_pfn, start_pfn + nr_pages); | |
bada68c4 | 501 | update_pgdat_span(pgdat); |
815121d2 YI |
502 | pgdat_resize_unlock(zone->zone_pgdat, &flags); |
503 | } | |
504 | ||
4b94ffdc DW |
505 | static int __remove_section(struct zone *zone, struct mem_section *ms, |
506 | unsigned long map_offset) | |
ea01ea93 | 507 | { |
815121d2 YI |
508 | unsigned long start_pfn; |
509 | int scn_nr; | |
ea01ea93 BP |
510 | int ret = -EINVAL; |
511 | ||
512 | if (!valid_section(ms)) | |
513 | return ret; | |
514 | ||
515 | ret = unregister_memory_section(ms); | |
516 | if (ret) | |
517 | return ret; | |
518 | ||
815121d2 | 519 | scn_nr = __section_nr(ms); |
1dd2bfc8 | 520 | start_pfn = section_nr_to_pfn((unsigned long)scn_nr); |
815121d2 YI |
521 | __remove_zone(zone, start_pfn); |
522 | ||
4b94ffdc | 523 | sparse_remove_one_section(zone, ms, map_offset); |
ea01ea93 BP |
524 | return 0; |
525 | } | |
526 | ||
ea01ea93 BP |
527 | /** |
528 | * __remove_pages() - remove sections of pages from a zone | |
529 | * @zone: zone from which pages need to be removed | |
530 | * @phys_start_pfn: starting pageframe (must be aligned to start of a section) | |
531 | * @nr_pages: number of pages to remove (must be multiple of section size) | |
532 | * | |
533 | * Generic helper function to remove section mappings and sysfs entries | |
534 | * for the section of the memory we are removing. Caller needs to make | |
535 | * sure that pages are marked reserved and zones are adjust properly by | |
536 | * calling offline_pages(). | |
537 | */ | |
538 | int __remove_pages(struct zone *zone, unsigned long phys_start_pfn, | |
539 | unsigned long nr_pages) | |
540 | { | |
fe74ebb1 | 541 | unsigned long i; |
4b94ffdc DW |
542 | unsigned long map_offset = 0; |
543 | int sections_to_remove, ret = 0; | |
544 | ||
545 | /* In the ZONE_DEVICE case device driver owns the memory region */ | |
546 | if (is_dev_zone(zone)) { | |
547 | struct page *page = pfn_to_page(phys_start_pfn); | |
548 | struct vmem_altmap *altmap; | |
549 | ||
550 | altmap = to_vmem_altmap((unsigned long) page); | |
551 | if (altmap) | |
552 | map_offset = vmem_altmap_offset(altmap); | |
553 | } else { | |
554 | resource_size_t start, size; | |
555 | ||
556 | start = phys_start_pfn << PAGE_SHIFT; | |
557 | size = nr_pages * PAGE_SIZE; | |
558 | ||
559 | ret = release_mem_region_adjustable(&iomem_resource, start, | |
560 | size); | |
561 | if (ret) { | |
562 | resource_size_t endres = start + size - 1; | |
563 | ||
564 | pr_warn("Unable to release resource <%pa-%pa> (%d)\n", | |
565 | &start, &endres, ret); | |
566 | } | |
567 | } | |
ea01ea93 | 568 | |
7cf91a98 JK |
569 | clear_zone_contiguous(zone); |
570 | ||
ea01ea93 BP |
571 | /* |
572 | * We can only remove entire sections | |
573 | */ | |
574 | BUG_ON(phys_start_pfn & ~PAGE_SECTION_MASK); | |
575 | BUG_ON(nr_pages % PAGES_PER_SECTION); | |
576 | ||
ea01ea93 BP |
577 | sections_to_remove = nr_pages / PAGES_PER_SECTION; |
578 | for (i = 0; i < sections_to_remove; i++) { | |
579 | unsigned long pfn = phys_start_pfn + i*PAGES_PER_SECTION; | |
4b94ffdc DW |
580 | |
581 | ret = __remove_section(zone, __pfn_to_section(pfn), map_offset); | |
582 | map_offset = 0; | |
ea01ea93 BP |
583 | if (ret) |
584 | break; | |
585 | } | |
7cf91a98 JK |
586 | |
587 | set_zone_contiguous(zone); | |
588 | ||
ea01ea93 BP |
589 | return ret; |
590 | } | |
4edd7cef | 591 | #endif /* CONFIG_MEMORY_HOTREMOVE */ |
ea01ea93 | 592 | |
9d0ad8ca DK |
593 | int set_online_page_callback(online_page_callback_t callback) |
594 | { | |
595 | int rc = -EINVAL; | |
596 | ||
bfc8c901 VD |
597 | get_online_mems(); |
598 | mutex_lock(&online_page_callback_lock); | |
9d0ad8ca DK |
599 | |
600 | if (online_page_callback == generic_online_page) { | |
601 | online_page_callback = callback; | |
602 | rc = 0; | |
603 | } | |
604 | ||
bfc8c901 VD |
605 | mutex_unlock(&online_page_callback_lock); |
606 | put_online_mems(); | |
9d0ad8ca DK |
607 | |
608 | return rc; | |
609 | } | |
610 | EXPORT_SYMBOL_GPL(set_online_page_callback); | |
611 | ||
612 | int restore_online_page_callback(online_page_callback_t callback) | |
613 | { | |
614 | int rc = -EINVAL; | |
615 | ||
bfc8c901 VD |
616 | get_online_mems(); |
617 | mutex_lock(&online_page_callback_lock); | |
9d0ad8ca DK |
618 | |
619 | if (online_page_callback == callback) { | |
620 | online_page_callback = generic_online_page; | |
621 | rc = 0; | |
622 | } | |
623 | ||
bfc8c901 VD |
624 | mutex_unlock(&online_page_callback_lock); |
625 | put_online_mems(); | |
9d0ad8ca DK |
626 | |
627 | return rc; | |
628 | } | |
629 | EXPORT_SYMBOL_GPL(restore_online_page_callback); | |
630 | ||
631 | void __online_page_set_limits(struct page *page) | |
180c06ef | 632 | { |
9d0ad8ca DK |
633 | } |
634 | EXPORT_SYMBOL_GPL(__online_page_set_limits); | |
635 | ||
636 | void __online_page_increment_counters(struct page *page) | |
637 | { | |
3dcc0571 | 638 | adjust_managed_page_count(page, 1); |
9d0ad8ca DK |
639 | } |
640 | EXPORT_SYMBOL_GPL(__online_page_increment_counters); | |
180c06ef | 641 | |
9d0ad8ca DK |
642 | void __online_page_free(struct page *page) |
643 | { | |
3dcc0571 | 644 | __free_reserved_page(page); |
180c06ef | 645 | } |
9d0ad8ca DK |
646 | EXPORT_SYMBOL_GPL(__online_page_free); |
647 | ||
648 | static void generic_online_page(struct page *page) | |
649 | { | |
650 | __online_page_set_limits(page); | |
651 | __online_page_increment_counters(page); | |
652 | __online_page_free(page); | |
653 | } | |
180c06ef | 654 | |
75884fb1 KH |
655 | static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages, |
656 | void *arg) | |
3947be19 DH |
657 | { |
658 | unsigned long i; | |
75884fb1 KH |
659 | unsigned long onlined_pages = *(unsigned long *)arg; |
660 | struct page *page; | |
2d070eab | 661 | |
75884fb1 KH |
662 | if (PageReserved(pfn_to_page(start_pfn))) |
663 | for (i = 0; i < nr_pages; i++) { | |
664 | page = pfn_to_page(start_pfn + i); | |
9d0ad8ca | 665 | (*online_page_callback)(page); |
75884fb1 KH |
666 | onlined_pages++; |
667 | } | |
2d070eab MH |
668 | |
669 | online_mem_sections(start_pfn, start_pfn + nr_pages); | |
670 | ||
75884fb1 KH |
671 | *(unsigned long *)arg = onlined_pages; |
672 | return 0; | |
673 | } | |
674 | ||
d9713679 LJ |
675 | /* check which state of node_states will be changed when online memory */ |
676 | static void node_states_check_changes_online(unsigned long nr_pages, | |
677 | struct zone *zone, struct memory_notify *arg) | |
678 | { | |
679 | int nid = zone_to_nid(zone); | |
680 | enum zone_type zone_last = ZONE_NORMAL; | |
681 | ||
682 | /* | |
6715ddf9 LJ |
683 | * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY] |
684 | * contains nodes which have zones of 0...ZONE_NORMAL, | |
685 | * set zone_last to ZONE_NORMAL. | |
d9713679 | 686 | * |
6715ddf9 LJ |
687 | * If we don't have HIGHMEM nor movable node, |
688 | * node_states[N_NORMAL_MEMORY] contains nodes which have zones of | |
689 | * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE. | |
d9713679 | 690 | */ |
6715ddf9 | 691 | if (N_MEMORY == N_NORMAL_MEMORY) |
d9713679 LJ |
692 | zone_last = ZONE_MOVABLE; |
693 | ||
694 | /* | |
695 | * if the memory to be online is in a zone of 0...zone_last, and | |
696 | * the zones of 0...zone_last don't have memory before online, we will | |
697 | * need to set the node to node_states[N_NORMAL_MEMORY] after | |
698 | * the memory is online. | |
699 | */ | |
700 | if (zone_idx(zone) <= zone_last && !node_state(nid, N_NORMAL_MEMORY)) | |
701 | arg->status_change_nid_normal = nid; | |
702 | else | |
703 | arg->status_change_nid_normal = -1; | |
704 | ||
6715ddf9 LJ |
705 | #ifdef CONFIG_HIGHMEM |
706 | /* | |
707 | * If we have movable node, node_states[N_HIGH_MEMORY] | |
708 | * contains nodes which have zones of 0...ZONE_HIGHMEM, | |
709 | * set zone_last to ZONE_HIGHMEM. | |
710 | * | |
711 | * If we don't have movable node, node_states[N_NORMAL_MEMORY] | |
712 | * contains nodes which have zones of 0...ZONE_MOVABLE, | |
713 | * set zone_last to ZONE_MOVABLE. | |
714 | */ | |
715 | zone_last = ZONE_HIGHMEM; | |
716 | if (N_MEMORY == N_HIGH_MEMORY) | |
717 | zone_last = ZONE_MOVABLE; | |
718 | ||
719 | if (zone_idx(zone) <= zone_last && !node_state(nid, N_HIGH_MEMORY)) | |
720 | arg->status_change_nid_high = nid; | |
721 | else | |
722 | arg->status_change_nid_high = -1; | |
723 | #else | |
724 | arg->status_change_nid_high = arg->status_change_nid_normal; | |
725 | #endif | |
726 | ||
d9713679 LJ |
727 | /* |
728 | * if the node don't have memory befor online, we will need to | |
6715ddf9 | 729 | * set the node to node_states[N_MEMORY] after the memory |
d9713679 LJ |
730 | * is online. |
731 | */ | |
6715ddf9 | 732 | if (!node_state(nid, N_MEMORY)) |
d9713679 LJ |
733 | arg->status_change_nid = nid; |
734 | else | |
735 | arg->status_change_nid = -1; | |
736 | } | |
737 | ||
738 | static void node_states_set_node(int node, struct memory_notify *arg) | |
739 | { | |
740 | if (arg->status_change_nid_normal >= 0) | |
741 | node_set_state(node, N_NORMAL_MEMORY); | |
742 | ||
6715ddf9 LJ |
743 | if (arg->status_change_nid_high >= 0) |
744 | node_set_state(node, N_HIGH_MEMORY); | |
745 | ||
746 | node_set_state(node, N_MEMORY); | |
d9713679 LJ |
747 | } |
748 | ||
f1dd2cd1 MH |
749 | static void __meminit resize_zone_range(struct zone *zone, unsigned long start_pfn, |
750 | unsigned long nr_pages) | |
751 | { | |
752 | unsigned long old_end_pfn = zone_end_pfn(zone); | |
753 | ||
754 | if (zone_is_empty(zone) || start_pfn < zone->zone_start_pfn) | |
755 | zone->zone_start_pfn = start_pfn; | |
756 | ||
757 | zone->spanned_pages = max(start_pfn + nr_pages, old_end_pfn) - zone->zone_start_pfn; | |
758 | } | |
759 | ||
760 | static void __meminit resize_pgdat_range(struct pglist_data *pgdat, unsigned long start_pfn, | |
761 | unsigned long nr_pages) | |
762 | { | |
763 | unsigned long old_end_pfn = pgdat_end_pfn(pgdat); | |
764 | ||
765 | if (!pgdat->node_spanned_pages || start_pfn < pgdat->node_start_pfn) | |
766 | pgdat->node_start_pfn = start_pfn; | |
767 | ||
768 | pgdat->node_spanned_pages = max(start_pfn + nr_pages, old_end_pfn) - pgdat->node_start_pfn; | |
769 | } | |
770 | ||
cdf72f25 | 771 | void __ref move_pfn_range_to_zone(struct zone *zone, |
f1dd2cd1 MH |
772 | unsigned long start_pfn, unsigned long nr_pages) |
773 | { | |
774 | struct pglist_data *pgdat = zone->zone_pgdat; | |
775 | int nid = pgdat->node_id; | |
776 | unsigned long flags; | |
df429ac0 | 777 | |
f1dd2cd1 MH |
778 | if (zone_is_empty(zone)) |
779 | init_currently_empty_zone(zone, start_pfn, nr_pages); | |
df429ac0 | 780 | |
f1dd2cd1 MH |
781 | clear_zone_contiguous(zone); |
782 | ||
783 | /* TODO Huh pgdat is irqsave while zone is not. It used to be like that before */ | |
784 | pgdat_resize_lock(pgdat, &flags); | |
785 | zone_span_writelock(zone); | |
786 | resize_zone_range(zone, start_pfn, nr_pages); | |
787 | zone_span_writeunlock(zone); | |
788 | resize_pgdat_range(pgdat, start_pfn, nr_pages); | |
789 | pgdat_resize_unlock(pgdat, &flags); | |
790 | ||
791 | /* | |
792 | * TODO now we have a visible range of pages which are not associated | |
793 | * with their zone properly. Not nice but set_pfnblock_flags_mask | |
794 | * expects the zone spans the pfn range. All the pages in the range | |
795 | * are reserved so nobody should be touching them so we should be safe | |
796 | */ | |
797 | memmap_init_zone(nr_pages, nid, zone_idx(zone), start_pfn, MEMMAP_HOTPLUG); | |
798 | ||
799 | set_zone_contiguous(zone); | |
800 | } | |
801 | ||
46e81a18 SF |
802 | void set_default_mem_hotplug_zone(enum zone_type zone) |
803 | { | |
804 | default_kernel_zone = zone; | |
805 | } | |
806 | ||
807 | #ifdef CONFIG_HIGHMEM | |
808 | #define MAX_KERNEL_ZONE ZONE_HIGHMEM | |
809 | #else | |
810 | #define MAX_KERNEL_ZONE ZONE_NORMAL | |
811 | #endif | |
812 | ||
c246a213 MH |
813 | /* |
814 | * Returns a default kernel memory zone for the given pfn range. | |
815 | * If no kernel zone covers this pfn range it will automatically go | |
46e81a18 | 816 | * to the MAX_KERNEL_ZONE. |
c246a213 | 817 | */ |
c6f03e29 | 818 | static struct zone *default_kernel_zone_for_pfn(int nid, unsigned long start_pfn, |
c246a213 MH |
819 | unsigned long nr_pages) |
820 | { | |
821 | struct pglist_data *pgdat = NODE_DATA(nid); | |
822 | int zid; | |
823 | ||
46e81a18 | 824 | for (zid = 0; zid <= MAX_KERNEL_ZONE; zid++) { |
c246a213 MH |
825 | struct zone *zone = &pgdat->node_zones[zid]; |
826 | ||
827 | if (zone_intersects(zone, start_pfn, nr_pages)) | |
828 | return zone; | |
829 | } | |
830 | ||
46e81a18 | 831 | return &pgdat->node_zones[default_kernel_zone]; |
c246a213 MH |
832 | } |
833 | ||
c6f03e29 MH |
834 | static inline struct zone *default_zone_for_pfn(int nid, unsigned long start_pfn, |
835 | unsigned long nr_pages) | |
e5e68930 | 836 | { |
c6f03e29 MH |
837 | struct zone *kernel_zone = default_kernel_zone_for_pfn(nid, start_pfn, |
838 | nr_pages); | |
839 | struct zone *movable_zone = &NODE_DATA(nid)->node_zones[ZONE_MOVABLE]; | |
840 | bool in_kernel = zone_intersects(kernel_zone, start_pfn, nr_pages); | |
841 | bool in_movable = zone_intersects(movable_zone, start_pfn, nr_pages); | |
e5e68930 MH |
842 | |
843 | /* | |
c6f03e29 MH |
844 | * We inherit the existing zone in a simple case where zones do not |
845 | * overlap in the given range | |
e5e68930 | 846 | */ |
c6f03e29 MH |
847 | if (in_kernel ^ in_movable) |
848 | return (in_kernel) ? kernel_zone : movable_zone; | |
9f123ab5 | 849 | |
c6f03e29 MH |
850 | /* |
851 | * If the range doesn't belong to any zone or two zones overlap in the | |
852 | * given range then we use movable zone only if movable_node is | |
853 | * enabled because we always online to a kernel zone by default. | |
854 | */ | |
855 | return movable_node_enabled ? movable_zone : kernel_zone; | |
9f123ab5 MH |
856 | } |
857 | ||
e5e68930 MH |
858 | struct zone * zone_for_pfn_range(int online_type, int nid, unsigned start_pfn, |
859 | unsigned long nr_pages) | |
f1dd2cd1 | 860 | { |
c6f03e29 MH |
861 | if (online_type == MMOP_ONLINE_KERNEL) |
862 | return default_kernel_zone_for_pfn(nid, start_pfn, nr_pages); | |
f1dd2cd1 | 863 | |
c6f03e29 MH |
864 | if (online_type == MMOP_ONLINE_MOVABLE) |
865 | return &NODE_DATA(nid)->node_zones[ZONE_MOVABLE]; | |
df429ac0 | 866 | |
c6f03e29 | 867 | return default_zone_for_pfn(nid, start_pfn, nr_pages); |
e5e68930 MH |
868 | } |
869 | ||
870 | /* | |
871 | * Associates the given pfn range with the given node and the zone appropriate | |
872 | * for the given online type. | |
873 | */ | |
874 | static struct zone * __meminit move_pfn_range(int online_type, int nid, | |
875 | unsigned long start_pfn, unsigned long nr_pages) | |
876 | { | |
877 | struct zone *zone; | |
878 | ||
879 | zone = zone_for_pfn_range(online_type, nid, start_pfn, nr_pages); | |
f1dd2cd1 MH |
880 | move_pfn_range_to_zone(zone, start_pfn, nr_pages); |
881 | return zone; | |
df429ac0 | 882 | } |
75884fb1 | 883 | |
511c2aba | 884 | int __ref online_pages(unsigned long pfn, unsigned long nr_pages, int online_type) |
75884fb1 | 885 | { |
aa47228a | 886 | unsigned long flags; |
3947be19 DH |
887 | unsigned long onlined_pages = 0; |
888 | struct zone *zone; | |
6811378e | 889 | int need_zonelists_rebuild = 0; |
7b78d335 YG |
890 | int nid; |
891 | int ret; | |
892 | struct memory_notify arg; | |
893 | ||
1679e5d6 DH |
894 | mem_hotplug_begin(); |
895 | ||
f1dd2cd1 | 896 | nid = pfn_to_nid(pfn); |
f1dd2cd1 MH |
897 | /* associate pfn range with the zone */ |
898 | zone = move_pfn_range(online_type, nid, pfn, nr_pages); | |
899 | ||
7b78d335 YG |
900 | arg.start_pfn = pfn; |
901 | arg.nr_pages = nr_pages; | |
d9713679 | 902 | node_states_check_changes_online(nr_pages, zone, &arg); |
7b78d335 | 903 | |
7b78d335 YG |
904 | ret = memory_notify(MEM_GOING_ONLINE, &arg); |
905 | ret = notifier_to_errno(ret); | |
e33e33b4 CY |
906 | if (ret) |
907 | goto failed_addition; | |
908 | ||
6811378e YG |
909 | /* |
910 | * If this zone is not populated, then it is not in zonelist. | |
911 | * This means the page allocator ignores this zone. | |
912 | * So, zonelist must be updated after online. | |
913 | */ | |
6dcd73d7 | 914 | if (!populated_zone(zone)) { |
6811378e | 915 | need_zonelists_rebuild = 1; |
72675e13 | 916 | setup_zone_pageset(zone); |
6dcd73d7 | 917 | } |
6811378e | 918 | |
908eedc6 | 919 | ret = walk_system_ram_range(pfn, nr_pages, &onlined_pages, |
75884fb1 | 920 | online_pages_range); |
fd8a4221 | 921 | if (ret) { |
6dcd73d7 WC |
922 | if (need_zonelists_rebuild) |
923 | zone_pcp_reset(zone); | |
e33e33b4 | 924 | goto failed_addition; |
fd8a4221 GL |
925 | } |
926 | ||
3947be19 | 927 | zone->present_pages += onlined_pages; |
aa47228a CS |
928 | |
929 | pgdat_resize_lock(zone->zone_pgdat, &flags); | |
f2937be5 | 930 | zone->zone_pgdat->node_present_pages += onlined_pages; |
aa47228a CS |
931 | pgdat_resize_unlock(zone->zone_pgdat, &flags); |
932 | ||
08dff7b7 | 933 | if (onlined_pages) { |
e888ca35 | 934 | node_states_set_node(nid, &arg); |
08dff7b7 | 935 | if (need_zonelists_rebuild) |
72675e13 | 936 | build_all_zonelists(NULL); |
08dff7b7 JL |
937 | else |
938 | zone_pcp_update(zone); | |
939 | } | |
3947be19 | 940 | |
1b79acc9 KM |
941 | init_per_zone_wmark_min(); |
942 | ||
698b1b30 | 943 | if (onlined_pages) { |
e888ca35 | 944 | kswapd_run(nid); |
698b1b30 VB |
945 | kcompactd_run(nid); |
946 | } | |
61b13993 | 947 | |
1f522509 | 948 | vm_total_pages = nr_free_pagecache_pages(); |
2f7f24ec | 949 | |
2d1d43f6 | 950 | writeback_set_ratelimit(); |
7b78d335 YG |
951 | |
952 | if (onlined_pages) | |
953 | memory_notify(MEM_ONLINE, &arg); | |
1679e5d6 | 954 | mem_hotplug_done(); |
30467e0b | 955 | return 0; |
e33e33b4 CY |
956 | |
957 | failed_addition: | |
958 | pr_debug("online_pages [mem %#010llx-%#010llx] failed\n", | |
959 | (unsigned long long) pfn << PAGE_SHIFT, | |
960 | (((unsigned long long) pfn + nr_pages) << PAGE_SHIFT) - 1); | |
961 | memory_notify(MEM_CANCEL_ONLINE, &arg); | |
1679e5d6 | 962 | mem_hotplug_done(); |
e33e33b4 | 963 | return ret; |
3947be19 | 964 | } |
53947027 | 965 | #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */ |
bc02af93 | 966 | |
0bd85420 TC |
967 | static void reset_node_present_pages(pg_data_t *pgdat) |
968 | { | |
969 | struct zone *z; | |
970 | ||
971 | for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++) | |
972 | z->present_pages = 0; | |
973 | ||
974 | pgdat->node_present_pages = 0; | |
975 | } | |
976 | ||
e1319331 HS |
977 | /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */ |
978 | static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start) | |
9af3c2de YG |
979 | { |
980 | struct pglist_data *pgdat; | |
981 | unsigned long zones_size[MAX_NR_ZONES] = {0}; | |
982 | unsigned long zholes_size[MAX_NR_ZONES] = {0}; | |
c8e861a5 | 983 | unsigned long start_pfn = PFN_DOWN(start); |
9af3c2de | 984 | |
a1e565aa TC |
985 | pgdat = NODE_DATA(nid); |
986 | if (!pgdat) { | |
987 | pgdat = arch_alloc_nodedata(nid); | |
988 | if (!pgdat) | |
989 | return NULL; | |
9af3c2de | 990 | |
a1e565aa | 991 | arch_refresh_nodedata(nid, pgdat); |
b0dc3a34 | 992 | } else { |
e716f2eb MG |
993 | /* |
994 | * Reset the nr_zones, order and classzone_idx before reuse. | |
995 | * Note that kswapd will init kswapd_classzone_idx properly | |
996 | * when it starts in the near future. | |
997 | */ | |
b0dc3a34 | 998 | pgdat->nr_zones = 0; |
38087d9b MG |
999 | pgdat->kswapd_order = 0; |
1000 | pgdat->kswapd_classzone_idx = 0; | |
a1e565aa | 1001 | } |
9af3c2de YG |
1002 | |
1003 | /* we can use NODE_DATA(nid) from here */ | |
1004 | ||
1005 | /* init node's zones as empty zones, we don't have any present pages.*/ | |
9109fb7b | 1006 | free_area_init_node(nid, zones_size, start_pfn, zholes_size); |
5830169f | 1007 | pgdat->per_cpu_nodestats = alloc_percpu(struct per_cpu_nodestat); |
9af3c2de | 1008 | |
959ecc48 KH |
1009 | /* |
1010 | * The node we allocated has no zone fallback lists. For avoiding | |
1011 | * to access not-initialized zonelist, build here. | |
1012 | */ | |
72675e13 | 1013 | build_all_zonelists(pgdat); |
959ecc48 | 1014 | |
f784a3f1 TC |
1015 | /* |
1016 | * zone->managed_pages is set to an approximate value in | |
1017 | * free_area_init_core(), which will cause | |
1018 | * /sys/device/system/node/nodeX/meminfo has wrong data. | |
1019 | * So reset it to 0 before any memory is onlined. | |
1020 | */ | |
1021 | reset_node_managed_pages(pgdat); | |
1022 | ||
0bd85420 TC |
1023 | /* |
1024 | * When memory is hot-added, all the memory is in offline state. So | |
1025 | * clear all zones' present_pages because they will be updated in | |
1026 | * online_pages() and offline_pages(). | |
1027 | */ | |
1028 | reset_node_present_pages(pgdat); | |
1029 | ||
9af3c2de YG |
1030 | return pgdat; |
1031 | } | |
1032 | ||
1033 | static void rollback_node_hotadd(int nid, pg_data_t *pgdat) | |
1034 | { | |
1035 | arch_refresh_nodedata(nid, NULL); | |
5830169f | 1036 | free_percpu(pgdat->per_cpu_nodestats); |
9af3c2de YG |
1037 | arch_free_nodedata(pgdat); |
1038 | return; | |
1039 | } | |
1040 | ||
0a547039 | 1041 | |
01b0f197 TK |
1042 | /** |
1043 | * try_online_node - online a node if offlined | |
1044 | * | |
cf23422b | 1045 | * called by cpu_up() to online a node without onlined memory. |
1046 | */ | |
01b0f197 | 1047 | int try_online_node(int nid) |
cf23422b | 1048 | { |
1049 | pg_data_t *pgdat; | |
1050 | int ret; | |
1051 | ||
01b0f197 TK |
1052 | if (node_online(nid)) |
1053 | return 0; | |
1054 | ||
bfc8c901 | 1055 | mem_hotplug_begin(); |
cf23422b | 1056 | pgdat = hotadd_new_pgdat(nid, 0); |
7553e8f2 | 1057 | if (!pgdat) { |
01b0f197 | 1058 | pr_err("Cannot online node %d due to NULL pgdat\n", nid); |
cf23422b | 1059 | ret = -ENOMEM; |
1060 | goto out; | |
1061 | } | |
1062 | node_set_online(nid); | |
1063 | ret = register_one_node(nid); | |
1064 | BUG_ON(ret); | |
cf23422b | 1065 | out: |
bfc8c901 | 1066 | mem_hotplug_done(); |
cf23422b | 1067 | return ret; |
1068 | } | |
1069 | ||
27356f54 TK |
1070 | static int check_hotplug_memory_range(u64 start, u64 size) |
1071 | { | |
c8e861a5 | 1072 | u64 start_pfn = PFN_DOWN(start); |
27356f54 TK |
1073 | u64 nr_pages = size >> PAGE_SHIFT; |
1074 | ||
1075 | /* Memory range must be aligned with section */ | |
1076 | if ((start_pfn & ~PAGE_SECTION_MASK) || | |
1077 | (nr_pages % PAGES_PER_SECTION) || (!nr_pages)) { | |
1078 | pr_err("Section-unaligned hotplug range: start 0x%llx, size 0x%llx\n", | |
1079 | (unsigned long long)start, | |
1080 | (unsigned long long)size); | |
1081 | return -EINVAL; | |
1082 | } | |
1083 | ||
1084 | return 0; | |
1085 | } | |
1086 | ||
31bc3858 VK |
1087 | static int online_memory_block(struct memory_block *mem, void *arg) |
1088 | { | |
dc18d706 | 1089 | return device_online(&mem->dev); |
31bc3858 VK |
1090 | } |
1091 | ||
de29654b DH |
1092 | /* |
1093 | * NOTE: The caller must call lock_device_hotplug() to serialize hotplug | |
1094 | * and online/offline operations (triggered e.g. by sysfs). | |
1095 | * | |
1096 | * we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG | |
1097 | */ | |
31bc3858 | 1098 | int __ref add_memory_resource(int nid, struct resource *res, bool online) |
bc02af93 | 1099 | { |
62cedb9f | 1100 | u64 start, size; |
9af3c2de | 1101 | pg_data_t *pgdat = NULL; |
a1e565aa TC |
1102 | bool new_pgdat; |
1103 | bool new_node; | |
bc02af93 YG |
1104 | int ret; |
1105 | ||
62cedb9f DV |
1106 | start = res->start; |
1107 | size = resource_size(res); | |
1108 | ||
27356f54 TK |
1109 | ret = check_hotplug_memory_range(start, size); |
1110 | if (ret) | |
1111 | return ret; | |
1112 | ||
a1e565aa TC |
1113 | { /* Stupid hack to suppress address-never-null warning */ |
1114 | void *p = NODE_DATA(nid); | |
1115 | new_pgdat = !p; | |
1116 | } | |
ac13c462 | 1117 | |
bfc8c901 | 1118 | mem_hotplug_begin(); |
ac13c462 | 1119 | |
7f36e3e5 TC |
1120 | /* |
1121 | * Add new range to memblock so that when hotadd_new_pgdat() is called | |
1122 | * to allocate new pgdat, get_pfn_range_for_nid() will be able to find | |
1123 | * this new range and calculate total pages correctly. The range will | |
1124 | * be removed at hot-remove time. | |
1125 | */ | |
1126 | memblock_add_node(start, size, nid); | |
1127 | ||
a1e565aa TC |
1128 | new_node = !node_online(nid); |
1129 | if (new_node) { | |
9af3c2de | 1130 | pgdat = hotadd_new_pgdat(nid, start); |
6ad696d2 | 1131 | ret = -ENOMEM; |
9af3c2de | 1132 | if (!pgdat) |
41b9e2d7 | 1133 | goto error; |
9af3c2de YG |
1134 | } |
1135 | ||
bc02af93 | 1136 | /* call arch's memory hotadd */ |
3d79a728 | 1137 | ret = arch_add_memory(nid, start, size, true); |
bc02af93 | 1138 | |
9af3c2de YG |
1139 | if (ret < 0) |
1140 | goto error; | |
1141 | ||
0fc44159 | 1142 | /* we online node here. we can't roll back from here. */ |
9af3c2de YG |
1143 | node_set_online(nid); |
1144 | ||
a1e565aa | 1145 | if (new_node) { |
9037a993 MH |
1146 | unsigned long start_pfn = start >> PAGE_SHIFT; |
1147 | unsigned long nr_pages = size >> PAGE_SHIFT; | |
1148 | ||
1149 | ret = __register_one_node(nid); | |
1150 | if (ret) | |
1151 | goto register_fail; | |
1152 | ||
1153 | /* | |
1154 | * link memory sections under this node. This is already | |
1155 | * done when creatig memory section in register_new_memory | |
1156 | * but that depends to have the node registered so offline | |
1157 | * nodes have to go through register_node. | |
1158 | * TODO clean up this mess. | |
1159 | */ | |
1160 | ret = link_mem_sections(nid, start_pfn, nr_pages); | |
1161 | register_fail: | |
0fc44159 YG |
1162 | /* |
1163 | * If sysfs file of new node can't create, cpu on the node | |
1164 | * can't be hot-added. There is no rollback way now. | |
1165 | * So, check by BUG_ON() to catch it reluctantly.. | |
1166 | */ | |
1167 | BUG_ON(ret); | |
1168 | } | |
1169 | ||
d96ae530 AM |
1170 | /* create new memmap entry */ |
1171 | firmware_map_add_hotplug(start, start + size, "System RAM"); | |
1172 | ||
1679e5d6 DH |
1173 | /* device_online() will take the lock when calling online_pages() */ |
1174 | mem_hotplug_done(); | |
1175 | ||
31bc3858 VK |
1176 | /* online pages if requested */ |
1177 | if (online) | |
1178 | walk_memory_range(PFN_DOWN(start), PFN_UP(start + size - 1), | |
1179 | NULL, online_memory_block); | |
1180 | ||
1679e5d6 | 1181 | return ret; |
9af3c2de YG |
1182 | error: |
1183 | /* rollback pgdat allocation and others */ | |
dbac61a3 | 1184 | if (new_pgdat && pgdat) |
9af3c2de | 1185 | rollback_node_hotadd(nid, pgdat); |
7f36e3e5 | 1186 | memblock_remove(start, size); |
bfc8c901 | 1187 | mem_hotplug_done(); |
bc02af93 YG |
1188 | return ret; |
1189 | } | |
62cedb9f | 1190 | |
de29654b DH |
1191 | /* requires device_hotplug_lock, see add_memory_resource() */ |
1192 | int __ref __add_memory(int nid, u64 start, u64 size) | |
62cedb9f DV |
1193 | { |
1194 | struct resource *res; | |
1195 | int ret; | |
1196 | ||
1197 | res = register_memory_resource(start, size); | |
6f754ba4 VK |
1198 | if (IS_ERR(res)) |
1199 | return PTR_ERR(res); | |
62cedb9f | 1200 | |
31bc3858 | 1201 | ret = add_memory_resource(nid, res, memhp_auto_online); |
62cedb9f DV |
1202 | if (ret < 0) |
1203 | release_memory_resource(res); | |
1204 | return ret; | |
1205 | } | |
de29654b DH |
1206 | |
1207 | int add_memory(int nid, u64 start, u64 size) | |
1208 | { | |
1209 | int rc; | |
1210 | ||
1211 | lock_device_hotplug(); | |
1212 | rc = __add_memory(nid, start, size); | |
1213 | unlock_device_hotplug(); | |
1214 | ||
1215 | return rc; | |
1216 | } | |
bc02af93 | 1217 | EXPORT_SYMBOL_GPL(add_memory); |
0c0e6195 KH |
1218 | |
1219 | #ifdef CONFIG_MEMORY_HOTREMOVE | |
5c755e9f BP |
1220 | /* |
1221 | * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy | |
1222 | * set and the size of the free page is given by page_order(). Using this, | |
1223 | * the function determines if the pageblock contains only free pages. | |
1224 | * Due to buddy contraints, a free page at least the size of a pageblock will | |
1225 | * be located at the start of the pageblock | |
1226 | */ | |
1227 | static inline int pageblock_free(struct page *page) | |
1228 | { | |
1229 | return PageBuddy(page) && page_order(page) >= pageblock_order; | |
1230 | } | |
1231 | ||
1232 | /* Return the start of the next active pageblock after a given page */ | |
1233 | static struct page *next_active_pageblock(struct page *page) | |
1234 | { | |
5c755e9f BP |
1235 | /* Ensure the starting page is pageblock-aligned */ |
1236 | BUG_ON(page_to_pfn(page) & (pageblock_nr_pages - 1)); | |
1237 | ||
5c755e9f | 1238 | /* If the entire pageblock is free, move to the end of free page */ |
0dcc48c1 KH |
1239 | if (pageblock_free(page)) { |
1240 | int order; | |
1241 | /* be careful. we don't have locks, page_order can be changed.*/ | |
1242 | order = page_order(page); | |
1243 | if ((order < MAX_ORDER) && (order >= pageblock_order)) | |
1244 | return page + (1 << order); | |
1245 | } | |
5c755e9f | 1246 | |
0dcc48c1 | 1247 | return page + pageblock_nr_pages; |
5c755e9f BP |
1248 | } |
1249 | ||
1250 | /* Checks if this range of memory is likely to be hot-removable. */ | |
c98940f6 | 1251 | bool is_mem_section_removable(unsigned long start_pfn, unsigned long nr_pages) |
5c755e9f | 1252 | { |
5c755e9f | 1253 | struct page *page = pfn_to_page(start_pfn); |
b4351d28 MH |
1254 | unsigned long end_pfn = min(start_pfn + nr_pages, zone_end_pfn(page_zone(page))); |
1255 | struct page *end_page = pfn_to_page(end_pfn); | |
5c755e9f BP |
1256 | |
1257 | /* Check the starting page of each pageblock within the range */ | |
1258 | for (; page < end_page; page = next_active_pageblock(page)) { | |
49ac8255 | 1259 | if (!is_pageblock_removable_nolock(page)) |
c98940f6 | 1260 | return false; |
49ac8255 | 1261 | cond_resched(); |
5c755e9f BP |
1262 | } |
1263 | ||
1264 | /* All pageblocks in the memory block are likely to be hot-removable */ | |
c98940f6 | 1265 | return true; |
5c755e9f BP |
1266 | } |
1267 | ||
0c0e6195 | 1268 | /* |
deb88a2a | 1269 | * Confirm all pages in a range [start, end) belong to the same zone. |
a96dfddb | 1270 | * When true, return its valid [start, end). |
0c0e6195 | 1271 | */ |
a96dfddb TK |
1272 | int test_pages_in_a_zone(unsigned long start_pfn, unsigned long end_pfn, |
1273 | unsigned long *valid_start, unsigned long *valid_end) | |
0c0e6195 | 1274 | { |
5f0f2887 | 1275 | unsigned long pfn, sec_end_pfn; |
a96dfddb | 1276 | unsigned long start, end; |
0c0e6195 KH |
1277 | struct zone *zone = NULL; |
1278 | struct page *page; | |
1279 | int i; | |
deb88a2a | 1280 | for (pfn = start_pfn, sec_end_pfn = SECTION_ALIGN_UP(start_pfn + 1); |
0c0e6195 | 1281 | pfn < end_pfn; |
deb88a2a | 1282 | pfn = sec_end_pfn, sec_end_pfn += PAGES_PER_SECTION) { |
5f0f2887 AB |
1283 | /* Make sure the memory section is present first */ |
1284 | if (!present_section_nr(pfn_to_section_nr(pfn))) | |
0c0e6195 | 1285 | continue; |
5f0f2887 AB |
1286 | for (; pfn < sec_end_pfn && pfn < end_pfn; |
1287 | pfn += MAX_ORDER_NR_PAGES) { | |
1288 | i = 0; | |
1289 | /* This is just a CONFIG_HOLES_IN_ZONE check.*/ | |
1290 | while ((i < MAX_ORDER_NR_PAGES) && | |
1291 | !pfn_valid_within(pfn + i)) | |
1292 | i++; | |
d6d8c8a4 | 1293 | if (i == MAX_ORDER_NR_PAGES || pfn + i >= end_pfn) |
5f0f2887 | 1294 | continue; |
1e0a98da MZ |
1295 | /* Check if we got outside of the zone */ |
1296 | if (zone && !zone_spans_pfn(zone, pfn + i)) | |
1297 | return 0; | |
5f0f2887 AB |
1298 | page = pfn_to_page(pfn + i); |
1299 | if (zone && page_zone(page) != zone) | |
1300 | return 0; | |
a96dfddb TK |
1301 | if (!zone) |
1302 | start = pfn + i; | |
5f0f2887 | 1303 | zone = page_zone(page); |
a96dfddb | 1304 | end = pfn + MAX_ORDER_NR_PAGES; |
5f0f2887 | 1305 | } |
0c0e6195 | 1306 | } |
deb88a2a | 1307 | |
a96dfddb TK |
1308 | if (zone) { |
1309 | *valid_start = start; | |
d6d8c8a4 | 1310 | *valid_end = min(end, end_pfn); |
deb88a2a | 1311 | return 1; |
a96dfddb | 1312 | } else { |
deb88a2a | 1313 | return 0; |
a96dfddb | 1314 | } |
0c0e6195 KH |
1315 | } |
1316 | ||
1317 | /* | |
0efadf48 YX |
1318 | * Scan pfn range [start,end) to find movable/migratable pages (LRU pages, |
1319 | * non-lru movable pages and hugepages). We scan pfn because it's much | |
1320 | * easier than scanning over linked list. This function returns the pfn | |
1321 | * of the first found movable page if it's found, otherwise 0. | |
0c0e6195 | 1322 | */ |
c8721bbb | 1323 | static unsigned long scan_movable_pages(unsigned long start, unsigned long end) |
0c0e6195 KH |
1324 | { |
1325 | unsigned long pfn; | |
88f24bbf | 1326 | |
0c0e6195 | 1327 | for (pfn = start; pfn < end; pfn++) { |
88f24bbf OS |
1328 | struct page *page, *head; |
1329 | unsigned long skip; | |
1330 | ||
1331 | if (!pfn_valid(pfn)) | |
1332 | continue; | |
1333 | page = pfn_to_page(pfn); | |
1334 | if (PageLRU(page)) | |
1335 | return pfn; | |
1336 | if (__PageMovable(page)) | |
1337 | return pfn; | |
1338 | ||
1339 | if (!PageHuge(page)) | |
1340 | continue; | |
1341 | head = compound_head(page); | |
1342 | if (page_huge_active(head)) | |
1343 | return pfn; | |
1344 | skip = (1 << compound_order(head)) - (page - head); | |
1345 | pfn += skip - 1; | |
0c0e6195 KH |
1346 | } |
1347 | return 0; | |
1348 | } | |
1349 | ||
394e31d2 XQ |
1350 | static struct page *new_node_page(struct page *page, unsigned long private, |
1351 | int **result) | |
1352 | { | |
394e31d2 | 1353 | int nid = page_to_nid(page); |
231e97e2 | 1354 | nodemask_t nmask = node_states[N_MEMORY]; |
7f252f27 MH |
1355 | |
1356 | /* | |
1357 | * try to allocate from a different node but reuse this node if there | |
1358 | * are no other online nodes to be used (e.g. we are offlining a part | |
1359 | * of the only existing node) | |
1360 | */ | |
1361 | node_clear(nid, nmask); | |
1362 | if (nodes_empty(nmask)) | |
1363 | node_set(nid, nmask); | |
394e31d2 | 1364 | |
8b913238 | 1365 | return new_page_nodemask(page, nid, &nmask); |
394e31d2 XQ |
1366 | } |
1367 | ||
0c0e6195 KH |
1368 | #define NR_OFFLINE_AT_ONCE_PAGES (256) |
1369 | static int | |
1370 | do_migrate_range(unsigned long start_pfn, unsigned long end_pfn) | |
1371 | { | |
1372 | unsigned long pfn; | |
1373 | struct page *page; | |
1374 | int move_pages = NR_OFFLINE_AT_ONCE_PAGES; | |
1375 | int not_managed = 0; | |
1376 | int ret = 0; | |
1377 | LIST_HEAD(source); | |
1378 | ||
1379 | for (pfn = start_pfn; pfn < end_pfn && move_pages > 0; pfn++) { | |
1380 | if (!pfn_valid(pfn)) | |
1381 | continue; | |
1382 | page = pfn_to_page(pfn); | |
c8721bbb NH |
1383 | |
1384 | if (PageHuge(page)) { | |
1385 | struct page *head = compound_head(page); | |
1386 | pfn = page_to_pfn(head) + (1<<compound_order(head)) - 1; | |
1387 | if (compound_order(head) > PFN_SECTION_SHIFT) { | |
1388 | ret = -EBUSY; | |
1389 | break; | |
1390 | } | |
1391 | if (isolate_huge_page(page, &source)) | |
1392 | move_pages -= 1 << compound_order(head); | |
1393 | continue; | |
8135d892 NH |
1394 | } else if (thp_migration_supported() && PageTransHuge(page)) |
1395 | pfn = page_to_pfn(compound_head(page)) | |
1396 | + hpage_nr_pages(page) - 1; | |
c8721bbb | 1397 | |
85953e73 MH |
1398 | /* |
1399 | * HWPoison pages have elevated reference counts so the migration would | |
1400 | * fail on them. It also doesn't make any sense to migrate them in the | |
1401 | * first place. Still try to unmap such a page in case it is still mapped | |
1402 | * (e.g. current hwpoison implementation doesn't unmap KSM pages but keep | |
1403 | * the unmap as the catch all safety net). | |
1404 | */ | |
1405 | if (PageHWPoison(page)) { | |
1406 | if (WARN_ON(PageLRU(page))) | |
1407 | isolate_lru_page(page); | |
1408 | if (page_mapped(page)) | |
1409 | try_to_unmap(page, TTU_IGNORE_MLOCK | TTU_IGNORE_ACCESS); | |
1410 | continue; | |
1411 | } | |
1412 | ||
700c2a46 | 1413 | if (!get_page_unless_zero(page)) |
0c0e6195 KH |
1414 | continue; |
1415 | /* | |
0efadf48 YX |
1416 | * We can skip free pages. And we can deal with pages on |
1417 | * LRU and non-lru movable pages. | |
0c0e6195 | 1418 | */ |
0efadf48 YX |
1419 | if (PageLRU(page)) |
1420 | ret = isolate_lru_page(page); | |
1421 | else | |
1422 | ret = isolate_movable_page(page, ISOLATE_UNEVICTABLE); | |
0c0e6195 | 1423 | if (!ret) { /* Success */ |
700c2a46 | 1424 | put_page(page); |
62695a84 | 1425 | list_add_tail(&page->lru, &source); |
0c0e6195 | 1426 | move_pages--; |
0efadf48 YX |
1427 | if (!__PageMovable(page)) |
1428 | inc_node_page_state(page, NR_ISOLATED_ANON + | |
1429 | page_is_file_cache(page)); | |
6d9c285a | 1430 | |
0c0e6195 | 1431 | } else { |
0c0e6195 | 1432 | #ifdef CONFIG_DEBUG_VM |
0efadf48 YX |
1433 | pr_alert("failed to isolate pfn %lx\n", pfn); |
1434 | dump_page(page, "isolation failed"); | |
0c0e6195 | 1435 | #endif |
700c2a46 | 1436 | put_page(page); |
25985edc | 1437 | /* Because we don't have big zone->lock. we should |
809c4449 BL |
1438 | check this again here. */ |
1439 | if (page_count(page)) { | |
1440 | not_managed++; | |
f3ab2636 | 1441 | ret = -EBUSY; |
809c4449 BL |
1442 | break; |
1443 | } | |
0c0e6195 KH |
1444 | } |
1445 | } | |
f3ab2636 BL |
1446 | if (!list_empty(&source)) { |
1447 | if (not_managed) { | |
c8721bbb | 1448 | putback_movable_pages(&source); |
f3ab2636 BL |
1449 | goto out; |
1450 | } | |
74c08f98 | 1451 | |
394e31d2 XQ |
1452 | /* Allocate a new page from the nearest neighbor node */ |
1453 | ret = migrate_pages(&source, new_node_page, NULL, 0, | |
9c620e2b | 1454 | MIGRATE_SYNC, MR_MEMORY_HOTPLUG); |
f3ab2636 | 1455 | if (ret) |
c8721bbb | 1456 | putback_movable_pages(&source); |
0c0e6195 | 1457 | } |
0c0e6195 KH |
1458 | out: |
1459 | return ret; | |
1460 | } | |
1461 | ||
1462 | /* | |
1463 | * remove from free_area[] and mark all as Reserved. | |
1464 | */ | |
1465 | static int | |
1466 | offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages, | |
1467 | void *data) | |
1468 | { | |
1469 | __offline_isolated_pages(start, start + nr_pages); | |
1470 | return 0; | |
1471 | } | |
1472 | ||
1473 | static void | |
1474 | offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn) | |
1475 | { | |
908eedc6 | 1476 | walk_system_ram_range(start_pfn, end_pfn - start_pfn, NULL, |
0c0e6195 KH |
1477 | offline_isolated_pages_cb); |
1478 | } | |
1479 | ||
1480 | /* | |
1481 | * Check all pages in range, recoreded as memory resource, are isolated. | |
1482 | */ | |
1483 | static int | |
1484 | check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages, | |
1485 | void *data) | |
1486 | { | |
1487 | int ret; | |
1488 | long offlined = *(long *)data; | |
b023f468 | 1489 | ret = test_pages_isolated(start_pfn, start_pfn + nr_pages, true); |
0c0e6195 KH |
1490 | offlined = nr_pages; |
1491 | if (!ret) | |
1492 | *(long *)data += offlined; | |
1493 | return ret; | |
1494 | } | |
1495 | ||
1496 | static long | |
1497 | check_pages_isolated(unsigned long start_pfn, unsigned long end_pfn) | |
1498 | { | |
1499 | long offlined = 0; | |
1500 | int ret; | |
1501 | ||
908eedc6 | 1502 | ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn, &offlined, |
0c0e6195 KH |
1503 | check_pages_isolated_cb); |
1504 | if (ret < 0) | |
1505 | offlined = (long)ret; | |
1506 | return offlined; | |
1507 | } | |
1508 | ||
c5320926 TC |
1509 | static int __init cmdline_parse_movable_node(char *p) |
1510 | { | |
4932381e | 1511 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
55ac590c | 1512 | movable_node_enabled = true; |
4932381e MH |
1513 | #else |
1514 | pr_warn("movable_node parameter depends on CONFIG_HAVE_MEMBLOCK_NODE_MAP to work properly\n"); | |
1515 | #endif | |
c5320926 TC |
1516 | return 0; |
1517 | } | |
1518 | early_param("movable_node", cmdline_parse_movable_node); | |
1519 | ||
d9713679 LJ |
1520 | /* check which state of node_states will be changed when offline memory */ |
1521 | static void node_states_check_changes_offline(unsigned long nr_pages, | |
1522 | struct zone *zone, struct memory_notify *arg) | |
1523 | { | |
1524 | struct pglist_data *pgdat = zone->zone_pgdat; | |
1525 | unsigned long present_pages = 0; | |
1526 | enum zone_type zt, zone_last = ZONE_NORMAL; | |
1527 | ||
1528 | /* | |
6715ddf9 LJ |
1529 | * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY] |
1530 | * contains nodes which have zones of 0...ZONE_NORMAL, | |
1531 | * set zone_last to ZONE_NORMAL. | |
d9713679 | 1532 | * |
6715ddf9 LJ |
1533 | * If we don't have HIGHMEM nor movable node, |
1534 | * node_states[N_NORMAL_MEMORY] contains nodes which have zones of | |
1535 | * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE. | |
d9713679 | 1536 | */ |
6715ddf9 | 1537 | if (N_MEMORY == N_NORMAL_MEMORY) |
d9713679 LJ |
1538 | zone_last = ZONE_MOVABLE; |
1539 | ||
1540 | /* | |
1541 | * check whether node_states[N_NORMAL_MEMORY] will be changed. | |
1542 | * If the memory to be offline is in a zone of 0...zone_last, | |
1543 | * and it is the last present memory, 0...zone_last will | |
1544 | * become empty after offline , thus we can determind we will | |
1545 | * need to clear the node from node_states[N_NORMAL_MEMORY]. | |
1546 | */ | |
1547 | for (zt = 0; zt <= zone_last; zt++) | |
1548 | present_pages += pgdat->node_zones[zt].present_pages; | |
1549 | if (zone_idx(zone) <= zone_last && nr_pages >= present_pages) | |
1550 | arg->status_change_nid_normal = zone_to_nid(zone); | |
1551 | else | |
1552 | arg->status_change_nid_normal = -1; | |
1553 | ||
6715ddf9 LJ |
1554 | #ifdef CONFIG_HIGHMEM |
1555 | /* | |
1556 | * If we have movable node, node_states[N_HIGH_MEMORY] | |
1557 | * contains nodes which have zones of 0...ZONE_HIGHMEM, | |
1558 | * set zone_last to ZONE_HIGHMEM. | |
1559 | * | |
1560 | * If we don't have movable node, node_states[N_NORMAL_MEMORY] | |
1561 | * contains nodes which have zones of 0...ZONE_MOVABLE, | |
1562 | * set zone_last to ZONE_MOVABLE. | |
1563 | */ | |
1564 | zone_last = ZONE_HIGHMEM; | |
1565 | if (N_MEMORY == N_HIGH_MEMORY) | |
1566 | zone_last = ZONE_MOVABLE; | |
1567 | ||
1568 | for (; zt <= zone_last; zt++) | |
1569 | present_pages += pgdat->node_zones[zt].present_pages; | |
1570 | if (zone_idx(zone) <= zone_last && nr_pages >= present_pages) | |
1571 | arg->status_change_nid_high = zone_to_nid(zone); | |
1572 | else | |
1573 | arg->status_change_nid_high = -1; | |
1574 | #else | |
1575 | arg->status_change_nid_high = arg->status_change_nid_normal; | |
1576 | #endif | |
1577 | ||
d9713679 LJ |
1578 | /* |
1579 | * node_states[N_HIGH_MEMORY] contains nodes which have 0...ZONE_MOVABLE | |
1580 | */ | |
1581 | zone_last = ZONE_MOVABLE; | |
1582 | ||
1583 | /* | |
1584 | * check whether node_states[N_HIGH_MEMORY] will be changed | |
1585 | * If we try to offline the last present @nr_pages from the node, | |
1586 | * we can determind we will need to clear the node from | |
1587 | * node_states[N_HIGH_MEMORY]. | |
1588 | */ | |
1589 | for (; zt <= zone_last; zt++) | |
1590 | present_pages += pgdat->node_zones[zt].present_pages; | |
1591 | if (nr_pages >= present_pages) | |
1592 | arg->status_change_nid = zone_to_nid(zone); | |
1593 | else | |
1594 | arg->status_change_nid = -1; | |
1595 | } | |
1596 | ||
1597 | static void node_states_clear_node(int node, struct memory_notify *arg) | |
1598 | { | |
1599 | if (arg->status_change_nid_normal >= 0) | |
1600 | node_clear_state(node, N_NORMAL_MEMORY); | |
1601 | ||
6715ddf9 LJ |
1602 | if ((N_MEMORY != N_NORMAL_MEMORY) && |
1603 | (arg->status_change_nid_high >= 0)) | |
d9713679 | 1604 | node_clear_state(node, N_HIGH_MEMORY); |
6715ddf9 LJ |
1605 | |
1606 | if ((N_MEMORY != N_HIGH_MEMORY) && | |
1607 | (arg->status_change_nid >= 0)) | |
1608 | node_clear_state(node, N_MEMORY); | |
d9713679 LJ |
1609 | } |
1610 | ||
a16cee10 | 1611 | static int __ref __offline_pages(unsigned long start_pfn, |
ecde0f3e | 1612 | unsigned long end_pfn) |
0c0e6195 | 1613 | { |
ecde0f3e | 1614 | unsigned long pfn, nr_pages; |
0c0e6195 | 1615 | long offlined_pages; |
72b39cfc | 1616 | int ret, node; |
d702909f | 1617 | unsigned long flags; |
a96dfddb | 1618 | unsigned long valid_start, valid_end; |
0c0e6195 | 1619 | struct zone *zone; |
7b78d335 | 1620 | struct memory_notify arg; |
0c0e6195 | 1621 | |
0c0e6195 KH |
1622 | /* at least, alignment against pageblock is necessary */ |
1623 | if (!IS_ALIGNED(start_pfn, pageblock_nr_pages)) | |
1624 | return -EINVAL; | |
1625 | if (!IS_ALIGNED(end_pfn, pageblock_nr_pages)) | |
1626 | return -EINVAL; | |
1679e5d6 DH |
1627 | |
1628 | mem_hotplug_begin(); | |
1629 | ||
0c0e6195 KH |
1630 | /* This makes hotplug much easier...and readable. |
1631 | we assume this for now. .*/ | |
1679e5d6 DH |
1632 | if (!test_pages_in_a_zone(start_pfn, end_pfn, &valid_start, |
1633 | &valid_end)) { | |
1634 | mem_hotplug_done(); | |
0c0e6195 | 1635 | return -EINVAL; |
1679e5d6 | 1636 | } |
7b78d335 | 1637 | |
a96dfddb | 1638 | zone = page_zone(pfn_to_page(valid_start)); |
7b78d335 YG |
1639 | node = zone_to_nid(zone); |
1640 | nr_pages = end_pfn - start_pfn; | |
1641 | ||
0c0e6195 | 1642 | /* set above range as isolated */ |
b023f468 WC |
1643 | ret = start_isolate_page_range(start_pfn, end_pfn, |
1644 | MIGRATE_MOVABLE, true); | |
1679e5d6 DH |
1645 | if (ret) { |
1646 | mem_hotplug_done(); | |
30467e0b | 1647 | return ret; |
1679e5d6 | 1648 | } |
7b78d335 YG |
1649 | |
1650 | arg.start_pfn = start_pfn; | |
1651 | arg.nr_pages = nr_pages; | |
d9713679 | 1652 | node_states_check_changes_offline(nr_pages, zone, &arg); |
7b78d335 YG |
1653 | |
1654 | ret = memory_notify(MEM_GOING_OFFLINE, &arg); | |
1655 | ret = notifier_to_errno(ret); | |
1656 | if (ret) | |
1657 | goto failed_removal; | |
1658 | ||
0c0e6195 | 1659 | pfn = start_pfn; |
0c0e6195 KH |
1660 | repeat: |
1661 | /* start memory hot removal */ | |
0c0e6195 KH |
1662 | ret = -EINTR; |
1663 | if (signal_pending(current)) | |
1664 | goto failed_removal; | |
72b39cfc MH |
1665 | |
1666 | cond_resched(); | |
1667 | lru_add_drain_all_cpuslocked(); | |
1668 | drain_all_pages(zone); | |
0c0e6195 | 1669 | |
c8721bbb NH |
1670 | pfn = scan_movable_pages(start_pfn, end_pfn); |
1671 | if (pfn) { /* We have movable pages */ | |
0c0e6195 | 1672 | ret = do_migrate_range(pfn, end_pfn); |
72b39cfc | 1673 | goto repeat; |
0c0e6195 | 1674 | } |
72b39cfc | 1675 | |
c8721bbb NH |
1676 | /* |
1677 | * dissolve free hugepages in the memory block before doing offlining | |
1678 | * actually in order to make hugetlbfs's object counting consistent. | |
1679 | */ | |
082d5b6b GS |
1680 | ret = dissolve_free_huge_pages(start_pfn, end_pfn); |
1681 | if (ret) | |
1682 | goto failed_removal; | |
0c0e6195 KH |
1683 | /* check again */ |
1684 | offlined_pages = check_pages_isolated(start_pfn, end_pfn); | |
72b39cfc MH |
1685 | if (offlined_pages < 0) |
1686 | goto repeat; | |
e33e33b4 | 1687 | pr_info("Offlined Pages %ld\n", offlined_pages); |
b3834be5 | 1688 | /* Ok, all of our target is isolated. |
0c0e6195 KH |
1689 | We cannot do rollback at this point. */ |
1690 | offline_isolated_pages(start_pfn, end_pfn); | |
dbc0e4ce | 1691 | /* reset pagetype flags and makes migrate type to be MOVABLE */ |
0815f3d8 | 1692 | undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE); |
0c0e6195 | 1693 | /* removal success */ |
3dcc0571 | 1694 | adjust_managed_page_count(pfn_to_page(start_pfn), -offlined_pages); |
0c0e6195 | 1695 | zone->present_pages -= offlined_pages; |
d702909f CS |
1696 | |
1697 | pgdat_resize_lock(zone->zone_pgdat, &flags); | |
0c0e6195 | 1698 | zone->zone_pgdat->node_present_pages -= offlined_pages; |
d702909f | 1699 | pgdat_resize_unlock(zone->zone_pgdat, &flags); |
7b78d335 | 1700 | |
1b79acc9 KM |
1701 | init_per_zone_wmark_min(); |
1702 | ||
1e8537ba | 1703 | if (!populated_zone(zone)) { |
340175b7 | 1704 | zone_pcp_reset(zone); |
72675e13 | 1705 | build_all_zonelists(NULL); |
1e8537ba XQ |
1706 | } else |
1707 | zone_pcp_update(zone); | |
340175b7 | 1708 | |
d9713679 | 1709 | node_states_clear_node(node, &arg); |
698b1b30 | 1710 | if (arg.status_change_nid >= 0) { |
8fe23e05 | 1711 | kswapd_stop(node); |
698b1b30 VB |
1712 | kcompactd_stop(node); |
1713 | } | |
bce7394a | 1714 | |
0c0e6195 KH |
1715 | vm_total_pages = nr_free_pagecache_pages(); |
1716 | writeback_set_ratelimit(); | |
7b78d335 YG |
1717 | |
1718 | memory_notify(MEM_OFFLINE, &arg); | |
1679e5d6 | 1719 | mem_hotplug_done(); |
0c0e6195 KH |
1720 | return 0; |
1721 | ||
1722 | failed_removal: | |
e33e33b4 CY |
1723 | pr_debug("memory offlining [mem %#010llx-%#010llx] failed\n", |
1724 | (unsigned long long) start_pfn << PAGE_SHIFT, | |
1725 | ((unsigned long long) end_pfn << PAGE_SHIFT) - 1); | |
7b78d335 | 1726 | memory_notify(MEM_CANCEL_OFFLINE, &arg); |
0c0e6195 | 1727 | /* pushback to free area */ |
0815f3d8 | 1728 | undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE); |
1679e5d6 | 1729 | mem_hotplug_done(); |
0c0e6195 KH |
1730 | return ret; |
1731 | } | |
71088785 | 1732 | |
a16cee10 WC |
1733 | int offline_pages(unsigned long start_pfn, unsigned long nr_pages) |
1734 | { | |
ecde0f3e | 1735 | return __offline_pages(start_pfn, start_pfn + nr_pages); |
a16cee10 | 1736 | } |
e2ff3940 | 1737 | #endif /* CONFIG_MEMORY_HOTREMOVE */ |
a16cee10 | 1738 | |
bbc76be6 WC |
1739 | /** |
1740 | * walk_memory_range - walks through all mem sections in [start_pfn, end_pfn) | |
1741 | * @start_pfn: start pfn of the memory range | |
e05c4bbf | 1742 | * @end_pfn: end pfn of the memory range |
bbc76be6 WC |
1743 | * @arg: argument passed to func |
1744 | * @func: callback for each memory section walked | |
1745 | * | |
1746 | * This function walks through all present mem sections in range | |
1747 | * [start_pfn, end_pfn) and call func on each mem section. | |
1748 | * | |
1749 | * Returns the return value of func. | |
1750 | */ | |
e2ff3940 | 1751 | int walk_memory_range(unsigned long start_pfn, unsigned long end_pfn, |
bbc76be6 | 1752 | void *arg, int (*func)(struct memory_block *, void *)) |
71088785 | 1753 | { |
e90bdb7f WC |
1754 | struct memory_block *mem = NULL; |
1755 | struct mem_section *section; | |
e90bdb7f WC |
1756 | unsigned long pfn, section_nr; |
1757 | int ret; | |
e90bdb7f WC |
1758 | |
1759 | for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) { | |
1760 | section_nr = pfn_to_section_nr(pfn); | |
1761 | if (!present_section_nr(section_nr)) | |
1762 | continue; | |
1763 | ||
1764 | section = __nr_to_section(section_nr); | |
1765 | /* same memblock? */ | |
1766 | if (mem) | |
1767 | if ((section_nr >= mem->start_section_nr) && | |
1768 | (section_nr <= mem->end_section_nr)) | |
1769 | continue; | |
1770 | ||
1771 | mem = find_memory_block_hinted(section, mem); | |
1772 | if (!mem) | |
1773 | continue; | |
1774 | ||
bbc76be6 | 1775 | ret = func(mem, arg); |
e90bdb7f | 1776 | if (ret) { |
bbc76be6 WC |
1777 | kobject_put(&mem->dev.kobj); |
1778 | return ret; | |
e90bdb7f WC |
1779 | } |
1780 | } | |
1781 | ||
1782 | if (mem) | |
1783 | kobject_put(&mem->dev.kobj); | |
1784 | ||
bbc76be6 WC |
1785 | return 0; |
1786 | } | |
1787 | ||
e2ff3940 | 1788 | #ifdef CONFIG_MEMORY_HOTREMOVE |
d6de9d53 | 1789 | static int check_memblock_offlined_cb(struct memory_block *mem, void *arg) |
bbc76be6 WC |
1790 | { |
1791 | int ret = !is_memblock_offlined(mem); | |
1792 | ||
349daa0f RD |
1793 | if (unlikely(ret)) { |
1794 | phys_addr_t beginpa, endpa; | |
1795 | ||
1796 | beginpa = PFN_PHYS(section_nr_to_pfn(mem->start_section_nr)); | |
1797 | endpa = PFN_PHYS(section_nr_to_pfn(mem->end_section_nr + 1))-1; | |
756a025f | 1798 | pr_warn("removing memory fails, because memory [%pa-%pa] is onlined\n", |
349daa0f RD |
1799 | &beginpa, &endpa); |
1800 | } | |
bbc76be6 WC |
1801 | |
1802 | return ret; | |
1803 | } | |
1804 | ||
0f1cfe9d | 1805 | static int check_cpu_on_node(pg_data_t *pgdat) |
60a5a19e | 1806 | { |
60a5a19e TC |
1807 | int cpu; |
1808 | ||
1809 | for_each_present_cpu(cpu) { | |
1810 | if (cpu_to_node(cpu) == pgdat->node_id) | |
1811 | /* | |
1812 | * the cpu on this node isn't removed, and we can't | |
1813 | * offline this node. | |
1814 | */ | |
1815 | return -EBUSY; | |
1816 | } | |
1817 | ||
1818 | return 0; | |
1819 | } | |
1820 | ||
0f1cfe9d | 1821 | static void unmap_cpu_on_node(pg_data_t *pgdat) |
e13fe869 WC |
1822 | { |
1823 | #ifdef CONFIG_ACPI_NUMA | |
e13fe869 WC |
1824 | int cpu; |
1825 | ||
1826 | for_each_possible_cpu(cpu) | |
1827 | if (cpu_to_node(cpu) == pgdat->node_id) | |
1828 | numa_clear_node(cpu); | |
1829 | #endif | |
1830 | } | |
1831 | ||
0f1cfe9d | 1832 | static int check_and_unmap_cpu_on_node(pg_data_t *pgdat) |
e13fe869 | 1833 | { |
0f1cfe9d | 1834 | int ret; |
e13fe869 | 1835 | |
0f1cfe9d | 1836 | ret = check_cpu_on_node(pgdat); |
e13fe869 WC |
1837 | if (ret) |
1838 | return ret; | |
1839 | ||
1840 | /* | |
1841 | * the node will be offlined when we come here, so we can clear | |
1842 | * the cpu_to_node() now. | |
1843 | */ | |
1844 | ||
0f1cfe9d | 1845 | unmap_cpu_on_node(pgdat); |
e13fe869 WC |
1846 | return 0; |
1847 | } | |
1848 | ||
0f1cfe9d TK |
1849 | /** |
1850 | * try_offline_node | |
1851 | * | |
1852 | * Offline a node if all memory sections and cpus of the node are removed. | |
1853 | * | |
1854 | * NOTE: The caller must call lock_device_hotplug() to serialize hotplug | |
1855 | * and online/offline operations before this call. | |
1856 | */ | |
90b30cdc | 1857 | void try_offline_node(int nid) |
60a5a19e | 1858 | { |
d822b86a WC |
1859 | pg_data_t *pgdat = NODE_DATA(nid); |
1860 | unsigned long start_pfn = pgdat->node_start_pfn; | |
1861 | unsigned long end_pfn = start_pfn + pgdat->node_spanned_pages; | |
60a5a19e TC |
1862 | unsigned long pfn; |
1863 | ||
1864 | for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) { | |
1865 | unsigned long section_nr = pfn_to_section_nr(pfn); | |
1866 | ||
1867 | if (!present_section_nr(section_nr)) | |
1868 | continue; | |
1869 | ||
1870 | if (pfn_to_nid(pfn) != nid) | |
1871 | continue; | |
1872 | ||
1873 | /* | |
1874 | * some memory sections of this node are not removed, and we | |
1875 | * can't offline node now. | |
1876 | */ | |
1877 | return; | |
1878 | } | |
1879 | ||
0f1cfe9d | 1880 | if (check_and_unmap_cpu_on_node(pgdat)) |
60a5a19e TC |
1881 | return; |
1882 | ||
1883 | /* | |
1884 | * all memory/cpu of this node are removed, we can offline this | |
1885 | * node now. | |
1886 | */ | |
1887 | node_set_offline(nid); | |
1888 | unregister_one_node(nid); | |
1889 | } | |
90b30cdc | 1890 | EXPORT_SYMBOL(try_offline_node); |
60a5a19e | 1891 | |
0f1cfe9d TK |
1892 | /** |
1893 | * remove_memory | |
1894 | * | |
1895 | * NOTE: The caller must call lock_device_hotplug() to serialize hotplug | |
1896 | * and online/offline operations before this call, as required by | |
1897 | * try_offline_node(). | |
1898 | */ | |
242831eb | 1899 | void __ref remove_memory(int nid, u64 start, u64 size) |
bbc76be6 | 1900 | { |
242831eb | 1901 | int ret; |
993c1aad | 1902 | |
27356f54 TK |
1903 | BUG_ON(check_hotplug_memory_range(start, size)); |
1904 | ||
bfc8c901 | 1905 | mem_hotplug_begin(); |
6677e3ea YI |
1906 | |
1907 | /* | |
242831eb RW |
1908 | * All memory blocks must be offlined before removing memory. Check |
1909 | * whether all memory blocks in question are offline and trigger a BUG() | |
1910 | * if this is not the case. | |
6677e3ea | 1911 | */ |
242831eb | 1912 | ret = walk_memory_range(PFN_DOWN(start), PFN_UP(start + size - 1), NULL, |
d6de9d53 | 1913 | check_memblock_offlined_cb); |
bfc8c901 | 1914 | if (ret) |
242831eb | 1915 | BUG(); |
6677e3ea | 1916 | |
46c66c4b YI |
1917 | /* remove memmap entry */ |
1918 | firmware_map_remove(start, start + size, "System RAM"); | |
f9126ab9 XQ |
1919 | memblock_free(start, size); |
1920 | memblock_remove(start, size); | |
46c66c4b | 1921 | |
24d335ca WC |
1922 | arch_remove_memory(start, size); |
1923 | ||
60a5a19e TC |
1924 | try_offline_node(nid); |
1925 | ||
bfc8c901 | 1926 | mem_hotplug_done(); |
71088785 | 1927 | } |
71088785 | 1928 | EXPORT_SYMBOL_GPL(remove_memory); |
aba6efc4 | 1929 | #endif /* CONFIG_MEMORY_HOTREMOVE */ |