]>
Commit | Line | Data |
---|---|---|
457c8996 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
3947be19 DH |
2 | /* |
3 | * linux/mm/memory_hotplug.c | |
4 | * | |
5 | * Copyright (C) | |
6 | */ | |
7 | ||
3947be19 DH |
8 | #include <linux/stddef.h> |
9 | #include <linux/mm.h> | |
174cd4b1 | 10 | #include <linux/sched/signal.h> |
3947be19 DH |
11 | #include <linux/swap.h> |
12 | #include <linux/interrupt.h> | |
13 | #include <linux/pagemap.h> | |
3947be19 | 14 | #include <linux/compiler.h> |
b95f1b31 | 15 | #include <linux/export.h> |
3947be19 | 16 | #include <linux/pagevec.h> |
2d1d43f6 | 17 | #include <linux/writeback.h> |
3947be19 DH |
18 | #include <linux/slab.h> |
19 | #include <linux/sysctl.h> | |
20 | #include <linux/cpu.h> | |
21 | #include <linux/memory.h> | |
4b94ffdc | 22 | #include <linux/memremap.h> |
3947be19 DH |
23 | #include <linux/memory_hotplug.h> |
24 | #include <linux/highmem.h> | |
25 | #include <linux/vmalloc.h> | |
0a547039 | 26 | #include <linux/ioport.h> |
0c0e6195 KH |
27 | #include <linux/delay.h> |
28 | #include <linux/migrate.h> | |
29 | #include <linux/page-isolation.h> | |
71088785 | 30 | #include <linux/pfn.h> |
6ad696d2 | 31 | #include <linux/suspend.h> |
6d9c285a | 32 | #include <linux/mm_inline.h> |
d96ae530 | 33 | #include <linux/firmware-map.h> |
60a5a19e | 34 | #include <linux/stop_machine.h> |
c8721bbb | 35 | #include <linux/hugetlb.h> |
c5320926 | 36 | #include <linux/memblock.h> |
698b1b30 | 37 | #include <linux/compaction.h> |
b15c8726 | 38 | #include <linux/rmap.h> |
3947be19 DH |
39 | |
40 | #include <asm/tlbflush.h> | |
41 | ||
1e5ad9a3 | 42 | #include "internal.h" |
e900a918 | 43 | #include "shuffle.h" |
1e5ad9a3 | 44 | |
9d0ad8ca DK |
45 | /* |
46 | * online_page_callback contains pointer to current page onlining function. | |
47 | * Initially it is generic_online_page(). If it is required it could be | |
48 | * changed by calling set_online_page_callback() for callback registration | |
49 | * and restore_online_page_callback() for generic callback restore. | |
50 | */ | |
51 | ||
9d0ad8ca | 52 | static online_page_callback_t online_page_callback = generic_online_page; |
bfc8c901 | 53 | static DEFINE_MUTEX(online_page_callback_lock); |
9d0ad8ca | 54 | |
3f906ba2 | 55 | DEFINE_STATIC_PERCPU_RWSEM(mem_hotplug_lock); |
bfc8c901 | 56 | |
3f906ba2 TG |
57 | void get_online_mems(void) |
58 | { | |
59 | percpu_down_read(&mem_hotplug_lock); | |
60 | } | |
bfc8c901 | 61 | |
3f906ba2 TG |
62 | void put_online_mems(void) |
63 | { | |
64 | percpu_up_read(&mem_hotplug_lock); | |
65 | } | |
bfc8c901 | 66 | |
4932381e MH |
67 | bool movable_node_enabled = false; |
68 | ||
8604d9e5 | 69 | #ifndef CONFIG_MEMORY_HOTPLUG_DEFAULT_ONLINE |
1adf8b46 | 70 | int mhp_default_online_type = MMOP_OFFLINE; |
8604d9e5 | 71 | #else |
1adf8b46 | 72 | int mhp_default_online_type = MMOP_ONLINE; |
8604d9e5 | 73 | #endif |
31bc3858 | 74 | |
86dd995d VK |
75 | static int __init setup_memhp_default_state(char *str) |
76 | { | |
1adf8b46 | 77 | const int online_type = mhp_online_type_from_str(str); |
5f47adf7 DH |
78 | |
79 | if (online_type >= 0) | |
1adf8b46 | 80 | mhp_default_online_type = online_type; |
86dd995d VK |
81 | |
82 | return 1; | |
83 | } | |
84 | __setup("memhp_default_state=", setup_memhp_default_state); | |
85 | ||
30467e0b | 86 | void mem_hotplug_begin(void) |
20d6c96b | 87 | { |
3f906ba2 TG |
88 | cpus_read_lock(); |
89 | percpu_down_write(&mem_hotplug_lock); | |
20d6c96b KM |
90 | } |
91 | ||
30467e0b | 92 | void mem_hotplug_done(void) |
bfc8c901 | 93 | { |
3f906ba2 TG |
94 | percpu_up_write(&mem_hotplug_lock); |
95 | cpus_read_unlock(); | |
bfc8c901 | 96 | } |
20d6c96b | 97 | |
357b4da5 JG |
98 | u64 max_mem_size = U64_MAX; |
99 | ||
45e0b78b | 100 | /* add this memory to iomem resource */ |
7b7b2721 DH |
101 | static struct resource *register_memory_resource(u64 start, u64 size, |
102 | const char *resource_name) | |
45e0b78b | 103 | { |
2794129e DH |
104 | struct resource *res; |
105 | unsigned long flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY; | |
7b7b2721 DH |
106 | |
107 | if (strcmp(resource_name, "System RAM")) | |
7cf603d1 | 108 | flags |= IORESOURCE_SYSRAM_DRIVER_MANAGED; |
357b4da5 | 109 | |
bca3feaa AK |
110 | if (!mhp_range_allowed(start, size, true)) |
111 | return ERR_PTR(-E2BIG); | |
112 | ||
f3cd4c86 BH |
113 | /* |
114 | * Make sure value parsed from 'mem=' only restricts memory adding | |
115 | * while booting, so that memory hotplug won't be impacted. Please | |
116 | * refer to document of 'mem=' in kernel-parameters.txt for more | |
117 | * details. | |
118 | */ | |
119 | if (start + size > max_mem_size && system_state < SYSTEM_RUNNING) | |
357b4da5 JG |
120 | return ERR_PTR(-E2BIG); |
121 | ||
2794129e DH |
122 | /* |
123 | * Request ownership of the new memory range. This might be | |
124 | * a child of an existing resource that was present but | |
125 | * not marked as busy. | |
126 | */ | |
127 | res = __request_region(&iomem_resource, start, size, | |
128 | resource_name, flags); | |
129 | ||
130 | if (!res) { | |
131 | pr_debug("Unable to reserve System RAM region: %016llx->%016llx\n", | |
132 | start, start + size); | |
6f754ba4 | 133 | return ERR_PTR(-EEXIST); |
45e0b78b KM |
134 | } |
135 | return res; | |
136 | } | |
137 | ||
138 | static void release_memory_resource(struct resource *res) | |
139 | { | |
140 | if (!res) | |
141 | return; | |
142 | release_resource(res); | |
143 | kfree(res); | |
45e0b78b KM |
144 | } |
145 | ||
53947027 | 146 | #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE |
46723bfa YI |
147 | void get_page_bootmem(unsigned long info, struct page *page, |
148 | unsigned long type) | |
04753278 | 149 | { |
ddffe98d | 150 | page->freelist = (void *)type; |
04753278 YG |
151 | SetPagePrivate(page); |
152 | set_page_private(page, info); | |
fe896d18 | 153 | page_ref_inc(page); |
04753278 YG |
154 | } |
155 | ||
170a5a7e | 156 | void put_page_bootmem(struct page *page) |
04753278 | 157 | { |
5f24ce5f | 158 | unsigned long type; |
04753278 | 159 | |
ddffe98d | 160 | type = (unsigned long) page->freelist; |
5f24ce5f AA |
161 | BUG_ON(type < MEMORY_HOTPLUG_MIN_BOOTMEM_TYPE || |
162 | type > MEMORY_HOTPLUG_MAX_BOOTMEM_TYPE); | |
04753278 | 163 | |
fe896d18 | 164 | if (page_ref_dec_return(page) == 1) { |
ddffe98d | 165 | page->freelist = NULL; |
04753278 YG |
166 | ClearPagePrivate(page); |
167 | set_page_private(page, 0); | |
5f24ce5f | 168 | INIT_LIST_HEAD(&page->lru); |
170a5a7e | 169 | free_reserved_page(page); |
04753278 | 170 | } |
04753278 YG |
171 | } |
172 | ||
46723bfa YI |
173 | #ifdef CONFIG_HAVE_BOOTMEM_INFO_NODE |
174 | #ifndef CONFIG_SPARSEMEM_VMEMMAP | |
d92bc318 | 175 | static void register_page_bootmem_info_section(unsigned long start_pfn) |
04753278 | 176 | { |
f1eca35a | 177 | unsigned long mapsize, section_nr, i; |
04753278 YG |
178 | struct mem_section *ms; |
179 | struct page *page, *memmap; | |
f1eca35a | 180 | struct mem_section_usage *usage; |
04753278 | 181 | |
04753278 YG |
182 | section_nr = pfn_to_section_nr(start_pfn); |
183 | ms = __nr_to_section(section_nr); | |
184 | ||
185 | /* Get section's memmap address */ | |
186 | memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr); | |
187 | ||
188 | /* | |
189 | * Get page for the memmap's phys address | |
190 | * XXX: need more consideration for sparse_vmemmap... | |
191 | */ | |
192 | page = virt_to_page(memmap); | |
193 | mapsize = sizeof(struct page) * PAGES_PER_SECTION; | |
194 | mapsize = PAGE_ALIGN(mapsize) >> PAGE_SHIFT; | |
195 | ||
196 | /* remember memmap's page */ | |
197 | for (i = 0; i < mapsize; i++, page++) | |
198 | get_page_bootmem(section_nr, page, SECTION_INFO); | |
199 | ||
f1eca35a DW |
200 | usage = ms->usage; |
201 | page = virt_to_page(usage); | |
04753278 | 202 | |
f1eca35a | 203 | mapsize = PAGE_ALIGN(mem_section_usage_size()) >> PAGE_SHIFT; |
04753278 YG |
204 | |
205 | for (i = 0; i < mapsize; i++, page++) | |
af370fb8 | 206 | get_page_bootmem(section_nr, page, MIX_SECTION_INFO); |
04753278 YG |
207 | |
208 | } | |
46723bfa YI |
209 | #else /* CONFIG_SPARSEMEM_VMEMMAP */ |
210 | static void register_page_bootmem_info_section(unsigned long start_pfn) | |
211 | { | |
f1eca35a | 212 | unsigned long mapsize, section_nr, i; |
46723bfa YI |
213 | struct mem_section *ms; |
214 | struct page *page, *memmap; | |
f1eca35a | 215 | struct mem_section_usage *usage; |
46723bfa | 216 | |
46723bfa YI |
217 | section_nr = pfn_to_section_nr(start_pfn); |
218 | ms = __nr_to_section(section_nr); | |
219 | ||
220 | memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr); | |
221 | ||
222 | register_page_bootmem_memmap(section_nr, memmap, PAGES_PER_SECTION); | |
223 | ||
f1eca35a DW |
224 | usage = ms->usage; |
225 | page = virt_to_page(usage); | |
46723bfa | 226 | |
f1eca35a | 227 | mapsize = PAGE_ALIGN(mem_section_usage_size()) >> PAGE_SHIFT; |
46723bfa YI |
228 | |
229 | for (i = 0; i < mapsize; i++, page++) | |
230 | get_page_bootmem(section_nr, page, MIX_SECTION_INFO); | |
231 | } | |
232 | #endif /* !CONFIG_SPARSEMEM_VMEMMAP */ | |
04753278 | 233 | |
7ded384a | 234 | void __init register_page_bootmem_info_node(struct pglist_data *pgdat) |
04753278 YG |
235 | { |
236 | unsigned long i, pfn, end_pfn, nr_pages; | |
237 | int node = pgdat->node_id; | |
238 | struct page *page; | |
04753278 YG |
239 | |
240 | nr_pages = PAGE_ALIGN(sizeof(struct pglist_data)) >> PAGE_SHIFT; | |
241 | page = virt_to_page(pgdat); | |
242 | ||
243 | for (i = 0; i < nr_pages; i++, page++) | |
244 | get_page_bootmem(node, page, NODE_INFO); | |
245 | ||
04753278 | 246 | pfn = pgdat->node_start_pfn; |
c1f19495 | 247 | end_pfn = pgdat_end_pfn(pgdat); |
04753278 | 248 | |
7e9f5eb0 | 249 | /* register section info */ |
f14851af | 250 | for (; pfn < end_pfn; pfn += PAGES_PER_SECTION) { |
251 | /* | |
252 | * Some platforms can assign the same pfn to multiple nodes - on | |
253 | * node0 as well as nodeN. To avoid registering a pfn against | |
254 | * multiple nodes we check that this pfn does not already | |
7e9f5eb0 | 255 | * reside in some other nodes. |
f14851af | 256 | */ |
f65e91df | 257 | if (pfn_valid(pfn) && (early_pfn_to_nid(pfn) == node)) |
f14851af | 258 | register_page_bootmem_info_section(pfn); |
259 | } | |
04753278 | 260 | } |
46723bfa | 261 | #endif /* CONFIG_HAVE_BOOTMEM_INFO_NODE */ |
04753278 | 262 | |
7ea62160 DW |
263 | static int check_pfn_span(unsigned long pfn, unsigned long nr_pages, |
264 | const char *reason) | |
265 | { | |
266 | /* | |
267 | * Disallow all operations smaller than a sub-section and only | |
268 | * allow operations smaller than a section for | |
269 | * SPARSEMEM_VMEMMAP. Note that check_hotplug_memory_range() | |
270 | * enforces a larger memory_block_size_bytes() granularity for | |
271 | * memory that will be marked online, so this check should only | |
272 | * fire for direct arch_{add,remove}_memory() users outside of | |
273 | * add_memory_resource(). | |
274 | */ | |
275 | unsigned long min_align; | |
276 | ||
277 | if (IS_ENABLED(CONFIG_SPARSEMEM_VMEMMAP)) | |
278 | min_align = PAGES_PER_SUBSECTION; | |
279 | else | |
280 | min_align = PAGES_PER_SECTION; | |
281 | if (!IS_ALIGNED(pfn, min_align) | |
282 | || !IS_ALIGNED(nr_pages, min_align)) { | |
283 | WARN(1, "Misaligned __%s_pages start: %#lx end: #%lx\n", | |
284 | reason, pfn, pfn + nr_pages - 1); | |
285 | return -EINVAL; | |
286 | } | |
287 | return 0; | |
288 | } | |
289 | ||
9f605f26 DW |
290 | /* |
291 | * Return page for the valid pfn only if the page is online. All pfn | |
292 | * walkers which rely on the fully initialized page->flags and others | |
293 | * should use this rather than pfn_valid && pfn_to_page | |
294 | */ | |
295 | struct page *pfn_to_online_page(unsigned long pfn) | |
296 | { | |
297 | unsigned long nr = pfn_to_section_nr(pfn); | |
1f90a347 | 298 | struct dev_pagemap *pgmap; |
9f9b02e5 DW |
299 | struct mem_section *ms; |
300 | ||
301 | if (nr >= NR_MEM_SECTIONS) | |
302 | return NULL; | |
303 | ||
304 | ms = __nr_to_section(nr); | |
305 | if (!online_section(ms)) | |
306 | return NULL; | |
307 | ||
308 | /* | |
309 | * Save some code text when online_section() + | |
310 | * pfn_section_valid() are sufficient. | |
311 | */ | |
312 | if (IS_ENABLED(CONFIG_HAVE_ARCH_PFN_VALID) && !pfn_valid(pfn)) | |
313 | return NULL; | |
314 | ||
315 | if (!pfn_section_valid(ms, pfn)) | |
316 | return NULL; | |
9f605f26 | 317 | |
1f90a347 DW |
318 | if (!online_device_section(ms)) |
319 | return pfn_to_page(pfn); | |
320 | ||
321 | /* | |
322 | * Slowpath: when ZONE_DEVICE collides with | |
323 | * ZONE_{NORMAL,MOVABLE} within the same section some pfns in | |
324 | * the section may be 'offline' but 'valid'. Only | |
325 | * get_dev_pagemap() can determine sub-section online status. | |
326 | */ | |
327 | pgmap = get_dev_pagemap(pfn, NULL); | |
328 | put_dev_pagemap(pgmap); | |
329 | ||
330 | /* The presence of a pgmap indicates ZONE_DEVICE offline pfn */ | |
331 | if (pgmap) | |
332 | return NULL; | |
333 | ||
9f9b02e5 | 334 | return pfn_to_page(pfn); |
9f605f26 DW |
335 | } |
336 | EXPORT_SYMBOL_GPL(pfn_to_online_page); | |
337 | ||
4edd7cef DR |
338 | /* |
339 | * Reasonably generic function for adding memory. It is | |
340 | * expected that archs that support memory hotplug will | |
341 | * call this function after deciding the zone to which to | |
342 | * add the new pages. | |
343 | */ | |
7ea62160 | 344 | int __ref __add_pages(int nid, unsigned long pfn, unsigned long nr_pages, |
f5637d3b | 345 | struct mhp_params *params) |
4edd7cef | 346 | { |
6cdd0b30 DH |
347 | const unsigned long end_pfn = pfn + nr_pages; |
348 | unsigned long cur_nr_pages; | |
9a845030 | 349 | int err; |
f5637d3b | 350 | struct vmem_altmap *altmap = params->altmap; |
4b94ffdc | 351 | |
bfeb022f LG |
352 | if (WARN_ON_ONCE(!params->pgprot.pgprot)) |
353 | return -EINVAL; | |
354 | ||
bca3feaa | 355 | VM_BUG_ON(!mhp_range_allowed(PFN_PHYS(pfn), nr_pages * PAGE_SIZE, false)); |
dca4436d | 356 | |
4b94ffdc DW |
357 | if (altmap) { |
358 | /* | |
359 | * Validate altmap is within bounds of the total request | |
360 | */ | |
7ea62160 | 361 | if (altmap->base_pfn != pfn |
4b94ffdc DW |
362 | || vmem_altmap_offset(altmap) > nr_pages) { |
363 | pr_warn_once("memory add fail, invalid altmap\n"); | |
7ea62160 | 364 | return -EINVAL; |
4b94ffdc DW |
365 | } |
366 | altmap->alloc = 0; | |
367 | } | |
368 | ||
7ea62160 DW |
369 | err = check_pfn_span(pfn, nr_pages, "add"); |
370 | if (err) | |
371 | return err; | |
372 | ||
6cdd0b30 DH |
373 | for (; pfn < end_pfn; pfn += cur_nr_pages) { |
374 | /* Select all remaining pages up to the next section boundary */ | |
375 | cur_nr_pages = min(end_pfn - pfn, | |
376 | SECTION_ALIGN_UP(pfn + 1) - pfn); | |
377 | err = sparse_add_section(nid, pfn, cur_nr_pages, altmap); | |
ba72b4c8 DW |
378 | if (err) |
379 | break; | |
f64ac5e6 | 380 | cond_resched(); |
4edd7cef | 381 | } |
c435a390 | 382 | vmemmap_populate_print_last(); |
4edd7cef DR |
383 | return err; |
384 | } | |
4edd7cef | 385 | |
815121d2 | 386 | /* find the smallest valid pfn in the range [start_pfn, end_pfn) */ |
d09b0137 | 387 | static unsigned long find_smallest_section_pfn(int nid, struct zone *zone, |
815121d2 YI |
388 | unsigned long start_pfn, |
389 | unsigned long end_pfn) | |
390 | { | |
49ba3c6b | 391 | for (; start_pfn < end_pfn; start_pfn += PAGES_PER_SUBSECTION) { |
7ce700bf | 392 | if (unlikely(!pfn_to_online_page(start_pfn))) |
815121d2 YI |
393 | continue; |
394 | ||
395 | if (unlikely(pfn_to_nid(start_pfn) != nid)) | |
396 | continue; | |
397 | ||
9b05158f | 398 | if (zone != page_zone(pfn_to_page(start_pfn))) |
815121d2 YI |
399 | continue; |
400 | ||
401 | return start_pfn; | |
402 | } | |
403 | ||
404 | return 0; | |
405 | } | |
406 | ||
407 | /* find the biggest valid pfn in the range [start_pfn, end_pfn). */ | |
d09b0137 | 408 | static unsigned long find_biggest_section_pfn(int nid, struct zone *zone, |
815121d2 YI |
409 | unsigned long start_pfn, |
410 | unsigned long end_pfn) | |
411 | { | |
815121d2 YI |
412 | unsigned long pfn; |
413 | ||
414 | /* pfn is the end pfn of a memory section. */ | |
415 | pfn = end_pfn - 1; | |
49ba3c6b | 416 | for (; pfn >= start_pfn; pfn -= PAGES_PER_SUBSECTION) { |
7ce700bf | 417 | if (unlikely(!pfn_to_online_page(pfn))) |
815121d2 YI |
418 | continue; |
419 | ||
420 | if (unlikely(pfn_to_nid(pfn) != nid)) | |
421 | continue; | |
422 | ||
9b05158f | 423 | if (zone != page_zone(pfn_to_page(pfn))) |
815121d2 YI |
424 | continue; |
425 | ||
426 | return pfn; | |
427 | } | |
428 | ||
429 | return 0; | |
430 | } | |
431 | ||
432 | static void shrink_zone_span(struct zone *zone, unsigned long start_pfn, | |
433 | unsigned long end_pfn) | |
434 | { | |
815121d2 | 435 | unsigned long pfn; |
815121d2 YI |
436 | int nid = zone_to_nid(zone); |
437 | ||
438 | zone_span_writelock(zone); | |
5d12071c | 439 | if (zone->zone_start_pfn == start_pfn) { |
815121d2 YI |
440 | /* |
441 | * If the section is smallest section in the zone, it need | |
442 | * shrink zone->zone_start_pfn and zone->zone_spanned_pages. | |
443 | * In this case, we find second smallest valid mem_section | |
444 | * for shrinking zone. | |
445 | */ | |
446 | pfn = find_smallest_section_pfn(nid, zone, end_pfn, | |
5d12071c | 447 | zone_end_pfn(zone)); |
815121d2 | 448 | if (pfn) { |
5d12071c | 449 | zone->spanned_pages = zone_end_pfn(zone) - pfn; |
815121d2 | 450 | zone->zone_start_pfn = pfn; |
950b68d9 DH |
451 | } else { |
452 | zone->zone_start_pfn = 0; | |
453 | zone->spanned_pages = 0; | |
815121d2 | 454 | } |
5d12071c | 455 | } else if (zone_end_pfn(zone) == end_pfn) { |
815121d2 YI |
456 | /* |
457 | * If the section is biggest section in the zone, it need | |
458 | * shrink zone->spanned_pages. | |
459 | * In this case, we find second biggest valid mem_section for | |
460 | * shrinking zone. | |
461 | */ | |
5d12071c | 462 | pfn = find_biggest_section_pfn(nid, zone, zone->zone_start_pfn, |
815121d2 YI |
463 | start_pfn); |
464 | if (pfn) | |
5d12071c | 465 | zone->spanned_pages = pfn - zone->zone_start_pfn + 1; |
950b68d9 DH |
466 | else { |
467 | zone->zone_start_pfn = 0; | |
468 | zone->spanned_pages = 0; | |
469 | } | |
815121d2 | 470 | } |
815121d2 YI |
471 | zone_span_writeunlock(zone); |
472 | } | |
473 | ||
00d6c019 | 474 | static void update_pgdat_span(struct pglist_data *pgdat) |
815121d2 | 475 | { |
00d6c019 DH |
476 | unsigned long node_start_pfn = 0, node_end_pfn = 0; |
477 | struct zone *zone; | |
478 | ||
479 | for (zone = pgdat->node_zones; | |
480 | zone < pgdat->node_zones + MAX_NR_ZONES; zone++) { | |
6c922cf7 | 481 | unsigned long end_pfn = zone_end_pfn(zone); |
00d6c019 DH |
482 | |
483 | /* No need to lock the zones, they can't change. */ | |
656d5711 DH |
484 | if (!zone->spanned_pages) |
485 | continue; | |
486 | if (!node_end_pfn) { | |
487 | node_start_pfn = zone->zone_start_pfn; | |
6c922cf7 | 488 | node_end_pfn = end_pfn; |
656d5711 DH |
489 | continue; |
490 | } | |
491 | ||
6c922cf7 ML |
492 | if (end_pfn > node_end_pfn) |
493 | node_end_pfn = end_pfn; | |
00d6c019 DH |
494 | if (zone->zone_start_pfn < node_start_pfn) |
495 | node_start_pfn = zone->zone_start_pfn; | |
815121d2 YI |
496 | } |
497 | ||
00d6c019 DH |
498 | pgdat->node_start_pfn = node_start_pfn; |
499 | pgdat->node_spanned_pages = node_end_pfn - node_start_pfn; | |
815121d2 YI |
500 | } |
501 | ||
feee6b29 DH |
502 | void __ref remove_pfn_range_from_zone(struct zone *zone, |
503 | unsigned long start_pfn, | |
504 | unsigned long nr_pages) | |
815121d2 | 505 | { |
b7e3debd | 506 | const unsigned long end_pfn = start_pfn + nr_pages; |
815121d2 | 507 | struct pglist_data *pgdat = zone->zone_pgdat; |
b7e3debd | 508 | unsigned long pfn, cur_nr_pages, flags; |
815121d2 | 509 | |
d33695b1 | 510 | /* Poison struct pages because they are now uninitialized again. */ |
b7e3debd BW |
511 | for (pfn = start_pfn; pfn < end_pfn; pfn += cur_nr_pages) { |
512 | cond_resched(); | |
513 | ||
514 | /* Select all remaining pages up to the next section boundary */ | |
515 | cur_nr_pages = | |
516 | min(end_pfn - pfn, SECTION_ALIGN_UP(pfn + 1) - pfn); | |
517 | page_init_poison(pfn_to_page(pfn), | |
518 | sizeof(struct page) * cur_nr_pages); | |
519 | } | |
d33695b1 | 520 | |
7ce700bf DH |
521 | #ifdef CONFIG_ZONE_DEVICE |
522 | /* | |
523 | * Zone shrinking code cannot properly deal with ZONE_DEVICE. So | |
524 | * we will not try to shrink the zones - which is okay as | |
525 | * set_zone_contiguous() cannot deal with ZONE_DEVICE either way. | |
526 | */ | |
527 | if (zone_idx(zone) == ZONE_DEVICE) | |
528 | return; | |
529 | #endif | |
530 | ||
feee6b29 DH |
531 | clear_zone_contiguous(zone); |
532 | ||
815121d2 YI |
533 | pgdat_resize_lock(zone->zone_pgdat, &flags); |
534 | shrink_zone_span(zone, start_pfn, start_pfn + nr_pages); | |
00d6c019 | 535 | update_pgdat_span(pgdat); |
815121d2 | 536 | pgdat_resize_unlock(zone->zone_pgdat, &flags); |
feee6b29 DH |
537 | |
538 | set_zone_contiguous(zone); | |
815121d2 YI |
539 | } |
540 | ||
feee6b29 DH |
541 | static void __remove_section(unsigned long pfn, unsigned long nr_pages, |
542 | unsigned long map_offset, | |
543 | struct vmem_altmap *altmap) | |
ea01ea93 | 544 | { |
10404901 | 545 | struct mem_section *ms = __pfn_to_section(pfn); |
ea01ea93 | 546 | |
9d1d887d DH |
547 | if (WARN_ON_ONCE(!valid_section(ms))) |
548 | return; | |
ea01ea93 | 549 | |
ba72b4c8 | 550 | sparse_remove_section(ms, pfn, nr_pages, map_offset, altmap); |
ea01ea93 BP |
551 | } |
552 | ||
ea01ea93 | 553 | /** |
feee6b29 | 554 | * __remove_pages() - remove sections of pages |
7ea62160 | 555 | * @pfn: starting pageframe (must be aligned to start of a section) |
ea01ea93 | 556 | * @nr_pages: number of pages to remove (must be multiple of section size) |
e8b098fc | 557 | * @altmap: alternative device page map or %NULL if default memmap is used |
ea01ea93 BP |
558 | * |
559 | * Generic helper function to remove section mappings and sysfs entries | |
560 | * for the section of the memory we are removing. Caller needs to make | |
561 | * sure that pages are marked reserved and zones are adjust properly by | |
562 | * calling offline_pages(). | |
563 | */ | |
feee6b29 DH |
564 | void __remove_pages(unsigned long pfn, unsigned long nr_pages, |
565 | struct vmem_altmap *altmap) | |
ea01ea93 | 566 | { |
52fb87c8 DH |
567 | const unsigned long end_pfn = pfn + nr_pages; |
568 | unsigned long cur_nr_pages; | |
4b94ffdc | 569 | unsigned long map_offset = 0; |
4b94ffdc | 570 | |
96da4350 | 571 | map_offset = vmem_altmap_offset(altmap); |
ea01ea93 | 572 | |
7ea62160 DW |
573 | if (check_pfn_span(pfn, nr_pages, "remove")) |
574 | return; | |
ea01ea93 | 575 | |
52fb87c8 | 576 | for (; pfn < end_pfn; pfn += cur_nr_pages) { |
dd33ad7b | 577 | cond_resched(); |
52fb87c8 | 578 | /* Select all remaining pages up to the next section boundary */ |
a11b9419 DH |
579 | cur_nr_pages = min(end_pfn - pfn, |
580 | SECTION_ALIGN_UP(pfn + 1) - pfn); | |
52fb87c8 | 581 | __remove_section(pfn, cur_nr_pages, map_offset, altmap); |
4b94ffdc | 582 | map_offset = 0; |
ea01ea93 | 583 | } |
ea01ea93 | 584 | } |
ea01ea93 | 585 | |
9d0ad8ca DK |
586 | int set_online_page_callback(online_page_callback_t callback) |
587 | { | |
588 | int rc = -EINVAL; | |
589 | ||
bfc8c901 VD |
590 | get_online_mems(); |
591 | mutex_lock(&online_page_callback_lock); | |
9d0ad8ca DK |
592 | |
593 | if (online_page_callback == generic_online_page) { | |
594 | online_page_callback = callback; | |
595 | rc = 0; | |
596 | } | |
597 | ||
bfc8c901 VD |
598 | mutex_unlock(&online_page_callback_lock); |
599 | put_online_mems(); | |
9d0ad8ca DK |
600 | |
601 | return rc; | |
602 | } | |
603 | EXPORT_SYMBOL_GPL(set_online_page_callback); | |
604 | ||
605 | int restore_online_page_callback(online_page_callback_t callback) | |
606 | { | |
607 | int rc = -EINVAL; | |
608 | ||
bfc8c901 VD |
609 | get_online_mems(); |
610 | mutex_lock(&online_page_callback_lock); | |
9d0ad8ca DK |
611 | |
612 | if (online_page_callback == callback) { | |
613 | online_page_callback = generic_online_page; | |
614 | rc = 0; | |
615 | } | |
616 | ||
bfc8c901 VD |
617 | mutex_unlock(&online_page_callback_lock); |
618 | put_online_mems(); | |
9d0ad8ca DK |
619 | |
620 | return rc; | |
621 | } | |
622 | EXPORT_SYMBOL_GPL(restore_online_page_callback); | |
623 | ||
18db1491 | 624 | void generic_online_page(struct page *page, unsigned int order) |
9d0ad8ca | 625 | { |
c87cbc1f VB |
626 | /* |
627 | * Freeing the page with debug_pagealloc enabled will try to unmap it, | |
628 | * so we should map it first. This is better than introducing a special | |
629 | * case in page freeing fast path. | |
630 | */ | |
77bc7fd6 | 631 | debug_pagealloc_map_pages(page, 1 << order); |
a9cd410a AK |
632 | __free_pages_core(page, order); |
633 | totalram_pages_add(1UL << order); | |
634 | #ifdef CONFIG_HIGHMEM | |
635 | if (PageHighMem(page)) | |
636 | totalhigh_pages_add(1UL << order); | |
637 | #endif | |
638 | } | |
18db1491 | 639 | EXPORT_SYMBOL_GPL(generic_online_page); |
a9cd410a | 640 | |
aac65321 | 641 | static void online_pages_range(unsigned long start_pfn, unsigned long nr_pages) |
3947be19 | 642 | { |
b2c2ab20 DH |
643 | const unsigned long end_pfn = start_pfn + nr_pages; |
644 | unsigned long pfn; | |
b2c2ab20 DH |
645 | |
646 | /* | |
aac65321 DH |
647 | * Online the pages in MAX_ORDER - 1 aligned chunks. The callback might |
648 | * decide to not expose all pages to the buddy (e.g., expose them | |
649 | * later). We account all pages as being online and belonging to this | |
650 | * zone ("present"). | |
b2c2ab20 | 651 | */ |
aac65321 DH |
652 | for (pfn = start_pfn; pfn < end_pfn; pfn += MAX_ORDER_NR_PAGES) |
653 | (*online_page_callback)(pfn_to_page(pfn), MAX_ORDER - 1); | |
2d070eab | 654 | |
b2c2ab20 DH |
655 | /* mark all involved sections as online */ |
656 | online_mem_sections(start_pfn, end_pfn); | |
75884fb1 KH |
657 | } |
658 | ||
d9713679 LJ |
659 | /* check which state of node_states will be changed when online memory */ |
660 | static void node_states_check_changes_online(unsigned long nr_pages, | |
661 | struct zone *zone, struct memory_notify *arg) | |
662 | { | |
663 | int nid = zone_to_nid(zone); | |
d9713679 | 664 | |
98fa15f3 AK |
665 | arg->status_change_nid = NUMA_NO_NODE; |
666 | arg->status_change_nid_normal = NUMA_NO_NODE; | |
667 | arg->status_change_nid_high = NUMA_NO_NODE; | |
d9713679 | 668 | |
8efe33f4 OS |
669 | if (!node_state(nid, N_MEMORY)) |
670 | arg->status_change_nid = nid; | |
671 | if (zone_idx(zone) <= ZONE_NORMAL && !node_state(nid, N_NORMAL_MEMORY)) | |
d9713679 | 672 | arg->status_change_nid_normal = nid; |
6715ddf9 | 673 | #ifdef CONFIG_HIGHMEM |
d3ba3ae1 | 674 | if (zone_idx(zone) <= ZONE_HIGHMEM && !node_state(nid, N_HIGH_MEMORY)) |
6715ddf9 | 675 | arg->status_change_nid_high = nid; |
6715ddf9 | 676 | #endif |
d9713679 LJ |
677 | } |
678 | ||
679 | static void node_states_set_node(int node, struct memory_notify *arg) | |
680 | { | |
681 | if (arg->status_change_nid_normal >= 0) | |
682 | node_set_state(node, N_NORMAL_MEMORY); | |
683 | ||
6715ddf9 LJ |
684 | if (arg->status_change_nid_high >= 0) |
685 | node_set_state(node, N_HIGH_MEMORY); | |
686 | ||
83d83612 OS |
687 | if (arg->status_change_nid >= 0) |
688 | node_set_state(node, N_MEMORY); | |
d9713679 LJ |
689 | } |
690 | ||
f1dd2cd1 MH |
691 | static void __meminit resize_zone_range(struct zone *zone, unsigned long start_pfn, |
692 | unsigned long nr_pages) | |
693 | { | |
694 | unsigned long old_end_pfn = zone_end_pfn(zone); | |
695 | ||
696 | if (zone_is_empty(zone) || start_pfn < zone->zone_start_pfn) | |
697 | zone->zone_start_pfn = start_pfn; | |
698 | ||
699 | zone->spanned_pages = max(start_pfn + nr_pages, old_end_pfn) - zone->zone_start_pfn; | |
700 | } | |
701 | ||
702 | static void __meminit resize_pgdat_range(struct pglist_data *pgdat, unsigned long start_pfn, | |
703 | unsigned long nr_pages) | |
704 | { | |
705 | unsigned long old_end_pfn = pgdat_end_pfn(pgdat); | |
706 | ||
707 | if (!pgdat->node_spanned_pages || start_pfn < pgdat->node_start_pfn) | |
708 | pgdat->node_start_pfn = start_pfn; | |
709 | ||
710 | pgdat->node_spanned_pages = max(start_pfn + nr_pages, old_end_pfn) - pgdat->node_start_pfn; | |
f1dd2cd1 | 711 | |
3fccb74c | 712 | } |
1f90a347 DW |
713 | |
714 | static void section_taint_zone_device(unsigned long pfn) | |
715 | { | |
716 | struct mem_section *ms = __pfn_to_section(pfn); | |
717 | ||
718 | ms->section_mem_map |= SECTION_TAINT_ZONE_DEVICE; | |
719 | } | |
720 | ||
3fccb74c DH |
721 | /* |
722 | * Associate the pfn range with the given zone, initializing the memmaps | |
723 | * and resizing the pgdat/zone data to span the added pages. After this | |
724 | * call, all affected pages are PG_reserved. | |
d882c006 DH |
725 | * |
726 | * All aligned pageblocks are initialized to the specified migratetype | |
727 | * (usually MIGRATE_MOVABLE). Besides setting the migratetype, no related | |
728 | * zone stats (e.g., nr_isolate_pageblock) are touched. | |
3fccb74c | 729 | */ |
a99583e7 | 730 | void __ref move_pfn_range_to_zone(struct zone *zone, unsigned long start_pfn, |
d882c006 DH |
731 | unsigned long nr_pages, |
732 | struct vmem_altmap *altmap, int migratetype) | |
f1dd2cd1 MH |
733 | { |
734 | struct pglist_data *pgdat = zone->zone_pgdat; | |
735 | int nid = pgdat->node_id; | |
736 | unsigned long flags; | |
df429ac0 | 737 | |
f1dd2cd1 MH |
738 | clear_zone_contiguous(zone); |
739 | ||
740 | /* TODO Huh pgdat is irqsave while zone is not. It used to be like that before */ | |
741 | pgdat_resize_lock(pgdat, &flags); | |
742 | zone_span_writelock(zone); | |
fa004ab7 WY |
743 | if (zone_is_empty(zone)) |
744 | init_currently_empty_zone(zone, start_pfn, nr_pages); | |
f1dd2cd1 MH |
745 | resize_zone_range(zone, start_pfn, nr_pages); |
746 | zone_span_writeunlock(zone); | |
747 | resize_pgdat_range(pgdat, start_pfn, nr_pages); | |
748 | pgdat_resize_unlock(pgdat, &flags); | |
749 | ||
1f90a347 DW |
750 | /* |
751 | * Subsection population requires care in pfn_to_online_page(). | |
752 | * Set the taint to enable the slow path detection of | |
753 | * ZONE_DEVICE pages in an otherwise ZONE_{NORMAL,MOVABLE} | |
754 | * section. | |
755 | */ | |
756 | if (zone_is_zone_device(zone)) { | |
757 | if (!IS_ALIGNED(start_pfn, PAGES_PER_SECTION)) | |
758 | section_taint_zone_device(start_pfn); | |
759 | if (!IS_ALIGNED(start_pfn + nr_pages, PAGES_PER_SECTION)) | |
760 | section_taint_zone_device(start_pfn + nr_pages); | |
761 | } | |
762 | ||
f1dd2cd1 MH |
763 | /* |
764 | * TODO now we have a visible range of pages which are not associated | |
765 | * with their zone properly. Not nice but set_pfnblock_flags_mask | |
766 | * expects the zone spans the pfn range. All the pages in the range | |
767 | * are reserved so nobody should be touching them so we should be safe | |
768 | */ | |
ab28cb6e | 769 | memmap_init_range(nr_pages, nid, zone_idx(zone), start_pfn, 0, |
d882c006 | 770 | MEMINIT_HOTPLUG, altmap, migratetype); |
f1dd2cd1 MH |
771 | |
772 | set_zone_contiguous(zone); | |
773 | } | |
774 | ||
c246a213 MH |
775 | /* |
776 | * Returns a default kernel memory zone for the given pfn range. | |
777 | * If no kernel zone covers this pfn range it will automatically go | |
778 | * to the ZONE_NORMAL. | |
779 | */ | |
c6f03e29 | 780 | static struct zone *default_kernel_zone_for_pfn(int nid, unsigned long start_pfn, |
c246a213 MH |
781 | unsigned long nr_pages) |
782 | { | |
783 | struct pglist_data *pgdat = NODE_DATA(nid); | |
784 | int zid; | |
785 | ||
786 | for (zid = 0; zid <= ZONE_NORMAL; zid++) { | |
787 | struct zone *zone = &pgdat->node_zones[zid]; | |
788 | ||
789 | if (zone_intersects(zone, start_pfn, nr_pages)) | |
790 | return zone; | |
791 | } | |
792 | ||
793 | return &pgdat->node_zones[ZONE_NORMAL]; | |
794 | } | |
795 | ||
c6f03e29 MH |
796 | static inline struct zone *default_zone_for_pfn(int nid, unsigned long start_pfn, |
797 | unsigned long nr_pages) | |
e5e68930 | 798 | { |
c6f03e29 MH |
799 | struct zone *kernel_zone = default_kernel_zone_for_pfn(nid, start_pfn, |
800 | nr_pages); | |
801 | struct zone *movable_zone = &NODE_DATA(nid)->node_zones[ZONE_MOVABLE]; | |
802 | bool in_kernel = zone_intersects(kernel_zone, start_pfn, nr_pages); | |
803 | bool in_movable = zone_intersects(movable_zone, start_pfn, nr_pages); | |
e5e68930 MH |
804 | |
805 | /* | |
c6f03e29 MH |
806 | * We inherit the existing zone in a simple case where zones do not |
807 | * overlap in the given range | |
e5e68930 | 808 | */ |
c6f03e29 MH |
809 | if (in_kernel ^ in_movable) |
810 | return (in_kernel) ? kernel_zone : movable_zone; | |
9f123ab5 | 811 | |
c6f03e29 MH |
812 | /* |
813 | * If the range doesn't belong to any zone or two zones overlap in the | |
814 | * given range then we use movable zone only if movable_node is | |
815 | * enabled because we always online to a kernel zone by default. | |
816 | */ | |
817 | return movable_node_enabled ? movable_zone : kernel_zone; | |
9f123ab5 MH |
818 | } |
819 | ||
e5e68930 MH |
820 | struct zone * zone_for_pfn_range(int online_type, int nid, unsigned start_pfn, |
821 | unsigned long nr_pages) | |
f1dd2cd1 | 822 | { |
c6f03e29 MH |
823 | if (online_type == MMOP_ONLINE_KERNEL) |
824 | return default_kernel_zone_for_pfn(nid, start_pfn, nr_pages); | |
f1dd2cd1 | 825 | |
c6f03e29 MH |
826 | if (online_type == MMOP_ONLINE_MOVABLE) |
827 | return &NODE_DATA(nid)->node_zones[ZONE_MOVABLE]; | |
df429ac0 | 828 | |
c6f03e29 | 829 | return default_zone_for_pfn(nid, start_pfn, nr_pages); |
e5e68930 MH |
830 | } |
831 | ||
bd5c2344 DH |
832 | int __ref online_pages(unsigned long pfn, unsigned long nr_pages, |
833 | int online_type, int nid) | |
75884fb1 | 834 | { |
aa47228a | 835 | unsigned long flags; |
3947be19 | 836 | struct zone *zone; |
6811378e | 837 | int need_zonelists_rebuild = 0; |
7b78d335 YG |
838 | int ret; |
839 | struct memory_notify arg; | |
d0dc12e8 | 840 | |
4986fac1 DH |
841 | /* We can only online full sections (e.g., SECTION_IS_ONLINE) */ |
842 | if (WARN_ON_ONCE(!nr_pages || | |
843 | !IS_ALIGNED(pfn | nr_pages, PAGES_PER_SECTION))) | |
844 | return -EINVAL; | |
845 | ||
381eab4a DH |
846 | mem_hotplug_begin(); |
847 | ||
f1dd2cd1 | 848 | /* associate pfn range with the zone */ |
3fccb74c | 849 | zone = zone_for_pfn_range(online_type, nid, pfn, nr_pages); |
b30c5927 | 850 | move_pfn_range_to_zone(zone, pfn, nr_pages, NULL, MIGRATE_ISOLATE); |
f1dd2cd1 | 851 | |
7b78d335 YG |
852 | arg.start_pfn = pfn; |
853 | arg.nr_pages = nr_pages; | |
d9713679 | 854 | node_states_check_changes_online(nr_pages, zone, &arg); |
7b78d335 | 855 | |
7b78d335 YG |
856 | ret = memory_notify(MEM_GOING_ONLINE, &arg); |
857 | ret = notifier_to_errno(ret); | |
e33e33b4 CY |
858 | if (ret) |
859 | goto failed_addition; | |
860 | ||
b30c5927 DH |
861 | /* |
862 | * Fixup the number of isolated pageblocks before marking the sections | |
863 | * onlining, such that undo_isolate_page_range() works correctly. | |
864 | */ | |
865 | spin_lock_irqsave(&zone->lock, flags); | |
866 | zone->nr_isolate_pageblock += nr_pages / pageblock_nr_pages; | |
867 | spin_unlock_irqrestore(&zone->lock, flags); | |
868 | ||
6811378e YG |
869 | /* |
870 | * If this zone is not populated, then it is not in zonelist. | |
871 | * This means the page allocator ignores this zone. | |
872 | * So, zonelist must be updated after online. | |
873 | */ | |
6dcd73d7 | 874 | if (!populated_zone(zone)) { |
6811378e | 875 | need_zonelists_rebuild = 1; |
72675e13 | 876 | setup_zone_pageset(zone); |
6dcd73d7 | 877 | } |
6811378e | 878 | |
aac65321 DH |
879 | online_pages_range(pfn, nr_pages); |
880 | zone->present_pages += nr_pages; | |
aa47228a CS |
881 | |
882 | pgdat_resize_lock(zone->zone_pgdat, &flags); | |
aac65321 | 883 | zone->zone_pgdat->node_present_pages += nr_pages; |
aa47228a CS |
884 | pgdat_resize_unlock(zone->zone_pgdat, &flags); |
885 | ||
b30c5927 DH |
886 | node_states_set_node(nid, &arg); |
887 | if (need_zonelists_rebuild) | |
888 | build_all_zonelists(NULL); | |
889 | zone_pcp_update(zone); | |
890 | ||
891 | /* Basic onlining is complete, allow allocation of onlined pages. */ | |
892 | undo_isolate_page_range(pfn, pfn + nr_pages, MIGRATE_MOVABLE); | |
893 | ||
93146d98 | 894 | /* |
b86c5fc4 DH |
895 | * Freshly onlined pages aren't shuffled (e.g., all pages are placed to |
896 | * the tail of the freelist when undoing isolation). Shuffle the whole | |
897 | * zone to make sure the just onlined pages are properly distributed | |
898 | * across the whole freelist - to create an initial shuffle. | |
93146d98 | 899 | */ |
e900a918 DW |
900 | shuffle_zone(zone); |
901 | ||
1b79acc9 KM |
902 | init_per_zone_wmark_min(); |
903 | ||
ca9a46f8 DH |
904 | kswapd_run(nid); |
905 | kcompactd_run(nid); | |
61b13993 | 906 | |
2d1d43f6 | 907 | writeback_set_ratelimit(); |
7b78d335 | 908 | |
ca9a46f8 | 909 | memory_notify(MEM_ONLINE, &arg); |
381eab4a | 910 | mem_hotplug_done(); |
30467e0b | 911 | return 0; |
e33e33b4 CY |
912 | |
913 | failed_addition: | |
914 | pr_debug("online_pages [mem %#010llx-%#010llx] failed\n", | |
915 | (unsigned long long) pfn << PAGE_SHIFT, | |
916 | (((unsigned long long) pfn + nr_pages) << PAGE_SHIFT) - 1); | |
917 | memory_notify(MEM_CANCEL_ONLINE, &arg); | |
feee6b29 | 918 | remove_pfn_range_from_zone(zone, pfn, nr_pages); |
381eab4a | 919 | mem_hotplug_done(); |
e33e33b4 | 920 | return ret; |
3947be19 | 921 | } |
53947027 | 922 | #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */ |
bc02af93 | 923 | |
0bd85420 TC |
924 | static void reset_node_present_pages(pg_data_t *pgdat) |
925 | { | |
926 | struct zone *z; | |
927 | ||
928 | for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++) | |
929 | z->present_pages = 0; | |
930 | ||
931 | pgdat->node_present_pages = 0; | |
932 | } | |
933 | ||
e1319331 | 934 | /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */ |
c68ab18c | 935 | static pg_data_t __ref *hotadd_new_pgdat(int nid) |
9af3c2de YG |
936 | { |
937 | struct pglist_data *pgdat; | |
9af3c2de | 938 | |
a1e565aa TC |
939 | pgdat = NODE_DATA(nid); |
940 | if (!pgdat) { | |
941 | pgdat = arch_alloc_nodedata(nid); | |
942 | if (!pgdat) | |
943 | return NULL; | |
9af3c2de | 944 | |
33fce011 WY |
945 | pgdat->per_cpu_nodestats = |
946 | alloc_percpu(struct per_cpu_nodestat); | |
a1e565aa | 947 | arch_refresh_nodedata(nid, pgdat); |
b0dc3a34 | 948 | } else { |
33fce011 | 949 | int cpu; |
e716f2eb | 950 | /* |
97a225e6 JK |
951 | * Reset the nr_zones, order and highest_zoneidx before reuse. |
952 | * Note that kswapd will init kswapd_highest_zoneidx properly | |
e716f2eb MG |
953 | * when it starts in the near future. |
954 | */ | |
b0dc3a34 | 955 | pgdat->nr_zones = 0; |
38087d9b | 956 | pgdat->kswapd_order = 0; |
97a225e6 | 957 | pgdat->kswapd_highest_zoneidx = 0; |
33fce011 WY |
958 | for_each_online_cpu(cpu) { |
959 | struct per_cpu_nodestat *p; | |
960 | ||
961 | p = per_cpu_ptr(pgdat->per_cpu_nodestats, cpu); | |
962 | memset(p, 0, sizeof(*p)); | |
963 | } | |
a1e565aa | 964 | } |
9af3c2de YG |
965 | |
966 | /* we can use NODE_DATA(nid) from here */ | |
03e85f9d | 967 | pgdat->node_id = nid; |
c68ab18c | 968 | pgdat->node_start_pfn = 0; |
03e85f9d | 969 | |
9af3c2de | 970 | /* init node's zones as empty zones, we don't have any present pages.*/ |
03e85f9d | 971 | free_area_init_core_hotplug(nid); |
9af3c2de | 972 | |
959ecc48 KH |
973 | /* |
974 | * The node we allocated has no zone fallback lists. For avoiding | |
975 | * to access not-initialized zonelist, build here. | |
976 | */ | |
72675e13 | 977 | build_all_zonelists(pgdat); |
959ecc48 | 978 | |
0bd85420 TC |
979 | /* |
980 | * When memory is hot-added, all the memory is in offline state. So | |
981 | * clear all zones' present_pages because they will be updated in | |
982 | * online_pages() and offline_pages(). | |
983 | */ | |
03e85f9d | 984 | reset_node_managed_pages(pgdat); |
0bd85420 TC |
985 | reset_node_present_pages(pgdat); |
986 | ||
9af3c2de YG |
987 | return pgdat; |
988 | } | |
989 | ||
b9ff0360 | 990 | static void rollback_node_hotadd(int nid) |
9af3c2de | 991 | { |
b9ff0360 OS |
992 | pg_data_t *pgdat = NODE_DATA(nid); |
993 | ||
9af3c2de | 994 | arch_refresh_nodedata(nid, NULL); |
5830169f | 995 | free_percpu(pgdat->per_cpu_nodestats); |
9af3c2de | 996 | arch_free_nodedata(pgdat); |
9af3c2de YG |
997 | } |
998 | ||
0a547039 | 999 | |
01b0f197 TK |
1000 | /** |
1001 | * try_online_node - online a node if offlined | |
e8b098fc | 1002 | * @nid: the node ID |
b9ff0360 | 1003 | * @set_node_online: Whether we want to online the node |
cf23422b | 1004 | * called by cpu_up() to online a node without onlined memory. |
b9ff0360 OS |
1005 | * |
1006 | * Returns: | |
1007 | * 1 -> a new node has been allocated | |
1008 | * 0 -> the node is already online | |
1009 | * -ENOMEM -> the node could not be allocated | |
cf23422b | 1010 | */ |
c68ab18c | 1011 | static int __try_online_node(int nid, bool set_node_online) |
cf23422b | 1012 | { |
b9ff0360 OS |
1013 | pg_data_t *pgdat; |
1014 | int ret = 1; | |
cf23422b | 1015 | |
01b0f197 TK |
1016 | if (node_online(nid)) |
1017 | return 0; | |
1018 | ||
c68ab18c | 1019 | pgdat = hotadd_new_pgdat(nid); |
7553e8f2 | 1020 | if (!pgdat) { |
01b0f197 | 1021 | pr_err("Cannot online node %d due to NULL pgdat\n", nid); |
cf23422b | 1022 | ret = -ENOMEM; |
1023 | goto out; | |
1024 | } | |
b9ff0360 OS |
1025 | |
1026 | if (set_node_online) { | |
1027 | node_set_online(nid); | |
1028 | ret = register_one_node(nid); | |
1029 | BUG_ON(ret); | |
1030 | } | |
cf23422b | 1031 | out: |
b9ff0360 OS |
1032 | return ret; |
1033 | } | |
1034 | ||
1035 | /* | |
1036 | * Users of this function always want to online/register the node | |
1037 | */ | |
1038 | int try_online_node(int nid) | |
1039 | { | |
1040 | int ret; | |
1041 | ||
1042 | mem_hotplug_begin(); | |
c68ab18c | 1043 | ret = __try_online_node(nid, true); |
bfc8c901 | 1044 | mem_hotplug_done(); |
cf23422b | 1045 | return ret; |
1046 | } | |
1047 | ||
27356f54 TK |
1048 | static int check_hotplug_memory_range(u64 start, u64 size) |
1049 | { | |
ba325585 | 1050 | /* memory range must be block size aligned */ |
cec3ebd0 DH |
1051 | if (!size || !IS_ALIGNED(start, memory_block_size_bytes()) || |
1052 | !IS_ALIGNED(size, memory_block_size_bytes())) { | |
ba325585 | 1053 | pr_err("Block size [%#lx] unaligned hotplug range: start %#llx, size %#llx", |
cec3ebd0 | 1054 | memory_block_size_bytes(), start, size); |
27356f54 TK |
1055 | return -EINVAL; |
1056 | } | |
1057 | ||
1058 | return 0; | |
1059 | } | |
1060 | ||
31bc3858 VK |
1061 | static int online_memory_block(struct memory_block *mem, void *arg) |
1062 | { | |
1adf8b46 | 1063 | mem->online_type = mhp_default_online_type; |
dc18d706 | 1064 | return device_online(&mem->dev); |
31bc3858 VK |
1065 | } |
1066 | ||
8df1d0e4 DH |
1067 | /* |
1068 | * NOTE: The caller must call lock_device_hotplug() to serialize hotplug | |
1069 | * and online/offline operations (triggered e.g. by sysfs). | |
1070 | * | |
1071 | * we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG | |
1072 | */ | |
b6117199 | 1073 | int __ref add_memory_resource(int nid, struct resource *res, mhp_t mhp_flags) |
bc02af93 | 1074 | { |
d15dfd31 | 1075 | struct mhp_params params = { .pgprot = pgprot_mhp(PAGE_KERNEL) }; |
62cedb9f | 1076 | u64 start, size; |
b9ff0360 | 1077 | bool new_node = false; |
bc02af93 YG |
1078 | int ret; |
1079 | ||
62cedb9f DV |
1080 | start = res->start; |
1081 | size = resource_size(res); | |
1082 | ||
27356f54 TK |
1083 | ret = check_hotplug_memory_range(start, size); |
1084 | if (ret) | |
1085 | return ret; | |
1086 | ||
fa6d9ec7 VV |
1087 | if (!node_possible(nid)) { |
1088 | WARN(1, "node %d was absent from the node_possible_map\n", nid); | |
1089 | return -EINVAL; | |
1090 | } | |
1091 | ||
bfc8c901 | 1092 | mem_hotplug_begin(); |
ac13c462 | 1093 | |
52219aea DH |
1094 | if (IS_ENABLED(CONFIG_ARCH_KEEP_MEMBLOCK)) |
1095 | memblock_add_node(start, size, nid); | |
7f36e3e5 | 1096 | |
c68ab18c | 1097 | ret = __try_online_node(nid, false); |
b9ff0360 OS |
1098 | if (ret < 0) |
1099 | goto error; | |
1100 | new_node = ret; | |
9af3c2de | 1101 | |
bc02af93 | 1102 | /* call arch's memory hotadd */ |
f5637d3b | 1103 | ret = arch_add_memory(nid, start, size, ¶ms); |
9af3c2de YG |
1104 | if (ret < 0) |
1105 | goto error; | |
1106 | ||
db051a0d DH |
1107 | /* create memory block devices after memory was added */ |
1108 | ret = create_memory_block_devices(start, size); | |
1109 | if (ret) { | |
1110 | arch_remove_memory(nid, start, size, NULL); | |
1111 | goto error; | |
1112 | } | |
1113 | ||
a1e565aa | 1114 | if (new_node) { |
d5b6f6a3 | 1115 | /* If sysfs file of new node can't be created, cpu on the node |
0fc44159 YG |
1116 | * can't be hot-added. There is no rollback way now. |
1117 | * So, check by BUG_ON() to catch it reluctantly.. | |
d5b6f6a3 | 1118 | * We online node here. We can't roll back from here. |
0fc44159 | 1119 | */ |
d5b6f6a3 OS |
1120 | node_set_online(nid); |
1121 | ret = __register_one_node(nid); | |
0fc44159 YG |
1122 | BUG_ON(ret); |
1123 | } | |
1124 | ||
d5b6f6a3 | 1125 | /* link memory sections under this node.*/ |
90c7eaeb LD |
1126 | link_mem_sections(nid, PFN_DOWN(start), PFN_UP(start + size - 1), |
1127 | MEMINIT_HOTPLUG); | |
d5b6f6a3 | 1128 | |
d96ae530 | 1129 | /* create new memmap entry */ |
7b7b2721 DH |
1130 | if (!strcmp(res->name, "System RAM")) |
1131 | firmware_map_add_hotplug(start, start + size, "System RAM"); | |
d96ae530 | 1132 | |
381eab4a DH |
1133 | /* device_online() will take the lock when calling online_pages() */ |
1134 | mem_hotplug_done(); | |
1135 | ||
9ca6551e DH |
1136 | /* |
1137 | * In case we're allowed to merge the resource, flag it and trigger | |
1138 | * merging now that adding succeeded. | |
1139 | */ | |
26011267 | 1140 | if (mhp_flags & MHP_MERGE_RESOURCE) |
9ca6551e DH |
1141 | merge_system_ram_resource(res); |
1142 | ||
31bc3858 | 1143 | /* online pages if requested */ |
1adf8b46 | 1144 | if (mhp_default_online_type != MMOP_OFFLINE) |
fbcf73ce | 1145 | walk_memory_blocks(start, size, NULL, online_memory_block); |
31bc3858 | 1146 | |
381eab4a | 1147 | return ret; |
9af3c2de YG |
1148 | error: |
1149 | /* rollback pgdat allocation and others */ | |
b9ff0360 OS |
1150 | if (new_node) |
1151 | rollback_node_hotadd(nid); | |
52219aea DH |
1152 | if (IS_ENABLED(CONFIG_ARCH_KEEP_MEMBLOCK)) |
1153 | memblock_remove(start, size); | |
bfc8c901 | 1154 | mem_hotplug_done(); |
bc02af93 YG |
1155 | return ret; |
1156 | } | |
62cedb9f | 1157 | |
8df1d0e4 | 1158 | /* requires device_hotplug_lock, see add_memory_resource() */ |
b6117199 | 1159 | int __ref __add_memory(int nid, u64 start, u64 size, mhp_t mhp_flags) |
62cedb9f DV |
1160 | { |
1161 | struct resource *res; | |
1162 | int ret; | |
1163 | ||
7b7b2721 | 1164 | res = register_memory_resource(start, size, "System RAM"); |
6f754ba4 VK |
1165 | if (IS_ERR(res)) |
1166 | return PTR_ERR(res); | |
62cedb9f | 1167 | |
b6117199 | 1168 | ret = add_memory_resource(nid, res, mhp_flags); |
62cedb9f DV |
1169 | if (ret < 0) |
1170 | release_memory_resource(res); | |
1171 | return ret; | |
1172 | } | |
8df1d0e4 | 1173 | |
b6117199 | 1174 | int add_memory(int nid, u64 start, u64 size, mhp_t mhp_flags) |
8df1d0e4 DH |
1175 | { |
1176 | int rc; | |
1177 | ||
1178 | lock_device_hotplug(); | |
b6117199 | 1179 | rc = __add_memory(nid, start, size, mhp_flags); |
8df1d0e4 DH |
1180 | unlock_device_hotplug(); |
1181 | ||
1182 | return rc; | |
1183 | } | |
bc02af93 | 1184 | EXPORT_SYMBOL_GPL(add_memory); |
0c0e6195 | 1185 | |
7b7b2721 DH |
1186 | /* |
1187 | * Add special, driver-managed memory to the system as system RAM. Such | |
1188 | * memory is not exposed via the raw firmware-provided memmap as system | |
1189 | * RAM, instead, it is detected and added by a driver - during cold boot, | |
1190 | * after a reboot, and after kexec. | |
1191 | * | |
1192 | * Reasons why this memory should not be used for the initial memmap of a | |
1193 | * kexec kernel or for placing kexec images: | |
1194 | * - The booting kernel is in charge of determining how this memory will be | |
1195 | * used (e.g., use persistent memory as system RAM) | |
1196 | * - Coordination with a hypervisor is required before this memory | |
1197 | * can be used (e.g., inaccessible parts). | |
1198 | * | |
1199 | * For this memory, no entries in /sys/firmware/memmap ("raw firmware-provided | |
1200 | * memory map") are created. Also, the created memory resource is flagged | |
7cf603d1 | 1201 | * with IORESOURCE_SYSRAM_DRIVER_MANAGED, so in-kernel users can special-case |
7b7b2721 DH |
1202 | * this memory as well (esp., not place kexec images onto it). |
1203 | * | |
1204 | * The resource_name (visible via /proc/iomem) has to have the format | |
1205 | * "System RAM ($DRIVER)". | |
1206 | */ | |
1207 | int add_memory_driver_managed(int nid, u64 start, u64 size, | |
b6117199 | 1208 | const char *resource_name, mhp_t mhp_flags) |
7b7b2721 DH |
1209 | { |
1210 | struct resource *res; | |
1211 | int rc; | |
1212 | ||
1213 | if (!resource_name || | |
1214 | strstr(resource_name, "System RAM (") != resource_name || | |
1215 | resource_name[strlen(resource_name) - 1] != ')') | |
1216 | return -EINVAL; | |
1217 | ||
1218 | lock_device_hotplug(); | |
1219 | ||
1220 | res = register_memory_resource(start, size, resource_name); | |
1221 | if (IS_ERR(res)) { | |
1222 | rc = PTR_ERR(res); | |
1223 | goto out_unlock; | |
1224 | } | |
1225 | ||
b6117199 | 1226 | rc = add_memory_resource(nid, res, mhp_flags); |
7b7b2721 DH |
1227 | if (rc < 0) |
1228 | release_memory_resource(res); | |
1229 | ||
1230 | out_unlock: | |
1231 | unlock_device_hotplug(); | |
1232 | return rc; | |
1233 | } | |
1234 | EXPORT_SYMBOL_GPL(add_memory_driver_managed); | |
1235 | ||
bca3feaa AK |
1236 | /* |
1237 | * Platforms should define arch_get_mappable_range() that provides | |
1238 | * maximum possible addressable physical memory range for which the | |
1239 | * linear mapping could be created. The platform returned address | |
1240 | * range must adhere to these following semantics. | |
1241 | * | |
1242 | * - range.start <= range.end | |
1243 | * - Range includes both end points [range.start..range.end] | |
1244 | * | |
1245 | * There is also a fallback definition provided here, allowing the | |
1246 | * entire possible physical address range in case any platform does | |
1247 | * not define arch_get_mappable_range(). | |
1248 | */ | |
1249 | struct range __weak arch_get_mappable_range(void) | |
1250 | { | |
1251 | struct range mhp_range = { | |
1252 | .start = 0UL, | |
1253 | .end = -1ULL, | |
1254 | }; | |
1255 | return mhp_range; | |
1256 | } | |
1257 | ||
1258 | struct range mhp_get_pluggable_range(bool need_mapping) | |
1259 | { | |
1260 | const u64 max_phys = (1ULL << MAX_PHYSMEM_BITS) - 1; | |
1261 | struct range mhp_range; | |
1262 | ||
1263 | if (need_mapping) { | |
1264 | mhp_range = arch_get_mappable_range(); | |
1265 | if (mhp_range.start > max_phys) { | |
1266 | mhp_range.start = 0; | |
1267 | mhp_range.end = 0; | |
1268 | } | |
1269 | mhp_range.end = min_t(u64, mhp_range.end, max_phys); | |
1270 | } else { | |
1271 | mhp_range.start = 0; | |
1272 | mhp_range.end = max_phys; | |
1273 | } | |
1274 | return mhp_range; | |
1275 | } | |
1276 | EXPORT_SYMBOL_GPL(mhp_get_pluggable_range); | |
1277 | ||
1278 | bool mhp_range_allowed(u64 start, u64 size, bool need_mapping) | |
1279 | { | |
1280 | struct range mhp_range = mhp_get_pluggable_range(need_mapping); | |
1281 | u64 end = start + size; | |
1282 | ||
1283 | if (start < end && start >= mhp_range.start && (end - 1) <= mhp_range.end) | |
1284 | return true; | |
1285 | ||
1286 | pr_warn("Hotplug memory [%#llx-%#llx] exceeds maximum addressable range [%#llx-%#llx]\n", | |
1287 | start, end, mhp_range.start, mhp_range.end); | |
1288 | return false; | |
1289 | } | |
1290 | ||
0c0e6195 KH |
1291 | #ifdef CONFIG_MEMORY_HOTREMOVE |
1292 | /* | |
92917998 DH |
1293 | * Confirm all pages in a range [start, end) belong to the same zone (skipping |
1294 | * memory holes). When true, return the zone. | |
0c0e6195 | 1295 | */ |
92917998 DH |
1296 | struct zone *test_pages_in_a_zone(unsigned long start_pfn, |
1297 | unsigned long end_pfn) | |
0c0e6195 | 1298 | { |
5f0f2887 | 1299 | unsigned long pfn, sec_end_pfn; |
0c0e6195 KH |
1300 | struct zone *zone = NULL; |
1301 | struct page *page; | |
1302 | int i; | |
deb88a2a | 1303 | for (pfn = start_pfn, sec_end_pfn = SECTION_ALIGN_UP(start_pfn + 1); |
0c0e6195 | 1304 | pfn < end_pfn; |
deb88a2a | 1305 | pfn = sec_end_pfn, sec_end_pfn += PAGES_PER_SECTION) { |
5f0f2887 AB |
1306 | /* Make sure the memory section is present first */ |
1307 | if (!present_section_nr(pfn_to_section_nr(pfn))) | |
0c0e6195 | 1308 | continue; |
5f0f2887 AB |
1309 | for (; pfn < sec_end_pfn && pfn < end_pfn; |
1310 | pfn += MAX_ORDER_NR_PAGES) { | |
1311 | i = 0; | |
1312 | /* This is just a CONFIG_HOLES_IN_ZONE check.*/ | |
1313 | while ((i < MAX_ORDER_NR_PAGES) && | |
1314 | !pfn_valid_within(pfn + i)) | |
1315 | i++; | |
d6d8c8a4 | 1316 | if (i == MAX_ORDER_NR_PAGES || pfn + i >= end_pfn) |
5f0f2887 | 1317 | continue; |
24feb47c MZ |
1318 | /* Check if we got outside of the zone */ |
1319 | if (zone && !zone_spans_pfn(zone, pfn + i)) | |
92917998 | 1320 | return NULL; |
5f0f2887 AB |
1321 | page = pfn_to_page(pfn + i); |
1322 | if (zone && page_zone(page) != zone) | |
92917998 | 1323 | return NULL; |
5f0f2887 AB |
1324 | zone = page_zone(page); |
1325 | } | |
0c0e6195 | 1326 | } |
deb88a2a | 1327 | |
92917998 | 1328 | return zone; |
0c0e6195 KH |
1329 | } |
1330 | ||
1331 | /* | |
0efadf48 | 1332 | * Scan pfn range [start,end) to find movable/migratable pages (LRU pages, |
aa218795 DH |
1333 | * non-lru movable pages and hugepages). Will skip over most unmovable |
1334 | * pages (esp., pages that can be skipped when offlining), but bail out on | |
1335 | * definitely unmovable pages. | |
1336 | * | |
1337 | * Returns: | |
1338 | * 0 in case a movable page is found and movable_pfn was updated. | |
1339 | * -ENOENT in case no movable page was found. | |
1340 | * -EBUSY in case a definitely unmovable page was found. | |
0c0e6195 | 1341 | */ |
aa218795 DH |
1342 | static int scan_movable_pages(unsigned long start, unsigned long end, |
1343 | unsigned long *movable_pfn) | |
0c0e6195 KH |
1344 | { |
1345 | unsigned long pfn; | |
eeb0efd0 | 1346 | |
0c0e6195 | 1347 | for (pfn = start; pfn < end; pfn++) { |
eeb0efd0 OS |
1348 | struct page *page, *head; |
1349 | unsigned long skip; | |
1350 | ||
1351 | if (!pfn_valid(pfn)) | |
1352 | continue; | |
1353 | page = pfn_to_page(pfn); | |
1354 | if (PageLRU(page)) | |
aa218795 | 1355 | goto found; |
eeb0efd0 | 1356 | if (__PageMovable(page)) |
aa218795 DH |
1357 | goto found; |
1358 | ||
1359 | /* | |
1360 | * PageOffline() pages that are not marked __PageMovable() and | |
1361 | * have a reference count > 0 (after MEM_GOING_OFFLINE) are | |
1362 | * definitely unmovable. If their reference count would be 0, | |
1363 | * they could at least be skipped when offlining memory. | |
1364 | */ | |
1365 | if (PageOffline(page) && page_count(page)) | |
1366 | return -EBUSY; | |
eeb0efd0 OS |
1367 | |
1368 | if (!PageHuge(page)) | |
1369 | continue; | |
1370 | head = compound_head(page); | |
8f251a3d MK |
1371 | /* |
1372 | * This test is racy as we hold no reference or lock. The | |
1373 | * hugetlb page could have been free'ed and head is no longer | |
1374 | * a hugetlb page before the following check. In such unlikely | |
1375 | * cases false positives and negatives are possible. Calling | |
1376 | * code must deal with these scenarios. | |
1377 | */ | |
1378 | if (HPageMigratable(head)) | |
aa218795 | 1379 | goto found; |
d8c6546b | 1380 | skip = compound_nr(head) - (page - head); |
eeb0efd0 | 1381 | pfn += skip - 1; |
0c0e6195 | 1382 | } |
aa218795 DH |
1383 | return -ENOENT; |
1384 | found: | |
1385 | *movable_pfn = pfn; | |
0c0e6195 KH |
1386 | return 0; |
1387 | } | |
1388 | ||
0c0e6195 KH |
1389 | static int |
1390 | do_migrate_range(unsigned long start_pfn, unsigned long end_pfn) | |
1391 | { | |
1392 | unsigned long pfn; | |
6c357848 | 1393 | struct page *page, *head; |
0c0e6195 KH |
1394 | int ret = 0; |
1395 | LIST_HEAD(source); | |
1396 | ||
a85009c3 | 1397 | for (pfn = start_pfn; pfn < end_pfn; pfn++) { |
0c0e6195 KH |
1398 | if (!pfn_valid(pfn)) |
1399 | continue; | |
1400 | page = pfn_to_page(pfn); | |
6c357848 | 1401 | head = compound_head(page); |
c8721bbb NH |
1402 | |
1403 | if (PageHuge(page)) { | |
d8c6546b | 1404 | pfn = page_to_pfn(head) + compound_nr(head) - 1; |
daf3538a | 1405 | isolate_huge_page(head, &source); |
c8721bbb | 1406 | continue; |
94723aaf | 1407 | } else if (PageTransHuge(page)) |
6c357848 | 1408 | pfn = page_to_pfn(head) + thp_nr_pages(page) - 1; |
c8721bbb | 1409 | |
b15c8726 MH |
1410 | /* |
1411 | * HWPoison pages have elevated reference counts so the migration would | |
1412 | * fail on them. It also doesn't make any sense to migrate them in the | |
1413 | * first place. Still try to unmap such a page in case it is still mapped | |
1414 | * (e.g. current hwpoison implementation doesn't unmap KSM pages but keep | |
1415 | * the unmap as the catch all safety net). | |
1416 | */ | |
1417 | if (PageHWPoison(page)) { | |
1418 | if (WARN_ON(PageLRU(page))) | |
1419 | isolate_lru_page(page); | |
1420 | if (page_mapped(page)) | |
013339df | 1421 | try_to_unmap(page, TTU_IGNORE_MLOCK); |
b15c8726 MH |
1422 | continue; |
1423 | } | |
1424 | ||
700c2a46 | 1425 | if (!get_page_unless_zero(page)) |
0c0e6195 KH |
1426 | continue; |
1427 | /* | |
0efadf48 YX |
1428 | * We can skip free pages. And we can deal with pages on |
1429 | * LRU and non-lru movable pages. | |
0c0e6195 | 1430 | */ |
0efadf48 YX |
1431 | if (PageLRU(page)) |
1432 | ret = isolate_lru_page(page); | |
1433 | else | |
1434 | ret = isolate_movable_page(page, ISOLATE_UNEVICTABLE); | |
0c0e6195 | 1435 | if (!ret) { /* Success */ |
62695a84 | 1436 | list_add_tail(&page->lru, &source); |
0efadf48 YX |
1437 | if (!__PageMovable(page)) |
1438 | inc_node_page_state(page, NR_ISOLATED_ANON + | |
9de4f22a | 1439 | page_is_file_lru(page)); |
6d9c285a | 1440 | |
0c0e6195 | 1441 | } else { |
2932c8b0 | 1442 | pr_warn("failed to isolate pfn %lx\n", pfn); |
0efadf48 | 1443 | dump_page(page, "isolation failed"); |
0c0e6195 | 1444 | } |
1723058e | 1445 | put_page(page); |
0c0e6195 | 1446 | } |
f3ab2636 | 1447 | if (!list_empty(&source)) { |
203e6e5c JK |
1448 | nodemask_t nmask = node_states[N_MEMORY]; |
1449 | struct migration_target_control mtc = { | |
1450 | .nmask = &nmask, | |
1451 | .gfp_mask = GFP_USER | __GFP_MOVABLE | __GFP_RETRY_MAYFAIL, | |
1452 | }; | |
1453 | ||
1454 | /* | |
1455 | * We have checked that migration range is on a single zone so | |
1456 | * we can use the nid of the first page to all the others. | |
1457 | */ | |
1458 | mtc.nid = page_to_nid(list_first_entry(&source, struct page, lru)); | |
1459 | ||
1460 | /* | |
1461 | * try to allocate from a different node but reuse this node | |
1462 | * if there are no other online nodes to be used (e.g. we are | |
1463 | * offlining a part of the only existing node) | |
1464 | */ | |
1465 | node_clear(mtc.nid, nmask); | |
1466 | if (nodes_empty(nmask)) | |
1467 | node_set(mtc.nid, nmask); | |
1468 | ret = migrate_pages(&source, alloc_migration_target, NULL, | |
1469 | (unsigned long)&mtc, MIGRATE_SYNC, MR_MEMORY_HOTPLUG); | |
2932c8b0 MH |
1470 | if (ret) { |
1471 | list_for_each_entry(page, &source, lru) { | |
1472 | pr_warn("migrating pfn %lx failed ret:%d ", | |
1473 | page_to_pfn(page), ret); | |
1474 | dump_page(page, "migration failure"); | |
1475 | } | |
c8721bbb | 1476 | putback_movable_pages(&source); |
2932c8b0 | 1477 | } |
0c0e6195 | 1478 | } |
1723058e | 1479 | |
0c0e6195 KH |
1480 | return ret; |
1481 | } | |
1482 | ||
c5320926 TC |
1483 | static int __init cmdline_parse_movable_node(char *p) |
1484 | { | |
55ac590c | 1485 | movable_node_enabled = true; |
c5320926 TC |
1486 | return 0; |
1487 | } | |
1488 | early_param("movable_node", cmdline_parse_movable_node); | |
1489 | ||
d9713679 LJ |
1490 | /* check which state of node_states will be changed when offline memory */ |
1491 | static void node_states_check_changes_offline(unsigned long nr_pages, | |
1492 | struct zone *zone, struct memory_notify *arg) | |
1493 | { | |
1494 | struct pglist_data *pgdat = zone->zone_pgdat; | |
1495 | unsigned long present_pages = 0; | |
86b27bea | 1496 | enum zone_type zt; |
d9713679 | 1497 | |
98fa15f3 AK |
1498 | arg->status_change_nid = NUMA_NO_NODE; |
1499 | arg->status_change_nid_normal = NUMA_NO_NODE; | |
1500 | arg->status_change_nid_high = NUMA_NO_NODE; | |
d9713679 LJ |
1501 | |
1502 | /* | |
86b27bea OS |
1503 | * Check whether node_states[N_NORMAL_MEMORY] will be changed. |
1504 | * If the memory to be offline is within the range | |
1505 | * [0..ZONE_NORMAL], and it is the last present memory there, | |
1506 | * the zones in that range will become empty after the offlining, | |
1507 | * thus we can determine that we need to clear the node from | |
1508 | * node_states[N_NORMAL_MEMORY]. | |
d9713679 | 1509 | */ |
86b27bea | 1510 | for (zt = 0; zt <= ZONE_NORMAL; zt++) |
d9713679 | 1511 | present_pages += pgdat->node_zones[zt].present_pages; |
86b27bea | 1512 | if (zone_idx(zone) <= ZONE_NORMAL && nr_pages >= present_pages) |
d9713679 | 1513 | arg->status_change_nid_normal = zone_to_nid(zone); |
d9713679 | 1514 | |
6715ddf9 LJ |
1515 | #ifdef CONFIG_HIGHMEM |
1516 | /* | |
86b27bea OS |
1517 | * node_states[N_HIGH_MEMORY] contains nodes which |
1518 | * have normal memory or high memory. | |
1519 | * Here we add the present_pages belonging to ZONE_HIGHMEM. | |
1520 | * If the zone is within the range of [0..ZONE_HIGHMEM), and | |
1521 | * we determine that the zones in that range become empty, | |
1522 | * we need to clear the node for N_HIGH_MEMORY. | |
6715ddf9 | 1523 | */ |
86b27bea OS |
1524 | present_pages += pgdat->node_zones[ZONE_HIGHMEM].present_pages; |
1525 | if (zone_idx(zone) <= ZONE_HIGHMEM && nr_pages >= present_pages) | |
6715ddf9 | 1526 | arg->status_change_nid_high = zone_to_nid(zone); |
6715ddf9 LJ |
1527 | #endif |
1528 | ||
d9713679 | 1529 | /* |
86b27bea OS |
1530 | * We have accounted the pages from [0..ZONE_NORMAL), and |
1531 | * in case of CONFIG_HIGHMEM the pages from ZONE_HIGHMEM | |
1532 | * as well. | |
1533 | * Here we count the possible pages from ZONE_MOVABLE. | |
1534 | * If after having accounted all the pages, we see that the nr_pages | |
1535 | * to be offlined is over or equal to the accounted pages, | |
1536 | * we know that the node will become empty, and so, we can clear | |
1537 | * it for N_MEMORY as well. | |
d9713679 | 1538 | */ |
86b27bea | 1539 | present_pages += pgdat->node_zones[ZONE_MOVABLE].present_pages; |
d9713679 | 1540 | |
d9713679 LJ |
1541 | if (nr_pages >= present_pages) |
1542 | arg->status_change_nid = zone_to_nid(zone); | |
d9713679 LJ |
1543 | } |
1544 | ||
1545 | static void node_states_clear_node(int node, struct memory_notify *arg) | |
1546 | { | |
1547 | if (arg->status_change_nid_normal >= 0) | |
1548 | node_clear_state(node, N_NORMAL_MEMORY); | |
1549 | ||
cf01f6f5 | 1550 | if (arg->status_change_nid_high >= 0) |
d9713679 | 1551 | node_clear_state(node, N_HIGH_MEMORY); |
6715ddf9 | 1552 | |
cf01f6f5 | 1553 | if (arg->status_change_nid >= 0) |
6715ddf9 | 1554 | node_clear_state(node, N_MEMORY); |
d9713679 LJ |
1555 | } |
1556 | ||
c5e79ef5 DH |
1557 | static int count_system_ram_pages_cb(unsigned long start_pfn, |
1558 | unsigned long nr_pages, void *data) | |
1559 | { | |
1560 | unsigned long *nr_system_ram_pages = data; | |
1561 | ||
1562 | *nr_system_ram_pages += nr_pages; | |
1563 | return 0; | |
1564 | } | |
1565 | ||
73a11c96 | 1566 | int __ref offline_pages(unsigned long start_pfn, unsigned long nr_pages) |
0c0e6195 | 1567 | { |
73a11c96 | 1568 | const unsigned long end_pfn = start_pfn + nr_pages; |
0a1a9a00 | 1569 | unsigned long pfn, system_ram_pages = 0; |
d702909f | 1570 | unsigned long flags; |
0c0e6195 | 1571 | struct zone *zone; |
7b78d335 | 1572 | struct memory_notify arg; |
ea15153c | 1573 | int ret, node; |
79605093 | 1574 | char *reason; |
0c0e6195 | 1575 | |
4986fac1 DH |
1576 | /* We can only offline full sections (e.g., SECTION_IS_ONLINE) */ |
1577 | if (WARN_ON_ONCE(!nr_pages || | |
1578 | !IS_ALIGNED(start_pfn | nr_pages, PAGES_PER_SECTION))) | |
1579 | return -EINVAL; | |
1580 | ||
381eab4a DH |
1581 | mem_hotplug_begin(); |
1582 | ||
c5e79ef5 DH |
1583 | /* |
1584 | * Don't allow to offline memory blocks that contain holes. | |
1585 | * Consequently, memory blocks with holes can never get onlined | |
1586 | * via the hotplug path - online_pages() - as hotplugged memory has | |
1587 | * no holes. This way, we e.g., don't have to worry about marking | |
1588 | * memory holes PG_reserved, don't need pfn_valid() checks, and can | |
1589 | * avoid using walk_system_ram_range() later. | |
1590 | */ | |
73a11c96 | 1591 | walk_system_ram_range(start_pfn, nr_pages, &system_ram_pages, |
c5e79ef5 | 1592 | count_system_ram_pages_cb); |
73a11c96 | 1593 | if (system_ram_pages != nr_pages) { |
c5e79ef5 DH |
1594 | ret = -EINVAL; |
1595 | reason = "memory holes"; | |
1596 | goto failed_removal; | |
1597 | } | |
1598 | ||
0c0e6195 KH |
1599 | /* This makes hotplug much easier...and readable. |
1600 | we assume this for now. .*/ | |
92917998 DH |
1601 | zone = test_pages_in_a_zone(start_pfn, end_pfn); |
1602 | if (!zone) { | |
79605093 MH |
1603 | ret = -EINVAL; |
1604 | reason = "multizone range"; | |
1605 | goto failed_removal; | |
381eab4a | 1606 | } |
7b78d335 | 1607 | node = zone_to_nid(zone); |
7b78d335 | 1608 | |
ec6e8c7e VB |
1609 | /* |
1610 | * Disable pcplists so that page isolation cannot race with freeing | |
1611 | * in a way that pages from isolated pageblock are left on pcplists. | |
1612 | */ | |
1613 | zone_pcp_disable(zone); | |
1614 | ||
0c0e6195 | 1615 | /* set above range as isolated */ |
b023f468 | 1616 | ret = start_isolate_page_range(start_pfn, end_pfn, |
d381c547 | 1617 | MIGRATE_MOVABLE, |
756d25be | 1618 | MEMORY_OFFLINE | REPORT_FAILURE); |
3fa0c7c7 | 1619 | if (ret) { |
79605093 | 1620 | reason = "failure to isolate range"; |
ec6e8c7e | 1621 | goto failed_removal_pcplists_disabled; |
381eab4a | 1622 | } |
7b78d335 YG |
1623 | |
1624 | arg.start_pfn = start_pfn; | |
1625 | arg.nr_pages = nr_pages; | |
d9713679 | 1626 | node_states_check_changes_offline(nr_pages, zone, &arg); |
7b78d335 YG |
1627 | |
1628 | ret = memory_notify(MEM_GOING_OFFLINE, &arg); | |
1629 | ret = notifier_to_errno(ret); | |
79605093 MH |
1630 | if (ret) { |
1631 | reason = "notifier failure"; | |
1632 | goto failed_removal_isolated; | |
1633 | } | |
7b78d335 | 1634 | |
bb8965bd | 1635 | do { |
aa218795 DH |
1636 | pfn = start_pfn; |
1637 | do { | |
bb8965bd MH |
1638 | if (signal_pending(current)) { |
1639 | ret = -EINTR; | |
1640 | reason = "signal backoff"; | |
1641 | goto failed_removal_isolated; | |
1642 | } | |
72b39cfc | 1643 | |
bb8965bd MH |
1644 | cond_resched(); |
1645 | lru_add_drain_all(); | |
bb8965bd | 1646 | |
aa218795 DH |
1647 | ret = scan_movable_pages(pfn, end_pfn, &pfn); |
1648 | if (!ret) { | |
bb8965bd MH |
1649 | /* |
1650 | * TODO: fatal migration failures should bail | |
1651 | * out | |
1652 | */ | |
1653 | do_migrate_range(pfn, end_pfn); | |
1654 | } | |
aa218795 DH |
1655 | } while (!ret); |
1656 | ||
1657 | if (ret != -ENOENT) { | |
1658 | reason = "unmovable page"; | |
1659 | goto failed_removal_isolated; | |
bb8965bd | 1660 | } |
0c0e6195 | 1661 | |
bb8965bd MH |
1662 | /* |
1663 | * Dissolve free hugepages in the memory block before doing | |
1664 | * offlining actually in order to make hugetlbfs's object | |
1665 | * counting consistent. | |
1666 | */ | |
1667 | ret = dissolve_free_huge_pages(start_pfn, end_pfn); | |
1668 | if (ret) { | |
1669 | reason = "failure to dissolve huge pages"; | |
1670 | goto failed_removal_isolated; | |
1671 | } | |
0a1a9a00 | 1672 | |
0a1a9a00 | 1673 | ret = test_pages_isolated(start_pfn, end_pfn, MEMORY_OFFLINE); |
ec6e8c7e | 1674 | |
5557c766 | 1675 | } while (ret); |
72b39cfc | 1676 | |
0a1a9a00 DH |
1677 | /* Mark all sections offline and remove free pages from the buddy. */ |
1678 | __offline_isolated_pages(start_pfn, end_pfn); | |
7c33023a | 1679 | pr_debug("Offlined Pages %ld\n", nr_pages); |
0a1a9a00 | 1680 | |
9b7ea46a | 1681 | /* |
b30c5927 DH |
1682 | * The memory sections are marked offline, and the pageblock flags |
1683 | * effectively stale; nobody should be touching them. Fixup the number | |
1684 | * of isolated pageblocks, memory onlining will properly revert this. | |
9b7ea46a QC |
1685 | */ |
1686 | spin_lock_irqsave(&zone->lock, flags); | |
ea15153c | 1687 | zone->nr_isolate_pageblock -= nr_pages / pageblock_nr_pages; |
9b7ea46a QC |
1688 | spin_unlock_irqrestore(&zone->lock, flags); |
1689 | ||
ec6e8c7e VB |
1690 | zone_pcp_enable(zone); |
1691 | ||
0c0e6195 | 1692 | /* removal success */ |
0a1a9a00 DH |
1693 | adjust_managed_page_count(pfn_to_page(start_pfn), -nr_pages); |
1694 | zone->present_pages -= nr_pages; | |
d702909f CS |
1695 | |
1696 | pgdat_resize_lock(zone->zone_pgdat, &flags); | |
0a1a9a00 | 1697 | zone->zone_pgdat->node_present_pages -= nr_pages; |
d702909f | 1698 | pgdat_resize_unlock(zone->zone_pgdat, &flags); |
7b78d335 | 1699 | |
1b79acc9 KM |
1700 | init_per_zone_wmark_min(); |
1701 | ||
1e8537ba | 1702 | if (!populated_zone(zone)) { |
340175b7 | 1703 | zone_pcp_reset(zone); |
72675e13 | 1704 | build_all_zonelists(NULL); |
1e8537ba XQ |
1705 | } else |
1706 | zone_pcp_update(zone); | |
340175b7 | 1707 | |
d9713679 | 1708 | node_states_clear_node(node, &arg); |
698b1b30 | 1709 | if (arg.status_change_nid >= 0) { |
8fe23e05 | 1710 | kswapd_stop(node); |
698b1b30 VB |
1711 | kcompactd_stop(node); |
1712 | } | |
bce7394a | 1713 | |
0c0e6195 | 1714 | writeback_set_ratelimit(); |
7b78d335 YG |
1715 | |
1716 | memory_notify(MEM_OFFLINE, &arg); | |
feee6b29 | 1717 | remove_pfn_range_from_zone(zone, start_pfn, nr_pages); |
381eab4a | 1718 | mem_hotplug_done(); |
0c0e6195 KH |
1719 | return 0; |
1720 | ||
79605093 MH |
1721 | failed_removal_isolated: |
1722 | undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE); | |
c4efe484 | 1723 | memory_notify(MEM_CANCEL_OFFLINE, &arg); |
ec6e8c7e VB |
1724 | failed_removal_pcplists_disabled: |
1725 | zone_pcp_enable(zone); | |
0c0e6195 | 1726 | failed_removal: |
79605093 | 1727 | pr_debug("memory offlining [mem %#010llx-%#010llx] failed due to %s\n", |
e33e33b4 | 1728 | (unsigned long long) start_pfn << PAGE_SHIFT, |
79605093 MH |
1729 | ((unsigned long long) end_pfn << PAGE_SHIFT) - 1, |
1730 | reason); | |
0c0e6195 | 1731 | /* pushback to free area */ |
381eab4a | 1732 | mem_hotplug_done(); |
0c0e6195 KH |
1733 | return ret; |
1734 | } | |
71088785 | 1735 | |
d6de9d53 | 1736 | static int check_memblock_offlined_cb(struct memory_block *mem, void *arg) |
bbc76be6 WC |
1737 | { |
1738 | int ret = !is_memblock_offlined(mem); | |
1739 | ||
349daa0f RD |
1740 | if (unlikely(ret)) { |
1741 | phys_addr_t beginpa, endpa; | |
1742 | ||
1743 | beginpa = PFN_PHYS(section_nr_to_pfn(mem->start_section_nr)); | |
b6c88d3b | 1744 | endpa = beginpa + memory_block_size_bytes() - 1; |
756a025f | 1745 | pr_warn("removing memory fails, because memory [%pa-%pa] is onlined\n", |
349daa0f | 1746 | &beginpa, &endpa); |
bbc76be6 | 1747 | |
eca499ab PT |
1748 | return -EBUSY; |
1749 | } | |
1750 | return 0; | |
bbc76be6 WC |
1751 | } |
1752 | ||
0f1cfe9d | 1753 | static int check_cpu_on_node(pg_data_t *pgdat) |
60a5a19e | 1754 | { |
60a5a19e TC |
1755 | int cpu; |
1756 | ||
1757 | for_each_present_cpu(cpu) { | |
1758 | if (cpu_to_node(cpu) == pgdat->node_id) | |
1759 | /* | |
1760 | * the cpu on this node isn't removed, and we can't | |
1761 | * offline this node. | |
1762 | */ | |
1763 | return -EBUSY; | |
1764 | } | |
1765 | ||
1766 | return 0; | |
1767 | } | |
1768 | ||
2c91f8fc DH |
1769 | static int check_no_memblock_for_node_cb(struct memory_block *mem, void *arg) |
1770 | { | |
1771 | int nid = *(int *)arg; | |
1772 | ||
1773 | /* | |
1774 | * If a memory block belongs to multiple nodes, the stored nid is not | |
1775 | * reliable. However, such blocks are always online (e.g., cannot get | |
1776 | * offlined) and, therefore, are still spanned by the node. | |
1777 | */ | |
1778 | return mem->nid == nid ? -EEXIST : 0; | |
1779 | } | |
1780 | ||
0f1cfe9d TK |
1781 | /** |
1782 | * try_offline_node | |
e8b098fc | 1783 | * @nid: the node ID |
0f1cfe9d TK |
1784 | * |
1785 | * Offline a node if all memory sections and cpus of the node are removed. | |
1786 | * | |
1787 | * NOTE: The caller must call lock_device_hotplug() to serialize hotplug | |
1788 | * and online/offline operations before this call. | |
1789 | */ | |
90b30cdc | 1790 | void try_offline_node(int nid) |
60a5a19e | 1791 | { |
d822b86a | 1792 | pg_data_t *pgdat = NODE_DATA(nid); |
2c91f8fc | 1793 | int rc; |
60a5a19e | 1794 | |
2c91f8fc DH |
1795 | /* |
1796 | * If the node still spans pages (especially ZONE_DEVICE), don't | |
1797 | * offline it. A node spans memory after move_pfn_range_to_zone(), | |
1798 | * e.g., after the memory block was onlined. | |
1799 | */ | |
1800 | if (pgdat->node_spanned_pages) | |
1801 | return; | |
60a5a19e | 1802 | |
2c91f8fc DH |
1803 | /* |
1804 | * Especially offline memory blocks might not be spanned by the | |
1805 | * node. They will get spanned by the node once they get onlined. | |
1806 | * However, they link to the node in sysfs and can get onlined later. | |
1807 | */ | |
1808 | rc = for_each_memory_block(&nid, check_no_memblock_for_node_cb); | |
1809 | if (rc) | |
60a5a19e | 1810 | return; |
60a5a19e | 1811 | |
46a3679b | 1812 | if (check_cpu_on_node(pgdat)) |
60a5a19e TC |
1813 | return; |
1814 | ||
1815 | /* | |
1816 | * all memory/cpu of this node are removed, we can offline this | |
1817 | * node now. | |
1818 | */ | |
1819 | node_set_offline(nid); | |
1820 | unregister_one_node(nid); | |
1821 | } | |
90b30cdc | 1822 | EXPORT_SYMBOL(try_offline_node); |
60a5a19e | 1823 | |
eca499ab | 1824 | static int __ref try_remove_memory(int nid, u64 start, u64 size) |
bbc76be6 | 1825 | { |
eca499ab | 1826 | int rc = 0; |
993c1aad | 1827 | |
27356f54 TK |
1828 | BUG_ON(check_hotplug_memory_range(start, size)); |
1829 | ||
6677e3ea | 1830 | /* |
242831eb | 1831 | * All memory blocks must be offlined before removing memory. Check |
eca499ab | 1832 | * whether all memory blocks in question are offline and return error |
242831eb | 1833 | * if this is not the case. |
6677e3ea | 1834 | */ |
fbcf73ce | 1835 | rc = walk_memory_blocks(start, size, NULL, check_memblock_offlined_cb); |
eca499ab | 1836 | if (rc) |
b4223a51 | 1837 | return rc; |
6677e3ea | 1838 | |
46c66c4b YI |
1839 | /* remove memmap entry */ |
1840 | firmware_map_remove(start, start + size, "System RAM"); | |
4c4b7f9b | 1841 | |
f1037ec0 DW |
1842 | /* |
1843 | * Memory block device removal under the device_hotplug_lock is | |
1844 | * a barrier against racing online attempts. | |
1845 | */ | |
4c4b7f9b | 1846 | remove_memory_block_devices(start, size); |
46c66c4b | 1847 | |
f1037ec0 DW |
1848 | mem_hotplug_begin(); |
1849 | ||
2c2a5af6 | 1850 | arch_remove_memory(nid, start, size, NULL); |
52219aea DH |
1851 | |
1852 | if (IS_ENABLED(CONFIG_ARCH_KEEP_MEMBLOCK)) { | |
1853 | memblock_free(start, size); | |
1854 | memblock_remove(start, size); | |
1855 | } | |
1856 | ||
cb8e3c8b | 1857 | release_mem_region_adjustable(start, size); |
24d335ca | 1858 | |
60a5a19e TC |
1859 | try_offline_node(nid); |
1860 | ||
bfc8c901 | 1861 | mem_hotplug_done(); |
b4223a51 | 1862 | return 0; |
71088785 | 1863 | } |
d15e5926 | 1864 | |
eca499ab PT |
1865 | /** |
1866 | * remove_memory | |
1867 | * @nid: the node ID | |
1868 | * @start: physical address of the region to remove | |
1869 | * @size: size of the region to remove | |
1870 | * | |
1871 | * NOTE: The caller must call lock_device_hotplug() to serialize hotplug | |
1872 | * and online/offline operations before this call, as required by | |
1873 | * try_offline_node(). | |
1874 | */ | |
1875 | void __remove_memory(int nid, u64 start, u64 size) | |
1876 | { | |
1877 | ||
1878 | /* | |
29a90db9 | 1879 | * trigger BUG() if some memory is not offlined prior to calling this |
eca499ab PT |
1880 | * function |
1881 | */ | |
1882 | if (try_remove_memory(nid, start, size)) | |
1883 | BUG(); | |
1884 | } | |
1885 | ||
1886 | /* | |
1887 | * Remove memory if every memory block is offline, otherwise return -EBUSY is | |
1888 | * some memory is not offline | |
1889 | */ | |
1890 | int remove_memory(int nid, u64 start, u64 size) | |
d15e5926 | 1891 | { |
eca499ab PT |
1892 | int rc; |
1893 | ||
d15e5926 | 1894 | lock_device_hotplug(); |
eca499ab | 1895 | rc = try_remove_memory(nid, start, size); |
d15e5926 | 1896 | unlock_device_hotplug(); |
eca499ab PT |
1897 | |
1898 | return rc; | |
d15e5926 | 1899 | } |
71088785 | 1900 | EXPORT_SYMBOL_GPL(remove_memory); |
08b3acd7 | 1901 | |
8dc4bb58 DH |
1902 | static int try_offline_memory_block(struct memory_block *mem, void *arg) |
1903 | { | |
1904 | uint8_t online_type = MMOP_ONLINE_KERNEL; | |
1905 | uint8_t **online_types = arg; | |
1906 | struct page *page; | |
1907 | int rc; | |
1908 | ||
1909 | /* | |
1910 | * Sense the online_type via the zone of the memory block. Offlining | |
1911 | * with multiple zones within one memory block will be rejected | |
1912 | * by offlining code ... so we don't care about that. | |
1913 | */ | |
1914 | page = pfn_to_online_page(section_nr_to_pfn(mem->start_section_nr)); | |
1915 | if (page && zone_idx(page_zone(page)) == ZONE_MOVABLE) | |
1916 | online_type = MMOP_ONLINE_MOVABLE; | |
1917 | ||
1918 | rc = device_offline(&mem->dev); | |
1919 | /* | |
1920 | * Default is MMOP_OFFLINE - change it only if offlining succeeded, | |
1921 | * so try_reonline_memory_block() can do the right thing. | |
1922 | */ | |
1923 | if (!rc) | |
1924 | **online_types = online_type; | |
1925 | ||
1926 | (*online_types)++; | |
1927 | /* Ignore if already offline. */ | |
1928 | return rc < 0 ? rc : 0; | |
1929 | } | |
1930 | ||
1931 | static int try_reonline_memory_block(struct memory_block *mem, void *arg) | |
1932 | { | |
1933 | uint8_t **online_types = arg; | |
1934 | int rc; | |
1935 | ||
1936 | if (**online_types != MMOP_OFFLINE) { | |
1937 | mem->online_type = **online_types; | |
1938 | rc = device_online(&mem->dev); | |
1939 | if (rc < 0) | |
1940 | pr_warn("%s: Failed to re-online memory: %d", | |
1941 | __func__, rc); | |
1942 | } | |
1943 | ||
1944 | /* Continue processing all remaining memory blocks. */ | |
1945 | (*online_types)++; | |
1946 | return 0; | |
1947 | } | |
1948 | ||
08b3acd7 | 1949 | /* |
8dc4bb58 DH |
1950 | * Try to offline and remove memory. Might take a long time to finish in case |
1951 | * memory is still in use. Primarily useful for memory devices that logically | |
1952 | * unplugged all memory (so it's no longer in use) and want to offline + remove | |
1953 | * that memory. | |
08b3acd7 DH |
1954 | */ |
1955 | int offline_and_remove_memory(int nid, u64 start, u64 size) | |
1956 | { | |
8dc4bb58 DH |
1957 | const unsigned long mb_count = size / memory_block_size_bytes(); |
1958 | uint8_t *online_types, *tmp; | |
1959 | int rc; | |
08b3acd7 DH |
1960 | |
1961 | if (!IS_ALIGNED(start, memory_block_size_bytes()) || | |
8dc4bb58 DH |
1962 | !IS_ALIGNED(size, memory_block_size_bytes()) || !size) |
1963 | return -EINVAL; | |
1964 | ||
1965 | /* | |
1966 | * We'll remember the old online type of each memory block, so we can | |
1967 | * try to revert whatever we did when offlining one memory block fails | |
1968 | * after offlining some others succeeded. | |
1969 | */ | |
1970 | online_types = kmalloc_array(mb_count, sizeof(*online_types), | |
1971 | GFP_KERNEL); | |
1972 | if (!online_types) | |
1973 | return -ENOMEM; | |
1974 | /* | |
1975 | * Initialize all states to MMOP_OFFLINE, so when we abort processing in | |
1976 | * try_offline_memory_block(), we'll skip all unprocessed blocks in | |
1977 | * try_reonline_memory_block(). | |
1978 | */ | |
1979 | memset(online_types, MMOP_OFFLINE, mb_count); | |
08b3acd7 DH |
1980 | |
1981 | lock_device_hotplug(); | |
8dc4bb58 DH |
1982 | |
1983 | tmp = online_types; | |
1984 | rc = walk_memory_blocks(start, size, &tmp, try_offline_memory_block); | |
08b3acd7 DH |
1985 | |
1986 | /* | |
8dc4bb58 | 1987 | * In case we succeeded to offline all memory, remove it. |
08b3acd7 DH |
1988 | * This cannot fail as it cannot get onlined in the meantime. |
1989 | */ | |
1990 | if (!rc) { | |
1991 | rc = try_remove_memory(nid, start, size); | |
8dc4bb58 DH |
1992 | if (rc) |
1993 | pr_err("%s: Failed to remove memory: %d", __func__, rc); | |
1994 | } | |
1995 | ||
1996 | /* | |
1997 | * Rollback what we did. While memory onlining might theoretically fail | |
1998 | * (nacked by a notifier), it barely ever happens. | |
1999 | */ | |
2000 | if (rc) { | |
2001 | tmp = online_types; | |
2002 | walk_memory_blocks(start, size, &tmp, | |
2003 | try_reonline_memory_block); | |
08b3acd7 DH |
2004 | } |
2005 | unlock_device_hotplug(); | |
2006 | ||
8dc4bb58 | 2007 | kfree(online_types); |
08b3acd7 DH |
2008 | return rc; |
2009 | } | |
2010 | EXPORT_SYMBOL_GPL(offline_and_remove_memory); | |
aba6efc4 | 2011 | #endif /* CONFIG_MEMORY_HOTREMOVE */ |