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 | |
e3a9d9fc OS |
45 | |
46 | /* | |
47 | * memory_hotplug.memmap_on_memory parameter | |
48 | */ | |
49 | static bool memmap_on_memory __ro_after_init; | |
50 | #ifdef CONFIG_MHP_MEMMAP_ON_MEMORY | |
51 | module_param(memmap_on_memory, bool, 0444); | |
52 | MODULE_PARM_DESC(memmap_on_memory, "Enable memmap on memory for memory hotplug"); | |
53 | #endif | |
a08a2ae3 | 54 | |
e83a437f DH |
55 | enum { |
56 | ONLINE_POLICY_CONTIG_ZONES = 0, | |
57 | ONLINE_POLICY_AUTO_MOVABLE, | |
58 | }; | |
59 | ||
60 | const char *online_policy_to_str[] = { | |
61 | [ONLINE_POLICY_CONTIG_ZONES] = "contig-zones", | |
62 | [ONLINE_POLICY_AUTO_MOVABLE] = "auto-movable", | |
63 | }; | |
64 | ||
65 | static int set_online_policy(const char *val, const struct kernel_param *kp) | |
66 | { | |
67 | int ret = sysfs_match_string(online_policy_to_str, val); | |
68 | ||
69 | if (ret < 0) | |
70 | return ret; | |
71 | *((int *)kp->arg) = ret; | |
72 | return 0; | |
73 | } | |
74 | ||
75 | static int get_online_policy(char *buffer, const struct kernel_param *kp) | |
76 | { | |
77 | return sprintf(buffer, "%s\n", online_policy_to_str[*((int *)kp->arg)]); | |
78 | } | |
79 | ||
80 | /* | |
81 | * memory_hotplug.online_policy: configure online behavior when onlining without | |
82 | * specifying a zone (MMOP_ONLINE) | |
83 | * | |
84 | * "contig-zones": keep zone contiguous | |
85 | * "auto-movable": online memory to ZONE_MOVABLE if the configuration | |
86 | * (auto_movable_ratio, auto_movable_numa_aware) allows for it | |
87 | */ | |
88 | static int online_policy __read_mostly = ONLINE_POLICY_CONTIG_ZONES; | |
89 | static const struct kernel_param_ops online_policy_ops = { | |
90 | .set = set_online_policy, | |
91 | .get = get_online_policy, | |
92 | }; | |
93 | module_param_cb(online_policy, &online_policy_ops, &online_policy, 0644); | |
94 | MODULE_PARM_DESC(online_policy, | |
95 | "Set the online policy (\"contig-zones\", \"auto-movable\") " | |
96 | "Default: \"contig-zones\""); | |
97 | ||
98 | /* | |
99 | * memory_hotplug.auto_movable_ratio: specify maximum MOVABLE:KERNEL ratio | |
100 | * | |
101 | * The ratio represent an upper limit and the kernel might decide to not | |
102 | * online some memory to ZONE_MOVABLE -- e.g., because hotplugged KERNEL memory | |
103 | * doesn't allow for more MOVABLE memory. | |
104 | */ | |
105 | static unsigned int auto_movable_ratio __read_mostly = 301; | |
106 | module_param(auto_movable_ratio, uint, 0644); | |
107 | MODULE_PARM_DESC(auto_movable_ratio, | |
108 | "Set the maximum ratio of MOVABLE:KERNEL memory in the system " | |
109 | "in percent for \"auto-movable\" online policy. Default: 301"); | |
110 | ||
111 | /* | |
112 | * memory_hotplug.auto_movable_numa_aware: consider numa node stats | |
113 | */ | |
114 | #ifdef CONFIG_NUMA | |
115 | static bool auto_movable_numa_aware __read_mostly = true; | |
116 | module_param(auto_movable_numa_aware, bool, 0644); | |
117 | MODULE_PARM_DESC(auto_movable_numa_aware, | |
118 | "Consider numa node stats in addition to global stats in " | |
119 | "\"auto-movable\" online policy. Default: true"); | |
120 | #endif /* CONFIG_NUMA */ | |
121 | ||
9d0ad8ca DK |
122 | /* |
123 | * online_page_callback contains pointer to current page onlining function. | |
124 | * Initially it is generic_online_page(). If it is required it could be | |
125 | * changed by calling set_online_page_callback() for callback registration | |
126 | * and restore_online_page_callback() for generic callback restore. | |
127 | */ | |
128 | ||
9d0ad8ca | 129 | static online_page_callback_t online_page_callback = generic_online_page; |
bfc8c901 | 130 | static DEFINE_MUTEX(online_page_callback_lock); |
9d0ad8ca | 131 | |
3f906ba2 | 132 | DEFINE_STATIC_PERCPU_RWSEM(mem_hotplug_lock); |
bfc8c901 | 133 | |
3f906ba2 TG |
134 | void get_online_mems(void) |
135 | { | |
136 | percpu_down_read(&mem_hotplug_lock); | |
137 | } | |
bfc8c901 | 138 | |
3f906ba2 TG |
139 | void put_online_mems(void) |
140 | { | |
141 | percpu_up_read(&mem_hotplug_lock); | |
142 | } | |
bfc8c901 | 143 | |
4932381e MH |
144 | bool movable_node_enabled = false; |
145 | ||
8604d9e5 | 146 | #ifndef CONFIG_MEMORY_HOTPLUG_DEFAULT_ONLINE |
1adf8b46 | 147 | int mhp_default_online_type = MMOP_OFFLINE; |
8604d9e5 | 148 | #else |
1adf8b46 | 149 | int mhp_default_online_type = MMOP_ONLINE; |
8604d9e5 | 150 | #endif |
31bc3858 | 151 | |
86dd995d VK |
152 | static int __init setup_memhp_default_state(char *str) |
153 | { | |
1adf8b46 | 154 | const int online_type = mhp_online_type_from_str(str); |
5f47adf7 DH |
155 | |
156 | if (online_type >= 0) | |
1adf8b46 | 157 | mhp_default_online_type = online_type; |
86dd995d VK |
158 | |
159 | return 1; | |
160 | } | |
161 | __setup("memhp_default_state=", setup_memhp_default_state); | |
162 | ||
30467e0b | 163 | void mem_hotplug_begin(void) |
20d6c96b | 164 | { |
3f906ba2 TG |
165 | cpus_read_lock(); |
166 | percpu_down_write(&mem_hotplug_lock); | |
20d6c96b KM |
167 | } |
168 | ||
30467e0b | 169 | void mem_hotplug_done(void) |
bfc8c901 | 170 | { |
3f906ba2 TG |
171 | percpu_up_write(&mem_hotplug_lock); |
172 | cpus_read_unlock(); | |
bfc8c901 | 173 | } |
20d6c96b | 174 | |
357b4da5 JG |
175 | u64 max_mem_size = U64_MAX; |
176 | ||
45e0b78b | 177 | /* add this memory to iomem resource */ |
7b7b2721 DH |
178 | static struct resource *register_memory_resource(u64 start, u64 size, |
179 | const char *resource_name) | |
45e0b78b | 180 | { |
2794129e DH |
181 | struct resource *res; |
182 | unsigned long flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY; | |
7b7b2721 DH |
183 | |
184 | if (strcmp(resource_name, "System RAM")) | |
7cf603d1 | 185 | flags |= IORESOURCE_SYSRAM_DRIVER_MANAGED; |
357b4da5 | 186 | |
bca3feaa AK |
187 | if (!mhp_range_allowed(start, size, true)) |
188 | return ERR_PTR(-E2BIG); | |
189 | ||
f3cd4c86 BH |
190 | /* |
191 | * Make sure value parsed from 'mem=' only restricts memory adding | |
192 | * while booting, so that memory hotplug won't be impacted. Please | |
193 | * refer to document of 'mem=' in kernel-parameters.txt for more | |
194 | * details. | |
195 | */ | |
196 | if (start + size > max_mem_size && system_state < SYSTEM_RUNNING) | |
357b4da5 JG |
197 | return ERR_PTR(-E2BIG); |
198 | ||
2794129e DH |
199 | /* |
200 | * Request ownership of the new memory range. This might be | |
201 | * a child of an existing resource that was present but | |
202 | * not marked as busy. | |
203 | */ | |
204 | res = __request_region(&iomem_resource, start, size, | |
205 | resource_name, flags); | |
206 | ||
207 | if (!res) { | |
208 | pr_debug("Unable to reserve System RAM region: %016llx->%016llx\n", | |
209 | start, start + size); | |
6f754ba4 | 210 | return ERR_PTR(-EEXIST); |
45e0b78b KM |
211 | } |
212 | return res; | |
213 | } | |
214 | ||
215 | static void release_memory_resource(struct resource *res) | |
216 | { | |
217 | if (!res) | |
218 | return; | |
219 | release_resource(res); | |
220 | kfree(res); | |
45e0b78b KM |
221 | } |
222 | ||
53947027 | 223 | #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE |
7ea62160 DW |
224 | static int check_pfn_span(unsigned long pfn, unsigned long nr_pages, |
225 | const char *reason) | |
226 | { | |
227 | /* | |
228 | * Disallow all operations smaller than a sub-section and only | |
229 | * allow operations smaller than a section for | |
230 | * SPARSEMEM_VMEMMAP. Note that check_hotplug_memory_range() | |
231 | * enforces a larger memory_block_size_bytes() granularity for | |
232 | * memory that will be marked online, so this check should only | |
233 | * fire for direct arch_{add,remove}_memory() users outside of | |
234 | * add_memory_resource(). | |
235 | */ | |
236 | unsigned long min_align; | |
237 | ||
238 | if (IS_ENABLED(CONFIG_SPARSEMEM_VMEMMAP)) | |
239 | min_align = PAGES_PER_SUBSECTION; | |
240 | else | |
241 | min_align = PAGES_PER_SECTION; | |
242 | if (!IS_ALIGNED(pfn, min_align) | |
243 | || !IS_ALIGNED(nr_pages, min_align)) { | |
244 | WARN(1, "Misaligned __%s_pages start: %#lx end: #%lx\n", | |
245 | reason, pfn, pfn + nr_pages - 1); | |
246 | return -EINVAL; | |
247 | } | |
248 | return 0; | |
249 | } | |
250 | ||
9f605f26 DW |
251 | /* |
252 | * Return page for the valid pfn only if the page is online. All pfn | |
253 | * walkers which rely on the fully initialized page->flags and others | |
254 | * should use this rather than pfn_valid && pfn_to_page | |
255 | */ | |
256 | struct page *pfn_to_online_page(unsigned long pfn) | |
257 | { | |
258 | unsigned long nr = pfn_to_section_nr(pfn); | |
1f90a347 | 259 | struct dev_pagemap *pgmap; |
9f9b02e5 DW |
260 | struct mem_section *ms; |
261 | ||
262 | if (nr >= NR_MEM_SECTIONS) | |
263 | return NULL; | |
264 | ||
265 | ms = __nr_to_section(nr); | |
266 | if (!online_section(ms)) | |
267 | return NULL; | |
268 | ||
269 | /* | |
270 | * Save some code text when online_section() + | |
271 | * pfn_section_valid() are sufficient. | |
272 | */ | |
273 | if (IS_ENABLED(CONFIG_HAVE_ARCH_PFN_VALID) && !pfn_valid(pfn)) | |
274 | return NULL; | |
275 | ||
276 | if (!pfn_section_valid(ms, pfn)) | |
277 | return NULL; | |
9f605f26 | 278 | |
1f90a347 DW |
279 | if (!online_device_section(ms)) |
280 | return pfn_to_page(pfn); | |
281 | ||
282 | /* | |
283 | * Slowpath: when ZONE_DEVICE collides with | |
284 | * ZONE_{NORMAL,MOVABLE} within the same section some pfns in | |
285 | * the section may be 'offline' but 'valid'. Only | |
286 | * get_dev_pagemap() can determine sub-section online status. | |
287 | */ | |
288 | pgmap = get_dev_pagemap(pfn, NULL); | |
289 | put_dev_pagemap(pgmap); | |
290 | ||
291 | /* The presence of a pgmap indicates ZONE_DEVICE offline pfn */ | |
292 | if (pgmap) | |
293 | return NULL; | |
294 | ||
9f9b02e5 | 295 | return pfn_to_page(pfn); |
9f605f26 DW |
296 | } |
297 | EXPORT_SYMBOL_GPL(pfn_to_online_page); | |
298 | ||
4edd7cef DR |
299 | /* |
300 | * Reasonably generic function for adding memory. It is | |
301 | * expected that archs that support memory hotplug will | |
302 | * call this function after deciding the zone to which to | |
303 | * add the new pages. | |
304 | */ | |
7ea62160 | 305 | int __ref __add_pages(int nid, unsigned long pfn, unsigned long nr_pages, |
f5637d3b | 306 | struct mhp_params *params) |
4edd7cef | 307 | { |
6cdd0b30 DH |
308 | const unsigned long end_pfn = pfn + nr_pages; |
309 | unsigned long cur_nr_pages; | |
9a845030 | 310 | int err; |
f5637d3b | 311 | struct vmem_altmap *altmap = params->altmap; |
4b94ffdc | 312 | |
bfeb022f LG |
313 | if (WARN_ON_ONCE(!params->pgprot.pgprot)) |
314 | return -EINVAL; | |
315 | ||
bca3feaa | 316 | VM_BUG_ON(!mhp_range_allowed(PFN_PHYS(pfn), nr_pages * PAGE_SIZE, false)); |
dca4436d | 317 | |
4b94ffdc DW |
318 | if (altmap) { |
319 | /* | |
320 | * Validate altmap is within bounds of the total request | |
321 | */ | |
7ea62160 | 322 | if (altmap->base_pfn != pfn |
4b94ffdc DW |
323 | || vmem_altmap_offset(altmap) > nr_pages) { |
324 | pr_warn_once("memory add fail, invalid altmap\n"); | |
7ea62160 | 325 | return -EINVAL; |
4b94ffdc DW |
326 | } |
327 | altmap->alloc = 0; | |
328 | } | |
329 | ||
7ea62160 DW |
330 | err = check_pfn_span(pfn, nr_pages, "add"); |
331 | if (err) | |
332 | return err; | |
333 | ||
6cdd0b30 DH |
334 | for (; pfn < end_pfn; pfn += cur_nr_pages) { |
335 | /* Select all remaining pages up to the next section boundary */ | |
336 | cur_nr_pages = min(end_pfn - pfn, | |
337 | SECTION_ALIGN_UP(pfn + 1) - pfn); | |
338 | err = sparse_add_section(nid, pfn, cur_nr_pages, altmap); | |
ba72b4c8 DW |
339 | if (err) |
340 | break; | |
f64ac5e6 | 341 | cond_resched(); |
4edd7cef | 342 | } |
c435a390 | 343 | vmemmap_populate_print_last(); |
4edd7cef DR |
344 | return err; |
345 | } | |
4edd7cef | 346 | |
815121d2 | 347 | /* find the smallest valid pfn in the range [start_pfn, end_pfn) */ |
d09b0137 | 348 | static unsigned long find_smallest_section_pfn(int nid, struct zone *zone, |
815121d2 YI |
349 | unsigned long start_pfn, |
350 | unsigned long end_pfn) | |
351 | { | |
49ba3c6b | 352 | for (; start_pfn < end_pfn; start_pfn += PAGES_PER_SUBSECTION) { |
7ce700bf | 353 | if (unlikely(!pfn_to_online_page(start_pfn))) |
815121d2 YI |
354 | continue; |
355 | ||
356 | if (unlikely(pfn_to_nid(start_pfn) != nid)) | |
357 | continue; | |
358 | ||
9b05158f | 359 | if (zone != page_zone(pfn_to_page(start_pfn))) |
815121d2 YI |
360 | continue; |
361 | ||
362 | return start_pfn; | |
363 | } | |
364 | ||
365 | return 0; | |
366 | } | |
367 | ||
368 | /* find the biggest valid pfn in the range [start_pfn, end_pfn). */ | |
d09b0137 | 369 | static unsigned long find_biggest_section_pfn(int nid, struct zone *zone, |
815121d2 YI |
370 | unsigned long start_pfn, |
371 | unsigned long end_pfn) | |
372 | { | |
815121d2 YI |
373 | unsigned long pfn; |
374 | ||
375 | /* pfn is the end pfn of a memory section. */ | |
376 | pfn = end_pfn - 1; | |
49ba3c6b | 377 | for (; pfn >= start_pfn; pfn -= PAGES_PER_SUBSECTION) { |
7ce700bf | 378 | if (unlikely(!pfn_to_online_page(pfn))) |
815121d2 YI |
379 | continue; |
380 | ||
381 | if (unlikely(pfn_to_nid(pfn) != nid)) | |
382 | continue; | |
383 | ||
9b05158f | 384 | if (zone != page_zone(pfn_to_page(pfn))) |
815121d2 YI |
385 | continue; |
386 | ||
387 | return pfn; | |
388 | } | |
389 | ||
390 | return 0; | |
391 | } | |
392 | ||
393 | static void shrink_zone_span(struct zone *zone, unsigned long start_pfn, | |
394 | unsigned long end_pfn) | |
395 | { | |
815121d2 | 396 | unsigned long pfn; |
815121d2 YI |
397 | int nid = zone_to_nid(zone); |
398 | ||
5d12071c | 399 | if (zone->zone_start_pfn == start_pfn) { |
815121d2 YI |
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, | |
5d12071c | 407 | zone_end_pfn(zone)); |
815121d2 | 408 | if (pfn) { |
5d12071c | 409 | zone->spanned_pages = zone_end_pfn(zone) - pfn; |
815121d2 | 410 | zone->zone_start_pfn = pfn; |
950b68d9 DH |
411 | } else { |
412 | zone->zone_start_pfn = 0; | |
413 | zone->spanned_pages = 0; | |
815121d2 | 414 | } |
5d12071c | 415 | } else if (zone_end_pfn(zone) == end_pfn) { |
815121d2 YI |
416 | /* |
417 | * If the section is biggest section in the zone, it need | |
418 | * shrink zone->spanned_pages. | |
419 | * In this case, we find second biggest valid mem_section for | |
420 | * shrinking zone. | |
421 | */ | |
5d12071c | 422 | pfn = find_biggest_section_pfn(nid, zone, zone->zone_start_pfn, |
815121d2 YI |
423 | start_pfn); |
424 | if (pfn) | |
5d12071c | 425 | zone->spanned_pages = pfn - zone->zone_start_pfn + 1; |
950b68d9 DH |
426 | else { |
427 | zone->zone_start_pfn = 0; | |
428 | zone->spanned_pages = 0; | |
429 | } | |
815121d2 | 430 | } |
815121d2 YI |
431 | } |
432 | ||
00d6c019 | 433 | static void update_pgdat_span(struct pglist_data *pgdat) |
815121d2 | 434 | { |
00d6c019 DH |
435 | unsigned long node_start_pfn = 0, node_end_pfn = 0; |
436 | struct zone *zone; | |
437 | ||
438 | for (zone = pgdat->node_zones; | |
439 | zone < pgdat->node_zones + MAX_NR_ZONES; zone++) { | |
6c922cf7 | 440 | unsigned long end_pfn = zone_end_pfn(zone); |
00d6c019 DH |
441 | |
442 | /* No need to lock the zones, they can't change. */ | |
656d5711 DH |
443 | if (!zone->spanned_pages) |
444 | continue; | |
445 | if (!node_end_pfn) { | |
446 | node_start_pfn = zone->zone_start_pfn; | |
6c922cf7 | 447 | node_end_pfn = end_pfn; |
656d5711 DH |
448 | continue; |
449 | } | |
450 | ||
6c922cf7 ML |
451 | if (end_pfn > node_end_pfn) |
452 | node_end_pfn = end_pfn; | |
00d6c019 DH |
453 | if (zone->zone_start_pfn < node_start_pfn) |
454 | node_start_pfn = zone->zone_start_pfn; | |
815121d2 YI |
455 | } |
456 | ||
00d6c019 DH |
457 | pgdat->node_start_pfn = node_start_pfn; |
458 | pgdat->node_spanned_pages = node_end_pfn - node_start_pfn; | |
815121d2 YI |
459 | } |
460 | ||
feee6b29 DH |
461 | void __ref remove_pfn_range_from_zone(struct zone *zone, |
462 | unsigned long start_pfn, | |
463 | unsigned long nr_pages) | |
815121d2 | 464 | { |
b7e3debd | 465 | const unsigned long end_pfn = start_pfn + nr_pages; |
815121d2 | 466 | struct pglist_data *pgdat = zone->zone_pgdat; |
27cacaad | 467 | unsigned long pfn, cur_nr_pages; |
815121d2 | 468 | |
d33695b1 | 469 | /* Poison struct pages because they are now uninitialized again. */ |
b7e3debd BW |
470 | for (pfn = start_pfn; pfn < end_pfn; pfn += cur_nr_pages) { |
471 | cond_resched(); | |
472 | ||
473 | /* Select all remaining pages up to the next section boundary */ | |
474 | cur_nr_pages = | |
475 | min(end_pfn - pfn, SECTION_ALIGN_UP(pfn + 1) - pfn); | |
476 | page_init_poison(pfn_to_page(pfn), | |
477 | sizeof(struct page) * cur_nr_pages); | |
478 | } | |
d33695b1 | 479 | |
7ce700bf DH |
480 | /* |
481 | * Zone shrinking code cannot properly deal with ZONE_DEVICE. So | |
482 | * we will not try to shrink the zones - which is okay as | |
483 | * set_zone_contiguous() cannot deal with ZONE_DEVICE either way. | |
484 | */ | |
5ef5f810 | 485 | if (zone_is_zone_device(zone)) |
7ce700bf | 486 | return; |
7ce700bf | 487 | |
feee6b29 DH |
488 | clear_zone_contiguous(zone); |
489 | ||
815121d2 | 490 | shrink_zone_span(zone, start_pfn, start_pfn + nr_pages); |
00d6c019 | 491 | update_pgdat_span(pgdat); |
feee6b29 DH |
492 | |
493 | set_zone_contiguous(zone); | |
815121d2 YI |
494 | } |
495 | ||
feee6b29 DH |
496 | static void __remove_section(unsigned long pfn, unsigned long nr_pages, |
497 | unsigned long map_offset, | |
498 | struct vmem_altmap *altmap) | |
ea01ea93 | 499 | { |
10404901 | 500 | struct mem_section *ms = __pfn_to_section(pfn); |
ea01ea93 | 501 | |
9d1d887d DH |
502 | if (WARN_ON_ONCE(!valid_section(ms))) |
503 | return; | |
ea01ea93 | 504 | |
ba72b4c8 | 505 | sparse_remove_section(ms, pfn, nr_pages, map_offset, altmap); |
ea01ea93 BP |
506 | } |
507 | ||
ea01ea93 | 508 | /** |
feee6b29 | 509 | * __remove_pages() - remove sections of pages |
7ea62160 | 510 | * @pfn: starting pageframe (must be aligned to start of a section) |
ea01ea93 | 511 | * @nr_pages: number of pages to remove (must be multiple of section size) |
e8b098fc | 512 | * @altmap: alternative device page map or %NULL if default memmap is used |
ea01ea93 BP |
513 | * |
514 | * Generic helper function to remove section mappings and sysfs entries | |
515 | * for the section of the memory we are removing. Caller needs to make | |
516 | * sure that pages are marked reserved and zones are adjust properly by | |
517 | * calling offline_pages(). | |
518 | */ | |
feee6b29 DH |
519 | void __remove_pages(unsigned long pfn, unsigned long nr_pages, |
520 | struct vmem_altmap *altmap) | |
ea01ea93 | 521 | { |
52fb87c8 DH |
522 | const unsigned long end_pfn = pfn + nr_pages; |
523 | unsigned long cur_nr_pages; | |
4b94ffdc | 524 | unsigned long map_offset = 0; |
4b94ffdc | 525 | |
96da4350 | 526 | map_offset = vmem_altmap_offset(altmap); |
ea01ea93 | 527 | |
7ea62160 DW |
528 | if (check_pfn_span(pfn, nr_pages, "remove")) |
529 | return; | |
ea01ea93 | 530 | |
52fb87c8 | 531 | for (; pfn < end_pfn; pfn += cur_nr_pages) { |
dd33ad7b | 532 | cond_resched(); |
52fb87c8 | 533 | /* Select all remaining pages up to the next section boundary */ |
a11b9419 DH |
534 | cur_nr_pages = min(end_pfn - pfn, |
535 | SECTION_ALIGN_UP(pfn + 1) - pfn); | |
52fb87c8 | 536 | __remove_section(pfn, cur_nr_pages, map_offset, altmap); |
4b94ffdc | 537 | map_offset = 0; |
ea01ea93 | 538 | } |
ea01ea93 | 539 | } |
ea01ea93 | 540 | |
9d0ad8ca DK |
541 | int set_online_page_callback(online_page_callback_t callback) |
542 | { | |
543 | int rc = -EINVAL; | |
544 | ||
bfc8c901 VD |
545 | get_online_mems(); |
546 | mutex_lock(&online_page_callback_lock); | |
9d0ad8ca DK |
547 | |
548 | if (online_page_callback == generic_online_page) { | |
549 | online_page_callback = callback; | |
550 | rc = 0; | |
551 | } | |
552 | ||
bfc8c901 VD |
553 | mutex_unlock(&online_page_callback_lock); |
554 | put_online_mems(); | |
9d0ad8ca DK |
555 | |
556 | return rc; | |
557 | } | |
558 | EXPORT_SYMBOL_GPL(set_online_page_callback); | |
559 | ||
560 | int restore_online_page_callback(online_page_callback_t callback) | |
561 | { | |
562 | int rc = -EINVAL; | |
563 | ||
bfc8c901 VD |
564 | get_online_mems(); |
565 | mutex_lock(&online_page_callback_lock); | |
9d0ad8ca DK |
566 | |
567 | if (online_page_callback == callback) { | |
568 | online_page_callback = generic_online_page; | |
569 | rc = 0; | |
570 | } | |
571 | ||
bfc8c901 VD |
572 | mutex_unlock(&online_page_callback_lock); |
573 | put_online_mems(); | |
9d0ad8ca DK |
574 | |
575 | return rc; | |
576 | } | |
577 | EXPORT_SYMBOL_GPL(restore_online_page_callback); | |
578 | ||
18db1491 | 579 | void generic_online_page(struct page *page, unsigned int order) |
9d0ad8ca | 580 | { |
c87cbc1f VB |
581 | /* |
582 | * Freeing the page with debug_pagealloc enabled will try to unmap it, | |
583 | * so we should map it first. This is better than introducing a special | |
584 | * case in page freeing fast path. | |
585 | */ | |
77bc7fd6 | 586 | debug_pagealloc_map_pages(page, 1 << order); |
a9cd410a AK |
587 | __free_pages_core(page, order); |
588 | totalram_pages_add(1UL << order); | |
589 | #ifdef CONFIG_HIGHMEM | |
590 | if (PageHighMem(page)) | |
591 | totalhigh_pages_add(1UL << order); | |
592 | #endif | |
593 | } | |
18db1491 | 594 | EXPORT_SYMBOL_GPL(generic_online_page); |
a9cd410a | 595 | |
aac65321 | 596 | static void online_pages_range(unsigned long start_pfn, unsigned long nr_pages) |
3947be19 | 597 | { |
b2c2ab20 DH |
598 | const unsigned long end_pfn = start_pfn + nr_pages; |
599 | unsigned long pfn; | |
b2c2ab20 DH |
600 | |
601 | /* | |
aac65321 DH |
602 | * Online the pages in MAX_ORDER - 1 aligned chunks. The callback might |
603 | * decide to not expose all pages to the buddy (e.g., expose them | |
604 | * later). We account all pages as being online and belonging to this | |
605 | * zone ("present"). | |
a08a2ae3 OS |
606 | * When using memmap_on_memory, the range might not be aligned to |
607 | * MAX_ORDER_NR_PAGES - 1, but pageblock aligned. __ffs() will detect | |
608 | * this and the first chunk to online will be pageblock_nr_pages. | |
b2c2ab20 | 609 | */ |
a08a2ae3 OS |
610 | for (pfn = start_pfn; pfn < end_pfn;) { |
611 | int order = min(MAX_ORDER - 1UL, __ffs(pfn)); | |
612 | ||
613 | (*online_page_callback)(pfn_to_page(pfn), order); | |
614 | pfn += (1UL << order); | |
615 | } | |
2d070eab | 616 | |
b2c2ab20 DH |
617 | /* mark all involved sections as online */ |
618 | online_mem_sections(start_pfn, end_pfn); | |
75884fb1 KH |
619 | } |
620 | ||
d9713679 LJ |
621 | /* check which state of node_states will be changed when online memory */ |
622 | static void node_states_check_changes_online(unsigned long nr_pages, | |
623 | struct zone *zone, struct memory_notify *arg) | |
624 | { | |
625 | int nid = zone_to_nid(zone); | |
d9713679 | 626 | |
98fa15f3 AK |
627 | arg->status_change_nid = NUMA_NO_NODE; |
628 | arg->status_change_nid_normal = NUMA_NO_NODE; | |
629 | arg->status_change_nid_high = NUMA_NO_NODE; | |
d9713679 | 630 | |
8efe33f4 OS |
631 | if (!node_state(nid, N_MEMORY)) |
632 | arg->status_change_nid = nid; | |
633 | if (zone_idx(zone) <= ZONE_NORMAL && !node_state(nid, N_NORMAL_MEMORY)) | |
d9713679 | 634 | arg->status_change_nid_normal = nid; |
6715ddf9 | 635 | #ifdef CONFIG_HIGHMEM |
d3ba3ae1 | 636 | if (zone_idx(zone) <= ZONE_HIGHMEM && !node_state(nid, N_HIGH_MEMORY)) |
6715ddf9 | 637 | arg->status_change_nid_high = nid; |
6715ddf9 | 638 | #endif |
d9713679 LJ |
639 | } |
640 | ||
641 | static void node_states_set_node(int node, struct memory_notify *arg) | |
642 | { | |
643 | if (arg->status_change_nid_normal >= 0) | |
644 | node_set_state(node, N_NORMAL_MEMORY); | |
645 | ||
6715ddf9 LJ |
646 | if (arg->status_change_nid_high >= 0) |
647 | node_set_state(node, N_HIGH_MEMORY); | |
648 | ||
83d83612 OS |
649 | if (arg->status_change_nid >= 0) |
650 | node_set_state(node, N_MEMORY); | |
d9713679 LJ |
651 | } |
652 | ||
f1dd2cd1 MH |
653 | static void __meminit resize_zone_range(struct zone *zone, unsigned long start_pfn, |
654 | unsigned long nr_pages) | |
655 | { | |
656 | unsigned long old_end_pfn = zone_end_pfn(zone); | |
657 | ||
658 | if (zone_is_empty(zone) || start_pfn < zone->zone_start_pfn) | |
659 | zone->zone_start_pfn = start_pfn; | |
660 | ||
661 | zone->spanned_pages = max(start_pfn + nr_pages, old_end_pfn) - zone->zone_start_pfn; | |
662 | } | |
663 | ||
664 | static void __meminit resize_pgdat_range(struct pglist_data *pgdat, unsigned long start_pfn, | |
665 | unsigned long nr_pages) | |
666 | { | |
667 | unsigned long old_end_pfn = pgdat_end_pfn(pgdat); | |
668 | ||
669 | if (!pgdat->node_spanned_pages || start_pfn < pgdat->node_start_pfn) | |
670 | pgdat->node_start_pfn = start_pfn; | |
671 | ||
672 | pgdat->node_spanned_pages = max(start_pfn + nr_pages, old_end_pfn) - pgdat->node_start_pfn; | |
f1dd2cd1 | 673 | |
3fccb74c | 674 | } |
1f90a347 DW |
675 | |
676 | static void section_taint_zone_device(unsigned long pfn) | |
677 | { | |
678 | struct mem_section *ms = __pfn_to_section(pfn); | |
679 | ||
680 | ms->section_mem_map |= SECTION_TAINT_ZONE_DEVICE; | |
681 | } | |
682 | ||
3fccb74c DH |
683 | /* |
684 | * Associate the pfn range with the given zone, initializing the memmaps | |
685 | * and resizing the pgdat/zone data to span the added pages. After this | |
686 | * call, all affected pages are PG_reserved. | |
d882c006 DH |
687 | * |
688 | * All aligned pageblocks are initialized to the specified migratetype | |
689 | * (usually MIGRATE_MOVABLE). Besides setting the migratetype, no related | |
690 | * zone stats (e.g., nr_isolate_pageblock) are touched. | |
3fccb74c | 691 | */ |
a99583e7 | 692 | void __ref move_pfn_range_to_zone(struct zone *zone, unsigned long start_pfn, |
d882c006 DH |
693 | unsigned long nr_pages, |
694 | struct vmem_altmap *altmap, int migratetype) | |
f1dd2cd1 MH |
695 | { |
696 | struct pglist_data *pgdat = zone->zone_pgdat; | |
697 | int nid = pgdat->node_id; | |
df429ac0 | 698 | |
f1dd2cd1 MH |
699 | clear_zone_contiguous(zone); |
700 | ||
fa004ab7 WY |
701 | if (zone_is_empty(zone)) |
702 | init_currently_empty_zone(zone, start_pfn, nr_pages); | |
f1dd2cd1 | 703 | resize_zone_range(zone, start_pfn, nr_pages); |
f1dd2cd1 | 704 | resize_pgdat_range(pgdat, start_pfn, nr_pages); |
f1dd2cd1 | 705 | |
1f90a347 DW |
706 | /* |
707 | * Subsection population requires care in pfn_to_online_page(). | |
708 | * Set the taint to enable the slow path detection of | |
709 | * ZONE_DEVICE pages in an otherwise ZONE_{NORMAL,MOVABLE} | |
710 | * section. | |
711 | */ | |
712 | if (zone_is_zone_device(zone)) { | |
713 | if (!IS_ALIGNED(start_pfn, PAGES_PER_SECTION)) | |
714 | section_taint_zone_device(start_pfn); | |
715 | if (!IS_ALIGNED(start_pfn + nr_pages, PAGES_PER_SECTION)) | |
716 | section_taint_zone_device(start_pfn + nr_pages); | |
717 | } | |
718 | ||
f1dd2cd1 MH |
719 | /* |
720 | * TODO now we have a visible range of pages which are not associated | |
721 | * with their zone properly. Not nice but set_pfnblock_flags_mask | |
722 | * expects the zone spans the pfn range. All the pages in the range | |
723 | * are reserved so nobody should be touching them so we should be safe | |
724 | */ | |
ab28cb6e | 725 | memmap_init_range(nr_pages, nid, zone_idx(zone), start_pfn, 0, |
d882c006 | 726 | MEMINIT_HOTPLUG, altmap, migratetype); |
f1dd2cd1 MH |
727 | |
728 | set_zone_contiguous(zone); | |
729 | } | |
730 | ||
e83a437f DH |
731 | struct auto_movable_stats { |
732 | unsigned long kernel_early_pages; | |
733 | unsigned long movable_pages; | |
734 | }; | |
735 | ||
736 | static void auto_movable_stats_account_zone(struct auto_movable_stats *stats, | |
737 | struct zone *zone) | |
738 | { | |
739 | if (zone_idx(zone) == ZONE_MOVABLE) { | |
740 | stats->movable_pages += zone->present_pages; | |
741 | } else { | |
742 | stats->kernel_early_pages += zone->present_early_pages; | |
743 | #ifdef CONFIG_CMA | |
744 | /* | |
745 | * CMA pages (never on hotplugged memory) behave like | |
746 | * ZONE_MOVABLE. | |
747 | */ | |
748 | stats->movable_pages += zone->cma_pages; | |
749 | stats->kernel_early_pages -= zone->cma_pages; | |
750 | #endif /* CONFIG_CMA */ | |
751 | } | |
752 | } | |
3fcebf90 DH |
753 | struct auto_movable_group_stats { |
754 | unsigned long movable_pages; | |
755 | unsigned long req_kernel_early_pages; | |
756 | }; | |
e83a437f | 757 | |
3fcebf90 DH |
758 | static int auto_movable_stats_account_group(struct memory_group *group, |
759 | void *arg) | |
760 | { | |
761 | const int ratio = READ_ONCE(auto_movable_ratio); | |
762 | struct auto_movable_group_stats *stats = arg; | |
763 | long pages; | |
764 | ||
765 | /* | |
766 | * We don't support modifying the config while the auto-movable online | |
767 | * policy is already enabled. Just avoid the division by zero below. | |
768 | */ | |
769 | if (!ratio) | |
770 | return 0; | |
771 | ||
772 | /* | |
773 | * Calculate how many early kernel pages this group requires to | |
774 | * satisfy the configured zone ratio. | |
775 | */ | |
776 | pages = group->present_movable_pages * 100 / ratio; | |
777 | pages -= group->present_kernel_pages; | |
778 | ||
779 | if (pages > 0) | |
780 | stats->req_kernel_early_pages += pages; | |
781 | stats->movable_pages += group->present_movable_pages; | |
782 | return 0; | |
783 | } | |
784 | ||
785 | static bool auto_movable_can_online_movable(int nid, struct memory_group *group, | |
786 | unsigned long nr_pages) | |
e83a437f | 787 | { |
e83a437f | 788 | unsigned long kernel_early_pages, movable_pages; |
3fcebf90 DH |
789 | struct auto_movable_group_stats group_stats = {}; |
790 | struct auto_movable_stats stats = {}; | |
e83a437f DH |
791 | pg_data_t *pgdat = NODE_DATA(nid); |
792 | struct zone *zone; | |
793 | int i; | |
794 | ||
795 | /* Walk all relevant zones and collect MOVABLE vs. KERNEL stats. */ | |
796 | if (nid == NUMA_NO_NODE) { | |
797 | /* TODO: cache values */ | |
798 | for_each_populated_zone(zone) | |
799 | auto_movable_stats_account_zone(&stats, zone); | |
800 | } else { | |
801 | for (i = 0; i < MAX_NR_ZONES; i++) { | |
802 | zone = pgdat->node_zones + i; | |
803 | if (populated_zone(zone)) | |
804 | auto_movable_stats_account_zone(&stats, zone); | |
805 | } | |
806 | } | |
807 | ||
808 | kernel_early_pages = stats.kernel_early_pages; | |
809 | movable_pages = stats.movable_pages; | |
810 | ||
3fcebf90 DH |
811 | /* |
812 | * Kernel memory inside dynamic memory group allows for more MOVABLE | |
813 | * memory within the same group. Remove the effect of all but the | |
814 | * current group from the stats. | |
815 | */ | |
816 | walk_dynamic_memory_groups(nid, auto_movable_stats_account_group, | |
817 | group, &group_stats); | |
818 | if (kernel_early_pages <= group_stats.req_kernel_early_pages) | |
819 | return false; | |
820 | kernel_early_pages -= group_stats.req_kernel_early_pages; | |
821 | movable_pages -= group_stats.movable_pages; | |
822 | ||
823 | if (group && group->is_dynamic) | |
824 | kernel_early_pages += group->present_kernel_pages; | |
825 | ||
e83a437f DH |
826 | /* |
827 | * Test if we could online the given number of pages to ZONE_MOVABLE | |
828 | * and still stay in the configured ratio. | |
829 | */ | |
830 | movable_pages += nr_pages; | |
831 | return movable_pages <= (auto_movable_ratio * kernel_early_pages) / 100; | |
832 | } | |
833 | ||
c246a213 MH |
834 | /* |
835 | * Returns a default kernel memory zone for the given pfn range. | |
836 | * If no kernel zone covers this pfn range it will automatically go | |
837 | * to the ZONE_NORMAL. | |
838 | */ | |
c6f03e29 | 839 | static struct zone *default_kernel_zone_for_pfn(int nid, unsigned long start_pfn, |
c246a213 MH |
840 | unsigned long nr_pages) |
841 | { | |
842 | struct pglist_data *pgdat = NODE_DATA(nid); | |
843 | int zid; | |
844 | ||
845 | for (zid = 0; zid <= ZONE_NORMAL; zid++) { | |
846 | struct zone *zone = &pgdat->node_zones[zid]; | |
847 | ||
848 | if (zone_intersects(zone, start_pfn, nr_pages)) | |
849 | return zone; | |
850 | } | |
851 | ||
852 | return &pgdat->node_zones[ZONE_NORMAL]; | |
853 | } | |
854 | ||
e83a437f DH |
855 | /* |
856 | * Determine to which zone to online memory dynamically based on user | |
857 | * configuration and system stats. We care about the following ratio: | |
858 | * | |
859 | * MOVABLE : KERNEL | |
860 | * | |
861 | * Whereby MOVABLE is memory in ZONE_MOVABLE and KERNEL is memory in | |
862 | * one of the kernel zones. CMA pages inside one of the kernel zones really | |
863 | * behaves like ZONE_MOVABLE, so we treat them accordingly. | |
864 | * | |
865 | * We don't allow for hotplugged memory in a KERNEL zone to increase the | |
866 | * amount of MOVABLE memory we can have, so we end up with: | |
867 | * | |
868 | * MOVABLE : KERNEL_EARLY | |
869 | * | |
870 | * Whereby KERNEL_EARLY is memory in one of the kernel zones, available sinze | |
871 | * boot. We base our calculation on KERNEL_EARLY internally, because: | |
872 | * | |
873 | * a) Hotplugged memory in one of the kernel zones can sometimes still get | |
874 | * hotunplugged, especially when hot(un)plugging individual memory blocks. | |
875 | * There is no coordination across memory devices, therefore "automatic" | |
876 | * hotunplugging, as implemented in hypervisors, could result in zone | |
877 | * imbalances. | |
878 | * b) Early/boot memory in one of the kernel zones can usually not get | |
879 | * hotunplugged again (e.g., no firmware interface to unplug, fragmented | |
880 | * with unmovable allocations). While there are corner cases where it might | |
881 | * still work, it is barely relevant in practice. | |
882 | * | |
3fcebf90 DH |
883 | * Exceptions are dynamic memory groups, which allow for more MOVABLE |
884 | * memory within the same memory group -- because in that case, there is | |
885 | * coordination within the single memory device managed by a single driver. | |
886 | * | |
e83a437f DH |
887 | * We rely on "present pages" instead of "managed pages", as the latter is |
888 | * highly unreliable and dynamic in virtualized environments, and does not | |
889 | * consider boot time allocations. For example, memory ballooning adjusts the | |
890 | * managed pages when inflating/deflating the balloon, and balloon compaction | |
891 | * can even migrate inflated pages between zones. | |
892 | * | |
893 | * Using "present pages" is better but some things to keep in mind are: | |
894 | * | |
895 | * a) Some memblock allocations, such as for the crashkernel area, are | |
896 | * effectively unused by the kernel, yet they account to "present pages". | |
897 | * Fortunately, these allocations are comparatively small in relevant setups | |
898 | * (e.g., fraction of system memory). | |
899 | * b) Some hotplugged memory blocks in virtualized environments, esecially | |
900 | * hotplugged by virtio-mem, look like they are completely present, however, | |
901 | * only parts of the memory block are actually currently usable. | |
902 | * "present pages" is an upper limit that can get reached at runtime. As | |
903 | * we base our calculations on KERNEL_EARLY, this is not an issue. | |
904 | */ | |
445fcf7c DH |
905 | static struct zone *auto_movable_zone_for_pfn(int nid, |
906 | struct memory_group *group, | |
907 | unsigned long pfn, | |
e83a437f DH |
908 | unsigned long nr_pages) |
909 | { | |
445fcf7c DH |
910 | unsigned long online_pages = 0, max_pages, end_pfn; |
911 | struct page *page; | |
912 | ||
e83a437f DH |
913 | if (!auto_movable_ratio) |
914 | goto kernel_zone; | |
915 | ||
445fcf7c DH |
916 | if (group && !group->is_dynamic) { |
917 | max_pages = group->s.max_pages; | |
918 | online_pages = group->present_movable_pages; | |
919 | ||
920 | /* If anything is !MOVABLE online the rest !MOVABLE. */ | |
921 | if (group->present_kernel_pages) | |
922 | goto kernel_zone; | |
923 | } else if (!group || group->d.unit_pages == nr_pages) { | |
924 | max_pages = nr_pages; | |
925 | } else { | |
926 | max_pages = group->d.unit_pages; | |
927 | /* | |
928 | * Take a look at all online sections in the current unit. | |
929 | * We can safely assume that all pages within a section belong | |
930 | * to the same zone, because dynamic memory groups only deal | |
931 | * with hotplugged memory. | |
932 | */ | |
933 | pfn = ALIGN_DOWN(pfn, group->d.unit_pages); | |
934 | end_pfn = pfn + group->d.unit_pages; | |
935 | for (; pfn < end_pfn; pfn += PAGES_PER_SECTION) { | |
936 | page = pfn_to_online_page(pfn); | |
937 | if (!page) | |
938 | continue; | |
939 | /* If anything is !MOVABLE online the rest !MOVABLE. */ | |
940 | if (page_zonenum(page) != ZONE_MOVABLE) | |
941 | goto kernel_zone; | |
942 | online_pages += PAGES_PER_SECTION; | |
943 | } | |
944 | } | |
945 | ||
946 | /* | |
947 | * Online MOVABLE if we could *currently* online all remaining parts | |
948 | * MOVABLE. We expect to (add+) online them immediately next, so if | |
949 | * nobody interferes, all will be MOVABLE if possible. | |
950 | */ | |
951 | nr_pages = max_pages - online_pages; | |
3fcebf90 | 952 | if (!auto_movable_can_online_movable(NUMA_NO_NODE, group, nr_pages)) |
e83a437f DH |
953 | goto kernel_zone; |
954 | ||
955 | #ifdef CONFIG_NUMA | |
956 | if (auto_movable_numa_aware && | |
3fcebf90 | 957 | !auto_movable_can_online_movable(nid, group, nr_pages)) |
e83a437f DH |
958 | goto kernel_zone; |
959 | #endif /* CONFIG_NUMA */ | |
960 | ||
961 | return &NODE_DATA(nid)->node_zones[ZONE_MOVABLE]; | |
962 | kernel_zone: | |
963 | return default_kernel_zone_for_pfn(nid, pfn, nr_pages); | |
964 | } | |
965 | ||
c6f03e29 MH |
966 | static inline struct zone *default_zone_for_pfn(int nid, unsigned long start_pfn, |
967 | unsigned long nr_pages) | |
e5e68930 | 968 | { |
c6f03e29 MH |
969 | struct zone *kernel_zone = default_kernel_zone_for_pfn(nid, start_pfn, |
970 | nr_pages); | |
971 | struct zone *movable_zone = &NODE_DATA(nid)->node_zones[ZONE_MOVABLE]; | |
972 | bool in_kernel = zone_intersects(kernel_zone, start_pfn, nr_pages); | |
973 | bool in_movable = zone_intersects(movable_zone, start_pfn, nr_pages); | |
e5e68930 MH |
974 | |
975 | /* | |
c6f03e29 MH |
976 | * We inherit the existing zone in a simple case where zones do not |
977 | * overlap in the given range | |
e5e68930 | 978 | */ |
c6f03e29 MH |
979 | if (in_kernel ^ in_movable) |
980 | return (in_kernel) ? kernel_zone : movable_zone; | |
9f123ab5 | 981 | |
c6f03e29 MH |
982 | /* |
983 | * If the range doesn't belong to any zone or two zones overlap in the | |
984 | * given range then we use movable zone only if movable_node is | |
985 | * enabled because we always online to a kernel zone by default. | |
986 | */ | |
987 | return movable_node_enabled ? movable_zone : kernel_zone; | |
9f123ab5 MH |
988 | } |
989 | ||
7cf209ba | 990 | struct zone *zone_for_pfn_range(int online_type, int nid, |
445fcf7c | 991 | struct memory_group *group, unsigned long start_pfn, |
e5e68930 | 992 | unsigned long nr_pages) |
f1dd2cd1 | 993 | { |
c6f03e29 MH |
994 | if (online_type == MMOP_ONLINE_KERNEL) |
995 | return default_kernel_zone_for_pfn(nid, start_pfn, nr_pages); | |
f1dd2cd1 | 996 | |
c6f03e29 MH |
997 | if (online_type == MMOP_ONLINE_MOVABLE) |
998 | return &NODE_DATA(nid)->node_zones[ZONE_MOVABLE]; | |
df429ac0 | 999 | |
e83a437f | 1000 | if (online_policy == ONLINE_POLICY_AUTO_MOVABLE) |
445fcf7c | 1001 | return auto_movable_zone_for_pfn(nid, group, start_pfn, nr_pages); |
e83a437f | 1002 | |
c6f03e29 | 1003 | return default_zone_for_pfn(nid, start_pfn, nr_pages); |
e5e68930 MH |
1004 | } |
1005 | ||
a08a2ae3 OS |
1006 | /* |
1007 | * This function should only be called by memory_block_{online,offline}, | |
1008 | * and {online,offline}_pages. | |
1009 | */ | |
836809ec DH |
1010 | void adjust_present_page_count(struct page *page, struct memory_group *group, |
1011 | long nr_pages) | |
f9901144 | 1012 | { |
4b097002 | 1013 | struct zone *zone = page_zone(page); |
836809ec | 1014 | const bool movable = zone_idx(zone) == ZONE_MOVABLE; |
4b097002 DH |
1015 | |
1016 | /* | |
1017 | * We only support onlining/offlining/adding/removing of complete | |
1018 | * memory blocks; therefore, either all is either early or hotplugged. | |
1019 | */ | |
1020 | if (early_section(__pfn_to_section(page_to_pfn(page)))) | |
1021 | zone->present_early_pages += nr_pages; | |
f9901144 | 1022 | zone->present_pages += nr_pages; |
f9901144 | 1023 | zone->zone_pgdat->node_present_pages += nr_pages; |
836809ec DH |
1024 | |
1025 | if (group && movable) | |
1026 | group->present_movable_pages += nr_pages; | |
1027 | else if (group && !movable) | |
1028 | group->present_kernel_pages += nr_pages; | |
f9901144 DH |
1029 | } |
1030 | ||
a08a2ae3 OS |
1031 | int mhp_init_memmap_on_memory(unsigned long pfn, unsigned long nr_pages, |
1032 | struct zone *zone) | |
1033 | { | |
1034 | unsigned long end_pfn = pfn + nr_pages; | |
1035 | int ret; | |
1036 | ||
1037 | ret = kasan_add_zero_shadow(__va(PFN_PHYS(pfn)), PFN_PHYS(nr_pages)); | |
1038 | if (ret) | |
1039 | return ret; | |
1040 | ||
1041 | move_pfn_range_to_zone(zone, pfn, nr_pages, NULL, MIGRATE_UNMOVABLE); | |
1042 | ||
1043 | /* | |
1044 | * It might be that the vmemmap_pages fully span sections. If that is | |
1045 | * the case, mark those sections online here as otherwise they will be | |
1046 | * left offline. | |
1047 | */ | |
1048 | if (nr_pages >= PAGES_PER_SECTION) | |
1049 | online_mem_sections(pfn, ALIGN_DOWN(end_pfn, PAGES_PER_SECTION)); | |
1050 | ||
1051 | return ret; | |
1052 | } | |
1053 | ||
1054 | void mhp_deinit_memmap_on_memory(unsigned long pfn, unsigned long nr_pages) | |
1055 | { | |
1056 | unsigned long end_pfn = pfn + nr_pages; | |
1057 | ||
1058 | /* | |
1059 | * It might be that the vmemmap_pages fully span sections. If that is | |
1060 | * the case, mark those sections offline here as otherwise they will be | |
1061 | * left online. | |
1062 | */ | |
1063 | if (nr_pages >= PAGES_PER_SECTION) | |
1064 | offline_mem_sections(pfn, ALIGN_DOWN(end_pfn, PAGES_PER_SECTION)); | |
1065 | ||
1066 | /* | |
1067 | * The pages associated with this vmemmap have been offlined, so | |
1068 | * we can reset its state here. | |
1069 | */ | |
1070 | remove_pfn_range_from_zone(page_zone(pfn_to_page(pfn)), pfn, nr_pages); | |
1071 | kasan_remove_zero_shadow(__va(PFN_PHYS(pfn)), PFN_PHYS(nr_pages)); | |
1072 | } | |
1073 | ||
836809ec DH |
1074 | int __ref online_pages(unsigned long pfn, unsigned long nr_pages, |
1075 | struct zone *zone, struct memory_group *group) | |
75884fb1 | 1076 | { |
aa47228a | 1077 | unsigned long flags; |
6811378e | 1078 | int need_zonelists_rebuild = 0; |
a08a2ae3 | 1079 | const int nid = zone_to_nid(zone); |
7b78d335 YG |
1080 | int ret; |
1081 | struct memory_notify arg; | |
d0dc12e8 | 1082 | |
dd8e2f23 OS |
1083 | /* |
1084 | * {on,off}lining is constrained to full memory sections (or more | |
041711ce | 1085 | * precisely to memory blocks from the user space POV). |
dd8e2f23 OS |
1086 | * memmap_on_memory is an exception because it reserves initial part |
1087 | * of the physical memory space for vmemmaps. That space is pageblock | |
1088 | * aligned. | |
1089 | */ | |
4986fac1 | 1090 | if (WARN_ON_ONCE(!nr_pages || |
dd8e2f23 OS |
1091 | !IS_ALIGNED(pfn, pageblock_nr_pages) || |
1092 | !IS_ALIGNED(pfn + nr_pages, PAGES_PER_SECTION))) | |
4986fac1 DH |
1093 | return -EINVAL; |
1094 | ||
381eab4a DH |
1095 | mem_hotplug_begin(); |
1096 | ||
f1dd2cd1 | 1097 | /* associate pfn range with the zone */ |
b30c5927 | 1098 | move_pfn_range_to_zone(zone, pfn, nr_pages, NULL, MIGRATE_ISOLATE); |
f1dd2cd1 | 1099 | |
7b78d335 YG |
1100 | arg.start_pfn = pfn; |
1101 | arg.nr_pages = nr_pages; | |
d9713679 | 1102 | node_states_check_changes_online(nr_pages, zone, &arg); |
7b78d335 | 1103 | |
7b78d335 YG |
1104 | ret = memory_notify(MEM_GOING_ONLINE, &arg); |
1105 | ret = notifier_to_errno(ret); | |
e33e33b4 CY |
1106 | if (ret) |
1107 | goto failed_addition; | |
1108 | ||
b30c5927 DH |
1109 | /* |
1110 | * Fixup the number of isolated pageblocks before marking the sections | |
1111 | * onlining, such that undo_isolate_page_range() works correctly. | |
1112 | */ | |
1113 | spin_lock_irqsave(&zone->lock, flags); | |
1114 | zone->nr_isolate_pageblock += nr_pages / pageblock_nr_pages; | |
1115 | spin_unlock_irqrestore(&zone->lock, flags); | |
1116 | ||
6811378e YG |
1117 | /* |
1118 | * If this zone is not populated, then it is not in zonelist. | |
1119 | * This means the page allocator ignores this zone. | |
1120 | * So, zonelist must be updated after online. | |
1121 | */ | |
6dcd73d7 | 1122 | if (!populated_zone(zone)) { |
6811378e | 1123 | need_zonelists_rebuild = 1; |
72675e13 | 1124 | setup_zone_pageset(zone); |
6dcd73d7 | 1125 | } |
6811378e | 1126 | |
aac65321 | 1127 | online_pages_range(pfn, nr_pages); |
836809ec | 1128 | adjust_present_page_count(pfn_to_page(pfn), group, nr_pages); |
aa47228a | 1129 | |
b30c5927 DH |
1130 | node_states_set_node(nid, &arg); |
1131 | if (need_zonelists_rebuild) | |
1132 | build_all_zonelists(NULL); | |
b30c5927 DH |
1133 | |
1134 | /* Basic onlining is complete, allow allocation of onlined pages. */ | |
1135 | undo_isolate_page_range(pfn, pfn + nr_pages, MIGRATE_MOVABLE); | |
1136 | ||
93146d98 | 1137 | /* |
b86c5fc4 DH |
1138 | * Freshly onlined pages aren't shuffled (e.g., all pages are placed to |
1139 | * the tail of the freelist when undoing isolation). Shuffle the whole | |
1140 | * zone to make sure the just onlined pages are properly distributed | |
1141 | * across the whole freelist - to create an initial shuffle. | |
93146d98 | 1142 | */ |
e900a918 DW |
1143 | shuffle_zone(zone); |
1144 | ||
b92ca18e | 1145 | /* reinitialise watermarks and update pcp limits */ |
1b79acc9 KM |
1146 | init_per_zone_wmark_min(); |
1147 | ||
ca9a46f8 DH |
1148 | kswapd_run(nid); |
1149 | kcompactd_run(nid); | |
61b13993 | 1150 | |
2d1d43f6 | 1151 | writeback_set_ratelimit(); |
7b78d335 | 1152 | |
ca9a46f8 | 1153 | memory_notify(MEM_ONLINE, &arg); |
381eab4a | 1154 | mem_hotplug_done(); |
30467e0b | 1155 | return 0; |
e33e33b4 CY |
1156 | |
1157 | failed_addition: | |
1158 | pr_debug("online_pages [mem %#010llx-%#010llx] failed\n", | |
1159 | (unsigned long long) pfn << PAGE_SHIFT, | |
1160 | (((unsigned long long) pfn + nr_pages) << PAGE_SHIFT) - 1); | |
1161 | memory_notify(MEM_CANCEL_ONLINE, &arg); | |
feee6b29 | 1162 | remove_pfn_range_from_zone(zone, pfn, nr_pages); |
381eab4a | 1163 | mem_hotplug_done(); |
e33e33b4 | 1164 | return ret; |
3947be19 | 1165 | } |
53947027 | 1166 | #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */ |
bc02af93 | 1167 | |
0bd85420 TC |
1168 | static void reset_node_present_pages(pg_data_t *pgdat) |
1169 | { | |
1170 | struct zone *z; | |
1171 | ||
1172 | for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++) | |
1173 | z->present_pages = 0; | |
1174 | ||
1175 | pgdat->node_present_pages = 0; | |
1176 | } | |
1177 | ||
e1319331 | 1178 | /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */ |
c68ab18c | 1179 | static pg_data_t __ref *hotadd_new_pgdat(int nid) |
9af3c2de YG |
1180 | { |
1181 | struct pglist_data *pgdat; | |
9af3c2de | 1182 | |
a1e565aa TC |
1183 | pgdat = NODE_DATA(nid); |
1184 | if (!pgdat) { | |
1185 | pgdat = arch_alloc_nodedata(nid); | |
1186 | if (!pgdat) | |
1187 | return NULL; | |
9af3c2de | 1188 | |
33fce011 WY |
1189 | pgdat->per_cpu_nodestats = |
1190 | alloc_percpu(struct per_cpu_nodestat); | |
a1e565aa | 1191 | arch_refresh_nodedata(nid, pgdat); |
b0dc3a34 | 1192 | } else { |
33fce011 | 1193 | int cpu; |
e716f2eb | 1194 | /* |
97a225e6 JK |
1195 | * Reset the nr_zones, order and highest_zoneidx before reuse. |
1196 | * Note that kswapd will init kswapd_highest_zoneidx properly | |
e716f2eb MG |
1197 | * when it starts in the near future. |
1198 | */ | |
b0dc3a34 | 1199 | pgdat->nr_zones = 0; |
38087d9b | 1200 | pgdat->kswapd_order = 0; |
97a225e6 | 1201 | pgdat->kswapd_highest_zoneidx = 0; |
33fce011 WY |
1202 | for_each_online_cpu(cpu) { |
1203 | struct per_cpu_nodestat *p; | |
1204 | ||
1205 | p = per_cpu_ptr(pgdat->per_cpu_nodestats, cpu); | |
1206 | memset(p, 0, sizeof(*p)); | |
1207 | } | |
a1e565aa | 1208 | } |
9af3c2de YG |
1209 | |
1210 | /* we can use NODE_DATA(nid) from here */ | |
03e85f9d | 1211 | pgdat->node_id = nid; |
c68ab18c | 1212 | pgdat->node_start_pfn = 0; |
03e85f9d | 1213 | |
9af3c2de | 1214 | /* init node's zones as empty zones, we don't have any present pages.*/ |
03e85f9d | 1215 | free_area_init_core_hotplug(nid); |
9af3c2de | 1216 | |
959ecc48 KH |
1217 | /* |
1218 | * The node we allocated has no zone fallback lists. For avoiding | |
1219 | * to access not-initialized zonelist, build here. | |
1220 | */ | |
72675e13 | 1221 | build_all_zonelists(pgdat); |
959ecc48 | 1222 | |
0bd85420 TC |
1223 | /* |
1224 | * When memory is hot-added, all the memory is in offline state. So | |
1225 | * clear all zones' present_pages because they will be updated in | |
1226 | * online_pages() and offline_pages(). | |
1227 | */ | |
03e85f9d | 1228 | reset_node_managed_pages(pgdat); |
0bd85420 TC |
1229 | reset_node_present_pages(pgdat); |
1230 | ||
9af3c2de YG |
1231 | return pgdat; |
1232 | } | |
1233 | ||
b9ff0360 | 1234 | static void rollback_node_hotadd(int nid) |
9af3c2de | 1235 | { |
b9ff0360 OS |
1236 | pg_data_t *pgdat = NODE_DATA(nid); |
1237 | ||
9af3c2de | 1238 | arch_refresh_nodedata(nid, NULL); |
5830169f | 1239 | free_percpu(pgdat->per_cpu_nodestats); |
9af3c2de | 1240 | arch_free_nodedata(pgdat); |
9af3c2de YG |
1241 | } |
1242 | ||
0a547039 | 1243 | |
ba2d2666 MG |
1244 | /* |
1245 | * __try_online_node - online a node if offlined | |
e8b098fc | 1246 | * @nid: the node ID |
b9ff0360 | 1247 | * @set_node_online: Whether we want to online the node |
cf23422b | 1248 | * called by cpu_up() to online a node without onlined memory. |
b9ff0360 OS |
1249 | * |
1250 | * Returns: | |
1251 | * 1 -> a new node has been allocated | |
1252 | * 0 -> the node is already online | |
1253 | * -ENOMEM -> the node could not be allocated | |
cf23422b | 1254 | */ |
c68ab18c | 1255 | static int __try_online_node(int nid, bool set_node_online) |
cf23422b | 1256 | { |
b9ff0360 OS |
1257 | pg_data_t *pgdat; |
1258 | int ret = 1; | |
cf23422b | 1259 | |
01b0f197 TK |
1260 | if (node_online(nid)) |
1261 | return 0; | |
1262 | ||
c68ab18c | 1263 | pgdat = hotadd_new_pgdat(nid); |
7553e8f2 | 1264 | if (!pgdat) { |
01b0f197 | 1265 | pr_err("Cannot online node %d due to NULL pgdat\n", nid); |
cf23422b | 1266 | ret = -ENOMEM; |
1267 | goto out; | |
1268 | } | |
b9ff0360 OS |
1269 | |
1270 | if (set_node_online) { | |
1271 | node_set_online(nid); | |
1272 | ret = register_one_node(nid); | |
1273 | BUG_ON(ret); | |
1274 | } | |
cf23422b | 1275 | out: |
b9ff0360 OS |
1276 | return ret; |
1277 | } | |
1278 | ||
1279 | /* | |
1280 | * Users of this function always want to online/register the node | |
1281 | */ | |
1282 | int try_online_node(int nid) | |
1283 | { | |
1284 | int ret; | |
1285 | ||
1286 | mem_hotplug_begin(); | |
c68ab18c | 1287 | ret = __try_online_node(nid, true); |
bfc8c901 | 1288 | mem_hotplug_done(); |
cf23422b | 1289 | return ret; |
1290 | } | |
1291 | ||
27356f54 TK |
1292 | static int check_hotplug_memory_range(u64 start, u64 size) |
1293 | { | |
ba325585 | 1294 | /* memory range must be block size aligned */ |
cec3ebd0 DH |
1295 | if (!size || !IS_ALIGNED(start, memory_block_size_bytes()) || |
1296 | !IS_ALIGNED(size, memory_block_size_bytes())) { | |
ba325585 | 1297 | pr_err("Block size [%#lx] unaligned hotplug range: start %#llx, size %#llx", |
cec3ebd0 | 1298 | memory_block_size_bytes(), start, size); |
27356f54 TK |
1299 | return -EINVAL; |
1300 | } | |
1301 | ||
1302 | return 0; | |
1303 | } | |
1304 | ||
31bc3858 VK |
1305 | static int online_memory_block(struct memory_block *mem, void *arg) |
1306 | { | |
1adf8b46 | 1307 | mem->online_type = mhp_default_online_type; |
dc18d706 | 1308 | return device_online(&mem->dev); |
31bc3858 VK |
1309 | } |
1310 | ||
a08a2ae3 OS |
1311 | bool mhp_supports_memmap_on_memory(unsigned long size) |
1312 | { | |
1313 | unsigned long nr_vmemmap_pages = size / PAGE_SIZE; | |
1314 | unsigned long vmemmap_size = nr_vmemmap_pages * sizeof(struct page); | |
1315 | unsigned long remaining_size = size - vmemmap_size; | |
1316 | ||
1317 | /* | |
1318 | * Besides having arch support and the feature enabled at runtime, we | |
1319 | * need a few more assumptions to hold true: | |
1320 | * | |
1321 | * a) We span a single memory block: memory onlining/offlinin;g happens | |
1322 | * in memory block granularity. We don't want the vmemmap of online | |
1323 | * memory blocks to reside on offline memory blocks. In the future, | |
1324 | * we might want to support variable-sized memory blocks to make the | |
1325 | * feature more versatile. | |
1326 | * | |
1327 | * b) The vmemmap pages span complete PMDs: We don't want vmemmap code | |
1328 | * to populate memory from the altmap for unrelated parts (i.e., | |
1329 | * other memory blocks) | |
1330 | * | |
1331 | * c) The vmemmap pages (and thereby the pages that will be exposed to | |
1332 | * the buddy) have to cover full pageblocks: memory onlining/offlining | |
1333 | * code requires applicable ranges to be page-aligned, for example, to | |
1334 | * set the migratetypes properly. | |
1335 | * | |
1336 | * TODO: Although we have a check here to make sure that vmemmap pages | |
1337 | * fully populate a PMD, it is not the right place to check for | |
1338 | * this. A much better solution involves improving vmemmap code | |
1339 | * to fallback to base pages when trying to populate vmemmap using | |
1340 | * altmap as an alternative source of memory, and we do not exactly | |
1341 | * populate a single PMD. | |
1342 | */ | |
1343 | return memmap_on_memory && | |
2d7a2171 | 1344 | !hugetlb_free_vmemmap_enabled && |
a08a2ae3 OS |
1345 | IS_ENABLED(CONFIG_MHP_MEMMAP_ON_MEMORY) && |
1346 | size == memory_block_size_bytes() && | |
1347 | IS_ALIGNED(vmemmap_size, PMD_SIZE) && | |
1348 | IS_ALIGNED(remaining_size, (pageblock_nr_pages << PAGE_SHIFT)); | |
1349 | } | |
1350 | ||
8df1d0e4 DH |
1351 | /* |
1352 | * NOTE: The caller must call lock_device_hotplug() to serialize hotplug | |
1353 | * and online/offline operations (triggered e.g. by sysfs). | |
1354 | * | |
1355 | * we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG | |
1356 | */ | |
b6117199 | 1357 | int __ref add_memory_resource(int nid, struct resource *res, mhp_t mhp_flags) |
bc02af93 | 1358 | { |
d15dfd31 | 1359 | struct mhp_params params = { .pgprot = pgprot_mhp(PAGE_KERNEL) }; |
a08a2ae3 | 1360 | struct vmem_altmap mhp_altmap = {}; |
028fc57a | 1361 | struct memory_group *group = NULL; |
62cedb9f | 1362 | u64 start, size; |
b9ff0360 | 1363 | bool new_node = false; |
bc02af93 YG |
1364 | int ret; |
1365 | ||
62cedb9f DV |
1366 | start = res->start; |
1367 | size = resource_size(res); | |
1368 | ||
27356f54 TK |
1369 | ret = check_hotplug_memory_range(start, size); |
1370 | if (ret) | |
1371 | return ret; | |
1372 | ||
028fc57a DH |
1373 | if (mhp_flags & MHP_NID_IS_MGID) { |
1374 | group = memory_group_find_by_id(nid); | |
1375 | if (!group) | |
1376 | return -EINVAL; | |
1377 | nid = group->nid; | |
1378 | } | |
1379 | ||
fa6d9ec7 VV |
1380 | if (!node_possible(nid)) { |
1381 | WARN(1, "node %d was absent from the node_possible_map\n", nid); | |
1382 | return -EINVAL; | |
1383 | } | |
1384 | ||
bfc8c901 | 1385 | mem_hotplug_begin(); |
ac13c462 | 1386 | |
52219aea DH |
1387 | if (IS_ENABLED(CONFIG_ARCH_KEEP_MEMBLOCK)) |
1388 | memblock_add_node(start, size, nid); | |
7f36e3e5 | 1389 | |
c68ab18c | 1390 | ret = __try_online_node(nid, false); |
b9ff0360 OS |
1391 | if (ret < 0) |
1392 | goto error; | |
1393 | new_node = ret; | |
9af3c2de | 1394 | |
a08a2ae3 OS |
1395 | /* |
1396 | * Self hosted memmap array | |
1397 | */ | |
1398 | if (mhp_flags & MHP_MEMMAP_ON_MEMORY) { | |
1399 | if (!mhp_supports_memmap_on_memory(size)) { | |
1400 | ret = -EINVAL; | |
1401 | goto error; | |
1402 | } | |
1403 | mhp_altmap.free = PHYS_PFN(size); | |
1404 | mhp_altmap.base_pfn = PHYS_PFN(start); | |
1405 | params.altmap = &mhp_altmap; | |
1406 | } | |
1407 | ||
bc02af93 | 1408 | /* call arch's memory hotadd */ |
f5637d3b | 1409 | ret = arch_add_memory(nid, start, size, ¶ms); |
9af3c2de YG |
1410 | if (ret < 0) |
1411 | goto error; | |
1412 | ||
db051a0d | 1413 | /* create memory block devices after memory was added */ |
028fc57a DH |
1414 | ret = create_memory_block_devices(start, size, mhp_altmap.alloc, |
1415 | group); | |
db051a0d | 1416 | if (ret) { |
65a2aa5f | 1417 | arch_remove_memory(start, size, NULL); |
db051a0d DH |
1418 | goto error; |
1419 | } | |
1420 | ||
a1e565aa | 1421 | if (new_node) { |
d5b6f6a3 | 1422 | /* If sysfs file of new node can't be created, cpu on the node |
0fc44159 YG |
1423 | * can't be hot-added. There is no rollback way now. |
1424 | * So, check by BUG_ON() to catch it reluctantly.. | |
d5b6f6a3 | 1425 | * We online node here. We can't roll back from here. |
0fc44159 | 1426 | */ |
d5b6f6a3 OS |
1427 | node_set_online(nid); |
1428 | ret = __register_one_node(nid); | |
0fc44159 YG |
1429 | BUG_ON(ret); |
1430 | } | |
1431 | ||
d5b6f6a3 | 1432 | /* link memory sections under this node.*/ |
90c7eaeb LD |
1433 | link_mem_sections(nid, PFN_DOWN(start), PFN_UP(start + size - 1), |
1434 | MEMINIT_HOTPLUG); | |
d5b6f6a3 | 1435 | |
d96ae530 | 1436 | /* create new memmap entry */ |
7b7b2721 DH |
1437 | if (!strcmp(res->name, "System RAM")) |
1438 | firmware_map_add_hotplug(start, start + size, "System RAM"); | |
d96ae530 | 1439 | |
381eab4a DH |
1440 | /* device_online() will take the lock when calling online_pages() */ |
1441 | mem_hotplug_done(); | |
1442 | ||
9ca6551e DH |
1443 | /* |
1444 | * In case we're allowed to merge the resource, flag it and trigger | |
1445 | * merging now that adding succeeded. | |
1446 | */ | |
26011267 | 1447 | if (mhp_flags & MHP_MERGE_RESOURCE) |
9ca6551e DH |
1448 | merge_system_ram_resource(res); |
1449 | ||
31bc3858 | 1450 | /* online pages if requested */ |
1adf8b46 | 1451 | if (mhp_default_online_type != MMOP_OFFLINE) |
fbcf73ce | 1452 | walk_memory_blocks(start, size, NULL, online_memory_block); |
31bc3858 | 1453 | |
381eab4a | 1454 | return ret; |
9af3c2de YG |
1455 | error: |
1456 | /* rollback pgdat allocation and others */ | |
b9ff0360 OS |
1457 | if (new_node) |
1458 | rollback_node_hotadd(nid); | |
52219aea DH |
1459 | if (IS_ENABLED(CONFIG_ARCH_KEEP_MEMBLOCK)) |
1460 | memblock_remove(start, size); | |
bfc8c901 | 1461 | mem_hotplug_done(); |
bc02af93 YG |
1462 | return ret; |
1463 | } | |
62cedb9f | 1464 | |
8df1d0e4 | 1465 | /* requires device_hotplug_lock, see add_memory_resource() */ |
b6117199 | 1466 | int __ref __add_memory(int nid, u64 start, u64 size, mhp_t mhp_flags) |
62cedb9f DV |
1467 | { |
1468 | struct resource *res; | |
1469 | int ret; | |
1470 | ||
7b7b2721 | 1471 | res = register_memory_resource(start, size, "System RAM"); |
6f754ba4 VK |
1472 | if (IS_ERR(res)) |
1473 | return PTR_ERR(res); | |
62cedb9f | 1474 | |
b6117199 | 1475 | ret = add_memory_resource(nid, res, mhp_flags); |
62cedb9f DV |
1476 | if (ret < 0) |
1477 | release_memory_resource(res); | |
1478 | return ret; | |
1479 | } | |
8df1d0e4 | 1480 | |
b6117199 | 1481 | int add_memory(int nid, u64 start, u64 size, mhp_t mhp_flags) |
8df1d0e4 DH |
1482 | { |
1483 | int rc; | |
1484 | ||
1485 | lock_device_hotplug(); | |
b6117199 | 1486 | rc = __add_memory(nid, start, size, mhp_flags); |
8df1d0e4 DH |
1487 | unlock_device_hotplug(); |
1488 | ||
1489 | return rc; | |
1490 | } | |
bc02af93 | 1491 | EXPORT_SYMBOL_GPL(add_memory); |
0c0e6195 | 1492 | |
7b7b2721 DH |
1493 | /* |
1494 | * Add special, driver-managed memory to the system as system RAM. Such | |
1495 | * memory is not exposed via the raw firmware-provided memmap as system | |
1496 | * RAM, instead, it is detected and added by a driver - during cold boot, | |
1497 | * after a reboot, and after kexec. | |
1498 | * | |
1499 | * Reasons why this memory should not be used for the initial memmap of a | |
1500 | * kexec kernel or for placing kexec images: | |
1501 | * - The booting kernel is in charge of determining how this memory will be | |
1502 | * used (e.g., use persistent memory as system RAM) | |
1503 | * - Coordination with a hypervisor is required before this memory | |
1504 | * can be used (e.g., inaccessible parts). | |
1505 | * | |
1506 | * For this memory, no entries in /sys/firmware/memmap ("raw firmware-provided | |
1507 | * memory map") are created. Also, the created memory resource is flagged | |
7cf603d1 | 1508 | * with IORESOURCE_SYSRAM_DRIVER_MANAGED, so in-kernel users can special-case |
7b7b2721 DH |
1509 | * this memory as well (esp., not place kexec images onto it). |
1510 | * | |
1511 | * The resource_name (visible via /proc/iomem) has to have the format | |
1512 | * "System RAM ($DRIVER)". | |
1513 | */ | |
1514 | int add_memory_driver_managed(int nid, u64 start, u64 size, | |
b6117199 | 1515 | const char *resource_name, mhp_t mhp_flags) |
7b7b2721 DH |
1516 | { |
1517 | struct resource *res; | |
1518 | int rc; | |
1519 | ||
1520 | if (!resource_name || | |
1521 | strstr(resource_name, "System RAM (") != resource_name || | |
1522 | resource_name[strlen(resource_name) - 1] != ')') | |
1523 | return -EINVAL; | |
1524 | ||
1525 | lock_device_hotplug(); | |
1526 | ||
1527 | res = register_memory_resource(start, size, resource_name); | |
1528 | if (IS_ERR(res)) { | |
1529 | rc = PTR_ERR(res); | |
1530 | goto out_unlock; | |
1531 | } | |
1532 | ||
b6117199 | 1533 | rc = add_memory_resource(nid, res, mhp_flags); |
7b7b2721 DH |
1534 | if (rc < 0) |
1535 | release_memory_resource(res); | |
1536 | ||
1537 | out_unlock: | |
1538 | unlock_device_hotplug(); | |
1539 | return rc; | |
1540 | } | |
1541 | EXPORT_SYMBOL_GPL(add_memory_driver_managed); | |
1542 | ||
bca3feaa AK |
1543 | /* |
1544 | * Platforms should define arch_get_mappable_range() that provides | |
1545 | * maximum possible addressable physical memory range for which the | |
1546 | * linear mapping could be created. The platform returned address | |
1547 | * range must adhere to these following semantics. | |
1548 | * | |
1549 | * - range.start <= range.end | |
1550 | * - Range includes both end points [range.start..range.end] | |
1551 | * | |
1552 | * There is also a fallback definition provided here, allowing the | |
1553 | * entire possible physical address range in case any platform does | |
1554 | * not define arch_get_mappable_range(). | |
1555 | */ | |
1556 | struct range __weak arch_get_mappable_range(void) | |
1557 | { | |
1558 | struct range mhp_range = { | |
1559 | .start = 0UL, | |
1560 | .end = -1ULL, | |
1561 | }; | |
1562 | return mhp_range; | |
1563 | } | |
1564 | ||
1565 | struct range mhp_get_pluggable_range(bool need_mapping) | |
1566 | { | |
1567 | const u64 max_phys = (1ULL << MAX_PHYSMEM_BITS) - 1; | |
1568 | struct range mhp_range; | |
1569 | ||
1570 | if (need_mapping) { | |
1571 | mhp_range = arch_get_mappable_range(); | |
1572 | if (mhp_range.start > max_phys) { | |
1573 | mhp_range.start = 0; | |
1574 | mhp_range.end = 0; | |
1575 | } | |
1576 | mhp_range.end = min_t(u64, mhp_range.end, max_phys); | |
1577 | } else { | |
1578 | mhp_range.start = 0; | |
1579 | mhp_range.end = max_phys; | |
1580 | } | |
1581 | return mhp_range; | |
1582 | } | |
1583 | EXPORT_SYMBOL_GPL(mhp_get_pluggable_range); | |
1584 | ||
1585 | bool mhp_range_allowed(u64 start, u64 size, bool need_mapping) | |
1586 | { | |
1587 | struct range mhp_range = mhp_get_pluggable_range(need_mapping); | |
1588 | u64 end = start + size; | |
1589 | ||
1590 | if (start < end && start >= mhp_range.start && (end - 1) <= mhp_range.end) | |
1591 | return true; | |
1592 | ||
1593 | pr_warn("Hotplug memory [%#llx-%#llx] exceeds maximum addressable range [%#llx-%#llx]\n", | |
1594 | start, end, mhp_range.start, mhp_range.end); | |
1595 | return false; | |
1596 | } | |
1597 | ||
0c0e6195 KH |
1598 | #ifdef CONFIG_MEMORY_HOTREMOVE |
1599 | /* | |
92917998 DH |
1600 | * Confirm all pages in a range [start, end) belong to the same zone (skipping |
1601 | * memory holes). When true, return the zone. | |
0c0e6195 | 1602 | */ |
92917998 DH |
1603 | struct zone *test_pages_in_a_zone(unsigned long start_pfn, |
1604 | unsigned long end_pfn) | |
0c0e6195 | 1605 | { |
5f0f2887 | 1606 | unsigned long pfn, sec_end_pfn; |
0c0e6195 KH |
1607 | struct zone *zone = NULL; |
1608 | struct page *page; | |
673d40c8 | 1609 | |
deb88a2a | 1610 | for (pfn = start_pfn, sec_end_pfn = SECTION_ALIGN_UP(start_pfn + 1); |
0c0e6195 | 1611 | pfn < end_pfn; |
deb88a2a | 1612 | pfn = sec_end_pfn, sec_end_pfn += PAGES_PER_SECTION) { |
5f0f2887 AB |
1613 | /* Make sure the memory section is present first */ |
1614 | if (!present_section_nr(pfn_to_section_nr(pfn))) | |
0c0e6195 | 1615 | continue; |
5f0f2887 AB |
1616 | for (; pfn < sec_end_pfn && pfn < end_pfn; |
1617 | pfn += MAX_ORDER_NR_PAGES) { | |
24feb47c | 1618 | /* Check if we got outside of the zone */ |
673d40c8 | 1619 | if (zone && !zone_spans_pfn(zone, pfn)) |
92917998 | 1620 | return NULL; |
673d40c8 | 1621 | page = pfn_to_page(pfn); |
5f0f2887 | 1622 | if (zone && page_zone(page) != zone) |
92917998 | 1623 | return NULL; |
5f0f2887 AB |
1624 | zone = page_zone(page); |
1625 | } | |
0c0e6195 | 1626 | } |
deb88a2a | 1627 | |
92917998 | 1628 | return zone; |
0c0e6195 KH |
1629 | } |
1630 | ||
1631 | /* | |
0efadf48 | 1632 | * Scan pfn range [start,end) to find movable/migratable pages (LRU pages, |
aa218795 DH |
1633 | * non-lru movable pages and hugepages). Will skip over most unmovable |
1634 | * pages (esp., pages that can be skipped when offlining), but bail out on | |
1635 | * definitely unmovable pages. | |
1636 | * | |
1637 | * Returns: | |
1638 | * 0 in case a movable page is found and movable_pfn was updated. | |
1639 | * -ENOENT in case no movable page was found. | |
1640 | * -EBUSY in case a definitely unmovable page was found. | |
0c0e6195 | 1641 | */ |
aa218795 DH |
1642 | static int scan_movable_pages(unsigned long start, unsigned long end, |
1643 | unsigned long *movable_pfn) | |
0c0e6195 KH |
1644 | { |
1645 | unsigned long pfn; | |
eeb0efd0 | 1646 | |
0c0e6195 | 1647 | for (pfn = start; pfn < end; pfn++) { |
eeb0efd0 OS |
1648 | struct page *page, *head; |
1649 | unsigned long skip; | |
1650 | ||
1651 | if (!pfn_valid(pfn)) | |
1652 | continue; | |
1653 | page = pfn_to_page(pfn); | |
1654 | if (PageLRU(page)) | |
aa218795 | 1655 | goto found; |
eeb0efd0 | 1656 | if (__PageMovable(page)) |
aa218795 DH |
1657 | goto found; |
1658 | ||
1659 | /* | |
1660 | * PageOffline() pages that are not marked __PageMovable() and | |
1661 | * have a reference count > 0 (after MEM_GOING_OFFLINE) are | |
1662 | * definitely unmovable. If their reference count would be 0, | |
1663 | * they could at least be skipped when offlining memory. | |
1664 | */ | |
1665 | if (PageOffline(page) && page_count(page)) | |
1666 | return -EBUSY; | |
eeb0efd0 OS |
1667 | |
1668 | if (!PageHuge(page)) | |
1669 | continue; | |
1670 | head = compound_head(page); | |
8f251a3d MK |
1671 | /* |
1672 | * This test is racy as we hold no reference or lock. The | |
1673 | * hugetlb page could have been free'ed and head is no longer | |
1674 | * a hugetlb page before the following check. In such unlikely | |
1675 | * cases false positives and negatives are possible. Calling | |
1676 | * code must deal with these scenarios. | |
1677 | */ | |
1678 | if (HPageMigratable(head)) | |
aa218795 | 1679 | goto found; |
d8c6546b | 1680 | skip = compound_nr(head) - (page - head); |
eeb0efd0 | 1681 | pfn += skip - 1; |
0c0e6195 | 1682 | } |
aa218795 DH |
1683 | return -ENOENT; |
1684 | found: | |
1685 | *movable_pfn = pfn; | |
0c0e6195 KH |
1686 | return 0; |
1687 | } | |
1688 | ||
0c0e6195 KH |
1689 | static int |
1690 | do_migrate_range(unsigned long start_pfn, unsigned long end_pfn) | |
1691 | { | |
1692 | unsigned long pfn; | |
6c357848 | 1693 | struct page *page, *head; |
0c0e6195 KH |
1694 | int ret = 0; |
1695 | LIST_HEAD(source); | |
786dee86 LM |
1696 | static DEFINE_RATELIMIT_STATE(migrate_rs, DEFAULT_RATELIMIT_INTERVAL, |
1697 | DEFAULT_RATELIMIT_BURST); | |
0c0e6195 | 1698 | |
a85009c3 | 1699 | for (pfn = start_pfn; pfn < end_pfn; pfn++) { |
0c0e6195 KH |
1700 | if (!pfn_valid(pfn)) |
1701 | continue; | |
1702 | page = pfn_to_page(pfn); | |
6c357848 | 1703 | head = compound_head(page); |
c8721bbb NH |
1704 | |
1705 | if (PageHuge(page)) { | |
d8c6546b | 1706 | pfn = page_to_pfn(head) + compound_nr(head) - 1; |
daf3538a | 1707 | isolate_huge_page(head, &source); |
c8721bbb | 1708 | continue; |
94723aaf | 1709 | } else if (PageTransHuge(page)) |
6c357848 | 1710 | pfn = page_to_pfn(head) + thp_nr_pages(page) - 1; |
c8721bbb | 1711 | |
b15c8726 MH |
1712 | /* |
1713 | * HWPoison pages have elevated reference counts so the migration would | |
1714 | * fail on them. It also doesn't make any sense to migrate them in the | |
1715 | * first place. Still try to unmap such a page in case it is still mapped | |
1716 | * (e.g. current hwpoison implementation doesn't unmap KSM pages but keep | |
1717 | * the unmap as the catch all safety net). | |
1718 | */ | |
1719 | if (PageHWPoison(page)) { | |
1720 | if (WARN_ON(PageLRU(page))) | |
1721 | isolate_lru_page(page); | |
1722 | if (page_mapped(page)) | |
013339df | 1723 | try_to_unmap(page, TTU_IGNORE_MLOCK); |
b15c8726 MH |
1724 | continue; |
1725 | } | |
1726 | ||
700c2a46 | 1727 | if (!get_page_unless_zero(page)) |
0c0e6195 KH |
1728 | continue; |
1729 | /* | |
0efadf48 YX |
1730 | * We can skip free pages. And we can deal with pages on |
1731 | * LRU and non-lru movable pages. | |
0c0e6195 | 1732 | */ |
0efadf48 YX |
1733 | if (PageLRU(page)) |
1734 | ret = isolate_lru_page(page); | |
1735 | else | |
1736 | ret = isolate_movable_page(page, ISOLATE_UNEVICTABLE); | |
0c0e6195 | 1737 | if (!ret) { /* Success */ |
62695a84 | 1738 | list_add_tail(&page->lru, &source); |
0efadf48 YX |
1739 | if (!__PageMovable(page)) |
1740 | inc_node_page_state(page, NR_ISOLATED_ANON + | |
9de4f22a | 1741 | page_is_file_lru(page)); |
6d9c285a | 1742 | |
0c0e6195 | 1743 | } else { |
786dee86 LM |
1744 | if (__ratelimit(&migrate_rs)) { |
1745 | pr_warn("failed to isolate pfn %lx\n", pfn); | |
1746 | dump_page(page, "isolation failed"); | |
1747 | } | |
0c0e6195 | 1748 | } |
1723058e | 1749 | put_page(page); |
0c0e6195 | 1750 | } |
f3ab2636 | 1751 | if (!list_empty(&source)) { |
203e6e5c JK |
1752 | nodemask_t nmask = node_states[N_MEMORY]; |
1753 | struct migration_target_control mtc = { | |
1754 | .nmask = &nmask, | |
1755 | .gfp_mask = GFP_USER | __GFP_MOVABLE | __GFP_RETRY_MAYFAIL, | |
1756 | }; | |
1757 | ||
1758 | /* | |
1759 | * We have checked that migration range is on a single zone so | |
1760 | * we can use the nid of the first page to all the others. | |
1761 | */ | |
1762 | mtc.nid = page_to_nid(list_first_entry(&source, struct page, lru)); | |
1763 | ||
1764 | /* | |
1765 | * try to allocate from a different node but reuse this node | |
1766 | * if there are no other online nodes to be used (e.g. we are | |
1767 | * offlining a part of the only existing node) | |
1768 | */ | |
1769 | node_clear(mtc.nid, nmask); | |
1770 | if (nodes_empty(nmask)) | |
1771 | node_set(mtc.nid, nmask); | |
1772 | ret = migrate_pages(&source, alloc_migration_target, NULL, | |
5ac95884 | 1773 | (unsigned long)&mtc, MIGRATE_SYNC, MR_MEMORY_HOTPLUG, NULL); |
2932c8b0 MH |
1774 | if (ret) { |
1775 | list_for_each_entry(page, &source, lru) { | |
786dee86 LM |
1776 | if (__ratelimit(&migrate_rs)) { |
1777 | pr_warn("migrating pfn %lx failed ret:%d\n", | |
1778 | page_to_pfn(page), ret); | |
1779 | dump_page(page, "migration failure"); | |
1780 | } | |
2932c8b0 | 1781 | } |
c8721bbb | 1782 | putback_movable_pages(&source); |
2932c8b0 | 1783 | } |
0c0e6195 | 1784 | } |
1723058e | 1785 | |
0c0e6195 KH |
1786 | return ret; |
1787 | } | |
1788 | ||
c5320926 TC |
1789 | static int __init cmdline_parse_movable_node(char *p) |
1790 | { | |
55ac590c | 1791 | movable_node_enabled = true; |
c5320926 TC |
1792 | return 0; |
1793 | } | |
1794 | early_param("movable_node", cmdline_parse_movable_node); | |
1795 | ||
d9713679 LJ |
1796 | /* check which state of node_states will be changed when offline memory */ |
1797 | static void node_states_check_changes_offline(unsigned long nr_pages, | |
1798 | struct zone *zone, struct memory_notify *arg) | |
1799 | { | |
1800 | struct pglist_data *pgdat = zone->zone_pgdat; | |
1801 | unsigned long present_pages = 0; | |
86b27bea | 1802 | enum zone_type zt; |
d9713679 | 1803 | |
98fa15f3 AK |
1804 | arg->status_change_nid = NUMA_NO_NODE; |
1805 | arg->status_change_nid_normal = NUMA_NO_NODE; | |
1806 | arg->status_change_nid_high = NUMA_NO_NODE; | |
d9713679 LJ |
1807 | |
1808 | /* | |
86b27bea OS |
1809 | * Check whether node_states[N_NORMAL_MEMORY] will be changed. |
1810 | * If the memory to be offline is within the range | |
1811 | * [0..ZONE_NORMAL], and it is the last present memory there, | |
1812 | * the zones in that range will become empty after the offlining, | |
1813 | * thus we can determine that we need to clear the node from | |
1814 | * node_states[N_NORMAL_MEMORY]. | |
d9713679 | 1815 | */ |
86b27bea | 1816 | for (zt = 0; zt <= ZONE_NORMAL; zt++) |
d9713679 | 1817 | present_pages += pgdat->node_zones[zt].present_pages; |
86b27bea | 1818 | if (zone_idx(zone) <= ZONE_NORMAL && nr_pages >= present_pages) |
d9713679 | 1819 | arg->status_change_nid_normal = zone_to_nid(zone); |
d9713679 | 1820 | |
6715ddf9 LJ |
1821 | #ifdef CONFIG_HIGHMEM |
1822 | /* | |
86b27bea OS |
1823 | * node_states[N_HIGH_MEMORY] contains nodes which |
1824 | * have normal memory or high memory. | |
1825 | * Here we add the present_pages belonging to ZONE_HIGHMEM. | |
1826 | * If the zone is within the range of [0..ZONE_HIGHMEM), and | |
1827 | * we determine that the zones in that range become empty, | |
1828 | * we need to clear the node for N_HIGH_MEMORY. | |
6715ddf9 | 1829 | */ |
86b27bea OS |
1830 | present_pages += pgdat->node_zones[ZONE_HIGHMEM].present_pages; |
1831 | if (zone_idx(zone) <= ZONE_HIGHMEM && nr_pages >= present_pages) | |
6715ddf9 | 1832 | arg->status_change_nid_high = zone_to_nid(zone); |
6715ddf9 LJ |
1833 | #endif |
1834 | ||
d9713679 | 1835 | /* |
86b27bea OS |
1836 | * We have accounted the pages from [0..ZONE_NORMAL), and |
1837 | * in case of CONFIG_HIGHMEM the pages from ZONE_HIGHMEM | |
1838 | * as well. | |
1839 | * Here we count the possible pages from ZONE_MOVABLE. | |
1840 | * If after having accounted all the pages, we see that the nr_pages | |
1841 | * to be offlined is over or equal to the accounted pages, | |
1842 | * we know that the node will become empty, and so, we can clear | |
1843 | * it for N_MEMORY as well. | |
d9713679 | 1844 | */ |
86b27bea | 1845 | present_pages += pgdat->node_zones[ZONE_MOVABLE].present_pages; |
d9713679 | 1846 | |
d9713679 LJ |
1847 | if (nr_pages >= present_pages) |
1848 | arg->status_change_nid = zone_to_nid(zone); | |
d9713679 LJ |
1849 | } |
1850 | ||
1851 | static void node_states_clear_node(int node, struct memory_notify *arg) | |
1852 | { | |
1853 | if (arg->status_change_nid_normal >= 0) | |
1854 | node_clear_state(node, N_NORMAL_MEMORY); | |
1855 | ||
cf01f6f5 | 1856 | if (arg->status_change_nid_high >= 0) |
d9713679 | 1857 | node_clear_state(node, N_HIGH_MEMORY); |
6715ddf9 | 1858 | |
cf01f6f5 | 1859 | if (arg->status_change_nid >= 0) |
6715ddf9 | 1860 | node_clear_state(node, N_MEMORY); |
d9713679 LJ |
1861 | } |
1862 | ||
c5e79ef5 DH |
1863 | static int count_system_ram_pages_cb(unsigned long start_pfn, |
1864 | unsigned long nr_pages, void *data) | |
1865 | { | |
1866 | unsigned long *nr_system_ram_pages = data; | |
1867 | ||
1868 | *nr_system_ram_pages += nr_pages; | |
1869 | return 0; | |
1870 | } | |
1871 | ||
836809ec DH |
1872 | int __ref offline_pages(unsigned long start_pfn, unsigned long nr_pages, |
1873 | struct memory_group *group) | |
0c0e6195 | 1874 | { |
73a11c96 | 1875 | const unsigned long end_pfn = start_pfn + nr_pages; |
0a1a9a00 | 1876 | unsigned long pfn, system_ram_pages = 0; |
d702909f | 1877 | unsigned long flags; |
0c0e6195 | 1878 | struct zone *zone; |
7b78d335 | 1879 | struct memory_notify arg; |
ea15153c | 1880 | int ret, node; |
79605093 | 1881 | char *reason; |
0c0e6195 | 1882 | |
dd8e2f23 OS |
1883 | /* |
1884 | * {on,off}lining is constrained to full memory sections (or more | |
041711ce | 1885 | * precisely to memory blocks from the user space POV). |
dd8e2f23 OS |
1886 | * memmap_on_memory is an exception because it reserves initial part |
1887 | * of the physical memory space for vmemmaps. That space is pageblock | |
1888 | * aligned. | |
1889 | */ | |
4986fac1 | 1890 | if (WARN_ON_ONCE(!nr_pages || |
dd8e2f23 OS |
1891 | !IS_ALIGNED(start_pfn, pageblock_nr_pages) || |
1892 | !IS_ALIGNED(start_pfn + nr_pages, PAGES_PER_SECTION))) | |
4986fac1 DH |
1893 | return -EINVAL; |
1894 | ||
381eab4a DH |
1895 | mem_hotplug_begin(); |
1896 | ||
c5e79ef5 DH |
1897 | /* |
1898 | * Don't allow to offline memory blocks that contain holes. | |
1899 | * Consequently, memory blocks with holes can never get onlined | |
1900 | * via the hotplug path - online_pages() - as hotplugged memory has | |
1901 | * no holes. This way, we e.g., don't have to worry about marking | |
1902 | * memory holes PG_reserved, don't need pfn_valid() checks, and can | |
1903 | * avoid using walk_system_ram_range() later. | |
1904 | */ | |
73a11c96 | 1905 | walk_system_ram_range(start_pfn, nr_pages, &system_ram_pages, |
c5e79ef5 | 1906 | count_system_ram_pages_cb); |
73a11c96 | 1907 | if (system_ram_pages != nr_pages) { |
c5e79ef5 DH |
1908 | ret = -EINVAL; |
1909 | reason = "memory holes"; | |
1910 | goto failed_removal; | |
1911 | } | |
1912 | ||
0c0e6195 KH |
1913 | /* This makes hotplug much easier...and readable. |
1914 | we assume this for now. .*/ | |
92917998 DH |
1915 | zone = test_pages_in_a_zone(start_pfn, end_pfn); |
1916 | if (!zone) { | |
79605093 MH |
1917 | ret = -EINVAL; |
1918 | reason = "multizone range"; | |
1919 | goto failed_removal; | |
381eab4a | 1920 | } |
7b78d335 | 1921 | node = zone_to_nid(zone); |
7b78d335 | 1922 | |
ec6e8c7e VB |
1923 | /* |
1924 | * Disable pcplists so that page isolation cannot race with freeing | |
1925 | * in a way that pages from isolated pageblock are left on pcplists. | |
1926 | */ | |
1927 | zone_pcp_disable(zone); | |
d479960e | 1928 | lru_cache_disable(); |
ec6e8c7e | 1929 | |
0c0e6195 | 1930 | /* set above range as isolated */ |
b023f468 | 1931 | ret = start_isolate_page_range(start_pfn, end_pfn, |
d381c547 | 1932 | MIGRATE_MOVABLE, |
756d25be | 1933 | MEMORY_OFFLINE | REPORT_FAILURE); |
3fa0c7c7 | 1934 | if (ret) { |
79605093 | 1935 | reason = "failure to isolate range"; |
ec6e8c7e | 1936 | goto failed_removal_pcplists_disabled; |
381eab4a | 1937 | } |
7b78d335 YG |
1938 | |
1939 | arg.start_pfn = start_pfn; | |
1940 | arg.nr_pages = nr_pages; | |
d9713679 | 1941 | node_states_check_changes_offline(nr_pages, zone, &arg); |
7b78d335 YG |
1942 | |
1943 | ret = memory_notify(MEM_GOING_OFFLINE, &arg); | |
1944 | ret = notifier_to_errno(ret); | |
79605093 MH |
1945 | if (ret) { |
1946 | reason = "notifier failure"; | |
1947 | goto failed_removal_isolated; | |
1948 | } | |
7b78d335 | 1949 | |
bb8965bd | 1950 | do { |
aa218795 DH |
1951 | pfn = start_pfn; |
1952 | do { | |
bb8965bd MH |
1953 | if (signal_pending(current)) { |
1954 | ret = -EINTR; | |
1955 | reason = "signal backoff"; | |
1956 | goto failed_removal_isolated; | |
1957 | } | |
72b39cfc | 1958 | |
bb8965bd | 1959 | cond_resched(); |
bb8965bd | 1960 | |
aa218795 DH |
1961 | ret = scan_movable_pages(pfn, end_pfn, &pfn); |
1962 | if (!ret) { | |
bb8965bd MH |
1963 | /* |
1964 | * TODO: fatal migration failures should bail | |
1965 | * out | |
1966 | */ | |
1967 | do_migrate_range(pfn, end_pfn); | |
1968 | } | |
aa218795 DH |
1969 | } while (!ret); |
1970 | ||
1971 | if (ret != -ENOENT) { | |
1972 | reason = "unmovable page"; | |
1973 | goto failed_removal_isolated; | |
bb8965bd | 1974 | } |
0c0e6195 | 1975 | |
bb8965bd MH |
1976 | /* |
1977 | * Dissolve free hugepages in the memory block before doing | |
1978 | * offlining actually in order to make hugetlbfs's object | |
1979 | * counting consistent. | |
1980 | */ | |
1981 | ret = dissolve_free_huge_pages(start_pfn, end_pfn); | |
1982 | if (ret) { | |
1983 | reason = "failure to dissolve huge pages"; | |
1984 | goto failed_removal_isolated; | |
1985 | } | |
0a1a9a00 | 1986 | |
0a1a9a00 | 1987 | ret = test_pages_isolated(start_pfn, end_pfn, MEMORY_OFFLINE); |
ec6e8c7e | 1988 | |
5557c766 | 1989 | } while (ret); |
72b39cfc | 1990 | |
0a1a9a00 DH |
1991 | /* Mark all sections offline and remove free pages from the buddy. */ |
1992 | __offline_isolated_pages(start_pfn, end_pfn); | |
7c33023a | 1993 | pr_debug("Offlined Pages %ld\n", nr_pages); |
0a1a9a00 | 1994 | |
9b7ea46a | 1995 | /* |
b30c5927 DH |
1996 | * The memory sections are marked offline, and the pageblock flags |
1997 | * effectively stale; nobody should be touching them. Fixup the number | |
1998 | * of isolated pageblocks, memory onlining will properly revert this. | |
9b7ea46a QC |
1999 | */ |
2000 | spin_lock_irqsave(&zone->lock, flags); | |
ea15153c | 2001 | zone->nr_isolate_pageblock -= nr_pages / pageblock_nr_pages; |
9b7ea46a QC |
2002 | spin_unlock_irqrestore(&zone->lock, flags); |
2003 | ||
d479960e | 2004 | lru_cache_enable(); |
ec6e8c7e VB |
2005 | zone_pcp_enable(zone); |
2006 | ||
0c0e6195 | 2007 | /* removal success */ |
0a1a9a00 | 2008 | adjust_managed_page_count(pfn_to_page(start_pfn), -nr_pages); |
836809ec | 2009 | adjust_present_page_count(pfn_to_page(start_pfn), group, -nr_pages); |
7b78d335 | 2010 | |
b92ca18e | 2011 | /* reinitialise watermarks and update pcp limits */ |
1b79acc9 KM |
2012 | init_per_zone_wmark_min(); |
2013 | ||
1e8537ba | 2014 | if (!populated_zone(zone)) { |
340175b7 | 2015 | zone_pcp_reset(zone); |
72675e13 | 2016 | build_all_zonelists(NULL); |
b92ca18e | 2017 | } |
340175b7 | 2018 | |
d9713679 | 2019 | node_states_clear_node(node, &arg); |
698b1b30 | 2020 | if (arg.status_change_nid >= 0) { |
8fe23e05 | 2021 | kswapd_stop(node); |
698b1b30 VB |
2022 | kcompactd_stop(node); |
2023 | } | |
bce7394a | 2024 | |
0c0e6195 | 2025 | writeback_set_ratelimit(); |
7b78d335 YG |
2026 | |
2027 | memory_notify(MEM_OFFLINE, &arg); | |
feee6b29 | 2028 | remove_pfn_range_from_zone(zone, start_pfn, nr_pages); |
381eab4a | 2029 | mem_hotplug_done(); |
0c0e6195 KH |
2030 | return 0; |
2031 | ||
79605093 MH |
2032 | failed_removal_isolated: |
2033 | undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE); | |
c4efe484 | 2034 | memory_notify(MEM_CANCEL_OFFLINE, &arg); |
ec6e8c7e | 2035 | failed_removal_pcplists_disabled: |
946746d1 | 2036 | lru_cache_enable(); |
ec6e8c7e | 2037 | zone_pcp_enable(zone); |
0c0e6195 | 2038 | failed_removal: |
79605093 | 2039 | pr_debug("memory offlining [mem %#010llx-%#010llx] failed due to %s\n", |
e33e33b4 | 2040 | (unsigned long long) start_pfn << PAGE_SHIFT, |
79605093 MH |
2041 | ((unsigned long long) end_pfn << PAGE_SHIFT) - 1, |
2042 | reason); | |
0c0e6195 | 2043 | /* pushback to free area */ |
381eab4a | 2044 | mem_hotplug_done(); |
0c0e6195 KH |
2045 | return ret; |
2046 | } | |
71088785 | 2047 | |
d6de9d53 | 2048 | static int check_memblock_offlined_cb(struct memory_block *mem, void *arg) |
bbc76be6 WC |
2049 | { |
2050 | int ret = !is_memblock_offlined(mem); | |
e1c158e4 | 2051 | int *nid = arg; |
bbc76be6 | 2052 | |
e1c158e4 | 2053 | *nid = mem->nid; |
349daa0f RD |
2054 | if (unlikely(ret)) { |
2055 | phys_addr_t beginpa, endpa; | |
2056 | ||
2057 | beginpa = PFN_PHYS(section_nr_to_pfn(mem->start_section_nr)); | |
b6c88d3b | 2058 | endpa = beginpa + memory_block_size_bytes() - 1; |
756a025f | 2059 | pr_warn("removing memory fails, because memory [%pa-%pa] is onlined\n", |
349daa0f | 2060 | &beginpa, &endpa); |
bbc76be6 | 2061 | |
eca499ab PT |
2062 | return -EBUSY; |
2063 | } | |
2064 | return 0; | |
bbc76be6 WC |
2065 | } |
2066 | ||
a08a2ae3 OS |
2067 | static int get_nr_vmemmap_pages_cb(struct memory_block *mem, void *arg) |
2068 | { | |
2069 | /* | |
2070 | * If not set, continue with the next block. | |
2071 | */ | |
2072 | return mem->nr_vmemmap_pages; | |
2073 | } | |
2074 | ||
0f1cfe9d | 2075 | static int check_cpu_on_node(pg_data_t *pgdat) |
60a5a19e | 2076 | { |
60a5a19e TC |
2077 | int cpu; |
2078 | ||
2079 | for_each_present_cpu(cpu) { | |
2080 | if (cpu_to_node(cpu) == pgdat->node_id) | |
2081 | /* | |
2082 | * the cpu on this node isn't removed, and we can't | |
2083 | * offline this node. | |
2084 | */ | |
2085 | return -EBUSY; | |
2086 | } | |
2087 | ||
2088 | return 0; | |
2089 | } | |
2090 | ||
2c91f8fc DH |
2091 | static int check_no_memblock_for_node_cb(struct memory_block *mem, void *arg) |
2092 | { | |
2093 | int nid = *(int *)arg; | |
2094 | ||
2095 | /* | |
2096 | * If a memory block belongs to multiple nodes, the stored nid is not | |
2097 | * reliable. However, such blocks are always online (e.g., cannot get | |
2098 | * offlined) and, therefore, are still spanned by the node. | |
2099 | */ | |
2100 | return mem->nid == nid ? -EEXIST : 0; | |
2101 | } | |
2102 | ||
0f1cfe9d TK |
2103 | /** |
2104 | * try_offline_node | |
e8b098fc | 2105 | * @nid: the node ID |
0f1cfe9d TK |
2106 | * |
2107 | * Offline a node if all memory sections and cpus of the node are removed. | |
2108 | * | |
2109 | * NOTE: The caller must call lock_device_hotplug() to serialize hotplug | |
2110 | * and online/offline operations before this call. | |
2111 | */ | |
90b30cdc | 2112 | void try_offline_node(int nid) |
60a5a19e | 2113 | { |
d822b86a | 2114 | pg_data_t *pgdat = NODE_DATA(nid); |
2c91f8fc | 2115 | int rc; |
60a5a19e | 2116 | |
2c91f8fc DH |
2117 | /* |
2118 | * If the node still spans pages (especially ZONE_DEVICE), don't | |
2119 | * offline it. A node spans memory after move_pfn_range_to_zone(), | |
2120 | * e.g., after the memory block was onlined. | |
2121 | */ | |
2122 | if (pgdat->node_spanned_pages) | |
2123 | return; | |
60a5a19e | 2124 | |
2c91f8fc DH |
2125 | /* |
2126 | * Especially offline memory blocks might not be spanned by the | |
2127 | * node. They will get spanned by the node once they get onlined. | |
2128 | * However, they link to the node in sysfs and can get onlined later. | |
2129 | */ | |
2130 | rc = for_each_memory_block(&nid, check_no_memblock_for_node_cb); | |
2131 | if (rc) | |
60a5a19e | 2132 | return; |
60a5a19e | 2133 | |
46a3679b | 2134 | if (check_cpu_on_node(pgdat)) |
60a5a19e TC |
2135 | return; |
2136 | ||
2137 | /* | |
2138 | * all memory/cpu of this node are removed, we can offline this | |
2139 | * node now. | |
2140 | */ | |
2141 | node_set_offline(nid); | |
2142 | unregister_one_node(nid); | |
2143 | } | |
90b30cdc | 2144 | EXPORT_SYMBOL(try_offline_node); |
60a5a19e | 2145 | |
e1c158e4 | 2146 | static int __ref try_remove_memory(u64 start, u64 size) |
bbc76be6 | 2147 | { |
a08a2ae3 OS |
2148 | struct vmem_altmap mhp_altmap = {}; |
2149 | struct vmem_altmap *altmap = NULL; | |
2150 | unsigned long nr_vmemmap_pages; | |
e1c158e4 | 2151 | int rc = 0, nid = NUMA_NO_NODE; |
993c1aad | 2152 | |
27356f54 TK |
2153 | BUG_ON(check_hotplug_memory_range(start, size)); |
2154 | ||
6677e3ea | 2155 | /* |
242831eb | 2156 | * All memory blocks must be offlined before removing memory. Check |
eca499ab | 2157 | * whether all memory blocks in question are offline and return error |
242831eb | 2158 | * if this is not the case. |
e1c158e4 DH |
2159 | * |
2160 | * While at it, determine the nid. Note that if we'd have mixed nodes, | |
2161 | * we'd only try to offline the last determined one -- which is good | |
2162 | * enough for the cases we care about. | |
6677e3ea | 2163 | */ |
e1c158e4 | 2164 | rc = walk_memory_blocks(start, size, &nid, check_memblock_offlined_cb); |
eca499ab | 2165 | if (rc) |
b4223a51 | 2166 | return rc; |
6677e3ea | 2167 | |
a08a2ae3 OS |
2168 | /* |
2169 | * We only support removing memory added with MHP_MEMMAP_ON_MEMORY in | |
2170 | * the same granularity it was added - a single memory block. | |
2171 | */ | |
2172 | if (memmap_on_memory) { | |
2173 | nr_vmemmap_pages = walk_memory_blocks(start, size, NULL, | |
2174 | get_nr_vmemmap_pages_cb); | |
2175 | if (nr_vmemmap_pages) { | |
2176 | if (size != memory_block_size_bytes()) { | |
2177 | pr_warn("Refuse to remove %#llx - %#llx," | |
2178 | "wrong granularity\n", | |
2179 | start, start + size); | |
2180 | return -EINVAL; | |
2181 | } | |
2182 | ||
2183 | /* | |
2184 | * Let remove_pmd_table->free_hugepage_table do the | |
2185 | * right thing if we used vmem_altmap when hot-adding | |
2186 | * the range. | |
2187 | */ | |
2188 | mhp_altmap.alloc = nr_vmemmap_pages; | |
2189 | altmap = &mhp_altmap; | |
2190 | } | |
2191 | } | |
2192 | ||
46c66c4b YI |
2193 | /* remove memmap entry */ |
2194 | firmware_map_remove(start, start + size, "System RAM"); | |
4c4b7f9b | 2195 | |
f1037ec0 DW |
2196 | /* |
2197 | * Memory block device removal under the device_hotplug_lock is | |
2198 | * a barrier against racing online attempts. | |
2199 | */ | |
4c4b7f9b | 2200 | remove_memory_block_devices(start, size); |
46c66c4b | 2201 | |
f1037ec0 DW |
2202 | mem_hotplug_begin(); |
2203 | ||
65a2aa5f | 2204 | arch_remove_memory(start, size, altmap); |
52219aea DH |
2205 | |
2206 | if (IS_ENABLED(CONFIG_ARCH_KEEP_MEMBLOCK)) { | |
2207 | memblock_free(start, size); | |
2208 | memblock_remove(start, size); | |
2209 | } | |
2210 | ||
cb8e3c8b | 2211 | release_mem_region_adjustable(start, size); |
24d335ca | 2212 | |
e1c158e4 DH |
2213 | if (nid != NUMA_NO_NODE) |
2214 | try_offline_node(nid); | |
60a5a19e | 2215 | |
bfc8c901 | 2216 | mem_hotplug_done(); |
b4223a51 | 2217 | return 0; |
71088785 | 2218 | } |
d15e5926 | 2219 | |
eca499ab | 2220 | /** |
5640c9ca | 2221 | * __remove_memory - Remove memory if every memory block is offline |
eca499ab PT |
2222 | * @start: physical address of the region to remove |
2223 | * @size: size of the region to remove | |
2224 | * | |
2225 | * NOTE: The caller must call lock_device_hotplug() to serialize hotplug | |
2226 | * and online/offline operations before this call, as required by | |
2227 | * try_offline_node(). | |
2228 | */ | |
e1c158e4 | 2229 | void __remove_memory(u64 start, u64 size) |
eca499ab PT |
2230 | { |
2231 | ||
2232 | /* | |
29a90db9 | 2233 | * trigger BUG() if some memory is not offlined prior to calling this |
eca499ab PT |
2234 | * function |
2235 | */ | |
e1c158e4 | 2236 | if (try_remove_memory(start, size)) |
eca499ab PT |
2237 | BUG(); |
2238 | } | |
2239 | ||
2240 | /* | |
2241 | * Remove memory if every memory block is offline, otherwise return -EBUSY is | |
2242 | * some memory is not offline | |
2243 | */ | |
e1c158e4 | 2244 | int remove_memory(u64 start, u64 size) |
d15e5926 | 2245 | { |
eca499ab PT |
2246 | int rc; |
2247 | ||
d15e5926 | 2248 | lock_device_hotplug(); |
e1c158e4 | 2249 | rc = try_remove_memory(start, size); |
d15e5926 | 2250 | unlock_device_hotplug(); |
eca499ab PT |
2251 | |
2252 | return rc; | |
d15e5926 | 2253 | } |
71088785 | 2254 | EXPORT_SYMBOL_GPL(remove_memory); |
08b3acd7 | 2255 | |
8dc4bb58 DH |
2256 | static int try_offline_memory_block(struct memory_block *mem, void *arg) |
2257 | { | |
2258 | uint8_t online_type = MMOP_ONLINE_KERNEL; | |
2259 | uint8_t **online_types = arg; | |
2260 | struct page *page; | |
2261 | int rc; | |
2262 | ||
2263 | /* | |
2264 | * Sense the online_type via the zone of the memory block. Offlining | |
2265 | * with multiple zones within one memory block will be rejected | |
2266 | * by offlining code ... so we don't care about that. | |
2267 | */ | |
2268 | page = pfn_to_online_page(section_nr_to_pfn(mem->start_section_nr)); | |
2269 | if (page && zone_idx(page_zone(page)) == ZONE_MOVABLE) | |
2270 | online_type = MMOP_ONLINE_MOVABLE; | |
2271 | ||
2272 | rc = device_offline(&mem->dev); | |
2273 | /* | |
2274 | * Default is MMOP_OFFLINE - change it only if offlining succeeded, | |
2275 | * so try_reonline_memory_block() can do the right thing. | |
2276 | */ | |
2277 | if (!rc) | |
2278 | **online_types = online_type; | |
2279 | ||
2280 | (*online_types)++; | |
2281 | /* Ignore if already offline. */ | |
2282 | return rc < 0 ? rc : 0; | |
2283 | } | |
2284 | ||
2285 | static int try_reonline_memory_block(struct memory_block *mem, void *arg) | |
2286 | { | |
2287 | uint8_t **online_types = arg; | |
2288 | int rc; | |
2289 | ||
2290 | if (**online_types != MMOP_OFFLINE) { | |
2291 | mem->online_type = **online_types; | |
2292 | rc = device_online(&mem->dev); | |
2293 | if (rc < 0) | |
2294 | pr_warn("%s: Failed to re-online memory: %d", | |
2295 | __func__, rc); | |
2296 | } | |
2297 | ||
2298 | /* Continue processing all remaining memory blocks. */ | |
2299 | (*online_types)++; | |
2300 | return 0; | |
2301 | } | |
2302 | ||
08b3acd7 | 2303 | /* |
8dc4bb58 DH |
2304 | * Try to offline and remove memory. Might take a long time to finish in case |
2305 | * memory is still in use. Primarily useful for memory devices that logically | |
2306 | * unplugged all memory (so it's no longer in use) and want to offline + remove | |
2307 | * that memory. | |
08b3acd7 | 2308 | */ |
e1c158e4 | 2309 | int offline_and_remove_memory(u64 start, u64 size) |
08b3acd7 | 2310 | { |
8dc4bb58 DH |
2311 | const unsigned long mb_count = size / memory_block_size_bytes(); |
2312 | uint8_t *online_types, *tmp; | |
2313 | int rc; | |
08b3acd7 DH |
2314 | |
2315 | if (!IS_ALIGNED(start, memory_block_size_bytes()) || | |
8dc4bb58 DH |
2316 | !IS_ALIGNED(size, memory_block_size_bytes()) || !size) |
2317 | return -EINVAL; | |
2318 | ||
2319 | /* | |
2320 | * We'll remember the old online type of each memory block, so we can | |
2321 | * try to revert whatever we did when offlining one memory block fails | |
2322 | * after offlining some others succeeded. | |
2323 | */ | |
2324 | online_types = kmalloc_array(mb_count, sizeof(*online_types), | |
2325 | GFP_KERNEL); | |
2326 | if (!online_types) | |
2327 | return -ENOMEM; | |
2328 | /* | |
2329 | * Initialize all states to MMOP_OFFLINE, so when we abort processing in | |
2330 | * try_offline_memory_block(), we'll skip all unprocessed blocks in | |
2331 | * try_reonline_memory_block(). | |
2332 | */ | |
2333 | memset(online_types, MMOP_OFFLINE, mb_count); | |
08b3acd7 DH |
2334 | |
2335 | lock_device_hotplug(); | |
8dc4bb58 DH |
2336 | |
2337 | tmp = online_types; | |
2338 | rc = walk_memory_blocks(start, size, &tmp, try_offline_memory_block); | |
08b3acd7 DH |
2339 | |
2340 | /* | |
8dc4bb58 | 2341 | * In case we succeeded to offline all memory, remove it. |
08b3acd7 DH |
2342 | * This cannot fail as it cannot get onlined in the meantime. |
2343 | */ | |
2344 | if (!rc) { | |
e1c158e4 | 2345 | rc = try_remove_memory(start, size); |
8dc4bb58 DH |
2346 | if (rc) |
2347 | pr_err("%s: Failed to remove memory: %d", __func__, rc); | |
2348 | } | |
2349 | ||
2350 | /* | |
2351 | * Rollback what we did. While memory onlining might theoretically fail | |
2352 | * (nacked by a notifier), it barely ever happens. | |
2353 | */ | |
2354 | if (rc) { | |
2355 | tmp = online_types; | |
2356 | walk_memory_blocks(start, size, &tmp, | |
2357 | try_reonline_memory_block); | |
08b3acd7 DH |
2358 | } |
2359 | unlock_device_hotplug(); | |
2360 | ||
8dc4bb58 | 2361 | kfree(online_types); |
08b3acd7 DH |
2362 | return rc; |
2363 | } | |
2364 | EXPORT_SYMBOL_GPL(offline_and_remove_memory); | |
aba6efc4 | 2365 | #endif /* CONFIG_MEMORY_HOTREMOVE */ |